4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
41 #include <linux/kasan.h>
45 #ifdef CONFIG_RKP_NS_PROT
51 * dcache->d_inode->i_lock protects:
52 * - i_dentry, d_u.d_alias, d_inode of aliases
53 * dcache_hash_bucket lock protects:
54 * - the dcache hash table
55 * s_anon bl list spinlock protects:
56 * - the s_anon list (see __d_drop)
57 * dentry->d_sb->s_dentry_lru_lock protects:
58 * - the dcache lru lists and counters
65 * - d_parent and d_subdirs
66 * - childrens' d_child and d_parent
67 * - d_u.d_alias, d_inode
70 * dentry->d_inode->i_lock
72 * dentry->d_sb->s_dentry_lru_lock
73 * dcache_hash_bucket lock
76 * If there is an ancestor relationship:
77 * dentry->d_parent->...->d_parent->d_lock
79 * dentry->d_parent->d_lock
82 * If no ancestor relationship:
83 * if (dentry1 < dentry2)
87 int sysctl_vfs_cache_pressure __read_mostly
= 100;
88 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure
);
90 __cacheline_aligned_in_smp
DEFINE_SEQLOCK(rename_lock
);
92 EXPORT_SYMBOL(rename_lock
);
94 static struct kmem_cache
*dentry_cache __read_mostly
;
97 * This is the single most critical data structure when it comes
98 * to the dcache: the hashtable for lookups. Somebody should try
99 * to make this good - I've just made it work.
101 * This hash-function tries to avoid losing too many bits of hash
102 * information, yet avoid using a prime hash-size or similar.
105 static unsigned int d_hash_mask __read_mostly
;
106 static unsigned int d_hash_shift __read_mostly
;
108 static struct hlist_bl_head
*dentry_hashtable __read_mostly
;
110 static inline struct hlist_bl_head
*d_hash(const struct dentry
*parent
,
113 hash
+= (unsigned long) parent
/ L1_CACHE_BYTES
;
114 return dentry_hashtable
+ hash_32(hash
, d_hash_shift
);
117 /* Statistics gathering. */
118 struct dentry_stat_t dentry_stat
= {
122 static DEFINE_PER_CPU(long, nr_dentry
);
123 static DEFINE_PER_CPU(long, nr_dentry_unused
);
125 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
128 * Here we resort to our own counters instead of using generic per-cpu counters
129 * for consistency with what the vfs inode code does. We are expected to harvest
130 * better code and performance by having our own specialized counters.
132 * Please note that the loop is done over all possible CPUs, not over all online
133 * CPUs. The reason for this is that we don't want to play games with CPUs going
134 * on and off. If one of them goes off, we will just keep their counters.
136 * glommer: See cffbc8a for details, and if you ever intend to change this,
137 * please update all vfs counters to match.
139 static long get_nr_dentry(void)
143 for_each_possible_cpu(i
)
144 sum
+= per_cpu(nr_dentry
, i
);
145 return sum
< 0 ? 0 : sum
;
148 static long get_nr_dentry_unused(void)
152 for_each_possible_cpu(i
)
153 sum
+= per_cpu(nr_dentry_unused
, i
);
154 return sum
< 0 ? 0 : sum
;
157 int proc_nr_dentry(struct ctl_table
*table
, int write
, void __user
*buffer
,
158 size_t *lenp
, loff_t
*ppos
)
160 dentry_stat
.nr_dentry
= get_nr_dentry();
161 dentry_stat
.nr_unused
= get_nr_dentry_unused();
162 return proc_doulongvec_minmax(table
, write
, buffer
, lenp
, ppos
);
167 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
168 * The strings are both count bytes long, and count is non-zero.
170 #ifdef CONFIG_DCACHE_WORD_ACCESS
172 #include <asm/word-at-a-time.h>
174 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
175 * aligned allocation for this particular component. We don't
176 * strictly need the load_unaligned_zeropad() safety, but it
177 * doesn't hurt either.
179 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
180 * need the careful unaligned handling.
182 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
184 unsigned long a
,b
,mask
;
187 a
= *(unsigned long *)cs
;
188 b
= load_unaligned_zeropad(ct
);
189 if (tcount
< sizeof(unsigned long))
191 if (unlikely(a
!= b
))
193 cs
+= sizeof(unsigned long);
194 ct
+= sizeof(unsigned long);
195 tcount
-= sizeof(unsigned long);
199 mask
= bytemask_from_count(tcount
);
200 return unlikely(!!((a
^ b
) & mask
));
205 static inline int dentry_string_cmp(const unsigned char *cs
, const unsigned char *ct
, unsigned tcount
)
219 static inline int dentry_cmp(const struct dentry
*dentry
, const unsigned char *ct
, unsigned tcount
)
221 const unsigned char *cs
;
223 * Be careful about RCU walk racing with rename:
224 * use ACCESS_ONCE to fetch the name pointer.
226 * NOTE! Even if a rename will mean that the length
227 * was not loaded atomically, we don't care. The
228 * RCU walk will check the sequence count eventually,
229 * and catch it. And we won't overrun the buffer,
230 * because we're reading the name pointer atomically,
231 * and a dentry name is guaranteed to be properly
232 * terminated with a NUL byte.
234 * End result: even if 'len' is wrong, we'll exit
235 * early because the data cannot match (there can
236 * be no NUL in the ct/tcount data)
238 cs
= ACCESS_ONCE(dentry
->d_name
.name
);
239 smp_read_barrier_depends();
240 return dentry_string_cmp(cs
, ct
, tcount
);
243 struct external_name
{
246 struct rcu_head head
;
248 unsigned char name
[];
251 static inline struct external_name
*external_name(struct dentry
*dentry
)
253 return container_of(dentry
->d_name
.name
, struct external_name
, name
[0]);
256 static void __d_free(struct rcu_head
*head
)
258 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
260 kmem_cache_free(dentry_cache
, dentry
);
263 static void __d_free_external(struct rcu_head
*head
)
265 struct dentry
*dentry
= container_of(head
, struct dentry
, d_u
.d_rcu
);
266 kfree(external_name(dentry
));
267 kmem_cache_free(dentry_cache
, dentry
);
270 static inline int dname_external(const struct dentry
*dentry
)
272 return dentry
->d_name
.name
!= dentry
->d_iname
;
275 void take_dentry_name_snapshot(struct name_snapshot
*name
, struct dentry
*dentry
)
277 spin_lock(&dentry
->d_lock
);
278 if (unlikely(dname_external(dentry
))) {
279 struct external_name
*p
= external_name(dentry
);
280 atomic_inc(&p
->u
.count
);
281 spin_unlock(&dentry
->d_lock
);
282 name
->name
= p
->name
;
284 memcpy(name
->inline_name
, dentry
->d_iname
, DNAME_INLINE_LEN
);
285 spin_unlock(&dentry
->d_lock
);
286 name
->name
= name
->inline_name
;
289 EXPORT_SYMBOL(take_dentry_name_snapshot
);
291 void release_dentry_name_snapshot(struct name_snapshot
*name
)
293 if (unlikely(name
->name
!= name
->inline_name
)) {
294 struct external_name
*p
;
295 p
= container_of(name
->name
, struct external_name
, name
[0]);
296 if (unlikely(atomic_dec_and_test(&p
->u
.count
)))
297 kfree_rcu(p
, u
.head
);
300 EXPORT_SYMBOL(release_dentry_name_snapshot
);
302 static inline void __d_set_inode_and_type(struct dentry
*dentry
,
308 dentry
->d_inode
= inode
;
309 flags
= READ_ONCE(dentry
->d_flags
);
310 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
312 WRITE_ONCE(dentry
->d_flags
, flags
);
315 static inline void __d_clear_type_and_inode(struct dentry
*dentry
)
317 unsigned flags
= READ_ONCE(dentry
->d_flags
);
319 flags
&= ~(DCACHE_ENTRY_TYPE
| DCACHE_FALLTHRU
);
320 WRITE_ONCE(dentry
->d_flags
, flags
);
321 dentry
->d_inode
= NULL
;
324 static void dentry_free(struct dentry
*dentry
)
326 WARN_ON(!hlist_unhashed(&dentry
->d_u
.d_alias
));
327 if (unlikely(dname_external(dentry
))) {
328 struct external_name
*p
= external_name(dentry
);
329 if (likely(atomic_dec_and_test(&p
->u
.count
))) {
330 call_rcu(&dentry
->d_u
.d_rcu
, __d_free_external
);
334 /* if dentry was never visible to RCU, immediate free is OK */
335 if (!(dentry
->d_flags
& DCACHE_RCUACCESS
))
336 __d_free(&dentry
->d_u
.d_rcu
);
338 call_rcu(&dentry
->d_u
.d_rcu
, __d_free
);
342 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
343 * @dentry: the target dentry
344 * After this call, in-progress rcu-walk path lookup will fail. This
345 * should be called after unhashing, and after changing d_inode (if
346 * the dentry has not already been unhashed).
348 static inline void dentry_rcuwalk_invalidate(struct dentry
*dentry
)
350 lockdep_assert_held(&dentry
->d_lock
);
351 /* Go through am invalidation barrier */
352 write_seqcount_invalidate(&dentry
->d_seq
);
356 * Release the dentry's inode, using the filesystem
357 * d_iput() operation if defined. Dentry has no refcount
360 static void dentry_iput(struct dentry
* dentry
)
361 __releases(dentry
->d_lock
)
362 __releases(dentry
->d_inode
->i_lock
)
364 struct inode
*inode
= dentry
->d_inode
;
366 __d_clear_type_and_inode(dentry
);
367 hlist_del_init(&dentry
->d_u
.d_alias
);
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
);
377 spin_unlock(&dentry
->d_lock
);
382 * Release the dentry's inode, using the filesystem
383 * d_iput() operation if defined. dentry remains in-use.
385 static void dentry_unlink_inode(struct dentry
* dentry
)
386 __releases(dentry
->d_lock
)
387 __releases(dentry
->d_inode
->i_lock
)
389 struct inode
*inode
= dentry
->d_inode
;
391 raw_write_seqcount_begin(&dentry
->d_seq
);
392 __d_clear_type_and_inode(dentry
);
393 hlist_del_init(&dentry
->d_u
.d_alias
);
394 raw_write_seqcount_end(&dentry
->d_seq
);
395 spin_unlock(&dentry
->d_lock
);
396 spin_unlock(&inode
->i_lock
);
398 fsnotify_inoderemove(inode
);
399 if (dentry
->d_op
&& dentry
->d_op
->d_iput
)
400 dentry
->d_op
->d_iput(dentry
, inode
);
406 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
407 * is in use - which includes both the "real" per-superblock
408 * LRU list _and_ the DCACHE_SHRINK_LIST use.
410 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
411 * on the shrink list (ie not on the superblock LRU list).
413 * The per-cpu "nr_dentry_unused" counters are updated with
414 * the DCACHE_LRU_LIST bit.
416 * These helper functions make sure we always follow the
417 * rules. d_lock must be held by the caller.
419 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
420 static void d_lru_add(struct dentry
*dentry
)
422 D_FLAG_VERIFY(dentry
, 0);
423 dentry
->d_flags
|= DCACHE_LRU_LIST
;
424 this_cpu_inc(nr_dentry_unused
);
425 WARN_ON_ONCE(!list_lru_add(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
428 static void d_lru_del(struct dentry
*dentry
)
430 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
431 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
432 this_cpu_dec(nr_dentry_unused
);
433 WARN_ON_ONCE(!list_lru_del(&dentry
->d_sb
->s_dentry_lru
, &dentry
->d_lru
));
436 static void d_shrink_del(struct dentry
*dentry
)
438 D_FLAG_VERIFY(dentry
, DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
439 list_del_init(&dentry
->d_lru
);
440 dentry
->d_flags
&= ~(DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
);
441 this_cpu_dec(nr_dentry_unused
);
444 static void d_shrink_add(struct dentry
*dentry
, struct list_head
*list
)
446 D_FLAG_VERIFY(dentry
, 0);
447 list_add(&dentry
->d_lru
, list
);
448 dentry
->d_flags
|= DCACHE_SHRINK_LIST
| DCACHE_LRU_LIST
;
449 this_cpu_inc(nr_dentry_unused
);
453 * These can only be called under the global LRU lock, ie during the
454 * callback for freeing the LRU list. "isolate" removes it from the
455 * LRU lists entirely, while shrink_move moves it to the indicated
458 static void d_lru_isolate(struct list_lru_one
*lru
, struct dentry
*dentry
)
460 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
461 dentry
->d_flags
&= ~DCACHE_LRU_LIST
;
462 this_cpu_dec(nr_dentry_unused
);
463 list_lru_isolate(lru
, &dentry
->d_lru
);
466 static void d_lru_shrink_move(struct list_lru_one
*lru
, struct dentry
*dentry
,
467 struct list_head
*list
)
469 D_FLAG_VERIFY(dentry
, DCACHE_LRU_LIST
);
470 dentry
->d_flags
|= DCACHE_SHRINK_LIST
;
471 list_lru_isolate_move(lru
, &dentry
->d_lru
, list
);
475 * dentry_lru_(add|del)_list) must be called with d_lock held.
477 static void dentry_lru_add(struct dentry
*dentry
)
479 if (unlikely(!(dentry
->d_flags
& DCACHE_LRU_LIST
)))
484 * d_drop - drop a dentry
485 * @dentry: dentry to drop
487 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
488 * be found through a VFS lookup any more. Note that this is different from
489 * deleting the dentry - d_delete will try to mark the dentry negative if
490 * possible, giving a successful _negative_ lookup, while d_drop will
491 * just make the cache lookup fail.
493 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
494 * reason (NFS timeouts or autofs deletes).
496 * __d_drop requires dentry->d_lock.
498 void __d_drop(struct dentry
*dentry
)
500 if (!d_unhashed(dentry
)) {
501 struct hlist_bl_head
*b
;
503 * Hashed dentries are normally on the dentry hashtable,
504 * with the exception of those newly allocated by
505 * d_obtain_alias, which are always IS_ROOT:
507 if (unlikely(IS_ROOT(dentry
)))
508 b
= &dentry
->d_sb
->s_anon
;
510 b
= d_hash(dentry
->d_parent
, dentry
->d_name
.hash
);
513 __hlist_bl_del(&dentry
->d_hash
);
514 dentry
->d_hash
.pprev
= NULL
;
516 dentry_rcuwalk_invalidate(dentry
);
519 EXPORT_SYMBOL(__d_drop
);
521 void d_drop(struct dentry
*dentry
)
523 spin_lock(&dentry
->d_lock
);
525 spin_unlock(&dentry
->d_lock
);
527 EXPORT_SYMBOL(d_drop
);
529 static void __dentry_kill(struct dentry
*dentry
)
531 struct dentry
*parent
= NULL
;
532 bool can_free
= true;
533 if (!IS_ROOT(dentry
))
534 parent
= dentry
->d_parent
;
537 * The dentry is now unrecoverably dead to the world.
539 lockref_mark_dead(&dentry
->d_lockref
);
542 * inform the fs via d_prune that this dentry is about to be
543 * unhashed and destroyed.
545 if (dentry
->d_flags
& DCACHE_OP_PRUNE
)
546 dentry
->d_op
->d_prune(dentry
);
548 if (dentry
->d_flags
& DCACHE_LRU_LIST
) {
549 if (!(dentry
->d_flags
& DCACHE_SHRINK_LIST
))
552 /* if it was on the hash then remove it */
554 __list_del_entry(&dentry
->d_child
);
556 * Inform d_walk() that we are no longer attached to the
559 dentry
->d_flags
|= DCACHE_DENTRY_KILLED
;
561 spin_unlock(&parent
->d_lock
);
564 * dentry_iput drops the locks, at which point nobody (except
565 * transient RCU lookups) can reach this dentry.
567 BUG_ON(dentry
->d_lockref
.count
> 0);
568 this_cpu_dec(nr_dentry
);
569 if (dentry
->d_op
&& dentry
->d_op
->d_release
)
570 dentry
->d_op
->d_release(dentry
);
572 spin_lock(&dentry
->d_lock
);
573 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
574 dentry
->d_flags
|= DCACHE_MAY_FREE
;
577 spin_unlock(&dentry
->d_lock
);
578 if (likely(can_free
))
583 * Finish off a dentry we've decided to kill.
584 * dentry->d_lock must be held, returns with it unlocked.
585 * If ref is non-zero, then decrement the refcount too.
586 * Returns dentry requiring refcount drop, or NULL if we're done.
588 static struct dentry
*dentry_kill(struct dentry
*dentry
)
589 __releases(dentry
->d_lock
)
591 struct inode
*inode
= dentry
->d_inode
;
592 struct dentry
*parent
= NULL
;
594 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
)))
597 if (!IS_ROOT(dentry
)) {
598 parent
= dentry
->d_parent
;
599 if (unlikely(!spin_trylock(&parent
->d_lock
))) {
601 spin_unlock(&inode
->i_lock
);
606 __dentry_kill(dentry
);
610 spin_unlock(&dentry
->d_lock
);
611 return dentry
; /* try again with same dentry */
614 static inline struct dentry
*lock_parent(struct dentry
*dentry
)
616 struct dentry
*parent
= dentry
->d_parent
;
619 if (unlikely(dentry
->d_lockref
.count
< 0))
621 if (likely(spin_trylock(&parent
->d_lock
)))
624 spin_unlock(&dentry
->d_lock
);
626 parent
= ACCESS_ONCE(dentry
->d_parent
);
627 spin_lock(&parent
->d_lock
);
629 * We can't blindly lock dentry until we are sure
630 * that we won't violate the locking order.
631 * Any changes of dentry->d_parent must have
632 * been done with parent->d_lock held, so
633 * spin_lock() above is enough of a barrier
634 * for checking if it's still our child.
636 if (unlikely(parent
!= dentry
->d_parent
)) {
637 spin_unlock(&parent
->d_lock
);
641 if (parent
!= dentry
)
642 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
649 * Try to do a lockless dput(), and return whether that was successful.
651 * If unsuccessful, we return false, having already taken the dentry lock.
653 * The caller needs to hold the RCU read lock, so that the dentry is
654 * guaranteed to stay around even if the refcount goes down to zero!
656 static inline bool fast_dput(struct dentry
*dentry
)
659 unsigned int d_flags
;
662 * If we have a d_op->d_delete() operation, we sould not
663 * let the dentry count go to zero, so use "put_or_lock".
665 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
))
666 return lockref_put_or_lock(&dentry
->d_lockref
);
669 * .. otherwise, we can try to just decrement the
670 * lockref optimistically.
672 ret
= lockref_put_return(&dentry
->d_lockref
);
675 * If the lockref_put_return() failed due to the lock being held
676 * by somebody else, the fast path has failed. We will need to
677 * get the lock, and then check the count again.
679 if (unlikely(ret
< 0)) {
680 spin_lock(&dentry
->d_lock
);
681 if (dentry
->d_lockref
.count
> 1) {
682 dentry
->d_lockref
.count
--;
683 spin_unlock(&dentry
->d_lock
);
690 * If we weren't the last ref, we're done.
696 * Careful, careful. The reference count went down
697 * to zero, but we don't hold the dentry lock, so
698 * somebody else could get it again, and do another
699 * dput(), and we need to not race with that.
701 * However, there is a very special and common case
702 * where we don't care, because there is nothing to
703 * do: the dentry is still hashed, it does not have
704 * a 'delete' op, and it's referenced and already on
707 * NOTE! Since we aren't locked, these values are
708 * not "stable". However, it is sufficient that at
709 * some point after we dropped the reference the
710 * dentry was hashed and the flags had the proper
711 * value. Other dentry users may have re-gotten
712 * a reference to the dentry and change that, but
713 * our work is done - we can leave the dentry
714 * around with a zero refcount.
717 d_flags
= ACCESS_ONCE(dentry
->d_flags
);
718 d_flags
&= DCACHE_REFERENCED
| DCACHE_LRU_LIST
| DCACHE_DISCONNECTED
;
720 /* Nothing to do? Dropping the reference was all we needed? */
721 if (d_flags
== (DCACHE_REFERENCED
| DCACHE_LRU_LIST
) && !d_unhashed(dentry
))
725 * Not the fast normal case? Get the lock. We've already decremented
726 * the refcount, but we'll need to re-check the situation after
729 spin_lock(&dentry
->d_lock
);
732 * Did somebody else grab a reference to it in the meantime, and
733 * we're no longer the last user after all? Alternatively, somebody
734 * else could have killed it and marked it dead. Either way, we
735 * don't need to do anything else.
737 if (dentry
->d_lockref
.count
) {
738 spin_unlock(&dentry
->d_lock
);
743 * Re-get the reference we optimistically dropped. We hold the
744 * lock, and we just tested that it was zero, so we can just
747 dentry
->d_lockref
.count
= 1;
755 * This is complicated by the fact that we do not want to put
756 * dentries that are no longer on any hash chain on the unused
757 * list: we'd much rather just get rid of them immediately.
759 * However, that implies that we have to traverse the dentry
760 * tree upwards to the parents which might _also_ now be
761 * scheduled for deletion (it may have been only waiting for
762 * its last child to go away).
764 * This tail recursion is done by hand as we don't want to depend
765 * on the compiler to always get this right (gcc generally doesn't).
766 * Real recursion would eat up our stack space.
770 * dput - release a dentry
771 * @dentry: dentry to release
773 * Release a dentry. This will drop the usage count and if appropriate
774 * call the dentry unlink method as well as removing it from the queues and
775 * releasing its resources. If the parent dentries were scheduled for release
776 * they too may now get deleted.
778 void dput(struct dentry
*dentry
)
780 if (unlikely(!dentry
))
787 if (likely(fast_dput(dentry
))) {
792 /* Slow case: now with the dentry lock held */
795 /* Unreachable? Get rid of it */
796 if (unlikely(d_unhashed(dentry
)))
799 if (unlikely(dentry
->d_flags
& DCACHE_DISCONNECTED
))
802 if (unlikely(dentry
->d_flags
& DCACHE_OP_DELETE
)) {
803 if (dentry
->d_op
->d_delete(dentry
))
807 if (!(dentry
->d_flags
& DCACHE_REFERENCED
))
808 dentry
->d_flags
|= DCACHE_REFERENCED
;
809 dentry_lru_add(dentry
);
811 dentry
->d_lockref
.count
--;
812 spin_unlock(&dentry
->d_lock
);
816 dentry
= dentry_kill(dentry
);
825 /* This must be called with d_lock held */
826 static inline void __dget_dlock(struct dentry
*dentry
)
828 dentry
->d_lockref
.count
++;
831 static inline void __dget(struct dentry
*dentry
)
833 lockref_get(&dentry
->d_lockref
);
836 struct dentry
*dget_parent(struct dentry
*dentry
)
842 * Do optimistic parent lookup without any
846 ret
= ACCESS_ONCE(dentry
->d_parent
);
847 gotref
= lockref_get_not_zero(&ret
->d_lockref
);
849 if (likely(gotref
)) {
850 if (likely(ret
== ACCESS_ONCE(dentry
->d_parent
)))
857 * Don't need rcu_dereference because we re-check it was correct under
861 ret
= dentry
->d_parent
;
862 spin_lock(&ret
->d_lock
);
863 if (unlikely(ret
!= dentry
->d_parent
)) {
864 spin_unlock(&ret
->d_lock
);
869 BUG_ON(!ret
->d_lockref
.count
);
870 ret
->d_lockref
.count
++;
871 spin_unlock(&ret
->d_lock
);
874 EXPORT_SYMBOL(dget_parent
);
877 * d_find_alias - grab a hashed alias of inode
878 * @inode: inode in question
880 * If inode has a hashed alias, or is a directory and has any alias,
881 * acquire the reference to alias and return it. Otherwise return NULL.
882 * Notice that if inode is a directory there can be only one alias and
883 * it can be unhashed only if it has no children, or if it is the root
884 * of a filesystem, or if the directory was renamed and d_revalidate
885 * was the first vfs operation to notice.
887 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
888 * any other hashed alias over that one.
890 static struct dentry
*__d_find_alias(struct inode
*inode
)
892 struct dentry
*alias
, *discon_alias
;
896 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
897 spin_lock(&alias
->d_lock
);
898 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
899 if (IS_ROOT(alias
) &&
900 (alias
->d_flags
& DCACHE_DISCONNECTED
)) {
901 discon_alias
= alias
;
904 spin_unlock(&alias
->d_lock
);
908 spin_unlock(&alias
->d_lock
);
911 alias
= discon_alias
;
912 spin_lock(&alias
->d_lock
);
913 if (S_ISDIR(inode
->i_mode
) || !d_unhashed(alias
)) {
915 spin_unlock(&alias
->d_lock
);
918 spin_unlock(&alias
->d_lock
);
924 struct dentry
*d_find_alias(struct inode
*inode
)
926 struct dentry
*de
= NULL
;
928 if (!hlist_empty(&inode
->i_dentry
)) {
929 spin_lock(&inode
->i_lock
);
930 de
= __d_find_alias(inode
);
931 spin_unlock(&inode
->i_lock
);
935 EXPORT_SYMBOL(d_find_alias
);
938 * Try to kill dentries associated with this inode.
939 * WARNING: you must own a reference to inode.
941 void d_prune_aliases(struct inode
*inode
)
943 struct dentry
*dentry
;
945 spin_lock(&inode
->i_lock
);
946 hlist_for_each_entry(dentry
, &inode
->i_dentry
, d_u
.d_alias
) {
947 spin_lock(&dentry
->d_lock
);
948 if (!dentry
->d_lockref
.count
) {
949 struct dentry
*parent
= lock_parent(dentry
);
950 if (likely(!dentry
->d_lockref
.count
)) {
951 __dentry_kill(dentry
);
956 spin_unlock(&parent
->d_lock
);
958 spin_unlock(&dentry
->d_lock
);
960 spin_unlock(&inode
->i_lock
);
962 EXPORT_SYMBOL(d_prune_aliases
);
964 static void shrink_dentry_list(struct list_head
*list
)
966 struct dentry
*dentry
, *parent
;
968 while (!list_empty(list
)) {
970 dentry
= list_entry(list
->prev
, struct dentry
, d_lru
);
971 spin_lock(&dentry
->d_lock
);
972 parent
= lock_parent(dentry
);
975 * The dispose list is isolated and dentries are not accounted
976 * to the LRU here, so we can simply remove it from the list
977 * here regardless of whether it is referenced or not.
979 d_shrink_del(dentry
);
982 * We found an inuse dentry which was not removed from
983 * the LRU because of laziness during lookup. Do not free it.
985 if (dentry
->d_lockref
.count
> 0) {
986 spin_unlock(&dentry
->d_lock
);
988 spin_unlock(&parent
->d_lock
);
993 if (unlikely(dentry
->d_flags
& DCACHE_DENTRY_KILLED
)) {
994 bool can_free
= dentry
->d_flags
& DCACHE_MAY_FREE
;
995 spin_unlock(&dentry
->d_lock
);
997 spin_unlock(&parent
->d_lock
);
1003 inode
= dentry
->d_inode
;
1004 if (inode
&& unlikely(!spin_trylock(&inode
->i_lock
))) {
1005 d_shrink_add(dentry
, list
);
1006 spin_unlock(&dentry
->d_lock
);
1008 spin_unlock(&parent
->d_lock
);
1012 __dentry_kill(dentry
);
1015 * We need to prune ancestors too. This is necessary to prevent
1016 * quadratic behavior of shrink_dcache_parent(), but is also
1017 * expected to be beneficial in reducing dentry cache
1021 while (dentry
&& !lockref_put_or_lock(&dentry
->d_lockref
)) {
1022 parent
= lock_parent(dentry
);
1023 if (dentry
->d_lockref
.count
!= 1) {
1024 dentry
->d_lockref
.count
--;
1025 spin_unlock(&dentry
->d_lock
);
1027 spin_unlock(&parent
->d_lock
);
1030 inode
= dentry
->d_inode
; /* can't be NULL */
1031 if (unlikely(!spin_trylock(&inode
->i_lock
))) {
1032 spin_unlock(&dentry
->d_lock
);
1034 spin_unlock(&parent
->d_lock
);
1038 __dentry_kill(dentry
);
1044 static enum lru_status
dentry_lru_isolate(struct list_head
*item
,
1045 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1047 struct list_head
*freeable
= arg
;
1048 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1052 * we are inverting the lru lock/dentry->d_lock here,
1053 * so use a trylock. If we fail to get the lock, just skip
1056 if (!spin_trylock(&dentry
->d_lock
))
1060 * Referenced dentries are still in use. If they have active
1061 * counts, just remove them from the LRU. Otherwise give them
1062 * another pass through the LRU.
1064 if (dentry
->d_lockref
.count
) {
1065 d_lru_isolate(lru
, dentry
);
1066 spin_unlock(&dentry
->d_lock
);
1070 if (dentry
->d_flags
& DCACHE_REFERENCED
) {
1071 dentry
->d_flags
&= ~DCACHE_REFERENCED
;
1072 spin_unlock(&dentry
->d_lock
);
1075 * The list move itself will be made by the common LRU code. At
1076 * this point, we've dropped the dentry->d_lock but keep the
1077 * lru lock. This is safe to do, since every list movement is
1078 * protected by the lru lock even if both locks are held.
1080 * This is guaranteed by the fact that all LRU management
1081 * functions are intermediated by the LRU API calls like
1082 * list_lru_add and list_lru_del. List movement in this file
1083 * only ever occur through this functions or through callbacks
1084 * like this one, that are called from the LRU API.
1086 * The only exceptions to this are functions like
1087 * shrink_dentry_list, and code that first checks for the
1088 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1089 * operating only with stack provided lists after they are
1090 * properly isolated from the main list. It is thus, always a
1096 d_lru_shrink_move(lru
, dentry
, freeable
);
1097 spin_unlock(&dentry
->d_lock
);
1103 * prune_dcache_sb - shrink the dcache
1105 * @sc: shrink control, passed to list_lru_shrink_walk()
1107 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1108 * is done when we need more memory and called from the superblock shrinker
1111 * This function may fail to free any resources if all the dentries are in
1114 long prune_dcache_sb(struct super_block
*sb
, struct shrink_control
*sc
)
1119 freed
= list_lru_shrink_walk(&sb
->s_dentry_lru
, sc
,
1120 dentry_lru_isolate
, &dispose
);
1121 shrink_dentry_list(&dispose
);
1125 static enum lru_status
dentry_lru_isolate_shrink(struct list_head
*item
,
1126 struct list_lru_one
*lru
, spinlock_t
*lru_lock
, void *arg
)
1128 struct list_head
*freeable
= arg
;
1129 struct dentry
*dentry
= container_of(item
, struct dentry
, d_lru
);
1132 * we are inverting the lru lock/dentry->d_lock here,
1133 * so use a trylock. If we fail to get the lock, just skip
1136 if (!spin_trylock(&dentry
->d_lock
))
1139 d_lru_shrink_move(lru
, dentry
, freeable
);
1140 spin_unlock(&dentry
->d_lock
);
1147 * shrink_dcache_sb - shrink dcache for a superblock
1150 * Shrink the dcache for the specified super block. This is used to free
1151 * the dcache before unmounting a file system.
1153 void shrink_dcache_sb(struct super_block
*sb
)
1160 freed
= list_lru_walk(&sb
->s_dentry_lru
,
1161 dentry_lru_isolate_shrink
, &dispose
, 1024);
1163 this_cpu_sub(nr_dentry_unused
, freed
);
1164 shrink_dentry_list(&dispose
);
1166 } while (list_lru_count(&sb
->s_dentry_lru
) > 0);
1168 EXPORT_SYMBOL(shrink_dcache_sb
);
1171 * enum d_walk_ret - action to talke during tree walk
1172 * @D_WALK_CONTINUE: contrinue walk
1173 * @D_WALK_QUIT: quit walk
1174 * @D_WALK_NORETRY: quit when retry is needed
1175 * @D_WALK_SKIP: skip this dentry and its children
1185 * d_walk - walk the dentry tree
1186 * @parent: start of walk
1187 * @data: data passed to @enter() and @finish()
1188 * @enter: callback when first entering the dentry
1189 * @finish: callback when successfully finished the walk
1191 * The @enter() and @finish() callbacks are called with d_lock held.
1193 static void d_walk(struct dentry
*parent
, void *data
,
1194 enum d_walk_ret (*enter
)(void *, struct dentry
*),
1195 void (*finish
)(void *))
1197 struct dentry
*this_parent
;
1198 struct list_head
*next
;
1200 enum d_walk_ret ret
;
1204 read_seqbegin_or_lock(&rename_lock
, &seq
);
1205 this_parent
= parent
;
1206 spin_lock(&this_parent
->d_lock
);
1208 ret
= enter(data
, this_parent
);
1210 case D_WALK_CONTINUE
:
1215 case D_WALK_NORETRY
:
1220 next
= this_parent
->d_subdirs
.next
;
1222 while (next
!= &this_parent
->d_subdirs
) {
1223 struct list_head
*tmp
= next
;
1224 struct dentry
*dentry
= list_entry(tmp
, struct dentry
, d_child
);
1227 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
1229 ret
= enter(data
, dentry
);
1231 case D_WALK_CONTINUE
:
1234 spin_unlock(&dentry
->d_lock
);
1236 case D_WALK_NORETRY
:
1240 spin_unlock(&dentry
->d_lock
);
1244 if (!list_empty(&dentry
->d_subdirs
)) {
1245 spin_unlock(&this_parent
->d_lock
);
1246 spin_release(&dentry
->d_lock
.dep_map
, 1, _RET_IP_
);
1247 this_parent
= dentry
;
1248 spin_acquire(&this_parent
->d_lock
.dep_map
, 0, 1, _RET_IP_
);
1251 spin_unlock(&dentry
->d_lock
);
1254 * All done at this level ... ascend and resume the search.
1258 if (this_parent
!= parent
) {
1259 struct dentry
*child
= this_parent
;
1260 this_parent
= child
->d_parent
;
1262 spin_unlock(&child
->d_lock
);
1263 spin_lock(&this_parent
->d_lock
);
1265 /* might go back up the wrong parent if we have had a rename. */
1266 if (need_seqretry(&rename_lock
, seq
))
1268 /* go into the first sibling still alive */
1270 next
= child
->d_child
.next
;
1271 if (next
== &this_parent
->d_subdirs
)
1273 child
= list_entry(next
, struct dentry
, d_child
);
1274 } while (unlikely(child
->d_flags
& DCACHE_DENTRY_KILLED
));
1278 if (need_seqretry(&rename_lock
, seq
))
1285 spin_unlock(&this_parent
->d_lock
);
1286 done_seqretry(&rename_lock
, seq
);
1290 spin_unlock(&this_parent
->d_lock
);
1300 * Search for at least 1 mount point in the dentry's subdirs.
1301 * We descend to the next level whenever the d_subdirs
1302 * list is non-empty and continue searching.
1305 static enum d_walk_ret
check_mount(void *data
, struct dentry
*dentry
)
1308 if (d_mountpoint(dentry
)) {
1312 return D_WALK_CONTINUE
;
1316 * have_submounts - check for mounts over a dentry
1317 * @parent: dentry to check.
1319 * Return true if the parent or its subdirectories contain
1322 int have_submounts(struct dentry
*parent
)
1326 d_walk(parent
, &ret
, check_mount
, NULL
);
1330 EXPORT_SYMBOL(have_submounts
);
1333 * Called by mount code to set a mountpoint and check if the mountpoint is
1334 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1335 * subtree can become unreachable).
1337 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1338 * this reason take rename_lock and d_lock on dentry and ancestors.
1340 int d_set_mounted(struct dentry
*dentry
)
1344 write_seqlock(&rename_lock
);
1345 for (p
= dentry
->d_parent
; !IS_ROOT(p
); p
= p
->d_parent
) {
1346 /* Need exclusion wrt. d_invalidate() */
1347 spin_lock(&p
->d_lock
);
1348 if (unlikely(d_unhashed(p
))) {
1349 spin_unlock(&p
->d_lock
);
1352 spin_unlock(&p
->d_lock
);
1354 spin_lock(&dentry
->d_lock
);
1355 if (!d_unlinked(dentry
)) {
1357 if (!d_mountpoint(dentry
)) {
1358 dentry
->d_flags
|= DCACHE_MOUNTED
;
1362 spin_unlock(&dentry
->d_lock
);
1364 write_sequnlock(&rename_lock
);
1369 * Search the dentry child list of the specified parent,
1370 * and move any unused dentries to the end of the unused
1371 * list for prune_dcache(). We descend to the next level
1372 * whenever the d_subdirs list is non-empty and continue
1375 * It returns zero iff there are no unused children,
1376 * otherwise it returns the number of children moved to
1377 * the end of the unused list. This may not be the total
1378 * number of unused children, because select_parent can
1379 * drop the lock and return early due to latency
1383 struct select_data
{
1384 struct dentry
*start
;
1385 struct list_head dispose
;
1389 static enum d_walk_ret
select_collect(void *_data
, struct dentry
*dentry
)
1391 struct select_data
*data
= _data
;
1392 enum d_walk_ret ret
= D_WALK_CONTINUE
;
1394 if (data
->start
== dentry
)
1397 if (dentry
->d_flags
& DCACHE_SHRINK_LIST
) {
1400 if (dentry
->d_flags
& DCACHE_LRU_LIST
)
1402 if (!dentry
->d_lockref
.count
) {
1403 d_shrink_add(dentry
, &data
->dispose
);
1408 * We can return to the caller if we have found some (this
1409 * ensures forward progress). We'll be coming back to find
1412 if (!list_empty(&data
->dispose
))
1413 ret
= need_resched() ? D_WALK_QUIT
: D_WALK_NORETRY
;
1419 * shrink_dcache_parent - prune dcache
1420 * @parent: parent of entries to prune
1422 * Prune the dcache to remove unused children of the parent dentry.
1424 void shrink_dcache_parent(struct dentry
*parent
)
1427 struct select_data data
;
1429 INIT_LIST_HEAD(&data
.dispose
);
1430 data
.start
= parent
;
1433 d_walk(parent
, &data
, select_collect
, NULL
);
1437 shrink_dentry_list(&data
.dispose
);
1441 EXPORT_SYMBOL(shrink_dcache_parent
);
1443 static enum d_walk_ret
umount_check(void *_data
, struct dentry
*dentry
)
1445 /* it has busy descendents; complain about those instead */
1446 if (!list_empty(&dentry
->d_subdirs
))
1447 return D_WALK_CONTINUE
;
1449 /* root with refcount 1 is fine */
1450 if (dentry
== _data
&& dentry
->d_lockref
.count
== 1)
1451 return D_WALK_CONTINUE
;
1453 printk(KERN_ERR
"BUG: Dentry %p{i=%lx,n=%pd} "
1454 " still in use (%d) [unmount of %s %s]\n",
1457 dentry
->d_inode
->i_ino
: 0UL,
1459 dentry
->d_lockref
.count
,
1460 dentry
->d_sb
->s_type
->name
,
1461 dentry
->d_sb
->s_id
);
1463 return D_WALK_CONTINUE
;
1466 static void do_one_tree(struct dentry
*dentry
)
1468 shrink_dcache_parent(dentry
);
1469 d_walk(dentry
, dentry
, umount_check
, NULL
);
1475 * destroy the dentries attached to a superblock on unmounting
1477 void shrink_dcache_for_umount(struct super_block
*sb
)
1479 struct dentry
*dentry
;
1481 WARN(down_read_trylock(&sb
->s_umount
), "s_umount should've been locked");
1483 dentry
= sb
->s_root
;
1485 do_one_tree(dentry
);
1487 while (!hlist_bl_empty(&sb
->s_anon
)) {
1488 dentry
= dget(hlist_bl_entry(hlist_bl_first(&sb
->s_anon
), struct dentry
, d_hash
));
1489 do_one_tree(dentry
);
1493 struct detach_data
{
1494 struct select_data select
;
1495 struct dentry
*mountpoint
;
1497 static enum d_walk_ret
detach_and_collect(void *_data
, struct dentry
*dentry
)
1499 struct detach_data
*data
= _data
;
1501 if (d_mountpoint(dentry
)) {
1502 __dget_dlock(dentry
);
1503 data
->mountpoint
= dentry
;
1507 return select_collect(&data
->select
, dentry
);
1510 static void check_and_drop(void *_data
)
1512 struct detach_data
*data
= _data
;
1514 if (!data
->mountpoint
&& list_empty(&data
->select
.dispose
))
1515 __d_drop(data
->select
.start
);
1519 * d_invalidate - detach submounts, prune dcache, and drop
1520 * @dentry: dentry to invalidate (aka detach, prune and drop)
1524 * The final d_drop is done as an atomic operation relative to
1525 * rename_lock ensuring there are no races with d_set_mounted. This
1526 * ensures there are no unhashed dentries on the path to a mountpoint.
1528 void d_invalidate(struct dentry
*dentry
)
1531 * If it's already been dropped, return OK.
1533 spin_lock(&dentry
->d_lock
);
1534 if (d_unhashed(dentry
)) {
1535 spin_unlock(&dentry
->d_lock
);
1538 spin_unlock(&dentry
->d_lock
);
1540 /* Negative dentries can be dropped without further checks */
1541 if (!dentry
->d_inode
) {
1547 struct detach_data data
;
1549 data
.mountpoint
= NULL
;
1550 INIT_LIST_HEAD(&data
.select
.dispose
);
1551 data
.select
.start
= dentry
;
1552 data
.select
.found
= 0;
1554 d_walk(dentry
, &data
, detach_and_collect
, check_and_drop
);
1556 if (!list_empty(&data
.select
.dispose
))
1557 shrink_dentry_list(&data
.select
.dispose
);
1558 else if (!data
.mountpoint
)
1561 if (data
.mountpoint
) {
1562 detach_mounts(data
.mountpoint
);
1563 dput(data
.mountpoint
);
1568 EXPORT_SYMBOL(d_invalidate
);
1571 * __d_alloc - allocate a dcache entry
1572 * @sb: filesystem it will belong to
1573 * @name: qstr of the name
1575 * Allocates a dentry. It returns %NULL if there is insufficient memory
1576 * available. On a success the dentry is returned. The name passed in is
1577 * copied and the copy passed in may be reused after this call.
1580 struct dentry
*__d_alloc(struct super_block
*sb
, const struct qstr
*name
)
1582 struct dentry
*dentry
;
1585 dentry
= kmem_cache_alloc(dentry_cache
, GFP_KERNEL
);
1590 * We guarantee that the inline name is always NUL-terminated.
1591 * This way the memcpy() done by the name switching in rename
1592 * will still always have a NUL at the end, even if we might
1593 * be overwriting an internal NUL character
1595 dentry
->d_iname
[DNAME_INLINE_LEN
-1] = 0;
1596 if (name
->len
> DNAME_INLINE_LEN
-1) {
1597 size_t size
= offsetof(struct external_name
, name
[1]);
1598 struct external_name
*p
= kmalloc(size
+ name
->len
, GFP_KERNEL
);
1600 kmem_cache_free(dentry_cache
, dentry
);
1603 atomic_set(&p
->u
.count
, 1);
1605 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS
))
1606 kasan_unpoison_shadow(dname
,
1607 round_up(name
->len
+ 1, sizeof(unsigned long)));
1609 dname
= dentry
->d_iname
;
1612 dentry
->d_name
.len
= name
->len
;
1613 dentry
->d_name
.hash
= name
->hash
;
1614 memcpy(dname
, name
->name
, name
->len
);
1615 dname
[name
->len
] = 0;
1617 /* Make sure we always see the terminating NUL character */
1619 dentry
->d_name
.name
= dname
;
1621 dentry
->d_lockref
.count
= 1;
1622 dentry
->d_flags
= 0;
1623 spin_lock_init(&dentry
->d_lock
);
1624 seqcount_init(&dentry
->d_seq
);
1625 dentry
->d_inode
= NULL
;
1626 dentry
->d_parent
= dentry
;
1628 dentry
->d_op
= NULL
;
1629 dentry
->d_fsdata
= NULL
;
1630 INIT_HLIST_BL_NODE(&dentry
->d_hash
);
1631 INIT_LIST_HEAD(&dentry
->d_lru
);
1632 INIT_LIST_HEAD(&dentry
->d_subdirs
);
1633 INIT_HLIST_NODE(&dentry
->d_u
.d_alias
);
1634 INIT_LIST_HEAD(&dentry
->d_child
);
1635 d_set_d_op(dentry
, dentry
->d_sb
->s_d_op
);
1637 this_cpu_inc(nr_dentry
);
1643 * d_alloc - allocate a dcache entry
1644 * @parent: parent of entry to allocate
1645 * @name: qstr of the name
1647 * Allocates a dentry. It returns %NULL if there is insufficient memory
1648 * available. On a success the dentry is returned. The name passed in is
1649 * copied and the copy passed in may be reused after this call.
1651 struct dentry
*d_alloc(struct dentry
* parent
, const struct qstr
*name
)
1653 struct dentry
*dentry
= __d_alloc(parent
->d_sb
, name
);
1656 dentry
->d_flags
|= DCACHE_RCUACCESS
;
1657 spin_lock(&parent
->d_lock
);
1659 * don't need child lock because it is not subject
1660 * to concurrency here
1662 __dget_dlock(parent
);
1663 dentry
->d_parent
= parent
;
1664 list_add(&dentry
->d_child
, &parent
->d_subdirs
);
1665 spin_unlock(&parent
->d_lock
);
1669 EXPORT_SYMBOL(d_alloc
);
1672 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1673 * @sb: the superblock
1674 * @name: qstr of the name
1676 * For a filesystem that just pins its dentries in memory and never
1677 * performs lookups at all, return an unhashed IS_ROOT dentry.
1679 struct dentry
*d_alloc_pseudo(struct super_block
*sb
, const struct qstr
*name
)
1681 return __d_alloc(sb
, name
);
1683 EXPORT_SYMBOL(d_alloc_pseudo
);
1685 struct dentry
*d_alloc_name(struct dentry
*parent
, const char *name
)
1690 q
.len
= strlen(name
);
1691 q
.hash
= full_name_hash(q
.name
, q
.len
);
1692 return d_alloc(parent
, &q
);
1694 EXPORT_SYMBOL(d_alloc_name
);
1696 void d_set_d_op(struct dentry
*dentry
, const struct dentry_operations
*op
)
1698 WARN_ON_ONCE(dentry
->d_op
);
1699 WARN_ON_ONCE(dentry
->d_flags
& (DCACHE_OP_HASH
|
1701 DCACHE_OP_REVALIDATE
|
1702 DCACHE_OP_WEAK_REVALIDATE
|
1704 DCACHE_OP_SELECT_INODE
|
1710 dentry
->d_flags
|= DCACHE_OP_HASH
;
1712 dentry
->d_flags
|= DCACHE_OP_COMPARE
;
1713 if (op
->d_revalidate
)
1714 dentry
->d_flags
|= DCACHE_OP_REVALIDATE
;
1715 if (op
->d_weak_revalidate
)
1716 dentry
->d_flags
|= DCACHE_OP_WEAK_REVALIDATE
;
1718 dentry
->d_flags
|= DCACHE_OP_DELETE
;
1720 dentry
->d_flags
|= DCACHE_OP_PRUNE
;
1721 if (op
->d_select_inode
)
1722 dentry
->d_flags
|= DCACHE_OP_SELECT_INODE
;
1724 dentry
->d_flags
|= DCACHE_OP_REAL
;
1727 EXPORT_SYMBOL(d_set_d_op
);
1731 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1732 * @dentry - The dentry to mark
1734 * Mark a dentry as falling through to the lower layer (as set with
1735 * d_pin_lower()). This flag may be recorded on the medium.
1737 void d_set_fallthru(struct dentry
*dentry
)
1739 spin_lock(&dentry
->d_lock
);
1740 dentry
->d_flags
|= DCACHE_FALLTHRU
;
1741 spin_unlock(&dentry
->d_lock
);
1743 EXPORT_SYMBOL(d_set_fallthru
);
1745 static unsigned d_flags_for_inode(struct inode
*inode
)
1747 unsigned add_flags
= DCACHE_REGULAR_TYPE
;
1750 return DCACHE_MISS_TYPE
;
1752 if (S_ISDIR(inode
->i_mode
)) {
1753 add_flags
= DCACHE_DIRECTORY_TYPE
;
1754 if (unlikely(!(inode
->i_opflags
& IOP_LOOKUP
))) {
1755 if (unlikely(!inode
->i_op
->lookup
))
1756 add_flags
= DCACHE_AUTODIR_TYPE
;
1758 inode
->i_opflags
|= IOP_LOOKUP
;
1760 goto type_determined
;
1763 if (unlikely(!(inode
->i_opflags
& IOP_NOFOLLOW
))) {
1764 if (unlikely(inode
->i_op
->follow_link
)) {
1765 add_flags
= DCACHE_SYMLINK_TYPE
;
1766 goto type_determined
;
1768 inode
->i_opflags
|= IOP_NOFOLLOW
;
1771 if (unlikely(!S_ISREG(inode
->i_mode
)))
1772 add_flags
= DCACHE_SPECIAL_TYPE
;
1775 if (unlikely(IS_AUTOMOUNT(inode
)))
1776 add_flags
|= DCACHE_NEED_AUTOMOUNT
;
1780 static void __d_instantiate(struct dentry
*dentry
, struct inode
*inode
)
1782 unsigned add_flags
= d_flags_for_inode(inode
);
1784 spin_lock(&dentry
->d_lock
);
1786 hlist_add_head(&dentry
->d_u
.d_alias
, &inode
->i_dentry
);
1787 raw_write_seqcount_begin(&dentry
->d_seq
);
1788 __d_set_inode_and_type(dentry
, inode
, add_flags
);
1789 raw_write_seqcount_end(&dentry
->d_seq
);
1790 spin_unlock(&dentry
->d_lock
);
1791 fsnotify_d_instantiate(dentry
, inode
);
1795 * d_instantiate - fill in inode information for a dentry
1796 * @entry: dentry to complete
1797 * @inode: inode to attach to this dentry
1799 * Fill in inode information in the entry.
1801 * This turns negative dentries into productive full members
1804 * NOTE! This assumes that the inode count has been incremented
1805 * (or otherwise set) by the caller to indicate that it is now
1806 * in use by the dcache.
1809 void d_instantiate(struct dentry
*entry
, struct inode
* inode
)
1811 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1813 spin_lock(&inode
->i_lock
);
1814 __d_instantiate(entry
, inode
);
1816 spin_unlock(&inode
->i_lock
);
1817 security_d_instantiate(entry
, inode
);
1819 EXPORT_SYMBOL(d_instantiate
);
1822 * d_instantiate_unique - instantiate a non-aliased dentry
1823 * @entry: dentry to instantiate
1824 * @inode: inode to attach to this dentry
1826 * Fill in inode information in the entry. On success, it returns NULL.
1827 * If an unhashed alias of "entry" already exists, then we return the
1828 * aliased dentry instead and drop one reference to inode.
1830 * Note that in order to avoid conflicts with rename() etc, the caller
1831 * had better be holding the parent directory semaphore.
1833 * This also assumes that the inode count has been incremented
1834 * (or otherwise set) by the caller to indicate that it is now
1835 * in use by the dcache.
1837 static struct dentry
*__d_instantiate_unique(struct dentry
*entry
,
1838 struct inode
*inode
)
1840 struct dentry
*alias
;
1841 int len
= entry
->d_name
.len
;
1842 const char *name
= entry
->d_name
.name
;
1843 unsigned int hash
= entry
->d_name
.hash
;
1846 __d_instantiate(entry
, NULL
);
1850 hlist_for_each_entry(alias
, &inode
->i_dentry
, d_u
.d_alias
) {
1852 * Don't need alias->d_lock here, because aliases with
1853 * d_parent == entry->d_parent are not subject to name or
1854 * parent changes, because the parent inode i_mutex is held.
1856 if (alias
->d_name
.hash
!= hash
)
1858 if (alias
->d_parent
!= entry
->d_parent
)
1860 if (alias
->d_name
.len
!= len
)
1862 if (dentry_cmp(alias
, name
, len
))
1868 __d_instantiate(entry
, inode
);
1872 struct dentry
*d_instantiate_unique(struct dentry
*entry
, struct inode
*inode
)
1874 struct dentry
*result
;
1876 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1879 spin_lock(&inode
->i_lock
);
1880 result
= __d_instantiate_unique(entry
, inode
);
1882 spin_unlock(&inode
->i_lock
);
1885 security_d_instantiate(entry
, inode
);
1889 BUG_ON(!d_unhashed(result
));
1894 EXPORT_SYMBOL(d_instantiate_unique
);
1897 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1898 * @entry: dentry to complete
1899 * @inode: inode to attach to this dentry
1901 * Fill in inode information in the entry. If a directory alias is found, then
1902 * return an error (and drop inode). Together with d_materialise_unique() this
1903 * guarantees that a directory inode may never have more than one alias.
1905 int d_instantiate_no_diralias(struct dentry
*entry
, struct inode
*inode
)
1907 BUG_ON(!hlist_unhashed(&entry
->d_u
.d_alias
));
1909 spin_lock(&inode
->i_lock
);
1910 if (S_ISDIR(inode
->i_mode
) && !hlist_empty(&inode
->i_dentry
)) {
1911 spin_unlock(&inode
->i_lock
);
1915 __d_instantiate(entry
, inode
);
1916 spin_unlock(&inode
->i_lock
);
1917 security_d_instantiate(entry
, inode
);
1921 EXPORT_SYMBOL(d_instantiate_no_diralias
);
1923 struct dentry
*d_make_root(struct inode
*root_inode
)
1925 struct dentry
*res
= NULL
;
1928 static const struct qstr name
= QSTR_INIT("/", 1);
1930 res
= __d_alloc(root_inode
->i_sb
, &name
);
1932 d_instantiate(res
, root_inode
);
1938 EXPORT_SYMBOL(d_make_root
);
1940 static struct dentry
* __d_find_any_alias(struct inode
*inode
)
1942 struct dentry
*alias
;
1944 if (hlist_empty(&inode
->i_dentry
))
1946 alias
= hlist_entry(inode
->i_dentry
.first
, struct dentry
, d_u
.d_alias
);
1952 * d_find_any_alias - find any alias for a given inode
1953 * @inode: inode to find an alias for
1955 * If any aliases exist for the given inode, take and return a
1956 * reference for one of them. If no aliases exist, return %NULL.
1958 struct dentry
*d_find_any_alias(struct inode
*inode
)
1962 spin_lock(&inode
->i_lock
);
1963 de
= __d_find_any_alias(inode
);
1964 spin_unlock(&inode
->i_lock
);
1967 EXPORT_SYMBOL(d_find_any_alias
);
1969 static struct dentry
*__d_obtain_alias(struct inode
*inode
, int disconnected
)
1971 static const struct qstr anonstring
= QSTR_INIT("/", 1);
1977 return ERR_PTR(-ESTALE
);
1979 return ERR_CAST(inode
);
1981 res
= d_find_any_alias(inode
);
1985 tmp
= __d_alloc(inode
->i_sb
, &anonstring
);
1987 res
= ERR_PTR(-ENOMEM
);
1991 spin_lock(&inode
->i_lock
);
1992 res
= __d_find_any_alias(inode
);
1994 spin_unlock(&inode
->i_lock
);
1999 /* attach a disconnected dentry */
2000 add_flags
= d_flags_for_inode(inode
);
2003 add_flags
|= DCACHE_DISCONNECTED
;
2005 spin_lock(&tmp
->d_lock
);
2006 __d_set_inode_and_type(tmp
, inode
, add_flags
);
2007 hlist_add_head(&tmp
->d_u
.d_alias
, &inode
->i_dentry
);
2008 hlist_bl_lock(&tmp
->d_sb
->s_anon
);
2009 hlist_bl_add_head(&tmp
->d_hash
, &tmp
->d_sb
->s_anon
);
2010 hlist_bl_unlock(&tmp
->d_sb
->s_anon
);
2011 spin_unlock(&tmp
->d_lock
);
2012 spin_unlock(&inode
->i_lock
);
2013 security_d_instantiate(tmp
, inode
);
2018 if (res
&& !IS_ERR(res
))
2019 security_d_instantiate(res
, inode
);
2025 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2026 * @inode: inode to allocate the dentry for
2028 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2029 * similar open by handle operations. The returned dentry may be anonymous,
2030 * or may have a full name (if the inode was already in the cache).
2032 * When called on a directory inode, we must ensure that the inode only ever
2033 * has one dentry. If a dentry is found, that is returned instead of
2034 * allocating a new one.
2036 * On successful return, the reference to the inode has been transferred
2037 * to the dentry. In case of an error the reference on the inode is released.
2038 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2039 * be passed in and the error will be propagated to the return value,
2040 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2042 struct dentry
*d_obtain_alias(struct inode
*inode
)
2044 return __d_obtain_alias(inode
, 1);
2046 EXPORT_SYMBOL(d_obtain_alias
);
2049 * d_obtain_root - find or allocate a dentry for a given inode
2050 * @inode: inode to allocate the dentry for
2052 * Obtain an IS_ROOT dentry for the root of a filesystem.
2054 * We must ensure that directory inodes only ever have one dentry. If a
2055 * dentry is found, that is returned instead of allocating a new one.
2057 * On successful return, the reference to the inode has been transferred
2058 * to the dentry. In case of an error the reference on the inode is
2059 * released. A %NULL or IS_ERR inode may be passed in and will be the
2060 * error will be propagate to the return value, with a %NULL @inode
2061 * replaced by ERR_PTR(-ESTALE).
2063 struct dentry
*d_obtain_root(struct inode
*inode
)
2065 return __d_obtain_alias(inode
, 0);
2067 EXPORT_SYMBOL(d_obtain_root
);
2070 * d_add_ci - lookup or allocate new dentry with case-exact name
2071 * @inode: the inode case-insensitive lookup has found
2072 * @dentry: the negative dentry that was passed to the parent's lookup func
2073 * @name: the case-exact name to be associated with the returned dentry
2075 * This is to avoid filling the dcache with case-insensitive names to the
2076 * same inode, only the actual correct case is stored in the dcache for
2077 * case-insensitive filesystems.
2079 * For a case-insensitive lookup match and if the the case-exact dentry
2080 * already exists in in the dcache, use it and return it.
2082 * If no entry exists with the exact case name, allocate new dentry with
2083 * the exact case, and return the spliced entry.
2085 struct dentry
*d_add_ci(struct dentry
*dentry
, struct inode
*inode
,
2088 struct dentry
*found
;
2092 * First check if a dentry matching the name already exists,
2093 * if not go ahead and create it now.
2095 found
= d_hash_and_lookup(dentry
->d_parent
, name
);
2097 new = d_alloc(dentry
->d_parent
, name
);
2099 found
= ERR_PTR(-ENOMEM
);
2101 found
= d_splice_alias(inode
, new);
2112 EXPORT_SYMBOL(d_add_ci
);
2115 * Do the slow-case of the dentry name compare.
2117 * Unlike the dentry_cmp() function, we need to atomically
2118 * load the name and length information, so that the
2119 * filesystem can rely on them, and can use the 'name' and
2120 * 'len' information without worrying about walking off the
2121 * end of memory etc.
2123 * Thus the read_seqcount_retry() and the "duplicate" info
2124 * in arguments (the low-level filesystem should not look
2125 * at the dentry inode or name contents directly, since
2126 * rename can change them while we're in RCU mode).
2128 enum slow_d_compare
{
2134 static noinline
enum slow_d_compare
slow_dentry_cmp(
2135 const struct dentry
*parent
,
2136 struct dentry
*dentry
,
2138 const struct qstr
*name
)
2140 int tlen
= dentry
->d_name
.len
;
2141 const char *tname
= dentry
->d_name
.name
;
2143 if (read_seqcount_retry(&dentry
->d_seq
, seq
)) {
2145 return D_COMP_SEQRETRY
;
2147 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2148 return D_COMP_NOMATCH
;
2153 * __d_lookup_rcu - search for a dentry (racy, store-free)
2154 * @parent: parent dentry
2155 * @name: qstr of name we wish to find
2156 * @seqp: returns d_seq value at the point where the dentry was found
2157 * Returns: dentry, or NULL
2159 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2160 * resolution (store-free path walking) design described in
2161 * Documentation/filesystems/path-lookup.txt.
2163 * This is not to be used outside core vfs.
2165 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2166 * held, and rcu_read_lock held. The returned dentry must not be stored into
2167 * without taking d_lock and checking d_seq sequence count against @seq
2170 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2173 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2174 * the returned dentry, so long as its parent's seqlock is checked after the
2175 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2176 * is formed, giving integrity down the path walk.
2178 * NOTE! The caller *has* to check the resulting dentry against the sequence
2179 * number we've returned before using any of the resulting dentry state!
2181 struct dentry
*__d_lookup_rcu(const struct dentry
*parent
,
2182 const struct qstr
*name
,
2185 u64 hashlen
= name
->hash_len
;
2186 const unsigned char *str
= name
->name
;
2187 struct hlist_bl_head
*b
= d_hash(parent
, hashlen_hash(hashlen
));
2188 struct hlist_bl_node
*node
;
2189 struct dentry
*dentry
;
2192 * Note: There is significant duplication with __d_lookup_rcu which is
2193 * required to prevent single threaded performance regressions
2194 * especially on architectures where smp_rmb (in seqcounts) are costly.
2195 * Keep the two functions in sync.
2199 * The hash list is protected using RCU.
2201 * Carefully use d_seq when comparing a candidate dentry, to avoid
2202 * races with d_move().
2204 * It is possible that concurrent renames can mess up our list
2205 * walk here and result in missing our dentry, resulting in the
2206 * false-negative result. d_lookup() protects against concurrent
2207 * renames using rename_lock seqlock.
2209 * See Documentation/filesystems/path-lookup.txt for more details.
2211 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2216 * The dentry sequence count protects us from concurrent
2217 * renames, and thus protects parent and name fields.
2219 * The caller must perform a seqcount check in order
2220 * to do anything useful with the returned dentry.
2222 * NOTE! We do a "raw" seqcount_begin here. That means that
2223 * we don't wait for the sequence count to stabilize if it
2224 * is in the middle of a sequence change. If we do the slow
2225 * dentry compare, we will do seqretries until it is stable,
2226 * and if we end up with a successful lookup, we actually
2227 * want to exit RCU lookup anyway.
2229 seq
= raw_seqcount_begin(&dentry
->d_seq
);
2230 if (dentry
->d_parent
!= parent
)
2232 if (d_unhashed(dentry
))
2235 if (unlikely(parent
->d_flags
& DCACHE_OP_COMPARE
)) {
2236 if (dentry
->d_name
.hash
!= hashlen_hash(hashlen
))
2239 switch (slow_dentry_cmp(parent
, dentry
, seq
, name
)) {
2242 case D_COMP_NOMATCH
:
2249 if (dentry
->d_name
.hash_len
!= hashlen
)
2252 if (!dentry_cmp(dentry
, str
, hashlen_len(hashlen
)))
2259 * d_lookup - search for a dentry
2260 * @parent: parent dentry
2261 * @name: qstr of name we wish to find
2262 * Returns: dentry, or NULL
2264 * d_lookup searches the children of the parent dentry for the name in
2265 * question. If the dentry is found its reference count is incremented and the
2266 * dentry is returned. The caller must use dput to free the entry when it has
2267 * finished using it. %NULL is returned if the dentry does not exist.
2269 struct dentry
*d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2271 struct dentry
*dentry
;
2275 seq
= read_seqbegin(&rename_lock
);
2276 dentry
= __d_lookup(parent
, name
);
2279 } while (read_seqretry(&rename_lock
, seq
));
2282 EXPORT_SYMBOL(d_lookup
);
2285 * __d_lookup - search for a dentry (racy)
2286 * @parent: parent dentry
2287 * @name: qstr of name we wish to find
2288 * Returns: dentry, or NULL
2290 * __d_lookup is like d_lookup, however it may (rarely) return a
2291 * false-negative result due to unrelated rename activity.
2293 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2294 * however it must be used carefully, eg. with a following d_lookup in
2295 * the case of failure.
2297 * __d_lookup callers must be commented.
2299 struct dentry
*__d_lookup(const struct dentry
*parent
, const struct qstr
*name
)
2301 unsigned int len
= name
->len
;
2302 unsigned int hash
= name
->hash
;
2303 const unsigned char *str
= name
->name
;
2304 struct hlist_bl_head
*b
= d_hash(parent
, hash
);
2305 struct hlist_bl_node
*node
;
2306 struct dentry
*found
= NULL
;
2307 struct dentry
*dentry
;
2310 * Note: There is significant duplication with __d_lookup_rcu which is
2311 * required to prevent single threaded performance regressions
2312 * especially on architectures where smp_rmb (in seqcounts) are costly.
2313 * Keep the two functions in sync.
2317 * The hash list is protected using RCU.
2319 * Take d_lock when comparing a candidate dentry, to avoid races
2322 * It is possible that concurrent renames can mess up our list
2323 * walk here and result in missing our dentry, resulting in the
2324 * false-negative result. d_lookup() protects against concurrent
2325 * renames using rename_lock seqlock.
2327 * See Documentation/filesystems/path-lookup.txt for more details.
2331 hlist_bl_for_each_entry_rcu(dentry
, node
, b
, d_hash
) {
2333 if (dentry
->d_name
.hash
!= hash
)
2336 spin_lock(&dentry
->d_lock
);
2337 if (dentry
->d_parent
!= parent
)
2339 if (d_unhashed(dentry
))
2343 * It is safe to compare names since d_move() cannot
2344 * change the qstr (protected by d_lock).
2346 if (parent
->d_flags
& DCACHE_OP_COMPARE
) {
2347 int tlen
= dentry
->d_name
.len
;
2348 const char *tname
= dentry
->d_name
.name
;
2349 if (parent
->d_op
->d_compare(parent
, dentry
, tlen
, tname
, name
))
2352 if (dentry
->d_name
.len
!= len
)
2354 if (dentry_cmp(dentry
, str
, len
))
2358 dentry
->d_lockref
.count
++;
2360 spin_unlock(&dentry
->d_lock
);
2363 spin_unlock(&dentry
->d_lock
);
2371 * d_hash_and_lookup - hash the qstr then search for a dentry
2372 * @dir: Directory to search in
2373 * @name: qstr of name we wish to find
2375 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2377 struct dentry
*d_hash_and_lookup(struct dentry
*dir
, struct qstr
*name
)
2380 * Check for a fs-specific hash function. Note that we must
2381 * calculate the standard hash first, as the d_op->d_hash()
2382 * routine may choose to leave the hash value unchanged.
2384 name
->hash
= full_name_hash(name
->name
, name
->len
);
2385 if (dir
->d_flags
& DCACHE_OP_HASH
) {
2386 int err
= dir
->d_op
->d_hash(dir
, name
);
2387 if (unlikely(err
< 0))
2388 return ERR_PTR(err
);
2390 return d_lookup(dir
, name
);
2392 EXPORT_SYMBOL(d_hash_and_lookup
);
2395 * When a file is deleted, we have two options:
2396 * - turn this dentry into a negative dentry
2397 * - unhash this dentry and free it.
2399 * Usually, we want to just turn this into
2400 * a negative dentry, but if anybody else is
2401 * currently using the dentry or the inode
2402 * we can't do that and we fall back on removing
2403 * it from the hash queues and waiting for
2404 * it to be deleted later when it has no users
2408 * d_delete - delete a dentry
2409 * @dentry: The dentry to delete
2411 * Turn the dentry into a negative dentry if possible, otherwise
2412 * remove it from the hash queues so it can be deleted later
2415 void d_delete(struct dentry
* dentry
)
2417 struct inode
*inode
;
2420 * Are we the only user?
2423 spin_lock(&dentry
->d_lock
);
2424 inode
= dentry
->d_inode
;
2425 isdir
= S_ISDIR(inode
->i_mode
);
2426 if (dentry
->d_lockref
.count
== 1) {
2427 if (!spin_trylock(&inode
->i_lock
)) {
2428 spin_unlock(&dentry
->d_lock
);
2432 dentry
->d_flags
&= ~DCACHE_CANT_MOUNT
;
2433 dentry_unlink_inode(dentry
);
2434 fsnotify_nameremove(dentry
, isdir
);
2438 if (!d_unhashed(dentry
))
2441 spin_unlock(&dentry
->d_lock
);
2443 fsnotify_nameremove(dentry
, isdir
);
2445 EXPORT_SYMBOL(d_delete
);
2447 static void __d_rehash(struct dentry
* entry
, struct hlist_bl_head
*b
)
2449 BUG_ON(!d_unhashed(entry
));
2451 hlist_bl_add_head_rcu(&entry
->d_hash
, b
);
2455 static void _d_rehash(struct dentry
* entry
)
2457 __d_rehash(entry
, d_hash(entry
->d_parent
, entry
->d_name
.hash
));
2461 * d_rehash - add an entry back to the hash
2462 * @entry: dentry to add to the hash
2464 * Adds a dentry to the hash according to its name.
2467 void d_rehash(struct dentry
* entry
)
2469 spin_lock(&entry
->d_lock
);
2471 spin_unlock(&entry
->d_lock
);
2473 EXPORT_SYMBOL(d_rehash
);
2476 * dentry_update_name_case - update case insensitive dentry with a new name
2477 * @dentry: dentry to be updated
2480 * Update a case insensitive dentry with new case of name.
2482 * dentry must have been returned by d_lookup with name @name. Old and new
2483 * name lengths must match (ie. no d_compare which allows mismatched name
2486 * Parent inode i_mutex must be held over d_lookup and into this call (to
2487 * keep renames and concurrent inserts, and readdir(2) away).
2489 void dentry_update_name_case(struct dentry
*dentry
, struct qstr
*name
)
2491 BUG_ON(!mutex_is_locked(&dentry
->d_parent
->d_inode
->i_mutex
));
2492 BUG_ON(dentry
->d_name
.len
!= name
->len
); /* d_lookup gives this */
2494 spin_lock(&dentry
->d_lock
);
2495 write_seqcount_begin(&dentry
->d_seq
);
2496 memcpy((unsigned char *)dentry
->d_name
.name
, name
->name
, name
->len
);
2497 write_seqcount_end(&dentry
->d_seq
);
2498 spin_unlock(&dentry
->d_lock
);
2500 EXPORT_SYMBOL(dentry_update_name_case
);
2502 static void swap_names(struct dentry
*dentry
, struct dentry
*target
)
2504 if (unlikely(dname_external(target
))) {
2505 if (unlikely(dname_external(dentry
))) {
2507 * Both external: swap the pointers
2509 swap(target
->d_name
.name
, dentry
->d_name
.name
);
2512 * dentry:internal, target:external. Steal target's
2513 * storage and make target internal.
2515 memcpy(target
->d_iname
, dentry
->d_name
.name
,
2516 dentry
->d_name
.len
+ 1);
2517 dentry
->d_name
.name
= target
->d_name
.name
;
2518 target
->d_name
.name
= target
->d_iname
;
2521 if (unlikely(dname_external(dentry
))) {
2523 * dentry:external, target:internal. Give dentry's
2524 * storage to target and make dentry internal
2526 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2527 target
->d_name
.len
+ 1);
2528 target
->d_name
.name
= dentry
->d_name
.name
;
2529 dentry
->d_name
.name
= dentry
->d_iname
;
2532 * Both are internal.
2535 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN
, sizeof(long)));
2536 kmemcheck_mark_initialized(dentry
->d_iname
, DNAME_INLINE_LEN
);
2537 kmemcheck_mark_initialized(target
->d_iname
, DNAME_INLINE_LEN
);
2538 for (i
= 0; i
< DNAME_INLINE_LEN
/ sizeof(long); i
++) {
2539 swap(((long *) &dentry
->d_iname
)[i
],
2540 ((long *) &target
->d_iname
)[i
]);
2544 swap(dentry
->d_name
.hash_len
, target
->d_name
.hash_len
);
2547 static void copy_name(struct dentry
*dentry
, struct dentry
*target
)
2549 struct external_name
*old_name
= NULL
;
2550 if (unlikely(dname_external(dentry
)))
2551 old_name
= external_name(dentry
);
2552 if (unlikely(dname_external(target
))) {
2553 atomic_inc(&external_name(target
)->u
.count
);
2554 dentry
->d_name
= target
->d_name
;
2556 memcpy(dentry
->d_iname
, target
->d_name
.name
,
2557 target
->d_name
.len
+ 1);
2558 dentry
->d_name
.name
= dentry
->d_iname
;
2559 dentry
->d_name
.hash_len
= target
->d_name
.hash_len
;
2561 if (old_name
&& likely(atomic_dec_and_test(&old_name
->u
.count
)))
2562 kfree_rcu(old_name
, u
.head
);
2565 static void dentry_lock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2568 * XXXX: do we really need to take target->d_lock?
2570 if (IS_ROOT(dentry
) || dentry
->d_parent
== target
->d_parent
)
2571 spin_lock(&target
->d_parent
->d_lock
);
2573 if (d_ancestor(dentry
->d_parent
, target
->d_parent
)) {
2574 spin_lock(&dentry
->d_parent
->d_lock
);
2575 spin_lock_nested(&target
->d_parent
->d_lock
,
2576 DENTRY_D_LOCK_NESTED
);
2578 spin_lock(&target
->d_parent
->d_lock
);
2579 spin_lock_nested(&dentry
->d_parent
->d_lock
,
2580 DENTRY_D_LOCK_NESTED
);
2583 if (target
< dentry
) {
2584 spin_lock_nested(&target
->d_lock
, 2);
2585 spin_lock_nested(&dentry
->d_lock
, 3);
2587 spin_lock_nested(&dentry
->d_lock
, 2);
2588 spin_lock_nested(&target
->d_lock
, 3);
2592 static void dentry_unlock_for_move(struct dentry
*dentry
, struct dentry
*target
)
2594 if (target
->d_parent
!= dentry
->d_parent
)
2595 spin_unlock(&dentry
->d_parent
->d_lock
);
2596 if (target
->d_parent
!= target
)
2597 spin_unlock(&target
->d_parent
->d_lock
);
2598 spin_unlock(&target
->d_lock
);
2599 spin_unlock(&dentry
->d_lock
);
2603 * When switching names, the actual string doesn't strictly have to
2604 * be preserved in the target - because we're dropping the target
2605 * anyway. As such, we can just do a simple memcpy() to copy over
2606 * the new name before we switch, unless we are going to rehash
2607 * it. Note that if we *do* unhash the target, we are not allowed
2608 * to rehash it without giving it a new name/hash key - whether
2609 * we swap or overwrite the names here, resulting name won't match
2610 * the reality in filesystem; it's only there for d_path() purposes.
2611 * Note that all of this is happening under rename_lock, so the
2612 * any hash lookup seeing it in the middle of manipulations will
2613 * be discarded anyway. So we do not care what happens to the hash
2617 * __d_move - move a dentry
2618 * @dentry: entry to move
2619 * @target: new dentry
2620 * @exchange: exchange the two dentries
2622 * Update the dcache to reflect the move of a file name. Negative
2623 * dcache entries should not be moved in this way. Caller must hold
2624 * rename_lock, the i_mutex of the source and target directories,
2625 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2627 static void __d_move(struct dentry
*dentry
, struct dentry
*target
,
2630 if (!dentry
->d_inode
)
2631 printk(KERN_WARNING
"VFS: moving negative dcache entry\n");
2633 BUG_ON(d_ancestor(dentry
, target
));
2634 BUG_ON(d_ancestor(target
, dentry
));
2636 dentry_lock_for_move(dentry
, target
);
2638 write_seqcount_begin(&dentry
->d_seq
);
2639 write_seqcount_begin_nested(&target
->d_seq
, DENTRY_D_LOCK_NESTED
);
2641 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2644 * Move the dentry to the target hash queue. Don't bother checking
2645 * for the same hash queue because of how unlikely it is.
2648 __d_rehash(dentry
, d_hash(target
->d_parent
, target
->d_name
.hash
));
2651 * Unhash the target (d_delete() is not usable here). If exchanging
2652 * the two dentries, then rehash onto the other's hash queue.
2657 d_hash(dentry
->d_parent
, dentry
->d_name
.hash
));
2660 /* Switch the names.. */
2662 swap_names(dentry
, target
);
2664 copy_name(dentry
, target
);
2666 /* ... and switch them in the tree */
2667 if (IS_ROOT(dentry
)) {
2668 /* splicing a tree */
2669 dentry
->d_flags
|= DCACHE_RCUACCESS
;
2670 dentry
->d_parent
= target
->d_parent
;
2671 target
->d_parent
= target
;
2672 list_del_init(&target
->d_child
);
2673 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2675 /* swapping two dentries */
2676 swap(dentry
->d_parent
, target
->d_parent
);
2677 list_move(&target
->d_child
, &target
->d_parent
->d_subdirs
);
2678 list_move(&dentry
->d_child
, &dentry
->d_parent
->d_subdirs
);
2680 fsnotify_d_move(target
);
2681 fsnotify_d_move(dentry
);
2684 write_seqcount_end(&target
->d_seq
);
2685 write_seqcount_end(&dentry
->d_seq
);
2687 dentry_unlock_for_move(dentry
, target
);
2691 * d_move - move a dentry
2692 * @dentry: entry to move
2693 * @target: new dentry
2695 * Update the dcache to reflect the move of a file name. Negative
2696 * dcache entries should not be moved in this way. See the locking
2697 * requirements for __d_move.
2699 void d_move(struct dentry
*dentry
, struct dentry
*target
)
2701 write_seqlock(&rename_lock
);
2702 __d_move(dentry
, target
, false);
2703 write_sequnlock(&rename_lock
);
2705 EXPORT_SYMBOL(d_move
);
2708 * d_exchange - exchange two dentries
2709 * @dentry1: first dentry
2710 * @dentry2: second dentry
2712 void d_exchange(struct dentry
*dentry1
, struct dentry
*dentry2
)
2714 write_seqlock(&rename_lock
);
2716 WARN_ON(!dentry1
->d_inode
);
2717 WARN_ON(!dentry2
->d_inode
);
2718 WARN_ON(IS_ROOT(dentry1
));
2719 WARN_ON(IS_ROOT(dentry2
));
2721 __d_move(dentry1
, dentry2
, true);
2723 write_sequnlock(&rename_lock
);
2727 * d_ancestor - search for an ancestor
2728 * @p1: ancestor dentry
2731 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2732 * an ancestor of p2, else NULL.
2734 struct dentry
*d_ancestor(struct dentry
*p1
, struct dentry
*p2
)
2738 for (p
= p2
; !IS_ROOT(p
); p
= p
->d_parent
) {
2739 if (p
->d_parent
== p1
)
2746 * This helper attempts to cope with remotely renamed directories
2748 * It assumes that the caller is already holding
2749 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2751 * Note: If ever the locking in lock_rename() changes, then please
2752 * remember to update this too...
2754 static int __d_unalias(struct inode
*inode
,
2755 struct dentry
*dentry
, struct dentry
*alias
)
2757 struct mutex
*m1
= NULL
, *m2
= NULL
;
2760 /* If alias and dentry share a parent, then no extra locks required */
2761 if (alias
->d_parent
== dentry
->d_parent
)
2764 /* See lock_rename() */
2765 if (!mutex_trylock(&dentry
->d_sb
->s_vfs_rename_mutex
))
2767 m1
= &dentry
->d_sb
->s_vfs_rename_mutex
;
2768 if (!mutex_trylock(&alias
->d_parent
->d_inode
->i_mutex
))
2770 m2
= &alias
->d_parent
->d_inode
->i_mutex
;
2772 __d_move(alias
, dentry
, false);
2783 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2784 * @inode: the inode which may have a disconnected dentry
2785 * @dentry: a negative dentry which we want to point to the inode.
2787 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2788 * place of the given dentry and return it, else simply d_add the inode
2789 * to the dentry and return NULL.
2791 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2792 * we should error out: directories can't have multiple aliases.
2794 * This is needed in the lookup routine of any filesystem that is exportable
2795 * (via knfsd) so that we can build dcache paths to directories effectively.
2797 * If a dentry was found and moved, then it is returned. Otherwise NULL
2798 * is returned. This matches the expected return value of ->lookup.
2800 * Cluster filesystems may call this function with a negative, hashed dentry.
2801 * In that case, we know that the inode will be a regular file, and also this
2802 * will only occur during atomic_open. So we need to check for the dentry
2803 * being already hashed only in the final case.
2805 struct dentry
*d_splice_alias(struct inode
*inode
, struct dentry
*dentry
)
2808 return ERR_CAST(inode
);
2810 BUG_ON(!d_unhashed(dentry
));
2813 __d_instantiate(dentry
, NULL
);
2816 spin_lock(&inode
->i_lock
);
2817 if (S_ISDIR(inode
->i_mode
)) {
2818 struct dentry
*new = __d_find_any_alias(inode
);
2819 if (unlikely(new)) {
2820 /* The reference to new ensures it remains an alias */
2821 spin_unlock(&inode
->i_lock
);
2822 write_seqlock(&rename_lock
);
2823 if (unlikely(d_ancestor(new, dentry
))) {
2824 write_sequnlock(&rename_lock
);
2826 new = ERR_PTR(-ELOOP
);
2827 pr_warn_ratelimited(
2828 "VFS: Lookup of '%s' in %s %s"
2829 " would have caused loop\n",
2830 dentry
->d_name
.name
,
2831 inode
->i_sb
->s_type
->name
,
2833 } else if (!IS_ROOT(new)) {
2834 int err
= __d_unalias(inode
, dentry
, new);
2835 write_sequnlock(&rename_lock
);
2841 __d_move(new, dentry
, false);
2842 write_sequnlock(&rename_lock
);
2843 security_d_instantiate(new, inode
);
2849 /* already taking inode->i_lock, so d_add() by hand */
2850 __d_instantiate(dentry
, inode
);
2851 spin_unlock(&inode
->i_lock
);
2853 security_d_instantiate(dentry
, inode
);
2857 EXPORT_SYMBOL(d_splice_alias
);
2859 static int prepend(char **buffer
, int *buflen
, const char *str
, int namelen
)
2863 return -ENAMETOOLONG
;
2865 memcpy(*buffer
, str
, namelen
);
2870 * prepend_name - prepend a pathname in front of current buffer pointer
2871 * @buffer: buffer pointer
2872 * @buflen: allocated length of the buffer
2873 * @name: name string and length qstr structure
2875 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2876 * make sure that either the old or the new name pointer and length are
2877 * fetched. However, there may be mismatch between length and pointer.
2878 * The length cannot be trusted, we need to copy it byte-by-byte until
2879 * the length is reached or a null byte is found. It also prepends "/" at
2880 * the beginning of the name. The sequence number check at the caller will
2881 * retry it again when a d_move() does happen. So any garbage in the buffer
2882 * due to mismatched pointer and length will be discarded.
2884 * Data dependency barrier is needed to make sure that we see that terminating
2885 * NUL. Alpha strikes again, film at 11...
2887 static int prepend_name(char **buffer
, int *buflen
, struct qstr
*name
)
2889 const char *dname
= ACCESS_ONCE(name
->name
);
2890 u32 dlen
= ACCESS_ONCE(name
->len
);
2893 smp_read_barrier_depends();
2895 *buflen
-= dlen
+ 1;
2897 return -ENAMETOOLONG
;
2898 p
= *buffer
-= dlen
+ 1;
2910 * prepend_path - Prepend path string to a buffer
2911 * @path: the dentry/vfsmount to report
2912 * @root: root vfsmnt/dentry
2913 * @buffer: pointer to the end of the buffer
2914 * @buflen: pointer to buffer length
2916 * The function will first try to write out the pathname without taking any
2917 * lock other than the RCU read lock to make sure that dentries won't go away.
2918 * It only checks the sequence number of the global rename_lock as any change
2919 * in the dentry's d_seq will be preceded by changes in the rename_lock
2920 * sequence number. If the sequence number had been changed, it will restart
2921 * the whole pathname back-tracing sequence again by taking the rename_lock.
2922 * In this case, there is no need to take the RCU read lock as the recursive
2923 * parent pointer references will keep the dentry chain alive as long as no
2924 * rename operation is performed.
2926 static int prepend_path(const struct path
*path
,
2927 const struct path
*root
,
2928 char **buffer
, int *buflen
)
2930 struct dentry
*dentry
;
2931 struct vfsmount
*vfsmnt
;
2934 unsigned seq
, m_seq
= 0;
2940 read_seqbegin_or_lock(&mount_lock
, &m_seq
);
2947 dentry
= path
->dentry
;
2949 mnt
= real_mount(vfsmnt
);
2950 read_seqbegin_or_lock(&rename_lock
, &seq
);
2951 while (dentry
!= root
->dentry
|| vfsmnt
!= root
->mnt
) {
2952 struct dentry
* parent
;
2954 if (dentry
== vfsmnt
->mnt_root
|| IS_ROOT(dentry
)) {
2955 struct mount
*parent
= ACCESS_ONCE(mnt
->mnt_parent
);
2957 if (dentry
!= vfsmnt
->mnt_root
) {
2964 if (mnt
!= parent
) {
2965 dentry
= ACCESS_ONCE(mnt
->mnt_mountpoint
);
2967 #ifdef CONFIG_RKP_NS_PROT
2975 error
= is_mounted(vfsmnt
) ? 1 : 2;
2978 parent
= dentry
->d_parent
;
2980 error
= prepend_name(&bptr
, &blen
, &dentry
->d_name
);
2988 if (need_seqretry(&rename_lock
, seq
)) {
2992 done_seqretry(&rename_lock
, seq
);
2996 if (need_seqretry(&mount_lock
, m_seq
)) {
3000 done_seqretry(&mount_lock
, m_seq
);
3002 if (error
>= 0 && bptr
== *buffer
) {
3004 error
= -ENAMETOOLONG
;
3014 * __d_path - return the path of a dentry
3015 * @path: the dentry/vfsmount to report
3016 * @root: root vfsmnt/dentry
3017 * @buf: buffer to return value in
3018 * @buflen: buffer length
3020 * Convert a dentry into an ASCII path name.
3022 * Returns a pointer into the buffer or an error code if the
3023 * path was too long.
3025 * "buflen" should be positive.
3027 * If the path is not reachable from the supplied root, return %NULL.
3029 char *__d_path(const struct path
*path
,
3030 const struct path
*root
,
3031 char *buf
, int buflen
)
3033 char *res
= buf
+ buflen
;
3036 prepend(&res
, &buflen
, "\0", 1);
3037 error
= prepend_path(path
, root
, &res
, &buflen
);
3040 return ERR_PTR(error
);
3046 char *d_absolute_path(const struct path
*path
,
3047 char *buf
, int buflen
)
3049 struct path root
= {};
3050 char *res
= buf
+ buflen
;
3053 prepend(&res
, &buflen
, "\0", 1);
3054 error
= prepend_path(path
, &root
, &res
, &buflen
);
3059 return ERR_PTR(error
);
3062 EXPORT_SYMBOL(d_absolute_path
);
3065 * same as __d_path but appends "(deleted)" for unlinked files.
3067 static int path_with_deleted(const struct path
*path
,
3068 const struct path
*root
,
3069 char **buf
, int *buflen
)
3071 prepend(buf
, buflen
, "\0", 1);
3072 if (d_unlinked(path
->dentry
)) {
3073 int error
= prepend(buf
, buflen
, " (deleted)", 10);
3078 return prepend_path(path
, root
, buf
, buflen
);
3081 static int prepend_unreachable(char **buffer
, int *buflen
)
3083 return prepend(buffer
, buflen
, "(unreachable)", 13);
3086 static void get_fs_root_rcu(struct fs_struct
*fs
, struct path
*root
)
3091 seq
= read_seqcount_begin(&fs
->seq
);
3093 } while (read_seqcount_retry(&fs
->seq
, seq
));
3097 * d_path - return the path of a dentry
3098 * @path: path to report
3099 * @buf: buffer to return value in
3100 * @buflen: buffer length
3102 * Convert a dentry into an ASCII path name. If the entry has been deleted
3103 * the string " (deleted)" is appended. Note that this is ambiguous.
3105 * Returns a pointer into the buffer or an error code if the path was
3106 * too long. Note: Callers should use the returned pointer, not the passed
3107 * in buffer, to use the name! The implementation often starts at an offset
3108 * into the buffer, and may leave 0 bytes at the start.
3110 * "buflen" should be positive.
3112 char *d_path(const struct path
*path
, char *buf
, int buflen
)
3114 char *res
= buf
+ buflen
;
3119 * We have various synthetic filesystems that never get mounted. On
3120 * these filesystems dentries are never used for lookup purposes, and
3121 * thus don't need to be hashed. They also don't need a name until a
3122 * user wants to identify the object in /proc/pid/fd/. The little hack
3123 * below allows us to generate a name for these objects on demand:
3125 * Some pseudo inodes are mountable. When they are mounted
3126 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3127 * and instead have d_path return the mounted path.
3129 if (path
->dentry
->d_op
&& path
->dentry
->d_op
->d_dname
&&
3130 (!IS_ROOT(path
->dentry
) || path
->dentry
!= path
->mnt
->mnt_root
))
3131 return path
->dentry
->d_op
->d_dname(path
->dentry
, buf
, buflen
);
3134 get_fs_root_rcu(current
->fs
, &root
);
3135 error
= path_with_deleted(path
, &root
, &res
, &buflen
);
3139 res
= ERR_PTR(error
);
3142 EXPORT_SYMBOL(d_path
);
3145 * Helper function for dentry_operations.d_dname() members
3147 char *dynamic_dname(struct dentry
*dentry
, char *buffer
, int buflen
,
3148 const char *fmt
, ...)
3154 va_start(args
, fmt
);
3155 sz
= vsnprintf(temp
, sizeof(temp
), fmt
, args
) + 1;
3158 if (sz
> sizeof(temp
) || sz
> buflen
)
3159 return ERR_PTR(-ENAMETOOLONG
);
3161 buffer
+= buflen
- sz
;
3162 return memcpy(buffer
, temp
, sz
);
3165 char *simple_dname(struct dentry
*dentry
, char *buffer
, int buflen
)
3167 char *end
= buffer
+ buflen
;
3168 /* these dentries are never renamed, so d_lock is not needed */
3169 if (prepend(&end
, &buflen
, " (deleted)", 11) ||
3170 prepend(&end
, &buflen
, dentry
->d_name
.name
, dentry
->d_name
.len
) ||
3171 prepend(&end
, &buflen
, "/", 1))
3172 end
= ERR_PTR(-ENAMETOOLONG
);
3175 EXPORT_SYMBOL(simple_dname
);
3178 * Write full pathname from the root of the filesystem into the buffer.
3180 static char *__dentry_path(struct dentry
*d
, char *buf
, int buflen
)
3182 struct dentry
*dentry
;
3195 prepend(&end
, &len
, "\0", 1);
3199 read_seqbegin_or_lock(&rename_lock
, &seq
);
3200 while (!IS_ROOT(dentry
)) {
3201 struct dentry
*parent
= dentry
->d_parent
;
3204 error
= prepend_name(&end
, &len
, &dentry
->d_name
);
3213 if (need_seqretry(&rename_lock
, seq
)) {
3217 done_seqretry(&rename_lock
, seq
);
3222 return ERR_PTR(-ENAMETOOLONG
);
3225 char *dentry_path_raw(struct dentry
*dentry
, char *buf
, int buflen
)
3227 return __dentry_path(dentry
, buf
, buflen
);
3229 EXPORT_SYMBOL(dentry_path_raw
);
3231 char *dentry_path(struct dentry
*dentry
, char *buf
, int buflen
)
3236 if (d_unlinked(dentry
)) {
3238 if (prepend(&p
, &buflen
, "//deleted", 10) != 0)
3242 retval
= __dentry_path(dentry
, buf
, buflen
);
3243 if (!IS_ERR(retval
) && p
)
3244 *p
= '/'; /* restore '/' overriden with '\0' */
3247 return ERR_PTR(-ENAMETOOLONG
);
3250 static void get_fs_root_and_pwd_rcu(struct fs_struct
*fs
, struct path
*root
,
3256 seq
= read_seqcount_begin(&fs
->seq
);
3259 } while (read_seqcount_retry(&fs
->seq
, seq
));
3263 * NOTE! The user-level library version returns a
3264 * character pointer. The kernel system call just
3265 * returns the length of the buffer filled (which
3266 * includes the ending '\0' character), or a negative
3267 * error value. So libc would do something like
3269 * char *getcwd(char * buf, size_t size)
3273 * retval = sys_getcwd(buf, size);
3280 SYSCALL_DEFINE2(getcwd
, char __user
*, buf
, unsigned long, size
)
3283 struct path pwd
, root
;
3284 char *page
= __getname();
3290 get_fs_root_and_pwd_rcu(current
->fs
, &root
, &pwd
);
3293 if (!d_unlinked(pwd
.dentry
)) {
3295 char *cwd
= page
+ PATH_MAX
;
3296 int buflen
= PATH_MAX
;
3298 prepend(&cwd
, &buflen
, "\0", 1);
3299 error
= prepend_path(&pwd
, &root
, &cwd
, &buflen
);
3305 /* Unreachable from current root */
3307 error
= prepend_unreachable(&cwd
, &buflen
);
3313 len
= PATH_MAX
+ page
- cwd
;
3316 if (copy_to_user(buf
, cwd
, len
))
3329 * Test whether new_dentry is a subdirectory of old_dentry.
3331 * Trivially implemented using the dcache structure
3335 * is_subdir - is new dentry a subdirectory of old_dentry
3336 * @new_dentry: new dentry
3337 * @old_dentry: old dentry
3339 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3340 * Returns 0 otherwise.
3341 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3344 int is_subdir(struct dentry
*new_dentry
, struct dentry
*old_dentry
)
3349 if (new_dentry
== old_dentry
)
3353 /* for restarting inner loop in case of seq retry */
3354 seq
= read_seqbegin(&rename_lock
);
3356 * Need rcu_readlock to protect against the d_parent trashing
3360 if (d_ancestor(old_dentry
, new_dentry
))
3365 } while (read_seqretry(&rename_lock
, seq
));
3370 static enum d_walk_ret
d_genocide_kill(void *data
, struct dentry
*dentry
)
3372 struct dentry
*root
= data
;
3373 if (dentry
!= root
) {
3374 if (d_unhashed(dentry
) || !dentry
->d_inode
)
3377 if (!(dentry
->d_flags
& DCACHE_GENOCIDE
)) {
3378 dentry
->d_flags
|= DCACHE_GENOCIDE
;
3379 dentry
->d_lockref
.count
--;
3382 return D_WALK_CONTINUE
;
3385 void d_genocide(struct dentry
*parent
)
3387 d_walk(parent
, parent
, d_genocide_kill
, NULL
);
3390 void d_tmpfile(struct dentry
*dentry
, struct inode
*inode
)
3392 inode_dec_link_count(inode
);
3393 BUG_ON(dentry
->d_name
.name
!= dentry
->d_iname
||
3394 !hlist_unhashed(&dentry
->d_u
.d_alias
) ||
3395 !d_unlinked(dentry
));
3396 spin_lock(&dentry
->d_parent
->d_lock
);
3397 spin_lock_nested(&dentry
->d_lock
, DENTRY_D_LOCK_NESTED
);
3398 dentry
->d_name
.len
= sprintf(dentry
->d_iname
, "#%llu",
3399 (unsigned long long)inode
->i_ino
);
3400 spin_unlock(&dentry
->d_lock
);
3401 spin_unlock(&dentry
->d_parent
->d_lock
);
3402 d_instantiate(dentry
, inode
);
3404 EXPORT_SYMBOL(d_tmpfile
);
3406 static __initdata
unsigned long dhash_entries
;
3407 static int __init
set_dhash_entries(char *str
)
3411 dhash_entries
= simple_strtoul(str
, &str
, 0);
3414 __setup("dhash_entries=", set_dhash_entries
);
3416 static void __init
dcache_init_early(void)
3420 /* If hashes are distributed across NUMA nodes, defer
3421 * hash allocation until vmalloc space is available.
3427 alloc_large_system_hash("Dentry cache",
3428 sizeof(struct hlist_bl_head
),
3437 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3438 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3441 static void __init
dcache_init(void)
3446 * A constructor could be added for stable state like the lists,
3447 * but it is probably not worth it because of the cache nature
3450 dentry_cache
= KMEM_CACHE(dentry
,
3451 SLAB_RECLAIM_ACCOUNT
|SLAB_PANIC
|SLAB_MEM_SPREAD
);
3453 /* Hash may have been set up in dcache_init_early */
3458 alloc_large_system_hash("Dentry cache",
3459 sizeof(struct hlist_bl_head
),
3468 for (loop
= 0; loop
< (1U << d_hash_shift
); loop
++)
3469 INIT_HLIST_BL_HEAD(dentry_hashtable
+ loop
);
3472 /* SLAB cache for __getname() consumers */
3473 struct kmem_cache
*names_cachep __read_mostly
;
3474 EXPORT_SYMBOL(names_cachep
);
3476 EXPORT_SYMBOL(d_genocide
);
3478 void __init
vfs_caches_init_early(void)
3480 dcache_init_early();
3484 void __init
vfs_caches_init(void)
3486 names_cachep
= kmem_cache_create("names_cache", PATH_MAX
, 0,
3487 SLAB_HWCACHE_ALIGN
|SLAB_PANIC
, NULL
);
3492 files_maxfiles_init();
3496 #ifdef CONFIG_RKP_NS_PROT