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
;
93 const struct qstr empty_name
= QSTR_INIT("", 0);
94 EXPORT_SYMBOL(empty_name
);
95 const struct qstr slash_name
= QSTR_INIT("/", 1);
96 EXPORT_SYMBOL(slash_name
);
99 * This is the single most critical data structure when it comes
100 * to the dcache: the hashtable for lookups. Somebody should try
101 * to make this good - I've just made it work.
103 * This hash-function tries to avoid losing too many bits of hash
104 * information, yet avoid using a prime hash-size or similar.
107 static unsigned int d_hash_mask __read_mostly
;
108 static unsigned int d_hash_shift __read_mostly
;
110 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
112 static inline struct hlist_bl_head
*d_hash(unsigned int hash
)
114 return dentry_hashtable
+ (hash
>> (32 - d_hash_shift
));
117 #define IN_LOOKUP_SHIFT 10
118 static struct hlist_bl_head in_lookup_hashtable
[1 << IN_LOOKUP_SHIFT
];
120 static inline struct hlist_bl_head
*in_lookup_hash(const struct dentry
*parent
,
123 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
124 return in_lookup_hashtable
+ hash_32(hash
, IN_LOOKUP_SHIFT
);
128 /* Statistics gathering. */
129 struct dentry_stat_t dentry_stat
= {
133 static DEFINE_PER_CPU(long, nr_dentry
);
134 static DEFINE_PER_CPU(long, nr_dentry_unused
);
136 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
139 * Here we resort to our own counters instead of using generic per-cpu counters
140 * for consistency with what the vfs inode code does. We are expected to harvest
141 * better code and performance by having our own specialized counters.
143 * Please note that the loop is done over all possible CPUs, not over all online
144 * CPUs. The reason for this is that we don't want to play games with CPUs going
145 * on and off. If one of them goes off, we will just keep their counters.
147 * glommer: See cffbc8a for details, and if you ever intend to change this,
148 * please update all vfs counters to match.
150 static long get_nr_dentry(void)
154 for_each_possible_cpu(i
)
155 sum
+= per_cpu(nr_dentry
, i
);
156 return sum
< 0 ? 0 : sum
;
159 static long get_nr_dentry_unused(void)
163 for_each_possible_cpu(i
)
164 sum
+= per_cpu(nr_dentry_unused
, i
);
165 return sum
< 0 ? 0 : sum
;
168 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
169 size_t *lenp
, loff_t
*ppos
)
171 dentry_stat
.nr_dentry
= get_nr_dentry();
172 dentry_stat
.nr_unused
= get_nr_dentry_unused();
173 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
178 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
179 * The strings are both count bytes long, and count is non-zero.
181 #ifdef CONFIG_DCACHE_WORD_ACCESS
183 #include <asm/word-at-a-time.h>
185 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
186 * aligned allocation for this particular component. We don't
187 * strictly need the load_unaligned_zeropad() safety, but it
188 * doesn't hurt either.
190 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
191 * need the careful unaligned handling.
193 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
195 unsigned long a
,b
,mask
;
198 a
= *(unsigned long *)cs
;
199 b
= load_unaligned_zeropad(ct
);
200 if (tcount
< sizeof(unsigned long))
202 if (unlikely(a
!= b
))
204 cs
+= sizeof(unsigned long);
205 ct
+= sizeof(unsigned long);
206 tcount
-= sizeof(unsigned long);
210 mask
= bytemask_from_count(tcount
);
211 return unlikely(!!((a
^ b
) & mask
));
216 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
230 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
233 * Be careful about RCU walk racing with rename:
234 * use 'READ_ONCE' to fetch the name pointer.
236 * NOTE! Even if a rename will mean that the length
237 * was not loaded atomically, we don't care. The
238 * RCU walk will check the sequence count eventually,
239 * and catch it. And we won't overrun the buffer,
240 * because we're reading the name pointer atomically,
241 * and a dentry name is guaranteed to be properly
242 * terminated with a NUL byte.
244 * End result: even if 'len' is wrong, we'll exit
245 * early because the data cannot match (there can
246 * be no NUL in the ct/tcount data)
248 const unsigned char *cs
= READ_ONCE(dentry
->d_name
.name
);
250 return dentry_string_cmp(cs
, ct
, tcount
);
253 struct external_name
{
256 struct rcu_head head
;
258 unsigned char name
[];
261 static inline struct external_name
*external_name(struct dentry
*dentry
)
263 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
266 static void __d_free(struct rcu_head
*head
)
268 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
270 kmem_cache_free(dentry_cache
, dentry
);
273 static void __d_free_external(struct rcu_head
*head
)
275 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
276 kfree(external_name(dentry
));
277 kmem_cache_free(dentry_cache
, dentry
);
280 static inline int dname_external(const struct dentry
*dentry
)
282 return dentry
->d_name
.name
!= dentry
->d_iname
;
285 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
287 spin_lock(&dentry
->d_lock
);
288 if (unlikely(dname_external(dentry
))) {
289 struct external_name
*p
= external_name(dentry
);
290 atomic_inc(&p
->u
.count
);
291 spin_unlock(&dentry
->d_lock
);
292 name
->name
= p
->name
;
294 memcpy(name
->inline_name
, dentry
->d_iname
, DNAME_INLINE_LEN
);
295 spin_unlock(&dentry
->d_lock
);
296 name
->name
= name
->inline_name
;
299 EXPORT_SYMBOL(take_dentry_name_snapshot
);
301 void release_dentry_name_snapshot(struct name_snapshot
*name
)
303 if (unlikely(name
->name
!= name
->inline_name
)) {
304 struct external_name
*p
;
305 p
= container_of(name
->name
, struct external_name
, name
[0]);
306 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
307 kfree_rcu(p
, u
.head
);
310 EXPORT_SYMBOL(release_dentry_name_snapshot
);
312 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
318 dentry
->d_inode
= inode
;
319 flags
= READ_ONCE(dentry
->d_flags
);
320 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
322 WRITE_ONCE(dentry
->d_flags
, flags
);
325 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
327 unsigned flags
= READ_ONCE(dentry
->d_flags
);
329 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
330 WRITE_ONCE(dentry
->d_flags
, flags
);
331 dentry
->d_inode
= NULL
;
334 static void dentry_free(struct dentry
*dentry
)
336 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
337 if (unlikely(dname_external(dentry
))) {
338 struct external_name
*p
= external_name(dentry
);
339 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
340 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
344 /* if dentry was never visible to RCU, immediate free is OK */
345 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
346 __d_free(&dentry
->d_u
.d_rcu
);
348 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
352 * Release the dentry's inode, using the filesystem
353 * d_iput() operation if defined.
355 static void dentry_unlink_inode(struct dentry
* dentry
)
356 __releases(dentry
->d_lock
)
357 __releases(dentry
->d_inode
->i_lock
)
359 struct inode
*inode
= dentry
->d_inode
;
360 bool hashed
= !d_unhashed(dentry
);
363 raw_write_seqcount_begin(&dentry
->d_seq
);
364 __d_clear_type_and_inode(dentry
);
365 hlist_del_init(&dentry
->d_u
.d_alias
);
367 raw_write_seqcount_end(&dentry
->d_seq
);
368 spin_unlock(&dentry
->d_lock
);
369 spin_unlock(&inode
->i_lock
);
371 fsnotify_inoderemove(inode
);
372 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
373 dentry
->d_op
->d_iput(dentry
, inode
);
379 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
380 * is in use - which includes both the "real" per-superblock
381 * LRU list _and_ the DCACHE_SHRINK_LIST use.
383 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
384 * on the shrink list (ie not on the superblock LRU list).
386 * The per-cpu "nr_dentry_unused" counters are updated with
387 * the DCACHE_LRU_LIST bit.
389 * These helper functions make sure we always follow the
390 * rules. d_lock must be held by the caller.
392 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
393 static void d_lru_add(struct dentry
*dentry
)
395 D_FLAG_VERIFY(dentry
, 0);
396 dentry
->d_flags
|= DCACHE_LRU_LIST
;
397 this_cpu_inc(nr_dentry_unused
);
398 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
401 static void d_lru_del(struct dentry
*dentry
)
403 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
404 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
405 this_cpu_dec(nr_dentry_unused
);
406 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
409 static void d_shrink_del(struct dentry
*dentry
)
411 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
412 list_del_init(&dentry
->d_lru
);
413 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
414 this_cpu_dec(nr_dentry_unused
);
417 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
419 D_FLAG_VERIFY(dentry
, 0);
420 list_add(&dentry
->d_lru
, list
);
421 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
422 this_cpu_inc(nr_dentry_unused
);
426 * These can only be called under the global LRU lock, ie during the
427 * callback for freeing the LRU list. "isolate" removes it from the
428 * LRU lists entirely, while shrink_move moves it to the indicated
431 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
433 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
434 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
435 this_cpu_dec(nr_dentry_unused
);
436 list_lru_isolate(lru
, &dentry
->d_lru
);
439 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
440 struct list_head
*list
)
442 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
443 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
444 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
448 * dentry_lru_(add|del)_list) must be called with d_lock held.
450 static void dentry_lru_add(struct dentry
*dentry
)
452 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
454 else if (unlikely(!(dentry
->d_flags
& DCACHE_REFERENCED
)))
455 dentry
->d_flags
|= DCACHE_REFERENCED
;
459 * d_drop - drop a dentry
460 * @dentry: dentry to drop
462 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
463 * be found through a VFS lookup any more. Note that this is different from
464 * deleting the dentry - d_delete will try to mark the dentry negative if
465 * possible, giving a successful _negative_ lookup, while d_drop will
466 * just make the cache lookup fail.
468 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
469 * reason (NFS timeouts or autofs deletes).
471 * __d_drop requires dentry->d_lock
472 * ___d_drop doesn't mark dentry as "unhashed"
473 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
475 static void ___d_drop(struct dentry
*dentry
)
477 if (!d_unhashed(dentry
)) {
478 struct hlist_bl_head
*b
;
480 * Hashed dentries are normally on the dentry hashtable,
481 * with the exception of those newly allocated by
482 * d_obtain_alias, which are always IS_ROOT:
484 if (unlikely(IS_ROOT(dentry
)))
485 b
= &dentry
->d_sb
->s_anon
;
487 b
= d_hash(dentry
->d_name
.hash
);
490 __hlist_bl_del(&dentry
->d_hash
);
492 /* After this call, in-progress rcu-walk path lookup will fail. */
493 write_seqcount_invalidate(&dentry
->d_seq
);
497 void __d_drop(struct dentry
*dentry
)
500 dentry
->d_hash
.pprev
= NULL
;
502 EXPORT_SYMBOL(__d_drop
);
504 void d_drop(struct dentry
*dentry
)
506 spin_lock(&dentry
->d_lock
);
508 spin_unlock(&dentry
->d_lock
);
510 EXPORT_SYMBOL(d_drop
);
512 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
516 * Inform d_walk() and shrink_dentry_list() that we are no longer
517 * attached to the dentry tree
519 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
520 if (unlikely(list_empty(&dentry
->d_child
)))
522 __list_del_entry(&dentry
->d_child
);
524 * Cursors can move around the list of children. While we'd been
525 * a normal list member, it didn't matter - ->d_child.next would've
526 * been updated. However, from now on it won't be and for the
527 * things like d_walk() it might end up with a nasty surprise.
528 * Normally d_walk() doesn't care about cursors moving around -
529 * ->d_lock on parent prevents that and since a cursor has no children
530 * of its own, we get through it without ever unlocking the parent.
531 * There is one exception, though - if we ascend from a child that
532 * gets killed as soon as we unlock it, the next sibling is found
533 * using the value left in its ->d_child.next. And if _that_
534 * pointed to a cursor, and cursor got moved (e.g. by lseek())
535 * before d_walk() regains parent->d_lock, we'll end up skipping
536 * everything the cursor had been moved past.
538 * Solution: make sure that the pointer left behind in ->d_child.next
539 * points to something that won't be moving around. I.e. skip the
542 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
543 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
544 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
546 dentry
->d_child
.next
= next
->d_child
.next
;
550 static void __dentry_kill(struct dentry
*dentry
)
552 struct dentry
*parent
= NULL
;
553 bool can_free
= true;
554 if (!IS_ROOT(dentry
))
555 parent
= dentry
->d_parent
;
558 * The dentry is now unrecoverably dead to the world.
560 lockref_mark_dead(&dentry
->d_lockref
);
563 * inform the fs via d_prune that this dentry is about to be
564 * unhashed and destroyed.
566 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
567 dentry
->d_op
->d_prune(dentry
);
569 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
570 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
573 /* if it was on the hash then remove it */
575 dentry_unlist(dentry
, parent
);
577 spin_unlock(&parent
->d_lock
);
579 dentry_unlink_inode(dentry
);
581 spin_unlock(&dentry
->d_lock
);
582 this_cpu_dec(nr_dentry
);
583 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
584 dentry
->d_op
->d_release(dentry
);
586 spin_lock(&dentry
->d_lock
);
587 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
588 dentry
->d_flags
|= DCACHE_MAY_FREE
;
591 spin_unlock(&dentry
->d_lock
);
592 if (likely(can_free
))
597 * Finish off a dentry we've decided to kill.
598 * dentry->d_lock must be held, returns with it unlocked.
599 * If ref is non-zero, then decrement the refcount too.
600 * Returns dentry requiring refcount drop, or NULL if we're done.
602 static struct dentry
*dentry_kill(struct dentry
*dentry
)
603 __releases(dentry
->d_lock
)
605 struct inode
*inode
= dentry
->d_inode
;
606 struct dentry
*parent
= NULL
;
608 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
611 if (!IS_ROOT(dentry
)) {
612 parent
= dentry
->d_parent
;
613 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
615 spin_unlock(&inode
->i_lock
);
620 __dentry_kill(dentry
);
624 spin_unlock(&dentry
->d_lock
);
625 return dentry
; /* try again with same dentry */
628 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
630 struct dentry
*parent
= dentry
->d_parent
;
633 if (unlikely(dentry
->d_lockref
.count
< 0))
635 if (likely(spin_trylock(&parent
->d_lock
)))
638 spin_unlock(&dentry
->d_lock
);
640 parent
= READ_ONCE(dentry
->d_parent
);
641 spin_lock(&parent
->d_lock
);
643 * We can't blindly lock dentry until we are sure
644 * that we won't violate the locking order.
645 * Any changes of dentry->d_parent must have
646 * been done with parent->d_lock held, so
647 * spin_lock() above is enough of a barrier
648 * for checking if it's still our child.
650 if (unlikely(parent
!= dentry
->d_parent
)) {
651 spin_unlock(&parent
->d_lock
);
654 if (parent
!= dentry
) {
655 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
656 if (unlikely(dentry
->d_lockref
.count
< 0)) {
657 spin_unlock(&parent
->d_lock
);
668 * Try to do a lockless dput(), and return whether that was successful.
670 * If unsuccessful, we return false, having already taken the dentry lock.
672 * The caller needs to hold the RCU read lock, so that the dentry is
673 * guaranteed to stay around even if the refcount goes down to zero!
675 static inline bool fast_dput(struct dentry
*dentry
)
678 unsigned int d_flags
;
681 * If we have a d_op->d_delete() operation, we sould not
682 * let the dentry count go to zero, so use "put_or_lock".
684 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
685 return lockref_put_or_lock(&dentry
->d_lockref
);
688 * .. otherwise, we can try to just decrement the
689 * lockref optimistically.
691 ret
= lockref_put_return(&dentry
->d_lockref
);
694 * If the lockref_put_return() failed due to the lock being held
695 * by somebody else, the fast path has failed. We will need to
696 * get the lock, and then check the count again.
698 if (unlikely(ret
< 0)) {
699 spin_lock(&dentry
->d_lock
);
700 if (dentry
->d_lockref
.count
> 1) {
701 dentry
->d_lockref
.count
--;
702 spin_unlock(&dentry
->d_lock
);
709 * If we weren't the last ref, we're done.
715 * Careful, careful. The reference count went down
716 * to zero, but we don't hold the dentry lock, so
717 * somebody else could get it again, and do another
718 * dput(), and we need to not race with that.
720 * However, there is a very special and common case
721 * where we don't care, because there is nothing to
722 * do: the dentry is still hashed, it does not have
723 * a 'delete' op, and it's referenced and already on
726 * NOTE! Since we aren't locked, these values are
727 * not "stable". However, it is sufficient that at
728 * some point after we dropped the reference the
729 * dentry was hashed and the flags had the proper
730 * value. Other dentry users may have re-gotten
731 * a reference to the dentry and change that, but
732 * our work is done - we can leave the dentry
733 * around with a zero refcount.
736 d_flags
= READ_ONCE(dentry
->d_flags
);
737 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
739 /* Nothing to do? Dropping the reference was all we needed? */
740 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
744 * Not the fast normal case? Get the lock. We've already decremented
745 * the refcount, but we'll need to re-check the situation after
748 spin_lock(&dentry
->d_lock
);
751 * Did somebody else grab a reference to it in the meantime, and
752 * we're no longer the last user after all? Alternatively, somebody
753 * else could have killed it and marked it dead. Either way, we
754 * don't need to do anything else.
756 if (dentry
->d_lockref
.count
) {
757 spin_unlock(&dentry
->d_lock
);
762 * Re-get the reference we optimistically dropped. We hold the
763 * lock, and we just tested that it was zero, so we can just
766 dentry
->d_lockref
.count
= 1;
774 * This is complicated by the fact that we do not want to put
775 * dentries that are no longer on any hash chain on the unused
776 * list: we'd much rather just get rid of them immediately.
778 * However, that implies that we have to traverse the dentry
779 * tree upwards to the parents which might _also_ now be
780 * scheduled for deletion (it may have been only waiting for
781 * its last child to go away).
783 * This tail recursion is done by hand as we don't want to depend
784 * on the compiler to always get this right (gcc generally doesn't).
785 * Real recursion would eat up our stack space.
789 * dput - release a dentry
790 * @dentry: dentry to release
792 * Release a dentry. This will drop the usage count and if appropriate
793 * call the dentry unlink method as well as removing it from the queues and
794 * releasing its resources. If the parent dentries were scheduled for release
795 * they too may now get deleted.
797 void dput(struct dentry
*dentry
)
799 if (unlikely(!dentry
))
806 if (likely(fast_dput(dentry
))) {
811 /* Slow case: now with the dentry lock held */
814 WARN_ON(d_in_lookup(dentry
));
816 /* Unreachable? Get rid of it */
817 if (unlikely(d_unhashed(dentry
)))
820 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
823 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
824 if (dentry
->d_op
->d_delete(dentry
))
828 dentry_lru_add(dentry
);
830 dentry
->d_lockref
.count
--;
831 spin_unlock(&dentry
->d_lock
);
835 dentry
= dentry_kill(dentry
);
844 /* This must be called with d_lock held */
845 static inline void __dget_dlock(struct dentry
*dentry
)
847 dentry
->d_lockref
.count
++;
850 static inline void __dget(struct dentry
*dentry
)
852 lockref_get(&dentry
->d_lockref
);
855 struct dentry
*dget_parent(struct dentry
*dentry
)
861 * Do optimistic parent lookup without any
865 ret
= READ_ONCE(dentry
->d_parent
);
866 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
868 if (likely(gotref
)) {
869 if (likely(ret
== READ_ONCE(dentry
->d_parent
)))
876 * Don't need rcu_dereference because we re-check it was correct under
880 ret
= dentry
->d_parent
;
881 spin_lock(&ret
->d_lock
);
882 if (unlikely(ret
!= dentry
->d_parent
)) {
883 spin_unlock(&ret
->d_lock
);
888 BUG_ON(!ret
->d_lockref
.count
);
889 ret
->d_lockref
.count
++;
890 spin_unlock(&ret
->d_lock
);
893 EXPORT_SYMBOL(dget_parent
);
896 * d_find_alias - grab a hashed alias of inode
897 * @inode: inode in question
899 * If inode has a hashed alias, or is a directory and has any alias,
900 * acquire the reference to alias and return it. Otherwise return NULL.
901 * Notice that if inode is a directory there can be only one alias and
902 * it can be unhashed only if it has no children, or if it is the root
903 * of a filesystem, or if the directory was renamed and d_revalidate
904 * was the first vfs operation to notice.
906 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
907 * any other hashed alias over that one.
909 static struct dentry
*__d_find_alias(struct inode
*inode
)
911 struct dentry
*alias
, *discon_alias
;
915 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
916 spin_lock(&alias
->d_lock
);
917 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
918 if (IS_ROOT(alias
) &&
919 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
920 discon_alias
= alias
;
923 spin_unlock(&alias
->d_lock
);
927 spin_unlock(&alias
->d_lock
);
930 alias
= discon_alias
;
931 spin_lock(&alias
->d_lock
);
932 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
934 spin_unlock(&alias
->d_lock
);
937 spin_unlock(&alias
->d_lock
);
943 struct dentry
*d_find_alias(struct inode
*inode
)
945 struct dentry
*de
= NULL
;
947 if (!hlist_empty(&inode
->i_dentry
)) {
948 spin_lock(&inode
->i_lock
);
949 de
= __d_find_alias(inode
);
950 spin_unlock(&inode
->i_lock
);
954 EXPORT_SYMBOL(d_find_alias
);
957 * Try to kill dentries associated with this inode.
958 * WARNING: you must own a reference to inode.
960 void d_prune_aliases(struct inode
*inode
)
962 struct dentry
*dentry
;
964 spin_lock(&inode
->i_lock
);
965 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
966 spin_lock(&dentry
->d_lock
);
967 if (!dentry
->d_lockref
.count
) {
968 struct dentry
*parent
= lock_parent(dentry
);
969 if (likely(!dentry
->d_lockref
.count
)) {
970 __dentry_kill(dentry
);
975 spin_unlock(&parent
->d_lock
);
977 spin_unlock(&dentry
->d_lock
);
979 spin_unlock(&inode
->i_lock
);
981 EXPORT_SYMBOL(d_prune_aliases
);
983 static void shrink_dentry_list(struct list_head
*list
)
985 struct dentry
*dentry
, *parent
;
987 while (!list_empty(list
)) {
989 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
990 spin_lock(&dentry
->d_lock
);
991 parent
= lock_parent(dentry
);
994 * The dispose list is isolated and dentries are not accounted
995 * to the LRU here, so we can simply remove it from the list
996 * here regardless of whether it is referenced or not.
998 d_shrink_del(dentry
);
1001 * We found an inuse dentry which was not removed from
1002 * the LRU because of laziness during lookup. Do not free it.
1004 if (dentry
->d_lockref
.count
> 0) {
1005 spin_unlock(&dentry
->d_lock
);
1007 spin_unlock(&parent
->d_lock
);
1012 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
1013 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
1014 spin_unlock(&dentry
->d_lock
);
1016 spin_unlock(&parent
->d_lock
);
1018 dentry_free(dentry
);
1022 inode
= dentry
->d_inode
;
1023 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1024 d_shrink_add(dentry
, list
);
1025 spin_unlock(&dentry
->d_lock
);
1027 spin_unlock(&parent
->d_lock
);
1031 __dentry_kill(dentry
);
1034 * We need to prune ancestors too. This is necessary to prevent
1035 * quadratic behavior of shrink_dcache_parent(), but is also
1036 * expected to be beneficial in reducing dentry cache
1040 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1041 parent
= lock_parent(dentry
);
1042 if (dentry
->d_lockref
.count
!= 1) {
1043 dentry
->d_lockref
.count
--;
1044 spin_unlock(&dentry
->d_lock
);
1046 spin_unlock(&parent
->d_lock
);
1049 inode
= dentry
->d_inode
; /* can't be NULL */
1050 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1051 spin_unlock(&dentry
->d_lock
);
1053 spin_unlock(&parent
->d_lock
);
1057 __dentry_kill(dentry
);
1063 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1064 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1066 struct list_head
*freeable
= arg
;
1067 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1071 * we are inverting the lru lock/dentry->d_lock here,
1072 * so use a trylock. If we fail to get the lock, just skip
1075 if (!spin_trylock(&dentry
->d_lock
))
1079 * Referenced dentries are still in use. If they have active
1080 * counts, just remove them from the LRU. Otherwise give them
1081 * another pass through the LRU.
1083 if (dentry
->d_lockref
.count
) {
1084 d_lru_isolate(lru
, dentry
);
1085 spin_unlock(&dentry
->d_lock
);
1089 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1090 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1091 spin_unlock(&dentry
->d_lock
);
1094 * The list move itself will be made by the common LRU code. At
1095 * this point, we've dropped the dentry->d_lock but keep the
1096 * lru lock. This is safe to do, since every list movement is
1097 * protected by the lru lock even if both locks are held.
1099 * This is guaranteed by the fact that all LRU management
1100 * functions are intermediated by the LRU API calls like
1101 * list_lru_add and list_lru_del. List movement in this file
1102 * only ever occur through this functions or through callbacks
1103 * like this one, that are called from the LRU API.
1105 * The only exceptions to this are functions like
1106 * shrink_dentry_list, and code that first checks for the
1107 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1108 * operating only with stack provided lists after they are
1109 * properly isolated from the main list. It is thus, always a
1115 d_lru_shrink_move(lru
, dentry
, freeable
);
1116 spin_unlock(&dentry
->d_lock
);
1122 * prune_dcache_sb - shrink the dcache
1124 * @sc: shrink control, passed to list_lru_shrink_walk()
1126 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1127 * is done when we need more memory and called from the superblock shrinker
1130 * This function may fail to free any resources if all the dentries are in
1133 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1138 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1139 dentry_lru_isolate
, &dispose
);
1140 shrink_dentry_list(&dispose
);
1144 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1145 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1147 struct list_head
*freeable
= arg
;
1148 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1151 * we are inverting the lru lock/dentry->d_lock here,
1152 * so use a trylock. If we fail to get the lock, just skip
1155 if (!spin_trylock(&dentry
->d_lock
))
1158 d_lru_shrink_move(lru
, dentry
, freeable
);
1159 spin_unlock(&dentry
->d_lock
);
1166 * shrink_dcache_sb - shrink dcache for a superblock
1169 * Shrink the dcache for the specified super block. This is used to free
1170 * the dcache before unmounting a file system.
1172 void shrink_dcache_sb(struct super_block
*sb
)
1179 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1180 dentry_lru_isolate_shrink
, &dispose
, 1024);
1182 this_cpu_sub(nr_dentry_unused
, freed
);
1183 shrink_dentry_list(&dispose
);
1185 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1187 EXPORT_SYMBOL(shrink_dcache_sb
);
1190 * enum d_walk_ret - action to talke during tree walk
1191 * @D_WALK_CONTINUE: contrinue walk
1192 * @D_WALK_QUIT: quit walk
1193 * @D_WALK_NORETRY: quit when retry is needed
1194 * @D_WALK_SKIP: skip this dentry and its children
1204 * d_walk - walk the dentry tree
1205 * @parent: start of walk
1206 * @data: data passed to @enter() and @finish()
1207 * @enter: callback when first entering the dentry
1208 * @finish: callback when successfully finished the walk
1210 * The @enter() and @finish() callbacks are called with d_lock held.
1212 void d_walk(struct dentry
*parent
, void *data
,
1213 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1214 void (*finish
)(void *))
1216 struct dentry
*this_parent
;
1217 struct list_head
*next
;
1219 enum d_walk_ret ret
;
1223 read_seqbegin_or_lock(&rename_lock
, &seq
);
1224 this_parent
= parent
;
1225 spin_lock(&this_parent
->d_lock
);
1227 ret
= enter(data
, this_parent
);
1229 case D_WALK_CONTINUE
:
1234 case D_WALK_NORETRY
:
1239 next
= this_parent
->d_subdirs
.next
;
1241 while (next
!= &this_parent
->d_subdirs
) {
1242 struct list_head
*tmp
= next
;
1243 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1246 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1249 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1251 ret
= enter(data
, dentry
);
1253 case D_WALK_CONTINUE
:
1256 spin_unlock(&dentry
->d_lock
);
1258 case D_WALK_NORETRY
:
1262 spin_unlock(&dentry
->d_lock
);
1266 if (!list_empty(&dentry
->d_subdirs
)) {
1267 spin_unlock(&this_parent
->d_lock
);
1268 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1269 this_parent
= dentry
;
1270 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1273 spin_unlock(&dentry
->d_lock
);
1276 * All done at this level ... ascend and resume the search.
1280 if (this_parent
!= parent
) {
1281 struct dentry
*child
= this_parent
;
1282 this_parent
= child
->d_parent
;
1284 spin_unlock(&child
->d_lock
);
1285 spin_lock(&this_parent
->d_lock
);
1287 /* might go back up the wrong parent if we have had a rename. */
1288 if (need_seqretry(&rename_lock
, seq
))
1290 /* go into the first sibling still alive */
1292 next
= child
->d_child
.next
;
1293 if (next
== &this_parent
->d_subdirs
)
1295 child
= list_entry(next
, struct dentry
, d_child
);
1296 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1300 if (need_seqretry(&rename_lock
, seq
))
1307 spin_unlock(&this_parent
->d_lock
);
1308 done_seqretry(&rename_lock
, seq
);
1312 spin_unlock(&this_parent
->d_lock
);
1320 EXPORT_SYMBOL_GPL(d_walk
);
1322 struct check_mount
{
1323 struct vfsmount
*mnt
;
1324 unsigned int mounted
;
1327 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1329 struct check_mount
*info
= data
;
1330 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1332 if (likely(!d_mountpoint(dentry
)))
1333 return D_WALK_CONTINUE
;
1334 if (__path_is_mountpoint(&path
)) {
1338 return D_WALK_CONTINUE
;
1342 * path_has_submounts - check for mounts over a dentry in the
1343 * current namespace.
1344 * @parent: path to check.
1346 * Return true if the parent or its subdirectories contain
1347 * a mount point in the current namespace.
1349 int path_has_submounts(const struct path
*parent
)
1351 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1353 read_seqlock_excl(&mount_lock
);
1354 d_walk(parent
->dentry
, &data
, path_check_mount
, NULL
);
1355 read_sequnlock_excl(&mount_lock
);
1357 return data
.mounted
;
1359 EXPORT_SYMBOL(path_has_submounts
);
1362 * Called by mount code to set a mountpoint and check if the mountpoint is
1363 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1364 * subtree can become unreachable).
1366 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1367 * this reason take rename_lock and d_lock on dentry and ancestors.
1369 int d_set_mounted(struct dentry
*dentry
)
1373 write_seqlock(&rename_lock
);
1374 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1375 /* Need exclusion wrt. d_invalidate() */
1376 spin_lock(&p
->d_lock
);
1377 if (unlikely(d_unhashed(p
))) {
1378 spin_unlock(&p
->d_lock
);
1381 spin_unlock(&p
->d_lock
);
1383 spin_lock(&dentry
->d_lock
);
1384 if (!d_unlinked(dentry
)) {
1386 if (!d_mountpoint(dentry
)) {
1387 dentry
->d_flags
|= DCACHE_MOUNTED
;
1391 spin_unlock(&dentry
->d_lock
);
1393 write_sequnlock(&rename_lock
);
1398 * Search the dentry child list of the specified parent,
1399 * and move any unused dentries to the end of the unused
1400 * list for prune_dcache(). We descend to the next level
1401 * whenever the d_subdirs list is non-empty and continue
1404 * It returns zero iff there are no unused children,
1405 * otherwise it returns the number of children moved to
1406 * the end of the unused list. This may not be the total
1407 * number of unused children, because select_parent can
1408 * drop the lock and return early due to latency
1412 struct select_data
{
1413 struct dentry
*start
;
1414 struct list_head dispose
;
1418 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1420 struct select_data
*data
= _data
;
1421 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1423 if (data
->start
== dentry
)
1426 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1429 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1431 if (!dentry
->d_lockref
.count
) {
1432 d_shrink_add(dentry
, &data
->dispose
);
1437 * We can return to the caller if we have found some (this
1438 * ensures forward progress). We'll be coming back to find
1441 if (!list_empty(&data
->dispose
))
1442 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1448 * shrink_dcache_parent - prune dcache
1449 * @parent: parent of entries to prune
1451 * Prune the dcache to remove unused children of the parent dentry.
1453 void shrink_dcache_parent(struct dentry
*parent
)
1456 struct select_data data
;
1458 INIT_LIST_HEAD(&data
.dispose
);
1459 data
.start
= parent
;
1462 d_walk(parent
, &data
, select_collect
, NULL
);
1466 shrink_dentry_list(&data
.dispose
);
1470 EXPORT_SYMBOL(shrink_dcache_parent
);
1472 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1474 /* it has busy descendents; complain about those instead */
1475 if (!list_empty(&dentry
->d_subdirs
))
1476 return D_WALK_CONTINUE
;
1478 /* root with refcount 1 is fine */
1479 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1480 return D_WALK_CONTINUE
;
1482 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1483 " still in use (%d) [unmount of %s %s]\n",
1486 dentry
->d_inode
->i_ino
: 0UL,
1488 dentry
->d_lockref
.count
,
1489 dentry
->d_sb
->s_type
->name
,
1490 dentry
->d_sb
->s_id
);
1492 return D_WALK_CONTINUE
;
1495 static void do_one_tree(struct dentry
*dentry
)
1497 shrink_dcache_parent(dentry
);
1498 d_walk(dentry
, dentry
, umount_check
, NULL
);
1504 * destroy the dentries attached to a superblock on unmounting
1506 void shrink_dcache_for_umount(struct super_block
*sb
)
1508 struct dentry
*dentry
;
1510 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1512 dentry
= sb
->s_root
;
1514 do_one_tree(dentry
);
1516 while (!hlist_bl_empty(&sb
->s_anon
)) {
1517 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1518 do_one_tree(dentry
);
1522 struct detach_data
{
1523 struct select_data select
;
1524 struct dentry
*mountpoint
;
1526 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1528 struct detach_data
*data
= _data
;
1530 if (d_mountpoint(dentry
)) {
1531 __dget_dlock(dentry
);
1532 data
->mountpoint
= dentry
;
1536 return select_collect(&data
->select
, dentry
);
1539 static void check_and_drop(void *_data
)
1541 struct detach_data
*data
= _data
;
1543 if (!data
->mountpoint
&& list_empty(&data
->select
.dispose
))
1544 __d_drop(data
->select
.start
);
1548 * d_invalidate - detach submounts, prune dcache, and drop
1549 * @dentry: dentry to invalidate (aka detach, prune and drop)
1553 * The final d_drop is done as an atomic operation relative to
1554 * rename_lock ensuring there are no races with d_set_mounted. This
1555 * ensures there are no unhashed dentries on the path to a mountpoint.
1557 void d_invalidate(struct dentry
*dentry
)
1560 * If it's already been dropped, return OK.
1562 spin_lock(&dentry
->d_lock
);
1563 if (d_unhashed(dentry
)) {
1564 spin_unlock(&dentry
->d_lock
);
1567 spin_unlock(&dentry
->d_lock
);
1569 /* Negative dentries can be dropped without further checks */
1570 if (!dentry
->d_inode
) {
1576 struct detach_data data
;
1578 data
.mountpoint
= NULL
;
1579 INIT_LIST_HEAD(&data
.select
.dispose
);
1580 data
.select
.start
= dentry
;
1581 data
.select
.found
= 0;
1583 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1585 if (!list_empty(&data
.select
.dispose
))
1586 shrink_dentry_list(&data
.select
.dispose
);
1587 else if (!data
.mountpoint
)
1590 if (data
.mountpoint
) {
1591 detach_mounts(data
.mountpoint
);
1592 dput(data
.mountpoint
);
1597 EXPORT_SYMBOL(d_invalidate
);
1600 * __d_alloc - allocate a dcache entry
1601 * @sb: filesystem it will belong to
1602 * @name: qstr of the name
1604 * Allocates a dentry. It returns %NULL if there is insufficient memory
1605 * available. On a success the dentry is returned. The name passed in is
1606 * copied and the copy passed in may be reused after this call.
1609 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1611 struct dentry
*dentry
;
1615 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1620 * We guarantee that the inline name is always NUL-terminated.
1621 * This way the memcpy() done by the name switching in rename
1622 * will still always have a NUL at the end, even if we might
1623 * be overwriting an internal NUL character
1625 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1626 if (unlikely(!name
)) {
1628 dname
= dentry
->d_iname
;
1629 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1630 size_t size
= offsetof(struct external_name
, name
[1]);
1631 struct external_name
*p
= kmalloc(size
+ name
->len
,
1632 GFP_KERNEL_ACCOUNT
);
1634 kmem_cache_free(dentry_cache
, dentry
);
1637 atomic_set(&p
->u
.count
, 1);
1639 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1640 kasan_unpoison_shadow(dname
,
1641 round_up(name
->len
+ 1, sizeof(unsigned long)));
1643 dname
= dentry
->d_iname
;
1646 dentry
->d_name
.len
= name
->len
;
1647 dentry
->d_name
.hash
= name
->hash
;
1648 memcpy(dname
, name
->name
, name
->len
);
1649 dname
[name
->len
] = 0;
1651 /* Make sure we always see the terminating NUL character */
1653 dentry
->d_name
.name
= dname
;
1655 dentry
->d_lockref
.count
= 1;
1656 dentry
->d_flags
= 0;
1657 spin_lock_init(&dentry
->d_lock
);
1658 seqcount_init(&dentry
->d_seq
);
1659 dentry
->d_inode
= NULL
;
1660 dentry
->d_parent
= dentry
;
1662 dentry
->d_op
= NULL
;
1663 dentry
->d_fsdata
= NULL
;
1664 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1665 INIT_LIST_HEAD(&dentry
->d_lru
);
1666 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1667 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1668 INIT_LIST_HEAD(&dentry
->d_child
);
1669 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1671 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1672 err
= dentry
->d_op
->d_init(dentry
);
1674 if (dname_external(dentry
))
1675 kfree(external_name(dentry
));
1676 kmem_cache_free(dentry_cache
, dentry
);
1681 this_cpu_inc(nr_dentry
);
1687 * d_alloc - allocate a dcache entry
1688 * @parent: parent of entry to allocate
1689 * @name: qstr of the name
1691 * Allocates a dentry. It returns %NULL if there is insufficient memory
1692 * available. On a success the dentry is returned. The name passed in is
1693 * copied and the copy passed in may be reused after this call.
1695 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1697 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1700 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1701 spin_lock(&parent
->d_lock
);
1703 * don't need child lock because it is not subject
1704 * to concurrency here
1706 __dget_dlock(parent
);
1707 dentry
->d_parent
= parent
;
1708 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1709 spin_unlock(&parent
->d_lock
);
1713 EXPORT_SYMBOL(d_alloc
);
1715 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1717 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1719 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1720 dentry
->d_parent
= dget(parent
);
1726 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1727 * @sb: the superblock
1728 * @name: qstr of the name
1730 * For a filesystem that just pins its dentries in memory and never
1731 * performs lookups at all, return an unhashed IS_ROOT dentry.
1733 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1735 return __d_alloc(sb
, name
);
1737 EXPORT_SYMBOL(d_alloc_pseudo
);
1739 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1744 q
.hash_len
= hashlen_string(parent
, name
);
1745 return d_alloc(parent
, &q
);
1747 EXPORT_SYMBOL(d_alloc_name
);
1749 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1751 WARN_ON_ONCE(dentry
->d_op
);
1752 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1754 DCACHE_OP_REVALIDATE
|
1755 DCACHE_OP_WEAK_REVALIDATE
|
1762 dentry
->d_flags
|= DCACHE_OP_HASH
;
1764 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1765 if (op
->d_revalidate
)
1766 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1767 if (op
->d_weak_revalidate
)
1768 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1770 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1772 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1774 dentry
->d_flags
|= DCACHE_OP_REAL
;
1777 EXPORT_SYMBOL(d_set_d_op
);
1781 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1782 * @dentry - The dentry to mark
1784 * Mark a dentry as falling through to the lower layer (as set with
1785 * d_pin_lower()). This flag may be recorded on the medium.
1787 void d_set_fallthru(struct dentry
*dentry
)
1789 spin_lock(&dentry
->d_lock
);
1790 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1791 spin_unlock(&dentry
->d_lock
);
1793 EXPORT_SYMBOL(d_set_fallthru
);
1795 static unsigned d_flags_for_inode(struct inode
*inode
)
1797 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1800 return DCACHE_MISS_TYPE
;
1802 if (S_ISDIR(inode
->i_mode
)) {
1803 add_flags
= DCACHE_DIRECTORY_TYPE
;
1804 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1805 if (unlikely(!inode
->i_op
->lookup
))
1806 add_flags
= DCACHE_AUTODIR_TYPE
;
1808 inode
->i_opflags
|= IOP_LOOKUP
;
1810 goto type_determined
;
1813 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1814 if (unlikely(inode
->i_op
->get_link
)) {
1815 add_flags
= DCACHE_SYMLINK_TYPE
;
1816 goto type_determined
;
1818 inode
->i_opflags
|= IOP_NOFOLLOW
;
1821 if (unlikely(!S_ISREG(inode
->i_mode
)))
1822 add_flags
= DCACHE_SPECIAL_TYPE
;
1825 if (unlikely(IS_AUTOMOUNT(inode
)))
1826 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1830 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1832 unsigned add_flags
= d_flags_for_inode(inode
);
1833 WARN_ON(d_in_lookup(dentry
));
1835 spin_lock(&dentry
->d_lock
);
1836 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1837 raw_write_seqcount_begin(&dentry
->d_seq
);
1838 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1839 raw_write_seqcount_end(&dentry
->d_seq
);
1840 fsnotify_update_flags(dentry
);
1841 spin_unlock(&dentry
->d_lock
);
1845 * d_instantiate - fill in inode information for a dentry
1846 * @entry: dentry to complete
1847 * @inode: inode to attach to this dentry
1849 * Fill in inode information in the entry.
1851 * This turns negative dentries into productive full members
1854 * NOTE! This assumes that the inode count has been incremented
1855 * (or otherwise set) by the caller to indicate that it is now
1856 * in use by the dcache.
1859 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1861 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1863 security_d_instantiate(entry
, inode
);
1864 spin_lock(&inode
->i_lock
);
1865 __d_instantiate(entry
, inode
);
1866 spin_unlock(&inode
->i_lock
);
1869 EXPORT_SYMBOL(d_instantiate
);
1872 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1873 * with lockdep-related part of unlock_new_inode() done before
1874 * anything else. Use that instead of open-coding d_instantiate()/
1875 * unlock_new_inode() combinations.
1877 void d_instantiate_new(struct dentry
*entry
, struct inode
*inode
)
1879 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1881 lockdep_annotate_inode_mutex_key(inode
);
1882 security_d_instantiate(entry
, inode
);
1883 spin_lock(&inode
->i_lock
);
1884 __d_instantiate(entry
, inode
);
1885 WARN_ON(!(inode
->i_state
& I_NEW
));
1886 inode
->i_state
&= ~I_NEW
;
1888 wake_up_bit(&inode
->i_state
, __I_NEW
);
1889 spin_unlock(&inode
->i_lock
);
1891 EXPORT_SYMBOL(d_instantiate_new
);
1894 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1895 * @entry: dentry to complete
1896 * @inode: inode to attach to this dentry
1898 * Fill in inode information in the entry. If a directory alias is found, then
1899 * return an error (and drop inode). Together with d_materialise_unique() this
1900 * guarantees that a directory inode may never have more than one alias.
1902 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1904 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1906 security_d_instantiate(entry
, inode
);
1907 spin_lock(&inode
->i_lock
);
1908 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1909 spin_unlock(&inode
->i_lock
);
1913 __d_instantiate(entry
, inode
);
1914 spin_unlock(&inode
->i_lock
);
1918 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1920 struct dentry
*d_make_root(struct inode
*root_inode
)
1922 struct dentry
*res
= NULL
;
1925 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1927 res
->d_flags
|= DCACHE_RCUACCESS
;
1928 d_instantiate(res
, root_inode
);
1935 EXPORT_SYMBOL(d_make_root
);
1937 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1939 struct dentry
*alias
;
1941 if (hlist_empty(&inode
->i_dentry
))
1943 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1949 * d_find_any_alias - find any alias for a given inode
1950 * @inode: inode to find an alias for
1952 * If any aliases exist for the given inode, take and return a
1953 * reference for one of them. If no aliases exist, return %NULL.
1955 struct dentry
*d_find_any_alias(struct inode
*inode
)
1959 spin_lock(&inode
->i_lock
);
1960 de
= __d_find_any_alias(inode
);
1961 spin_unlock(&inode
->i_lock
);
1964 EXPORT_SYMBOL(d_find_any_alias
);
1966 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1973 return ERR_PTR(-ESTALE
);
1975 return ERR_CAST(inode
);
1977 res
= d_find_any_alias(inode
);
1981 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1983 res
= ERR_PTR(-ENOMEM
);
1987 security_d_instantiate(tmp
, inode
);
1988 spin_lock(&inode
->i_lock
);
1989 res
= __d_find_any_alias(inode
);
1991 spin_unlock(&inode
->i_lock
);
1996 /* attach a disconnected dentry */
1997 add_flags
= d_flags_for_inode(inode
);
2000 add_flags
|= DCACHE_DISCONNECTED
;
2002 spin_lock(&tmp
->d_lock
);
2003 __d_set_inode_and_type(tmp
, inode
, add_flags
);
2004 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
2005 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
2006 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
2007 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
2008 spin_unlock(&tmp
->d_lock
);
2009 spin_unlock(&inode
->i_lock
);
2019 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2020 * @inode: inode to allocate the dentry for
2022 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2023 * similar open by handle operations. The returned dentry may be anonymous,
2024 * or may have a full name (if the inode was already in the cache).
2026 * When called on a directory inode, we must ensure that the inode only ever
2027 * has one dentry. If a dentry is found, that is returned instead of
2028 * allocating a new one.
2030 * On successful return, the reference to the inode has been transferred
2031 * to the dentry. In case of an error the reference on the inode is released.
2032 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2033 * be passed in and the error will be propagated to the return value,
2034 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2036 struct dentry
*d_obtain_alias(struct inode
*inode
)
2038 return __d_obtain_alias(inode
, 1);
2040 EXPORT_SYMBOL(d_obtain_alias
);
2043 * d_obtain_root - find or allocate a dentry for a given inode
2044 * @inode: inode to allocate the dentry for
2046 * Obtain an IS_ROOT dentry for the root of a filesystem.
2048 * We must ensure that directory inodes only ever have one dentry. If a
2049 * dentry is found, that is returned instead of allocating a new one.
2051 * On successful return, the reference to the inode has been transferred
2052 * to the dentry. In case of an error the reference on the inode is
2053 * released. A %NULL or IS_ERR inode may be passed in and will be the
2054 * error will be propagate to the return value, with a %NULL @inode
2055 * replaced by ERR_PTR(-ESTALE).
2057 struct dentry
*d_obtain_root(struct inode
*inode
)
2059 return __d_obtain_alias(inode
, 0);
2061 EXPORT_SYMBOL(d_obtain_root
);
2064 * d_add_ci - lookup or allocate new dentry with case-exact name
2065 * @inode: the inode case-insensitive lookup has found
2066 * @dentry: the negative dentry that was passed to the parent's lookup func
2067 * @name: the case-exact name to be associated with the returned dentry
2069 * This is to avoid filling the dcache with case-insensitive names to the
2070 * same inode, only the actual correct case is stored in the dcache for
2071 * case-insensitive filesystems.
2073 * For a case-insensitive lookup match and if the the case-exact dentry
2074 * already exists in in the dcache, use it and return it.
2076 * If no entry exists with the exact case name, allocate new dentry with
2077 * the exact case, and return the spliced entry.
2079 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2082 struct dentry
*found
, *res
;
2085 * First check if a dentry matching the name already exists,
2086 * if not go ahead and create it now.
2088 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2093 if (d_in_lookup(dentry
)) {
2094 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2096 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2101 found
= d_alloc(dentry
->d_parent
, name
);
2104 return ERR_PTR(-ENOMEM
);
2107 res
= d_splice_alias(inode
, found
);
2114 EXPORT_SYMBOL(d_add_ci
);
2117 static inline bool d_same_name(const struct dentry
*dentry
,
2118 const struct dentry
*parent
,
2119 const struct qstr
*name
)
2121 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2122 if (dentry
->d_name
.len
!= name
->len
)
2124 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2126 return parent
->d_op
->d_compare(dentry
,
2127 dentry
->d_name
.len
, dentry
->d_name
.name
,
2132 * __d_lookup_rcu - search for a dentry (racy, store-free)
2133 * @parent: parent dentry
2134 * @name: qstr of name we wish to find
2135 * @seqp: returns d_seq value at the point where the dentry was found
2136 * Returns: dentry, or NULL
2138 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2139 * resolution (store-free path walking) design described in
2140 * Documentation/filesystems/path-lookup.txt.
2142 * This is not to be used outside core vfs.
2144 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2145 * held, and rcu_read_lock held. The returned dentry must not be stored into
2146 * without taking d_lock and checking d_seq sequence count against @seq
2149 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2152 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2153 * the returned dentry, so long as its parent's seqlock is checked after the
2154 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2155 * is formed, giving integrity down the path walk.
2157 * NOTE! The caller *has* to check the resulting dentry against the sequence
2158 * number we've returned before using any of the resulting dentry state!
2160 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2161 const struct qstr
*name
,
2164 u64 hashlen
= name
->hash_len
;
2165 const unsigned char *str
= name
->name
;
2166 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2167 struct hlist_bl_node
*node
;
2168 struct dentry
*dentry
;
2171 * Note: There is significant duplication with __d_lookup_rcu which is
2172 * required to prevent single threaded performance regressions
2173 * especially on architectures where smp_rmb (in seqcounts) are costly.
2174 * Keep the two functions in sync.
2178 * The hash list is protected using RCU.
2180 * Carefully use d_seq when comparing a candidate dentry, to avoid
2181 * races with d_move().
2183 * It is possible that concurrent renames can mess up our list
2184 * walk here and result in missing our dentry, resulting in the
2185 * false-negative result. d_lookup() protects against concurrent
2186 * renames using rename_lock seqlock.
2188 * See Documentation/filesystems/path-lookup.txt for more details.
2190 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2195 * The dentry sequence count protects us from concurrent
2196 * renames, and thus protects parent and name fields.
2198 * The caller must perform a seqcount check in order
2199 * to do anything useful with the returned dentry.
2201 * NOTE! We do a "raw" seqcount_begin here. That means that
2202 * we don't wait for the sequence count to stabilize if it
2203 * is in the middle of a sequence change. If we do the slow
2204 * dentry compare, we will do seqretries until it is stable,
2205 * and if we end up with a successful lookup, we actually
2206 * want to exit RCU lookup anyway.
2208 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2209 * we are still guaranteed NUL-termination of ->d_name.name.
2211 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2212 if (dentry
->d_parent
!= parent
)
2214 if (d_unhashed(dentry
))
2217 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2220 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2222 tlen
= dentry
->d_name
.len
;
2223 tname
= dentry
->d_name
.name
;
2224 /* we want a consistent (name,len) pair */
2225 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2229 if (parent
->d_op
->d_compare(dentry
,
2230 tlen
, tname
, name
) != 0)
2233 if (dentry
->d_name
.hash_len
!= hashlen
)
2235 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2245 * d_lookup - search for a dentry
2246 * @parent: parent dentry
2247 * @name: qstr of name we wish to find
2248 * Returns: dentry, or NULL
2250 * d_lookup searches the children of the parent dentry for the name in
2251 * question. If the dentry is found its reference count is incremented and the
2252 * dentry is returned. The caller must use dput to free the entry when it has
2253 * finished using it. %NULL is returned if the dentry does not exist.
2255 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2257 struct dentry
*dentry
;
2261 seq
= read_seqbegin(&rename_lock
);
2262 dentry
= __d_lookup(parent
, name
);
2265 } while (read_seqretry(&rename_lock
, seq
));
2268 EXPORT_SYMBOL(d_lookup
);
2271 * __d_lookup - search for a dentry (racy)
2272 * @parent: parent dentry
2273 * @name: qstr of name we wish to find
2274 * Returns: dentry, or NULL
2276 * __d_lookup is like d_lookup, however it may (rarely) return a
2277 * false-negative result due to unrelated rename activity.
2279 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2280 * however it must be used carefully, eg. with a following d_lookup in
2281 * the case of failure.
2283 * __d_lookup callers must be commented.
2285 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2287 unsigned int hash
= name
->hash
;
2288 struct hlist_bl_head
*b
= d_hash(hash
);
2289 struct hlist_bl_node
*node
;
2290 struct dentry
*found
= NULL
;
2291 struct dentry
*dentry
;
2294 * Note: There is significant duplication with __d_lookup_rcu which is
2295 * required to prevent single threaded performance regressions
2296 * especially on architectures where smp_rmb (in seqcounts) are costly.
2297 * Keep the two functions in sync.
2301 * The hash list is protected using RCU.
2303 * Take d_lock when comparing a candidate dentry, to avoid races
2306 * It is possible that concurrent renames can mess up our list
2307 * walk here and result in missing our dentry, resulting in the
2308 * false-negative result. d_lookup() protects against concurrent
2309 * renames using rename_lock seqlock.
2311 * See Documentation/filesystems/path-lookup.txt for more details.
2315 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2317 if (dentry
->d_name
.hash
!= hash
)
2320 spin_lock(&dentry
->d_lock
);
2321 if (dentry
->d_parent
!= parent
)
2323 if (d_unhashed(dentry
))
2326 if (!d_same_name(dentry
, parent
, name
))
2329 dentry
->d_lockref
.count
++;
2331 spin_unlock(&dentry
->d_lock
);
2334 spin_unlock(&dentry
->d_lock
);
2342 * d_hash_and_lookup - hash the qstr then search for a dentry
2343 * @dir: Directory to search in
2344 * @name: qstr of name we wish to find
2346 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2348 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2351 * Check for a fs-specific hash function. Note that we must
2352 * calculate the standard hash first, as the d_op->d_hash()
2353 * routine may choose to leave the hash value unchanged.
2355 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2356 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2357 int err
= dir
->d_op
->d_hash(dir
, name
);
2358 if (unlikely(err
< 0))
2359 return ERR_PTR(err
);
2361 return d_lookup(dir
, name
);
2363 EXPORT_SYMBOL(d_hash_and_lookup
);
2366 * When a file is deleted, we have two options:
2367 * - turn this dentry into a negative dentry
2368 * - unhash this dentry and free it.
2370 * Usually, we want to just turn this into
2371 * a negative dentry, but if anybody else is
2372 * currently using the dentry or the inode
2373 * we can't do that and we fall back on removing
2374 * it from the hash queues and waiting for
2375 * it to be deleted later when it has no users
2379 * d_delete - delete a dentry
2380 * @dentry: The dentry to delete
2382 * Turn the dentry into a negative dentry if possible, otherwise
2383 * remove it from the hash queues so it can be deleted later
2386 void d_delete(struct dentry
* dentry
)
2388 struct inode
*inode
;
2391 * Are we the only user?
2394 spin_lock(&dentry
->d_lock
);
2395 inode
= dentry
->d_inode
;
2396 isdir
= S_ISDIR(inode
->i_mode
);
2397 if (dentry
->d_lockref
.count
== 1) {
2398 if (!spin_trylock(&inode
->i_lock
)) {
2399 spin_unlock(&dentry
->d_lock
);
2403 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2404 dentry_unlink_inode(dentry
);
2405 fsnotify_nameremove(dentry
, isdir
);
2409 if (!d_unhashed(dentry
))
2412 spin_unlock(&dentry
->d_lock
);
2414 fsnotify_nameremove(dentry
, isdir
);
2416 EXPORT_SYMBOL(d_delete
);
2418 static void __d_rehash(struct dentry
*entry
)
2420 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2423 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2428 * d_rehash - add an entry back to the hash
2429 * @entry: dentry to add to the hash
2431 * Adds a dentry to the hash according to its name.
2434 void d_rehash(struct dentry
* entry
)
2436 spin_lock(&entry
->d_lock
);
2438 spin_unlock(&entry
->d_lock
);
2440 EXPORT_SYMBOL(d_rehash
);
2442 static inline unsigned start_dir_add(struct inode
*dir
)
2446 unsigned n
= dir
->i_dir_seq
;
2447 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2453 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2455 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2458 static void d_wait_lookup(struct dentry
*dentry
)
2460 if (d_in_lookup(dentry
)) {
2461 DECLARE_WAITQUEUE(wait
, current
);
2462 add_wait_queue(dentry
->d_wait
, &wait
);
2464 set_current_state(TASK_UNINTERRUPTIBLE
);
2465 spin_unlock(&dentry
->d_lock
);
2467 spin_lock(&dentry
->d_lock
);
2468 } while (d_in_lookup(dentry
));
2472 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2473 const struct qstr
*name
,
2474 wait_queue_head_t
*wq
)
2476 unsigned int hash
= name
->hash
;
2477 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2478 struct hlist_bl_node
*node
;
2479 struct dentry
*new = d_alloc(parent
, name
);
2480 struct dentry
*dentry
;
2481 unsigned seq
, r_seq
, d_seq
;
2484 return ERR_PTR(-ENOMEM
);
2488 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
);
2489 r_seq
= read_seqbegin(&rename_lock
);
2490 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2491 if (unlikely(dentry
)) {
2492 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2496 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2505 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2510 if (unlikely(seq
& 1)) {
2516 if (unlikely(READ_ONCE(parent
->d_inode
->i_dir_seq
) != seq
)) {
2522 * No changes for the parent since the beginning of d_lookup().
2523 * Since all removals from the chain happen with hlist_bl_lock(),
2524 * any potential in-lookup matches are going to stay here until
2525 * we unlock the chain. All fields are stable in everything
2528 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2529 if (dentry
->d_name
.hash
!= hash
)
2531 if (dentry
->d_parent
!= parent
)
2533 if (!d_same_name(dentry
, parent
, name
))
2536 /* now we can try to grab a reference */
2537 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2544 * somebody is likely to be still doing lookup for it;
2545 * wait for them to finish
2547 spin_lock(&dentry
->d_lock
);
2548 d_wait_lookup(dentry
);
2550 * it's not in-lookup anymore; in principle we should repeat
2551 * everything from dcache lookup, but it's likely to be what
2552 * d_lookup() would've found anyway. If it is, just return it;
2553 * otherwise we really have to repeat the whole thing.
2555 if (unlikely(dentry
->d_name
.hash
!= hash
))
2557 if (unlikely(dentry
->d_parent
!= parent
))
2559 if (unlikely(d_unhashed(dentry
)))
2561 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2563 /* OK, it *is* a hashed match; return it */
2564 spin_unlock(&dentry
->d_lock
);
2569 /* we can't take ->d_lock here; it's OK, though. */
2570 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2572 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2576 spin_unlock(&dentry
->d_lock
);
2580 EXPORT_SYMBOL(d_alloc_parallel
);
2582 void __d_lookup_done(struct dentry
*dentry
)
2584 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2585 dentry
->d_name
.hash
);
2587 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2588 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2589 wake_up_all(dentry
->d_wait
);
2590 dentry
->d_wait
= NULL
;
2592 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2593 INIT_LIST_HEAD(&dentry
->d_lru
);
2595 EXPORT_SYMBOL(__d_lookup_done
);
2597 /* inode->i_lock held if inode is non-NULL */
2599 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2601 struct inode
*dir
= NULL
;
2603 spin_lock(&dentry
->d_lock
);
2604 if (unlikely(d_in_lookup(dentry
))) {
2605 dir
= dentry
->d_parent
->d_inode
;
2606 n
= start_dir_add(dir
);
2607 __d_lookup_done(dentry
);
2610 unsigned add_flags
= d_flags_for_inode(inode
);
2611 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2612 raw_write_seqcount_begin(&dentry
->d_seq
);
2613 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2614 raw_write_seqcount_end(&dentry
->d_seq
);
2615 fsnotify_update_flags(dentry
);
2619 end_dir_add(dir
, n
);
2620 spin_unlock(&dentry
->d_lock
);
2622 spin_unlock(&inode
->i_lock
);
2626 * d_add - add dentry to hash queues
2627 * @entry: dentry to add
2628 * @inode: The inode to attach to this dentry
2630 * This adds the entry to the hash queues and initializes @inode.
2631 * The entry was actually filled in earlier during d_alloc().
2634 void d_add(struct dentry
*entry
, struct inode
*inode
)
2637 security_d_instantiate(entry
, inode
);
2638 spin_lock(&inode
->i_lock
);
2640 __d_add(entry
, inode
);
2642 EXPORT_SYMBOL(d_add
);
2645 * d_exact_alias - find and hash an exact unhashed alias
2646 * @entry: dentry to add
2647 * @inode: The inode to go with this dentry
2649 * If an unhashed dentry with the same name/parent and desired
2650 * inode already exists, hash and return it. Otherwise, return
2653 * Parent directory should be locked.
2655 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2657 struct dentry
*alias
;
2658 unsigned int hash
= entry
->d_name
.hash
;
2660 spin_lock(&inode
->i_lock
);
2661 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2663 * Don't need alias->d_lock here, because aliases with
2664 * d_parent == entry->d_parent are not subject to name or
2665 * parent changes, because the parent inode i_mutex is held.
2667 if (alias
->d_name
.hash
!= hash
)
2669 if (alias
->d_parent
!= entry
->d_parent
)
2671 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2673 spin_lock(&alias
->d_lock
);
2674 if (!d_unhashed(alias
)) {
2675 spin_unlock(&alias
->d_lock
);
2678 __dget_dlock(alias
);
2680 spin_unlock(&alias
->d_lock
);
2682 spin_unlock(&inode
->i_lock
);
2685 spin_unlock(&inode
->i_lock
);
2688 EXPORT_SYMBOL(d_exact_alias
);
2691 * dentry_update_name_case - update case insensitive dentry with a new name
2692 * @dentry: dentry to be updated
2695 * Update a case insensitive dentry with new case of name.
2697 * dentry must have been returned by d_lookup with name @name. Old and new
2698 * name lengths must match (ie. no d_compare which allows mismatched name
2701 * Parent inode i_mutex must be held over d_lookup and into this call (to
2702 * keep renames and concurrent inserts, and readdir(2) away).
2704 void dentry_update_name_case(struct dentry
*dentry
, const struct qstr
*name
)
2706 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2707 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2709 spin_lock(&dentry
->d_lock
);
2710 write_seqcount_begin(&dentry
->d_seq
);
2711 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2712 write_seqcount_end(&dentry
->d_seq
);
2713 spin_unlock(&dentry
->d_lock
);
2715 EXPORT_SYMBOL(dentry_update_name_case
);
2717 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2719 if (unlikely(dname_external(target
))) {
2720 if (unlikely(dname_external(dentry
))) {
2722 * Both external: swap the pointers
2724 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2727 * dentry:internal, target:external. Steal target's
2728 * storage and make target internal.
2730 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2731 dentry
->d_name
.len
+ 1);
2732 dentry
->d_name
.name
= target
->d_name
.name
;
2733 target
->d_name
.name
= target
->d_iname
;
2736 if (unlikely(dname_external(dentry
))) {
2738 * dentry:external, target:internal. Give dentry's
2739 * storage to target and make dentry internal
2741 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2742 target
->d_name
.len
+ 1);
2743 target
->d_name
.name
= dentry
->d_name
.name
;
2744 dentry
->d_name
.name
= dentry
->d_iname
;
2747 * Both are internal.
2750 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2751 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2752 swap(((long *) &dentry
->d_iname
)[i
],
2753 ((long *) &target
->d_iname
)[i
]);
2757 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2760 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2762 struct external_name
*old_name
= NULL
;
2763 if (unlikely(dname_external(dentry
)))
2764 old_name
= external_name(dentry
);
2765 if (unlikely(dname_external(target
))) {
2766 atomic_inc(&external_name(target
)->u
.count
);
2767 dentry
->d_name
= target
->d_name
;
2769 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2770 target
->d_name
.len
+ 1);
2771 dentry
->d_name
.name
= dentry
->d_iname
;
2772 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2774 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2775 kfree_rcu(old_name
, u
.head
);
2778 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2781 * XXXX: do we really need to take target->d_lock?
2783 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2784 spin_lock(&target
->d_parent
->d_lock
);
2786 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2787 spin_lock(&dentry
->d_parent
->d_lock
);
2788 spin_lock_nested(&target
->d_parent
->d_lock
,
2789 DENTRY_D_LOCK_NESTED
);
2791 spin_lock(&target
->d_parent
->d_lock
);
2792 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2793 DENTRY_D_LOCK_NESTED
);
2796 if (target
< dentry
) {
2797 spin_lock_nested(&target
->d_lock
, 2);
2798 spin_lock_nested(&dentry
->d_lock
, 3);
2800 spin_lock_nested(&dentry
->d_lock
, 2);
2801 spin_lock_nested(&target
->d_lock
, 3);
2805 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2807 if (target
->d_parent
!= dentry
->d_parent
)
2808 spin_unlock(&dentry
->d_parent
->d_lock
);
2809 if (target
->d_parent
!= target
)
2810 spin_unlock(&target
->d_parent
->d_lock
);
2811 spin_unlock(&target
->d_lock
);
2812 spin_unlock(&dentry
->d_lock
);
2816 * When switching names, the actual string doesn't strictly have to
2817 * be preserved in the target - because we're dropping the target
2818 * anyway. As such, we can just do a simple memcpy() to copy over
2819 * the new name before we switch, unless we are going to rehash
2820 * it. Note that if we *do* unhash the target, we are not allowed
2821 * to rehash it without giving it a new name/hash key - whether
2822 * we swap or overwrite the names here, resulting name won't match
2823 * the reality in filesystem; it's only there for d_path() purposes.
2824 * Note that all of this is happening under rename_lock, so the
2825 * any hash lookup seeing it in the middle of manipulations will
2826 * be discarded anyway. So we do not care what happens to the hash
2830 * __d_move - move a dentry
2831 * @dentry: entry to move
2832 * @target: new dentry
2833 * @exchange: exchange the two dentries
2835 * Update the dcache to reflect the move of a file name. Negative
2836 * dcache entries should not be moved in this way. Caller must hold
2837 * rename_lock, the i_mutex of the source and target directories,
2838 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2840 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2843 struct inode
*dir
= NULL
;
2845 if (!dentry
->d_inode
)
2846 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2848 BUG_ON(d_ancestor(dentry
, target
));
2849 BUG_ON(d_ancestor(target
, dentry
));
2851 dentry_lock_for_move(dentry
, target
);
2852 if (unlikely(d_in_lookup(target
))) {
2853 dir
= target
->d_parent
->d_inode
;
2854 n
= start_dir_add(dir
);
2855 __d_lookup_done(target
);
2858 write_seqcount_begin(&dentry
->d_seq
);
2859 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2862 /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2866 /* Switch the names.. */
2868 swap_names(dentry
, target
);
2870 copy_name(dentry
, target
);
2872 /* rehash in new place(s) */
2877 target
->d_hash
.pprev
= NULL
;
2879 /* ... and switch them in the tree */
2880 if (IS_ROOT(dentry
)) {
2881 /* splicing a tree */
2882 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2883 dentry
->d_parent
= target
->d_parent
;
2884 target
->d_parent
= target
;
2885 list_del_init(&target
->d_child
);
2886 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2888 /* swapping two dentries */
2889 swap(dentry
->d_parent
, target
->d_parent
);
2890 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2891 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2893 fsnotify_update_flags(target
);
2894 fsnotify_update_flags(dentry
);
2897 write_seqcount_end(&target
->d_seq
);
2898 write_seqcount_end(&dentry
->d_seq
);
2901 end_dir_add(dir
, n
);
2902 dentry_unlock_for_move(dentry
, target
);
2906 * d_move - move a dentry
2907 * @dentry: entry to move
2908 * @target: new dentry
2910 * Update the dcache to reflect the move of a file name. Negative
2911 * dcache entries should not be moved in this way. See the locking
2912 * requirements for __d_move.
2914 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2916 write_seqlock(&rename_lock
);
2917 __d_move(dentry
, target
, false);
2918 write_sequnlock(&rename_lock
);
2920 EXPORT_SYMBOL(d_move
);
2923 * d_exchange - exchange two dentries
2924 * @dentry1: first dentry
2925 * @dentry2: second dentry
2927 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2929 write_seqlock(&rename_lock
);
2931 WARN_ON(!dentry1
->d_inode
);
2932 WARN_ON(!dentry2
->d_inode
);
2933 WARN_ON(IS_ROOT(dentry1
));
2934 WARN_ON(IS_ROOT(dentry2
));
2936 __d_move(dentry1
, dentry2
, true);
2938 write_sequnlock(&rename_lock
);
2940 EXPORT_SYMBOL_GPL(d_exchange
);
2943 * d_ancestor - search for an ancestor
2944 * @p1: ancestor dentry
2947 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2948 * an ancestor of p2, else NULL.
2950 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2954 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2955 if (p
->d_parent
== p1
)
2962 * This helper attempts to cope with remotely renamed directories
2964 * It assumes that the caller is already holding
2965 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2967 * Note: If ever the locking in lock_rename() changes, then please
2968 * remember to update this too...
2970 static int __d_unalias(struct inode
*inode
,
2971 struct dentry
*dentry
, struct dentry
*alias
)
2973 struct mutex
*m1
= NULL
;
2974 struct rw_semaphore
*m2
= NULL
;
2977 /* If alias and dentry share a parent, then no extra locks required */
2978 if (alias
->d_parent
== dentry
->d_parent
)
2981 /* See lock_rename() */
2982 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2984 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2985 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
2987 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
2989 __d_move(alias
, dentry
, false);
3000 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3001 * @inode: the inode which may have a disconnected dentry
3002 * @dentry: a negative dentry which we want to point to the inode.
3004 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3005 * place of the given dentry and return it, else simply d_add the inode
3006 * to the dentry and return NULL.
3008 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3009 * we should error out: directories can't have multiple aliases.
3011 * This is needed in the lookup routine of any filesystem that is exportable
3012 * (via knfsd) so that we can build dcache paths to directories effectively.
3014 * If a dentry was found and moved, then it is returned. Otherwise NULL
3015 * is returned. This matches the expected return value of ->lookup.
3017 * Cluster filesystems may call this function with a negative, hashed dentry.
3018 * In that case, we know that the inode will be a regular file, and also this
3019 * will only occur during atomic_open. So we need to check for the dentry
3020 * being already hashed only in the final case.
3022 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
3025 return ERR_CAST(inode
);
3027 BUG_ON(!d_unhashed(dentry
));
3032 security_d_instantiate(dentry
, inode
);
3033 spin_lock(&inode
->i_lock
);
3034 if (S_ISDIR(inode
->i_mode
)) {
3035 struct dentry
*new = __d_find_any_alias(inode
);
3036 if (unlikely(new)) {
3037 /* The reference to new ensures it remains an alias */
3038 spin_unlock(&inode
->i_lock
);
3039 write_seqlock(&rename_lock
);
3040 if (unlikely(d_ancestor(new, dentry
))) {
3041 write_sequnlock(&rename_lock
);
3043 new = ERR_PTR(-ELOOP
);
3044 pr_warn_ratelimited(
3045 "VFS: Lookup of '%s' in %s %s"
3046 " would have caused loop\n",
3047 dentry
->d_name
.name
,
3048 inode
->i_sb
->s_type
->name
,
3050 } else if (!IS_ROOT(new)) {
3051 int err
= __d_unalias(inode
, dentry
, new);
3052 write_sequnlock(&rename_lock
);
3058 __d_move(new, dentry
, false);
3059 write_sequnlock(&rename_lock
);
3066 __d_add(dentry
, inode
);
3069 EXPORT_SYMBOL(d_splice_alias
);
3071 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
3075 return -ENAMETOOLONG
;
3077 memcpy(*buffer
, str
, namelen
);
3082 * prepend_name - prepend a pathname in front of current buffer pointer
3083 * @buffer: buffer pointer
3084 * @buflen: allocated length of the buffer
3085 * @name: name string and length qstr structure
3087 * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
3088 * make sure that either the old or the new name pointer and length are
3089 * fetched. However, there may be mismatch between length and pointer.
3090 * The length cannot be trusted, we need to copy it byte-by-byte until
3091 * the length is reached or a null byte is found. It also prepends "/" at
3092 * the beginning of the name. The sequence number check at the caller will
3093 * retry it again when a d_move() does happen. So any garbage in the buffer
3094 * due to mismatched pointer and length will be discarded.
3096 * Data dependency barrier is needed to make sure that we see that terminating
3097 * NUL. Alpha strikes again, film at 11...
3099 static int prepend_name(char **buffer
, int *buflen
, const struct qstr
*name
)
3101 const char *dname
= READ_ONCE(name
->name
);
3102 u32 dlen
= READ_ONCE(name
->len
);
3105 smp_read_barrier_depends();
3107 *buflen
-= dlen
+ 1;
3109 return -ENAMETOOLONG
;
3110 p
= *buffer
-= dlen
+ 1;
3122 * prepend_path - Prepend path string to a buffer
3123 * @path: the dentry/vfsmount to report
3124 * @root: root vfsmnt/dentry
3125 * @buffer: pointer to the end of the buffer
3126 * @buflen: pointer to buffer length
3128 * The function will first try to write out the pathname without taking any
3129 * lock other than the RCU read lock to make sure that dentries won't go away.
3130 * It only checks the sequence number of the global rename_lock as any change
3131 * in the dentry's d_seq will be preceded by changes in the rename_lock
3132 * sequence number. If the sequence number had been changed, it will restart
3133 * the whole pathname back-tracing sequence again by taking the rename_lock.
3134 * In this case, there is no need to take the RCU read lock as the recursive
3135 * parent pointer references will keep the dentry chain alive as long as no
3136 * rename operation is performed.
3138 static int prepend_path(const struct path
*path
,
3139 const struct path
*root
,
3140 char **buffer
, int *buflen
)
3142 struct dentry
*dentry
;
3143 struct vfsmount
*vfsmnt
;
3146 unsigned seq
, m_seq
= 0;
3152 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3159 dentry
= path
->dentry
;
3161 mnt
= real_mount(vfsmnt
);
3162 read_seqbegin_or_lock(&rename_lock
, &seq
);
3163 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3164 struct dentry
* parent
;
3166 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3167 struct mount
*parent
= READ_ONCE(mnt
->mnt_parent
);
3169 if (dentry
!= vfsmnt
->mnt_root
) {
3176 if (mnt
!= parent
) {
3177 dentry
= READ_ONCE(mnt
->mnt_mountpoint
);
3183 error
= is_mounted(vfsmnt
) ? 1 : 2;
3186 parent
= dentry
->d_parent
;
3188 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3196 if (need_seqretry(&rename_lock
, seq
)) {
3200 done_seqretry(&rename_lock
, seq
);
3204 if (need_seqretry(&mount_lock
, m_seq
)) {
3208 done_seqretry(&mount_lock
, m_seq
);
3210 if (error
>= 0 && bptr
== *buffer
) {
3212 error
= -ENAMETOOLONG
;
3222 * __d_path - return the path of a dentry
3223 * @path: the dentry/vfsmount to report
3224 * @root: root vfsmnt/dentry
3225 * @buf: buffer to return value in
3226 * @buflen: buffer length
3228 * Convert a dentry into an ASCII path name.
3230 * Returns a pointer into the buffer or an error code if the
3231 * path was too long.
3233 * "buflen" should be positive.
3235 * If the path is not reachable from the supplied root, return %NULL.
3237 char *__d_path(const struct path
*path
,
3238 const struct path
*root
,
3239 char *buf
, int buflen
)
3241 char *res
= buf
+ buflen
;
3244 prepend(&res
, &buflen
, "\0", 1);
3245 error
= prepend_path(path
, root
, &res
, &buflen
);
3248 return ERR_PTR(error
);
3254 char *d_absolute_path(const struct path
*path
,
3255 char *buf
, int buflen
)
3257 struct path root
= {};
3258 char *res
= buf
+ buflen
;
3261 prepend(&res
, &buflen
, "\0", 1);
3262 error
= prepend_path(path
, &root
, &res
, &buflen
);
3267 return ERR_PTR(error
);
3272 * same as __d_path but appends "(deleted)" for unlinked files.
3274 static int path_with_deleted(const struct path
*path
,
3275 const struct path
*root
,
3276 char **buf
, int *buflen
)
3278 prepend(buf
, buflen
, "\0", 1);
3279 if (d_unlinked(path
->dentry
)) {
3280 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3285 return prepend_path(path
, root
, buf
, buflen
);
3288 static int prepend_unreachable(char **buffer
, int *buflen
)
3290 return prepend(buffer
, buflen
, "(unreachable)", 13);
3293 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3298 seq
= read_seqcount_begin(&fs
->seq
);
3300 } while (read_seqcount_retry(&fs
->seq
, seq
));
3304 * d_path - return the path of a dentry
3305 * @path: path to report
3306 * @buf: buffer to return value in
3307 * @buflen: buffer length
3309 * Convert a dentry into an ASCII path name. If the entry has been deleted
3310 * the string " (deleted)" is appended. Note that this is ambiguous.
3312 * Returns a pointer into the buffer or an error code if the path was
3313 * too long. Note: Callers should use the returned pointer, not the passed
3314 * in buffer, to use the name! The implementation often starts at an offset
3315 * into the buffer, and may leave 0 bytes at the start.
3317 * "buflen" should be positive.
3319 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3321 char *res
= buf
+ buflen
;
3326 * We have various synthetic filesystems that never get mounted. On
3327 * these filesystems dentries are never used for lookup purposes, and
3328 * thus don't need to be hashed. They also don't need a name until a
3329 * user wants to identify the object in /proc/pid/fd/. The little hack
3330 * below allows us to generate a name for these objects on demand:
3332 * Some pseudo inodes are mountable. When they are mounted
3333 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3334 * and instead have d_path return the mounted path.
3336 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3337 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3338 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3341 get_fs_root_rcu(current
->fs
, &root
);
3342 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3346 res
= ERR_PTR(error
);
3349 EXPORT_SYMBOL(d_path
);
3352 * Helper function for dentry_operations.d_dname() members
3354 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3355 const char *fmt
, ...)
3361 va_start(args
, fmt
);
3362 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3365 if (sz
> sizeof(temp
) || sz
> buflen
)
3366 return ERR_PTR(-ENAMETOOLONG
);
3368 buffer
+= buflen
- sz
;
3369 return memcpy(buffer
, temp
, sz
);
3372 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3374 char *end
= buffer
+ buflen
;
3375 /* these dentries are never renamed, so d_lock is not needed */
3376 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3377 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3378 prepend(&end
, &buflen
, "/", 1))
3379 end
= ERR_PTR(-ENAMETOOLONG
);
3382 EXPORT_SYMBOL(simple_dname
);
3385 * Write full pathname from the root of the filesystem into the buffer.
3387 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3389 struct dentry
*dentry
;
3402 prepend(&end
, &len
, "\0", 1);
3406 read_seqbegin_or_lock(&rename_lock
, &seq
);
3407 while (!IS_ROOT(dentry
)) {
3408 struct dentry
*parent
= dentry
->d_parent
;
3411 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3420 if (need_seqretry(&rename_lock
, seq
)) {
3424 done_seqretry(&rename_lock
, seq
);
3429 return ERR_PTR(-ENAMETOOLONG
);
3432 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3434 return __dentry_path(dentry
, buf
, buflen
);
3436 EXPORT_SYMBOL(dentry_path_raw
);
3438 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3443 if (d_unlinked(dentry
)) {
3445 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3449 retval
= __dentry_path(dentry
, buf
, buflen
);
3450 if (!IS_ERR(retval
) && p
)
3451 *p
= '/'; /* restore '/' overriden with '\0' */
3454 return ERR_PTR(-ENAMETOOLONG
);
3457 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3463 seq
= read_seqcount_begin(&fs
->seq
);
3466 } while (read_seqcount_retry(&fs
->seq
, seq
));
3470 * NOTE! The user-level library version returns a
3471 * character pointer. The kernel system call just
3472 * returns the length of the buffer filled (which
3473 * includes the ending '\0' character), or a negative
3474 * error value. So libc would do something like
3476 * char *getcwd(char * buf, size_t size)
3480 * retval = sys_getcwd(buf, size);
3487 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3490 struct path pwd
, root
;
3491 char *page
= __getname();
3497 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3500 if (!d_unlinked(pwd
.dentry
)) {
3502 char *cwd
= page
+ PATH_MAX
;
3503 int buflen
= PATH_MAX
;
3505 prepend(&cwd
, &buflen
, "\0", 1);
3506 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3512 /* Unreachable from current root */
3514 error
= prepend_unreachable(&cwd
, &buflen
);
3520 len
= PATH_MAX
+ page
- cwd
;
3523 if (copy_to_user(buf
, cwd
, len
))
3536 * Test whether new_dentry is a subdirectory of old_dentry.
3538 * Trivially implemented using the dcache structure
3542 * is_subdir - is new dentry a subdirectory of old_dentry
3543 * @new_dentry: new dentry
3544 * @old_dentry: old dentry
3546 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3547 * Returns false otherwise.
3548 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3551 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3556 if (new_dentry
== old_dentry
)
3560 /* for restarting inner loop in case of seq retry */
3561 seq
= read_seqbegin(&rename_lock
);
3563 * Need rcu_readlock to protect against the d_parent trashing
3567 if (d_ancestor(old_dentry
, new_dentry
))
3572 } while (read_seqretry(&rename_lock
, seq
));
3577 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3579 struct dentry
*root
= data
;
3580 if (dentry
!= root
) {
3581 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3584 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3585 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3586 dentry
->d_lockref
.count
--;
3589 return D_WALK_CONTINUE
;
3592 void d_genocide(struct dentry
*parent
)
3594 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3597 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3599 inode_dec_link_count(inode
);
3600 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3601 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3602 !d_unlinked(dentry
));
3603 spin_lock(&dentry
->d_parent
->d_lock
);
3604 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3605 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3606 (unsigned long long)inode
->i_ino
);
3607 spin_unlock(&dentry
->d_lock
);
3608 spin_unlock(&dentry
->d_parent
->d_lock
);
3609 d_instantiate(dentry
, inode
);
3611 EXPORT_SYMBOL(d_tmpfile
);
3613 static __initdata
unsigned long dhash_entries
;
3614 static int __init
set_dhash_entries(char *str
)
3618 dhash_entries
= simple_strtoul(str
, &str
, 0);
3621 __setup("dhash_entries=", set_dhash_entries
);
3623 static void __init
dcache_init_early(void)
3625 /* If hashes are distributed across NUMA nodes, defer
3626 * hash allocation until vmalloc space is available.
3632 alloc_large_system_hash("Dentry cache",
3633 sizeof(struct hlist_bl_head
),
3636 HASH_EARLY
| HASH_ZERO
,
3643 static void __init
dcache_init(void)
3646 * A constructor could be added for stable state like the lists,
3647 * but it is probably not worth it because of the cache nature
3650 dentry_cache
= KMEM_CACHE(dentry
,
3651 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
);
3653 /* Hash may have been set up in dcache_init_early */
3658 alloc_large_system_hash("Dentry cache",
3659 sizeof(struct hlist_bl_head
),
3669 /* SLAB cache for __getname() consumers */
3670 struct kmem_cache
*names_cachep __read_mostly
;
3671 EXPORT_SYMBOL(names_cachep
);
3673 EXPORT_SYMBOL(d_genocide
);
3675 void __init
vfs_caches_init_early(void)
3679 for (i
= 0; i
< ARRAY_SIZE(in_lookup_hashtable
); i
++)
3680 INIT_HLIST_BL_HEAD(&in_lookup_hashtable
[i
]);
3682 dcache_init_early();
3686 void __init
vfs_caches_init(void)
3688 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3689 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3694 files_maxfiles_init();