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 *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))
651 if (unlikely(dentry->d_flags & DCACHE_DONTCACHE))
654 /* retain; LRU fodder */
655 dentry->d_lockref.count--;
656 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
658 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
659 dentry->d_flags |= DCACHE_REFERENCED;
663 void d_mark_dontcache(struct inode *inode)
667 spin_lock(&inode->i_lock);
668 hlist_for_each_entry(de, &inode->i_dentry, d_u.d_alias) {
669 spin_lock(&de->d_lock);
670 de->d_flags |= DCACHE_DONTCACHE;
671 spin_unlock(&de->d_lock);
673 inode->i_state |= I_DONTCACHE;
674 spin_unlock(&inode->i_lock);
676 EXPORT_SYMBOL(d_mark_dontcache);
679 * Finish off a dentry we've decided to kill.
680 * dentry->d_lock must be held, returns with it unlocked.
681 * Returns dentry requiring refcount drop, or NULL if we're done.
683 static struct dentry *dentry_kill(struct dentry *dentry)
684 __releases(dentry->d_lock)
686 struct inode *inode = dentry->d_inode;
687 struct dentry *parent = NULL;
689 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
692 if (!IS_ROOT(dentry)) {
693 parent = dentry->d_parent;
694 if (unlikely(!spin_trylock(&parent->d_lock))) {
695 parent = __lock_parent(dentry);
696 if (likely(inode || !dentry->d_inode))
698 /* negative that became positive */
700 spin_unlock(&parent->d_lock);
701 inode = dentry->d_inode;
705 __dentry_kill(dentry);
709 spin_unlock(&dentry->d_lock);
710 spin_lock(&inode->i_lock);
711 spin_lock(&dentry->d_lock);
712 parent = lock_parent(dentry);
714 if (unlikely(dentry->d_lockref.count != 1)) {
715 dentry->d_lockref.count--;
716 } else if (likely(!retain_dentry(dentry))) {
717 __dentry_kill(dentry);
720 /* we are keeping it, after all */
722 spin_unlock(&inode->i_lock);
724 spin_unlock(&parent->d_lock);
725 spin_unlock(&dentry->d_lock);
730 * Try to do a lockless dput(), and return whether that was successful.
732 * If unsuccessful, we return false, having already taken the dentry lock.
734 * The caller needs to hold the RCU read lock, so that the dentry is
735 * guaranteed to stay around even if the refcount goes down to zero!
737 static inline bool fast_dput(struct dentry *dentry)
740 unsigned int d_flags;
743 * If we have a d_op->d_delete() operation, we sould not
744 * let the dentry count go to zero, so use "put_or_lock".
746 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
747 return lockref_put_or_lock(&dentry->d_lockref);
750 * .. otherwise, we can try to just decrement the
751 * lockref optimistically.
753 ret = lockref_put_return(&dentry->d_lockref);
756 * If the lockref_put_return() failed due to the lock being held
757 * by somebody else, the fast path has failed. We will need to
758 * get the lock, and then check the count again.
760 if (unlikely(ret < 0)) {
761 spin_lock(&dentry->d_lock);
762 if (WARN_ON_ONCE(dentry->d_lockref.count <= 0)) {
763 spin_unlock(&dentry->d_lock);
766 dentry->d_lockref.count--;
771 * If we weren't the last ref, we're done.
777 * Careful, careful. The reference count went down
778 * to zero, but we don't hold the dentry lock, so
779 * somebody else could get it again, and do another
780 * dput(), and we need to not race with that.
782 * However, there is a very special and common case
783 * where we don't care, because there is nothing to
784 * do: the dentry is still hashed, it does not have
785 * a 'delete' op, and it's referenced and already on
788 * NOTE! Since we aren't locked, these values are
789 * not "stable". However, it is sufficient that at
790 * some point after we dropped the reference the
791 * dentry was hashed and the flags had the proper
792 * value. Other dentry users may have re-gotten
793 * a reference to the dentry and change that, but
794 * our work is done - we can leave the dentry
795 * around with a zero refcount.
798 d_flags = READ_ONCE(dentry->d_flags);
799 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
801 /* Nothing to do? Dropping the reference was all we needed? */
802 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
806 * Not the fast normal case? Get the lock. We've already decremented
807 * the refcount, but we'll need to re-check the situation after
810 spin_lock(&dentry->d_lock);
813 * Did somebody else grab a reference to it in the meantime, and
814 * we're no longer the last user after all? Alternatively, somebody
815 * else could have killed it and marked it dead. Either way, we
816 * don't need to do anything else.
819 if (dentry->d_lockref.count) {
820 spin_unlock(&dentry->d_lock);
825 * Re-get the reference we optimistically dropped. We hold the
826 * lock, and we just tested that it was zero, so we can just
829 dentry->d_lockref.count = 1;
837 * This is complicated by the fact that we do not want to put
838 * dentries that are no longer on any hash chain on the unused
839 * list: we'd much rather just get rid of them immediately.
841 * However, that implies that we have to traverse the dentry
842 * tree upwards to the parents which might _also_ now be
843 * scheduled for deletion (it may have been only waiting for
844 * its last child to go away).
846 * This tail recursion is done by hand as we don't want to depend
847 * on the compiler to always get this right (gcc generally doesn't).
848 * Real recursion would eat up our stack space.
852 * dput - release a dentry
853 * @dentry: dentry to release
855 * Release a dentry. This will drop the usage count and if appropriate
856 * call the dentry unlink method as well as removing it from the queues and
857 * releasing its resources. If the parent dentries were scheduled for release
858 * they too may now get deleted.
860 void dput(struct dentry *dentry)
866 if (likely(fast_dput(dentry))) {
871 /* Slow case: now with the dentry lock held */
874 if (likely(retain_dentry(dentry))) {
875 spin_unlock(&dentry->d_lock);
879 dentry = dentry_kill(dentry);
884 static void __dput_to_list(struct dentry *dentry, struct list_head *list)
885 __must_hold(&dentry->d_lock)
887 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
888 /* let the owner of the list it's on deal with it */
889 --dentry->d_lockref.count;
891 if (dentry->d_flags & DCACHE_LRU_LIST)
893 if (!--dentry->d_lockref.count)
894 d_shrink_add(dentry, list);
898 void dput_to_list(struct dentry *dentry, struct list_head *list)
901 if (likely(fast_dput(dentry))) {
906 if (!retain_dentry(dentry))
907 __dput_to_list(dentry, list);
908 spin_unlock(&dentry->d_lock);
911 /* This must be called with d_lock held */
912 static inline void __dget_dlock(struct dentry *dentry)
914 dentry->d_lockref.count++;
917 static inline void __dget(struct dentry *dentry)
919 lockref_get(&dentry->d_lockref);
922 struct dentry *dget_parent(struct dentry *dentry)
929 * Do optimistic parent lookup without any
933 seq = raw_seqcount_begin(&dentry->d_seq);
934 ret = READ_ONCE(dentry->d_parent);
935 gotref = lockref_get_not_zero(&ret->d_lockref);
937 if (likely(gotref)) {
938 if (!read_seqcount_retry(&dentry->d_seq, seq))
945 * Don't need rcu_dereference because we re-check it was correct under
949 ret = dentry->d_parent;
950 spin_lock(&ret->d_lock);
951 if (unlikely(ret != dentry->d_parent)) {
952 spin_unlock(&ret->d_lock);
957 BUG_ON(!ret->d_lockref.count);
958 ret->d_lockref.count++;
959 spin_unlock(&ret->d_lock);
962 EXPORT_SYMBOL(dget_parent);
964 static struct dentry * __d_find_any_alias(struct inode *inode)
966 struct dentry *alias;
968 if (hlist_empty(&inode->i_dentry))
970 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
976 * d_find_any_alias - find any alias for a given inode
977 * @inode: inode to find an alias for
979 * If any aliases exist for the given inode, take and return a
980 * reference for one of them. If no aliases exist, return %NULL.
982 struct dentry *d_find_any_alias(struct inode *inode)
986 spin_lock(&inode->i_lock);
987 de = __d_find_any_alias(inode);
988 spin_unlock(&inode->i_lock);
991 EXPORT_SYMBOL(d_find_any_alias);
994 * d_find_alias - grab a hashed alias of inode
995 * @inode: inode in question
997 * If inode has a hashed alias, or is a directory and has any alias,
998 * acquire the reference to alias and return it. Otherwise return NULL.
999 * Notice that if inode is a directory there can be only one alias and
1000 * it can be unhashed only if it has no children, or if it is the root
1001 * of a filesystem, or if the directory was renamed and d_revalidate
1002 * was the first vfs operation to notice.
1004 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1005 * any other hashed alias over that one.
1007 static struct dentry *__d_find_alias(struct inode *inode)
1009 struct dentry *alias;
1011 if (S_ISDIR(inode->i_mode))
1012 return __d_find_any_alias(inode);
1014 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1015 spin_lock(&alias->d_lock);
1016 if (!d_unhashed(alias)) {
1017 __dget_dlock(alias);
1018 spin_unlock(&alias->d_lock);
1021 spin_unlock(&alias->d_lock);
1026 struct dentry *d_find_alias(struct inode *inode)
1028 struct dentry *de = NULL;
1030 if (!hlist_empty(&inode->i_dentry)) {
1031 spin_lock(&inode->i_lock);
1032 de = __d_find_alias(inode);
1033 spin_unlock(&inode->i_lock);
1037 EXPORT_SYMBOL(d_find_alias);
1040 * Try to kill dentries associated with this inode.
1041 * WARNING: you must own a reference to inode.
1043 void d_prune_aliases(struct inode *inode)
1045 struct dentry *dentry;
1047 spin_lock(&inode->i_lock);
1048 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
1049 spin_lock(&dentry->d_lock);
1050 if (!dentry->d_lockref.count) {
1051 struct dentry *parent = lock_parent(dentry);
1052 if (likely(!dentry->d_lockref.count)) {
1053 __dentry_kill(dentry);
1058 spin_unlock(&parent->d_lock);
1060 spin_unlock(&dentry->d_lock);
1062 spin_unlock(&inode->i_lock);
1064 EXPORT_SYMBOL(d_prune_aliases);
1067 * Lock a dentry from shrink list.
1068 * Called under rcu_read_lock() and dentry->d_lock; the former
1069 * guarantees that nothing we access will be freed under us.
1070 * Note that dentry is *not* protected from concurrent dentry_kill(),
1073 * Return false if dentry has been disrupted or grabbed, leaving
1074 * the caller to kick it off-list. Otherwise, return true and have
1075 * that dentry's inode and parent both locked.
1077 static bool shrink_lock_dentry(struct dentry *dentry)
1079 struct inode *inode;
1080 struct dentry *parent;
1082 if (dentry->d_lockref.count)
1085 inode = dentry->d_inode;
1086 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1087 spin_unlock(&dentry->d_lock);
1088 spin_lock(&inode->i_lock);
1089 spin_lock(&dentry->d_lock);
1090 if (unlikely(dentry->d_lockref.count))
1092 /* changed inode means that somebody had grabbed it */
1093 if (unlikely(inode != dentry->d_inode))
1097 parent = dentry->d_parent;
1098 if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1101 spin_unlock(&dentry->d_lock);
1102 spin_lock(&parent->d_lock);
1103 if (unlikely(parent != dentry->d_parent)) {
1104 spin_unlock(&parent->d_lock);
1105 spin_lock(&dentry->d_lock);
1108 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1109 if (likely(!dentry->d_lockref.count))
1111 spin_unlock(&parent->d_lock);
1114 spin_unlock(&inode->i_lock);
1118 void shrink_dentry_list(struct list_head *list)
1120 while (!list_empty(list)) {
1121 struct dentry *dentry, *parent;
1123 dentry = list_entry(list->prev, struct dentry, d_lru);
1124 spin_lock(&dentry->d_lock);
1126 if (!shrink_lock_dentry(dentry)) {
1127 bool can_free = false;
1129 d_shrink_del(dentry);
1130 if (dentry->d_lockref.count < 0)
1131 can_free = dentry->d_flags & DCACHE_MAY_FREE;
1132 spin_unlock(&dentry->d_lock);
1134 dentry_free(dentry);
1138 d_shrink_del(dentry);
1139 parent = dentry->d_parent;
1140 if (parent != dentry)
1141 __dput_to_list(parent, list);
1142 __dentry_kill(dentry);
1146 static enum lru_status dentry_lru_isolate(struct list_head *item,
1147 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1149 struct list_head *freeable = arg;
1150 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1154 * we are inverting the lru lock/dentry->d_lock here,
1155 * so use a trylock. If we fail to get the lock, just skip
1158 if (!spin_trylock(&dentry->d_lock))
1162 * Referenced dentries are still in use. If they have active
1163 * counts, just remove them from the LRU. Otherwise give them
1164 * another pass through the LRU.
1166 if (dentry->d_lockref.count) {
1167 d_lru_isolate(lru, dentry);
1168 spin_unlock(&dentry->d_lock);
1172 if (dentry->d_flags & DCACHE_REFERENCED) {
1173 dentry->d_flags &= ~DCACHE_REFERENCED;
1174 spin_unlock(&dentry->d_lock);
1177 * The list move itself will be made by the common LRU code. At
1178 * this point, we've dropped the dentry->d_lock but keep the
1179 * lru lock. This is safe to do, since every list movement is
1180 * protected by the lru lock even if both locks are held.
1182 * This is guaranteed by the fact that all LRU management
1183 * functions are intermediated by the LRU API calls like
1184 * list_lru_add and list_lru_del. List movement in this file
1185 * only ever occur through this functions or through callbacks
1186 * like this one, that are called from the LRU API.
1188 * The only exceptions to this are functions like
1189 * shrink_dentry_list, and code that first checks for the
1190 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1191 * operating only with stack provided lists after they are
1192 * properly isolated from the main list. It is thus, always a
1198 d_lru_shrink_move(lru, dentry, freeable);
1199 spin_unlock(&dentry->d_lock);
1205 * prune_dcache_sb - shrink the dcache
1207 * @sc: shrink control, passed to list_lru_shrink_walk()
1209 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1210 * is done when we need more memory and called from the superblock shrinker
1213 * This function may fail to free any resources if all the dentries are in
1216 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1221 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1222 dentry_lru_isolate, &dispose);
1223 shrink_dentry_list(&dispose);
1227 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1228 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1230 struct list_head *freeable = arg;
1231 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1234 * we are inverting the lru lock/dentry->d_lock here,
1235 * so use a trylock. If we fail to get the lock, just skip
1238 if (!spin_trylock(&dentry->d_lock))
1241 d_lru_shrink_move(lru, dentry, freeable);
1242 spin_unlock(&dentry->d_lock);
1249 * shrink_dcache_sb - shrink dcache for a superblock
1252 * Shrink the dcache for the specified super block. This is used to free
1253 * the dcache before unmounting a file system.
1255 void shrink_dcache_sb(struct super_block *sb)
1260 list_lru_walk(&sb->s_dentry_lru,
1261 dentry_lru_isolate_shrink, &dispose, 1024);
1262 shrink_dentry_list(&dispose);
1263 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1265 EXPORT_SYMBOL(shrink_dcache_sb);
1268 * enum d_walk_ret - action to talke during tree walk
1269 * @D_WALK_CONTINUE: contrinue walk
1270 * @D_WALK_QUIT: quit walk
1271 * @D_WALK_NORETRY: quit when retry is needed
1272 * @D_WALK_SKIP: skip this dentry and its children
1282 * d_walk - walk the dentry tree
1283 * @parent: start of walk
1284 * @data: data passed to @enter() and @finish()
1285 * @enter: callback when first entering the dentry
1287 * The @enter() callbacks are called with d_lock held.
1289 static void d_walk(struct dentry *parent, void *data,
1290 enum d_walk_ret (*enter)(void *, struct dentry *))
1292 struct dentry *this_parent;
1293 struct list_head *next;
1295 enum d_walk_ret ret;
1299 read_seqbegin_or_lock(&rename_lock, &seq);
1300 this_parent = parent;
1301 spin_lock(&this_parent->d_lock);
1303 ret = enter(data, this_parent);
1305 case D_WALK_CONTINUE:
1310 case D_WALK_NORETRY:
1315 next = this_parent->d_subdirs.next;
1317 while (next != &this_parent->d_subdirs) {
1318 struct list_head *tmp = next;
1319 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1322 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1325 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1327 ret = enter(data, dentry);
1329 case D_WALK_CONTINUE:
1332 spin_unlock(&dentry->d_lock);
1334 case D_WALK_NORETRY:
1338 spin_unlock(&dentry->d_lock);
1342 if (!list_empty(&dentry->d_subdirs)) {
1343 spin_unlock(&this_parent->d_lock);
1344 spin_release(&dentry->d_lock.dep_map, _RET_IP_);
1345 this_parent = dentry;
1346 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1349 spin_unlock(&dentry->d_lock);
1352 * All done at this level ... ascend and resume the search.
1356 if (this_parent != parent) {
1357 struct dentry *child = this_parent;
1358 this_parent = child->d_parent;
1360 spin_unlock(&child->d_lock);
1361 spin_lock(&this_parent->d_lock);
1363 /* might go back up the wrong parent if we have had a rename. */
1364 if (need_seqretry(&rename_lock, seq))
1366 /* go into the first sibling still alive */
1368 next = child->d_child.next;
1369 if (next == &this_parent->d_subdirs)
1371 child = list_entry(next, struct dentry, d_child);
1372 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1376 if (need_seqretry(&rename_lock, seq))
1381 spin_unlock(&this_parent->d_lock);
1382 done_seqretry(&rename_lock, seq);
1386 spin_unlock(&this_parent->d_lock);
1395 struct check_mount {
1396 struct vfsmount *mnt;
1397 unsigned int mounted;
1400 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1402 struct check_mount *info = data;
1403 struct path path = { .mnt = info->mnt, .dentry = dentry };
1405 if (likely(!d_mountpoint(dentry)))
1406 return D_WALK_CONTINUE;
1407 if (__path_is_mountpoint(&path)) {
1411 return D_WALK_CONTINUE;
1415 * path_has_submounts - check for mounts over a dentry in the
1416 * current namespace.
1417 * @parent: path to check.
1419 * Return true if the parent or its subdirectories contain
1420 * a mount point in the current namespace.
1422 int path_has_submounts(const struct path *parent)
1424 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1426 read_seqlock_excl(&mount_lock);
1427 d_walk(parent->dentry, &data, path_check_mount);
1428 read_sequnlock_excl(&mount_lock);
1430 return data.mounted;
1432 EXPORT_SYMBOL(path_has_submounts);
1435 * Called by mount code to set a mountpoint and check if the mountpoint is
1436 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1437 * subtree can become unreachable).
1439 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1440 * this reason take rename_lock and d_lock on dentry and ancestors.
1442 int d_set_mounted(struct dentry *dentry)
1446 write_seqlock(&rename_lock);
1447 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1448 /* Need exclusion wrt. d_invalidate() */
1449 spin_lock(&p->d_lock);
1450 if (unlikely(d_unhashed(p))) {
1451 spin_unlock(&p->d_lock);
1454 spin_unlock(&p->d_lock);
1456 spin_lock(&dentry->d_lock);
1457 if (!d_unlinked(dentry)) {
1459 if (!d_mountpoint(dentry)) {
1460 dentry->d_flags |= DCACHE_MOUNTED;
1464 spin_unlock(&dentry->d_lock);
1466 write_sequnlock(&rename_lock);
1471 * Search the dentry child list of the specified parent,
1472 * and move any unused dentries to the end of the unused
1473 * list for prune_dcache(). We descend to the next level
1474 * whenever the d_subdirs list is non-empty and continue
1477 * It returns zero iff there are no unused children,
1478 * otherwise it returns the number of children moved to
1479 * the end of the unused list. This may not be the total
1480 * number of unused children, because select_parent can
1481 * drop the lock and return early due to latency
1485 struct select_data {
1486 struct dentry *start;
1489 struct dentry *victim;
1491 struct list_head dispose;
1494 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1496 struct select_data *data = _data;
1497 enum d_walk_ret ret = D_WALK_CONTINUE;
1499 if (data->start == dentry)
1502 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1505 if (dentry->d_flags & DCACHE_LRU_LIST)
1507 if (!dentry->d_lockref.count) {
1508 d_shrink_add(dentry, &data->dispose);
1513 * We can return to the caller if we have found some (this
1514 * ensures forward progress). We'll be coming back to find
1517 if (!list_empty(&data->dispose))
1518 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1523 static enum d_walk_ret select_collect2(void *_data, struct dentry *dentry)
1525 struct select_data *data = _data;
1526 enum d_walk_ret ret = D_WALK_CONTINUE;
1528 if (data->start == dentry)
1531 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1532 if (!dentry->d_lockref.count) {
1534 data->victim = dentry;
1538 if (dentry->d_flags & DCACHE_LRU_LIST)
1540 if (!dentry->d_lockref.count)
1541 d_shrink_add(dentry, &data->dispose);
1544 * We can return to the caller if we have found some (this
1545 * ensures forward progress). We'll be coming back to find
1548 if (!list_empty(&data->dispose))
1549 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1555 * shrink_dcache_parent - prune dcache
1556 * @parent: parent of entries to prune
1558 * Prune the dcache to remove unused children of the parent dentry.
1560 void shrink_dcache_parent(struct dentry *parent)
1563 struct select_data data = {.start = parent};
1565 INIT_LIST_HEAD(&data.dispose);
1566 d_walk(parent, &data, select_collect);
1568 if (!list_empty(&data.dispose)) {
1569 shrink_dentry_list(&data.dispose);
1577 d_walk(parent, &data, select_collect2);
1579 struct dentry *parent;
1580 spin_lock(&data.victim->d_lock);
1581 if (!shrink_lock_dentry(data.victim)) {
1582 spin_unlock(&data.victim->d_lock);
1586 parent = data.victim->d_parent;
1587 if (parent != data.victim)
1588 __dput_to_list(parent, &data.dispose);
1589 __dentry_kill(data.victim);
1592 if (!list_empty(&data.dispose))
1593 shrink_dentry_list(&data.dispose);
1596 EXPORT_SYMBOL(shrink_dcache_parent);
1598 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1600 /* it has busy descendents; complain about those instead */
1601 if (!list_empty(&dentry->d_subdirs))
1602 return D_WALK_CONTINUE;
1604 /* root with refcount 1 is fine */
1605 if (dentry == _data && dentry->d_lockref.count == 1)
1606 return D_WALK_CONTINUE;
1608 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1609 " still in use (%d) [unmount of %s %s]\n",
1612 dentry->d_inode->i_ino : 0UL,
1614 dentry->d_lockref.count,
1615 dentry->d_sb->s_type->name,
1616 dentry->d_sb->s_id);
1618 return D_WALK_CONTINUE;
1621 static void do_one_tree(struct dentry *dentry)
1623 shrink_dcache_parent(dentry);
1624 d_walk(dentry, dentry, umount_check);
1630 * destroy the dentries attached to a superblock on unmounting
1632 void shrink_dcache_for_umount(struct super_block *sb)
1634 struct dentry *dentry;
1636 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1638 dentry = sb->s_root;
1640 do_one_tree(dentry);
1642 while (!hlist_bl_empty(&sb->s_roots)) {
1643 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1644 do_one_tree(dentry);
1648 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1650 struct dentry **victim = _data;
1651 if (d_mountpoint(dentry)) {
1652 __dget_dlock(dentry);
1656 return D_WALK_CONTINUE;
1660 * d_invalidate - detach submounts, prune dcache, and drop
1661 * @dentry: dentry to invalidate (aka detach, prune and drop)
1663 void d_invalidate(struct dentry *dentry)
1665 bool had_submounts = false;
1666 spin_lock(&dentry->d_lock);
1667 if (d_unhashed(dentry)) {
1668 spin_unlock(&dentry->d_lock);
1672 spin_unlock(&dentry->d_lock);
1674 /* Negative dentries can be dropped without further checks */
1675 if (!dentry->d_inode)
1678 shrink_dcache_parent(dentry);
1680 struct dentry *victim = NULL;
1681 d_walk(dentry, &victim, find_submount);
1684 shrink_dcache_parent(dentry);
1687 had_submounts = true;
1688 detach_mounts(victim);
1692 EXPORT_SYMBOL(d_invalidate);
1695 * __d_alloc - allocate a dcache entry
1696 * @sb: filesystem it will belong to
1697 * @name: qstr of the name
1699 * Allocates a dentry. It returns %NULL if there is insufficient memory
1700 * available. On a success the dentry is returned. The name passed in is
1701 * copied and the copy passed in may be reused after this call.
1704 static struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1706 struct dentry *dentry;
1710 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1715 * We guarantee that the inline name is always NUL-terminated.
1716 * This way the memcpy() done by the name switching in rename
1717 * will still always have a NUL at the end, even if we might
1718 * be overwriting an internal NUL character
1720 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1721 if (unlikely(!name)) {
1723 dname = dentry->d_iname;
1724 } else if (name->len > DNAME_INLINE_LEN-1) {
1725 size_t size = offsetof(struct external_name, name[1]);
1726 struct external_name *p = kmalloc(size + name->len,
1727 GFP_KERNEL_ACCOUNT |
1730 kmem_cache_free(dentry_cache, dentry);
1733 atomic_set(&p->u.count, 1);
1736 dname = dentry->d_iname;
1739 dentry->d_name.len = name->len;
1740 dentry->d_name.hash = name->hash;
1741 memcpy(dname, name->name, name->len);
1742 dname[name->len] = 0;
1744 /* Make sure we always see the terminating NUL character */
1745 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1747 dentry->d_lockref.count = 1;
1748 dentry->d_flags = 0;
1749 spin_lock_init(&dentry->d_lock);
1750 seqcount_spinlock_init(&dentry->d_seq, &dentry->d_lock);
1751 dentry->d_inode = NULL;
1752 dentry->d_parent = dentry;
1754 dentry->d_op = NULL;
1755 dentry->d_fsdata = NULL;
1756 INIT_HLIST_BL_NODE(&dentry->d_hash);
1757 INIT_LIST_HEAD(&dentry->d_lru);
1758 INIT_LIST_HEAD(&dentry->d_subdirs);
1759 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1760 INIT_LIST_HEAD(&dentry->d_child);
1761 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1763 if (dentry->d_op && dentry->d_op->d_init) {
1764 err = dentry->d_op->d_init(dentry);
1766 if (dname_external(dentry))
1767 kfree(external_name(dentry));
1768 kmem_cache_free(dentry_cache, dentry);
1773 this_cpu_inc(nr_dentry);
1779 * d_alloc - allocate a dcache entry
1780 * @parent: parent of entry to allocate
1781 * @name: qstr of the name
1783 * Allocates a dentry. It returns %NULL if there is insufficient memory
1784 * available. On a success the dentry is returned. The name passed in is
1785 * copied and the copy passed in may be reused after this call.
1787 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1789 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1792 spin_lock(&parent->d_lock);
1794 * don't need child lock because it is not subject
1795 * to concurrency here
1797 __dget_dlock(parent);
1798 dentry->d_parent = parent;
1799 list_add(&dentry->d_child, &parent->d_subdirs);
1800 spin_unlock(&parent->d_lock);
1804 EXPORT_SYMBOL(d_alloc);
1806 struct dentry *d_alloc_anon(struct super_block *sb)
1808 return __d_alloc(sb, NULL);
1810 EXPORT_SYMBOL(d_alloc_anon);
1812 struct dentry *d_alloc_cursor(struct dentry * parent)
1814 struct dentry *dentry = d_alloc_anon(parent->d_sb);
1816 dentry->d_flags |= DCACHE_DENTRY_CURSOR;
1817 dentry->d_parent = dget(parent);
1823 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1824 * @sb: the superblock
1825 * @name: qstr of the name
1827 * For a filesystem that just pins its dentries in memory and never
1828 * performs lookups at all, return an unhashed IS_ROOT dentry.
1829 * This is used for pipes, sockets et.al. - the stuff that should
1830 * never be anyone's children or parents. Unlike all other
1831 * dentries, these will not have RCU delay between dropping the
1832 * last reference and freeing them.
1834 * The only user is alloc_file_pseudo() and that's what should
1835 * be considered a public interface. Don't use directly.
1837 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1839 struct dentry *dentry = __d_alloc(sb, name);
1841 dentry->d_flags |= DCACHE_NORCU;
1845 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1850 q.hash_len = hashlen_string(parent, name);
1851 return d_alloc(parent, &q);
1853 EXPORT_SYMBOL(d_alloc_name);
1855 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1857 WARN_ON_ONCE(dentry->d_op);
1858 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1860 DCACHE_OP_REVALIDATE |
1861 DCACHE_OP_WEAK_REVALIDATE |
1868 dentry->d_flags |= DCACHE_OP_HASH;
1870 dentry->d_flags |= DCACHE_OP_COMPARE;
1871 if (op->d_revalidate)
1872 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1873 if (op->d_weak_revalidate)
1874 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1876 dentry->d_flags |= DCACHE_OP_DELETE;
1878 dentry->d_flags |= DCACHE_OP_PRUNE;
1880 dentry->d_flags |= DCACHE_OP_REAL;
1883 EXPORT_SYMBOL(d_set_d_op);
1887 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1888 * @dentry - The dentry to mark
1890 * Mark a dentry as falling through to the lower layer (as set with
1891 * d_pin_lower()). This flag may be recorded on the medium.
1893 void d_set_fallthru(struct dentry *dentry)
1895 spin_lock(&dentry->d_lock);
1896 dentry->d_flags |= DCACHE_FALLTHRU;
1897 spin_unlock(&dentry->d_lock);
1899 EXPORT_SYMBOL(d_set_fallthru);
1901 static unsigned d_flags_for_inode(struct inode *inode)
1903 unsigned add_flags = DCACHE_REGULAR_TYPE;
1906 return DCACHE_MISS_TYPE;
1908 if (S_ISDIR(inode->i_mode)) {
1909 add_flags = DCACHE_DIRECTORY_TYPE;
1910 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1911 if (unlikely(!inode->i_op->lookup))
1912 add_flags = DCACHE_AUTODIR_TYPE;
1914 inode->i_opflags |= IOP_LOOKUP;
1916 goto type_determined;
1919 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1920 if (unlikely(inode->i_op->get_link)) {
1921 add_flags = DCACHE_SYMLINK_TYPE;
1922 goto type_determined;
1924 inode->i_opflags |= IOP_NOFOLLOW;
1927 if (unlikely(!S_ISREG(inode->i_mode)))
1928 add_flags = DCACHE_SPECIAL_TYPE;
1931 if (unlikely(IS_AUTOMOUNT(inode)))
1932 add_flags |= DCACHE_NEED_AUTOMOUNT;
1936 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1938 unsigned add_flags = d_flags_for_inode(inode);
1939 WARN_ON(d_in_lookup(dentry));
1941 spin_lock(&dentry->d_lock);
1943 * Decrement negative dentry count if it was in the LRU list.
1945 if (dentry->d_flags & DCACHE_LRU_LIST)
1946 this_cpu_dec(nr_dentry_negative);
1947 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1948 raw_write_seqcount_begin(&dentry->d_seq);
1949 __d_set_inode_and_type(dentry, inode, add_flags);
1950 raw_write_seqcount_end(&dentry->d_seq);
1951 fsnotify_update_flags(dentry);
1952 spin_unlock(&dentry->d_lock);
1956 * d_instantiate - fill in inode information for a dentry
1957 * @entry: dentry to complete
1958 * @inode: inode to attach to this dentry
1960 * Fill in inode information in the entry.
1962 * This turns negative dentries into productive full members
1965 * NOTE! This assumes that the inode count has been incremented
1966 * (or otherwise set) by the caller to indicate that it is now
1967 * in use by the dcache.
1970 void d_instantiate(struct dentry *entry, struct inode * inode)
1972 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1974 security_d_instantiate(entry, inode);
1975 spin_lock(&inode->i_lock);
1976 __d_instantiate(entry, inode);
1977 spin_unlock(&inode->i_lock);
1980 EXPORT_SYMBOL(d_instantiate);
1983 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1984 * with lockdep-related part of unlock_new_inode() done before
1985 * anything else. Use that instead of open-coding d_instantiate()/
1986 * unlock_new_inode() combinations.
1988 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1990 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1992 lockdep_annotate_inode_mutex_key(inode);
1993 security_d_instantiate(entry, inode);
1994 spin_lock(&inode->i_lock);
1995 __d_instantiate(entry, inode);
1996 WARN_ON(!(inode->i_state & I_NEW));
1997 inode->i_state &= ~I_NEW & ~I_CREATING;
1999 wake_up_bit(&inode->i_state, __I_NEW);
2000 spin_unlock(&inode->i_lock);
2002 EXPORT_SYMBOL(d_instantiate_new);
2004 struct dentry *d_make_root(struct inode *root_inode)
2006 struct dentry *res = NULL;
2009 res = d_alloc_anon(root_inode->i_sb);
2011 d_instantiate(res, root_inode);
2017 EXPORT_SYMBOL(d_make_root);
2019 static struct dentry *__d_instantiate_anon(struct dentry *dentry,
2020 struct inode *inode,
2026 security_d_instantiate(dentry, inode);
2027 spin_lock(&inode->i_lock);
2028 res = __d_find_any_alias(inode);
2030 spin_unlock(&inode->i_lock);
2035 /* attach a disconnected dentry */
2036 add_flags = d_flags_for_inode(inode);
2039 add_flags |= DCACHE_DISCONNECTED;
2041 spin_lock(&dentry->d_lock);
2042 __d_set_inode_and_type(dentry, inode, add_flags);
2043 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2044 if (!disconnected) {
2045 hlist_bl_lock(&dentry->d_sb->s_roots);
2046 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
2047 hlist_bl_unlock(&dentry->d_sb->s_roots);
2049 spin_unlock(&dentry->d_lock);
2050 spin_unlock(&inode->i_lock);
2059 struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
2061 return __d_instantiate_anon(dentry, inode, true);
2063 EXPORT_SYMBOL(d_instantiate_anon);
2065 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
2071 return ERR_PTR(-ESTALE);
2073 return ERR_CAST(inode);
2075 res = d_find_any_alias(inode);
2079 tmp = d_alloc_anon(inode->i_sb);
2081 res = ERR_PTR(-ENOMEM);
2085 return __d_instantiate_anon(tmp, inode, disconnected);
2093 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2094 * @inode: inode to allocate the dentry for
2096 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2097 * similar open by handle operations. The returned dentry may be anonymous,
2098 * or may have a full name (if the inode was already in the cache).
2100 * When called on a directory inode, we must ensure that the inode only ever
2101 * has one dentry. If a dentry is found, that is returned instead of
2102 * allocating a new one.
2104 * On successful return, the reference to the inode has been transferred
2105 * to the dentry. In case of an error the reference on the inode is released.
2106 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2107 * be passed in and the error will be propagated to the return value,
2108 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2110 struct dentry *d_obtain_alias(struct inode *inode)
2112 return __d_obtain_alias(inode, true);
2114 EXPORT_SYMBOL(d_obtain_alias);
2117 * d_obtain_root - find or allocate a dentry for a given inode
2118 * @inode: inode to allocate the dentry for
2120 * Obtain an IS_ROOT dentry for the root of a filesystem.
2122 * We must ensure that directory inodes only ever have one dentry. If a
2123 * dentry is found, that is returned instead of allocating a new one.
2125 * On successful return, the reference to the inode has been transferred
2126 * to the dentry. In case of an error the reference on the inode is
2127 * released. A %NULL or IS_ERR inode may be passed in and will be the
2128 * error will be propagate to the return value, with a %NULL @inode
2129 * replaced by ERR_PTR(-ESTALE).
2131 struct dentry *d_obtain_root(struct inode *inode)
2133 return __d_obtain_alias(inode, false);
2135 EXPORT_SYMBOL(d_obtain_root);
2138 * d_add_ci - lookup or allocate new dentry with case-exact name
2139 * @inode: the inode case-insensitive lookup has found
2140 * @dentry: the negative dentry that was passed to the parent's lookup func
2141 * @name: the case-exact name to be associated with the returned dentry
2143 * This is to avoid filling the dcache with case-insensitive names to the
2144 * same inode, only the actual correct case is stored in the dcache for
2145 * case-insensitive filesystems.
2147 * For a case-insensitive lookup match and if the the case-exact dentry
2148 * already exists in in the dcache, use it and return it.
2150 * If no entry exists with the exact case name, allocate new dentry with
2151 * the exact case, and return the spliced entry.
2153 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2156 struct dentry *found, *res;
2159 * First check if a dentry matching the name already exists,
2160 * if not go ahead and create it now.
2162 found = d_hash_and_lookup(dentry->d_parent, name);
2167 if (d_in_lookup(dentry)) {
2168 found = d_alloc_parallel(dentry->d_parent, name,
2170 if (IS_ERR(found) || !d_in_lookup(found)) {
2175 found = d_alloc(dentry->d_parent, name);
2178 return ERR_PTR(-ENOMEM);
2181 res = d_splice_alias(inode, found);
2188 EXPORT_SYMBOL(d_add_ci);
2191 static inline bool d_same_name(const struct dentry *dentry,
2192 const struct dentry *parent,
2193 const struct qstr *name)
2195 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2196 if (dentry->d_name.len != name->len)
2198 return dentry_cmp(dentry, name->name, name->len) == 0;
2200 return parent->d_op->d_compare(dentry,
2201 dentry->d_name.len, dentry->d_name.name,
2206 * __d_lookup_rcu - search for a dentry (racy, store-free)
2207 * @parent: parent dentry
2208 * @name: qstr of name we wish to find
2209 * @seqp: returns d_seq value at the point where the dentry was found
2210 * Returns: dentry, or NULL
2212 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2213 * resolution (store-free path walking) design described in
2214 * Documentation/filesystems/path-lookup.txt.
2216 * This is not to be used outside core vfs.
2218 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2219 * held, and rcu_read_lock held. The returned dentry must not be stored into
2220 * without taking d_lock and checking d_seq sequence count against @seq
2223 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2226 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2227 * the returned dentry, so long as its parent's seqlock is checked after the
2228 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2229 * is formed, giving integrity down the path walk.
2231 * NOTE! The caller *has* to check the resulting dentry against the sequence
2232 * number we've returned before using any of the resulting dentry state!
2234 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2235 const struct qstr *name,
2238 u64 hashlen = name->hash_len;
2239 const unsigned char *str = name->name;
2240 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2241 struct hlist_bl_node *node;
2242 struct dentry *dentry;
2245 * Note: There is significant duplication with __d_lookup_rcu which is
2246 * required to prevent single threaded performance regressions
2247 * especially on architectures where smp_rmb (in seqcounts) are costly.
2248 * Keep the two functions in sync.
2252 * The hash list is protected using RCU.
2254 * Carefully use d_seq when comparing a candidate dentry, to avoid
2255 * races with d_move().
2257 * It is possible that concurrent renames can mess up our list
2258 * walk here and result in missing our dentry, resulting in the
2259 * false-negative result. d_lookup() protects against concurrent
2260 * renames using rename_lock seqlock.
2262 * See Documentation/filesystems/path-lookup.txt for more details.
2264 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2269 * The dentry sequence count protects us from concurrent
2270 * renames, and thus protects parent and name fields.
2272 * The caller must perform a seqcount check in order
2273 * to do anything useful with the returned dentry.
2275 * NOTE! We do a "raw" seqcount_begin here. That means that
2276 * we don't wait for the sequence count to stabilize if it
2277 * is in the middle of a sequence change. If we do the slow
2278 * dentry compare, we will do seqretries until it is stable,
2279 * and if we end up with a successful lookup, we actually
2280 * want to exit RCU lookup anyway.
2282 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2283 * we are still guaranteed NUL-termination of ->d_name.name.
2285 seq = raw_seqcount_begin(&dentry->d_seq);
2286 if (dentry->d_parent != parent)
2288 if (d_unhashed(dentry))
2291 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2294 if (dentry->d_name.hash != hashlen_hash(hashlen))
2296 tlen = dentry->d_name.len;
2297 tname = dentry->d_name.name;
2298 /* we want a consistent (name,len) pair */
2299 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2303 if (parent->d_op->d_compare(dentry,
2304 tlen, tname, name) != 0)
2307 if (dentry->d_name.hash_len != hashlen)
2309 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2319 * d_lookup - search for a dentry
2320 * @parent: parent dentry
2321 * @name: qstr of name we wish to find
2322 * Returns: dentry, or NULL
2324 * d_lookup searches the children of the parent dentry for the name in
2325 * question. If the dentry is found its reference count is incremented and the
2326 * dentry is returned. The caller must use dput to free the entry when it has
2327 * finished using it. %NULL is returned if the dentry does not exist.
2329 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2331 struct dentry *dentry;
2335 seq = read_seqbegin(&rename_lock);
2336 dentry = __d_lookup(parent, name);
2339 } while (read_seqretry(&rename_lock, seq));
2342 EXPORT_SYMBOL(d_lookup);
2345 * __d_lookup - search for a dentry (racy)
2346 * @parent: parent dentry
2347 * @name: qstr of name we wish to find
2348 * Returns: dentry, or NULL
2350 * __d_lookup is like d_lookup, however it may (rarely) return a
2351 * false-negative result due to unrelated rename activity.
2353 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2354 * however it must be used carefully, eg. with a following d_lookup in
2355 * the case of failure.
2357 * __d_lookup callers must be commented.
2359 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2361 unsigned int hash = name->hash;
2362 struct hlist_bl_head *b = d_hash(hash);
2363 struct hlist_bl_node *node;
2364 struct dentry *found = NULL;
2365 struct dentry *dentry;
2368 * Note: There is significant duplication with __d_lookup_rcu which is
2369 * required to prevent single threaded performance regressions
2370 * especially on architectures where smp_rmb (in seqcounts) are costly.
2371 * Keep the two functions in sync.
2375 * The hash list is protected using RCU.
2377 * Take d_lock when comparing a candidate dentry, to avoid races
2380 * It is possible that concurrent renames can mess up our list
2381 * walk here and result in missing our dentry, resulting in the
2382 * false-negative result. d_lookup() protects against concurrent
2383 * renames using rename_lock seqlock.
2385 * See Documentation/filesystems/path-lookup.txt for more details.
2389 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2391 if (dentry->d_name.hash != hash)
2394 spin_lock(&dentry->d_lock);
2395 if (dentry->d_parent != parent)
2397 if (d_unhashed(dentry))
2400 if (!d_same_name(dentry, parent, name))
2403 dentry->d_lockref.count++;
2405 spin_unlock(&dentry->d_lock);
2408 spin_unlock(&dentry->d_lock);
2416 * d_hash_and_lookup - hash the qstr then search for a dentry
2417 * @dir: Directory to search in
2418 * @name: qstr of name we wish to find
2420 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2422 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2425 * Check for a fs-specific hash function. Note that we must
2426 * calculate the standard hash first, as the d_op->d_hash()
2427 * routine may choose to leave the hash value unchanged.
2429 name->hash = full_name_hash(dir, name->name, name->len);
2430 if (dir->d_flags & DCACHE_OP_HASH) {
2431 int err = dir->d_op->d_hash(dir, name);
2432 if (unlikely(err < 0))
2433 return ERR_PTR(err);
2435 return d_lookup(dir, name);
2437 EXPORT_SYMBOL(d_hash_and_lookup);
2440 * When a file is deleted, we have two options:
2441 * - turn this dentry into a negative dentry
2442 * - unhash this dentry and free it.
2444 * Usually, we want to just turn this into
2445 * a negative dentry, but if anybody else is
2446 * currently using the dentry or the inode
2447 * we can't do that and we fall back on removing
2448 * it from the hash queues and waiting for
2449 * it to be deleted later when it has no users
2453 * d_delete - delete a dentry
2454 * @dentry: The dentry to delete
2456 * Turn the dentry into a negative dentry if possible, otherwise
2457 * remove it from the hash queues so it can be deleted later
2460 void d_delete(struct dentry * dentry)
2462 struct inode *inode = dentry->d_inode;
2464 spin_lock(&inode->i_lock);
2465 spin_lock(&dentry->d_lock);
2467 * Are we the only user?
2469 if (dentry->d_lockref.count == 1) {
2470 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2471 dentry_unlink_inode(dentry);
2474 spin_unlock(&dentry->d_lock);
2475 spin_unlock(&inode->i_lock);
2478 EXPORT_SYMBOL(d_delete);
2480 static void __d_rehash(struct dentry *entry)
2482 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2485 hlist_bl_add_head_rcu(&entry->d_hash, b);
2490 * d_rehash - add an entry back to the hash
2491 * @entry: dentry to add to the hash
2493 * Adds a dentry to the hash according to its name.
2496 void d_rehash(struct dentry * entry)
2498 spin_lock(&entry->d_lock);
2500 spin_unlock(&entry->d_lock);
2502 EXPORT_SYMBOL(d_rehash);
2504 static inline unsigned start_dir_add(struct inode *dir)
2508 unsigned n = dir->i_dir_seq;
2509 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2515 static inline void end_dir_add(struct inode *dir, unsigned n)
2517 smp_store_release(&dir->i_dir_seq, n + 2);
2520 static void d_wait_lookup(struct dentry *dentry)
2522 if (d_in_lookup(dentry)) {
2523 DECLARE_WAITQUEUE(wait, current);
2524 add_wait_queue(dentry->d_wait, &wait);
2526 set_current_state(TASK_UNINTERRUPTIBLE);
2527 spin_unlock(&dentry->d_lock);
2529 spin_lock(&dentry->d_lock);
2530 } while (d_in_lookup(dentry));
2534 struct dentry *d_alloc_parallel(struct dentry *parent,
2535 const struct qstr *name,
2536 wait_queue_head_t *wq)
2538 unsigned int hash = name->hash;
2539 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2540 struct hlist_bl_node *node;
2541 struct dentry *new = d_alloc(parent, name);
2542 struct dentry *dentry;
2543 unsigned seq, r_seq, d_seq;
2546 return ERR_PTR(-ENOMEM);
2550 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2551 r_seq = read_seqbegin(&rename_lock);
2552 dentry = __d_lookup_rcu(parent, name, &d_seq);
2553 if (unlikely(dentry)) {
2554 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2558 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2567 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2572 if (unlikely(seq & 1)) {
2578 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2584 * No changes for the parent since the beginning of d_lookup().
2585 * Since all removals from the chain happen with hlist_bl_lock(),
2586 * any potential in-lookup matches are going to stay here until
2587 * we unlock the chain. All fields are stable in everything
2590 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2591 if (dentry->d_name.hash != hash)
2593 if (dentry->d_parent != parent)
2595 if (!d_same_name(dentry, parent, name))
2598 /* now we can try to grab a reference */
2599 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2606 * somebody is likely to be still doing lookup for it;
2607 * wait for them to finish
2609 spin_lock(&dentry->d_lock);
2610 d_wait_lookup(dentry);
2612 * it's not in-lookup anymore; in principle we should repeat
2613 * everything from dcache lookup, but it's likely to be what
2614 * d_lookup() would've found anyway. If it is, just return it;
2615 * otherwise we really have to repeat the whole thing.
2617 if (unlikely(dentry->d_name.hash != hash))
2619 if (unlikely(dentry->d_parent != parent))
2621 if (unlikely(d_unhashed(dentry)))
2623 if (unlikely(!d_same_name(dentry, parent, name)))
2625 /* OK, it *is* a hashed match; return it */
2626 spin_unlock(&dentry->d_lock);
2631 /* we can't take ->d_lock here; it's OK, though. */
2632 new->d_flags |= DCACHE_PAR_LOOKUP;
2634 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2638 spin_unlock(&dentry->d_lock);
2642 EXPORT_SYMBOL(d_alloc_parallel);
2644 void __d_lookup_done(struct dentry *dentry)
2646 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2647 dentry->d_name.hash);
2649 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2650 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2651 wake_up_all(dentry->d_wait);
2652 dentry->d_wait = NULL;
2654 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2655 INIT_LIST_HEAD(&dentry->d_lru);
2657 EXPORT_SYMBOL(__d_lookup_done);
2659 /* inode->i_lock held if inode is non-NULL */
2661 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2663 struct inode *dir = NULL;
2665 spin_lock(&dentry->d_lock);
2666 if (unlikely(d_in_lookup(dentry))) {
2667 dir = dentry->d_parent->d_inode;
2668 n = start_dir_add(dir);
2669 __d_lookup_done(dentry);
2672 unsigned add_flags = d_flags_for_inode(inode);
2673 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2674 raw_write_seqcount_begin(&dentry->d_seq);
2675 __d_set_inode_and_type(dentry, inode, add_flags);
2676 raw_write_seqcount_end(&dentry->d_seq);
2677 fsnotify_update_flags(dentry);
2681 end_dir_add(dir, n);
2682 spin_unlock(&dentry->d_lock);
2684 spin_unlock(&inode->i_lock);
2688 * d_add - add dentry to hash queues
2689 * @entry: dentry to add
2690 * @inode: The inode to attach to this dentry
2692 * This adds the entry to the hash queues and initializes @inode.
2693 * The entry was actually filled in earlier during d_alloc().
2696 void d_add(struct dentry *entry, struct inode *inode)
2699 security_d_instantiate(entry, inode);
2700 spin_lock(&inode->i_lock);
2702 __d_add(entry, inode);
2704 EXPORT_SYMBOL(d_add);
2707 * d_exact_alias - find and hash an exact unhashed alias
2708 * @entry: dentry to add
2709 * @inode: The inode to go with this dentry
2711 * If an unhashed dentry with the same name/parent and desired
2712 * inode already exists, hash and return it. Otherwise, return
2715 * Parent directory should be locked.
2717 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2719 struct dentry *alias;
2720 unsigned int hash = entry->d_name.hash;
2722 spin_lock(&inode->i_lock);
2723 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2725 * Don't need alias->d_lock here, because aliases with
2726 * d_parent == entry->d_parent are not subject to name or
2727 * parent changes, because the parent inode i_mutex is held.
2729 if (alias->d_name.hash != hash)
2731 if (alias->d_parent != entry->d_parent)
2733 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2735 spin_lock(&alias->d_lock);
2736 if (!d_unhashed(alias)) {
2737 spin_unlock(&alias->d_lock);
2740 __dget_dlock(alias);
2742 spin_unlock(&alias->d_lock);
2744 spin_unlock(&inode->i_lock);
2747 spin_unlock(&inode->i_lock);
2750 EXPORT_SYMBOL(d_exact_alias);
2752 static void swap_names(struct dentry *dentry, struct dentry *target)
2754 if (unlikely(dname_external(target))) {
2755 if (unlikely(dname_external(dentry))) {
2757 * Both external: swap the pointers
2759 swap(target->d_name.name, dentry->d_name.name);
2762 * dentry:internal, target:external. Steal target's
2763 * storage and make target internal.
2765 memcpy(target->d_iname, dentry->d_name.name,
2766 dentry->d_name.len + 1);
2767 dentry->d_name.name = target->d_name.name;
2768 target->d_name.name = target->d_iname;
2771 if (unlikely(dname_external(dentry))) {
2773 * dentry:external, target:internal. Give dentry's
2774 * storage to target and make dentry internal
2776 memcpy(dentry->d_iname, target->d_name.name,
2777 target->d_name.len + 1);
2778 target->d_name.name = dentry->d_name.name;
2779 dentry->d_name.name = dentry->d_iname;
2782 * Both are internal.
2785 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2786 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2787 swap(((long *) &dentry->d_iname)[i],
2788 ((long *) &target->d_iname)[i]);
2792 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2795 static void copy_name(struct dentry *dentry, struct dentry *target)
2797 struct external_name *old_name = NULL;
2798 if (unlikely(dname_external(dentry)))
2799 old_name = external_name(dentry);
2800 if (unlikely(dname_external(target))) {
2801 atomic_inc(&external_name(target)->u.count);
2802 dentry->d_name = target->d_name;
2804 memcpy(dentry->d_iname, target->d_name.name,
2805 target->d_name.len + 1);
2806 dentry->d_name.name = dentry->d_iname;
2807 dentry->d_name.hash_len = target->d_name.hash_len;
2809 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2810 kfree_rcu(old_name, u.head);
2814 * __d_move - move a dentry
2815 * @dentry: entry to move
2816 * @target: new dentry
2817 * @exchange: exchange the two dentries
2819 * Update the dcache to reflect the move of a file name. Negative
2820 * dcache entries should not be moved in this way. Caller must hold
2821 * rename_lock, the i_mutex of the source and target directories,
2822 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2824 static void __d_move(struct dentry *dentry, struct dentry *target,
2827 struct dentry *old_parent, *p;
2828 struct inode *dir = NULL;
2831 WARN_ON(!dentry->d_inode);
2832 if (WARN_ON(dentry == target))
2835 BUG_ON(d_ancestor(target, dentry));
2836 old_parent = dentry->d_parent;
2837 p = d_ancestor(old_parent, target);
2838 if (IS_ROOT(dentry)) {
2840 spin_lock(&target->d_parent->d_lock);
2842 /* target is not a descendent of dentry->d_parent */
2843 spin_lock(&target->d_parent->d_lock);
2844 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2846 BUG_ON(p == dentry);
2847 spin_lock(&old_parent->d_lock);
2849 spin_lock_nested(&target->d_parent->d_lock,
2850 DENTRY_D_LOCK_NESTED);
2852 spin_lock_nested(&dentry->d_lock, 2);
2853 spin_lock_nested(&target->d_lock, 3);
2855 if (unlikely(d_in_lookup(target))) {
2856 dir = target->d_parent->d_inode;
2857 n = start_dir_add(dir);
2858 __d_lookup_done(target);
2861 write_seqcount_begin(&dentry->d_seq);
2862 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2865 if (!d_unhashed(dentry))
2867 if (!d_unhashed(target))
2870 /* ... and switch them in the tree */
2871 dentry->d_parent = target->d_parent;
2873 copy_name(dentry, target);
2874 target->d_hash.pprev = NULL;
2875 dentry->d_parent->d_lockref.count++;
2876 if (dentry != old_parent) /* wasn't IS_ROOT */
2877 WARN_ON(!--old_parent->d_lockref.count);
2879 target->d_parent = old_parent;
2880 swap_names(dentry, target);
2881 list_move(&target->d_child, &target->d_parent->d_subdirs);
2883 fsnotify_update_flags(target);
2885 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2887 fsnotify_update_flags(dentry);
2888 fscrypt_handle_d_move(dentry);
2890 write_seqcount_end(&target->d_seq);
2891 write_seqcount_end(&dentry->d_seq);
2894 end_dir_add(dir, n);
2896 if (dentry->d_parent != old_parent)
2897 spin_unlock(&dentry->d_parent->d_lock);
2898 if (dentry != old_parent)
2899 spin_unlock(&old_parent->d_lock);
2900 spin_unlock(&target->d_lock);
2901 spin_unlock(&dentry->d_lock);
2905 * d_move - move a dentry
2906 * @dentry: entry to move
2907 * @target: new dentry
2909 * Update the dcache to reflect the move of a file name. Negative
2910 * dcache entries should not be moved in this way. See the locking
2911 * requirements for __d_move.
2913 void d_move(struct dentry *dentry, struct dentry *target)
2915 write_seqlock(&rename_lock);
2916 __d_move(dentry, target, false);
2917 write_sequnlock(&rename_lock);
2919 EXPORT_SYMBOL(d_move);
2922 * d_exchange - exchange two dentries
2923 * @dentry1: first dentry
2924 * @dentry2: second dentry
2926 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2928 write_seqlock(&rename_lock);
2930 WARN_ON(!dentry1->d_inode);
2931 WARN_ON(!dentry2->d_inode);
2932 WARN_ON(IS_ROOT(dentry1));
2933 WARN_ON(IS_ROOT(dentry2));
2935 __d_move(dentry1, dentry2, true);
2937 write_sequnlock(&rename_lock);
2941 * d_ancestor - search for an ancestor
2942 * @p1: ancestor dentry
2945 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2946 * an ancestor of p2, else NULL.
2948 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2952 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2953 if (p->d_parent == p1)
2960 * This helper attempts to cope with remotely renamed directories
2962 * It assumes that the caller is already holding
2963 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2965 * Note: If ever the locking in lock_rename() changes, then please
2966 * remember to update this too...
2968 static int __d_unalias(struct inode *inode,
2969 struct dentry *dentry, struct dentry *alias)
2971 struct mutex *m1 = NULL;
2972 struct rw_semaphore *m2 = NULL;
2975 /* If alias and dentry share a parent, then no extra locks required */
2976 if (alias->d_parent == dentry->d_parent)
2979 /* See lock_rename() */
2980 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2982 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2983 if (!inode_trylock_shared(alias->d_parent->d_inode))
2985 m2 = &alias->d_parent->d_inode->i_rwsem;
2987 __d_move(alias, dentry, false);
2998 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2999 * @inode: the inode which may have a disconnected dentry
3000 * @dentry: a negative dentry which we want to point to the inode.
3002 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3003 * place of the given dentry and return it, else simply d_add the inode
3004 * to the dentry and return NULL.
3006 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3007 * we should error out: directories can't have multiple aliases.
3009 * This is needed in the lookup routine of any filesystem that is exportable
3010 * (via knfsd) so that we can build dcache paths to directories effectively.
3012 * If a dentry was found and moved, then it is returned. Otherwise NULL
3013 * is returned. This matches the expected return value of ->lookup.
3015 * Cluster filesystems may call this function with a negative, hashed dentry.
3016 * In that case, we know that the inode will be a regular file, and also this
3017 * will only occur during atomic_open. So we need to check for the dentry
3018 * being already hashed only in the final case.
3020 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3023 return ERR_CAST(inode);
3025 BUG_ON(!d_unhashed(dentry));
3030 security_d_instantiate(dentry, inode);
3031 spin_lock(&inode->i_lock);
3032 if (S_ISDIR(inode->i_mode)) {
3033 struct dentry *new = __d_find_any_alias(inode);
3034 if (unlikely(new)) {
3035 /* The reference to new ensures it remains an alias */
3036 spin_unlock(&inode->i_lock);
3037 write_seqlock(&rename_lock);
3038 if (unlikely(d_ancestor(new, dentry))) {
3039 write_sequnlock(&rename_lock);
3041 new = ERR_PTR(-ELOOP);
3042 pr_warn_ratelimited(
3043 "VFS: Lookup of '%s' in %s %s"
3044 " would have caused loop\n",
3045 dentry->d_name.name,
3046 inode->i_sb->s_type->name,
3048 } else if (!IS_ROOT(new)) {
3049 struct dentry *old_parent = dget(new->d_parent);
3050 int err = __d_unalias(inode, dentry, new);
3051 write_sequnlock(&rename_lock);
3058 __d_move(new, dentry, false);
3059 write_sequnlock(&rename_lock);
3066 __d_add(dentry, inode);
3069 EXPORT_SYMBOL(d_splice_alias);
3072 * Test whether new_dentry is a subdirectory of old_dentry.
3074 * Trivially implemented using the dcache structure
3078 * is_subdir - is new dentry a subdirectory of old_dentry
3079 * @new_dentry: new dentry
3080 * @old_dentry: old dentry
3082 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3083 * Returns false otherwise.
3084 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3087 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3092 if (new_dentry == old_dentry)
3096 /* for restarting inner loop in case of seq retry */
3097 seq = read_seqbegin(&rename_lock);
3099 * Need rcu_readlock to protect against the d_parent trashing
3103 if (d_ancestor(old_dentry, new_dentry))
3108 } while (read_seqretry(&rename_lock, seq));
3112 EXPORT_SYMBOL(is_subdir);
3114 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3116 struct dentry *root = data;
3117 if (dentry != root) {
3118 if (d_unhashed(dentry) || !dentry->d_inode)
3121 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3122 dentry->d_flags |= DCACHE_GENOCIDE;
3123 dentry->d_lockref.count--;
3126 return D_WALK_CONTINUE;
3129 void d_genocide(struct dentry *parent)
3131 d_walk(parent, parent, d_genocide_kill);
3134 EXPORT_SYMBOL(d_genocide);
3136 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3138 inode_dec_link_count(inode);
3139 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3140 !hlist_unhashed(&dentry->d_u.d_alias) ||
3141 !d_unlinked(dentry));
3142 spin_lock(&dentry->d_parent->d_lock);
3143 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3144 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3145 (unsigned long long)inode->i_ino);
3146 spin_unlock(&dentry->d_lock);
3147 spin_unlock(&dentry->d_parent->d_lock);
3148 d_instantiate(dentry, inode);
3150 EXPORT_SYMBOL(d_tmpfile);
3152 static __initdata unsigned long dhash_entries;
3153 static int __init set_dhash_entries(char *str)
3157 dhash_entries = simple_strtoul(str, &str, 0);
3160 __setup("dhash_entries=", set_dhash_entries);
3162 static void __init dcache_init_early(void)
3164 /* If hashes are distributed across NUMA nodes, defer
3165 * hash allocation until vmalloc space is available.
3171 alloc_large_system_hash("Dentry cache",
3172 sizeof(struct hlist_bl_head),
3175 HASH_EARLY | HASH_ZERO,
3180 d_hash_shift = 32 - d_hash_shift;
3183 static void __init dcache_init(void)
3186 * A constructor could be added for stable state like the lists,
3187 * but it is probably not worth it because of the cache nature
3190 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3191 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3194 /* Hash may have been set up in dcache_init_early */
3199 alloc_large_system_hash("Dentry cache",
3200 sizeof(struct hlist_bl_head),
3208 d_hash_shift = 32 - d_hash_shift;
3211 /* SLAB cache for __getname() consumers */
3212 struct kmem_cache *names_cachep __read_mostly;
3213 EXPORT_SYMBOL(names_cachep);
3215 void __init vfs_caches_init_early(void)
3219 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3220 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3222 dcache_init_early();
3226 void __init vfs_caches_init(void)
3228 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3229 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3234 files_maxfiles_init();