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 static void dentry_free(struct dentry *dentry)
274 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
275 if (unlikely(dname_external(dentry))) {
276 struct external_name *p = external_name(dentry);
277 if (likely(atomic_dec_and_test(&p->u.count))) {
278 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
282 /* if dentry was never visible to RCU, immediate free is OK */
283 if (!(dentry->d_flags & DCACHE_RCUACCESS))
284 __d_free(&dentry->d_u.d_rcu);
286 call_rcu(&dentry->d_u.d_rcu, __d_free);
290 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
291 * @dentry: the target dentry
292 * After this call, in-progress rcu-walk path lookup will fail. This
293 * should be called after unhashing, and after changing d_inode (if
294 * the dentry has not already been unhashed).
296 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
298 assert_spin_locked(&dentry->d_lock);
299 /* Go through a barrier */
300 write_seqcount_barrier(&dentry->d_seq);
304 * Release the dentry's inode, using the filesystem
305 * d_iput() operation if defined. Dentry has no refcount
308 static void dentry_iput(struct dentry * dentry)
309 __releases(dentry->d_lock)
310 __releases(dentry->d_inode->i_lock)
312 struct inode *inode = dentry->d_inode;
314 dentry->d_inode = NULL;
315 hlist_del_init(&dentry->d_u.d_alias);
316 spin_unlock(&dentry->d_lock);
317 spin_unlock(&inode->i_lock);
319 fsnotify_inoderemove(inode);
320 if (dentry->d_op && dentry->d_op->d_iput)
321 dentry->d_op->d_iput(dentry, inode);
325 spin_unlock(&dentry->d_lock);
330 * Release the dentry's inode, using the filesystem
331 * d_iput() operation if defined. dentry remains in-use.
333 static void dentry_unlink_inode(struct dentry * dentry)
334 __releases(dentry->d_lock)
335 __releases(dentry->d_inode->i_lock)
337 struct inode *inode = dentry->d_inode;
338 __d_clear_type(dentry);
339 dentry->d_inode = NULL;
340 hlist_del_init(&dentry->d_u.d_alias);
341 dentry_rcuwalk_barrier(dentry);
342 spin_unlock(&dentry->d_lock);
343 spin_unlock(&inode->i_lock);
345 fsnotify_inoderemove(inode);
346 if (dentry->d_op && dentry->d_op->d_iput)
347 dentry->d_op->d_iput(dentry, inode);
353 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
354 * is in use - which includes both the "real" per-superblock
355 * LRU list _and_ the DCACHE_SHRINK_LIST use.
357 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
358 * on the shrink list (ie not on the superblock LRU list).
360 * The per-cpu "nr_dentry_unused" counters are updated with
361 * the DCACHE_LRU_LIST bit.
363 * These helper functions make sure we always follow the
364 * rules. d_lock must be held by the caller.
366 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
367 static void d_lru_add(struct dentry *dentry)
369 D_FLAG_VERIFY(dentry, 0);
370 dentry->d_flags |= DCACHE_LRU_LIST;
371 this_cpu_inc(nr_dentry_unused);
372 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
375 static void d_lru_del(struct dentry *dentry)
377 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
378 dentry->d_flags &= ~DCACHE_LRU_LIST;
379 this_cpu_dec(nr_dentry_unused);
380 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
383 static void d_shrink_del(struct dentry *dentry)
385 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
386 list_del_init(&dentry->d_lru);
387 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
388 this_cpu_dec(nr_dentry_unused);
391 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
393 D_FLAG_VERIFY(dentry, 0);
394 list_add(&dentry->d_lru, list);
395 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
396 this_cpu_inc(nr_dentry_unused);
400 * These can only be called under the global LRU lock, ie during the
401 * callback for freeing the LRU list. "isolate" removes it from the
402 * LRU lists entirely, while shrink_move moves it to the indicated
405 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
407 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
408 dentry->d_flags &= ~DCACHE_LRU_LIST;
409 this_cpu_dec(nr_dentry_unused);
410 list_lru_isolate(lru, &dentry->d_lru);
413 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
414 struct list_head *list)
416 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
417 dentry->d_flags |= DCACHE_SHRINK_LIST;
418 list_lru_isolate_move(lru, &dentry->d_lru, list);
422 * dentry_lru_(add|del)_list) must be called with d_lock held.
424 static void dentry_lru_add(struct dentry *dentry)
426 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
431 * d_drop - drop a dentry
432 * @dentry: dentry to drop
434 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
435 * be found through a VFS lookup any more. Note that this is different from
436 * deleting the dentry - d_delete will try to mark the dentry negative if
437 * possible, giving a successful _negative_ lookup, while d_drop will
438 * just make the cache lookup fail.
440 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
441 * reason (NFS timeouts or autofs deletes).
443 * __d_drop requires dentry->d_lock.
445 void __d_drop(struct dentry *dentry)
447 if (!d_unhashed(dentry)) {
448 struct hlist_bl_head *b;
450 * Hashed dentries are normally on the dentry hashtable,
451 * with the exception of those newly allocated by
452 * d_obtain_alias, which are always IS_ROOT:
454 if (unlikely(IS_ROOT(dentry)))
455 b = &dentry->d_sb->s_anon;
457 b = d_hash(dentry->d_parent, dentry->d_name.hash);
460 __hlist_bl_del(&dentry->d_hash);
461 dentry->d_hash.pprev = NULL;
463 dentry_rcuwalk_barrier(dentry);
466 EXPORT_SYMBOL(__d_drop);
468 void d_drop(struct dentry *dentry)
470 spin_lock(&dentry->d_lock);
472 spin_unlock(&dentry->d_lock);
474 EXPORT_SYMBOL(d_drop);
476 static void __dentry_kill(struct dentry *dentry)
478 struct dentry *parent = NULL;
479 bool can_free = true;
480 if (!IS_ROOT(dentry))
481 parent = dentry->d_parent;
484 * The dentry is now unrecoverably dead to the world.
486 lockref_mark_dead(&dentry->d_lockref);
489 * inform the fs via d_prune that this dentry is about to be
490 * unhashed and destroyed.
492 if (dentry->d_flags & DCACHE_OP_PRUNE)
493 dentry->d_op->d_prune(dentry);
495 if (dentry->d_flags & DCACHE_LRU_LIST) {
496 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
499 /* if it was on the hash then remove it */
501 __list_del_entry(&dentry->d_child);
503 * Inform d_walk() that we are no longer attached to the
506 dentry->d_flags |= DCACHE_DENTRY_KILLED;
508 spin_unlock(&parent->d_lock);
511 * dentry_iput drops the locks, at which point nobody (except
512 * transient RCU lookups) can reach this dentry.
514 BUG_ON(dentry->d_lockref.count > 0);
515 this_cpu_dec(nr_dentry);
516 if (dentry->d_op && dentry->d_op->d_release)
517 dentry->d_op->d_release(dentry);
519 spin_lock(&dentry->d_lock);
520 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
521 dentry->d_flags |= DCACHE_MAY_FREE;
524 spin_unlock(&dentry->d_lock);
525 if (likely(can_free))
530 * Finish off a dentry we've decided to kill.
531 * dentry->d_lock must be held, returns with it unlocked.
532 * If ref is non-zero, then decrement the refcount too.
533 * Returns dentry requiring refcount drop, or NULL if we're done.
535 static struct dentry *dentry_kill(struct dentry *dentry)
536 __releases(dentry->d_lock)
538 struct inode *inode = dentry->d_inode;
539 struct dentry *parent = NULL;
541 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
544 if (!IS_ROOT(dentry)) {
545 parent = dentry->d_parent;
546 if (unlikely(!spin_trylock(&parent->d_lock))) {
548 spin_unlock(&inode->i_lock);
553 __dentry_kill(dentry);
557 spin_unlock(&dentry->d_lock);
559 return dentry; /* try again with same dentry */
562 static inline struct dentry *lock_parent(struct dentry *dentry)
564 struct dentry *parent = dentry->d_parent;
567 if (unlikely(dentry->d_lockref.count < 0))
569 if (likely(spin_trylock(&parent->d_lock)))
572 spin_unlock(&dentry->d_lock);
574 parent = ACCESS_ONCE(dentry->d_parent);
575 spin_lock(&parent->d_lock);
577 * We can't blindly lock dentry until we are sure
578 * that we won't violate the locking order.
579 * Any changes of dentry->d_parent must have
580 * been done with parent->d_lock held, so
581 * spin_lock() above is enough of a barrier
582 * for checking if it's still our child.
584 if (unlikely(parent != dentry->d_parent)) {
585 spin_unlock(&parent->d_lock);
589 if (parent != dentry)
590 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
597 * Try to do a lockless dput(), and return whether that was successful.
599 * If unsuccessful, we return false, having already taken the dentry lock.
601 * The caller needs to hold the RCU read lock, so that the dentry is
602 * guaranteed to stay around even if the refcount goes down to zero!
604 static inline bool fast_dput(struct dentry *dentry)
607 unsigned int d_flags;
610 * If we have a d_op->d_delete() operation, we sould not
611 * let the dentry count go to zero, so use "put__or_lock".
613 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
614 return lockref_put_or_lock(&dentry->d_lockref);
617 * .. otherwise, we can try to just decrement the
618 * lockref optimistically.
620 ret = lockref_put_return(&dentry->d_lockref);
623 * If the lockref_put_return() failed due to the lock being held
624 * by somebody else, the fast path has failed. We will need to
625 * get the lock, and then check the count again.
627 if (unlikely(ret < 0)) {
628 spin_lock(&dentry->d_lock);
629 if (dentry->d_lockref.count > 1) {
630 dentry->d_lockref.count--;
631 spin_unlock(&dentry->d_lock);
638 * If we weren't the last ref, we're done.
644 * Careful, careful. The reference count went down
645 * to zero, but we don't hold the dentry lock, so
646 * somebody else could get it again, and do another
647 * dput(), and we need to not race with that.
649 * However, there is a very special and common case
650 * where we don't care, because there is nothing to
651 * do: the dentry is still hashed, it does not have
652 * a 'delete' op, and it's referenced and already on
655 * NOTE! Since we aren't locked, these values are
656 * not "stable". However, it is sufficient that at
657 * some point after we dropped the reference the
658 * dentry was hashed and the flags had the proper
659 * value. Other dentry users may have re-gotten
660 * a reference to the dentry and change that, but
661 * our work is done - we can leave the dentry
662 * around with a zero refcount.
665 d_flags = ACCESS_ONCE(dentry->d_flags);
666 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST;
668 /* Nothing to do? Dropping the reference was all we needed? */
669 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
673 * Not the fast normal case? Get the lock. We've already decremented
674 * the refcount, but we'll need to re-check the situation after
677 spin_lock(&dentry->d_lock);
680 * Did somebody else grab a reference to it in the meantime, and
681 * we're no longer the last user after all? Alternatively, somebody
682 * else could have killed it and marked it dead. Either way, we
683 * don't need to do anything else.
685 if (dentry->d_lockref.count) {
686 spin_unlock(&dentry->d_lock);
691 * Re-get the reference we optimistically dropped. We hold the
692 * lock, and we just tested that it was zero, so we can just
695 dentry->d_lockref.count = 1;
703 * This is complicated by the fact that we do not want to put
704 * dentries that are no longer on any hash chain on the unused
705 * list: we'd much rather just get rid of them immediately.
707 * However, that implies that we have to traverse the dentry
708 * tree upwards to the parents which might _also_ now be
709 * scheduled for deletion (it may have been only waiting for
710 * its last child to go away).
712 * This tail recursion is done by hand as we don't want to depend
713 * on the compiler to always get this right (gcc generally doesn't).
714 * Real recursion would eat up our stack space.
718 * dput - release a dentry
719 * @dentry: dentry to release
721 * Release a dentry. This will drop the usage count and if appropriate
722 * call the dentry unlink method as well as removing it from the queues and
723 * releasing its resources. If the parent dentries were scheduled for release
724 * they too may now get deleted.
726 void dput(struct dentry *dentry)
728 if (unlikely(!dentry))
733 if (likely(fast_dput(dentry))) {
738 /* Slow case: now with the dentry lock held */
741 /* Unreachable? Get rid of it */
742 if (unlikely(d_unhashed(dentry)))
745 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
746 if (dentry->d_op->d_delete(dentry))
750 if (!(dentry->d_flags & DCACHE_REFERENCED))
751 dentry->d_flags |= DCACHE_REFERENCED;
752 dentry_lru_add(dentry);
754 dentry->d_lockref.count--;
755 spin_unlock(&dentry->d_lock);
759 dentry = dentry_kill(dentry);
766 /* This must be called with d_lock held */
767 static inline void __dget_dlock(struct dentry *dentry)
769 dentry->d_lockref.count++;
772 static inline void __dget(struct dentry *dentry)
774 lockref_get(&dentry->d_lockref);
777 struct dentry *dget_parent(struct dentry *dentry)
783 * Do optimistic parent lookup without any
787 ret = ACCESS_ONCE(dentry->d_parent);
788 gotref = lockref_get_not_zero(&ret->d_lockref);
790 if (likely(gotref)) {
791 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
798 * Don't need rcu_dereference because we re-check it was correct under
802 ret = dentry->d_parent;
803 spin_lock(&ret->d_lock);
804 if (unlikely(ret != dentry->d_parent)) {
805 spin_unlock(&ret->d_lock);
810 BUG_ON(!ret->d_lockref.count);
811 ret->d_lockref.count++;
812 spin_unlock(&ret->d_lock);
815 EXPORT_SYMBOL(dget_parent);
818 * d_find_alias - grab a hashed alias of inode
819 * @inode: inode in question
821 * If inode has a hashed alias, or is a directory and has any alias,
822 * acquire the reference to alias and return it. Otherwise return NULL.
823 * Notice that if inode is a directory there can be only one alias and
824 * it can be unhashed only if it has no children, or if it is the root
825 * of a filesystem, or if the directory was renamed and d_revalidate
826 * was the first vfs operation to notice.
828 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
829 * any other hashed alias over that one.
831 static struct dentry *__d_find_alias(struct inode *inode)
833 struct dentry *alias, *discon_alias;
837 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
838 spin_lock(&alias->d_lock);
839 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
840 if (IS_ROOT(alias) &&
841 (alias->d_flags & DCACHE_DISCONNECTED)) {
842 discon_alias = alias;
845 spin_unlock(&alias->d_lock);
849 spin_unlock(&alias->d_lock);
852 alias = discon_alias;
853 spin_lock(&alias->d_lock);
854 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
856 spin_unlock(&alias->d_lock);
859 spin_unlock(&alias->d_lock);
865 struct dentry *d_find_alias(struct inode *inode)
867 struct dentry *de = NULL;
869 if (!hlist_empty(&inode->i_dentry)) {
870 spin_lock(&inode->i_lock);
871 de = __d_find_alias(inode);
872 spin_unlock(&inode->i_lock);
876 EXPORT_SYMBOL(d_find_alias);
879 * Try to kill dentries associated with this inode.
880 * WARNING: you must own a reference to inode.
882 void d_prune_aliases(struct inode *inode)
884 struct dentry *dentry;
886 spin_lock(&inode->i_lock);
887 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
888 spin_lock(&dentry->d_lock);
889 if (!dentry->d_lockref.count) {
890 struct dentry *parent = lock_parent(dentry);
891 if (likely(!dentry->d_lockref.count)) {
892 __dentry_kill(dentry);
897 spin_unlock(&parent->d_lock);
899 spin_unlock(&dentry->d_lock);
901 spin_unlock(&inode->i_lock);
903 EXPORT_SYMBOL(d_prune_aliases);
905 static void shrink_dentry_list(struct list_head *list)
907 struct dentry *dentry, *parent;
909 while (!list_empty(list)) {
911 dentry = list_entry(list->prev, struct dentry, d_lru);
912 spin_lock(&dentry->d_lock);
913 parent = lock_parent(dentry);
916 * The dispose list is isolated and dentries are not accounted
917 * to the LRU here, so we can simply remove it from the list
918 * here regardless of whether it is referenced or not.
920 d_shrink_del(dentry);
923 * We found an inuse dentry which was not removed from
924 * the LRU because of laziness during lookup. Do not free it.
926 if (dentry->d_lockref.count > 0) {
927 spin_unlock(&dentry->d_lock);
929 spin_unlock(&parent->d_lock);
934 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
935 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
936 spin_unlock(&dentry->d_lock);
938 spin_unlock(&parent->d_lock);
944 inode = dentry->d_inode;
945 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
946 d_shrink_add(dentry, list);
947 spin_unlock(&dentry->d_lock);
949 spin_unlock(&parent->d_lock);
953 __dentry_kill(dentry);
956 * We need to prune ancestors too. This is necessary to prevent
957 * quadratic behavior of shrink_dcache_parent(), but is also
958 * expected to be beneficial in reducing dentry cache
962 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
963 parent = lock_parent(dentry);
964 if (dentry->d_lockref.count != 1) {
965 dentry->d_lockref.count--;
966 spin_unlock(&dentry->d_lock);
968 spin_unlock(&parent->d_lock);
971 inode = dentry->d_inode; /* can't be NULL */
972 if (unlikely(!spin_trylock(&inode->i_lock))) {
973 spin_unlock(&dentry->d_lock);
975 spin_unlock(&parent->d_lock);
979 __dentry_kill(dentry);
985 static enum lru_status dentry_lru_isolate(struct list_head *item,
986 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
988 struct list_head *freeable = arg;
989 struct dentry *dentry = container_of(item, struct dentry, d_lru);
993 * we are inverting the lru lock/dentry->d_lock here,
994 * so use a trylock. If we fail to get the lock, just skip
997 if (!spin_trylock(&dentry->d_lock))
1001 * Referenced dentries are still in use. If they have active
1002 * counts, just remove them from the LRU. Otherwise give them
1003 * another pass through the LRU.
1005 if (dentry->d_lockref.count) {
1006 d_lru_isolate(lru, dentry);
1007 spin_unlock(&dentry->d_lock);
1011 if (dentry->d_flags & DCACHE_REFERENCED) {
1012 dentry->d_flags &= ~DCACHE_REFERENCED;
1013 spin_unlock(&dentry->d_lock);
1016 * The list move itself will be made by the common LRU code. At
1017 * this point, we've dropped the dentry->d_lock but keep the
1018 * lru lock. This is safe to do, since every list movement is
1019 * protected by the lru lock even if both locks are held.
1021 * This is guaranteed by the fact that all LRU management
1022 * functions are intermediated by the LRU API calls like
1023 * list_lru_add and list_lru_del. List movement in this file
1024 * only ever occur through this functions or through callbacks
1025 * like this one, that are called from the LRU API.
1027 * The only exceptions to this are functions like
1028 * shrink_dentry_list, and code that first checks for the
1029 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1030 * operating only with stack provided lists after they are
1031 * properly isolated from the main list. It is thus, always a
1037 d_lru_shrink_move(lru, dentry, freeable);
1038 spin_unlock(&dentry->d_lock);
1044 * prune_dcache_sb - shrink the dcache
1046 * @sc: shrink control, passed to list_lru_shrink_walk()
1048 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1049 * is done when we need more memory and called from the superblock shrinker
1052 * This function may fail to free any resources if all the dentries are in
1055 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1060 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1061 dentry_lru_isolate, &dispose);
1062 shrink_dentry_list(&dispose);
1066 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1067 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1069 struct list_head *freeable = arg;
1070 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1073 * we are inverting the lru lock/dentry->d_lock here,
1074 * so use a trylock. If we fail to get the lock, just skip
1077 if (!spin_trylock(&dentry->d_lock))
1080 d_lru_shrink_move(lru, dentry, freeable);
1081 spin_unlock(&dentry->d_lock);
1088 * shrink_dcache_sb - shrink dcache for a superblock
1091 * Shrink the dcache for the specified super block. This is used to free
1092 * the dcache before unmounting a file system.
1094 void shrink_dcache_sb(struct super_block *sb)
1101 freed = list_lru_walk(&sb->s_dentry_lru,
1102 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1104 this_cpu_sub(nr_dentry_unused, freed);
1105 shrink_dentry_list(&dispose);
1106 } while (freed > 0);
1108 EXPORT_SYMBOL(shrink_dcache_sb);
1111 * enum d_walk_ret - action to talke during tree walk
1112 * @D_WALK_CONTINUE: contrinue walk
1113 * @D_WALK_QUIT: quit walk
1114 * @D_WALK_NORETRY: quit when retry is needed
1115 * @D_WALK_SKIP: skip this dentry and its children
1125 * d_walk - walk the dentry tree
1126 * @parent: start of walk
1127 * @data: data passed to @enter() and @finish()
1128 * @enter: callback when first entering the dentry
1129 * @finish: callback when successfully finished the walk
1131 * The @enter() and @finish() callbacks are called with d_lock held.
1133 static void d_walk(struct dentry *parent, void *data,
1134 enum d_walk_ret (*enter)(void *, struct dentry *),
1135 void (*finish)(void *))
1137 struct dentry *this_parent;
1138 struct list_head *next;
1140 enum d_walk_ret ret;
1144 read_seqbegin_or_lock(&rename_lock, &seq);
1145 this_parent = parent;
1146 spin_lock(&this_parent->d_lock);
1148 ret = enter(data, this_parent);
1150 case D_WALK_CONTINUE:
1155 case D_WALK_NORETRY:
1160 next = this_parent->d_subdirs.next;
1162 while (next != &this_parent->d_subdirs) {
1163 struct list_head *tmp = next;
1164 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1167 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1169 ret = enter(data, dentry);
1171 case D_WALK_CONTINUE:
1174 spin_unlock(&dentry->d_lock);
1176 case D_WALK_NORETRY:
1180 spin_unlock(&dentry->d_lock);
1184 if (!list_empty(&dentry->d_subdirs)) {
1185 spin_unlock(&this_parent->d_lock);
1186 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1187 this_parent = dentry;
1188 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1191 spin_unlock(&dentry->d_lock);
1194 * All done at this level ... ascend and resume the search.
1198 if (this_parent != parent) {
1199 struct dentry *child = this_parent;
1200 this_parent = child->d_parent;
1202 spin_unlock(&child->d_lock);
1203 spin_lock(&this_parent->d_lock);
1205 /* might go back up the wrong parent if we have had a rename. */
1206 if (need_seqretry(&rename_lock, seq))
1208 next = child->d_child.next;
1209 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1210 if (next == &this_parent->d_subdirs)
1212 child = list_entry(next, struct dentry, d_child);
1218 if (need_seqretry(&rename_lock, seq))
1225 spin_unlock(&this_parent->d_lock);
1226 done_seqretry(&rename_lock, seq);
1230 spin_unlock(&this_parent->d_lock);
1240 * Search for at least 1 mount point in the dentry's subdirs.
1241 * We descend to the next level whenever the d_subdirs
1242 * list is non-empty and continue searching.
1245 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1248 if (d_mountpoint(dentry)) {
1252 return D_WALK_CONTINUE;
1256 * have_submounts - check for mounts over a dentry
1257 * @parent: dentry to check.
1259 * Return true if the parent or its subdirectories contain
1262 int have_submounts(struct dentry *parent)
1266 d_walk(parent, &ret, check_mount, NULL);
1270 EXPORT_SYMBOL(have_submounts);
1273 * Called by mount code to set a mountpoint and check if the mountpoint is
1274 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1275 * subtree can become unreachable).
1277 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1278 * this reason take rename_lock and d_lock on dentry and ancestors.
1280 int d_set_mounted(struct dentry *dentry)
1284 write_seqlock(&rename_lock);
1285 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1286 /* Need exclusion wrt. d_invalidate() */
1287 spin_lock(&p->d_lock);
1288 if (unlikely(d_unhashed(p))) {
1289 spin_unlock(&p->d_lock);
1292 spin_unlock(&p->d_lock);
1294 spin_lock(&dentry->d_lock);
1295 if (!d_unlinked(dentry)) {
1296 dentry->d_flags |= DCACHE_MOUNTED;
1299 spin_unlock(&dentry->d_lock);
1301 write_sequnlock(&rename_lock);
1306 * Search the dentry child list of the specified parent,
1307 * and move any unused dentries to the end of the unused
1308 * list for prune_dcache(). We descend to the next level
1309 * whenever the d_subdirs list is non-empty and continue
1312 * It returns zero iff there are no unused children,
1313 * otherwise it returns the number of children moved to
1314 * the end of the unused list. This may not be the total
1315 * number of unused children, because select_parent can
1316 * drop the lock and return early due to latency
1320 struct select_data {
1321 struct dentry *start;
1322 struct list_head dispose;
1326 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1328 struct select_data *data = _data;
1329 enum d_walk_ret ret = D_WALK_CONTINUE;
1331 if (data->start == dentry)
1334 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1337 if (dentry->d_flags & DCACHE_LRU_LIST)
1339 if (!dentry->d_lockref.count) {
1340 d_shrink_add(dentry, &data->dispose);
1345 * We can return to the caller if we have found some (this
1346 * ensures forward progress). We'll be coming back to find
1349 if (!list_empty(&data->dispose))
1350 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1356 * shrink_dcache_parent - prune dcache
1357 * @parent: parent of entries to prune
1359 * Prune the dcache to remove unused children of the parent dentry.
1361 void shrink_dcache_parent(struct dentry *parent)
1364 struct select_data data;
1366 INIT_LIST_HEAD(&data.dispose);
1367 data.start = parent;
1370 d_walk(parent, &data, select_collect, NULL);
1374 shrink_dentry_list(&data.dispose);
1378 EXPORT_SYMBOL(shrink_dcache_parent);
1380 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1382 /* it has busy descendents; complain about those instead */
1383 if (!list_empty(&dentry->d_subdirs))
1384 return D_WALK_CONTINUE;
1386 /* root with refcount 1 is fine */
1387 if (dentry == _data && dentry->d_lockref.count == 1)
1388 return D_WALK_CONTINUE;
1390 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1391 " still in use (%d) [unmount of %s %s]\n",
1394 dentry->d_inode->i_ino : 0UL,
1396 dentry->d_lockref.count,
1397 dentry->d_sb->s_type->name,
1398 dentry->d_sb->s_id);
1400 return D_WALK_CONTINUE;
1403 static void do_one_tree(struct dentry *dentry)
1405 shrink_dcache_parent(dentry);
1406 d_walk(dentry, dentry, umount_check, NULL);
1412 * destroy the dentries attached to a superblock on unmounting
1414 void shrink_dcache_for_umount(struct super_block *sb)
1416 struct dentry *dentry;
1418 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1420 dentry = sb->s_root;
1422 do_one_tree(dentry);
1424 while (!hlist_bl_empty(&sb->s_anon)) {
1425 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1426 do_one_tree(dentry);
1430 struct detach_data {
1431 struct select_data select;
1432 struct dentry *mountpoint;
1434 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1436 struct detach_data *data = _data;
1438 if (d_mountpoint(dentry)) {
1439 __dget_dlock(dentry);
1440 data->mountpoint = dentry;
1444 return select_collect(&data->select, dentry);
1447 static void check_and_drop(void *_data)
1449 struct detach_data *data = _data;
1451 if (!data->mountpoint && !data->select.found)
1452 __d_drop(data->select.start);
1456 * d_invalidate - detach submounts, prune dcache, and drop
1457 * @dentry: dentry to invalidate (aka detach, prune and drop)
1461 * The final d_drop is done as an atomic operation relative to
1462 * rename_lock ensuring there are no races with d_set_mounted. This
1463 * ensures there are no unhashed dentries on the path to a mountpoint.
1465 void d_invalidate(struct dentry *dentry)
1468 * If it's already been dropped, return OK.
1470 spin_lock(&dentry->d_lock);
1471 if (d_unhashed(dentry)) {
1472 spin_unlock(&dentry->d_lock);
1475 spin_unlock(&dentry->d_lock);
1477 /* Negative dentries can be dropped without further checks */
1478 if (!dentry->d_inode) {
1484 struct detach_data data;
1486 data.mountpoint = NULL;
1487 INIT_LIST_HEAD(&data.select.dispose);
1488 data.select.start = dentry;
1489 data.select.found = 0;
1491 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1493 if (data.select.found)
1494 shrink_dentry_list(&data.select.dispose);
1496 if (data.mountpoint) {
1497 detach_mounts(data.mountpoint);
1498 dput(data.mountpoint);
1501 if (!data.mountpoint && !data.select.found)
1507 EXPORT_SYMBOL(d_invalidate);
1510 * __d_alloc - allocate a dcache entry
1511 * @sb: filesystem it will belong to
1512 * @name: qstr of the name
1514 * Allocates a dentry. It returns %NULL if there is insufficient memory
1515 * available. On a success the dentry is returned. The name passed in is
1516 * copied and the copy passed in may be reused after this call.
1519 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1521 struct dentry *dentry;
1524 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1529 * We guarantee that the inline name is always NUL-terminated.
1530 * This way the memcpy() done by the name switching in rename
1531 * will still always have a NUL at the end, even if we might
1532 * be overwriting an internal NUL character
1534 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1535 if (name->len > DNAME_INLINE_LEN-1) {
1536 size_t size = offsetof(struct external_name, name[1]);
1537 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1539 kmem_cache_free(dentry_cache, dentry);
1542 atomic_set(&p->u.count, 1);
1544 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1545 kasan_unpoison_shadow(dname,
1546 round_up(name->len + 1, sizeof(unsigned long)));
1548 dname = dentry->d_iname;
1551 dentry->d_name.len = name->len;
1552 dentry->d_name.hash = name->hash;
1553 memcpy(dname, name->name, name->len);
1554 dname[name->len] = 0;
1556 /* Make sure we always see the terminating NUL character */
1558 dentry->d_name.name = dname;
1560 dentry->d_lockref.count = 1;
1561 dentry->d_flags = 0;
1562 spin_lock_init(&dentry->d_lock);
1563 seqcount_init(&dentry->d_seq);
1564 dentry->d_inode = NULL;
1565 dentry->d_parent = dentry;
1567 dentry->d_op = NULL;
1568 dentry->d_fsdata = NULL;
1569 INIT_HLIST_BL_NODE(&dentry->d_hash);
1570 INIT_LIST_HEAD(&dentry->d_lru);
1571 INIT_LIST_HEAD(&dentry->d_subdirs);
1572 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1573 INIT_LIST_HEAD(&dentry->d_child);
1574 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1576 this_cpu_inc(nr_dentry);
1582 * d_alloc - allocate a dcache entry
1583 * @parent: parent of entry to allocate
1584 * @name: qstr of the name
1586 * Allocates a dentry. It returns %NULL if there is insufficient memory
1587 * available. On a success the dentry is returned. The name passed in is
1588 * copied and the copy passed in may be reused after this call.
1590 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1592 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1596 spin_lock(&parent->d_lock);
1598 * don't need child lock because it is not subject
1599 * to concurrency here
1601 __dget_dlock(parent);
1602 dentry->d_parent = parent;
1603 list_add(&dentry->d_child, &parent->d_subdirs);
1604 spin_unlock(&parent->d_lock);
1608 EXPORT_SYMBOL(d_alloc);
1611 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1612 * @sb: the superblock
1613 * @name: qstr of the name
1615 * For a filesystem that just pins its dentries in memory and never
1616 * performs lookups at all, return an unhashed IS_ROOT dentry.
1618 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1620 return __d_alloc(sb, name);
1622 EXPORT_SYMBOL(d_alloc_pseudo);
1624 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1629 q.len = strlen(name);
1630 q.hash = full_name_hash(q.name, q.len);
1631 return d_alloc(parent, &q);
1633 EXPORT_SYMBOL(d_alloc_name);
1635 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1637 WARN_ON_ONCE(dentry->d_op);
1638 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1640 DCACHE_OP_REVALIDATE |
1641 DCACHE_OP_WEAK_REVALIDATE |
1642 DCACHE_OP_DELETE ));
1647 dentry->d_flags |= DCACHE_OP_HASH;
1649 dentry->d_flags |= DCACHE_OP_COMPARE;
1650 if (op->d_revalidate)
1651 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1652 if (op->d_weak_revalidate)
1653 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1655 dentry->d_flags |= DCACHE_OP_DELETE;
1657 dentry->d_flags |= DCACHE_OP_PRUNE;
1660 EXPORT_SYMBOL(d_set_d_op);
1662 static unsigned d_flags_for_inode(struct inode *inode)
1664 unsigned add_flags = DCACHE_FILE_TYPE;
1667 return DCACHE_MISS_TYPE;
1669 if (S_ISDIR(inode->i_mode)) {
1670 add_flags = DCACHE_DIRECTORY_TYPE;
1671 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1672 if (unlikely(!inode->i_op->lookup))
1673 add_flags = DCACHE_AUTODIR_TYPE;
1675 inode->i_opflags |= IOP_LOOKUP;
1677 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1678 if (unlikely(inode->i_op->follow_link))
1679 add_flags = DCACHE_SYMLINK_TYPE;
1681 inode->i_opflags |= IOP_NOFOLLOW;
1684 if (unlikely(IS_AUTOMOUNT(inode)))
1685 add_flags |= DCACHE_NEED_AUTOMOUNT;
1689 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1691 unsigned add_flags = d_flags_for_inode(inode);
1693 spin_lock(&dentry->d_lock);
1694 __d_set_type(dentry, add_flags);
1696 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1697 dentry->d_inode = inode;
1698 dentry_rcuwalk_barrier(dentry);
1699 spin_unlock(&dentry->d_lock);
1700 fsnotify_d_instantiate(dentry, inode);
1704 * d_instantiate - fill in inode information for a dentry
1705 * @entry: dentry to complete
1706 * @inode: inode to attach to this dentry
1708 * Fill in inode information in the entry.
1710 * This turns negative dentries into productive full members
1713 * NOTE! This assumes that the inode count has been incremented
1714 * (or otherwise set) by the caller to indicate that it is now
1715 * in use by the dcache.
1718 void d_instantiate(struct dentry *entry, struct inode * inode)
1720 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1722 spin_lock(&inode->i_lock);
1723 __d_instantiate(entry, inode);
1725 spin_unlock(&inode->i_lock);
1726 security_d_instantiate(entry, inode);
1728 EXPORT_SYMBOL(d_instantiate);
1731 * d_instantiate_unique - instantiate a non-aliased dentry
1732 * @entry: dentry to instantiate
1733 * @inode: inode to attach to this dentry
1735 * Fill in inode information in the entry. On success, it returns NULL.
1736 * If an unhashed alias of "entry" already exists, then we return the
1737 * aliased dentry instead and drop one reference to inode.
1739 * Note that in order to avoid conflicts with rename() etc, the caller
1740 * had better be holding the parent directory semaphore.
1742 * This also assumes that the inode count has been incremented
1743 * (or otherwise set) by the caller to indicate that it is now
1744 * in use by the dcache.
1746 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1747 struct inode *inode)
1749 struct dentry *alias;
1750 int len = entry->d_name.len;
1751 const char *name = entry->d_name.name;
1752 unsigned int hash = entry->d_name.hash;
1755 __d_instantiate(entry, NULL);
1759 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1761 * Don't need alias->d_lock here, because aliases with
1762 * d_parent == entry->d_parent are not subject to name or
1763 * parent changes, because the parent inode i_mutex is held.
1765 if (alias->d_name.hash != hash)
1767 if (alias->d_parent != entry->d_parent)
1769 if (alias->d_name.len != len)
1771 if (dentry_cmp(alias, name, len))
1777 __d_instantiate(entry, inode);
1781 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1783 struct dentry *result;
1785 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1788 spin_lock(&inode->i_lock);
1789 result = __d_instantiate_unique(entry, inode);
1791 spin_unlock(&inode->i_lock);
1794 security_d_instantiate(entry, inode);
1798 BUG_ON(!d_unhashed(result));
1803 EXPORT_SYMBOL(d_instantiate_unique);
1806 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1807 * @entry: dentry to complete
1808 * @inode: inode to attach to this dentry
1810 * Fill in inode information in the entry. If a directory alias is found, then
1811 * return an error (and drop inode). Together with d_materialise_unique() this
1812 * guarantees that a directory inode may never have more than one alias.
1814 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1816 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1818 spin_lock(&inode->i_lock);
1819 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1820 spin_unlock(&inode->i_lock);
1824 __d_instantiate(entry, inode);
1825 spin_unlock(&inode->i_lock);
1826 security_d_instantiate(entry, inode);
1830 EXPORT_SYMBOL(d_instantiate_no_diralias);
1832 struct dentry *d_make_root(struct inode *root_inode)
1834 struct dentry *res = NULL;
1837 static const struct qstr name = QSTR_INIT("/", 1);
1839 res = __d_alloc(root_inode->i_sb, &name);
1841 d_instantiate(res, root_inode);
1847 EXPORT_SYMBOL(d_make_root);
1849 static struct dentry * __d_find_any_alias(struct inode *inode)
1851 struct dentry *alias;
1853 if (hlist_empty(&inode->i_dentry))
1855 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1861 * d_find_any_alias - find any alias for a given inode
1862 * @inode: inode to find an alias for
1864 * If any aliases exist for the given inode, take and return a
1865 * reference for one of them. If no aliases exist, return %NULL.
1867 struct dentry *d_find_any_alias(struct inode *inode)
1871 spin_lock(&inode->i_lock);
1872 de = __d_find_any_alias(inode);
1873 spin_unlock(&inode->i_lock);
1876 EXPORT_SYMBOL(d_find_any_alias);
1878 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1880 static const struct qstr anonstring = QSTR_INIT("/", 1);
1886 return ERR_PTR(-ESTALE);
1888 return ERR_CAST(inode);
1890 res = d_find_any_alias(inode);
1894 tmp = __d_alloc(inode->i_sb, &anonstring);
1896 res = ERR_PTR(-ENOMEM);
1900 spin_lock(&inode->i_lock);
1901 res = __d_find_any_alias(inode);
1903 spin_unlock(&inode->i_lock);
1908 /* attach a disconnected dentry */
1909 add_flags = d_flags_for_inode(inode);
1912 add_flags |= DCACHE_DISCONNECTED;
1914 spin_lock(&tmp->d_lock);
1915 tmp->d_inode = inode;
1916 tmp->d_flags |= add_flags;
1917 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1918 hlist_bl_lock(&tmp->d_sb->s_anon);
1919 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1920 hlist_bl_unlock(&tmp->d_sb->s_anon);
1921 spin_unlock(&tmp->d_lock);
1922 spin_unlock(&inode->i_lock);
1923 security_d_instantiate(tmp, inode);
1928 if (res && !IS_ERR(res))
1929 security_d_instantiate(res, inode);
1935 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1936 * @inode: inode to allocate the dentry for
1938 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1939 * similar open by handle operations. The returned dentry may be anonymous,
1940 * or may have a full name (if the inode was already in the cache).
1942 * When called on a directory inode, we must ensure that the inode only ever
1943 * has one dentry. If a dentry is found, that is returned instead of
1944 * allocating a new one.
1946 * On successful return, the reference to the inode has been transferred
1947 * to the dentry. In case of an error the reference on the inode is released.
1948 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1949 * be passed in and the error will be propagated to the return value,
1950 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1952 struct dentry *d_obtain_alias(struct inode *inode)
1954 return __d_obtain_alias(inode, 1);
1956 EXPORT_SYMBOL(d_obtain_alias);
1959 * d_obtain_root - find or allocate a dentry for a given inode
1960 * @inode: inode to allocate the dentry for
1962 * Obtain an IS_ROOT dentry for the root of a filesystem.
1964 * We must ensure that directory inodes only ever have one dentry. If a
1965 * dentry is found, that is returned instead of allocating a new one.
1967 * On successful return, the reference to the inode has been transferred
1968 * to the dentry. In case of an error the reference on the inode is
1969 * released. A %NULL or IS_ERR inode may be passed in and will be the
1970 * error will be propagate to the return value, with a %NULL @inode
1971 * replaced by ERR_PTR(-ESTALE).
1973 struct dentry *d_obtain_root(struct inode *inode)
1975 return __d_obtain_alias(inode, 0);
1977 EXPORT_SYMBOL(d_obtain_root);
1980 * d_add_ci - lookup or allocate new dentry with case-exact name
1981 * @inode: the inode case-insensitive lookup has found
1982 * @dentry: the negative dentry that was passed to the parent's lookup func
1983 * @name: the case-exact name to be associated with the returned dentry
1985 * This is to avoid filling the dcache with case-insensitive names to the
1986 * same inode, only the actual correct case is stored in the dcache for
1987 * case-insensitive filesystems.
1989 * For a case-insensitive lookup match and if the the case-exact dentry
1990 * already exists in in the dcache, use it and return it.
1992 * If no entry exists with the exact case name, allocate new dentry with
1993 * the exact case, and return the spliced entry.
1995 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1998 struct dentry *found;
2002 * First check if a dentry matching the name already exists,
2003 * if not go ahead and create it now.
2005 found = d_hash_and_lookup(dentry->d_parent, name);
2007 new = d_alloc(dentry->d_parent, name);
2009 found = ERR_PTR(-ENOMEM);
2011 found = d_splice_alias(inode, new);
2022 EXPORT_SYMBOL(d_add_ci);
2025 * Do the slow-case of the dentry name compare.
2027 * Unlike the dentry_cmp() function, we need to atomically
2028 * load the name and length information, so that the
2029 * filesystem can rely on them, and can use the 'name' and
2030 * 'len' information without worrying about walking off the
2031 * end of memory etc.
2033 * Thus the read_seqcount_retry() and the "duplicate" info
2034 * in arguments (the low-level filesystem should not look
2035 * at the dentry inode or name contents directly, since
2036 * rename can change them while we're in RCU mode).
2038 enum slow_d_compare {
2044 static noinline enum slow_d_compare slow_dentry_cmp(
2045 const struct dentry *parent,
2046 struct dentry *dentry,
2048 const struct qstr *name)
2050 int tlen = dentry->d_name.len;
2051 const char *tname = dentry->d_name.name;
2053 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2055 return D_COMP_SEQRETRY;
2057 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2058 return D_COMP_NOMATCH;
2063 * __d_lookup_rcu - search for a dentry (racy, store-free)
2064 * @parent: parent dentry
2065 * @name: qstr of name we wish to find
2066 * @seqp: returns d_seq value at the point where the dentry was found
2067 * Returns: dentry, or NULL
2069 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2070 * resolution (store-free path walking) design described in
2071 * Documentation/filesystems/path-lookup.txt.
2073 * This is not to be used outside core vfs.
2075 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2076 * held, and rcu_read_lock held. The returned dentry must not be stored into
2077 * without taking d_lock and checking d_seq sequence count against @seq
2080 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2083 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2084 * the returned dentry, so long as its parent's seqlock is checked after the
2085 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2086 * is formed, giving integrity down the path walk.
2088 * NOTE! The caller *has* to check the resulting dentry against the sequence
2089 * number we've returned before using any of the resulting dentry state!
2091 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2092 const struct qstr *name,
2095 u64 hashlen = name->hash_len;
2096 const unsigned char *str = name->name;
2097 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2098 struct hlist_bl_node *node;
2099 struct dentry *dentry;
2102 * Note: There is significant duplication with __d_lookup_rcu which is
2103 * required to prevent single threaded performance regressions
2104 * especially on architectures where smp_rmb (in seqcounts) are costly.
2105 * Keep the two functions in sync.
2109 * The hash list is protected using RCU.
2111 * Carefully use d_seq when comparing a candidate dentry, to avoid
2112 * races with d_move().
2114 * It is possible that concurrent renames can mess up our list
2115 * walk here and result in missing our dentry, resulting in the
2116 * false-negative result. d_lookup() protects against concurrent
2117 * renames using rename_lock seqlock.
2119 * See Documentation/filesystems/path-lookup.txt for more details.
2121 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2126 * The dentry sequence count protects us from concurrent
2127 * renames, and thus protects parent and name fields.
2129 * The caller must perform a seqcount check in order
2130 * to do anything useful with the returned dentry.
2132 * NOTE! We do a "raw" seqcount_begin here. That means that
2133 * we don't wait for the sequence count to stabilize if it
2134 * is in the middle of a sequence change. If we do the slow
2135 * dentry compare, we will do seqretries until it is stable,
2136 * and if we end up with a successful lookup, we actually
2137 * want to exit RCU lookup anyway.
2139 seq = raw_seqcount_begin(&dentry->d_seq);
2140 if (dentry->d_parent != parent)
2142 if (d_unhashed(dentry))
2145 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2146 if (dentry->d_name.hash != hashlen_hash(hashlen))
2149 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2152 case D_COMP_NOMATCH:
2159 if (dentry->d_name.hash_len != hashlen)
2162 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2169 * d_lookup - search for a dentry
2170 * @parent: parent dentry
2171 * @name: qstr of name we wish to find
2172 * Returns: dentry, or NULL
2174 * d_lookup searches the children of the parent dentry for the name in
2175 * question. If the dentry is found its reference count is incremented and the
2176 * dentry is returned. The caller must use dput to free the entry when it has
2177 * finished using it. %NULL is returned if the dentry does not exist.
2179 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2181 struct dentry *dentry;
2185 seq = read_seqbegin(&rename_lock);
2186 dentry = __d_lookup(parent, name);
2189 } while (read_seqretry(&rename_lock, seq));
2192 EXPORT_SYMBOL(d_lookup);
2195 * __d_lookup - search for a dentry (racy)
2196 * @parent: parent dentry
2197 * @name: qstr of name we wish to find
2198 * Returns: dentry, or NULL
2200 * __d_lookup is like d_lookup, however it may (rarely) return a
2201 * false-negative result due to unrelated rename activity.
2203 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2204 * however it must be used carefully, eg. with a following d_lookup in
2205 * the case of failure.
2207 * __d_lookup callers must be commented.
2209 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2211 unsigned int len = name->len;
2212 unsigned int hash = name->hash;
2213 const unsigned char *str = name->name;
2214 struct hlist_bl_head *b = d_hash(parent, hash);
2215 struct hlist_bl_node *node;
2216 struct dentry *found = NULL;
2217 struct dentry *dentry;
2220 * Note: There is significant duplication with __d_lookup_rcu which is
2221 * required to prevent single threaded performance regressions
2222 * especially on architectures where smp_rmb (in seqcounts) are costly.
2223 * Keep the two functions in sync.
2227 * The hash list is protected using RCU.
2229 * Take d_lock when comparing a candidate dentry, to avoid races
2232 * It is possible that concurrent renames can mess up our list
2233 * walk here and result in missing our dentry, resulting in the
2234 * false-negative result. d_lookup() protects against concurrent
2235 * renames using rename_lock seqlock.
2237 * See Documentation/filesystems/path-lookup.txt for more details.
2241 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2243 if (dentry->d_name.hash != hash)
2246 spin_lock(&dentry->d_lock);
2247 if (dentry->d_parent != parent)
2249 if (d_unhashed(dentry))
2253 * It is safe to compare names since d_move() cannot
2254 * change the qstr (protected by d_lock).
2256 if (parent->d_flags & DCACHE_OP_COMPARE) {
2257 int tlen = dentry->d_name.len;
2258 const char *tname = dentry->d_name.name;
2259 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2262 if (dentry->d_name.len != len)
2264 if (dentry_cmp(dentry, str, len))
2268 dentry->d_lockref.count++;
2270 spin_unlock(&dentry->d_lock);
2273 spin_unlock(&dentry->d_lock);
2281 * d_hash_and_lookup - hash the qstr then search for a dentry
2282 * @dir: Directory to search in
2283 * @name: qstr of name we wish to find
2285 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2287 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2290 * Check for a fs-specific hash function. Note that we must
2291 * calculate the standard hash first, as the d_op->d_hash()
2292 * routine may choose to leave the hash value unchanged.
2294 name->hash = full_name_hash(name->name, name->len);
2295 if (dir->d_flags & DCACHE_OP_HASH) {
2296 int err = dir->d_op->d_hash(dir, name);
2297 if (unlikely(err < 0))
2298 return ERR_PTR(err);
2300 return d_lookup(dir, name);
2302 EXPORT_SYMBOL(d_hash_and_lookup);
2305 * When a file is deleted, we have two options:
2306 * - turn this dentry into a negative dentry
2307 * - unhash this dentry and free it.
2309 * Usually, we want to just turn this into
2310 * a negative dentry, but if anybody else is
2311 * currently using the dentry or the inode
2312 * we can't do that and we fall back on removing
2313 * it from the hash queues and waiting for
2314 * it to be deleted later when it has no users
2318 * d_delete - delete a dentry
2319 * @dentry: The dentry to delete
2321 * Turn the dentry into a negative dentry if possible, otherwise
2322 * remove it from the hash queues so it can be deleted later
2325 void d_delete(struct dentry * dentry)
2327 struct inode *inode;
2330 * Are we the only user?
2333 spin_lock(&dentry->d_lock);
2334 inode = dentry->d_inode;
2335 isdir = S_ISDIR(inode->i_mode);
2336 if (dentry->d_lockref.count == 1) {
2337 if (!spin_trylock(&inode->i_lock)) {
2338 spin_unlock(&dentry->d_lock);
2342 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2343 dentry_unlink_inode(dentry);
2344 fsnotify_nameremove(dentry, isdir);
2348 if (!d_unhashed(dentry))
2351 spin_unlock(&dentry->d_lock);
2353 fsnotify_nameremove(dentry, isdir);
2355 EXPORT_SYMBOL(d_delete);
2357 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2359 BUG_ON(!d_unhashed(entry));
2361 entry->d_flags |= DCACHE_RCUACCESS;
2362 hlist_bl_add_head_rcu(&entry->d_hash, b);
2366 static void _d_rehash(struct dentry * entry)
2368 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2372 * d_rehash - add an entry back to the hash
2373 * @entry: dentry to add to the hash
2375 * Adds a dentry to the hash according to its name.
2378 void d_rehash(struct dentry * entry)
2380 spin_lock(&entry->d_lock);
2382 spin_unlock(&entry->d_lock);
2384 EXPORT_SYMBOL(d_rehash);
2387 * dentry_update_name_case - update case insensitive dentry with a new name
2388 * @dentry: dentry to be updated
2391 * Update a case insensitive dentry with new case of name.
2393 * dentry must have been returned by d_lookup with name @name. Old and new
2394 * name lengths must match (ie. no d_compare which allows mismatched name
2397 * Parent inode i_mutex must be held over d_lookup and into this call (to
2398 * keep renames and concurrent inserts, and readdir(2) away).
2400 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2402 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2403 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2405 spin_lock(&dentry->d_lock);
2406 write_seqcount_begin(&dentry->d_seq);
2407 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2408 write_seqcount_end(&dentry->d_seq);
2409 spin_unlock(&dentry->d_lock);
2411 EXPORT_SYMBOL(dentry_update_name_case);
2413 static void swap_names(struct dentry *dentry, struct dentry *target)
2415 if (unlikely(dname_external(target))) {
2416 if (unlikely(dname_external(dentry))) {
2418 * Both external: swap the pointers
2420 swap(target->d_name.name, dentry->d_name.name);
2423 * dentry:internal, target:external. Steal target's
2424 * storage and make target internal.
2426 memcpy(target->d_iname, dentry->d_name.name,
2427 dentry->d_name.len + 1);
2428 dentry->d_name.name = target->d_name.name;
2429 target->d_name.name = target->d_iname;
2432 if (unlikely(dname_external(dentry))) {
2434 * dentry:external, target:internal. Give dentry's
2435 * storage to target and make dentry internal
2437 memcpy(dentry->d_iname, target->d_name.name,
2438 target->d_name.len + 1);
2439 target->d_name.name = dentry->d_name.name;
2440 dentry->d_name.name = dentry->d_iname;
2443 * Both are internal.
2446 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2447 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2448 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2449 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2450 swap(((long *) &dentry->d_iname)[i],
2451 ((long *) &target->d_iname)[i]);
2455 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2458 static void copy_name(struct dentry *dentry, struct dentry *target)
2460 struct external_name *old_name = NULL;
2461 if (unlikely(dname_external(dentry)))
2462 old_name = external_name(dentry);
2463 if (unlikely(dname_external(target))) {
2464 atomic_inc(&external_name(target)->u.count);
2465 dentry->d_name = target->d_name;
2467 memcpy(dentry->d_iname, target->d_name.name,
2468 target->d_name.len + 1);
2469 dentry->d_name.name = dentry->d_iname;
2470 dentry->d_name.hash_len = target->d_name.hash_len;
2472 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2473 kfree_rcu(old_name, u.head);
2476 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2479 * XXXX: do we really need to take target->d_lock?
2481 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2482 spin_lock(&target->d_parent->d_lock);
2484 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2485 spin_lock(&dentry->d_parent->d_lock);
2486 spin_lock_nested(&target->d_parent->d_lock,
2487 DENTRY_D_LOCK_NESTED);
2489 spin_lock(&target->d_parent->d_lock);
2490 spin_lock_nested(&dentry->d_parent->d_lock,
2491 DENTRY_D_LOCK_NESTED);
2494 if (target < dentry) {
2495 spin_lock_nested(&target->d_lock, 2);
2496 spin_lock_nested(&dentry->d_lock, 3);
2498 spin_lock_nested(&dentry->d_lock, 2);
2499 spin_lock_nested(&target->d_lock, 3);
2503 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2505 if (target->d_parent != dentry->d_parent)
2506 spin_unlock(&dentry->d_parent->d_lock);
2507 if (target->d_parent != target)
2508 spin_unlock(&target->d_parent->d_lock);
2509 spin_unlock(&target->d_lock);
2510 spin_unlock(&dentry->d_lock);
2514 * When switching names, the actual string doesn't strictly have to
2515 * be preserved in the target - because we're dropping the target
2516 * anyway. As such, we can just do a simple memcpy() to copy over
2517 * the new name before we switch, unless we are going to rehash
2518 * it. Note that if we *do* unhash the target, we are not allowed
2519 * to rehash it without giving it a new name/hash key - whether
2520 * we swap or overwrite the names here, resulting name won't match
2521 * the reality in filesystem; it's only there for d_path() purposes.
2522 * Note that all of this is happening under rename_lock, so the
2523 * any hash lookup seeing it in the middle of manipulations will
2524 * be discarded anyway. So we do not care what happens to the hash
2528 * __d_move - move a dentry
2529 * @dentry: entry to move
2530 * @target: new dentry
2531 * @exchange: exchange the two dentries
2533 * Update the dcache to reflect the move of a file name. Negative
2534 * dcache entries should not be moved in this way. Caller must hold
2535 * rename_lock, the i_mutex of the source and target directories,
2536 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2538 static void __d_move(struct dentry *dentry, struct dentry *target,
2541 if (!dentry->d_inode)
2542 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2544 BUG_ON(d_ancestor(dentry, target));
2545 BUG_ON(d_ancestor(target, dentry));
2547 dentry_lock_for_move(dentry, target);
2549 write_seqcount_begin(&dentry->d_seq);
2550 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2552 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2555 * Move the dentry to the target hash queue. Don't bother checking
2556 * for the same hash queue because of how unlikely it is.
2559 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2562 * Unhash the target (d_delete() is not usable here). If exchanging
2563 * the two dentries, then rehash onto the other's hash queue.
2568 d_hash(dentry->d_parent, dentry->d_name.hash));
2571 /* Switch the names.. */
2573 swap_names(dentry, target);
2575 copy_name(dentry, target);
2577 /* ... and switch them in the tree */
2578 if (IS_ROOT(dentry)) {
2579 /* splicing a tree */
2580 dentry->d_parent = target->d_parent;
2581 target->d_parent = target;
2582 list_del_init(&target->d_child);
2583 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2585 /* swapping two dentries */
2586 swap(dentry->d_parent, target->d_parent);
2587 list_move(&target->d_child, &target->d_parent->d_subdirs);
2588 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2590 fsnotify_d_move(target);
2591 fsnotify_d_move(dentry);
2594 write_seqcount_end(&target->d_seq);
2595 write_seqcount_end(&dentry->d_seq);
2597 dentry_unlock_for_move(dentry, target);
2601 * d_move - move a dentry
2602 * @dentry: entry to move
2603 * @target: new dentry
2605 * Update the dcache to reflect the move of a file name. Negative
2606 * dcache entries should not be moved in this way. See the locking
2607 * requirements for __d_move.
2609 void d_move(struct dentry *dentry, struct dentry *target)
2611 write_seqlock(&rename_lock);
2612 __d_move(dentry, target, false);
2613 write_sequnlock(&rename_lock);
2615 EXPORT_SYMBOL(d_move);
2618 * d_exchange - exchange two dentries
2619 * @dentry1: first dentry
2620 * @dentry2: second dentry
2622 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2624 write_seqlock(&rename_lock);
2626 WARN_ON(!dentry1->d_inode);
2627 WARN_ON(!dentry2->d_inode);
2628 WARN_ON(IS_ROOT(dentry1));
2629 WARN_ON(IS_ROOT(dentry2));
2631 __d_move(dentry1, dentry2, true);
2633 write_sequnlock(&rename_lock);
2637 * d_ancestor - search for an ancestor
2638 * @p1: ancestor dentry
2641 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2642 * an ancestor of p2, else NULL.
2644 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2648 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2649 if (p->d_parent == p1)
2656 * This helper attempts to cope with remotely renamed directories
2658 * It assumes that the caller is already holding
2659 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2661 * Note: If ever the locking in lock_rename() changes, then please
2662 * remember to update this too...
2664 static int __d_unalias(struct inode *inode,
2665 struct dentry *dentry, struct dentry *alias)
2667 struct mutex *m1 = NULL, *m2 = NULL;
2670 /* If alias and dentry share a parent, then no extra locks required */
2671 if (alias->d_parent == dentry->d_parent)
2674 /* See lock_rename() */
2675 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2677 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2678 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2680 m2 = &alias->d_parent->d_inode->i_mutex;
2682 __d_move(alias, dentry, false);
2685 spin_unlock(&inode->i_lock);
2694 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2695 * @inode: the inode which may have a disconnected dentry
2696 * @dentry: a negative dentry which we want to point to the inode.
2698 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2699 * place of the given dentry and return it, else simply d_add the inode
2700 * to the dentry and return NULL.
2702 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2703 * we should error out: directories can't have multiple aliases.
2705 * This is needed in the lookup routine of any filesystem that is exportable
2706 * (via knfsd) so that we can build dcache paths to directories effectively.
2708 * If a dentry was found and moved, then it is returned. Otherwise NULL
2709 * is returned. This matches the expected return value of ->lookup.
2711 * Cluster filesystems may call this function with a negative, hashed dentry.
2712 * In that case, we know that the inode will be a regular file, and also this
2713 * will only occur during atomic_open. So we need to check for the dentry
2714 * being already hashed only in the final case.
2716 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2719 return ERR_CAST(inode);
2721 BUG_ON(!d_unhashed(dentry));
2724 __d_instantiate(dentry, NULL);
2727 spin_lock(&inode->i_lock);
2728 if (S_ISDIR(inode->i_mode)) {
2729 struct dentry *new = __d_find_any_alias(inode);
2730 if (unlikely(new)) {
2731 write_seqlock(&rename_lock);
2732 if (unlikely(d_ancestor(new, dentry))) {
2733 write_sequnlock(&rename_lock);
2734 spin_unlock(&inode->i_lock);
2736 new = ERR_PTR(-ELOOP);
2737 pr_warn_ratelimited(
2738 "VFS: Lookup of '%s' in %s %s"
2739 " would have caused loop\n",
2740 dentry->d_name.name,
2741 inode->i_sb->s_type->name,
2743 } else if (!IS_ROOT(new)) {
2744 int err = __d_unalias(inode, dentry, new);
2745 write_sequnlock(&rename_lock);
2751 __d_move(new, dentry, false);
2752 write_sequnlock(&rename_lock);
2753 spin_unlock(&inode->i_lock);
2754 security_d_instantiate(new, inode);
2760 /* already taking inode->i_lock, so d_add() by hand */
2761 __d_instantiate(dentry, inode);
2762 spin_unlock(&inode->i_lock);
2764 security_d_instantiate(dentry, inode);
2768 EXPORT_SYMBOL(d_splice_alias);
2770 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2774 return -ENAMETOOLONG;
2776 memcpy(*buffer, str, namelen);
2781 * prepend_name - prepend a pathname in front of current buffer pointer
2782 * @buffer: buffer pointer
2783 * @buflen: allocated length of the buffer
2784 * @name: name string and length qstr structure
2786 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2787 * make sure that either the old or the new name pointer and length are
2788 * fetched. However, there may be mismatch between length and pointer.
2789 * The length cannot be trusted, we need to copy it byte-by-byte until
2790 * the length is reached or a null byte is found. It also prepends "/" at
2791 * the beginning of the name. The sequence number check at the caller will
2792 * retry it again when a d_move() does happen. So any garbage in the buffer
2793 * due to mismatched pointer and length will be discarded.
2795 * Data dependency barrier is needed to make sure that we see that terminating
2796 * NUL. Alpha strikes again, film at 11...
2798 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2800 const char *dname = ACCESS_ONCE(name->name);
2801 u32 dlen = ACCESS_ONCE(name->len);
2804 smp_read_barrier_depends();
2806 *buflen -= dlen + 1;
2808 return -ENAMETOOLONG;
2809 p = *buffer -= dlen + 1;
2821 * prepend_path - Prepend path string to a buffer
2822 * @path: the dentry/vfsmount to report
2823 * @root: root vfsmnt/dentry
2824 * @buffer: pointer to the end of the buffer
2825 * @buflen: pointer to buffer length
2827 * The function will first try to write out the pathname without taking any
2828 * lock other than the RCU read lock to make sure that dentries won't go away.
2829 * It only checks the sequence number of the global rename_lock as any change
2830 * in the dentry's d_seq will be preceded by changes in the rename_lock
2831 * sequence number. If the sequence number had been changed, it will restart
2832 * the whole pathname back-tracing sequence again by taking the rename_lock.
2833 * In this case, there is no need to take the RCU read lock as the recursive
2834 * parent pointer references will keep the dentry chain alive as long as no
2835 * rename operation is performed.
2837 static int prepend_path(const struct path *path,
2838 const struct path *root,
2839 char **buffer, int *buflen)
2841 struct dentry *dentry;
2842 struct vfsmount *vfsmnt;
2845 unsigned seq, m_seq = 0;
2851 read_seqbegin_or_lock(&mount_lock, &m_seq);
2858 dentry = path->dentry;
2860 mnt = real_mount(vfsmnt);
2861 read_seqbegin_or_lock(&rename_lock, &seq);
2862 while (dentry != root->dentry || vfsmnt != root->mnt) {
2863 struct dentry * parent;
2865 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2866 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2868 if (mnt != parent) {
2869 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2875 * Filesystems needing to implement special "root names"
2876 * should do so with ->d_dname()
2878 if (IS_ROOT(dentry) &&
2879 (dentry->d_name.len != 1 ||
2880 dentry->d_name.name[0] != '/')) {
2881 WARN(1, "Root dentry has weird name <%.*s>\n",
2882 (int) dentry->d_name.len,
2883 dentry->d_name.name);
2886 error = is_mounted(vfsmnt) ? 1 : 2;
2889 parent = dentry->d_parent;
2891 error = prepend_name(&bptr, &blen, &dentry->d_name);
2899 if (need_seqretry(&rename_lock, seq)) {
2903 done_seqretry(&rename_lock, seq);
2907 if (need_seqretry(&mount_lock, m_seq)) {
2911 done_seqretry(&mount_lock, m_seq);
2913 if (error >= 0 && bptr == *buffer) {
2915 error = -ENAMETOOLONG;
2925 * __d_path - return the path of a dentry
2926 * @path: the dentry/vfsmount to report
2927 * @root: root vfsmnt/dentry
2928 * @buf: buffer to return value in
2929 * @buflen: buffer length
2931 * Convert a dentry into an ASCII path name.
2933 * Returns a pointer into the buffer or an error code if the
2934 * path was too long.
2936 * "buflen" should be positive.
2938 * If the path is not reachable from the supplied root, return %NULL.
2940 char *__d_path(const struct path *path,
2941 const struct path *root,
2942 char *buf, int buflen)
2944 char *res = buf + buflen;
2947 prepend(&res, &buflen, "\0", 1);
2948 error = prepend_path(path, root, &res, &buflen);
2951 return ERR_PTR(error);
2957 char *d_absolute_path(const struct path *path,
2958 char *buf, int buflen)
2960 struct path root = {};
2961 char *res = buf + buflen;
2964 prepend(&res, &buflen, "\0", 1);
2965 error = prepend_path(path, &root, &res, &buflen);
2970 return ERR_PTR(error);
2975 * same as __d_path but appends "(deleted)" for unlinked files.
2977 static int path_with_deleted(const struct path *path,
2978 const struct path *root,
2979 char **buf, int *buflen)
2981 prepend(buf, buflen, "\0", 1);
2982 if (d_unlinked(path->dentry)) {
2983 int error = prepend(buf, buflen, " (deleted)", 10);
2988 return prepend_path(path, root, buf, buflen);
2991 static int prepend_unreachable(char **buffer, int *buflen)
2993 return prepend(buffer, buflen, "(unreachable)", 13);
2996 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3001 seq = read_seqcount_begin(&fs->seq);
3003 } while (read_seqcount_retry(&fs->seq, seq));
3007 * d_path - return the path of a dentry
3008 * @path: path to report
3009 * @buf: buffer to return value in
3010 * @buflen: buffer length
3012 * Convert a dentry into an ASCII path name. If the entry has been deleted
3013 * the string " (deleted)" is appended. Note that this is ambiguous.
3015 * Returns a pointer into the buffer or an error code if the path was
3016 * too long. Note: Callers should use the returned pointer, not the passed
3017 * in buffer, to use the name! The implementation often starts at an offset
3018 * into the buffer, and may leave 0 bytes at the start.
3020 * "buflen" should be positive.
3022 char *d_path(const struct path *path, char *buf, int buflen)
3024 char *res = buf + buflen;
3029 * We have various synthetic filesystems that never get mounted. On
3030 * these filesystems dentries are never used for lookup purposes, and
3031 * thus don't need to be hashed. They also don't need a name until a
3032 * user wants to identify the object in /proc/pid/fd/. The little hack
3033 * below allows us to generate a name for these objects on demand:
3035 * Some pseudo inodes are mountable. When they are mounted
3036 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3037 * and instead have d_path return the mounted path.
3039 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3040 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3041 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3044 get_fs_root_rcu(current->fs, &root);
3045 error = path_with_deleted(path, &root, &res, &buflen);
3049 res = ERR_PTR(error);
3052 EXPORT_SYMBOL(d_path);
3055 * Helper function for dentry_operations.d_dname() members
3057 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3058 const char *fmt, ...)
3064 va_start(args, fmt);
3065 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3068 if (sz > sizeof(temp) || sz > buflen)
3069 return ERR_PTR(-ENAMETOOLONG);
3071 buffer += buflen - sz;
3072 return memcpy(buffer, temp, sz);
3075 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3077 char *end = buffer + buflen;
3078 /* these dentries are never renamed, so d_lock is not needed */
3079 if (prepend(&end, &buflen, " (deleted)", 11) ||
3080 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3081 prepend(&end, &buflen, "/", 1))
3082 end = ERR_PTR(-ENAMETOOLONG);
3085 EXPORT_SYMBOL(simple_dname);
3088 * Write full pathname from the root of the filesystem into the buffer.
3090 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3092 struct dentry *dentry;
3105 prepend(&end, &len, "\0", 1);
3109 read_seqbegin_or_lock(&rename_lock, &seq);
3110 while (!IS_ROOT(dentry)) {
3111 struct dentry *parent = dentry->d_parent;
3114 error = prepend_name(&end, &len, &dentry->d_name);
3123 if (need_seqretry(&rename_lock, seq)) {
3127 done_seqretry(&rename_lock, seq);
3132 return ERR_PTR(-ENAMETOOLONG);
3135 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3137 return __dentry_path(dentry, buf, buflen);
3139 EXPORT_SYMBOL(dentry_path_raw);
3141 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3146 if (d_unlinked(dentry)) {
3148 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3152 retval = __dentry_path(dentry, buf, buflen);
3153 if (!IS_ERR(retval) && p)
3154 *p = '/'; /* restore '/' overriden with '\0' */
3157 return ERR_PTR(-ENAMETOOLONG);
3160 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3166 seq = read_seqcount_begin(&fs->seq);
3169 } while (read_seqcount_retry(&fs->seq, seq));
3173 * NOTE! The user-level library version returns a
3174 * character pointer. The kernel system call just
3175 * returns the length of the buffer filled (which
3176 * includes the ending '\0' character), or a negative
3177 * error value. So libc would do something like
3179 * char *getcwd(char * buf, size_t size)
3183 * retval = sys_getcwd(buf, size);
3190 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3193 struct path pwd, root;
3194 char *page = __getname();
3200 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3203 if (!d_unlinked(pwd.dentry)) {
3205 char *cwd = page + PATH_MAX;
3206 int buflen = PATH_MAX;
3208 prepend(&cwd, &buflen, "\0", 1);
3209 error = prepend_path(&pwd, &root, &cwd, &buflen);
3215 /* Unreachable from current root */
3217 error = prepend_unreachable(&cwd, &buflen);
3223 len = PATH_MAX + page - cwd;
3226 if (copy_to_user(buf, cwd, len))
3239 * Test whether new_dentry is a subdirectory of old_dentry.
3241 * Trivially implemented using the dcache structure
3245 * is_subdir - is new dentry a subdirectory of old_dentry
3246 * @new_dentry: new dentry
3247 * @old_dentry: old dentry
3249 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3250 * Returns 0 otherwise.
3251 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3254 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3259 if (new_dentry == old_dentry)
3263 /* for restarting inner loop in case of seq retry */
3264 seq = read_seqbegin(&rename_lock);
3266 * Need rcu_readlock to protect against the d_parent trashing
3270 if (d_ancestor(old_dentry, new_dentry))
3275 } while (read_seqretry(&rename_lock, seq));
3280 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3282 struct dentry *root = data;
3283 if (dentry != root) {
3284 if (d_unhashed(dentry) || !dentry->d_inode)
3287 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3288 dentry->d_flags |= DCACHE_GENOCIDE;
3289 dentry->d_lockref.count--;
3292 return D_WALK_CONTINUE;
3295 void d_genocide(struct dentry *parent)
3297 d_walk(parent, parent, d_genocide_kill, NULL);
3300 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3302 inode_dec_link_count(inode);
3303 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3304 !hlist_unhashed(&dentry->d_u.d_alias) ||
3305 !d_unlinked(dentry));
3306 spin_lock(&dentry->d_parent->d_lock);
3307 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3308 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3309 (unsigned long long)inode->i_ino);
3310 spin_unlock(&dentry->d_lock);
3311 spin_unlock(&dentry->d_parent->d_lock);
3312 d_instantiate(dentry, inode);
3314 EXPORT_SYMBOL(d_tmpfile);
3316 static __initdata unsigned long dhash_entries;
3317 static int __init set_dhash_entries(char *str)
3321 dhash_entries = simple_strtoul(str, &str, 0);
3324 __setup("dhash_entries=", set_dhash_entries);
3326 static void __init dcache_init_early(void)
3330 /* If hashes are distributed across NUMA nodes, defer
3331 * hash allocation until vmalloc space is available.
3337 alloc_large_system_hash("Dentry cache",
3338 sizeof(struct hlist_bl_head),
3347 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3348 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3351 static void __init dcache_init(void)
3356 * A constructor could be added for stable state like the lists,
3357 * but it is probably not worth it because of the cache nature
3360 dentry_cache = KMEM_CACHE(dentry,
3361 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3363 /* Hash may have been set up in dcache_init_early */
3368 alloc_large_system_hash("Dentry cache",
3369 sizeof(struct hlist_bl_head),
3378 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3379 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3382 /* SLAB cache for __getname() consumers */
3383 struct kmem_cache *names_cachep __read_mostly;
3384 EXPORT_SYMBOL(names_cachep);
3386 EXPORT_SYMBOL(d_genocide);
3388 void __init vfs_caches_init_early(void)
3390 dcache_init_early();
3394 void __init vfs_caches_init(unsigned long mempages)
3396 unsigned long reserve;
3398 /* Base hash sizes on available memory, with a reserve equal to
3399 150% of current kernel size */
3401 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3402 mempages -= reserve;
3404 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3405 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3409 files_init(mempages);