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 'lockless_dereference' 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
= lockless_dereference(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.
473 void __d_drop(struct dentry
*dentry
)
475 if (!d_unhashed(dentry
)) {
476 struct hlist_bl_head
*b
;
478 * Hashed dentries are normally on the dentry hashtable,
479 * with the exception of those newly allocated by
480 * d_obtain_alias, which are always IS_ROOT:
482 if (unlikely(IS_ROOT(dentry
)))
483 b
= &dentry
->d_sb
->s_anon
;
485 b
= d_hash(dentry
->d_name
.hash
);
488 __hlist_bl_del(&dentry
->d_hash
);
489 dentry
->d_hash
.pprev
= NULL
;
491 /* After this call, in-progress rcu-walk path lookup will fail. */
492 write_seqcount_invalidate(&dentry
->d_seq
);
495 EXPORT_SYMBOL(__d_drop
);
497 void d_drop(struct dentry
*dentry
)
499 spin_lock(&dentry
->d_lock
);
501 spin_unlock(&dentry
->d_lock
);
503 EXPORT_SYMBOL(d_drop
);
505 static inline void dentry_unlist(struct dentry
*dentry
, struct dentry
*parent
)
509 * Inform d_walk() and shrink_dentry_list() that we are no longer
510 * attached to the dentry tree
512 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
513 if (unlikely(list_empty(&dentry
->d_child
)))
515 __list_del_entry(&dentry
->d_child
);
517 * Cursors can move around the list of children. While we'd been
518 * a normal list member, it didn't matter - ->d_child.next would've
519 * been updated. However, from now on it won't be and for the
520 * things like d_walk() it might end up with a nasty surprise.
521 * Normally d_walk() doesn't care about cursors moving around -
522 * ->d_lock on parent prevents that and since a cursor has no children
523 * of its own, we get through it without ever unlocking the parent.
524 * There is one exception, though - if we ascend from a child that
525 * gets killed as soon as we unlock it, the next sibling is found
526 * using the value left in its ->d_child.next. And if _that_
527 * pointed to a cursor, and cursor got moved (e.g. by lseek())
528 * before d_walk() regains parent->d_lock, we'll end up skipping
529 * everything the cursor had been moved past.
531 * Solution: make sure that the pointer left behind in ->d_child.next
532 * points to something that won't be moving around. I.e. skip the
535 while (dentry
->d_child
.next
!= &parent
->d_subdirs
) {
536 next
= list_entry(dentry
->d_child
.next
, struct dentry
, d_child
);
537 if (likely(!(next
->d_flags
& DCACHE_DENTRY_CURSOR
)))
539 dentry
->d_child
.next
= next
->d_child
.next
;
543 static void __dentry_kill(struct dentry
*dentry
)
545 struct dentry
*parent
= NULL
;
546 bool can_free
= true;
547 if (!IS_ROOT(dentry
))
548 parent
= dentry
->d_parent
;
551 * The dentry is now unrecoverably dead to the world.
553 lockref_mark_dead(&dentry
->d_lockref
);
556 * inform the fs via d_prune that this dentry is about to be
557 * unhashed and destroyed.
559 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
560 dentry
->d_op
->d_prune(dentry
);
562 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
563 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
566 /* if it was on the hash then remove it */
568 dentry_unlist(dentry
, parent
);
570 spin_unlock(&parent
->d_lock
);
572 dentry_unlink_inode(dentry
);
574 spin_unlock(&dentry
->d_lock
);
575 this_cpu_dec(nr_dentry
);
576 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
577 dentry
->d_op
->d_release(dentry
);
579 spin_lock(&dentry
->d_lock
);
580 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
581 dentry
->d_flags
|= DCACHE_MAY_FREE
;
584 spin_unlock(&dentry
->d_lock
);
585 if (likely(can_free
))
590 * Finish off a dentry we've decided to kill.
591 * dentry->d_lock must be held, returns with it unlocked.
592 * If ref is non-zero, then decrement the refcount too.
593 * Returns dentry requiring refcount drop, or NULL if we're done.
595 static struct dentry
*dentry_kill(struct dentry
*dentry
)
596 __releases(dentry
->d_lock
)
598 struct inode
*inode
= dentry
->d_inode
;
599 struct dentry
*parent
= NULL
;
601 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
604 if (!IS_ROOT(dentry
)) {
605 parent
= dentry
->d_parent
;
606 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
608 spin_unlock(&inode
->i_lock
);
613 __dentry_kill(dentry
);
617 spin_unlock(&dentry
->d_lock
);
618 return dentry
; /* try again with same dentry */
621 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
623 struct dentry
*parent
= dentry
->d_parent
;
626 if (unlikely(dentry
->d_lockref
.count
< 0))
628 if (likely(spin_trylock(&parent
->d_lock
)))
631 spin_unlock(&dentry
->d_lock
);
633 parent
= ACCESS_ONCE(dentry
->d_parent
);
634 spin_lock(&parent
->d_lock
);
636 * We can't blindly lock dentry until we are sure
637 * that we won't violate the locking order.
638 * Any changes of dentry->d_parent must have
639 * been done with parent->d_lock held, so
640 * spin_lock() above is enough of a barrier
641 * for checking if it's still our child.
643 if (unlikely(parent
!= dentry
->d_parent
)) {
644 spin_unlock(&parent
->d_lock
);
648 if (parent
!= dentry
)
649 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
656 * Try to do a lockless dput(), and return whether that was successful.
658 * If unsuccessful, we return false, having already taken the dentry lock.
660 * The caller needs to hold the RCU read lock, so that the dentry is
661 * guaranteed to stay around even if the refcount goes down to zero!
663 static inline bool fast_dput(struct dentry
*dentry
)
666 unsigned int d_flags
;
669 * If we have a d_op->d_delete() operation, we sould not
670 * let the dentry count go to zero, so use "put_or_lock".
672 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
673 return lockref_put_or_lock(&dentry
->d_lockref
);
676 * .. otherwise, we can try to just decrement the
677 * lockref optimistically.
679 ret
= lockref_put_return(&dentry
->d_lockref
);
682 * If the lockref_put_return() failed due to the lock being held
683 * by somebody else, the fast path has failed. We will need to
684 * get the lock, and then check the count again.
686 if (unlikely(ret
< 0)) {
687 spin_lock(&dentry
->d_lock
);
688 if (dentry
->d_lockref
.count
> 1) {
689 dentry
->d_lockref
.count
--;
690 spin_unlock(&dentry
->d_lock
);
697 * If we weren't the last ref, we're done.
703 * Careful, careful. The reference count went down
704 * to zero, but we don't hold the dentry lock, so
705 * somebody else could get it again, and do another
706 * dput(), and we need to not race with that.
708 * However, there is a very special and common case
709 * where we don't care, because there is nothing to
710 * do: the dentry is still hashed, it does not have
711 * a 'delete' op, and it's referenced and already on
714 * NOTE! Since we aren't locked, these values are
715 * not "stable". However, it is sufficient that at
716 * some point after we dropped the reference the
717 * dentry was hashed and the flags had the proper
718 * value. Other dentry users may have re-gotten
719 * a reference to the dentry and change that, but
720 * our work is done - we can leave the dentry
721 * around with a zero refcount.
724 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
725 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
727 /* Nothing to do? Dropping the reference was all we needed? */
728 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
732 * Not the fast normal case? Get the lock. We've already decremented
733 * the refcount, but we'll need to re-check the situation after
736 spin_lock(&dentry
->d_lock
);
739 * Did somebody else grab a reference to it in the meantime, and
740 * we're no longer the last user after all? Alternatively, somebody
741 * else could have killed it and marked it dead. Either way, we
742 * don't need to do anything else.
744 if (dentry
->d_lockref
.count
) {
745 spin_unlock(&dentry
->d_lock
);
750 * Re-get the reference we optimistically dropped. We hold the
751 * lock, and we just tested that it was zero, so we can just
754 dentry
->d_lockref
.count
= 1;
762 * This is complicated by the fact that we do not want to put
763 * dentries that are no longer on any hash chain on the unused
764 * list: we'd much rather just get rid of them immediately.
766 * However, that implies that we have to traverse the dentry
767 * tree upwards to the parents which might _also_ now be
768 * scheduled for deletion (it may have been only waiting for
769 * its last child to go away).
771 * This tail recursion is done by hand as we don't want to depend
772 * on the compiler to always get this right (gcc generally doesn't).
773 * Real recursion would eat up our stack space.
777 * dput - release a dentry
778 * @dentry: dentry to release
780 * Release a dentry. This will drop the usage count and if appropriate
781 * call the dentry unlink method as well as removing it from the queues and
782 * releasing its resources. If the parent dentries were scheduled for release
783 * they too may now get deleted.
785 void dput(struct dentry
*dentry
)
787 if (unlikely(!dentry
))
794 if (likely(fast_dput(dentry
))) {
799 /* Slow case: now with the dentry lock held */
802 WARN_ON(d_in_lookup(dentry
));
804 /* Unreachable? Get rid of it */
805 if (unlikely(d_unhashed(dentry
)))
808 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
811 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
812 if (dentry
->d_op
->d_delete(dentry
))
816 dentry_lru_add(dentry
);
818 dentry
->d_lockref
.count
--;
819 spin_unlock(&dentry
->d_lock
);
823 dentry
= dentry_kill(dentry
);
832 /* This must be called with d_lock held */
833 static inline void __dget_dlock(struct dentry
*dentry
)
835 dentry
->d_lockref
.count
++;
838 static inline void __dget(struct dentry
*dentry
)
840 lockref_get(&dentry
->d_lockref
);
843 struct dentry
*dget_parent(struct dentry
*dentry
)
849 * Do optimistic parent lookup without any
853 ret
= ACCESS_ONCE(dentry
->d_parent
);
854 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
856 if (likely(gotref
)) {
857 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
864 * Don't need rcu_dereference because we re-check it was correct under
868 ret
= dentry
->d_parent
;
869 spin_lock(&ret
->d_lock
);
870 if (unlikely(ret
!= dentry
->d_parent
)) {
871 spin_unlock(&ret
->d_lock
);
876 BUG_ON(!ret
->d_lockref
.count
);
877 ret
->d_lockref
.count
++;
878 spin_unlock(&ret
->d_lock
);
881 EXPORT_SYMBOL(dget_parent
);
884 * d_find_alias - grab a hashed alias of inode
885 * @inode: inode in question
887 * If inode has a hashed alias, or is a directory and has any alias,
888 * acquire the reference to alias and return it. Otherwise return NULL.
889 * Notice that if inode is a directory there can be only one alias and
890 * it can be unhashed only if it has no children, or if it is the root
891 * of a filesystem, or if the directory was renamed and d_revalidate
892 * was the first vfs operation to notice.
894 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
895 * any other hashed alias over that one.
897 static struct dentry
*__d_find_alias(struct inode
*inode
)
899 struct dentry
*alias
, *discon_alias
;
903 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
904 spin_lock(&alias
->d_lock
);
905 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
906 if (IS_ROOT(alias
) &&
907 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
908 discon_alias
= alias
;
911 spin_unlock(&alias
->d_lock
);
915 spin_unlock(&alias
->d_lock
);
918 alias
= discon_alias
;
919 spin_lock(&alias
->d_lock
);
920 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
922 spin_unlock(&alias
->d_lock
);
925 spin_unlock(&alias
->d_lock
);
931 struct dentry
*d_find_alias(struct inode
*inode
)
933 struct dentry
*de
= NULL
;
935 if (!hlist_empty(&inode
->i_dentry
)) {
936 spin_lock(&inode
->i_lock
);
937 de
= __d_find_alias(inode
);
938 spin_unlock(&inode
->i_lock
);
942 EXPORT_SYMBOL(d_find_alias
);
945 * Try to kill dentries associated with this inode.
946 * WARNING: you must own a reference to inode.
948 void d_prune_aliases(struct inode
*inode
)
950 struct dentry
*dentry
;
952 spin_lock(&inode
->i_lock
);
953 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
954 spin_lock(&dentry
->d_lock
);
955 if (!dentry
->d_lockref
.count
) {
956 struct dentry
*parent
= lock_parent(dentry
);
957 if (likely(!dentry
->d_lockref
.count
)) {
958 __dentry_kill(dentry
);
963 spin_unlock(&parent
->d_lock
);
965 spin_unlock(&dentry
->d_lock
);
967 spin_unlock(&inode
->i_lock
);
969 EXPORT_SYMBOL(d_prune_aliases
);
971 static void shrink_dentry_list(struct list_head
*list
)
973 struct dentry
*dentry
, *parent
;
975 while (!list_empty(list
)) {
977 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
978 spin_lock(&dentry
->d_lock
);
979 parent
= lock_parent(dentry
);
982 * The dispose list is isolated and dentries are not accounted
983 * to the LRU here, so we can simply remove it from the list
984 * here regardless of whether it is referenced or not.
986 d_shrink_del(dentry
);
989 * We found an inuse dentry which was not removed from
990 * the LRU because of laziness during lookup. Do not free it.
992 if (dentry
->d_lockref
.count
> 0) {
993 spin_unlock(&dentry
->d_lock
);
995 spin_unlock(&parent
->d_lock
);
1000 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
1001 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
1002 spin_unlock(&dentry
->d_lock
);
1004 spin_unlock(&parent
->d_lock
);
1006 dentry_free(dentry
);
1010 inode
= dentry
->d_inode
;
1011 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1012 d_shrink_add(dentry
, list
);
1013 spin_unlock(&dentry
->d_lock
);
1015 spin_unlock(&parent
->d_lock
);
1019 __dentry_kill(dentry
);
1022 * We need to prune ancestors too. This is necessary to prevent
1023 * quadratic behavior of shrink_dcache_parent(), but is also
1024 * expected to be beneficial in reducing dentry cache
1028 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1029 parent
= lock_parent(dentry
);
1030 if (dentry
->d_lockref
.count
!= 1) {
1031 dentry
->d_lockref
.count
--;
1032 spin_unlock(&dentry
->d_lock
);
1034 spin_unlock(&parent
->d_lock
);
1037 inode
= dentry
->d_inode
; /* can't be NULL */
1038 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1039 spin_unlock(&dentry
->d_lock
);
1041 spin_unlock(&parent
->d_lock
);
1045 __dentry_kill(dentry
);
1051 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1052 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1054 struct list_head
*freeable
= arg
;
1055 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1059 * we are inverting the lru lock/dentry->d_lock here,
1060 * so use a trylock. If we fail to get the lock, just skip
1063 if (!spin_trylock(&dentry
->d_lock
))
1067 * Referenced dentries are still in use. If they have active
1068 * counts, just remove them from the LRU. Otherwise give them
1069 * another pass through the LRU.
1071 if (dentry
->d_lockref
.count
) {
1072 d_lru_isolate(lru
, dentry
);
1073 spin_unlock(&dentry
->d_lock
);
1077 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1078 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1079 spin_unlock(&dentry
->d_lock
);
1082 * The list move itself will be made by the common LRU code. At
1083 * this point, we've dropped the dentry->d_lock but keep the
1084 * lru lock. This is safe to do, since every list movement is
1085 * protected by the lru lock even if both locks are held.
1087 * This is guaranteed by the fact that all LRU management
1088 * functions are intermediated by the LRU API calls like
1089 * list_lru_add and list_lru_del. List movement in this file
1090 * only ever occur through this functions or through callbacks
1091 * like this one, that are called from the LRU API.
1093 * The only exceptions to this are functions like
1094 * shrink_dentry_list, and code that first checks for the
1095 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1096 * operating only with stack provided lists after they are
1097 * properly isolated from the main list. It is thus, always a
1103 d_lru_shrink_move(lru
, dentry
, freeable
);
1104 spin_unlock(&dentry
->d_lock
);
1110 * prune_dcache_sb - shrink the dcache
1112 * @sc: shrink control, passed to list_lru_shrink_walk()
1114 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1115 * is done when we need more memory and called from the superblock shrinker
1118 * This function may fail to free any resources if all the dentries are in
1121 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1126 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1127 dentry_lru_isolate
, &dispose
);
1128 shrink_dentry_list(&dispose
);
1132 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1133 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1135 struct list_head
*freeable
= arg
;
1136 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1139 * we are inverting the lru lock/dentry->d_lock here,
1140 * so use a trylock. If we fail to get the lock, just skip
1143 if (!spin_trylock(&dentry
->d_lock
))
1146 d_lru_shrink_move(lru
, dentry
, freeable
);
1147 spin_unlock(&dentry
->d_lock
);
1154 * shrink_dcache_sb - shrink dcache for a superblock
1157 * Shrink the dcache for the specified super block. This is used to free
1158 * the dcache before unmounting a file system.
1160 void shrink_dcache_sb(struct super_block
*sb
)
1167 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1168 dentry_lru_isolate_shrink
, &dispose
, 1024);
1170 this_cpu_sub(nr_dentry_unused
, freed
);
1171 shrink_dentry_list(&dispose
);
1173 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1175 EXPORT_SYMBOL(shrink_dcache_sb
);
1178 * enum d_walk_ret - action to talke during tree walk
1179 * @D_WALK_CONTINUE: contrinue walk
1180 * @D_WALK_QUIT: quit walk
1181 * @D_WALK_NORETRY: quit when retry is needed
1182 * @D_WALK_SKIP: skip this dentry and its children
1192 * d_walk - walk the dentry tree
1193 * @parent: start of walk
1194 * @data: data passed to @enter() and @finish()
1195 * @enter: callback when first entering the dentry
1196 * @finish: callback when successfully finished the walk
1198 * The @enter() and @finish() callbacks are called with d_lock held.
1200 void d_walk(struct dentry
*parent
, void *data
,
1201 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1202 void (*finish
)(void *))
1204 struct dentry
*this_parent
;
1205 struct list_head
*next
;
1207 enum d_walk_ret ret
;
1211 read_seqbegin_or_lock(&rename_lock
, &seq
);
1212 this_parent
= parent
;
1213 spin_lock(&this_parent
->d_lock
);
1215 ret
= enter(data
, this_parent
);
1217 case D_WALK_CONTINUE
:
1222 case D_WALK_NORETRY
:
1227 next
= this_parent
->d_subdirs
.next
;
1229 while (next
!= &this_parent
->d_subdirs
) {
1230 struct list_head
*tmp
= next
;
1231 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1234 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_CURSOR
))
1237 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1239 ret
= enter(data
, dentry
);
1241 case D_WALK_CONTINUE
:
1244 spin_unlock(&dentry
->d_lock
);
1246 case D_WALK_NORETRY
:
1250 spin_unlock(&dentry
->d_lock
);
1254 if (!list_empty(&dentry
->d_subdirs
)) {
1255 spin_unlock(&this_parent
->d_lock
);
1256 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1257 this_parent
= dentry
;
1258 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1261 spin_unlock(&dentry
->d_lock
);
1264 * All done at this level ... ascend and resume the search.
1268 if (this_parent
!= parent
) {
1269 struct dentry
*child
= this_parent
;
1270 this_parent
= child
->d_parent
;
1272 spin_unlock(&child
->d_lock
);
1273 spin_lock(&this_parent
->d_lock
);
1275 /* might go back up the wrong parent if we have had a rename. */
1276 if (need_seqretry(&rename_lock
, seq
))
1278 /* go into the first sibling still alive */
1280 next
= child
->d_child
.next
;
1281 if (next
== &this_parent
->d_subdirs
)
1283 child
= list_entry(next
, struct dentry
, d_child
);
1284 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1288 if (need_seqretry(&rename_lock
, seq
))
1295 spin_unlock(&this_parent
->d_lock
);
1296 done_seqretry(&rename_lock
, seq
);
1300 spin_unlock(&this_parent
->d_lock
);
1308 EXPORT_SYMBOL_GPL(d_walk
);
1310 struct check_mount
{
1311 struct vfsmount
*mnt
;
1312 unsigned int mounted
;
1315 static enum d_walk_ret
path_check_mount(void *data
, struct dentry
*dentry
)
1317 struct check_mount
*info
= data
;
1318 struct path path
= { .mnt
= info
->mnt
, .dentry
= dentry
};
1320 if (likely(!d_mountpoint(dentry
)))
1321 return D_WALK_CONTINUE
;
1322 if (__path_is_mountpoint(&path
)) {
1326 return D_WALK_CONTINUE
;
1330 * path_has_submounts - check for mounts over a dentry in the
1331 * current namespace.
1332 * @parent: path to check.
1334 * Return true if the parent or its subdirectories contain
1335 * a mount point in the current namespace.
1337 int path_has_submounts(const struct path
*parent
)
1339 struct check_mount data
= { .mnt
= parent
->mnt
, .mounted
= 0 };
1341 read_seqlock_excl(&mount_lock
);
1342 d_walk(parent
->dentry
, &data
, path_check_mount
, NULL
);
1343 read_sequnlock_excl(&mount_lock
);
1345 return data
.mounted
;
1347 EXPORT_SYMBOL(path_has_submounts
);
1350 * Called by mount code to set a mountpoint and check if the mountpoint is
1351 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1352 * subtree can become unreachable).
1354 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1355 * this reason take rename_lock and d_lock on dentry and ancestors.
1357 int d_set_mounted(struct dentry
*dentry
)
1361 write_seqlock(&rename_lock
);
1362 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1363 /* Need exclusion wrt. d_invalidate() */
1364 spin_lock(&p
->d_lock
);
1365 if (unlikely(d_unhashed(p
))) {
1366 spin_unlock(&p
->d_lock
);
1369 spin_unlock(&p
->d_lock
);
1371 spin_lock(&dentry
->d_lock
);
1372 if (!d_unlinked(dentry
)) {
1374 if (!d_mountpoint(dentry
)) {
1375 dentry
->d_flags
|= DCACHE_MOUNTED
;
1379 spin_unlock(&dentry
->d_lock
);
1381 write_sequnlock(&rename_lock
);
1386 * Search the dentry child list of the specified parent,
1387 * and move any unused dentries to the end of the unused
1388 * list for prune_dcache(). We descend to the next level
1389 * whenever the d_subdirs list is non-empty and continue
1392 * It returns zero iff there are no unused children,
1393 * otherwise it returns the number of children moved to
1394 * the end of the unused list. This may not be the total
1395 * number of unused children, because select_parent can
1396 * drop the lock and return early due to latency
1400 struct select_data
{
1401 struct dentry
*start
;
1402 struct list_head dispose
;
1406 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1408 struct select_data
*data
= _data
;
1409 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1411 if (data
->start
== dentry
)
1414 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1417 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1419 if (!dentry
->d_lockref
.count
) {
1420 d_shrink_add(dentry
, &data
->dispose
);
1425 * We can return to the caller if we have found some (this
1426 * ensures forward progress). We'll be coming back to find
1429 if (!list_empty(&data
->dispose
))
1430 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1436 * shrink_dcache_parent - prune dcache
1437 * @parent: parent of entries to prune
1439 * Prune the dcache to remove unused children of the parent dentry.
1441 void shrink_dcache_parent(struct dentry
*parent
)
1444 struct select_data data
;
1446 INIT_LIST_HEAD(&data
.dispose
);
1447 data
.start
= parent
;
1450 d_walk(parent
, &data
, select_collect
, NULL
);
1454 shrink_dentry_list(&data
.dispose
);
1458 EXPORT_SYMBOL(shrink_dcache_parent
);
1460 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1462 /* it has busy descendents; complain about those instead */
1463 if (!list_empty(&dentry
->d_subdirs
))
1464 return D_WALK_CONTINUE
;
1466 /* root with refcount 1 is fine */
1467 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1468 return D_WALK_CONTINUE
;
1470 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1471 " still in use (%d) [unmount of %s %s]\n",
1474 dentry
->d_inode
->i_ino
: 0UL,
1476 dentry
->d_lockref
.count
,
1477 dentry
->d_sb
->s_type
->name
,
1478 dentry
->d_sb
->s_id
);
1480 return D_WALK_CONTINUE
;
1483 static void do_one_tree(struct dentry
*dentry
)
1485 shrink_dcache_parent(dentry
);
1486 d_walk(dentry
, dentry
, umount_check
, NULL
);
1492 * destroy the dentries attached to a superblock on unmounting
1494 void shrink_dcache_for_umount(struct super_block
*sb
)
1496 struct dentry
*dentry
;
1498 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1500 dentry
= sb
->s_root
;
1502 do_one_tree(dentry
);
1504 while (!hlist_bl_empty(&sb
->s_anon
)) {
1505 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1506 do_one_tree(dentry
);
1510 struct detach_data
{
1511 struct select_data select
;
1512 struct dentry
*mountpoint
;
1514 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1516 struct detach_data
*data
= _data
;
1518 if (d_mountpoint(dentry
)) {
1519 __dget_dlock(dentry
);
1520 data
->mountpoint
= dentry
;
1524 return select_collect(&data
->select
, dentry
);
1527 static void check_and_drop(void *_data
)
1529 struct detach_data
*data
= _data
;
1531 if (!data
->mountpoint
&& list_empty(&data
->select
.dispose
))
1532 __d_drop(data
->select
.start
);
1536 * d_invalidate - detach submounts, prune dcache, and drop
1537 * @dentry: dentry to invalidate (aka detach, prune and drop)
1541 * The final d_drop is done as an atomic operation relative to
1542 * rename_lock ensuring there are no races with d_set_mounted. This
1543 * ensures there are no unhashed dentries on the path to a mountpoint.
1545 void d_invalidate(struct dentry
*dentry
)
1548 * If it's already been dropped, return OK.
1550 spin_lock(&dentry
->d_lock
);
1551 if (d_unhashed(dentry
)) {
1552 spin_unlock(&dentry
->d_lock
);
1555 spin_unlock(&dentry
->d_lock
);
1557 /* Negative dentries can be dropped without further checks */
1558 if (!dentry
->d_inode
) {
1564 struct detach_data data
;
1566 data
.mountpoint
= NULL
;
1567 INIT_LIST_HEAD(&data
.select
.dispose
);
1568 data
.select
.start
= dentry
;
1569 data
.select
.found
= 0;
1571 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1573 if (!list_empty(&data
.select
.dispose
))
1574 shrink_dentry_list(&data
.select
.dispose
);
1575 else if (!data
.mountpoint
)
1578 if (data
.mountpoint
) {
1579 detach_mounts(data
.mountpoint
);
1580 dput(data
.mountpoint
);
1585 EXPORT_SYMBOL(d_invalidate
);
1588 * __d_alloc - allocate a dcache entry
1589 * @sb: filesystem it will belong to
1590 * @name: qstr of the name
1592 * Allocates a dentry. It returns %NULL if there is insufficient memory
1593 * available. On a success the dentry is returned. The name passed in is
1594 * copied and the copy passed in may be reused after this call.
1597 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1599 struct dentry
*dentry
;
1603 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1608 * We guarantee that the inline name is always NUL-terminated.
1609 * This way the memcpy() done by the name switching in rename
1610 * will still always have a NUL at the end, even if we might
1611 * be overwriting an internal NUL character
1613 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1614 if (unlikely(!name
)) {
1616 dname
= dentry
->d_iname
;
1617 } else if (name
->len
> DNAME_INLINE_LEN
-1) {
1618 size_t size
= offsetof(struct external_name
, name
[1]);
1619 struct external_name
*p
= kmalloc(size
+ name
->len
,
1620 GFP_KERNEL_ACCOUNT
);
1622 kmem_cache_free(dentry_cache
, dentry
);
1625 atomic_set(&p
->u
.count
, 1);
1627 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1628 kasan_unpoison_shadow(dname
,
1629 round_up(name
->len
+ 1, sizeof(unsigned long)));
1631 dname
= dentry
->d_iname
;
1634 dentry
->d_name
.len
= name
->len
;
1635 dentry
->d_name
.hash
= name
->hash
;
1636 memcpy(dname
, name
->name
, name
->len
);
1637 dname
[name
->len
] = 0;
1639 /* Make sure we always see the terminating NUL character */
1641 dentry
->d_name
.name
= dname
;
1643 dentry
->d_lockref
.count
= 1;
1644 dentry
->d_flags
= 0;
1645 spin_lock_init(&dentry
->d_lock
);
1646 seqcount_init(&dentry
->d_seq
);
1647 dentry
->d_inode
= NULL
;
1648 dentry
->d_parent
= dentry
;
1650 dentry
->d_op
= NULL
;
1651 dentry
->d_fsdata
= NULL
;
1652 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1653 INIT_LIST_HEAD(&dentry
->d_lru
);
1654 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1655 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1656 INIT_LIST_HEAD(&dentry
->d_child
);
1657 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1659 if (dentry
->d_op
&& dentry
->d_op
->d_init
) {
1660 err
= dentry
->d_op
->d_init(dentry
);
1662 if (dname_external(dentry
))
1663 kfree(external_name(dentry
));
1664 kmem_cache_free(dentry_cache
, dentry
);
1669 this_cpu_inc(nr_dentry
);
1675 * d_alloc - allocate a dcache entry
1676 * @parent: parent of entry to allocate
1677 * @name: qstr of the name
1679 * Allocates a dentry. It returns %NULL if there is insufficient memory
1680 * available. On a success the dentry is returned. The name passed in is
1681 * copied and the copy passed in may be reused after this call.
1683 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1685 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1688 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1689 spin_lock(&parent
->d_lock
);
1691 * don't need child lock because it is not subject
1692 * to concurrency here
1694 __dget_dlock(parent
);
1695 dentry
->d_parent
= parent
;
1696 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1697 spin_unlock(&parent
->d_lock
);
1701 EXPORT_SYMBOL(d_alloc
);
1703 struct dentry
*d_alloc_cursor(struct dentry
* parent
)
1705 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, NULL
);
1707 dentry
->d_flags
|= DCACHE_RCUACCESS
| DCACHE_DENTRY_CURSOR
;
1708 dentry
->d_parent
= dget(parent
);
1714 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1715 * @sb: the superblock
1716 * @name: qstr of the name
1718 * For a filesystem that just pins its dentries in memory and never
1719 * performs lookups at all, return an unhashed IS_ROOT dentry.
1721 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1723 return __d_alloc(sb
, name
);
1725 EXPORT_SYMBOL(d_alloc_pseudo
);
1727 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1732 q
.hash_len
= hashlen_string(parent
, name
);
1733 return d_alloc(parent
, &q
);
1735 EXPORT_SYMBOL(d_alloc_name
);
1737 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1739 WARN_ON_ONCE(dentry
->d_op
);
1740 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1742 DCACHE_OP_REVALIDATE
|
1743 DCACHE_OP_WEAK_REVALIDATE
|
1750 dentry
->d_flags
|= DCACHE_OP_HASH
;
1752 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1753 if (op
->d_revalidate
)
1754 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1755 if (op
->d_weak_revalidate
)
1756 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1758 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1760 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1762 dentry
->d_flags
|= DCACHE_OP_REAL
;
1765 EXPORT_SYMBOL(d_set_d_op
);
1769 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1770 * @dentry - The dentry to mark
1772 * Mark a dentry as falling through to the lower layer (as set with
1773 * d_pin_lower()). This flag may be recorded on the medium.
1775 void d_set_fallthru(struct dentry
*dentry
)
1777 spin_lock(&dentry
->d_lock
);
1778 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1779 spin_unlock(&dentry
->d_lock
);
1781 EXPORT_SYMBOL(d_set_fallthru
);
1783 static unsigned d_flags_for_inode(struct inode
*inode
)
1785 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1788 return DCACHE_MISS_TYPE
;
1790 if (S_ISDIR(inode
->i_mode
)) {
1791 add_flags
= DCACHE_DIRECTORY_TYPE
;
1792 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1793 if (unlikely(!inode
->i_op
->lookup
))
1794 add_flags
= DCACHE_AUTODIR_TYPE
;
1796 inode
->i_opflags
|= IOP_LOOKUP
;
1798 goto type_determined
;
1801 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1802 if (unlikely(inode
->i_op
->get_link
)) {
1803 add_flags
= DCACHE_SYMLINK_TYPE
;
1804 goto type_determined
;
1806 inode
->i_opflags
|= IOP_NOFOLLOW
;
1809 if (unlikely(!S_ISREG(inode
->i_mode
)))
1810 add_flags
= DCACHE_SPECIAL_TYPE
;
1813 if (unlikely(IS_AUTOMOUNT(inode
)))
1814 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1818 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1820 unsigned add_flags
= d_flags_for_inode(inode
);
1821 WARN_ON(d_in_lookup(dentry
));
1823 spin_lock(&dentry
->d_lock
);
1824 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1825 raw_write_seqcount_begin(&dentry
->d_seq
);
1826 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1827 raw_write_seqcount_end(&dentry
->d_seq
);
1828 fsnotify_update_flags(dentry
);
1829 spin_unlock(&dentry
->d_lock
);
1833 * d_instantiate - fill in inode information for a dentry
1834 * @entry: dentry to complete
1835 * @inode: inode to attach to this dentry
1837 * Fill in inode information in the entry.
1839 * This turns negative dentries into productive full members
1842 * NOTE! This assumes that the inode count has been incremented
1843 * (or otherwise set) by the caller to indicate that it is now
1844 * in use by the dcache.
1847 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1849 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1851 security_d_instantiate(entry
, inode
);
1852 spin_lock(&inode
->i_lock
);
1853 __d_instantiate(entry
, inode
);
1854 spin_unlock(&inode
->i_lock
);
1857 EXPORT_SYMBOL(d_instantiate
);
1860 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1861 * @entry: dentry to complete
1862 * @inode: inode to attach to this dentry
1864 * Fill in inode information in the entry. If a directory alias is found, then
1865 * return an error (and drop inode). Together with d_materialise_unique() this
1866 * guarantees that a directory inode may never have more than one alias.
1868 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1870 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1872 security_d_instantiate(entry
, inode
);
1873 spin_lock(&inode
->i_lock
);
1874 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1875 spin_unlock(&inode
->i_lock
);
1879 __d_instantiate(entry
, inode
);
1880 spin_unlock(&inode
->i_lock
);
1884 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1886 struct dentry
*d_make_root(struct inode
*root_inode
)
1888 struct dentry
*res
= NULL
;
1891 res
= __d_alloc(root_inode
->i_sb
, NULL
);
1893 d_instantiate(res
, root_inode
);
1899 EXPORT_SYMBOL(d_make_root
);
1901 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1903 struct dentry
*alias
;
1905 if (hlist_empty(&inode
->i_dentry
))
1907 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1913 * d_find_any_alias - find any alias for a given inode
1914 * @inode: inode to find an alias for
1916 * If any aliases exist for the given inode, take and return a
1917 * reference for one of them. If no aliases exist, return %NULL.
1919 struct dentry
*d_find_any_alias(struct inode
*inode
)
1923 spin_lock(&inode
->i_lock
);
1924 de
= __d_find_any_alias(inode
);
1925 spin_unlock(&inode
->i_lock
);
1928 EXPORT_SYMBOL(d_find_any_alias
);
1930 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1937 return ERR_PTR(-ESTALE
);
1939 return ERR_CAST(inode
);
1941 res
= d_find_any_alias(inode
);
1945 tmp
= __d_alloc(inode
->i_sb
, NULL
);
1947 res
= ERR_PTR(-ENOMEM
);
1951 security_d_instantiate(tmp
, inode
);
1952 spin_lock(&inode
->i_lock
);
1953 res
= __d_find_any_alias(inode
);
1955 spin_unlock(&inode
->i_lock
);
1960 /* attach a disconnected dentry */
1961 add_flags
= d_flags_for_inode(inode
);
1964 add_flags
|= DCACHE_DISCONNECTED
;
1966 spin_lock(&tmp
->d_lock
);
1967 __d_set_inode_and_type(tmp
, inode
, add_flags
);
1968 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
1969 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
1970 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
1971 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
1972 spin_unlock(&tmp
->d_lock
);
1973 spin_unlock(&inode
->i_lock
);
1983 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1984 * @inode: inode to allocate the dentry for
1986 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1987 * similar open by handle operations. The returned dentry may be anonymous,
1988 * or may have a full name (if the inode was already in the cache).
1990 * When called on a directory inode, we must ensure that the inode only ever
1991 * has one dentry. If a dentry is found, that is returned instead of
1992 * allocating a new one.
1994 * On successful return, the reference to the inode has been transferred
1995 * to the dentry. In case of an error the reference on the inode is released.
1996 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1997 * be passed in and the error will be propagated to the return value,
1998 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2000 struct dentry
*d_obtain_alias(struct inode
*inode
)
2002 return __d_obtain_alias(inode
, 1);
2004 EXPORT_SYMBOL(d_obtain_alias
);
2007 * d_obtain_root - find or allocate a dentry for a given inode
2008 * @inode: inode to allocate the dentry for
2010 * Obtain an IS_ROOT dentry for the root of a filesystem.
2012 * We must ensure that directory inodes only ever have one dentry. If a
2013 * dentry is found, that is returned instead of allocating a new one.
2015 * On successful return, the reference to the inode has been transferred
2016 * to the dentry. In case of an error the reference on the inode is
2017 * released. A %NULL or IS_ERR inode may be passed in and will be the
2018 * error will be propagate to the return value, with a %NULL @inode
2019 * replaced by ERR_PTR(-ESTALE).
2021 struct dentry
*d_obtain_root(struct inode
*inode
)
2023 return __d_obtain_alias(inode
, 0);
2025 EXPORT_SYMBOL(d_obtain_root
);
2028 * d_add_ci - lookup or allocate new dentry with case-exact name
2029 * @inode: the inode case-insensitive lookup has found
2030 * @dentry: the negative dentry that was passed to the parent's lookup func
2031 * @name: the case-exact name to be associated with the returned dentry
2033 * This is to avoid filling the dcache with case-insensitive names to the
2034 * same inode, only the actual correct case is stored in the dcache for
2035 * case-insensitive filesystems.
2037 * For a case-insensitive lookup match and if the the case-exact dentry
2038 * already exists in in the dcache, use it and return it.
2040 * If no entry exists with the exact case name, allocate new dentry with
2041 * the exact case, and return the spliced entry.
2043 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2046 struct dentry
*found
, *res
;
2049 * First check if a dentry matching the name already exists,
2050 * if not go ahead and create it now.
2052 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2057 if (d_in_lookup(dentry
)) {
2058 found
= d_alloc_parallel(dentry
->d_parent
, name
,
2060 if (IS_ERR(found
) || !d_in_lookup(found
)) {
2065 found
= d_alloc(dentry
->d_parent
, name
);
2068 return ERR_PTR(-ENOMEM
);
2071 res
= d_splice_alias(inode
, found
);
2078 EXPORT_SYMBOL(d_add_ci
);
2081 static inline bool d_same_name(const struct dentry
*dentry
,
2082 const struct dentry
*parent
,
2083 const struct qstr
*name
)
2085 if (likely(!(parent
->d_flags
& DCACHE_OP_COMPARE
))) {
2086 if (dentry
->d_name
.len
!= name
->len
)
2088 return dentry_cmp(dentry
, name
->name
, name
->len
) == 0;
2090 return parent
->d_op
->d_compare(dentry
,
2091 dentry
->d_name
.len
, dentry
->d_name
.name
,
2096 * __d_lookup_rcu - search for a dentry (racy, store-free)
2097 * @parent: parent dentry
2098 * @name: qstr of name we wish to find
2099 * @seqp: returns d_seq value at the point where the dentry was found
2100 * Returns: dentry, or NULL
2102 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2103 * resolution (store-free path walking) design described in
2104 * Documentation/filesystems/path-lookup.txt.
2106 * This is not to be used outside core vfs.
2108 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2109 * held, and rcu_read_lock held. The returned dentry must not be stored into
2110 * without taking d_lock and checking d_seq sequence count against @seq
2113 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2116 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2117 * the returned dentry, so long as its parent's seqlock is checked after the
2118 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2119 * is formed, giving integrity down the path walk.
2121 * NOTE! The caller *has* to check the resulting dentry against the sequence
2122 * number we've returned before using any of the resulting dentry state!
2124 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2125 const struct qstr
*name
,
2128 u64 hashlen
= name
->hash_len
;
2129 const unsigned char *str
= name
->name
;
2130 struct hlist_bl_head
*b
= d_hash(hashlen_hash(hashlen
));
2131 struct hlist_bl_node
*node
;
2132 struct dentry
*dentry
;
2135 * Note: There is significant duplication with __d_lookup_rcu which is
2136 * required to prevent single threaded performance regressions
2137 * especially on architectures where smp_rmb (in seqcounts) are costly.
2138 * Keep the two functions in sync.
2142 * The hash list is protected using RCU.
2144 * Carefully use d_seq when comparing a candidate dentry, to avoid
2145 * races with d_move().
2147 * It is possible that concurrent renames can mess up our list
2148 * walk here and result in missing our dentry, resulting in the
2149 * false-negative result. d_lookup() protects against concurrent
2150 * renames using rename_lock seqlock.
2152 * See Documentation/filesystems/path-lookup.txt for more details.
2154 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2159 * The dentry sequence count protects us from concurrent
2160 * renames, and thus protects parent and name fields.
2162 * The caller must perform a seqcount check in order
2163 * to do anything useful with the returned dentry.
2165 * NOTE! We do a "raw" seqcount_begin here. That means that
2166 * we don't wait for the sequence count to stabilize if it
2167 * is in the middle of a sequence change. If we do the slow
2168 * dentry compare, we will do seqretries until it is stable,
2169 * and if we end up with a successful lookup, we actually
2170 * want to exit RCU lookup anyway.
2172 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2173 * we are still guaranteed NUL-termination of ->d_name.name.
2175 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2176 if (dentry
->d_parent
!= parent
)
2178 if (d_unhashed(dentry
))
2181 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2184 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2186 tlen
= dentry
->d_name
.len
;
2187 tname
= dentry
->d_name
.name
;
2188 /* we want a consistent (name,len) pair */
2189 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2193 if (parent
->d_op
->d_compare(dentry
,
2194 tlen
, tname
, name
) != 0)
2197 if (dentry
->d_name
.hash_len
!= hashlen
)
2199 if (dentry_cmp(dentry
, str
, hashlen_len(hashlen
)) != 0)
2209 * d_lookup - search for a dentry
2210 * @parent: parent dentry
2211 * @name: qstr of name we wish to find
2212 * Returns: dentry, or NULL
2214 * d_lookup searches the children of the parent dentry for the name in
2215 * question. If the dentry is found its reference count is incremented and the
2216 * dentry is returned. The caller must use dput to free the entry when it has
2217 * finished using it. %NULL is returned if the dentry does not exist.
2219 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2221 struct dentry
*dentry
;
2225 seq
= read_seqbegin(&rename_lock
);
2226 dentry
= __d_lookup(parent
, name
);
2229 } while (read_seqretry(&rename_lock
, seq
));
2232 EXPORT_SYMBOL(d_lookup
);
2235 * __d_lookup - search for a dentry (racy)
2236 * @parent: parent dentry
2237 * @name: qstr of name we wish to find
2238 * Returns: dentry, or NULL
2240 * __d_lookup is like d_lookup, however it may (rarely) return a
2241 * false-negative result due to unrelated rename activity.
2243 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2244 * however it must be used carefully, eg. with a following d_lookup in
2245 * the case of failure.
2247 * __d_lookup callers must be commented.
2249 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2251 unsigned int hash
= name
->hash
;
2252 struct hlist_bl_head
*b
= d_hash(hash
);
2253 struct hlist_bl_node
*node
;
2254 struct dentry
*found
= NULL
;
2255 struct dentry
*dentry
;
2258 * Note: There is significant duplication with __d_lookup_rcu which is
2259 * required to prevent single threaded performance regressions
2260 * especially on architectures where smp_rmb (in seqcounts) are costly.
2261 * Keep the two functions in sync.
2265 * The hash list is protected using RCU.
2267 * Take d_lock when comparing a candidate dentry, to avoid races
2270 * It is possible that concurrent renames can mess up our list
2271 * walk here and result in missing our dentry, resulting in the
2272 * false-negative result. d_lookup() protects against concurrent
2273 * renames using rename_lock seqlock.
2275 * See Documentation/filesystems/path-lookup.txt for more details.
2279 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2281 if (dentry
->d_name
.hash
!= hash
)
2284 spin_lock(&dentry
->d_lock
);
2285 if (dentry
->d_parent
!= parent
)
2287 if (d_unhashed(dentry
))
2290 if (!d_same_name(dentry
, parent
, name
))
2293 dentry
->d_lockref
.count
++;
2295 spin_unlock(&dentry
->d_lock
);
2298 spin_unlock(&dentry
->d_lock
);
2306 * d_hash_and_lookup - hash the qstr then search for a dentry
2307 * @dir: Directory to search in
2308 * @name: qstr of name we wish to find
2310 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2312 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2315 * Check for a fs-specific hash function. Note that we must
2316 * calculate the standard hash first, as the d_op->d_hash()
2317 * routine may choose to leave the hash value unchanged.
2319 name
->hash
= full_name_hash(dir
, name
->name
, name
->len
);
2320 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2321 int err
= dir
->d_op
->d_hash(dir
, name
);
2322 if (unlikely(err
< 0))
2323 return ERR_PTR(err
);
2325 return d_lookup(dir
, name
);
2327 EXPORT_SYMBOL(d_hash_and_lookup
);
2330 * When a file is deleted, we have two options:
2331 * - turn this dentry into a negative dentry
2332 * - unhash this dentry and free it.
2334 * Usually, we want to just turn this into
2335 * a negative dentry, but if anybody else is
2336 * currently using the dentry or the inode
2337 * we can't do that and we fall back on removing
2338 * it from the hash queues and waiting for
2339 * it to be deleted later when it has no users
2343 * d_delete - delete a dentry
2344 * @dentry: The dentry to delete
2346 * Turn the dentry into a negative dentry if possible, otherwise
2347 * remove it from the hash queues so it can be deleted later
2350 void d_delete(struct dentry
* dentry
)
2352 struct inode
*inode
;
2355 * Are we the only user?
2358 spin_lock(&dentry
->d_lock
);
2359 inode
= dentry
->d_inode
;
2360 isdir
= S_ISDIR(inode
->i_mode
);
2361 if (dentry
->d_lockref
.count
== 1) {
2362 if (!spin_trylock(&inode
->i_lock
)) {
2363 spin_unlock(&dentry
->d_lock
);
2367 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2368 dentry_unlink_inode(dentry
);
2369 fsnotify_nameremove(dentry
, isdir
);
2373 if (!d_unhashed(dentry
))
2376 spin_unlock(&dentry
->d_lock
);
2378 fsnotify_nameremove(dentry
, isdir
);
2380 EXPORT_SYMBOL(d_delete
);
2382 static void __d_rehash(struct dentry
*entry
)
2384 struct hlist_bl_head
*b
= d_hash(entry
->d_name
.hash
);
2385 BUG_ON(!d_unhashed(entry
));
2387 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2392 * d_rehash - add an entry back to the hash
2393 * @entry: dentry to add to the hash
2395 * Adds a dentry to the hash according to its name.
2398 void d_rehash(struct dentry
* entry
)
2400 spin_lock(&entry
->d_lock
);
2402 spin_unlock(&entry
->d_lock
);
2404 EXPORT_SYMBOL(d_rehash
);
2406 static inline unsigned start_dir_add(struct inode
*dir
)
2410 unsigned n
= dir
->i_dir_seq
;
2411 if (!(n
& 1) && cmpxchg(&dir
->i_dir_seq
, n
, n
+ 1) == n
)
2417 static inline void end_dir_add(struct inode
*dir
, unsigned n
)
2419 smp_store_release(&dir
->i_dir_seq
, n
+ 2);
2422 static void d_wait_lookup(struct dentry
*dentry
)
2424 if (d_in_lookup(dentry
)) {
2425 DECLARE_WAITQUEUE(wait
, current
);
2426 add_wait_queue(dentry
->d_wait
, &wait
);
2428 set_current_state(TASK_UNINTERRUPTIBLE
);
2429 spin_unlock(&dentry
->d_lock
);
2431 spin_lock(&dentry
->d_lock
);
2432 } while (d_in_lookup(dentry
));
2436 struct dentry
*d_alloc_parallel(struct dentry
*parent
,
2437 const struct qstr
*name
,
2438 wait_queue_head_t
*wq
)
2440 unsigned int hash
= name
->hash
;
2441 struct hlist_bl_head
*b
= in_lookup_hash(parent
, hash
);
2442 struct hlist_bl_node
*node
;
2443 struct dentry
*new = d_alloc(parent
, name
);
2444 struct dentry
*dentry
;
2445 unsigned seq
, r_seq
, d_seq
;
2448 return ERR_PTR(-ENOMEM
);
2452 seq
= smp_load_acquire(&parent
->d_inode
->i_dir_seq
) & ~1;
2453 r_seq
= read_seqbegin(&rename_lock
);
2454 dentry
= __d_lookup_rcu(parent
, name
, &d_seq
);
2455 if (unlikely(dentry
)) {
2456 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2460 if (read_seqcount_retry(&dentry
->d_seq
, d_seq
)) {
2469 if (unlikely(read_seqretry(&rename_lock
, r_seq
))) {
2474 if (unlikely(parent
->d_inode
->i_dir_seq
!= seq
)) {
2480 * No changes for the parent since the beginning of d_lookup().
2481 * Since all removals from the chain happen with hlist_bl_lock(),
2482 * any potential in-lookup matches are going to stay here until
2483 * we unlock the chain. All fields are stable in everything
2486 hlist_bl_for_each_entry(dentry
, node
, b
, d_u
.d_in_lookup_hash
) {
2487 if (dentry
->d_name
.hash
!= hash
)
2489 if (dentry
->d_parent
!= parent
)
2491 if (!d_same_name(dentry
, parent
, name
))
2494 /* now we can try to grab a reference */
2495 if (!lockref_get_not_dead(&dentry
->d_lockref
)) {
2502 * somebody is likely to be still doing lookup for it;
2503 * wait for them to finish
2505 spin_lock(&dentry
->d_lock
);
2506 d_wait_lookup(dentry
);
2508 * it's not in-lookup anymore; in principle we should repeat
2509 * everything from dcache lookup, but it's likely to be what
2510 * d_lookup() would've found anyway. If it is, just return it;
2511 * otherwise we really have to repeat the whole thing.
2513 if (unlikely(dentry
->d_name
.hash
!= hash
))
2515 if (unlikely(dentry
->d_parent
!= parent
))
2517 if (unlikely(d_unhashed(dentry
)))
2519 if (unlikely(!d_same_name(dentry
, parent
, name
)))
2521 /* OK, it *is* a hashed match; return it */
2522 spin_unlock(&dentry
->d_lock
);
2527 /* we can't take ->d_lock here; it's OK, though. */
2528 new->d_flags
|= DCACHE_PAR_LOOKUP
;
2530 hlist_bl_add_head_rcu(&new->d_u
.d_in_lookup_hash
, b
);
2534 spin_unlock(&dentry
->d_lock
);
2538 EXPORT_SYMBOL(d_alloc_parallel
);
2540 void __d_lookup_done(struct dentry
*dentry
)
2542 struct hlist_bl_head
*b
= in_lookup_hash(dentry
->d_parent
,
2543 dentry
->d_name
.hash
);
2545 dentry
->d_flags
&= ~DCACHE_PAR_LOOKUP
;
2546 __hlist_bl_del(&dentry
->d_u
.d_in_lookup_hash
);
2547 wake_up_all(dentry
->d_wait
);
2548 dentry
->d_wait
= NULL
;
2550 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
2551 INIT_LIST_HEAD(&dentry
->d_lru
);
2553 EXPORT_SYMBOL(__d_lookup_done
);
2555 /* inode->i_lock held if inode is non-NULL */
2557 static inline void __d_add(struct dentry
*dentry
, struct inode
*inode
)
2559 struct inode
*dir
= NULL
;
2561 spin_lock(&dentry
->d_lock
);
2562 if (unlikely(d_in_lookup(dentry
))) {
2563 dir
= dentry
->d_parent
->d_inode
;
2564 n
= start_dir_add(dir
);
2565 __d_lookup_done(dentry
);
2568 unsigned add_flags
= d_flags_for_inode(inode
);
2569 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
2570 raw_write_seqcount_begin(&dentry
->d_seq
);
2571 __d_set_inode_and_type(dentry
, inode
, add_flags
);
2572 raw_write_seqcount_end(&dentry
->d_seq
);
2573 fsnotify_update_flags(dentry
);
2577 end_dir_add(dir
, n
);
2578 spin_unlock(&dentry
->d_lock
);
2580 spin_unlock(&inode
->i_lock
);
2584 * d_add - add dentry to hash queues
2585 * @entry: dentry to add
2586 * @inode: The inode to attach to this dentry
2588 * This adds the entry to the hash queues and initializes @inode.
2589 * The entry was actually filled in earlier during d_alloc().
2592 void d_add(struct dentry
*entry
, struct inode
*inode
)
2595 security_d_instantiate(entry
, inode
);
2596 spin_lock(&inode
->i_lock
);
2598 __d_add(entry
, inode
);
2600 EXPORT_SYMBOL(d_add
);
2603 * d_exact_alias - find and hash an exact unhashed alias
2604 * @entry: dentry to add
2605 * @inode: The inode to go with this dentry
2607 * If an unhashed dentry with the same name/parent and desired
2608 * inode already exists, hash and return it. Otherwise, return
2611 * Parent directory should be locked.
2613 struct dentry
*d_exact_alias(struct dentry
*entry
, struct inode
*inode
)
2615 struct dentry
*alias
;
2616 unsigned int hash
= entry
->d_name
.hash
;
2618 spin_lock(&inode
->i_lock
);
2619 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
2621 * Don't need alias->d_lock here, because aliases with
2622 * d_parent == entry->d_parent are not subject to name or
2623 * parent changes, because the parent inode i_mutex is held.
2625 if (alias
->d_name
.hash
!= hash
)
2627 if (alias
->d_parent
!= entry
->d_parent
)
2629 if (!d_same_name(alias
, entry
->d_parent
, &entry
->d_name
))
2631 spin_lock(&alias
->d_lock
);
2632 if (!d_unhashed(alias
)) {
2633 spin_unlock(&alias
->d_lock
);
2636 __dget_dlock(alias
);
2638 spin_unlock(&alias
->d_lock
);
2640 spin_unlock(&inode
->i_lock
);
2643 spin_unlock(&inode
->i_lock
);
2646 EXPORT_SYMBOL(d_exact_alias
);
2649 * dentry_update_name_case - update case insensitive dentry with a new name
2650 * @dentry: dentry to be updated
2653 * Update a case insensitive dentry with new case of name.
2655 * dentry must have been returned by d_lookup with name @name. Old and new
2656 * name lengths must match (ie. no d_compare which allows mismatched name
2659 * Parent inode i_mutex must be held over d_lookup and into this call (to
2660 * keep renames and concurrent inserts, and readdir(2) away).
2662 void dentry_update_name_case(struct dentry
*dentry
, const struct qstr
*name
)
2664 BUG_ON(!inode_is_locked(dentry
->d_parent
->d_inode
));
2665 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2667 spin_lock(&dentry
->d_lock
);
2668 write_seqcount_begin(&dentry
->d_seq
);
2669 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2670 write_seqcount_end(&dentry
->d_seq
);
2671 spin_unlock(&dentry
->d_lock
);
2673 EXPORT_SYMBOL(dentry_update_name_case
);
2675 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2677 if (unlikely(dname_external(target
))) {
2678 if (unlikely(dname_external(dentry
))) {
2680 * Both external: swap the pointers
2682 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2685 * dentry:internal, target:external. Steal target's
2686 * storage and make target internal.
2688 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2689 dentry
->d_name
.len
+ 1);
2690 dentry
->d_name
.name
= target
->d_name
.name
;
2691 target
->d_name
.name
= target
->d_iname
;
2694 if (unlikely(dname_external(dentry
))) {
2696 * dentry:external, target:internal. Give dentry's
2697 * storage to target and make dentry internal
2699 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2700 target
->d_name
.len
+ 1);
2701 target
->d_name
.name
= dentry
->d_name
.name
;
2702 dentry
->d_name
.name
= dentry
->d_iname
;
2705 * Both are internal.
2708 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2709 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2710 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2711 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2712 swap(((long *) &dentry
->d_iname
)[i
],
2713 ((long *) &target
->d_iname
)[i
]);
2717 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2720 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2722 struct external_name
*old_name
= NULL
;
2723 if (unlikely(dname_external(dentry
)))
2724 old_name
= external_name(dentry
);
2725 if (unlikely(dname_external(target
))) {
2726 atomic_inc(&external_name(target
)->u
.count
);
2727 dentry
->d_name
= target
->d_name
;
2729 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2730 target
->d_name
.len
+ 1);
2731 dentry
->d_name
.name
= dentry
->d_iname
;
2732 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2734 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2735 kfree_rcu(old_name
, u
.head
);
2738 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2741 * XXXX: do we really need to take target->d_lock?
2743 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2744 spin_lock(&target
->d_parent
->d_lock
);
2746 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2747 spin_lock(&dentry
->d_parent
->d_lock
);
2748 spin_lock_nested(&target
->d_parent
->d_lock
,
2749 DENTRY_D_LOCK_NESTED
);
2751 spin_lock(&target
->d_parent
->d_lock
);
2752 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2753 DENTRY_D_LOCK_NESTED
);
2756 if (target
< dentry
) {
2757 spin_lock_nested(&target
->d_lock
, 2);
2758 spin_lock_nested(&dentry
->d_lock
, 3);
2760 spin_lock_nested(&dentry
->d_lock
, 2);
2761 spin_lock_nested(&target
->d_lock
, 3);
2765 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2767 if (target
->d_parent
!= dentry
->d_parent
)
2768 spin_unlock(&dentry
->d_parent
->d_lock
);
2769 if (target
->d_parent
!= target
)
2770 spin_unlock(&target
->d_parent
->d_lock
);
2771 spin_unlock(&target
->d_lock
);
2772 spin_unlock(&dentry
->d_lock
);
2776 * When switching names, the actual string doesn't strictly have to
2777 * be preserved in the target - because we're dropping the target
2778 * anyway. As such, we can just do a simple memcpy() to copy over
2779 * the new name before we switch, unless we are going to rehash
2780 * it. Note that if we *do* unhash the target, we are not allowed
2781 * to rehash it without giving it a new name/hash key - whether
2782 * we swap or overwrite the names here, resulting name won't match
2783 * the reality in filesystem; it's only there for d_path() purposes.
2784 * Note that all of this is happening under rename_lock, so the
2785 * any hash lookup seeing it in the middle of manipulations will
2786 * be discarded anyway. So we do not care what happens to the hash
2790 * __d_move - move a dentry
2791 * @dentry: entry to move
2792 * @target: new dentry
2793 * @exchange: exchange the two dentries
2795 * Update the dcache to reflect the move of a file name. Negative
2796 * dcache entries should not be moved in this way. Caller must hold
2797 * rename_lock, the i_mutex of the source and target directories,
2798 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2800 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2803 struct inode
*dir
= NULL
;
2805 if (!dentry
->d_inode
)
2806 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2808 BUG_ON(d_ancestor(dentry
, target
));
2809 BUG_ON(d_ancestor(target
, dentry
));
2811 dentry_lock_for_move(dentry
, target
);
2812 if (unlikely(d_in_lookup(target
))) {
2813 dir
= target
->d_parent
->d_inode
;
2814 n
= start_dir_add(dir
);
2815 __d_lookup_done(target
);
2818 write_seqcount_begin(&dentry
->d_seq
);
2819 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2822 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2826 /* Switch the names.. */
2828 swap_names(dentry
, target
);
2830 copy_name(dentry
, target
);
2832 /* rehash in new place(s) */
2837 /* ... and switch them in the tree */
2838 if (IS_ROOT(dentry
)) {
2839 /* splicing a tree */
2840 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2841 dentry
->d_parent
= target
->d_parent
;
2842 target
->d_parent
= target
;
2843 list_del_init(&target
->d_child
);
2844 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2846 /* swapping two dentries */
2847 swap(dentry
->d_parent
, target
->d_parent
);
2848 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2849 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2851 fsnotify_update_flags(target
);
2852 fsnotify_update_flags(dentry
);
2855 write_seqcount_end(&target
->d_seq
);
2856 write_seqcount_end(&dentry
->d_seq
);
2859 end_dir_add(dir
, n
);
2860 dentry_unlock_for_move(dentry
, target
);
2864 * d_move - move a dentry
2865 * @dentry: entry to move
2866 * @target: new dentry
2868 * Update the dcache to reflect the move of a file name. Negative
2869 * dcache entries should not be moved in this way. See the locking
2870 * requirements for __d_move.
2872 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2874 write_seqlock(&rename_lock
);
2875 __d_move(dentry
, target
, false);
2876 write_sequnlock(&rename_lock
);
2878 EXPORT_SYMBOL(d_move
);
2881 * d_exchange - exchange two dentries
2882 * @dentry1: first dentry
2883 * @dentry2: second dentry
2885 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2887 write_seqlock(&rename_lock
);
2889 WARN_ON(!dentry1
->d_inode
);
2890 WARN_ON(!dentry2
->d_inode
);
2891 WARN_ON(IS_ROOT(dentry1
));
2892 WARN_ON(IS_ROOT(dentry2
));
2894 __d_move(dentry1
, dentry2
, true);
2896 write_sequnlock(&rename_lock
);
2898 EXPORT_SYMBOL_GPL(d_exchange
);
2901 * d_ancestor - search for an ancestor
2902 * @p1: ancestor dentry
2905 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2906 * an ancestor of p2, else NULL.
2908 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2912 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2913 if (p
->d_parent
== p1
)
2920 * This helper attempts to cope with remotely renamed directories
2922 * It assumes that the caller is already holding
2923 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2925 * Note: If ever the locking in lock_rename() changes, then please
2926 * remember to update this too...
2928 static int __d_unalias(struct inode
*inode
,
2929 struct dentry
*dentry
, struct dentry
*alias
)
2931 struct mutex
*m1
= NULL
;
2932 struct rw_semaphore
*m2
= NULL
;
2935 /* If alias and dentry share a parent, then no extra locks required */
2936 if (alias
->d_parent
== dentry
->d_parent
)
2939 /* See lock_rename() */
2940 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2942 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2943 if (!inode_trylock_shared(alias
->d_parent
->d_inode
))
2945 m2
= &alias
->d_parent
->d_inode
->i_rwsem
;
2947 __d_move(alias
, dentry
, false);
2958 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2959 * @inode: the inode which may have a disconnected dentry
2960 * @dentry: a negative dentry which we want to point to the inode.
2962 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2963 * place of the given dentry and return it, else simply d_add the inode
2964 * to the dentry and return NULL.
2966 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2967 * we should error out: directories can't have multiple aliases.
2969 * This is needed in the lookup routine of any filesystem that is exportable
2970 * (via knfsd) so that we can build dcache paths to directories effectively.
2972 * If a dentry was found and moved, then it is returned. Otherwise NULL
2973 * is returned. This matches the expected return value of ->lookup.
2975 * Cluster filesystems may call this function with a negative, hashed dentry.
2976 * In that case, we know that the inode will be a regular file, and also this
2977 * will only occur during atomic_open. So we need to check for the dentry
2978 * being already hashed only in the final case.
2980 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2983 return ERR_CAST(inode
);
2985 BUG_ON(!d_unhashed(dentry
));
2990 security_d_instantiate(dentry
, inode
);
2991 spin_lock(&inode
->i_lock
);
2992 if (S_ISDIR(inode
->i_mode
)) {
2993 struct dentry
*new = __d_find_any_alias(inode
);
2994 if (unlikely(new)) {
2995 /* The reference to new ensures it remains an alias */
2996 spin_unlock(&inode
->i_lock
);
2997 write_seqlock(&rename_lock
);
2998 if (unlikely(d_ancestor(new, dentry
))) {
2999 write_sequnlock(&rename_lock
);
3001 new = ERR_PTR(-ELOOP
);
3002 pr_warn_ratelimited(
3003 "VFS: Lookup of '%s' in %s %s"
3004 " would have caused loop\n",
3005 dentry
->d_name
.name
,
3006 inode
->i_sb
->s_type
->name
,
3008 } else if (!IS_ROOT(new)) {
3009 int err
= __d_unalias(inode
, dentry
, new);
3010 write_sequnlock(&rename_lock
);
3016 __d_move(new, dentry
, false);
3017 write_sequnlock(&rename_lock
);
3024 __d_add(dentry
, inode
);
3027 EXPORT_SYMBOL(d_splice_alias
);
3029 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
3033 return -ENAMETOOLONG
;
3035 memcpy(*buffer
, str
, namelen
);
3040 * prepend_name - prepend a pathname in front of current buffer pointer
3041 * @buffer: buffer pointer
3042 * @buflen: allocated length of the buffer
3043 * @name: name string and length qstr structure
3045 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3046 * make sure that either the old or the new name pointer and length are
3047 * fetched. However, there may be mismatch between length and pointer.
3048 * The length cannot be trusted, we need to copy it byte-by-byte until
3049 * the length is reached or a null byte is found. It also prepends "/" at
3050 * the beginning of the name. The sequence number check at the caller will
3051 * retry it again when a d_move() does happen. So any garbage in the buffer
3052 * due to mismatched pointer and length will be discarded.
3054 * Data dependency barrier is needed to make sure that we see that terminating
3055 * NUL. Alpha strikes again, film at 11...
3057 static int prepend_name(char **buffer
, int *buflen
, const struct qstr
*name
)
3059 const char *dname
= ACCESS_ONCE(name
->name
);
3060 u32 dlen
= ACCESS_ONCE(name
->len
);
3063 smp_read_barrier_depends();
3065 *buflen
-= dlen
+ 1;
3067 return -ENAMETOOLONG
;
3068 p
= *buffer
-= dlen
+ 1;
3080 * prepend_path - Prepend path string to a buffer
3081 * @path: the dentry/vfsmount to report
3082 * @root: root vfsmnt/dentry
3083 * @buffer: pointer to the end of the buffer
3084 * @buflen: pointer to buffer length
3086 * The function will first try to write out the pathname without taking any
3087 * lock other than the RCU read lock to make sure that dentries won't go away.
3088 * It only checks the sequence number of the global rename_lock as any change
3089 * in the dentry's d_seq will be preceded by changes in the rename_lock
3090 * sequence number. If the sequence number had been changed, it will restart
3091 * the whole pathname back-tracing sequence again by taking the rename_lock.
3092 * In this case, there is no need to take the RCU read lock as the recursive
3093 * parent pointer references will keep the dentry chain alive as long as no
3094 * rename operation is performed.
3096 static int prepend_path(const struct path
*path
,
3097 const struct path
*root
,
3098 char **buffer
, int *buflen
)
3100 struct dentry
*dentry
;
3101 struct vfsmount
*vfsmnt
;
3104 unsigned seq
, m_seq
= 0;
3110 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
3117 dentry
= path
->dentry
;
3119 mnt
= real_mount(vfsmnt
);
3120 read_seqbegin_or_lock(&rename_lock
, &seq
);
3121 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
3122 struct dentry
* parent
;
3124 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
3125 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
3127 if (dentry
!= vfsmnt
->mnt_root
) {
3134 if (mnt
!= parent
) {
3135 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
3141 error
= is_mounted(vfsmnt
) ? 1 : 2;
3144 parent
= dentry
->d_parent
;
3146 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
3154 if (need_seqretry(&rename_lock
, seq
)) {
3158 done_seqretry(&rename_lock
, seq
);
3162 if (need_seqretry(&mount_lock
, m_seq
)) {
3166 done_seqretry(&mount_lock
, m_seq
);
3168 if (error
>= 0 && bptr
== *buffer
) {
3170 error
= -ENAMETOOLONG
;
3180 * __d_path - return the path of a dentry
3181 * @path: the dentry/vfsmount to report
3182 * @root: root vfsmnt/dentry
3183 * @buf: buffer to return value in
3184 * @buflen: buffer length
3186 * Convert a dentry into an ASCII path name.
3188 * Returns a pointer into the buffer or an error code if the
3189 * path was too long.
3191 * "buflen" should be positive.
3193 * If the path is not reachable from the supplied root, return %NULL.
3195 char *__d_path(const struct path
*path
,
3196 const struct path
*root
,
3197 char *buf
, int buflen
)
3199 char *res
= buf
+ buflen
;
3202 prepend(&res
, &buflen
, "\0", 1);
3203 error
= prepend_path(path
, root
, &res
, &buflen
);
3206 return ERR_PTR(error
);
3212 char *d_absolute_path(const struct path
*path
,
3213 char *buf
, int buflen
)
3215 struct path root
= {};
3216 char *res
= buf
+ buflen
;
3219 prepend(&res
, &buflen
, "\0", 1);
3220 error
= prepend_path(path
, &root
, &res
, &buflen
);
3225 return ERR_PTR(error
);
3230 * same as __d_path but appends "(deleted)" for unlinked files.
3232 static int path_with_deleted(const struct path
*path
,
3233 const struct path
*root
,
3234 char **buf
, int *buflen
)
3236 prepend(buf
, buflen
, "\0", 1);
3237 if (d_unlinked(path
->dentry
)) {
3238 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3243 return prepend_path(path
, root
, buf
, buflen
);
3246 static int prepend_unreachable(char **buffer
, int *buflen
)
3248 return prepend(buffer
, buflen
, "(unreachable)", 13);
3251 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3256 seq
= read_seqcount_begin(&fs
->seq
);
3258 } while (read_seqcount_retry(&fs
->seq
, seq
));
3262 * d_path - return the path of a dentry
3263 * @path: path to report
3264 * @buf: buffer to return value in
3265 * @buflen: buffer length
3267 * Convert a dentry into an ASCII path name. If the entry has been deleted
3268 * the string " (deleted)" is appended. Note that this is ambiguous.
3270 * Returns a pointer into the buffer or an error code if the path was
3271 * too long. Note: Callers should use the returned pointer, not the passed
3272 * in buffer, to use the name! The implementation often starts at an offset
3273 * into the buffer, and may leave 0 bytes at the start.
3275 * "buflen" should be positive.
3277 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3279 char *res
= buf
+ buflen
;
3284 * We have various synthetic filesystems that never get mounted. On
3285 * these filesystems dentries are never used for lookup purposes, and
3286 * thus don't need to be hashed. They also don't need a name until a
3287 * user wants to identify the object in /proc/pid/fd/. The little hack
3288 * below allows us to generate a name for these objects on demand:
3290 * Some pseudo inodes are mountable. When they are mounted
3291 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3292 * and instead have d_path return the mounted path.
3294 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3295 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3296 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3299 get_fs_root_rcu(current
->fs
, &root
);
3300 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3304 res
= ERR_PTR(error
);
3307 EXPORT_SYMBOL(d_path
);
3310 * Helper function for dentry_operations.d_dname() members
3312 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3313 const char *fmt
, ...)
3319 va_start(args
, fmt
);
3320 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3323 if (sz
> sizeof(temp
) || sz
> buflen
)
3324 return ERR_PTR(-ENAMETOOLONG
);
3326 buffer
+= buflen
- sz
;
3327 return memcpy(buffer
, temp
, sz
);
3330 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3332 char *end
= buffer
+ buflen
;
3333 /* these dentries are never renamed, so d_lock is not needed */
3334 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3335 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3336 prepend(&end
, &buflen
, "/", 1))
3337 end
= ERR_PTR(-ENAMETOOLONG
);
3340 EXPORT_SYMBOL(simple_dname
);
3343 * Write full pathname from the root of the filesystem into the buffer.
3345 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3347 struct dentry
*dentry
;
3360 prepend(&end
, &len
, "\0", 1);
3364 read_seqbegin_or_lock(&rename_lock
, &seq
);
3365 while (!IS_ROOT(dentry
)) {
3366 struct dentry
*parent
= dentry
->d_parent
;
3369 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3378 if (need_seqretry(&rename_lock
, seq
)) {
3382 done_seqretry(&rename_lock
, seq
);
3387 return ERR_PTR(-ENAMETOOLONG
);
3390 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3392 return __dentry_path(dentry
, buf
, buflen
);
3394 EXPORT_SYMBOL(dentry_path_raw
);
3396 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3401 if (d_unlinked(dentry
)) {
3403 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3407 retval
= __dentry_path(dentry
, buf
, buflen
);
3408 if (!IS_ERR(retval
) && p
)
3409 *p
= '/'; /* restore '/' overriden with '\0' */
3412 return ERR_PTR(-ENAMETOOLONG
);
3415 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3421 seq
= read_seqcount_begin(&fs
->seq
);
3424 } while (read_seqcount_retry(&fs
->seq
, seq
));
3428 * NOTE! The user-level library version returns a
3429 * character pointer. The kernel system call just
3430 * returns the length of the buffer filled (which
3431 * includes the ending '\0' character), or a negative
3432 * error value. So libc would do something like
3434 * char *getcwd(char * buf, size_t size)
3438 * retval = sys_getcwd(buf, size);
3445 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3448 struct path pwd
, root
;
3449 char *page
= __getname();
3455 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3458 if (!d_unlinked(pwd
.dentry
)) {
3460 char *cwd
= page
+ PATH_MAX
;
3461 int buflen
= PATH_MAX
;
3463 prepend(&cwd
, &buflen
, "\0", 1);
3464 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3470 /* Unreachable from current root */
3472 error
= prepend_unreachable(&cwd
, &buflen
);
3478 len
= PATH_MAX
+ page
- cwd
;
3481 if (copy_to_user(buf
, cwd
, len
))
3494 * Test whether new_dentry is a subdirectory of old_dentry.
3496 * Trivially implemented using the dcache structure
3500 * is_subdir - is new dentry a subdirectory of old_dentry
3501 * @new_dentry: new dentry
3502 * @old_dentry: old dentry
3504 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3505 * Returns false otherwise.
3506 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3509 bool is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3514 if (new_dentry
== old_dentry
)
3518 /* for restarting inner loop in case of seq retry */
3519 seq
= read_seqbegin(&rename_lock
);
3521 * Need rcu_readlock to protect against the d_parent trashing
3525 if (d_ancestor(old_dentry
, new_dentry
))
3530 } while (read_seqretry(&rename_lock
, seq
));
3535 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3537 struct dentry
*root
= data
;
3538 if (dentry
!= root
) {
3539 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3542 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3543 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3544 dentry
->d_lockref
.count
--;
3547 return D_WALK_CONTINUE
;
3550 void d_genocide(struct dentry
*parent
)
3552 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3555 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3557 inode_dec_link_count(inode
);
3558 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3559 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3560 !d_unlinked(dentry
));
3561 spin_lock(&dentry
->d_parent
->d_lock
);
3562 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3563 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3564 (unsigned long long)inode
->i_ino
);
3565 spin_unlock(&dentry
->d_lock
);
3566 spin_unlock(&dentry
->d_parent
->d_lock
);
3567 d_instantiate(dentry
, inode
);
3569 EXPORT_SYMBOL(d_tmpfile
);
3571 static __initdata
unsigned long dhash_entries
;
3572 static int __init
set_dhash_entries(char *str
)
3576 dhash_entries
= simple_strtoul(str
, &str
, 0);
3579 __setup("dhash_entries=", set_dhash_entries
);
3581 static void __init
dcache_init_early(void)
3583 /* If hashes are distributed across NUMA nodes, defer
3584 * hash allocation until vmalloc space is available.
3590 alloc_large_system_hash("Dentry cache",
3591 sizeof(struct hlist_bl_head
),
3594 HASH_EARLY
| HASH_ZERO
,
3601 static void __init
dcache_init(void)
3604 * A constructor could be added for stable state like the lists,
3605 * but it is probably not worth it because of the cache nature
3608 dentry_cache
= KMEM_CACHE(dentry
,
3609 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
|SLAB_ACCOUNT
);
3611 /* Hash may have been set up in dcache_init_early */
3616 alloc_large_system_hash("Dentry cache",
3617 sizeof(struct hlist_bl_head
),
3627 /* SLAB cache for __getname() consumers */
3628 struct kmem_cache
*names_cachep __read_mostly
;
3629 EXPORT_SYMBOL(names_cachep
);
3631 EXPORT_SYMBOL(d_genocide
);
3633 void __init
vfs_caches_init_early(void)
3637 for (i
= 0; i
< ARRAY_SIZE(in_lookup_hashtable
); i
++)
3638 INIT_HLIST_BL_HEAD(&in_lookup_hashtable
[i
]);
3640 dcache_init_early();
3644 void __init
vfs_caches_init(void)
3646 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3647 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3652 files_maxfiles_init();