4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/ratelimit.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/security.h>
28 #include <linux/seqlock.h>
29 #include <linux/bootmem.h>
30 #include <linux/bit_spinlock.h>
31 #include <linux/rculist_bl.h>
32 #include <linux/list_lru.h>
38 * dcache->d_inode->i_lock protects:
39 * - i_dentry, d_u.d_alias, d_inode of aliases
40 * dcache_hash_bucket lock protects:
41 * - the dcache hash table
42 * s_roots bl list spinlock protects:
43 * - the s_roots list (see __d_drop)
44 * dentry->d_sb->s_dentry_lru_lock protects:
45 * - the dcache lru lists and counters
52 * - d_parent and d_subdirs
53 * - childrens' d_child and d_parent
54 * - d_u.d_alias, d_inode
57 * dentry->d_inode->i_lock
59 * dentry->d_sb->s_dentry_lru_lock
60 * dcache_hash_bucket lock
63 * If there is an ancestor relationship:
64 * dentry->d_parent->...->d_parent->d_lock
66 * dentry->d_parent->d_lock
69 * If no ancestor relationship:
70 * arbitrary, since it's serialized on rename_lock
72 int sysctl_vfs_cache_pressure __read_mostly = 100;
73 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
75 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
77 EXPORT_SYMBOL(rename_lock);
79 static struct kmem_cache *dentry_cache __read_mostly;
81 const struct qstr empty_name = QSTR_INIT("", 0);
82 EXPORT_SYMBOL(empty_name);
83 const struct qstr slash_name = QSTR_INIT("/", 1);
84 EXPORT_SYMBOL(slash_name);
87 * This is the single most critical data structure when it comes
88 * to the dcache: the hashtable for lookups. Somebody should try
89 * to make this good - I've just made it work.
91 * This hash-function tries to avoid losing too many bits of hash
92 * information, yet avoid using a prime hash-size or similar.
95 static unsigned int d_hash_shift __read_mostly;
97 static struct hlist_bl_head *dentry_hashtable __read_mostly;
99 static inline struct hlist_bl_head *d_hash(unsigned int hash)
101 return dentry_hashtable + (hash >> d_hash_shift);
104 #define IN_LOOKUP_SHIFT 10
105 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
107 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat = {
120 static DEFINE_PER_CPU(long, nr_dentry);
121 static DEFINE_PER_CPU(long, nr_dentry_unused);
123 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
126 * Here we resort to our own counters instead of using generic per-cpu counters
127 * for consistency with what the vfs inode code does. We are expected to harvest
128 * better code and performance by having our own specialized counters.
130 * Please note that the loop is done over all possible CPUs, not over all online
131 * CPUs. The reason for this is that we don't want to play games with CPUs going
132 * on and off. If one of them goes off, we will just keep their counters.
134 * glommer: See cffbc8a for details, and if you ever intend to change this,
135 * please update all vfs counters to match.
137 static long get_nr_dentry(void)
141 for_each_possible_cpu(i)
142 sum += per_cpu(nr_dentry, i);
143 return sum < 0 ? 0 : sum;
146 static long get_nr_dentry_unused(void)
150 for_each_possible_cpu(i)
151 sum += per_cpu(nr_dentry_unused, i);
152 return sum < 0 ? 0 : sum;
155 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
156 size_t *lenp, loff_t *ppos)
158 dentry_stat.nr_dentry = get_nr_dentry();
159 dentry_stat.nr_unused = get_nr_dentry_unused();
160 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
165 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
166 * The strings are both count bytes long, and count is non-zero.
168 #ifdef CONFIG_DCACHE_WORD_ACCESS
170 #include <asm/word-at-a-time.h>
172 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
173 * aligned allocation for this particular component. We don't
174 * strictly need the load_unaligned_zeropad() safety, but it
175 * doesn't hurt either.
177 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
178 * need the careful unaligned handling.
180 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
182 unsigned long a,b,mask;
185 a = read_word_at_a_time(cs);
186 b = load_unaligned_zeropad(ct);
187 if (tcount < sizeof(unsigned long))
189 if (unlikely(a != b))
191 cs += sizeof(unsigned long);
192 ct += sizeof(unsigned long);
193 tcount -= sizeof(unsigned long);
197 mask = bytemask_from_count(tcount);
198 return unlikely(!!((a ^ b) & mask));
203 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
217 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
220 * Be careful about RCU walk racing with rename:
221 * use 'READ_ONCE' to fetch the name pointer.
223 * NOTE! Even if a rename will mean that the length
224 * was not loaded atomically, we don't care. The
225 * RCU walk will check the sequence count eventually,
226 * and catch it. And we won't overrun the buffer,
227 * because we're reading the name pointer atomically,
228 * and a dentry name is guaranteed to be properly
229 * terminated with a NUL byte.
231 * End result: even if 'len' is wrong, we'll exit
232 * early because the data cannot match (there can
233 * be no NUL in the ct/tcount data)
235 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
237 return dentry_string_cmp(cs, ct, tcount);
240 struct external_name {
243 struct rcu_head head;
245 unsigned char name[];
248 static inline struct external_name *external_name(struct dentry *dentry)
250 return container_of(dentry->d_name.name, struct external_name, name[0]);
253 static void __d_free(struct rcu_head *head)
255 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
257 kmem_cache_free(dentry_cache, dentry);
260 static void __d_free_external_name(struct rcu_head *head)
262 struct external_name *name = container_of(head, struct external_name,
265 mod_node_page_state(page_pgdat(virt_to_page(name)),
266 NR_INDIRECTLY_RECLAIMABLE_BYTES,
272 static void __d_free_external(struct rcu_head *head)
274 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
276 __d_free_external_name(&external_name(dentry)->u.head);
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 if (unlikely(dname_external(dentry))) {
290 struct external_name *p = external_name(dentry);
291 atomic_inc(&p->u.count);
292 spin_unlock(&dentry->d_lock);
293 name->name = p->name;
295 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
296 spin_unlock(&dentry->d_lock);
297 name->name = name->inline_name;
300 EXPORT_SYMBOL(take_dentry_name_snapshot);
302 void release_dentry_name_snapshot(struct name_snapshot *name)
304 if (unlikely(name->name != name->inline_name)) {
305 struct external_name *p;
306 p = container_of(name->name, struct external_name, name[0]);
307 if (unlikely(atomic_dec_and_test(&p->u.count)))
308 call_rcu(&p->u.head, __d_free_external_name);
311 EXPORT_SYMBOL(release_dentry_name_snapshot);
313 static inline void __d_set_inode_and_type(struct dentry *dentry,
319 dentry->d_inode = inode;
320 flags = READ_ONCE(dentry->d_flags);
321 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
323 WRITE_ONCE(dentry->d_flags, flags);
326 static inline void __d_clear_type_and_inode(struct dentry *dentry)
328 unsigned flags = READ_ONCE(dentry->d_flags);
330 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
331 WRITE_ONCE(dentry->d_flags, flags);
332 dentry->d_inode = NULL;
335 static void dentry_free(struct dentry *dentry)
337 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
338 if (unlikely(dname_external(dentry))) {
339 struct external_name *p = external_name(dentry);
340 if (likely(atomic_dec_and_test(&p->u.count))) {
341 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
345 /* if dentry was never visible to RCU, immediate free is OK */
346 if (!(dentry->d_flags & DCACHE_RCUACCESS))
347 __d_free(&dentry->d_u.d_rcu);
349 call_rcu(&dentry->d_u.d_rcu, __d_free);
353 * Release the dentry's inode, using the filesystem
354 * d_iput() operation if defined.
356 static void dentry_unlink_inode(struct dentry * dentry)
357 __releases(dentry->d_lock)
358 __releases(dentry->d_inode->i_lock)
360 struct inode *inode = dentry->d_inode;
362 raw_write_seqcount_begin(&dentry->d_seq);
363 __d_clear_type_and_inode(dentry);
364 hlist_del_init(&dentry->d_u.d_alias);
365 raw_write_seqcount_end(&dentry->d_seq);
366 spin_unlock(&dentry->d_lock);
367 spin_unlock(&inode->i_lock);
369 fsnotify_inoderemove(inode);
370 if (dentry->d_op && dentry->d_op->d_iput)
371 dentry->d_op->d_iput(dentry, inode);
377 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
378 * is in use - which includes both the "real" per-superblock
379 * LRU list _and_ the DCACHE_SHRINK_LIST use.
381 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
382 * on the shrink list (ie not on the superblock LRU list).
384 * The per-cpu "nr_dentry_unused" counters are updated with
385 * the DCACHE_LRU_LIST bit.
387 * These helper functions make sure we always follow the
388 * rules. d_lock must be held by the caller.
390 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
391 static void d_lru_add(struct dentry *dentry)
393 D_FLAG_VERIFY(dentry, 0);
394 dentry->d_flags |= DCACHE_LRU_LIST;
395 this_cpu_inc(nr_dentry_unused);
396 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
399 static void d_lru_del(struct dentry *dentry)
401 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
402 dentry->d_flags &= ~DCACHE_LRU_LIST;
403 this_cpu_dec(nr_dentry_unused);
404 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
407 static void d_shrink_del(struct dentry *dentry)
409 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
410 list_del_init(&dentry->d_lru);
411 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
412 this_cpu_dec(nr_dentry_unused);
415 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
417 D_FLAG_VERIFY(dentry, 0);
418 list_add(&dentry->d_lru, list);
419 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
420 this_cpu_inc(nr_dentry_unused);
424 * These can only be called under the global LRU lock, ie during the
425 * callback for freeing the LRU list. "isolate" removes it from the
426 * LRU lists entirely, while shrink_move moves it to the indicated
429 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
431 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
432 dentry->d_flags &= ~DCACHE_LRU_LIST;
433 this_cpu_dec(nr_dentry_unused);
434 list_lru_isolate(lru, &dentry->d_lru);
437 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
438 struct list_head *list)
440 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
441 dentry->d_flags |= DCACHE_SHRINK_LIST;
442 list_lru_isolate_move(lru, &dentry->d_lru, list);
446 * d_drop - drop a dentry
447 * @dentry: dentry to drop
449 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
450 * be found through a VFS lookup any more. Note that this is different from
451 * deleting the dentry - d_delete will try to mark the dentry negative if
452 * possible, giving a successful _negative_ lookup, while d_drop will
453 * just make the cache lookup fail.
455 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
456 * reason (NFS timeouts or autofs deletes).
458 * __d_drop requires dentry->d_lock
459 * ___d_drop doesn't mark dentry as "unhashed"
460 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
462 static void ___d_drop(struct dentry *dentry)
464 struct hlist_bl_head *b;
466 * Hashed dentries are normally on the dentry hashtable,
467 * with the exception of those newly allocated by
468 * d_obtain_root, which are always IS_ROOT:
470 if (unlikely(IS_ROOT(dentry)))
471 b = &dentry->d_sb->s_roots;
473 b = d_hash(dentry->d_name.hash);
476 __hlist_bl_del(&dentry->d_hash);
480 void __d_drop(struct dentry *dentry)
482 if (!d_unhashed(dentry)) {
484 dentry->d_hash.pprev = NULL;
485 write_seqcount_invalidate(&dentry->d_seq);
488 EXPORT_SYMBOL(__d_drop);
490 void d_drop(struct dentry *dentry)
492 spin_lock(&dentry->d_lock);
494 spin_unlock(&dentry->d_lock);
496 EXPORT_SYMBOL(d_drop);
498 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
502 * Inform d_walk() and shrink_dentry_list() that we are no longer
503 * attached to the dentry tree
505 dentry->d_flags |= DCACHE_DENTRY_KILLED;
506 if (unlikely(list_empty(&dentry->d_child)))
508 __list_del_entry(&dentry->d_child);
510 * Cursors can move around the list of children. While we'd been
511 * a normal list member, it didn't matter - ->d_child.next would've
512 * been updated. However, from now on it won't be and for the
513 * things like d_walk() it might end up with a nasty surprise.
514 * Normally d_walk() doesn't care about cursors moving around -
515 * ->d_lock on parent prevents that and since a cursor has no children
516 * of its own, we get through it without ever unlocking the parent.
517 * There is one exception, though - if we ascend from a child that
518 * gets killed as soon as we unlock it, the next sibling is found
519 * using the value left in its ->d_child.next. And if _that_
520 * pointed to a cursor, and cursor got moved (e.g. by lseek())
521 * before d_walk() regains parent->d_lock, we'll end up skipping
522 * everything the cursor had been moved past.
524 * Solution: make sure that the pointer left behind in ->d_child.next
525 * points to something that won't be moving around. I.e. skip the
528 while (dentry->d_child.next != &parent->d_subdirs) {
529 next = list_entry(dentry->d_child.next, struct dentry, d_child);
530 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
532 dentry->d_child.next = next->d_child.next;
536 static void __dentry_kill(struct dentry *dentry)
538 struct dentry *parent = NULL;
539 bool can_free = true;
540 if (!IS_ROOT(dentry))
541 parent = dentry->d_parent;
544 * The dentry is now unrecoverably dead to the world.
546 lockref_mark_dead(&dentry->d_lockref);
549 * inform the fs via d_prune that this dentry is about to be
550 * unhashed and destroyed.
552 if (dentry->d_flags & DCACHE_OP_PRUNE)
553 dentry->d_op->d_prune(dentry);
555 if (dentry->d_flags & DCACHE_LRU_LIST) {
556 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
559 /* if it was on the hash then remove it */
561 dentry_unlist(dentry, parent);
563 spin_unlock(&parent->d_lock);
565 dentry_unlink_inode(dentry);
567 spin_unlock(&dentry->d_lock);
568 this_cpu_dec(nr_dentry);
569 if (dentry->d_op && dentry->d_op->d_release)
570 dentry->d_op->d_release(dentry);
572 spin_lock(&dentry->d_lock);
573 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
574 dentry->d_flags |= DCACHE_MAY_FREE;
577 spin_unlock(&dentry->d_lock);
578 if (likely(can_free))
583 static struct dentry *__lock_parent(struct dentry *dentry)
585 struct dentry *parent;
587 spin_unlock(&dentry->d_lock);
589 parent = READ_ONCE(dentry->d_parent);
590 spin_lock(&parent->d_lock);
592 * We can't blindly lock dentry until we are sure
593 * that we won't violate the locking order.
594 * Any changes of dentry->d_parent must have
595 * been done with parent->d_lock held, so
596 * spin_lock() above is enough of a barrier
597 * for checking if it's still our child.
599 if (unlikely(parent != dentry->d_parent)) {
600 spin_unlock(&parent->d_lock);
604 if (parent != dentry)
605 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
611 static inline struct dentry *lock_parent(struct dentry *dentry)
613 struct dentry *parent = dentry->d_parent;
616 if (likely(spin_trylock(&parent->d_lock)))
618 return __lock_parent(dentry);
621 static inline bool retain_dentry(struct dentry *dentry)
623 WARN_ON(d_in_lookup(dentry));
625 /* Unreachable? Get rid of it */
626 if (unlikely(d_unhashed(dentry)))
629 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
632 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
633 if (dentry->d_op->d_delete(dentry))
636 /* retain; LRU fodder */
637 dentry->d_lockref.count--;
638 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
640 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
641 dentry->d_flags |= DCACHE_REFERENCED;
646 * Finish off a dentry we've decided to kill.
647 * dentry->d_lock must be held, returns with it unlocked.
648 * Returns dentry requiring refcount drop, or NULL if we're done.
650 static struct dentry *dentry_kill(struct dentry *dentry)
651 __releases(dentry->d_lock)
653 struct inode *inode = dentry->d_inode;
654 struct dentry *parent = NULL;
656 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
659 if (!IS_ROOT(dentry)) {
660 parent = dentry->d_parent;
661 if (unlikely(!spin_trylock(&parent->d_lock))) {
662 parent = __lock_parent(dentry);
663 if (likely(inode || !dentry->d_inode))
665 /* negative that became positive */
667 spin_unlock(&parent->d_lock);
668 inode = dentry->d_inode;
672 __dentry_kill(dentry);
676 spin_unlock(&dentry->d_lock);
677 spin_lock(&inode->i_lock);
678 spin_lock(&dentry->d_lock);
679 parent = lock_parent(dentry);
681 if (unlikely(dentry->d_lockref.count != 1)) {
682 dentry->d_lockref.count--;
683 } else if (likely(!retain_dentry(dentry))) {
684 __dentry_kill(dentry);
687 /* we are keeping it, after all */
689 spin_unlock(&inode->i_lock);
691 spin_unlock(&parent->d_lock);
692 spin_unlock(&dentry->d_lock);
697 * Try to do a lockless dput(), and return whether that was successful.
699 * If unsuccessful, we return false, having already taken the dentry lock.
701 * The caller needs to hold the RCU read lock, so that the dentry is
702 * guaranteed to stay around even if the refcount goes down to zero!
704 static inline bool fast_dput(struct dentry *dentry)
707 unsigned int d_flags;
710 * If we have a d_op->d_delete() operation, we sould not
711 * let the dentry count go to zero, so use "put_or_lock".
713 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
714 return lockref_put_or_lock(&dentry->d_lockref);
717 * .. otherwise, we can try to just decrement the
718 * lockref optimistically.
720 ret = lockref_put_return(&dentry->d_lockref);
723 * If the lockref_put_return() failed due to the lock being held
724 * by somebody else, the fast path has failed. We will need to
725 * get the lock, and then check the count again.
727 if (unlikely(ret < 0)) {
728 spin_lock(&dentry->d_lock);
729 if (dentry->d_lockref.count > 1) {
730 dentry->d_lockref.count--;
731 spin_unlock(&dentry->d_lock);
738 * If we weren't the last ref, we're done.
744 * Careful, careful. The reference count went down
745 * to zero, but we don't hold the dentry lock, so
746 * somebody else could get it again, and do another
747 * dput(), and we need to not race with that.
749 * However, there is a very special and common case
750 * where we don't care, because there is nothing to
751 * do: the dentry is still hashed, it does not have
752 * a 'delete' op, and it's referenced and already on
755 * NOTE! Since we aren't locked, these values are
756 * not "stable". However, it is sufficient that at
757 * some point after we dropped the reference the
758 * dentry was hashed and the flags had the proper
759 * value. Other dentry users may have re-gotten
760 * a reference to the dentry and change that, but
761 * our work is done - we can leave the dentry
762 * around with a zero refcount.
765 d_flags = READ_ONCE(dentry->d_flags);
766 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
768 /* Nothing to do? Dropping the reference was all we needed? */
769 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
773 * Not the fast normal case? Get the lock. We've already decremented
774 * the refcount, but we'll need to re-check the situation after
777 spin_lock(&dentry->d_lock);
780 * Did somebody else grab a reference to it in the meantime, and
781 * we're no longer the last user after all? Alternatively, somebody
782 * else could have killed it and marked it dead. Either way, we
783 * don't need to do anything else.
785 if (dentry->d_lockref.count) {
786 spin_unlock(&dentry->d_lock);
791 * Re-get the reference we optimistically dropped. We hold the
792 * lock, and we just tested that it was zero, so we can just
795 dentry->d_lockref.count = 1;
803 * This is complicated by the fact that we do not want to put
804 * dentries that are no longer on any hash chain on the unused
805 * list: we'd much rather just get rid of them immediately.
807 * However, that implies that we have to traverse the dentry
808 * tree upwards to the parents which might _also_ now be
809 * scheduled for deletion (it may have been only waiting for
810 * its last child to go away).
812 * This tail recursion is done by hand as we don't want to depend
813 * on the compiler to always get this right (gcc generally doesn't).
814 * Real recursion would eat up our stack space.
818 * dput - release a dentry
819 * @dentry: dentry to release
821 * Release a dentry. This will drop the usage count and if appropriate
822 * call the dentry unlink method as well as removing it from the queues and
823 * releasing its resources. If the parent dentries were scheduled for release
824 * they too may now get deleted.
826 void dput(struct dentry *dentry)
832 if (likely(fast_dput(dentry))) {
837 /* Slow case: now with the dentry lock held */
840 if (likely(retain_dentry(dentry))) {
841 spin_unlock(&dentry->d_lock);
845 dentry = dentry_kill(dentry);
851 /* This must be called with d_lock held */
852 static inline void __dget_dlock(struct dentry *dentry)
854 dentry->d_lockref.count++;
857 static inline void __dget(struct dentry *dentry)
859 lockref_get(&dentry->d_lockref);
862 struct dentry *dget_parent(struct dentry *dentry)
868 * Do optimistic parent lookup without any
872 ret = READ_ONCE(dentry->d_parent);
873 gotref = lockref_get_not_zero(&ret->d_lockref);
875 if (likely(gotref)) {
876 if (likely(ret == READ_ONCE(dentry->d_parent)))
883 * Don't need rcu_dereference because we re-check it was correct under
887 ret = dentry->d_parent;
888 spin_lock(&ret->d_lock);
889 if (unlikely(ret != dentry->d_parent)) {
890 spin_unlock(&ret->d_lock);
895 BUG_ON(!ret->d_lockref.count);
896 ret->d_lockref.count++;
897 spin_unlock(&ret->d_lock);
900 EXPORT_SYMBOL(dget_parent);
902 static struct dentry * __d_find_any_alias(struct inode *inode)
904 struct dentry *alias;
906 if (hlist_empty(&inode->i_dentry))
908 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
914 * d_find_any_alias - find any alias for a given inode
915 * @inode: inode to find an alias for
917 * If any aliases exist for the given inode, take and return a
918 * reference for one of them. If no aliases exist, return %NULL.
920 struct dentry *d_find_any_alias(struct inode *inode)
924 spin_lock(&inode->i_lock);
925 de = __d_find_any_alias(inode);
926 spin_unlock(&inode->i_lock);
929 EXPORT_SYMBOL(d_find_any_alias);
932 * d_find_alias - grab a hashed alias of inode
933 * @inode: inode in question
935 * If inode has a hashed alias, or is a directory and has any alias,
936 * acquire the reference to alias and return it. Otherwise return NULL.
937 * Notice that if inode is a directory there can be only one alias and
938 * it can be unhashed only if it has no children, or if it is the root
939 * of a filesystem, or if the directory was renamed and d_revalidate
940 * was the first vfs operation to notice.
942 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
943 * any other hashed alias over that one.
945 static struct dentry *__d_find_alias(struct inode *inode)
947 struct dentry *alias;
949 if (S_ISDIR(inode->i_mode))
950 return __d_find_any_alias(inode);
952 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
953 spin_lock(&alias->d_lock);
954 if (!d_unhashed(alias)) {
956 spin_unlock(&alias->d_lock);
959 spin_unlock(&alias->d_lock);
964 struct dentry *d_find_alias(struct inode *inode)
966 struct dentry *de = NULL;
968 if (!hlist_empty(&inode->i_dentry)) {
969 spin_lock(&inode->i_lock);
970 de = __d_find_alias(inode);
971 spin_unlock(&inode->i_lock);
975 EXPORT_SYMBOL(d_find_alias);
978 * Try to kill dentries associated with this inode.
979 * WARNING: you must own a reference to inode.
981 void d_prune_aliases(struct inode *inode)
983 struct dentry *dentry;
985 spin_lock(&inode->i_lock);
986 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
987 spin_lock(&dentry->d_lock);
988 if (!dentry->d_lockref.count) {
989 struct dentry *parent = lock_parent(dentry);
990 if (likely(!dentry->d_lockref.count)) {
991 __dentry_kill(dentry);
996 spin_unlock(&parent->d_lock);
998 spin_unlock(&dentry->d_lock);
1000 spin_unlock(&inode->i_lock);
1002 EXPORT_SYMBOL(d_prune_aliases);
1005 * Lock a dentry from shrink list.
1006 * Called under rcu_read_lock() and dentry->d_lock; the former
1007 * guarantees that nothing we access will be freed under us.
1008 * Note that dentry is *not* protected from concurrent dentry_kill(),
1011 * Return false if dentry has been disrupted or grabbed, leaving
1012 * the caller to kick it off-list. Otherwise, return true and have
1013 * that dentry's inode and parent both locked.
1015 static bool shrink_lock_dentry(struct dentry *dentry)
1017 struct inode *inode;
1018 struct dentry *parent;
1020 if (dentry->d_lockref.count)
1023 inode = dentry->d_inode;
1024 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1025 spin_unlock(&dentry->d_lock);
1026 spin_lock(&inode->i_lock);
1027 spin_lock(&dentry->d_lock);
1028 if (unlikely(dentry->d_lockref.count))
1030 /* changed inode means that somebody had grabbed it */
1031 if (unlikely(inode != dentry->d_inode))
1035 parent = dentry->d_parent;
1036 if (IS_ROOT(dentry) || likely(spin_trylock(&parent->d_lock)))
1039 spin_unlock(&dentry->d_lock);
1040 spin_lock(&parent->d_lock);
1041 if (unlikely(parent != dentry->d_parent)) {
1042 spin_unlock(&parent->d_lock);
1043 spin_lock(&dentry->d_lock);
1046 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1047 if (likely(!dentry->d_lockref.count))
1049 spin_unlock(&parent->d_lock);
1052 spin_unlock(&inode->i_lock);
1056 static void shrink_dentry_list(struct list_head *list)
1058 while (!list_empty(list)) {
1059 struct dentry *dentry, *parent;
1061 dentry = list_entry(list->prev, struct dentry, d_lru);
1062 spin_lock(&dentry->d_lock);
1064 if (!shrink_lock_dentry(dentry)) {
1065 bool can_free = false;
1067 d_shrink_del(dentry);
1068 if (dentry->d_lockref.count < 0)
1069 can_free = dentry->d_flags & DCACHE_MAY_FREE;
1070 spin_unlock(&dentry->d_lock);
1072 dentry_free(dentry);
1076 d_shrink_del(dentry);
1077 parent = dentry->d_parent;
1078 __dentry_kill(dentry);
1079 if (parent == dentry)
1082 * We need to prune ancestors too. This is necessary to prevent
1083 * quadratic behavior of shrink_dcache_parent(), but is also
1084 * expected to be beneficial in reducing dentry cache
1088 while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
1089 dentry = dentry_kill(dentry);
1093 static enum lru_status dentry_lru_isolate(struct list_head *item,
1094 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1096 struct list_head *freeable = arg;
1097 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1101 * we are inverting the lru lock/dentry->d_lock here,
1102 * so use a trylock. If we fail to get the lock, just skip
1105 if (!spin_trylock(&dentry->d_lock))
1109 * Referenced dentries are still in use. If they have active
1110 * counts, just remove them from the LRU. Otherwise give them
1111 * another pass through the LRU.
1113 if (dentry->d_lockref.count) {
1114 d_lru_isolate(lru, dentry);
1115 spin_unlock(&dentry->d_lock);
1119 if (dentry->d_flags & DCACHE_REFERENCED) {
1120 dentry->d_flags &= ~DCACHE_REFERENCED;
1121 spin_unlock(&dentry->d_lock);
1124 * The list move itself will be made by the common LRU code. At
1125 * this point, we've dropped the dentry->d_lock but keep the
1126 * lru lock. This is safe to do, since every list movement is
1127 * protected by the lru lock even if both locks are held.
1129 * This is guaranteed by the fact that all LRU management
1130 * functions are intermediated by the LRU API calls like
1131 * list_lru_add and list_lru_del. List movement in this file
1132 * only ever occur through this functions or through callbacks
1133 * like this one, that are called from the LRU API.
1135 * The only exceptions to this are functions like
1136 * shrink_dentry_list, and code that first checks for the
1137 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1138 * operating only with stack provided lists after they are
1139 * properly isolated from the main list. It is thus, always a
1145 d_lru_shrink_move(lru, dentry, freeable);
1146 spin_unlock(&dentry->d_lock);
1152 * prune_dcache_sb - shrink the dcache
1154 * @sc: shrink control, passed to list_lru_shrink_walk()
1156 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1157 * is done when we need more memory and called from the superblock shrinker
1160 * This function may fail to free any resources if all the dentries are in
1163 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1168 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1169 dentry_lru_isolate, &dispose);
1170 shrink_dentry_list(&dispose);
1174 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1175 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1177 struct list_head *freeable = arg;
1178 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1181 * we are inverting the lru lock/dentry->d_lock here,
1182 * so use a trylock. If we fail to get the lock, just skip
1185 if (!spin_trylock(&dentry->d_lock))
1188 d_lru_shrink_move(lru, dentry, freeable);
1189 spin_unlock(&dentry->d_lock);
1196 * shrink_dcache_sb - shrink dcache for a superblock
1199 * Shrink the dcache for the specified super block. This is used to free
1200 * the dcache before unmounting a file system.
1202 void shrink_dcache_sb(struct super_block *sb)
1209 freed = list_lru_walk(&sb->s_dentry_lru,
1210 dentry_lru_isolate_shrink, &dispose, 1024);
1212 this_cpu_sub(nr_dentry_unused, freed);
1213 shrink_dentry_list(&dispose);
1214 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1216 EXPORT_SYMBOL(shrink_dcache_sb);
1219 * enum d_walk_ret - action to talke during tree walk
1220 * @D_WALK_CONTINUE: contrinue walk
1221 * @D_WALK_QUIT: quit walk
1222 * @D_WALK_NORETRY: quit when retry is needed
1223 * @D_WALK_SKIP: skip this dentry and its children
1233 * d_walk - walk the dentry tree
1234 * @parent: start of walk
1235 * @data: data passed to @enter() and @finish()
1236 * @enter: callback when first entering the dentry
1238 * The @enter() callbacks are called with d_lock held.
1240 static void d_walk(struct dentry *parent, void *data,
1241 enum d_walk_ret (*enter)(void *, struct dentry *))
1243 struct dentry *this_parent;
1244 struct list_head *next;
1246 enum d_walk_ret ret;
1250 read_seqbegin_or_lock(&rename_lock, &seq);
1251 this_parent = parent;
1252 spin_lock(&this_parent->d_lock);
1254 ret = enter(data, this_parent);
1256 case D_WALK_CONTINUE:
1261 case D_WALK_NORETRY:
1266 next = this_parent->d_subdirs.next;
1268 while (next != &this_parent->d_subdirs) {
1269 struct list_head *tmp = next;
1270 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1273 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1276 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1278 ret = enter(data, dentry);
1280 case D_WALK_CONTINUE:
1283 spin_unlock(&dentry->d_lock);
1285 case D_WALK_NORETRY:
1289 spin_unlock(&dentry->d_lock);
1293 if (!list_empty(&dentry->d_subdirs)) {
1294 spin_unlock(&this_parent->d_lock);
1295 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1296 this_parent = dentry;
1297 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1300 spin_unlock(&dentry->d_lock);
1303 * All done at this level ... ascend and resume the search.
1307 if (this_parent != parent) {
1308 struct dentry *child = this_parent;
1309 this_parent = child->d_parent;
1311 spin_unlock(&child->d_lock);
1312 spin_lock(&this_parent->d_lock);
1314 /* might go back up the wrong parent if we have had a rename. */
1315 if (need_seqretry(&rename_lock, seq))
1317 /* go into the first sibling still alive */
1319 next = child->d_child.next;
1320 if (next == &this_parent->d_subdirs)
1322 child = list_entry(next, struct dentry, d_child);
1323 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1327 if (need_seqretry(&rename_lock, seq))
1332 spin_unlock(&this_parent->d_lock);
1333 done_seqretry(&rename_lock, seq);
1337 spin_unlock(&this_parent->d_lock);
1346 struct check_mount {
1347 struct vfsmount *mnt;
1348 unsigned int mounted;
1351 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1353 struct check_mount *info = data;
1354 struct path path = { .mnt = info->mnt, .dentry = dentry };
1356 if (likely(!d_mountpoint(dentry)))
1357 return D_WALK_CONTINUE;
1358 if (__path_is_mountpoint(&path)) {
1362 return D_WALK_CONTINUE;
1366 * path_has_submounts - check for mounts over a dentry in the
1367 * current namespace.
1368 * @parent: path to check.
1370 * Return true if the parent or its subdirectories contain
1371 * a mount point in the current namespace.
1373 int path_has_submounts(const struct path *parent)
1375 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1377 read_seqlock_excl(&mount_lock);
1378 d_walk(parent->dentry, &data, path_check_mount);
1379 read_sequnlock_excl(&mount_lock);
1381 return data.mounted;
1383 EXPORT_SYMBOL(path_has_submounts);
1386 * Called by mount code to set a mountpoint and check if the mountpoint is
1387 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1388 * subtree can become unreachable).
1390 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1391 * this reason take rename_lock and d_lock on dentry and ancestors.
1393 int d_set_mounted(struct dentry *dentry)
1397 write_seqlock(&rename_lock);
1398 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1399 /* Need exclusion wrt. d_invalidate() */
1400 spin_lock(&p->d_lock);
1401 if (unlikely(d_unhashed(p))) {
1402 spin_unlock(&p->d_lock);
1405 spin_unlock(&p->d_lock);
1407 spin_lock(&dentry->d_lock);
1408 if (!d_unlinked(dentry)) {
1410 if (!d_mountpoint(dentry)) {
1411 dentry->d_flags |= DCACHE_MOUNTED;
1415 spin_unlock(&dentry->d_lock);
1417 write_sequnlock(&rename_lock);
1422 * Search the dentry child list of the specified parent,
1423 * and move any unused dentries to the end of the unused
1424 * list for prune_dcache(). We descend to the next level
1425 * whenever the d_subdirs list is non-empty and continue
1428 * It returns zero iff there are no unused children,
1429 * otherwise it returns the number of children moved to
1430 * the end of the unused list. This may not be the total
1431 * number of unused children, because select_parent can
1432 * drop the lock and return early due to latency
1436 struct select_data {
1437 struct dentry *start;
1438 struct list_head dispose;
1442 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1444 struct select_data *data = _data;
1445 enum d_walk_ret ret = D_WALK_CONTINUE;
1447 if (data->start == dentry)
1450 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1453 if (dentry->d_flags & DCACHE_LRU_LIST)
1455 if (!dentry->d_lockref.count) {
1456 d_shrink_add(dentry, &data->dispose);
1461 * We can return to the caller if we have found some (this
1462 * ensures forward progress). We'll be coming back to find
1465 if (!list_empty(&data->dispose))
1466 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1472 * shrink_dcache_parent - prune dcache
1473 * @parent: parent of entries to prune
1475 * Prune the dcache to remove unused children of the parent dentry.
1477 void shrink_dcache_parent(struct dentry *parent)
1480 struct select_data data;
1482 INIT_LIST_HEAD(&data.dispose);
1483 data.start = parent;
1486 d_walk(parent, &data, select_collect);
1488 if (!list_empty(&data.dispose)) {
1489 shrink_dentry_list(&data.dispose);
1498 EXPORT_SYMBOL(shrink_dcache_parent);
1500 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1502 /* it has busy descendents; complain about those instead */
1503 if (!list_empty(&dentry->d_subdirs))
1504 return D_WALK_CONTINUE;
1506 /* root with refcount 1 is fine */
1507 if (dentry == _data && dentry->d_lockref.count == 1)
1508 return D_WALK_CONTINUE;
1510 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1511 " still in use (%d) [unmount of %s %s]\n",
1514 dentry->d_inode->i_ino : 0UL,
1516 dentry->d_lockref.count,
1517 dentry->d_sb->s_type->name,
1518 dentry->d_sb->s_id);
1520 return D_WALK_CONTINUE;
1523 static void do_one_tree(struct dentry *dentry)
1525 shrink_dcache_parent(dentry);
1526 d_walk(dentry, dentry, umount_check);
1532 * destroy the dentries attached to a superblock on unmounting
1534 void shrink_dcache_for_umount(struct super_block *sb)
1536 struct dentry *dentry;
1538 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1540 dentry = sb->s_root;
1542 do_one_tree(dentry);
1544 while (!hlist_bl_empty(&sb->s_roots)) {
1545 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1546 do_one_tree(dentry);
1550 static enum d_walk_ret find_submount(void *_data, struct dentry *dentry)
1552 struct dentry **victim = _data;
1553 if (d_mountpoint(dentry)) {
1554 __dget_dlock(dentry);
1558 return D_WALK_CONTINUE;
1562 * d_invalidate - detach submounts, prune dcache, and drop
1563 * @dentry: dentry to invalidate (aka detach, prune and drop)
1565 void d_invalidate(struct dentry *dentry)
1567 bool had_submounts = false;
1568 spin_lock(&dentry->d_lock);
1569 if (d_unhashed(dentry)) {
1570 spin_unlock(&dentry->d_lock);
1574 spin_unlock(&dentry->d_lock);
1576 /* Negative dentries can be dropped without further checks */
1577 if (!dentry->d_inode)
1580 shrink_dcache_parent(dentry);
1582 struct dentry *victim = NULL;
1583 d_walk(dentry, &victim, find_submount);
1586 shrink_dcache_parent(dentry);
1589 had_submounts = true;
1590 detach_mounts(victim);
1594 EXPORT_SYMBOL(d_invalidate);
1597 * __d_alloc - allocate a dcache entry
1598 * @sb: filesystem it will belong to
1599 * @name: qstr of the name
1601 * Allocates a dentry. It returns %NULL if there is insufficient memory
1602 * available. On a success the dentry is returned. The name passed in is
1603 * copied and the copy passed in may be reused after this call.
1606 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1608 struct external_name *ext = NULL;
1609 struct dentry *dentry;
1613 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1618 * We guarantee that the inline name is always NUL-terminated.
1619 * This way the memcpy() done by the name switching in rename
1620 * will still always have a NUL at the end, even if we might
1621 * be overwriting an internal NUL character
1623 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1624 if (unlikely(!name)) {
1626 dname = dentry->d_iname;
1627 } else if (name->len > DNAME_INLINE_LEN-1) {
1628 size_t size = offsetof(struct external_name, name[1]);
1630 ext = kmalloc(size + name->len, GFP_KERNEL_ACCOUNT);
1632 kmem_cache_free(dentry_cache, dentry);
1635 atomic_set(&ext->u.count, 1);
1638 dname = dentry->d_iname;
1641 dentry->d_name.len = name->len;
1642 dentry->d_name.hash = name->hash;
1643 memcpy(dname, name->name, name->len);
1644 dname[name->len] = 0;
1646 /* Make sure we always see the terminating NUL character */
1647 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1649 dentry->d_lockref.count = 1;
1650 dentry->d_flags = 0;
1651 spin_lock_init(&dentry->d_lock);
1652 seqcount_init(&dentry->d_seq);
1653 dentry->d_inode = NULL;
1654 dentry->d_parent = dentry;
1656 dentry->d_op = NULL;
1657 dentry->d_fsdata = NULL;
1658 INIT_HLIST_BL_NODE(&dentry->d_hash);
1659 INIT_LIST_HEAD(&dentry->d_lru);
1660 INIT_LIST_HEAD(&dentry->d_subdirs);
1661 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1662 INIT_LIST_HEAD(&dentry->d_child);
1663 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1665 if (dentry->d_op && dentry->d_op->d_init) {
1666 err = dentry->d_op->d_init(dentry);
1668 if (dname_external(dentry))
1669 kfree(external_name(dentry));
1670 kmem_cache_free(dentry_cache, dentry);
1675 if (unlikely(ext)) {
1676 pg_data_t *pgdat = page_pgdat(virt_to_page(ext));
1677 mod_node_page_state(pgdat, NR_INDIRECTLY_RECLAIMABLE_BYTES,
1681 this_cpu_inc(nr_dentry);
1687 * d_alloc - allocate a dcache entry
1688 * @parent: parent of entry to allocate
1689 * @name: qstr of the name
1691 * Allocates a dentry. It returns %NULL if there is insufficient memory
1692 * available. On a success the dentry is returned. The name passed in is
1693 * copied and the copy passed in may be reused after this call.
1695 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1697 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1700 dentry->d_flags |= DCACHE_RCUACCESS;
1701 spin_lock(&parent->d_lock);
1703 * don't need child lock because it is not subject
1704 * to concurrency here
1706 __dget_dlock(parent);
1707 dentry->d_parent = parent;
1708 list_add(&dentry->d_child, &parent->d_subdirs);
1709 spin_unlock(&parent->d_lock);
1713 EXPORT_SYMBOL(d_alloc);
1715 struct dentry *d_alloc_anon(struct super_block *sb)
1717 return __d_alloc(sb, NULL);
1719 EXPORT_SYMBOL(d_alloc_anon);
1721 struct dentry *d_alloc_cursor(struct dentry * parent)
1723 struct dentry *dentry = d_alloc_anon(parent->d_sb);
1725 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1726 dentry->d_parent = dget(parent);
1732 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1733 * @sb: the superblock
1734 * @name: qstr of the name
1736 * For a filesystem that just pins its dentries in memory and never
1737 * performs lookups at all, return an unhashed IS_ROOT dentry.
1739 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1741 return __d_alloc(sb, name);
1743 EXPORT_SYMBOL(d_alloc_pseudo);
1745 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1750 q.hash_len = hashlen_string(parent, name);
1751 return d_alloc(parent, &q);
1753 EXPORT_SYMBOL(d_alloc_name);
1755 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1757 WARN_ON_ONCE(dentry->d_op);
1758 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1760 DCACHE_OP_REVALIDATE |
1761 DCACHE_OP_WEAK_REVALIDATE |
1768 dentry->d_flags |= DCACHE_OP_HASH;
1770 dentry->d_flags |= DCACHE_OP_COMPARE;
1771 if (op->d_revalidate)
1772 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1773 if (op->d_weak_revalidate)
1774 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1776 dentry->d_flags |= DCACHE_OP_DELETE;
1778 dentry->d_flags |= DCACHE_OP_PRUNE;
1780 dentry->d_flags |= DCACHE_OP_REAL;
1783 EXPORT_SYMBOL(d_set_d_op);
1787 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1788 * @dentry - The dentry to mark
1790 * Mark a dentry as falling through to the lower layer (as set with
1791 * d_pin_lower()). This flag may be recorded on the medium.
1793 void d_set_fallthru(struct dentry *dentry)
1795 spin_lock(&dentry->d_lock);
1796 dentry->d_flags |= DCACHE_FALLTHRU;
1797 spin_unlock(&dentry->d_lock);
1799 EXPORT_SYMBOL(d_set_fallthru);
1801 static unsigned d_flags_for_inode(struct inode *inode)
1803 unsigned add_flags = DCACHE_REGULAR_TYPE;
1806 return DCACHE_MISS_TYPE;
1808 if (S_ISDIR(inode->i_mode)) {
1809 add_flags = DCACHE_DIRECTORY_TYPE;
1810 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1811 if (unlikely(!inode->i_op->lookup))
1812 add_flags = DCACHE_AUTODIR_TYPE;
1814 inode->i_opflags |= IOP_LOOKUP;
1816 goto type_determined;
1819 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1820 if (unlikely(inode->i_op->get_link)) {
1821 add_flags = DCACHE_SYMLINK_TYPE;
1822 goto type_determined;
1824 inode->i_opflags |= IOP_NOFOLLOW;
1827 if (unlikely(!S_ISREG(inode->i_mode)))
1828 add_flags = DCACHE_SPECIAL_TYPE;
1831 if (unlikely(IS_AUTOMOUNT(inode)))
1832 add_flags |= DCACHE_NEED_AUTOMOUNT;
1836 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1838 unsigned add_flags = d_flags_for_inode(inode);
1839 WARN_ON(d_in_lookup(dentry));
1841 spin_lock(&dentry->d_lock);
1842 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1843 raw_write_seqcount_begin(&dentry->d_seq);
1844 __d_set_inode_and_type(dentry, inode, add_flags);
1845 raw_write_seqcount_end(&dentry->d_seq);
1846 fsnotify_update_flags(dentry);
1847 spin_unlock(&dentry->d_lock);
1851 * d_instantiate - fill in inode information for a dentry
1852 * @entry: dentry to complete
1853 * @inode: inode to attach to this dentry
1855 * Fill in inode information in the entry.
1857 * This turns negative dentries into productive full members
1860 * NOTE! This assumes that the inode count has been incremented
1861 * (or otherwise set) by the caller to indicate that it is now
1862 * in use by the dcache.
1865 void d_instantiate(struct dentry *entry, struct inode * inode)
1867 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1869 security_d_instantiate(entry, inode);
1870 spin_lock(&inode->i_lock);
1871 __d_instantiate(entry, inode);
1872 spin_unlock(&inode->i_lock);
1875 EXPORT_SYMBOL(d_instantiate);
1878 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1879 * with lockdep-related part of unlock_new_inode() done before
1880 * anything else. Use that instead of open-coding d_instantiate()/
1881 * unlock_new_inode() combinations.
1883 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1885 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1887 lockdep_annotate_inode_mutex_key(inode);
1888 security_d_instantiate(entry, inode);
1889 spin_lock(&inode->i_lock);
1890 __d_instantiate(entry, inode);
1891 WARN_ON(!(inode->i_state & I_NEW));
1892 inode->i_state &= ~I_NEW;
1894 wake_up_bit(&inode->i_state, __I_NEW);
1895 spin_unlock(&inode->i_lock);
1897 EXPORT_SYMBOL(d_instantiate_new);
1900 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1901 * @entry: dentry to complete
1902 * @inode: inode to attach to this dentry
1904 * Fill in inode information in the entry. If a directory alias is found, then
1905 * return an error (and drop inode). Together with d_materialise_unique() this
1906 * guarantees that a directory inode may never have more than one alias.
1908 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1910 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1912 security_d_instantiate(entry, inode);
1913 spin_lock(&inode->i_lock);
1914 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1915 spin_unlock(&inode->i_lock);
1919 __d_instantiate(entry, inode);
1920 spin_unlock(&inode->i_lock);
1924 EXPORT_SYMBOL(d_instantiate_no_diralias);
1926 struct dentry *d_make_root(struct inode *root_inode)
1928 struct dentry *res = NULL;
1931 res = d_alloc_anon(root_inode->i_sb);
1933 res->d_flags |= DCACHE_RCUACCESS;
1934 d_instantiate(res, root_inode);
1941 EXPORT_SYMBOL(d_make_root);
1943 static struct dentry *__d_instantiate_anon(struct dentry *dentry,
1944 struct inode *inode,
1950 security_d_instantiate(dentry, inode);
1951 spin_lock(&inode->i_lock);
1952 res = __d_find_any_alias(inode);
1954 spin_unlock(&inode->i_lock);
1959 /* attach a disconnected dentry */
1960 add_flags = d_flags_for_inode(inode);
1963 add_flags |= DCACHE_DISCONNECTED;
1965 spin_lock(&dentry->d_lock);
1966 __d_set_inode_and_type(dentry, inode, add_flags);
1967 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1968 if (!disconnected) {
1969 hlist_bl_lock(&dentry->d_sb->s_roots);
1970 hlist_bl_add_head(&dentry->d_hash, &dentry->d_sb->s_roots);
1971 hlist_bl_unlock(&dentry->d_sb->s_roots);
1973 spin_unlock(&dentry->d_lock);
1974 spin_unlock(&inode->i_lock);
1983 struct dentry *d_instantiate_anon(struct dentry *dentry, struct inode *inode)
1985 return __d_instantiate_anon(dentry, inode, true);
1987 EXPORT_SYMBOL(d_instantiate_anon);
1989 static struct dentry *__d_obtain_alias(struct inode *inode, bool disconnected)
1995 return ERR_PTR(-ESTALE);
1997 return ERR_CAST(inode);
1999 res = d_find_any_alias(inode);
2003 tmp = d_alloc_anon(inode->i_sb);
2005 res = ERR_PTR(-ENOMEM);
2009 return __d_instantiate_anon(tmp, inode, disconnected);
2017 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2018 * @inode: inode to allocate the dentry for
2020 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2021 * similar open by handle operations. The returned dentry may be anonymous,
2022 * or may have a full name (if the inode was already in the cache).
2024 * When called on a directory inode, we must ensure that the inode only ever
2025 * has one dentry. If a dentry is found, that is returned instead of
2026 * allocating a new one.
2028 * On successful return, the reference to the inode has been transferred
2029 * to the dentry. In case of an error the reference on the inode is released.
2030 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2031 * be passed in and the error will be propagated to the return value,
2032 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2034 struct dentry *d_obtain_alias(struct inode *inode)
2036 return __d_obtain_alias(inode, true);
2038 EXPORT_SYMBOL(d_obtain_alias);
2041 * d_obtain_root - find or allocate a dentry for a given inode
2042 * @inode: inode to allocate the dentry for
2044 * Obtain an IS_ROOT dentry for the root of a filesystem.
2046 * We must ensure that directory inodes only ever have one dentry. If a
2047 * dentry is found, that is returned instead of allocating a new one.
2049 * On successful return, the reference to the inode has been transferred
2050 * to the dentry. In case of an error the reference on the inode is
2051 * released. A %NULL or IS_ERR inode may be passed in and will be the
2052 * error will be propagate to the return value, with a %NULL @inode
2053 * replaced by ERR_PTR(-ESTALE).
2055 struct dentry *d_obtain_root(struct inode *inode)
2057 return __d_obtain_alias(inode, false);
2059 EXPORT_SYMBOL(d_obtain_root);
2062 * d_add_ci - lookup or allocate new dentry with case-exact name
2063 * @inode: the inode case-insensitive lookup has found
2064 * @dentry: the negative dentry that was passed to the parent's lookup func
2065 * @name: the case-exact name to be associated with the returned dentry
2067 * This is to avoid filling the dcache with case-insensitive names to the
2068 * same inode, only the actual correct case is stored in the dcache for
2069 * case-insensitive filesystems.
2071 * For a case-insensitive lookup match and if the the case-exact dentry
2072 * already exists in in the dcache, use it and return it.
2074 * If no entry exists with the exact case name, allocate new dentry with
2075 * the exact case, and return the spliced entry.
2077 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2080 struct dentry *found, *res;
2083 * First check if a dentry matching the name already exists,
2084 * if not go ahead and create it now.
2086 found = d_hash_and_lookup(dentry->d_parent, name);
2091 if (d_in_lookup(dentry)) {
2092 found = d_alloc_parallel(dentry->d_parent, name,
2094 if (IS_ERR(found) || !d_in_lookup(found)) {
2099 found = d_alloc(dentry->d_parent, name);
2102 return ERR_PTR(-ENOMEM);
2105 res = d_splice_alias(inode, found);
2112 EXPORT_SYMBOL(d_add_ci);
2115 static inline bool d_same_name(const struct dentry *dentry,
2116 const struct dentry *parent,
2117 const struct qstr *name)
2119 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2120 if (dentry->d_name.len != name->len)
2122 return dentry_cmp(dentry, name->name, name->len) == 0;
2124 return parent->d_op->d_compare(dentry,
2125 dentry->d_name.len, dentry->d_name.name,
2130 * __d_lookup_rcu - search for a dentry (racy, store-free)
2131 * @parent: parent dentry
2132 * @name: qstr of name we wish to find
2133 * @seqp: returns d_seq value at the point where the dentry was found
2134 * Returns: dentry, or NULL
2136 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2137 * resolution (store-free path walking) design described in
2138 * Documentation/filesystems/path-lookup.txt.
2140 * This is not to be used outside core vfs.
2142 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2143 * held, and rcu_read_lock held. The returned dentry must not be stored into
2144 * without taking d_lock and checking d_seq sequence count against @seq
2147 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2150 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2151 * the returned dentry, so long as its parent's seqlock is checked after the
2152 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2153 * is formed, giving integrity down the path walk.
2155 * NOTE! The caller *has* to check the resulting dentry against the sequence
2156 * number we've returned before using any of the resulting dentry state!
2158 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2159 const struct qstr *name,
2162 u64 hashlen = name->hash_len;
2163 const unsigned char *str = name->name;
2164 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2165 struct hlist_bl_node *node;
2166 struct dentry *dentry;
2169 * Note: There is significant duplication with __d_lookup_rcu which is
2170 * required to prevent single threaded performance regressions
2171 * especially on architectures where smp_rmb (in seqcounts) are costly.
2172 * Keep the two functions in sync.
2176 * The hash list is protected using RCU.
2178 * Carefully use d_seq when comparing a candidate dentry, to avoid
2179 * races with d_move().
2181 * It is possible that concurrent renames can mess up our list
2182 * walk here and result in missing our dentry, resulting in the
2183 * false-negative result. d_lookup() protects against concurrent
2184 * renames using rename_lock seqlock.
2186 * See Documentation/filesystems/path-lookup.txt for more details.
2188 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2193 * The dentry sequence count protects us from concurrent
2194 * renames, and thus protects parent and name fields.
2196 * The caller must perform a seqcount check in order
2197 * to do anything useful with the returned dentry.
2199 * NOTE! We do a "raw" seqcount_begin here. That means that
2200 * we don't wait for the sequence count to stabilize if it
2201 * is in the middle of a sequence change. If we do the slow
2202 * dentry compare, we will do seqretries until it is stable,
2203 * and if we end up with a successful lookup, we actually
2204 * want to exit RCU lookup anyway.
2206 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2207 * we are still guaranteed NUL-termination of ->d_name.name.
2209 seq = raw_seqcount_begin(&dentry->d_seq);
2210 if (dentry->d_parent != parent)
2212 if (d_unhashed(dentry))
2215 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2218 if (dentry->d_name.hash != hashlen_hash(hashlen))
2220 tlen = dentry->d_name.len;
2221 tname = dentry->d_name.name;
2222 /* we want a consistent (name,len) pair */
2223 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2227 if (parent->d_op->d_compare(dentry,
2228 tlen, tname, name) != 0)
2231 if (dentry->d_name.hash_len != hashlen)
2233 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2243 * d_lookup - search for a dentry
2244 * @parent: parent dentry
2245 * @name: qstr of name we wish to find
2246 * Returns: dentry, or NULL
2248 * d_lookup searches the children of the parent dentry for the name in
2249 * question. If the dentry is found its reference count is incremented and the
2250 * dentry is returned. The caller must use dput to free the entry when it has
2251 * finished using it. %NULL is returned if the dentry does not exist.
2253 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2255 struct dentry *dentry;
2259 seq = read_seqbegin(&rename_lock);
2260 dentry = __d_lookup(parent, name);
2263 } while (read_seqretry(&rename_lock, seq));
2266 EXPORT_SYMBOL(d_lookup);
2269 * __d_lookup - search for a dentry (racy)
2270 * @parent: parent dentry
2271 * @name: qstr of name we wish to find
2272 * Returns: dentry, or NULL
2274 * __d_lookup is like d_lookup, however it may (rarely) return a
2275 * false-negative result due to unrelated rename activity.
2277 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2278 * however it must be used carefully, eg. with a following d_lookup in
2279 * the case of failure.
2281 * __d_lookup callers must be commented.
2283 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2285 unsigned int hash = name->hash;
2286 struct hlist_bl_head *b = d_hash(hash);
2287 struct hlist_bl_node *node;
2288 struct dentry *found = NULL;
2289 struct dentry *dentry;
2292 * Note: There is significant duplication with __d_lookup_rcu which is
2293 * required to prevent single threaded performance regressions
2294 * especially on architectures where smp_rmb (in seqcounts) are costly.
2295 * Keep the two functions in sync.
2299 * The hash list is protected using RCU.
2301 * Take d_lock when comparing a candidate dentry, to avoid races
2304 * It is possible that concurrent renames can mess up our list
2305 * walk here and result in missing our dentry, resulting in the
2306 * false-negative result. d_lookup() protects against concurrent
2307 * renames using rename_lock seqlock.
2309 * See Documentation/filesystems/path-lookup.txt for more details.
2313 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2315 if (dentry->d_name.hash != hash)
2318 spin_lock(&dentry->d_lock);
2319 if (dentry->d_parent != parent)
2321 if (d_unhashed(dentry))
2324 if (!d_same_name(dentry, parent, name))
2327 dentry->d_lockref.count++;
2329 spin_unlock(&dentry->d_lock);
2332 spin_unlock(&dentry->d_lock);
2340 * d_hash_and_lookup - hash the qstr then search for a dentry
2341 * @dir: Directory to search in
2342 * @name: qstr of name we wish to find
2344 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2346 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2349 * Check for a fs-specific hash function. Note that we must
2350 * calculate the standard hash first, as the d_op->d_hash()
2351 * routine may choose to leave the hash value unchanged.
2353 name->hash = full_name_hash(dir, name->name, name->len);
2354 if (dir->d_flags & DCACHE_OP_HASH) {
2355 int err = dir->d_op->d_hash(dir, name);
2356 if (unlikely(err < 0))
2357 return ERR_PTR(err);
2359 return d_lookup(dir, name);
2361 EXPORT_SYMBOL(d_hash_and_lookup);
2364 * When a file is deleted, we have two options:
2365 * - turn this dentry into a negative dentry
2366 * - unhash this dentry and free it.
2368 * Usually, we want to just turn this into
2369 * a negative dentry, but if anybody else is
2370 * currently using the dentry or the inode
2371 * we can't do that and we fall back on removing
2372 * it from the hash queues and waiting for
2373 * it to be deleted later when it has no users
2377 * d_delete - delete a dentry
2378 * @dentry: The dentry to delete
2380 * Turn the dentry into a negative dentry if possible, otherwise
2381 * remove it from the hash queues so it can be deleted later
2384 void d_delete(struct dentry * dentry)
2386 struct inode *inode = dentry->d_inode;
2387 int isdir = d_is_dir(dentry);
2389 spin_lock(&inode->i_lock);
2390 spin_lock(&dentry->d_lock);
2392 * Are we the only user?
2394 if (dentry->d_lockref.count == 1) {
2395 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2396 dentry_unlink_inode(dentry);
2399 spin_unlock(&dentry->d_lock);
2400 spin_unlock(&inode->i_lock);
2402 fsnotify_nameremove(dentry, isdir);
2404 EXPORT_SYMBOL(d_delete);
2406 static void __d_rehash(struct dentry *entry)
2408 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2411 hlist_bl_add_head_rcu(&entry->d_hash, b);
2416 * d_rehash - add an entry back to the hash
2417 * @entry: dentry to add to the hash
2419 * Adds a dentry to the hash according to its name.
2422 void d_rehash(struct dentry * entry)
2424 spin_lock(&entry->d_lock);
2426 spin_unlock(&entry->d_lock);
2428 EXPORT_SYMBOL(d_rehash);
2430 static inline unsigned start_dir_add(struct inode *dir)
2434 unsigned n = dir->i_dir_seq;
2435 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2441 static inline void end_dir_add(struct inode *dir, unsigned n)
2443 smp_store_release(&dir->i_dir_seq, n + 2);
2446 static void d_wait_lookup(struct dentry *dentry)
2448 if (d_in_lookup(dentry)) {
2449 DECLARE_WAITQUEUE(wait, current);
2450 add_wait_queue(dentry->d_wait, &wait);
2452 set_current_state(TASK_UNINTERRUPTIBLE);
2453 spin_unlock(&dentry->d_lock);
2455 spin_lock(&dentry->d_lock);
2456 } while (d_in_lookup(dentry));
2460 struct dentry *d_alloc_parallel(struct dentry *parent,
2461 const struct qstr *name,
2462 wait_queue_head_t *wq)
2464 unsigned int hash = name->hash;
2465 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2466 struct hlist_bl_node *node;
2467 struct dentry *new = d_alloc(parent, name);
2468 struct dentry *dentry;
2469 unsigned seq, r_seq, d_seq;
2472 return ERR_PTR(-ENOMEM);
2476 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2477 r_seq = read_seqbegin(&rename_lock);
2478 dentry = __d_lookup_rcu(parent, name, &d_seq);
2479 if (unlikely(dentry)) {
2480 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2484 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2493 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2498 if (unlikely(seq & 1)) {
2504 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2510 * No changes for the parent since the beginning of d_lookup().
2511 * Since all removals from the chain happen with hlist_bl_lock(),
2512 * any potential in-lookup matches are going to stay here until
2513 * we unlock the chain. All fields are stable in everything
2516 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2517 if (dentry->d_name.hash != hash)
2519 if (dentry->d_parent != parent)
2521 if (!d_same_name(dentry, parent, name))
2524 /* now we can try to grab a reference */
2525 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2532 * somebody is likely to be still doing lookup for it;
2533 * wait for them to finish
2535 spin_lock(&dentry->d_lock);
2536 d_wait_lookup(dentry);
2538 * it's not in-lookup anymore; in principle we should repeat
2539 * everything from dcache lookup, but it's likely to be what
2540 * d_lookup() would've found anyway. If it is, just return it;
2541 * otherwise we really have to repeat the whole thing.
2543 if (unlikely(dentry->d_name.hash != hash))
2545 if (unlikely(dentry->d_parent != parent))
2547 if (unlikely(d_unhashed(dentry)))
2549 if (unlikely(!d_same_name(dentry, parent, name)))
2551 /* OK, it *is* a hashed match; return it */
2552 spin_unlock(&dentry->d_lock);
2557 /* we can't take ->d_lock here; it's OK, though. */
2558 new->d_flags |= DCACHE_PAR_LOOKUP;
2560 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2564 spin_unlock(&dentry->d_lock);
2568 EXPORT_SYMBOL(d_alloc_parallel);
2570 void __d_lookup_done(struct dentry *dentry)
2572 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2573 dentry->d_name.hash);
2575 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2576 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2577 wake_up_all(dentry->d_wait);
2578 dentry->d_wait = NULL;
2580 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2581 INIT_LIST_HEAD(&dentry->d_lru);
2583 EXPORT_SYMBOL(__d_lookup_done);
2585 /* inode->i_lock held if inode is non-NULL */
2587 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2589 struct inode *dir = NULL;
2591 spin_lock(&dentry->d_lock);
2592 if (unlikely(d_in_lookup(dentry))) {
2593 dir = dentry->d_parent->d_inode;
2594 n = start_dir_add(dir);
2595 __d_lookup_done(dentry);
2598 unsigned add_flags = d_flags_for_inode(inode);
2599 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2600 raw_write_seqcount_begin(&dentry->d_seq);
2601 __d_set_inode_and_type(dentry, inode, add_flags);
2602 raw_write_seqcount_end(&dentry->d_seq);
2603 fsnotify_update_flags(dentry);
2607 end_dir_add(dir, n);
2608 spin_unlock(&dentry->d_lock);
2610 spin_unlock(&inode->i_lock);
2614 * d_add - add dentry to hash queues
2615 * @entry: dentry to add
2616 * @inode: The inode to attach to this dentry
2618 * This adds the entry to the hash queues and initializes @inode.
2619 * The entry was actually filled in earlier during d_alloc().
2622 void d_add(struct dentry *entry, struct inode *inode)
2625 security_d_instantiate(entry, inode);
2626 spin_lock(&inode->i_lock);
2628 __d_add(entry, inode);
2630 EXPORT_SYMBOL(d_add);
2633 * d_exact_alias - find and hash an exact unhashed alias
2634 * @entry: dentry to add
2635 * @inode: The inode to go with this dentry
2637 * If an unhashed dentry with the same name/parent and desired
2638 * inode already exists, hash and return it. Otherwise, return
2641 * Parent directory should be locked.
2643 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2645 struct dentry *alias;
2646 unsigned int hash = entry->d_name.hash;
2648 spin_lock(&inode->i_lock);
2649 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2651 * Don't need alias->d_lock here, because aliases with
2652 * d_parent == entry->d_parent are not subject to name or
2653 * parent changes, because the parent inode i_mutex is held.
2655 if (alias->d_name.hash != hash)
2657 if (alias->d_parent != entry->d_parent)
2659 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2661 spin_lock(&alias->d_lock);
2662 if (!d_unhashed(alias)) {
2663 spin_unlock(&alias->d_lock);
2666 __dget_dlock(alias);
2668 spin_unlock(&alias->d_lock);
2670 spin_unlock(&inode->i_lock);
2673 spin_unlock(&inode->i_lock);
2676 EXPORT_SYMBOL(d_exact_alias);
2679 * dentry_update_name_case - update case insensitive dentry with a new name
2680 * @dentry: dentry to be updated
2683 * Update a case insensitive dentry with new case of name.
2685 * dentry must have been returned by d_lookup with name @name. Old and new
2686 * name lengths must match (ie. no d_compare which allows mismatched name
2689 * Parent inode i_mutex must be held over d_lookup and into this call (to
2690 * keep renames and concurrent inserts, and readdir(2) away).
2692 void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
2694 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
2695 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2697 spin_lock(&dentry->d_lock);
2698 write_seqcount_begin(&dentry->d_seq);
2699 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2700 write_seqcount_end(&dentry->d_seq);
2701 spin_unlock(&dentry->d_lock);
2703 EXPORT_SYMBOL(dentry_update_name_case);
2705 static void swap_names(struct dentry *dentry, struct dentry *target)
2707 if (unlikely(dname_external(target))) {
2708 if (unlikely(dname_external(dentry))) {
2710 * Both external: swap the pointers
2712 swap(target->d_name.name, dentry->d_name.name);
2715 * dentry:internal, target:external. Steal target's
2716 * storage and make target internal.
2718 memcpy(target->d_iname, dentry->d_name.name,
2719 dentry->d_name.len + 1);
2720 dentry->d_name.name = target->d_name.name;
2721 target->d_name.name = target->d_iname;
2724 if (unlikely(dname_external(dentry))) {
2726 * dentry:external, target:internal. Give dentry's
2727 * storage to target and make dentry internal
2729 memcpy(dentry->d_iname, target->d_name.name,
2730 target->d_name.len + 1);
2731 target->d_name.name = dentry->d_name.name;
2732 dentry->d_name.name = dentry->d_iname;
2735 * Both are internal.
2738 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2739 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2740 swap(((long *) &dentry->d_iname)[i],
2741 ((long *) &target->d_iname)[i]);
2745 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2748 static void copy_name(struct dentry *dentry, struct dentry *target)
2750 struct external_name *old_name = NULL;
2751 if (unlikely(dname_external(dentry)))
2752 old_name = external_name(dentry);
2753 if (unlikely(dname_external(target))) {
2754 atomic_inc(&external_name(target)->u.count);
2755 dentry->d_name = target->d_name;
2757 memcpy(dentry->d_iname, target->d_name.name,
2758 target->d_name.len + 1);
2759 dentry->d_name.name = dentry->d_iname;
2760 dentry->d_name.hash_len = target->d_name.hash_len;
2762 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2763 call_rcu(&old_name->u.head, __d_free_external_name);
2767 * __d_move - move a dentry
2768 * @dentry: entry to move
2769 * @target: new dentry
2770 * @exchange: exchange the two dentries
2772 * Update the dcache to reflect the move of a file name. Negative
2773 * dcache entries should not be moved in this way. Caller must hold
2774 * rename_lock, the i_mutex of the source and target directories,
2775 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2777 static void __d_move(struct dentry *dentry, struct dentry *target,
2780 struct dentry *old_parent, *p;
2781 struct inode *dir = NULL;
2784 WARN_ON(!dentry->d_inode);
2785 if (WARN_ON(dentry == target))
2788 BUG_ON(d_ancestor(target, dentry));
2789 old_parent = dentry->d_parent;
2790 p = d_ancestor(old_parent, target);
2791 if (IS_ROOT(dentry)) {
2793 spin_lock(&target->d_parent->d_lock);
2795 /* target is not a descendent of dentry->d_parent */
2796 spin_lock(&target->d_parent->d_lock);
2797 spin_lock_nested(&old_parent->d_lock, DENTRY_D_LOCK_NESTED);
2799 BUG_ON(p == dentry);
2800 spin_lock(&old_parent->d_lock);
2802 spin_lock_nested(&target->d_parent->d_lock,
2803 DENTRY_D_LOCK_NESTED);
2805 spin_lock_nested(&dentry->d_lock, 2);
2806 spin_lock_nested(&target->d_lock, 3);
2808 if (unlikely(d_in_lookup(target))) {
2809 dir = target->d_parent->d_inode;
2810 n = start_dir_add(dir);
2811 __d_lookup_done(target);
2814 write_seqcount_begin(&dentry->d_seq);
2815 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2818 if (!d_unhashed(dentry))
2820 if (!d_unhashed(target))
2823 /* ... and switch them in the tree */
2824 dentry->d_parent = target->d_parent;
2826 copy_name(dentry, target);
2827 target->d_hash.pprev = NULL;
2828 dentry->d_parent->d_lockref.count++;
2829 if (dentry == old_parent)
2830 dentry->d_flags |= DCACHE_RCUACCESS;
2832 WARN_ON(!--old_parent->d_lockref.count);
2834 target->d_parent = old_parent;
2835 swap_names(dentry, target);
2836 list_move(&target->d_child, &target->d_parent->d_subdirs);
2838 fsnotify_update_flags(target);
2840 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2842 fsnotify_update_flags(dentry);
2844 write_seqcount_end(&target->d_seq);
2845 write_seqcount_end(&dentry->d_seq);
2848 end_dir_add(dir, n);
2850 if (dentry->d_parent != old_parent)
2851 spin_unlock(&dentry->d_parent->d_lock);
2852 if (dentry != old_parent)
2853 spin_unlock(&old_parent->d_lock);
2854 spin_unlock(&target->d_lock);
2855 spin_unlock(&dentry->d_lock);
2859 * d_move - move a dentry
2860 * @dentry: entry to move
2861 * @target: new dentry
2863 * Update the dcache to reflect the move of a file name. Negative
2864 * dcache entries should not be moved in this way. See the locking
2865 * requirements for __d_move.
2867 void d_move(struct dentry *dentry, struct dentry *target)
2869 write_seqlock(&rename_lock);
2870 __d_move(dentry, target, false);
2871 write_sequnlock(&rename_lock);
2873 EXPORT_SYMBOL(d_move);
2876 * d_exchange - exchange two dentries
2877 * @dentry1: first dentry
2878 * @dentry2: second dentry
2880 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2882 write_seqlock(&rename_lock);
2884 WARN_ON(!dentry1->d_inode);
2885 WARN_ON(!dentry2->d_inode);
2886 WARN_ON(IS_ROOT(dentry1));
2887 WARN_ON(IS_ROOT(dentry2));
2889 __d_move(dentry1, dentry2, true);
2891 write_sequnlock(&rename_lock);
2895 * d_ancestor - search for an ancestor
2896 * @p1: ancestor dentry
2899 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2900 * an ancestor of p2, else NULL.
2902 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2906 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2907 if (p->d_parent == p1)
2914 * This helper attempts to cope with remotely renamed directories
2916 * It assumes that the caller is already holding
2917 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2919 * Note: If ever the locking in lock_rename() changes, then please
2920 * remember to update this too...
2922 static int __d_unalias(struct inode *inode,
2923 struct dentry *dentry, struct dentry *alias)
2925 struct mutex *m1 = NULL;
2926 struct rw_semaphore *m2 = NULL;
2929 /* If alias and dentry share a parent, then no extra locks required */
2930 if (alias->d_parent == dentry->d_parent)
2933 /* See lock_rename() */
2934 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2936 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2937 if (!inode_trylock_shared(alias->d_parent->d_inode))
2939 m2 = &alias->d_parent->d_inode->i_rwsem;
2941 __d_move(alias, dentry, false);
2952 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2953 * @inode: the inode which may have a disconnected dentry
2954 * @dentry: a negative dentry which we want to point to the inode.
2956 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2957 * place of the given dentry and return it, else simply d_add the inode
2958 * to the dentry and return NULL.
2960 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2961 * we should error out: directories can't have multiple aliases.
2963 * This is needed in the lookup routine of any filesystem that is exportable
2964 * (via knfsd) so that we can build dcache paths to directories effectively.
2966 * If a dentry was found and moved, then it is returned. Otherwise NULL
2967 * is returned. This matches the expected return value of ->lookup.
2969 * Cluster filesystems may call this function with a negative, hashed dentry.
2970 * In that case, we know that the inode will be a regular file, and also this
2971 * will only occur during atomic_open. So we need to check for the dentry
2972 * being already hashed only in the final case.
2974 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2977 return ERR_CAST(inode);
2979 BUG_ON(!d_unhashed(dentry));
2984 security_d_instantiate(dentry, inode);
2985 spin_lock(&inode->i_lock);
2986 if (S_ISDIR(inode->i_mode)) {
2987 struct dentry *new = __d_find_any_alias(inode);
2988 if (unlikely(new)) {
2989 /* The reference to new ensures it remains an alias */
2990 spin_unlock(&inode->i_lock);
2991 write_seqlock(&rename_lock);
2992 if (unlikely(d_ancestor(new, dentry))) {
2993 write_sequnlock(&rename_lock);
2995 new = ERR_PTR(-ELOOP);
2996 pr_warn_ratelimited(
2997 "VFS: Lookup of '%s' in %s %s"
2998 " would have caused loop\n",
2999 dentry->d_name.name,
3000 inode->i_sb->s_type->name,
3002 } else if (!IS_ROOT(new)) {
3003 struct dentry *old_parent = dget(new->d_parent);
3004 int err = __d_unalias(inode, dentry, new);
3005 write_sequnlock(&rename_lock);
3012 __d_move(new, dentry, false);
3013 write_sequnlock(&rename_lock);
3020 __d_add(dentry, inode);
3023 EXPORT_SYMBOL(d_splice_alias);
3026 * Test whether new_dentry is a subdirectory of old_dentry.
3028 * Trivially implemented using the dcache structure
3032 * is_subdir - is new dentry a subdirectory of old_dentry
3033 * @new_dentry: new dentry
3034 * @old_dentry: old dentry
3036 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3037 * Returns false otherwise.
3038 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3041 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3046 if (new_dentry == old_dentry)
3050 /* for restarting inner loop in case of seq retry */
3051 seq = read_seqbegin(&rename_lock);
3053 * Need rcu_readlock to protect against the d_parent trashing
3057 if (d_ancestor(old_dentry, new_dentry))
3062 } while (read_seqretry(&rename_lock, seq));
3066 EXPORT_SYMBOL(is_subdir);
3068 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3070 struct dentry *root = data;
3071 if (dentry != root) {
3072 if (d_unhashed(dentry) || !dentry->d_inode)
3075 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3076 dentry->d_flags |= DCACHE_GENOCIDE;
3077 dentry->d_lockref.count--;
3080 return D_WALK_CONTINUE;
3083 void d_genocide(struct dentry *parent)
3085 d_walk(parent, parent, d_genocide_kill);
3088 EXPORT_SYMBOL(d_genocide);
3090 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3092 inode_dec_link_count(inode);
3093 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3094 !hlist_unhashed(&dentry->d_u.d_alias) ||
3095 !d_unlinked(dentry));
3096 spin_lock(&dentry->d_parent->d_lock);
3097 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3098 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3099 (unsigned long long)inode->i_ino);
3100 spin_unlock(&dentry->d_lock);
3101 spin_unlock(&dentry->d_parent->d_lock);
3102 d_instantiate(dentry, inode);
3104 EXPORT_SYMBOL(d_tmpfile);
3106 static __initdata unsigned long dhash_entries;
3107 static int __init set_dhash_entries(char *str)
3111 dhash_entries = simple_strtoul(str, &str, 0);
3114 __setup("dhash_entries=", set_dhash_entries);
3116 static void __init dcache_init_early(void)
3118 /* If hashes are distributed across NUMA nodes, defer
3119 * hash allocation until vmalloc space is available.
3125 alloc_large_system_hash("Dentry cache",
3126 sizeof(struct hlist_bl_head),
3129 HASH_EARLY | HASH_ZERO,
3134 d_hash_shift = 32 - d_hash_shift;
3137 static void __init dcache_init(void)
3140 * A constructor could be added for stable state like the lists,
3141 * but it is probably not worth it because of the cache nature
3144 dentry_cache = KMEM_CACHE_USERCOPY(dentry,
3145 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT,
3148 /* Hash may have been set up in dcache_init_early */
3153 alloc_large_system_hash("Dentry cache",
3154 sizeof(struct hlist_bl_head),
3162 d_hash_shift = 32 - d_hash_shift;
3165 /* SLAB cache for __getname() consumers */
3166 struct kmem_cache *names_cachep __read_mostly;
3167 EXPORT_SYMBOL(names_cachep);
3169 void __init vfs_caches_init_early(void)
3173 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3174 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3176 dcache_init_early();
3180 void __init vfs_caches_init(void)
3182 names_cachep = kmem_cache_create_usercopy("names_cache", PATH_MAX, 0,
3183 SLAB_HWCACHE_ALIGN|SLAB_PANIC, 0, PATH_MAX, NULL);
3188 files_maxfiles_init();