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((int)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((int)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);
599 * This is complicated by the fact that we do not want to put
600 * dentries that are no longer on any hash chain on the unused
601 * list: we'd much rather just get rid of them immediately.
603 * However, that implies that we have to traverse the dentry
604 * tree upwards to the parents which might _also_ now be
605 * scheduled for deletion (it may have been only waiting for
606 * its last child to go away).
608 * This tail recursion is done by hand as we don't want to depend
609 * on the compiler to always get this right (gcc generally doesn't).
610 * Real recursion would eat up our stack space.
614 * dput - release a dentry
615 * @dentry: dentry to release
617 * Release a dentry. This will drop the usage count and if appropriate
618 * call the dentry unlink method as well as removing it from the queues and
619 * releasing its resources. If the parent dentries were scheduled for release
620 * they too may now get deleted.
622 void dput(struct dentry *dentry)
624 if (unlikely(!dentry))
628 if (lockref_put_or_lock(&dentry->d_lockref))
631 /* Unreachable? Get rid of it */
632 if (unlikely(d_unhashed(dentry)))
635 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
636 if (dentry->d_op->d_delete(dentry))
640 if (!(dentry->d_flags & DCACHE_REFERENCED))
641 dentry->d_flags |= DCACHE_REFERENCED;
642 dentry_lru_add(dentry);
644 dentry->d_lockref.count--;
645 spin_unlock(&dentry->d_lock);
649 dentry = dentry_kill(dentry);
656 /* This must be called with d_lock held */
657 static inline void __dget_dlock(struct dentry *dentry)
659 dentry->d_lockref.count++;
662 static inline void __dget(struct dentry *dentry)
664 lockref_get(&dentry->d_lockref);
667 struct dentry *dget_parent(struct dentry *dentry)
673 * Do optimistic parent lookup without any
677 ret = ACCESS_ONCE(dentry->d_parent);
678 gotref = lockref_get_not_zero(&ret->d_lockref);
680 if (likely(gotref)) {
681 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
688 * Don't need rcu_dereference because we re-check it was correct under
692 ret = dentry->d_parent;
693 spin_lock(&ret->d_lock);
694 if (unlikely(ret != dentry->d_parent)) {
695 spin_unlock(&ret->d_lock);
700 BUG_ON(!ret->d_lockref.count);
701 ret->d_lockref.count++;
702 spin_unlock(&ret->d_lock);
705 EXPORT_SYMBOL(dget_parent);
708 * d_find_alias - grab a hashed alias of inode
709 * @inode: inode in question
711 * If inode has a hashed alias, or is a directory and has any alias,
712 * acquire the reference to alias and return it. Otherwise return NULL.
713 * Notice that if inode is a directory there can be only one alias and
714 * it can be unhashed only if it has no children, or if it is the root
715 * of a filesystem, or if the directory was renamed and d_revalidate
716 * was the first vfs operation to notice.
718 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
719 * any other hashed alias over that one.
721 static struct dentry *__d_find_alias(struct inode *inode)
723 struct dentry *alias, *discon_alias;
727 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
728 spin_lock(&alias->d_lock);
729 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
730 if (IS_ROOT(alias) &&
731 (alias->d_flags & DCACHE_DISCONNECTED)) {
732 discon_alias = alias;
735 spin_unlock(&alias->d_lock);
739 spin_unlock(&alias->d_lock);
742 alias = discon_alias;
743 spin_lock(&alias->d_lock);
744 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
746 spin_unlock(&alias->d_lock);
749 spin_unlock(&alias->d_lock);
755 struct dentry *d_find_alias(struct inode *inode)
757 struct dentry *de = NULL;
759 if (!hlist_empty(&inode->i_dentry)) {
760 spin_lock(&inode->i_lock);
761 de = __d_find_alias(inode);
762 spin_unlock(&inode->i_lock);
766 EXPORT_SYMBOL(d_find_alias);
769 * Try to kill dentries associated with this inode.
770 * WARNING: you must own a reference to inode.
772 void d_prune_aliases(struct inode *inode)
774 struct dentry *dentry;
776 spin_lock(&inode->i_lock);
777 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
778 spin_lock(&dentry->d_lock);
779 if (!dentry->d_lockref.count) {
780 struct dentry *parent = lock_parent(dentry);
781 if (likely(!dentry->d_lockref.count)) {
782 __dentry_kill(dentry);
787 spin_unlock(&parent->d_lock);
789 spin_unlock(&dentry->d_lock);
791 spin_unlock(&inode->i_lock);
793 EXPORT_SYMBOL(d_prune_aliases);
795 static void shrink_dentry_list(struct list_head *list)
797 struct dentry *dentry, *parent;
799 while (!list_empty(list)) {
801 dentry = list_entry(list->prev, struct dentry, d_lru);
802 spin_lock(&dentry->d_lock);
803 parent = lock_parent(dentry);
806 * The dispose list is isolated and dentries are not accounted
807 * to the LRU here, so we can simply remove it from the list
808 * here regardless of whether it is referenced or not.
810 d_shrink_del(dentry);
813 * We found an inuse dentry which was not removed from
814 * the LRU because of laziness during lookup. Do not free it.
816 if ((int)dentry->d_lockref.count > 0) {
817 spin_unlock(&dentry->d_lock);
819 spin_unlock(&parent->d_lock);
824 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
825 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
826 spin_unlock(&dentry->d_lock);
828 spin_unlock(&parent->d_lock);
834 inode = dentry->d_inode;
835 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
836 d_shrink_add(dentry, list);
837 spin_unlock(&dentry->d_lock);
839 spin_unlock(&parent->d_lock);
843 __dentry_kill(dentry);
846 * We need to prune ancestors too. This is necessary to prevent
847 * quadratic behavior of shrink_dcache_parent(), but is also
848 * expected to be beneficial in reducing dentry cache
852 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
853 parent = lock_parent(dentry);
854 if (dentry->d_lockref.count != 1) {
855 dentry->d_lockref.count--;
856 spin_unlock(&dentry->d_lock);
858 spin_unlock(&parent->d_lock);
861 inode = dentry->d_inode; /* can't be NULL */
862 if (unlikely(!spin_trylock(&inode->i_lock))) {
863 spin_unlock(&dentry->d_lock);
865 spin_unlock(&parent->d_lock);
869 __dentry_kill(dentry);
875 static enum lru_status dentry_lru_isolate(struct list_head *item,
876 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
878 struct list_head *freeable = arg;
879 struct dentry *dentry = container_of(item, struct dentry, d_lru);
883 * we are inverting the lru lock/dentry->d_lock here,
884 * so use a trylock. If we fail to get the lock, just skip
887 if (!spin_trylock(&dentry->d_lock))
891 * Referenced dentries are still in use. If they have active
892 * counts, just remove them from the LRU. Otherwise give them
893 * another pass through the LRU.
895 if (dentry->d_lockref.count) {
896 d_lru_isolate(lru, dentry);
897 spin_unlock(&dentry->d_lock);
901 if (dentry->d_flags & DCACHE_REFERENCED) {
902 dentry->d_flags &= ~DCACHE_REFERENCED;
903 spin_unlock(&dentry->d_lock);
906 * The list move itself will be made by the common LRU code. At
907 * this point, we've dropped the dentry->d_lock but keep the
908 * lru lock. This is safe to do, since every list movement is
909 * protected by the lru lock even if both locks are held.
911 * This is guaranteed by the fact that all LRU management
912 * functions are intermediated by the LRU API calls like
913 * list_lru_add and list_lru_del. List movement in this file
914 * only ever occur through this functions or through callbacks
915 * like this one, that are called from the LRU API.
917 * The only exceptions to this are functions like
918 * shrink_dentry_list, and code that first checks for the
919 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
920 * operating only with stack provided lists after they are
921 * properly isolated from the main list. It is thus, always a
927 d_lru_shrink_move(lru, dentry, freeable);
928 spin_unlock(&dentry->d_lock);
934 * prune_dcache_sb - shrink the dcache
936 * @sc: shrink control, passed to list_lru_shrink_walk()
938 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
939 * is done when we need more memory and called from the superblock shrinker
942 * This function may fail to free any resources if all the dentries are in
945 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
950 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
951 dentry_lru_isolate, &dispose);
952 shrink_dentry_list(&dispose);
956 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
957 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
959 struct list_head *freeable = arg;
960 struct dentry *dentry = container_of(item, struct dentry, d_lru);
963 * we are inverting the lru lock/dentry->d_lock here,
964 * so use a trylock. If we fail to get the lock, just skip
967 if (!spin_trylock(&dentry->d_lock))
970 d_lru_shrink_move(lru, dentry, freeable);
971 spin_unlock(&dentry->d_lock);
978 * shrink_dcache_sb - shrink dcache for a superblock
981 * Shrink the dcache for the specified super block. This is used to free
982 * the dcache before unmounting a file system.
984 void shrink_dcache_sb(struct super_block *sb)
991 freed = list_lru_walk(&sb->s_dentry_lru,
992 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
994 this_cpu_sub(nr_dentry_unused, freed);
995 shrink_dentry_list(&dispose);
998 EXPORT_SYMBOL(shrink_dcache_sb);
1001 * enum d_walk_ret - action to talke during tree walk
1002 * @D_WALK_CONTINUE: contrinue walk
1003 * @D_WALK_QUIT: quit walk
1004 * @D_WALK_NORETRY: quit when retry is needed
1005 * @D_WALK_SKIP: skip this dentry and its children
1015 * d_walk - walk the dentry tree
1016 * @parent: start of walk
1017 * @data: data passed to @enter() and @finish()
1018 * @enter: callback when first entering the dentry
1019 * @finish: callback when successfully finished the walk
1021 * The @enter() and @finish() callbacks are called with d_lock held.
1023 static void d_walk(struct dentry *parent, void *data,
1024 enum d_walk_ret (*enter)(void *, struct dentry *),
1025 void (*finish)(void *))
1027 struct dentry *this_parent;
1028 struct list_head *next;
1030 enum d_walk_ret ret;
1034 read_seqbegin_or_lock(&rename_lock, &seq);
1035 this_parent = parent;
1036 spin_lock(&this_parent->d_lock);
1038 ret = enter(data, this_parent);
1040 case D_WALK_CONTINUE:
1045 case D_WALK_NORETRY:
1050 next = this_parent->d_subdirs.next;
1052 while (next != &this_parent->d_subdirs) {
1053 struct list_head *tmp = next;
1054 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1057 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1059 ret = enter(data, dentry);
1061 case D_WALK_CONTINUE:
1064 spin_unlock(&dentry->d_lock);
1066 case D_WALK_NORETRY:
1070 spin_unlock(&dentry->d_lock);
1074 if (!list_empty(&dentry->d_subdirs)) {
1075 spin_unlock(&this_parent->d_lock);
1076 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1077 this_parent = dentry;
1078 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1081 spin_unlock(&dentry->d_lock);
1084 * All done at this level ... ascend and resume the search.
1088 if (this_parent != parent) {
1089 struct dentry *child = this_parent;
1090 this_parent = child->d_parent;
1092 spin_unlock(&child->d_lock);
1093 spin_lock(&this_parent->d_lock);
1095 /* might go back up the wrong parent if we have had a rename. */
1096 if (need_seqretry(&rename_lock, seq))
1098 next = child->d_child.next;
1099 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1100 if (next == &this_parent->d_subdirs)
1102 child = list_entry(next, struct dentry, d_child);
1108 if (need_seqretry(&rename_lock, seq))
1115 spin_unlock(&this_parent->d_lock);
1116 done_seqretry(&rename_lock, seq);
1120 spin_unlock(&this_parent->d_lock);
1130 * Search for at least 1 mount point in the dentry's subdirs.
1131 * We descend to the next level whenever the d_subdirs
1132 * list is non-empty and continue searching.
1135 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1138 if (d_mountpoint(dentry)) {
1142 return D_WALK_CONTINUE;
1146 * have_submounts - check for mounts over a dentry
1147 * @parent: dentry to check.
1149 * Return true if the parent or its subdirectories contain
1152 int have_submounts(struct dentry *parent)
1156 d_walk(parent, &ret, check_mount, NULL);
1160 EXPORT_SYMBOL(have_submounts);
1163 * Called by mount code to set a mountpoint and check if the mountpoint is
1164 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1165 * subtree can become unreachable).
1167 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1168 * this reason take rename_lock and d_lock on dentry and ancestors.
1170 int d_set_mounted(struct dentry *dentry)
1174 write_seqlock(&rename_lock);
1175 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1176 /* Need exclusion wrt. d_invalidate() */
1177 spin_lock(&p->d_lock);
1178 if (unlikely(d_unhashed(p))) {
1179 spin_unlock(&p->d_lock);
1182 spin_unlock(&p->d_lock);
1184 spin_lock(&dentry->d_lock);
1185 if (!d_unlinked(dentry)) {
1186 dentry->d_flags |= DCACHE_MOUNTED;
1189 spin_unlock(&dentry->d_lock);
1191 write_sequnlock(&rename_lock);
1196 * Search the dentry child list of the specified parent,
1197 * and move any unused dentries to the end of the unused
1198 * list for prune_dcache(). We descend to the next level
1199 * whenever the d_subdirs list is non-empty and continue
1202 * It returns zero iff there are no unused children,
1203 * otherwise it returns the number of children moved to
1204 * the end of the unused list. This may not be the total
1205 * number of unused children, because select_parent can
1206 * drop the lock and return early due to latency
1210 struct select_data {
1211 struct dentry *start;
1212 struct list_head dispose;
1216 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1218 struct select_data *data = _data;
1219 enum d_walk_ret ret = D_WALK_CONTINUE;
1221 if (data->start == dentry)
1224 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1227 if (dentry->d_flags & DCACHE_LRU_LIST)
1229 if (!dentry->d_lockref.count) {
1230 d_shrink_add(dentry, &data->dispose);
1235 * We can return to the caller if we have found some (this
1236 * ensures forward progress). We'll be coming back to find
1239 if (!list_empty(&data->dispose))
1240 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1246 * shrink_dcache_parent - prune dcache
1247 * @parent: parent of entries to prune
1249 * Prune the dcache to remove unused children of the parent dentry.
1251 void shrink_dcache_parent(struct dentry *parent)
1254 struct select_data data;
1256 INIT_LIST_HEAD(&data.dispose);
1257 data.start = parent;
1260 d_walk(parent, &data, select_collect, NULL);
1264 shrink_dentry_list(&data.dispose);
1268 EXPORT_SYMBOL(shrink_dcache_parent);
1270 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1272 /* it has busy descendents; complain about those instead */
1273 if (!list_empty(&dentry->d_subdirs))
1274 return D_WALK_CONTINUE;
1276 /* root with refcount 1 is fine */
1277 if (dentry == _data && dentry->d_lockref.count == 1)
1278 return D_WALK_CONTINUE;
1280 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1281 " still in use (%d) [unmount of %s %s]\n",
1284 dentry->d_inode->i_ino : 0UL,
1286 dentry->d_lockref.count,
1287 dentry->d_sb->s_type->name,
1288 dentry->d_sb->s_id);
1290 return D_WALK_CONTINUE;
1293 static void do_one_tree(struct dentry *dentry)
1295 shrink_dcache_parent(dentry);
1296 d_walk(dentry, dentry, umount_check, NULL);
1302 * destroy the dentries attached to a superblock on unmounting
1304 void shrink_dcache_for_umount(struct super_block *sb)
1306 struct dentry *dentry;
1308 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1310 dentry = sb->s_root;
1312 do_one_tree(dentry);
1314 while (!hlist_bl_empty(&sb->s_anon)) {
1315 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1316 do_one_tree(dentry);
1320 struct detach_data {
1321 struct select_data select;
1322 struct dentry *mountpoint;
1324 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1326 struct detach_data *data = _data;
1328 if (d_mountpoint(dentry)) {
1329 __dget_dlock(dentry);
1330 data->mountpoint = dentry;
1334 return select_collect(&data->select, dentry);
1337 static void check_and_drop(void *_data)
1339 struct detach_data *data = _data;
1341 if (!data->mountpoint && !data->select.found)
1342 __d_drop(data->select.start);
1346 * d_invalidate - detach submounts, prune dcache, and drop
1347 * @dentry: dentry to invalidate (aka detach, prune and drop)
1351 * The final d_drop is done as an atomic operation relative to
1352 * rename_lock ensuring there are no races with d_set_mounted. This
1353 * ensures there are no unhashed dentries on the path to a mountpoint.
1355 void d_invalidate(struct dentry *dentry)
1358 * If it's already been dropped, return OK.
1360 spin_lock(&dentry->d_lock);
1361 if (d_unhashed(dentry)) {
1362 spin_unlock(&dentry->d_lock);
1365 spin_unlock(&dentry->d_lock);
1367 /* Negative dentries can be dropped without further checks */
1368 if (!dentry->d_inode) {
1374 struct detach_data data;
1376 data.mountpoint = NULL;
1377 INIT_LIST_HEAD(&data.select.dispose);
1378 data.select.start = dentry;
1379 data.select.found = 0;
1381 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1383 if (data.select.found)
1384 shrink_dentry_list(&data.select.dispose);
1386 if (data.mountpoint) {
1387 detach_mounts(data.mountpoint);
1388 dput(data.mountpoint);
1391 if (!data.mountpoint && !data.select.found)
1397 EXPORT_SYMBOL(d_invalidate);
1400 * __d_alloc - allocate a dcache entry
1401 * @sb: filesystem it will belong to
1402 * @name: qstr of the name
1404 * Allocates a dentry. It returns %NULL if there is insufficient memory
1405 * available. On a success the dentry is returned. The name passed in is
1406 * copied and the copy passed in may be reused after this call.
1409 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1411 struct dentry *dentry;
1414 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1419 * We guarantee that the inline name is always NUL-terminated.
1420 * This way the memcpy() done by the name switching in rename
1421 * will still always have a NUL at the end, even if we might
1422 * be overwriting an internal NUL character
1424 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1425 if (name->len > DNAME_INLINE_LEN-1) {
1426 size_t size = offsetof(struct external_name, name[1]);
1427 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1429 kmem_cache_free(dentry_cache, dentry);
1432 atomic_set(&p->u.count, 1);
1434 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1435 kasan_unpoison_shadow(dname,
1436 round_up(name->len + 1, sizeof(unsigned long)));
1438 dname = dentry->d_iname;
1441 dentry->d_name.len = name->len;
1442 dentry->d_name.hash = name->hash;
1443 memcpy(dname, name->name, name->len);
1444 dname[name->len] = 0;
1446 /* Make sure we always see the terminating NUL character */
1448 dentry->d_name.name = dname;
1450 dentry->d_lockref.count = 1;
1451 dentry->d_flags = 0;
1452 spin_lock_init(&dentry->d_lock);
1453 seqcount_init(&dentry->d_seq);
1454 dentry->d_inode = NULL;
1455 dentry->d_parent = dentry;
1457 dentry->d_op = NULL;
1458 dentry->d_fsdata = NULL;
1459 INIT_HLIST_BL_NODE(&dentry->d_hash);
1460 INIT_LIST_HEAD(&dentry->d_lru);
1461 INIT_LIST_HEAD(&dentry->d_subdirs);
1462 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1463 INIT_LIST_HEAD(&dentry->d_child);
1464 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1466 this_cpu_inc(nr_dentry);
1472 * d_alloc - allocate a dcache entry
1473 * @parent: parent of entry to allocate
1474 * @name: qstr of the name
1476 * Allocates a dentry. It returns %NULL if there is insufficient memory
1477 * available. On a success the dentry is returned. The name passed in is
1478 * copied and the copy passed in may be reused after this call.
1480 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1482 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1486 spin_lock(&parent->d_lock);
1488 * don't need child lock because it is not subject
1489 * to concurrency here
1491 __dget_dlock(parent);
1492 dentry->d_parent = parent;
1493 list_add(&dentry->d_child, &parent->d_subdirs);
1494 spin_unlock(&parent->d_lock);
1498 EXPORT_SYMBOL(d_alloc);
1501 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1502 * @sb: the superblock
1503 * @name: qstr of the name
1505 * For a filesystem that just pins its dentries in memory and never
1506 * performs lookups at all, return an unhashed IS_ROOT dentry.
1508 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1510 return __d_alloc(sb, name);
1512 EXPORT_SYMBOL(d_alloc_pseudo);
1514 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1519 q.len = strlen(name);
1520 q.hash = full_name_hash(q.name, q.len);
1521 return d_alloc(parent, &q);
1523 EXPORT_SYMBOL(d_alloc_name);
1525 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1527 WARN_ON_ONCE(dentry->d_op);
1528 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1530 DCACHE_OP_REVALIDATE |
1531 DCACHE_OP_WEAK_REVALIDATE |
1532 DCACHE_OP_DELETE ));
1537 dentry->d_flags |= DCACHE_OP_HASH;
1539 dentry->d_flags |= DCACHE_OP_COMPARE;
1540 if (op->d_revalidate)
1541 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1542 if (op->d_weak_revalidate)
1543 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1545 dentry->d_flags |= DCACHE_OP_DELETE;
1547 dentry->d_flags |= DCACHE_OP_PRUNE;
1550 EXPORT_SYMBOL(d_set_d_op);
1552 static unsigned d_flags_for_inode(struct inode *inode)
1554 unsigned add_flags = DCACHE_FILE_TYPE;
1557 return DCACHE_MISS_TYPE;
1559 if (S_ISDIR(inode->i_mode)) {
1560 add_flags = DCACHE_DIRECTORY_TYPE;
1561 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1562 if (unlikely(!inode->i_op->lookup))
1563 add_flags = DCACHE_AUTODIR_TYPE;
1565 inode->i_opflags |= IOP_LOOKUP;
1567 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1568 if (unlikely(inode->i_op->follow_link))
1569 add_flags = DCACHE_SYMLINK_TYPE;
1571 inode->i_opflags |= IOP_NOFOLLOW;
1574 if (unlikely(IS_AUTOMOUNT(inode)))
1575 add_flags |= DCACHE_NEED_AUTOMOUNT;
1579 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1581 unsigned add_flags = d_flags_for_inode(inode);
1583 spin_lock(&dentry->d_lock);
1584 __d_set_type(dentry, add_flags);
1586 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1587 dentry->d_inode = inode;
1588 dentry_rcuwalk_barrier(dentry);
1589 spin_unlock(&dentry->d_lock);
1590 fsnotify_d_instantiate(dentry, inode);
1594 * d_instantiate - fill in inode information for a dentry
1595 * @entry: dentry to complete
1596 * @inode: inode to attach to this dentry
1598 * Fill in inode information in the entry.
1600 * This turns negative dentries into productive full members
1603 * NOTE! This assumes that the inode count has been incremented
1604 * (or otherwise set) by the caller to indicate that it is now
1605 * in use by the dcache.
1608 void d_instantiate(struct dentry *entry, struct inode * inode)
1610 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1612 spin_lock(&inode->i_lock);
1613 __d_instantiate(entry, inode);
1615 spin_unlock(&inode->i_lock);
1616 security_d_instantiate(entry, inode);
1618 EXPORT_SYMBOL(d_instantiate);
1621 * d_instantiate_unique - instantiate a non-aliased dentry
1622 * @entry: dentry to instantiate
1623 * @inode: inode to attach to this dentry
1625 * Fill in inode information in the entry. On success, it returns NULL.
1626 * If an unhashed alias of "entry" already exists, then we return the
1627 * aliased dentry instead and drop one reference to inode.
1629 * Note that in order to avoid conflicts with rename() etc, the caller
1630 * had better be holding the parent directory semaphore.
1632 * This also assumes that the inode count has been incremented
1633 * (or otherwise set) by the caller to indicate that it is now
1634 * in use by the dcache.
1636 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1637 struct inode *inode)
1639 struct dentry *alias;
1640 int len = entry->d_name.len;
1641 const char *name = entry->d_name.name;
1642 unsigned int hash = entry->d_name.hash;
1645 __d_instantiate(entry, NULL);
1649 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1651 * Don't need alias->d_lock here, because aliases with
1652 * d_parent == entry->d_parent are not subject to name or
1653 * parent changes, because the parent inode i_mutex is held.
1655 if (alias->d_name.hash != hash)
1657 if (alias->d_parent != entry->d_parent)
1659 if (alias->d_name.len != len)
1661 if (dentry_cmp(alias, name, len))
1667 __d_instantiate(entry, inode);
1671 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1673 struct dentry *result;
1675 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1678 spin_lock(&inode->i_lock);
1679 result = __d_instantiate_unique(entry, inode);
1681 spin_unlock(&inode->i_lock);
1684 security_d_instantiate(entry, inode);
1688 BUG_ON(!d_unhashed(result));
1693 EXPORT_SYMBOL(d_instantiate_unique);
1696 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1697 * @entry: dentry to complete
1698 * @inode: inode to attach to this dentry
1700 * Fill in inode information in the entry. If a directory alias is found, then
1701 * return an error (and drop inode). Together with d_materialise_unique() this
1702 * guarantees that a directory inode may never have more than one alias.
1704 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1706 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1708 spin_lock(&inode->i_lock);
1709 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1710 spin_unlock(&inode->i_lock);
1714 __d_instantiate(entry, inode);
1715 spin_unlock(&inode->i_lock);
1716 security_d_instantiate(entry, inode);
1720 EXPORT_SYMBOL(d_instantiate_no_diralias);
1722 struct dentry *d_make_root(struct inode *root_inode)
1724 struct dentry *res = NULL;
1727 static const struct qstr name = QSTR_INIT("/", 1);
1729 res = __d_alloc(root_inode->i_sb, &name);
1731 d_instantiate(res, root_inode);
1737 EXPORT_SYMBOL(d_make_root);
1739 static struct dentry * __d_find_any_alias(struct inode *inode)
1741 struct dentry *alias;
1743 if (hlist_empty(&inode->i_dentry))
1745 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1751 * d_find_any_alias - find any alias for a given inode
1752 * @inode: inode to find an alias for
1754 * If any aliases exist for the given inode, take and return a
1755 * reference for one of them. If no aliases exist, return %NULL.
1757 struct dentry *d_find_any_alias(struct inode *inode)
1761 spin_lock(&inode->i_lock);
1762 de = __d_find_any_alias(inode);
1763 spin_unlock(&inode->i_lock);
1766 EXPORT_SYMBOL(d_find_any_alias);
1768 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1770 static const struct qstr anonstring = QSTR_INIT("/", 1);
1776 return ERR_PTR(-ESTALE);
1778 return ERR_CAST(inode);
1780 res = d_find_any_alias(inode);
1784 tmp = __d_alloc(inode->i_sb, &anonstring);
1786 res = ERR_PTR(-ENOMEM);
1790 spin_lock(&inode->i_lock);
1791 res = __d_find_any_alias(inode);
1793 spin_unlock(&inode->i_lock);
1798 /* attach a disconnected dentry */
1799 add_flags = d_flags_for_inode(inode);
1802 add_flags |= DCACHE_DISCONNECTED;
1804 spin_lock(&tmp->d_lock);
1805 tmp->d_inode = inode;
1806 tmp->d_flags |= add_flags;
1807 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1808 hlist_bl_lock(&tmp->d_sb->s_anon);
1809 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1810 hlist_bl_unlock(&tmp->d_sb->s_anon);
1811 spin_unlock(&tmp->d_lock);
1812 spin_unlock(&inode->i_lock);
1813 security_d_instantiate(tmp, inode);
1818 if (res && !IS_ERR(res))
1819 security_d_instantiate(res, inode);
1825 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1826 * @inode: inode to allocate the dentry for
1828 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1829 * similar open by handle operations. The returned dentry may be anonymous,
1830 * or may have a full name (if the inode was already in the cache).
1832 * When called on a directory inode, we must ensure that the inode only ever
1833 * has one dentry. If a dentry is found, that is returned instead of
1834 * allocating a new one.
1836 * On successful return, the reference to the inode has been transferred
1837 * to the dentry. In case of an error the reference on the inode is released.
1838 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1839 * be passed in and the error will be propagated to the return value,
1840 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1842 struct dentry *d_obtain_alias(struct inode *inode)
1844 return __d_obtain_alias(inode, 1);
1846 EXPORT_SYMBOL(d_obtain_alias);
1849 * d_obtain_root - find or allocate a dentry for a given inode
1850 * @inode: inode to allocate the dentry for
1852 * Obtain an IS_ROOT dentry for the root of a filesystem.
1854 * We must ensure that directory inodes only ever have one dentry. If a
1855 * dentry is found, that is returned instead of allocating a new one.
1857 * On successful return, the reference to the inode has been transferred
1858 * to the dentry. In case of an error the reference on the inode is
1859 * released. A %NULL or IS_ERR inode may be passed in and will be the
1860 * error will be propagate to the return value, with a %NULL @inode
1861 * replaced by ERR_PTR(-ESTALE).
1863 struct dentry *d_obtain_root(struct inode *inode)
1865 return __d_obtain_alias(inode, 0);
1867 EXPORT_SYMBOL(d_obtain_root);
1870 * d_add_ci - lookup or allocate new dentry with case-exact name
1871 * @inode: the inode case-insensitive lookup has found
1872 * @dentry: the negative dentry that was passed to the parent's lookup func
1873 * @name: the case-exact name to be associated with the returned dentry
1875 * This is to avoid filling the dcache with case-insensitive names to the
1876 * same inode, only the actual correct case is stored in the dcache for
1877 * case-insensitive filesystems.
1879 * For a case-insensitive lookup match and if the the case-exact dentry
1880 * already exists in in the dcache, use it and return it.
1882 * If no entry exists with the exact case name, allocate new dentry with
1883 * the exact case, and return the spliced entry.
1885 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1888 struct dentry *found;
1892 * First check if a dentry matching the name already exists,
1893 * if not go ahead and create it now.
1895 found = d_hash_and_lookup(dentry->d_parent, name);
1897 new = d_alloc(dentry->d_parent, name);
1899 found = ERR_PTR(-ENOMEM);
1901 found = d_splice_alias(inode, new);
1912 EXPORT_SYMBOL(d_add_ci);
1915 * Do the slow-case of the dentry name compare.
1917 * Unlike the dentry_cmp() function, we need to atomically
1918 * load the name and length information, so that the
1919 * filesystem can rely on them, and can use the 'name' and
1920 * 'len' information without worrying about walking off the
1921 * end of memory etc.
1923 * Thus the read_seqcount_retry() and the "duplicate" info
1924 * in arguments (the low-level filesystem should not look
1925 * at the dentry inode or name contents directly, since
1926 * rename can change them while we're in RCU mode).
1928 enum slow_d_compare {
1934 static noinline enum slow_d_compare slow_dentry_cmp(
1935 const struct dentry *parent,
1936 struct dentry *dentry,
1938 const struct qstr *name)
1940 int tlen = dentry->d_name.len;
1941 const char *tname = dentry->d_name.name;
1943 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1945 return D_COMP_SEQRETRY;
1947 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1948 return D_COMP_NOMATCH;
1953 * __d_lookup_rcu - search for a dentry (racy, store-free)
1954 * @parent: parent dentry
1955 * @name: qstr of name we wish to find
1956 * @seqp: returns d_seq value at the point where the dentry was found
1957 * Returns: dentry, or NULL
1959 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1960 * resolution (store-free path walking) design described in
1961 * Documentation/filesystems/path-lookup.txt.
1963 * This is not to be used outside core vfs.
1965 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1966 * held, and rcu_read_lock held. The returned dentry must not be stored into
1967 * without taking d_lock and checking d_seq sequence count against @seq
1970 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1973 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1974 * the returned dentry, so long as its parent's seqlock is checked after the
1975 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1976 * is formed, giving integrity down the path walk.
1978 * NOTE! The caller *has* to check the resulting dentry against the sequence
1979 * number we've returned before using any of the resulting dentry state!
1981 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1982 const struct qstr *name,
1985 u64 hashlen = name->hash_len;
1986 const unsigned char *str = name->name;
1987 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1988 struct hlist_bl_node *node;
1989 struct dentry *dentry;
1992 * Note: There is significant duplication with __d_lookup_rcu which is
1993 * required to prevent single threaded performance regressions
1994 * especially on architectures where smp_rmb (in seqcounts) are costly.
1995 * Keep the two functions in sync.
1999 * The hash list is protected using RCU.
2001 * Carefully use d_seq when comparing a candidate dentry, to avoid
2002 * races with d_move().
2004 * It is possible that concurrent renames can mess up our list
2005 * walk here and result in missing our dentry, resulting in the
2006 * false-negative result. d_lookup() protects against concurrent
2007 * renames using rename_lock seqlock.
2009 * See Documentation/filesystems/path-lookup.txt for more details.
2011 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2016 * The dentry sequence count protects us from concurrent
2017 * renames, and thus protects parent and name fields.
2019 * The caller must perform a seqcount check in order
2020 * to do anything useful with the returned dentry.
2022 * NOTE! We do a "raw" seqcount_begin here. That means that
2023 * we don't wait for the sequence count to stabilize if it
2024 * is in the middle of a sequence change. If we do the slow
2025 * dentry compare, we will do seqretries until it is stable,
2026 * and if we end up with a successful lookup, we actually
2027 * want to exit RCU lookup anyway.
2029 seq = raw_seqcount_begin(&dentry->d_seq);
2030 if (dentry->d_parent != parent)
2032 if (d_unhashed(dentry))
2035 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2036 if (dentry->d_name.hash != hashlen_hash(hashlen))
2039 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2042 case D_COMP_NOMATCH:
2049 if (dentry->d_name.hash_len != hashlen)
2052 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2059 * d_lookup - search for a dentry
2060 * @parent: parent dentry
2061 * @name: qstr of name we wish to find
2062 * Returns: dentry, or NULL
2064 * d_lookup searches the children of the parent dentry for the name in
2065 * question. If the dentry is found its reference count is incremented and the
2066 * dentry is returned. The caller must use dput to free the entry when it has
2067 * finished using it. %NULL is returned if the dentry does not exist.
2069 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2071 struct dentry *dentry;
2075 seq = read_seqbegin(&rename_lock);
2076 dentry = __d_lookup(parent, name);
2079 } while (read_seqretry(&rename_lock, seq));
2082 EXPORT_SYMBOL(d_lookup);
2085 * __d_lookup - search for a dentry (racy)
2086 * @parent: parent dentry
2087 * @name: qstr of name we wish to find
2088 * Returns: dentry, or NULL
2090 * __d_lookup is like d_lookup, however it may (rarely) return a
2091 * false-negative result due to unrelated rename activity.
2093 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2094 * however it must be used carefully, eg. with a following d_lookup in
2095 * the case of failure.
2097 * __d_lookup callers must be commented.
2099 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2101 unsigned int len = name->len;
2102 unsigned int hash = name->hash;
2103 const unsigned char *str = name->name;
2104 struct hlist_bl_head *b = d_hash(parent, hash);
2105 struct hlist_bl_node *node;
2106 struct dentry *found = NULL;
2107 struct dentry *dentry;
2110 * Note: There is significant duplication with __d_lookup_rcu which is
2111 * required to prevent single threaded performance regressions
2112 * especially on architectures where smp_rmb (in seqcounts) are costly.
2113 * Keep the two functions in sync.
2117 * The hash list is protected using RCU.
2119 * Take d_lock when comparing a candidate dentry, to avoid races
2122 * It is possible that concurrent renames can mess up our list
2123 * walk here and result in missing our dentry, resulting in the
2124 * false-negative result. d_lookup() protects against concurrent
2125 * renames using rename_lock seqlock.
2127 * See Documentation/filesystems/path-lookup.txt for more details.
2131 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2133 if (dentry->d_name.hash != hash)
2136 spin_lock(&dentry->d_lock);
2137 if (dentry->d_parent != parent)
2139 if (d_unhashed(dentry))
2143 * It is safe to compare names since d_move() cannot
2144 * change the qstr (protected by d_lock).
2146 if (parent->d_flags & DCACHE_OP_COMPARE) {
2147 int tlen = dentry->d_name.len;
2148 const char *tname = dentry->d_name.name;
2149 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2152 if (dentry->d_name.len != len)
2154 if (dentry_cmp(dentry, str, len))
2158 dentry->d_lockref.count++;
2160 spin_unlock(&dentry->d_lock);
2163 spin_unlock(&dentry->d_lock);
2171 * d_hash_and_lookup - hash the qstr then search for a dentry
2172 * @dir: Directory to search in
2173 * @name: qstr of name we wish to find
2175 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2177 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2180 * Check for a fs-specific hash function. Note that we must
2181 * calculate the standard hash first, as the d_op->d_hash()
2182 * routine may choose to leave the hash value unchanged.
2184 name->hash = full_name_hash(name->name, name->len);
2185 if (dir->d_flags & DCACHE_OP_HASH) {
2186 int err = dir->d_op->d_hash(dir, name);
2187 if (unlikely(err < 0))
2188 return ERR_PTR(err);
2190 return d_lookup(dir, name);
2192 EXPORT_SYMBOL(d_hash_and_lookup);
2195 * d_validate - verify dentry provided from insecure source (deprecated)
2196 * @dentry: The dentry alleged to be valid child of @dparent
2197 * @dparent: The parent dentry (known to be valid)
2199 * An insecure source has sent us a dentry, here we verify it and dget() it.
2200 * This is used by ncpfs in its readdir implementation.
2201 * Zero is returned in the dentry is invalid.
2203 * This function is slow for big directories, and deprecated, do not use it.
2205 int d_validate(struct dentry *dentry, struct dentry *dparent)
2207 struct dentry *child;
2209 spin_lock(&dparent->d_lock);
2210 list_for_each_entry(child, &dparent->d_subdirs, d_child) {
2211 if (dentry == child) {
2212 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2213 __dget_dlock(dentry);
2214 spin_unlock(&dentry->d_lock);
2215 spin_unlock(&dparent->d_lock);
2219 spin_unlock(&dparent->d_lock);
2223 EXPORT_SYMBOL(d_validate);
2226 * When a file is deleted, we have two options:
2227 * - turn this dentry into a negative dentry
2228 * - unhash this dentry and free it.
2230 * Usually, we want to just turn this into
2231 * a negative dentry, but if anybody else is
2232 * currently using the dentry or the inode
2233 * we can't do that and we fall back on removing
2234 * it from the hash queues and waiting for
2235 * it to be deleted later when it has no users
2239 * d_delete - delete a dentry
2240 * @dentry: The dentry to delete
2242 * Turn the dentry into a negative dentry if possible, otherwise
2243 * remove it from the hash queues so it can be deleted later
2246 void d_delete(struct dentry * dentry)
2248 struct inode *inode;
2251 * Are we the only user?
2254 spin_lock(&dentry->d_lock);
2255 inode = dentry->d_inode;
2256 isdir = S_ISDIR(inode->i_mode);
2257 if (dentry->d_lockref.count == 1) {
2258 if (!spin_trylock(&inode->i_lock)) {
2259 spin_unlock(&dentry->d_lock);
2263 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2264 dentry_unlink_inode(dentry);
2265 fsnotify_nameremove(dentry, isdir);
2269 if (!d_unhashed(dentry))
2272 spin_unlock(&dentry->d_lock);
2274 fsnotify_nameremove(dentry, isdir);
2276 EXPORT_SYMBOL(d_delete);
2278 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2280 BUG_ON(!d_unhashed(entry));
2282 entry->d_flags |= DCACHE_RCUACCESS;
2283 hlist_bl_add_head_rcu(&entry->d_hash, b);
2287 static void _d_rehash(struct dentry * entry)
2289 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2293 * d_rehash - add an entry back to the hash
2294 * @entry: dentry to add to the hash
2296 * Adds a dentry to the hash according to its name.
2299 void d_rehash(struct dentry * entry)
2301 spin_lock(&entry->d_lock);
2303 spin_unlock(&entry->d_lock);
2305 EXPORT_SYMBOL(d_rehash);
2308 * dentry_update_name_case - update case insensitive dentry with a new name
2309 * @dentry: dentry to be updated
2312 * Update a case insensitive dentry with new case of name.
2314 * dentry must have been returned by d_lookup with name @name. Old and new
2315 * name lengths must match (ie. no d_compare which allows mismatched name
2318 * Parent inode i_mutex must be held over d_lookup and into this call (to
2319 * keep renames and concurrent inserts, and readdir(2) away).
2321 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2323 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2324 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2326 spin_lock(&dentry->d_lock);
2327 write_seqcount_begin(&dentry->d_seq);
2328 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2329 write_seqcount_end(&dentry->d_seq);
2330 spin_unlock(&dentry->d_lock);
2332 EXPORT_SYMBOL(dentry_update_name_case);
2334 static void swap_names(struct dentry *dentry, struct dentry *target)
2336 if (unlikely(dname_external(target))) {
2337 if (unlikely(dname_external(dentry))) {
2339 * Both external: swap the pointers
2341 swap(target->d_name.name, dentry->d_name.name);
2344 * dentry:internal, target:external. Steal target's
2345 * storage and make target internal.
2347 memcpy(target->d_iname, dentry->d_name.name,
2348 dentry->d_name.len + 1);
2349 dentry->d_name.name = target->d_name.name;
2350 target->d_name.name = target->d_iname;
2353 if (unlikely(dname_external(dentry))) {
2355 * dentry:external, target:internal. Give dentry's
2356 * storage to target and make dentry internal
2358 memcpy(dentry->d_iname, target->d_name.name,
2359 target->d_name.len + 1);
2360 target->d_name.name = dentry->d_name.name;
2361 dentry->d_name.name = dentry->d_iname;
2364 * Both are internal.
2367 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2368 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2369 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2370 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2371 swap(((long *) &dentry->d_iname)[i],
2372 ((long *) &target->d_iname)[i]);
2376 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2379 static void copy_name(struct dentry *dentry, struct dentry *target)
2381 struct external_name *old_name = NULL;
2382 if (unlikely(dname_external(dentry)))
2383 old_name = external_name(dentry);
2384 if (unlikely(dname_external(target))) {
2385 atomic_inc(&external_name(target)->u.count);
2386 dentry->d_name = target->d_name;
2388 memcpy(dentry->d_iname, target->d_name.name,
2389 target->d_name.len + 1);
2390 dentry->d_name.name = dentry->d_iname;
2391 dentry->d_name.hash_len = target->d_name.hash_len;
2393 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2394 kfree_rcu(old_name, u.head);
2397 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2400 * XXXX: do we really need to take target->d_lock?
2402 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2403 spin_lock(&target->d_parent->d_lock);
2405 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2406 spin_lock(&dentry->d_parent->d_lock);
2407 spin_lock_nested(&target->d_parent->d_lock,
2408 DENTRY_D_LOCK_NESTED);
2410 spin_lock(&target->d_parent->d_lock);
2411 spin_lock_nested(&dentry->d_parent->d_lock,
2412 DENTRY_D_LOCK_NESTED);
2415 if (target < dentry) {
2416 spin_lock_nested(&target->d_lock, 2);
2417 spin_lock_nested(&dentry->d_lock, 3);
2419 spin_lock_nested(&dentry->d_lock, 2);
2420 spin_lock_nested(&target->d_lock, 3);
2424 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2426 if (target->d_parent != dentry->d_parent)
2427 spin_unlock(&dentry->d_parent->d_lock);
2428 if (target->d_parent != target)
2429 spin_unlock(&target->d_parent->d_lock);
2430 spin_unlock(&target->d_lock);
2431 spin_unlock(&dentry->d_lock);
2435 * When switching names, the actual string doesn't strictly have to
2436 * be preserved in the target - because we're dropping the target
2437 * anyway. As such, we can just do a simple memcpy() to copy over
2438 * the new name before we switch, unless we are going to rehash
2439 * it. Note that if we *do* unhash the target, we are not allowed
2440 * to rehash it without giving it a new name/hash key - whether
2441 * we swap or overwrite the names here, resulting name won't match
2442 * the reality in filesystem; it's only there for d_path() purposes.
2443 * Note that all of this is happening under rename_lock, so the
2444 * any hash lookup seeing it in the middle of manipulations will
2445 * be discarded anyway. So we do not care what happens to the hash
2449 * __d_move - move a dentry
2450 * @dentry: entry to move
2451 * @target: new dentry
2452 * @exchange: exchange the two dentries
2454 * Update the dcache to reflect the move of a file name. Negative
2455 * dcache entries should not be moved in this way. Caller must hold
2456 * rename_lock, the i_mutex of the source and target directories,
2457 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2459 static void __d_move(struct dentry *dentry, struct dentry *target,
2462 if (!dentry->d_inode)
2463 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2465 BUG_ON(d_ancestor(dentry, target));
2466 BUG_ON(d_ancestor(target, dentry));
2468 dentry_lock_for_move(dentry, target);
2470 write_seqcount_begin(&dentry->d_seq);
2471 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2473 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2476 * Move the dentry to the target hash queue. Don't bother checking
2477 * for the same hash queue because of how unlikely it is.
2480 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2483 * Unhash the target (d_delete() is not usable here). If exchanging
2484 * the two dentries, then rehash onto the other's hash queue.
2489 d_hash(dentry->d_parent, dentry->d_name.hash));
2492 /* Switch the names.. */
2494 swap_names(dentry, target);
2496 copy_name(dentry, target);
2498 /* ... and switch them in the tree */
2499 if (IS_ROOT(dentry)) {
2500 /* splicing a tree */
2501 dentry->d_parent = target->d_parent;
2502 target->d_parent = target;
2503 list_del_init(&target->d_child);
2504 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2506 /* swapping two dentries */
2507 swap(dentry->d_parent, target->d_parent);
2508 list_move(&target->d_child, &target->d_parent->d_subdirs);
2509 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2511 fsnotify_d_move(target);
2512 fsnotify_d_move(dentry);
2515 write_seqcount_end(&target->d_seq);
2516 write_seqcount_end(&dentry->d_seq);
2518 dentry_unlock_for_move(dentry, target);
2522 * d_move - move a dentry
2523 * @dentry: entry to move
2524 * @target: new dentry
2526 * Update the dcache to reflect the move of a file name. Negative
2527 * dcache entries should not be moved in this way. See the locking
2528 * requirements for __d_move.
2530 void d_move(struct dentry *dentry, struct dentry *target)
2532 write_seqlock(&rename_lock);
2533 __d_move(dentry, target, false);
2534 write_sequnlock(&rename_lock);
2536 EXPORT_SYMBOL(d_move);
2539 * d_exchange - exchange two dentries
2540 * @dentry1: first dentry
2541 * @dentry2: second dentry
2543 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2545 write_seqlock(&rename_lock);
2547 WARN_ON(!dentry1->d_inode);
2548 WARN_ON(!dentry2->d_inode);
2549 WARN_ON(IS_ROOT(dentry1));
2550 WARN_ON(IS_ROOT(dentry2));
2552 __d_move(dentry1, dentry2, true);
2554 write_sequnlock(&rename_lock);
2558 * d_ancestor - search for an ancestor
2559 * @p1: ancestor dentry
2562 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2563 * an ancestor of p2, else NULL.
2565 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2569 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2570 if (p->d_parent == p1)
2577 * This helper attempts to cope with remotely renamed directories
2579 * It assumes that the caller is already holding
2580 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2582 * Note: If ever the locking in lock_rename() changes, then please
2583 * remember to update this too...
2585 static int __d_unalias(struct inode *inode,
2586 struct dentry *dentry, struct dentry *alias)
2588 struct mutex *m1 = NULL, *m2 = NULL;
2591 /* If alias and dentry share a parent, then no extra locks required */
2592 if (alias->d_parent == dentry->d_parent)
2595 /* See lock_rename() */
2596 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2598 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2599 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2601 m2 = &alias->d_parent->d_inode->i_mutex;
2603 __d_move(alias, dentry, false);
2606 spin_unlock(&inode->i_lock);
2615 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2616 * @inode: the inode which may have a disconnected dentry
2617 * @dentry: a negative dentry which we want to point to the inode.
2619 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2620 * place of the given dentry and return it, else simply d_add the inode
2621 * to the dentry and return NULL.
2623 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2624 * we should error out: directories can't have multiple aliases.
2626 * This is needed in the lookup routine of any filesystem that is exportable
2627 * (via knfsd) so that we can build dcache paths to directories effectively.
2629 * If a dentry was found and moved, then it is returned. Otherwise NULL
2630 * is returned. This matches the expected return value of ->lookup.
2632 * Cluster filesystems may call this function with a negative, hashed dentry.
2633 * In that case, we know that the inode will be a regular file, and also this
2634 * will only occur during atomic_open. So we need to check for the dentry
2635 * being already hashed only in the final case.
2637 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2640 return ERR_CAST(inode);
2642 BUG_ON(!d_unhashed(dentry));
2645 __d_instantiate(dentry, NULL);
2648 spin_lock(&inode->i_lock);
2649 if (S_ISDIR(inode->i_mode)) {
2650 struct dentry *new = __d_find_any_alias(inode);
2651 if (unlikely(new)) {
2652 write_seqlock(&rename_lock);
2653 if (unlikely(d_ancestor(new, dentry))) {
2654 write_sequnlock(&rename_lock);
2655 spin_unlock(&inode->i_lock);
2657 new = ERR_PTR(-ELOOP);
2658 pr_warn_ratelimited(
2659 "VFS: Lookup of '%s' in %s %s"
2660 " would have caused loop\n",
2661 dentry->d_name.name,
2662 inode->i_sb->s_type->name,
2664 } else if (!IS_ROOT(new)) {
2665 int err = __d_unalias(inode, dentry, new);
2666 write_sequnlock(&rename_lock);
2672 __d_move(new, dentry, false);
2673 write_sequnlock(&rename_lock);
2674 spin_unlock(&inode->i_lock);
2675 security_d_instantiate(new, inode);
2681 /* already taking inode->i_lock, so d_add() by hand */
2682 __d_instantiate(dentry, inode);
2683 spin_unlock(&inode->i_lock);
2685 security_d_instantiate(dentry, inode);
2689 EXPORT_SYMBOL(d_splice_alias);
2691 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2695 return -ENAMETOOLONG;
2697 memcpy(*buffer, str, namelen);
2702 * prepend_name - prepend a pathname in front of current buffer pointer
2703 * @buffer: buffer pointer
2704 * @buflen: allocated length of the buffer
2705 * @name: name string and length qstr structure
2707 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2708 * make sure that either the old or the new name pointer and length are
2709 * fetched. However, there may be mismatch between length and pointer.
2710 * The length cannot be trusted, we need to copy it byte-by-byte until
2711 * the length is reached or a null byte is found. It also prepends "/" at
2712 * the beginning of the name. The sequence number check at the caller will
2713 * retry it again when a d_move() does happen. So any garbage in the buffer
2714 * due to mismatched pointer and length will be discarded.
2716 * Data dependency barrier is needed to make sure that we see that terminating
2717 * NUL. Alpha strikes again, film at 11...
2719 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2721 const char *dname = ACCESS_ONCE(name->name);
2722 u32 dlen = ACCESS_ONCE(name->len);
2725 smp_read_barrier_depends();
2727 *buflen -= dlen + 1;
2729 return -ENAMETOOLONG;
2730 p = *buffer -= dlen + 1;
2742 * prepend_path - Prepend path string to a buffer
2743 * @path: the dentry/vfsmount to report
2744 * @root: root vfsmnt/dentry
2745 * @buffer: pointer to the end of the buffer
2746 * @buflen: pointer to buffer length
2748 * The function will first try to write out the pathname without taking any
2749 * lock other than the RCU read lock to make sure that dentries won't go away.
2750 * It only checks the sequence number of the global rename_lock as any change
2751 * in the dentry's d_seq will be preceded by changes in the rename_lock
2752 * sequence number. If the sequence number had been changed, it will restart
2753 * the whole pathname back-tracing sequence again by taking the rename_lock.
2754 * In this case, there is no need to take the RCU read lock as the recursive
2755 * parent pointer references will keep the dentry chain alive as long as no
2756 * rename operation is performed.
2758 static int prepend_path(const struct path *path,
2759 const struct path *root,
2760 char **buffer, int *buflen)
2762 struct dentry *dentry;
2763 struct vfsmount *vfsmnt;
2766 unsigned seq, m_seq = 0;
2772 read_seqbegin_or_lock(&mount_lock, &m_seq);
2779 dentry = path->dentry;
2781 mnt = real_mount(vfsmnt);
2782 read_seqbegin_or_lock(&rename_lock, &seq);
2783 while (dentry != root->dentry || vfsmnt != root->mnt) {
2784 struct dentry * parent;
2786 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2787 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2789 if (mnt != parent) {
2790 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2796 * Filesystems needing to implement special "root names"
2797 * should do so with ->d_dname()
2799 if (IS_ROOT(dentry) &&
2800 (dentry->d_name.len != 1 ||
2801 dentry->d_name.name[0] != '/')) {
2802 WARN(1, "Root dentry has weird name <%.*s>\n",
2803 (int) dentry->d_name.len,
2804 dentry->d_name.name);
2807 error = is_mounted(vfsmnt) ? 1 : 2;
2810 parent = dentry->d_parent;
2812 error = prepend_name(&bptr, &blen, &dentry->d_name);
2820 if (need_seqretry(&rename_lock, seq)) {
2824 done_seqretry(&rename_lock, seq);
2828 if (need_seqretry(&mount_lock, m_seq)) {
2832 done_seqretry(&mount_lock, m_seq);
2834 if (error >= 0 && bptr == *buffer) {
2836 error = -ENAMETOOLONG;
2846 * __d_path - return the path of a dentry
2847 * @path: the dentry/vfsmount to report
2848 * @root: root vfsmnt/dentry
2849 * @buf: buffer to return value in
2850 * @buflen: buffer length
2852 * Convert a dentry into an ASCII path name.
2854 * Returns a pointer into the buffer or an error code if the
2855 * path was too long.
2857 * "buflen" should be positive.
2859 * If the path is not reachable from the supplied root, return %NULL.
2861 char *__d_path(const struct path *path,
2862 const struct path *root,
2863 char *buf, int buflen)
2865 char *res = buf + buflen;
2868 prepend(&res, &buflen, "\0", 1);
2869 error = prepend_path(path, root, &res, &buflen);
2872 return ERR_PTR(error);
2878 char *d_absolute_path(const struct path *path,
2879 char *buf, int buflen)
2881 struct path root = {};
2882 char *res = buf + buflen;
2885 prepend(&res, &buflen, "\0", 1);
2886 error = prepend_path(path, &root, &res, &buflen);
2891 return ERR_PTR(error);
2896 * same as __d_path but appends "(deleted)" for unlinked files.
2898 static int path_with_deleted(const struct path *path,
2899 const struct path *root,
2900 char **buf, int *buflen)
2902 prepend(buf, buflen, "\0", 1);
2903 if (d_unlinked(path->dentry)) {
2904 int error = prepend(buf, buflen, " (deleted)", 10);
2909 return prepend_path(path, root, buf, buflen);
2912 static int prepend_unreachable(char **buffer, int *buflen)
2914 return prepend(buffer, buflen, "(unreachable)", 13);
2917 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2922 seq = read_seqcount_begin(&fs->seq);
2924 } while (read_seqcount_retry(&fs->seq, seq));
2928 * d_path - return the path of a dentry
2929 * @path: path to report
2930 * @buf: buffer to return value in
2931 * @buflen: buffer length
2933 * Convert a dentry into an ASCII path name. If the entry has been deleted
2934 * the string " (deleted)" is appended. Note that this is ambiguous.
2936 * Returns a pointer into the buffer or an error code if the path was
2937 * too long. Note: Callers should use the returned pointer, not the passed
2938 * in buffer, to use the name! The implementation often starts at an offset
2939 * into the buffer, and may leave 0 bytes at the start.
2941 * "buflen" should be positive.
2943 char *d_path(const struct path *path, char *buf, int buflen)
2945 char *res = buf + buflen;
2950 * We have various synthetic filesystems that never get mounted. On
2951 * these filesystems dentries are never used for lookup purposes, and
2952 * thus don't need to be hashed. They also don't need a name until a
2953 * user wants to identify the object in /proc/pid/fd/. The little hack
2954 * below allows us to generate a name for these objects on demand:
2956 * Some pseudo inodes are mountable. When they are mounted
2957 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
2958 * and instead have d_path return the mounted path.
2960 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
2961 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
2962 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2965 get_fs_root_rcu(current->fs, &root);
2966 error = path_with_deleted(path, &root, &res, &buflen);
2970 res = ERR_PTR(error);
2973 EXPORT_SYMBOL(d_path);
2976 * Helper function for dentry_operations.d_dname() members
2978 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2979 const char *fmt, ...)
2985 va_start(args, fmt);
2986 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2989 if (sz > sizeof(temp) || sz > buflen)
2990 return ERR_PTR(-ENAMETOOLONG);
2992 buffer += buflen - sz;
2993 return memcpy(buffer, temp, sz);
2996 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2998 char *end = buffer + buflen;
2999 /* these dentries are never renamed, so d_lock is not needed */
3000 if (prepend(&end, &buflen, " (deleted)", 11) ||
3001 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3002 prepend(&end, &buflen, "/", 1))
3003 end = ERR_PTR(-ENAMETOOLONG);
3006 EXPORT_SYMBOL(simple_dname);
3009 * Write full pathname from the root of the filesystem into the buffer.
3011 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3013 struct dentry *dentry;
3026 prepend(&end, &len, "\0", 1);
3030 read_seqbegin_or_lock(&rename_lock, &seq);
3031 while (!IS_ROOT(dentry)) {
3032 struct dentry *parent = dentry->d_parent;
3035 error = prepend_name(&end, &len, &dentry->d_name);
3044 if (need_seqretry(&rename_lock, seq)) {
3048 done_seqretry(&rename_lock, seq);
3053 return ERR_PTR(-ENAMETOOLONG);
3056 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3058 return __dentry_path(dentry, buf, buflen);
3060 EXPORT_SYMBOL(dentry_path_raw);
3062 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3067 if (d_unlinked(dentry)) {
3069 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3073 retval = __dentry_path(dentry, buf, buflen);
3074 if (!IS_ERR(retval) && p)
3075 *p = '/'; /* restore '/' overriden with '\0' */
3078 return ERR_PTR(-ENAMETOOLONG);
3081 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3087 seq = read_seqcount_begin(&fs->seq);
3090 } while (read_seqcount_retry(&fs->seq, seq));
3094 * NOTE! The user-level library version returns a
3095 * character pointer. The kernel system call just
3096 * returns the length of the buffer filled (which
3097 * includes the ending '\0' character), or a negative
3098 * error value. So libc would do something like
3100 * char *getcwd(char * buf, size_t size)
3104 * retval = sys_getcwd(buf, size);
3111 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3114 struct path pwd, root;
3115 char *page = __getname();
3121 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3124 if (!d_unlinked(pwd.dentry)) {
3126 char *cwd = page + PATH_MAX;
3127 int buflen = PATH_MAX;
3129 prepend(&cwd, &buflen, "\0", 1);
3130 error = prepend_path(&pwd, &root, &cwd, &buflen);
3136 /* Unreachable from current root */
3138 error = prepend_unreachable(&cwd, &buflen);
3144 len = PATH_MAX + page - cwd;
3147 if (copy_to_user(buf, cwd, len))
3160 * Test whether new_dentry is a subdirectory of old_dentry.
3162 * Trivially implemented using the dcache structure
3166 * is_subdir - is new dentry a subdirectory of old_dentry
3167 * @new_dentry: new dentry
3168 * @old_dentry: old dentry
3170 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3171 * Returns 0 otherwise.
3172 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3175 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3180 if (new_dentry == old_dentry)
3184 /* for restarting inner loop in case of seq retry */
3185 seq = read_seqbegin(&rename_lock);
3187 * Need rcu_readlock to protect against the d_parent trashing
3191 if (d_ancestor(old_dentry, new_dentry))
3196 } while (read_seqretry(&rename_lock, seq));
3201 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3203 struct dentry *root = data;
3204 if (dentry != root) {
3205 if (d_unhashed(dentry) || !dentry->d_inode)
3208 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3209 dentry->d_flags |= DCACHE_GENOCIDE;
3210 dentry->d_lockref.count--;
3213 return D_WALK_CONTINUE;
3216 void d_genocide(struct dentry *parent)
3218 d_walk(parent, parent, d_genocide_kill, NULL);
3221 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3223 inode_dec_link_count(inode);
3224 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3225 !hlist_unhashed(&dentry->d_u.d_alias) ||
3226 !d_unlinked(dentry));
3227 spin_lock(&dentry->d_parent->d_lock);
3228 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3229 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3230 (unsigned long long)inode->i_ino);
3231 spin_unlock(&dentry->d_lock);
3232 spin_unlock(&dentry->d_parent->d_lock);
3233 d_instantiate(dentry, inode);
3235 EXPORT_SYMBOL(d_tmpfile);
3237 static __initdata unsigned long dhash_entries;
3238 static int __init set_dhash_entries(char *str)
3242 dhash_entries = simple_strtoul(str, &str, 0);
3245 __setup("dhash_entries=", set_dhash_entries);
3247 static void __init dcache_init_early(void)
3251 /* If hashes are distributed across NUMA nodes, defer
3252 * hash allocation until vmalloc space is available.
3258 alloc_large_system_hash("Dentry cache",
3259 sizeof(struct hlist_bl_head),
3268 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3269 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3272 static void __init dcache_init(void)
3277 * A constructor could be added for stable state like the lists,
3278 * but it is probably not worth it because of the cache nature
3281 dentry_cache = KMEM_CACHE(dentry,
3282 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3284 /* Hash may have been set up in dcache_init_early */
3289 alloc_large_system_hash("Dentry cache",
3290 sizeof(struct hlist_bl_head),
3299 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3300 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3303 /* SLAB cache for __getname() consumers */
3304 struct kmem_cache *names_cachep __read_mostly;
3305 EXPORT_SYMBOL(names_cachep);
3307 EXPORT_SYMBOL(d_genocide);
3309 void __init vfs_caches_init_early(void)
3311 dcache_init_early();
3315 void __init vfs_caches_init(unsigned long mempages)
3317 unsigned long reserve;
3319 /* Base hash sizes on available memory, with a reserve equal to
3320 150% of current kernel size */
3322 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3323 mempages -= reserve;
3325 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3326 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3330 files_init(mempages);