4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <linux/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/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;
93 const struct qstr empty_name = QSTR_INIT("", 0);
94 EXPORT_SYMBOL(empty_name);
95 const struct qstr slash_name = QSTR_INIT("/", 1);
96 EXPORT_SYMBOL(slash_name);
99 * This is the single most critical data structure when it comes
100 * to the dcache: the hashtable for lookups. Somebody should try
101 * to make this good - I've just made it work.
103 * This hash-function tries to avoid losing too many bits of hash
104 * information, yet avoid using a prime hash-size or similar.
107 static unsigned int d_hash_mask __read_mostly;
108 static unsigned int d_hash_shift __read_mostly;
110 static struct hlist_bl_head *dentry_hashtable __read_mostly;
112 static inline struct hlist_bl_head *d_hash(unsigned int hash)
114 return dentry_hashtable + (hash >> (32 - d_hash_shift));
117 #define IN_LOOKUP_SHIFT 10
118 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
120 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
123 hash += (unsigned long) parent / L1_CACHE_BYTES;
124 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
128 /* Statistics gathering. */
129 struct dentry_stat_t dentry_stat = {
133 static DEFINE_PER_CPU(long, nr_dentry);
134 static DEFINE_PER_CPU(long, nr_dentry_unused);
136 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
139 * Here we resort to our own counters instead of using generic per-cpu counters
140 * for consistency with what the vfs inode code does. We are expected to harvest
141 * better code and performance by having our own specialized counters.
143 * Please note that the loop is done over all possible CPUs, not over all online
144 * CPUs. The reason for this is that we don't want to play games with CPUs going
145 * on and off. If one of them goes off, we will just keep their counters.
147 * glommer: See cffbc8a for details, and if you ever intend to change this,
148 * please update all vfs counters to match.
150 static long get_nr_dentry(void)
154 for_each_possible_cpu(i)
155 sum += per_cpu(nr_dentry, i);
156 return sum < 0 ? 0 : sum;
159 static long get_nr_dentry_unused(void)
163 for_each_possible_cpu(i)
164 sum += per_cpu(nr_dentry_unused, i);
165 return sum < 0 ? 0 : sum;
168 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
169 size_t *lenp, loff_t *ppos)
171 dentry_stat.nr_dentry = get_nr_dentry();
172 dentry_stat.nr_unused = get_nr_dentry_unused();
173 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
178 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
179 * The strings are both count bytes long, and count is non-zero.
181 #ifdef CONFIG_DCACHE_WORD_ACCESS
183 #include <asm/word-at-a-time.h>
185 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
186 * aligned allocation for this particular component. We don't
187 * strictly need the load_unaligned_zeropad() safety, but it
188 * doesn't hurt either.
190 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
191 * need the careful unaligned handling.
193 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
195 unsigned long a,b,mask;
198 a = *(unsigned long *)cs;
199 b = load_unaligned_zeropad(ct);
200 if (tcount < sizeof(unsigned long))
202 if (unlikely(a != b))
204 cs += sizeof(unsigned long);
205 ct += sizeof(unsigned long);
206 tcount -= sizeof(unsigned long);
210 mask = bytemask_from_count(tcount);
211 return unlikely(!!((a ^ b) & mask));
216 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
230 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
233 * Be careful about RCU walk racing with rename:
234 * use 'READ_ONCE' to fetch the name pointer.
236 * NOTE! Even if a rename will mean that the length
237 * was not loaded atomically, we don't care. The
238 * RCU walk will check the sequence count eventually,
239 * and catch it. And we won't overrun the buffer,
240 * because we're reading the name pointer atomically,
241 * and a dentry name is guaranteed to be properly
242 * terminated with a NUL byte.
244 * End result: even if 'len' is wrong, we'll exit
245 * early because the data cannot match (there can
246 * be no NUL in the ct/tcount data)
248 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
250 return dentry_string_cmp(cs, ct, tcount);
253 struct external_name {
256 struct rcu_head head;
258 unsigned char name[];
261 static inline struct external_name *external_name(struct dentry *dentry)
263 return container_of(dentry->d_name.name, struct external_name, name[0]);
266 static void __d_free(struct rcu_head *head)
268 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
270 kmem_cache_free(dentry_cache, dentry);
273 static void __d_free_external(struct rcu_head *head)
275 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
276 kfree(external_name(dentry));
277 kmem_cache_free(dentry_cache, dentry);
280 static inline int dname_external(const struct dentry *dentry)
282 return dentry->d_name.name != dentry->d_iname;
285 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
287 spin_lock(&dentry->d_lock);
288 if (unlikely(dname_external(dentry))) {
289 struct external_name *p = external_name(dentry);
290 atomic_inc(&p->u.count);
291 spin_unlock(&dentry->d_lock);
292 name->name = p->name;
294 memcpy(name->inline_name, dentry->d_iname, DNAME_INLINE_LEN);
295 spin_unlock(&dentry->d_lock);
296 name->name = name->inline_name;
299 EXPORT_SYMBOL(take_dentry_name_snapshot);
301 void release_dentry_name_snapshot(struct name_snapshot *name)
303 if (unlikely(name->name != name->inline_name)) {
304 struct external_name *p;
305 p = container_of(name->name, struct external_name, name[0]);
306 if (unlikely(atomic_dec_and_test(&p->u.count)))
307 kfree_rcu(p, u.head);
310 EXPORT_SYMBOL(release_dentry_name_snapshot);
312 static inline void __d_set_inode_and_type(struct dentry *dentry,
318 dentry->d_inode = inode;
319 flags = READ_ONCE(dentry->d_flags);
320 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
322 WRITE_ONCE(dentry->d_flags, flags);
325 static inline void __d_clear_type_and_inode(struct dentry *dentry)
327 unsigned flags = READ_ONCE(dentry->d_flags);
329 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
330 WRITE_ONCE(dentry->d_flags, flags);
331 dentry->d_inode = NULL;
334 static void dentry_free(struct dentry *dentry)
336 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
337 if (unlikely(dname_external(dentry))) {
338 struct external_name *p = external_name(dentry);
339 if (likely(atomic_dec_and_test(&p->u.count))) {
340 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
344 /* if dentry was never visible to RCU, immediate free is OK */
345 if (!(dentry->d_flags & DCACHE_RCUACCESS))
346 __d_free(&dentry->d_u.d_rcu);
348 call_rcu(&dentry->d_u.d_rcu, __d_free);
352 * Release the dentry's inode, using the filesystem
353 * d_iput() operation if defined.
355 static void dentry_unlink_inode(struct dentry * dentry)
356 __releases(dentry->d_lock)
357 __releases(dentry->d_inode->i_lock)
359 struct inode *inode = dentry->d_inode;
360 bool hashed = !d_unhashed(dentry);
363 raw_write_seqcount_begin(&dentry->d_seq);
364 __d_clear_type_and_inode(dentry);
365 hlist_del_init(&dentry->d_u.d_alias);
367 raw_write_seqcount_end(&dentry->d_seq);
368 spin_unlock(&dentry->d_lock);
369 spin_unlock(&inode->i_lock);
371 fsnotify_inoderemove(inode);
372 if (dentry->d_op && dentry->d_op->d_iput)
373 dentry->d_op->d_iput(dentry, inode);
379 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
380 * is in use - which includes both the "real" per-superblock
381 * LRU list _and_ the DCACHE_SHRINK_LIST use.
383 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
384 * on the shrink list (ie not on the superblock LRU list).
386 * The per-cpu "nr_dentry_unused" counters are updated with
387 * the DCACHE_LRU_LIST bit.
389 * These helper functions make sure we always follow the
390 * rules. d_lock must be held by the caller.
392 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
393 static void d_lru_add(struct dentry *dentry)
395 D_FLAG_VERIFY(dentry, 0);
396 dentry->d_flags |= DCACHE_LRU_LIST;
397 this_cpu_inc(nr_dentry_unused);
398 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
401 static void d_lru_del(struct dentry *dentry)
403 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
404 dentry->d_flags &= ~DCACHE_LRU_LIST;
405 this_cpu_dec(nr_dentry_unused);
406 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
409 static void d_shrink_del(struct dentry *dentry)
411 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
412 list_del_init(&dentry->d_lru);
413 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
414 this_cpu_dec(nr_dentry_unused);
417 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
419 D_FLAG_VERIFY(dentry, 0);
420 list_add(&dentry->d_lru, list);
421 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
422 this_cpu_inc(nr_dentry_unused);
426 * These can only be called under the global LRU lock, ie during the
427 * callback for freeing the LRU list. "isolate" removes it from the
428 * LRU lists entirely, while shrink_move moves it to the indicated
431 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
433 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
434 dentry->d_flags &= ~DCACHE_LRU_LIST;
435 this_cpu_dec(nr_dentry_unused);
436 list_lru_isolate(lru, &dentry->d_lru);
439 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
440 struct list_head *list)
442 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
443 dentry->d_flags |= DCACHE_SHRINK_LIST;
444 list_lru_isolate_move(lru, &dentry->d_lru, list);
448 * dentry_lru_(add|del)_list) must be called with d_lock held.
450 static void dentry_lru_add(struct dentry *dentry)
452 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
454 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
455 dentry->d_flags |= DCACHE_REFERENCED;
459 * d_drop - drop a dentry
460 * @dentry: dentry to drop
462 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
463 * be found through a VFS lookup any more. Note that this is different from
464 * deleting the dentry - d_delete will try to mark the dentry negative if
465 * possible, giving a successful _negative_ lookup, while d_drop will
466 * just make the cache lookup fail.
468 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
469 * reason (NFS timeouts or autofs deletes).
471 * __d_drop requires dentry->d_lock.
473 void __d_drop(struct dentry *dentry)
475 if (!d_unhashed(dentry)) {
476 struct hlist_bl_head *b;
478 * Hashed dentries are normally on the dentry hashtable,
479 * with the exception of those newly allocated by
480 * d_obtain_alias, which are always IS_ROOT:
482 if (unlikely(IS_ROOT(dentry)))
483 b = &dentry->d_sb->s_anon;
485 b = d_hash(dentry->d_name.hash);
488 __hlist_bl_del(&dentry->d_hash);
489 dentry->d_hash.pprev = NULL;
491 /* After this call, in-progress rcu-walk path lookup will fail. */
492 write_seqcount_invalidate(&dentry->d_seq);
495 EXPORT_SYMBOL(__d_drop);
497 void d_drop(struct dentry *dentry)
499 spin_lock(&dentry->d_lock);
501 spin_unlock(&dentry->d_lock);
503 EXPORT_SYMBOL(d_drop);
505 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
509 * Inform d_walk() and shrink_dentry_list() that we are no longer
510 * attached to the dentry tree
512 dentry->d_flags |= DCACHE_DENTRY_KILLED;
513 if (unlikely(list_empty(&dentry->d_child)))
515 __list_del_entry(&dentry->d_child);
517 * Cursors can move around the list of children. While we'd been
518 * a normal list member, it didn't matter - ->d_child.next would've
519 * been updated. However, from now on it won't be and for the
520 * things like d_walk() it might end up with a nasty surprise.
521 * Normally d_walk() doesn't care about cursors moving around -
522 * ->d_lock on parent prevents that and since a cursor has no children
523 * of its own, we get through it without ever unlocking the parent.
524 * There is one exception, though - if we ascend from a child that
525 * gets killed as soon as we unlock it, the next sibling is found
526 * using the value left in its ->d_child.next. And if _that_
527 * pointed to a cursor, and cursor got moved (e.g. by lseek())
528 * before d_walk() regains parent->d_lock, we'll end up skipping
529 * everything the cursor had been moved past.
531 * Solution: make sure that the pointer left behind in ->d_child.next
532 * points to something that won't be moving around. I.e. skip the
535 while (dentry->d_child.next != &parent->d_subdirs) {
536 next = list_entry(dentry->d_child.next, struct dentry, d_child);
537 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
539 dentry->d_child.next = next->d_child.next;
543 static void __dentry_kill(struct dentry *dentry)
545 struct dentry *parent = NULL;
546 bool can_free = true;
547 if (!IS_ROOT(dentry))
548 parent = dentry->d_parent;
551 * The dentry is now unrecoverably dead to the world.
553 lockref_mark_dead(&dentry->d_lockref);
556 * inform the fs via d_prune that this dentry is about to be
557 * unhashed and destroyed.
559 if (dentry->d_flags & DCACHE_OP_PRUNE)
560 dentry->d_op->d_prune(dentry);
562 if (dentry->d_flags & DCACHE_LRU_LIST) {
563 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
566 /* if it was on the hash then remove it */
568 dentry_unlist(dentry, parent);
570 spin_unlock(&parent->d_lock);
572 dentry_unlink_inode(dentry);
574 spin_unlock(&dentry->d_lock);
575 this_cpu_dec(nr_dentry);
576 if (dentry->d_op && dentry->d_op->d_release)
577 dentry->d_op->d_release(dentry);
579 spin_lock(&dentry->d_lock);
580 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
581 dentry->d_flags |= DCACHE_MAY_FREE;
584 spin_unlock(&dentry->d_lock);
585 if (likely(can_free))
590 * Finish off a dentry we've decided to kill.
591 * dentry->d_lock must be held, returns with it unlocked.
592 * If ref is non-zero, then decrement the refcount too.
593 * Returns dentry requiring refcount drop, or NULL if we're done.
595 static struct dentry *dentry_kill(struct dentry *dentry)
596 __releases(dentry->d_lock)
598 struct inode *inode = dentry->d_inode;
599 struct dentry *parent = NULL;
601 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
604 if (!IS_ROOT(dentry)) {
605 parent = dentry->d_parent;
606 if (unlikely(!spin_trylock(&parent->d_lock))) {
608 spin_unlock(&inode->i_lock);
613 __dentry_kill(dentry);
617 spin_unlock(&dentry->d_lock);
618 return dentry; /* try again with same dentry */
621 static inline struct dentry *lock_parent(struct dentry *dentry)
623 struct dentry *parent = dentry->d_parent;
626 if (unlikely(dentry->d_lockref.count < 0))
628 if (likely(spin_trylock(&parent->d_lock)))
631 spin_unlock(&dentry->d_lock);
633 parent = READ_ONCE(dentry->d_parent);
634 spin_lock(&parent->d_lock);
636 * We can't blindly lock dentry until we are sure
637 * that we won't violate the locking order.
638 * Any changes of dentry->d_parent must have
639 * been done with parent->d_lock held, so
640 * spin_lock() above is enough of a barrier
641 * for checking if it's still our child.
643 if (unlikely(parent != dentry->d_parent)) {
644 spin_unlock(&parent->d_lock);
648 if (parent != dentry)
649 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
656 * Try to do a lockless dput(), and return whether that was successful.
658 * If unsuccessful, we return false, having already taken the dentry lock.
660 * The caller needs to hold the RCU read lock, so that the dentry is
661 * guaranteed to stay around even if the refcount goes down to zero!
663 static inline bool fast_dput(struct dentry *dentry)
666 unsigned int d_flags;
669 * If we have a d_op->d_delete() operation, we sould not
670 * let the dentry count go to zero, so use "put_or_lock".
672 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
673 return lockref_put_or_lock(&dentry->d_lockref);
676 * .. otherwise, we can try to just decrement the
677 * lockref optimistically.
679 ret = lockref_put_return(&dentry->d_lockref);
682 * If the lockref_put_return() failed due to the lock being held
683 * by somebody else, the fast path has failed. We will need to
684 * get the lock, and then check the count again.
686 if (unlikely(ret < 0)) {
687 spin_lock(&dentry->d_lock);
688 if (dentry->d_lockref.count > 1) {
689 dentry->d_lockref.count--;
690 spin_unlock(&dentry->d_lock);
697 * If we weren't the last ref, we're done.
703 * Careful, careful. The reference count went down
704 * to zero, but we don't hold the dentry lock, so
705 * somebody else could get it again, and do another
706 * dput(), and we need to not race with that.
708 * However, there is a very special and common case
709 * where we don't care, because there is nothing to
710 * do: the dentry is still hashed, it does not have
711 * a 'delete' op, and it's referenced and already on
714 * NOTE! Since we aren't locked, these values are
715 * not "stable". However, it is sufficient that at
716 * some point after we dropped the reference the
717 * dentry was hashed and the flags had the proper
718 * value. Other dentry users may have re-gotten
719 * a reference to the dentry and change that, but
720 * our work is done - we can leave the dentry
721 * around with a zero refcount.
724 d_flags = READ_ONCE(dentry->d_flags);
725 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
727 /* Nothing to do? Dropping the reference was all we needed? */
728 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
732 * Not the fast normal case? Get the lock. We've already decremented
733 * the refcount, but we'll need to re-check the situation after
736 spin_lock(&dentry->d_lock);
739 * Did somebody else grab a reference to it in the meantime, and
740 * we're no longer the last user after all? Alternatively, somebody
741 * else could have killed it and marked it dead. Either way, we
742 * don't need to do anything else.
744 if (dentry->d_lockref.count) {
745 spin_unlock(&dentry->d_lock);
750 * Re-get the reference we optimistically dropped. We hold the
751 * lock, and we just tested that it was zero, so we can just
754 dentry->d_lockref.count = 1;
762 * This is complicated by the fact that we do not want to put
763 * dentries that are no longer on any hash chain on the unused
764 * list: we'd much rather just get rid of them immediately.
766 * However, that implies that we have to traverse the dentry
767 * tree upwards to the parents which might _also_ now be
768 * scheduled for deletion (it may have been only waiting for
769 * its last child to go away).
771 * This tail recursion is done by hand as we don't want to depend
772 * on the compiler to always get this right (gcc generally doesn't).
773 * Real recursion would eat up our stack space.
777 * dput - release a dentry
778 * @dentry: dentry to release
780 * Release a dentry. This will drop the usage count and if appropriate
781 * call the dentry unlink method as well as removing it from the queues and
782 * releasing its resources. If the parent dentries were scheduled for release
783 * they too may now get deleted.
785 void dput(struct dentry *dentry)
787 if (unlikely(!dentry))
794 if (likely(fast_dput(dentry))) {
799 /* Slow case: now with the dentry lock held */
802 WARN_ON(d_in_lookup(dentry));
804 /* Unreachable? Get rid of it */
805 if (unlikely(d_unhashed(dentry)))
808 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
811 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
812 if (dentry->d_op->d_delete(dentry))
816 dentry_lru_add(dentry);
818 dentry->d_lockref.count--;
819 spin_unlock(&dentry->d_lock);
823 dentry = dentry_kill(dentry);
832 /* This must be called with d_lock held */
833 static inline void __dget_dlock(struct dentry *dentry)
835 dentry->d_lockref.count++;
838 static inline void __dget(struct dentry *dentry)
840 lockref_get(&dentry->d_lockref);
843 struct dentry *dget_parent(struct dentry *dentry)
849 * Do optimistic parent lookup without any
853 ret = READ_ONCE(dentry->d_parent);
854 gotref = lockref_get_not_zero(&ret->d_lockref);
856 if (likely(gotref)) {
857 if (likely(ret == READ_ONCE(dentry->d_parent)))
864 * Don't need rcu_dereference because we re-check it was correct under
868 ret = dentry->d_parent;
869 spin_lock(&ret->d_lock);
870 if (unlikely(ret != dentry->d_parent)) {
871 spin_unlock(&ret->d_lock);
876 BUG_ON(!ret->d_lockref.count);
877 ret->d_lockref.count++;
878 spin_unlock(&ret->d_lock);
881 EXPORT_SYMBOL(dget_parent);
884 * d_find_alias - grab a hashed alias of inode
885 * @inode: inode in question
887 * If inode has a hashed alias, or is a directory and has any alias,
888 * acquire the reference to alias and return it. Otherwise return NULL.
889 * Notice that if inode is a directory there can be only one alias and
890 * it can be unhashed only if it has no children, or if it is the root
891 * of a filesystem, or if the directory was renamed and d_revalidate
892 * was the first vfs operation to notice.
894 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
895 * any other hashed alias over that one.
897 static struct dentry *__d_find_alias(struct inode *inode)
899 struct dentry *alias, *discon_alias;
903 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
904 spin_lock(&alias->d_lock);
905 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
906 if (IS_ROOT(alias) &&
907 (alias->d_flags & DCACHE_DISCONNECTED)) {
908 discon_alias = alias;
911 spin_unlock(&alias->d_lock);
915 spin_unlock(&alias->d_lock);
918 alias = discon_alias;
919 spin_lock(&alias->d_lock);
920 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
922 spin_unlock(&alias->d_lock);
925 spin_unlock(&alias->d_lock);
931 struct dentry *d_find_alias(struct inode *inode)
933 struct dentry *de = NULL;
935 if (!hlist_empty(&inode->i_dentry)) {
936 spin_lock(&inode->i_lock);
937 de = __d_find_alias(inode);
938 spin_unlock(&inode->i_lock);
942 EXPORT_SYMBOL(d_find_alias);
945 * Try to kill dentries associated with this inode.
946 * WARNING: you must own a reference to inode.
948 void d_prune_aliases(struct inode *inode)
950 struct dentry *dentry;
952 spin_lock(&inode->i_lock);
953 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
954 spin_lock(&dentry->d_lock);
955 if (!dentry->d_lockref.count) {
956 struct dentry *parent = lock_parent(dentry);
957 if (likely(!dentry->d_lockref.count)) {
958 __dentry_kill(dentry);
963 spin_unlock(&parent->d_lock);
965 spin_unlock(&dentry->d_lock);
967 spin_unlock(&inode->i_lock);
969 EXPORT_SYMBOL(d_prune_aliases);
971 static void shrink_dentry_list(struct list_head *list)
973 struct dentry *dentry, *parent;
975 while (!list_empty(list)) {
977 dentry = list_entry(list->prev, struct dentry, d_lru);
978 spin_lock(&dentry->d_lock);
979 parent = lock_parent(dentry);
982 * The dispose list is isolated and dentries are not accounted
983 * to the LRU here, so we can simply remove it from the list
984 * here regardless of whether it is referenced or not.
986 d_shrink_del(dentry);
989 * We found an inuse dentry which was not removed from
990 * the LRU because of laziness during lookup. Do not free it.
992 if (dentry->d_lockref.count > 0) {
993 spin_unlock(&dentry->d_lock);
995 spin_unlock(&parent->d_lock);
1000 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1001 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1002 spin_unlock(&dentry->d_lock);
1004 spin_unlock(&parent->d_lock);
1006 dentry_free(dentry);
1010 inode = dentry->d_inode;
1011 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1012 d_shrink_add(dentry, list);
1013 spin_unlock(&dentry->d_lock);
1015 spin_unlock(&parent->d_lock);
1019 __dentry_kill(dentry);
1022 * We need to prune ancestors too. This is necessary to prevent
1023 * quadratic behavior of shrink_dcache_parent(), but is also
1024 * expected to be beneficial in reducing dentry cache
1028 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1029 parent = lock_parent(dentry);
1030 if (dentry->d_lockref.count != 1) {
1031 dentry->d_lockref.count--;
1032 spin_unlock(&dentry->d_lock);
1034 spin_unlock(&parent->d_lock);
1037 inode = dentry->d_inode; /* can't be NULL */
1038 if (unlikely(!spin_trylock(&inode->i_lock))) {
1039 spin_unlock(&dentry->d_lock);
1041 spin_unlock(&parent->d_lock);
1045 __dentry_kill(dentry);
1051 static enum lru_status dentry_lru_isolate(struct list_head *item,
1052 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1054 struct list_head *freeable = arg;
1055 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1059 * we are inverting the lru lock/dentry->d_lock here,
1060 * so use a trylock. If we fail to get the lock, just skip
1063 if (!spin_trylock(&dentry->d_lock))
1067 * Referenced dentries are still in use. If they have active
1068 * counts, just remove them from the LRU. Otherwise give them
1069 * another pass through the LRU.
1071 if (dentry->d_lockref.count) {
1072 d_lru_isolate(lru, dentry);
1073 spin_unlock(&dentry->d_lock);
1077 if (dentry->d_flags & DCACHE_REFERENCED) {
1078 dentry->d_flags &= ~DCACHE_REFERENCED;
1079 spin_unlock(&dentry->d_lock);
1082 * The list move itself will be made by the common LRU code. At
1083 * this point, we've dropped the dentry->d_lock but keep the
1084 * lru lock. This is safe to do, since every list movement is
1085 * protected by the lru lock even if both locks are held.
1087 * This is guaranteed by the fact that all LRU management
1088 * functions are intermediated by the LRU API calls like
1089 * list_lru_add and list_lru_del. List movement in this file
1090 * only ever occur through this functions or through callbacks
1091 * like this one, that are called from the LRU API.
1093 * The only exceptions to this are functions like
1094 * shrink_dentry_list, and code that first checks for the
1095 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1096 * operating only with stack provided lists after they are
1097 * properly isolated from the main list. It is thus, always a
1103 d_lru_shrink_move(lru, dentry, freeable);
1104 spin_unlock(&dentry->d_lock);
1110 * prune_dcache_sb - shrink the dcache
1112 * @sc: shrink control, passed to list_lru_shrink_walk()
1114 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1115 * is done when we need more memory and called from the superblock shrinker
1118 * This function may fail to free any resources if all the dentries are in
1121 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1126 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1127 dentry_lru_isolate, &dispose);
1128 shrink_dentry_list(&dispose);
1132 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1133 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1135 struct list_head *freeable = arg;
1136 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1139 * we are inverting the lru lock/dentry->d_lock here,
1140 * so use a trylock. If we fail to get the lock, just skip
1143 if (!spin_trylock(&dentry->d_lock))
1146 d_lru_shrink_move(lru, dentry, freeable);
1147 spin_unlock(&dentry->d_lock);
1154 * shrink_dcache_sb - shrink dcache for a superblock
1157 * Shrink the dcache for the specified super block. This is used to free
1158 * the dcache before unmounting a file system.
1160 void shrink_dcache_sb(struct super_block *sb)
1167 freed = list_lru_walk(&sb->s_dentry_lru,
1168 dentry_lru_isolate_shrink, &dispose, 1024);
1170 this_cpu_sub(nr_dentry_unused, freed);
1171 shrink_dentry_list(&dispose);
1173 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1175 EXPORT_SYMBOL(shrink_dcache_sb);
1178 * enum d_walk_ret - action to talke during tree walk
1179 * @D_WALK_CONTINUE: contrinue walk
1180 * @D_WALK_QUIT: quit walk
1181 * @D_WALK_NORETRY: quit when retry is needed
1182 * @D_WALK_SKIP: skip this dentry and its children
1192 * d_walk - walk the dentry tree
1193 * @parent: start of walk
1194 * @data: data passed to @enter() and @finish()
1195 * @enter: callback when first entering the dentry
1196 * @finish: callback when successfully finished the walk
1198 * The @enter() and @finish() callbacks are called with d_lock held.
1200 static void d_walk(struct dentry *parent, void *data,
1201 enum d_walk_ret (*enter)(void *, struct dentry *),
1202 void (*finish)(void *))
1204 struct dentry *this_parent;
1205 struct list_head *next;
1207 enum d_walk_ret ret;
1211 read_seqbegin_or_lock(&rename_lock, &seq);
1212 this_parent = parent;
1213 spin_lock(&this_parent->d_lock);
1215 ret = enter(data, this_parent);
1217 case D_WALK_CONTINUE:
1222 case D_WALK_NORETRY:
1227 next = this_parent->d_subdirs.next;
1229 while (next != &this_parent->d_subdirs) {
1230 struct list_head *tmp = next;
1231 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1234 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1237 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1239 ret = enter(data, dentry);
1241 case D_WALK_CONTINUE:
1244 spin_unlock(&dentry->d_lock);
1246 case D_WALK_NORETRY:
1250 spin_unlock(&dentry->d_lock);
1254 if (!list_empty(&dentry->d_subdirs)) {
1255 spin_unlock(&this_parent->d_lock);
1256 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1257 this_parent = dentry;
1258 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1261 spin_unlock(&dentry->d_lock);
1264 * All done at this level ... ascend and resume the search.
1268 if (this_parent != parent) {
1269 struct dentry *child = this_parent;
1270 this_parent = child->d_parent;
1272 spin_unlock(&child->d_lock);
1273 spin_lock(&this_parent->d_lock);
1275 /* might go back up the wrong parent if we have had a rename. */
1276 if (need_seqretry(&rename_lock, seq))
1278 /* go into the first sibling still alive */
1280 next = child->d_child.next;
1281 if (next == &this_parent->d_subdirs)
1283 child = list_entry(next, struct dentry, d_child);
1284 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1288 if (need_seqretry(&rename_lock, seq))
1295 spin_unlock(&this_parent->d_lock);
1296 done_seqretry(&rename_lock, seq);
1300 spin_unlock(&this_parent->d_lock);
1309 struct check_mount {
1310 struct vfsmount *mnt;
1311 unsigned int mounted;
1314 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1316 struct check_mount *info = data;
1317 struct path path = { .mnt = info->mnt, .dentry = dentry };
1319 if (likely(!d_mountpoint(dentry)))
1320 return D_WALK_CONTINUE;
1321 if (__path_is_mountpoint(&path)) {
1325 return D_WALK_CONTINUE;
1329 * path_has_submounts - check for mounts over a dentry in the
1330 * current namespace.
1331 * @parent: path to check.
1333 * Return true if the parent or its subdirectories contain
1334 * a mount point in the current namespace.
1336 int path_has_submounts(const struct path *parent)
1338 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1340 read_seqlock_excl(&mount_lock);
1341 d_walk(parent->dentry, &data, path_check_mount, NULL);
1342 read_sequnlock_excl(&mount_lock);
1344 return data.mounted;
1346 EXPORT_SYMBOL(path_has_submounts);
1349 * Called by mount code to set a mountpoint and check if the mountpoint is
1350 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1351 * subtree can become unreachable).
1353 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1354 * this reason take rename_lock and d_lock on dentry and ancestors.
1356 int d_set_mounted(struct dentry *dentry)
1360 write_seqlock(&rename_lock);
1361 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1362 /* Need exclusion wrt. d_invalidate() */
1363 spin_lock(&p->d_lock);
1364 if (unlikely(d_unhashed(p))) {
1365 spin_unlock(&p->d_lock);
1368 spin_unlock(&p->d_lock);
1370 spin_lock(&dentry->d_lock);
1371 if (!d_unlinked(dentry)) {
1373 if (!d_mountpoint(dentry)) {
1374 dentry->d_flags |= DCACHE_MOUNTED;
1378 spin_unlock(&dentry->d_lock);
1380 write_sequnlock(&rename_lock);
1385 * Search the dentry child list of the specified parent,
1386 * and move any unused dentries to the end of the unused
1387 * list for prune_dcache(). We descend to the next level
1388 * whenever the d_subdirs list is non-empty and continue
1391 * It returns zero iff there are no unused children,
1392 * otherwise it returns the number of children moved to
1393 * the end of the unused list. This may not be the total
1394 * number of unused children, because select_parent can
1395 * drop the lock and return early due to latency
1399 struct select_data {
1400 struct dentry *start;
1401 struct list_head dispose;
1405 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1407 struct select_data *data = _data;
1408 enum d_walk_ret ret = D_WALK_CONTINUE;
1410 if (data->start == dentry)
1413 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1416 if (dentry->d_flags & DCACHE_LRU_LIST)
1418 if (!dentry->d_lockref.count) {
1419 d_shrink_add(dentry, &data->dispose);
1424 * We can return to the caller if we have found some (this
1425 * ensures forward progress). We'll be coming back to find
1428 if (!list_empty(&data->dispose))
1429 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1435 * shrink_dcache_parent - prune dcache
1436 * @parent: parent of entries to prune
1438 * Prune the dcache to remove unused children of the parent dentry.
1440 void shrink_dcache_parent(struct dentry *parent)
1443 struct select_data data;
1445 INIT_LIST_HEAD(&data.dispose);
1446 data.start = parent;
1449 d_walk(parent, &data, select_collect, NULL);
1453 shrink_dentry_list(&data.dispose);
1457 EXPORT_SYMBOL(shrink_dcache_parent);
1459 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1461 /* it has busy descendents; complain about those instead */
1462 if (!list_empty(&dentry->d_subdirs))
1463 return D_WALK_CONTINUE;
1465 /* root with refcount 1 is fine */
1466 if (dentry == _data && dentry->d_lockref.count == 1)
1467 return D_WALK_CONTINUE;
1469 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1470 " still in use (%d) [unmount of %s %s]\n",
1473 dentry->d_inode->i_ino : 0UL,
1475 dentry->d_lockref.count,
1476 dentry->d_sb->s_type->name,
1477 dentry->d_sb->s_id);
1479 return D_WALK_CONTINUE;
1482 static void do_one_tree(struct dentry *dentry)
1484 shrink_dcache_parent(dentry);
1485 d_walk(dentry, dentry, umount_check, NULL);
1491 * destroy the dentries attached to a superblock on unmounting
1493 void shrink_dcache_for_umount(struct super_block *sb)
1495 struct dentry *dentry;
1497 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1499 dentry = sb->s_root;
1501 do_one_tree(dentry);
1503 while (!hlist_bl_empty(&sb->s_anon)) {
1504 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1505 do_one_tree(dentry);
1509 struct detach_data {
1510 struct select_data select;
1511 struct dentry *mountpoint;
1513 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1515 struct detach_data *data = _data;
1517 if (d_mountpoint(dentry)) {
1518 __dget_dlock(dentry);
1519 data->mountpoint = dentry;
1523 return select_collect(&data->select, dentry);
1526 static void check_and_drop(void *_data)
1528 struct detach_data *data = _data;
1530 if (!data->mountpoint && list_empty(&data->select.dispose))
1531 __d_drop(data->select.start);
1535 * d_invalidate - detach submounts, prune dcache, and drop
1536 * @dentry: dentry to invalidate (aka detach, prune and drop)
1540 * The final d_drop is done as an atomic operation relative to
1541 * rename_lock ensuring there are no races with d_set_mounted. This
1542 * ensures there are no unhashed dentries on the path to a mountpoint.
1544 void d_invalidate(struct dentry *dentry)
1547 * If it's already been dropped, return OK.
1549 spin_lock(&dentry->d_lock);
1550 if (d_unhashed(dentry)) {
1551 spin_unlock(&dentry->d_lock);
1554 spin_unlock(&dentry->d_lock);
1556 /* Negative dentries can be dropped without further checks */
1557 if (!dentry->d_inode) {
1563 struct detach_data data;
1565 data.mountpoint = NULL;
1566 INIT_LIST_HEAD(&data.select.dispose);
1567 data.select.start = dentry;
1568 data.select.found = 0;
1570 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1572 if (!list_empty(&data.select.dispose))
1573 shrink_dentry_list(&data.select.dispose);
1574 else if (!data.mountpoint)
1577 if (data.mountpoint) {
1578 detach_mounts(data.mountpoint);
1579 dput(data.mountpoint);
1584 EXPORT_SYMBOL(d_invalidate);
1587 * __d_alloc - allocate a dcache entry
1588 * @sb: filesystem it will belong to
1589 * @name: qstr of the name
1591 * Allocates a dentry. It returns %NULL if there is insufficient memory
1592 * available. On a success the dentry is returned. The name passed in is
1593 * copied and the copy passed in may be reused after this call.
1596 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1598 struct dentry *dentry;
1602 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1607 * We guarantee that the inline name is always NUL-terminated.
1608 * This way the memcpy() done by the name switching in rename
1609 * will still always have a NUL at the end, even if we might
1610 * be overwriting an internal NUL character
1612 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1613 if (unlikely(!name)) {
1615 dname = dentry->d_iname;
1616 } else if (name->len > DNAME_INLINE_LEN-1) {
1617 size_t size = offsetof(struct external_name, name[1]);
1618 struct external_name *p = kmalloc(size + name->len,
1619 GFP_KERNEL_ACCOUNT);
1621 kmem_cache_free(dentry_cache, dentry);
1624 atomic_set(&p->u.count, 1);
1626 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1627 kasan_unpoison_shadow(dname,
1628 round_up(name->len + 1, sizeof(unsigned long)));
1630 dname = dentry->d_iname;
1633 dentry->d_name.len = name->len;
1634 dentry->d_name.hash = name->hash;
1635 memcpy(dname, name->name, name->len);
1636 dname[name->len] = 0;
1638 /* Make sure we always see the terminating NUL character */
1639 smp_store_release(&dentry->d_name.name, dname); /* ^^^ */
1641 dentry->d_lockref.count = 1;
1642 dentry->d_flags = 0;
1643 spin_lock_init(&dentry->d_lock);
1644 seqcount_init(&dentry->d_seq);
1645 dentry->d_inode = NULL;
1646 dentry->d_parent = dentry;
1648 dentry->d_op = NULL;
1649 dentry->d_fsdata = NULL;
1650 INIT_HLIST_BL_NODE(&dentry->d_hash);
1651 INIT_LIST_HEAD(&dentry->d_lru);
1652 INIT_LIST_HEAD(&dentry->d_subdirs);
1653 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1654 INIT_LIST_HEAD(&dentry->d_child);
1655 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1657 if (dentry->d_op && dentry->d_op->d_init) {
1658 err = dentry->d_op->d_init(dentry);
1660 if (dname_external(dentry))
1661 kfree(external_name(dentry));
1662 kmem_cache_free(dentry_cache, dentry);
1667 this_cpu_inc(nr_dentry);
1673 * d_alloc - allocate a dcache entry
1674 * @parent: parent of entry to allocate
1675 * @name: qstr of the name
1677 * Allocates a dentry. It returns %NULL if there is insufficient memory
1678 * available. On a success the dentry is returned. The name passed in is
1679 * copied and the copy passed in may be reused after this call.
1681 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1683 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1686 dentry->d_flags |= DCACHE_RCUACCESS;
1687 spin_lock(&parent->d_lock);
1689 * don't need child lock because it is not subject
1690 * to concurrency here
1692 __dget_dlock(parent);
1693 dentry->d_parent = parent;
1694 list_add(&dentry->d_child, &parent->d_subdirs);
1695 spin_unlock(&parent->d_lock);
1699 EXPORT_SYMBOL(d_alloc);
1701 struct dentry *d_alloc_cursor(struct dentry * parent)
1703 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1705 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1706 dentry->d_parent = dget(parent);
1712 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1713 * @sb: the superblock
1714 * @name: qstr of the name
1716 * For a filesystem that just pins its dentries in memory and never
1717 * performs lookups at all, return an unhashed IS_ROOT dentry.
1719 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1721 return __d_alloc(sb, name);
1723 EXPORT_SYMBOL(d_alloc_pseudo);
1725 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1730 q.hash_len = hashlen_string(parent, name);
1731 return d_alloc(parent, &q);
1733 EXPORT_SYMBOL(d_alloc_name);
1735 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1737 WARN_ON_ONCE(dentry->d_op);
1738 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1740 DCACHE_OP_REVALIDATE |
1741 DCACHE_OP_WEAK_REVALIDATE |
1748 dentry->d_flags |= DCACHE_OP_HASH;
1750 dentry->d_flags |= DCACHE_OP_COMPARE;
1751 if (op->d_revalidate)
1752 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1753 if (op->d_weak_revalidate)
1754 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1756 dentry->d_flags |= DCACHE_OP_DELETE;
1758 dentry->d_flags |= DCACHE_OP_PRUNE;
1760 dentry->d_flags |= DCACHE_OP_REAL;
1763 EXPORT_SYMBOL(d_set_d_op);
1767 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1768 * @dentry - The dentry to mark
1770 * Mark a dentry as falling through to the lower layer (as set with
1771 * d_pin_lower()). This flag may be recorded on the medium.
1773 void d_set_fallthru(struct dentry *dentry)
1775 spin_lock(&dentry->d_lock);
1776 dentry->d_flags |= DCACHE_FALLTHRU;
1777 spin_unlock(&dentry->d_lock);
1779 EXPORT_SYMBOL(d_set_fallthru);
1781 static unsigned d_flags_for_inode(struct inode *inode)
1783 unsigned add_flags = DCACHE_REGULAR_TYPE;
1786 return DCACHE_MISS_TYPE;
1788 if (S_ISDIR(inode->i_mode)) {
1789 add_flags = DCACHE_DIRECTORY_TYPE;
1790 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1791 if (unlikely(!inode->i_op->lookup))
1792 add_flags = DCACHE_AUTODIR_TYPE;
1794 inode->i_opflags |= IOP_LOOKUP;
1796 goto type_determined;
1799 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1800 if (unlikely(inode->i_op->get_link)) {
1801 add_flags = DCACHE_SYMLINK_TYPE;
1802 goto type_determined;
1804 inode->i_opflags |= IOP_NOFOLLOW;
1807 if (unlikely(!S_ISREG(inode->i_mode)))
1808 add_flags = DCACHE_SPECIAL_TYPE;
1811 if (unlikely(IS_AUTOMOUNT(inode)))
1812 add_flags |= DCACHE_NEED_AUTOMOUNT;
1816 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1818 unsigned add_flags = d_flags_for_inode(inode);
1819 WARN_ON(d_in_lookup(dentry));
1821 spin_lock(&dentry->d_lock);
1822 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1823 raw_write_seqcount_begin(&dentry->d_seq);
1824 __d_set_inode_and_type(dentry, inode, add_flags);
1825 raw_write_seqcount_end(&dentry->d_seq);
1826 fsnotify_update_flags(dentry);
1827 spin_unlock(&dentry->d_lock);
1831 * d_instantiate - fill in inode information for a dentry
1832 * @entry: dentry to complete
1833 * @inode: inode to attach to this dentry
1835 * Fill in inode information in the entry.
1837 * This turns negative dentries into productive full members
1840 * NOTE! This assumes that the inode count has been incremented
1841 * (or otherwise set) by the caller to indicate that it is now
1842 * in use by the dcache.
1845 void d_instantiate(struct dentry *entry, struct inode * inode)
1847 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1849 security_d_instantiate(entry, inode);
1850 spin_lock(&inode->i_lock);
1851 __d_instantiate(entry, inode);
1852 spin_unlock(&inode->i_lock);
1855 EXPORT_SYMBOL(d_instantiate);
1858 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1859 * @entry: dentry to complete
1860 * @inode: inode to attach to this dentry
1862 * Fill in inode information in the entry. If a directory alias is found, then
1863 * return an error (and drop inode). Together with d_materialise_unique() this
1864 * guarantees that a directory inode may never have more than one alias.
1866 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1868 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1870 security_d_instantiate(entry, inode);
1871 spin_lock(&inode->i_lock);
1872 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1873 spin_unlock(&inode->i_lock);
1877 __d_instantiate(entry, inode);
1878 spin_unlock(&inode->i_lock);
1882 EXPORT_SYMBOL(d_instantiate_no_diralias);
1884 struct dentry *d_make_root(struct inode *root_inode)
1886 struct dentry *res = NULL;
1889 res = __d_alloc(root_inode->i_sb, NULL);
1891 d_instantiate(res, root_inode);
1897 EXPORT_SYMBOL(d_make_root);
1899 static struct dentry * __d_find_any_alias(struct inode *inode)
1901 struct dentry *alias;
1903 if (hlist_empty(&inode->i_dentry))
1905 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1911 * d_find_any_alias - find any alias for a given inode
1912 * @inode: inode to find an alias for
1914 * If any aliases exist for the given inode, take and return a
1915 * reference for one of them. If no aliases exist, return %NULL.
1917 struct dentry *d_find_any_alias(struct inode *inode)
1921 spin_lock(&inode->i_lock);
1922 de = __d_find_any_alias(inode);
1923 spin_unlock(&inode->i_lock);
1926 EXPORT_SYMBOL(d_find_any_alias);
1928 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1935 return ERR_PTR(-ESTALE);
1937 return ERR_CAST(inode);
1939 res = d_find_any_alias(inode);
1943 tmp = __d_alloc(inode->i_sb, NULL);
1945 res = ERR_PTR(-ENOMEM);
1949 security_d_instantiate(tmp, inode);
1950 spin_lock(&inode->i_lock);
1951 res = __d_find_any_alias(inode);
1953 spin_unlock(&inode->i_lock);
1958 /* attach a disconnected dentry */
1959 add_flags = d_flags_for_inode(inode);
1962 add_flags |= DCACHE_DISCONNECTED;
1964 spin_lock(&tmp->d_lock);
1965 __d_set_inode_and_type(tmp, inode, add_flags);
1966 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1967 hlist_bl_lock(&tmp->d_sb->s_anon);
1968 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1969 hlist_bl_unlock(&tmp->d_sb->s_anon);
1970 spin_unlock(&tmp->d_lock);
1971 spin_unlock(&inode->i_lock);
1981 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1982 * @inode: inode to allocate the dentry for
1984 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1985 * similar open by handle operations. The returned dentry may be anonymous,
1986 * or may have a full name (if the inode was already in the cache).
1988 * When called on a directory inode, we must ensure that the inode only ever
1989 * has one dentry. If a dentry is found, that is returned instead of
1990 * allocating a new one.
1992 * On successful return, the reference to the inode has been transferred
1993 * to the dentry. In case of an error the reference on the inode is released.
1994 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1995 * be passed in and the error will be propagated to the return value,
1996 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1998 struct dentry *d_obtain_alias(struct inode *inode)
2000 return __d_obtain_alias(inode, 1);
2002 EXPORT_SYMBOL(d_obtain_alias);
2005 * d_obtain_root - find or allocate a dentry for a given inode
2006 * @inode: inode to allocate the dentry for
2008 * Obtain an IS_ROOT dentry for the root of a filesystem.
2010 * We must ensure that directory inodes only ever have one dentry. If a
2011 * dentry is found, that is returned instead of allocating a new one.
2013 * On successful return, the reference to the inode has been transferred
2014 * to the dentry. In case of an error the reference on the inode is
2015 * released. A %NULL or IS_ERR inode may be passed in and will be the
2016 * error will be propagate to the return value, with a %NULL @inode
2017 * replaced by ERR_PTR(-ESTALE).
2019 struct dentry *d_obtain_root(struct inode *inode)
2021 return __d_obtain_alias(inode, 0);
2023 EXPORT_SYMBOL(d_obtain_root);
2026 * d_add_ci - lookup or allocate new dentry with case-exact name
2027 * @inode: the inode case-insensitive lookup has found
2028 * @dentry: the negative dentry that was passed to the parent's lookup func
2029 * @name: the case-exact name to be associated with the returned dentry
2031 * This is to avoid filling the dcache with case-insensitive names to the
2032 * same inode, only the actual correct case is stored in the dcache for
2033 * case-insensitive filesystems.
2035 * For a case-insensitive lookup match and if the the case-exact dentry
2036 * already exists in in the dcache, use it and return it.
2038 * If no entry exists with the exact case name, allocate new dentry with
2039 * the exact case, and return the spliced entry.
2041 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2044 struct dentry *found, *res;
2047 * First check if a dentry matching the name already exists,
2048 * if not go ahead and create it now.
2050 found = d_hash_and_lookup(dentry->d_parent, name);
2055 if (d_in_lookup(dentry)) {
2056 found = d_alloc_parallel(dentry->d_parent, name,
2058 if (IS_ERR(found) || !d_in_lookup(found)) {
2063 found = d_alloc(dentry->d_parent, name);
2066 return ERR_PTR(-ENOMEM);
2069 res = d_splice_alias(inode, found);
2076 EXPORT_SYMBOL(d_add_ci);
2079 static inline bool d_same_name(const struct dentry *dentry,
2080 const struct dentry *parent,
2081 const struct qstr *name)
2083 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2084 if (dentry->d_name.len != name->len)
2086 return dentry_cmp(dentry, name->name, name->len) == 0;
2088 return parent->d_op->d_compare(dentry,
2089 dentry->d_name.len, dentry->d_name.name,
2094 * __d_lookup_rcu - search for a dentry (racy, store-free)
2095 * @parent: parent dentry
2096 * @name: qstr of name we wish to find
2097 * @seqp: returns d_seq value at the point where the dentry was found
2098 * Returns: dentry, or NULL
2100 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2101 * resolution (store-free path walking) design described in
2102 * Documentation/filesystems/path-lookup.txt.
2104 * This is not to be used outside core vfs.
2106 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2107 * held, and rcu_read_lock held. The returned dentry must not be stored into
2108 * without taking d_lock and checking d_seq sequence count against @seq
2111 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2114 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2115 * the returned dentry, so long as its parent's seqlock is checked after the
2116 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2117 * is formed, giving integrity down the path walk.
2119 * NOTE! The caller *has* to check the resulting dentry against the sequence
2120 * number we've returned before using any of the resulting dentry state!
2122 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2123 const struct qstr *name,
2126 u64 hashlen = name->hash_len;
2127 const unsigned char *str = name->name;
2128 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2129 struct hlist_bl_node *node;
2130 struct dentry *dentry;
2133 * Note: There is significant duplication with __d_lookup_rcu which is
2134 * required to prevent single threaded performance regressions
2135 * especially on architectures where smp_rmb (in seqcounts) are costly.
2136 * Keep the two functions in sync.
2140 * The hash list is protected using RCU.
2142 * Carefully use d_seq when comparing a candidate dentry, to avoid
2143 * races with d_move().
2145 * It is possible that concurrent renames can mess up our list
2146 * walk here and result in missing our dentry, resulting in the
2147 * false-negative result. d_lookup() protects against concurrent
2148 * renames using rename_lock seqlock.
2150 * See Documentation/filesystems/path-lookup.txt for more details.
2152 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2157 * The dentry sequence count protects us from concurrent
2158 * renames, and thus protects parent and name fields.
2160 * The caller must perform a seqcount check in order
2161 * to do anything useful with the returned dentry.
2163 * NOTE! We do a "raw" seqcount_begin here. That means that
2164 * we don't wait for the sequence count to stabilize if it
2165 * is in the middle of a sequence change. If we do the slow
2166 * dentry compare, we will do seqretries until it is stable,
2167 * and if we end up with a successful lookup, we actually
2168 * want to exit RCU lookup anyway.
2170 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2171 * we are still guaranteed NUL-termination of ->d_name.name.
2173 seq = raw_seqcount_begin(&dentry->d_seq);
2174 if (dentry->d_parent != parent)
2176 if (d_unhashed(dentry))
2179 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2182 if (dentry->d_name.hash != hashlen_hash(hashlen))
2184 tlen = dentry->d_name.len;
2185 tname = dentry->d_name.name;
2186 /* we want a consistent (name,len) pair */
2187 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2191 if (parent->d_op->d_compare(dentry,
2192 tlen, tname, name) != 0)
2195 if (dentry->d_name.hash_len != hashlen)
2197 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2207 * d_lookup - search for a dentry
2208 * @parent: parent dentry
2209 * @name: qstr of name we wish to find
2210 * Returns: dentry, or NULL
2212 * d_lookup searches the children of the parent dentry for the name in
2213 * question. If the dentry is found its reference count is incremented and the
2214 * dentry is returned. The caller must use dput to free the entry when it has
2215 * finished using it. %NULL is returned if the dentry does not exist.
2217 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2219 struct dentry *dentry;
2223 seq = read_seqbegin(&rename_lock);
2224 dentry = __d_lookup(parent, name);
2227 } while (read_seqretry(&rename_lock, seq));
2230 EXPORT_SYMBOL(d_lookup);
2233 * __d_lookup - search for a dentry (racy)
2234 * @parent: parent dentry
2235 * @name: qstr of name we wish to find
2236 * Returns: dentry, or NULL
2238 * __d_lookup is like d_lookup, however it may (rarely) return a
2239 * false-negative result due to unrelated rename activity.
2241 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2242 * however it must be used carefully, eg. with a following d_lookup in
2243 * the case of failure.
2245 * __d_lookup callers must be commented.
2247 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2249 unsigned int hash = name->hash;
2250 struct hlist_bl_head *b = d_hash(hash);
2251 struct hlist_bl_node *node;
2252 struct dentry *found = NULL;
2253 struct dentry *dentry;
2256 * Note: There is significant duplication with __d_lookup_rcu which is
2257 * required to prevent single threaded performance regressions
2258 * especially on architectures where smp_rmb (in seqcounts) are costly.
2259 * Keep the two functions in sync.
2263 * The hash list is protected using RCU.
2265 * Take d_lock when comparing a candidate dentry, to avoid races
2268 * It is possible that concurrent renames can mess up our list
2269 * walk here and result in missing our dentry, resulting in the
2270 * false-negative result. d_lookup() protects against concurrent
2271 * renames using rename_lock seqlock.
2273 * See Documentation/filesystems/path-lookup.txt for more details.
2277 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2279 if (dentry->d_name.hash != hash)
2282 spin_lock(&dentry->d_lock);
2283 if (dentry->d_parent != parent)
2285 if (d_unhashed(dentry))
2288 if (!d_same_name(dentry, parent, name))
2291 dentry->d_lockref.count++;
2293 spin_unlock(&dentry->d_lock);
2296 spin_unlock(&dentry->d_lock);
2304 * d_hash_and_lookup - hash the qstr then search for a dentry
2305 * @dir: Directory to search in
2306 * @name: qstr of name we wish to find
2308 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2310 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2313 * Check for a fs-specific hash function. Note that we must
2314 * calculate the standard hash first, as the d_op->d_hash()
2315 * routine may choose to leave the hash value unchanged.
2317 name->hash = full_name_hash(dir, name->name, name->len);
2318 if (dir->d_flags & DCACHE_OP_HASH) {
2319 int err = dir->d_op->d_hash(dir, name);
2320 if (unlikely(err < 0))
2321 return ERR_PTR(err);
2323 return d_lookup(dir, name);
2325 EXPORT_SYMBOL(d_hash_and_lookup);
2328 * When a file is deleted, we have two options:
2329 * - turn this dentry into a negative dentry
2330 * - unhash this dentry and free it.
2332 * Usually, we want to just turn this into
2333 * a negative dentry, but if anybody else is
2334 * currently using the dentry or the inode
2335 * we can't do that and we fall back on removing
2336 * it from the hash queues and waiting for
2337 * it to be deleted later when it has no users
2341 * d_delete - delete a dentry
2342 * @dentry: The dentry to delete
2344 * Turn the dentry into a negative dentry if possible, otherwise
2345 * remove it from the hash queues so it can be deleted later
2348 void d_delete(struct dentry * dentry)
2350 struct inode *inode;
2353 * Are we the only user?
2356 spin_lock(&dentry->d_lock);
2357 inode = dentry->d_inode;
2358 isdir = S_ISDIR(inode->i_mode);
2359 if (dentry->d_lockref.count == 1) {
2360 if (!spin_trylock(&inode->i_lock)) {
2361 spin_unlock(&dentry->d_lock);
2365 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2366 dentry_unlink_inode(dentry);
2367 fsnotify_nameremove(dentry, isdir);
2371 if (!d_unhashed(dentry))
2374 spin_unlock(&dentry->d_lock);
2376 fsnotify_nameremove(dentry, isdir);
2378 EXPORT_SYMBOL(d_delete);
2380 static void __d_rehash(struct dentry *entry)
2382 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2383 BUG_ON(!d_unhashed(entry));
2385 hlist_bl_add_head_rcu(&entry->d_hash, b);
2390 * d_rehash - add an entry back to the hash
2391 * @entry: dentry to add to the hash
2393 * Adds a dentry to the hash according to its name.
2396 void d_rehash(struct dentry * entry)
2398 spin_lock(&entry->d_lock);
2400 spin_unlock(&entry->d_lock);
2402 EXPORT_SYMBOL(d_rehash);
2404 static inline unsigned start_dir_add(struct inode *dir)
2408 unsigned n = dir->i_dir_seq;
2409 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2415 static inline void end_dir_add(struct inode *dir, unsigned n)
2417 smp_store_release(&dir->i_dir_seq, n + 2);
2420 static void d_wait_lookup(struct dentry *dentry)
2422 if (d_in_lookup(dentry)) {
2423 DECLARE_WAITQUEUE(wait, current);
2424 add_wait_queue(dentry->d_wait, &wait);
2426 set_current_state(TASK_UNINTERRUPTIBLE);
2427 spin_unlock(&dentry->d_lock);
2429 spin_lock(&dentry->d_lock);
2430 } while (d_in_lookup(dentry));
2434 struct dentry *d_alloc_parallel(struct dentry *parent,
2435 const struct qstr *name,
2436 wait_queue_head_t *wq)
2438 unsigned int hash = name->hash;
2439 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2440 struct hlist_bl_node *node;
2441 struct dentry *new = d_alloc(parent, name);
2442 struct dentry *dentry;
2443 unsigned seq, r_seq, d_seq;
2446 return ERR_PTR(-ENOMEM);
2450 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2451 r_seq = read_seqbegin(&rename_lock);
2452 dentry = __d_lookup_rcu(parent, name, &d_seq);
2453 if (unlikely(dentry)) {
2454 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2458 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2467 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2472 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2478 * No changes for the parent since the beginning of d_lookup().
2479 * Since all removals from the chain happen with hlist_bl_lock(),
2480 * any potential in-lookup matches are going to stay here until
2481 * we unlock the chain. All fields are stable in everything
2484 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2485 if (dentry->d_name.hash != hash)
2487 if (dentry->d_parent != parent)
2489 if (!d_same_name(dentry, parent, name))
2492 /* now we can try to grab a reference */
2493 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2500 * somebody is likely to be still doing lookup for it;
2501 * wait for them to finish
2503 spin_lock(&dentry->d_lock);
2504 d_wait_lookup(dentry);
2506 * it's not in-lookup anymore; in principle we should repeat
2507 * everything from dcache lookup, but it's likely to be what
2508 * d_lookup() would've found anyway. If it is, just return it;
2509 * otherwise we really have to repeat the whole thing.
2511 if (unlikely(dentry->d_name.hash != hash))
2513 if (unlikely(dentry->d_parent != parent))
2515 if (unlikely(d_unhashed(dentry)))
2517 if (unlikely(!d_same_name(dentry, parent, name)))
2519 /* OK, it *is* a hashed match; return it */
2520 spin_unlock(&dentry->d_lock);
2525 /* we can't take ->d_lock here; it's OK, though. */
2526 new->d_flags |= DCACHE_PAR_LOOKUP;
2528 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2532 spin_unlock(&dentry->d_lock);
2536 EXPORT_SYMBOL(d_alloc_parallel);
2538 void __d_lookup_done(struct dentry *dentry)
2540 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2541 dentry->d_name.hash);
2543 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2544 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2545 wake_up_all(dentry->d_wait);
2546 dentry->d_wait = NULL;
2548 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2549 INIT_LIST_HEAD(&dentry->d_lru);
2551 EXPORT_SYMBOL(__d_lookup_done);
2553 /* inode->i_lock held if inode is non-NULL */
2555 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2557 struct inode *dir = NULL;
2559 spin_lock(&dentry->d_lock);
2560 if (unlikely(d_in_lookup(dentry))) {
2561 dir = dentry->d_parent->d_inode;
2562 n = start_dir_add(dir);
2563 __d_lookup_done(dentry);
2566 unsigned add_flags = d_flags_for_inode(inode);
2567 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2568 raw_write_seqcount_begin(&dentry->d_seq);
2569 __d_set_inode_and_type(dentry, inode, add_flags);
2570 raw_write_seqcount_end(&dentry->d_seq);
2571 fsnotify_update_flags(dentry);
2575 end_dir_add(dir, n);
2576 spin_unlock(&dentry->d_lock);
2578 spin_unlock(&inode->i_lock);
2582 * d_add - add dentry to hash queues
2583 * @entry: dentry to add
2584 * @inode: The inode to attach to this dentry
2586 * This adds the entry to the hash queues and initializes @inode.
2587 * The entry was actually filled in earlier during d_alloc().
2590 void d_add(struct dentry *entry, struct inode *inode)
2593 security_d_instantiate(entry, inode);
2594 spin_lock(&inode->i_lock);
2596 __d_add(entry, inode);
2598 EXPORT_SYMBOL(d_add);
2601 * d_exact_alias - find and hash an exact unhashed alias
2602 * @entry: dentry to add
2603 * @inode: The inode to go with this dentry
2605 * If an unhashed dentry with the same name/parent and desired
2606 * inode already exists, hash and return it. Otherwise, return
2609 * Parent directory should be locked.
2611 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2613 struct dentry *alias;
2614 unsigned int hash = entry->d_name.hash;
2616 spin_lock(&inode->i_lock);
2617 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2619 * Don't need alias->d_lock here, because aliases with
2620 * d_parent == entry->d_parent are not subject to name or
2621 * parent changes, because the parent inode i_mutex is held.
2623 if (alias->d_name.hash != hash)
2625 if (alias->d_parent != entry->d_parent)
2627 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2629 spin_lock(&alias->d_lock);
2630 if (!d_unhashed(alias)) {
2631 spin_unlock(&alias->d_lock);
2634 __dget_dlock(alias);
2636 spin_unlock(&alias->d_lock);
2638 spin_unlock(&inode->i_lock);
2641 spin_unlock(&inode->i_lock);
2644 EXPORT_SYMBOL(d_exact_alias);
2647 * dentry_update_name_case - update case insensitive dentry with a new name
2648 * @dentry: dentry to be updated
2651 * Update a case insensitive dentry with new case of name.
2653 * dentry must have been returned by d_lookup with name @name. Old and new
2654 * name lengths must match (ie. no d_compare which allows mismatched name
2657 * Parent inode i_mutex must be held over d_lookup and into this call (to
2658 * keep renames and concurrent inserts, and readdir(2) away).
2660 void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
2662 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
2663 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2665 spin_lock(&dentry->d_lock);
2666 write_seqcount_begin(&dentry->d_seq);
2667 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2668 write_seqcount_end(&dentry->d_seq);
2669 spin_unlock(&dentry->d_lock);
2671 EXPORT_SYMBOL(dentry_update_name_case);
2673 static void swap_names(struct dentry *dentry, struct dentry *target)
2675 if (unlikely(dname_external(target))) {
2676 if (unlikely(dname_external(dentry))) {
2678 * Both external: swap the pointers
2680 swap(target->d_name.name, dentry->d_name.name);
2683 * dentry:internal, target:external. Steal target's
2684 * storage and make target internal.
2686 memcpy(target->d_iname, dentry->d_name.name,
2687 dentry->d_name.len + 1);
2688 dentry->d_name.name = target->d_name.name;
2689 target->d_name.name = target->d_iname;
2692 if (unlikely(dname_external(dentry))) {
2694 * dentry:external, target:internal. Give dentry's
2695 * storage to target and make dentry internal
2697 memcpy(dentry->d_iname, target->d_name.name,
2698 target->d_name.len + 1);
2699 target->d_name.name = dentry->d_name.name;
2700 dentry->d_name.name = dentry->d_iname;
2703 * Both are internal.
2706 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2707 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2708 swap(((long *) &dentry->d_iname)[i],
2709 ((long *) &target->d_iname)[i]);
2713 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2716 static void copy_name(struct dentry *dentry, struct dentry *target)
2718 struct external_name *old_name = NULL;
2719 if (unlikely(dname_external(dentry)))
2720 old_name = external_name(dentry);
2721 if (unlikely(dname_external(target))) {
2722 atomic_inc(&external_name(target)->u.count);
2723 dentry->d_name = target->d_name;
2725 memcpy(dentry->d_iname, target->d_name.name,
2726 target->d_name.len + 1);
2727 dentry->d_name.name = dentry->d_iname;
2728 dentry->d_name.hash_len = target->d_name.hash_len;
2730 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2731 kfree_rcu(old_name, u.head);
2734 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2737 * XXXX: do we really need to take target->d_lock?
2739 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2740 spin_lock(&target->d_parent->d_lock);
2742 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2743 spin_lock(&dentry->d_parent->d_lock);
2744 spin_lock_nested(&target->d_parent->d_lock,
2745 DENTRY_D_LOCK_NESTED);
2747 spin_lock(&target->d_parent->d_lock);
2748 spin_lock_nested(&dentry->d_parent->d_lock,
2749 DENTRY_D_LOCK_NESTED);
2752 if (target < dentry) {
2753 spin_lock_nested(&target->d_lock, 2);
2754 spin_lock_nested(&dentry->d_lock, 3);
2756 spin_lock_nested(&dentry->d_lock, 2);
2757 spin_lock_nested(&target->d_lock, 3);
2761 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2763 if (target->d_parent != dentry->d_parent)
2764 spin_unlock(&dentry->d_parent->d_lock);
2765 if (target->d_parent != target)
2766 spin_unlock(&target->d_parent->d_lock);
2767 spin_unlock(&target->d_lock);
2768 spin_unlock(&dentry->d_lock);
2772 * When switching names, the actual string doesn't strictly have to
2773 * be preserved in the target - because we're dropping the target
2774 * anyway. As such, we can just do a simple memcpy() to copy over
2775 * the new name before we switch, unless we are going to rehash
2776 * it. Note that if we *do* unhash the target, we are not allowed
2777 * to rehash it without giving it a new name/hash key - whether
2778 * we swap or overwrite the names here, resulting name won't match
2779 * the reality in filesystem; it's only there for d_path() purposes.
2780 * Note that all of this is happening under rename_lock, so the
2781 * any hash lookup seeing it in the middle of manipulations will
2782 * be discarded anyway. So we do not care what happens to the hash
2786 * __d_move - move a dentry
2787 * @dentry: entry to move
2788 * @target: new dentry
2789 * @exchange: exchange the two dentries
2791 * Update the dcache to reflect the move of a file name. Negative
2792 * dcache entries should not be moved in this way. Caller must hold
2793 * rename_lock, the i_mutex of the source and target directories,
2794 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2796 static void __d_move(struct dentry *dentry, struct dentry *target,
2799 struct inode *dir = NULL;
2801 if (!dentry->d_inode)
2802 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2804 BUG_ON(d_ancestor(dentry, target));
2805 BUG_ON(d_ancestor(target, dentry));
2807 dentry_lock_for_move(dentry, target);
2808 if (unlikely(d_in_lookup(target))) {
2809 dir = target->d_parent->d_inode;
2810 n = start_dir_add(dir);
2811 __d_lookup_done(target);
2814 write_seqcount_begin(&dentry->d_seq);
2815 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2818 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2822 /* Switch the names.. */
2824 swap_names(dentry, target);
2826 copy_name(dentry, target);
2828 /* rehash in new place(s) */
2833 /* ... and switch them in the tree */
2834 if (IS_ROOT(dentry)) {
2835 /* splicing a tree */
2836 dentry->d_flags |= DCACHE_RCUACCESS;
2837 dentry->d_parent = target->d_parent;
2838 target->d_parent = target;
2839 list_del_init(&target->d_child);
2840 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2842 /* swapping two dentries */
2843 swap(dentry->d_parent, target->d_parent);
2844 list_move(&target->d_child, &target->d_parent->d_subdirs);
2845 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2847 fsnotify_update_flags(target);
2848 fsnotify_update_flags(dentry);
2851 write_seqcount_end(&target->d_seq);
2852 write_seqcount_end(&dentry->d_seq);
2855 end_dir_add(dir, n);
2856 dentry_unlock_for_move(dentry, target);
2860 * d_move - move a dentry
2861 * @dentry: entry to move
2862 * @target: new dentry
2864 * Update the dcache to reflect the move of a file name. Negative
2865 * dcache entries should not be moved in this way. See the locking
2866 * requirements for __d_move.
2868 void d_move(struct dentry *dentry, struct dentry *target)
2870 write_seqlock(&rename_lock);
2871 __d_move(dentry, target, false);
2872 write_sequnlock(&rename_lock);
2874 EXPORT_SYMBOL(d_move);
2877 * d_exchange - exchange two dentries
2878 * @dentry1: first dentry
2879 * @dentry2: second dentry
2881 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2883 write_seqlock(&rename_lock);
2885 WARN_ON(!dentry1->d_inode);
2886 WARN_ON(!dentry2->d_inode);
2887 WARN_ON(IS_ROOT(dentry1));
2888 WARN_ON(IS_ROOT(dentry2));
2890 __d_move(dentry1, dentry2, true);
2892 write_sequnlock(&rename_lock);
2896 * d_ancestor - search for an ancestor
2897 * @p1: ancestor dentry
2900 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2901 * an ancestor of p2, else NULL.
2903 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2907 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2908 if (p->d_parent == p1)
2915 * This helper attempts to cope with remotely renamed directories
2917 * It assumes that the caller is already holding
2918 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2920 * Note: If ever the locking in lock_rename() changes, then please
2921 * remember to update this too...
2923 static int __d_unalias(struct inode *inode,
2924 struct dentry *dentry, struct dentry *alias)
2926 struct mutex *m1 = NULL;
2927 struct rw_semaphore *m2 = NULL;
2930 /* If alias and dentry share a parent, then no extra locks required */
2931 if (alias->d_parent == dentry->d_parent)
2934 /* See lock_rename() */
2935 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2937 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2938 if (!inode_trylock_shared(alias->d_parent->d_inode))
2940 m2 = &alias->d_parent->d_inode->i_rwsem;
2942 __d_move(alias, dentry, false);
2953 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2954 * @inode: the inode which may have a disconnected dentry
2955 * @dentry: a negative dentry which we want to point to the inode.
2957 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2958 * place of the given dentry and return it, else simply d_add the inode
2959 * to the dentry and return NULL.
2961 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2962 * we should error out: directories can't have multiple aliases.
2964 * This is needed in the lookup routine of any filesystem that is exportable
2965 * (via knfsd) so that we can build dcache paths to directories effectively.
2967 * If a dentry was found and moved, then it is returned. Otherwise NULL
2968 * is returned. This matches the expected return value of ->lookup.
2970 * Cluster filesystems may call this function with a negative, hashed dentry.
2971 * In that case, we know that the inode will be a regular file, and also this
2972 * will only occur during atomic_open. So we need to check for the dentry
2973 * being already hashed only in the final case.
2975 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2978 return ERR_CAST(inode);
2980 BUG_ON(!d_unhashed(dentry));
2985 security_d_instantiate(dentry, inode);
2986 spin_lock(&inode->i_lock);
2987 if (S_ISDIR(inode->i_mode)) {
2988 struct dentry *new = __d_find_any_alias(inode);
2989 if (unlikely(new)) {
2990 /* The reference to new ensures it remains an alias */
2991 spin_unlock(&inode->i_lock);
2992 write_seqlock(&rename_lock);
2993 if (unlikely(d_ancestor(new, dentry))) {
2994 write_sequnlock(&rename_lock);
2996 new = ERR_PTR(-ELOOP);
2997 pr_warn_ratelimited(
2998 "VFS: Lookup of '%s' in %s %s"
2999 " would have caused loop\n",
3000 dentry->d_name.name,
3001 inode->i_sb->s_type->name,
3003 } else if (!IS_ROOT(new)) {
3004 int err = __d_unalias(inode, dentry, new);
3005 write_sequnlock(&rename_lock);
3011 __d_move(new, dentry, false);
3012 write_sequnlock(&rename_lock);
3019 __d_add(dentry, inode);
3022 EXPORT_SYMBOL(d_splice_alias);
3024 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
3028 return -ENAMETOOLONG;
3030 memcpy(*buffer, str, namelen);
3035 * prepend_name - prepend a pathname in front of current buffer pointer
3036 * @buffer: buffer pointer
3037 * @buflen: allocated length of the buffer
3038 * @name: name string and length qstr structure
3040 * With RCU path tracing, it may race with d_move(). Use READ_ONCE() to
3041 * make sure that either the old or the new name pointer and length are
3042 * fetched. However, there may be mismatch between length and pointer.
3043 * The length cannot be trusted, we need to copy it byte-by-byte until
3044 * the length is reached or a null byte is found. It also prepends "/" at
3045 * the beginning of the name. The sequence number check at the caller will
3046 * retry it again when a d_move() does happen. So any garbage in the buffer
3047 * due to mismatched pointer and length will be discarded.
3049 * Load acquire is needed to make sure that we see that terminating NUL.
3051 static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
3053 const char *dname = smp_load_acquire(&name->name); /* ^^^ */
3054 u32 dlen = READ_ONCE(name->len);
3057 *buflen -= dlen + 1;
3059 return -ENAMETOOLONG;
3060 p = *buffer -= dlen + 1;
3072 * prepend_path - Prepend path string to a buffer
3073 * @path: the dentry/vfsmount to report
3074 * @root: root vfsmnt/dentry
3075 * @buffer: pointer to the end of the buffer
3076 * @buflen: pointer to buffer length
3078 * The function will first try to write out the pathname without taking any
3079 * lock other than the RCU read lock to make sure that dentries won't go away.
3080 * It only checks the sequence number of the global rename_lock as any change
3081 * in the dentry's d_seq will be preceded by changes in the rename_lock
3082 * sequence number. If the sequence number had been changed, it will restart
3083 * the whole pathname back-tracing sequence again by taking the rename_lock.
3084 * In this case, there is no need to take the RCU read lock as the recursive
3085 * parent pointer references will keep the dentry chain alive as long as no
3086 * rename operation is performed.
3088 static int prepend_path(const struct path *path,
3089 const struct path *root,
3090 char **buffer, int *buflen)
3092 struct dentry *dentry;
3093 struct vfsmount *vfsmnt;
3096 unsigned seq, m_seq = 0;
3102 read_seqbegin_or_lock(&mount_lock, &m_seq);
3109 dentry = path->dentry;
3111 mnt = real_mount(vfsmnt);
3112 read_seqbegin_or_lock(&rename_lock, &seq);
3113 while (dentry != root->dentry || vfsmnt != root->mnt) {
3114 struct dentry * parent;
3116 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
3117 struct mount *parent = READ_ONCE(mnt->mnt_parent);
3119 if (dentry != vfsmnt->mnt_root) {
3126 if (mnt != parent) {
3127 dentry = READ_ONCE(mnt->mnt_mountpoint);
3133 error = is_mounted(vfsmnt) ? 1 : 2;
3136 parent = dentry->d_parent;
3138 error = prepend_name(&bptr, &blen, &dentry->d_name);
3146 if (need_seqretry(&rename_lock, seq)) {
3150 done_seqretry(&rename_lock, seq);
3154 if (need_seqretry(&mount_lock, m_seq)) {
3158 done_seqretry(&mount_lock, m_seq);
3160 if (error >= 0 && bptr == *buffer) {
3162 error = -ENAMETOOLONG;
3172 * __d_path - return the path of a dentry
3173 * @path: the dentry/vfsmount to report
3174 * @root: root vfsmnt/dentry
3175 * @buf: buffer to return value in
3176 * @buflen: buffer length
3178 * Convert a dentry into an ASCII path name.
3180 * Returns a pointer into the buffer or an error code if the
3181 * path was too long.
3183 * "buflen" should be positive.
3185 * If the path is not reachable from the supplied root, return %NULL.
3187 char *__d_path(const struct path *path,
3188 const struct path *root,
3189 char *buf, int buflen)
3191 char *res = buf + buflen;
3194 prepend(&res, &buflen, "\0", 1);
3195 error = prepend_path(path, root, &res, &buflen);
3198 return ERR_PTR(error);
3204 char *d_absolute_path(const struct path *path,
3205 char *buf, int buflen)
3207 struct path root = {};
3208 char *res = buf + buflen;
3211 prepend(&res, &buflen, "\0", 1);
3212 error = prepend_path(path, &root, &res, &buflen);
3217 return ERR_PTR(error);
3222 * same as __d_path but appends "(deleted)" for unlinked files.
3224 static int path_with_deleted(const struct path *path,
3225 const struct path *root,
3226 char **buf, int *buflen)
3228 prepend(buf, buflen, "\0", 1);
3229 if (d_unlinked(path->dentry)) {
3230 int error = prepend(buf, buflen, " (deleted)", 10);
3235 return prepend_path(path, root, buf, buflen);
3238 static int prepend_unreachable(char **buffer, int *buflen)
3240 return prepend(buffer, buflen, "(unreachable)", 13);
3243 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3248 seq = read_seqcount_begin(&fs->seq);
3250 } while (read_seqcount_retry(&fs->seq, seq));
3254 * d_path - return the path of a dentry
3255 * @path: path to report
3256 * @buf: buffer to return value in
3257 * @buflen: buffer length
3259 * Convert a dentry into an ASCII path name. If the entry has been deleted
3260 * the string " (deleted)" is appended. Note that this is ambiguous.
3262 * Returns a pointer into the buffer or an error code if the path was
3263 * too long. Note: Callers should use the returned pointer, not the passed
3264 * in buffer, to use the name! The implementation often starts at an offset
3265 * into the buffer, and may leave 0 bytes at the start.
3267 * "buflen" should be positive.
3269 char *d_path(const struct path *path, char *buf, int buflen)
3271 char *res = buf + buflen;
3276 * We have various synthetic filesystems that never get mounted. On
3277 * these filesystems dentries are never used for lookup purposes, and
3278 * thus don't need to be hashed. They also don't need a name until a
3279 * user wants to identify the object in /proc/pid/fd/. The little hack
3280 * below allows us to generate a name for these objects on demand:
3282 * Some pseudo inodes are mountable. When they are mounted
3283 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3284 * and instead have d_path return the mounted path.
3286 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3287 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3288 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3291 get_fs_root_rcu(current->fs, &root);
3292 error = path_with_deleted(path, &root, &res, &buflen);
3296 res = ERR_PTR(error);
3299 EXPORT_SYMBOL(d_path);
3302 * Helper function for dentry_operations.d_dname() members
3304 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3305 const char *fmt, ...)
3311 va_start(args, fmt);
3312 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3315 if (sz > sizeof(temp) || sz > buflen)
3316 return ERR_PTR(-ENAMETOOLONG);
3318 buffer += buflen - sz;
3319 return memcpy(buffer, temp, sz);
3322 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3324 char *end = buffer + buflen;
3325 /* these dentries are never renamed, so d_lock is not needed */
3326 if (prepend(&end, &buflen, " (deleted)", 11) ||
3327 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3328 prepend(&end, &buflen, "/", 1))
3329 end = ERR_PTR(-ENAMETOOLONG);
3332 EXPORT_SYMBOL(simple_dname);
3335 * Write full pathname from the root of the filesystem into the buffer.
3337 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3339 struct dentry *dentry;
3352 prepend(&end, &len, "\0", 1);
3356 read_seqbegin_or_lock(&rename_lock, &seq);
3357 while (!IS_ROOT(dentry)) {
3358 struct dentry *parent = dentry->d_parent;
3361 error = prepend_name(&end, &len, &dentry->d_name);
3370 if (need_seqretry(&rename_lock, seq)) {
3374 done_seqretry(&rename_lock, seq);
3379 return ERR_PTR(-ENAMETOOLONG);
3382 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3384 return __dentry_path(dentry, buf, buflen);
3386 EXPORT_SYMBOL(dentry_path_raw);
3388 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3393 if (d_unlinked(dentry)) {
3395 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3399 retval = __dentry_path(dentry, buf, buflen);
3400 if (!IS_ERR(retval) && p)
3401 *p = '/'; /* restore '/' overriden with '\0' */
3404 return ERR_PTR(-ENAMETOOLONG);
3407 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3413 seq = read_seqcount_begin(&fs->seq);
3416 } while (read_seqcount_retry(&fs->seq, seq));
3420 * NOTE! The user-level library version returns a
3421 * character pointer. The kernel system call just
3422 * returns the length of the buffer filled (which
3423 * includes the ending '\0' character), or a negative
3424 * error value. So libc would do something like
3426 * char *getcwd(char * buf, size_t size)
3430 * retval = sys_getcwd(buf, size);
3437 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3440 struct path pwd, root;
3441 char *page = __getname();
3447 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3450 if (!d_unlinked(pwd.dentry)) {
3452 char *cwd = page + PATH_MAX;
3453 int buflen = PATH_MAX;
3455 prepend(&cwd, &buflen, "\0", 1);
3456 error = prepend_path(&pwd, &root, &cwd, &buflen);
3462 /* Unreachable from current root */
3464 error = prepend_unreachable(&cwd, &buflen);
3470 len = PATH_MAX + page - cwd;
3473 if (copy_to_user(buf, cwd, len))
3486 * Test whether new_dentry is a subdirectory of old_dentry.
3488 * Trivially implemented using the dcache structure
3492 * is_subdir - is new dentry a subdirectory of old_dentry
3493 * @new_dentry: new dentry
3494 * @old_dentry: old dentry
3496 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3497 * Returns false otherwise.
3498 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3501 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3506 if (new_dentry == old_dentry)
3510 /* for restarting inner loop in case of seq retry */
3511 seq = read_seqbegin(&rename_lock);
3513 * Need rcu_readlock to protect against the d_parent trashing
3517 if (d_ancestor(old_dentry, new_dentry))
3522 } while (read_seqretry(&rename_lock, seq));
3527 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3529 struct dentry *root = data;
3530 if (dentry != root) {
3531 if (d_unhashed(dentry) || !dentry->d_inode)
3534 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3535 dentry->d_flags |= DCACHE_GENOCIDE;
3536 dentry->d_lockref.count--;
3539 return D_WALK_CONTINUE;
3542 void d_genocide(struct dentry *parent)
3544 d_walk(parent, parent, d_genocide_kill, NULL);
3547 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3549 inode_dec_link_count(inode);
3550 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3551 !hlist_unhashed(&dentry->d_u.d_alias) ||
3552 !d_unlinked(dentry));
3553 spin_lock(&dentry->d_parent->d_lock);
3554 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3555 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3556 (unsigned long long)inode->i_ino);
3557 spin_unlock(&dentry->d_lock);
3558 spin_unlock(&dentry->d_parent->d_lock);
3559 d_instantiate(dentry, inode);
3561 EXPORT_SYMBOL(d_tmpfile);
3563 static __initdata unsigned long dhash_entries;
3564 static int __init set_dhash_entries(char *str)
3568 dhash_entries = simple_strtoul(str, &str, 0);
3571 __setup("dhash_entries=", set_dhash_entries);
3573 static void __init dcache_init_early(void)
3575 /* If hashes are distributed across NUMA nodes, defer
3576 * hash allocation until vmalloc space is available.
3582 alloc_large_system_hash("Dentry cache",
3583 sizeof(struct hlist_bl_head),
3586 HASH_EARLY | HASH_ZERO,
3593 static void __init dcache_init(void)
3596 * A constructor could be added for stable state like the lists,
3597 * but it is probably not worth it because of the cache nature
3600 dentry_cache = KMEM_CACHE(dentry,
3601 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
3603 /* Hash may have been set up in dcache_init_early */
3608 alloc_large_system_hash("Dentry cache",
3609 sizeof(struct hlist_bl_head),
3619 /* SLAB cache for __getname() consumers */
3620 struct kmem_cache *names_cachep __read_mostly;
3621 EXPORT_SYMBOL(names_cachep);
3623 EXPORT_SYMBOL(d_genocide);
3625 void __init vfs_caches_init_early(void)
3629 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3630 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3632 dcache_init_early();
3636 void __init vfs_caches_init(void)
3638 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3639 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3644 files_maxfiles_init();