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/delay.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/hash.h>
26 #include <linux/cache.h>
27 #include <linux/export.h>
28 #include <linux/mount.h>
29 #include <linux/file.h>
30 #include <asm/uaccess.h>
31 #include <linux/security.h>
32 #include <linux/seqlock.h>
33 #include <linux/swap.h>
34 #include <linux/bootmem.h>
35 #include <linux/fs_struct.h>
36 #include <linux/hardirq.h>
37 #include <linux/bit_spinlock.h>
38 #include <linux/rculist_bl.h>
39 #include <linux/prefetch.h>
40 #include <linux/ratelimit.h>
41 #include <linux/list_lru.h>
42 #include <linux/kasan.h>
49 * dcache->d_inode->i_lock protects:
50 * - i_dentry, d_u.d_alias, d_inode of aliases
51 * dcache_hash_bucket lock protects:
52 * - the dcache hash table
53 * s_anon bl list spinlock protects:
54 * - the s_anon list (see __d_drop)
55 * dentry->d_sb->s_dentry_lru_lock protects:
56 * - the dcache lru lists and counters
63 * - d_parent and d_subdirs
64 * - childrens' d_child and d_parent
65 * - d_u.d_alias, d_inode
68 * dentry->d_inode->i_lock
70 * dentry->d_sb->s_dentry_lru_lock
71 * dcache_hash_bucket lock
74 * If there is an ancestor relationship:
75 * dentry->d_parent->...->d_parent->d_lock
77 * dentry->d_parent->d_lock
80 * If no ancestor relationship:
81 * if (dentry1 < dentry2)
85 int sysctl_vfs_cache_pressure __read_mostly = 100;
86 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
88 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
90 EXPORT_SYMBOL(rename_lock);
92 static struct kmem_cache *dentry_cache __read_mostly;
95 * This is the single most critical data structure when it comes
96 * to the dcache: the hashtable for lookups. Somebody should try
97 * to make this good - I've just made it work.
99 * This hash-function tries to avoid losing too many bits of hash
100 * information, yet avoid using a prime hash-size or similar.
103 static unsigned int d_hash_mask __read_mostly;
104 static unsigned int d_hash_shift __read_mostly;
106 static struct hlist_bl_head *dentry_hashtable __read_mostly;
108 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
111 hash += (unsigned long) parent / L1_CACHE_BYTES;
112 return dentry_hashtable + hash_32(hash, d_hash_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 = *(unsigned long *)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)
219 const unsigned char *cs;
221 * Be careful about RCU walk racing with rename:
222 * use ACCESS_ONCE to fetch the name pointer.
224 * NOTE! Even if a rename will mean that the length
225 * was not loaded atomically, we don't care. The
226 * RCU walk will check the sequence count eventually,
227 * and catch it. And we won't overrun the buffer,
228 * because we're reading the name pointer atomically,
229 * and a dentry name is guaranteed to be properly
230 * terminated with a NUL byte.
232 * End result: even if 'len' is wrong, we'll exit
233 * early because the data cannot match (there can
234 * be no NUL in the ct/tcount data)
236 cs = ACCESS_ONCE(dentry->d_name.name);
237 smp_read_barrier_depends();
238 return dentry_string_cmp(cs, ct, tcount);
241 struct external_name {
244 struct rcu_head head;
246 unsigned char name[];
249 static inline struct external_name *external_name(struct dentry *dentry)
251 return container_of(dentry->d_name.name, struct external_name, name[0]);
254 static void __d_free(struct rcu_head *head)
256 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
258 kmem_cache_free(dentry_cache, dentry);
261 static void __d_free_external(struct rcu_head *head)
263 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
264 kfree(external_name(dentry));
265 kmem_cache_free(dentry_cache, dentry);
268 static inline int dname_external(const struct dentry *dentry)
270 return dentry->d_name.name != dentry->d_iname;
273 static inline void __d_set_inode_and_type(struct dentry *dentry,
279 dentry->d_inode = inode;
280 flags = READ_ONCE(dentry->d_flags);
281 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
283 WRITE_ONCE(dentry->d_flags, flags);
286 static inline void __d_clear_type_and_inode(struct dentry *dentry)
288 unsigned flags = READ_ONCE(dentry->d_flags);
290 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
291 WRITE_ONCE(dentry->d_flags, flags);
292 dentry->d_inode = NULL;
295 static void dentry_free(struct dentry *dentry)
297 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
298 if (unlikely(dname_external(dentry))) {
299 struct external_name *p = external_name(dentry);
300 if (likely(atomic_dec_and_test(&p->u.count))) {
301 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
305 /* if dentry was never visible to RCU, immediate free is OK */
306 if (!(dentry->d_flags & DCACHE_RCUACCESS))
307 __d_free(&dentry->d_u.d_rcu);
309 call_rcu(&dentry->d_u.d_rcu, __d_free);
313 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
314 * @dentry: the target dentry
315 * After this call, in-progress rcu-walk path lookup will fail. This
316 * should be called after unhashing, and after changing d_inode (if
317 * the dentry has not already been unhashed).
319 static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
321 lockdep_assert_held(&dentry->d_lock);
322 /* Go through am invalidation barrier */
323 write_seqcount_invalidate(&dentry->d_seq);
327 * Release the dentry's inode, using the filesystem
328 * d_iput() operation if defined. Dentry has no refcount
331 static void dentry_iput(struct dentry * dentry)
332 __releases(dentry->d_lock)
333 __releases(dentry->d_inode->i_lock)
335 struct inode *inode = dentry->d_inode;
337 __d_clear_type_and_inode(dentry);
338 hlist_del_init(&dentry->d_u.d_alias);
339 spin_unlock(&dentry->d_lock);
340 spin_unlock(&inode->i_lock);
342 fsnotify_inoderemove(inode);
343 if (dentry->d_op && dentry->d_op->d_iput)
344 dentry->d_op->d_iput(dentry, inode);
348 spin_unlock(&dentry->d_lock);
353 * Release the dentry's inode, using the filesystem
354 * d_iput() operation if defined. dentry remains in-use.
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 * 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)))
455 * d_drop - drop a dentry
456 * @dentry: dentry to drop
458 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
459 * be found through a VFS lookup any more. Note that this is different from
460 * deleting the dentry - d_delete will try to mark the dentry negative if
461 * possible, giving a successful _negative_ lookup, while d_drop will
462 * just make the cache lookup fail.
464 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
465 * reason (NFS timeouts or autofs deletes).
467 * __d_drop requires dentry->d_lock.
469 void __d_drop(struct dentry *dentry)
471 if (!d_unhashed(dentry)) {
472 struct hlist_bl_head *b;
474 * Hashed dentries are normally on the dentry hashtable,
475 * with the exception of those newly allocated by
476 * d_obtain_alias, which are always IS_ROOT:
478 if (unlikely(IS_ROOT(dentry)))
479 b = &dentry->d_sb->s_anon;
481 b = d_hash(dentry->d_parent, dentry->d_name.hash);
484 __hlist_bl_del(&dentry->d_hash);
485 dentry->d_hash.pprev = NULL;
487 dentry_rcuwalk_invalidate(dentry);
490 EXPORT_SYMBOL(__d_drop);
492 void d_drop(struct dentry *dentry)
494 spin_lock(&dentry->d_lock);
496 spin_unlock(&dentry->d_lock);
498 EXPORT_SYMBOL(d_drop);
500 static void __dentry_kill(struct dentry *dentry)
502 struct dentry *parent = NULL;
503 bool can_free = true;
504 if (!IS_ROOT(dentry))
505 parent = dentry->d_parent;
508 * The dentry is now unrecoverably dead to the world.
510 lockref_mark_dead(&dentry->d_lockref);
513 * inform the fs via d_prune that this dentry is about to be
514 * unhashed and destroyed.
516 if (dentry->d_flags & DCACHE_OP_PRUNE)
517 dentry->d_op->d_prune(dentry);
519 if (dentry->d_flags & DCACHE_LRU_LIST) {
520 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
523 /* if it was on the hash then remove it */
525 __list_del_entry(&dentry->d_child);
527 * Inform d_walk() that we are no longer attached to the
530 dentry->d_flags |= DCACHE_DENTRY_KILLED;
532 spin_unlock(&parent->d_lock);
535 * dentry_iput drops the locks, at which point nobody (except
536 * transient RCU lookups) can reach this dentry.
538 BUG_ON(dentry->d_lockref.count > 0);
539 this_cpu_dec(nr_dentry);
540 if (dentry->d_op && dentry->d_op->d_release)
541 dentry->d_op->d_release(dentry);
543 spin_lock(&dentry->d_lock);
544 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
545 dentry->d_flags |= DCACHE_MAY_FREE;
548 spin_unlock(&dentry->d_lock);
549 if (likely(can_free))
554 * Finish off a dentry we've decided to kill.
555 * dentry->d_lock must be held, returns with it unlocked.
556 * If ref is non-zero, then decrement the refcount too.
557 * Returns dentry requiring refcount drop, or NULL if we're done.
559 static struct dentry *dentry_kill(struct dentry *dentry)
560 __releases(dentry->d_lock)
562 struct inode *inode = dentry->d_inode;
563 struct dentry *parent = NULL;
565 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
568 if (!IS_ROOT(dentry)) {
569 parent = dentry->d_parent;
570 if (unlikely(!spin_trylock(&parent->d_lock))) {
572 spin_unlock(&inode->i_lock);
577 __dentry_kill(dentry);
581 spin_unlock(&dentry->d_lock);
583 return dentry; /* try again with same dentry */
586 static inline struct dentry *lock_parent(struct dentry *dentry)
588 struct dentry *parent = dentry->d_parent;
591 if (unlikely(dentry->d_lockref.count < 0))
593 if (likely(spin_trylock(&parent->d_lock)))
596 spin_unlock(&dentry->d_lock);
598 parent = ACCESS_ONCE(dentry->d_parent);
599 spin_lock(&parent->d_lock);
601 * We can't blindly lock dentry until we are sure
602 * that we won't violate the locking order.
603 * Any changes of dentry->d_parent must have
604 * been done with parent->d_lock held, so
605 * spin_lock() above is enough of a barrier
606 * for checking if it's still our child.
608 if (unlikely(parent != dentry->d_parent)) {
609 spin_unlock(&parent->d_lock);
613 if (parent != dentry)
614 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
621 * Try to do a lockless dput(), and return whether that was successful.
623 * If unsuccessful, we return false, having already taken the dentry lock.
625 * The caller needs to hold the RCU read lock, so that the dentry is
626 * guaranteed to stay around even if the refcount goes down to zero!
628 static inline bool fast_dput(struct dentry *dentry)
631 unsigned int d_flags;
634 * If we have a d_op->d_delete() operation, we sould not
635 * let the dentry count go to zero, so use "put_or_lock".
637 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
638 return lockref_put_or_lock(&dentry->d_lockref);
641 * .. otherwise, we can try to just decrement the
642 * lockref optimistically.
644 ret = lockref_put_return(&dentry->d_lockref);
647 * If the lockref_put_return() failed due to the lock being held
648 * by somebody else, the fast path has failed. We will need to
649 * get the lock, and then check the count again.
651 if (unlikely(ret < 0)) {
652 spin_lock(&dentry->d_lock);
653 if (dentry->d_lockref.count > 1) {
654 dentry->d_lockref.count--;
655 spin_unlock(&dentry->d_lock);
662 * If we weren't the last ref, we're done.
668 * Careful, careful. The reference count went down
669 * to zero, but we don't hold the dentry lock, so
670 * somebody else could get it again, and do another
671 * dput(), and we need to not race with that.
673 * However, there is a very special and common case
674 * where we don't care, because there is nothing to
675 * do: the dentry is still hashed, it does not have
676 * a 'delete' op, and it's referenced and already on
679 * NOTE! Since we aren't locked, these values are
680 * not "stable". However, it is sufficient that at
681 * some point after we dropped the reference the
682 * dentry was hashed and the flags had the proper
683 * value. Other dentry users may have re-gotten
684 * a reference to the dentry and change that, but
685 * our work is done - we can leave the dentry
686 * around with a zero refcount.
689 d_flags = ACCESS_ONCE(dentry->d_flags);
690 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
692 /* Nothing to do? Dropping the reference was all we needed? */
693 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
697 * Not the fast normal case? Get the lock. We've already decremented
698 * the refcount, but we'll need to re-check the situation after
701 spin_lock(&dentry->d_lock);
704 * Did somebody else grab a reference to it in the meantime, and
705 * we're no longer the last user after all? Alternatively, somebody
706 * else could have killed it and marked it dead. Either way, we
707 * don't need to do anything else.
709 if (dentry->d_lockref.count) {
710 spin_unlock(&dentry->d_lock);
715 * Re-get the reference we optimistically dropped. We hold the
716 * lock, and we just tested that it was zero, so we can just
719 dentry->d_lockref.count = 1;
727 * This is complicated by the fact that we do not want to put
728 * dentries that are no longer on any hash chain on the unused
729 * list: we'd much rather just get rid of them immediately.
731 * However, that implies that we have to traverse the dentry
732 * tree upwards to the parents which might _also_ now be
733 * scheduled for deletion (it may have been only waiting for
734 * its last child to go away).
736 * This tail recursion is done by hand as we don't want to depend
737 * on the compiler to always get this right (gcc generally doesn't).
738 * Real recursion would eat up our stack space.
742 * dput - release a dentry
743 * @dentry: dentry to release
745 * Release a dentry. This will drop the usage count and if appropriate
746 * call the dentry unlink method as well as removing it from the queues and
747 * releasing its resources. If the parent dentries were scheduled for release
748 * they too may now get deleted.
750 void dput(struct dentry *dentry)
752 if (unlikely(!dentry))
757 if (likely(fast_dput(dentry))) {
762 /* Slow case: now with the dentry lock held */
765 /* Unreachable? Get rid of it */
766 if (unlikely(d_unhashed(dentry)))
769 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
772 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
773 if (dentry->d_op->d_delete(dentry))
777 if (!(dentry->d_flags & DCACHE_REFERENCED))
778 dentry->d_flags |= DCACHE_REFERENCED;
779 dentry_lru_add(dentry);
781 dentry->d_lockref.count--;
782 spin_unlock(&dentry->d_lock);
786 dentry = dentry_kill(dentry);
793 /* This must be called with d_lock held */
794 static inline void __dget_dlock(struct dentry *dentry)
796 dentry->d_lockref.count++;
799 static inline void __dget(struct dentry *dentry)
801 lockref_get(&dentry->d_lockref);
804 struct dentry *dget_parent(struct dentry *dentry)
810 * Do optimistic parent lookup without any
814 ret = ACCESS_ONCE(dentry->d_parent);
815 gotref = lockref_get_not_zero(&ret->d_lockref);
817 if (likely(gotref)) {
818 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
825 * Don't need rcu_dereference because we re-check it was correct under
829 ret = dentry->d_parent;
830 spin_lock(&ret->d_lock);
831 if (unlikely(ret != dentry->d_parent)) {
832 spin_unlock(&ret->d_lock);
837 BUG_ON(!ret->d_lockref.count);
838 ret->d_lockref.count++;
839 spin_unlock(&ret->d_lock);
842 EXPORT_SYMBOL(dget_parent);
845 * d_find_alias - grab a hashed alias of inode
846 * @inode: inode in question
848 * If inode has a hashed alias, or is a directory and has any alias,
849 * acquire the reference to alias and return it. Otherwise return NULL.
850 * Notice that if inode is a directory there can be only one alias and
851 * it can be unhashed only if it has no children, or if it is the root
852 * of a filesystem, or if the directory was renamed and d_revalidate
853 * was the first vfs operation to notice.
855 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
856 * any other hashed alias over that one.
858 static struct dentry *__d_find_alias(struct inode *inode)
860 struct dentry *alias, *discon_alias;
864 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
865 spin_lock(&alias->d_lock);
866 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
867 if (IS_ROOT(alias) &&
868 (alias->d_flags & DCACHE_DISCONNECTED)) {
869 discon_alias = alias;
872 spin_unlock(&alias->d_lock);
876 spin_unlock(&alias->d_lock);
879 alias = discon_alias;
880 spin_lock(&alias->d_lock);
881 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
883 spin_unlock(&alias->d_lock);
886 spin_unlock(&alias->d_lock);
892 struct dentry *d_find_alias(struct inode *inode)
894 struct dentry *de = NULL;
896 if (!hlist_empty(&inode->i_dentry)) {
897 spin_lock(&inode->i_lock);
898 de = __d_find_alias(inode);
899 spin_unlock(&inode->i_lock);
903 EXPORT_SYMBOL(d_find_alias);
906 * Try to kill dentries associated with this inode.
907 * WARNING: you must own a reference to inode.
909 void d_prune_aliases(struct inode *inode)
911 struct dentry *dentry;
913 spin_lock(&inode->i_lock);
914 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
915 spin_lock(&dentry->d_lock);
916 if (!dentry->d_lockref.count) {
917 struct dentry *parent = lock_parent(dentry);
918 if (likely(!dentry->d_lockref.count)) {
919 __dentry_kill(dentry);
924 spin_unlock(&parent->d_lock);
926 spin_unlock(&dentry->d_lock);
928 spin_unlock(&inode->i_lock);
930 EXPORT_SYMBOL(d_prune_aliases);
932 static void shrink_dentry_list(struct list_head *list)
934 struct dentry *dentry, *parent;
936 while (!list_empty(list)) {
938 dentry = list_entry(list->prev, struct dentry, d_lru);
939 spin_lock(&dentry->d_lock);
940 parent = lock_parent(dentry);
943 * The dispose list is isolated and dentries are not accounted
944 * to the LRU here, so we can simply remove it from the list
945 * here regardless of whether it is referenced or not.
947 d_shrink_del(dentry);
950 * We found an inuse dentry which was not removed from
951 * the LRU because of laziness during lookup. Do not free it.
953 if (dentry->d_lockref.count > 0) {
954 spin_unlock(&dentry->d_lock);
956 spin_unlock(&parent->d_lock);
961 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
962 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
963 spin_unlock(&dentry->d_lock);
965 spin_unlock(&parent->d_lock);
971 inode = dentry->d_inode;
972 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
973 d_shrink_add(dentry, list);
974 spin_unlock(&dentry->d_lock);
976 spin_unlock(&parent->d_lock);
980 __dentry_kill(dentry);
983 * We need to prune ancestors too. This is necessary to prevent
984 * quadratic behavior of shrink_dcache_parent(), but is also
985 * expected to be beneficial in reducing dentry cache
989 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
990 parent = lock_parent(dentry);
991 if (dentry->d_lockref.count != 1) {
992 dentry->d_lockref.count--;
993 spin_unlock(&dentry->d_lock);
995 spin_unlock(&parent->d_lock);
998 inode = dentry->d_inode; /* can't be NULL */
999 if (unlikely(!spin_trylock(&inode->i_lock))) {
1000 spin_unlock(&dentry->d_lock);
1002 spin_unlock(&parent->d_lock);
1006 __dentry_kill(dentry);
1012 static enum lru_status dentry_lru_isolate(struct list_head *item,
1013 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1015 struct list_head *freeable = arg;
1016 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1020 * we are inverting the lru lock/dentry->d_lock here,
1021 * so use a trylock. If we fail to get the lock, just skip
1024 if (!spin_trylock(&dentry->d_lock))
1028 * Referenced dentries are still in use. If they have active
1029 * counts, just remove them from the LRU. Otherwise give them
1030 * another pass through the LRU.
1032 if (dentry->d_lockref.count) {
1033 d_lru_isolate(lru, dentry);
1034 spin_unlock(&dentry->d_lock);
1038 if (dentry->d_flags & DCACHE_REFERENCED) {
1039 dentry->d_flags &= ~DCACHE_REFERENCED;
1040 spin_unlock(&dentry->d_lock);
1043 * The list move itself will be made by the common LRU code. At
1044 * this point, we've dropped the dentry->d_lock but keep the
1045 * lru lock. This is safe to do, since every list movement is
1046 * protected by the lru lock even if both locks are held.
1048 * This is guaranteed by the fact that all LRU management
1049 * functions are intermediated by the LRU API calls like
1050 * list_lru_add and list_lru_del. List movement in this file
1051 * only ever occur through this functions or through callbacks
1052 * like this one, that are called from the LRU API.
1054 * The only exceptions to this are functions like
1055 * shrink_dentry_list, and code that first checks for the
1056 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1057 * operating only with stack provided lists after they are
1058 * properly isolated from the main list. It is thus, always a
1064 d_lru_shrink_move(lru, dentry, freeable);
1065 spin_unlock(&dentry->d_lock);
1071 * prune_dcache_sb - shrink the dcache
1073 * @sc: shrink control, passed to list_lru_shrink_walk()
1075 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1076 * is done when we need more memory and called from the superblock shrinker
1079 * This function may fail to free any resources if all the dentries are in
1082 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1087 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1088 dentry_lru_isolate, &dispose);
1089 shrink_dentry_list(&dispose);
1093 static enum lru_status dentry_lru_isolate_shrink(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);
1100 * we are inverting the lru lock/dentry->d_lock here,
1101 * so use a trylock. If we fail to get the lock, just skip
1104 if (!spin_trylock(&dentry->d_lock))
1107 d_lru_shrink_move(lru, dentry, freeable);
1108 spin_unlock(&dentry->d_lock);
1115 * shrink_dcache_sb - shrink dcache for a superblock
1118 * Shrink the dcache for the specified super block. This is used to free
1119 * the dcache before unmounting a file system.
1121 void shrink_dcache_sb(struct super_block *sb)
1128 freed = list_lru_walk(&sb->s_dentry_lru,
1129 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1131 this_cpu_sub(nr_dentry_unused, freed);
1132 shrink_dentry_list(&dispose);
1133 } while (freed > 0);
1135 EXPORT_SYMBOL(shrink_dcache_sb);
1138 * enum d_walk_ret - action to talke during tree walk
1139 * @D_WALK_CONTINUE: contrinue walk
1140 * @D_WALK_QUIT: quit walk
1141 * @D_WALK_NORETRY: quit when retry is needed
1142 * @D_WALK_SKIP: skip this dentry and its children
1152 * d_walk - walk the dentry tree
1153 * @parent: start of walk
1154 * @data: data passed to @enter() and @finish()
1155 * @enter: callback when first entering the dentry
1156 * @finish: callback when successfully finished the walk
1158 * The @enter() and @finish() callbacks are called with d_lock held.
1160 static void d_walk(struct dentry *parent, void *data,
1161 enum d_walk_ret (*enter)(void *, struct dentry *),
1162 void (*finish)(void *))
1164 struct dentry *this_parent;
1165 struct list_head *next;
1167 enum d_walk_ret ret;
1171 read_seqbegin_or_lock(&rename_lock, &seq);
1172 this_parent = parent;
1173 spin_lock(&this_parent->d_lock);
1175 ret = enter(data, this_parent);
1177 case D_WALK_CONTINUE:
1182 case D_WALK_NORETRY:
1187 next = this_parent->d_subdirs.next;
1189 while (next != &this_parent->d_subdirs) {
1190 struct list_head *tmp = next;
1191 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1194 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1196 ret = enter(data, dentry);
1198 case D_WALK_CONTINUE:
1201 spin_unlock(&dentry->d_lock);
1203 case D_WALK_NORETRY:
1207 spin_unlock(&dentry->d_lock);
1211 if (!list_empty(&dentry->d_subdirs)) {
1212 spin_unlock(&this_parent->d_lock);
1213 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1214 this_parent = dentry;
1215 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1218 spin_unlock(&dentry->d_lock);
1221 * All done at this level ... ascend and resume the search.
1225 if (this_parent != parent) {
1226 struct dentry *child = this_parent;
1227 this_parent = child->d_parent;
1229 spin_unlock(&child->d_lock);
1230 spin_lock(&this_parent->d_lock);
1232 /* might go back up the wrong parent if we have had a rename. */
1233 if (need_seqretry(&rename_lock, seq))
1235 /* go into the first sibling still alive */
1237 next = child->d_child.next;
1238 if (next == &this_parent->d_subdirs)
1240 child = list_entry(next, struct dentry, d_child);
1241 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1245 if (need_seqretry(&rename_lock, seq))
1252 spin_unlock(&this_parent->d_lock);
1253 done_seqretry(&rename_lock, seq);
1257 spin_unlock(&this_parent->d_lock);
1267 * Search for at least 1 mount point in the dentry's subdirs.
1268 * We descend to the next level whenever the d_subdirs
1269 * list is non-empty and continue searching.
1272 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1275 if (d_mountpoint(dentry)) {
1279 return D_WALK_CONTINUE;
1283 * have_submounts - check for mounts over a dentry
1284 * @parent: dentry to check.
1286 * Return true if the parent or its subdirectories contain
1289 int have_submounts(struct dentry *parent)
1293 d_walk(parent, &ret, check_mount, NULL);
1297 EXPORT_SYMBOL(have_submounts);
1300 * Called by mount code to set a mountpoint and check if the mountpoint is
1301 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1302 * subtree can become unreachable).
1304 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1305 * this reason take rename_lock and d_lock on dentry and ancestors.
1307 int d_set_mounted(struct dentry *dentry)
1311 write_seqlock(&rename_lock);
1312 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1313 /* Need exclusion wrt. d_invalidate() */
1314 spin_lock(&p->d_lock);
1315 if (unlikely(d_unhashed(p))) {
1316 spin_unlock(&p->d_lock);
1319 spin_unlock(&p->d_lock);
1321 spin_lock(&dentry->d_lock);
1322 if (!d_unlinked(dentry)) {
1323 dentry->d_flags |= DCACHE_MOUNTED;
1326 spin_unlock(&dentry->d_lock);
1328 write_sequnlock(&rename_lock);
1333 * Search the dentry child list of the specified parent,
1334 * and move any unused dentries to the end of the unused
1335 * list for prune_dcache(). We descend to the next level
1336 * whenever the d_subdirs list is non-empty and continue
1339 * It returns zero iff there are no unused children,
1340 * otherwise it returns the number of children moved to
1341 * the end of the unused list. This may not be the total
1342 * number of unused children, because select_parent can
1343 * drop the lock and return early due to latency
1347 struct select_data {
1348 struct dentry *start;
1349 struct list_head dispose;
1353 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1355 struct select_data *data = _data;
1356 enum d_walk_ret ret = D_WALK_CONTINUE;
1358 if (data->start == dentry)
1361 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1364 if (dentry->d_flags & DCACHE_LRU_LIST)
1366 if (!dentry->d_lockref.count) {
1367 d_shrink_add(dentry, &data->dispose);
1372 * We can return to the caller if we have found some (this
1373 * ensures forward progress). We'll be coming back to find
1376 if (!list_empty(&data->dispose))
1377 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1383 * shrink_dcache_parent - prune dcache
1384 * @parent: parent of entries to prune
1386 * Prune the dcache to remove unused children of the parent dentry.
1388 void shrink_dcache_parent(struct dentry *parent)
1391 struct select_data data;
1393 INIT_LIST_HEAD(&data.dispose);
1394 data.start = parent;
1397 d_walk(parent, &data, select_collect, NULL);
1401 shrink_dentry_list(&data.dispose);
1405 EXPORT_SYMBOL(shrink_dcache_parent);
1407 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1409 /* it has busy descendents; complain about those instead */
1410 if (!list_empty(&dentry->d_subdirs))
1411 return D_WALK_CONTINUE;
1413 /* root with refcount 1 is fine */
1414 if (dentry == _data && dentry->d_lockref.count == 1)
1415 return D_WALK_CONTINUE;
1417 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1418 " still in use (%d) [unmount of %s %s]\n",
1421 dentry->d_inode->i_ino : 0UL,
1423 dentry->d_lockref.count,
1424 dentry->d_sb->s_type->name,
1425 dentry->d_sb->s_id);
1427 return D_WALK_CONTINUE;
1430 static void do_one_tree(struct dentry *dentry)
1432 shrink_dcache_parent(dentry);
1433 d_walk(dentry, dentry, umount_check, NULL);
1439 * destroy the dentries attached to a superblock on unmounting
1441 void shrink_dcache_for_umount(struct super_block *sb)
1443 struct dentry *dentry;
1445 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1447 dentry = sb->s_root;
1449 do_one_tree(dentry);
1451 while (!hlist_bl_empty(&sb->s_anon)) {
1452 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1453 do_one_tree(dentry);
1457 struct detach_data {
1458 struct select_data select;
1459 struct dentry *mountpoint;
1461 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1463 struct detach_data *data = _data;
1465 if (d_mountpoint(dentry)) {
1466 __dget_dlock(dentry);
1467 data->mountpoint = dentry;
1471 return select_collect(&data->select, dentry);
1474 static void check_and_drop(void *_data)
1476 struct detach_data *data = _data;
1478 if (!data->mountpoint && !data->select.found)
1479 __d_drop(data->select.start);
1483 * d_invalidate - detach submounts, prune dcache, and drop
1484 * @dentry: dentry to invalidate (aka detach, prune and drop)
1488 * The final d_drop is done as an atomic operation relative to
1489 * rename_lock ensuring there are no races with d_set_mounted. This
1490 * ensures there are no unhashed dentries on the path to a mountpoint.
1492 void d_invalidate(struct dentry *dentry)
1495 * If it's already been dropped, return OK.
1497 spin_lock(&dentry->d_lock);
1498 if (d_unhashed(dentry)) {
1499 spin_unlock(&dentry->d_lock);
1502 spin_unlock(&dentry->d_lock);
1504 /* Negative dentries can be dropped without further checks */
1505 if (!dentry->d_inode) {
1511 struct detach_data data;
1513 data.mountpoint = NULL;
1514 INIT_LIST_HEAD(&data.select.dispose);
1515 data.select.start = dentry;
1516 data.select.found = 0;
1518 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1520 if (data.select.found)
1521 shrink_dentry_list(&data.select.dispose);
1523 if (data.mountpoint) {
1524 detach_mounts(data.mountpoint);
1525 dput(data.mountpoint);
1528 if (!data.mountpoint && !data.select.found)
1534 EXPORT_SYMBOL(d_invalidate);
1537 * __d_alloc - allocate a dcache entry
1538 * @sb: filesystem it will belong to
1539 * @name: qstr of the name
1541 * Allocates a dentry. It returns %NULL if there is insufficient memory
1542 * available. On a success the dentry is returned. The name passed in is
1543 * copied and the copy passed in may be reused after this call.
1546 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1548 struct dentry *dentry;
1551 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1556 * We guarantee that the inline name is always NUL-terminated.
1557 * This way the memcpy() done by the name switching in rename
1558 * will still always have a NUL at the end, even if we might
1559 * be overwriting an internal NUL character
1561 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1562 if (name->len > DNAME_INLINE_LEN-1) {
1563 size_t size = offsetof(struct external_name, name[1]);
1564 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1566 kmem_cache_free(dentry_cache, dentry);
1569 atomic_set(&p->u.count, 1);
1571 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1572 kasan_unpoison_shadow(dname,
1573 round_up(name->len + 1, sizeof(unsigned long)));
1575 dname = dentry->d_iname;
1578 dentry->d_name.len = name->len;
1579 dentry->d_name.hash = name->hash;
1580 memcpy(dname, name->name, name->len);
1581 dname[name->len] = 0;
1583 /* Make sure we always see the terminating NUL character */
1585 dentry->d_name.name = dname;
1587 dentry->d_lockref.count = 1;
1588 dentry->d_flags = 0;
1589 spin_lock_init(&dentry->d_lock);
1590 seqcount_init(&dentry->d_seq);
1591 dentry->d_inode = NULL;
1592 dentry->d_parent = dentry;
1594 dentry->d_op = NULL;
1595 dentry->d_fsdata = NULL;
1596 INIT_HLIST_BL_NODE(&dentry->d_hash);
1597 INIT_LIST_HEAD(&dentry->d_lru);
1598 INIT_LIST_HEAD(&dentry->d_subdirs);
1599 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1600 INIT_LIST_HEAD(&dentry->d_child);
1601 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1603 this_cpu_inc(nr_dentry);
1609 * d_alloc - allocate a dcache entry
1610 * @parent: parent of entry to allocate
1611 * @name: qstr of the name
1613 * Allocates a dentry. It returns %NULL if there is insufficient memory
1614 * available. On a success the dentry is returned. The name passed in is
1615 * copied and the copy passed in may be reused after this call.
1617 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1619 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1623 spin_lock(&parent->d_lock);
1625 * don't need child lock because it is not subject
1626 * to concurrency here
1628 __dget_dlock(parent);
1629 dentry->d_parent = parent;
1630 list_add(&dentry->d_child, &parent->d_subdirs);
1631 spin_unlock(&parent->d_lock);
1635 EXPORT_SYMBOL(d_alloc);
1638 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1639 * @sb: the superblock
1640 * @name: qstr of the name
1642 * For a filesystem that just pins its dentries in memory and never
1643 * performs lookups at all, return an unhashed IS_ROOT dentry.
1645 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1647 return __d_alloc(sb, name);
1649 EXPORT_SYMBOL(d_alloc_pseudo);
1651 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1656 q.len = strlen(name);
1657 q.hash = full_name_hash(q.name, q.len);
1658 return d_alloc(parent, &q);
1660 EXPORT_SYMBOL(d_alloc_name);
1662 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1664 WARN_ON_ONCE(dentry->d_op);
1665 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1667 DCACHE_OP_REVALIDATE |
1668 DCACHE_OP_WEAK_REVALIDATE |
1670 DCACHE_OP_SELECT_INODE));
1675 dentry->d_flags |= DCACHE_OP_HASH;
1677 dentry->d_flags |= DCACHE_OP_COMPARE;
1678 if (op->d_revalidate)
1679 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1680 if (op->d_weak_revalidate)
1681 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1683 dentry->d_flags |= DCACHE_OP_DELETE;
1685 dentry->d_flags |= DCACHE_OP_PRUNE;
1686 if (op->d_select_inode)
1687 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
1690 EXPORT_SYMBOL(d_set_d_op);
1694 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1695 * @dentry - The dentry to mark
1697 * Mark a dentry as falling through to the lower layer (as set with
1698 * d_pin_lower()). This flag may be recorded on the medium.
1700 void d_set_fallthru(struct dentry *dentry)
1702 spin_lock(&dentry->d_lock);
1703 dentry->d_flags |= DCACHE_FALLTHRU;
1704 spin_unlock(&dentry->d_lock);
1706 EXPORT_SYMBOL(d_set_fallthru);
1708 static unsigned d_flags_for_inode(struct inode *inode)
1710 unsigned add_flags = DCACHE_REGULAR_TYPE;
1713 return DCACHE_MISS_TYPE;
1715 if (S_ISDIR(inode->i_mode)) {
1716 add_flags = DCACHE_DIRECTORY_TYPE;
1717 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1718 if (unlikely(!inode->i_op->lookup))
1719 add_flags = DCACHE_AUTODIR_TYPE;
1721 inode->i_opflags |= IOP_LOOKUP;
1723 goto type_determined;
1726 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1727 if (unlikely(inode->i_op->follow_link)) {
1728 add_flags = DCACHE_SYMLINK_TYPE;
1729 goto type_determined;
1731 inode->i_opflags |= IOP_NOFOLLOW;
1734 if (unlikely(!S_ISREG(inode->i_mode)))
1735 add_flags = DCACHE_SPECIAL_TYPE;
1738 if (unlikely(IS_AUTOMOUNT(inode)))
1739 add_flags |= DCACHE_NEED_AUTOMOUNT;
1743 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1745 unsigned add_flags = d_flags_for_inode(inode);
1747 spin_lock(&dentry->d_lock);
1749 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1750 raw_write_seqcount_begin(&dentry->d_seq);
1751 __d_set_inode_and_type(dentry, inode, add_flags);
1752 raw_write_seqcount_end(&dentry->d_seq);
1753 spin_unlock(&dentry->d_lock);
1754 fsnotify_d_instantiate(dentry, inode);
1758 * d_instantiate - fill in inode information for a dentry
1759 * @entry: dentry to complete
1760 * @inode: inode to attach to this dentry
1762 * Fill in inode information in the entry.
1764 * This turns negative dentries into productive full members
1767 * NOTE! This assumes that the inode count has been incremented
1768 * (or otherwise set) by the caller to indicate that it is now
1769 * in use by the dcache.
1772 void d_instantiate(struct dentry *entry, struct inode * inode)
1774 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1776 spin_lock(&inode->i_lock);
1777 __d_instantiate(entry, inode);
1779 spin_unlock(&inode->i_lock);
1780 security_d_instantiate(entry, inode);
1782 EXPORT_SYMBOL(d_instantiate);
1785 * d_instantiate_unique - instantiate a non-aliased dentry
1786 * @entry: dentry to instantiate
1787 * @inode: inode to attach to this dentry
1789 * Fill in inode information in the entry. On success, it returns NULL.
1790 * If an unhashed alias of "entry" already exists, then we return the
1791 * aliased dentry instead and drop one reference to inode.
1793 * Note that in order to avoid conflicts with rename() etc, the caller
1794 * had better be holding the parent directory semaphore.
1796 * This also assumes that the inode count has been incremented
1797 * (or otherwise set) by the caller to indicate that it is now
1798 * in use by the dcache.
1800 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1801 struct inode *inode)
1803 struct dentry *alias;
1804 int len = entry->d_name.len;
1805 const char *name = entry->d_name.name;
1806 unsigned int hash = entry->d_name.hash;
1809 __d_instantiate(entry, NULL);
1813 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1815 * Don't need alias->d_lock here, because aliases with
1816 * d_parent == entry->d_parent are not subject to name or
1817 * parent changes, because the parent inode i_mutex is held.
1819 if (alias->d_name.hash != hash)
1821 if (alias->d_parent != entry->d_parent)
1823 if (alias->d_name.len != len)
1825 if (dentry_cmp(alias, name, len))
1831 __d_instantiate(entry, inode);
1835 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1837 struct dentry *result;
1839 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1842 spin_lock(&inode->i_lock);
1843 result = __d_instantiate_unique(entry, inode);
1845 spin_unlock(&inode->i_lock);
1848 security_d_instantiate(entry, inode);
1852 BUG_ON(!d_unhashed(result));
1857 EXPORT_SYMBOL(d_instantiate_unique);
1860 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1861 * @entry: dentry to complete
1862 * @inode: inode to attach to this dentry
1864 * Fill in inode information in the entry. If a directory alias is found, then
1865 * return an error (and drop inode). Together with d_materialise_unique() this
1866 * guarantees that a directory inode may never have more than one alias.
1868 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1870 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1872 spin_lock(&inode->i_lock);
1873 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1874 spin_unlock(&inode->i_lock);
1878 __d_instantiate(entry, inode);
1879 spin_unlock(&inode->i_lock);
1880 security_d_instantiate(entry, inode);
1884 EXPORT_SYMBOL(d_instantiate_no_diralias);
1886 struct dentry *d_make_root(struct inode *root_inode)
1888 struct dentry *res = NULL;
1891 static const struct qstr name = QSTR_INIT("/", 1);
1893 res = __d_alloc(root_inode->i_sb, &name);
1895 d_instantiate(res, root_inode);
1901 EXPORT_SYMBOL(d_make_root);
1903 static struct dentry * __d_find_any_alias(struct inode *inode)
1905 struct dentry *alias;
1907 if (hlist_empty(&inode->i_dentry))
1909 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1915 * d_find_any_alias - find any alias for a given inode
1916 * @inode: inode to find an alias for
1918 * If any aliases exist for the given inode, take and return a
1919 * reference for one of them. If no aliases exist, return %NULL.
1921 struct dentry *d_find_any_alias(struct inode *inode)
1925 spin_lock(&inode->i_lock);
1926 de = __d_find_any_alias(inode);
1927 spin_unlock(&inode->i_lock);
1930 EXPORT_SYMBOL(d_find_any_alias);
1932 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1934 static const struct qstr anonstring = QSTR_INIT("/", 1);
1940 return ERR_PTR(-ESTALE);
1942 return ERR_CAST(inode);
1944 res = d_find_any_alias(inode);
1948 tmp = __d_alloc(inode->i_sb, &anonstring);
1950 res = ERR_PTR(-ENOMEM);
1954 spin_lock(&inode->i_lock);
1955 res = __d_find_any_alias(inode);
1957 spin_unlock(&inode->i_lock);
1962 /* attach a disconnected dentry */
1963 add_flags = d_flags_for_inode(inode);
1966 add_flags |= DCACHE_DISCONNECTED;
1968 spin_lock(&tmp->d_lock);
1969 __d_set_inode_and_type(tmp, inode, add_flags);
1970 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1971 hlist_bl_lock(&tmp->d_sb->s_anon);
1972 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1973 hlist_bl_unlock(&tmp->d_sb->s_anon);
1974 spin_unlock(&tmp->d_lock);
1975 spin_unlock(&inode->i_lock);
1976 security_d_instantiate(tmp, inode);
1981 if (res && !IS_ERR(res))
1982 security_d_instantiate(res, inode);
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;
2055 * First check if a dentry matching the name already exists,
2056 * if not go ahead and create it now.
2058 found = d_hash_and_lookup(dentry->d_parent, name);
2060 new = d_alloc(dentry->d_parent, name);
2062 found = ERR_PTR(-ENOMEM);
2064 found = d_splice_alias(inode, new);
2075 EXPORT_SYMBOL(d_add_ci);
2078 * Do the slow-case of the dentry name compare.
2080 * Unlike the dentry_cmp() function, we need to atomically
2081 * load the name and length information, so that the
2082 * filesystem can rely on them, and can use the 'name' and
2083 * 'len' information without worrying about walking off the
2084 * end of memory etc.
2086 * Thus the read_seqcount_retry() and the "duplicate" info
2087 * in arguments (the low-level filesystem should not look
2088 * at the dentry inode or name contents directly, since
2089 * rename can change them while we're in RCU mode).
2091 enum slow_d_compare {
2097 static noinline enum slow_d_compare slow_dentry_cmp(
2098 const struct dentry *parent,
2099 struct dentry *dentry,
2101 const struct qstr *name)
2103 int tlen = dentry->d_name.len;
2104 const char *tname = dentry->d_name.name;
2106 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2108 return D_COMP_SEQRETRY;
2110 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2111 return D_COMP_NOMATCH;
2116 * __d_lookup_rcu - search for a dentry (racy, store-free)
2117 * @parent: parent dentry
2118 * @name: qstr of name we wish to find
2119 * @seqp: returns d_seq value at the point where the dentry was found
2120 * Returns: dentry, or NULL
2122 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2123 * resolution (store-free path walking) design described in
2124 * Documentation/filesystems/path-lookup.txt.
2126 * This is not to be used outside core vfs.
2128 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2129 * held, and rcu_read_lock held. The returned dentry must not be stored into
2130 * without taking d_lock and checking d_seq sequence count against @seq
2133 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2136 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2137 * the returned dentry, so long as its parent's seqlock is checked after the
2138 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2139 * is formed, giving integrity down the path walk.
2141 * NOTE! The caller *has* to check the resulting dentry against the sequence
2142 * number we've returned before using any of the resulting dentry state!
2144 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2145 const struct qstr *name,
2148 u64 hashlen = name->hash_len;
2149 const unsigned char *str = name->name;
2150 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2151 struct hlist_bl_node *node;
2152 struct dentry *dentry;
2155 * Note: There is significant duplication with __d_lookup_rcu which is
2156 * required to prevent single threaded performance regressions
2157 * especially on architectures where smp_rmb (in seqcounts) are costly.
2158 * Keep the two functions in sync.
2162 * The hash list is protected using RCU.
2164 * Carefully use d_seq when comparing a candidate dentry, to avoid
2165 * races with d_move().
2167 * It is possible that concurrent renames can mess up our list
2168 * walk here and result in missing our dentry, resulting in the
2169 * false-negative result. d_lookup() protects against concurrent
2170 * renames using rename_lock seqlock.
2172 * See Documentation/filesystems/path-lookup.txt for more details.
2174 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2179 * The dentry sequence count protects us from concurrent
2180 * renames, and thus protects parent and name fields.
2182 * The caller must perform a seqcount check in order
2183 * to do anything useful with the returned dentry.
2185 * NOTE! We do a "raw" seqcount_begin here. That means that
2186 * we don't wait for the sequence count to stabilize if it
2187 * is in the middle of a sequence change. If we do the slow
2188 * dentry compare, we will do seqretries until it is stable,
2189 * and if we end up with a successful lookup, we actually
2190 * want to exit RCU lookup anyway.
2192 seq = raw_seqcount_begin(&dentry->d_seq);
2193 if (dentry->d_parent != parent)
2195 if (d_unhashed(dentry))
2198 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2199 if (dentry->d_name.hash != hashlen_hash(hashlen))
2202 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2205 case D_COMP_NOMATCH:
2212 if (dentry->d_name.hash_len != hashlen)
2215 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2222 * d_lookup - search for a dentry
2223 * @parent: parent dentry
2224 * @name: qstr of name we wish to find
2225 * Returns: dentry, or NULL
2227 * d_lookup searches the children of the parent dentry for the name in
2228 * question. If the dentry is found its reference count is incremented and the
2229 * dentry is returned. The caller must use dput to free the entry when it has
2230 * finished using it. %NULL is returned if the dentry does not exist.
2232 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2234 struct dentry *dentry;
2238 seq = read_seqbegin(&rename_lock);
2239 dentry = __d_lookup(parent, name);
2242 } while (read_seqretry(&rename_lock, seq));
2245 EXPORT_SYMBOL(d_lookup);
2248 * __d_lookup - search for a dentry (racy)
2249 * @parent: parent dentry
2250 * @name: qstr of name we wish to find
2251 * Returns: dentry, or NULL
2253 * __d_lookup is like d_lookup, however it may (rarely) return a
2254 * false-negative result due to unrelated rename activity.
2256 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2257 * however it must be used carefully, eg. with a following d_lookup in
2258 * the case of failure.
2260 * __d_lookup callers must be commented.
2262 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2264 unsigned int len = name->len;
2265 unsigned int hash = name->hash;
2266 const unsigned char *str = name->name;
2267 struct hlist_bl_head *b = d_hash(parent, hash);
2268 struct hlist_bl_node *node;
2269 struct dentry *found = NULL;
2270 struct dentry *dentry;
2273 * Note: There is significant duplication with __d_lookup_rcu which is
2274 * required to prevent single threaded performance regressions
2275 * especially on architectures where smp_rmb (in seqcounts) are costly.
2276 * Keep the two functions in sync.
2280 * The hash list is protected using RCU.
2282 * Take d_lock when comparing a candidate dentry, to avoid races
2285 * It is possible that concurrent renames can mess up our list
2286 * walk here and result in missing our dentry, resulting in the
2287 * false-negative result. d_lookup() protects against concurrent
2288 * renames using rename_lock seqlock.
2290 * See Documentation/filesystems/path-lookup.txt for more details.
2294 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2296 if (dentry->d_name.hash != hash)
2299 spin_lock(&dentry->d_lock);
2300 if (dentry->d_parent != parent)
2302 if (d_unhashed(dentry))
2306 * It is safe to compare names since d_move() cannot
2307 * change the qstr (protected by d_lock).
2309 if (parent->d_flags & DCACHE_OP_COMPARE) {
2310 int tlen = dentry->d_name.len;
2311 const char *tname = dentry->d_name.name;
2312 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2315 if (dentry->d_name.len != len)
2317 if (dentry_cmp(dentry, str, len))
2321 dentry->d_lockref.count++;
2323 spin_unlock(&dentry->d_lock);
2326 spin_unlock(&dentry->d_lock);
2334 * d_hash_and_lookup - hash the qstr then search for a dentry
2335 * @dir: Directory to search in
2336 * @name: qstr of name we wish to find
2338 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2340 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2343 * Check for a fs-specific hash function. Note that we must
2344 * calculate the standard hash first, as the d_op->d_hash()
2345 * routine may choose to leave the hash value unchanged.
2347 name->hash = full_name_hash(name->name, name->len);
2348 if (dir->d_flags & DCACHE_OP_HASH) {
2349 int err = dir->d_op->d_hash(dir, name);
2350 if (unlikely(err < 0))
2351 return ERR_PTR(err);
2353 return d_lookup(dir, name);
2355 EXPORT_SYMBOL(d_hash_and_lookup);
2358 * When a file is deleted, we have two options:
2359 * - turn this dentry into a negative dentry
2360 * - unhash this dentry and free it.
2362 * Usually, we want to just turn this into
2363 * a negative dentry, but if anybody else is
2364 * currently using the dentry or the inode
2365 * we can't do that and we fall back on removing
2366 * it from the hash queues and waiting for
2367 * it to be deleted later when it has no users
2371 * d_delete - delete a dentry
2372 * @dentry: The dentry to delete
2374 * Turn the dentry into a negative dentry if possible, otherwise
2375 * remove it from the hash queues so it can be deleted later
2378 void d_delete(struct dentry * dentry)
2380 struct inode *inode;
2383 * Are we the only user?
2386 spin_lock(&dentry->d_lock);
2387 inode = dentry->d_inode;
2388 isdir = S_ISDIR(inode->i_mode);
2389 if (dentry->d_lockref.count == 1) {
2390 if (!spin_trylock(&inode->i_lock)) {
2391 spin_unlock(&dentry->d_lock);
2395 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2396 dentry_unlink_inode(dentry);
2397 fsnotify_nameremove(dentry, isdir);
2401 if (!d_unhashed(dentry))
2404 spin_unlock(&dentry->d_lock);
2406 fsnotify_nameremove(dentry, isdir);
2408 EXPORT_SYMBOL(d_delete);
2410 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2412 BUG_ON(!d_unhashed(entry));
2414 entry->d_flags |= DCACHE_RCUACCESS;
2415 hlist_bl_add_head_rcu(&entry->d_hash, b);
2419 static void _d_rehash(struct dentry * entry)
2421 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2425 * d_rehash - add an entry back to the hash
2426 * @entry: dentry to add to the hash
2428 * Adds a dentry to the hash according to its name.
2431 void d_rehash(struct dentry * entry)
2433 spin_lock(&entry->d_lock);
2435 spin_unlock(&entry->d_lock);
2437 EXPORT_SYMBOL(d_rehash);
2440 * dentry_update_name_case - update case insensitive dentry with a new name
2441 * @dentry: dentry to be updated
2444 * Update a case insensitive dentry with new case of name.
2446 * dentry must have been returned by d_lookup with name @name. Old and new
2447 * name lengths must match (ie. no d_compare which allows mismatched name
2450 * Parent inode i_mutex must be held over d_lookup and into this call (to
2451 * keep renames and concurrent inserts, and readdir(2) away).
2453 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2455 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2456 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2458 spin_lock(&dentry->d_lock);
2459 write_seqcount_begin(&dentry->d_seq);
2460 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2461 write_seqcount_end(&dentry->d_seq);
2462 spin_unlock(&dentry->d_lock);
2464 EXPORT_SYMBOL(dentry_update_name_case);
2466 static void swap_names(struct dentry *dentry, struct dentry *target)
2468 if (unlikely(dname_external(target))) {
2469 if (unlikely(dname_external(dentry))) {
2471 * Both external: swap the pointers
2473 swap(target->d_name.name, dentry->d_name.name);
2476 * dentry:internal, target:external. Steal target's
2477 * storage and make target internal.
2479 memcpy(target->d_iname, dentry->d_name.name,
2480 dentry->d_name.len + 1);
2481 dentry->d_name.name = target->d_name.name;
2482 target->d_name.name = target->d_iname;
2485 if (unlikely(dname_external(dentry))) {
2487 * dentry:external, target:internal. Give dentry's
2488 * storage to target and make dentry internal
2490 memcpy(dentry->d_iname, target->d_name.name,
2491 target->d_name.len + 1);
2492 target->d_name.name = dentry->d_name.name;
2493 dentry->d_name.name = dentry->d_iname;
2496 * Both are internal.
2499 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2500 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2501 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2502 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2503 swap(((long *) &dentry->d_iname)[i],
2504 ((long *) &target->d_iname)[i]);
2508 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2511 static void copy_name(struct dentry *dentry, struct dentry *target)
2513 struct external_name *old_name = NULL;
2514 if (unlikely(dname_external(dentry)))
2515 old_name = external_name(dentry);
2516 if (unlikely(dname_external(target))) {
2517 atomic_inc(&external_name(target)->u.count);
2518 dentry->d_name = target->d_name;
2520 memcpy(dentry->d_iname, target->d_name.name,
2521 target->d_name.len + 1);
2522 dentry->d_name.name = dentry->d_iname;
2523 dentry->d_name.hash_len = target->d_name.hash_len;
2525 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2526 kfree_rcu(old_name, u.head);
2529 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2532 * XXXX: do we really need to take target->d_lock?
2534 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2535 spin_lock(&target->d_parent->d_lock);
2537 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2538 spin_lock(&dentry->d_parent->d_lock);
2539 spin_lock_nested(&target->d_parent->d_lock,
2540 DENTRY_D_LOCK_NESTED);
2542 spin_lock(&target->d_parent->d_lock);
2543 spin_lock_nested(&dentry->d_parent->d_lock,
2544 DENTRY_D_LOCK_NESTED);
2547 if (target < dentry) {
2548 spin_lock_nested(&target->d_lock, 2);
2549 spin_lock_nested(&dentry->d_lock, 3);
2551 spin_lock_nested(&dentry->d_lock, 2);
2552 spin_lock_nested(&target->d_lock, 3);
2556 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2558 if (target->d_parent != dentry->d_parent)
2559 spin_unlock(&dentry->d_parent->d_lock);
2560 if (target->d_parent != target)
2561 spin_unlock(&target->d_parent->d_lock);
2562 spin_unlock(&target->d_lock);
2563 spin_unlock(&dentry->d_lock);
2567 * When switching names, the actual string doesn't strictly have to
2568 * be preserved in the target - because we're dropping the target
2569 * anyway. As such, we can just do a simple memcpy() to copy over
2570 * the new name before we switch, unless we are going to rehash
2571 * it. Note that if we *do* unhash the target, we are not allowed
2572 * to rehash it without giving it a new name/hash key - whether
2573 * we swap or overwrite the names here, resulting name won't match
2574 * the reality in filesystem; it's only there for d_path() purposes.
2575 * Note that all of this is happening under rename_lock, so the
2576 * any hash lookup seeing it in the middle of manipulations will
2577 * be discarded anyway. So we do not care what happens to the hash
2581 * __d_move - move a dentry
2582 * @dentry: entry to move
2583 * @target: new dentry
2584 * @exchange: exchange the two dentries
2586 * Update the dcache to reflect the move of a file name. Negative
2587 * dcache entries should not be moved in this way. Caller must hold
2588 * rename_lock, the i_mutex of the source and target directories,
2589 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2591 static void __d_move(struct dentry *dentry, struct dentry *target,
2594 if (!dentry->d_inode)
2595 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2597 BUG_ON(d_ancestor(dentry, target));
2598 BUG_ON(d_ancestor(target, dentry));
2600 dentry_lock_for_move(dentry, target);
2602 write_seqcount_begin(&dentry->d_seq);
2603 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2605 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2608 * Move the dentry to the target hash queue. Don't bother checking
2609 * for the same hash queue because of how unlikely it is.
2612 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2615 * Unhash the target (d_delete() is not usable here). If exchanging
2616 * the two dentries, then rehash onto the other's hash queue.
2621 d_hash(dentry->d_parent, dentry->d_name.hash));
2624 /* Switch the names.. */
2626 swap_names(dentry, target);
2628 copy_name(dentry, target);
2630 /* ... and switch them in the tree */
2631 if (IS_ROOT(dentry)) {
2632 /* splicing a tree */
2633 dentry->d_parent = target->d_parent;
2634 target->d_parent = target;
2635 list_del_init(&target->d_child);
2636 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2638 /* swapping two dentries */
2639 swap(dentry->d_parent, target->d_parent);
2640 list_move(&target->d_child, &target->d_parent->d_subdirs);
2641 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2643 fsnotify_d_move(target);
2644 fsnotify_d_move(dentry);
2647 write_seqcount_end(&target->d_seq);
2648 write_seqcount_end(&dentry->d_seq);
2650 dentry_unlock_for_move(dentry, target);
2654 * d_move - move a dentry
2655 * @dentry: entry to move
2656 * @target: new dentry
2658 * Update the dcache to reflect the move of a file name. Negative
2659 * dcache entries should not be moved in this way. See the locking
2660 * requirements for __d_move.
2662 void d_move(struct dentry *dentry, struct dentry *target)
2664 write_seqlock(&rename_lock);
2665 __d_move(dentry, target, false);
2666 write_sequnlock(&rename_lock);
2668 EXPORT_SYMBOL(d_move);
2671 * d_exchange - exchange two dentries
2672 * @dentry1: first dentry
2673 * @dentry2: second dentry
2675 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2677 write_seqlock(&rename_lock);
2679 WARN_ON(!dentry1->d_inode);
2680 WARN_ON(!dentry2->d_inode);
2681 WARN_ON(IS_ROOT(dentry1));
2682 WARN_ON(IS_ROOT(dentry2));
2684 __d_move(dentry1, dentry2, true);
2686 write_sequnlock(&rename_lock);
2690 * d_ancestor - search for an ancestor
2691 * @p1: ancestor dentry
2694 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2695 * an ancestor of p2, else NULL.
2697 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2701 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2702 if (p->d_parent == p1)
2709 * This helper attempts to cope with remotely renamed directories
2711 * It assumes that the caller is already holding
2712 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2714 * Note: If ever the locking in lock_rename() changes, then please
2715 * remember to update this too...
2717 static int __d_unalias(struct inode *inode,
2718 struct dentry *dentry, struct dentry *alias)
2720 struct mutex *m1 = NULL, *m2 = NULL;
2723 /* If alias and dentry share a parent, then no extra locks required */
2724 if (alias->d_parent == dentry->d_parent)
2727 /* See lock_rename() */
2728 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2730 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2731 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2733 m2 = &alias->d_parent->d_inode->i_mutex;
2735 __d_move(alias, dentry, false);
2746 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2747 * @inode: the inode which may have a disconnected dentry
2748 * @dentry: a negative dentry which we want to point to the inode.
2750 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2751 * place of the given dentry and return it, else simply d_add the inode
2752 * to the dentry and return NULL.
2754 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2755 * we should error out: directories can't have multiple aliases.
2757 * This is needed in the lookup routine of any filesystem that is exportable
2758 * (via knfsd) so that we can build dcache paths to directories effectively.
2760 * If a dentry was found and moved, then it is returned. Otherwise NULL
2761 * is returned. This matches the expected return value of ->lookup.
2763 * Cluster filesystems may call this function with a negative, hashed dentry.
2764 * In that case, we know that the inode will be a regular file, and also this
2765 * will only occur during atomic_open. So we need to check for the dentry
2766 * being already hashed only in the final case.
2768 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2771 return ERR_CAST(inode);
2773 BUG_ON(!d_unhashed(dentry));
2776 __d_instantiate(dentry, NULL);
2779 spin_lock(&inode->i_lock);
2780 if (S_ISDIR(inode->i_mode)) {
2781 struct dentry *new = __d_find_any_alias(inode);
2782 if (unlikely(new)) {
2783 /* The reference to new ensures it remains an alias */
2784 spin_unlock(&inode->i_lock);
2785 write_seqlock(&rename_lock);
2786 if (unlikely(d_ancestor(new, dentry))) {
2787 write_sequnlock(&rename_lock);
2789 new = ERR_PTR(-ELOOP);
2790 pr_warn_ratelimited(
2791 "VFS: Lookup of '%s' in %s %s"
2792 " would have caused loop\n",
2793 dentry->d_name.name,
2794 inode->i_sb->s_type->name,
2796 } else if (!IS_ROOT(new)) {
2797 int err = __d_unalias(inode, dentry, new);
2798 write_sequnlock(&rename_lock);
2804 __d_move(new, dentry, false);
2805 write_sequnlock(&rename_lock);
2806 security_d_instantiate(new, inode);
2812 /* already taking inode->i_lock, so d_add() by hand */
2813 __d_instantiate(dentry, inode);
2814 spin_unlock(&inode->i_lock);
2816 security_d_instantiate(dentry, inode);
2820 EXPORT_SYMBOL(d_splice_alias);
2822 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2826 return -ENAMETOOLONG;
2828 memcpy(*buffer, str, namelen);
2833 * prepend_name - prepend a pathname in front of current buffer pointer
2834 * @buffer: buffer pointer
2835 * @buflen: allocated length of the buffer
2836 * @name: name string and length qstr structure
2838 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2839 * make sure that either the old or the new name pointer and length are
2840 * fetched. However, there may be mismatch between length and pointer.
2841 * The length cannot be trusted, we need to copy it byte-by-byte until
2842 * the length is reached or a null byte is found. It also prepends "/" at
2843 * the beginning of the name. The sequence number check at the caller will
2844 * retry it again when a d_move() does happen. So any garbage in the buffer
2845 * due to mismatched pointer and length will be discarded.
2847 * Data dependency barrier is needed to make sure that we see that terminating
2848 * NUL. Alpha strikes again, film at 11...
2850 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2852 const char *dname = ACCESS_ONCE(name->name);
2853 u32 dlen = ACCESS_ONCE(name->len);
2856 smp_read_barrier_depends();
2858 *buflen -= dlen + 1;
2860 return -ENAMETOOLONG;
2861 p = *buffer -= dlen + 1;
2873 * prepend_path - Prepend path string to a buffer
2874 * @path: the dentry/vfsmount to report
2875 * @root: root vfsmnt/dentry
2876 * @buffer: pointer to the end of the buffer
2877 * @buflen: pointer to buffer length
2879 * The function will first try to write out the pathname without taking any
2880 * lock other than the RCU read lock to make sure that dentries won't go away.
2881 * It only checks the sequence number of the global rename_lock as any change
2882 * in the dentry's d_seq will be preceded by changes in the rename_lock
2883 * sequence number. If the sequence number had been changed, it will restart
2884 * the whole pathname back-tracing sequence again by taking the rename_lock.
2885 * In this case, there is no need to take the RCU read lock as the recursive
2886 * parent pointer references will keep the dentry chain alive as long as no
2887 * rename operation is performed.
2889 static int prepend_path(const struct path *path,
2890 const struct path *root,
2891 char **buffer, int *buflen)
2893 struct dentry *dentry;
2894 struct vfsmount *vfsmnt;
2897 unsigned seq, m_seq = 0;
2903 read_seqbegin_or_lock(&mount_lock, &m_seq);
2910 dentry = path->dentry;
2912 mnt = real_mount(vfsmnt);
2913 read_seqbegin_or_lock(&rename_lock, &seq);
2914 while (dentry != root->dentry || vfsmnt != root->mnt) {
2915 struct dentry * parent;
2917 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2918 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2920 if (dentry != vfsmnt->mnt_root) {
2927 if (mnt != parent) {
2928 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2934 error = is_mounted(vfsmnt) ? 1 : 2;
2937 parent = dentry->d_parent;
2939 error = prepend_name(&bptr, &blen, &dentry->d_name);
2947 if (need_seqretry(&rename_lock, seq)) {
2951 done_seqretry(&rename_lock, seq);
2955 if (need_seqretry(&mount_lock, m_seq)) {
2959 done_seqretry(&mount_lock, m_seq);
2961 if (error >= 0 && bptr == *buffer) {
2963 error = -ENAMETOOLONG;
2973 * __d_path - return the path of a dentry
2974 * @path: the dentry/vfsmount to report
2975 * @root: root vfsmnt/dentry
2976 * @buf: buffer to return value in
2977 * @buflen: buffer length
2979 * Convert a dentry into an ASCII path name.
2981 * Returns a pointer into the buffer or an error code if the
2982 * path was too long.
2984 * "buflen" should be positive.
2986 * If the path is not reachable from the supplied root, return %NULL.
2988 char *__d_path(const struct path *path,
2989 const struct path *root,
2990 char *buf, int buflen)
2992 char *res = buf + buflen;
2995 prepend(&res, &buflen, "\0", 1);
2996 error = prepend_path(path, root, &res, &buflen);
2999 return ERR_PTR(error);
3005 char *d_absolute_path(const struct path *path,
3006 char *buf, int buflen)
3008 struct path root = {};
3009 char *res = buf + buflen;
3012 prepend(&res, &buflen, "\0", 1);
3013 error = prepend_path(path, &root, &res, &buflen);
3018 return ERR_PTR(error);
3023 * same as __d_path but appends "(deleted)" for unlinked files.
3025 static int path_with_deleted(const struct path *path,
3026 const struct path *root,
3027 char **buf, int *buflen)
3029 prepend(buf, buflen, "\0", 1);
3030 if (d_unlinked(path->dentry)) {
3031 int error = prepend(buf, buflen, " (deleted)", 10);
3036 return prepend_path(path, root, buf, buflen);
3039 static int prepend_unreachable(char **buffer, int *buflen)
3041 return prepend(buffer, buflen, "(unreachable)", 13);
3044 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3049 seq = read_seqcount_begin(&fs->seq);
3051 } while (read_seqcount_retry(&fs->seq, seq));
3055 * d_path - return the path of a dentry
3056 * @path: path to report
3057 * @buf: buffer to return value in
3058 * @buflen: buffer length
3060 * Convert a dentry into an ASCII path name. If the entry has been deleted
3061 * the string " (deleted)" is appended. Note that this is ambiguous.
3063 * Returns a pointer into the buffer or an error code if the path was
3064 * too long. Note: Callers should use the returned pointer, not the passed
3065 * in buffer, to use the name! The implementation often starts at an offset
3066 * into the buffer, and may leave 0 bytes at the start.
3068 * "buflen" should be positive.
3070 char *d_path(const struct path *path, char *buf, int buflen)
3072 char *res = buf + buflen;
3077 * We have various synthetic filesystems that never get mounted. On
3078 * these filesystems dentries are never used for lookup purposes, and
3079 * thus don't need to be hashed. They also don't need a name until a
3080 * user wants to identify the object in /proc/pid/fd/. The little hack
3081 * below allows us to generate a name for these objects on demand:
3083 * Some pseudo inodes are mountable. When they are mounted
3084 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3085 * and instead have d_path return the mounted path.
3087 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3088 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3089 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3092 get_fs_root_rcu(current->fs, &root);
3093 error = path_with_deleted(path, &root, &res, &buflen);
3097 res = ERR_PTR(error);
3100 EXPORT_SYMBOL(d_path);
3103 * Helper function for dentry_operations.d_dname() members
3105 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3106 const char *fmt, ...)
3112 va_start(args, fmt);
3113 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3116 if (sz > sizeof(temp) || sz > buflen)
3117 return ERR_PTR(-ENAMETOOLONG);
3119 buffer += buflen - sz;
3120 return memcpy(buffer, temp, sz);
3123 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3125 char *end = buffer + buflen;
3126 /* these dentries are never renamed, so d_lock is not needed */
3127 if (prepend(&end, &buflen, " (deleted)", 11) ||
3128 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3129 prepend(&end, &buflen, "/", 1))
3130 end = ERR_PTR(-ENAMETOOLONG);
3133 EXPORT_SYMBOL(simple_dname);
3136 * Write full pathname from the root of the filesystem into the buffer.
3138 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3140 struct dentry *dentry;
3153 prepend(&end, &len, "\0", 1);
3157 read_seqbegin_or_lock(&rename_lock, &seq);
3158 while (!IS_ROOT(dentry)) {
3159 struct dentry *parent = dentry->d_parent;
3162 error = prepend_name(&end, &len, &dentry->d_name);
3171 if (need_seqretry(&rename_lock, seq)) {
3175 done_seqretry(&rename_lock, seq);
3180 return ERR_PTR(-ENAMETOOLONG);
3183 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3185 return __dentry_path(dentry, buf, buflen);
3187 EXPORT_SYMBOL(dentry_path_raw);
3189 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3194 if (d_unlinked(dentry)) {
3196 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3200 retval = __dentry_path(dentry, buf, buflen);
3201 if (!IS_ERR(retval) && p)
3202 *p = '/'; /* restore '/' overriden with '\0' */
3205 return ERR_PTR(-ENAMETOOLONG);
3208 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3214 seq = read_seqcount_begin(&fs->seq);
3217 } while (read_seqcount_retry(&fs->seq, seq));
3221 * NOTE! The user-level library version returns a
3222 * character pointer. The kernel system call just
3223 * returns the length of the buffer filled (which
3224 * includes the ending '\0' character), or a negative
3225 * error value. So libc would do something like
3227 * char *getcwd(char * buf, size_t size)
3231 * retval = sys_getcwd(buf, size);
3238 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3241 struct path pwd, root;
3242 char *page = __getname();
3248 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3251 if (!d_unlinked(pwd.dentry)) {
3253 char *cwd = page + PATH_MAX;
3254 int buflen = PATH_MAX;
3256 prepend(&cwd, &buflen, "\0", 1);
3257 error = prepend_path(&pwd, &root, &cwd, &buflen);
3263 /* Unreachable from current root */
3265 error = prepend_unreachable(&cwd, &buflen);
3271 len = PATH_MAX + page - cwd;
3274 if (copy_to_user(buf, cwd, len))
3287 * Test whether new_dentry is a subdirectory of old_dentry.
3289 * Trivially implemented using the dcache structure
3293 * is_subdir - is new dentry a subdirectory of old_dentry
3294 * @new_dentry: new dentry
3295 * @old_dentry: old dentry
3297 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3298 * Returns 0 otherwise.
3299 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3302 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3307 if (new_dentry == old_dentry)
3311 /* for restarting inner loop in case of seq retry */
3312 seq = read_seqbegin(&rename_lock);
3314 * Need rcu_readlock to protect against the d_parent trashing
3318 if (d_ancestor(old_dentry, new_dentry))
3323 } while (read_seqretry(&rename_lock, seq));
3328 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3330 struct dentry *root = data;
3331 if (dentry != root) {
3332 if (d_unhashed(dentry) || !dentry->d_inode)
3335 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3336 dentry->d_flags |= DCACHE_GENOCIDE;
3337 dentry->d_lockref.count--;
3340 return D_WALK_CONTINUE;
3343 void d_genocide(struct dentry *parent)
3345 d_walk(parent, parent, d_genocide_kill, NULL);
3348 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3350 inode_dec_link_count(inode);
3351 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3352 !hlist_unhashed(&dentry->d_u.d_alias) ||
3353 !d_unlinked(dentry));
3354 spin_lock(&dentry->d_parent->d_lock);
3355 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3356 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3357 (unsigned long long)inode->i_ino);
3358 spin_unlock(&dentry->d_lock);
3359 spin_unlock(&dentry->d_parent->d_lock);
3360 d_instantiate(dentry, inode);
3362 EXPORT_SYMBOL(d_tmpfile);
3364 static __initdata unsigned long dhash_entries;
3365 static int __init set_dhash_entries(char *str)
3369 dhash_entries = simple_strtoul(str, &str, 0);
3372 __setup("dhash_entries=", set_dhash_entries);
3374 static void __init dcache_init_early(void)
3378 /* If hashes are distributed across NUMA nodes, defer
3379 * hash allocation until vmalloc space is available.
3385 alloc_large_system_hash("Dentry cache",
3386 sizeof(struct hlist_bl_head),
3395 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3396 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3399 static void __init dcache_init(void)
3404 * A constructor could be added for stable state like the lists,
3405 * but it is probably not worth it because of the cache nature
3408 dentry_cache = KMEM_CACHE(dentry,
3409 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3411 /* Hash may have been set up in dcache_init_early */
3416 alloc_large_system_hash("Dentry cache",
3417 sizeof(struct hlist_bl_head),
3426 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3427 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3430 /* SLAB cache for __getname() consumers */
3431 struct kmem_cache *names_cachep __read_mostly;
3432 EXPORT_SYMBOL(names_cachep);
3434 EXPORT_SYMBOL(d_genocide);
3436 void __init vfs_caches_init_early(void)
3438 dcache_init_early();
3442 void __init vfs_caches_init(void)
3444 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3445 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3450 files_maxfiles_init();