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 <asm/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/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
41 #include <linux/kasan.h>
48 * dcache->d_inode->i_lock protects:
49 * - i_dentry, d_u.d_alias, d_inode of aliases
50 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
54 * dentry->d_sb->s_dentry_lru_lock protects:
55 * - the dcache lru lists and counters
62 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
64 * - d_u.d_alias, d_inode
67 * dentry->d_inode->i_lock
69 * dentry->d_sb->s_dentry_lru_lock
70 * dcache_hash_bucket lock
73 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
76 * dentry->d_parent->d_lock
79 * If no ancestor relationship:
80 * if (dentry1 < dentry2)
84 int sysctl_vfs_cache_pressure __read_mostly = 100;
85 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
87 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
89 EXPORT_SYMBOL(rename_lock);
91 static struct kmem_cache *dentry_cache __read_mostly;
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
102 static unsigned int d_hash_mask __read_mostly;
103 static unsigned int d_hash_shift __read_mostly;
105 static struct hlist_bl_head *dentry_hashtable __read_mostly;
107 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 return dentry_hashtable + hash_32(hash, d_hash_shift);
114 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat = {
119 static DEFINE_PER_CPU(long, nr_dentry);
120 static DEFINE_PER_CPU(long, nr_dentry_unused);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
125 * Here we resort to our own counters instead of using generic per-cpu counters
126 * for consistency with what the vfs inode code does. We are expected to harvest
127 * better code and performance by having our own specialized counters.
129 * Please note that the loop is done over all possible CPUs, not over all online
130 * CPUs. The reason for this is that we don't want to play games with CPUs going
131 * on and off. If one of them goes off, we will just keep their counters.
133 * glommer: See cffbc8a for details, and if you ever intend to change this,
134 * please update all vfs counters to match.
136 static long get_nr_dentry(void)
140 for_each_possible_cpu(i)
141 sum += per_cpu(nr_dentry, i);
142 return sum < 0 ? 0 : sum;
145 static long get_nr_dentry_unused(void)
149 for_each_possible_cpu(i)
150 sum += per_cpu(nr_dentry_unused, i);
151 return sum < 0 ? 0 : sum;
154 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
155 size_t *lenp, loff_t *ppos)
157 dentry_stat.nr_dentry = get_nr_dentry();
158 dentry_stat.nr_unused = get_nr_dentry_unused();
159 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
164 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
165 * The strings are both count bytes long, and count is non-zero.
167 #ifdef CONFIG_DCACHE_WORD_ACCESS
169 #include <asm/word-at-a-time.h>
171 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
172 * aligned allocation for this particular component. We don't
173 * strictly need the load_unaligned_zeropad() safety, but it
174 * doesn't hurt either.
176 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
177 * need the careful unaligned handling.
179 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
181 unsigned long a,b,mask;
184 a = *(unsigned long *)cs;
185 b = load_unaligned_zeropad(ct);
186 if (tcount < sizeof(unsigned long))
188 if (unlikely(a != b))
190 cs += sizeof(unsigned long);
191 ct += sizeof(unsigned long);
192 tcount -= sizeof(unsigned long);
196 mask = bytemask_from_count(tcount);
197 return unlikely(!!((a ^ b) & mask));
202 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
216 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
218 const unsigned char *cs;
220 * Be careful about RCU walk racing with rename:
221 * use ACCESS_ONCE to fetch the name pointer.
223 * NOTE! Even if a rename will mean that the length
224 * was not loaded atomically, we don't care. The
225 * RCU walk will check the sequence count eventually,
226 * and catch it. And we won't overrun the buffer,
227 * because we're reading the name pointer atomically,
228 * and a dentry name is guaranteed to be properly
229 * terminated with a NUL byte.
231 * End result: even if 'len' is wrong, we'll exit
232 * early because the data cannot match (there can
233 * be no NUL in the ct/tcount data)
235 cs = ACCESS_ONCE(dentry->d_name.name);
236 smp_read_barrier_depends();
237 return dentry_string_cmp(cs, ct, tcount);
240 struct external_name {
243 struct rcu_head head;
245 unsigned char name[];
248 static inline struct external_name *external_name(struct dentry *dentry)
250 return container_of(dentry->d_name.name, struct external_name, name[0]);
253 static void __d_free(struct rcu_head *head)
255 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
257 kmem_cache_free(dentry_cache, dentry);
260 static void __d_free_external(struct rcu_head *head)
262 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
263 kfree(external_name(dentry));
264 kmem_cache_free(dentry_cache, dentry);
267 static inline int dname_external(const struct dentry *dentry)
269 return dentry->d_name.name != dentry->d_iname;
272 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
274 spin_lock(&dentry->d_lock);
275 if (unlikely(dname_external(dentry))) {
276 struct external_name *p = external_name(dentry);
277 atomic_inc(&p->u.count);
278 spin_unlock(&dentry->d_lock);
279 name->name = p->name;
281 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
282 spin_unlock(&dentry->d_lock);
283 name->name = name->inline_name;
286 EXPORT_SYMBOL(take_dentry_name_snapshot);
288 void release_dentry_name_snapshot(struct name_snapshot *name)
290 if (unlikely(name->name != name->inline_name)) {
291 struct external_name *p;
292 p = container_of(name->name, struct external_name, name[0]);
293 if (unlikely(atomic_dec_and_test(&p->u.count)))
294 kfree_rcu(p, u.head);
297 EXPORT_SYMBOL(release_dentry_name_snapshot);
299 static inline void __d_set_inode_and_type(struct dentry *dentry,
305 dentry->d_inode = inode;
306 flags = READ_ONCE(dentry->d_flags);
307 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
309 WRITE_ONCE(dentry->d_flags, flags);
312 static inline void __d_clear_type_and_inode(struct dentry *dentry)
314 unsigned flags = READ_ONCE(dentry->d_flags);
316 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
317 WRITE_ONCE(dentry->d_flags, flags);
318 dentry->d_inode = NULL;
321 static void dentry_free(struct dentry *dentry)
323 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
324 if (unlikely(dname_external(dentry))) {
325 struct external_name *p = external_name(dentry);
326 if (likely(atomic_dec_and_test(&p->u.count))) {
327 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
331 /* if dentry was never visible to RCU, immediate free is OK */
332 if (!(dentry->d_flags & DCACHE_RCUACCESS))
333 __d_free(&dentry->d_u.d_rcu);
335 call_rcu(&dentry->d_u.d_rcu, __d_free);
339 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
340 * @dentry: the target dentry
341 * After this call, in-progress rcu-walk path lookup will fail. This
342 * should be called after unhashing, and after changing d_inode (if
343 * the dentry has not already been unhashed).
345 static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
347 lockdep_assert_held(&dentry->d_lock);
348 /* Go through am invalidation barrier */
349 write_seqcount_invalidate(&dentry->d_seq);
353 * Release the dentry's inode, using the filesystem
354 * d_iput() operation if defined. Dentry has no refcount
357 static void dentry_iput(struct dentry * dentry)
358 __releases(dentry->d_lock)
359 __releases(dentry->d_inode->i_lock)
361 struct inode *inode = dentry->d_inode;
363 __d_clear_type_and_inode(dentry);
364 hlist_del_init(&dentry->d_u.d_alias);
365 spin_unlock(&dentry->d_lock);
366 spin_unlock(&inode->i_lock);
368 fsnotify_inoderemove(inode);
369 if (dentry->d_op && dentry->d_op->d_iput)
370 dentry->d_op->d_iput(dentry, inode);
374 spin_unlock(&dentry->d_lock);
379 * Release the dentry's inode, using the filesystem
380 * d_iput() operation if defined. dentry remains in-use.
382 static void dentry_unlink_inode(struct dentry * dentry)
383 __releases(dentry->d_lock)
384 __releases(dentry->d_inode->i_lock)
386 struct inode *inode = dentry->d_inode;
388 raw_write_seqcount_begin(&dentry->d_seq);
389 __d_clear_type_and_inode(dentry);
390 hlist_del_init(&dentry->d_u.d_alias);
391 raw_write_seqcount_end(&dentry->d_seq);
392 spin_unlock(&dentry->d_lock);
393 spin_unlock(&inode->i_lock);
395 fsnotify_inoderemove(inode);
396 if (dentry->d_op && dentry->d_op->d_iput)
397 dentry->d_op->d_iput(dentry, inode);
403 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
404 * is in use - which includes both the "real" per-superblock
405 * LRU list _and_ the DCACHE_SHRINK_LIST use.
407 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
408 * on the shrink list (ie not on the superblock LRU list).
410 * The per-cpu "nr_dentry_unused" counters are updated with
411 * the DCACHE_LRU_LIST bit.
413 * These helper functions make sure we always follow the
414 * rules. d_lock must be held by the caller.
416 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
417 static void d_lru_add(struct dentry *dentry)
419 D_FLAG_VERIFY(dentry, 0);
420 dentry->d_flags |= DCACHE_LRU_LIST;
421 this_cpu_inc(nr_dentry_unused);
422 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
425 static void d_lru_del(struct dentry *dentry)
427 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
428 dentry->d_flags &= ~DCACHE_LRU_LIST;
429 this_cpu_dec(nr_dentry_unused);
430 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
433 static void d_shrink_del(struct dentry *dentry)
435 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
436 list_del_init(&dentry->d_lru);
437 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
438 this_cpu_dec(nr_dentry_unused);
441 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
443 D_FLAG_VERIFY(dentry, 0);
444 list_add(&dentry->d_lru, list);
445 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
446 this_cpu_inc(nr_dentry_unused);
450 * These can only be called under the global LRU lock, ie during the
451 * callback for freeing the LRU list. "isolate" removes it from the
452 * LRU lists entirely, while shrink_move moves it to the indicated
455 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
457 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
458 dentry->d_flags &= ~DCACHE_LRU_LIST;
459 this_cpu_dec(nr_dentry_unused);
460 list_lru_isolate(lru, &dentry->d_lru);
463 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
464 struct list_head *list)
466 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
467 dentry->d_flags |= DCACHE_SHRINK_LIST;
468 list_lru_isolate_move(lru, &dentry->d_lru, list);
472 * dentry_lru_(add|del)_list) must be called with d_lock held.
474 static void dentry_lru_add(struct dentry *dentry)
476 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
481 * d_drop - drop a dentry
482 * @dentry: dentry to drop
484 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
485 * be found through a VFS lookup any more. Note that this is different from
486 * deleting the dentry - d_delete will try to mark the dentry negative if
487 * possible, giving a successful _negative_ lookup, while d_drop will
488 * just make the cache lookup fail.
490 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
491 * reason (NFS timeouts or autofs deletes).
493 * __d_drop requires dentry->d_lock.
495 void __d_drop(struct dentry *dentry)
497 if (!d_unhashed(dentry)) {
498 struct hlist_bl_head *b;
500 * Hashed dentries are normally on the dentry hashtable,
501 * with the exception of those newly allocated by
502 * d_obtain_alias, which are always IS_ROOT:
504 if (unlikely(IS_ROOT(dentry)))
505 b = &dentry->d_sb->s_anon;
507 b = d_hash(dentry->d_parent, dentry->d_name.hash);
510 __hlist_bl_del(&dentry->d_hash);
511 dentry->d_hash.pprev = NULL;
513 dentry_rcuwalk_invalidate(dentry);
516 EXPORT_SYMBOL(__d_drop);
518 void d_drop(struct dentry *dentry)
520 spin_lock(&dentry->d_lock);
522 spin_unlock(&dentry->d_lock);
524 EXPORT_SYMBOL(d_drop);
526 static void __dentry_kill(struct dentry *dentry)
528 struct dentry *parent = NULL;
529 bool can_free = true;
530 if (!IS_ROOT(dentry))
531 parent = dentry->d_parent;
534 * The dentry is now unrecoverably dead to the world.
536 lockref_mark_dead(&dentry->d_lockref);
539 * inform the fs via d_prune that this dentry is about to be
540 * unhashed and destroyed.
542 if (dentry->d_flags & DCACHE_OP_PRUNE)
543 dentry->d_op->d_prune(dentry);
545 if (dentry->d_flags & DCACHE_LRU_LIST) {
546 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
549 /* if it was on the hash then remove it */
551 __list_del_entry(&dentry->d_child);
553 * Inform d_walk() that we are no longer attached to the
556 dentry->d_flags |= DCACHE_DENTRY_KILLED;
558 spin_unlock(&parent->d_lock);
561 * dentry_iput drops the locks, at which point nobody (except
562 * transient RCU lookups) can reach this dentry.
564 BUG_ON(dentry->d_lockref.count > 0);
565 this_cpu_dec(nr_dentry);
566 if (dentry->d_op && dentry->d_op->d_release)
567 dentry->d_op->d_release(dentry);
569 spin_lock(&dentry->d_lock);
570 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
571 dentry->d_flags |= DCACHE_MAY_FREE;
574 spin_unlock(&dentry->d_lock);
575 if (likely(can_free))
580 * Finish off a dentry we've decided to kill.
581 * dentry->d_lock must be held, returns with it unlocked.
582 * If ref is non-zero, then decrement the refcount too.
583 * Returns dentry requiring refcount drop, or NULL if we're done.
585 static struct dentry *dentry_kill(struct dentry *dentry)
586 __releases(dentry->d_lock)
588 struct inode *inode = dentry->d_inode;
589 struct dentry *parent = NULL;
591 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
594 if (!IS_ROOT(dentry)) {
595 parent = dentry->d_parent;
596 if (unlikely(!spin_trylock(&parent->d_lock))) {
598 spin_unlock(&inode->i_lock);
603 __dentry_kill(dentry);
607 spin_unlock(&dentry->d_lock);
608 return dentry; /* try again with same dentry */
611 static inline struct dentry *lock_parent(struct dentry *dentry)
613 struct dentry *parent = dentry->d_parent;
616 if (unlikely(dentry->d_lockref.count < 0))
618 if (likely(spin_trylock(&parent->d_lock)))
621 spin_unlock(&dentry->d_lock);
623 parent = ACCESS_ONCE(dentry->d_parent);
624 spin_lock(&parent->d_lock);
626 * We can't blindly lock dentry until we are sure
627 * that we won't violate the locking order.
628 * Any changes of dentry->d_parent must have
629 * been done with parent->d_lock held, so
630 * spin_lock() above is enough of a barrier
631 * for checking if it's still our child.
633 if (unlikely(parent != dentry->d_parent)) {
634 spin_unlock(&parent->d_lock);
637 if (parent != dentry) {
638 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
639 if (unlikely(dentry->d_lockref.count < 0)) {
640 spin_unlock(&parent->d_lock);
651 * Try to do a lockless dput(), and return whether that was successful.
653 * If unsuccessful, we return false, having already taken the dentry lock.
655 * The caller needs to hold the RCU read lock, so that the dentry is
656 * guaranteed to stay around even if the refcount goes down to zero!
658 static inline bool fast_dput(struct dentry *dentry)
661 unsigned int d_flags;
664 * If we have a d_op->d_delete() operation, we sould not
665 * let the dentry count go to zero, so use "put_or_lock".
667 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
668 return lockref_put_or_lock(&dentry->d_lockref);
671 * .. otherwise, we can try to just decrement the
672 * lockref optimistically.
674 ret = lockref_put_return(&dentry->d_lockref);
677 * If the lockref_put_return() failed due to the lock being held
678 * by somebody else, the fast path has failed. We will need to
679 * get the lock, and then check the count again.
681 if (unlikely(ret < 0)) {
682 spin_lock(&dentry->d_lock);
683 if (dentry->d_lockref.count > 1) {
684 dentry->d_lockref.count--;
685 spin_unlock(&dentry->d_lock);
692 * If we weren't the last ref, we're done.
698 * Careful, careful. The reference count went down
699 * to zero, but we don't hold the dentry lock, so
700 * somebody else could get it again, and do another
701 * dput(), and we need to not race with that.
703 * However, there is a very special and common case
704 * where we don't care, because there is nothing to
705 * do: the dentry is still hashed, it does not have
706 * a 'delete' op, and it's referenced and already on
709 * NOTE! Since we aren't locked, these values are
710 * not "stable". However, it is sufficient that at
711 * some point after we dropped the reference the
712 * dentry was hashed and the flags had the proper
713 * value. Other dentry users may have re-gotten
714 * a reference to the dentry and change that, but
715 * our work is done - we can leave the dentry
716 * around with a zero refcount.
719 d_flags = ACCESS_ONCE(dentry->d_flags);
720 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
722 /* Nothing to do? Dropping the reference was all we needed? */
723 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
727 * Not the fast normal case? Get the lock. We've already decremented
728 * the refcount, but we'll need to re-check the situation after
731 spin_lock(&dentry->d_lock);
734 * Did somebody else grab a reference to it in the meantime, and
735 * we're no longer the last user after all? Alternatively, somebody
736 * else could have killed it and marked it dead. Either way, we
737 * don't need to do anything else.
739 if (dentry->d_lockref.count) {
740 spin_unlock(&dentry->d_lock);
745 * Re-get the reference we optimistically dropped. We hold the
746 * lock, and we just tested that it was zero, so we can just
749 dentry->d_lockref.count = 1;
757 * This is complicated by the fact that we do not want to put
758 * dentries that are no longer on any hash chain on the unused
759 * list: we'd much rather just get rid of them immediately.
761 * However, that implies that we have to traverse the dentry
762 * tree upwards to the parents which might _also_ now be
763 * scheduled for deletion (it may have been only waiting for
764 * its last child to go away).
766 * This tail recursion is done by hand as we don't want to depend
767 * on the compiler to always get this right (gcc generally doesn't).
768 * Real recursion would eat up our stack space.
772 * dput - release a dentry
773 * @dentry: dentry to release
775 * Release a dentry. This will drop the usage count and if appropriate
776 * call the dentry unlink method as well as removing it from the queues and
777 * releasing its resources. If the parent dentries were scheduled for release
778 * they too may now get deleted.
780 void dput(struct dentry *dentry)
782 if (unlikely(!dentry))
789 if (likely(fast_dput(dentry))) {
794 /* Slow case: now with the dentry lock held */
797 /* Unreachable? Get rid of it */
798 if (unlikely(d_unhashed(dentry)))
801 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
804 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
805 if (dentry->d_op->d_delete(dentry))
809 if (!(dentry->d_flags & DCACHE_REFERENCED))
810 dentry->d_flags |= DCACHE_REFERENCED;
811 dentry_lru_add(dentry);
813 dentry->d_lockref.count--;
814 spin_unlock(&dentry->d_lock);
818 dentry = dentry_kill(dentry);
827 /* This must be called with d_lock held */
828 static inline void __dget_dlock(struct dentry *dentry)
830 dentry->d_lockref.count++;
833 static inline void __dget(struct dentry *dentry)
835 lockref_get(&dentry->d_lockref);
838 struct dentry *dget_parent(struct dentry *dentry)
844 * Do optimistic parent lookup without any
848 ret = ACCESS_ONCE(dentry->d_parent);
849 gotref = lockref_get_not_zero(&ret->d_lockref);
851 if (likely(gotref)) {
852 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
859 * Don't need rcu_dereference because we re-check it was correct under
863 ret = dentry->d_parent;
864 spin_lock(&ret->d_lock);
865 if (unlikely(ret != dentry->d_parent)) {
866 spin_unlock(&ret->d_lock);
871 BUG_ON(!ret->d_lockref.count);
872 ret->d_lockref.count++;
873 spin_unlock(&ret->d_lock);
876 EXPORT_SYMBOL(dget_parent);
879 * d_find_alias - grab a hashed alias of inode
880 * @inode: inode in question
882 * If inode has a hashed alias, or is a directory and has any alias,
883 * acquire the reference to alias and return it. Otherwise return NULL.
884 * Notice that if inode is a directory there can be only one alias and
885 * it can be unhashed only if it has no children, or if it is the root
886 * of a filesystem, or if the directory was renamed and d_revalidate
887 * was the first vfs operation to notice.
889 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
890 * any other hashed alias over that one.
892 static struct dentry *__d_find_alias(struct inode *inode)
894 struct dentry *alias, *discon_alias;
898 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
899 spin_lock(&alias->d_lock);
900 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
901 if (IS_ROOT(alias) &&
902 (alias->d_flags & DCACHE_DISCONNECTED)) {
903 discon_alias = alias;
906 spin_unlock(&alias->d_lock);
910 spin_unlock(&alias->d_lock);
913 alias = discon_alias;
914 spin_lock(&alias->d_lock);
915 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
917 spin_unlock(&alias->d_lock);
920 spin_unlock(&alias->d_lock);
926 struct dentry *d_find_alias(struct inode *inode)
928 struct dentry *de = NULL;
930 if (!hlist_empty(&inode->i_dentry)) {
931 spin_lock(&inode->i_lock);
932 de = __d_find_alias(inode);
933 spin_unlock(&inode->i_lock);
937 EXPORT_SYMBOL(d_find_alias);
940 * Try to kill dentries associated with this inode.
941 * WARNING: you must own a reference to inode.
943 void d_prune_aliases(struct inode *inode)
945 struct dentry *dentry;
947 spin_lock(&inode->i_lock);
948 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
949 spin_lock(&dentry->d_lock);
950 if (!dentry->d_lockref.count) {
951 struct dentry *parent = lock_parent(dentry);
952 if (likely(!dentry->d_lockref.count)) {
953 __dentry_kill(dentry);
958 spin_unlock(&parent->d_lock);
960 spin_unlock(&dentry->d_lock);
962 spin_unlock(&inode->i_lock);
964 EXPORT_SYMBOL(d_prune_aliases);
966 static void shrink_dentry_list(struct list_head *list)
968 struct dentry *dentry, *parent;
970 while (!list_empty(list)) {
972 dentry = list_entry(list->prev, struct dentry, d_lru);
973 spin_lock(&dentry->d_lock);
974 parent = lock_parent(dentry);
977 * The dispose list is isolated and dentries are not accounted
978 * to the LRU here, so we can simply remove it from the list
979 * here regardless of whether it is referenced or not.
981 d_shrink_del(dentry);
984 * We found an inuse dentry which was not removed from
985 * the LRU because of laziness during lookup. Do not free it.
987 if (dentry->d_lockref.count > 0) {
988 spin_unlock(&dentry->d_lock);
990 spin_unlock(&parent->d_lock);
995 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
996 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
997 spin_unlock(&dentry->d_lock);
999 spin_unlock(&parent->d_lock);
1001 dentry_free(dentry);
1005 inode = dentry->d_inode;
1006 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1007 d_shrink_add(dentry, list);
1008 spin_unlock(&dentry->d_lock);
1010 spin_unlock(&parent->d_lock);
1014 __dentry_kill(dentry);
1017 * We need to prune ancestors too. This is necessary to prevent
1018 * quadratic behavior of shrink_dcache_parent(), but is also
1019 * expected to be beneficial in reducing dentry cache
1023 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1024 parent = lock_parent(dentry);
1025 if (dentry->d_lockref.count != 1) {
1026 dentry->d_lockref.count--;
1027 spin_unlock(&dentry->d_lock);
1029 spin_unlock(&parent->d_lock);
1032 inode = dentry->d_inode; /* can't be NULL */
1033 if (unlikely(!spin_trylock(&inode->i_lock))) {
1034 spin_unlock(&dentry->d_lock);
1036 spin_unlock(&parent->d_lock);
1040 __dentry_kill(dentry);
1046 static enum lru_status dentry_lru_isolate(struct list_head *item,
1047 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1049 struct list_head *freeable = arg;
1050 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1054 * we are inverting the lru lock/dentry->d_lock here,
1055 * so use a trylock. If we fail to get the lock, just skip
1058 if (!spin_trylock(&dentry->d_lock))
1062 * Referenced dentries are still in use. If they have active
1063 * counts, just remove them from the LRU. Otherwise give them
1064 * another pass through the LRU.
1066 if (dentry->d_lockref.count) {
1067 d_lru_isolate(lru, dentry);
1068 spin_unlock(&dentry->d_lock);
1072 if (dentry->d_flags & DCACHE_REFERENCED) {
1073 dentry->d_flags &= ~DCACHE_REFERENCED;
1074 spin_unlock(&dentry->d_lock);
1077 * The list move itself will be made by the common LRU code. At
1078 * this point, we've dropped the dentry->d_lock but keep the
1079 * lru lock. This is safe to do, since every list movement is
1080 * protected by the lru lock even if both locks are held.
1082 * This is guaranteed by the fact that all LRU management
1083 * functions are intermediated by the LRU API calls like
1084 * list_lru_add and list_lru_del. List movement in this file
1085 * only ever occur through this functions or through callbacks
1086 * like this one, that are called from the LRU API.
1088 * The only exceptions to this are functions like
1089 * shrink_dentry_list, and code that first checks for the
1090 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1091 * operating only with stack provided lists after they are
1092 * properly isolated from the main list. It is thus, always a
1098 d_lru_shrink_move(lru, dentry, freeable);
1099 spin_unlock(&dentry->d_lock);
1105 * prune_dcache_sb - shrink the dcache
1107 * @sc: shrink control, passed to list_lru_shrink_walk()
1109 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1110 * is done when we need more memory and called from the superblock shrinker
1113 * This function may fail to free any resources if all the dentries are in
1116 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1121 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1122 dentry_lru_isolate, &dispose);
1123 shrink_dentry_list(&dispose);
1127 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1128 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1130 struct list_head *freeable = arg;
1131 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1134 * we are inverting the lru lock/dentry->d_lock here,
1135 * so use a trylock. If we fail to get the lock, just skip
1138 if (!spin_trylock(&dentry->d_lock))
1141 d_lru_shrink_move(lru, dentry, freeable);
1142 spin_unlock(&dentry->d_lock);
1149 * shrink_dcache_sb - shrink dcache for a superblock
1152 * Shrink the dcache for the specified super block. This is used to free
1153 * the dcache before unmounting a file system.
1155 void shrink_dcache_sb(struct super_block *sb)
1162 freed = list_lru_walk(&sb->s_dentry_lru,
1163 dentry_lru_isolate_shrink, &dispose, 1024);
1165 this_cpu_sub(nr_dentry_unused, freed);
1166 shrink_dentry_list(&dispose);
1168 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1170 EXPORT_SYMBOL(shrink_dcache_sb);
1173 * enum d_walk_ret - action to talke during tree walk
1174 * @D_WALK_CONTINUE: contrinue walk
1175 * @D_WALK_QUIT: quit walk
1176 * @D_WALK_NORETRY: quit when retry is needed
1177 * @D_WALK_SKIP: skip this dentry and its children
1187 * d_walk - walk the dentry tree
1188 * @parent: start of walk
1189 * @data: data passed to @enter() and @finish()
1190 * @enter: callback when first entering the dentry
1191 * @finish: callback when successfully finished the walk
1193 * The @enter() and @finish() callbacks are called with d_lock held.
1195 static void d_walk(struct dentry *parent, void *data,
1196 enum d_walk_ret (*enter)(void *, struct dentry *),
1197 void (*finish)(void *))
1199 struct dentry *this_parent;
1200 struct list_head *next;
1202 enum d_walk_ret ret;
1206 read_seqbegin_or_lock(&rename_lock, &seq);
1207 this_parent = parent;
1208 spin_lock(&this_parent->d_lock);
1210 ret = enter(data, this_parent);
1212 case D_WALK_CONTINUE:
1217 case D_WALK_NORETRY:
1222 next = this_parent->d_subdirs.next;
1224 while (next != &this_parent->d_subdirs) {
1225 struct list_head *tmp = next;
1226 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1229 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1231 ret = enter(data, dentry);
1233 case D_WALK_CONTINUE:
1236 spin_unlock(&dentry->d_lock);
1238 case D_WALK_NORETRY:
1242 spin_unlock(&dentry->d_lock);
1246 if (!list_empty(&dentry->d_subdirs)) {
1247 spin_unlock(&this_parent->d_lock);
1248 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1249 this_parent = dentry;
1250 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1253 spin_unlock(&dentry->d_lock);
1256 * All done at this level ... ascend and resume the search.
1260 if (this_parent != parent) {
1261 struct dentry *child = this_parent;
1262 this_parent = child->d_parent;
1264 spin_unlock(&child->d_lock);
1265 spin_lock(&this_parent->d_lock);
1267 /* might go back up the wrong parent if we have had a rename. */
1268 if (need_seqretry(&rename_lock, seq))
1270 /* go into the first sibling still alive */
1272 next = child->d_child.next;
1273 if (next == &this_parent->d_subdirs)
1275 child = list_entry(next, struct dentry, d_child);
1276 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1280 if (need_seqretry(&rename_lock, seq))
1287 spin_unlock(&this_parent->d_lock);
1288 done_seqretry(&rename_lock, seq);
1292 spin_unlock(&this_parent->d_lock);
1302 * Search for at least 1 mount point in the dentry's subdirs.
1303 * We descend to the next level whenever the d_subdirs
1304 * list is non-empty and continue searching.
1307 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1310 if (d_mountpoint(dentry)) {
1314 return D_WALK_CONTINUE;
1318 * have_submounts - check for mounts over a dentry
1319 * @parent: dentry to check.
1321 * Return true if the parent or its subdirectories contain
1324 int have_submounts(struct dentry *parent)
1328 d_walk(parent, &ret, check_mount, NULL);
1332 EXPORT_SYMBOL(have_submounts);
1335 * Called by mount code to set a mountpoint and check if the mountpoint is
1336 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1337 * subtree can become unreachable).
1339 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1340 * this reason take rename_lock and d_lock on dentry and ancestors.
1342 int d_set_mounted(struct dentry *dentry)
1346 write_seqlock(&rename_lock);
1347 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1348 /* Need exclusion wrt. d_invalidate() */
1349 spin_lock(&p->d_lock);
1350 if (unlikely(d_unhashed(p))) {
1351 spin_unlock(&p->d_lock);
1354 spin_unlock(&p->d_lock);
1356 spin_lock(&dentry->d_lock);
1357 if (!d_unlinked(dentry)) {
1359 if (!d_mountpoint(dentry)) {
1360 dentry->d_flags |= DCACHE_MOUNTED;
1364 spin_unlock(&dentry->d_lock);
1366 write_sequnlock(&rename_lock);
1371 * Search the dentry child list of the specified parent,
1372 * and move any unused dentries to the end of the unused
1373 * list for prune_dcache(). We descend to the next level
1374 * whenever the d_subdirs list is non-empty and continue
1377 * It returns zero iff there are no unused children,
1378 * otherwise it returns the number of children moved to
1379 * the end of the unused list. This may not be the total
1380 * number of unused children, because select_parent can
1381 * drop the lock and return early due to latency
1385 struct select_data {
1386 struct dentry *start;
1387 struct list_head dispose;
1391 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1393 struct select_data *data = _data;
1394 enum d_walk_ret ret = D_WALK_CONTINUE;
1396 if (data->start == dentry)
1399 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1402 if (dentry->d_flags & DCACHE_LRU_LIST)
1404 if (!dentry->d_lockref.count) {
1405 d_shrink_add(dentry, &data->dispose);
1410 * We can return to the caller if we have found some (this
1411 * ensures forward progress). We'll be coming back to find
1414 if (!list_empty(&data->dispose))
1415 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1421 * shrink_dcache_parent - prune dcache
1422 * @parent: parent of entries to prune
1424 * Prune the dcache to remove unused children of the parent dentry.
1426 void shrink_dcache_parent(struct dentry *parent)
1429 struct select_data data;
1431 INIT_LIST_HEAD(&data.dispose);
1432 data.start = parent;
1435 d_walk(parent, &data, select_collect, NULL);
1439 shrink_dentry_list(&data.dispose);
1443 EXPORT_SYMBOL(shrink_dcache_parent);
1445 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1447 /* it has busy descendents; complain about those instead */
1448 if (!list_empty(&dentry->d_subdirs))
1449 return D_WALK_CONTINUE;
1451 /* root with refcount 1 is fine */
1452 if (dentry == _data && dentry->d_lockref.count == 1)
1453 return D_WALK_CONTINUE;
1455 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1456 " still in use (%d) [unmount of %s %s]\n",
1459 dentry->d_inode->i_ino : 0UL,
1461 dentry->d_lockref.count,
1462 dentry->d_sb->s_type->name,
1463 dentry->d_sb->s_id);
1465 return D_WALK_CONTINUE;
1468 static void do_one_tree(struct dentry *dentry)
1470 shrink_dcache_parent(dentry);
1471 d_walk(dentry, dentry, umount_check, NULL);
1477 * destroy the dentries attached to a superblock on unmounting
1479 void shrink_dcache_for_umount(struct super_block *sb)
1481 struct dentry *dentry;
1483 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1485 dentry = sb->s_root;
1487 do_one_tree(dentry);
1489 while (!hlist_bl_empty(&sb->s_anon)) {
1490 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1491 do_one_tree(dentry);
1495 struct detach_data {
1496 struct select_data select;
1497 struct dentry *mountpoint;
1499 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1501 struct detach_data *data = _data;
1503 if (d_mountpoint(dentry)) {
1504 __dget_dlock(dentry);
1505 data->mountpoint = dentry;
1509 return select_collect(&data->select, dentry);
1512 static void check_and_drop(void *_data)
1514 struct detach_data *data = _data;
1516 if (!data->mountpoint && !data->select.found)
1517 __d_drop(data->select.start);
1521 * d_invalidate - detach submounts, prune dcache, and drop
1522 * @dentry: dentry to invalidate (aka detach, prune and drop)
1526 * The final d_drop is done as an atomic operation relative to
1527 * rename_lock ensuring there are no races with d_set_mounted. This
1528 * ensures there are no unhashed dentries on the path to a mountpoint.
1530 void d_invalidate(struct dentry *dentry)
1533 * If it's already been dropped, return OK.
1535 spin_lock(&dentry->d_lock);
1536 if (d_unhashed(dentry)) {
1537 spin_unlock(&dentry->d_lock);
1540 spin_unlock(&dentry->d_lock);
1542 /* Negative dentries can be dropped without further checks */
1543 if (!dentry->d_inode) {
1549 struct detach_data data;
1551 data.mountpoint = NULL;
1552 INIT_LIST_HEAD(&data.select.dispose);
1553 data.select.start = dentry;
1554 data.select.found = 0;
1556 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1558 if (data.select.found)
1559 shrink_dentry_list(&data.select.dispose);
1561 if (data.mountpoint) {
1562 detach_mounts(data.mountpoint);
1563 dput(data.mountpoint);
1566 if (!data.mountpoint && !data.select.found)
1572 EXPORT_SYMBOL(d_invalidate);
1575 * __d_alloc - allocate a dcache entry
1576 * @sb: filesystem it will belong to
1577 * @name: qstr of the name
1579 * Allocates a dentry. It returns %NULL if there is insufficient memory
1580 * available. On a success the dentry is returned. The name passed in is
1581 * copied and the copy passed in may be reused after this call.
1584 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1586 struct dentry *dentry;
1589 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1594 * We guarantee that the inline name is always NUL-terminated.
1595 * This way the memcpy() done by the name switching in rename
1596 * will still always have a NUL at the end, even if we might
1597 * be overwriting an internal NUL character
1599 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1600 if (name->len > DNAME_INLINE_LEN-1) {
1601 size_t size = offsetof(struct external_name, name[1]);
1602 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1604 kmem_cache_free(dentry_cache, dentry);
1607 atomic_set(&p->u.count, 1);
1609 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1610 kasan_unpoison_shadow(dname,
1611 round_up(name->len + 1, sizeof(unsigned long)));
1613 dname = dentry->d_iname;
1616 dentry->d_name.len = name->len;
1617 dentry->d_name.hash = name->hash;
1618 memcpy(dname, name->name, name->len);
1619 dname[name->len] = 0;
1621 /* Make sure we always see the terminating NUL character */
1623 dentry->d_name.name = dname;
1625 dentry->d_lockref.count = 1;
1626 dentry->d_flags = 0;
1627 spin_lock_init(&dentry->d_lock);
1628 seqcount_init(&dentry->d_seq);
1629 dentry->d_inode = NULL;
1630 dentry->d_parent = dentry;
1632 dentry->d_op = NULL;
1633 dentry->d_fsdata = NULL;
1634 INIT_HLIST_BL_NODE(&dentry->d_hash);
1635 INIT_LIST_HEAD(&dentry->d_lru);
1636 INIT_LIST_HEAD(&dentry->d_subdirs);
1637 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1638 INIT_LIST_HEAD(&dentry->d_child);
1639 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1641 this_cpu_inc(nr_dentry);
1647 * d_alloc - allocate a dcache entry
1648 * @parent: parent of entry to allocate
1649 * @name: qstr of the name
1651 * Allocates a dentry. It returns %NULL if there is insufficient memory
1652 * available. On a success the dentry is returned. The name passed in is
1653 * copied and the copy passed in may be reused after this call.
1655 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1657 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1660 dentry->d_flags |= DCACHE_RCUACCESS;
1661 spin_lock(&parent->d_lock);
1663 * don't need child lock because it is not subject
1664 * to concurrency here
1666 __dget_dlock(parent);
1667 dentry->d_parent = parent;
1668 list_add(&dentry->d_child, &parent->d_subdirs);
1669 spin_unlock(&parent->d_lock);
1673 EXPORT_SYMBOL(d_alloc);
1676 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1677 * @sb: the superblock
1678 * @name: qstr of the name
1680 * For a filesystem that just pins its dentries in memory and never
1681 * performs lookups at all, return an unhashed IS_ROOT dentry.
1683 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1685 return __d_alloc(sb, name);
1687 EXPORT_SYMBOL(d_alloc_pseudo);
1689 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1694 q.len = strlen(name);
1695 q.hash = full_name_hash(q.name, q.len);
1696 return d_alloc(parent, &q);
1698 EXPORT_SYMBOL(d_alloc_name);
1700 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1702 WARN_ON_ONCE(dentry->d_op);
1703 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1705 DCACHE_OP_REVALIDATE |
1706 DCACHE_OP_WEAK_REVALIDATE |
1708 DCACHE_OP_SELECT_INODE |
1714 dentry->d_flags |= DCACHE_OP_HASH;
1716 dentry->d_flags |= DCACHE_OP_COMPARE;
1717 if (op->d_revalidate)
1718 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1719 if (op->d_weak_revalidate)
1720 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1722 dentry->d_flags |= DCACHE_OP_DELETE;
1724 dentry->d_flags |= DCACHE_OP_PRUNE;
1725 if (op->d_select_inode)
1726 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
1728 dentry->d_flags |= DCACHE_OP_REAL;
1731 EXPORT_SYMBOL(d_set_d_op);
1735 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1736 * @dentry - The dentry to mark
1738 * Mark a dentry as falling through to the lower layer (as set with
1739 * d_pin_lower()). This flag may be recorded on the medium.
1741 void d_set_fallthru(struct dentry *dentry)
1743 spin_lock(&dentry->d_lock);
1744 dentry->d_flags |= DCACHE_FALLTHRU;
1745 spin_unlock(&dentry->d_lock);
1747 EXPORT_SYMBOL(d_set_fallthru);
1749 static unsigned d_flags_for_inode(struct inode *inode)
1751 unsigned add_flags = DCACHE_REGULAR_TYPE;
1754 return DCACHE_MISS_TYPE;
1756 if (S_ISDIR(inode->i_mode)) {
1757 add_flags = DCACHE_DIRECTORY_TYPE;
1758 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1759 if (unlikely(!inode->i_op->lookup))
1760 add_flags = DCACHE_AUTODIR_TYPE;
1762 inode->i_opflags |= IOP_LOOKUP;
1764 goto type_determined;
1767 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1768 if (unlikely(inode->i_op->follow_link)) {
1769 add_flags = DCACHE_SYMLINK_TYPE;
1770 goto type_determined;
1772 inode->i_opflags |= IOP_NOFOLLOW;
1775 if (unlikely(!S_ISREG(inode->i_mode)))
1776 add_flags = DCACHE_SPECIAL_TYPE;
1779 if (unlikely(IS_AUTOMOUNT(inode)))
1780 add_flags |= DCACHE_NEED_AUTOMOUNT;
1784 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1786 unsigned add_flags = d_flags_for_inode(inode);
1788 spin_lock(&dentry->d_lock);
1790 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1791 raw_write_seqcount_begin(&dentry->d_seq);
1792 __d_set_inode_and_type(dentry, inode, add_flags);
1793 raw_write_seqcount_end(&dentry->d_seq);
1794 spin_unlock(&dentry->d_lock);
1795 fsnotify_d_instantiate(dentry, inode);
1799 * d_instantiate - fill in inode information for a dentry
1800 * @entry: dentry to complete
1801 * @inode: inode to attach to this dentry
1803 * Fill in inode information in the entry.
1805 * This turns negative dentries into productive full members
1808 * NOTE! This assumes that the inode count has been incremented
1809 * (or otherwise set) by the caller to indicate that it is now
1810 * in use by the dcache.
1813 void d_instantiate(struct dentry *entry, struct inode * inode)
1815 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1817 spin_lock(&inode->i_lock);
1818 __d_instantiate(entry, inode);
1820 spin_unlock(&inode->i_lock);
1821 security_d_instantiate(entry, inode);
1823 EXPORT_SYMBOL(d_instantiate);
1826 * d_instantiate_unique - instantiate a non-aliased dentry
1827 * @entry: dentry to instantiate
1828 * @inode: inode to attach to this dentry
1830 * Fill in inode information in the entry. On success, it returns NULL.
1831 * If an unhashed alias of "entry" already exists, then we return the
1832 * aliased dentry instead and drop one reference to inode.
1834 * Note that in order to avoid conflicts with rename() etc, the caller
1835 * had better be holding the parent directory semaphore.
1837 * This also assumes that the inode count has been incremented
1838 * (or otherwise set) by the caller to indicate that it is now
1839 * in use by the dcache.
1841 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1842 struct inode *inode)
1844 struct dentry *alias;
1845 int len = entry->d_name.len;
1846 const char *name = entry->d_name.name;
1847 unsigned int hash = entry->d_name.hash;
1850 __d_instantiate(entry, NULL);
1854 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1856 * Don't need alias->d_lock here, because aliases with
1857 * d_parent == entry->d_parent are not subject to name or
1858 * parent changes, because the parent inode i_mutex is held.
1860 if (alias->d_name.hash != hash)
1862 if (alias->d_parent != entry->d_parent)
1864 if (alias->d_name.len != len)
1866 if (dentry_cmp(alias, name, len))
1872 __d_instantiate(entry, inode);
1876 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1878 struct dentry *result;
1880 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1883 spin_lock(&inode->i_lock);
1884 result = __d_instantiate_unique(entry, inode);
1886 spin_unlock(&inode->i_lock);
1889 security_d_instantiate(entry, inode);
1893 BUG_ON(!d_unhashed(result));
1898 EXPORT_SYMBOL(d_instantiate_unique);
1901 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1902 * @entry: dentry to complete
1903 * @inode: inode to attach to this dentry
1905 * Fill in inode information in the entry. If a directory alias is found, then
1906 * return an error (and drop inode). Together with d_materialise_unique() this
1907 * guarantees that a directory inode may never have more than one alias.
1909 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1911 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1913 spin_lock(&inode->i_lock);
1914 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1915 spin_unlock(&inode->i_lock);
1919 __d_instantiate(entry, inode);
1920 spin_unlock(&inode->i_lock);
1921 security_d_instantiate(entry, inode);
1925 EXPORT_SYMBOL(d_instantiate_no_diralias);
1927 struct dentry *d_make_root(struct inode *root_inode)
1929 struct dentry *res = NULL;
1932 static const struct qstr name = QSTR_INIT("/", 1);
1934 res = __d_alloc(root_inode->i_sb, &name);
1936 d_instantiate(res, root_inode);
1942 EXPORT_SYMBOL(d_make_root);
1944 static struct dentry * __d_find_any_alias(struct inode *inode)
1946 struct dentry *alias;
1948 if (hlist_empty(&inode->i_dentry))
1950 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1956 * d_find_any_alias - find any alias for a given inode
1957 * @inode: inode to find an alias for
1959 * If any aliases exist for the given inode, take and return a
1960 * reference for one of them. If no aliases exist, return %NULL.
1962 struct dentry *d_find_any_alias(struct inode *inode)
1966 spin_lock(&inode->i_lock);
1967 de = __d_find_any_alias(inode);
1968 spin_unlock(&inode->i_lock);
1971 EXPORT_SYMBOL(d_find_any_alias);
1973 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1975 static const struct qstr anonstring = QSTR_INIT("/", 1);
1981 return ERR_PTR(-ESTALE);
1983 return ERR_CAST(inode);
1985 res = d_find_any_alias(inode);
1989 tmp = __d_alloc(inode->i_sb, &anonstring);
1991 res = ERR_PTR(-ENOMEM);
1995 spin_lock(&inode->i_lock);
1996 res = __d_find_any_alias(inode);
1998 spin_unlock(&inode->i_lock);
2003 /* attach a disconnected dentry */
2004 add_flags = d_flags_for_inode(inode);
2007 add_flags |= DCACHE_DISCONNECTED;
2009 spin_lock(&tmp->d_lock);
2010 __d_set_inode_and_type(tmp, inode, add_flags);
2011 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
2012 hlist_bl_lock(&tmp->d_sb->s_anon);
2013 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
2014 hlist_bl_unlock(&tmp->d_sb->s_anon);
2015 spin_unlock(&tmp->d_lock);
2016 spin_unlock(&inode->i_lock);
2017 security_d_instantiate(tmp, inode);
2022 if (res && !IS_ERR(res))
2023 security_d_instantiate(res, inode);
2029 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2030 * @inode: inode to allocate the dentry for
2032 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2033 * similar open by handle operations. The returned dentry may be anonymous,
2034 * or may have a full name (if the inode was already in the cache).
2036 * When called on a directory inode, we must ensure that the inode only ever
2037 * has one dentry. If a dentry is found, that is returned instead of
2038 * allocating a new one.
2040 * On successful return, the reference to the inode has been transferred
2041 * to the dentry. In case of an error the reference on the inode is released.
2042 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2043 * be passed in and the error will be propagated to the return value,
2044 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2046 struct dentry *d_obtain_alias(struct inode *inode)
2048 return __d_obtain_alias(inode, 1);
2050 EXPORT_SYMBOL(d_obtain_alias);
2053 * d_obtain_root - find or allocate a dentry for a given inode
2054 * @inode: inode to allocate the dentry for
2056 * Obtain an IS_ROOT dentry for the root of a filesystem.
2058 * We must ensure that directory inodes only ever have one dentry. If a
2059 * dentry is found, that is returned instead of allocating a new one.
2061 * On successful return, the reference to the inode has been transferred
2062 * to the dentry. In case of an error the reference on the inode is
2063 * released. A %NULL or IS_ERR inode may be passed in and will be the
2064 * error will be propagate to the return value, with a %NULL @inode
2065 * replaced by ERR_PTR(-ESTALE).
2067 struct dentry *d_obtain_root(struct inode *inode)
2069 return __d_obtain_alias(inode, 0);
2071 EXPORT_SYMBOL(d_obtain_root);
2074 * d_add_ci - lookup or allocate new dentry with case-exact name
2075 * @inode: the inode case-insensitive lookup has found
2076 * @dentry: the negative dentry that was passed to the parent's lookup func
2077 * @name: the case-exact name to be associated with the returned dentry
2079 * This is to avoid filling the dcache with case-insensitive names to the
2080 * same inode, only the actual correct case is stored in the dcache for
2081 * case-insensitive filesystems.
2083 * For a case-insensitive lookup match and if the the case-exact dentry
2084 * already exists in in the dcache, use it and return it.
2086 * If no entry exists with the exact case name, allocate new dentry with
2087 * the exact case, and return the spliced entry.
2089 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2092 struct dentry *found;
2096 * First check if a dentry matching the name already exists,
2097 * if not go ahead and create it now.
2099 found = d_hash_and_lookup(dentry->d_parent, name);
2101 new = d_alloc(dentry->d_parent, name);
2103 found = ERR_PTR(-ENOMEM);
2105 found = d_splice_alias(inode, new);
2116 EXPORT_SYMBOL(d_add_ci);
2119 * Do the slow-case of the dentry name compare.
2121 * Unlike the dentry_cmp() function, we need to atomically
2122 * load the name and length information, so that the
2123 * filesystem can rely on them, and can use the 'name' and
2124 * 'len' information without worrying about walking off the
2125 * end of memory etc.
2127 * Thus the read_seqcount_retry() and the "duplicate" info
2128 * in arguments (the low-level filesystem should not look
2129 * at the dentry inode or name contents directly, since
2130 * rename can change them while we're in RCU mode).
2132 enum slow_d_compare {
2138 static noinline enum slow_d_compare slow_dentry_cmp(
2139 const struct dentry *parent,
2140 struct dentry *dentry,
2142 const struct qstr *name)
2144 int tlen = dentry->d_name.len;
2145 const char *tname = dentry->d_name.name;
2147 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2149 return D_COMP_SEQRETRY;
2151 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2152 return D_COMP_NOMATCH;
2157 * __d_lookup_rcu - search for a dentry (racy, store-free)
2158 * @parent: parent dentry
2159 * @name: qstr of name we wish to find
2160 * @seqp: returns d_seq value at the point where the dentry was found
2161 * Returns: dentry, or NULL
2163 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2164 * resolution (store-free path walking) design described in
2165 * Documentation/filesystems/path-lookup.txt.
2167 * This is not to be used outside core vfs.
2169 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2170 * held, and rcu_read_lock held. The returned dentry must not be stored into
2171 * without taking d_lock and checking d_seq sequence count against @seq
2174 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2177 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2178 * the returned dentry, so long as its parent's seqlock is checked after the
2179 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2180 * is formed, giving integrity down the path walk.
2182 * NOTE! The caller *has* to check the resulting dentry against the sequence
2183 * number we've returned before using any of the resulting dentry state!
2185 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2186 const struct qstr *name,
2189 u64 hashlen = name->hash_len;
2190 const unsigned char *str = name->name;
2191 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2192 struct hlist_bl_node *node;
2193 struct dentry *dentry;
2196 * Note: There is significant duplication with __d_lookup_rcu which is
2197 * required to prevent single threaded performance regressions
2198 * especially on architectures where smp_rmb (in seqcounts) are costly.
2199 * Keep the two functions in sync.
2203 * The hash list is protected using RCU.
2205 * Carefully use d_seq when comparing a candidate dentry, to avoid
2206 * races with d_move().
2208 * It is possible that concurrent renames can mess up our list
2209 * walk here and result in missing our dentry, resulting in the
2210 * false-negative result. d_lookup() protects against concurrent
2211 * renames using rename_lock seqlock.
2213 * See Documentation/filesystems/path-lookup.txt for more details.
2215 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2220 * The dentry sequence count protects us from concurrent
2221 * renames, and thus protects parent and name fields.
2223 * The caller must perform a seqcount check in order
2224 * to do anything useful with the returned dentry.
2226 * NOTE! We do a "raw" seqcount_begin here. That means that
2227 * we don't wait for the sequence count to stabilize if it
2228 * is in the middle of a sequence change. If we do the slow
2229 * dentry compare, we will do seqretries until it is stable,
2230 * and if we end up with a successful lookup, we actually
2231 * want to exit RCU lookup anyway.
2233 seq = raw_seqcount_begin(&dentry->d_seq);
2234 if (dentry->d_parent != parent)
2236 if (d_unhashed(dentry))
2239 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2240 if (dentry->d_name.hash != hashlen_hash(hashlen))
2243 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2246 case D_COMP_NOMATCH:
2253 if (dentry->d_name.hash_len != hashlen)
2256 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2263 * d_lookup - search for a dentry
2264 * @parent: parent dentry
2265 * @name: qstr of name we wish to find
2266 * Returns: dentry, or NULL
2268 * d_lookup searches the children of the parent dentry for the name in
2269 * question. If the dentry is found its reference count is incremented and the
2270 * dentry is returned. The caller must use dput to free the entry when it has
2271 * finished using it. %NULL is returned if the dentry does not exist.
2273 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2275 struct dentry *dentry;
2279 seq = read_seqbegin(&rename_lock);
2280 dentry = __d_lookup(parent, name);
2283 } while (read_seqretry(&rename_lock, seq));
2286 EXPORT_SYMBOL(d_lookup);
2289 * __d_lookup - search for a dentry (racy)
2290 * @parent: parent dentry
2291 * @name: qstr of name we wish to find
2292 * Returns: dentry, or NULL
2294 * __d_lookup is like d_lookup, however it may (rarely) return a
2295 * false-negative result due to unrelated rename activity.
2297 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2298 * however it must be used carefully, eg. with a following d_lookup in
2299 * the case of failure.
2301 * __d_lookup callers must be commented.
2303 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2305 unsigned int len = name->len;
2306 unsigned int hash = name->hash;
2307 const unsigned char *str = name->name;
2308 struct hlist_bl_head *b = d_hash(parent, hash);
2309 struct hlist_bl_node *node;
2310 struct dentry *found = NULL;
2311 struct dentry *dentry;
2314 * Note: There is significant duplication with __d_lookup_rcu which is
2315 * required to prevent single threaded performance regressions
2316 * especially on architectures where smp_rmb (in seqcounts) are costly.
2317 * Keep the two functions in sync.
2321 * The hash list is protected using RCU.
2323 * Take d_lock when comparing a candidate dentry, to avoid races
2326 * It is possible that concurrent renames can mess up our list
2327 * walk here and result in missing our dentry, resulting in the
2328 * false-negative result. d_lookup() protects against concurrent
2329 * renames using rename_lock seqlock.
2331 * See Documentation/filesystems/path-lookup.txt for more details.
2335 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2337 if (dentry->d_name.hash != hash)
2340 spin_lock(&dentry->d_lock);
2341 if (dentry->d_parent != parent)
2343 if (d_unhashed(dentry))
2347 * It is safe to compare names since d_move() cannot
2348 * change the qstr (protected by d_lock).
2350 if (parent->d_flags & DCACHE_OP_COMPARE) {
2351 int tlen = dentry->d_name.len;
2352 const char *tname = dentry->d_name.name;
2353 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2356 if (dentry->d_name.len != len)
2358 if (dentry_cmp(dentry, str, len))
2362 dentry->d_lockref.count++;
2364 spin_unlock(&dentry->d_lock);
2367 spin_unlock(&dentry->d_lock);
2375 * d_hash_and_lookup - hash the qstr then search for a dentry
2376 * @dir: Directory to search in
2377 * @name: qstr of name we wish to find
2379 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2381 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2384 * Check for a fs-specific hash function. Note that we must
2385 * calculate the standard hash first, as the d_op->d_hash()
2386 * routine may choose to leave the hash value unchanged.
2388 name->hash = full_name_hash(name->name, name->len);
2389 if (dir->d_flags & DCACHE_OP_HASH) {
2390 int err = dir->d_op->d_hash(dir, name);
2391 if (unlikely(err < 0))
2392 return ERR_PTR(err);
2394 return d_lookup(dir, name);
2396 EXPORT_SYMBOL(d_hash_and_lookup);
2399 * When a file is deleted, we have two options:
2400 * - turn this dentry into a negative dentry
2401 * - unhash this dentry and free it.
2403 * Usually, we want to just turn this into
2404 * a negative dentry, but if anybody else is
2405 * currently using the dentry or the inode
2406 * we can't do that and we fall back on removing
2407 * it from the hash queues and waiting for
2408 * it to be deleted later when it has no users
2412 * d_delete - delete a dentry
2413 * @dentry: The dentry to delete
2415 * Turn the dentry into a negative dentry if possible, otherwise
2416 * remove it from the hash queues so it can be deleted later
2419 void d_delete(struct dentry * dentry)
2421 struct inode *inode;
2424 * Are we the only user?
2427 spin_lock(&dentry->d_lock);
2428 inode = dentry->d_inode;
2429 isdir = S_ISDIR(inode->i_mode);
2430 if (dentry->d_lockref.count == 1) {
2431 if (!spin_trylock(&inode->i_lock)) {
2432 spin_unlock(&dentry->d_lock);
2436 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2437 dentry_unlink_inode(dentry);
2438 fsnotify_nameremove(dentry, isdir);
2442 if (!d_unhashed(dentry))
2445 spin_unlock(&dentry->d_lock);
2447 fsnotify_nameremove(dentry, isdir);
2449 EXPORT_SYMBOL(d_delete);
2451 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2453 BUG_ON(!d_unhashed(entry));
2455 hlist_bl_add_head_rcu(&entry->d_hash, b);
2459 static void _d_rehash(struct dentry * entry)
2461 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2465 * d_rehash - add an entry back to the hash
2466 * @entry: dentry to add to the hash
2468 * Adds a dentry to the hash according to its name.
2471 void d_rehash(struct dentry * entry)
2473 spin_lock(&entry->d_lock);
2475 spin_unlock(&entry->d_lock);
2477 EXPORT_SYMBOL(d_rehash);
2480 * dentry_update_name_case - update case insensitive dentry with a new name
2481 * @dentry: dentry to be updated
2484 * Update a case insensitive dentry with new case of name.
2486 * dentry must have been returned by d_lookup with name @name. Old and new
2487 * name lengths must match (ie. no d_compare which allows mismatched name
2490 * Parent inode i_mutex must be held over d_lookup and into this call (to
2491 * keep renames and concurrent inserts, and readdir(2) away).
2493 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2495 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2496 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2498 spin_lock(&dentry->d_lock);
2499 write_seqcount_begin(&dentry->d_seq);
2500 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2501 write_seqcount_end(&dentry->d_seq);
2502 spin_unlock(&dentry->d_lock);
2504 EXPORT_SYMBOL(dentry_update_name_case);
2506 static void swap_names(struct dentry *dentry, struct dentry *target)
2508 if (unlikely(dname_external(target))) {
2509 if (unlikely(dname_external(dentry))) {
2511 * Both external: swap the pointers
2513 swap(target->d_name.name, dentry->d_name.name);
2516 * dentry:internal, target:external. Steal target's
2517 * storage and make target internal.
2519 memcpy(target->d_iname, dentry->d_name.name,
2520 dentry->d_name.len + 1);
2521 dentry->d_name.name = target->d_name.name;
2522 target->d_name.name = target->d_iname;
2525 if (unlikely(dname_external(dentry))) {
2527 * dentry:external, target:internal. Give dentry's
2528 * storage to target and make dentry internal
2530 memcpy(dentry->d_iname, target->d_name.name,
2531 target->d_name.len + 1);
2532 target->d_name.name = dentry->d_name.name;
2533 dentry->d_name.name = dentry->d_iname;
2536 * Both are internal.
2539 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2540 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2541 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2542 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2543 swap(((long *) &dentry->d_iname)[i],
2544 ((long *) &target->d_iname)[i]);
2548 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2551 static void copy_name(struct dentry *dentry, struct dentry *target)
2553 struct external_name *old_name = NULL;
2554 if (unlikely(dname_external(dentry)))
2555 old_name = external_name(dentry);
2556 if (unlikely(dname_external(target))) {
2557 atomic_inc(&external_name(target)->u.count);
2558 dentry->d_name = target->d_name;
2560 memcpy(dentry->d_iname, target->d_name.name,
2561 target->d_name.len + 1);
2562 dentry->d_name.name = dentry->d_iname;
2563 dentry->d_name.hash_len = target->d_name.hash_len;
2565 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2566 kfree_rcu(old_name, u.head);
2569 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2572 * XXXX: do we really need to take target->d_lock?
2574 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2575 spin_lock(&target->d_parent->d_lock);
2577 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2578 spin_lock(&dentry->d_parent->d_lock);
2579 spin_lock_nested(&target->d_parent->d_lock,
2580 DENTRY_D_LOCK_NESTED);
2582 spin_lock(&target->d_parent->d_lock);
2583 spin_lock_nested(&dentry->d_parent->d_lock,
2584 DENTRY_D_LOCK_NESTED);
2587 if (target < dentry) {
2588 spin_lock_nested(&target->d_lock, 2);
2589 spin_lock_nested(&dentry->d_lock, 3);
2591 spin_lock_nested(&dentry->d_lock, 2);
2592 spin_lock_nested(&target->d_lock, 3);
2596 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2598 if (target->d_parent != dentry->d_parent)
2599 spin_unlock(&dentry->d_parent->d_lock);
2600 if (target->d_parent != target)
2601 spin_unlock(&target->d_parent->d_lock);
2602 spin_unlock(&target->d_lock);
2603 spin_unlock(&dentry->d_lock);
2607 * When switching names, the actual string doesn't strictly have to
2608 * be preserved in the target - because we're dropping the target
2609 * anyway. As such, we can just do a simple memcpy() to copy over
2610 * the new name before we switch, unless we are going to rehash
2611 * it. Note that if we *do* unhash the target, we are not allowed
2612 * to rehash it without giving it a new name/hash key - whether
2613 * we swap or overwrite the names here, resulting name won't match
2614 * the reality in filesystem; it's only there for d_path() purposes.
2615 * Note that all of this is happening under rename_lock, so the
2616 * any hash lookup seeing it in the middle of manipulations will
2617 * be discarded anyway. So we do not care what happens to the hash
2621 * __d_move - move a dentry
2622 * @dentry: entry to move
2623 * @target: new dentry
2624 * @exchange: exchange the two dentries
2626 * Update the dcache to reflect the move of a file name. Negative
2627 * dcache entries should not be moved in this way. Caller must hold
2628 * rename_lock, the i_mutex of the source and target directories,
2629 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2631 static void __d_move(struct dentry *dentry, struct dentry *target,
2634 if (!dentry->d_inode)
2635 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2637 BUG_ON(d_ancestor(dentry, target));
2638 BUG_ON(d_ancestor(target, dentry));
2640 dentry_lock_for_move(dentry, target);
2642 write_seqcount_begin(&dentry->d_seq);
2643 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2645 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2648 * Move the dentry to the target hash queue. Don't bother checking
2649 * for the same hash queue because of how unlikely it is.
2652 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2655 * Unhash the target (d_delete() is not usable here). If exchanging
2656 * the two dentries, then rehash onto the other's hash queue.
2661 d_hash(dentry->d_parent, dentry->d_name.hash));
2664 /* Switch the names.. */
2666 swap_names(dentry, target);
2668 copy_name(dentry, target);
2670 /* ... and switch them in the tree */
2671 if (IS_ROOT(dentry)) {
2672 /* splicing a tree */
2673 dentry->d_flags |= DCACHE_RCUACCESS;
2674 dentry->d_parent = target->d_parent;
2675 target->d_parent = target;
2676 list_del_init(&target->d_child);
2677 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2679 /* swapping two dentries */
2680 swap(dentry->d_parent, target->d_parent);
2681 list_move(&target->d_child, &target->d_parent->d_subdirs);
2682 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2684 fsnotify_d_move(target);
2685 fsnotify_d_move(dentry);
2688 write_seqcount_end(&target->d_seq);
2689 write_seqcount_end(&dentry->d_seq);
2691 dentry_unlock_for_move(dentry, target);
2695 * d_move - move a dentry
2696 * @dentry: entry to move
2697 * @target: new dentry
2699 * Update the dcache to reflect the move of a file name. Negative
2700 * dcache entries should not be moved in this way. See the locking
2701 * requirements for __d_move.
2703 void d_move(struct dentry *dentry, struct dentry *target)
2705 write_seqlock(&rename_lock);
2706 __d_move(dentry, target, false);
2707 write_sequnlock(&rename_lock);
2709 EXPORT_SYMBOL(d_move);
2712 * d_exchange - exchange two dentries
2713 * @dentry1: first dentry
2714 * @dentry2: second dentry
2716 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2718 write_seqlock(&rename_lock);
2720 WARN_ON(!dentry1->d_inode);
2721 WARN_ON(!dentry2->d_inode);
2722 WARN_ON(IS_ROOT(dentry1));
2723 WARN_ON(IS_ROOT(dentry2));
2725 __d_move(dentry1, dentry2, true);
2727 write_sequnlock(&rename_lock);
2731 * d_ancestor - search for an ancestor
2732 * @p1: ancestor dentry
2735 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2736 * an ancestor of p2, else NULL.
2738 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2742 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2743 if (p->d_parent == p1)
2750 * This helper attempts to cope with remotely renamed directories
2752 * It assumes that the caller is already holding
2753 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2755 * Note: If ever the locking in lock_rename() changes, then please
2756 * remember to update this too...
2758 static int __d_unalias(struct inode *inode,
2759 struct dentry *dentry, struct dentry *alias)
2761 struct mutex *m1 = NULL, *m2 = NULL;
2764 /* If alias and dentry share a parent, then no extra locks required */
2765 if (alias->d_parent == dentry->d_parent)
2768 /* See lock_rename() */
2769 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2771 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2772 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2774 m2 = &alias->d_parent->d_inode->i_mutex;
2776 __d_move(alias, dentry, false);
2787 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2788 * @inode: the inode which may have a disconnected dentry
2789 * @dentry: a negative dentry which we want to point to the inode.
2791 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2792 * place of the given dentry and return it, else simply d_add the inode
2793 * to the dentry and return NULL.
2795 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2796 * we should error out: directories can't have multiple aliases.
2798 * This is needed in the lookup routine of any filesystem that is exportable
2799 * (via knfsd) so that we can build dcache paths to directories effectively.
2801 * If a dentry was found and moved, then it is returned. Otherwise NULL
2802 * is returned. This matches the expected return value of ->lookup.
2804 * Cluster filesystems may call this function with a negative, hashed dentry.
2805 * In that case, we know that the inode will be a regular file, and also this
2806 * will only occur during atomic_open. So we need to check for the dentry
2807 * being already hashed only in the final case.
2809 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2812 return ERR_CAST(inode);
2814 BUG_ON(!d_unhashed(dentry));
2817 __d_instantiate(dentry, NULL);
2820 spin_lock(&inode->i_lock);
2821 if (S_ISDIR(inode->i_mode)) {
2822 struct dentry *new = __d_find_any_alias(inode);
2823 if (unlikely(new)) {
2824 /* The reference to new ensures it remains an alias */
2825 spin_unlock(&inode->i_lock);
2826 write_seqlock(&rename_lock);
2827 if (unlikely(d_ancestor(new, dentry))) {
2828 write_sequnlock(&rename_lock);
2830 new = ERR_PTR(-ELOOP);
2831 pr_warn_ratelimited(
2832 "VFS: Lookup of '%s' in %s %s"
2833 " would have caused loop\n",
2834 dentry->d_name.name,
2835 inode->i_sb->s_type->name,
2837 } else if (!IS_ROOT(new)) {
2838 int err = __d_unalias(inode, dentry, new);
2839 write_sequnlock(&rename_lock);
2845 __d_move(new, dentry, false);
2846 write_sequnlock(&rename_lock);
2847 security_d_instantiate(new, inode);
2853 /* already taking inode->i_lock, so d_add() by hand */
2854 __d_instantiate(dentry, inode);
2855 spin_unlock(&inode->i_lock);
2857 security_d_instantiate(dentry, inode);
2861 EXPORT_SYMBOL(d_splice_alias);
2863 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2867 return -ENAMETOOLONG;
2869 memcpy(*buffer, str, namelen);
2874 * prepend_name - prepend a pathname in front of current buffer pointer
2875 * @buffer: buffer pointer
2876 * @buflen: allocated length of the buffer
2877 * @name: name string and length qstr structure
2879 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2880 * make sure that either the old or the new name pointer and length are
2881 * fetched. However, there may be mismatch between length and pointer.
2882 * The length cannot be trusted, we need to copy it byte-by-byte until
2883 * the length is reached or a null byte is found. It also prepends "/" at
2884 * the beginning of the name. The sequence number check at the caller will
2885 * retry it again when a d_move() does happen. So any garbage in the buffer
2886 * due to mismatched pointer and length will be discarded.
2888 * Data dependency barrier is needed to make sure that we see that terminating
2889 * NUL. Alpha strikes again, film at 11...
2891 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2893 const char *dname = ACCESS_ONCE(name->name);
2894 u32 dlen = ACCESS_ONCE(name->len);
2897 smp_read_barrier_depends();
2899 *buflen -= dlen + 1;
2901 return -ENAMETOOLONG;
2902 p = *buffer -= dlen + 1;
2914 * prepend_path - Prepend path string to a buffer
2915 * @path: the dentry/vfsmount to report
2916 * @root: root vfsmnt/dentry
2917 * @buffer: pointer to the end of the buffer
2918 * @buflen: pointer to buffer length
2920 * The function will first try to write out the pathname without taking any
2921 * lock other than the RCU read lock to make sure that dentries won't go away.
2922 * It only checks the sequence number of the global rename_lock as any change
2923 * in the dentry's d_seq will be preceded by changes in the rename_lock
2924 * sequence number. If the sequence number had been changed, it will restart
2925 * the whole pathname back-tracing sequence again by taking the rename_lock.
2926 * In this case, there is no need to take the RCU read lock as the recursive
2927 * parent pointer references will keep the dentry chain alive as long as no
2928 * rename operation is performed.
2930 static int prepend_path(const struct path *path,
2931 const struct path *root,
2932 char **buffer, int *buflen)
2934 struct dentry *dentry;
2935 struct vfsmount *vfsmnt;
2938 unsigned seq, m_seq = 0;
2944 read_seqbegin_or_lock(&mount_lock, &m_seq);
2951 dentry = path->dentry;
2953 mnt = real_mount(vfsmnt);
2954 read_seqbegin_or_lock(&rename_lock, &seq);
2955 while (dentry != root->dentry || vfsmnt != root->mnt) {
2956 struct dentry * parent;
2958 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2959 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2961 if (dentry != vfsmnt->mnt_root) {
2968 if (mnt != parent) {
2969 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2975 error = is_mounted(vfsmnt) ? 1 : 2;
2978 parent = dentry->d_parent;
2980 error = prepend_name(&bptr, &blen, &dentry->d_name);
2988 if (need_seqretry(&rename_lock, seq)) {
2992 done_seqretry(&rename_lock, seq);
2996 if (need_seqretry(&mount_lock, m_seq)) {
3000 done_seqretry(&mount_lock, m_seq);
3002 if (error >= 0 && bptr == *buffer) {
3004 error = -ENAMETOOLONG;
3014 * __d_path - return the path of a dentry
3015 * @path: the dentry/vfsmount to report
3016 * @root: root vfsmnt/dentry
3017 * @buf: buffer to return value in
3018 * @buflen: buffer length
3020 * Convert a dentry into an ASCII path name.
3022 * Returns a pointer into the buffer or an error code if the
3023 * path was too long.
3025 * "buflen" should be positive.
3027 * If the path is not reachable from the supplied root, return %NULL.
3029 char *__d_path(const struct path *path,
3030 const struct path *root,
3031 char *buf, int buflen)
3033 char *res = buf + buflen;
3036 prepend(&res, &buflen, "\0", 1);
3037 error = prepend_path(path, root, &res, &buflen);
3040 return ERR_PTR(error);
3046 char *d_absolute_path(const struct path *path,
3047 char *buf, int buflen)
3049 struct path root = {};
3050 char *res = buf + buflen;
3053 prepend(&res, &buflen, "\0", 1);
3054 error = prepend_path(path, &root, &res, &buflen);
3059 return ERR_PTR(error);
3062 EXPORT_SYMBOL(d_absolute_path);
3065 * same as __d_path but appends "(deleted)" for unlinked files.
3067 static int path_with_deleted(const struct path *path,
3068 const struct path *root,
3069 char **buf, int *buflen)
3071 prepend(buf, buflen, "\0", 1);
3072 if (d_unlinked(path->dentry)) {
3073 int error = prepend(buf, buflen, " (deleted)", 10);
3078 return prepend_path(path, root, buf, buflen);
3081 static int prepend_unreachable(char **buffer, int *buflen)
3083 return prepend(buffer, buflen, "(unreachable)", 13);
3086 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3091 seq = read_seqcount_begin(&fs->seq);
3093 } while (read_seqcount_retry(&fs->seq, seq));
3097 * d_path - return the path of a dentry
3098 * @path: path to report
3099 * @buf: buffer to return value in
3100 * @buflen: buffer length
3102 * Convert a dentry into an ASCII path name. If the entry has been deleted
3103 * the string " (deleted)" is appended. Note that this is ambiguous.
3105 * Returns a pointer into the buffer or an error code if the path was
3106 * too long. Note: Callers should use the returned pointer, not the passed
3107 * in buffer, to use the name! The implementation often starts at an offset
3108 * into the buffer, and may leave 0 bytes at the start.
3110 * "buflen" should be positive.
3112 char *d_path(const struct path *path, char *buf, int buflen)
3114 char *res = buf + buflen;
3119 * We have various synthetic filesystems that never get mounted. On
3120 * these filesystems dentries are never used for lookup purposes, and
3121 * thus don't need to be hashed. They also don't need a name until a
3122 * user wants to identify the object in /proc/pid/fd/. The little hack
3123 * below allows us to generate a name for these objects on demand:
3125 * Some pseudo inodes are mountable. When they are mounted
3126 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3127 * and instead have d_path return the mounted path.
3129 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3130 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3131 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3134 get_fs_root_rcu(current->fs, &root);
3135 error = path_with_deleted(path, &root, &res, &buflen);
3139 res = ERR_PTR(error);
3142 EXPORT_SYMBOL(d_path);
3145 * Helper function for dentry_operations.d_dname() members
3147 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3148 const char *fmt, ...)
3154 va_start(args, fmt);
3155 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3158 if (sz > sizeof(temp) || sz > buflen)
3159 return ERR_PTR(-ENAMETOOLONG);
3161 buffer += buflen - sz;
3162 return memcpy(buffer, temp, sz);
3165 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3167 char *end = buffer + buflen;
3168 /* these dentries are never renamed, so d_lock is not needed */
3169 if (prepend(&end, &buflen, " (deleted)", 11) ||
3170 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3171 prepend(&end, &buflen, "/", 1))
3172 end = ERR_PTR(-ENAMETOOLONG);
3175 EXPORT_SYMBOL(simple_dname);
3178 * Write full pathname from the root of the filesystem into the buffer.
3180 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3182 struct dentry *dentry;
3195 prepend(&end, &len, "\0", 1);
3199 read_seqbegin_or_lock(&rename_lock, &seq);
3200 while (!IS_ROOT(dentry)) {
3201 struct dentry *parent = dentry->d_parent;
3204 error = prepend_name(&end, &len, &dentry->d_name);
3213 if (need_seqretry(&rename_lock, seq)) {
3217 done_seqretry(&rename_lock, seq);
3222 return ERR_PTR(-ENAMETOOLONG);
3225 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3227 return __dentry_path(dentry, buf, buflen);
3229 EXPORT_SYMBOL(dentry_path_raw);
3231 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3236 if (d_unlinked(dentry)) {
3238 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3242 retval = __dentry_path(dentry, buf, buflen);
3243 if (!IS_ERR(retval) && p)
3244 *p = '/'; /* restore '/' overriden with '\0' */
3247 return ERR_PTR(-ENAMETOOLONG);
3250 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3256 seq = read_seqcount_begin(&fs->seq);
3259 } while (read_seqcount_retry(&fs->seq, seq));
3263 * NOTE! The user-level library version returns a
3264 * character pointer. The kernel system call just
3265 * returns the length of the buffer filled (which
3266 * includes the ending '\0' character), or a negative
3267 * error value. So libc would do something like
3269 * char *getcwd(char * buf, size_t size)
3273 * retval = sys_getcwd(buf, size);
3280 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3283 struct path pwd, root;
3284 char *page = __getname();
3290 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3293 if (!d_unlinked(pwd.dentry)) {
3295 char *cwd = page + PATH_MAX;
3296 int buflen = PATH_MAX;
3298 prepend(&cwd, &buflen, "\0", 1);
3299 error = prepend_path(&pwd, &root, &cwd, &buflen);
3305 /* Unreachable from current root */
3307 error = prepend_unreachable(&cwd, &buflen);
3313 len = PATH_MAX + page - cwd;
3316 if (copy_to_user(buf, cwd, len))
3329 * Test whether new_dentry is a subdirectory of old_dentry.
3331 * Trivially implemented using the dcache structure
3335 * is_subdir - is new dentry a subdirectory of old_dentry
3336 * @new_dentry: new dentry
3337 * @old_dentry: old dentry
3339 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3340 * Returns 0 otherwise.
3341 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3344 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3349 if (new_dentry == old_dentry)
3353 /* for restarting inner loop in case of seq retry */
3354 seq = read_seqbegin(&rename_lock);
3356 * Need rcu_readlock to protect against the d_parent trashing
3360 if (d_ancestor(old_dentry, new_dentry))
3365 } while (read_seqretry(&rename_lock, seq));
3370 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3372 struct dentry *root = data;
3373 if (dentry != root) {
3374 if (d_unhashed(dentry) || !dentry->d_inode)
3377 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3378 dentry->d_flags |= DCACHE_GENOCIDE;
3379 dentry->d_lockref.count--;
3382 return D_WALK_CONTINUE;
3385 void d_genocide(struct dentry *parent)
3387 d_walk(parent, parent, d_genocide_kill, NULL);
3390 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3392 inode_dec_link_count(inode);
3393 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3394 !hlist_unhashed(&dentry->d_u.d_alias) ||
3395 !d_unlinked(dentry));
3396 spin_lock(&dentry->d_parent->d_lock);
3397 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3398 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3399 (unsigned long long)inode->i_ino);
3400 spin_unlock(&dentry->d_lock);
3401 spin_unlock(&dentry->d_parent->d_lock);
3402 d_instantiate(dentry, inode);
3404 EXPORT_SYMBOL(d_tmpfile);
3406 static __initdata unsigned long dhash_entries;
3407 static int __init set_dhash_entries(char *str)
3411 dhash_entries = simple_strtoul(str, &str, 0);
3414 __setup("dhash_entries=", set_dhash_entries);
3416 static void __init dcache_init_early(void)
3420 /* If hashes are distributed across NUMA nodes, defer
3421 * hash allocation until vmalloc space is available.
3427 alloc_large_system_hash("Dentry cache",
3428 sizeof(struct hlist_bl_head),
3437 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3438 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3441 static void __init dcache_init(void)
3446 * A constructor could be added for stable state like the lists,
3447 * but it is probably not worth it because of the cache nature
3450 dentry_cache = KMEM_CACHE(dentry,
3451 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3453 /* Hash may have been set up in dcache_init_early */
3458 alloc_large_system_hash("Dentry cache",
3459 sizeof(struct hlist_bl_head),
3468 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3469 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3472 /* SLAB cache for __getname() consumers */
3473 struct kmem_cache *names_cachep __read_mostly;
3474 EXPORT_SYMBOL(names_cachep);
3476 EXPORT_SYMBOL(d_genocide);
3478 void __init vfs_caches_init_early(void)
3480 dcache_init_early();
3484 void __init vfs_caches_init(void)
3486 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3487 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3492 files_maxfiles_init();