1 // SPDX-License-Identifier: GPL-2.0-only
5 * Complete reimplementation
6 * (C) 1997 Thomas Schoebel-Theuer,
7 * with heavy changes by Linus Torvalds
11 * Notes on the allocation strategy:
13 * The dcache is a master of the icache - whenever a dcache entry
14 * exists, the inode will always exist. "iput()" is done either when
15 * the dcache entry is deleted or garbage collected.
18 #include <linux/ratelimit.h>
19 #include <linux/string.h>
22 #include <linux/fscrypt.h>
23 #include <linux/fsnotify.h>
24 #include <linux/slab.h>
25 #include <linux/init.h>
26 #include <linux/hash.h>
27 #include <linux/cache.h>
28 #include <linux/export.h>
29 #include <linux/security.h>
30 #include <linux/seqlock.h>
31 #include <linux/memblock.h>
32 #include <linux/bit_spinlock.h>
33 #include <linux/rculist_bl.h>
34 #include <linux/list_lru.h>
40 * dcache->d_inode->i_lock protects:
41 * - i_dentry, d_u.d_alias, d_inode of aliases
42 * dcache_hash_bucket lock protects:
43 * - the dcache hash table
44 * s_roots bl list spinlock protects:
45 * - the s_roots list (see __d_drop)
46 * dentry->d_sb->s_dentry_lru_lock protects:
47 * - the dcache lru lists and counters
54 * - d_parent and d_subdirs
55 * - childrens' d_child and d_parent
56 * - d_u.d_alias, d_inode
59 * dentry->d_inode->i_lock
61 * dentry->d_sb->s_dentry_lru_lock
62 * dcache_hash_bucket lock
65 * If there is an ancestor relationship:
66 * dentry->d_parent->...->d_parent->d_lock
68 * dentry->d_parent->d_lock
71 * If no ancestor relationship:
72 * arbitrary, since it's serialized on rename_lock
74 int sysctl_vfs_cache_pressure __read_mostly = 100;
75 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
77 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
79 EXPORT_SYMBOL(rename_lock);
81 static struct kmem_cache *dentry_cache __read_mostly;
83 const struct qstr empty_name = QSTR_INIT("", 0);
84 EXPORT_SYMBOL(empty_name);
85 const struct qstr slash_name = QSTR_INIT("/", 1);
86 EXPORT_SYMBOL(slash_name);
89 * This is the single most critical data structure when it comes
90 * to the dcache: the hashtable for lookups. Somebody should try
91 * to make this good - I've just made it work.
93 * This hash-function tries to avoid losing too many bits of hash
94 * information, yet avoid using a prime hash-size or similar.
97 static unsigned int d_hash_shift __read_mostly;
99 static struct hlist_bl_head *dentry_hashtable __read_mostly;
101 static inline struct hlist_bl_head *d_hash(unsigned int hash)
103 return dentry_hashtable + (hash >> d_hash_shift);
106 #define IN_LOOKUP_SHIFT 10
107 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
109 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
112 hash += (unsigned long) parent / L1_CACHE_BYTES;
113 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_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);
124 static DEFINE_PER_CPU(long, nr_dentry_negative);
126 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
129 * Here we resort to our own counters instead of using generic per-cpu counters
130 * for consistency with what the vfs inode code does. We are expected to harvest
131 * better code and performance by having our own specialized counters.
133 * Please note that the loop is done over all possible CPUs, not over all online
134 * CPUs. The reason for this is that we don't want to play games with CPUs going
135 * on and off. If one of them goes off, we will just keep their counters.
137 * glommer: See cffbc8a for details, and if you ever intend to change this,
138 * please update all vfs counters to match.
140 static long get_nr_dentry(void)
144 for_each_possible_cpu(i)
145 sum += per_cpu(nr_dentry, i);
146 return sum < 0 ? 0 : sum;
149 static long get_nr_dentry_unused(void)
153 for_each_possible_cpu(i)
154 sum += per_cpu(nr_dentry_unused, i);
155 return sum < 0 ? 0 : sum;
158 static long get_nr_dentry_negative(void)
163 for_each_possible_cpu(i)
164 sum += per_cpu(nr_dentry_negative, 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 dentry_stat.nr_negative = get_nr_dentry_negative();
174 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
179 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
180 * The strings are both count bytes long, and count is non-zero.
182 #ifdef CONFIG_DCACHE_WORD_ACCESS
184 #include <asm/word-at-a-time.h>
186 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
187 * aligned allocation for this particular component. We don't
188 * strictly need the load_unaligned_zeropad() safety, but it
189 * doesn't hurt either.
191 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
192 * need the careful unaligned handling.
194 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
196 unsigned long a,b,mask;
199 a = read_word_at_a_time(cs);
200 b = load_unaligned_zeropad(ct);
201 if (tcount < sizeof(unsigned long))
203 if (unlikely(a != b))
205 cs += sizeof(unsigned long);
206 ct += sizeof(unsigned long);
207 tcount -= sizeof(unsigned long);
211 mask = bytemask_from_count(tcount);
212 return unlikely(!!((a ^ b) & mask));
217 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
231 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
234 * Be careful about RCU walk racing with rename:
235 * use 'READ_ONCE' to fetch the name pointer.
237 * NOTE! Even if a rename will mean that the length
238 * was not loaded atomically, we don't care. The
239 * RCU walk will check the sequence count eventually,
240 * and catch it. And we won't overrun the buffer,
241 * because we're reading the name pointer atomically,
242 * and a dentry name is guaranteed to be properly
243 * terminated with a NUL byte.
245 * End result: even if 'len' is wrong, we'll exit
246 * early because the data cannot match (there can
247 * be no NUL in the ct/tcount data)
249 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
251 return dentry_string_cmp(cs, ct, tcount);
254 struct external_name {
257 struct rcu_head head;
259 unsigned char name[];
262 static inline struct external_name *external_name(struct dentry *dentry)
264 return container_of(dentry->d_name.name, struct external_name, name[0]);
267 static void __d_free(struct rcu_head *head)
269 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
271 kmem_cache_free(dentry_cache, dentry);
274 static void __d_free_external(struct rcu_head *head)
276 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
277 kfree(external_name(dentry));
278 kmem_cache_free(dentry_cache, dentry);
281 static inline int dname_external(const struct dentry *dentry)
283 return dentry->d_name.name != dentry->d_iname;
286 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
288 spin_lock(&dentry->d_lock);
289 name->name = dentry->d_name;
290 if (unlikely(dname_external(dentry))) {
291 atomic_inc(&external_name(dentry)->u.count);
293 memcpy(name->inline_name, dentry->d_iname,
294 dentry->d_name.len + 1);
295 name->name.name = name->inline_name;
297 spin_unlock(&dentry->d_lock);
299 EXPORT_SYMBOL(take_dentry_name_snapshot);
301 void release_dentry_name_snapshot(struct name_snapshot *name)
303 if (unlikely(name->name.name != name->inline_name)) {
304 struct external_name *p;
305 p = container_of(name->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 smp_store_release(&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;
332 if (dentry->d_flags & DCACHE_LRU_LIST)
333 this_cpu_inc(nr_dentry_negative);
336 static void dentry_free(struct dentry *dentry)
338 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
339 if (unlikely(dname_external(dentry))) {
340 struct external_name *p = external_name(dentry);
341 if (likely(atomic_dec_and_test(&p->u.count))) {
342 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
346 /* if dentry was never visible to RCU, immediate free is OK */
347 if (dentry->d_flags & DCACHE_NORCU)
348 __d_free(&dentry->d_u.d_rcu);
350 call_rcu(&dentry->d_u.d_rcu, __d_free);
354 * Release the dentry's inode, using the filesystem
355 * d_iput() operation if defined.
357 static void dentry_unlink_inode(struct dentry * dentry)
358 __releases(dentry->d_lock)
359 __releases(dentry->d_inode->i_lock)
361 struct inode *inode = dentry->d_inode;
363 raw_write_seqcount_begin(&dentry->d_seq);
364 __d_clear_type_and_inode(dentry);
365 hlist_del_init(&dentry->d_u.d_alias);
366 raw_write_seqcount_end(&dentry->d_seq);
367 spin_unlock(&dentry->d_lock);
368 spin_unlock(&inode->i_lock);
370 fsnotify_inoderemove(inode);
371 if (dentry->d_op && dentry->d_op->d_iput)
372 dentry->d_op->d_iput(dentry, inode);
378 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
379 * is in use - which includes both the "real" per-superblock
380 * LRU list _and_ the DCACHE_SHRINK_LIST use.
382 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
383 * on the shrink list (ie not on the superblock LRU list).
385 * The per-cpu "nr_dentry_unused" counters are updated with
386 * the DCACHE_LRU_LIST bit.
388 * The per-cpu "nr_dentry_negative" counters are only updated
389 * when deleted from or added to the per-superblock LRU list, not
390 * from/to the shrink list. That is to avoid an unneeded dec/inc
391 * pair when moving from LRU to shrink list in select_collect().
393 * These helper functions make sure we always follow the
394 * rules. d_lock must be held by the caller.
396 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
397 static void d_lru_add(struct dentry *dentry)
399 D_FLAG_VERIFY(dentry, 0);
400 dentry->d_flags |= DCACHE_LRU_LIST;
401 this_cpu_inc(nr_dentry_unused);
402 if (d_is_negative(dentry))
403 this_cpu_inc(nr_dentry_negative);
404 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
407 static void d_lru_del(struct dentry *dentry)
409 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
410 dentry->d_flags &= ~DCACHE_LRU_LIST;
411 this_cpu_dec(nr_dentry_unused);
412 if (d_is_negative(dentry))
413 this_cpu_dec(nr_dentry_negative);
414 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
417 static void d_shrink_del(struct dentry *dentry)
419 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
420 list_del_init(&dentry->d_lru);
421 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
422 this_cpu_dec(nr_dentry_unused);
425 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
427 D_FLAG_VERIFY(dentry, 0);
428 list_add(&dentry->d_lru, list);
429 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
430 this_cpu_inc(nr_dentry_unused);
434 * These can only be called under the global LRU lock, ie during the
435 * callback for freeing the LRU list. "isolate" removes it from the
436 * LRU lists entirely, while shrink_move moves it to the indicated
439 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
441 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
442 dentry->d_flags &= ~DCACHE_LRU_LIST;
443 this_cpu_dec(nr_dentry_unused);
444 if (d_is_negative(dentry))
445 this_cpu_dec(nr_dentry_negative);
446 list_lru_isolate(lru, &dentry->d_lru);
449 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
450 struct list_head *list)
452 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
453 dentry->d_flags |= DCACHE_SHRINK_LIST;
454 if (d_is_negative(dentry))
455 this_cpu_dec(nr_dentry_negative);
456 list_lru_isolate_move(lru, &dentry->d_lru, list);
460 * d_drop - drop a dentry
461 * @dentry: dentry to drop
463 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
464 * be found through a VFS lookup any more. Note that this is different from
465 * deleting the dentry - d_delete will try to mark the dentry negative if
466 * possible, giving a successful _negative_ lookup, while d_drop will
467 * just make the cache lookup fail.
469 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
470 * reason (NFS timeouts or autofs deletes).
472 * __d_drop requires dentry->d_lock
473 * ___d_drop doesn't mark dentry as "unhashed"
474 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
476 static void ___d_drop(struct dentry *dentry)
478 struct hlist_bl_head *b;
480 * Hashed dentries are normally on the dentry hashtable,
481 * with the exception of those newly allocated by
482 * d_obtain_root, which are always IS_ROOT:
484 if (unlikely(IS_ROOT(dentry)))
485 b = &dentry->d_sb->s_roots;
487 b = d_hash(dentry->d_name.hash);
490 __hlist_bl_del(&dentry->d_hash);
494 void __d_drop(struct dentry *dentry)
496 if (!d_unhashed(dentry)) {
498 dentry->d_hash.pprev = NULL;
499 write_seqcount_invalidate(&dentry->d_seq);
502 EXPORT_SYMBOL(__d_drop);
504 void d_drop(struct dentry *dentry)
506 spin_lock(&dentry->d_lock);
508 spin_unlock(&dentry->d_lock);
510 EXPORT_SYMBOL(d_drop);
512 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
516 * Inform d_walk() and shrink_dentry_list() that we are no longer
517 * attached to the dentry tree
519 dentry->d_flags |= DCACHE_DENTRY_KILLED;
520 if (unlikely(list_empty(&dentry->d_child)))
522 __list_del_entry(&dentry->d_child);
524 * Cursors can move around the list of children. While we'd been
525 * a normal list member, it didn't matter - ->d_child.next would've
526 * been updated. However, from now on it won't be and for the
527 * things like d_walk() it might end up with a nasty surprise.
528 * Normally d_walk() doesn't care about cursors moving around -
529 * ->d_lock on parent prevents that and since a cursor has no children
530 * of its own, we get through it without ever unlocking the parent.
531 * There is one exception, though - if we ascend from a child that
532 * gets killed as soon as we unlock it, the next sibling is found
533 * using the value left in its ->d_child.next. And if _that_
534 * pointed to a cursor, and cursor got moved (e.g. by lseek())
535 * before d_walk() regains parent->d_lock, we'll end up skipping
536 * everything the cursor had been moved past.
538 * Solution: make sure that the pointer left behind in ->d_child.next
539 * points to something that won't be moving around. I.e. skip the
542 while (dentry->d_child.next != &parent->d_subdirs) {
543 next = list_entry(dentry->d_child.next, struct dentry, d_child);
544 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
546 dentry->d_child.next = next->d_child.next;
550 static void __dentry_kill(struct dentry *dentry)
552 struct dentry *parent = NULL;
553 bool can_free = true;
554 if (!IS_ROOT(dentry))
555 parent = dentry->d_parent;
558 * The dentry is now unrecoverably dead to the world.
560 lockref_mark_dead(&dentry->d_lockref);
563 * inform the fs via d_prune that this dentry is about to be
564 * unhashed and destroyed.
566 if (dentry->d_flags & DCACHE_OP_PRUNE)
567 dentry->d_op->d_prune(dentry);
569 if (dentry->d_flags & DCACHE_LRU_LIST) {
570 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
573 /* if it was on the hash then remove it */
575 dentry_unlist(dentry, parent);
577 spin_unlock(&parent->d_lock);
579 dentry_unlink_inode(dentry);
581 spin_unlock(&dentry->d_lock);
582 this_cpu_dec(nr_dentry);
583 if (dentry->d_op && dentry->d_op->d_release)
584 dentry->d_op->d_release(dentry);
586 spin_lock(&dentry->d_lock);
587 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
588 dentry->d_flags |= DCACHE_MAY_FREE;
591 spin_unlock(&dentry->d_lock);
592 if (likely(can_free))
597 static struct dentry *__lock_parent(struct dentry *dentry)
599 struct dentry *parent;
601 spin_unlock(&dentry->d_lock);
603 parent = READ_ONCE(dentry->d_parent);
604 spin_lock(&parent->d_lock);
606 * We can't blindly lock dentry until we are sure
607 * that we won't violate the locking order.
608 * Any changes of dentry->d_parent must have
609 * been done with parent->d_lock held, so
610 * spin_lock() above is enough of a barrier
611 * for checking if it's still our child.
613 if (unlikely(parent != dentry->d_parent)) {
614 spin_unlock(&parent->d_lock);
618 if (parent != dentry)
619 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
625 static inline struct dentry *lock_parent(struct dentry *dentry)
627 struct dentry *parent = dentry->d_parent;
630 if (likely(spin_trylock(&parent->d_lock)))
632 return __lock_parent(dentry);
635 static inline bool retain_dentry(struct dentry *dentry)
637 WARN_ON(d_in_lookup(dentry));
639 /* Unreachable? Get rid of it */
640 if (unlikely(d_unhashed(dentry)))
643 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
646 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
647 if (dentry->d_op->d_delete(dentry))
650 /* retain; LRU fodder */
651 dentry->d_lockref.count--;
652 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
654 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
655 dentry->d_flags |= DCACHE_REFERENCED;
660 * Finish off a dentry we've decided to kill.
661 * dentry->d_lock must be held, returns with it unlocked.
662 * Returns dentry requiring refcount drop, or NULL if we're done.
664 static struct dentry *dentry_kill(struct dentry *dentry)
665 __releases(dentry->d_lock)
667 struct inode *inode = dentry->d_inode;
668 struct dentry *parent = NULL;
670 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
673 if (!IS_ROOT(dentry)) {
674 parent = dentry->d_parent;
675 if (unlikely(!spin_trylock(&parent->d_lock))) {
676 parent = __lock_parent(dentry);
677 if (likely(inode || !dentry->d_inode))
679 /* negative that became positive */
681 spin_unlock(&parent->d_lock);
682 inode = dentry->d_inode;
686 __dentry_kill(dentry);
690 spin_unlock(&dentry->d_lock);
691 spin_lock(&inode->i_lock);
692 spin_lock(&dentry->d_lock);
693 parent = lock_parent(dentry);
695 if (unlikely(dentry->d_lockref.count != 1)) {
696 dentry->d_lockref.count--;
697 } else if (likely(!retain_dentry(dentry))) {
698 __dentry_kill(dentry);
701 /* we are keeping it, after all */
703 spin_unlock(&inode->i_lock);
705 spin_unlock(&parent->d_lock);
706 spin_unlock(&dentry->d_lock);
711 * Try to do a lockless dput(), and return whether that was successful.
713 * If unsuccessful, we return false, having already taken the dentry lock.
715 * The caller needs to hold the RCU read lock, so that the dentry is
716 * guaranteed to stay around even if the refcount goes down to zero!
718 static inline bool fast_dput(struct dentry *dentry)
721 unsigned int d_flags;
724 * If we have a d_op->d_delete() operation, we sould not
725 * let the dentry count go to zero, so use "put_or_lock".
727 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
728 return lockref_put_or_lock(&dentry->d_lockref);
731 * .. otherwise, we can try to just decrement the
732 * lockref optimistically.
734 ret = lockref_put_return(&dentry->d_lockref);
737 * If the lockref_put_return() failed due to the lock being held
738 * by somebody else, the fast path has failed. We will need to
739 * get the lock, and then check the count again.
741 if (unlikely(ret < 0)) {
742 spin_lock(&dentry->d_lock);
743 if (dentry->d_lockref.count > 1) {
744 dentry->d_lockref.count--;
745 spin_unlock(&dentry->d_lock);
752 * If we weren't the last ref, we're done.
758 * Careful, careful. The reference count went down
759 * to zero, but we don't hold the dentry lock, so
760 * somebody else could get it again, and do another
761 * dput(), and we need to not race with that.
763 * However, there is a very special and common case
764 * where we don't care, because there is nothing to
765 * do: the dentry is still hashed, it does not have
766 * a 'delete' op, and it's referenced and already on
769 * NOTE! Since we aren't locked, these values are
770 * not "stable". However, it is sufficient that at
771 * some point after we dropped the reference the
772 * dentry was hashed and the flags had the proper
773 * value. Other dentry users may have re-gotten
774 * a reference to the dentry and change that, but
775 * our work is done - we can leave the dentry
776 * around with a zero refcount.
779 d_flags = READ_ONCE(dentry->d_flags);
780 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
782 /* Nothing to do? Dropping the reference was all we needed? */
783 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
787 * Not the fast normal case? Get the lock. We've already decremented
788 * the refcount, but we'll need to re-check the situation after
791 spin_lock(&dentry->d_lock);
794 * Did somebody else grab a reference to it in the meantime, and
795 * we're no longer the last user after all? Alternatively, somebody
796 * else could have killed it and marked it dead. Either way, we
797 * don't need to do anything else.
799 if (dentry->d_lockref.count) {
800 spin_unlock(&dentry->d_lock);
805 * Re-get the reference we optimistically dropped. We hold the
806 * lock, and we just tested that it was zero, so we can just
809 dentry->d_lockref.count = 1;
817 * This is complicated by the fact that we do not want to put
818 * dentries that are no longer on any hash chain on the unused
819 * list: we'd much rather just get rid of them immediately.
821 * However, that implies that we have to traverse the dentry
822 * tree upwards to the parents which might _also_ now be
823 * scheduled for deletion (it may have been only waiting for
824 * its last child to go away).
826 * This tail recursion is done by hand as we don't want to depend
827 * on the compiler to always get this right (gcc generally doesn't).
828 * Real recursion would eat up our stack space.
832 * dput - release a dentry
833 * @dentry: dentry to release
835 * Release a dentry. This will drop the usage count and if appropriate
836 * call the dentry unlink method as well as removing it from the queues and
837 * releasing its resources. If the parent dentries were scheduled for release
838 * they too may now get deleted.
840 void dput(struct dentry *dentry)
846 if (likely(fast_dput(dentry))) {
851 /* Slow case: now with the dentry lock held */
854 if (likely(retain_dentry(dentry))) {
855 spin_unlock(&dentry->d_lock);
859 dentry = dentry_kill(dentry);
864 static void __dput_to_list(struct dentry *dentry, struct list_head *list)
865 __must_hold(&dentry->d_lock)
867 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
868 /* let the owner of the list it's on deal with it */
869 --dentry->d_lockref.count;
871 if (dentry->d_flags & DCACHE_LRU_LIST)
873 if (!--dentry->d_lockref.count)
874 d_shrink_add(dentry, list);
878 void dput_to_list(struct dentry *dentry, struct list_head *list)
881 if (likely(fast_dput(dentry))) {
886 if (!retain_dentry(dentry))
887 __dput_to_list(dentry, list);
888 spin_unlock(&dentry->d_lock);
891 /* This must be called with d_lock held */
892 static inline void __dget_dlock(struct dentry *dentry)
894 dentry->d_lockref.count++;
897 static inline void __dget(struct dentry *dentry)
899 lockref_get(&dentry->d_lockref);
902 struct dentry *dget_parent(struct dentry *dentry)
909 * Do optimistic parent lookup without any
913 seq = raw_seqcount_begin(&dentry->d_seq);
914 ret = READ_ONCE(dentry->d_parent);
915 gotref = lockref_get_not_zero(&ret->d_lockref);
917 if (likely(gotref)) {
918 if (!read_seqcount_retry(&dentry->d_seq, seq))
925 * Don't need rcu_dereference because we re-check it was correct under
929 ret = dentry->d_parent;
930 spin_lock(&ret->d_lock);
931 if (unlikely(ret != dentry->d_parent)) {
932 spin_unlock(&ret->d_lock);
937 BUG_ON(!ret->d_lockref.count);
938 ret->d_lockref.count++;
939 spin_unlock(&ret->d_lock);
942 EXPORT_SYMBOL(dget_parent);
944 static struct dentry * __d_find_any_alias(struct inode *inode)
946 struct dentry *alias;
948 if (hlist_empty(&inode->i_dentry))
950 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
956 * d_find_any_alias - find any alias for a given inode
957 * @inode: inode to find an alias for
959 * If any aliases exist for the given inode, take and return a
960 * reference for one of them. If no aliases exist, return %NULL.
962 struct dentry *d_find_any_alias(struct inode *inode)
966 spin_lock(&inode->i_lock);
967 de = __d_find_any_alias(inode);
968 spin_unlock(&inode->i_lock);
971 EXPORT_SYMBOL(d_find_any_alias);
974 * d_find_alias - grab a hashed alias of inode
975 * @inode: inode in question
977 * If inode has a hashed alias, or is a directory and has any alias,
978 * acquire the reference to alias and return it. Otherwise return NULL.
979 * Notice that if inode is a directory there can be only one alias and
980 * it can be unhashed only if it has no children, or if it is the root
981 * of a filesystem, or if the directory was renamed and d_revalidate
982 * was the first vfs operation to notice.
984 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
985 * any other hashed alias over that one.
987 static struct dentry *__d_find_alias(struct inode *inode)
989 struct dentry *alias;
991 if (S_ISDIR(inode->i_mode))
992 return __d_find_any_alias(inode);
994 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
995 spin_lock(&alias->d_lock);
996 if (!d_unhashed(alias)) {
998 spin_unlock(&alias->d_lock);
1001 spin_unlock(&alias->d_lock);
1006 struct dentry *d_find_alias(struct inode *inode)
1008 struct dentry *de = NULL;
1010 if (!hlist_empty(&inode->i_dentry)) {
1011 spin_lock(&inode->i_lock);
1012 de = __d_find_alias(inode);
1013 spin_unlock(&inode->i_lock);
1017 EXPORT_SYMBOL(d_find_alias);
1020 * Try to kill dentries associated with this inode.
1021 * WARNING: you must own a reference to inode.
1023 void d_prune_aliases(struct inode *inode)
1025 struct dentry *dentry;
1027 spin_lock(&inode->i_lock);
1028 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1029 spin_lock(&dentry->d_lock);
1030 if (!dentry->d_lockref.count) {
1031 struct dentry *parent = lock_parent(dentry);
1032 if (likely(!dentry->d_lockref.count)) {
1033 __dentry_kill(dentry);
1038 spin_unlock(&parent->d_lock);
1040 spin_unlock(&dentry->d_lock);
1042 spin_unlock(&inode->i_lock);
1044 EXPORT_SYMBOL(d_prune_aliases);
1047 * Lock a dentry from shrink list.
1048 * Called under rcu_read_lock() and dentry->d_lock; the former
1049 * guarantees that nothing we access will be freed under us.
1050 * Note that dentry is *not* protected from concurrent dentry_kill(),
1053 * Return false if dentry has been disrupted or grabbed, leaving
1054 * the caller to kick it off-list. Otherwise, return true and have
1055 * that dentry's inode and parent both locked.
1057 static bool shrink_lock_dentry(struct dentry *dentry)
1059 struct inode *inode;
1060 struct dentry *parent;
1062 if (dentry->d_lockref.count)
1065 inode = dentry->d_inode;
1066 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1067 spin_unlock(&dentry->d_lock);
1068 spin_lock(&inode->i_lock);
1069 spin_lock(&dentry->d_lock);
1070 if (unlikely(dentry->d_lockref.count))
1072 /* changed inode means that somebody had grabbed it */
1073 if (unlikely(inode != dentry->d_inode))
1077 parent = dentry->d_parent;
1078 if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1081 spin_unlock(&dentry->d_lock);
1082 spin_lock(&parent->d_lock);
1083 if (unlikely(parent != dentry->d_parent)) {
1084 spin_unlock(&parent->d_lock);
1085 spin_lock(&dentry->d_lock);
1088 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1089 if (likely(!dentry->d_lockref.count))
1091 spin_unlock(&parent->d_lock);
1094 spin_unlock(&inode->i_lock);
1098 void shrink_dentry_list(struct list_head *list)
1100 while (!list_empty(list)) {
1101 struct dentry *dentry, *parent;
1103 dentry = list_entry(list->prev, struct dentry, d_lru);
1104 spin_lock(&dentry->d_lock);
1106 if (!shrink_lock_dentry(dentry)) {
1107 bool can_free = false;
1109 d_shrink_del(dentry);
1110 if (dentry->d_lockref.count < 0)
1111 can_free = dentry->d_flags & DCACHE_MAY_FREE;
1112 spin_unlock(&dentry->d_lock);
1114 dentry_free(dentry);
1118 d_shrink_del(dentry);
1119 parent = dentry->d_parent;
1120 if (parent != dentry)
1121 __dput_to_list(parent, list);
1122 __dentry_kill(dentry);
1126 static enum lru_status dentry_lru_isolate(struct list_head *item,
1127 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1129 struct list_head *freeable = arg;
1130 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1134 * we are inverting the lru lock/dentry->d_lock here,
1135 * so use a trylock. If we fail to get the lock, just skip
1138 if (!spin_trylock(&dentry->d_lock))
1142 * Referenced dentries are still in use. If they have active
1143 * counts, just remove them from the LRU. Otherwise give them
1144 * another pass through the LRU.
1146 if (dentry->d_lockref.count) {
1147 d_lru_isolate(lru, dentry);
1148 spin_unlock(&dentry->d_lock);
1152 if (dentry->d_flags & DCACHE_REFERENCED) {
1153 dentry->d_flags &= ~DCACHE_REFERENCED;
1154 spin_unlock(&dentry->d_lock);
1157 * The list move itself will be made by the common LRU code. At
1158 * this point, we've dropped the dentry->d_lock but keep the
1159 * lru lock. This is safe to do, since every list movement is
1160 * protected by the lru lock even if both locks are held.
1162 * This is guaranteed by the fact that all LRU management
1163 * functions are intermediated by the LRU API calls like
1164 * list_lru_add and list_lru_del. List movement in this file
1165 * only ever occur through this functions or through callbacks
1166 * like this one, that are called from the LRU API.
1168 * The only exceptions to this are functions like
1169 * shrink_dentry_list, and code that first checks for the
1170 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1171 * operating only with stack provided lists after they are
1172 * properly isolated from the main list. It is thus, always a
1178 d_lru_shrink_move(lru, dentry, freeable);
1179 spin_unlock(&dentry->d_lock);
1185 * prune_dcache_sb - shrink the dcache
1187 * @sc: shrink control, passed to list_lru_shrink_walk()
1189 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1190 * is done when we need more memory and called from the superblock shrinker
1193 * This function may fail to free any resources if all the dentries are in
1196 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1201 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1202 dentry_lru_isolate, &dispose);
1203 shrink_dentry_list(&dispose);
1207 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1208 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1210 struct list_head *freeable = arg;
1211 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1214 * we are inverting the lru lock/dentry->d_lock here,
1215 * so use a trylock. If we fail to get the lock, just skip
1218 if (!spin_trylock(&dentry->d_lock))
1221 d_lru_shrink_move(lru, dentry, freeable);
1222 spin_unlock(&dentry->d_lock);
1229 * shrink_dcache_sb - shrink dcache for a superblock
1232 * Shrink the dcache for the specified super block. This is used to free
1233 * the dcache before unmounting a file system.
1235 void shrink_dcache_sb(struct super_block *sb)
1240 list_lru_walk(&sb->s_dentry_lru,
1241 dentry_lru_isolate_shrink, &dispose, 1024);
1242 shrink_dentry_list(&dispose);
1243 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1245 EXPORT_SYMBOL(shrink_dcache_sb);
1248 * enum d_walk_ret - action to talke during tree walk
1249 * @D_WALK_CONTINUE: contrinue walk
1250 * @D_WALK_QUIT: quit walk
1251 * @D_WALK_NORETRY: quit when retry is needed
1252 * @D_WALK_SKIP: skip this dentry and its children
1262 * d_walk - walk the dentry tree
1263 * @parent: start of walk
1264 * @data: data passed to @enter() and @finish()
1265 * @enter: callback when first entering the dentry
1267 * The @enter() callbacks are called with d_lock held.
1269 static void d_walk(struct dentry *parent, void *data,
1270 enum d_walk_ret (*enter)(void *, struct dentry *))
1272 struct dentry *this_parent;
1273 struct list_head *next;
1275 enum d_walk_ret ret;
1279 read_seqbegin_or_lock(&rename_lock, &seq);
1280 this_parent = parent;
1281 spin_lock(&this_parent->d_lock);
1283 ret = enter(data, this_parent);
1285 case D_WALK_CONTINUE:
1290 case D_WALK_NORETRY:
1295 next = this_parent->d_subdirs.next;
1297 while (next != &this_parent->d_subdirs) {
1298 struct list_head *tmp = next;
1299 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1302 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1305 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1307 ret = enter(data, dentry);
1309 case D_WALK_CONTINUE:
1312 spin_unlock(&dentry->d_lock);
1314 case D_WALK_NORETRY:
1318 spin_unlock(&dentry->d_lock);
1322 if (!list_empty(&dentry->d_subdirs)) {
1323 spin_unlock(&this_parent->d_lock);
1324 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1325 this_parent = dentry;
1326 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1329 spin_unlock(&dentry->d_lock);
1332 * All done at this level ... ascend and resume the search.
1336 if (this_parent != parent) {
1337 struct dentry *child = this_parent;
1338 this_parent = child->d_parent;
1340 spin_unlock(&child->d_lock);
1341 spin_lock(&this_parent->d_lock);
1343 /* might go back up the wrong parent if we have had a rename. */
1344 if (need_seqretry(&rename_lock, seq))
1346 /* go into the first sibling still alive */
1348 next = child->d_child.next;
1349 if (next == &this_parent->d_subdirs)
1351 child = list_entry(next, struct dentry, d_child);
1352 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1356 if (need_seqretry(&rename_lock, seq))
1361 spin_unlock(&this_parent->d_lock);
1362 done_seqretry(&rename_lock, seq);
1366 spin_unlock(&this_parent->d_lock);
1375 struct check_mount {
1376 struct vfsmount *mnt;
1377 unsigned int mounted;
1380 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1382 struct check_mount *info = data;
1383 struct path path = { .mnt = info->mnt, .dentry = dentry };
1385 if (likely(!d_mountpoint(dentry)))
1386 return D_WALK_CONTINUE;
1387 if (__path_is_mountpoint(&path)) {
1391 return D_WALK_CONTINUE;
1395 * path_has_submounts - check for mounts over a dentry in the
1396 * current namespace.
1397 * @parent: path to check.
1399 * Return true if the parent or its subdirectories contain
1400 * a mount point in the current namespace.
1402 int path_has_submounts(const struct path *parent)
1404 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1406 read_seqlock_excl(&mount_lock);
1407 d_walk(parent->dentry, &data, path_check_mount);
1408 read_sequnlock_excl(&mount_lock);
1410 return data.mounted;
1412 EXPORT_SYMBOL(path_has_submounts);
1415 * Called by mount code to set a mountpoint and check if the mountpoint is
1416 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1417 * subtree can become unreachable).
1419 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1420 * this reason take rename_lock and d_lock on dentry and ancestors.
1422 int d_set_mounted(struct dentry *dentry)
1426 write_seqlock(&rename_lock);
1427 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1428 /* Need exclusion wrt. d_invalidate() */
1429 spin_lock(&p->d_lock);
1430 if (unlikely(d_unhashed(p))) {
1431 spin_unlock(&p->d_lock);
1434 spin_unlock(&p->d_lock);
1436 spin_lock(&dentry->d_lock);
1437 if (!d_unlinked(dentry)) {
1439 if (!d_mountpoint(dentry)) {
1440 dentry->d_flags |= DCACHE_MOUNTED;
1444 spin_unlock(&dentry->d_lock);
1446 write_sequnlock(&rename_lock);
1451 * Search the dentry child list of the specified parent,
1452 * and move any unused dentries to the end of the unused
1453 * list for prune_dcache(). We descend to the next level
1454 * whenever the d_subdirs list is non-empty and continue
1457 * It returns zero iff there are no unused children,
1458 * otherwise it returns the number of children moved to
1459 * the end of the unused list. This may not be the total
1460 * number of unused children, because select_parent can
1461 * drop the lock and return early due to latency
1465 struct select_data {
1466 struct dentry *start;
1469 struct dentry *victim;
1471 struct list_head dispose;
1474 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1476 struct select_data *data = _data;
1477 enum d_walk_ret ret = D_WALK_CONTINUE;
1479 if (data->start == dentry)
1482 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1485 if (dentry->d_flags & DCACHE_LRU_LIST)
1487 if (!dentry->d_lockref.count) {
1488 d_shrink_add(dentry, &data->dispose);
1493 * We can return to the caller if we have found some (this
1494 * ensures forward progress). We'll be coming back to find
1497 if (!list_empty(&data->dispose))
1498 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1503 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1505 struct select_data *data = _data;
1506 enum d_walk_ret ret = D_WALK_CONTINUE;
1508 if (data->start == dentry)
1511 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1512 if (!dentry->d_lockref.count) {
1514 data->victim = dentry;
1518 if (dentry->d_flags & DCACHE_LRU_LIST)
1520 if (!dentry->d_lockref.count)
1521 d_shrink_add(dentry, &data->dispose);
1524 * We can return to the caller if we have found some (this
1525 * ensures forward progress). We'll be coming back to find
1528 if (!list_empty(&data->dispose))
1529 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1535 * shrink_dcache_parent - prune dcache
1536 * @parent: parent of entries to prune
1538 * Prune the dcache to remove unused children of the parent dentry.
1540 void shrink_dcache_parent(struct dentry *parent)
1543 struct select_data data = {.start = parent};
1545 INIT_LIST_HEAD(&data.dispose);
1546 d_walk(parent, &data, select_collect);
1548 if (!list_empty(&data.dispose)) {
1549 shrink_dentry_list(&data.dispose);
1557 d_walk(parent, &data, select_collect2);
1559 struct dentry *parent;
1560 spin_lock(&data.victim->d_lock);
1561 if (!shrink_lock_dentry(data.victim)) {
1562 spin_unlock(&data.victim->d_lock);
1566 parent = data.victim->d_parent;
1567 if (parent != data.victim)
1568 __dput_to_list(parent, &data.dispose);
1569 __dentry_kill(data.victim);
1572 if (!list_empty(&data.dispose))
1573 shrink_dentry_list(&data.dispose);
1576 EXPORT_SYMBOL(shrink_dcache_parent);
1578 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1580 /* it has busy descendents; complain about those instead */
1581 if (!list_empty(&dentry->d_subdirs))
1582 return D_WALK_CONTINUE;
1584 /* root with refcount 1 is fine */
1585 if (dentry == _data && dentry->d_lockref.count == 1)
1586 return D_WALK_CONTINUE;
1588 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1589 " still in use (%d) [unmount of %s %s]\n",
1592 dentry->d_inode->i_ino : 0UL,
1594 dentry->d_lockref.count,
1595 dentry->d_sb->s_type->name,
1596 dentry->d_sb->s_id);
1598 return D_WALK_CONTINUE;
1601 static void do_one_tree(struct dentry *dentry)
1603 shrink_dcache_parent(dentry);
1604 d_walk(dentry, dentry, umount_check);
1610 * destroy the dentries attached to a superblock on unmounting
1612 void shrink_dcache_for_umount(struct super_block *sb)
1614 struct dentry *dentry;
1616 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1618 dentry = sb->s_root;
1620 do_one_tree(dentry);
1622 while (!hlist_bl_empty(&sb->s_roots)) {
1623 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1624 do_one_tree(dentry);
1628 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1630 struct dentry **victim = _data;
1631 if (d_mountpoint(dentry)) {
1632 __dget_dlock(dentry);
1636 return D_WALK_CONTINUE;
1640 * d_invalidate - detach submounts, prune dcache, and drop
1641 * @dentry: dentry to invalidate (aka detach, prune and drop)
1643 void d_invalidate(struct dentry *dentry)
1645 bool had_submounts = false;
1646 spin_lock(&dentry->d_lock);
1647 if (d_unhashed(dentry)) {
1648 spin_unlock(&dentry->d_lock);
1652 spin_unlock(&dentry->d_lock);
1654 /* Negative dentries can be dropped without further checks */
1655 if (!dentry->d_inode)
1658 shrink_dcache_parent(dentry);
1660 struct dentry *victim = NULL;
1661 d_walk(dentry, &victim, find_submount);
1664 shrink_dcache_parent(dentry);
1667 had_submounts = true;
1668 detach_mounts(victim);
1672 EXPORT_SYMBOL(d_invalidate);
1675 * __d_alloc - allocate a dcache entry
1676 * @sb: filesystem it will belong to
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.
1684 static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1686 struct dentry *dentry;
1690 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1695 * We guarantee that the inline name is always NUL-terminated.
1696 * This way the memcpy() done by the name switching in rename
1697 * will still always have a NUL at the end, even if we might
1698 * be overwriting an internal NUL character
1700 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1701 if (unlikely(!name)) {
1703 dname = dentry->d_iname;
1704 } else if (name->len > DNAME_INLINE_LEN-1) {
1705 size_t size = offsetof(struct external_name, name[1]);
1706 struct external_name *p = kmalloc(size + name->len,
1707 GFP_KERNEL_ACCOUNT |
1710 kmem_cache_free(dentry_cache, dentry);
1713 atomic_set(&p->u.count, 1);
1716 dname = dentry->d_iname;
1719 dentry->d_name.len = name->len;
1720 dentry->d_name.hash = name->hash;
1721 memcpy(dname, name->name, name->len);
1722 dname[name->len] = 0;
1724 /* Make sure we always see the terminating NUL character */
1725 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1727 dentry->d_lockref.count = 1;
1728 dentry->d_flags = 0;
1729 spin_lock_init(&dentry->d_lock);
1730 seqcount_init(&dentry->d_seq);
1731 dentry->d_inode = NULL;
1732 dentry->d_parent = dentry;
1734 dentry->d_op = NULL;
1735 dentry->d_fsdata = NULL;
1736 INIT_HLIST_BL_NODE(&dentry->d_hash);
1737 INIT_LIST_HEAD(&dentry->d_lru);
1738 INIT_LIST_HEAD(&dentry->d_subdirs);
1739 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1740 INIT_LIST_HEAD(&dentry->d_child);
1741 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1743 if (dentry->d_op && dentry->d_op->d_init) {
1744 err = dentry->d_op->d_init(dentry);
1746 if (dname_external(dentry))
1747 kfree(external_name(dentry));
1748 kmem_cache_free(dentry_cache, dentry);
1753 this_cpu_inc(nr_dentry);
1759 * d_alloc - allocate a dcache entry
1760 * @parent: parent of entry to allocate
1761 * @name: qstr of the name
1763 * Allocates a dentry. It returns %NULL if there is insufficient memory
1764 * available. On a success the dentry is returned. The name passed in is
1765 * copied and the copy passed in may be reused after this call.
1767 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1769 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1772 spin_lock(&parent->d_lock);
1774 * don't need child lock because it is not subject
1775 * to concurrency here
1777 __dget_dlock(parent);
1778 dentry->d_parent = parent;
1779 list_add(&dentry->d_child, &parent->d_subdirs);
1780 spin_unlock(&parent->d_lock);
1784 EXPORT_SYMBOL(d_alloc);
1786 struct dentry *d_alloc_anon(struct super_block *sb)
1788 return __d_alloc(sb, NULL);
1790 EXPORT_SYMBOL(d_alloc_anon);
1792 struct dentry *d_alloc_cursor(struct dentry * parent)
1794 struct dentry *dentry = d_alloc_anon(parent->d_sb);
1796 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1797 dentry->d_parent = dget(parent);
1803 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1804 * @sb: the superblock
1805 * @name: qstr of the name
1807 * For a filesystem that just pins its dentries in memory and never
1808 * performs lookups at all, return an unhashed IS_ROOT dentry.
1809 * This is used for pipes, sockets et.al. - the stuff that should
1810 * never be anyone's children or parents. Unlike all other
1811 * dentries, these will not have RCU delay between dropping the
1812 * last reference and freeing them.
1814 * The only user is alloc_file_pseudo() and that's what should
1815 * be considered a public interface. Don't use directly.
1817 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1819 struct dentry *dentry = __d_alloc(sb, name);
1821 dentry->d_flags |= DCACHE_NORCU;
1825 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1830 q.hash_len = hashlen_string(parent, name);
1831 return d_alloc(parent, &q);
1833 EXPORT_SYMBOL(d_alloc_name);
1835 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1837 WARN_ON_ONCE(dentry->d_op);
1838 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1840 DCACHE_OP_REVALIDATE |
1841 DCACHE_OP_WEAK_REVALIDATE |
1848 dentry->d_flags |= DCACHE_OP_HASH;
1850 dentry->d_flags |= DCACHE_OP_COMPARE;
1851 if (op->d_revalidate)
1852 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1853 if (op->d_weak_revalidate)
1854 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1856 dentry->d_flags |= DCACHE_OP_DELETE;
1858 dentry->d_flags |= DCACHE_OP_PRUNE;
1860 dentry->d_flags |= DCACHE_OP_REAL;
1863 EXPORT_SYMBOL(d_set_d_op);
1867 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1868 * @dentry - The dentry to mark
1870 * Mark a dentry as falling through to the lower layer (as set with
1871 * d_pin_lower()). This flag may be recorded on the medium.
1873 void d_set_fallthru(struct dentry *dentry)
1875 spin_lock(&dentry->d_lock);
1876 dentry->d_flags |= DCACHE_FALLTHRU;
1877 spin_unlock(&dentry->d_lock);
1879 EXPORT_SYMBOL(d_set_fallthru);
1881 static unsigned d_flags_for_inode(struct inode *inode)
1883 unsigned add_flags = DCACHE_REGULAR_TYPE;
1886 return DCACHE_MISS_TYPE;
1888 if (S_ISDIR(inode->i_mode)) {
1889 add_flags = DCACHE_DIRECTORY_TYPE;
1890 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1891 if (unlikely(!inode->i_op->lookup))
1892 add_flags = DCACHE_AUTODIR_TYPE;
1894 inode->i_opflags |= IOP_LOOKUP;
1896 goto type_determined;
1899 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1900 if (unlikely(inode->i_op->get_link)) {
1901 add_flags = DCACHE_SYMLINK_TYPE;
1902 goto type_determined;
1904 inode->i_opflags |= IOP_NOFOLLOW;
1907 if (unlikely(!S_ISREG(inode->i_mode)))
1908 add_flags = DCACHE_SPECIAL_TYPE;
1911 if (unlikely(IS_AUTOMOUNT(inode)))
1912 add_flags |= DCACHE_NEED_AUTOMOUNT;
1916 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1918 unsigned add_flags = d_flags_for_inode(inode);
1919 WARN_ON(d_in_lookup(dentry));
1921 spin_lock(&dentry->d_lock);
1923 * Decrement negative dentry count if it was in the LRU list.
1925 if (dentry->d_flags & DCACHE_LRU_LIST)
1926 this_cpu_dec(nr_dentry_negative);
1927 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1928 raw_write_seqcount_begin(&dentry->d_seq);
1929 __d_set_inode_and_type(dentry, inode, add_flags);
1930 raw_write_seqcount_end(&dentry->d_seq);
1931 fsnotify_update_flags(dentry);
1932 spin_unlock(&dentry->d_lock);
1936 * d_instantiate - fill in inode information for a dentry
1937 * @entry: dentry to complete
1938 * @inode: inode to attach to this dentry
1940 * Fill in inode information in the entry.
1942 * This turns negative dentries into productive full members
1945 * NOTE! This assumes that the inode count has been incremented
1946 * (or otherwise set) by the caller to indicate that it is now
1947 * in use by the dcache.
1950 void d_instantiate(struct dentry *entry, struct inode * inode)
1952 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1954 security_d_instantiate(entry, inode);
1955 spin_lock(&inode->i_lock);
1956 __d_instantiate(entry, inode);
1957 spin_unlock(&inode->i_lock);
1960 EXPORT_SYMBOL(d_instantiate);
1963 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1964 * with lockdep-related part of unlock_new_inode() done before
1965 * anything else. Use that instead of open-coding d_instantiate()/
1966 * unlock_new_inode() combinations.
1968 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1970 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1972 lockdep_annotate_inode_mutex_key(inode);
1973 security_d_instantiate(entry, inode);
1974 spin_lock(&inode->i_lock);
1975 __d_instantiate(entry, inode);
1976 WARN_ON(!(inode->i_state & I_NEW));
1977 inode->i_state &= ~I_NEW & ~I_CREATING;
1979 wake_up_bit(&inode->i_state, __I_NEW);
1980 spin_unlock(&inode->i_lock);
1982 EXPORT_SYMBOL(d_instantiate_new);
1984 struct dentry *d_make_root(struct inode *root_inode)
1986 struct dentry *res = NULL;
1989 res = d_alloc_anon(root_inode->i_sb);
1991 d_instantiate(res, root_inode);
1997 EXPORT_SYMBOL(d_make_root);
1999 static struct dentry *__d_instantiate_anon(struct dentry *dentry,
2000 struct inode *inode,
2006 security_d_instantiate(dentry, inode);
2007 spin_lock(&inode->i_lock);
2008 res = __d_find_any_alias(inode);
2010 spin_unlock(&inode->i_lock);
2015 /* attach a disconnected dentry */
2016 add_flags = d_flags_for_inode(inode);
2019 add_flags |= DCACHE_DISCONNECTED;
2021 spin_lock(&dentry->d_lock);
2022 __d_set_inode_and_type(dentry, inode, add_flags);
2023 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2024 if (!disconnected) {
2025 hlist_bl_lock(&dentry->d_sb->s_roots);
2026 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
2027 hlist_bl_unlock(&dentry->d_sb->s_roots);
2029 spin_unlock(&dentry->d_lock);
2030 spin_unlock(&inode->i_lock);
2039 struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
2041 return __d_instantiate_anon(dentry, inode, true);
2043 EXPORT_SYMBOL(d_instantiate_anon);
2045 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
2051 return ERR_PTR(-ESTALE);
2053 return ERR_CAST(inode);
2055 res = d_find_any_alias(inode);
2059 tmp = d_alloc_anon(inode->i_sb);
2061 res = ERR_PTR(-ENOMEM);
2065 return __d_instantiate_anon(tmp, inode, disconnected);
2073 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2074 * @inode: inode to allocate the dentry for
2076 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2077 * similar open by handle operations. The returned dentry may be anonymous,
2078 * or may have a full name (if the inode was already in the cache).
2080 * When called on a directory inode, we must ensure that the inode only ever
2081 * has one dentry. If a dentry is found, that is returned instead of
2082 * allocating a new one.
2084 * On successful return, the reference to the inode has been transferred
2085 * to the dentry. In case of an error the reference on the inode is released.
2086 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2087 * be passed in and the error will be propagated to the return value,
2088 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2090 struct dentry *d_obtain_alias(struct inode *inode)
2092 return __d_obtain_alias(inode, true);
2094 EXPORT_SYMBOL(d_obtain_alias);
2097 * d_obtain_root - find or allocate a dentry for a given inode
2098 * @inode: inode to allocate the dentry for
2100 * Obtain an IS_ROOT dentry for the root of a filesystem.
2102 * We must ensure that directory inodes only ever have one dentry. If a
2103 * dentry is found, that is returned instead of allocating a new one.
2105 * On successful return, the reference to the inode has been transferred
2106 * to the dentry. In case of an error the reference on the inode is
2107 * released. A %NULL or IS_ERR inode may be passed in and will be the
2108 * error will be propagate to the return value, with a %NULL @inode
2109 * replaced by ERR_PTR(-ESTALE).
2111 struct dentry *d_obtain_root(struct inode *inode)
2113 return __d_obtain_alias(inode, false);
2115 EXPORT_SYMBOL(d_obtain_root);
2118 * d_add_ci - lookup or allocate new dentry with case-exact name
2119 * @inode: the inode case-insensitive lookup has found
2120 * @dentry: the negative dentry that was passed to the parent's lookup func
2121 * @name: the case-exact name to be associated with the returned dentry
2123 * This is to avoid filling the dcache with case-insensitive names to the
2124 * same inode, only the actual correct case is stored in the dcache for
2125 * case-insensitive filesystems.
2127 * For a case-insensitive lookup match and if the the case-exact dentry
2128 * already exists in in the dcache, use it and return it.
2130 * If no entry exists with the exact case name, allocate new dentry with
2131 * the exact case, and return the spliced entry.
2133 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2136 struct dentry *found, *res;
2139 * First check if a dentry matching the name already exists,
2140 * if not go ahead and create it now.
2142 found = d_hash_and_lookup(dentry->d_parent, name);
2147 if (d_in_lookup(dentry)) {
2148 found = d_alloc_parallel(dentry->d_parent, name,
2150 if (IS_ERR(found) || !d_in_lookup(found)) {
2155 found = d_alloc(dentry->d_parent, name);
2158 return ERR_PTR(-ENOMEM);
2161 res = d_splice_alias(inode, found);
2168 EXPORT_SYMBOL(d_add_ci);
2171 static inline bool d_same_name(const struct dentry *dentry,
2172 const struct dentry *parent,
2173 const struct qstr *name)
2175 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2176 if (dentry->d_name.len != name->len)
2178 return dentry_cmp(dentry, name->name, name->len) == 0;
2180 return parent->d_op->d_compare(dentry,
2181 dentry->d_name.len, dentry->d_name.name,
2186 * __d_lookup_rcu - search for a dentry (racy, store-free)
2187 * @parent: parent dentry
2188 * @name: qstr of name we wish to find
2189 * @seqp: returns d_seq value at the point where the dentry was found
2190 * Returns: dentry, or NULL
2192 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2193 * resolution (store-free path walking) design described in
2194 * Documentation/filesystems/path-lookup.txt.
2196 * This is not to be used outside core vfs.
2198 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2199 * held, and rcu_read_lock held. The returned dentry must not be stored into
2200 * without taking d_lock and checking d_seq sequence count against @seq
2203 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2206 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2207 * the returned dentry, so long as its parent's seqlock is checked after the
2208 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2209 * is formed, giving integrity down the path walk.
2211 * NOTE! The caller *has* to check the resulting dentry against the sequence
2212 * number we've returned before using any of the resulting dentry state!
2214 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2215 const struct qstr *name,
2218 u64 hashlen = name->hash_len;
2219 const unsigned char *str = name->name;
2220 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2221 struct hlist_bl_node *node;
2222 struct dentry *dentry;
2225 * Note: There is significant duplication with __d_lookup_rcu which is
2226 * required to prevent single threaded performance regressions
2227 * especially on architectures where smp_rmb (in seqcounts) are costly.
2228 * Keep the two functions in sync.
2232 * The hash list is protected using RCU.
2234 * Carefully use d_seq when comparing a candidate dentry, to avoid
2235 * races with d_move().
2237 * It is possible that concurrent renames can mess up our list
2238 * walk here and result in missing our dentry, resulting in the
2239 * false-negative result. d_lookup() protects against concurrent
2240 * renames using rename_lock seqlock.
2242 * See Documentation/filesystems/path-lookup.txt for more details.
2244 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2249 * The dentry sequence count protects us from concurrent
2250 * renames, and thus protects parent and name fields.
2252 * The caller must perform a seqcount check in order
2253 * to do anything useful with the returned dentry.
2255 * NOTE! We do a "raw" seqcount_begin here. That means that
2256 * we don't wait for the sequence count to stabilize if it
2257 * is in the middle of a sequence change. If we do the slow
2258 * dentry compare, we will do seqretries until it is stable,
2259 * and if we end up with a successful lookup, we actually
2260 * want to exit RCU lookup anyway.
2262 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2263 * we are still guaranteed NUL-termination of ->d_name.name.
2265 seq = raw_seqcount_begin(&dentry->d_seq);
2266 if (dentry->d_parent != parent)
2268 if (d_unhashed(dentry))
2271 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2274 if (dentry->d_name.hash != hashlen_hash(hashlen))
2276 tlen = dentry->d_name.len;
2277 tname = dentry->d_name.name;
2278 /* we want a consistent (name,len) pair */
2279 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2283 if (parent->d_op->d_compare(dentry,
2284 tlen, tname, name) != 0)
2287 if (dentry->d_name.hash_len != hashlen)
2289 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2299 * d_lookup - search for a dentry
2300 * @parent: parent dentry
2301 * @name: qstr of name we wish to find
2302 * Returns: dentry, or NULL
2304 * d_lookup searches the children of the parent dentry for the name in
2305 * question. If the dentry is found its reference count is incremented and the
2306 * dentry is returned. The caller must use dput to free the entry when it has
2307 * finished using it. %NULL is returned if the dentry does not exist.
2309 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2311 struct dentry *dentry;
2315 seq = read_seqbegin(&rename_lock);
2316 dentry = __d_lookup(parent, name);
2319 } while (read_seqretry(&rename_lock, seq));
2322 EXPORT_SYMBOL(d_lookup);
2325 * __d_lookup - search for a dentry (racy)
2326 * @parent: parent dentry
2327 * @name: qstr of name we wish to find
2328 * Returns: dentry, or NULL
2330 * __d_lookup is like d_lookup, however it may (rarely) return a
2331 * false-negative result due to unrelated rename activity.
2333 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2334 * however it must be used carefully, eg. with a following d_lookup in
2335 * the case of failure.
2337 * __d_lookup callers must be commented.
2339 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2341 unsigned int hash = name->hash;
2342 struct hlist_bl_head *b = d_hash(hash);
2343 struct hlist_bl_node *node;
2344 struct dentry *found = NULL;
2345 struct dentry *dentry;
2348 * Note: There is significant duplication with __d_lookup_rcu which is
2349 * required to prevent single threaded performance regressions
2350 * especially on architectures where smp_rmb (in seqcounts) are costly.
2351 * Keep the two functions in sync.
2355 * The hash list is protected using RCU.
2357 * Take d_lock when comparing a candidate dentry, to avoid races
2360 * It is possible that concurrent renames can mess up our list
2361 * walk here and result in missing our dentry, resulting in the
2362 * false-negative result. d_lookup() protects against concurrent
2363 * renames using rename_lock seqlock.
2365 * See Documentation/filesystems/path-lookup.txt for more details.
2369 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2371 if (dentry->d_name.hash != hash)
2374 spin_lock(&dentry->d_lock);
2375 if (dentry->d_parent != parent)
2377 if (d_unhashed(dentry))
2380 if (!d_same_name(dentry, parent, name))
2383 dentry->d_lockref.count++;
2385 spin_unlock(&dentry->d_lock);
2388 spin_unlock(&dentry->d_lock);
2396 * d_hash_and_lookup - hash the qstr then search for a dentry
2397 * @dir: Directory to search in
2398 * @name: qstr of name we wish to find
2400 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2402 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2405 * Check for a fs-specific hash function. Note that we must
2406 * calculate the standard hash first, as the d_op->d_hash()
2407 * routine may choose to leave the hash value unchanged.
2409 name->hash = full_name_hash(dir, name->name, name->len);
2410 if (dir->d_flags & DCACHE_OP_HASH) {
2411 int err = dir->d_op->d_hash(dir, name);
2412 if (unlikely(err < 0))
2413 return ERR_PTR(err);
2415 return d_lookup(dir, name);
2417 EXPORT_SYMBOL(d_hash_and_lookup);
2420 * When a file is deleted, we have two options:
2421 * - turn this dentry into a negative dentry
2422 * - unhash this dentry and free it.
2424 * Usually, we want to just turn this into
2425 * a negative dentry, but if anybody else is
2426 * currently using the dentry or the inode
2427 * we can't do that and we fall back on removing
2428 * it from the hash queues and waiting for
2429 * it to be deleted later when it has no users
2433 * d_delete - delete a dentry
2434 * @dentry: The dentry to delete
2436 * Turn the dentry into a negative dentry if possible, otherwise
2437 * remove it from the hash queues so it can be deleted later
2440 void d_delete(struct dentry * dentry)
2442 struct inode *inode = dentry->d_inode;
2444 spin_lock(&inode->i_lock);
2445 spin_lock(&dentry->d_lock);
2447 * Are we the only user?
2449 if (dentry->d_lockref.count == 1) {
2450 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2451 dentry_unlink_inode(dentry);
2454 spin_unlock(&dentry->d_lock);
2455 spin_unlock(&inode->i_lock);
2458 EXPORT_SYMBOL(d_delete);
2460 static void __d_rehash(struct dentry *entry)
2462 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2465 hlist_bl_add_head_rcu(&entry->d_hash, b);
2470 * d_rehash - add an entry back to the hash
2471 * @entry: dentry to add to the hash
2473 * Adds a dentry to the hash according to its name.
2476 void d_rehash(struct dentry * entry)
2478 spin_lock(&entry->d_lock);
2480 spin_unlock(&entry->d_lock);
2482 EXPORT_SYMBOL(d_rehash);
2484 static inline unsigned start_dir_add(struct inode *dir)
2488 unsigned n = dir->i_dir_seq;
2489 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2495 static inline void end_dir_add(struct inode *dir, unsigned n)
2497 smp_store_release(&dir->i_dir_seq, n + 2);
2500 static void d_wait_lookup(struct dentry *dentry)
2502 if (d_in_lookup(dentry)) {
2503 DECLARE_WAITQUEUE(wait, current);
2504 add_wait_queue(dentry->d_wait, &wait);
2506 set_current_state(TASK_UNINTERRUPTIBLE);
2507 spin_unlock(&dentry->d_lock);
2509 spin_lock(&dentry->d_lock);
2510 } while (d_in_lookup(dentry));
2514 struct dentry *d_alloc_parallel(struct dentry *parent,
2515 const struct qstr *name,
2516 wait_queue_head_t *wq)
2518 unsigned int hash = name->hash;
2519 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2520 struct hlist_bl_node *node;
2521 struct dentry *new = d_alloc(parent, name);
2522 struct dentry *dentry;
2523 unsigned seq, r_seq, d_seq;
2526 return ERR_PTR(-ENOMEM);
2530 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2531 r_seq = read_seqbegin(&rename_lock);
2532 dentry = __d_lookup_rcu(parent, name, &d_seq);
2533 if (unlikely(dentry)) {
2534 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2538 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2547 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2552 if (unlikely(seq & 1)) {
2558 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2564 * No changes for the parent since the beginning of d_lookup().
2565 * Since all removals from the chain happen with hlist_bl_lock(),
2566 * any potential in-lookup matches are going to stay here until
2567 * we unlock the chain. All fields are stable in everything
2570 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2571 if (dentry->d_name.hash != hash)
2573 if (dentry->d_parent != parent)
2575 if (!d_same_name(dentry, parent, name))
2578 /* now we can try to grab a reference */
2579 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2586 * somebody is likely to be still doing lookup for it;
2587 * wait for them to finish
2589 spin_lock(&dentry->d_lock);
2590 d_wait_lookup(dentry);
2592 * it's not in-lookup anymore; in principle we should repeat
2593 * everything from dcache lookup, but it's likely to be what
2594 * d_lookup() would've found anyway. If it is, just return it;
2595 * otherwise we really have to repeat the whole thing.
2597 if (unlikely(dentry->d_name.hash != hash))
2599 if (unlikely(dentry->d_parent != parent))
2601 if (unlikely(d_unhashed(dentry)))
2603 if (unlikely(!d_same_name(dentry, parent, name)))
2605 /* OK, it *is* a hashed match; return it */
2606 spin_unlock(&dentry->d_lock);
2611 /* we can't take ->d_lock here; it's OK, though. */
2612 new->d_flags |= DCACHE_PAR_LOOKUP;
2614 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2618 spin_unlock(&dentry->d_lock);
2622 EXPORT_SYMBOL(d_alloc_parallel);
2624 void __d_lookup_done(struct dentry *dentry)
2626 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2627 dentry->d_name.hash);
2629 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2630 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2631 wake_up_all(dentry->d_wait);
2632 dentry->d_wait = NULL;
2634 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2635 INIT_LIST_HEAD(&dentry->d_lru);
2637 EXPORT_SYMBOL(__d_lookup_done);
2639 /* inode->i_lock held if inode is non-NULL */
2641 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2643 struct inode *dir = NULL;
2645 spin_lock(&dentry->d_lock);
2646 if (unlikely(d_in_lookup(dentry))) {
2647 dir = dentry->d_parent->d_inode;
2648 n = start_dir_add(dir);
2649 __d_lookup_done(dentry);
2652 unsigned add_flags = d_flags_for_inode(inode);
2653 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2654 raw_write_seqcount_begin(&dentry->d_seq);
2655 __d_set_inode_and_type(dentry, inode, add_flags);
2656 raw_write_seqcount_end(&dentry->d_seq);
2657 fsnotify_update_flags(dentry);
2661 end_dir_add(dir, n);
2662 spin_unlock(&dentry->d_lock);
2664 spin_unlock(&inode->i_lock);
2668 * d_add - add dentry to hash queues
2669 * @entry: dentry to add
2670 * @inode: The inode to attach to this dentry
2672 * This adds the entry to the hash queues and initializes @inode.
2673 * The entry was actually filled in earlier during d_alloc().
2676 void d_add(struct dentry *entry, struct inode *inode)
2679 security_d_instantiate(entry, inode);
2680 spin_lock(&inode->i_lock);
2682 __d_add(entry, inode);
2684 EXPORT_SYMBOL(d_add);
2687 * d_exact_alias - find and hash an exact unhashed alias
2688 * @entry: dentry to add
2689 * @inode: The inode to go with this dentry
2691 * If an unhashed dentry with the same name/parent and desired
2692 * inode already exists, hash and return it. Otherwise, return
2695 * Parent directory should be locked.
2697 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2699 struct dentry *alias;
2700 unsigned int hash = entry->d_name.hash;
2702 spin_lock(&inode->i_lock);
2703 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2705 * Don't need alias->d_lock here, because aliases with
2706 * d_parent == entry->d_parent are not subject to name or
2707 * parent changes, because the parent inode i_mutex is held.
2709 if (alias->d_name.hash != hash)
2711 if (alias->d_parent != entry->d_parent)
2713 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2715 spin_lock(&alias->d_lock);
2716 if (!d_unhashed(alias)) {
2717 spin_unlock(&alias->d_lock);
2720 __dget_dlock(alias);
2722 spin_unlock(&alias->d_lock);
2724 spin_unlock(&inode->i_lock);
2727 spin_unlock(&inode->i_lock);
2730 EXPORT_SYMBOL(d_exact_alias);
2732 static void swap_names(struct dentry *dentry, struct dentry *target)
2734 if (unlikely(dname_external(target))) {
2735 if (unlikely(dname_external(dentry))) {
2737 * Both external: swap the pointers
2739 swap(target->d_name.name, dentry->d_name.name);
2742 * dentry:internal, target:external. Steal target's
2743 * storage and make target internal.
2745 memcpy(target->d_iname, dentry->d_name.name,
2746 dentry->d_name.len + 1);
2747 dentry->d_name.name = target->d_name.name;
2748 target->d_name.name = target->d_iname;
2751 if (unlikely(dname_external(dentry))) {
2753 * dentry:external, target:internal. Give dentry's
2754 * storage to target and make dentry internal
2756 memcpy(dentry->d_iname, target->d_name.name,
2757 target->d_name.len + 1);
2758 target->d_name.name = dentry->d_name.name;
2759 dentry->d_name.name = dentry->d_iname;
2762 * Both are internal.
2765 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2766 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2767 swap(((long *) &dentry->d_iname)[i],
2768 ((long *) &target->d_iname)[i]);
2772 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2775 static void copy_name(struct dentry *dentry, struct dentry *target)
2777 struct external_name *old_name = NULL;
2778 if (unlikely(dname_external(dentry)))
2779 old_name = external_name(dentry);
2780 if (unlikely(dname_external(target))) {
2781 atomic_inc(&external_name(target)->u.count);
2782 dentry->d_name = target->d_name;
2784 memcpy(dentry->d_iname, target->d_name.name,
2785 target->d_name.len + 1);
2786 dentry->d_name.name = dentry->d_iname;
2787 dentry->d_name.hash_len = target->d_name.hash_len;
2789 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2790 kfree_rcu(old_name, u.head);
2794 * __d_move - move a dentry
2795 * @dentry: entry to move
2796 * @target: new dentry
2797 * @exchange: exchange the two dentries
2799 * Update the dcache to reflect the move of a file name. Negative
2800 * dcache entries should not be moved in this way. Caller must hold
2801 * rename_lock, the i_mutex of the source and target directories,
2802 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2804 static void __d_move(struct dentry *dentry, struct dentry *target,
2807 struct dentry *old_parent, *p;
2808 struct inode *dir = NULL;
2811 WARN_ON(!dentry->d_inode);
2812 if (WARN_ON(dentry == target))
2815 BUG_ON(d_ancestor(target, dentry));
2816 old_parent = dentry->d_parent;
2817 p = d_ancestor(old_parent, target);
2818 if (IS_ROOT(dentry)) {
2820 spin_lock(&target->d_parent->d_lock);
2822 /* target is not a descendent of dentry->d_parent */
2823 spin_lock(&target->d_parent->d_lock);
2824 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2826 BUG_ON(p == dentry);
2827 spin_lock(&old_parent->d_lock);
2829 spin_lock_nested(&target->d_parent->d_lock,
2830 DENTRY_D_LOCK_NESTED);
2832 spin_lock_nested(&dentry->d_lock, 2);
2833 spin_lock_nested(&target->d_lock, 3);
2835 if (unlikely(d_in_lookup(target))) {
2836 dir = target->d_parent->d_inode;
2837 n = start_dir_add(dir);
2838 __d_lookup_done(target);
2841 write_seqcount_begin(&dentry->d_seq);
2842 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2845 if (!d_unhashed(dentry))
2847 if (!d_unhashed(target))
2850 /* ... and switch them in the tree */
2851 dentry->d_parent = target->d_parent;
2853 copy_name(dentry, target);
2854 target->d_hash.pprev = NULL;
2855 dentry->d_parent->d_lockref.count++;
2856 if (dentry != old_parent) /* wasn't IS_ROOT */
2857 WARN_ON(!--old_parent->d_lockref.count);
2859 target->d_parent = old_parent;
2860 swap_names(dentry, target);
2861 list_move(&target->d_child, &target->d_parent->d_subdirs);
2863 fsnotify_update_flags(target);
2865 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2867 fsnotify_update_flags(dentry);
2868 fscrypt_handle_d_move(dentry);
2870 write_seqcount_end(&target->d_seq);
2871 write_seqcount_end(&dentry->d_seq);
2874 end_dir_add(dir, n);
2876 if (dentry->d_parent != old_parent)
2877 spin_unlock(&dentry->d_parent->d_lock);
2878 if (dentry != old_parent)
2879 spin_unlock(&old_parent->d_lock);
2880 spin_unlock(&target->d_lock);
2881 spin_unlock(&dentry->d_lock);
2885 * d_move - move a dentry
2886 * @dentry: entry to move
2887 * @target: new dentry
2889 * Update the dcache to reflect the move of a file name. Negative
2890 * dcache entries should not be moved in this way. See the locking
2891 * requirements for __d_move.
2893 void d_move(struct dentry *dentry, struct dentry *target)
2895 write_seqlock(&rename_lock);
2896 __d_move(dentry, target, false);
2897 write_sequnlock(&rename_lock);
2899 EXPORT_SYMBOL(d_move);
2902 * d_exchange - exchange two dentries
2903 * @dentry1: first dentry
2904 * @dentry2: second dentry
2906 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2908 write_seqlock(&rename_lock);
2910 WARN_ON(!dentry1->d_inode);
2911 WARN_ON(!dentry2->d_inode);
2912 WARN_ON(IS_ROOT(dentry1));
2913 WARN_ON(IS_ROOT(dentry2));
2915 __d_move(dentry1, dentry2, true);
2917 write_sequnlock(&rename_lock);
2921 * d_ancestor - search for an ancestor
2922 * @p1: ancestor dentry
2925 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2926 * an ancestor of p2, else NULL.
2928 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2932 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2933 if (p->d_parent == p1)
2940 * This helper attempts to cope with remotely renamed directories
2942 * It assumes that the caller is already holding
2943 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2945 * Note: If ever the locking in lock_rename() changes, then please
2946 * remember to update this too...
2948 static int __d_unalias(struct inode *inode,
2949 struct dentry *dentry, struct dentry *alias)
2951 struct mutex *m1 = NULL;
2952 struct rw_semaphore *m2 = NULL;
2955 /* If alias and dentry share a parent, then no extra locks required */
2956 if (alias->d_parent == dentry->d_parent)
2959 /* See lock_rename() */
2960 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2962 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2963 if (!inode_trylock_shared(alias->d_parent->d_inode))
2965 m2 = &alias->d_parent->d_inode->i_rwsem;
2967 __d_move(alias, dentry, false);
2978 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2979 * @inode: the inode which may have a disconnected dentry
2980 * @dentry: a negative dentry which we want to point to the inode.
2982 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2983 * place of the given dentry and return it, else simply d_add the inode
2984 * to the dentry and return NULL.
2986 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2987 * we should error out: directories can't have multiple aliases.
2989 * This is needed in the lookup routine of any filesystem that is exportable
2990 * (via knfsd) so that we can build dcache paths to directories effectively.
2992 * If a dentry was found and moved, then it is returned. Otherwise NULL
2993 * is returned. This matches the expected return value of ->lookup.
2995 * Cluster filesystems may call this function with a negative, hashed dentry.
2996 * In that case, we know that the inode will be a regular file, and also this
2997 * will only occur during atomic_open. So we need to check for the dentry
2998 * being already hashed only in the final case.
3000 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3003 return ERR_CAST(inode);
3005 BUG_ON(!d_unhashed(dentry));
3010 security_d_instantiate(dentry, inode);
3011 spin_lock(&inode->i_lock);
3012 if (S_ISDIR(inode->i_mode)) {
3013 struct dentry *new = __d_find_any_alias(inode);
3014 if (unlikely(new)) {
3015 /* The reference to new ensures it remains an alias */
3016 spin_unlock(&inode->i_lock);
3017 write_seqlock(&rename_lock);
3018 if (unlikely(d_ancestor(new, dentry))) {
3019 write_sequnlock(&rename_lock);
3021 new = ERR_PTR(-ELOOP);
3022 pr_warn_ratelimited(
3023 "VFS: Lookup of '%s' in %s %s"
3024 " would have caused loop\n",
3025 dentry->d_name.name,
3026 inode->i_sb->s_type->name,
3028 } else if (!IS_ROOT(new)) {
3029 struct dentry *old_parent = dget(new->d_parent);
3030 int err = __d_unalias(inode, dentry, new);
3031 write_sequnlock(&rename_lock);
3038 __d_move(new, dentry, false);
3039 write_sequnlock(&rename_lock);
3046 __d_add(dentry, inode);
3049 EXPORT_SYMBOL(d_splice_alias);
3052 * Test whether new_dentry is a subdirectory of old_dentry.
3054 * Trivially implemented using the dcache structure
3058 * is_subdir - is new dentry a subdirectory of old_dentry
3059 * @new_dentry: new dentry
3060 * @old_dentry: old dentry
3062 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3063 * Returns false otherwise.
3064 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3067 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3072 if (new_dentry == old_dentry)
3076 /* for restarting inner loop in case of seq retry */
3077 seq = read_seqbegin(&rename_lock);
3079 * Need rcu_readlock to protect against the d_parent trashing
3083 if (d_ancestor(old_dentry, new_dentry))
3088 } while (read_seqretry(&rename_lock, seq));
3092 EXPORT_SYMBOL(is_subdir);
3094 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3096 struct dentry *root = data;
3097 if (dentry != root) {
3098 if (d_unhashed(dentry) || !dentry->d_inode)
3101 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3102 dentry->d_flags |= DCACHE_GENOCIDE;
3103 dentry->d_lockref.count--;
3106 return D_WALK_CONTINUE;
3109 void d_genocide(struct dentry *parent)
3111 d_walk(parent, parent, d_genocide_kill);
3114 EXPORT_SYMBOL(d_genocide);
3116 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3118 inode_dec_link_count(inode);
3119 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3120 !hlist_unhashed(&dentry->d_u.d_alias) ||
3121 !d_unlinked(dentry));
3122 spin_lock(&dentry->d_parent->d_lock);
3123 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3124 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3125 (unsigned long long)inode->i_ino);
3126 spin_unlock(&dentry->d_lock);
3127 spin_unlock(&dentry->d_parent->d_lock);
3128 d_instantiate(dentry, inode);
3130 EXPORT_SYMBOL(d_tmpfile);
3132 static __initdata unsigned long dhash_entries;
3133 static int __init set_dhash_entries(char *str)
3137 dhash_entries = simple_strtoul(str, &str, 0);
3140 __setup("dhash_entries=", set_dhash_entries);
3142 static void __init dcache_init_early(void)
3144 /* If hashes are distributed across NUMA nodes, defer
3145 * hash allocation until vmalloc space is available.
3151 alloc_large_system_hash("Dentry cache",
3152 sizeof(struct hlist_bl_head),
3155 HASH_EARLY | HASH_ZERO,
3160 d_hash_shift = 32 - d_hash_shift;
3163 static void __init dcache_init(void)
3166 * A constructor could be added for stable state like the lists,
3167 * but it is probably not worth it because of the cache nature
3170 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3171 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3174 /* Hash may have been set up in dcache_init_early */
3179 alloc_large_system_hash("Dentry cache",
3180 sizeof(struct hlist_bl_head),
3188 d_hash_shift = 32 - d_hash_shift;
3191 /* SLAB cache for __getname() consumers */
3192 struct kmem_cache *names_cachep __read_mostly;
3193 EXPORT_SYMBOL(names_cachep);
3195 void __init vfs_caches_init_early(void)
3199 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3200 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3202 dcache_init_early();
3206 void __init vfs_caches_init(void)
3208 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3209 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3214 files_maxfiles_init();