4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <linux/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/bit_spinlock.h>
36 #include <linux/rculist_bl.h>
37 #include <linux/prefetch.h>
38 #include <linux/ratelimit.h>
39 #include <linux/list_lru.h>
40 #include <linux/kasan.h>
47 * dcache->d_inode->i_lock protects:
48 * - i_dentry, d_u.d_alias, d_inode of aliases
49 * dcache_hash_bucket lock protects:
50 * - the dcache hash table
51 * s_roots bl list spinlock protects:
52 * - the s_roots list (see __d_drop)
53 * dentry->d_sb->s_dentry_lru_lock protects:
54 * - the dcache lru lists and counters
61 * - d_parent and d_subdirs
62 * - childrens' d_child and d_parent
63 * - d_u.d_alias, d_inode
66 * dentry->d_inode->i_lock
68 * dentry->d_sb->s_dentry_lru_lock
69 * dcache_hash_bucket lock
72 * If there is an ancestor relationship:
73 * dentry->d_parent->...->d_parent->d_lock
75 * dentry->d_parent->d_lock
78 * If no ancestor relationship:
79 * if (dentry1 < dentry2)
83 int sysctl_vfs_cache_pressure __read_mostly = 100;
84 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
88 EXPORT_SYMBOL(rename_lock);
90 static struct kmem_cache *dentry_cache __read_mostly;
92 const struct qstr empty_name = QSTR_INIT("", 0);
93 EXPORT_SYMBOL(empty_name);
94 const struct qstr slash_name = QSTR_INIT("/", 1);
95 EXPORT_SYMBOL(slash_name);
98 * This is the single most critical data structure when it comes
99 * to the dcache: the hashtable for lookups. Somebody should try
100 * to make this good - I've just made it work.
102 * This hash-function tries to avoid losing too many bits of hash
103 * information, yet avoid using a prime hash-size or similar.
106 static unsigned int d_hash_shift __read_mostly;
108 static struct hlist_bl_head *dentry_hashtable __read_mostly;
110 static inline struct hlist_bl_head *d_hash(unsigned int hash)
112 return dentry_hashtable + (hash >> d_hash_shift);
115 #define IN_LOOKUP_SHIFT 10
116 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
118 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
121 hash += (unsigned long) parent / L1_CACHE_BYTES;
122 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
126 /* Statistics gathering. */
127 struct dentry_stat_t dentry_stat = {
131 static DEFINE_PER_CPU(long, nr_dentry);
132 static DEFINE_PER_CPU(long, nr_dentry_unused);
134 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
137 * Here we resort to our own counters instead of using generic per-cpu counters
138 * for consistency with what the vfs inode code does. We are expected to harvest
139 * better code and performance by having our own specialized counters.
141 * Please note that the loop is done over all possible CPUs, not over all online
142 * CPUs. The reason for this is that we don't want to play games with CPUs going
143 * on and off. If one of them goes off, we will just keep their counters.
145 * glommer: See cffbc8a for details, and if you ever intend to change this,
146 * please update all vfs counters to match.
148 static long get_nr_dentry(void)
152 for_each_possible_cpu(i)
153 sum += per_cpu(nr_dentry, i);
154 return sum < 0 ? 0 : sum;
157 static long get_nr_dentry_unused(void)
161 for_each_possible_cpu(i)
162 sum += per_cpu(nr_dentry_unused, i);
163 return sum < 0 ? 0 : sum;
166 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
167 size_t *lenp, loff_t *ppos)
169 dentry_stat.nr_dentry = get_nr_dentry();
170 dentry_stat.nr_unused = get_nr_dentry_unused();
171 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
176 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
177 * The strings are both count bytes long, and count is non-zero.
179 #ifdef CONFIG_DCACHE_WORD_ACCESS
181 #include <asm/word-at-a-time.h>
183 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
184 * aligned allocation for this particular component. We don't
185 * strictly need the load_unaligned_zeropad() safety, but it
186 * doesn't hurt either.
188 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
189 * need the careful unaligned handling.
191 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
193 unsigned long a,b,mask;
196 a = *(unsigned long *)cs;
197 b = load_unaligned_zeropad(ct);
198 if (tcount < sizeof(unsigned long))
200 if (unlikely(a != b))
202 cs += sizeof(unsigned long);
203 ct += sizeof(unsigned long);
204 tcount -= sizeof(unsigned long);
208 mask = bytemask_from_count(tcount);
209 return unlikely(!!((a ^ b) & mask));
214 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
228 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
231 * Be careful about RCU walk racing with rename:
232 * use 'READ_ONCE' to fetch the name pointer.
234 * NOTE! Even if a rename will mean that the length
235 * was not loaded atomically, we don't care. The
236 * RCU walk will check the sequence count eventually,
237 * and catch it. And we won't overrun the buffer,
238 * because we're reading the name pointer atomically,
239 * and a dentry name is guaranteed to be properly
240 * terminated with a NUL byte.
242 * End result: even if 'len' is wrong, we'll exit
243 * early because the data cannot match (there can
244 * be no NUL in the ct/tcount data)
246 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
248 return dentry_string_cmp(cs, ct, tcount);
251 struct external_name {
254 struct rcu_head head;
256 unsigned char name[];
259 static inline struct external_name *external_name(struct dentry *dentry)
261 return container_of(dentry->d_name.name, struct external_name, name[0]);
264 static void __d_free(struct rcu_head *head)
266 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
268 kmem_cache_free(dentry_cache, dentry);
271 static void __d_free_external(struct rcu_head *head)
273 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
274 kfree(external_name(dentry));
275 kmem_cache_free(dentry_cache, dentry);
278 static inline int dname_external(const struct dentry *dentry)
280 return dentry->d_name.name != dentry->d_iname;
283 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
285 spin_lock(&dentry->d_lock);
286 if (unlikely(dname_external(dentry))) {
287 struct external_name *p = external_name(dentry);
288 atomic_inc(&p->u.count);
289 spin_unlock(&dentry->d_lock);
290 name->name = p->name;
292 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
293 spin_unlock(&dentry->d_lock);
294 name->name = name->inline_name;
297 EXPORT_SYMBOL(take_dentry_name_snapshot);
299 void release_dentry_name_snapshot(struct name_snapshot *name)
301 if (unlikely(name->name != name->inline_name)) {
302 struct external_name *p;
303 p = container_of(name->name, struct external_name, name[0]);
304 if (unlikely(atomic_dec_and_test(&p->u.count)))
305 kfree_rcu(p, u.head);
308 EXPORT_SYMBOL(release_dentry_name_snapshot);
310 static inline void __d_set_inode_and_type(struct dentry *dentry,
316 dentry->d_inode = inode;
317 flags = READ_ONCE(dentry->d_flags);
318 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
320 WRITE_ONCE(dentry->d_flags, flags);
323 static inline void __d_clear_type_and_inode(struct dentry *dentry)
325 unsigned flags = READ_ONCE(dentry->d_flags);
327 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
328 WRITE_ONCE(dentry->d_flags, flags);
329 dentry->d_inode = NULL;
332 static void dentry_free(struct dentry *dentry)
334 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
335 if (unlikely(dname_external(dentry))) {
336 struct external_name *p = external_name(dentry);
337 if (likely(atomic_dec_and_test(&p->u.count))) {
338 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
342 /* if dentry was never visible to RCU, immediate free is OK */
343 if (!(dentry->d_flags & DCACHE_RCUACCESS))
344 __d_free(&dentry->d_u.d_rcu);
346 call_rcu(&dentry->d_u.d_rcu, __d_free);
350 * Release the dentry's inode, using the filesystem
351 * d_iput() operation if defined.
353 static void dentry_unlink_inode(struct dentry * dentry)
354 __releases(dentry->d_lock)
355 __releases(dentry->d_inode->i_lock)
357 struct inode *inode = dentry->d_inode;
358 bool hashed = !d_unhashed(dentry);
361 raw_write_seqcount_begin(&dentry->d_seq);
362 __d_clear_type_and_inode(dentry);
363 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 * dentry_lru_(add|del)_list) must be called with d_lock held.
448 static void dentry_lru_add(struct dentry *dentry)
450 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
452 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
453 dentry->d_flags |= DCACHE_REFERENCED;
457 * d_drop - drop a dentry
458 * @dentry: dentry to drop
460 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
461 * be found through a VFS lookup any more. Note that this is different from
462 * deleting the dentry - d_delete will try to mark the dentry negative if
463 * possible, giving a successful _negative_ lookup, while d_drop will
464 * just make the cache lookup fail.
466 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
467 * reason (NFS timeouts or autofs deletes).
469 * __d_drop requires dentry->d_lock
470 * ___d_drop doesn't mark dentry as "unhashed"
471 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
473 static void ___d_drop(struct dentry *dentry)
475 if (!d_unhashed(dentry)) {
476 struct hlist_bl_head *b;
478 * Hashed dentries are normally on the dentry hashtable,
479 * with the exception of those newly allocated by
480 * d_obtain_root, which are always IS_ROOT:
482 if (unlikely(IS_ROOT(dentry)))
483 b = &dentry->d_sb->s_roots;
485 b = d_hash(dentry->d_name.hash);
488 __hlist_bl_del(&dentry->d_hash);
490 /* After this call, in-progress rcu-walk path lookup will fail. */
491 write_seqcount_invalidate(&dentry->d_seq);
495 void __d_drop(struct dentry *dentry)
498 dentry->d_hash.pprev = NULL;
500 EXPORT_SYMBOL(__d_drop);
502 void d_drop(struct dentry *dentry)
504 spin_lock(&dentry->d_lock);
506 spin_unlock(&dentry->d_lock);
508 EXPORT_SYMBOL(d_drop);
510 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
514 * Inform d_walk() and shrink_dentry_list() that we are no longer
515 * attached to the dentry tree
517 dentry->d_flags |= DCACHE_DENTRY_KILLED;
518 if (unlikely(list_empty(&dentry->d_child)))
520 __list_del_entry(&dentry->d_child);
522 * Cursors can move around the list of children. While we'd been
523 * a normal list member, it didn't matter - ->d_child.next would've
524 * been updated. However, from now on it won't be and for the
525 * things like d_walk() it might end up with a nasty surprise.
526 * Normally d_walk() doesn't care about cursors moving around -
527 * ->d_lock on parent prevents that and since a cursor has no children
528 * of its own, we get through it without ever unlocking the parent.
529 * There is one exception, though - if we ascend from a child that
530 * gets killed as soon as we unlock it, the next sibling is found
531 * using the value left in its ->d_child.next. And if _that_
532 * pointed to a cursor, and cursor got moved (e.g. by lseek())
533 * before d_walk() regains parent->d_lock, we'll end up skipping
534 * everything the cursor had been moved past.
536 * Solution: make sure that the pointer left behind in ->d_child.next
537 * points to something that won't be moving around. I.e. skip the
540 while (dentry->d_child.next != &parent->d_subdirs) {
541 next = list_entry(dentry->d_child.next, struct dentry, d_child);
542 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
544 dentry->d_child.next = next->d_child.next;
548 static void __dentry_kill(struct dentry *dentry)
550 struct dentry *parent = NULL;
551 bool can_free = true;
552 if (!IS_ROOT(dentry))
553 parent = dentry->d_parent;
556 * The dentry is now unrecoverably dead to the world.
558 lockref_mark_dead(&dentry->d_lockref);
561 * inform the fs via d_prune that this dentry is about to be
562 * unhashed and destroyed.
564 if (dentry->d_flags & DCACHE_OP_PRUNE)
565 dentry->d_op->d_prune(dentry);
567 if (dentry->d_flags & DCACHE_LRU_LIST) {
568 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
571 /* if it was on the hash then remove it */
573 dentry_unlist(dentry, parent);
575 spin_unlock(&parent->d_lock);
577 dentry_unlink_inode(dentry);
579 spin_unlock(&dentry->d_lock);
580 this_cpu_dec(nr_dentry);
581 if (dentry->d_op && dentry->d_op->d_release)
582 dentry->d_op->d_release(dentry);
584 spin_lock(&dentry->d_lock);
585 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
586 dentry->d_flags |= DCACHE_MAY_FREE;
589 spin_unlock(&dentry->d_lock);
590 if (likely(can_free))
595 * Finish off a dentry we've decided to kill.
596 * dentry->d_lock must be held, returns with it unlocked.
597 * If ref is non-zero, then decrement the refcount too.
598 * Returns dentry requiring refcount drop, or NULL if we're done.
600 static struct dentry *dentry_kill(struct dentry *dentry)
601 __releases(dentry->d_lock)
603 struct inode *inode = dentry->d_inode;
604 struct dentry *parent = NULL;
606 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
609 if (!IS_ROOT(dentry)) {
610 parent = dentry->d_parent;
611 if (unlikely(!spin_trylock(&parent->d_lock))) {
613 spin_unlock(&inode->i_lock);
618 __dentry_kill(dentry);
622 spin_unlock(&dentry->d_lock);
623 return dentry; /* try again with same dentry */
626 static inline struct dentry *lock_parent(struct dentry *dentry)
628 struct dentry *parent = dentry->d_parent;
631 if (unlikely(dentry->d_lockref.count < 0))
633 if (likely(spin_trylock(&parent->d_lock)))
636 spin_unlock(&dentry->d_lock);
638 parent = READ_ONCE(dentry->d_parent);
639 spin_lock(&parent->d_lock);
641 * We can't blindly lock dentry until we are sure
642 * that we won't violate the locking order.
643 * Any changes of dentry->d_parent must have
644 * been done with parent->d_lock held, so
645 * spin_lock() above is enough of a barrier
646 * for checking if it's still our child.
648 if (unlikely(parent != dentry->d_parent)) {
649 spin_unlock(&parent->d_lock);
653 if (parent != dentry)
654 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
661 * Try to do a lockless dput(), and return whether that was successful.
663 * If unsuccessful, we return false, having already taken the dentry lock.
665 * The caller needs to hold the RCU read lock, so that the dentry is
666 * guaranteed to stay around even if the refcount goes down to zero!
668 static inline bool fast_dput(struct dentry *dentry)
671 unsigned int d_flags;
674 * If we have a d_op->d_delete() operation, we sould not
675 * let the dentry count go to zero, so use "put_or_lock".
677 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
678 return lockref_put_or_lock(&dentry->d_lockref);
681 * .. otherwise, we can try to just decrement the
682 * lockref optimistically.
684 ret = lockref_put_return(&dentry->d_lockref);
687 * If the lockref_put_return() failed due to the lock being held
688 * by somebody else, the fast path has failed. We will need to
689 * get the lock, and then check the count again.
691 if (unlikely(ret < 0)) {
692 spin_lock(&dentry->d_lock);
693 if (dentry->d_lockref.count > 1) {
694 dentry->d_lockref.count--;
695 spin_unlock(&dentry->d_lock);
702 * If we weren't the last ref, we're done.
708 * Careful, careful. The reference count went down
709 * to zero, but we don't hold the dentry lock, so
710 * somebody else could get it again, and do another
711 * dput(), and we need to not race with that.
713 * However, there is a very special and common case
714 * where we don't care, because there is nothing to
715 * do: the dentry is still hashed, it does not have
716 * a 'delete' op, and it's referenced and already on
719 * NOTE! Since we aren't locked, these values are
720 * not "stable". However, it is sufficient that at
721 * some point after we dropped the reference the
722 * dentry was hashed and the flags had the proper
723 * value. Other dentry users may have re-gotten
724 * a reference to the dentry and change that, but
725 * our work is done - we can leave the dentry
726 * around with a zero refcount.
729 d_flags = READ_ONCE(dentry->d_flags);
730 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
732 /* Nothing to do? Dropping the reference was all we needed? */
733 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
737 * Not the fast normal case? Get the lock. We've already decremented
738 * the refcount, but we'll need to re-check the situation after
741 spin_lock(&dentry->d_lock);
744 * Did somebody else grab a reference to it in the meantime, and
745 * we're no longer the last user after all? Alternatively, somebody
746 * else could have killed it and marked it dead. Either way, we
747 * don't need to do anything else.
749 if (dentry->d_lockref.count) {
750 spin_unlock(&dentry->d_lock);
755 * Re-get the reference we optimistically dropped. We hold the
756 * lock, and we just tested that it was zero, so we can just
759 dentry->d_lockref.count = 1;
767 * This is complicated by the fact that we do not want to put
768 * dentries that are no longer on any hash chain on the unused
769 * list: we'd much rather just get rid of them immediately.
771 * However, that implies that we have to traverse the dentry
772 * tree upwards to the parents which might _also_ now be
773 * scheduled for deletion (it may have been only waiting for
774 * its last child to go away).
776 * This tail recursion is done by hand as we don't want to depend
777 * on the compiler to always get this right (gcc generally doesn't).
778 * Real recursion would eat up our stack space.
782 * dput - release a dentry
783 * @dentry: dentry to release
785 * Release a dentry. This will drop the usage count and if appropriate
786 * call the dentry unlink method as well as removing it from the queues and
787 * releasing its resources. If the parent dentries were scheduled for release
788 * they too may now get deleted.
790 void dput(struct dentry *dentry)
792 if (unlikely(!dentry))
799 if (likely(fast_dput(dentry))) {
804 /* Slow case: now with the dentry lock held */
807 WARN_ON(d_in_lookup(dentry));
809 /* Unreachable? Get rid of it */
810 if (unlikely(d_unhashed(dentry)))
813 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
816 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
817 if (dentry->d_op->d_delete(dentry))
821 dentry_lru_add(dentry);
823 dentry->d_lockref.count--;
824 spin_unlock(&dentry->d_lock);
828 dentry = dentry_kill(dentry);
837 /* This must be called with d_lock held */
838 static inline void __dget_dlock(struct dentry *dentry)
840 dentry->d_lockref.count++;
843 static inline void __dget(struct dentry *dentry)
845 lockref_get(&dentry->d_lockref);
848 struct dentry *dget_parent(struct dentry *dentry)
854 * Do optimistic parent lookup without any
858 ret = READ_ONCE(dentry->d_parent);
859 gotref = lockref_get_not_zero(&ret->d_lockref);
861 if (likely(gotref)) {
862 if (likely(ret == READ_ONCE(dentry->d_parent)))
869 * Don't need rcu_dereference because we re-check it was correct under
873 ret = dentry->d_parent;
874 spin_lock(&ret->d_lock);
875 if (unlikely(ret != dentry->d_parent)) {
876 spin_unlock(&ret->d_lock);
881 BUG_ON(!ret->d_lockref.count);
882 ret->d_lockref.count++;
883 spin_unlock(&ret->d_lock);
886 EXPORT_SYMBOL(dget_parent);
889 * d_find_alias - grab a hashed alias of inode
890 * @inode: inode in question
892 * If inode has a hashed alias, or is a directory and has any alias,
893 * acquire the reference to alias and return it. Otherwise return NULL.
894 * Notice that if inode is a directory there can be only one alias and
895 * it can be unhashed only if it has no children, or if it is the root
896 * of a filesystem, or if the directory was renamed and d_revalidate
897 * was the first vfs operation to notice.
899 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
900 * any other hashed alias over that one.
902 static struct dentry *__d_find_alias(struct inode *inode)
904 struct dentry *alias, *discon_alias;
908 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
909 spin_lock(&alias->d_lock);
910 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
911 if (IS_ROOT(alias) &&
912 (alias->d_flags & DCACHE_DISCONNECTED)) {
913 discon_alias = alias;
916 spin_unlock(&alias->d_lock);
920 spin_unlock(&alias->d_lock);
923 alias = discon_alias;
924 spin_lock(&alias->d_lock);
925 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
927 spin_unlock(&alias->d_lock);
930 spin_unlock(&alias->d_lock);
936 struct dentry *d_find_alias(struct inode *inode)
938 struct dentry *de = NULL;
940 if (!hlist_empty(&inode->i_dentry)) {
941 spin_lock(&inode->i_lock);
942 de = __d_find_alias(inode);
943 spin_unlock(&inode->i_lock);
947 EXPORT_SYMBOL(d_find_alias);
950 * Try to kill dentries associated with this inode.
951 * WARNING: you must own a reference to inode.
953 void d_prune_aliases(struct inode *inode)
955 struct dentry *dentry;
957 spin_lock(&inode->i_lock);
958 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
959 spin_lock(&dentry->d_lock);
960 if (!dentry->d_lockref.count) {
961 struct dentry *parent = lock_parent(dentry);
962 if (likely(!dentry->d_lockref.count)) {
963 __dentry_kill(dentry);
968 spin_unlock(&parent->d_lock);
970 spin_unlock(&dentry->d_lock);
972 spin_unlock(&inode->i_lock);
974 EXPORT_SYMBOL(d_prune_aliases);
976 static void shrink_dentry_list(struct list_head *list)
978 struct dentry *dentry, *parent;
980 while (!list_empty(list)) {
982 dentry = list_entry(list->prev, struct dentry, d_lru);
983 spin_lock(&dentry->d_lock);
984 parent = lock_parent(dentry);
987 * The dispose list is isolated and dentries are not accounted
988 * to the LRU here, so we can simply remove it from the list
989 * here regardless of whether it is referenced or not.
991 d_shrink_del(dentry);
994 * We found an inuse dentry which was not removed from
995 * the LRU because of laziness during lookup. Do not free it.
997 if (dentry->d_lockref.count > 0) {
998 spin_unlock(&dentry->d_lock);
1000 spin_unlock(&parent->d_lock);
1005 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1006 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1007 spin_unlock(&dentry->d_lock);
1009 spin_unlock(&parent->d_lock);
1011 dentry_free(dentry);
1015 inode = dentry->d_inode;
1016 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1017 d_shrink_add(dentry, list);
1018 spin_unlock(&dentry->d_lock);
1020 spin_unlock(&parent->d_lock);
1024 __dentry_kill(dentry);
1027 * We need to prune ancestors too. This is necessary to prevent
1028 * quadratic behavior of shrink_dcache_parent(), but is also
1029 * expected to be beneficial in reducing dentry cache
1033 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1034 parent = lock_parent(dentry);
1035 if (dentry->d_lockref.count != 1) {
1036 dentry->d_lockref.count--;
1037 spin_unlock(&dentry->d_lock);
1039 spin_unlock(&parent->d_lock);
1042 inode = dentry->d_inode; /* can't be NULL */
1043 if (unlikely(!spin_trylock(&inode->i_lock))) {
1044 spin_unlock(&dentry->d_lock);
1046 spin_unlock(&parent->d_lock);
1050 __dentry_kill(dentry);
1056 static enum lru_status dentry_lru_isolate(struct list_head *item,
1057 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1059 struct list_head *freeable = arg;
1060 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1064 * we are inverting the lru lock/dentry->d_lock here,
1065 * so use a trylock. If we fail to get the lock, just skip
1068 if (!spin_trylock(&dentry->d_lock))
1072 * Referenced dentries are still in use. If they have active
1073 * counts, just remove them from the LRU. Otherwise give them
1074 * another pass through the LRU.
1076 if (dentry->d_lockref.count) {
1077 d_lru_isolate(lru, dentry);
1078 spin_unlock(&dentry->d_lock);
1082 if (dentry->d_flags & DCACHE_REFERENCED) {
1083 dentry->d_flags &= ~DCACHE_REFERENCED;
1084 spin_unlock(&dentry->d_lock);
1087 * The list move itself will be made by the common LRU code. At
1088 * this point, we've dropped the dentry->d_lock but keep the
1089 * lru lock. This is safe to do, since every list movement is
1090 * protected by the lru lock even if both locks are held.
1092 * This is guaranteed by the fact that all LRU management
1093 * functions are intermediated by the LRU API calls like
1094 * list_lru_add and list_lru_del. List movement in this file
1095 * only ever occur through this functions or through callbacks
1096 * like this one, that are called from the LRU API.
1098 * The only exceptions to this are functions like
1099 * shrink_dentry_list, and code that first checks for the
1100 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1101 * operating only with stack provided lists after they are
1102 * properly isolated from the main list. It is thus, always a
1108 d_lru_shrink_move(lru, dentry, freeable);
1109 spin_unlock(&dentry->d_lock);
1115 * prune_dcache_sb - shrink the dcache
1117 * @sc: shrink control, passed to list_lru_shrink_walk()
1119 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1120 * is done when we need more memory and called from the superblock shrinker
1123 * This function may fail to free any resources if all the dentries are in
1126 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1131 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1132 dentry_lru_isolate, &dispose);
1133 shrink_dentry_list(&dispose);
1137 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1138 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1140 struct list_head *freeable = arg;
1141 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1144 * we are inverting the lru lock/dentry->d_lock here,
1145 * so use a trylock. If we fail to get the lock, just skip
1148 if (!spin_trylock(&dentry->d_lock))
1151 d_lru_shrink_move(lru, dentry, freeable);
1152 spin_unlock(&dentry->d_lock);
1159 * shrink_dcache_sb - shrink dcache for a superblock
1162 * Shrink the dcache for the specified super block. This is used to free
1163 * the dcache before unmounting a file system.
1165 void shrink_dcache_sb(struct super_block *sb)
1172 freed = list_lru_walk(&sb->s_dentry_lru,
1173 dentry_lru_isolate_shrink, &dispose, 1024);
1175 this_cpu_sub(nr_dentry_unused, freed);
1176 shrink_dentry_list(&dispose);
1178 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1180 EXPORT_SYMBOL(shrink_dcache_sb);
1183 * enum d_walk_ret - action to talke during tree walk
1184 * @D_WALK_CONTINUE: contrinue walk
1185 * @D_WALK_QUIT: quit walk
1186 * @D_WALK_NORETRY: quit when retry is needed
1187 * @D_WALK_SKIP: skip this dentry and its children
1197 * d_walk - walk the dentry tree
1198 * @parent: start of walk
1199 * @data: data passed to @enter() and @finish()
1200 * @enter: callback when first entering the dentry
1201 * @finish: callback when successfully finished the walk
1203 * The @enter() and @finish() callbacks are called with d_lock held.
1205 static void d_walk(struct dentry *parent, void *data,
1206 enum d_walk_ret (*enter)(void *, struct dentry *),
1207 void (*finish)(void *))
1209 struct dentry *this_parent;
1210 struct list_head *next;
1212 enum d_walk_ret ret;
1216 read_seqbegin_or_lock(&rename_lock, &seq);
1217 this_parent = parent;
1218 spin_lock(&this_parent->d_lock);
1220 ret = enter(data, this_parent);
1222 case D_WALK_CONTINUE:
1227 case D_WALK_NORETRY:
1232 next = this_parent->d_subdirs.next;
1234 while (next != &this_parent->d_subdirs) {
1235 struct list_head *tmp = next;
1236 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1239 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1242 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1244 ret = enter(data, dentry);
1246 case D_WALK_CONTINUE:
1249 spin_unlock(&dentry->d_lock);
1251 case D_WALK_NORETRY:
1255 spin_unlock(&dentry->d_lock);
1259 if (!list_empty(&dentry->d_subdirs)) {
1260 spin_unlock(&this_parent->d_lock);
1261 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1262 this_parent = dentry;
1263 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1266 spin_unlock(&dentry->d_lock);
1269 * All done at this level ... ascend and resume the search.
1273 if (this_parent != parent) {
1274 struct dentry *child = this_parent;
1275 this_parent = child->d_parent;
1277 spin_unlock(&child->d_lock);
1278 spin_lock(&this_parent->d_lock);
1280 /* might go back up the wrong parent if we have had a rename. */
1281 if (need_seqretry(&rename_lock, seq))
1283 /* go into the first sibling still alive */
1285 next = child->d_child.next;
1286 if (next == &this_parent->d_subdirs)
1288 child = list_entry(next, struct dentry, d_child);
1289 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1293 if (need_seqretry(&rename_lock, seq))
1300 spin_unlock(&this_parent->d_lock);
1301 done_seqretry(&rename_lock, seq);
1305 spin_unlock(&this_parent->d_lock);
1314 struct check_mount {
1315 struct vfsmount *mnt;
1316 unsigned int mounted;
1319 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1321 struct check_mount *info = data;
1322 struct path path = { .mnt = info->mnt, .dentry = dentry };
1324 if (likely(!d_mountpoint(dentry)))
1325 return D_WALK_CONTINUE;
1326 if (__path_is_mountpoint(&path)) {
1330 return D_WALK_CONTINUE;
1334 * path_has_submounts - check for mounts over a dentry in the
1335 * current namespace.
1336 * @parent: path to check.
1338 * Return true if the parent or its subdirectories contain
1339 * a mount point in the current namespace.
1341 int path_has_submounts(const struct path *parent)
1343 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1345 read_seqlock_excl(&mount_lock);
1346 d_walk(parent->dentry, &data, path_check_mount, NULL);
1347 read_sequnlock_excl(&mount_lock);
1349 return data.mounted;
1351 EXPORT_SYMBOL(path_has_submounts);
1354 * Called by mount code to set a mountpoint and check if the mountpoint is
1355 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1356 * subtree can become unreachable).
1358 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1359 * this reason take rename_lock and d_lock on dentry and ancestors.
1361 int d_set_mounted(struct dentry *dentry)
1365 write_seqlock(&rename_lock);
1366 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1367 /* Need exclusion wrt. d_invalidate() */
1368 spin_lock(&p->d_lock);
1369 if (unlikely(d_unhashed(p))) {
1370 spin_unlock(&p->d_lock);
1373 spin_unlock(&p->d_lock);
1375 spin_lock(&dentry->d_lock);
1376 if (!d_unlinked(dentry)) {
1378 if (!d_mountpoint(dentry)) {
1379 dentry->d_flags |= DCACHE_MOUNTED;
1383 spin_unlock(&dentry->d_lock);
1385 write_sequnlock(&rename_lock);
1390 * Search the dentry child list of the specified parent,
1391 * and move any unused dentries to the end of the unused
1392 * list for prune_dcache(). We descend to the next level
1393 * whenever the d_subdirs list is non-empty and continue
1396 * It returns zero iff there are no unused children,
1397 * otherwise it returns the number of children moved to
1398 * the end of the unused list. This may not be the total
1399 * number of unused children, because select_parent can
1400 * drop the lock and return early due to latency
1404 struct select_data {
1405 struct dentry *start;
1406 struct list_head dispose;
1410 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1412 struct select_data *data = _data;
1413 enum d_walk_ret ret = D_WALK_CONTINUE;
1415 if (data->start == dentry)
1418 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1421 if (dentry->d_flags & DCACHE_LRU_LIST)
1423 if (!dentry->d_lockref.count) {
1424 d_shrink_add(dentry, &data->dispose);
1429 * We can return to the caller if we have found some (this
1430 * ensures forward progress). We'll be coming back to find
1433 if (!list_empty(&data->dispose))
1434 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1440 * shrink_dcache_parent - prune dcache
1441 * @parent: parent of entries to prune
1443 * Prune the dcache to remove unused children of the parent dentry.
1445 void shrink_dcache_parent(struct dentry *parent)
1448 struct select_data data;
1450 INIT_LIST_HEAD(&data.dispose);
1451 data.start = parent;
1454 d_walk(parent, &data, select_collect, NULL);
1458 shrink_dentry_list(&data.dispose);
1462 EXPORT_SYMBOL(shrink_dcache_parent);
1464 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1466 /* it has busy descendents; complain about those instead */
1467 if (!list_empty(&dentry->d_subdirs))
1468 return D_WALK_CONTINUE;
1470 /* root with refcount 1 is fine */
1471 if (dentry == _data && dentry->d_lockref.count == 1)
1472 return D_WALK_CONTINUE;
1474 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1475 " still in use (%d) [unmount of %s %s]\n",
1478 dentry->d_inode->i_ino : 0UL,
1480 dentry->d_lockref.count,
1481 dentry->d_sb->s_type->name,
1482 dentry->d_sb->s_id);
1484 return D_WALK_CONTINUE;
1487 static void do_one_tree(struct dentry *dentry)
1489 shrink_dcache_parent(dentry);
1490 d_walk(dentry, dentry, umount_check, NULL);
1496 * destroy the dentries attached to a superblock on unmounting
1498 void shrink_dcache_for_umount(struct super_block *sb)
1500 struct dentry *dentry;
1502 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1504 dentry = sb->s_root;
1506 do_one_tree(dentry);
1508 while (!hlist_bl_empty(&sb->s_roots)) {
1509 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_roots), struct dentry, d_hash));
1510 do_one_tree(dentry);
1514 struct detach_data {
1515 struct select_data select;
1516 struct dentry *mountpoint;
1518 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1520 struct detach_data *data = _data;
1522 if (d_mountpoint(dentry)) {
1523 __dget_dlock(dentry);
1524 data->mountpoint = dentry;
1528 return select_collect(&data->select, dentry);
1531 static void check_and_drop(void *_data)
1533 struct detach_data *data = _data;
1535 if (!data->mountpoint && list_empty(&data->select.dispose))
1536 __d_drop(data->select.start);
1540 * d_invalidate - detach submounts, prune dcache, and drop
1541 * @dentry: dentry to invalidate (aka detach, prune and drop)
1545 * The final d_drop is done as an atomic operation relative to
1546 * rename_lock ensuring there are no races with d_set_mounted. This
1547 * ensures there are no unhashed dentries on the path to a mountpoint.
1549 void d_invalidate(struct dentry *dentry)
1552 * If it's already been dropped, return OK.
1554 spin_lock(&dentry->d_lock);
1555 if (d_unhashed(dentry)) {
1556 spin_unlock(&dentry->d_lock);
1559 spin_unlock(&dentry->d_lock);
1561 /* Negative dentries can be dropped without further checks */
1562 if (!dentry->d_inode) {
1568 struct detach_data data;
1570 data.mountpoint = NULL;
1571 INIT_LIST_HEAD(&data.select.dispose);
1572 data.select.start = dentry;
1573 data.select.found = 0;
1575 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1577 if (!list_empty(&data.select.dispose))
1578 shrink_dentry_list(&data.select.dispose);
1579 else if (!data.mountpoint)
1582 if (data.mountpoint) {
1583 detach_mounts(data.mountpoint);
1584 dput(data.mountpoint);
1589 EXPORT_SYMBOL(d_invalidate);
1592 * __d_alloc - allocate a dcache entry
1593 * @sb: filesystem it will belong to
1594 * @name: qstr of the name
1596 * Allocates a dentry. It returns %NULL if there is insufficient memory
1597 * available. On a success the dentry is returned. The name passed in is
1598 * copied and the copy passed in may be reused after this call.
1601 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1603 struct dentry *dentry;
1607 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1612 * We guarantee that the inline name is always NUL-terminated.
1613 * This way the memcpy() done by the name switching in rename
1614 * will still always have a NUL at the end, even if we might
1615 * be overwriting an internal NUL character
1617 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1618 if (unlikely(!name)) {
1620 dname = dentry->d_iname;
1621 } else if (name->len > DNAME_INLINE_LEN-1) {
1622 size_t size = offsetof(struct external_name, name[1]);
1623 struct external_name *p = kmalloc(size + name->len,
1624 GFP_KERNEL_ACCOUNT);
1626 kmem_cache_free(dentry_cache, dentry);
1629 atomic_set(&p->u.count, 1);
1631 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1632 kasan_unpoison_shadow(dname,
1633 round_up(name->len + 1, sizeof(unsigned long)));
1635 dname = dentry->d_iname;
1638 dentry->d_name.len = name->len;
1639 dentry->d_name.hash = name->hash;
1640 memcpy(dname, name->name, name->len);
1641 dname[name->len] = 0;
1643 /* Make sure we always see the terminating NUL character */
1644 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1646 dentry->d_lockref.count = 1;
1647 dentry->d_flags = 0;
1648 spin_lock_init(&dentry->d_lock);
1649 seqcount_init(&dentry->d_seq);
1650 dentry->d_inode = NULL;
1651 dentry->d_parent = dentry;
1653 dentry->d_op = NULL;
1654 dentry->d_fsdata = NULL;
1655 INIT_HLIST_BL_NODE(&dentry->d_hash);
1656 INIT_LIST_HEAD(&dentry->d_lru);
1657 INIT_LIST_HEAD(&dentry->d_subdirs);
1658 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1659 INIT_LIST_HEAD(&dentry->d_child);
1660 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1662 if (dentry->d_op && dentry->d_op->d_init) {
1663 err = dentry->d_op->d_init(dentry);
1665 if (dname_external(dentry))
1666 kfree(external_name(dentry));
1667 kmem_cache_free(dentry_cache, dentry);
1672 this_cpu_inc(nr_dentry);
1678 * d_alloc - allocate a dcache entry
1679 * @parent: parent of entry to allocate
1680 * @name: qstr of the name
1682 * Allocates a dentry. It returns %NULL if there is insufficient memory
1683 * available. On a success the dentry is returned. The name passed in is
1684 * copied and the copy passed in may be reused after this call.
1686 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1688 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1691 dentry->d_flags |= DCACHE_RCUACCESS;
1692 spin_lock(&parent->d_lock);
1694 * don't need child lock because it is not subject
1695 * to concurrency here
1697 __dget_dlock(parent);
1698 dentry->d_parent = parent;
1699 list_add(&dentry->d_child, &parent->d_subdirs);
1700 spin_unlock(&parent->d_lock);
1704 EXPORT_SYMBOL(d_alloc);
1706 struct dentry *d_alloc_cursor(struct dentry * parent)
1708 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1710 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1711 dentry->d_parent = dget(parent);
1717 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1718 * @sb: the superblock
1719 * @name: qstr of the name
1721 * For a filesystem that just pins its dentries in memory and never
1722 * performs lookups at all, return an unhashed IS_ROOT dentry.
1724 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1726 return __d_alloc(sb, name);
1728 EXPORT_SYMBOL(d_alloc_pseudo);
1730 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1735 q.hash_len = hashlen_string(parent, name);
1736 return d_alloc(parent, &q);
1738 EXPORT_SYMBOL(d_alloc_name);
1740 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1742 WARN_ON_ONCE(dentry->d_op);
1743 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1745 DCACHE_OP_REVALIDATE |
1746 DCACHE_OP_WEAK_REVALIDATE |
1753 dentry->d_flags |= DCACHE_OP_HASH;
1755 dentry->d_flags |= DCACHE_OP_COMPARE;
1756 if (op->d_revalidate)
1757 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1758 if (op->d_weak_revalidate)
1759 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1761 dentry->d_flags |= DCACHE_OP_DELETE;
1763 dentry->d_flags |= DCACHE_OP_PRUNE;
1765 dentry->d_flags |= DCACHE_OP_REAL;
1768 EXPORT_SYMBOL(d_set_d_op);
1772 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1773 * @dentry - The dentry to mark
1775 * Mark a dentry as falling through to the lower layer (as set with
1776 * d_pin_lower()). This flag may be recorded on the medium.
1778 void d_set_fallthru(struct dentry *dentry)
1780 spin_lock(&dentry->d_lock);
1781 dentry->d_flags |= DCACHE_FALLTHRU;
1782 spin_unlock(&dentry->d_lock);
1784 EXPORT_SYMBOL(d_set_fallthru);
1786 static unsigned d_flags_for_inode(struct inode *inode)
1788 unsigned add_flags = DCACHE_REGULAR_TYPE;
1791 return DCACHE_MISS_TYPE;
1793 if (S_ISDIR(inode->i_mode)) {
1794 add_flags = DCACHE_DIRECTORY_TYPE;
1795 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1796 if (unlikely(!inode->i_op->lookup))
1797 add_flags = DCACHE_AUTODIR_TYPE;
1799 inode->i_opflags |= IOP_LOOKUP;
1801 goto type_determined;
1804 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1805 if (unlikely(inode->i_op->get_link)) {
1806 add_flags = DCACHE_SYMLINK_TYPE;
1807 goto type_determined;
1809 inode->i_opflags |= IOP_NOFOLLOW;
1812 if (unlikely(!S_ISREG(inode->i_mode)))
1813 add_flags = DCACHE_SPECIAL_TYPE;
1816 if (unlikely(IS_AUTOMOUNT(inode)))
1817 add_flags |= DCACHE_NEED_AUTOMOUNT;
1821 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1823 unsigned add_flags = d_flags_for_inode(inode);
1824 WARN_ON(d_in_lookup(dentry));
1826 spin_lock(&dentry->d_lock);
1827 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1828 raw_write_seqcount_begin(&dentry->d_seq);
1829 __d_set_inode_and_type(dentry, inode, add_flags);
1830 raw_write_seqcount_end(&dentry->d_seq);
1831 fsnotify_update_flags(dentry);
1832 spin_unlock(&dentry->d_lock);
1836 * d_instantiate - fill in inode information for a dentry
1837 * @entry: dentry to complete
1838 * @inode: inode to attach to this dentry
1840 * Fill in inode information in the entry.
1842 * This turns negative dentries into productive full members
1845 * NOTE! This assumes that the inode count has been incremented
1846 * (or otherwise set) by the caller to indicate that it is now
1847 * in use by the dcache.
1850 void d_instantiate(struct dentry *entry, struct inode * inode)
1852 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1854 security_d_instantiate(entry, inode);
1855 spin_lock(&inode->i_lock);
1856 __d_instantiate(entry, inode);
1857 spin_unlock(&inode->i_lock);
1860 EXPORT_SYMBOL(d_instantiate);
1863 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1864 * @entry: dentry to complete
1865 * @inode: inode to attach to this dentry
1867 * Fill in inode information in the entry. If a directory alias is found, then
1868 * return an error (and drop inode). Together with d_materialise_unique() this
1869 * guarantees that a directory inode may never have more than one alias.
1871 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1873 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1875 security_d_instantiate(entry, inode);
1876 spin_lock(&inode->i_lock);
1877 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1878 spin_unlock(&inode->i_lock);
1882 __d_instantiate(entry, inode);
1883 spin_unlock(&inode->i_lock);
1887 EXPORT_SYMBOL(d_instantiate_no_diralias);
1889 struct dentry *d_make_root(struct inode *root_inode)
1891 struct dentry *res = NULL;
1894 res = __d_alloc(root_inode->i_sb, NULL);
1896 d_instantiate(res, root_inode);
1902 EXPORT_SYMBOL(d_make_root);
1904 static struct dentry * __d_find_any_alias(struct inode *inode)
1906 struct dentry *alias;
1908 if (hlist_empty(&inode->i_dentry))
1910 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1916 * d_find_any_alias - find any alias for a given inode
1917 * @inode: inode to find an alias for
1919 * If any aliases exist for the given inode, take and return a
1920 * reference for one of them. If no aliases exist, return %NULL.
1922 struct dentry *d_find_any_alias(struct inode *inode)
1926 spin_lock(&inode->i_lock);
1927 de = __d_find_any_alias(inode);
1928 spin_unlock(&inode->i_lock);
1931 EXPORT_SYMBOL(d_find_any_alias);
1933 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1940 return ERR_PTR(-ESTALE);
1942 return ERR_CAST(inode);
1944 res = d_find_any_alias(inode);
1948 tmp = __d_alloc(inode->i_sb, NULL);
1950 res = ERR_PTR(-ENOMEM);
1954 security_d_instantiate(tmp, inode);
1955 spin_lock(&inode->i_lock);
1956 res = __d_find_any_alias(inode);
1958 spin_unlock(&inode->i_lock);
1963 /* attach a disconnected dentry */
1964 add_flags = d_flags_for_inode(inode);
1967 add_flags |= DCACHE_DISCONNECTED;
1969 spin_lock(&tmp->d_lock);
1970 __d_set_inode_and_type(tmp, inode, add_flags);
1971 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1972 if (!disconnected) {
1973 hlist_bl_lock(&tmp->d_sb->s_roots);
1974 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_roots);
1975 hlist_bl_unlock(&tmp->d_sb->s_roots);
1977 spin_unlock(&tmp->d_lock);
1978 spin_unlock(&inode->i_lock);
1988 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1989 * @inode: inode to allocate the dentry for
1991 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1992 * similar open by handle operations. The returned dentry may be anonymous,
1993 * or may have a full name (if the inode was already in the cache).
1995 * When called on a directory inode, we must ensure that the inode only ever
1996 * has one dentry. If a dentry is found, that is returned instead of
1997 * allocating a new one.
1999 * On successful return, the reference to the inode has been transferred
2000 * to the dentry. In case of an error the reference on the inode is released.
2001 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2002 * be passed in and the error will be propagated to the return value,
2003 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2005 struct dentry *d_obtain_alias(struct inode *inode)
2007 return __d_obtain_alias(inode, 1);
2009 EXPORT_SYMBOL(d_obtain_alias);
2012 * d_obtain_root - find or allocate a dentry for a given inode
2013 * @inode: inode to allocate the dentry for
2015 * Obtain an IS_ROOT dentry for the root of a filesystem.
2017 * We must ensure that directory inodes only ever have one dentry. If a
2018 * dentry is found, that is returned instead of allocating a new one.
2020 * On successful return, the reference to the inode has been transferred
2021 * to the dentry. In case of an error the reference on the inode is
2022 * released. A %NULL or IS_ERR inode may be passed in and will be the
2023 * error will be propagate to the return value, with a %NULL @inode
2024 * replaced by ERR_PTR(-ESTALE).
2026 struct dentry *d_obtain_root(struct inode *inode)
2028 return __d_obtain_alias(inode, 0);
2030 EXPORT_SYMBOL(d_obtain_root);
2033 * d_add_ci - lookup or allocate new dentry with case-exact name
2034 * @inode: the inode case-insensitive lookup has found
2035 * @dentry: the negative dentry that was passed to the parent's lookup func
2036 * @name: the case-exact name to be associated with the returned dentry
2038 * This is to avoid filling the dcache with case-insensitive names to the
2039 * same inode, only the actual correct case is stored in the dcache for
2040 * case-insensitive filesystems.
2042 * For a case-insensitive lookup match and if the the case-exact dentry
2043 * already exists in in the dcache, use it and return it.
2045 * If no entry exists with the exact case name, allocate new dentry with
2046 * the exact case, and return the spliced entry.
2048 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2051 struct dentry *found, *res;
2054 * First check if a dentry matching the name already exists,
2055 * if not go ahead and create it now.
2057 found = d_hash_and_lookup(dentry->d_parent, name);
2062 if (d_in_lookup(dentry)) {
2063 found = d_alloc_parallel(dentry->d_parent, name,
2065 if (IS_ERR(found) || !d_in_lookup(found)) {
2070 found = d_alloc(dentry->d_parent, name);
2073 return ERR_PTR(-ENOMEM);
2076 res = d_splice_alias(inode, found);
2083 EXPORT_SYMBOL(d_add_ci);
2086 static inline bool d_same_name(const struct dentry *dentry,
2087 const struct dentry *parent,
2088 const struct qstr *name)
2090 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2091 if (dentry->d_name.len != name->len)
2093 return dentry_cmp(dentry, name->name, name->len) == 0;
2095 return parent->d_op->d_compare(dentry,
2096 dentry->d_name.len, dentry->d_name.name,
2101 * __d_lookup_rcu - search for a dentry (racy, store-free)
2102 * @parent: parent dentry
2103 * @name: qstr of name we wish to find
2104 * @seqp: returns d_seq value at the point where the dentry was found
2105 * Returns: dentry, or NULL
2107 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2108 * resolution (store-free path walking) design described in
2109 * Documentation/filesystems/path-lookup.txt.
2111 * This is not to be used outside core vfs.
2113 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2114 * held, and rcu_read_lock held. The returned dentry must not be stored into
2115 * without taking d_lock and checking d_seq sequence count against @seq
2118 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2121 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2122 * the returned dentry, so long as its parent's seqlock is checked after the
2123 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2124 * is formed, giving integrity down the path walk.
2126 * NOTE! The caller *has* to check the resulting dentry against the sequence
2127 * number we've returned before using any of the resulting dentry state!
2129 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2130 const struct qstr *name,
2133 u64 hashlen = name->hash_len;
2134 const unsigned char *str = name->name;
2135 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2136 struct hlist_bl_node *node;
2137 struct dentry *dentry;
2140 * Note: There is significant duplication with __d_lookup_rcu which is
2141 * required to prevent single threaded performance regressions
2142 * especially on architectures where smp_rmb (in seqcounts) are costly.
2143 * Keep the two functions in sync.
2147 * The hash list is protected using RCU.
2149 * Carefully use d_seq when comparing a candidate dentry, to avoid
2150 * races with d_move().
2152 * It is possible that concurrent renames can mess up our list
2153 * walk here and result in missing our dentry, resulting in the
2154 * false-negative result. d_lookup() protects against concurrent
2155 * renames using rename_lock seqlock.
2157 * See Documentation/filesystems/path-lookup.txt for more details.
2159 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2164 * The dentry sequence count protects us from concurrent
2165 * renames, and thus protects parent and name fields.
2167 * The caller must perform a seqcount check in order
2168 * to do anything useful with the returned dentry.
2170 * NOTE! We do a "raw" seqcount_begin here. That means that
2171 * we don't wait for the sequence count to stabilize if it
2172 * is in the middle of a sequence change. If we do the slow
2173 * dentry compare, we will do seqretries until it is stable,
2174 * and if we end up with a successful lookup, we actually
2175 * want to exit RCU lookup anyway.
2177 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2178 * we are still guaranteed NUL-termination of ->d_name.name.
2180 seq = raw_seqcount_begin(&dentry->d_seq);
2181 if (dentry->d_parent != parent)
2183 if (d_unhashed(dentry))
2186 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2189 if (dentry->d_name.hash != hashlen_hash(hashlen))
2191 tlen = dentry->d_name.len;
2192 tname = dentry->d_name.name;
2193 /* we want a consistent (name,len) pair */
2194 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2198 if (parent->d_op->d_compare(dentry,
2199 tlen, tname, name) != 0)
2202 if (dentry->d_name.hash_len != hashlen)
2204 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2214 * d_lookup - search for a dentry
2215 * @parent: parent dentry
2216 * @name: qstr of name we wish to find
2217 * Returns: dentry, or NULL
2219 * d_lookup searches the children of the parent dentry for the name in
2220 * question. If the dentry is found its reference count is incremented and the
2221 * dentry is returned. The caller must use dput to free the entry when it has
2222 * finished using it. %NULL is returned if the dentry does not exist.
2224 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2226 struct dentry *dentry;
2230 seq = read_seqbegin(&rename_lock);
2231 dentry = __d_lookup(parent, name);
2234 } while (read_seqretry(&rename_lock, seq));
2237 EXPORT_SYMBOL(d_lookup);
2240 * __d_lookup - search for a dentry (racy)
2241 * @parent: parent dentry
2242 * @name: qstr of name we wish to find
2243 * Returns: dentry, or NULL
2245 * __d_lookup is like d_lookup, however it may (rarely) return a
2246 * false-negative result due to unrelated rename activity.
2248 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2249 * however it must be used carefully, eg. with a following d_lookup in
2250 * the case of failure.
2252 * __d_lookup callers must be commented.
2254 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2256 unsigned int hash = name->hash;
2257 struct hlist_bl_head *b = d_hash(hash);
2258 struct hlist_bl_node *node;
2259 struct dentry *found = NULL;
2260 struct dentry *dentry;
2263 * Note: There is significant duplication with __d_lookup_rcu which is
2264 * required to prevent single threaded performance regressions
2265 * especially on architectures where smp_rmb (in seqcounts) are costly.
2266 * Keep the two functions in sync.
2270 * The hash list is protected using RCU.
2272 * Take d_lock when comparing a candidate dentry, to avoid races
2275 * It is possible that concurrent renames can mess up our list
2276 * walk here and result in missing our dentry, resulting in the
2277 * false-negative result. d_lookup() protects against concurrent
2278 * renames using rename_lock seqlock.
2280 * See Documentation/filesystems/path-lookup.txt for more details.
2284 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2286 if (dentry->d_name.hash != hash)
2289 spin_lock(&dentry->d_lock);
2290 if (dentry->d_parent != parent)
2292 if (d_unhashed(dentry))
2295 if (!d_same_name(dentry, parent, name))
2298 dentry->d_lockref.count++;
2300 spin_unlock(&dentry->d_lock);
2303 spin_unlock(&dentry->d_lock);
2311 * d_hash_and_lookup - hash the qstr then search for a dentry
2312 * @dir: Directory to search in
2313 * @name: qstr of name we wish to find
2315 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2317 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2320 * Check for a fs-specific hash function. Note that we must
2321 * calculate the standard hash first, as the d_op->d_hash()
2322 * routine may choose to leave the hash value unchanged.
2324 name->hash = full_name_hash(dir, name->name, name->len);
2325 if (dir->d_flags & DCACHE_OP_HASH) {
2326 int err = dir->d_op->d_hash(dir, name);
2327 if (unlikely(err < 0))
2328 return ERR_PTR(err);
2330 return d_lookup(dir, name);
2332 EXPORT_SYMBOL(d_hash_and_lookup);
2335 * When a file is deleted, we have two options:
2336 * - turn this dentry into a negative dentry
2337 * - unhash this dentry and free it.
2339 * Usually, we want to just turn this into
2340 * a negative dentry, but if anybody else is
2341 * currently using the dentry or the inode
2342 * we can't do that and we fall back on removing
2343 * it from the hash queues and waiting for
2344 * it to be deleted later when it has no users
2348 * d_delete - delete a dentry
2349 * @dentry: The dentry to delete
2351 * Turn the dentry into a negative dentry if possible, otherwise
2352 * remove it from the hash queues so it can be deleted later
2355 void d_delete(struct dentry * dentry)
2357 struct inode *inode;
2360 * Are we the only user?
2363 spin_lock(&dentry->d_lock);
2364 inode = dentry->d_inode;
2365 isdir = S_ISDIR(inode->i_mode);
2366 if (dentry->d_lockref.count == 1) {
2367 if (!spin_trylock(&inode->i_lock)) {
2368 spin_unlock(&dentry->d_lock);
2372 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2373 dentry_unlink_inode(dentry);
2374 fsnotify_nameremove(dentry, isdir);
2378 if (!d_unhashed(dentry))
2381 spin_unlock(&dentry->d_lock);
2383 fsnotify_nameremove(dentry, isdir);
2385 EXPORT_SYMBOL(d_delete);
2387 static void __d_rehash(struct dentry *entry)
2389 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2392 hlist_bl_add_head_rcu(&entry->d_hash, b);
2397 * d_rehash - add an entry back to the hash
2398 * @entry: dentry to add to the hash
2400 * Adds a dentry to the hash according to its name.
2403 void d_rehash(struct dentry * entry)
2405 spin_lock(&entry->d_lock);
2407 spin_unlock(&entry->d_lock);
2409 EXPORT_SYMBOL(d_rehash);
2411 static inline unsigned start_dir_add(struct inode *dir)
2415 unsigned n = dir->i_dir_seq;
2416 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2422 static inline void end_dir_add(struct inode *dir, unsigned n)
2424 smp_store_release(&dir->i_dir_seq, n + 2);
2427 static void d_wait_lookup(struct dentry *dentry)
2429 if (d_in_lookup(dentry)) {
2430 DECLARE_WAITQUEUE(wait, current);
2431 add_wait_queue(dentry->d_wait, &wait);
2433 set_current_state(TASK_UNINTERRUPTIBLE);
2434 spin_unlock(&dentry->d_lock);
2436 spin_lock(&dentry->d_lock);
2437 } while (d_in_lookup(dentry));
2441 struct dentry *d_alloc_parallel(struct dentry *parent,
2442 const struct qstr *name,
2443 wait_queue_head_t *wq)
2445 unsigned int hash = name->hash;
2446 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2447 struct hlist_bl_node *node;
2448 struct dentry *new = d_alloc(parent, name);
2449 struct dentry *dentry;
2450 unsigned seq, r_seq, d_seq;
2453 return ERR_PTR(-ENOMEM);
2457 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2458 r_seq = read_seqbegin(&rename_lock);
2459 dentry = __d_lookup_rcu(parent, name, &d_seq);
2460 if (unlikely(dentry)) {
2461 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2465 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2474 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2479 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2485 * No changes for the parent since the beginning of d_lookup().
2486 * Since all removals from the chain happen with hlist_bl_lock(),
2487 * any potential in-lookup matches are going to stay here until
2488 * we unlock the chain. All fields are stable in everything
2491 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2492 if (dentry->d_name.hash != hash)
2494 if (dentry->d_parent != parent)
2496 if (!d_same_name(dentry, parent, name))
2499 /* now we can try to grab a reference */
2500 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2507 * somebody is likely to be still doing lookup for it;
2508 * wait for them to finish
2510 spin_lock(&dentry->d_lock);
2511 d_wait_lookup(dentry);
2513 * it's not in-lookup anymore; in principle we should repeat
2514 * everything from dcache lookup, but it's likely to be what
2515 * d_lookup() would've found anyway. If it is, just return it;
2516 * otherwise we really have to repeat the whole thing.
2518 if (unlikely(dentry->d_name.hash != hash))
2520 if (unlikely(dentry->d_parent != parent))
2522 if (unlikely(d_unhashed(dentry)))
2524 if (unlikely(!d_same_name(dentry, parent, name)))
2526 /* OK, it *is* a hashed match; return it */
2527 spin_unlock(&dentry->d_lock);
2532 /* we can't take ->d_lock here; it's OK, though. */
2533 new->d_flags |= DCACHE_PAR_LOOKUP;
2535 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2539 spin_unlock(&dentry->d_lock);
2543 EXPORT_SYMBOL(d_alloc_parallel);
2545 void __d_lookup_done(struct dentry *dentry)
2547 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2548 dentry->d_name.hash);
2550 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2551 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2552 wake_up_all(dentry->d_wait);
2553 dentry->d_wait = NULL;
2555 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2556 INIT_LIST_HEAD(&dentry->d_lru);
2558 EXPORT_SYMBOL(__d_lookup_done);
2560 /* inode->i_lock held if inode is non-NULL */
2562 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2564 struct inode *dir = NULL;
2566 spin_lock(&dentry->d_lock);
2567 if (unlikely(d_in_lookup(dentry))) {
2568 dir = dentry->d_parent->d_inode;
2569 n = start_dir_add(dir);
2570 __d_lookup_done(dentry);
2573 unsigned add_flags = d_flags_for_inode(inode);
2574 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2575 raw_write_seqcount_begin(&dentry->d_seq);
2576 __d_set_inode_and_type(dentry, inode, add_flags);
2577 raw_write_seqcount_end(&dentry->d_seq);
2578 fsnotify_update_flags(dentry);
2582 end_dir_add(dir, n);
2583 spin_unlock(&dentry->d_lock);
2585 spin_unlock(&inode->i_lock);
2589 * d_add - add dentry to hash queues
2590 * @entry: dentry to add
2591 * @inode: The inode to attach to this dentry
2593 * This adds the entry to the hash queues and initializes @inode.
2594 * The entry was actually filled in earlier during d_alloc().
2597 void d_add(struct dentry *entry, struct inode *inode)
2600 security_d_instantiate(entry, inode);
2601 spin_lock(&inode->i_lock);
2603 __d_add(entry, inode);
2605 EXPORT_SYMBOL(d_add);
2608 * d_exact_alias - find and hash an exact unhashed alias
2609 * @entry: dentry to add
2610 * @inode: The inode to go with this dentry
2612 * If an unhashed dentry with the same name/parent and desired
2613 * inode already exists, hash and return it. Otherwise, return
2616 * Parent directory should be locked.
2618 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2620 struct dentry *alias;
2621 unsigned int hash = entry->d_name.hash;
2623 spin_lock(&inode->i_lock);
2624 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2626 * Don't need alias->d_lock here, because aliases with
2627 * d_parent == entry->d_parent are not subject to name or
2628 * parent changes, because the parent inode i_mutex is held.
2630 if (alias->d_name.hash != hash)
2632 if (alias->d_parent != entry->d_parent)
2634 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2636 spin_lock(&alias->d_lock);
2637 if (!d_unhashed(alias)) {
2638 spin_unlock(&alias->d_lock);
2641 __dget_dlock(alias);
2643 spin_unlock(&alias->d_lock);
2645 spin_unlock(&inode->i_lock);
2648 spin_unlock(&inode->i_lock);
2651 EXPORT_SYMBOL(d_exact_alias);
2654 * dentry_update_name_case - update case insensitive dentry with a new name
2655 * @dentry: dentry to be updated
2658 * Update a case insensitive dentry with new case of name.
2660 * dentry must have been returned by d_lookup with name @name. Old and new
2661 * name lengths must match (ie. no d_compare which allows mismatched name
2664 * Parent inode i_mutex must be held over d_lookup and into this call (to
2665 * keep renames and concurrent inserts, and readdir(2) away).
2667 void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
2669 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
2670 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2672 spin_lock(&dentry->d_lock);
2673 write_seqcount_begin(&dentry->d_seq);
2674 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2675 write_seqcount_end(&dentry->d_seq);
2676 spin_unlock(&dentry->d_lock);
2678 EXPORT_SYMBOL(dentry_update_name_case);
2680 static void swap_names(struct dentry *dentry, struct dentry *target)
2682 if (unlikely(dname_external(target))) {
2683 if (unlikely(dname_external(dentry))) {
2685 * Both external: swap the pointers
2687 swap(target->d_name.name, dentry->d_name.name);
2690 * dentry:internal, target:external. Steal target's
2691 * storage and make target internal.
2693 memcpy(target->d_iname, dentry->d_name.name,
2694 dentry->d_name.len + 1);
2695 dentry->d_name.name = target->d_name.name;
2696 target->d_name.name = target->d_iname;
2699 if (unlikely(dname_external(dentry))) {
2701 * dentry:external, target:internal. Give dentry's
2702 * storage to target and make dentry internal
2704 memcpy(dentry->d_iname, target->d_name.name,
2705 target->d_name.len + 1);
2706 target->d_name.name = dentry->d_name.name;
2707 dentry->d_name.name = dentry->d_iname;
2710 * Both are internal.
2713 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2714 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2715 swap(((long *) &dentry->d_iname)[i],
2716 ((long *) &target->d_iname)[i]);
2720 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2723 static void copy_name(struct dentry *dentry, struct dentry *target)
2725 struct external_name *old_name = NULL;
2726 if (unlikely(dname_external(dentry)))
2727 old_name = external_name(dentry);
2728 if (unlikely(dname_external(target))) {
2729 atomic_inc(&external_name(target)->u.count);
2730 dentry->d_name = target->d_name;
2732 memcpy(dentry->d_iname, target->d_name.name,
2733 target->d_name.len + 1);
2734 dentry->d_name.name = dentry->d_iname;
2735 dentry->d_name.hash_len = target->d_name.hash_len;
2737 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2738 kfree_rcu(old_name, u.head);
2741 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2744 * XXXX: do we really need to take target->d_lock?
2746 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2747 spin_lock(&target->d_parent->d_lock);
2749 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2750 spin_lock(&dentry->d_parent->d_lock);
2751 spin_lock_nested(&target->d_parent->d_lock,
2752 DENTRY_D_LOCK_NESTED);
2754 spin_lock(&target->d_parent->d_lock);
2755 spin_lock_nested(&dentry->d_parent->d_lock,
2756 DENTRY_D_LOCK_NESTED);
2759 if (target < dentry) {
2760 spin_lock_nested(&target->d_lock, 2);
2761 spin_lock_nested(&dentry->d_lock, 3);
2763 spin_lock_nested(&dentry->d_lock, 2);
2764 spin_lock_nested(&target->d_lock, 3);
2768 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2770 if (target->d_parent != dentry->d_parent)
2771 spin_unlock(&dentry->d_parent->d_lock);
2772 if (target->d_parent != target)
2773 spin_unlock(&target->d_parent->d_lock);
2774 spin_unlock(&target->d_lock);
2775 spin_unlock(&dentry->d_lock);
2779 * When switching names, the actual string doesn't strictly have to
2780 * be preserved in the target - because we're dropping the target
2781 * anyway. As such, we can just do a simple memcpy() to copy over
2782 * the new name before we switch, unless we are going to rehash
2783 * it. Note that if we *do* unhash the target, we are not allowed
2784 * to rehash it without giving it a new name/hash key - whether
2785 * we swap or overwrite the names here, resulting name won't match
2786 * the reality in filesystem; it's only there for d_path() purposes.
2787 * Note that all of this is happening under rename_lock, so the
2788 * any hash lookup seeing it in the middle of manipulations will
2789 * be discarded anyway. So we do not care what happens to the hash
2793 * __d_move - move a dentry
2794 * @dentry: entry to move
2795 * @target: new dentry
2796 * @exchange: exchange the two dentries
2798 * Update the dcache to reflect the move of a file name. Negative
2799 * dcache entries should not be moved in this way. Caller must hold
2800 * rename_lock, the i_mutex of the source and target directories,
2801 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2803 static void __d_move(struct dentry *dentry, struct dentry *target,
2806 struct inode *dir = NULL;
2808 if (!dentry->d_inode)
2809 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2811 BUG_ON(d_ancestor(dentry, target));
2812 BUG_ON(d_ancestor(target, dentry));
2814 dentry_lock_for_move(dentry, target);
2815 if (unlikely(d_in_lookup(target))) {
2816 dir = target->d_parent->d_inode;
2817 n = start_dir_add(dir);
2818 __d_lookup_done(target);
2821 write_seqcount_begin(&dentry->d_seq);
2822 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2825 /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2829 /* Switch the names.. */
2831 swap_names(dentry, target);
2833 copy_name(dentry, target);
2835 /* rehash in new place(s) */
2840 target->d_hash.pprev = NULL;
2842 /* ... and switch them in the tree */
2843 if (IS_ROOT(dentry)) {
2844 /* splicing a tree */
2845 dentry->d_flags |= DCACHE_RCUACCESS;
2846 dentry->d_parent = target->d_parent;
2847 target->d_parent = target;
2848 list_del_init(&target->d_child);
2849 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2851 /* swapping two dentries */
2852 swap(dentry->d_parent, target->d_parent);
2853 list_move(&target->d_child, &target->d_parent->d_subdirs);
2854 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2856 fsnotify_update_flags(target);
2857 fsnotify_update_flags(dentry);
2860 write_seqcount_end(&target->d_seq);
2861 write_seqcount_end(&dentry->d_seq);
2864 end_dir_add(dir, n);
2865 dentry_unlock_for_move(dentry, target);
2869 * d_move - move a dentry
2870 * @dentry: entry to move
2871 * @target: new dentry
2873 * Update the dcache to reflect the move of a file name. Negative
2874 * dcache entries should not be moved in this way. See the locking
2875 * requirements for __d_move.
2877 void d_move(struct dentry *dentry, struct dentry *target)
2879 write_seqlock(&rename_lock);
2880 __d_move(dentry, target, false);
2881 write_sequnlock(&rename_lock);
2883 EXPORT_SYMBOL(d_move);
2886 * d_exchange - exchange two dentries
2887 * @dentry1: first dentry
2888 * @dentry2: second dentry
2890 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2892 write_seqlock(&rename_lock);
2894 WARN_ON(!dentry1->d_inode);
2895 WARN_ON(!dentry2->d_inode);
2896 WARN_ON(IS_ROOT(dentry1));
2897 WARN_ON(IS_ROOT(dentry2));
2899 __d_move(dentry1, dentry2, true);
2901 write_sequnlock(&rename_lock);
2905 * d_ancestor - search for an ancestor
2906 * @p1: ancestor dentry
2909 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2910 * an ancestor of p2, else NULL.
2912 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2916 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2917 if (p->d_parent == p1)
2924 * This helper attempts to cope with remotely renamed directories
2926 * It assumes that the caller is already holding
2927 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2929 * Note: If ever the locking in lock_rename() changes, then please
2930 * remember to update this too...
2932 static int __d_unalias(struct inode *inode,
2933 struct dentry *dentry, struct dentry *alias)
2935 struct mutex *m1 = NULL;
2936 struct rw_semaphore *m2 = NULL;
2939 /* If alias and dentry share a parent, then no extra locks required */
2940 if (alias->d_parent == dentry->d_parent)
2943 /* See lock_rename() */
2944 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2946 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2947 if (!inode_trylock_shared(alias->d_parent->d_inode))
2949 m2 = &alias->d_parent->d_inode->i_rwsem;
2951 __d_move(alias, dentry, false);
2962 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2963 * @inode: the inode which may have a disconnected dentry
2964 * @dentry: a negative dentry which we want to point to the inode.
2966 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2967 * place of the given dentry and return it, else simply d_add the inode
2968 * to the dentry and return NULL.
2970 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2971 * we should error out: directories can't have multiple aliases.
2973 * This is needed in the lookup routine of any filesystem that is exportable
2974 * (via knfsd) so that we can build dcache paths to directories effectively.
2976 * If a dentry was found and moved, then it is returned. Otherwise NULL
2977 * is returned. This matches the expected return value of ->lookup.
2979 * Cluster filesystems may call this function with a negative, hashed dentry.
2980 * In that case, we know that the inode will be a regular file, and also this
2981 * will only occur during atomic_open. So we need to check for the dentry
2982 * being already hashed only in the final case.
2984 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2987 return ERR_CAST(inode);
2989 BUG_ON(!d_unhashed(dentry));
2994 security_d_instantiate(dentry, inode);
2995 spin_lock(&inode->i_lock);
2996 if (S_ISDIR(inode->i_mode)) {
2997 struct dentry *new = __d_find_any_alias(inode);
2998 if (unlikely(new)) {
2999 /* The reference to new ensures it remains an alias */
3000 spin_unlock(&inode->i_lock);
3001 write_seqlock(&rename_lock);
3002 if (unlikely(d_ancestor(new, dentry))) {
3003 write_sequnlock(&rename_lock);
3005 new = ERR_PTR(-ELOOP);
3006 pr_warn_ratelimited(
3007 "VFS: Lookup of '%s' in %s %s"
3008 " would have caused loop\n",
3009 dentry->d_name.name,
3010 inode->i_sb->s_type->name,
3012 } else if (!IS_ROOT(new)) {
3013 int err = __d_unalias(inode, dentry, new);
3014 write_sequnlock(&rename_lock);
3020 __d_move(new, dentry, false);
3021 write_sequnlock(&rename_lock);
3028 __d_add(dentry, inode);
3031 EXPORT_SYMBOL(d_splice_alias);
3033 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
3037 return -ENAMETOOLONG;
3039 memcpy(*buffer, str, namelen);
3044 * prepend_name - prepend a pathname in front of current buffer pointer
3045 * @buffer: buffer pointer
3046 * @buflen: allocated length of the buffer
3047 * @name: name string and length qstr structure
3049 * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
3050 * make sure that either the old or the new name pointer and length are
3051 * fetched. However, there may be mismatch between length and pointer.
3052 * The length cannot be trusted, we need to copy it byte-by-byte until
3053 * the length is reached or a null byte is found. It also prepends "/" at
3054 * the beginning of the name. The sequence number check at the caller will
3055 * retry it again when a d_move() does happen. So any garbage in the buffer
3056 * due to mismatched pointer and length will be discarded.
3058 * Load acquire is needed to make sure that we see that terminating NUL.
3060 static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
3062 const char *dname = smp_load_acquire(&name->name); /* ^^^ */
3063 u32 dlen = READ_ONCE(name->len);
3066 *buflen -= dlen + 1;
3068 return -ENAMETOOLONG;
3069 p = *buffer -= dlen + 1;
3081 * prepend_path - Prepend path string to a buffer
3082 * @path: the dentry/vfsmount to report
3083 * @root: root vfsmnt/dentry
3084 * @buffer: pointer to the end of the buffer
3085 * @buflen: pointer to buffer length
3087 * The function will first try to write out the pathname without taking any
3088 * lock other than the RCU read lock to make sure that dentries won't go away.
3089 * It only checks the sequence number of the global rename_lock as any change
3090 * in the dentry's d_seq will be preceded by changes in the rename_lock
3091 * sequence number. If the sequence number had been changed, it will restart
3092 * the whole pathname back-tracing sequence again by taking the rename_lock.
3093 * In this case, there is no need to take the RCU read lock as the recursive
3094 * parent pointer references will keep the dentry chain alive as long as no
3095 * rename operation is performed.
3097 static int prepend_path(const struct path *path,
3098 const struct path *root,
3099 char **buffer, int *buflen)
3101 struct dentry *dentry;
3102 struct vfsmount *vfsmnt;
3105 unsigned seq, m_seq = 0;
3111 read_seqbegin_or_lock(&mount_lock, &m_seq);
3118 dentry = path->dentry;
3120 mnt = real_mount(vfsmnt);
3121 read_seqbegin_or_lock(&rename_lock, &seq);
3122 while (dentry != root->dentry || vfsmnt != root->mnt) {
3123 struct dentry * parent;
3125 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
3126 struct mount *parent = READ_ONCE(mnt->mnt_parent);
3128 if (dentry != vfsmnt->mnt_root) {
3135 if (mnt != parent) {
3136 dentry = READ_ONCE(mnt->mnt_mountpoint);
3142 error = is_mounted(vfsmnt) ? 1 : 2;
3145 parent = dentry->d_parent;
3147 error = prepend_name(&bptr, &blen, &dentry->d_name);
3155 if (need_seqretry(&rename_lock, seq)) {
3159 done_seqretry(&rename_lock, seq);
3163 if (need_seqretry(&mount_lock, m_seq)) {
3167 done_seqretry(&mount_lock, m_seq);
3169 if (error >= 0 && bptr == *buffer) {
3171 error = -ENAMETOOLONG;
3181 * __d_path - return the path of a dentry
3182 * @path: the dentry/vfsmount to report
3183 * @root: root vfsmnt/dentry
3184 * @buf: buffer to return value in
3185 * @buflen: buffer length
3187 * Convert a dentry into an ASCII path name.
3189 * Returns a pointer into the buffer or an error code if the
3190 * path was too long.
3192 * "buflen" should be positive.
3194 * If the path is not reachable from the supplied root, return %NULL.
3196 char *__d_path(const struct path *path,
3197 const struct path *root,
3198 char *buf, int buflen)
3200 char *res = buf + buflen;
3203 prepend(&res, &buflen, "\0", 1);
3204 error = prepend_path(path, root, &res, &buflen);
3207 return ERR_PTR(error);
3213 char *d_absolute_path(const struct path *path,
3214 char *buf, int buflen)
3216 struct path root = {};
3217 char *res = buf + buflen;
3220 prepend(&res, &buflen, "\0", 1);
3221 error = prepend_path(path, &root, &res, &buflen);
3226 return ERR_PTR(error);
3231 * same as __d_path but appends "(deleted)" for unlinked files.
3233 static int path_with_deleted(const struct path *path,
3234 const struct path *root,
3235 char **buf, int *buflen)
3237 prepend(buf, buflen, "\0", 1);
3238 if (d_unlinked(path->dentry)) {
3239 int error = prepend(buf, buflen, " (deleted)", 10);
3244 return prepend_path(path, root, buf, buflen);
3247 static int prepend_unreachable(char **buffer, int *buflen)
3249 return prepend(buffer, buflen, "(unreachable)", 13);
3252 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3257 seq = read_seqcount_begin(&fs->seq);
3259 } while (read_seqcount_retry(&fs->seq, seq));
3263 * d_path - return the path of a dentry
3264 * @path: path to report
3265 * @buf: buffer to return value in
3266 * @buflen: buffer length
3268 * Convert a dentry into an ASCII path name. If the entry has been deleted
3269 * the string " (deleted)" is appended. Note that this is ambiguous.
3271 * Returns a pointer into the buffer or an error code if the path was
3272 * too long. Note: Callers should use the returned pointer, not the passed
3273 * in buffer, to use the name! The implementation often starts at an offset
3274 * into the buffer, and may leave 0 bytes at the start.
3276 * "buflen" should be positive.
3278 char *d_path(const struct path *path, char *buf, int buflen)
3280 char *res = buf + buflen;
3285 * We have various synthetic filesystems that never get mounted. On
3286 * these filesystems dentries are never used for lookup purposes, and
3287 * thus don't need to be hashed. They also don't need a name until a
3288 * user wants to identify the object in /proc/pid/fd/. The little hack
3289 * below allows us to generate a name for these objects on demand:
3291 * Some pseudo inodes are mountable. When they are mounted
3292 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3293 * and instead have d_path return the mounted path.
3295 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3296 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3297 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3300 get_fs_root_rcu(current->fs, &root);
3301 error = path_with_deleted(path, &root, &res, &buflen);
3305 res = ERR_PTR(error);
3308 EXPORT_SYMBOL(d_path);
3311 * Helper function for dentry_operations.d_dname() members
3313 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3314 const char *fmt, ...)
3320 va_start(args, fmt);
3321 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3324 if (sz > sizeof(temp) || sz > buflen)
3325 return ERR_PTR(-ENAMETOOLONG);
3327 buffer += buflen - sz;
3328 return memcpy(buffer, temp, sz);
3331 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3333 char *end = buffer + buflen;
3334 /* these dentries are never renamed, so d_lock is not needed */
3335 if (prepend(&end, &buflen, " (deleted)", 11) ||
3336 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3337 prepend(&end, &buflen, "/", 1))
3338 end = ERR_PTR(-ENAMETOOLONG);
3341 EXPORT_SYMBOL(simple_dname);
3344 * Write full pathname from the root of the filesystem into the buffer.
3346 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3348 struct dentry *dentry;
3361 prepend(&end, &len, "\0", 1);
3365 read_seqbegin_or_lock(&rename_lock, &seq);
3366 while (!IS_ROOT(dentry)) {
3367 struct dentry *parent = dentry->d_parent;
3370 error = prepend_name(&end, &len, &dentry->d_name);
3379 if (need_seqretry(&rename_lock, seq)) {
3383 done_seqretry(&rename_lock, seq);
3388 return ERR_PTR(-ENAMETOOLONG);
3391 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3393 return __dentry_path(dentry, buf, buflen);
3395 EXPORT_SYMBOL(dentry_path_raw);
3397 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3402 if (d_unlinked(dentry)) {
3404 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3408 retval = __dentry_path(dentry, buf, buflen);
3409 if (!IS_ERR(retval) && p)
3410 *p = '/'; /* restore '/' overriden with '\0' */
3413 return ERR_PTR(-ENAMETOOLONG);
3416 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3422 seq = read_seqcount_begin(&fs->seq);
3425 } while (read_seqcount_retry(&fs->seq, seq));
3429 * NOTE! The user-level library version returns a
3430 * character pointer. The kernel system call just
3431 * returns the length of the buffer filled (which
3432 * includes the ending '\0' character), or a negative
3433 * error value. So libc would do something like
3435 * char *getcwd(char * buf, size_t size)
3439 * retval = sys_getcwd(buf, size);
3446 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3449 struct path pwd, root;
3450 char *page = __getname();
3456 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3459 if (!d_unlinked(pwd.dentry)) {
3461 char *cwd = page + PATH_MAX;
3462 int buflen = PATH_MAX;
3464 prepend(&cwd, &buflen, "\0", 1);
3465 error = prepend_path(&pwd, &root, &cwd, &buflen);
3471 /* Unreachable from current root */
3473 error = prepend_unreachable(&cwd, &buflen);
3479 len = PATH_MAX + page - cwd;
3482 if (copy_to_user(buf, cwd, len))
3495 * Test whether new_dentry is a subdirectory of old_dentry.
3497 * Trivially implemented using the dcache structure
3501 * is_subdir - is new dentry a subdirectory of old_dentry
3502 * @new_dentry: new dentry
3503 * @old_dentry: old dentry
3505 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3506 * Returns false otherwise.
3507 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3510 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3515 if (new_dentry == old_dentry)
3519 /* for restarting inner loop in case of seq retry */
3520 seq = read_seqbegin(&rename_lock);
3522 * Need rcu_readlock to protect against the d_parent trashing
3526 if (d_ancestor(old_dentry, new_dentry))
3531 } while (read_seqretry(&rename_lock, seq));
3536 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3538 struct dentry *root = data;
3539 if (dentry != root) {
3540 if (d_unhashed(dentry) || !dentry->d_inode)
3543 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3544 dentry->d_flags |= DCACHE_GENOCIDE;
3545 dentry->d_lockref.count--;
3548 return D_WALK_CONTINUE;
3551 void d_genocide(struct dentry *parent)
3553 d_walk(parent, parent, d_genocide_kill, NULL);
3556 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3558 inode_dec_link_count(inode);
3559 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3560 !hlist_unhashed(&dentry->d_u.d_alias) ||
3561 !d_unlinked(dentry));
3562 spin_lock(&dentry->d_parent->d_lock);
3563 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3564 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3565 (unsigned long long)inode->i_ino);
3566 spin_unlock(&dentry->d_lock);
3567 spin_unlock(&dentry->d_parent->d_lock);
3568 d_instantiate(dentry, inode);
3570 EXPORT_SYMBOL(d_tmpfile);
3572 static __initdata unsigned long dhash_entries;
3573 static int __init set_dhash_entries(char *str)
3577 dhash_entries = simple_strtoul(str, &str, 0);
3580 __setup("dhash_entries=", set_dhash_entries);
3582 static void __init dcache_init_early(void)
3584 /* If hashes are distributed across NUMA nodes, defer
3585 * hash allocation until vmalloc space is available.
3591 alloc_large_system_hash("Dentry cache",
3592 sizeof(struct hlist_bl_head),
3595 HASH_EARLY | HASH_ZERO,
3600 d_hash_shift = 32 - d_hash_shift;
3603 static void __init dcache_init(void)
3606 * A constructor could be added for stable state like the lists,
3607 * but it is probably not worth it because of the cache nature
3610 dentry_cache = KMEM_CACHE(dentry,
3611 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
3613 /* Hash may have been set up in dcache_init_early */
3618 alloc_large_system_hash("Dentry cache",
3619 sizeof(struct hlist_bl_head),
3627 d_hash_shift = 32 - d_hash_shift;
3630 /* SLAB cache for __getname() consumers */
3631 struct kmem_cache *names_cachep __read_mostly;
3632 EXPORT_SYMBOL(names_cachep);
3634 EXPORT_SYMBOL(d_genocide);
3636 void __init vfs_caches_init_early(void)
3640 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3641 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3643 dcache_init_early();
3647 void __init vfs_caches_init(void)
3649 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3650 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3655 files_maxfiles_init();