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 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
286 dentry
->d_inode
= inode
;
287 flags
= READ_ONCE(dentry
->d_flags
);
288 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
290 WRITE_ONCE(dentry
->d_flags
, flags
);
293 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
295 unsigned flags
= READ_ONCE(dentry
->d_flags
);
297 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
298 WRITE_ONCE(dentry
->d_flags
, flags
);
299 dentry
->d_inode
= NULL
;
302 static void dentry_free(struct dentry
*dentry
)
304 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
305 if (unlikely(dname_external(dentry
))) {
306 struct external_name
*p
= external_name(dentry
);
307 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
308 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
312 /* if dentry was never visible to RCU, immediate free is OK */
313 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
314 __d_free(&dentry
->d_u
.d_rcu
);
316 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
320 * Release the dentry's inode, using the filesystem
321 * d_iput() operation if defined.
323 static void dentry_unlink_inode(struct dentry
* dentry
)
324 __releases(dentry
->d_lock
)
325 __releases(dentry
->d_inode
->i_lock
)
327 struct inode
*inode
= dentry
->d_inode
;
328 bool hashed
= !d_unhashed(dentry
);
331 raw_write_seqcount_begin(&dentry
->d_seq
);
332 __d_clear_type_and_inode(dentry
);
333 hlist_del_init(&dentry
->d_u
.d_alias
);
335 raw_write_seqcount_end(&dentry
->d_seq
);
336 spin_unlock(&dentry
->d_lock
);
337 spin_unlock(&inode
->i_lock
);
339 fsnotify_inoderemove(inode
);
340 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
341 dentry
->d_op
->d_iput(dentry
, inode
);
347 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
348 * is in use - which includes both the "real" per-superblock
349 * LRU list _and_ the DCACHE_SHRINK_LIST use.
351 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
352 * on the shrink list (ie not on the superblock LRU list).
354 * The per-cpu "nr_dentry_unused" counters are updated with
355 * the DCACHE_LRU_LIST bit.
357 * These helper functions make sure we always follow the
358 * rules. d_lock must be held by the caller.
360 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
361 static void d_lru_add(struct dentry
*dentry
)
363 D_FLAG_VERIFY(dentry
, 0);
364 dentry
->d_flags
|= DCACHE_LRU_LIST
;
365 this_cpu_inc(nr_dentry_unused
);
366 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
369 static void d_lru_del(struct dentry
*dentry
)
371 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
372 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
373 this_cpu_dec(nr_dentry_unused
);
374 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
377 static void d_shrink_del(struct dentry
*dentry
)
379 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
380 list_del_init(&dentry
->d_lru
);
381 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
382 this_cpu_dec(nr_dentry_unused
);
385 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
387 D_FLAG_VERIFY(dentry
, 0);
388 list_add(&dentry
->d_lru
, list
);
389 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
390 this_cpu_inc(nr_dentry_unused
);
394 * These can only be called under the global LRU lock, ie during the
395 * callback for freeing the LRU list. "isolate" removes it from the
396 * LRU lists entirely, while shrink_move moves it to the indicated
399 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
401 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
402 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
403 this_cpu_dec(nr_dentry_unused
);
404 list_lru_isolate(lru
, &dentry
->d_lru
);
407 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
408 struct list_head
*list
)
410 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
411 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
412 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
416 * dentry_lru_(add|del)_list) must be called with d_lock held.
418 static void dentry_lru_add(struct dentry
*dentry
)
420 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
422 else if (unlikely(!(dentry
->d_flags
& DCACHE_REFERENCED
)))
423 dentry
->d_flags
|= DCACHE_REFERENCED
;
427 * d_drop - drop a dentry
428 * @dentry: dentry to drop
430 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
431 * be found through a VFS lookup any more. Note that this is different from
432 * deleting the dentry - d_delete will try to mark the dentry negative if
433 * possible, giving a successful _negative_ lookup, while d_drop will
434 * just make the cache lookup fail.
436 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
437 * reason (NFS timeouts or autofs deletes).
439 * __d_drop requires dentry->d_lock.
441 void __d_drop(struct dentry
*dentry
)
443 if (!d_unhashed(dentry
)) {
444 struct hlist_bl_head
*b
;
446 * Hashed dentries are normally on the dentry hashtable,
447 * with the exception of those newly allocated by
448 * d_obtain_alias, which are always IS_ROOT:
450 if (unlikely(IS_ROOT(dentry
)))
451 b
= &dentry
->d_sb
->s_anon
;
453 b
= d_hash(dentry
->d_name
.hash
);
456 __hlist_bl_del(&dentry
->d_hash
);
457 dentry
->d_hash
.pprev
= NULL
;
459 /* After this call, in-progress rcu-walk path lookup will fail. */
460 write_seqcount_invalidate(&dentry
->d_seq
);
463 EXPORT_SYMBOL(__d_drop
);
465 void d_drop(struct dentry
*dentry
)
467 spin_lock(&dentry
->d_lock
);
469 spin_unlock(&dentry
->d_lock
);
471 EXPORT_SYMBOL(d_drop
);
473 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
477 * Inform d_walk() and shrink_dentry_list() that we are no longer
478 * attached to the dentry tree
480 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
481 if (unlikely(list_empty(&dentry
->d_child
)))
483 __list_del_entry(&dentry
->d_child
);
485 * Cursors can move around the list of children. While we'd been
486 * a normal list member, it didn't matter - ->d_child.next would've
487 * been updated. However, from now on it won't be and for the
488 * things like d_walk() it might end up with a nasty surprise.
489 * Normally d_walk() doesn't care about cursors moving around -
490 * ->d_lock on parent prevents that and since a cursor has no children
491 * of its own, we get through it without ever unlocking the parent.
492 * There is one exception, though - if we ascend from a child that
493 * gets killed as soon as we unlock it, the next sibling is found
494 * using the value left in its ->d_child.next. And if _that_
495 * pointed to a cursor, and cursor got moved (e.g. by lseek())
496 * before d_walk() regains parent->d_lock, we'll end up skipping
497 * everything the cursor had been moved past.
499 * Solution: make sure that the pointer left behind in ->d_child.next
500 * points to something that won't be moving around. I.e. skip the
503 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
504 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
505 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
507 dentry
->d_child
.next
= next
->d_child
.next
;
511 static void __dentry_kill(struct dentry
*dentry
)
513 struct dentry
*parent
= NULL
;
514 bool can_free
= true;
515 if (!IS_ROOT(dentry
))
516 parent
= dentry
->d_parent
;
519 * The dentry is now unrecoverably dead to the world.
521 lockref_mark_dead(&dentry
->d_lockref
);
524 * inform the fs via d_prune that this dentry is about to be
525 * unhashed and destroyed.
527 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
528 dentry
->d_op
->d_prune(dentry
);
530 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
531 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
534 /* if it was on the hash then remove it */
536 dentry_unlist(dentry
, parent
);
538 spin_unlock(&parent
->d_lock
);
540 dentry_unlink_inode(dentry
);
542 spin_unlock(&dentry
->d_lock
);
543 this_cpu_dec(nr_dentry
);
544 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
545 dentry
->d_op
->d_release(dentry
);
547 spin_lock(&dentry
->d_lock
);
548 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
549 dentry
->d_flags
|= DCACHE_MAY_FREE
;
552 spin_unlock(&dentry
->d_lock
);
553 if (likely(can_free
))
558 * Finish off a dentry we've decided to kill.
559 * dentry->d_lock must be held, returns with it unlocked.
560 * If ref is non-zero, then decrement the refcount too.
561 * Returns dentry requiring refcount drop, or NULL if we're done.
563 static struct dentry
*dentry_kill(struct dentry
*dentry
)
564 __releases(dentry
->d_lock
)
566 struct inode
*inode
= dentry
->d_inode
;
567 struct dentry
*parent
= NULL
;
569 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
572 if (!IS_ROOT(dentry
)) {
573 parent
= dentry
->d_parent
;
574 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
576 spin_unlock(&inode
->i_lock
);
581 __dentry_kill(dentry
);
585 spin_unlock(&dentry
->d_lock
);
586 return dentry
; /* try again with same dentry */
589 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
591 struct dentry
*parent
= dentry
->d_parent
;
594 if (unlikely(dentry
->d_lockref
.count
< 0))
596 if (likely(spin_trylock(&parent
->d_lock
)))
599 spin_unlock(&dentry
->d_lock
);
601 parent
= ACCESS_ONCE(dentry
->d_parent
);
602 spin_lock(&parent
->d_lock
);
604 * We can't blindly lock dentry until we are sure
605 * that we won't violate the locking order.
606 * Any changes of dentry->d_parent must have
607 * been done with parent->d_lock held, so
608 * spin_lock() above is enough of a barrier
609 * for checking if it's still our child.
611 if (unlikely(parent
!= dentry
->d_parent
)) {
612 spin_unlock(&parent
->d_lock
);
616 if (parent
!= dentry
)
617 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
624 * Try to do a lockless dput(), and return whether that was successful.
626 * If unsuccessful, we return false, having already taken the dentry lock.
628 * The caller needs to hold the RCU read lock, so that the dentry is
629 * guaranteed to stay around even if the refcount goes down to zero!
631 static inline bool fast_dput(struct dentry
*dentry
)
634 unsigned int d_flags
;
637 * If we have a d_op->d_delete() operation, we sould not
638 * let the dentry count go to zero, so use "put_or_lock".
640 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
641 return lockref_put_or_lock(&dentry
->d_lockref
);
644 * .. otherwise, we can try to just decrement the
645 * lockref optimistically.
647 ret
= lockref_put_return(&dentry
->d_lockref
);
650 * If the lockref_put_return() failed due to the lock being held
651 * by somebody else, the fast path has failed. We will need to
652 * get the lock, and then check the count again.
654 if (unlikely(ret
< 0)) {
655 spin_lock(&dentry
->d_lock
);
656 if (dentry
->d_lockref
.count
> 1) {
657 dentry
->d_lockref
.count
--;
658 spin_unlock(&dentry
->d_lock
);
665 * If we weren't the last ref, we're done.
671 * Careful, careful. The reference count went down
672 * to zero, but we don't hold the dentry lock, so
673 * somebody else could get it again, and do another
674 * dput(), and we need to not race with that.
676 * However, there is a very special and common case
677 * where we don't care, because there is nothing to
678 * do: the dentry is still hashed, it does not have
679 * a 'delete' op, and it's referenced and already on
682 * NOTE! Since we aren't locked, these values are
683 * not "stable". However, it is sufficient that at
684 * some point after we dropped the reference the
685 * dentry was hashed and the flags had the proper
686 * value. Other dentry users may have re-gotten
687 * a reference to the dentry and change that, but
688 * our work is done - we can leave the dentry
689 * around with a zero refcount.
692 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
693 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
695 /* Nothing to do? Dropping the reference was all we needed? */
696 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
700 * Not the fast normal case? Get the lock. We've already decremented
701 * the refcount, but we'll need to re-check the situation after
704 spin_lock(&dentry
->d_lock
);
707 * Did somebody else grab a reference to it in the meantime, and
708 * we're no longer the last user after all? Alternatively, somebody
709 * else could have killed it and marked it dead. Either way, we
710 * don't need to do anything else.
712 if (dentry
->d_lockref
.count
) {
713 spin_unlock(&dentry
->d_lock
);
718 * Re-get the reference we optimistically dropped. We hold the
719 * lock, and we just tested that it was zero, so we can just
722 dentry
->d_lockref
.count
= 1;
730 * This is complicated by the fact that we do not want to put
731 * dentries that are no longer on any hash chain on the unused
732 * list: we'd much rather just get rid of them immediately.
734 * However, that implies that we have to traverse the dentry
735 * tree upwards to the parents which might _also_ now be
736 * scheduled for deletion (it may have been only waiting for
737 * its last child to go away).
739 * This tail recursion is done by hand as we don't want to depend
740 * on the compiler to always get this right (gcc generally doesn't).
741 * Real recursion would eat up our stack space.
745 * dput - release a dentry
746 * @dentry: dentry to release
748 * Release a dentry. This will drop the usage count and if appropriate
749 * call the dentry unlink method as well as removing it from the queues and
750 * releasing its resources. If the parent dentries were scheduled for release
751 * they too may now get deleted.
753 void dput(struct dentry
*dentry
)
755 if (unlikely(!dentry
))
762 if (likely(fast_dput(dentry
))) {
767 /* Slow case: now with the dentry lock held */
770 WARN_ON(d_in_lookup(dentry
));
772 /* Unreachable? Get rid of it */
773 if (unlikely(d_unhashed(dentry
)))
776 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
779 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
780 if (dentry
->d_op
->d_delete(dentry
))
784 dentry_lru_add(dentry
);
786 dentry
->d_lockref
.count
--;
787 spin_unlock(&dentry
->d_lock
);
791 dentry
= dentry_kill(dentry
);
800 /* This must be called with d_lock held */
801 static inline void __dget_dlock(struct dentry
*dentry
)
803 dentry
->d_lockref
.count
++;
806 static inline void __dget(struct dentry
*dentry
)
808 lockref_get(&dentry
->d_lockref
);
811 struct dentry
*dget_parent(struct dentry
*dentry
)
817 * Do optimistic parent lookup without any
821 ret
= ACCESS_ONCE(dentry
->d_parent
);
822 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
824 if (likely(gotref
)) {
825 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
832 * Don't need rcu_dereference because we re-check it was correct under
836 ret
= dentry
->d_parent
;
837 spin_lock(&ret
->d_lock
);
838 if (unlikely(ret
!= dentry
->d_parent
)) {
839 spin_unlock(&ret
->d_lock
);
844 BUG_ON(!ret
->d_lockref
.count
);
845 ret
->d_lockref
.count
++;
846 spin_unlock(&ret
->d_lock
);
849 EXPORT_SYMBOL(dget_parent
);
852 * d_find_alias - grab a hashed alias of inode
853 * @inode: inode in question
855 * If inode has a hashed alias, or is a directory and has any alias,
856 * acquire the reference to alias and return it. Otherwise return NULL.
857 * Notice that if inode is a directory there can be only one alias and
858 * it can be unhashed only if it has no children, or if it is the root
859 * of a filesystem, or if the directory was renamed and d_revalidate
860 * was the first vfs operation to notice.
862 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
863 * any other hashed alias over that one.
865 static struct dentry
*__d_find_alias(struct inode
*inode
)
867 struct dentry
*alias
, *discon_alias
;
871 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
872 spin_lock(&alias
->d_lock
);
873 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
874 if (IS_ROOT(alias
) &&
875 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
876 discon_alias
= alias
;
879 spin_unlock(&alias
->d_lock
);
883 spin_unlock(&alias
->d_lock
);
886 alias
= discon_alias
;
887 spin_lock(&alias
->d_lock
);
888 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
890 spin_unlock(&alias
->d_lock
);
893 spin_unlock(&alias
->d_lock
);
899 struct dentry
*d_find_alias(struct inode
*inode
)
901 struct dentry
*de
= NULL
;
903 if (!hlist_empty(&inode
->i_dentry
)) {
904 spin_lock(&inode
->i_lock
);
905 de
= __d_find_alias(inode
);
906 spin_unlock(&inode
->i_lock
);
910 EXPORT_SYMBOL(d_find_alias
);
913 * Try to kill dentries associated with this inode.
914 * WARNING: you must own a reference to inode.
916 void d_prune_aliases(struct inode
*inode
)
918 struct dentry
*dentry
;
920 spin_lock(&inode
->i_lock
);
921 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
922 spin_lock(&dentry
->d_lock
);
923 if (!dentry
->d_lockref
.count
) {
924 struct dentry
*parent
= lock_parent(dentry
);
925 if (likely(!dentry
->d_lockref
.count
)) {
926 __dentry_kill(dentry
);
931 spin_unlock(&parent
->d_lock
);
933 spin_unlock(&dentry
->d_lock
);
935 spin_unlock(&inode
->i_lock
);
937 EXPORT_SYMBOL(d_prune_aliases
);
939 static void shrink_dentry_list(struct list_head
*list
)
941 struct dentry
*dentry
, *parent
;
943 while (!list_empty(list
)) {
945 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
946 spin_lock(&dentry
->d_lock
);
947 parent
= lock_parent(dentry
);
950 * The dispose list is isolated and dentries are not accounted
951 * to the LRU here, so we can simply remove it from the list
952 * here regardless of whether it is referenced or not.
954 d_shrink_del(dentry
);
957 * We found an inuse dentry which was not removed from
958 * the LRU because of laziness during lookup. Do not free it.
960 if (dentry
->d_lockref
.count
> 0) {
961 spin_unlock(&dentry
->d_lock
);
963 spin_unlock(&parent
->d_lock
);
968 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
969 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
970 spin_unlock(&dentry
->d_lock
);
972 spin_unlock(&parent
->d_lock
);
978 inode
= dentry
->d_inode
;
979 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
980 d_shrink_add(dentry
, list
);
981 spin_unlock(&dentry
->d_lock
);
983 spin_unlock(&parent
->d_lock
);
987 __dentry_kill(dentry
);
990 * We need to prune ancestors too. This is necessary to prevent
991 * quadratic behavior of shrink_dcache_parent(), but is also
992 * expected to be beneficial in reducing dentry cache
996 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
997 parent
= lock_parent(dentry
);
998 if (dentry
->d_lockref
.count
!= 1) {
999 dentry
->d_lockref
.count
--;
1000 spin_unlock(&dentry
->d_lock
);
1002 spin_unlock(&parent
->d_lock
);
1005 inode
= dentry
->d_inode
; /* can't be NULL */
1006 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1007 spin_unlock(&dentry
->d_lock
);
1009 spin_unlock(&parent
->d_lock
);
1013 __dentry_kill(dentry
);
1019 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1020 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1022 struct list_head
*freeable
= arg
;
1023 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1027 * we are inverting the lru lock/dentry->d_lock here,
1028 * so use a trylock. If we fail to get the lock, just skip
1031 if (!spin_trylock(&dentry
->d_lock
))
1035 * Referenced dentries are still in use. If they have active
1036 * counts, just remove them from the LRU. Otherwise give them
1037 * another pass through the LRU.
1039 if (dentry
->d_lockref
.count
) {
1040 d_lru_isolate(lru
, dentry
);
1041 spin_unlock(&dentry
->d_lock
);
1045 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1046 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1047 spin_unlock(&dentry
->d_lock
);
1050 * The list move itself will be made by the common LRU code. At
1051 * this point, we've dropped the dentry->d_lock but keep the
1052 * lru lock. This is safe to do, since every list movement is
1053 * protected by the lru lock even if both locks are held.
1055 * This is guaranteed by the fact that all LRU management
1056 * functions are intermediated by the LRU API calls like
1057 * list_lru_add and list_lru_del. List movement in this file
1058 * only ever occur through this functions or through callbacks
1059 * like this one, that are called from the LRU API.
1061 * The only exceptions to this are functions like
1062 * shrink_dentry_list, and code that first checks for the
1063 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1064 * operating only with stack provided lists after they are
1065 * properly isolated from the main list. It is thus, always a
1071 d_lru_shrink_move(lru
, dentry
, freeable
);
1072 spin_unlock(&dentry
->d_lock
);
1078 * prune_dcache_sb - shrink the dcache
1080 * @sc: shrink control, passed to list_lru_shrink_walk()
1082 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1083 * is done when we need more memory and called from the superblock shrinker
1086 * This function may fail to free any resources if all the dentries are in
1089 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1094 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1095 dentry_lru_isolate
, &dispose
);
1096 shrink_dentry_list(&dispose
);
1100 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1101 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1103 struct list_head
*freeable
= arg
;
1104 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1107 * we are inverting the lru lock/dentry->d_lock here,
1108 * so use a trylock. If we fail to get the lock, just skip
1111 if (!spin_trylock(&dentry
->d_lock
))
1114 d_lru_shrink_move(lru
, dentry
, freeable
);
1115 spin_unlock(&dentry
->d_lock
);
1122 * shrink_dcache_sb - shrink dcache for a superblock
1125 * Shrink the dcache for the specified super block. This is used to free
1126 * the dcache before unmounting a file system.
1128 void shrink_dcache_sb(struct super_block
*sb
)
1135 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1136 dentry_lru_isolate_shrink
, &dispose
, UINT_MAX
);
1138 this_cpu_sub(nr_dentry_unused
, freed
);
1139 shrink_dentry_list(&dispose
);
1140 } while (freed
> 0);
1142 EXPORT_SYMBOL(shrink_dcache_sb
);
1145 * enum d_walk_ret - action to talke during tree walk
1146 * @D_WALK_CONTINUE: contrinue walk
1147 * @D_WALK_QUIT: quit walk
1148 * @D_WALK_NORETRY: quit when retry is needed
1149 * @D_WALK_SKIP: skip this dentry and its children
1159 * d_walk - walk the dentry tree
1160 * @parent: start of walk
1161 * @data: data passed to @enter() and @finish()
1162 * @enter: callback when first entering the dentry
1163 * @finish: callback when successfully finished the walk
1165 * The @enter() and @finish() callbacks are called with d_lock held.
1167 static void d_walk(struct dentry
*parent
, void *data
,
1168 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1169 void (*finish
)(void *))
1171 struct dentry
*this_parent
;
1172 struct list_head
*next
;
1174 enum d_walk_ret ret
;
1178 read_seqbegin_or_lock(&rename_lock
, &seq
);
1179 this_parent
= parent
;
1180 spin_lock(&this_parent
->d_lock
);
1182 ret
= enter(data
, this_parent
);
1184 case D_WALK_CONTINUE
:
1189 case D_WALK_NORETRY
:
1194 next
= this_parent
->d_subdirs
.next
;
1196 while (next
!= &this_parent
->d_subdirs
) {
1197 struct list_head
*tmp
= next
;
1198 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1201 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1204 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1206 ret
= enter(data
, dentry
);
1208 case D_WALK_CONTINUE
:
1211 spin_unlock(&dentry
->d_lock
);
1213 case D_WALK_NORETRY
:
1217 spin_unlock(&dentry
->d_lock
);
1221 if (!list_empty(&dentry
->d_subdirs
)) {
1222 spin_unlock(&this_parent
->d_lock
);
1223 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1224 this_parent
= dentry
;
1225 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1228 spin_unlock(&dentry
->d_lock
);
1231 * All done at this level ... ascend and resume the search.
1235 if (this_parent
!= parent
) {
1236 struct dentry
*child
= this_parent
;
1237 this_parent
= child
->d_parent
;
1239 spin_unlock(&child
->d_lock
);
1240 spin_lock(&this_parent
->d_lock
);
1242 /* might go back up the wrong parent if we have had a rename. */
1243 if (need_seqretry(&rename_lock
, seq
))
1245 /* go into the first sibling still alive */
1247 next
= child
->d_child
.next
;
1248 if (next
== &this_parent
->d_subdirs
)
1250 child
= list_entry(next
, struct dentry
, d_child
);
1251 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1255 if (need_seqretry(&rename_lock
, seq
))
1262 spin_unlock(&this_parent
->d_lock
);
1263 done_seqretry(&rename_lock
, seq
);
1267 spin_unlock(&this_parent
->d_lock
);
1276 struct check_mount
{
1277 struct vfsmount
*mnt
;
1278 unsigned int mounted
;
1281 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1283 struct check_mount
*info
= data
;
1284 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1286 if (likely(!d_mountpoint(dentry
)))
1287 return D_WALK_CONTINUE
;
1288 if (__path_is_mountpoint(&path
)) {
1292 return D_WALK_CONTINUE
;
1296 * path_has_submounts - check for mounts over a dentry in the
1297 * current namespace.
1298 * @parent: path to check.
1300 * Return true if the parent or its subdirectories contain
1301 * a mount point in the current namespace.
1303 int path_has_submounts(const struct path
*parent
)
1305 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1307 read_seqlock_excl(&mount_lock
);
1308 d_walk(parent
->dentry
, &data
, path_check_mount
, NULL
);
1309 read_sequnlock_excl(&mount_lock
);
1311 return data
.mounted
;
1313 EXPORT_SYMBOL(path_has_submounts
);
1316 * Called by mount code to set a mountpoint and check if the mountpoint is
1317 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1318 * subtree can become unreachable).
1320 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1321 * this reason take rename_lock and d_lock on dentry and ancestors.
1323 int d_set_mounted(struct dentry
*dentry
)
1327 write_seqlock(&rename_lock
);
1328 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1329 /* Need exclusion wrt. d_invalidate() */
1330 spin_lock(&p
->d_lock
);
1331 if (unlikely(d_unhashed(p
))) {
1332 spin_unlock(&p
->d_lock
);
1335 spin_unlock(&p
->d_lock
);
1337 spin_lock(&dentry
->d_lock
);
1338 if (!d_unlinked(dentry
)) {
1340 if (!d_mountpoint(dentry
)) {
1341 dentry
->d_flags
|= DCACHE_MOUNTED
;
1345 spin_unlock(&dentry
->d_lock
);
1347 write_sequnlock(&rename_lock
);
1352 * Search the dentry child list of the specified parent,
1353 * and move any unused dentries to the end of the unused
1354 * list for prune_dcache(). We descend to the next level
1355 * whenever the d_subdirs list is non-empty and continue
1358 * It returns zero iff there are no unused children,
1359 * otherwise it returns the number of children moved to
1360 * the end of the unused list. This may not be the total
1361 * number of unused children, because select_parent can
1362 * drop the lock and return early due to latency
1366 struct select_data
{
1367 struct dentry
*start
;
1368 struct list_head dispose
;
1372 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1374 struct select_data
*data
= _data
;
1375 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1377 if (data
->start
== dentry
)
1380 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1383 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1385 if (!dentry
->d_lockref
.count
) {
1386 d_shrink_add(dentry
, &data
->dispose
);
1391 * We can return to the caller if we have found some (this
1392 * ensures forward progress). We'll be coming back to find
1395 if (!list_empty(&data
->dispose
))
1396 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1402 * shrink_dcache_parent - prune dcache
1403 * @parent: parent of entries to prune
1405 * Prune the dcache to remove unused children of the parent dentry.
1407 void shrink_dcache_parent(struct dentry
*parent
)
1410 struct select_data data
;
1412 INIT_LIST_HEAD(&data
.dispose
);
1413 data
.start
= parent
;
1416 d_walk(parent
, &data
, select_collect
, NULL
);
1420 shrink_dentry_list(&data
.dispose
);
1424 EXPORT_SYMBOL(shrink_dcache_parent
);
1426 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1428 /* it has busy descendents; complain about those instead */
1429 if (!list_empty(&dentry
->d_subdirs
))
1430 return D_WALK_CONTINUE
;
1432 /* root with refcount 1 is fine */
1433 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1434 return D_WALK_CONTINUE
;
1436 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1437 " still in use (%d) [unmount of %s %s]\n",
1440 dentry
->d_inode
->i_ino
: 0UL,
1442 dentry
->d_lockref
.count
,
1443 dentry
->d_sb
->s_type
->name
,
1444 dentry
->d_sb
->s_id
);
1446 return D_WALK_CONTINUE
;
1449 static void do_one_tree(struct dentry
*dentry
)
1451 shrink_dcache_parent(dentry
);
1452 d_walk(dentry
, dentry
, umount_check
, NULL
);
1458 * destroy the dentries attached to a superblock on unmounting
1460 void shrink_dcache_for_umount(struct super_block
*sb
)
1462 struct dentry
*dentry
;
1464 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1466 dentry
= sb
->s_root
;
1468 do_one_tree(dentry
);
1470 while (!hlist_bl_empty(&sb
->s_anon
)) {
1471 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1472 do_one_tree(dentry
);
1476 struct detach_data
{
1477 struct select_data select
;
1478 struct dentry
*mountpoint
;
1480 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1482 struct detach_data
*data
= _data
;
1484 if (d_mountpoint(dentry
)) {
1485 __dget_dlock(dentry
);
1486 data
->mountpoint
= dentry
;
1490 return select_collect(&data
->select
, dentry
);
1493 static void check_and_drop(void *_data
)
1495 struct detach_data
*data
= _data
;
1497 if (!data
->mountpoint
&& list_empty(&data
->select
.dispose
))
1498 __d_drop(data
->select
.start
);
1502 * d_invalidate - detach submounts, prune dcache, and drop
1503 * @dentry: dentry to invalidate (aka detach, prune and drop)
1507 * The final d_drop is done as an atomic operation relative to
1508 * rename_lock ensuring there are no races with d_set_mounted. This
1509 * ensures there are no unhashed dentries on the path to a mountpoint.
1511 void d_invalidate(struct dentry
*dentry
)
1514 * If it's already been dropped, return OK.
1516 spin_lock(&dentry
->d_lock
);
1517 if (d_unhashed(dentry
)) {
1518 spin_unlock(&dentry
->d_lock
);
1521 spin_unlock(&dentry
->d_lock
);
1523 /* Negative dentries can be dropped without further checks */
1524 if (!dentry
->d_inode
) {
1530 struct detach_data data
;
1532 data
.mountpoint
= NULL
;
1533 INIT_LIST_HEAD(&data
.select
.dispose
);
1534 data
.select
.start
= dentry
;
1535 data
.select
.found
= 0;
1537 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1539 if (!list_empty(&data
.select
.dispose
))
1540 shrink_dentry_list(&data
.select
.dispose
);
1541 else if (!data
.mountpoint
)
1544 if (data
.mountpoint
) {
1545 detach_mounts(data
.mountpoint
);
1546 dput(data
.mountpoint
);
1551 EXPORT_SYMBOL(d_invalidate
);
1554 * __d_alloc - allocate a dcache entry
1555 * @sb: filesystem it will belong to
1556 * @name: qstr of the name
1558 * Allocates a dentry. It returns %NULL if there is insufficient memory
1559 * available. On a success the dentry is returned. The name passed in is
1560 * copied and the copy passed in may be reused after this call.
1563 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1565 struct dentry
*dentry
;
1569 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1574 * We guarantee that the inline name is always NUL-terminated.
1575 * This way the memcpy() done by the name switching in rename
1576 * will still always have a NUL at the end, even if we might
1577 * be overwriting an internal NUL character
1579 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1580 if (unlikely(!name
)) {
1581 static const struct qstr anon
= QSTR_INIT("/", 1);
1583 dname
= dentry
->d_iname
;
1584 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1585 size_t size
= offsetof(struct external_name
, name
[1]);
1586 struct external_name
*p
= kmalloc(size
+ name
->len
,
1587 GFP_KERNEL_ACCOUNT
);
1589 kmem_cache_free(dentry_cache
, dentry
);
1592 atomic_set(&p
->u
.count
, 1);
1594 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1595 kasan_unpoison_shadow(dname
,
1596 round_up(name
->len
+ 1, sizeof(unsigned long)));
1598 dname
= dentry
->d_iname
;
1601 dentry
->d_name
.len
= name
->len
;
1602 dentry
->d_name
.hash
= name
->hash
;
1603 memcpy(dname
, name
->name
, name
->len
);
1604 dname
[name
->len
] = 0;
1606 /* Make sure we always see the terminating NUL character */
1608 dentry
->d_name
.name
= dname
;
1610 dentry
->d_lockref
.count
= 1;
1611 dentry
->d_flags
= 0;
1612 spin_lock_init(&dentry
->d_lock
);
1613 seqcount_init(&dentry
->d_seq
);
1614 dentry
->d_inode
= NULL
;
1615 dentry
->d_parent
= dentry
;
1617 dentry
->d_op
= NULL
;
1618 dentry
->d_fsdata
= NULL
;
1619 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1620 INIT_LIST_HEAD(&dentry
->d_lru
);
1621 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1622 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1623 INIT_LIST_HEAD(&dentry
->d_child
);
1624 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1626 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1627 err
= dentry
->d_op
->d_init(dentry
);
1629 if (dname_external(dentry
))
1630 kfree(external_name(dentry
));
1631 kmem_cache_free(dentry_cache
, dentry
);
1636 this_cpu_inc(nr_dentry
);
1642 * d_alloc - allocate a dcache entry
1643 * @parent: parent of entry to allocate
1644 * @name: qstr of the name
1646 * Allocates a dentry. It returns %NULL if there is insufficient memory
1647 * available. On a success the dentry is returned. The name passed in is
1648 * copied and the copy passed in may be reused after this call.
1650 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1652 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1655 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1656 spin_lock(&parent
->d_lock
);
1658 * don't need child lock because it is not subject
1659 * to concurrency here
1661 __dget_dlock(parent
);
1662 dentry
->d_parent
= parent
;
1663 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1664 spin_unlock(&parent
->d_lock
);
1668 EXPORT_SYMBOL(d_alloc
);
1670 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1672 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1674 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1675 dentry
->d_parent
= dget(parent
);
1681 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1682 * @sb: the superblock
1683 * @name: qstr of the name
1685 * For a filesystem that just pins its dentries in memory and never
1686 * performs lookups at all, return an unhashed IS_ROOT dentry.
1688 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1690 return __d_alloc(sb
, name
);
1692 EXPORT_SYMBOL(d_alloc_pseudo
);
1694 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1699 q
.hash_len
= hashlen_string(parent
, name
);
1700 return d_alloc(parent
, &q
);
1702 EXPORT_SYMBOL(d_alloc_name
);
1704 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1706 WARN_ON_ONCE(dentry
->d_op
);
1707 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1709 DCACHE_OP_REVALIDATE
|
1710 DCACHE_OP_WEAK_REVALIDATE
|
1717 dentry
->d_flags
|= DCACHE_OP_HASH
;
1719 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1720 if (op
->d_revalidate
)
1721 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1722 if (op
->d_weak_revalidate
)
1723 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1725 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1727 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1729 dentry
->d_flags
|= DCACHE_OP_REAL
;
1732 EXPORT_SYMBOL(d_set_d_op
);
1736 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1737 * @dentry - The dentry to mark
1739 * Mark a dentry as falling through to the lower layer (as set with
1740 * d_pin_lower()). This flag may be recorded on the medium.
1742 void d_set_fallthru(struct dentry
*dentry
)
1744 spin_lock(&dentry
->d_lock
);
1745 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1746 spin_unlock(&dentry
->d_lock
);
1748 EXPORT_SYMBOL(d_set_fallthru
);
1750 static unsigned d_flags_for_inode(struct inode
*inode
)
1752 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1755 return DCACHE_MISS_TYPE
;
1757 if (S_ISDIR(inode
->i_mode
)) {
1758 add_flags
= DCACHE_DIRECTORY_TYPE
;
1759 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1760 if (unlikely(!inode
->i_op
->lookup
))
1761 add_flags
= DCACHE_AUTODIR_TYPE
;
1763 inode
->i_opflags
|= IOP_LOOKUP
;
1765 goto type_determined
;
1768 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1769 if (unlikely(inode
->i_op
->get_link
)) {
1770 add_flags
= DCACHE_SYMLINK_TYPE
;
1771 goto type_determined
;
1773 inode
->i_opflags
|= IOP_NOFOLLOW
;
1776 if (unlikely(!S_ISREG(inode
->i_mode
)))
1777 add_flags
= DCACHE_SPECIAL_TYPE
;
1780 if (unlikely(IS_AUTOMOUNT(inode
)))
1781 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1785 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1787 unsigned add_flags
= d_flags_for_inode(inode
);
1788 WARN_ON(d_in_lookup(dentry
));
1790 spin_lock(&dentry
->d_lock
);
1791 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1792 raw_write_seqcount_begin(&dentry
->d_seq
);
1793 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1794 raw_write_seqcount_end(&dentry
->d_seq
);
1795 fsnotify_update_flags(dentry
);
1796 spin_unlock(&dentry
->d_lock
);
1800 * d_instantiate - fill in inode information for a dentry
1801 * @entry: dentry to complete
1802 * @inode: inode to attach to this dentry
1804 * Fill in inode information in the entry.
1806 * This turns negative dentries into productive full members
1809 * NOTE! This assumes that the inode count has been incremented
1810 * (or otherwise set) by the caller to indicate that it is now
1811 * in use by the dcache.
1814 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1816 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1818 security_d_instantiate(entry
, inode
);
1819 spin_lock(&inode
->i_lock
);
1820 __d_instantiate(entry
, inode
);
1821 spin_unlock(&inode
->i_lock
);
1824 EXPORT_SYMBOL(d_instantiate
);
1827 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1828 * @entry: dentry to complete
1829 * @inode: inode to attach to this dentry
1831 * Fill in inode information in the entry. If a directory alias is found, then
1832 * return an error (and drop inode). Together with d_materialise_unique() this
1833 * guarantees that a directory inode may never have more than one alias.
1835 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1837 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1839 security_d_instantiate(entry
, inode
);
1840 spin_lock(&inode
->i_lock
);
1841 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1842 spin_unlock(&inode
->i_lock
);
1846 __d_instantiate(entry
, inode
);
1847 spin_unlock(&inode
->i_lock
);
1851 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1853 struct dentry
*d_make_root(struct inode
*root_inode
)
1855 struct dentry
*res
= NULL
;
1858 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1860 d_instantiate(res
, root_inode
);
1866 EXPORT_SYMBOL(d_make_root
);
1868 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1870 struct dentry
*alias
;
1872 if (hlist_empty(&inode
->i_dentry
))
1874 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1880 * d_find_any_alias - find any alias for a given inode
1881 * @inode: inode to find an alias for
1883 * If any aliases exist for the given inode, take and return a
1884 * reference for one of them. If no aliases exist, return %NULL.
1886 struct dentry
*d_find_any_alias(struct inode
*inode
)
1890 spin_lock(&inode
->i_lock
);
1891 de
= __d_find_any_alias(inode
);
1892 spin_unlock(&inode
->i_lock
);
1895 EXPORT_SYMBOL(d_find_any_alias
);
1897 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1904 return ERR_PTR(-ESTALE
);
1906 return ERR_CAST(inode
);
1908 res
= d_find_any_alias(inode
);
1912 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1914 res
= ERR_PTR(-ENOMEM
);
1918 security_d_instantiate(tmp
, inode
);
1919 spin_lock(&inode
->i_lock
);
1920 res
= __d_find_any_alias(inode
);
1922 spin_unlock(&inode
->i_lock
);
1927 /* attach a disconnected dentry */
1928 add_flags
= d_flags_for_inode(inode
);
1931 add_flags
|= DCACHE_DISCONNECTED
;
1933 spin_lock(&tmp
->d_lock
);
1934 __d_set_inode_and_type(tmp
, inode
, add_flags
);
1935 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1936 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1937 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1938 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1939 spin_unlock(&tmp
->d_lock
);
1940 spin_unlock(&inode
->i_lock
);
1950 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1951 * @inode: inode to allocate the dentry for
1953 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1954 * similar open by handle operations. The returned dentry may be anonymous,
1955 * or may have a full name (if the inode was already in the cache).
1957 * When called on a directory inode, we must ensure that the inode only ever
1958 * has one dentry. If a dentry is found, that is returned instead of
1959 * allocating a new one.
1961 * On successful return, the reference to the inode has been transferred
1962 * to the dentry. In case of an error the reference on the inode is released.
1963 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1964 * be passed in and the error will be propagated to the return value,
1965 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1967 struct dentry
*d_obtain_alias(struct inode
*inode
)
1969 return __d_obtain_alias(inode
, 1);
1971 EXPORT_SYMBOL(d_obtain_alias
);
1974 * d_obtain_root - find or allocate a dentry for a given inode
1975 * @inode: inode to allocate the dentry for
1977 * Obtain an IS_ROOT dentry for the root of a filesystem.
1979 * We must ensure that directory inodes only ever have one dentry. If a
1980 * dentry is found, that is returned instead of allocating a new one.
1982 * On successful return, the reference to the inode has been transferred
1983 * to the dentry. In case of an error the reference on the inode is
1984 * released. A %NULL or IS_ERR inode may be passed in and will be the
1985 * error will be propagate to the return value, with a %NULL @inode
1986 * replaced by ERR_PTR(-ESTALE).
1988 struct dentry
*d_obtain_root(struct inode
*inode
)
1990 return __d_obtain_alias(inode
, 0);
1992 EXPORT_SYMBOL(d_obtain_root
);
1995 * d_add_ci - lookup or allocate new dentry with case-exact name
1996 * @inode: the inode case-insensitive lookup has found
1997 * @dentry: the negative dentry that was passed to the parent's lookup func
1998 * @name: the case-exact name to be associated with the returned dentry
2000 * This is to avoid filling the dcache with case-insensitive names to the
2001 * same inode, only the actual correct case is stored in the dcache for
2002 * case-insensitive filesystems.
2004 * For a case-insensitive lookup match and if the the case-exact dentry
2005 * already exists in in the dcache, use it and return it.
2007 * If no entry exists with the exact case name, allocate new dentry with
2008 * the exact case, and return the spliced entry.
2010 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2013 struct dentry
*found
, *res
;
2016 * First check if a dentry matching the name already exists,
2017 * if not go ahead and create it now.
2019 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2024 if (d_in_lookup(dentry
)) {
2025 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2027 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2032 found
= d_alloc(dentry
->d_parent
, name
);
2035 return ERR_PTR(-ENOMEM
);
2038 res
= d_splice_alias(inode
, found
);
2045 EXPORT_SYMBOL(d_add_ci
);
2048 static inline bool d_same_name(const struct dentry
*dentry
,
2049 const struct dentry
*parent
,
2050 const struct qstr
*name
)
2052 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2053 if (dentry
->d_name
.len
!= name
->len
)
2055 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2057 return parent
->d_op
->d_compare(dentry
,
2058 dentry
->d_name
.len
, dentry
->d_name
.name
,
2063 * __d_lookup_rcu - search for a dentry (racy, store-free)
2064 * @parent: parent dentry
2065 * @name: qstr of name we wish to find
2066 * @seqp: returns d_seq value at the point where the dentry was found
2067 * Returns: dentry, or NULL
2069 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2070 * resolution (store-free path walking) design described in
2071 * Documentation/filesystems/path-lookup.txt.
2073 * This is not to be used outside core vfs.
2075 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2076 * held, and rcu_read_lock held. The returned dentry must not be stored into
2077 * without taking d_lock and checking d_seq sequence count against @seq
2080 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2083 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2084 * the returned dentry, so long as its parent's seqlock is checked after the
2085 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2086 * is formed, giving integrity down the path walk.
2088 * NOTE! The caller *has* to check the resulting dentry against the sequence
2089 * number we've returned before using any of the resulting dentry state!
2091 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2092 const struct qstr
*name
,
2095 u64 hashlen
= name
->hash_len
;
2096 const unsigned char *str
= name
->name
;
2097 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2098 struct hlist_bl_node
*node
;
2099 struct dentry
*dentry
;
2102 * Note: There is significant duplication with __d_lookup_rcu which is
2103 * required to prevent single threaded performance regressions
2104 * especially on architectures where smp_rmb (in seqcounts) are costly.
2105 * Keep the two functions in sync.
2109 * The hash list is protected using RCU.
2111 * Carefully use d_seq when comparing a candidate dentry, to avoid
2112 * races with d_move().
2114 * It is possible that concurrent renames can mess up our list
2115 * walk here and result in missing our dentry, resulting in the
2116 * false-negative result. d_lookup() protects against concurrent
2117 * renames using rename_lock seqlock.
2119 * See Documentation/filesystems/path-lookup.txt for more details.
2121 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2126 * The dentry sequence count protects us from concurrent
2127 * renames, and thus protects parent and name fields.
2129 * The caller must perform a seqcount check in order
2130 * to do anything useful with the returned dentry.
2132 * NOTE! We do a "raw" seqcount_begin here. That means that
2133 * we don't wait for the sequence count to stabilize if it
2134 * is in the middle of a sequence change. If we do the slow
2135 * dentry compare, we will do seqretries until it is stable,
2136 * and if we end up with a successful lookup, we actually
2137 * want to exit RCU lookup anyway.
2139 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2140 * we are still guaranteed NUL-termination of ->d_name.name.
2142 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2143 if (dentry
->d_parent
!= parent
)
2145 if (d_unhashed(dentry
))
2148 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2151 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2153 tlen
= dentry
->d_name
.len
;
2154 tname
= dentry
->d_name
.name
;
2155 /* we want a consistent (name,len) pair */
2156 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2160 if (parent
->d_op
->d_compare(dentry
,
2161 tlen
, tname
, name
) != 0)
2164 if (dentry
->d_name
.hash_len
!= hashlen
)
2166 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2176 * d_lookup - search for a dentry
2177 * @parent: parent dentry
2178 * @name: qstr of name we wish to find
2179 * Returns: dentry, or NULL
2181 * d_lookup searches the children of the parent dentry for the name in
2182 * question. If the dentry is found its reference count is incremented and the
2183 * dentry is returned. The caller must use dput to free the entry when it has
2184 * finished using it. %NULL is returned if the dentry does not exist.
2186 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2188 struct dentry
*dentry
;
2192 seq
= read_seqbegin(&rename_lock
);
2193 dentry
= __d_lookup(parent
, name
);
2196 } while (read_seqretry(&rename_lock
, seq
));
2199 EXPORT_SYMBOL(d_lookup
);
2202 * __d_lookup - search for a dentry (racy)
2203 * @parent: parent dentry
2204 * @name: qstr of name we wish to find
2205 * Returns: dentry, or NULL
2207 * __d_lookup is like d_lookup, however it may (rarely) return a
2208 * false-negative result due to unrelated rename activity.
2210 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2211 * however it must be used carefully, eg. with a following d_lookup in
2212 * the case of failure.
2214 * __d_lookup callers must be commented.
2216 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2218 unsigned int hash
= name
->hash
;
2219 struct hlist_bl_head
*b
= d_hash(hash
);
2220 struct hlist_bl_node
*node
;
2221 struct dentry
*found
= NULL
;
2222 struct dentry
*dentry
;
2225 * Note: There is significant duplication with __d_lookup_rcu which is
2226 * required to prevent single threaded performance regressions
2227 * especially on architectures where smp_rmb (in seqcounts) are costly.
2228 * Keep the two functions in sync.
2232 * The hash list is protected using RCU.
2234 * Take d_lock when comparing a candidate dentry, to avoid races
2237 * It is possible that concurrent renames can mess up our list
2238 * walk here and result in missing our dentry, resulting in the
2239 * false-negative result. d_lookup() protects against concurrent
2240 * renames using rename_lock seqlock.
2242 * See Documentation/filesystems/path-lookup.txt for more details.
2246 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2248 if (dentry
->d_name
.hash
!= hash
)
2251 spin_lock(&dentry
->d_lock
);
2252 if (dentry
->d_parent
!= parent
)
2254 if (d_unhashed(dentry
))
2257 if (!d_same_name(dentry
, parent
, name
))
2260 dentry
->d_lockref
.count
++;
2262 spin_unlock(&dentry
->d_lock
);
2265 spin_unlock(&dentry
->d_lock
);
2273 * d_hash_and_lookup - hash the qstr then search for a dentry
2274 * @dir: Directory to search in
2275 * @name: qstr of name we wish to find
2277 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2279 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2282 * Check for a fs-specific hash function. Note that we must
2283 * calculate the standard hash first, as the d_op->d_hash()
2284 * routine may choose to leave the hash value unchanged.
2286 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2287 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2288 int err
= dir
->d_op
->d_hash(dir
, name
);
2289 if (unlikely(err
< 0))
2290 return ERR_PTR(err
);
2292 return d_lookup(dir
, name
);
2294 EXPORT_SYMBOL(d_hash_and_lookup
);
2297 * When a file is deleted, we have two options:
2298 * - turn this dentry into a negative dentry
2299 * - unhash this dentry and free it.
2301 * Usually, we want to just turn this into
2302 * a negative dentry, but if anybody else is
2303 * currently using the dentry or the inode
2304 * we can't do that and we fall back on removing
2305 * it from the hash queues and waiting for
2306 * it to be deleted later when it has no users
2310 * d_delete - delete a dentry
2311 * @dentry: The dentry to delete
2313 * Turn the dentry into a negative dentry if possible, otherwise
2314 * remove it from the hash queues so it can be deleted later
2317 void d_delete(struct dentry
* dentry
)
2319 struct inode
*inode
;
2322 * Are we the only user?
2325 spin_lock(&dentry
->d_lock
);
2326 inode
= dentry
->d_inode
;
2327 isdir
= S_ISDIR(inode
->i_mode
);
2328 if (dentry
->d_lockref
.count
== 1) {
2329 if (!spin_trylock(&inode
->i_lock
)) {
2330 spin_unlock(&dentry
->d_lock
);
2334 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2335 dentry_unlink_inode(dentry
);
2336 fsnotify_nameremove(dentry
, isdir
);
2340 if (!d_unhashed(dentry
))
2343 spin_unlock(&dentry
->d_lock
);
2345 fsnotify_nameremove(dentry
, isdir
);
2347 EXPORT_SYMBOL(d_delete
);
2349 static void __d_rehash(struct dentry
*entry
)
2351 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2352 BUG_ON(!d_unhashed(entry
));
2354 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2359 * d_rehash - add an entry back to the hash
2360 * @entry: dentry to add to the hash
2362 * Adds a dentry to the hash according to its name.
2365 void d_rehash(struct dentry
* entry
)
2367 spin_lock(&entry
->d_lock
);
2369 spin_unlock(&entry
->d_lock
);
2371 EXPORT_SYMBOL(d_rehash
);
2373 static inline unsigned start_dir_add(struct inode
*dir
)
2377 unsigned n
= dir
->i_dir_seq
;
2378 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2384 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2386 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2389 static void d_wait_lookup(struct dentry
*dentry
)
2391 if (d_in_lookup(dentry
)) {
2392 DECLARE_WAITQUEUE(wait
, current
);
2393 add_wait_queue(dentry
->d_wait
, &wait
);
2395 set_current_state(TASK_UNINTERRUPTIBLE
);
2396 spin_unlock(&dentry
->d_lock
);
2398 spin_lock(&dentry
->d_lock
);
2399 } while (d_in_lookup(dentry
));
2403 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2404 const struct qstr
*name
,
2405 wait_queue_head_t
*wq
)
2407 unsigned int hash
= name
->hash
;
2408 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2409 struct hlist_bl_node
*node
;
2410 struct dentry
*new = d_alloc(parent
, name
);
2411 struct dentry
*dentry
;
2412 unsigned seq
, r_seq
, d_seq
;
2415 return ERR_PTR(-ENOMEM
);
2419 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
) & ~1;
2420 r_seq
= read_seqbegin(&rename_lock
);
2421 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2422 if (unlikely(dentry
)) {
2423 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2427 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2436 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2441 if (unlikely(parent
->d_inode
->i_dir_seq
!= seq
)) {
2447 * No changes for the parent since the beginning of d_lookup().
2448 * Since all removals from the chain happen with hlist_bl_lock(),
2449 * any potential in-lookup matches are going to stay here until
2450 * we unlock the chain. All fields are stable in everything
2453 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2454 if (dentry
->d_name
.hash
!= hash
)
2456 if (dentry
->d_parent
!= parent
)
2458 if (!d_same_name(dentry
, parent
, name
))
2461 /* now we can try to grab a reference */
2462 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2469 * somebody is likely to be still doing lookup for it;
2470 * wait for them to finish
2472 spin_lock(&dentry
->d_lock
);
2473 d_wait_lookup(dentry
);
2475 * it's not in-lookup anymore; in principle we should repeat
2476 * everything from dcache lookup, but it's likely to be what
2477 * d_lookup() would've found anyway. If it is, just return it;
2478 * otherwise we really have to repeat the whole thing.
2480 if (unlikely(dentry
->d_name
.hash
!= hash
))
2482 if (unlikely(dentry
->d_parent
!= parent
))
2484 if (unlikely(d_unhashed(dentry
)))
2486 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2488 /* OK, it *is* a hashed match; return it */
2489 spin_unlock(&dentry
->d_lock
);
2494 /* we can't take ->d_lock here; it's OK, though. */
2495 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2497 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2501 spin_unlock(&dentry
->d_lock
);
2505 EXPORT_SYMBOL(d_alloc_parallel
);
2507 void __d_lookup_done(struct dentry
*dentry
)
2509 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2510 dentry
->d_name
.hash
);
2512 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2513 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2514 wake_up_all(dentry
->d_wait
);
2515 dentry
->d_wait
= NULL
;
2517 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2518 INIT_LIST_HEAD(&dentry
->d_lru
);
2520 EXPORT_SYMBOL(__d_lookup_done
);
2522 /* inode->i_lock held if inode is non-NULL */
2524 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2526 struct inode
*dir
= NULL
;
2528 spin_lock(&dentry
->d_lock
);
2529 if (unlikely(d_in_lookup(dentry
))) {
2530 dir
= dentry
->d_parent
->d_inode
;
2531 n
= start_dir_add(dir
);
2532 __d_lookup_done(dentry
);
2535 unsigned add_flags
= d_flags_for_inode(inode
);
2536 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2537 raw_write_seqcount_begin(&dentry
->d_seq
);
2538 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2539 raw_write_seqcount_end(&dentry
->d_seq
);
2540 fsnotify_update_flags(dentry
);
2544 end_dir_add(dir
, n
);
2545 spin_unlock(&dentry
->d_lock
);
2547 spin_unlock(&inode
->i_lock
);
2551 * d_add - add dentry to hash queues
2552 * @entry: dentry to add
2553 * @inode: The inode to attach to this dentry
2555 * This adds the entry to the hash queues and initializes @inode.
2556 * The entry was actually filled in earlier during d_alloc().
2559 void d_add(struct dentry
*entry
, struct inode
*inode
)
2562 security_d_instantiate(entry
, inode
);
2563 spin_lock(&inode
->i_lock
);
2565 __d_add(entry
, inode
);
2567 EXPORT_SYMBOL(d_add
);
2570 * d_exact_alias - find and hash an exact unhashed alias
2571 * @entry: dentry to add
2572 * @inode: The inode to go with this dentry
2574 * If an unhashed dentry with the same name/parent and desired
2575 * inode already exists, hash and return it. Otherwise, return
2578 * Parent directory should be locked.
2580 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2582 struct dentry
*alias
;
2583 unsigned int hash
= entry
->d_name
.hash
;
2585 spin_lock(&inode
->i_lock
);
2586 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2588 * Don't need alias->d_lock here, because aliases with
2589 * d_parent == entry->d_parent are not subject to name or
2590 * parent changes, because the parent inode i_mutex is held.
2592 if (alias
->d_name
.hash
!= hash
)
2594 if (alias
->d_parent
!= entry
->d_parent
)
2596 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2598 spin_lock(&alias
->d_lock
);
2599 if (!d_unhashed(alias
)) {
2600 spin_unlock(&alias
->d_lock
);
2603 __dget_dlock(alias
);
2605 spin_unlock(&alias
->d_lock
);
2607 spin_unlock(&inode
->i_lock
);
2610 spin_unlock(&inode
->i_lock
);
2613 EXPORT_SYMBOL(d_exact_alias
);
2616 * dentry_update_name_case - update case insensitive dentry with a new name
2617 * @dentry: dentry to be updated
2620 * Update a case insensitive dentry with new case of name.
2622 * dentry must have been returned by d_lookup with name @name. Old and new
2623 * name lengths must match (ie. no d_compare which allows mismatched name
2626 * Parent inode i_mutex must be held over d_lookup and into this call (to
2627 * keep renames and concurrent inserts, and readdir(2) away).
2629 void dentry_update_name_case(struct dentry
*dentry
, const struct qstr
*name
)
2631 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2632 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2634 spin_lock(&dentry
->d_lock
);
2635 write_seqcount_begin(&dentry
->d_seq
);
2636 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2637 write_seqcount_end(&dentry
->d_seq
);
2638 spin_unlock(&dentry
->d_lock
);
2640 EXPORT_SYMBOL(dentry_update_name_case
);
2642 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2644 if (unlikely(dname_external(target
))) {
2645 if (unlikely(dname_external(dentry
))) {
2647 * Both external: swap the pointers
2649 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2652 * dentry:internal, target:external. Steal target's
2653 * storage and make target internal.
2655 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2656 dentry
->d_name
.len
+ 1);
2657 dentry
->d_name
.name
= target
->d_name
.name
;
2658 target
->d_name
.name
= target
->d_iname
;
2661 if (unlikely(dname_external(dentry
))) {
2663 * dentry:external, target:internal. Give dentry's
2664 * storage to target and make dentry internal
2666 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2667 target
->d_name
.len
+ 1);
2668 target
->d_name
.name
= dentry
->d_name
.name
;
2669 dentry
->d_name
.name
= dentry
->d_iname
;
2672 * Both are internal.
2675 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2676 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2677 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2678 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2679 swap(((long *) &dentry
->d_iname
)[i
],
2680 ((long *) &target
->d_iname
)[i
]);
2684 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2687 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2689 struct external_name
*old_name
= NULL
;
2690 if (unlikely(dname_external(dentry
)))
2691 old_name
= external_name(dentry
);
2692 if (unlikely(dname_external(target
))) {
2693 atomic_inc(&external_name(target
)->u
.count
);
2694 dentry
->d_name
= target
->d_name
;
2696 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2697 target
->d_name
.len
+ 1);
2698 dentry
->d_name
.name
= dentry
->d_iname
;
2699 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2701 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2702 kfree_rcu(old_name
, u
.head
);
2705 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2708 * XXXX: do we really need to take target->d_lock?
2710 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2711 spin_lock(&target
->d_parent
->d_lock
);
2713 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2714 spin_lock(&dentry
->d_parent
->d_lock
);
2715 spin_lock_nested(&target
->d_parent
->d_lock
,
2716 DENTRY_D_LOCK_NESTED
);
2718 spin_lock(&target
->d_parent
->d_lock
);
2719 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2720 DENTRY_D_LOCK_NESTED
);
2723 if (target
< dentry
) {
2724 spin_lock_nested(&target
->d_lock
, 2);
2725 spin_lock_nested(&dentry
->d_lock
, 3);
2727 spin_lock_nested(&dentry
->d_lock
, 2);
2728 spin_lock_nested(&target
->d_lock
, 3);
2732 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2734 if (target
->d_parent
!= dentry
->d_parent
)
2735 spin_unlock(&dentry
->d_parent
->d_lock
);
2736 if (target
->d_parent
!= target
)
2737 spin_unlock(&target
->d_parent
->d_lock
);
2738 spin_unlock(&target
->d_lock
);
2739 spin_unlock(&dentry
->d_lock
);
2743 * When switching names, the actual string doesn't strictly have to
2744 * be preserved in the target - because we're dropping the target
2745 * anyway. As such, we can just do a simple memcpy() to copy over
2746 * the new name before we switch, unless we are going to rehash
2747 * it. Note that if we *do* unhash the target, we are not allowed
2748 * to rehash it without giving it a new name/hash key - whether
2749 * we swap or overwrite the names here, resulting name won't match
2750 * the reality in filesystem; it's only there for d_path() purposes.
2751 * Note that all of this is happening under rename_lock, so the
2752 * any hash lookup seeing it in the middle of manipulations will
2753 * be discarded anyway. So we do not care what happens to the hash
2757 * __d_move - move a dentry
2758 * @dentry: entry to move
2759 * @target: new dentry
2760 * @exchange: exchange the two dentries
2762 * Update the dcache to reflect the move of a file name. Negative
2763 * dcache entries should not be moved in this way. Caller must hold
2764 * rename_lock, the i_mutex of the source and target directories,
2765 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2767 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2770 struct inode
*dir
= NULL
;
2772 if (!dentry
->d_inode
)
2773 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2775 BUG_ON(d_ancestor(dentry
, target
));
2776 BUG_ON(d_ancestor(target
, dentry
));
2778 dentry_lock_for_move(dentry
, target
);
2779 if (unlikely(d_in_lookup(target
))) {
2780 dir
= target
->d_parent
->d_inode
;
2781 n
= start_dir_add(dir
);
2782 __d_lookup_done(target
);
2785 write_seqcount_begin(&dentry
->d_seq
);
2786 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2789 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2793 /* Switch the names.. */
2795 swap_names(dentry
, target
);
2797 copy_name(dentry
, target
);
2799 /* rehash in new place(s) */
2804 /* ... and switch them in the tree */
2805 if (IS_ROOT(dentry
)) {
2806 /* splicing a tree */
2807 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2808 dentry
->d_parent
= target
->d_parent
;
2809 target
->d_parent
= target
;
2810 list_del_init(&target
->d_child
);
2811 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2813 /* swapping two dentries */
2814 swap(dentry
->d_parent
, target
->d_parent
);
2815 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2816 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2818 fsnotify_update_flags(target
);
2819 fsnotify_update_flags(dentry
);
2822 write_seqcount_end(&target
->d_seq
);
2823 write_seqcount_end(&dentry
->d_seq
);
2826 end_dir_add(dir
, n
);
2827 dentry_unlock_for_move(dentry
, target
);
2831 * d_move - move a dentry
2832 * @dentry: entry to move
2833 * @target: new dentry
2835 * Update the dcache to reflect the move of a file name. Negative
2836 * dcache entries should not be moved in this way. See the locking
2837 * requirements for __d_move.
2839 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2841 write_seqlock(&rename_lock
);
2842 __d_move(dentry
, target
, false);
2843 write_sequnlock(&rename_lock
);
2845 EXPORT_SYMBOL(d_move
);
2848 * d_exchange - exchange two dentries
2849 * @dentry1: first dentry
2850 * @dentry2: second dentry
2852 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2854 write_seqlock(&rename_lock
);
2856 WARN_ON(!dentry1
->d_inode
);
2857 WARN_ON(!dentry2
->d_inode
);
2858 WARN_ON(IS_ROOT(dentry1
));
2859 WARN_ON(IS_ROOT(dentry2
));
2861 __d_move(dentry1
, dentry2
, true);
2863 write_sequnlock(&rename_lock
);
2867 * d_ancestor - search for an ancestor
2868 * @p1: ancestor dentry
2871 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2872 * an ancestor of p2, else NULL.
2874 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2878 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2879 if (p
->d_parent
== p1
)
2886 * This helper attempts to cope with remotely renamed directories
2888 * It assumes that the caller is already holding
2889 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2891 * Note: If ever the locking in lock_rename() changes, then please
2892 * remember to update this too...
2894 static int __d_unalias(struct inode
*inode
,
2895 struct dentry
*dentry
, struct dentry
*alias
)
2897 struct mutex
*m1
= NULL
;
2898 struct rw_semaphore
*m2
= NULL
;
2901 /* If alias and dentry share a parent, then no extra locks required */
2902 if (alias
->d_parent
== dentry
->d_parent
)
2905 /* See lock_rename() */
2906 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2908 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2909 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
2911 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
2913 __d_move(alias
, dentry
, false);
2924 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2925 * @inode: the inode which may have a disconnected dentry
2926 * @dentry: a negative dentry which we want to point to the inode.
2928 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2929 * place of the given dentry and return it, else simply d_add the inode
2930 * to the dentry and return NULL.
2932 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2933 * we should error out: directories can't have multiple aliases.
2935 * This is needed in the lookup routine of any filesystem that is exportable
2936 * (via knfsd) so that we can build dcache paths to directories effectively.
2938 * If a dentry was found and moved, then it is returned. Otherwise NULL
2939 * is returned. This matches the expected return value of ->lookup.
2941 * Cluster filesystems may call this function with a negative, hashed dentry.
2942 * In that case, we know that the inode will be a regular file, and also this
2943 * will only occur during atomic_open. So we need to check for the dentry
2944 * being already hashed only in the final case.
2946 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2949 return ERR_CAST(inode
);
2951 BUG_ON(!d_unhashed(dentry
));
2956 security_d_instantiate(dentry
, inode
);
2957 spin_lock(&inode
->i_lock
);
2958 if (S_ISDIR(inode
->i_mode
)) {
2959 struct dentry
*new = __d_find_any_alias(inode
);
2960 if (unlikely(new)) {
2961 /* The reference to new ensures it remains an alias */
2962 spin_unlock(&inode
->i_lock
);
2963 write_seqlock(&rename_lock
);
2964 if (unlikely(d_ancestor(new, dentry
))) {
2965 write_sequnlock(&rename_lock
);
2967 new = ERR_PTR(-ELOOP
);
2968 pr_warn_ratelimited(
2969 "VFS: Lookup of '%s' in %s %s"
2970 " would have caused loop\n",
2971 dentry
->d_name
.name
,
2972 inode
->i_sb
->s_type
->name
,
2974 } else if (!IS_ROOT(new)) {
2975 int err
= __d_unalias(inode
, dentry
, new);
2976 write_sequnlock(&rename_lock
);
2982 __d_move(new, dentry
, false);
2983 write_sequnlock(&rename_lock
);
2990 __d_add(dentry
, inode
);
2993 EXPORT_SYMBOL(d_splice_alias
);
2995 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2999 return -ENAMETOOLONG
;
3001 memcpy(*buffer
, str
, namelen
);
3006 * prepend_name - prepend a pathname in front of current buffer pointer
3007 * @buffer: buffer pointer
3008 * @buflen: allocated length of the buffer
3009 * @name: name string and length qstr structure
3011 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3012 * make sure that either the old or the new name pointer and length are
3013 * fetched. However, there may be mismatch between length and pointer.
3014 * The length cannot be trusted, we need to copy it byte-by-byte until
3015 * the length is reached or a null byte is found. It also prepends "/" at
3016 * the beginning of the name. The sequence number check at the caller will
3017 * retry it again when a d_move() does happen. So any garbage in the buffer
3018 * due to mismatched pointer and length will be discarded.
3020 * Data dependency barrier is needed to make sure that we see that terminating
3021 * NUL. Alpha strikes again, film at 11...
3023 static int prepend_name(char **buffer
, int *buflen
, const struct qstr
*name
)
3025 const char *dname
= ACCESS_ONCE(name
->name
);
3026 u32 dlen
= ACCESS_ONCE(name
->len
);
3029 smp_read_barrier_depends();
3031 *buflen
-= dlen
+ 1;
3033 return -ENAMETOOLONG
;
3034 p
= *buffer
-= dlen
+ 1;
3046 * prepend_path - Prepend path string to a buffer
3047 * @path: the dentry/vfsmount to report
3048 * @root: root vfsmnt/dentry
3049 * @buffer: pointer to the end of the buffer
3050 * @buflen: pointer to buffer length
3052 * The function will first try to write out the pathname without taking any
3053 * lock other than the RCU read lock to make sure that dentries won't go away.
3054 * It only checks the sequence number of the global rename_lock as any change
3055 * in the dentry's d_seq will be preceded by changes in the rename_lock
3056 * sequence number. If the sequence number had been changed, it will restart
3057 * the whole pathname back-tracing sequence again by taking the rename_lock.
3058 * In this case, there is no need to take the RCU read lock as the recursive
3059 * parent pointer references will keep the dentry chain alive as long as no
3060 * rename operation is performed.
3062 static int prepend_path(const struct path
*path
,
3063 const struct path
*root
,
3064 char **buffer
, int *buflen
)
3066 struct dentry
*dentry
;
3067 struct vfsmount
*vfsmnt
;
3070 unsigned seq
, m_seq
= 0;
3076 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3083 dentry
= path
->dentry
;
3085 mnt
= real_mount(vfsmnt
);
3086 read_seqbegin_or_lock(&rename_lock
, &seq
);
3087 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3088 struct dentry
* parent
;
3090 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3091 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
3093 if (dentry
!= vfsmnt
->mnt_root
) {
3100 if (mnt
!= parent
) {
3101 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
3107 error
= is_mounted(vfsmnt
) ? 1 : 2;
3110 parent
= dentry
->d_parent
;
3112 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3120 if (need_seqretry(&rename_lock
, seq
)) {
3124 done_seqretry(&rename_lock
, seq
);
3128 if (need_seqretry(&mount_lock
, m_seq
)) {
3132 done_seqretry(&mount_lock
, m_seq
);
3134 if (error
>= 0 && bptr
== *buffer
) {
3136 error
= -ENAMETOOLONG
;
3146 * __d_path - return the path of a dentry
3147 * @path: the dentry/vfsmount to report
3148 * @root: root vfsmnt/dentry
3149 * @buf: buffer to return value in
3150 * @buflen: buffer length
3152 * Convert a dentry into an ASCII path name.
3154 * Returns a pointer into the buffer or an error code if the
3155 * path was too long.
3157 * "buflen" should be positive.
3159 * If the path is not reachable from the supplied root, return %NULL.
3161 char *__d_path(const struct path
*path
,
3162 const struct path
*root
,
3163 char *buf
, int buflen
)
3165 char *res
= buf
+ buflen
;
3168 prepend(&res
, &buflen
, "\0", 1);
3169 error
= prepend_path(path
, root
, &res
, &buflen
);
3172 return ERR_PTR(error
);
3178 char *d_absolute_path(const struct path
*path
,
3179 char *buf
, int buflen
)
3181 struct path root
= {};
3182 char *res
= buf
+ buflen
;
3185 prepend(&res
, &buflen
, "\0", 1);
3186 error
= prepend_path(path
, &root
, &res
, &buflen
);
3191 return ERR_PTR(error
);
3196 * same as __d_path but appends "(deleted)" for unlinked files.
3198 static int path_with_deleted(const struct path
*path
,
3199 const struct path
*root
,
3200 char **buf
, int *buflen
)
3202 prepend(buf
, buflen
, "\0", 1);
3203 if (d_unlinked(path
->dentry
)) {
3204 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3209 return prepend_path(path
, root
, buf
, buflen
);
3212 static int prepend_unreachable(char **buffer
, int *buflen
)
3214 return prepend(buffer
, buflen
, "(unreachable)", 13);
3217 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3222 seq
= read_seqcount_begin(&fs
->seq
);
3224 } while (read_seqcount_retry(&fs
->seq
, seq
));
3228 * d_path - return the path of a dentry
3229 * @path: path to report
3230 * @buf: buffer to return value in
3231 * @buflen: buffer length
3233 * Convert a dentry into an ASCII path name. If the entry has been deleted
3234 * the string " (deleted)" is appended. Note that this is ambiguous.
3236 * Returns a pointer into the buffer or an error code if the path was
3237 * too long. Note: Callers should use the returned pointer, not the passed
3238 * in buffer, to use the name! The implementation often starts at an offset
3239 * into the buffer, and may leave 0 bytes at the start.
3241 * "buflen" should be positive.
3243 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3245 char *res
= buf
+ buflen
;
3250 * We have various synthetic filesystems that never get mounted. On
3251 * these filesystems dentries are never used for lookup purposes, and
3252 * thus don't need to be hashed. They also don't need a name until a
3253 * user wants to identify the object in /proc/pid/fd/. The little hack
3254 * below allows us to generate a name for these objects on demand:
3256 * Some pseudo inodes are mountable. When they are mounted
3257 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3258 * and instead have d_path return the mounted path.
3260 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3261 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3262 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3265 get_fs_root_rcu(current
->fs
, &root
);
3266 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3270 res
= ERR_PTR(error
);
3273 EXPORT_SYMBOL(d_path
);
3276 * Helper function for dentry_operations.d_dname() members
3278 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3279 const char *fmt
, ...)
3285 va_start(args
, fmt
);
3286 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3289 if (sz
> sizeof(temp
) || sz
> buflen
)
3290 return ERR_PTR(-ENAMETOOLONG
);
3292 buffer
+= buflen
- sz
;
3293 return memcpy(buffer
, temp
, sz
);
3296 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3298 char *end
= buffer
+ buflen
;
3299 /* these dentries are never renamed, so d_lock is not needed */
3300 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3301 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3302 prepend(&end
, &buflen
, "/", 1))
3303 end
= ERR_PTR(-ENAMETOOLONG
);
3306 EXPORT_SYMBOL(simple_dname
);
3309 * Write full pathname from the root of the filesystem into the buffer.
3311 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3313 struct dentry
*dentry
;
3326 prepend(&end
, &len
, "\0", 1);
3330 read_seqbegin_or_lock(&rename_lock
, &seq
);
3331 while (!IS_ROOT(dentry
)) {
3332 struct dentry
*parent
= dentry
->d_parent
;
3335 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3344 if (need_seqretry(&rename_lock
, seq
)) {
3348 done_seqretry(&rename_lock
, seq
);
3353 return ERR_PTR(-ENAMETOOLONG
);
3356 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3358 return __dentry_path(dentry
, buf
, buflen
);
3360 EXPORT_SYMBOL(dentry_path_raw
);
3362 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3367 if (d_unlinked(dentry
)) {
3369 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3373 retval
= __dentry_path(dentry
, buf
, buflen
);
3374 if (!IS_ERR(retval
) && p
)
3375 *p
= '/'; /* restore '/' overriden with '\0' */
3378 return ERR_PTR(-ENAMETOOLONG
);
3381 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3387 seq
= read_seqcount_begin(&fs
->seq
);
3390 } while (read_seqcount_retry(&fs
->seq
, seq
));
3394 * NOTE! The user-level library version returns a
3395 * character pointer. The kernel system call just
3396 * returns the length of the buffer filled (which
3397 * includes the ending '\0' character), or a negative
3398 * error value. So libc would do something like
3400 * char *getcwd(char * buf, size_t size)
3404 * retval = sys_getcwd(buf, size);
3411 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3414 struct path pwd
, root
;
3415 char *page
= __getname();
3421 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3424 if (!d_unlinked(pwd
.dentry
)) {
3426 char *cwd
= page
+ PATH_MAX
;
3427 int buflen
= PATH_MAX
;
3429 prepend(&cwd
, &buflen
, "\0", 1);
3430 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3436 /* Unreachable from current root */
3438 error
= prepend_unreachable(&cwd
, &buflen
);
3444 len
= PATH_MAX
+ page
- cwd
;
3447 if (copy_to_user(buf
, cwd
, len
))
3460 * Test whether new_dentry is a subdirectory of old_dentry.
3462 * Trivially implemented using the dcache structure
3466 * is_subdir - is new dentry a subdirectory of old_dentry
3467 * @new_dentry: new dentry
3468 * @old_dentry: old dentry
3470 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3471 * Returns false otherwise.
3472 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3475 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3480 if (new_dentry
== old_dentry
)
3484 /* for restarting inner loop in case of seq retry */
3485 seq
= read_seqbegin(&rename_lock
);
3487 * Need rcu_readlock to protect against the d_parent trashing
3491 if (d_ancestor(old_dentry
, new_dentry
))
3496 } while (read_seqretry(&rename_lock
, seq
));
3501 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3503 struct dentry
*root
= data
;
3504 if (dentry
!= root
) {
3505 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3508 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3509 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3510 dentry
->d_lockref
.count
--;
3513 return D_WALK_CONTINUE
;
3516 void d_genocide(struct dentry
*parent
)
3518 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3521 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3523 inode_dec_link_count(inode
);
3524 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3525 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3526 !d_unlinked(dentry
));
3527 spin_lock(&dentry
->d_parent
->d_lock
);
3528 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3529 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3530 (unsigned long long)inode
->i_ino
);
3531 spin_unlock(&dentry
->d_lock
);
3532 spin_unlock(&dentry
->d_parent
->d_lock
);
3533 d_instantiate(dentry
, inode
);
3535 EXPORT_SYMBOL(d_tmpfile
);
3537 static __initdata
unsigned long dhash_entries
;
3538 static int __init
set_dhash_entries(char *str
)
3542 dhash_entries
= simple_strtoul(str
, &str
, 0);
3545 __setup("dhash_entries=", set_dhash_entries
);
3547 static void __init
dcache_init_early(void)
3551 /* If hashes are distributed across NUMA nodes, defer
3552 * hash allocation until vmalloc space is available.
3558 alloc_large_system_hash("Dentry cache",
3559 sizeof(struct hlist_bl_head
),
3568 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3569 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3572 static void __init
dcache_init(void)
3577 * A constructor could be added for stable state like the lists,
3578 * but it is probably not worth it because of the cache nature
3581 dentry_cache
= KMEM_CACHE(dentry
,
3582 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
);
3584 /* Hash may have been set up in dcache_init_early */
3589 alloc_large_system_hash("Dentry cache",
3590 sizeof(struct hlist_bl_head
),
3599 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3600 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3603 /* SLAB cache for __getname() consumers */
3604 struct kmem_cache
*names_cachep __read_mostly
;
3605 EXPORT_SYMBOL(names_cachep
);
3607 EXPORT_SYMBOL(d_genocide
);
3609 void __init
vfs_caches_init_early(void)
3611 dcache_init_early();
3615 void __init
vfs_caches_init(void)
3617 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3618 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3623 files_maxfiles_init();