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 <linux/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(unsigned int hash
)
109 return dentry_hashtable
+ (hash
>> (32 - d_hash_shift
));
112 #define IN_LOOKUP_SHIFT 10
113 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
115 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
118 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
119 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
123 /* Statistics gathering. */
124 struct dentry_stat_t dentry_stat
= {
128 static DEFINE_PER_CPU(long, nr_dentry
);
129 static DEFINE_PER_CPU(long, nr_dentry_unused
);
131 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
134 * Here we resort to our own counters instead of using generic per-cpu counters
135 * for consistency with what the vfs inode code does. We are expected to harvest
136 * better code and performance by having our own specialized counters.
138 * Please note that the loop is done over all possible CPUs, not over all online
139 * CPUs. The reason for this is that we don't want to play games with CPUs going
140 * on and off. If one of them goes off, we will just keep their counters.
142 * glommer: See cffbc8a for details, and if you ever intend to change this,
143 * please update all vfs counters to match.
145 static long get_nr_dentry(void)
149 for_each_possible_cpu(i
)
150 sum
+= per_cpu(nr_dentry
, i
);
151 return sum
< 0 ? 0 : sum
;
154 static long get_nr_dentry_unused(void)
158 for_each_possible_cpu(i
)
159 sum
+= per_cpu(nr_dentry_unused
, i
);
160 return sum
< 0 ? 0 : sum
;
163 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
164 size_t *lenp
, loff_t
*ppos
)
166 dentry_stat
.nr_dentry
= get_nr_dentry();
167 dentry_stat
.nr_unused
= get_nr_dentry_unused();
168 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
173 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
174 * The strings are both count bytes long, and count is non-zero.
176 #ifdef CONFIG_DCACHE_WORD_ACCESS
178 #include <asm/word-at-a-time.h>
180 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
181 * aligned allocation for this particular component. We don't
182 * strictly need the load_unaligned_zeropad() safety, but it
183 * doesn't hurt either.
185 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
186 * need the careful unaligned handling.
188 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
190 unsigned long a
,b
,mask
;
193 a
= *(unsigned long *)cs
;
194 b
= load_unaligned_zeropad(ct
);
195 if (tcount
< sizeof(unsigned long))
197 if (unlikely(a
!= b
))
199 cs
+= sizeof(unsigned long);
200 ct
+= sizeof(unsigned long);
201 tcount
-= sizeof(unsigned long);
205 mask
= bytemask_from_count(tcount
);
206 return unlikely(!!((a
^ b
) & mask
));
211 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
225 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
228 * Be careful about RCU walk racing with rename:
229 * use 'lockless_dereference' to fetch the name pointer.
231 * NOTE! Even if a rename will mean that the length
232 * was not loaded atomically, we don't care. The
233 * RCU walk will check the sequence count eventually,
234 * and catch it. And we won't overrun the buffer,
235 * because we're reading the name pointer atomically,
236 * and a dentry name is guaranteed to be properly
237 * terminated with a NUL byte.
239 * End result: even if 'len' is wrong, we'll exit
240 * early because the data cannot match (there can
241 * be no NUL in the ct/tcount data)
243 const unsigned char *cs
= lockless_dereference(dentry
->d_name
.name
);
245 return dentry_string_cmp(cs
, ct
, tcount
);
248 struct external_name
{
251 struct rcu_head head
;
253 unsigned char name
[];
256 static inline struct external_name
*external_name(struct dentry
*dentry
)
258 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
261 static void __d_free(struct rcu_head
*head
)
263 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
265 kmem_cache_free(dentry_cache
, dentry
);
268 static void __d_free_external(struct rcu_head
*head
)
270 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
271 kfree(external_name(dentry
));
272 kmem_cache_free(dentry_cache
, dentry
);
275 static inline int dname_external(const struct dentry
*dentry
)
277 return dentry
->d_name
.name
!= dentry
->d_iname
;
280 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
282 spin_lock(&dentry
->d_lock
);
283 if (unlikely(dname_external(dentry
))) {
284 struct external_name
*p
= external_name(dentry
);
285 atomic_inc(&p
->u
.count
);
286 spin_unlock(&dentry
->d_lock
);
287 name
->name
= p
->name
;
289 memcpy(name
->inline_name
, dentry
->d_iname
, DNAME_INLINE_LEN
);
290 spin_unlock(&dentry
->d_lock
);
291 name
->name
= name
->inline_name
;
294 EXPORT_SYMBOL(take_dentry_name_snapshot
);
296 void release_dentry_name_snapshot(struct name_snapshot
*name
)
298 if (unlikely(name
->name
!= name
->inline_name
)) {
299 struct external_name
*p
;
300 p
= container_of(name
->name
, struct external_name
, name
[0]);
301 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
302 kfree_rcu(p
, u
.head
);
305 EXPORT_SYMBOL(release_dentry_name_snapshot
);
307 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
313 dentry
->d_inode
= inode
;
314 flags
= READ_ONCE(dentry
->d_flags
);
315 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
317 WRITE_ONCE(dentry
->d_flags
, flags
);
320 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
322 unsigned flags
= READ_ONCE(dentry
->d_flags
);
324 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
325 WRITE_ONCE(dentry
->d_flags
, flags
);
326 dentry
->d_inode
= NULL
;
329 static void dentry_free(struct dentry
*dentry
)
331 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
332 if (unlikely(dname_external(dentry
))) {
333 struct external_name
*p
= external_name(dentry
);
334 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
335 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
339 /* if dentry was never visible to RCU, immediate free is OK */
340 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
341 __d_free(&dentry
->d_u
.d_rcu
);
343 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
347 * Release the dentry's inode, using the filesystem
348 * d_iput() operation if defined.
350 static void dentry_unlink_inode(struct dentry
* dentry
)
351 __releases(dentry
->d_lock
)
352 __releases(dentry
->d_inode
->i_lock
)
354 struct inode
*inode
= dentry
->d_inode
;
355 bool hashed
= !d_unhashed(dentry
);
358 raw_write_seqcount_begin(&dentry
->d_seq
);
359 __d_clear_type_and_inode(dentry
);
360 hlist_del_init(&dentry
->d_u
.d_alias
);
362 raw_write_seqcount_end(&dentry
->d_seq
);
363 spin_unlock(&dentry
->d_lock
);
364 spin_unlock(&inode
->i_lock
);
366 fsnotify_inoderemove(inode
);
367 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
368 dentry
->d_op
->d_iput(dentry
, inode
);
374 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
375 * is in use - which includes both the "real" per-superblock
376 * LRU list _and_ the DCACHE_SHRINK_LIST use.
378 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
379 * on the shrink list (ie not on the superblock LRU list).
381 * The per-cpu "nr_dentry_unused" counters are updated with
382 * the DCACHE_LRU_LIST bit.
384 * These helper functions make sure we always follow the
385 * rules. d_lock must be held by the caller.
387 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
388 static void d_lru_add(struct dentry
*dentry
)
390 D_FLAG_VERIFY(dentry
, 0);
391 dentry
->d_flags
|= DCACHE_LRU_LIST
;
392 this_cpu_inc(nr_dentry_unused
);
393 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
396 static void d_lru_del(struct dentry
*dentry
)
398 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
399 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
400 this_cpu_dec(nr_dentry_unused
);
401 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
404 static void d_shrink_del(struct dentry
*dentry
)
406 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
407 list_del_init(&dentry
->d_lru
);
408 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
409 this_cpu_dec(nr_dentry_unused
);
412 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
414 D_FLAG_VERIFY(dentry
, 0);
415 list_add(&dentry
->d_lru
, list
);
416 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
417 this_cpu_inc(nr_dentry_unused
);
421 * These can only be called under the global LRU lock, ie during the
422 * callback for freeing the LRU list. "isolate" removes it from the
423 * LRU lists entirely, while shrink_move moves it to the indicated
426 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
428 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
429 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
430 this_cpu_dec(nr_dentry_unused
);
431 list_lru_isolate(lru
, &dentry
->d_lru
);
434 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
435 struct list_head
*list
)
437 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
438 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
439 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
443 * dentry_lru_(add|del)_list) must be called with d_lock held.
445 static void dentry_lru_add(struct dentry
*dentry
)
447 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
449 else if (unlikely(!(dentry
->d_flags
& DCACHE_REFERENCED
)))
450 dentry
->d_flags
|= DCACHE_REFERENCED
;
454 * d_drop - drop a dentry
455 * @dentry: dentry to drop
457 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
458 * be found through a VFS lookup any more. Note that this is different from
459 * deleting the dentry - d_delete will try to mark the dentry negative if
460 * possible, giving a successful _negative_ lookup, while d_drop will
461 * just make the cache lookup fail.
463 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
464 * reason (NFS timeouts or autofs deletes).
466 * __d_drop requires dentry->d_lock.
468 void __d_drop(struct dentry
*dentry
)
470 if (!d_unhashed(dentry
)) {
471 struct hlist_bl_head
*b
;
473 * Hashed dentries are normally on the dentry hashtable,
474 * with the exception of those newly allocated by
475 * d_obtain_alias, which are always IS_ROOT:
477 if (unlikely(IS_ROOT(dentry
)))
478 b
= &dentry
->d_sb
->s_anon
;
480 b
= d_hash(dentry
->d_name
.hash
);
483 __hlist_bl_del(&dentry
->d_hash
);
484 dentry
->d_hash
.pprev
= NULL
;
486 /* After this call, in-progress rcu-walk path lookup will fail. */
487 write_seqcount_invalidate(&dentry
->d_seq
);
490 EXPORT_SYMBOL(__d_drop
);
492 void d_drop(struct dentry
*dentry
)
494 spin_lock(&dentry
->d_lock
);
496 spin_unlock(&dentry
->d_lock
);
498 EXPORT_SYMBOL(d_drop
);
500 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
504 * Inform d_walk() and shrink_dentry_list() that we are no longer
505 * attached to the dentry tree
507 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
508 if (unlikely(list_empty(&dentry
->d_child
)))
510 __list_del_entry(&dentry
->d_child
);
512 * Cursors can move around the list of children. While we'd been
513 * a normal list member, it didn't matter - ->d_child.next would've
514 * been updated. However, from now on it won't be and for the
515 * things like d_walk() it might end up with a nasty surprise.
516 * Normally d_walk() doesn't care about cursors moving around -
517 * ->d_lock on parent prevents that and since a cursor has no children
518 * of its own, we get through it without ever unlocking the parent.
519 * There is one exception, though - if we ascend from a child that
520 * gets killed as soon as we unlock it, the next sibling is found
521 * using the value left in its ->d_child.next. And if _that_
522 * pointed to a cursor, and cursor got moved (e.g. by lseek())
523 * before d_walk() regains parent->d_lock, we'll end up skipping
524 * everything the cursor had been moved past.
526 * Solution: make sure that the pointer left behind in ->d_child.next
527 * points to something that won't be moving around. I.e. skip the
530 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
531 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
532 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
534 dentry
->d_child
.next
= next
->d_child
.next
;
538 static void __dentry_kill(struct dentry
*dentry
)
540 struct dentry
*parent
= NULL
;
541 bool can_free
= true;
542 if (!IS_ROOT(dentry
))
543 parent
= dentry
->d_parent
;
546 * The dentry is now unrecoverably dead to the world.
548 lockref_mark_dead(&dentry
->d_lockref
);
551 * inform the fs via d_prune that this dentry is about to be
552 * unhashed and destroyed.
554 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
555 dentry
->d_op
->d_prune(dentry
);
557 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
558 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
561 /* if it was on the hash then remove it */
563 dentry_unlist(dentry
, parent
);
565 spin_unlock(&parent
->d_lock
);
567 dentry_unlink_inode(dentry
);
569 spin_unlock(&dentry
->d_lock
);
570 this_cpu_dec(nr_dentry
);
571 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
572 dentry
->d_op
->d_release(dentry
);
574 spin_lock(&dentry
->d_lock
);
575 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
576 dentry
->d_flags
|= DCACHE_MAY_FREE
;
579 spin_unlock(&dentry
->d_lock
);
580 if (likely(can_free
))
585 * Finish off a dentry we've decided to kill.
586 * dentry->d_lock must be held, returns with it unlocked.
587 * If ref is non-zero, then decrement the refcount too.
588 * Returns dentry requiring refcount drop, or NULL if we're done.
590 static struct dentry
*dentry_kill(struct dentry
*dentry
)
591 __releases(dentry
->d_lock
)
593 struct inode
*inode
= dentry
->d_inode
;
594 struct dentry
*parent
= NULL
;
596 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
599 if (!IS_ROOT(dentry
)) {
600 parent
= dentry
->d_parent
;
601 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
603 spin_unlock(&inode
->i_lock
);
608 __dentry_kill(dentry
);
612 spin_unlock(&dentry
->d_lock
);
613 return dentry
; /* try again with same dentry */
616 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
618 struct dentry
*parent
= dentry
->d_parent
;
621 if (unlikely(dentry
->d_lockref
.count
< 0))
623 if (likely(spin_trylock(&parent
->d_lock
)))
626 spin_unlock(&dentry
->d_lock
);
628 parent
= ACCESS_ONCE(dentry
->d_parent
);
629 spin_lock(&parent
->d_lock
);
631 * We can't blindly lock dentry until we are sure
632 * that we won't violate the locking order.
633 * Any changes of dentry->d_parent must have
634 * been done with parent->d_lock held, so
635 * spin_lock() above is enough of a barrier
636 * for checking if it's still our child.
638 if (unlikely(parent
!= dentry
->d_parent
)) {
639 spin_unlock(&parent
->d_lock
);
643 if (parent
!= dentry
)
644 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
651 * Try to do a lockless dput(), and return whether that was successful.
653 * If unsuccessful, we return false, having already taken the dentry lock.
655 * The caller needs to hold the RCU read lock, so that the dentry is
656 * guaranteed to stay around even if the refcount goes down to zero!
658 static inline bool fast_dput(struct dentry
*dentry
)
661 unsigned int d_flags
;
664 * If we have a d_op->d_delete() operation, we sould not
665 * let the dentry count go to zero, so use "put_or_lock".
667 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
668 return lockref_put_or_lock(&dentry
->d_lockref
);
671 * .. otherwise, we can try to just decrement the
672 * lockref optimistically.
674 ret
= lockref_put_return(&dentry
->d_lockref
);
677 * If the lockref_put_return() failed due to the lock being held
678 * by somebody else, the fast path has failed. We will need to
679 * get the lock, and then check the count again.
681 if (unlikely(ret
< 0)) {
682 spin_lock(&dentry
->d_lock
);
683 if (dentry
->d_lockref
.count
> 1) {
684 dentry
->d_lockref
.count
--;
685 spin_unlock(&dentry
->d_lock
);
692 * If we weren't the last ref, we're done.
698 * Careful, careful. The reference count went down
699 * to zero, but we don't hold the dentry lock, so
700 * somebody else could get it again, and do another
701 * dput(), and we need to not race with that.
703 * However, there is a very special and common case
704 * where we don't care, because there is nothing to
705 * do: the dentry is still hashed, it does not have
706 * a 'delete' op, and it's referenced and already on
709 * NOTE! Since we aren't locked, these values are
710 * not "stable". However, it is sufficient that at
711 * some point after we dropped the reference the
712 * dentry was hashed and the flags had the proper
713 * value. Other dentry users may have re-gotten
714 * a reference to the dentry and change that, but
715 * our work is done - we can leave the dentry
716 * around with a zero refcount.
719 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
720 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
722 /* Nothing to do? Dropping the reference was all we needed? */
723 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
727 * Not the fast normal case? Get the lock. We've already decremented
728 * the refcount, but we'll need to re-check the situation after
731 spin_lock(&dentry
->d_lock
);
734 * Did somebody else grab a reference to it in the meantime, and
735 * we're no longer the last user after all? Alternatively, somebody
736 * else could have killed it and marked it dead. Either way, we
737 * don't need to do anything else.
739 if (dentry
->d_lockref
.count
) {
740 spin_unlock(&dentry
->d_lock
);
745 * Re-get the reference we optimistically dropped. We hold the
746 * lock, and we just tested that it was zero, so we can just
749 dentry
->d_lockref
.count
= 1;
757 * This is complicated by the fact that we do not want to put
758 * dentries that are no longer on any hash chain on the unused
759 * list: we'd much rather just get rid of them immediately.
761 * However, that implies that we have to traverse the dentry
762 * tree upwards to the parents which might _also_ now be
763 * scheduled for deletion (it may have been only waiting for
764 * its last child to go away).
766 * This tail recursion is done by hand as we don't want to depend
767 * on the compiler to always get this right (gcc generally doesn't).
768 * Real recursion would eat up our stack space.
772 * dput - release a dentry
773 * @dentry: dentry to release
775 * Release a dentry. This will drop the usage count and if appropriate
776 * call the dentry unlink method as well as removing it from the queues and
777 * releasing its resources. If the parent dentries were scheduled for release
778 * they too may now get deleted.
780 void dput(struct dentry
*dentry
)
782 if (unlikely(!dentry
))
789 if (likely(fast_dput(dentry
))) {
794 /* Slow case: now with the dentry lock held */
797 WARN_ON(d_in_lookup(dentry
));
799 /* Unreachable? Get rid of it */
800 if (unlikely(d_unhashed(dentry
)))
803 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
806 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
807 if (dentry
->d_op
->d_delete(dentry
))
811 dentry_lru_add(dentry
);
813 dentry
->d_lockref
.count
--;
814 spin_unlock(&dentry
->d_lock
);
818 dentry
= dentry_kill(dentry
);
827 /* This must be called with d_lock held */
828 static inline void __dget_dlock(struct dentry
*dentry
)
830 dentry
->d_lockref
.count
++;
833 static inline void __dget(struct dentry
*dentry
)
835 lockref_get(&dentry
->d_lockref
);
838 struct dentry
*dget_parent(struct dentry
*dentry
)
844 * Do optimistic parent lookup without any
848 ret
= ACCESS_ONCE(dentry
->d_parent
);
849 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
851 if (likely(gotref
)) {
852 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
859 * Don't need rcu_dereference because we re-check it was correct under
863 ret
= dentry
->d_parent
;
864 spin_lock(&ret
->d_lock
);
865 if (unlikely(ret
!= dentry
->d_parent
)) {
866 spin_unlock(&ret
->d_lock
);
871 BUG_ON(!ret
->d_lockref
.count
);
872 ret
->d_lockref
.count
++;
873 spin_unlock(&ret
->d_lock
);
876 EXPORT_SYMBOL(dget_parent
);
879 * d_find_alias - grab a hashed alias of inode
880 * @inode: inode in question
882 * If inode has a hashed alias, or is a directory and has any alias,
883 * acquire the reference to alias and return it. Otherwise return NULL.
884 * Notice that if inode is a directory there can be only one alias and
885 * it can be unhashed only if it has no children, or if it is the root
886 * of a filesystem, or if the directory was renamed and d_revalidate
887 * was the first vfs operation to notice.
889 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
890 * any other hashed alias over that one.
892 static struct dentry
*__d_find_alias(struct inode
*inode
)
894 struct dentry
*alias
, *discon_alias
;
898 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
899 spin_lock(&alias
->d_lock
);
900 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
901 if (IS_ROOT(alias
) &&
902 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
903 discon_alias
= alias
;
906 spin_unlock(&alias
->d_lock
);
910 spin_unlock(&alias
->d_lock
);
913 alias
= discon_alias
;
914 spin_lock(&alias
->d_lock
);
915 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
917 spin_unlock(&alias
->d_lock
);
920 spin_unlock(&alias
->d_lock
);
926 struct dentry
*d_find_alias(struct inode
*inode
)
928 struct dentry
*de
= NULL
;
930 if (!hlist_empty(&inode
->i_dentry
)) {
931 spin_lock(&inode
->i_lock
);
932 de
= __d_find_alias(inode
);
933 spin_unlock(&inode
->i_lock
);
937 EXPORT_SYMBOL(d_find_alias
);
940 * Try to kill dentries associated with this inode.
941 * WARNING: you must own a reference to inode.
943 void d_prune_aliases(struct inode
*inode
)
945 struct dentry
*dentry
;
947 spin_lock(&inode
->i_lock
);
948 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
949 spin_lock(&dentry
->d_lock
);
950 if (!dentry
->d_lockref
.count
) {
951 struct dentry
*parent
= lock_parent(dentry
);
952 if (likely(!dentry
->d_lockref
.count
)) {
953 __dentry_kill(dentry
);
958 spin_unlock(&parent
->d_lock
);
960 spin_unlock(&dentry
->d_lock
);
962 spin_unlock(&inode
->i_lock
);
964 EXPORT_SYMBOL(d_prune_aliases
);
966 static void shrink_dentry_list(struct list_head
*list
)
968 struct dentry
*dentry
, *parent
;
970 while (!list_empty(list
)) {
972 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
973 spin_lock(&dentry
->d_lock
);
974 parent
= lock_parent(dentry
);
977 * The dispose list is isolated and dentries are not accounted
978 * to the LRU here, so we can simply remove it from the list
979 * here regardless of whether it is referenced or not.
981 d_shrink_del(dentry
);
984 * We found an inuse dentry which was not removed from
985 * the LRU because of laziness during lookup. Do not free it.
987 if (dentry
->d_lockref
.count
> 0) {
988 spin_unlock(&dentry
->d_lock
);
990 spin_unlock(&parent
->d_lock
);
995 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
996 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
997 spin_unlock(&dentry
->d_lock
);
999 spin_unlock(&parent
->d_lock
);
1001 dentry_free(dentry
);
1005 inode
= dentry
->d_inode
;
1006 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1007 d_shrink_add(dentry
, list
);
1008 spin_unlock(&dentry
->d_lock
);
1010 spin_unlock(&parent
->d_lock
);
1014 __dentry_kill(dentry
);
1017 * We need to prune ancestors too. This is necessary to prevent
1018 * quadratic behavior of shrink_dcache_parent(), but is also
1019 * expected to be beneficial in reducing dentry cache
1023 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1024 parent
= lock_parent(dentry
);
1025 if (dentry
->d_lockref
.count
!= 1) {
1026 dentry
->d_lockref
.count
--;
1027 spin_unlock(&dentry
->d_lock
);
1029 spin_unlock(&parent
->d_lock
);
1032 inode
= dentry
->d_inode
; /* can't be NULL */
1033 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1034 spin_unlock(&dentry
->d_lock
);
1036 spin_unlock(&parent
->d_lock
);
1040 __dentry_kill(dentry
);
1046 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1047 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1049 struct list_head
*freeable
= arg
;
1050 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1054 * we are inverting the lru lock/dentry->d_lock here,
1055 * so use a trylock. If we fail to get the lock, just skip
1058 if (!spin_trylock(&dentry
->d_lock
))
1062 * Referenced dentries are still in use. If they have active
1063 * counts, just remove them from the LRU. Otherwise give them
1064 * another pass through the LRU.
1066 if (dentry
->d_lockref
.count
) {
1067 d_lru_isolate(lru
, dentry
);
1068 spin_unlock(&dentry
->d_lock
);
1072 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1073 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1074 spin_unlock(&dentry
->d_lock
);
1077 * The list move itself will be made by the common LRU code. At
1078 * this point, we've dropped the dentry->d_lock but keep the
1079 * lru lock. This is safe to do, since every list movement is
1080 * protected by the lru lock even if both locks are held.
1082 * This is guaranteed by the fact that all LRU management
1083 * functions are intermediated by the LRU API calls like
1084 * list_lru_add and list_lru_del. List movement in this file
1085 * only ever occur through this functions or through callbacks
1086 * like this one, that are called from the LRU API.
1088 * The only exceptions to this are functions like
1089 * shrink_dentry_list, and code that first checks for the
1090 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1091 * operating only with stack provided lists after they are
1092 * properly isolated from the main list. It is thus, always a
1098 d_lru_shrink_move(lru
, dentry
, freeable
);
1099 spin_unlock(&dentry
->d_lock
);
1105 * prune_dcache_sb - shrink the dcache
1107 * @sc: shrink control, passed to list_lru_shrink_walk()
1109 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1110 * is done when we need more memory and called from the superblock shrinker
1113 * This function may fail to free any resources if all the dentries are in
1116 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1121 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1122 dentry_lru_isolate
, &dispose
);
1123 shrink_dentry_list(&dispose
);
1127 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1128 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1130 struct list_head
*freeable
= arg
;
1131 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1134 * we are inverting the lru lock/dentry->d_lock here,
1135 * so use a trylock. If we fail to get the lock, just skip
1138 if (!spin_trylock(&dentry
->d_lock
))
1141 d_lru_shrink_move(lru
, dentry
, freeable
);
1142 spin_unlock(&dentry
->d_lock
);
1149 * shrink_dcache_sb - shrink dcache for a superblock
1152 * Shrink the dcache for the specified super block. This is used to free
1153 * the dcache before unmounting a file system.
1155 void shrink_dcache_sb(struct super_block
*sb
)
1162 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1163 dentry_lru_isolate_shrink
, &dispose
, 1024);
1165 this_cpu_sub(nr_dentry_unused
, freed
);
1166 shrink_dentry_list(&dispose
);
1168 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1170 EXPORT_SYMBOL(shrink_dcache_sb
);
1173 * enum d_walk_ret - action to talke during tree walk
1174 * @D_WALK_CONTINUE: contrinue walk
1175 * @D_WALK_QUIT: quit walk
1176 * @D_WALK_NORETRY: quit when retry is needed
1177 * @D_WALK_SKIP: skip this dentry and its children
1187 * d_walk - walk the dentry tree
1188 * @parent: start of walk
1189 * @data: data passed to @enter() and @finish()
1190 * @enter: callback when first entering the dentry
1191 * @finish: callback when successfully finished the walk
1193 * The @enter() and @finish() callbacks are called with d_lock held.
1195 static void d_walk(struct dentry
*parent
, void *data
,
1196 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1197 void (*finish
)(void *))
1199 struct dentry
*this_parent
;
1200 struct list_head
*next
;
1202 enum d_walk_ret ret
;
1206 read_seqbegin_or_lock(&rename_lock
, &seq
);
1207 this_parent
= parent
;
1208 spin_lock(&this_parent
->d_lock
);
1210 ret
= enter(data
, this_parent
);
1212 case D_WALK_CONTINUE
:
1217 case D_WALK_NORETRY
:
1222 next
= this_parent
->d_subdirs
.next
;
1224 while (next
!= &this_parent
->d_subdirs
) {
1225 struct list_head
*tmp
= next
;
1226 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1229 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1232 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1234 ret
= enter(data
, dentry
);
1236 case D_WALK_CONTINUE
:
1239 spin_unlock(&dentry
->d_lock
);
1241 case D_WALK_NORETRY
:
1245 spin_unlock(&dentry
->d_lock
);
1249 if (!list_empty(&dentry
->d_subdirs
)) {
1250 spin_unlock(&this_parent
->d_lock
);
1251 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1252 this_parent
= dentry
;
1253 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1256 spin_unlock(&dentry
->d_lock
);
1259 * All done at this level ... ascend and resume the search.
1263 if (this_parent
!= parent
) {
1264 struct dentry
*child
= this_parent
;
1265 this_parent
= child
->d_parent
;
1267 spin_unlock(&child
->d_lock
);
1268 spin_lock(&this_parent
->d_lock
);
1270 /* might go back up the wrong parent if we have had a rename. */
1271 if (need_seqretry(&rename_lock
, seq
))
1273 /* go into the first sibling still alive */
1275 next
= child
->d_child
.next
;
1276 if (next
== &this_parent
->d_subdirs
)
1278 child
= list_entry(next
, struct dentry
, d_child
);
1279 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1283 if (need_seqretry(&rename_lock
, seq
))
1290 spin_unlock(&this_parent
->d_lock
);
1291 done_seqretry(&rename_lock
, seq
);
1295 spin_unlock(&this_parent
->d_lock
);
1304 struct check_mount
{
1305 struct vfsmount
*mnt
;
1306 unsigned int mounted
;
1309 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1311 struct check_mount
*info
= data
;
1312 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1314 if (likely(!d_mountpoint(dentry
)))
1315 return D_WALK_CONTINUE
;
1316 if (__path_is_mountpoint(&path
)) {
1320 return D_WALK_CONTINUE
;
1324 * path_has_submounts - check for mounts over a dentry in the
1325 * current namespace.
1326 * @parent: path to check.
1328 * Return true if the parent or its subdirectories contain
1329 * a mount point in the current namespace.
1331 int path_has_submounts(const struct path
*parent
)
1333 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1335 read_seqlock_excl(&mount_lock
);
1336 d_walk(parent
->dentry
, &data
, path_check_mount
, NULL
);
1337 read_sequnlock_excl(&mount_lock
);
1339 return data
.mounted
;
1341 EXPORT_SYMBOL(path_has_submounts
);
1344 * Called by mount code to set a mountpoint and check if the mountpoint is
1345 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1346 * subtree can become unreachable).
1348 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1349 * this reason take rename_lock and d_lock on dentry and ancestors.
1351 int d_set_mounted(struct dentry
*dentry
)
1355 write_seqlock(&rename_lock
);
1356 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1357 /* Need exclusion wrt. d_invalidate() */
1358 spin_lock(&p
->d_lock
);
1359 if (unlikely(d_unhashed(p
))) {
1360 spin_unlock(&p
->d_lock
);
1363 spin_unlock(&p
->d_lock
);
1365 spin_lock(&dentry
->d_lock
);
1366 if (!d_unlinked(dentry
)) {
1368 if (!d_mountpoint(dentry
)) {
1369 dentry
->d_flags
|= DCACHE_MOUNTED
;
1373 spin_unlock(&dentry
->d_lock
);
1375 write_sequnlock(&rename_lock
);
1380 * Search the dentry child list of the specified parent,
1381 * and move any unused dentries to the end of the unused
1382 * list for prune_dcache(). We descend to the next level
1383 * whenever the d_subdirs list is non-empty and continue
1386 * It returns zero iff there are no unused children,
1387 * otherwise it returns the number of children moved to
1388 * the end of the unused list. This may not be the total
1389 * number of unused children, because select_parent can
1390 * drop the lock and return early due to latency
1394 struct select_data
{
1395 struct dentry
*start
;
1396 struct list_head dispose
;
1400 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1402 struct select_data
*data
= _data
;
1403 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1405 if (data
->start
== dentry
)
1408 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1411 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1413 if (!dentry
->d_lockref
.count
) {
1414 d_shrink_add(dentry
, &data
->dispose
);
1419 * We can return to the caller if we have found some (this
1420 * ensures forward progress). We'll be coming back to find
1423 if (!list_empty(&data
->dispose
))
1424 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1430 * shrink_dcache_parent - prune dcache
1431 * @parent: parent of entries to prune
1433 * Prune the dcache to remove unused children of the parent dentry.
1435 void shrink_dcache_parent(struct dentry
*parent
)
1438 struct select_data data
;
1440 INIT_LIST_HEAD(&data
.dispose
);
1441 data
.start
= parent
;
1444 d_walk(parent
, &data
, select_collect
, NULL
);
1448 shrink_dentry_list(&data
.dispose
);
1452 EXPORT_SYMBOL(shrink_dcache_parent
);
1454 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1456 /* it has busy descendents; complain about those instead */
1457 if (!list_empty(&dentry
->d_subdirs
))
1458 return D_WALK_CONTINUE
;
1460 /* root with refcount 1 is fine */
1461 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1462 return D_WALK_CONTINUE
;
1464 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1465 " still in use (%d) [unmount of %s %s]\n",
1468 dentry
->d_inode
->i_ino
: 0UL,
1470 dentry
->d_lockref
.count
,
1471 dentry
->d_sb
->s_type
->name
,
1472 dentry
->d_sb
->s_id
);
1474 return D_WALK_CONTINUE
;
1477 static void do_one_tree(struct dentry
*dentry
)
1479 shrink_dcache_parent(dentry
);
1480 d_walk(dentry
, dentry
, umount_check
, NULL
);
1486 * destroy the dentries attached to a superblock on unmounting
1488 void shrink_dcache_for_umount(struct super_block
*sb
)
1490 struct dentry
*dentry
;
1492 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1494 dentry
= sb
->s_root
;
1496 do_one_tree(dentry
);
1498 while (!hlist_bl_empty(&sb
->s_anon
)) {
1499 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1500 do_one_tree(dentry
);
1504 struct detach_data
{
1505 struct select_data select
;
1506 struct dentry
*mountpoint
;
1508 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1510 struct detach_data
*data
= _data
;
1512 if (d_mountpoint(dentry
)) {
1513 __dget_dlock(dentry
);
1514 data
->mountpoint
= dentry
;
1518 return select_collect(&data
->select
, dentry
);
1521 static void check_and_drop(void *_data
)
1523 struct detach_data
*data
= _data
;
1525 if (!data
->mountpoint
&& list_empty(&data
->select
.dispose
))
1526 __d_drop(data
->select
.start
);
1530 * d_invalidate - detach submounts, prune dcache, and drop
1531 * @dentry: dentry to invalidate (aka detach, prune and drop)
1535 * The final d_drop is done as an atomic operation relative to
1536 * rename_lock ensuring there are no races with d_set_mounted. This
1537 * ensures there are no unhashed dentries on the path to a mountpoint.
1539 void d_invalidate(struct dentry
*dentry
)
1542 * If it's already been dropped, return OK.
1544 spin_lock(&dentry
->d_lock
);
1545 if (d_unhashed(dentry
)) {
1546 spin_unlock(&dentry
->d_lock
);
1549 spin_unlock(&dentry
->d_lock
);
1551 /* Negative dentries can be dropped without further checks */
1552 if (!dentry
->d_inode
) {
1558 struct detach_data data
;
1560 data
.mountpoint
= NULL
;
1561 INIT_LIST_HEAD(&data
.select
.dispose
);
1562 data
.select
.start
= dentry
;
1563 data
.select
.found
= 0;
1565 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1567 if (!list_empty(&data
.select
.dispose
))
1568 shrink_dentry_list(&data
.select
.dispose
);
1569 else if (!data
.mountpoint
)
1572 if (data
.mountpoint
) {
1573 detach_mounts(data
.mountpoint
);
1574 dput(data
.mountpoint
);
1579 EXPORT_SYMBOL(d_invalidate
);
1582 * __d_alloc - allocate a dcache entry
1583 * @sb: filesystem it will belong to
1584 * @name: qstr of the name
1586 * Allocates a dentry. It returns %NULL if there is insufficient memory
1587 * available. On a success the dentry is returned. The name passed in is
1588 * copied and the copy passed in may be reused after this call.
1591 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1593 struct dentry
*dentry
;
1597 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1602 * We guarantee that the inline name is always NUL-terminated.
1603 * This way the memcpy() done by the name switching in rename
1604 * will still always have a NUL at the end, even if we might
1605 * be overwriting an internal NUL character
1607 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1608 if (unlikely(!name
)) {
1609 static const struct qstr anon
= QSTR_INIT("/", 1);
1611 dname
= dentry
->d_iname
;
1612 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1613 size_t size
= offsetof(struct external_name
, name
[1]);
1614 struct external_name
*p
= kmalloc(size
+ name
->len
,
1615 GFP_KERNEL_ACCOUNT
);
1617 kmem_cache_free(dentry_cache
, dentry
);
1620 atomic_set(&p
->u
.count
, 1);
1622 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1623 kasan_unpoison_shadow(dname
,
1624 round_up(name
->len
+ 1, sizeof(unsigned long)));
1626 dname
= dentry
->d_iname
;
1629 dentry
->d_name
.len
= name
->len
;
1630 dentry
->d_name
.hash
= name
->hash
;
1631 memcpy(dname
, name
->name
, name
->len
);
1632 dname
[name
->len
] = 0;
1634 /* Make sure we always see the terminating NUL character */
1636 dentry
->d_name
.name
= dname
;
1638 dentry
->d_lockref
.count
= 1;
1639 dentry
->d_flags
= 0;
1640 spin_lock_init(&dentry
->d_lock
);
1641 seqcount_init(&dentry
->d_seq
);
1642 dentry
->d_inode
= NULL
;
1643 dentry
->d_parent
= dentry
;
1645 dentry
->d_op
= NULL
;
1646 dentry
->d_fsdata
= NULL
;
1647 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1648 INIT_LIST_HEAD(&dentry
->d_lru
);
1649 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1650 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1651 INIT_LIST_HEAD(&dentry
->d_child
);
1652 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1654 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1655 err
= dentry
->d_op
->d_init(dentry
);
1657 if (dname_external(dentry
))
1658 kfree(external_name(dentry
));
1659 kmem_cache_free(dentry_cache
, dentry
);
1664 this_cpu_inc(nr_dentry
);
1670 * d_alloc - allocate a dcache entry
1671 * @parent: parent of entry to allocate
1672 * @name: qstr of the name
1674 * Allocates a dentry. It returns %NULL if there is insufficient memory
1675 * available. On a success the dentry is returned. The name passed in is
1676 * copied and the copy passed in may be reused after this call.
1678 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1680 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1683 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1684 spin_lock(&parent
->d_lock
);
1686 * don't need child lock because it is not subject
1687 * to concurrency here
1689 __dget_dlock(parent
);
1690 dentry
->d_parent
= parent
;
1691 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1692 spin_unlock(&parent
->d_lock
);
1696 EXPORT_SYMBOL(d_alloc
);
1698 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1700 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1702 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1703 dentry
->d_parent
= dget(parent
);
1709 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1710 * @sb: the superblock
1711 * @name: qstr of the name
1713 * For a filesystem that just pins its dentries in memory and never
1714 * performs lookups at all, return an unhashed IS_ROOT dentry.
1716 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1718 return __d_alloc(sb
, name
);
1720 EXPORT_SYMBOL(d_alloc_pseudo
);
1722 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1727 q
.hash_len
= hashlen_string(parent
, name
);
1728 return d_alloc(parent
, &q
);
1730 EXPORT_SYMBOL(d_alloc_name
);
1732 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1734 WARN_ON_ONCE(dentry
->d_op
);
1735 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1737 DCACHE_OP_REVALIDATE
|
1738 DCACHE_OP_WEAK_REVALIDATE
|
1745 dentry
->d_flags
|= DCACHE_OP_HASH
;
1747 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1748 if (op
->d_revalidate
)
1749 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1750 if (op
->d_weak_revalidate
)
1751 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1753 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1755 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1757 dentry
->d_flags
|= DCACHE_OP_REAL
;
1760 EXPORT_SYMBOL(d_set_d_op
);
1764 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1765 * @dentry - The dentry to mark
1767 * Mark a dentry as falling through to the lower layer (as set with
1768 * d_pin_lower()). This flag may be recorded on the medium.
1770 void d_set_fallthru(struct dentry
*dentry
)
1772 spin_lock(&dentry
->d_lock
);
1773 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1774 spin_unlock(&dentry
->d_lock
);
1776 EXPORT_SYMBOL(d_set_fallthru
);
1778 static unsigned d_flags_for_inode(struct inode
*inode
)
1780 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1783 return DCACHE_MISS_TYPE
;
1785 if (S_ISDIR(inode
->i_mode
)) {
1786 add_flags
= DCACHE_DIRECTORY_TYPE
;
1787 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1788 if (unlikely(!inode
->i_op
->lookup
))
1789 add_flags
= DCACHE_AUTODIR_TYPE
;
1791 inode
->i_opflags
|= IOP_LOOKUP
;
1793 goto type_determined
;
1796 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1797 if (unlikely(inode
->i_op
->get_link
)) {
1798 add_flags
= DCACHE_SYMLINK_TYPE
;
1799 goto type_determined
;
1801 inode
->i_opflags
|= IOP_NOFOLLOW
;
1804 if (unlikely(!S_ISREG(inode
->i_mode
)))
1805 add_flags
= DCACHE_SPECIAL_TYPE
;
1808 if (unlikely(IS_AUTOMOUNT(inode
)))
1809 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1813 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1815 unsigned add_flags
= d_flags_for_inode(inode
);
1816 WARN_ON(d_in_lookup(dentry
));
1818 spin_lock(&dentry
->d_lock
);
1819 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1820 raw_write_seqcount_begin(&dentry
->d_seq
);
1821 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1822 raw_write_seqcount_end(&dentry
->d_seq
);
1823 fsnotify_update_flags(dentry
);
1824 spin_unlock(&dentry
->d_lock
);
1828 * d_instantiate - fill in inode information for a dentry
1829 * @entry: dentry to complete
1830 * @inode: inode to attach to this dentry
1832 * Fill in inode information in the entry.
1834 * This turns negative dentries into productive full members
1837 * NOTE! This assumes that the inode count has been incremented
1838 * (or otherwise set) by the caller to indicate that it is now
1839 * in use by the dcache.
1842 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1844 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1846 security_d_instantiate(entry
, inode
);
1847 spin_lock(&inode
->i_lock
);
1848 __d_instantiate(entry
, inode
);
1849 spin_unlock(&inode
->i_lock
);
1852 EXPORT_SYMBOL(d_instantiate
);
1855 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1856 * @entry: dentry to complete
1857 * @inode: inode to attach to this dentry
1859 * Fill in inode information in the entry. If a directory alias is found, then
1860 * return an error (and drop inode). Together with d_materialise_unique() this
1861 * guarantees that a directory inode may never have more than one alias.
1863 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1865 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1867 security_d_instantiate(entry
, inode
);
1868 spin_lock(&inode
->i_lock
);
1869 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1870 spin_unlock(&inode
->i_lock
);
1874 __d_instantiate(entry
, inode
);
1875 spin_unlock(&inode
->i_lock
);
1879 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1881 struct dentry
*d_make_root(struct inode
*root_inode
)
1883 struct dentry
*res
= NULL
;
1886 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1888 d_instantiate(res
, root_inode
);
1894 EXPORT_SYMBOL(d_make_root
);
1896 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1898 struct dentry
*alias
;
1900 if (hlist_empty(&inode
->i_dentry
))
1902 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1908 * d_find_any_alias - find any alias for a given inode
1909 * @inode: inode to find an alias for
1911 * If any aliases exist for the given inode, take and return a
1912 * reference for one of them. If no aliases exist, return %NULL.
1914 struct dentry
*d_find_any_alias(struct inode
*inode
)
1918 spin_lock(&inode
->i_lock
);
1919 de
= __d_find_any_alias(inode
);
1920 spin_unlock(&inode
->i_lock
);
1923 EXPORT_SYMBOL(d_find_any_alias
);
1925 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1932 return ERR_PTR(-ESTALE
);
1934 return ERR_CAST(inode
);
1936 res
= d_find_any_alias(inode
);
1940 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1942 res
= ERR_PTR(-ENOMEM
);
1946 security_d_instantiate(tmp
, inode
);
1947 spin_lock(&inode
->i_lock
);
1948 res
= __d_find_any_alias(inode
);
1950 spin_unlock(&inode
->i_lock
);
1955 /* attach a disconnected dentry */
1956 add_flags
= d_flags_for_inode(inode
);
1959 add_flags
|= DCACHE_DISCONNECTED
;
1961 spin_lock(&tmp
->d_lock
);
1962 __d_set_inode_and_type(tmp
, inode
, add_flags
);
1963 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1964 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1965 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1966 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1967 spin_unlock(&tmp
->d_lock
);
1968 spin_unlock(&inode
->i_lock
);
1978 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1979 * @inode: inode to allocate the dentry for
1981 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1982 * similar open by handle operations. The returned dentry may be anonymous,
1983 * or may have a full name (if the inode was already in the cache).
1985 * When called on a directory inode, we must ensure that the inode only ever
1986 * has one dentry. If a dentry is found, that is returned instead of
1987 * allocating a new one.
1989 * On successful return, the reference to the inode has been transferred
1990 * to the dentry. In case of an error the reference on the inode is released.
1991 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1992 * be passed in and the error will be propagated to the return value,
1993 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1995 struct dentry
*d_obtain_alias(struct inode
*inode
)
1997 return __d_obtain_alias(inode
, 1);
1999 EXPORT_SYMBOL(d_obtain_alias
);
2002 * d_obtain_root - find or allocate a dentry for a given inode
2003 * @inode: inode to allocate the dentry for
2005 * Obtain an IS_ROOT dentry for the root of a filesystem.
2007 * We must ensure that directory inodes only ever have one dentry. If a
2008 * dentry is found, that is returned instead of allocating a new one.
2010 * On successful return, the reference to the inode has been transferred
2011 * to the dentry. In case of an error the reference on the inode is
2012 * released. A %NULL or IS_ERR inode may be passed in and will be the
2013 * error will be propagate to the return value, with a %NULL @inode
2014 * replaced by ERR_PTR(-ESTALE).
2016 struct dentry
*d_obtain_root(struct inode
*inode
)
2018 return __d_obtain_alias(inode
, 0);
2020 EXPORT_SYMBOL(d_obtain_root
);
2023 * d_add_ci - lookup or allocate new dentry with case-exact name
2024 * @inode: the inode case-insensitive lookup has found
2025 * @dentry: the negative dentry that was passed to the parent's lookup func
2026 * @name: the case-exact name to be associated with the returned dentry
2028 * This is to avoid filling the dcache with case-insensitive names to the
2029 * same inode, only the actual correct case is stored in the dcache for
2030 * case-insensitive filesystems.
2032 * For a case-insensitive lookup match and if the the case-exact dentry
2033 * already exists in in the dcache, use it and return it.
2035 * If no entry exists with the exact case name, allocate new dentry with
2036 * the exact case, and return the spliced entry.
2038 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2041 struct dentry
*found
, *res
;
2044 * First check if a dentry matching the name already exists,
2045 * if not go ahead and create it now.
2047 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2052 if (d_in_lookup(dentry
)) {
2053 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2055 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2060 found
= d_alloc(dentry
->d_parent
, name
);
2063 return ERR_PTR(-ENOMEM
);
2066 res
= d_splice_alias(inode
, found
);
2073 EXPORT_SYMBOL(d_add_ci
);
2076 static inline bool d_same_name(const struct dentry
*dentry
,
2077 const struct dentry
*parent
,
2078 const struct qstr
*name
)
2080 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2081 if (dentry
->d_name
.len
!= name
->len
)
2083 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2085 return parent
->d_op
->d_compare(dentry
,
2086 dentry
->d_name
.len
, dentry
->d_name
.name
,
2091 * __d_lookup_rcu - search for a dentry (racy, store-free)
2092 * @parent: parent dentry
2093 * @name: qstr of name we wish to find
2094 * @seqp: returns d_seq value at the point where the dentry was found
2095 * Returns: dentry, or NULL
2097 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2098 * resolution (store-free path walking) design described in
2099 * Documentation/filesystems/path-lookup.txt.
2101 * This is not to be used outside core vfs.
2103 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2104 * held, and rcu_read_lock held. The returned dentry must not be stored into
2105 * without taking d_lock and checking d_seq sequence count against @seq
2108 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2111 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2112 * the returned dentry, so long as its parent's seqlock is checked after the
2113 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2114 * is formed, giving integrity down the path walk.
2116 * NOTE! The caller *has* to check the resulting dentry against the sequence
2117 * number we've returned before using any of the resulting dentry state!
2119 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2120 const struct qstr
*name
,
2123 u64 hashlen
= name
->hash_len
;
2124 const unsigned char *str
= name
->name
;
2125 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2126 struct hlist_bl_node
*node
;
2127 struct dentry
*dentry
;
2130 * Note: There is significant duplication with __d_lookup_rcu which is
2131 * required to prevent single threaded performance regressions
2132 * especially on architectures where smp_rmb (in seqcounts) are costly.
2133 * Keep the two functions in sync.
2137 * The hash list is protected using RCU.
2139 * Carefully use d_seq when comparing a candidate dentry, to avoid
2140 * races with d_move().
2142 * It is possible that concurrent renames can mess up our list
2143 * walk here and result in missing our dentry, resulting in the
2144 * false-negative result. d_lookup() protects against concurrent
2145 * renames using rename_lock seqlock.
2147 * See Documentation/filesystems/path-lookup.txt for more details.
2149 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2154 * The dentry sequence count protects us from concurrent
2155 * renames, and thus protects parent and name fields.
2157 * The caller must perform a seqcount check in order
2158 * to do anything useful with the returned dentry.
2160 * NOTE! We do a "raw" seqcount_begin here. That means that
2161 * we don't wait for the sequence count to stabilize if it
2162 * is in the middle of a sequence change. If we do the slow
2163 * dentry compare, we will do seqretries until it is stable,
2164 * and if we end up with a successful lookup, we actually
2165 * want to exit RCU lookup anyway.
2167 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2168 * we are still guaranteed NUL-termination of ->d_name.name.
2170 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2171 if (dentry
->d_parent
!= parent
)
2173 if (d_unhashed(dentry
))
2176 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2179 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2181 tlen
= dentry
->d_name
.len
;
2182 tname
= dentry
->d_name
.name
;
2183 /* we want a consistent (name,len) pair */
2184 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2188 if (parent
->d_op
->d_compare(dentry
,
2189 tlen
, tname
, name
) != 0)
2192 if (dentry
->d_name
.hash_len
!= hashlen
)
2194 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2204 * d_lookup - search for a dentry
2205 * @parent: parent dentry
2206 * @name: qstr of name we wish to find
2207 * Returns: dentry, or NULL
2209 * d_lookup searches the children of the parent dentry for the name in
2210 * question. If the dentry is found its reference count is incremented and the
2211 * dentry is returned. The caller must use dput to free the entry when it has
2212 * finished using it. %NULL is returned if the dentry does not exist.
2214 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2216 struct dentry
*dentry
;
2220 seq
= read_seqbegin(&rename_lock
);
2221 dentry
= __d_lookup(parent
, name
);
2224 } while (read_seqretry(&rename_lock
, seq
));
2227 EXPORT_SYMBOL(d_lookup
);
2230 * __d_lookup - search for a dentry (racy)
2231 * @parent: parent dentry
2232 * @name: qstr of name we wish to find
2233 * Returns: dentry, or NULL
2235 * __d_lookup is like d_lookup, however it may (rarely) return a
2236 * false-negative result due to unrelated rename activity.
2238 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2239 * however it must be used carefully, eg. with a following d_lookup in
2240 * the case of failure.
2242 * __d_lookup callers must be commented.
2244 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2246 unsigned int hash
= name
->hash
;
2247 struct hlist_bl_head
*b
= d_hash(hash
);
2248 struct hlist_bl_node
*node
;
2249 struct dentry
*found
= NULL
;
2250 struct dentry
*dentry
;
2253 * Note: There is significant duplication with __d_lookup_rcu which is
2254 * required to prevent single threaded performance regressions
2255 * especially on architectures where smp_rmb (in seqcounts) are costly.
2256 * Keep the two functions in sync.
2260 * The hash list is protected using RCU.
2262 * Take d_lock when comparing a candidate dentry, to avoid races
2265 * It is possible that concurrent renames can mess up our list
2266 * walk here and result in missing our dentry, resulting in the
2267 * false-negative result. d_lookup() protects against concurrent
2268 * renames using rename_lock seqlock.
2270 * See Documentation/filesystems/path-lookup.txt for more details.
2274 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2276 if (dentry
->d_name
.hash
!= hash
)
2279 spin_lock(&dentry
->d_lock
);
2280 if (dentry
->d_parent
!= parent
)
2282 if (d_unhashed(dentry
))
2285 if (!d_same_name(dentry
, parent
, name
))
2288 dentry
->d_lockref
.count
++;
2290 spin_unlock(&dentry
->d_lock
);
2293 spin_unlock(&dentry
->d_lock
);
2301 * d_hash_and_lookup - hash the qstr then search for a dentry
2302 * @dir: Directory to search in
2303 * @name: qstr of name we wish to find
2305 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2307 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2310 * Check for a fs-specific hash function. Note that we must
2311 * calculate the standard hash first, as the d_op->d_hash()
2312 * routine may choose to leave the hash value unchanged.
2314 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2315 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2316 int err
= dir
->d_op
->d_hash(dir
, name
);
2317 if (unlikely(err
< 0))
2318 return ERR_PTR(err
);
2320 return d_lookup(dir
, name
);
2322 EXPORT_SYMBOL(d_hash_and_lookup
);
2325 * When a file is deleted, we have two options:
2326 * - turn this dentry into a negative dentry
2327 * - unhash this dentry and free it.
2329 * Usually, we want to just turn this into
2330 * a negative dentry, but if anybody else is
2331 * currently using the dentry or the inode
2332 * we can't do that and we fall back on removing
2333 * it from the hash queues and waiting for
2334 * it to be deleted later when it has no users
2338 * d_delete - delete a dentry
2339 * @dentry: The dentry to delete
2341 * Turn the dentry into a negative dentry if possible, otherwise
2342 * remove it from the hash queues so it can be deleted later
2345 void d_delete(struct dentry
* dentry
)
2347 struct inode
*inode
;
2350 * Are we the only user?
2353 spin_lock(&dentry
->d_lock
);
2354 inode
= dentry
->d_inode
;
2355 isdir
= S_ISDIR(inode
->i_mode
);
2356 if (dentry
->d_lockref
.count
== 1) {
2357 if (!spin_trylock(&inode
->i_lock
)) {
2358 spin_unlock(&dentry
->d_lock
);
2362 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2363 dentry_unlink_inode(dentry
);
2364 fsnotify_nameremove(dentry
, isdir
);
2368 if (!d_unhashed(dentry
))
2371 spin_unlock(&dentry
->d_lock
);
2373 fsnotify_nameremove(dentry
, isdir
);
2375 EXPORT_SYMBOL(d_delete
);
2377 static void __d_rehash(struct dentry
*entry
)
2379 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2380 BUG_ON(!d_unhashed(entry
));
2382 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2387 * d_rehash - add an entry back to the hash
2388 * @entry: dentry to add to the hash
2390 * Adds a dentry to the hash according to its name.
2393 void d_rehash(struct dentry
* entry
)
2395 spin_lock(&entry
->d_lock
);
2397 spin_unlock(&entry
->d_lock
);
2399 EXPORT_SYMBOL(d_rehash
);
2401 static inline unsigned start_dir_add(struct inode
*dir
)
2405 unsigned n
= dir
->i_dir_seq
;
2406 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2412 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2414 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2417 static void d_wait_lookup(struct dentry
*dentry
)
2419 if (d_in_lookup(dentry
)) {
2420 DECLARE_WAITQUEUE(wait
, current
);
2421 add_wait_queue(dentry
->d_wait
, &wait
);
2423 set_current_state(TASK_UNINTERRUPTIBLE
);
2424 spin_unlock(&dentry
->d_lock
);
2426 spin_lock(&dentry
->d_lock
);
2427 } while (d_in_lookup(dentry
));
2431 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2432 const struct qstr
*name
,
2433 wait_queue_head_t
*wq
)
2435 unsigned int hash
= name
->hash
;
2436 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2437 struct hlist_bl_node
*node
;
2438 struct dentry
*new = d_alloc(parent
, name
);
2439 struct dentry
*dentry
;
2440 unsigned seq
, r_seq
, d_seq
;
2443 return ERR_PTR(-ENOMEM
);
2447 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
) & ~1;
2448 r_seq
= read_seqbegin(&rename_lock
);
2449 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2450 if (unlikely(dentry
)) {
2451 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2455 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2464 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2469 if (unlikely(parent
->d_inode
->i_dir_seq
!= seq
)) {
2475 * No changes for the parent since the beginning of d_lookup().
2476 * Since all removals from the chain happen with hlist_bl_lock(),
2477 * any potential in-lookup matches are going to stay here until
2478 * we unlock the chain. All fields are stable in everything
2481 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2482 if (dentry
->d_name
.hash
!= hash
)
2484 if (dentry
->d_parent
!= parent
)
2486 if (!d_same_name(dentry
, parent
, name
))
2489 /* now we can try to grab a reference */
2490 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2497 * somebody is likely to be still doing lookup for it;
2498 * wait for them to finish
2500 spin_lock(&dentry
->d_lock
);
2501 d_wait_lookup(dentry
);
2503 * it's not in-lookup anymore; in principle we should repeat
2504 * everything from dcache lookup, but it's likely to be what
2505 * d_lookup() would've found anyway. If it is, just return it;
2506 * otherwise we really have to repeat the whole thing.
2508 if (unlikely(dentry
->d_name
.hash
!= hash
))
2510 if (unlikely(dentry
->d_parent
!= parent
))
2512 if (unlikely(d_unhashed(dentry
)))
2514 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2516 /* OK, it *is* a hashed match; return it */
2517 spin_unlock(&dentry
->d_lock
);
2522 /* we can't take ->d_lock here; it's OK, though. */
2523 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2525 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2529 spin_unlock(&dentry
->d_lock
);
2533 EXPORT_SYMBOL(d_alloc_parallel
);
2535 void __d_lookup_done(struct dentry
*dentry
)
2537 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2538 dentry
->d_name
.hash
);
2540 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2541 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2542 wake_up_all(dentry
->d_wait
);
2543 dentry
->d_wait
= NULL
;
2545 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2546 INIT_LIST_HEAD(&dentry
->d_lru
);
2548 EXPORT_SYMBOL(__d_lookup_done
);
2550 /* inode->i_lock held if inode is non-NULL */
2552 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2554 struct inode
*dir
= NULL
;
2556 spin_lock(&dentry
->d_lock
);
2557 if (unlikely(d_in_lookup(dentry
))) {
2558 dir
= dentry
->d_parent
->d_inode
;
2559 n
= start_dir_add(dir
);
2560 __d_lookup_done(dentry
);
2563 unsigned add_flags
= d_flags_for_inode(inode
);
2564 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2565 raw_write_seqcount_begin(&dentry
->d_seq
);
2566 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2567 raw_write_seqcount_end(&dentry
->d_seq
);
2568 fsnotify_update_flags(dentry
);
2572 end_dir_add(dir
, n
);
2573 spin_unlock(&dentry
->d_lock
);
2575 spin_unlock(&inode
->i_lock
);
2579 * d_add - add dentry to hash queues
2580 * @entry: dentry to add
2581 * @inode: The inode to attach to this dentry
2583 * This adds the entry to the hash queues and initializes @inode.
2584 * The entry was actually filled in earlier during d_alloc().
2587 void d_add(struct dentry
*entry
, struct inode
*inode
)
2590 security_d_instantiate(entry
, inode
);
2591 spin_lock(&inode
->i_lock
);
2593 __d_add(entry
, inode
);
2595 EXPORT_SYMBOL(d_add
);
2598 * d_exact_alias - find and hash an exact unhashed alias
2599 * @entry: dentry to add
2600 * @inode: The inode to go with this dentry
2602 * If an unhashed dentry with the same name/parent and desired
2603 * inode already exists, hash and return it. Otherwise, return
2606 * Parent directory should be locked.
2608 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2610 struct dentry
*alias
;
2611 unsigned int hash
= entry
->d_name
.hash
;
2613 spin_lock(&inode
->i_lock
);
2614 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2616 * Don't need alias->d_lock here, because aliases with
2617 * d_parent == entry->d_parent are not subject to name or
2618 * parent changes, because the parent inode i_mutex is held.
2620 if (alias
->d_name
.hash
!= hash
)
2622 if (alias
->d_parent
!= entry
->d_parent
)
2624 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2626 spin_lock(&alias
->d_lock
);
2627 if (!d_unhashed(alias
)) {
2628 spin_unlock(&alias
->d_lock
);
2631 __dget_dlock(alias
);
2633 spin_unlock(&alias
->d_lock
);
2635 spin_unlock(&inode
->i_lock
);
2638 spin_unlock(&inode
->i_lock
);
2641 EXPORT_SYMBOL(d_exact_alias
);
2644 * dentry_update_name_case - update case insensitive dentry with a new name
2645 * @dentry: dentry to be updated
2648 * Update a case insensitive dentry with new case of name.
2650 * dentry must have been returned by d_lookup with name @name. Old and new
2651 * name lengths must match (ie. no d_compare which allows mismatched name
2654 * Parent inode i_mutex must be held over d_lookup and into this call (to
2655 * keep renames and concurrent inserts, and readdir(2) away).
2657 void dentry_update_name_case(struct dentry
*dentry
, const struct qstr
*name
)
2659 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2660 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2662 spin_lock(&dentry
->d_lock
);
2663 write_seqcount_begin(&dentry
->d_seq
);
2664 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2665 write_seqcount_end(&dentry
->d_seq
);
2666 spin_unlock(&dentry
->d_lock
);
2668 EXPORT_SYMBOL(dentry_update_name_case
);
2670 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2672 if (unlikely(dname_external(target
))) {
2673 if (unlikely(dname_external(dentry
))) {
2675 * Both external: swap the pointers
2677 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2680 * dentry:internal, target:external. Steal target's
2681 * storage and make target internal.
2683 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2684 dentry
->d_name
.len
+ 1);
2685 dentry
->d_name
.name
= target
->d_name
.name
;
2686 target
->d_name
.name
= target
->d_iname
;
2689 if (unlikely(dname_external(dentry
))) {
2691 * dentry:external, target:internal. Give dentry's
2692 * storage to target and make dentry internal
2694 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2695 target
->d_name
.len
+ 1);
2696 target
->d_name
.name
= dentry
->d_name
.name
;
2697 dentry
->d_name
.name
= dentry
->d_iname
;
2700 * Both are internal.
2703 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2704 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2705 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2706 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2707 swap(((long *) &dentry
->d_iname
)[i
],
2708 ((long *) &target
->d_iname
)[i
]);
2712 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2715 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2717 struct external_name
*old_name
= NULL
;
2718 if (unlikely(dname_external(dentry
)))
2719 old_name
= external_name(dentry
);
2720 if (unlikely(dname_external(target
))) {
2721 atomic_inc(&external_name(target
)->u
.count
);
2722 dentry
->d_name
= target
->d_name
;
2724 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2725 target
->d_name
.len
+ 1);
2726 dentry
->d_name
.name
= dentry
->d_iname
;
2727 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2729 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2730 kfree_rcu(old_name
, u
.head
);
2733 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2736 * XXXX: do we really need to take target->d_lock?
2738 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2739 spin_lock(&target
->d_parent
->d_lock
);
2741 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2742 spin_lock(&dentry
->d_parent
->d_lock
);
2743 spin_lock_nested(&target
->d_parent
->d_lock
,
2744 DENTRY_D_LOCK_NESTED
);
2746 spin_lock(&target
->d_parent
->d_lock
);
2747 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2748 DENTRY_D_LOCK_NESTED
);
2751 if (target
< dentry
) {
2752 spin_lock_nested(&target
->d_lock
, 2);
2753 spin_lock_nested(&dentry
->d_lock
, 3);
2755 spin_lock_nested(&dentry
->d_lock
, 2);
2756 spin_lock_nested(&target
->d_lock
, 3);
2760 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2762 if (target
->d_parent
!= dentry
->d_parent
)
2763 spin_unlock(&dentry
->d_parent
->d_lock
);
2764 if (target
->d_parent
!= target
)
2765 spin_unlock(&target
->d_parent
->d_lock
);
2766 spin_unlock(&target
->d_lock
);
2767 spin_unlock(&dentry
->d_lock
);
2771 * When switching names, the actual string doesn't strictly have to
2772 * be preserved in the target - because we're dropping the target
2773 * anyway. As such, we can just do a simple memcpy() to copy over
2774 * the new name before we switch, unless we are going to rehash
2775 * it. Note that if we *do* unhash the target, we are not allowed
2776 * to rehash it without giving it a new name/hash key - whether
2777 * we swap or overwrite the names here, resulting name won't match
2778 * the reality in filesystem; it's only there for d_path() purposes.
2779 * Note that all of this is happening under rename_lock, so the
2780 * any hash lookup seeing it in the middle of manipulations will
2781 * be discarded anyway. So we do not care what happens to the hash
2785 * __d_move - move a dentry
2786 * @dentry: entry to move
2787 * @target: new dentry
2788 * @exchange: exchange the two dentries
2790 * Update the dcache to reflect the move of a file name. Negative
2791 * dcache entries should not be moved in this way. Caller must hold
2792 * rename_lock, the i_mutex of the source and target directories,
2793 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2795 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2798 struct inode
*dir
= NULL
;
2800 if (!dentry
->d_inode
)
2801 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2803 BUG_ON(d_ancestor(dentry
, target
));
2804 BUG_ON(d_ancestor(target
, dentry
));
2806 dentry_lock_for_move(dentry
, target
);
2807 if (unlikely(d_in_lookup(target
))) {
2808 dir
= target
->d_parent
->d_inode
;
2809 n
= start_dir_add(dir
);
2810 __d_lookup_done(target
);
2813 write_seqcount_begin(&dentry
->d_seq
);
2814 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2817 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2821 /* Switch the names.. */
2823 swap_names(dentry
, target
);
2825 copy_name(dentry
, target
);
2827 /* rehash in new place(s) */
2832 /* ... and switch them in the tree */
2833 if (IS_ROOT(dentry
)) {
2834 /* splicing a tree */
2835 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2836 dentry
->d_parent
= target
->d_parent
;
2837 target
->d_parent
= target
;
2838 list_del_init(&target
->d_child
);
2839 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2841 /* swapping two dentries */
2842 swap(dentry
->d_parent
, target
->d_parent
);
2843 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2844 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2846 fsnotify_update_flags(target
);
2847 fsnotify_update_flags(dentry
);
2850 write_seqcount_end(&target
->d_seq
);
2851 write_seqcount_end(&dentry
->d_seq
);
2854 end_dir_add(dir
, n
);
2855 dentry_unlock_for_move(dentry
, target
);
2859 * d_move - move a dentry
2860 * @dentry: entry to move
2861 * @target: new dentry
2863 * Update the dcache to reflect the move of a file name. Negative
2864 * dcache entries should not be moved in this way. See the locking
2865 * requirements for __d_move.
2867 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2869 write_seqlock(&rename_lock
);
2870 __d_move(dentry
, target
, false);
2871 write_sequnlock(&rename_lock
);
2873 EXPORT_SYMBOL(d_move
);
2876 * d_exchange - exchange two dentries
2877 * @dentry1: first dentry
2878 * @dentry2: second dentry
2880 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2882 write_seqlock(&rename_lock
);
2884 WARN_ON(!dentry1
->d_inode
);
2885 WARN_ON(!dentry2
->d_inode
);
2886 WARN_ON(IS_ROOT(dentry1
));
2887 WARN_ON(IS_ROOT(dentry2
));
2889 __d_move(dentry1
, dentry2
, true);
2891 write_sequnlock(&rename_lock
);
2895 * d_ancestor - search for an ancestor
2896 * @p1: ancestor dentry
2899 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2900 * an ancestor of p2, else NULL.
2902 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2906 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2907 if (p
->d_parent
== p1
)
2914 * This helper attempts to cope with remotely renamed directories
2916 * It assumes that the caller is already holding
2917 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2919 * Note: If ever the locking in lock_rename() changes, then please
2920 * remember to update this too...
2922 static int __d_unalias(struct inode
*inode
,
2923 struct dentry
*dentry
, struct dentry
*alias
)
2925 struct mutex
*m1
= NULL
;
2926 struct rw_semaphore
*m2
= NULL
;
2929 /* If alias and dentry share a parent, then no extra locks required */
2930 if (alias
->d_parent
== dentry
->d_parent
)
2933 /* See lock_rename() */
2934 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2936 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2937 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
2939 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
2941 __d_move(alias
, dentry
, false);
2952 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2953 * @inode: the inode which may have a disconnected dentry
2954 * @dentry: a negative dentry which we want to point to the inode.
2956 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2957 * place of the given dentry and return it, else simply d_add the inode
2958 * to the dentry and return NULL.
2960 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2961 * we should error out: directories can't have multiple aliases.
2963 * This is needed in the lookup routine of any filesystem that is exportable
2964 * (via knfsd) so that we can build dcache paths to directories effectively.
2966 * If a dentry was found and moved, then it is returned. Otherwise NULL
2967 * is returned. This matches the expected return value of ->lookup.
2969 * Cluster filesystems may call this function with a negative, hashed dentry.
2970 * In that case, we know that the inode will be a regular file, and also this
2971 * will only occur during atomic_open. So we need to check for the dentry
2972 * being already hashed only in the final case.
2974 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2977 return ERR_CAST(inode
);
2979 BUG_ON(!d_unhashed(dentry
));
2984 security_d_instantiate(dentry
, inode
);
2985 spin_lock(&inode
->i_lock
);
2986 if (S_ISDIR(inode
->i_mode
)) {
2987 struct dentry
*new = __d_find_any_alias(inode
);
2988 if (unlikely(new)) {
2989 /* The reference to new ensures it remains an alias */
2990 spin_unlock(&inode
->i_lock
);
2991 write_seqlock(&rename_lock
);
2992 if (unlikely(d_ancestor(new, dentry
))) {
2993 write_sequnlock(&rename_lock
);
2995 new = ERR_PTR(-ELOOP
);
2996 pr_warn_ratelimited(
2997 "VFS: Lookup of '%s' in %s %s"
2998 " would have caused loop\n",
2999 dentry
->d_name
.name
,
3000 inode
->i_sb
->s_type
->name
,
3002 } else if (!IS_ROOT(new)) {
3003 int err
= __d_unalias(inode
, dentry
, new);
3004 write_sequnlock(&rename_lock
);
3010 __d_move(new, dentry
, false);
3011 write_sequnlock(&rename_lock
);
3018 __d_add(dentry
, inode
);
3021 EXPORT_SYMBOL(d_splice_alias
);
3023 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
3027 return -ENAMETOOLONG
;
3029 memcpy(*buffer
, str
, namelen
);
3034 * prepend_name - prepend a pathname in front of current buffer pointer
3035 * @buffer: buffer pointer
3036 * @buflen: allocated length of the buffer
3037 * @name: name string and length qstr structure
3039 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3040 * make sure that either the old or the new name pointer and length are
3041 * fetched. However, there may be mismatch between length and pointer.
3042 * The length cannot be trusted, we need to copy it byte-by-byte until
3043 * the length is reached or a null byte is found. It also prepends "/" at
3044 * the beginning of the name. The sequence number check at the caller will
3045 * retry it again when a d_move() does happen. So any garbage in the buffer
3046 * due to mismatched pointer and length will be discarded.
3048 * Data dependency barrier is needed to make sure that we see that terminating
3049 * NUL. Alpha strikes again, film at 11...
3051 static int prepend_name(char **buffer
, int *buflen
, const struct qstr
*name
)
3053 const char *dname
= ACCESS_ONCE(name
->name
);
3054 u32 dlen
= ACCESS_ONCE(name
->len
);
3057 smp_read_barrier_depends();
3059 *buflen
-= dlen
+ 1;
3061 return -ENAMETOOLONG
;
3062 p
= *buffer
-= dlen
+ 1;
3074 * prepend_path - Prepend path string to a buffer
3075 * @path: the dentry/vfsmount to report
3076 * @root: root vfsmnt/dentry
3077 * @buffer: pointer to the end of the buffer
3078 * @buflen: pointer to buffer length
3080 * The function will first try to write out the pathname without taking any
3081 * lock other than the RCU read lock to make sure that dentries won't go away.
3082 * It only checks the sequence number of the global rename_lock as any change
3083 * in the dentry's d_seq will be preceded by changes in the rename_lock
3084 * sequence number. If the sequence number had been changed, it will restart
3085 * the whole pathname back-tracing sequence again by taking the rename_lock.
3086 * In this case, there is no need to take the RCU read lock as the recursive
3087 * parent pointer references will keep the dentry chain alive as long as no
3088 * rename operation is performed.
3090 static int prepend_path(const struct path
*path
,
3091 const struct path
*root
,
3092 char **buffer
, int *buflen
)
3094 struct dentry
*dentry
;
3095 struct vfsmount
*vfsmnt
;
3098 unsigned seq
, m_seq
= 0;
3104 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3111 dentry
= path
->dentry
;
3113 mnt
= real_mount(vfsmnt
);
3114 read_seqbegin_or_lock(&rename_lock
, &seq
);
3115 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3116 struct dentry
* parent
;
3118 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3119 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
3121 if (dentry
!= vfsmnt
->mnt_root
) {
3128 if (mnt
!= parent
) {
3129 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
3135 error
= is_mounted(vfsmnt
) ? 1 : 2;
3138 parent
= dentry
->d_parent
;
3140 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3148 if (need_seqretry(&rename_lock
, seq
)) {
3152 done_seqretry(&rename_lock
, seq
);
3156 if (need_seqretry(&mount_lock
, m_seq
)) {
3160 done_seqretry(&mount_lock
, m_seq
);
3162 if (error
>= 0 && bptr
== *buffer
) {
3164 error
= -ENAMETOOLONG
;
3174 * __d_path - return the path of a dentry
3175 * @path: the dentry/vfsmount to report
3176 * @root: root vfsmnt/dentry
3177 * @buf: buffer to return value in
3178 * @buflen: buffer length
3180 * Convert a dentry into an ASCII path name.
3182 * Returns a pointer into the buffer or an error code if the
3183 * path was too long.
3185 * "buflen" should be positive.
3187 * If the path is not reachable from the supplied root, return %NULL.
3189 char *__d_path(const struct path
*path
,
3190 const struct path
*root
,
3191 char *buf
, int buflen
)
3193 char *res
= buf
+ buflen
;
3196 prepend(&res
, &buflen
, "\0", 1);
3197 error
= prepend_path(path
, root
, &res
, &buflen
);
3200 return ERR_PTR(error
);
3206 char *d_absolute_path(const struct path
*path
,
3207 char *buf
, int buflen
)
3209 struct path root
= {};
3210 char *res
= buf
+ buflen
;
3213 prepend(&res
, &buflen
, "\0", 1);
3214 error
= prepend_path(path
, &root
, &res
, &buflen
);
3219 return ERR_PTR(error
);
3224 * same as __d_path but appends "(deleted)" for unlinked files.
3226 static int path_with_deleted(const struct path
*path
,
3227 const struct path
*root
,
3228 char **buf
, int *buflen
)
3230 prepend(buf
, buflen
, "\0", 1);
3231 if (d_unlinked(path
->dentry
)) {
3232 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3237 return prepend_path(path
, root
, buf
, buflen
);
3240 static int prepend_unreachable(char **buffer
, int *buflen
)
3242 return prepend(buffer
, buflen
, "(unreachable)", 13);
3245 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3250 seq
= read_seqcount_begin(&fs
->seq
);
3252 } while (read_seqcount_retry(&fs
->seq
, seq
));
3256 * d_path - return the path of a dentry
3257 * @path: path to report
3258 * @buf: buffer to return value in
3259 * @buflen: buffer length
3261 * Convert a dentry into an ASCII path name. If the entry has been deleted
3262 * the string " (deleted)" is appended. Note that this is ambiguous.
3264 * Returns a pointer into the buffer or an error code if the path was
3265 * too long. Note: Callers should use the returned pointer, not the passed
3266 * in buffer, to use the name! The implementation often starts at an offset
3267 * into the buffer, and may leave 0 bytes at the start.
3269 * "buflen" should be positive.
3271 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3273 char *res
= buf
+ buflen
;
3278 * We have various synthetic filesystems that never get mounted. On
3279 * these filesystems dentries are never used for lookup purposes, and
3280 * thus don't need to be hashed. They also don't need a name until a
3281 * user wants to identify the object in /proc/pid/fd/. The little hack
3282 * below allows us to generate a name for these objects on demand:
3284 * Some pseudo inodes are mountable. When they are mounted
3285 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3286 * and instead have d_path return the mounted path.
3288 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3289 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3290 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3293 get_fs_root_rcu(current
->fs
, &root
);
3294 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3298 res
= ERR_PTR(error
);
3301 EXPORT_SYMBOL(d_path
);
3304 * Helper function for dentry_operations.d_dname() members
3306 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3307 const char *fmt
, ...)
3313 va_start(args
, fmt
);
3314 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3317 if (sz
> sizeof(temp
) || sz
> buflen
)
3318 return ERR_PTR(-ENAMETOOLONG
);
3320 buffer
+= buflen
- sz
;
3321 return memcpy(buffer
, temp
, sz
);
3324 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3326 char *end
= buffer
+ buflen
;
3327 /* these dentries are never renamed, so d_lock is not needed */
3328 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3329 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3330 prepend(&end
, &buflen
, "/", 1))
3331 end
= ERR_PTR(-ENAMETOOLONG
);
3334 EXPORT_SYMBOL(simple_dname
);
3337 * Write full pathname from the root of the filesystem into the buffer.
3339 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3341 struct dentry
*dentry
;
3354 prepend(&end
, &len
, "\0", 1);
3358 read_seqbegin_or_lock(&rename_lock
, &seq
);
3359 while (!IS_ROOT(dentry
)) {
3360 struct dentry
*parent
= dentry
->d_parent
;
3363 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3372 if (need_seqretry(&rename_lock
, seq
)) {
3376 done_seqretry(&rename_lock
, seq
);
3381 return ERR_PTR(-ENAMETOOLONG
);
3384 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3386 return __dentry_path(dentry
, buf
, buflen
);
3388 EXPORT_SYMBOL(dentry_path_raw
);
3390 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3395 if (d_unlinked(dentry
)) {
3397 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3401 retval
= __dentry_path(dentry
, buf
, buflen
);
3402 if (!IS_ERR(retval
) && p
)
3403 *p
= '/'; /* restore '/' overriden with '\0' */
3406 return ERR_PTR(-ENAMETOOLONG
);
3409 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3415 seq
= read_seqcount_begin(&fs
->seq
);
3418 } while (read_seqcount_retry(&fs
->seq
, seq
));
3422 * NOTE! The user-level library version returns a
3423 * character pointer. The kernel system call just
3424 * returns the length of the buffer filled (which
3425 * includes the ending '\0' character), or a negative
3426 * error value. So libc would do something like
3428 * char *getcwd(char * buf, size_t size)
3432 * retval = sys_getcwd(buf, size);
3439 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3442 struct path pwd
, root
;
3443 char *page
= __getname();
3449 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3452 if (!d_unlinked(pwd
.dentry
)) {
3454 char *cwd
= page
+ PATH_MAX
;
3455 int buflen
= PATH_MAX
;
3457 prepend(&cwd
, &buflen
, "\0", 1);
3458 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3464 /* Unreachable from current root */
3466 error
= prepend_unreachable(&cwd
, &buflen
);
3472 len
= PATH_MAX
+ page
- cwd
;
3475 if (copy_to_user(buf
, cwd
, len
))
3488 * Test whether new_dentry is a subdirectory of old_dentry.
3490 * Trivially implemented using the dcache structure
3494 * is_subdir - is new dentry a subdirectory of old_dentry
3495 * @new_dentry: new dentry
3496 * @old_dentry: old dentry
3498 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3499 * Returns false otherwise.
3500 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3503 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3508 if (new_dentry
== old_dentry
)
3512 /* for restarting inner loop in case of seq retry */
3513 seq
= read_seqbegin(&rename_lock
);
3515 * Need rcu_readlock to protect against the d_parent trashing
3519 if (d_ancestor(old_dentry
, new_dentry
))
3524 } while (read_seqretry(&rename_lock
, seq
));
3529 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3531 struct dentry
*root
= data
;
3532 if (dentry
!= root
) {
3533 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3536 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3537 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3538 dentry
->d_lockref
.count
--;
3541 return D_WALK_CONTINUE
;
3544 void d_genocide(struct dentry
*parent
)
3546 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3549 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3551 inode_dec_link_count(inode
);
3552 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3553 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3554 !d_unlinked(dentry
));
3555 spin_lock(&dentry
->d_parent
->d_lock
);
3556 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3557 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3558 (unsigned long long)inode
->i_ino
);
3559 spin_unlock(&dentry
->d_lock
);
3560 spin_unlock(&dentry
->d_parent
->d_lock
);
3561 d_instantiate(dentry
, inode
);
3563 EXPORT_SYMBOL(d_tmpfile
);
3565 static __initdata
unsigned long dhash_entries
;
3566 static int __init
set_dhash_entries(char *str
)
3570 dhash_entries
= simple_strtoul(str
, &str
, 0);
3573 __setup("dhash_entries=", set_dhash_entries
);
3575 static void __init
dcache_init_early(void)
3577 /* If hashes are distributed across NUMA nodes, defer
3578 * hash allocation until vmalloc space is available.
3584 alloc_large_system_hash("Dentry cache",
3585 sizeof(struct hlist_bl_head
),
3588 HASH_EARLY
| HASH_ZERO
,
3595 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
),
3621 /* SLAB cache for __getname() consumers */
3622 struct kmem_cache
*names_cachep __read_mostly
;
3623 EXPORT_SYMBOL(names_cachep
);
3625 EXPORT_SYMBOL(d_genocide
);
3627 void __init
vfs_caches_init_early(void)
3631 for (i
= 0; i
< ARRAY_SIZE(in_lookup_hashtable
); i
++)
3632 INIT_HLIST_BL_HEAD(&in_lookup_hashtable
[i
]);
3634 dcache_init_early();
3638 void __init
vfs_caches_init(void)
3640 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3641 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3646 files_maxfiles_init();