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 #define IN_LOOKUP_SHIFT 10
115 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
117 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
120 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
121 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
125 /* Statistics gathering. */
126 struct dentry_stat_t dentry_stat
= {
130 static DEFINE_PER_CPU(long, nr_dentry
);
131 static DEFINE_PER_CPU(long, nr_dentry_unused
);
133 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
136 * Here we resort to our own counters instead of using generic per-cpu counters
137 * for consistency with what the vfs inode code does. We are expected to harvest
138 * better code and performance by having our own specialized counters.
140 * Please note that the loop is done over all possible CPUs, not over all online
141 * CPUs. The reason for this is that we don't want to play games with CPUs going
142 * on and off. If one of them goes off, we will just keep their counters.
144 * glommer: See cffbc8a for details, and if you ever intend to change this,
145 * please update all vfs counters to match.
147 static long get_nr_dentry(void)
151 for_each_possible_cpu(i
)
152 sum
+= per_cpu(nr_dentry
, i
);
153 return sum
< 0 ? 0 : sum
;
156 static long get_nr_dentry_unused(void)
160 for_each_possible_cpu(i
)
161 sum
+= per_cpu(nr_dentry_unused
, i
);
162 return sum
< 0 ? 0 : sum
;
165 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
166 size_t *lenp
, loff_t
*ppos
)
168 dentry_stat
.nr_dentry
= get_nr_dentry();
169 dentry_stat
.nr_unused
= get_nr_dentry_unused();
170 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
175 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
176 * The strings are both count bytes long, and count is non-zero.
178 #ifdef CONFIG_DCACHE_WORD_ACCESS
180 #include <asm/word-at-a-time.h>
182 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
183 * aligned allocation for this particular component. We don't
184 * strictly need the load_unaligned_zeropad() safety, but it
185 * doesn't hurt either.
187 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
188 * need the careful unaligned handling.
190 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
192 unsigned long a
,b
,mask
;
195 a
= *(unsigned long *)cs
;
196 b
= load_unaligned_zeropad(ct
);
197 if (tcount
< sizeof(unsigned long))
199 if (unlikely(a
!= b
))
201 cs
+= sizeof(unsigned long);
202 ct
+= sizeof(unsigned long);
203 tcount
-= sizeof(unsigned long);
207 mask
= bytemask_from_count(tcount
);
208 return unlikely(!!((a
^ b
) & mask
));
213 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
227 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
230 * Be careful about RCU walk racing with rename:
231 * use 'lockless_dereference' to fetch the name pointer.
233 * NOTE! Even if a rename will mean that the length
234 * was not loaded atomically, we don't care. The
235 * RCU walk will check the sequence count eventually,
236 * and catch it. And we won't overrun the buffer,
237 * because we're reading the name pointer atomically,
238 * and a dentry name is guaranteed to be properly
239 * terminated with a NUL byte.
241 * End result: even if 'len' is wrong, we'll exit
242 * early because the data cannot match (there can
243 * be no NUL in the ct/tcount data)
245 const unsigned char *cs
= lockless_dereference(dentry
->d_name
.name
);
247 return dentry_string_cmp(cs
, ct
, tcount
);
250 struct external_name
{
253 struct rcu_head head
;
255 unsigned char name
[];
258 static inline struct external_name
*external_name(struct dentry
*dentry
)
260 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
263 static void __d_free(struct rcu_head
*head
)
265 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
267 kmem_cache_free(dentry_cache
, dentry
);
270 static void __d_free_external(struct rcu_head
*head
)
272 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
273 kfree(external_name(dentry
));
274 kmem_cache_free(dentry_cache
, dentry
);
277 static inline int dname_external(const struct dentry
*dentry
)
279 return dentry
->d_name
.name
!= dentry
->d_iname
;
282 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
288 dentry
->d_inode
= inode
;
289 flags
= READ_ONCE(dentry
->d_flags
);
290 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
292 WRITE_ONCE(dentry
->d_flags
, flags
);
295 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
297 unsigned flags
= READ_ONCE(dentry
->d_flags
);
299 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
300 WRITE_ONCE(dentry
->d_flags
, flags
);
301 dentry
->d_inode
= NULL
;
304 static void dentry_free(struct dentry
*dentry
)
306 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
307 if (unlikely(dname_external(dentry
))) {
308 struct external_name
*p
= external_name(dentry
);
309 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
310 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
314 /* if dentry was never visible to RCU, immediate free is OK */
315 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
316 __d_free(&dentry
->d_u
.d_rcu
);
318 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
322 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
323 * @dentry: the target dentry
324 * After this call, in-progress rcu-walk path lookup will fail. This
325 * should be called after unhashing, and after changing d_inode (if
326 * the dentry has not already been unhashed).
328 static inline void dentry_rcuwalk_invalidate(struct dentry
*dentry
)
330 lockdep_assert_held(&dentry
->d_lock
);
331 /* Go through am invalidation barrier */
332 write_seqcount_invalidate(&dentry
->d_seq
);
336 * Release the dentry's inode, using the filesystem
337 * d_iput() operation if defined.
339 static void dentry_unlink_inode(struct dentry
* dentry
)
340 __releases(dentry
->d_lock
)
341 __releases(dentry
->d_inode
->i_lock
)
343 struct inode
*inode
= dentry
->d_inode
;
344 bool hashed
= !d_unhashed(dentry
);
347 raw_write_seqcount_begin(&dentry
->d_seq
);
348 __d_clear_type_and_inode(dentry
);
349 hlist_del_init(&dentry
->d_u
.d_alias
);
351 raw_write_seqcount_end(&dentry
->d_seq
);
352 spin_unlock(&dentry
->d_lock
);
353 spin_unlock(&inode
->i_lock
);
355 fsnotify_inoderemove(inode
);
356 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
357 dentry
->d_op
->d_iput(dentry
, inode
);
363 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
364 * is in use - which includes both the "real" per-superblock
365 * LRU list _and_ the DCACHE_SHRINK_LIST use.
367 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
368 * on the shrink list (ie not on the superblock LRU list).
370 * The per-cpu "nr_dentry_unused" counters are updated with
371 * the DCACHE_LRU_LIST bit.
373 * These helper functions make sure we always follow the
374 * rules. d_lock must be held by the caller.
376 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
377 static void d_lru_add(struct dentry
*dentry
)
379 D_FLAG_VERIFY(dentry
, 0);
380 dentry
->d_flags
|= DCACHE_LRU_LIST
;
381 this_cpu_inc(nr_dentry_unused
);
382 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
385 static void d_lru_del(struct dentry
*dentry
)
387 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
388 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
389 this_cpu_dec(nr_dentry_unused
);
390 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
393 static void d_shrink_del(struct dentry
*dentry
)
395 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
396 list_del_init(&dentry
->d_lru
);
397 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
398 this_cpu_dec(nr_dentry_unused
);
401 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
403 D_FLAG_VERIFY(dentry
, 0);
404 list_add(&dentry
->d_lru
, list
);
405 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
406 this_cpu_inc(nr_dentry_unused
);
410 * These can only be called under the global LRU lock, ie during the
411 * callback for freeing the LRU list. "isolate" removes it from the
412 * LRU lists entirely, while shrink_move moves it to the indicated
415 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
417 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
418 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
419 this_cpu_dec(nr_dentry_unused
);
420 list_lru_isolate(lru
, &dentry
->d_lru
);
423 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
424 struct list_head
*list
)
426 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
427 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
428 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
432 * dentry_lru_(add|del)_list) must be called with d_lock held.
434 static void dentry_lru_add(struct dentry
*dentry
)
436 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
441 * d_drop - drop a dentry
442 * @dentry: dentry to drop
444 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
445 * be found through a VFS lookup any more. Note that this is different from
446 * deleting the dentry - d_delete will try to mark the dentry negative if
447 * possible, giving a successful _negative_ lookup, while d_drop will
448 * just make the cache lookup fail.
450 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
451 * reason (NFS timeouts or autofs deletes).
453 * __d_drop requires dentry->d_lock.
455 void __d_drop(struct dentry
*dentry
)
457 if (!d_unhashed(dentry
)) {
458 struct hlist_bl_head
*b
;
460 * Hashed dentries are normally on the dentry hashtable,
461 * with the exception of those newly allocated by
462 * d_obtain_alias, which are always IS_ROOT:
464 if (unlikely(IS_ROOT(dentry
)))
465 b
= &dentry
->d_sb
->s_anon
;
467 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
470 __hlist_bl_del(&dentry
->d_hash
);
471 dentry
->d_hash
.pprev
= NULL
;
473 dentry_rcuwalk_invalidate(dentry
);
476 EXPORT_SYMBOL(__d_drop
);
478 void d_drop(struct dentry
*dentry
)
480 spin_lock(&dentry
->d_lock
);
482 spin_unlock(&dentry
->d_lock
);
484 EXPORT_SYMBOL(d_drop
);
486 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
490 * Inform d_walk() and shrink_dentry_list() that we are no longer
491 * attached to the dentry tree
493 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
494 if (unlikely(list_empty(&dentry
->d_child
)))
496 __list_del_entry(&dentry
->d_child
);
498 * Cursors can move around the list of children. While we'd been
499 * a normal list member, it didn't matter - ->d_child.next would've
500 * been updated. However, from now on it won't be and for the
501 * things like d_walk() it might end up with a nasty surprise.
502 * Normally d_walk() doesn't care about cursors moving around -
503 * ->d_lock on parent prevents that and since a cursor has no children
504 * of its own, we get through it without ever unlocking the parent.
505 * There is one exception, though - if we ascend from a child that
506 * gets killed as soon as we unlock it, the next sibling is found
507 * using the value left in its ->d_child.next. And if _that_
508 * pointed to a cursor, and cursor got moved (e.g. by lseek())
509 * before d_walk() regains parent->d_lock, we'll end up skipping
510 * everything the cursor had been moved past.
512 * Solution: make sure that the pointer left behind in ->d_child.next
513 * points to something that won't be moving around. I.e. skip the
516 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
517 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
518 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
520 dentry
->d_child
.next
= next
->d_child
.next
;
524 static void __dentry_kill(struct dentry
*dentry
)
526 struct dentry
*parent
= NULL
;
527 bool can_free
= true;
528 if (!IS_ROOT(dentry
))
529 parent
= dentry
->d_parent
;
532 * The dentry is now unrecoverably dead to the world.
534 lockref_mark_dead(&dentry
->d_lockref
);
537 * inform the fs via d_prune that this dentry is about to be
538 * unhashed and destroyed.
540 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
541 dentry
->d_op
->d_prune(dentry
);
543 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
544 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
547 /* if it was on the hash then remove it */
549 dentry_unlist(dentry
, parent
);
551 spin_unlock(&parent
->d_lock
);
553 dentry_unlink_inode(dentry
);
555 spin_unlock(&dentry
->d_lock
);
556 this_cpu_dec(nr_dentry
);
557 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
558 dentry
->d_op
->d_release(dentry
);
560 spin_lock(&dentry
->d_lock
);
561 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
562 dentry
->d_flags
|= DCACHE_MAY_FREE
;
565 spin_unlock(&dentry
->d_lock
);
566 if (likely(can_free
))
571 * Finish off a dentry we've decided to kill.
572 * dentry->d_lock must be held, returns with it unlocked.
573 * If ref is non-zero, then decrement the refcount too.
574 * Returns dentry requiring refcount drop, or NULL if we're done.
576 static struct dentry
*dentry_kill(struct dentry
*dentry
)
577 __releases(dentry
->d_lock
)
579 struct inode
*inode
= dentry
->d_inode
;
580 struct dentry
*parent
= NULL
;
582 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
585 if (!IS_ROOT(dentry
)) {
586 parent
= dentry
->d_parent
;
587 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
589 spin_unlock(&inode
->i_lock
);
594 __dentry_kill(dentry
);
598 spin_unlock(&dentry
->d_lock
);
599 return dentry
; /* try again with same dentry */
602 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
604 struct dentry
*parent
= dentry
->d_parent
;
607 if (unlikely(dentry
->d_lockref
.count
< 0))
609 if (likely(spin_trylock(&parent
->d_lock
)))
612 spin_unlock(&dentry
->d_lock
);
614 parent
= ACCESS_ONCE(dentry
->d_parent
);
615 spin_lock(&parent
->d_lock
);
617 * We can't blindly lock dentry until we are sure
618 * that we won't violate the locking order.
619 * Any changes of dentry->d_parent must have
620 * been done with parent->d_lock held, so
621 * spin_lock() above is enough of a barrier
622 * for checking if it's still our child.
624 if (unlikely(parent
!= dentry
->d_parent
)) {
625 spin_unlock(&parent
->d_lock
);
629 if (parent
!= dentry
)
630 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
637 * Try to do a lockless dput(), and return whether that was successful.
639 * If unsuccessful, we return false, having already taken the dentry lock.
641 * The caller needs to hold the RCU read lock, so that the dentry is
642 * guaranteed to stay around even if the refcount goes down to zero!
644 static inline bool fast_dput(struct dentry
*dentry
)
647 unsigned int d_flags
;
650 * If we have a d_op->d_delete() operation, we sould not
651 * let the dentry count go to zero, so use "put_or_lock".
653 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
654 return lockref_put_or_lock(&dentry
->d_lockref
);
657 * .. otherwise, we can try to just decrement the
658 * lockref optimistically.
660 ret
= lockref_put_return(&dentry
->d_lockref
);
663 * If the lockref_put_return() failed due to the lock being held
664 * by somebody else, the fast path has failed. We will need to
665 * get the lock, and then check the count again.
667 if (unlikely(ret
< 0)) {
668 spin_lock(&dentry
->d_lock
);
669 if (dentry
->d_lockref
.count
> 1) {
670 dentry
->d_lockref
.count
--;
671 spin_unlock(&dentry
->d_lock
);
678 * If we weren't the last ref, we're done.
684 * Careful, careful. The reference count went down
685 * to zero, but we don't hold the dentry lock, so
686 * somebody else could get it again, and do another
687 * dput(), and we need to not race with that.
689 * However, there is a very special and common case
690 * where we don't care, because there is nothing to
691 * do: the dentry is still hashed, it does not have
692 * a 'delete' op, and it's referenced and already on
695 * NOTE! Since we aren't locked, these values are
696 * not "stable". However, it is sufficient that at
697 * some point after we dropped the reference the
698 * dentry was hashed and the flags had the proper
699 * value. Other dentry users may have re-gotten
700 * a reference to the dentry and change that, but
701 * our work is done - we can leave the dentry
702 * around with a zero refcount.
705 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
706 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
708 /* Nothing to do? Dropping the reference was all we needed? */
709 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
713 * Not the fast normal case? Get the lock. We've already decremented
714 * the refcount, but we'll need to re-check the situation after
717 spin_lock(&dentry
->d_lock
);
720 * Did somebody else grab a reference to it in the meantime, and
721 * we're no longer the last user after all? Alternatively, somebody
722 * else could have killed it and marked it dead. Either way, we
723 * don't need to do anything else.
725 if (dentry
->d_lockref
.count
) {
726 spin_unlock(&dentry
->d_lock
);
731 * Re-get the reference we optimistically dropped. We hold the
732 * lock, and we just tested that it was zero, so we can just
735 dentry
->d_lockref
.count
= 1;
743 * This is complicated by the fact that we do not want to put
744 * dentries that are no longer on any hash chain on the unused
745 * list: we'd much rather just get rid of them immediately.
747 * However, that implies that we have to traverse the dentry
748 * tree upwards to the parents which might _also_ now be
749 * scheduled for deletion (it may have been only waiting for
750 * its last child to go away).
752 * This tail recursion is done by hand as we don't want to depend
753 * on the compiler to always get this right (gcc generally doesn't).
754 * Real recursion would eat up our stack space.
758 * dput - release a dentry
759 * @dentry: dentry to release
761 * Release a dentry. This will drop the usage count and if appropriate
762 * call the dentry unlink method as well as removing it from the queues and
763 * releasing its resources. If the parent dentries were scheduled for release
764 * they too may now get deleted.
766 void dput(struct dentry
*dentry
)
768 if (unlikely(!dentry
))
775 if (likely(fast_dput(dentry
))) {
780 /* Slow case: now with the dentry lock held */
783 WARN_ON(d_in_lookup(dentry
));
785 /* Unreachable? Get rid of it */
786 if (unlikely(d_unhashed(dentry
)))
789 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
792 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
793 if (dentry
->d_op
->d_delete(dentry
))
797 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
798 dentry
->d_flags
|= DCACHE_REFERENCED
;
799 dentry_lru_add(dentry
);
801 dentry
->d_lockref
.count
--;
802 spin_unlock(&dentry
->d_lock
);
806 dentry
= dentry_kill(dentry
);
815 /* This must be called with d_lock held */
816 static inline void __dget_dlock(struct dentry
*dentry
)
818 dentry
->d_lockref
.count
++;
821 static inline void __dget(struct dentry
*dentry
)
823 lockref_get(&dentry
->d_lockref
);
826 struct dentry
*dget_parent(struct dentry
*dentry
)
832 * Do optimistic parent lookup without any
836 ret
= ACCESS_ONCE(dentry
->d_parent
);
837 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
839 if (likely(gotref
)) {
840 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
847 * Don't need rcu_dereference because we re-check it was correct under
851 ret
= dentry
->d_parent
;
852 spin_lock(&ret
->d_lock
);
853 if (unlikely(ret
!= dentry
->d_parent
)) {
854 spin_unlock(&ret
->d_lock
);
859 BUG_ON(!ret
->d_lockref
.count
);
860 ret
->d_lockref
.count
++;
861 spin_unlock(&ret
->d_lock
);
864 EXPORT_SYMBOL(dget_parent
);
867 * d_find_alias - grab a hashed alias of inode
868 * @inode: inode in question
870 * If inode has a hashed alias, or is a directory and has any alias,
871 * acquire the reference to alias and return it. Otherwise return NULL.
872 * Notice that if inode is a directory there can be only one alias and
873 * it can be unhashed only if it has no children, or if it is the root
874 * of a filesystem, or if the directory was renamed and d_revalidate
875 * was the first vfs operation to notice.
877 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
878 * any other hashed alias over that one.
880 static struct dentry
*__d_find_alias(struct inode
*inode
)
882 struct dentry
*alias
, *discon_alias
;
886 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
887 spin_lock(&alias
->d_lock
);
888 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
889 if (IS_ROOT(alias
) &&
890 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
891 discon_alias
= alias
;
894 spin_unlock(&alias
->d_lock
);
898 spin_unlock(&alias
->d_lock
);
901 alias
= discon_alias
;
902 spin_lock(&alias
->d_lock
);
903 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
905 spin_unlock(&alias
->d_lock
);
908 spin_unlock(&alias
->d_lock
);
914 struct dentry
*d_find_alias(struct inode
*inode
)
916 struct dentry
*de
= NULL
;
918 if (!hlist_empty(&inode
->i_dentry
)) {
919 spin_lock(&inode
->i_lock
);
920 de
= __d_find_alias(inode
);
921 spin_unlock(&inode
->i_lock
);
925 EXPORT_SYMBOL(d_find_alias
);
928 * Try to kill dentries associated with this inode.
929 * WARNING: you must own a reference to inode.
931 void d_prune_aliases(struct inode
*inode
)
933 struct dentry
*dentry
;
935 spin_lock(&inode
->i_lock
);
936 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
937 spin_lock(&dentry
->d_lock
);
938 if (!dentry
->d_lockref
.count
) {
939 struct dentry
*parent
= lock_parent(dentry
);
940 if (likely(!dentry
->d_lockref
.count
)) {
941 __dentry_kill(dentry
);
946 spin_unlock(&parent
->d_lock
);
948 spin_unlock(&dentry
->d_lock
);
950 spin_unlock(&inode
->i_lock
);
952 EXPORT_SYMBOL(d_prune_aliases
);
954 static void shrink_dentry_list(struct list_head
*list
)
956 struct dentry
*dentry
, *parent
;
958 while (!list_empty(list
)) {
960 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
961 spin_lock(&dentry
->d_lock
);
962 parent
= lock_parent(dentry
);
965 * The dispose list is isolated and dentries are not accounted
966 * to the LRU here, so we can simply remove it from the list
967 * here regardless of whether it is referenced or not.
969 d_shrink_del(dentry
);
972 * We found an inuse dentry which was not removed from
973 * the LRU because of laziness during lookup. Do not free it.
975 if (dentry
->d_lockref
.count
> 0) {
976 spin_unlock(&dentry
->d_lock
);
978 spin_unlock(&parent
->d_lock
);
983 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
984 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
985 spin_unlock(&dentry
->d_lock
);
987 spin_unlock(&parent
->d_lock
);
993 inode
= dentry
->d_inode
;
994 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
995 d_shrink_add(dentry
, list
);
996 spin_unlock(&dentry
->d_lock
);
998 spin_unlock(&parent
->d_lock
);
1002 __dentry_kill(dentry
);
1005 * We need to prune ancestors too. This is necessary to prevent
1006 * quadratic behavior of shrink_dcache_parent(), but is also
1007 * expected to be beneficial in reducing dentry cache
1011 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1012 parent
= lock_parent(dentry
);
1013 if (dentry
->d_lockref
.count
!= 1) {
1014 dentry
->d_lockref
.count
--;
1015 spin_unlock(&dentry
->d_lock
);
1017 spin_unlock(&parent
->d_lock
);
1020 inode
= dentry
->d_inode
; /* can't be NULL */
1021 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1022 spin_unlock(&dentry
->d_lock
);
1024 spin_unlock(&parent
->d_lock
);
1028 __dentry_kill(dentry
);
1034 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1035 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1037 struct list_head
*freeable
= arg
;
1038 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1042 * we are inverting the lru lock/dentry->d_lock here,
1043 * so use a trylock. If we fail to get the lock, just skip
1046 if (!spin_trylock(&dentry
->d_lock
))
1050 * Referenced dentries are still in use. If they have active
1051 * counts, just remove them from the LRU. Otherwise give them
1052 * another pass through the LRU.
1054 if (dentry
->d_lockref
.count
) {
1055 d_lru_isolate(lru
, dentry
);
1056 spin_unlock(&dentry
->d_lock
);
1060 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1061 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1062 spin_unlock(&dentry
->d_lock
);
1065 * The list move itself will be made by the common LRU code. At
1066 * this point, we've dropped the dentry->d_lock but keep the
1067 * lru lock. This is safe to do, since every list movement is
1068 * protected by the lru lock even if both locks are held.
1070 * This is guaranteed by the fact that all LRU management
1071 * functions are intermediated by the LRU API calls like
1072 * list_lru_add and list_lru_del. List movement in this file
1073 * only ever occur through this functions or through callbacks
1074 * like this one, that are called from the LRU API.
1076 * The only exceptions to this are functions like
1077 * shrink_dentry_list, and code that first checks for the
1078 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1079 * operating only with stack provided lists after they are
1080 * properly isolated from the main list. It is thus, always a
1086 d_lru_shrink_move(lru
, dentry
, freeable
);
1087 spin_unlock(&dentry
->d_lock
);
1093 * prune_dcache_sb - shrink the dcache
1095 * @sc: shrink control, passed to list_lru_shrink_walk()
1097 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1098 * is done when we need more memory and called from the superblock shrinker
1101 * This function may fail to free any resources if all the dentries are in
1104 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1109 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1110 dentry_lru_isolate
, &dispose
);
1111 shrink_dentry_list(&dispose
);
1115 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1116 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1118 struct list_head
*freeable
= arg
;
1119 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1122 * we are inverting the lru lock/dentry->d_lock here,
1123 * so use a trylock. If we fail to get the lock, just skip
1126 if (!spin_trylock(&dentry
->d_lock
))
1129 d_lru_shrink_move(lru
, dentry
, freeable
);
1130 spin_unlock(&dentry
->d_lock
);
1137 * shrink_dcache_sb - shrink dcache for a superblock
1140 * Shrink the dcache for the specified super block. This is used to free
1141 * the dcache before unmounting a file system.
1143 void shrink_dcache_sb(struct super_block
*sb
)
1150 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1151 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
1153 this_cpu_sub(nr_dentry_unused
, freed
);
1154 shrink_dentry_list(&dispose
);
1155 } while (freed
> 0);
1157 EXPORT_SYMBOL(shrink_dcache_sb
);
1160 * enum d_walk_ret - action to talke during tree walk
1161 * @D_WALK_CONTINUE: contrinue walk
1162 * @D_WALK_QUIT: quit walk
1163 * @D_WALK_NORETRY: quit when retry is needed
1164 * @D_WALK_SKIP: skip this dentry and its children
1174 * d_walk - walk the dentry tree
1175 * @parent: start of walk
1176 * @data: data passed to @enter() and @finish()
1177 * @enter: callback when first entering the dentry
1178 * @finish: callback when successfully finished the walk
1180 * The @enter() and @finish() callbacks are called with d_lock held.
1182 static void d_walk(struct dentry
*parent
, void *data
,
1183 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1184 void (*finish
)(void *))
1186 struct dentry
*this_parent
;
1187 struct list_head
*next
;
1189 enum d_walk_ret ret
;
1193 read_seqbegin_or_lock(&rename_lock
, &seq
);
1194 this_parent
= parent
;
1195 spin_lock(&this_parent
->d_lock
);
1197 ret
= enter(data
, this_parent
);
1199 case D_WALK_CONTINUE
:
1204 case D_WALK_NORETRY
:
1209 next
= this_parent
->d_subdirs
.next
;
1211 while (next
!= &this_parent
->d_subdirs
) {
1212 struct list_head
*tmp
= next
;
1213 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1216 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1219 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1221 ret
= enter(data
, dentry
);
1223 case D_WALK_CONTINUE
:
1226 spin_unlock(&dentry
->d_lock
);
1228 case D_WALK_NORETRY
:
1232 spin_unlock(&dentry
->d_lock
);
1236 if (!list_empty(&dentry
->d_subdirs
)) {
1237 spin_unlock(&this_parent
->d_lock
);
1238 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1239 this_parent
= dentry
;
1240 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1243 spin_unlock(&dentry
->d_lock
);
1246 * All done at this level ... ascend and resume the search.
1250 if (this_parent
!= parent
) {
1251 struct dentry
*child
= this_parent
;
1252 this_parent
= child
->d_parent
;
1254 spin_unlock(&child
->d_lock
);
1255 spin_lock(&this_parent
->d_lock
);
1257 /* might go back up the wrong parent if we have had a rename. */
1258 if (need_seqretry(&rename_lock
, seq
))
1260 /* go into the first sibling still alive */
1262 next
= child
->d_child
.next
;
1263 if (next
== &this_parent
->d_subdirs
)
1265 child
= list_entry(next
, struct dentry
, d_child
);
1266 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1270 if (need_seqretry(&rename_lock
, seq
))
1277 spin_unlock(&this_parent
->d_lock
);
1278 done_seqretry(&rename_lock
, seq
);
1282 spin_unlock(&this_parent
->d_lock
);
1292 * Search for at least 1 mount point in the dentry's subdirs.
1293 * We descend to the next level whenever the d_subdirs
1294 * list is non-empty and continue searching.
1297 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1300 if (d_mountpoint(dentry
)) {
1304 return D_WALK_CONTINUE
;
1308 * have_submounts - check for mounts over a dentry
1309 * @parent: dentry to check.
1311 * Return true if the parent or its subdirectories contain
1314 int have_submounts(struct dentry
*parent
)
1318 d_walk(parent
, &ret
, check_mount
, NULL
);
1322 EXPORT_SYMBOL(have_submounts
);
1325 * Called by mount code to set a mountpoint and check if the mountpoint is
1326 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1327 * subtree can become unreachable).
1329 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1330 * this reason take rename_lock and d_lock on dentry and ancestors.
1332 int d_set_mounted(struct dentry
*dentry
)
1336 write_seqlock(&rename_lock
);
1337 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1338 /* Need exclusion wrt. d_invalidate() */
1339 spin_lock(&p
->d_lock
);
1340 if (unlikely(d_unhashed(p
))) {
1341 spin_unlock(&p
->d_lock
);
1344 spin_unlock(&p
->d_lock
);
1346 spin_lock(&dentry
->d_lock
);
1347 if (!d_unlinked(dentry
)) {
1348 dentry
->d_flags
|= DCACHE_MOUNTED
;
1351 spin_unlock(&dentry
->d_lock
);
1353 write_sequnlock(&rename_lock
);
1358 * Search the dentry child list of the specified parent,
1359 * and move any unused dentries to the end of the unused
1360 * list for prune_dcache(). We descend to the next level
1361 * whenever the d_subdirs list is non-empty and continue
1364 * It returns zero iff there are no unused children,
1365 * otherwise it returns the number of children moved to
1366 * the end of the unused list. This may not be the total
1367 * number of unused children, because select_parent can
1368 * drop the lock and return early due to latency
1372 struct select_data
{
1373 struct dentry
*start
;
1374 struct list_head dispose
;
1378 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1380 struct select_data
*data
= _data
;
1381 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1383 if (data
->start
== dentry
)
1386 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1389 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1391 if (!dentry
->d_lockref
.count
) {
1392 d_shrink_add(dentry
, &data
->dispose
);
1397 * We can return to the caller if we have found some (this
1398 * ensures forward progress). We'll be coming back to find
1401 if (!list_empty(&data
->dispose
))
1402 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1408 * shrink_dcache_parent - prune dcache
1409 * @parent: parent of entries to prune
1411 * Prune the dcache to remove unused children of the parent dentry.
1413 void shrink_dcache_parent(struct dentry
*parent
)
1416 struct select_data data
;
1418 INIT_LIST_HEAD(&data
.dispose
);
1419 data
.start
= parent
;
1422 d_walk(parent
, &data
, select_collect
, NULL
);
1426 shrink_dentry_list(&data
.dispose
);
1430 EXPORT_SYMBOL(shrink_dcache_parent
);
1432 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1434 /* it has busy descendents; complain about those instead */
1435 if (!list_empty(&dentry
->d_subdirs
))
1436 return D_WALK_CONTINUE
;
1438 /* root with refcount 1 is fine */
1439 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1440 return D_WALK_CONTINUE
;
1442 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1443 " still in use (%d) [unmount of %s %s]\n",
1446 dentry
->d_inode
->i_ino
: 0UL,
1448 dentry
->d_lockref
.count
,
1449 dentry
->d_sb
->s_type
->name
,
1450 dentry
->d_sb
->s_id
);
1452 return D_WALK_CONTINUE
;
1455 static void do_one_tree(struct dentry
*dentry
)
1457 shrink_dcache_parent(dentry
);
1458 d_walk(dentry
, dentry
, umount_check
, NULL
);
1464 * destroy the dentries attached to a superblock on unmounting
1466 void shrink_dcache_for_umount(struct super_block
*sb
)
1468 struct dentry
*dentry
;
1470 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1472 dentry
= sb
->s_root
;
1474 do_one_tree(dentry
);
1476 while (!hlist_bl_empty(&sb
->s_anon
)) {
1477 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1478 do_one_tree(dentry
);
1482 struct detach_data
{
1483 struct select_data select
;
1484 struct dentry
*mountpoint
;
1486 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1488 struct detach_data
*data
= _data
;
1490 if (d_mountpoint(dentry
)) {
1491 __dget_dlock(dentry
);
1492 data
->mountpoint
= dentry
;
1496 return select_collect(&data
->select
, dentry
);
1499 static void check_and_drop(void *_data
)
1501 struct detach_data
*data
= _data
;
1503 if (!data
->mountpoint
&& !data
->select
.found
)
1504 __d_drop(data
->select
.start
);
1508 * d_invalidate - detach submounts, prune dcache, and drop
1509 * @dentry: dentry to invalidate (aka detach, prune and drop)
1513 * The final d_drop is done as an atomic operation relative to
1514 * rename_lock ensuring there are no races with d_set_mounted. This
1515 * ensures there are no unhashed dentries on the path to a mountpoint.
1517 void d_invalidate(struct dentry
*dentry
)
1520 * If it's already been dropped, return OK.
1522 spin_lock(&dentry
->d_lock
);
1523 if (d_unhashed(dentry
)) {
1524 spin_unlock(&dentry
->d_lock
);
1527 spin_unlock(&dentry
->d_lock
);
1529 /* Negative dentries can be dropped without further checks */
1530 if (!dentry
->d_inode
) {
1536 struct detach_data data
;
1538 data
.mountpoint
= NULL
;
1539 INIT_LIST_HEAD(&data
.select
.dispose
);
1540 data
.select
.start
= dentry
;
1541 data
.select
.found
= 0;
1543 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1545 if (data
.select
.found
)
1546 shrink_dentry_list(&data
.select
.dispose
);
1548 if (data
.mountpoint
) {
1549 detach_mounts(data
.mountpoint
);
1550 dput(data
.mountpoint
);
1553 if (!data
.mountpoint
&& !data
.select
.found
)
1559 EXPORT_SYMBOL(d_invalidate
);
1562 * __d_alloc - allocate a dcache entry
1563 * @sb: filesystem it will belong to
1564 * @name: qstr of the name
1566 * Allocates a dentry. It returns %NULL if there is insufficient memory
1567 * available. On a success the dentry is returned. The name passed in is
1568 * copied and the copy passed in may be reused after this call.
1571 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1573 struct dentry
*dentry
;
1577 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1582 * We guarantee that the inline name is always NUL-terminated.
1583 * This way the memcpy() done by the name switching in rename
1584 * will still always have a NUL at the end, even if we might
1585 * be overwriting an internal NUL character
1587 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1588 if (unlikely(!name
)) {
1589 static const struct qstr anon
= QSTR_INIT("/", 1);
1591 dname
= dentry
->d_iname
;
1592 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1593 size_t size
= offsetof(struct external_name
, name
[1]);
1594 struct external_name
*p
= kmalloc(size
+ name
->len
,
1595 GFP_KERNEL_ACCOUNT
);
1597 kmem_cache_free(dentry_cache
, dentry
);
1600 atomic_set(&p
->u
.count
, 1);
1602 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1603 kasan_unpoison_shadow(dname
,
1604 round_up(name
->len
+ 1, sizeof(unsigned long)));
1606 dname
= dentry
->d_iname
;
1609 dentry
->d_name
.len
= name
->len
;
1610 dentry
->d_name
.hash
= name
->hash
;
1611 memcpy(dname
, name
->name
, name
->len
);
1612 dname
[name
->len
] = 0;
1614 /* Make sure we always see the terminating NUL character */
1616 dentry
->d_name
.name
= dname
;
1618 dentry
->d_lockref
.count
= 1;
1619 dentry
->d_flags
= 0;
1620 spin_lock_init(&dentry
->d_lock
);
1621 seqcount_init(&dentry
->d_seq
);
1622 dentry
->d_inode
= NULL
;
1623 dentry
->d_parent
= dentry
;
1625 dentry
->d_op
= NULL
;
1626 dentry
->d_fsdata
= NULL
;
1627 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1628 INIT_LIST_HEAD(&dentry
->d_lru
);
1629 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1630 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1631 INIT_LIST_HEAD(&dentry
->d_child
);
1632 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1634 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1635 err
= dentry
->d_op
->d_init(dentry
);
1637 if (dname_external(dentry
))
1638 kfree(external_name(dentry
));
1639 kmem_cache_free(dentry_cache
, dentry
);
1644 this_cpu_inc(nr_dentry
);
1650 * d_alloc - allocate a dcache entry
1651 * @parent: parent of entry to allocate
1652 * @name: qstr of the name
1654 * Allocates a dentry. It returns %NULL if there is insufficient memory
1655 * available. On a success the dentry is returned. The name passed in is
1656 * copied and the copy passed in may be reused after this call.
1658 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1660 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1663 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1664 spin_lock(&parent
->d_lock
);
1666 * don't need child lock because it is not subject
1667 * to concurrency here
1669 __dget_dlock(parent
);
1670 dentry
->d_parent
= parent
;
1671 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1672 spin_unlock(&parent
->d_lock
);
1676 EXPORT_SYMBOL(d_alloc
);
1678 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1680 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1682 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1683 dentry
->d_parent
= dget(parent
);
1689 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1690 * @sb: the superblock
1691 * @name: qstr of the name
1693 * For a filesystem that just pins its dentries in memory and never
1694 * performs lookups at all, return an unhashed IS_ROOT dentry.
1696 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1698 return __d_alloc(sb
, name
);
1700 EXPORT_SYMBOL(d_alloc_pseudo
);
1702 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1707 q
.hash_len
= hashlen_string(name
);
1708 return d_alloc(parent
, &q
);
1710 EXPORT_SYMBOL(d_alloc_name
);
1712 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1714 WARN_ON_ONCE(dentry
->d_op
);
1715 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1717 DCACHE_OP_REVALIDATE
|
1718 DCACHE_OP_WEAK_REVALIDATE
|
1725 dentry
->d_flags
|= DCACHE_OP_HASH
;
1727 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1728 if (op
->d_revalidate
)
1729 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1730 if (op
->d_weak_revalidate
)
1731 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1733 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1735 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1737 dentry
->d_flags
|= DCACHE_OP_REAL
;
1740 EXPORT_SYMBOL(d_set_d_op
);
1744 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1745 * @dentry - The dentry to mark
1747 * Mark a dentry as falling through to the lower layer (as set with
1748 * d_pin_lower()). This flag may be recorded on the medium.
1750 void d_set_fallthru(struct dentry
*dentry
)
1752 spin_lock(&dentry
->d_lock
);
1753 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1754 spin_unlock(&dentry
->d_lock
);
1756 EXPORT_SYMBOL(d_set_fallthru
);
1758 static unsigned d_flags_for_inode(struct inode
*inode
)
1760 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1763 return DCACHE_MISS_TYPE
;
1765 if (S_ISDIR(inode
->i_mode
)) {
1766 add_flags
= DCACHE_DIRECTORY_TYPE
;
1767 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1768 if (unlikely(!inode
->i_op
->lookup
))
1769 add_flags
= DCACHE_AUTODIR_TYPE
;
1771 inode
->i_opflags
|= IOP_LOOKUP
;
1773 goto type_determined
;
1776 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1777 if (unlikely(inode
->i_op
->get_link
)) {
1778 add_flags
= DCACHE_SYMLINK_TYPE
;
1779 goto type_determined
;
1781 inode
->i_opflags
|= IOP_NOFOLLOW
;
1784 if (unlikely(!S_ISREG(inode
->i_mode
)))
1785 add_flags
= DCACHE_SPECIAL_TYPE
;
1788 if (unlikely(IS_AUTOMOUNT(inode
)))
1789 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1793 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1795 unsigned add_flags
= d_flags_for_inode(inode
);
1796 WARN_ON(d_in_lookup(dentry
));
1798 spin_lock(&dentry
->d_lock
);
1799 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1800 raw_write_seqcount_begin(&dentry
->d_seq
);
1801 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1802 raw_write_seqcount_end(&dentry
->d_seq
);
1803 fsnotify_update_flags(dentry
);
1804 spin_unlock(&dentry
->d_lock
);
1808 * d_instantiate - fill in inode information for a dentry
1809 * @entry: dentry to complete
1810 * @inode: inode to attach to this dentry
1812 * Fill in inode information in the entry.
1814 * This turns negative dentries into productive full members
1817 * NOTE! This assumes that the inode count has been incremented
1818 * (or otherwise set) by the caller to indicate that it is now
1819 * in use by the dcache.
1822 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1824 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1826 security_d_instantiate(entry
, inode
);
1827 spin_lock(&inode
->i_lock
);
1828 __d_instantiate(entry
, inode
);
1829 spin_unlock(&inode
->i_lock
);
1832 EXPORT_SYMBOL(d_instantiate
);
1835 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1836 * @entry: dentry to complete
1837 * @inode: inode to attach to this dentry
1839 * Fill in inode information in the entry. If a directory alias is found, then
1840 * return an error (and drop inode). Together with d_materialise_unique() this
1841 * guarantees that a directory inode may never have more than one alias.
1843 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1845 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1847 security_d_instantiate(entry
, inode
);
1848 spin_lock(&inode
->i_lock
);
1849 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1850 spin_unlock(&inode
->i_lock
);
1854 __d_instantiate(entry
, inode
);
1855 spin_unlock(&inode
->i_lock
);
1859 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1861 struct dentry
*d_make_root(struct inode
*root_inode
)
1863 struct dentry
*res
= NULL
;
1866 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1868 d_instantiate(res
, root_inode
);
1874 EXPORT_SYMBOL(d_make_root
);
1876 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1878 struct dentry
*alias
;
1880 if (hlist_empty(&inode
->i_dentry
))
1882 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1888 * d_find_any_alias - find any alias for a given inode
1889 * @inode: inode to find an alias for
1891 * If any aliases exist for the given inode, take and return a
1892 * reference for one of them. If no aliases exist, return %NULL.
1894 struct dentry
*d_find_any_alias(struct inode
*inode
)
1898 spin_lock(&inode
->i_lock
);
1899 de
= __d_find_any_alias(inode
);
1900 spin_unlock(&inode
->i_lock
);
1903 EXPORT_SYMBOL(d_find_any_alias
);
1905 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1912 return ERR_PTR(-ESTALE
);
1914 return ERR_CAST(inode
);
1916 res
= d_find_any_alias(inode
);
1920 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1922 res
= ERR_PTR(-ENOMEM
);
1926 security_d_instantiate(tmp
, inode
);
1927 spin_lock(&inode
->i_lock
);
1928 res
= __d_find_any_alias(inode
);
1930 spin_unlock(&inode
->i_lock
);
1935 /* attach a disconnected dentry */
1936 add_flags
= d_flags_for_inode(inode
);
1939 add_flags
|= DCACHE_DISCONNECTED
;
1941 spin_lock(&tmp
->d_lock
);
1942 __d_set_inode_and_type(tmp
, inode
, add_flags
);
1943 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1944 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1945 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1946 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1947 spin_unlock(&tmp
->d_lock
);
1948 spin_unlock(&inode
->i_lock
);
1958 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1959 * @inode: inode to allocate the dentry for
1961 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1962 * similar open by handle operations. The returned dentry may be anonymous,
1963 * or may have a full name (if the inode was already in the cache).
1965 * When called on a directory inode, we must ensure that the inode only ever
1966 * has one dentry. If a dentry is found, that is returned instead of
1967 * allocating a new one.
1969 * On successful return, the reference to the inode has been transferred
1970 * to the dentry. In case of an error the reference on the inode is released.
1971 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1972 * be passed in and the error will be propagated to the return value,
1973 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1975 struct dentry
*d_obtain_alias(struct inode
*inode
)
1977 return __d_obtain_alias(inode
, 1);
1979 EXPORT_SYMBOL(d_obtain_alias
);
1982 * d_obtain_root - find or allocate a dentry for a given inode
1983 * @inode: inode to allocate the dentry for
1985 * Obtain an IS_ROOT dentry for the root of a filesystem.
1987 * We must ensure that directory inodes only ever have one dentry. If a
1988 * dentry is found, that is returned instead of allocating a new one.
1990 * On successful return, the reference to the inode has been transferred
1991 * to the dentry. In case of an error the reference on the inode is
1992 * released. A %NULL or IS_ERR inode may be passed in and will be the
1993 * error will be propagate to the return value, with a %NULL @inode
1994 * replaced by ERR_PTR(-ESTALE).
1996 struct dentry
*d_obtain_root(struct inode
*inode
)
1998 return __d_obtain_alias(inode
, 0);
2000 EXPORT_SYMBOL(d_obtain_root
);
2003 * d_add_ci - lookup or allocate new dentry with case-exact name
2004 * @inode: the inode case-insensitive lookup has found
2005 * @dentry: the negative dentry that was passed to the parent's lookup func
2006 * @name: the case-exact name to be associated with the returned dentry
2008 * This is to avoid filling the dcache with case-insensitive names to the
2009 * same inode, only the actual correct case is stored in the dcache for
2010 * case-insensitive filesystems.
2012 * For a case-insensitive lookup match and if the the case-exact dentry
2013 * already exists in in the dcache, use it and return it.
2015 * If no entry exists with the exact case name, allocate new dentry with
2016 * the exact case, and return the spliced entry.
2018 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2021 struct dentry
*found
, *res
;
2024 * First check if a dentry matching the name already exists,
2025 * if not go ahead and create it now.
2027 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2032 if (d_in_lookup(dentry
)) {
2033 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2035 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2040 found
= d_alloc(dentry
->d_parent
, name
);
2043 return ERR_PTR(-ENOMEM
);
2046 res
= d_splice_alias(inode
, found
);
2053 EXPORT_SYMBOL(d_add_ci
);
2056 static inline bool d_same_name(const struct dentry
*dentry
,
2057 const struct dentry
*parent
,
2058 const struct qstr
*name
)
2060 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2061 if (dentry
->d_name
.len
!= name
->len
)
2063 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2065 return parent
->d_op
->d_compare(parent
, dentry
,
2066 dentry
->d_name
.len
, dentry
->d_name
.name
,
2071 * __d_lookup_rcu - search for a dentry (racy, store-free)
2072 * @parent: parent dentry
2073 * @name: qstr of name we wish to find
2074 * @seqp: returns d_seq value at the point where the dentry was found
2075 * Returns: dentry, or NULL
2077 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2078 * resolution (store-free path walking) design described in
2079 * Documentation/filesystems/path-lookup.txt.
2081 * This is not to be used outside core vfs.
2083 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2084 * held, and rcu_read_lock held. The returned dentry must not be stored into
2085 * without taking d_lock and checking d_seq sequence count against @seq
2088 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2091 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2092 * the returned dentry, so long as its parent's seqlock is checked after the
2093 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2094 * is formed, giving integrity down the path walk.
2096 * NOTE! The caller *has* to check the resulting dentry against the sequence
2097 * number we've returned before using any of the resulting dentry state!
2099 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2100 const struct qstr
*name
,
2103 u64 hashlen
= name
->hash_len
;
2104 const unsigned char *str
= name
->name
;
2105 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2106 struct hlist_bl_node
*node
;
2107 struct dentry
*dentry
;
2110 * Note: There is significant duplication with __d_lookup_rcu which is
2111 * required to prevent single threaded performance regressions
2112 * especially on architectures where smp_rmb (in seqcounts) are costly.
2113 * Keep the two functions in sync.
2117 * The hash list is protected using RCU.
2119 * Carefully use d_seq when comparing a candidate dentry, to avoid
2120 * races with d_move().
2122 * It is possible that concurrent renames can mess up our list
2123 * walk here and result in missing our dentry, resulting in the
2124 * false-negative result. d_lookup() protects against concurrent
2125 * renames using rename_lock seqlock.
2127 * See Documentation/filesystems/path-lookup.txt for more details.
2129 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2134 * The dentry sequence count protects us from concurrent
2135 * renames, and thus protects parent and name fields.
2137 * The caller must perform a seqcount check in order
2138 * to do anything useful with the returned dentry.
2140 * NOTE! We do a "raw" seqcount_begin here. That means that
2141 * we don't wait for the sequence count to stabilize if it
2142 * is in the middle of a sequence change. If we do the slow
2143 * dentry compare, we will do seqretries until it is stable,
2144 * and if we end up with a successful lookup, we actually
2145 * want to exit RCU lookup anyway.
2147 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2148 * we are still guaranteed NUL-termination of ->d_name.name.
2150 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2151 if (dentry
->d_parent
!= parent
)
2153 if (d_unhashed(dentry
))
2156 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2159 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2161 tlen
= dentry
->d_name
.len
;
2162 tname
= dentry
->d_name
.name
;
2163 /* we want a consistent (name,len) pair */
2164 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2168 if (parent
->d_op
->d_compare(parent
, dentry
,
2169 tlen
, tname
, name
) != 0)
2172 if (dentry
->d_name
.hash_len
!= hashlen
)
2174 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2184 * d_lookup - search for a dentry
2185 * @parent: parent dentry
2186 * @name: qstr of name we wish to find
2187 * Returns: dentry, or NULL
2189 * d_lookup searches the children of the parent dentry for the name in
2190 * question. If the dentry is found its reference count is incremented and the
2191 * dentry is returned. The caller must use dput to free the entry when it has
2192 * finished using it. %NULL is returned if the dentry does not exist.
2194 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2196 struct dentry
*dentry
;
2200 seq
= read_seqbegin(&rename_lock
);
2201 dentry
= __d_lookup(parent
, name
);
2204 } while (read_seqretry(&rename_lock
, seq
));
2207 EXPORT_SYMBOL(d_lookup
);
2210 * __d_lookup - search for a dentry (racy)
2211 * @parent: parent dentry
2212 * @name: qstr of name we wish to find
2213 * Returns: dentry, or NULL
2215 * __d_lookup is like d_lookup, however it may (rarely) return a
2216 * false-negative result due to unrelated rename activity.
2218 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2219 * however it must be used carefully, eg. with a following d_lookup in
2220 * the case of failure.
2222 * __d_lookup callers must be commented.
2224 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2226 unsigned int hash
= name
->hash
;
2227 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2228 struct hlist_bl_node
*node
;
2229 struct dentry
*found
= NULL
;
2230 struct dentry
*dentry
;
2233 * Note: There is significant duplication with __d_lookup_rcu which is
2234 * required to prevent single threaded performance regressions
2235 * especially on architectures where smp_rmb (in seqcounts) are costly.
2236 * Keep the two functions in sync.
2240 * The hash list is protected using RCU.
2242 * Take d_lock when comparing a candidate dentry, to avoid races
2245 * It is possible that concurrent renames can mess up our list
2246 * walk here and result in missing our dentry, resulting in the
2247 * false-negative result. d_lookup() protects against concurrent
2248 * renames using rename_lock seqlock.
2250 * See Documentation/filesystems/path-lookup.txt for more details.
2254 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2256 if (dentry
->d_name
.hash
!= hash
)
2259 spin_lock(&dentry
->d_lock
);
2260 if (dentry
->d_parent
!= parent
)
2262 if (d_unhashed(dentry
))
2265 if (!d_same_name(dentry
, parent
, name
))
2268 dentry
->d_lockref
.count
++;
2270 spin_unlock(&dentry
->d_lock
);
2273 spin_unlock(&dentry
->d_lock
);
2281 * d_hash_and_lookup - hash the qstr then search for a dentry
2282 * @dir: Directory to search in
2283 * @name: qstr of name we wish to find
2285 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2287 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2290 * Check for a fs-specific hash function. Note that we must
2291 * calculate the standard hash first, as the d_op->d_hash()
2292 * routine may choose to leave the hash value unchanged.
2294 name
->hash
= full_name_hash(name
->name
, name
->len
);
2295 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2296 int err
= dir
->d_op
->d_hash(dir
, name
);
2297 if (unlikely(err
< 0))
2298 return ERR_PTR(err
);
2300 return d_lookup(dir
, name
);
2302 EXPORT_SYMBOL(d_hash_and_lookup
);
2305 * When a file is deleted, we have two options:
2306 * - turn this dentry into a negative dentry
2307 * - unhash this dentry and free it.
2309 * Usually, we want to just turn this into
2310 * a negative dentry, but if anybody else is
2311 * currently using the dentry or the inode
2312 * we can't do that and we fall back on removing
2313 * it from the hash queues and waiting for
2314 * it to be deleted later when it has no users
2318 * d_delete - delete a dentry
2319 * @dentry: The dentry to delete
2321 * Turn the dentry into a negative dentry if possible, otherwise
2322 * remove it from the hash queues so it can be deleted later
2325 void d_delete(struct dentry
* dentry
)
2327 struct inode
*inode
;
2330 * Are we the only user?
2333 spin_lock(&dentry
->d_lock
);
2334 inode
= dentry
->d_inode
;
2335 isdir
= S_ISDIR(inode
->i_mode
);
2336 if (dentry
->d_lockref
.count
== 1) {
2337 if (!spin_trylock(&inode
->i_lock
)) {
2338 spin_unlock(&dentry
->d_lock
);
2342 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2343 dentry_unlink_inode(dentry
);
2344 fsnotify_nameremove(dentry
, isdir
);
2348 if (!d_unhashed(dentry
))
2351 spin_unlock(&dentry
->d_lock
);
2353 fsnotify_nameremove(dentry
, isdir
);
2355 EXPORT_SYMBOL(d_delete
);
2357 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2359 BUG_ON(!d_unhashed(entry
));
2361 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2365 static void _d_rehash(struct dentry
* entry
)
2367 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2371 * d_rehash - add an entry back to the hash
2372 * @entry: dentry to add to the hash
2374 * Adds a dentry to the hash according to its name.
2377 void d_rehash(struct dentry
* entry
)
2379 spin_lock(&entry
->d_lock
);
2381 spin_unlock(&entry
->d_lock
);
2383 EXPORT_SYMBOL(d_rehash
);
2385 static inline unsigned start_dir_add(struct inode
*dir
)
2389 unsigned n
= dir
->i_dir_seq
;
2390 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2396 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2398 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2401 static void d_wait_lookup(struct dentry
*dentry
)
2403 if (d_in_lookup(dentry
)) {
2404 DECLARE_WAITQUEUE(wait
, current
);
2405 add_wait_queue(dentry
->d_wait
, &wait
);
2407 set_current_state(TASK_UNINTERRUPTIBLE
);
2408 spin_unlock(&dentry
->d_lock
);
2410 spin_lock(&dentry
->d_lock
);
2411 } while (d_in_lookup(dentry
));
2415 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2416 const struct qstr
*name
,
2417 wait_queue_head_t
*wq
)
2419 unsigned int hash
= name
->hash
;
2420 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2421 struct hlist_bl_node
*node
;
2422 struct dentry
*new = d_alloc(parent
, name
);
2423 struct dentry
*dentry
;
2424 unsigned seq
, r_seq
, d_seq
;
2427 return ERR_PTR(-ENOMEM
);
2431 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
) & ~1;
2432 r_seq
= read_seqbegin(&rename_lock
);
2433 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2434 if (unlikely(dentry
)) {
2435 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2439 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2448 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2453 if (unlikely(parent
->d_inode
->i_dir_seq
!= seq
)) {
2459 * No changes for the parent since the beginning of d_lookup().
2460 * Since all removals from the chain happen with hlist_bl_lock(),
2461 * any potential in-lookup matches are going to stay here until
2462 * we unlock the chain. All fields are stable in everything
2465 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2466 if (dentry
->d_name
.hash
!= hash
)
2468 if (dentry
->d_parent
!= parent
)
2470 if (!d_same_name(dentry
, parent
, name
))
2473 /* now we can try to grab a reference */
2474 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2481 * somebody is likely to be still doing lookup for it;
2482 * wait for them to finish
2484 spin_lock(&dentry
->d_lock
);
2485 d_wait_lookup(dentry
);
2487 * it's not in-lookup anymore; in principle we should repeat
2488 * everything from dcache lookup, but it's likely to be what
2489 * d_lookup() would've found anyway. If it is, just return it;
2490 * otherwise we really have to repeat the whole thing.
2492 if (unlikely(dentry
->d_name
.hash
!= hash
))
2494 if (unlikely(dentry
->d_parent
!= parent
))
2496 if (unlikely(d_unhashed(dentry
)))
2498 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2500 /* OK, it *is* a hashed match; return it */
2501 spin_unlock(&dentry
->d_lock
);
2506 /* we can't take ->d_lock here; it's OK, though. */
2507 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2509 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2513 spin_unlock(&dentry
->d_lock
);
2517 EXPORT_SYMBOL(d_alloc_parallel
);
2519 void __d_lookup_done(struct dentry
*dentry
)
2521 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2522 dentry
->d_name
.hash
);
2524 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2525 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2526 wake_up_all(dentry
->d_wait
);
2527 dentry
->d_wait
= NULL
;
2529 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2530 INIT_LIST_HEAD(&dentry
->d_lru
);
2532 EXPORT_SYMBOL(__d_lookup_done
);
2534 /* inode->i_lock held if inode is non-NULL */
2536 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2538 struct inode
*dir
= NULL
;
2540 spin_lock(&dentry
->d_lock
);
2541 if (unlikely(d_in_lookup(dentry
))) {
2542 dir
= dentry
->d_parent
->d_inode
;
2543 n
= start_dir_add(dir
);
2544 __d_lookup_done(dentry
);
2547 unsigned add_flags
= d_flags_for_inode(inode
);
2548 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2549 raw_write_seqcount_begin(&dentry
->d_seq
);
2550 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2551 raw_write_seqcount_end(&dentry
->d_seq
);
2552 fsnotify_update_flags(dentry
);
2556 end_dir_add(dir
, n
);
2557 spin_unlock(&dentry
->d_lock
);
2559 spin_unlock(&inode
->i_lock
);
2563 * d_add - add dentry to hash queues
2564 * @entry: dentry to add
2565 * @inode: The inode to attach to this dentry
2567 * This adds the entry to the hash queues and initializes @inode.
2568 * The entry was actually filled in earlier during d_alloc().
2571 void d_add(struct dentry
*entry
, struct inode
*inode
)
2574 security_d_instantiate(entry
, inode
);
2575 spin_lock(&inode
->i_lock
);
2577 __d_add(entry
, inode
);
2579 EXPORT_SYMBOL(d_add
);
2582 * d_exact_alias - find and hash an exact unhashed alias
2583 * @entry: dentry to add
2584 * @inode: The inode to go with this dentry
2586 * If an unhashed dentry with the same name/parent and desired
2587 * inode already exists, hash and return it. Otherwise, return
2590 * Parent directory should be locked.
2592 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2594 struct dentry
*alias
;
2595 unsigned int hash
= entry
->d_name
.hash
;
2597 spin_lock(&inode
->i_lock
);
2598 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2600 * Don't need alias->d_lock here, because aliases with
2601 * d_parent == entry->d_parent are not subject to name or
2602 * parent changes, because the parent inode i_mutex is held.
2604 if (alias
->d_name
.hash
!= hash
)
2606 if (alias
->d_parent
!= entry
->d_parent
)
2608 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2610 spin_lock(&alias
->d_lock
);
2611 if (!d_unhashed(alias
)) {
2612 spin_unlock(&alias
->d_lock
);
2615 __dget_dlock(alias
);
2617 spin_unlock(&alias
->d_lock
);
2619 spin_unlock(&inode
->i_lock
);
2622 spin_unlock(&inode
->i_lock
);
2625 EXPORT_SYMBOL(d_exact_alias
);
2628 * dentry_update_name_case - update case insensitive dentry with a new name
2629 * @dentry: dentry to be updated
2632 * Update a case insensitive dentry with new case of name.
2634 * dentry must have been returned by d_lookup with name @name. Old and new
2635 * name lengths must match (ie. no d_compare which allows mismatched name
2638 * Parent inode i_mutex must be held over d_lookup and into this call (to
2639 * keep renames and concurrent inserts, and readdir(2) away).
2641 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2643 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2644 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2646 spin_lock(&dentry
->d_lock
);
2647 write_seqcount_begin(&dentry
->d_seq
);
2648 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2649 write_seqcount_end(&dentry
->d_seq
);
2650 spin_unlock(&dentry
->d_lock
);
2652 EXPORT_SYMBOL(dentry_update_name_case
);
2654 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2656 if (unlikely(dname_external(target
))) {
2657 if (unlikely(dname_external(dentry
))) {
2659 * Both external: swap the pointers
2661 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2664 * dentry:internal, target:external. Steal target's
2665 * storage and make target internal.
2667 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2668 dentry
->d_name
.len
+ 1);
2669 dentry
->d_name
.name
= target
->d_name
.name
;
2670 target
->d_name
.name
= target
->d_iname
;
2673 if (unlikely(dname_external(dentry
))) {
2675 * dentry:external, target:internal. Give dentry's
2676 * storage to target and make dentry internal
2678 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2679 target
->d_name
.len
+ 1);
2680 target
->d_name
.name
= dentry
->d_name
.name
;
2681 dentry
->d_name
.name
= dentry
->d_iname
;
2684 * Both are internal.
2687 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2688 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2689 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2690 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2691 swap(((long *) &dentry
->d_iname
)[i
],
2692 ((long *) &target
->d_iname
)[i
]);
2696 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2699 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2701 struct external_name
*old_name
= NULL
;
2702 if (unlikely(dname_external(dentry
)))
2703 old_name
= external_name(dentry
);
2704 if (unlikely(dname_external(target
))) {
2705 atomic_inc(&external_name(target
)->u
.count
);
2706 dentry
->d_name
= target
->d_name
;
2708 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2709 target
->d_name
.len
+ 1);
2710 dentry
->d_name
.name
= dentry
->d_iname
;
2711 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2713 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2714 kfree_rcu(old_name
, u
.head
);
2717 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2720 * XXXX: do we really need to take target->d_lock?
2722 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2723 spin_lock(&target
->d_parent
->d_lock
);
2725 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2726 spin_lock(&dentry
->d_parent
->d_lock
);
2727 spin_lock_nested(&target
->d_parent
->d_lock
,
2728 DENTRY_D_LOCK_NESTED
);
2730 spin_lock(&target
->d_parent
->d_lock
);
2731 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2732 DENTRY_D_LOCK_NESTED
);
2735 if (target
< dentry
) {
2736 spin_lock_nested(&target
->d_lock
, 2);
2737 spin_lock_nested(&dentry
->d_lock
, 3);
2739 spin_lock_nested(&dentry
->d_lock
, 2);
2740 spin_lock_nested(&target
->d_lock
, 3);
2744 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2746 if (target
->d_parent
!= dentry
->d_parent
)
2747 spin_unlock(&dentry
->d_parent
->d_lock
);
2748 if (target
->d_parent
!= target
)
2749 spin_unlock(&target
->d_parent
->d_lock
);
2750 spin_unlock(&target
->d_lock
);
2751 spin_unlock(&dentry
->d_lock
);
2755 * When switching names, the actual string doesn't strictly have to
2756 * be preserved in the target - because we're dropping the target
2757 * anyway. As such, we can just do a simple memcpy() to copy over
2758 * the new name before we switch, unless we are going to rehash
2759 * it. Note that if we *do* unhash the target, we are not allowed
2760 * to rehash it without giving it a new name/hash key - whether
2761 * we swap or overwrite the names here, resulting name won't match
2762 * the reality in filesystem; it's only there for d_path() purposes.
2763 * Note that all of this is happening under rename_lock, so the
2764 * any hash lookup seeing it in the middle of manipulations will
2765 * be discarded anyway. So we do not care what happens to the hash
2769 * __d_move - move a dentry
2770 * @dentry: entry to move
2771 * @target: new dentry
2772 * @exchange: exchange the two dentries
2774 * Update the dcache to reflect the move of a file name. Negative
2775 * dcache entries should not be moved in this way. Caller must hold
2776 * rename_lock, the i_mutex of the source and target directories,
2777 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2779 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2782 struct inode
*dir
= NULL
;
2784 if (!dentry
->d_inode
)
2785 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2787 BUG_ON(d_ancestor(dentry
, target
));
2788 BUG_ON(d_ancestor(target
, dentry
));
2790 dentry_lock_for_move(dentry
, target
);
2791 if (unlikely(d_in_lookup(target
))) {
2792 dir
= target
->d_parent
->d_inode
;
2793 n
= start_dir_add(dir
);
2794 __d_lookup_done(target
);
2797 write_seqcount_begin(&dentry
->d_seq
);
2798 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2800 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2803 * Move the dentry to the target hash queue. Don't bother checking
2804 * for the same hash queue because of how unlikely it is.
2807 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2810 * Unhash the target (d_delete() is not usable here). If exchanging
2811 * the two dentries, then rehash onto the other's hash queue.
2816 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2819 /* Switch the names.. */
2821 swap_names(dentry
, target
);
2823 copy_name(dentry
, target
);
2825 /* ... and switch them in the tree */
2826 if (IS_ROOT(dentry
)) {
2827 /* splicing a tree */
2828 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2829 dentry
->d_parent
= target
->d_parent
;
2830 target
->d_parent
= target
;
2831 list_del_init(&target
->d_child
);
2832 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2834 /* swapping two dentries */
2835 swap(dentry
->d_parent
, target
->d_parent
);
2836 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2837 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2839 fsnotify_update_flags(target
);
2840 fsnotify_update_flags(dentry
);
2843 write_seqcount_end(&target
->d_seq
);
2844 write_seqcount_end(&dentry
->d_seq
);
2847 end_dir_add(dir
, n
);
2848 dentry_unlock_for_move(dentry
, target
);
2852 * d_move - move a dentry
2853 * @dentry: entry to move
2854 * @target: new dentry
2856 * Update the dcache to reflect the move of a file name. Negative
2857 * dcache entries should not be moved in this way. See the locking
2858 * requirements for __d_move.
2860 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2862 write_seqlock(&rename_lock
);
2863 __d_move(dentry
, target
, false);
2864 write_sequnlock(&rename_lock
);
2866 EXPORT_SYMBOL(d_move
);
2869 * d_exchange - exchange two dentries
2870 * @dentry1: first dentry
2871 * @dentry2: second dentry
2873 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2875 write_seqlock(&rename_lock
);
2877 WARN_ON(!dentry1
->d_inode
);
2878 WARN_ON(!dentry2
->d_inode
);
2879 WARN_ON(IS_ROOT(dentry1
));
2880 WARN_ON(IS_ROOT(dentry2
));
2882 __d_move(dentry1
, dentry2
, true);
2884 write_sequnlock(&rename_lock
);
2888 * d_ancestor - search for an ancestor
2889 * @p1: ancestor dentry
2892 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2893 * an ancestor of p2, else NULL.
2895 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2899 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2900 if (p
->d_parent
== p1
)
2907 * This helper attempts to cope with remotely renamed directories
2909 * It assumes that the caller is already holding
2910 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2912 * Note: If ever the locking in lock_rename() changes, then please
2913 * remember to update this too...
2915 static int __d_unalias(struct inode
*inode
,
2916 struct dentry
*dentry
, struct dentry
*alias
)
2918 struct mutex
*m1
= NULL
;
2919 struct rw_semaphore
*m2
= NULL
;
2922 /* If alias and dentry share a parent, then no extra locks required */
2923 if (alias
->d_parent
== dentry
->d_parent
)
2926 /* See lock_rename() */
2927 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2929 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2930 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
2932 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
2934 __d_move(alias
, dentry
, false);
2945 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2946 * @inode: the inode which may have a disconnected dentry
2947 * @dentry: a negative dentry which we want to point to the inode.
2949 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2950 * place of the given dentry and return it, else simply d_add the inode
2951 * to the dentry and return NULL.
2953 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2954 * we should error out: directories can't have multiple aliases.
2956 * This is needed in the lookup routine of any filesystem that is exportable
2957 * (via knfsd) so that we can build dcache paths to directories effectively.
2959 * If a dentry was found and moved, then it is returned. Otherwise NULL
2960 * is returned. This matches the expected return value of ->lookup.
2962 * Cluster filesystems may call this function with a negative, hashed dentry.
2963 * In that case, we know that the inode will be a regular file, and also this
2964 * will only occur during atomic_open. So we need to check for the dentry
2965 * being already hashed only in the final case.
2967 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2970 return ERR_CAST(inode
);
2972 BUG_ON(!d_unhashed(dentry
));
2977 security_d_instantiate(dentry
, inode
);
2978 spin_lock(&inode
->i_lock
);
2979 if (S_ISDIR(inode
->i_mode
)) {
2980 struct dentry
*new = __d_find_any_alias(inode
);
2981 if (unlikely(new)) {
2982 /* The reference to new ensures it remains an alias */
2983 spin_unlock(&inode
->i_lock
);
2984 write_seqlock(&rename_lock
);
2985 if (unlikely(d_ancestor(new, dentry
))) {
2986 write_sequnlock(&rename_lock
);
2988 new = ERR_PTR(-ELOOP
);
2989 pr_warn_ratelimited(
2990 "VFS: Lookup of '%s' in %s %s"
2991 " would have caused loop\n",
2992 dentry
->d_name
.name
,
2993 inode
->i_sb
->s_type
->name
,
2995 } else if (!IS_ROOT(new)) {
2996 int err
= __d_unalias(inode
, dentry
, new);
2997 write_sequnlock(&rename_lock
);
3003 __d_move(new, dentry
, false);
3004 write_sequnlock(&rename_lock
);
3011 __d_add(dentry
, inode
);
3014 EXPORT_SYMBOL(d_splice_alias
);
3016 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
3020 return -ENAMETOOLONG
;
3022 memcpy(*buffer
, str
, namelen
);
3027 * prepend_name - prepend a pathname in front of current buffer pointer
3028 * @buffer: buffer pointer
3029 * @buflen: allocated length of the buffer
3030 * @name: name string and length qstr structure
3032 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3033 * make sure that either the old or the new name pointer and length are
3034 * fetched. However, there may be mismatch between length and pointer.
3035 * The length cannot be trusted, we need to copy it byte-by-byte until
3036 * the length is reached or a null byte is found. It also prepends "/" at
3037 * the beginning of the name. The sequence number check at the caller will
3038 * retry it again when a d_move() does happen. So any garbage in the buffer
3039 * due to mismatched pointer and length will be discarded.
3041 * Data dependency barrier is needed to make sure that we see that terminating
3042 * NUL. Alpha strikes again, film at 11...
3044 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
3046 const char *dname
= ACCESS_ONCE(name
->name
);
3047 u32 dlen
= ACCESS_ONCE(name
->len
);
3050 smp_read_barrier_depends();
3052 *buflen
-= dlen
+ 1;
3054 return -ENAMETOOLONG
;
3055 p
= *buffer
-= dlen
+ 1;
3067 * prepend_path - Prepend path string to a buffer
3068 * @path: the dentry/vfsmount to report
3069 * @root: root vfsmnt/dentry
3070 * @buffer: pointer to the end of the buffer
3071 * @buflen: pointer to buffer length
3073 * The function will first try to write out the pathname without taking any
3074 * lock other than the RCU read lock to make sure that dentries won't go away.
3075 * It only checks the sequence number of the global rename_lock as any change
3076 * in the dentry's d_seq will be preceded by changes in the rename_lock
3077 * sequence number. If the sequence number had been changed, it will restart
3078 * the whole pathname back-tracing sequence again by taking the rename_lock.
3079 * In this case, there is no need to take the RCU read lock as the recursive
3080 * parent pointer references will keep the dentry chain alive as long as no
3081 * rename operation is performed.
3083 static int prepend_path(const struct path
*path
,
3084 const struct path
*root
,
3085 char **buffer
, int *buflen
)
3087 struct dentry
*dentry
;
3088 struct vfsmount
*vfsmnt
;
3091 unsigned seq
, m_seq
= 0;
3097 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3104 dentry
= path
->dentry
;
3106 mnt
= real_mount(vfsmnt
);
3107 read_seqbegin_or_lock(&rename_lock
, &seq
);
3108 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3109 struct dentry
* parent
;
3111 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3112 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
3114 if (dentry
!= vfsmnt
->mnt_root
) {
3121 if (mnt
!= parent
) {
3122 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
3128 error
= is_mounted(vfsmnt
) ? 1 : 2;
3131 parent
= dentry
->d_parent
;
3133 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3141 if (need_seqretry(&rename_lock
, seq
)) {
3145 done_seqretry(&rename_lock
, seq
);
3149 if (need_seqretry(&mount_lock
, m_seq
)) {
3153 done_seqretry(&mount_lock
, m_seq
);
3155 if (error
>= 0 && bptr
== *buffer
) {
3157 error
= -ENAMETOOLONG
;
3167 * __d_path - return the path of a dentry
3168 * @path: the dentry/vfsmount to report
3169 * @root: root vfsmnt/dentry
3170 * @buf: buffer to return value in
3171 * @buflen: buffer length
3173 * Convert a dentry into an ASCII path name.
3175 * Returns a pointer into the buffer or an error code if the
3176 * path was too long.
3178 * "buflen" should be positive.
3180 * If the path is not reachable from the supplied root, return %NULL.
3182 char *__d_path(const struct path
*path
,
3183 const struct path
*root
,
3184 char *buf
, int buflen
)
3186 char *res
= buf
+ buflen
;
3189 prepend(&res
, &buflen
, "\0", 1);
3190 error
= prepend_path(path
, root
, &res
, &buflen
);
3193 return ERR_PTR(error
);
3199 char *d_absolute_path(const struct path
*path
,
3200 char *buf
, int buflen
)
3202 struct path root
= {};
3203 char *res
= buf
+ buflen
;
3206 prepend(&res
, &buflen
, "\0", 1);
3207 error
= prepend_path(path
, &root
, &res
, &buflen
);
3212 return ERR_PTR(error
);
3217 * same as __d_path but appends "(deleted)" for unlinked files.
3219 static int path_with_deleted(const struct path
*path
,
3220 const struct path
*root
,
3221 char **buf
, int *buflen
)
3223 prepend(buf
, buflen
, "\0", 1);
3224 if (d_unlinked(path
->dentry
)) {
3225 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3230 return prepend_path(path
, root
, buf
, buflen
);
3233 static int prepend_unreachable(char **buffer
, int *buflen
)
3235 return prepend(buffer
, buflen
, "(unreachable)", 13);
3238 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3243 seq
= read_seqcount_begin(&fs
->seq
);
3245 } while (read_seqcount_retry(&fs
->seq
, seq
));
3249 * d_path - return the path of a dentry
3250 * @path: path to report
3251 * @buf: buffer to return value in
3252 * @buflen: buffer length
3254 * Convert a dentry into an ASCII path name. If the entry has been deleted
3255 * the string " (deleted)" is appended. Note that this is ambiguous.
3257 * Returns a pointer into the buffer or an error code if the path was
3258 * too long. Note: Callers should use the returned pointer, not the passed
3259 * in buffer, to use the name! The implementation often starts at an offset
3260 * into the buffer, and may leave 0 bytes at the start.
3262 * "buflen" should be positive.
3264 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3266 char *res
= buf
+ buflen
;
3271 * We have various synthetic filesystems that never get mounted. On
3272 * these filesystems dentries are never used for lookup purposes, and
3273 * thus don't need to be hashed. They also don't need a name until a
3274 * user wants to identify the object in /proc/pid/fd/. The little hack
3275 * below allows us to generate a name for these objects on demand:
3277 * Some pseudo inodes are mountable. When they are mounted
3278 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3279 * and instead have d_path return the mounted path.
3281 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3282 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3283 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3286 get_fs_root_rcu(current
->fs
, &root
);
3287 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3291 res
= ERR_PTR(error
);
3294 EXPORT_SYMBOL(d_path
);
3297 * Helper function for dentry_operations.d_dname() members
3299 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3300 const char *fmt
, ...)
3306 va_start(args
, fmt
);
3307 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3310 if (sz
> sizeof(temp
) || sz
> buflen
)
3311 return ERR_PTR(-ENAMETOOLONG
);
3313 buffer
+= buflen
- sz
;
3314 return memcpy(buffer
, temp
, sz
);
3317 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3319 char *end
= buffer
+ buflen
;
3320 /* these dentries are never renamed, so d_lock is not needed */
3321 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3322 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3323 prepend(&end
, &buflen
, "/", 1))
3324 end
= ERR_PTR(-ENAMETOOLONG
);
3327 EXPORT_SYMBOL(simple_dname
);
3330 * Write full pathname from the root of the filesystem into the buffer.
3332 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3334 struct dentry
*dentry
;
3347 prepend(&end
, &len
, "\0", 1);
3351 read_seqbegin_or_lock(&rename_lock
, &seq
);
3352 while (!IS_ROOT(dentry
)) {
3353 struct dentry
*parent
= dentry
->d_parent
;
3356 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3365 if (need_seqretry(&rename_lock
, seq
)) {
3369 done_seqretry(&rename_lock
, seq
);
3374 return ERR_PTR(-ENAMETOOLONG
);
3377 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3379 return __dentry_path(dentry
, buf
, buflen
);
3381 EXPORT_SYMBOL(dentry_path_raw
);
3383 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3388 if (d_unlinked(dentry
)) {
3390 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3394 retval
= __dentry_path(dentry
, buf
, buflen
);
3395 if (!IS_ERR(retval
) && p
)
3396 *p
= '/'; /* restore '/' overriden with '\0' */
3399 return ERR_PTR(-ENAMETOOLONG
);
3402 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3408 seq
= read_seqcount_begin(&fs
->seq
);
3411 } while (read_seqcount_retry(&fs
->seq
, seq
));
3415 * NOTE! The user-level library version returns a
3416 * character pointer. The kernel system call just
3417 * returns the length of the buffer filled (which
3418 * includes the ending '\0' character), or a negative
3419 * error value. So libc would do something like
3421 * char *getcwd(char * buf, size_t size)
3425 * retval = sys_getcwd(buf, size);
3432 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3435 struct path pwd
, root
;
3436 char *page
= __getname();
3442 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3445 if (!d_unlinked(pwd
.dentry
)) {
3447 char *cwd
= page
+ PATH_MAX
;
3448 int buflen
= PATH_MAX
;
3450 prepend(&cwd
, &buflen
, "\0", 1);
3451 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3457 /* Unreachable from current root */
3459 error
= prepend_unreachable(&cwd
, &buflen
);
3465 len
= PATH_MAX
+ page
- cwd
;
3468 if (copy_to_user(buf
, cwd
, len
))
3481 * Test whether new_dentry is a subdirectory of old_dentry.
3483 * Trivially implemented using the dcache structure
3487 * is_subdir - is new dentry a subdirectory of old_dentry
3488 * @new_dentry: new dentry
3489 * @old_dentry: old dentry
3491 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3492 * Returns false otherwise.
3493 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3496 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3501 if (new_dentry
== old_dentry
)
3505 /* for restarting inner loop in case of seq retry */
3506 seq
= read_seqbegin(&rename_lock
);
3508 * Need rcu_readlock to protect against the d_parent trashing
3512 if (d_ancestor(old_dentry
, new_dentry
))
3517 } while (read_seqretry(&rename_lock
, seq
));
3522 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3524 struct dentry
*root
= data
;
3525 if (dentry
!= root
) {
3526 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3529 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3530 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3531 dentry
->d_lockref
.count
--;
3534 return D_WALK_CONTINUE
;
3537 void d_genocide(struct dentry
*parent
)
3539 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3542 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3544 inode_dec_link_count(inode
);
3545 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3546 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3547 !d_unlinked(dentry
));
3548 spin_lock(&dentry
->d_parent
->d_lock
);
3549 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3550 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3551 (unsigned long long)inode
->i_ino
);
3552 spin_unlock(&dentry
->d_lock
);
3553 spin_unlock(&dentry
->d_parent
->d_lock
);
3554 d_instantiate(dentry
, inode
);
3556 EXPORT_SYMBOL(d_tmpfile
);
3558 static __initdata
unsigned long dhash_entries
;
3559 static int __init
set_dhash_entries(char *str
)
3563 dhash_entries
= simple_strtoul(str
, &str
, 0);
3566 __setup("dhash_entries=", set_dhash_entries
);
3568 static void __init
dcache_init_early(void)
3572 /* If hashes are distributed across NUMA nodes, defer
3573 * hash allocation until vmalloc space is available.
3579 alloc_large_system_hash("Dentry cache",
3580 sizeof(struct hlist_bl_head
),
3589 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3590 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3593 static void __init
dcache_init(void)
3598 * A constructor could be added for stable state like the lists,
3599 * but it is probably not worth it because of the cache nature
3602 dentry_cache
= KMEM_CACHE(dentry
,
3603 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
);
3605 /* Hash may have been set up in dcache_init_early */
3610 alloc_large_system_hash("Dentry cache",
3611 sizeof(struct hlist_bl_head
),
3620 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3621 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3624 /* SLAB cache for __getname() consumers */
3625 struct kmem_cache
*names_cachep __read_mostly
;
3626 EXPORT_SYMBOL(names_cachep
);
3628 EXPORT_SYMBOL(d_genocide
);
3630 void __init
vfs_caches_init_early(void)
3632 dcache_init_early();
3636 void __init
vfs_caches_init(void)
3638 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3639 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3644 files_maxfiles_init();