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>
46 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_u.d_alias, d_inode of aliases
48 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
52 * dentry->d_sb->s_dentry_lru_lock protects:
53 * - the dcache lru lists and counters
60 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
62 * - d_u.d_alias, d_inode
65 * dentry->d_inode->i_lock
67 * dentry->d_sb->s_dentry_lru_lock
68 * dcache_hash_bucket lock
71 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
74 * dentry->d_parent->d_lock
77 * If no ancestor relationship:
78 * if (dentry1 < dentry2)
82 int sysctl_vfs_cache_pressure __read_mostly = 100;
83 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
87 EXPORT_SYMBOL(rename_lock);
89 static struct kmem_cache *dentry_cache __read_mostly;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
100 static unsigned int d_hash_mask __read_mostly;
101 static unsigned int d_hash_shift __read_mostly;
103 static struct hlist_bl_head *dentry_hashtable __read_mostly;
105 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
108 hash += (unsigned long) parent / L1_CACHE_BYTES;
109 return dentry_hashtable + hash_32(hash, d_hash_shift);
112 /* Statistics gathering. */
113 struct dentry_stat_t dentry_stat = {
117 static DEFINE_PER_CPU(long, nr_dentry);
118 static DEFINE_PER_CPU(long, nr_dentry_unused);
120 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 * Here we resort to our own counters instead of using generic per-cpu counters
124 * for consistency with what the vfs inode code does. We are expected to harvest
125 * better code and performance by having our own specialized counters.
127 * Please note that the loop is done over all possible CPUs, not over all online
128 * CPUs. The reason for this is that we don't want to play games with CPUs going
129 * on and off. If one of them goes off, we will just keep their counters.
131 * glommer: See cffbc8a for details, and if you ever intend to change this,
132 * please update all vfs counters to match.
134 static long get_nr_dentry(void)
138 for_each_possible_cpu(i)
139 sum += per_cpu(nr_dentry, i);
140 return sum < 0 ? 0 : sum;
143 static long get_nr_dentry_unused(void)
147 for_each_possible_cpu(i)
148 sum += per_cpu(nr_dentry_unused, i);
149 return sum < 0 ? 0 : sum;
152 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
153 size_t *lenp, loff_t *ppos)
155 dentry_stat.nr_dentry = get_nr_dentry();
156 dentry_stat.nr_unused = get_nr_dentry_unused();
157 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
162 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
163 * The strings are both count bytes long, and count is non-zero.
165 #ifdef CONFIG_DCACHE_WORD_ACCESS
167 #include <asm/word-at-a-time.h>
169 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
170 * aligned allocation for this particular component. We don't
171 * strictly need the load_unaligned_zeropad() safety, but it
172 * doesn't hurt either.
174 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
175 * need the careful unaligned handling.
177 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
179 unsigned long a,b,mask;
182 a = *(unsigned long *)cs;
183 b = load_unaligned_zeropad(ct);
184 if (tcount < sizeof(unsigned long))
186 if (unlikely(a != b))
188 cs += sizeof(unsigned long);
189 ct += sizeof(unsigned long);
190 tcount -= sizeof(unsigned long);
194 mask = bytemask_from_count(tcount);
195 return unlikely(!!((a ^ b) & mask));
200 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
214 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
216 const unsigned char *cs;
218 * Be careful about RCU walk racing with rename:
219 * use ACCESS_ONCE to fetch the name pointer.
221 * NOTE! Even if a rename will mean that the length
222 * was not loaded atomically, we don't care. The
223 * RCU walk will check the sequence count eventually,
224 * and catch it. And we won't overrun the buffer,
225 * because we're reading the name pointer atomically,
226 * and a dentry name is guaranteed to be properly
227 * terminated with a NUL byte.
229 * End result: even if 'len' is wrong, we'll exit
230 * early because the data cannot match (there can
231 * be no NUL in the ct/tcount data)
233 cs = ACCESS_ONCE(dentry->d_name.name);
234 smp_read_barrier_depends();
235 return dentry_string_cmp(cs, ct, tcount);
238 struct external_name {
241 struct rcu_head head;
243 unsigned char name[];
246 static inline struct external_name *external_name(struct dentry *dentry)
248 return container_of(dentry->d_name.name, struct external_name, name[0]);
251 static void __d_free(struct rcu_head *head)
253 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
255 kmem_cache_free(dentry_cache, dentry);
258 static void __d_free_external(struct rcu_head *head)
260 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
261 kfree(external_name(dentry));
262 kmem_cache_free(dentry_cache, dentry);
265 static inline int dname_external(const struct dentry *dentry)
267 return dentry->d_name.name != dentry->d_iname;
270 static void dentry_free(struct dentry *dentry)
272 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
273 if (unlikely(dname_external(dentry))) {
274 struct external_name *p = external_name(dentry);
275 if (likely(atomic_dec_and_test(&p->u.count))) {
276 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
280 /* if dentry was never visible to RCU, immediate free is OK */
281 if (!(dentry->d_flags & DCACHE_RCUACCESS))
282 __d_free(&dentry->d_u.d_rcu);
284 call_rcu(&dentry->d_u.d_rcu, __d_free);
288 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
289 * @dentry: the target dentry
290 * After this call, in-progress rcu-walk path lookup will fail. This
291 * should be called after unhashing, and after changing d_inode (if
292 * the dentry has not already been unhashed).
294 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
296 assert_spin_locked(&dentry->d_lock);
297 /* Go through a barrier */
298 write_seqcount_barrier(&dentry->d_seq);
302 * Release the dentry's inode, using the filesystem
303 * d_iput() operation if defined. Dentry has no refcount
306 static void dentry_iput(struct dentry * dentry)
307 __releases(dentry->d_lock)
308 __releases(dentry->d_inode->i_lock)
310 struct inode *inode = dentry->d_inode;
312 dentry->d_inode = NULL;
313 hlist_del_init(&dentry->d_u.d_alias);
314 spin_unlock(&dentry->d_lock);
315 spin_unlock(&inode->i_lock);
317 fsnotify_inoderemove(inode);
318 if (dentry->d_op && dentry->d_op->d_iput)
319 dentry->d_op->d_iput(dentry, inode);
323 spin_unlock(&dentry->d_lock);
328 * Release the dentry's inode, using the filesystem
329 * d_iput() operation if defined. dentry remains in-use.
331 static void dentry_unlink_inode(struct dentry * dentry)
332 __releases(dentry->d_lock)
333 __releases(dentry->d_inode->i_lock)
335 struct inode *inode = dentry->d_inode;
336 __d_clear_type(dentry);
337 dentry->d_inode = NULL;
338 hlist_del_init(&dentry->d_u.d_alias);
339 dentry_rcuwalk_barrier(dentry);
340 spin_unlock(&dentry->d_lock);
341 spin_unlock(&inode->i_lock);
343 fsnotify_inoderemove(inode);
344 if (dentry->d_op && dentry->d_op->d_iput)
345 dentry->d_op->d_iput(dentry, inode);
351 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
352 * is in use - which includes both the "real" per-superblock
353 * LRU list _and_ the DCACHE_SHRINK_LIST use.
355 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
356 * on the shrink list (ie not on the superblock LRU list).
358 * The per-cpu "nr_dentry_unused" counters are updated with
359 * the DCACHE_LRU_LIST bit.
361 * These helper functions make sure we always follow the
362 * rules. d_lock must be held by the caller.
364 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
365 static void d_lru_add(struct dentry *dentry)
367 D_FLAG_VERIFY(dentry, 0);
368 dentry->d_flags |= DCACHE_LRU_LIST;
369 this_cpu_inc(nr_dentry_unused);
370 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
373 static void d_lru_del(struct dentry *dentry)
375 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
376 dentry->d_flags &= ~DCACHE_LRU_LIST;
377 this_cpu_dec(nr_dentry_unused);
378 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
381 static void d_shrink_del(struct dentry *dentry)
383 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
384 list_del_init(&dentry->d_lru);
385 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
386 this_cpu_dec(nr_dentry_unused);
389 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
391 D_FLAG_VERIFY(dentry, 0);
392 list_add(&dentry->d_lru, list);
393 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
394 this_cpu_inc(nr_dentry_unused);
398 * These can only be called under the global LRU lock, ie during the
399 * callback for freeing the LRU list. "isolate" removes it from the
400 * LRU lists entirely, while shrink_move moves it to the indicated
403 static void d_lru_isolate(struct dentry *dentry)
405 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
406 dentry->d_flags &= ~DCACHE_LRU_LIST;
407 this_cpu_dec(nr_dentry_unused);
408 list_del_init(&dentry->d_lru);
411 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
413 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
414 dentry->d_flags |= DCACHE_SHRINK_LIST;
415 list_move_tail(&dentry->d_lru, list);
419 * dentry_lru_(add|del)_list) must be called with d_lock held.
421 static void dentry_lru_add(struct dentry *dentry)
423 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
428 * d_drop - drop a dentry
429 * @dentry: dentry to drop
431 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
432 * be found through a VFS lookup any more. Note that this is different from
433 * deleting the dentry - d_delete will try to mark the dentry negative if
434 * possible, giving a successful _negative_ lookup, while d_drop will
435 * just make the cache lookup fail.
437 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
438 * reason (NFS timeouts or autofs deletes).
440 * __d_drop requires dentry->d_lock.
442 void __d_drop(struct dentry *dentry)
444 if (!d_unhashed(dentry)) {
445 struct hlist_bl_head *b;
447 * Hashed dentries are normally on the dentry hashtable,
448 * with the exception of those newly allocated by
449 * d_obtain_alias, which are always IS_ROOT:
451 if (unlikely(IS_ROOT(dentry)))
452 b = &dentry->d_sb->s_anon;
454 b = d_hash(dentry->d_parent, dentry->d_name.hash);
457 __hlist_bl_del(&dentry->d_hash);
458 dentry->d_hash.pprev = NULL;
460 dentry_rcuwalk_barrier(dentry);
463 EXPORT_SYMBOL(__d_drop);
465 void d_drop(struct dentry *dentry)
467 spin_lock(&dentry->d_lock);
469 spin_unlock(&dentry->d_lock);
471 EXPORT_SYMBOL(d_drop);
473 static void __dentry_kill(struct dentry *dentry)
475 struct dentry *parent = NULL;
476 bool can_free = true;
477 if (!IS_ROOT(dentry))
478 parent = dentry->d_parent;
481 * The dentry is now unrecoverably dead to the world.
483 lockref_mark_dead(&dentry->d_lockref);
486 * inform the fs via d_prune that this dentry is about to be
487 * unhashed and destroyed.
489 if (dentry->d_flags & DCACHE_OP_PRUNE)
490 dentry->d_op->d_prune(dentry);
492 if (dentry->d_flags & DCACHE_LRU_LIST) {
493 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
496 /* if it was on the hash then remove it */
498 __list_del_entry(&dentry->d_child);
500 * Inform d_walk() that we are no longer attached to the
503 dentry->d_flags |= DCACHE_DENTRY_KILLED;
505 spin_unlock(&parent->d_lock);
508 * dentry_iput drops the locks, at which point nobody (except
509 * transient RCU lookups) can reach this dentry.
511 BUG_ON((int)dentry->d_lockref.count > 0);
512 this_cpu_dec(nr_dentry);
513 if (dentry->d_op && dentry->d_op->d_release)
514 dentry->d_op->d_release(dentry);
516 spin_lock(&dentry->d_lock);
517 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
518 dentry->d_flags |= DCACHE_MAY_FREE;
521 spin_unlock(&dentry->d_lock);
522 if (likely(can_free))
527 * Finish off a dentry we've decided to kill.
528 * dentry->d_lock must be held, returns with it unlocked.
529 * If ref is non-zero, then decrement the refcount too.
530 * Returns dentry requiring refcount drop, or NULL if we're done.
532 static struct dentry *dentry_kill(struct dentry *dentry)
533 __releases(dentry->d_lock)
535 struct inode *inode = dentry->d_inode;
536 struct dentry *parent = NULL;
538 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
541 if (!IS_ROOT(dentry)) {
542 parent = dentry->d_parent;
543 if (unlikely(!spin_trylock(&parent->d_lock))) {
545 spin_unlock(&inode->i_lock);
550 __dentry_kill(dentry);
554 spin_unlock(&dentry->d_lock);
556 return dentry; /* try again with same dentry */
559 static inline struct dentry *lock_parent(struct dentry *dentry)
561 struct dentry *parent = dentry->d_parent;
564 if (unlikely((int)dentry->d_lockref.count < 0))
566 if (likely(spin_trylock(&parent->d_lock)))
569 spin_unlock(&dentry->d_lock);
571 parent = ACCESS_ONCE(dentry->d_parent);
572 spin_lock(&parent->d_lock);
574 * We can't blindly lock dentry until we are sure
575 * that we won't violate the locking order.
576 * Any changes of dentry->d_parent must have
577 * been done with parent->d_lock held, so
578 * spin_lock() above is enough of a barrier
579 * for checking if it's still our child.
581 if (unlikely(parent != dentry->d_parent)) {
582 spin_unlock(&parent->d_lock);
586 if (parent != dentry)
587 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
596 * This is complicated by the fact that we do not want to put
597 * dentries that are no longer on any hash chain on the unused
598 * list: we'd much rather just get rid of them immediately.
600 * However, that implies that we have to traverse the dentry
601 * tree upwards to the parents which might _also_ now be
602 * scheduled for deletion (it may have been only waiting for
603 * its last child to go away).
605 * This tail recursion is done by hand as we don't want to depend
606 * on the compiler to always get this right (gcc generally doesn't).
607 * Real recursion would eat up our stack space.
611 * dput - release a dentry
612 * @dentry: dentry to release
614 * Release a dentry. This will drop the usage count and if appropriate
615 * call the dentry unlink method as well as removing it from the queues and
616 * releasing its resources. If the parent dentries were scheduled for release
617 * they too may now get deleted.
619 void dput(struct dentry *dentry)
621 if (unlikely(!dentry))
625 if (lockref_put_or_lock(&dentry->d_lockref))
628 /* Unreachable? Get rid of it */
629 if (unlikely(d_unhashed(dentry)))
632 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
633 if (dentry->d_op->d_delete(dentry))
637 if (!(dentry->d_flags & DCACHE_REFERENCED))
638 dentry->d_flags |= DCACHE_REFERENCED;
639 dentry_lru_add(dentry);
641 dentry->d_lockref.count--;
642 spin_unlock(&dentry->d_lock);
646 dentry = dentry_kill(dentry);
653 /* This must be called with d_lock held */
654 static inline void __dget_dlock(struct dentry *dentry)
656 dentry->d_lockref.count++;
659 static inline void __dget(struct dentry *dentry)
661 lockref_get(&dentry->d_lockref);
664 struct dentry *dget_parent(struct dentry *dentry)
670 * Do optimistic parent lookup without any
674 ret = ACCESS_ONCE(dentry->d_parent);
675 gotref = lockref_get_not_zero(&ret->d_lockref);
677 if (likely(gotref)) {
678 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
685 * Don't need rcu_dereference because we re-check it was correct under
689 ret = dentry->d_parent;
690 spin_lock(&ret->d_lock);
691 if (unlikely(ret != dentry->d_parent)) {
692 spin_unlock(&ret->d_lock);
697 BUG_ON(!ret->d_lockref.count);
698 ret->d_lockref.count++;
699 spin_unlock(&ret->d_lock);
702 EXPORT_SYMBOL(dget_parent);
705 * d_find_alias - grab a hashed alias of inode
706 * @inode: inode in question
708 * If inode has a hashed alias, or is a directory and has any alias,
709 * acquire the reference to alias and return it. Otherwise return NULL.
710 * Notice that if inode is a directory there can be only one alias and
711 * it can be unhashed only if it has no children, or if it is the root
712 * of a filesystem, or if the directory was renamed and d_revalidate
713 * was the first vfs operation to notice.
715 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
716 * any other hashed alias over that one.
718 static struct dentry *__d_find_alias(struct inode *inode)
720 struct dentry *alias, *discon_alias;
724 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
725 spin_lock(&alias->d_lock);
726 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
727 if (IS_ROOT(alias) &&
728 (alias->d_flags & DCACHE_DISCONNECTED)) {
729 discon_alias = alias;
732 spin_unlock(&alias->d_lock);
736 spin_unlock(&alias->d_lock);
739 alias = discon_alias;
740 spin_lock(&alias->d_lock);
741 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
743 spin_unlock(&alias->d_lock);
746 spin_unlock(&alias->d_lock);
752 struct dentry *d_find_alias(struct inode *inode)
754 struct dentry *de = NULL;
756 if (!hlist_empty(&inode->i_dentry)) {
757 spin_lock(&inode->i_lock);
758 de = __d_find_alias(inode);
759 spin_unlock(&inode->i_lock);
763 EXPORT_SYMBOL(d_find_alias);
766 * Try to kill dentries associated with this inode.
767 * WARNING: you must own a reference to inode.
769 void d_prune_aliases(struct inode *inode)
771 struct dentry *dentry;
773 spin_lock(&inode->i_lock);
774 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
775 spin_lock(&dentry->d_lock);
776 if (!dentry->d_lockref.count) {
777 struct dentry *parent = lock_parent(dentry);
778 if (likely(!dentry->d_lockref.count)) {
779 __dentry_kill(dentry);
784 spin_unlock(&parent->d_lock);
786 spin_unlock(&dentry->d_lock);
788 spin_unlock(&inode->i_lock);
790 EXPORT_SYMBOL(d_prune_aliases);
792 static void shrink_dentry_list(struct list_head *list)
794 struct dentry *dentry, *parent;
796 while (!list_empty(list)) {
798 dentry = list_entry(list->prev, struct dentry, d_lru);
799 spin_lock(&dentry->d_lock);
800 parent = lock_parent(dentry);
803 * The dispose list is isolated and dentries are not accounted
804 * to the LRU here, so we can simply remove it from the list
805 * here regardless of whether it is referenced or not.
807 d_shrink_del(dentry);
810 * We found an inuse dentry which was not removed from
811 * the LRU because of laziness during lookup. Do not free it.
813 if ((int)dentry->d_lockref.count > 0) {
814 spin_unlock(&dentry->d_lock);
816 spin_unlock(&parent->d_lock);
821 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
822 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
823 spin_unlock(&dentry->d_lock);
825 spin_unlock(&parent->d_lock);
831 inode = dentry->d_inode;
832 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
833 d_shrink_add(dentry, list);
834 spin_unlock(&dentry->d_lock);
836 spin_unlock(&parent->d_lock);
840 __dentry_kill(dentry);
843 * We need to prune ancestors too. This is necessary to prevent
844 * quadratic behavior of shrink_dcache_parent(), but is also
845 * expected to be beneficial in reducing dentry cache
849 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
850 parent = lock_parent(dentry);
851 if (dentry->d_lockref.count != 1) {
852 dentry->d_lockref.count--;
853 spin_unlock(&dentry->d_lock);
855 spin_unlock(&parent->d_lock);
858 inode = dentry->d_inode; /* can't be NULL */
859 if (unlikely(!spin_trylock(&inode->i_lock))) {
860 spin_unlock(&dentry->d_lock);
862 spin_unlock(&parent->d_lock);
866 __dentry_kill(dentry);
872 static enum lru_status
873 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
875 struct list_head *freeable = arg;
876 struct dentry *dentry = container_of(item, struct dentry, d_lru);
880 * we are inverting the lru lock/dentry->d_lock here,
881 * so use a trylock. If we fail to get the lock, just skip
884 if (!spin_trylock(&dentry->d_lock))
888 * Referenced dentries are still in use. If they have active
889 * counts, just remove them from the LRU. Otherwise give them
890 * another pass through the LRU.
892 if (dentry->d_lockref.count) {
893 d_lru_isolate(dentry);
894 spin_unlock(&dentry->d_lock);
898 if (dentry->d_flags & DCACHE_REFERENCED) {
899 dentry->d_flags &= ~DCACHE_REFERENCED;
900 spin_unlock(&dentry->d_lock);
903 * The list move itself will be made by the common LRU code. At
904 * this point, we've dropped the dentry->d_lock but keep the
905 * lru lock. This is safe to do, since every list movement is
906 * protected by the lru lock even if both locks are held.
908 * This is guaranteed by the fact that all LRU management
909 * functions are intermediated by the LRU API calls like
910 * list_lru_add and list_lru_del. List movement in this file
911 * only ever occur through this functions or through callbacks
912 * like this one, that are called from the LRU API.
914 * The only exceptions to this are functions like
915 * shrink_dentry_list, and code that first checks for the
916 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
917 * operating only with stack provided lists after they are
918 * properly isolated from the main list. It is thus, always a
924 d_lru_shrink_move(dentry, freeable);
925 spin_unlock(&dentry->d_lock);
931 * prune_dcache_sb - shrink the dcache
933 * @sc: shrink control, passed to list_lru_shrink_walk()
935 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
936 * is done when we need more memory and called from the superblock shrinker
939 * This function may fail to free any resources if all the dentries are in
942 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
947 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
948 dentry_lru_isolate, &dispose);
949 shrink_dentry_list(&dispose);
953 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
954 spinlock_t *lru_lock, void *arg)
956 struct list_head *freeable = arg;
957 struct dentry *dentry = container_of(item, struct dentry, d_lru);
960 * we are inverting the lru lock/dentry->d_lock here,
961 * so use a trylock. If we fail to get the lock, just skip
964 if (!spin_trylock(&dentry->d_lock))
967 d_lru_shrink_move(dentry, freeable);
968 spin_unlock(&dentry->d_lock);
975 * shrink_dcache_sb - shrink dcache for a superblock
978 * Shrink the dcache for the specified super block. This is used to free
979 * the dcache before unmounting a file system.
981 void shrink_dcache_sb(struct super_block *sb)
988 freed = list_lru_walk(&sb->s_dentry_lru,
989 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
991 this_cpu_sub(nr_dentry_unused, freed);
992 shrink_dentry_list(&dispose);
995 EXPORT_SYMBOL(shrink_dcache_sb);
998 * enum d_walk_ret - action to talke during tree walk
999 * @D_WALK_CONTINUE: contrinue walk
1000 * @D_WALK_QUIT: quit walk
1001 * @D_WALK_NORETRY: quit when retry is needed
1002 * @D_WALK_SKIP: skip this dentry and its children
1012 * d_walk - walk the dentry tree
1013 * @parent: start of walk
1014 * @data: data passed to @enter() and @finish()
1015 * @enter: callback when first entering the dentry
1016 * @finish: callback when successfully finished the walk
1018 * The @enter() and @finish() callbacks are called with d_lock held.
1020 static void d_walk(struct dentry *parent, void *data,
1021 enum d_walk_ret (*enter)(void *, struct dentry *),
1022 void (*finish)(void *))
1024 struct dentry *this_parent;
1025 struct list_head *next;
1027 enum d_walk_ret ret;
1031 read_seqbegin_or_lock(&rename_lock, &seq);
1032 this_parent = parent;
1033 spin_lock(&this_parent->d_lock);
1035 ret = enter(data, this_parent);
1037 case D_WALK_CONTINUE:
1042 case D_WALK_NORETRY:
1047 next = this_parent->d_subdirs.next;
1049 while (next != &this_parent->d_subdirs) {
1050 struct list_head *tmp = next;
1051 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1054 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1056 ret = enter(data, dentry);
1058 case D_WALK_CONTINUE:
1061 spin_unlock(&dentry->d_lock);
1063 case D_WALK_NORETRY:
1067 spin_unlock(&dentry->d_lock);
1071 if (!list_empty(&dentry->d_subdirs)) {
1072 spin_unlock(&this_parent->d_lock);
1073 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1074 this_parent = dentry;
1075 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1078 spin_unlock(&dentry->d_lock);
1081 * All done at this level ... ascend and resume the search.
1085 if (this_parent != parent) {
1086 struct dentry *child = this_parent;
1087 this_parent = child->d_parent;
1089 spin_unlock(&child->d_lock);
1090 spin_lock(&this_parent->d_lock);
1092 /* might go back up the wrong parent if we have had a rename. */
1093 if (need_seqretry(&rename_lock, seq))
1095 next = child->d_child.next;
1096 while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)) {
1097 if (next == &this_parent->d_subdirs)
1099 child = list_entry(next, struct dentry, d_child);
1105 if (need_seqretry(&rename_lock, seq))
1112 spin_unlock(&this_parent->d_lock);
1113 done_seqretry(&rename_lock, seq);
1117 spin_unlock(&this_parent->d_lock);
1127 * Search for at least 1 mount point in the dentry's subdirs.
1128 * We descend to the next level whenever the d_subdirs
1129 * list is non-empty and continue searching.
1132 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1135 if (d_mountpoint(dentry)) {
1139 return D_WALK_CONTINUE;
1143 * have_submounts - check for mounts over a dentry
1144 * @parent: dentry to check.
1146 * Return true if the parent or its subdirectories contain
1149 int have_submounts(struct dentry *parent)
1153 d_walk(parent, &ret, check_mount, NULL);
1157 EXPORT_SYMBOL(have_submounts);
1160 * Called by mount code to set a mountpoint and check if the mountpoint is
1161 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1162 * subtree can become unreachable).
1164 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1165 * this reason take rename_lock and d_lock on dentry and ancestors.
1167 int d_set_mounted(struct dentry *dentry)
1171 write_seqlock(&rename_lock);
1172 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1173 /* Need exclusion wrt. d_invalidate() */
1174 spin_lock(&p->d_lock);
1175 if (unlikely(d_unhashed(p))) {
1176 spin_unlock(&p->d_lock);
1179 spin_unlock(&p->d_lock);
1181 spin_lock(&dentry->d_lock);
1182 if (!d_unlinked(dentry)) {
1183 dentry->d_flags |= DCACHE_MOUNTED;
1186 spin_unlock(&dentry->d_lock);
1188 write_sequnlock(&rename_lock);
1193 * Search the dentry child list of the specified parent,
1194 * and move any unused dentries to the end of the unused
1195 * list for prune_dcache(). We descend to the next level
1196 * whenever the d_subdirs list is non-empty and continue
1199 * It returns zero iff there are no unused children,
1200 * otherwise it returns the number of children moved to
1201 * the end of the unused list. This may not be the total
1202 * number of unused children, because select_parent can
1203 * drop the lock and return early due to latency
1207 struct select_data {
1208 struct dentry *start;
1209 struct list_head dispose;
1213 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1215 struct select_data *data = _data;
1216 enum d_walk_ret ret = D_WALK_CONTINUE;
1218 if (data->start == dentry)
1221 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1224 if (dentry->d_flags & DCACHE_LRU_LIST)
1226 if (!dentry->d_lockref.count) {
1227 d_shrink_add(dentry, &data->dispose);
1232 * We can return to the caller if we have found some (this
1233 * ensures forward progress). We'll be coming back to find
1236 if (!list_empty(&data->dispose))
1237 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1243 * shrink_dcache_parent - prune dcache
1244 * @parent: parent of entries to prune
1246 * Prune the dcache to remove unused children of the parent dentry.
1248 void shrink_dcache_parent(struct dentry *parent)
1251 struct select_data data;
1253 INIT_LIST_HEAD(&data.dispose);
1254 data.start = parent;
1257 d_walk(parent, &data, select_collect, NULL);
1261 shrink_dentry_list(&data.dispose);
1265 EXPORT_SYMBOL(shrink_dcache_parent);
1267 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1269 /* it has busy descendents; complain about those instead */
1270 if (!list_empty(&dentry->d_subdirs))
1271 return D_WALK_CONTINUE;
1273 /* root with refcount 1 is fine */
1274 if (dentry == _data && dentry->d_lockref.count == 1)
1275 return D_WALK_CONTINUE;
1277 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1278 " still in use (%d) [unmount of %s %s]\n",
1281 dentry->d_inode->i_ino : 0UL,
1283 dentry->d_lockref.count,
1284 dentry->d_sb->s_type->name,
1285 dentry->d_sb->s_id);
1287 return D_WALK_CONTINUE;
1290 static void do_one_tree(struct dentry *dentry)
1292 shrink_dcache_parent(dentry);
1293 d_walk(dentry, dentry, umount_check, NULL);
1299 * destroy the dentries attached to a superblock on unmounting
1301 void shrink_dcache_for_umount(struct super_block *sb)
1303 struct dentry *dentry;
1305 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1307 dentry = sb->s_root;
1309 do_one_tree(dentry);
1311 while (!hlist_bl_empty(&sb->s_anon)) {
1312 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1313 do_one_tree(dentry);
1317 struct detach_data {
1318 struct select_data select;
1319 struct dentry *mountpoint;
1321 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1323 struct detach_data *data = _data;
1325 if (d_mountpoint(dentry)) {
1326 __dget_dlock(dentry);
1327 data->mountpoint = dentry;
1331 return select_collect(&data->select, dentry);
1334 static void check_and_drop(void *_data)
1336 struct detach_data *data = _data;
1338 if (!data->mountpoint && !data->select.found)
1339 __d_drop(data->select.start);
1343 * d_invalidate - detach submounts, prune dcache, and drop
1344 * @dentry: dentry to invalidate (aka detach, prune and drop)
1348 * The final d_drop is done as an atomic operation relative to
1349 * rename_lock ensuring there are no races with d_set_mounted. This
1350 * ensures there are no unhashed dentries on the path to a mountpoint.
1352 void d_invalidate(struct dentry *dentry)
1355 * If it's already been dropped, return OK.
1357 spin_lock(&dentry->d_lock);
1358 if (d_unhashed(dentry)) {
1359 spin_unlock(&dentry->d_lock);
1362 spin_unlock(&dentry->d_lock);
1364 /* Negative dentries can be dropped without further checks */
1365 if (!dentry->d_inode) {
1371 struct detach_data data;
1373 data.mountpoint = NULL;
1374 INIT_LIST_HEAD(&data.select.dispose);
1375 data.select.start = dentry;
1376 data.select.found = 0;
1378 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1380 if (data.select.found)
1381 shrink_dentry_list(&data.select.dispose);
1383 if (data.mountpoint) {
1384 detach_mounts(data.mountpoint);
1385 dput(data.mountpoint);
1388 if (!data.mountpoint && !data.select.found)
1394 EXPORT_SYMBOL(d_invalidate);
1397 * __d_alloc - allocate a dcache entry
1398 * @sb: filesystem it will belong to
1399 * @name: qstr of the name
1401 * Allocates a dentry. It returns %NULL if there is insufficient memory
1402 * available. On a success the dentry is returned. The name passed in is
1403 * copied and the copy passed in may be reused after this call.
1406 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1408 struct dentry *dentry;
1411 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1416 * We guarantee that the inline name is always NUL-terminated.
1417 * This way the memcpy() done by the name switching in rename
1418 * will still always have a NUL at the end, even if we might
1419 * be overwriting an internal NUL character
1421 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1422 if (name->len > DNAME_INLINE_LEN-1) {
1423 size_t size = offsetof(struct external_name, name[1]);
1424 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1426 kmem_cache_free(dentry_cache, dentry);
1429 atomic_set(&p->u.count, 1);
1432 dname = dentry->d_iname;
1435 dentry->d_name.len = name->len;
1436 dentry->d_name.hash = name->hash;
1437 memcpy(dname, name->name, name->len);
1438 dname[name->len] = 0;
1440 /* Make sure we always see the terminating NUL character */
1442 dentry->d_name.name = dname;
1444 dentry->d_lockref.count = 1;
1445 dentry->d_flags = 0;
1446 spin_lock_init(&dentry->d_lock);
1447 seqcount_init(&dentry->d_seq);
1448 dentry->d_inode = NULL;
1449 dentry->d_parent = dentry;
1451 dentry->d_op = NULL;
1452 dentry->d_fsdata = NULL;
1453 INIT_HLIST_BL_NODE(&dentry->d_hash);
1454 INIT_LIST_HEAD(&dentry->d_lru);
1455 INIT_LIST_HEAD(&dentry->d_subdirs);
1456 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1457 INIT_LIST_HEAD(&dentry->d_child);
1458 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1460 this_cpu_inc(nr_dentry);
1466 * d_alloc - allocate a dcache entry
1467 * @parent: parent of entry to allocate
1468 * @name: qstr of the name
1470 * Allocates a dentry. It returns %NULL if there is insufficient memory
1471 * available. On a success the dentry is returned. The name passed in is
1472 * copied and the copy passed in may be reused after this call.
1474 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1476 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1480 spin_lock(&parent->d_lock);
1482 * don't need child lock because it is not subject
1483 * to concurrency here
1485 __dget_dlock(parent);
1486 dentry->d_parent = parent;
1487 list_add(&dentry->d_child, &parent->d_subdirs);
1488 spin_unlock(&parent->d_lock);
1492 EXPORT_SYMBOL(d_alloc);
1495 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1496 * @sb: the superblock
1497 * @name: qstr of the name
1499 * For a filesystem that just pins its dentries in memory and never
1500 * performs lookups at all, return an unhashed IS_ROOT dentry.
1502 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1504 return __d_alloc(sb, name);
1506 EXPORT_SYMBOL(d_alloc_pseudo);
1508 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1513 q.len = strlen(name);
1514 q.hash = full_name_hash(q.name, q.len);
1515 return d_alloc(parent, &q);
1517 EXPORT_SYMBOL(d_alloc_name);
1519 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1521 WARN_ON_ONCE(dentry->d_op);
1522 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1524 DCACHE_OP_REVALIDATE |
1525 DCACHE_OP_WEAK_REVALIDATE |
1526 DCACHE_OP_DELETE ));
1531 dentry->d_flags |= DCACHE_OP_HASH;
1533 dentry->d_flags |= DCACHE_OP_COMPARE;
1534 if (op->d_revalidate)
1535 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1536 if (op->d_weak_revalidate)
1537 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1539 dentry->d_flags |= DCACHE_OP_DELETE;
1541 dentry->d_flags |= DCACHE_OP_PRUNE;
1544 EXPORT_SYMBOL(d_set_d_op);
1546 static unsigned d_flags_for_inode(struct inode *inode)
1548 unsigned add_flags = DCACHE_FILE_TYPE;
1551 return DCACHE_MISS_TYPE;
1553 if (S_ISDIR(inode->i_mode)) {
1554 add_flags = DCACHE_DIRECTORY_TYPE;
1555 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1556 if (unlikely(!inode->i_op->lookup))
1557 add_flags = DCACHE_AUTODIR_TYPE;
1559 inode->i_opflags |= IOP_LOOKUP;
1561 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1562 if (unlikely(inode->i_op->follow_link))
1563 add_flags = DCACHE_SYMLINK_TYPE;
1565 inode->i_opflags |= IOP_NOFOLLOW;
1568 if (unlikely(IS_AUTOMOUNT(inode)))
1569 add_flags |= DCACHE_NEED_AUTOMOUNT;
1573 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1575 unsigned add_flags = d_flags_for_inode(inode);
1577 spin_lock(&dentry->d_lock);
1578 __d_set_type(dentry, add_flags);
1580 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1581 dentry->d_inode = inode;
1582 dentry_rcuwalk_barrier(dentry);
1583 spin_unlock(&dentry->d_lock);
1584 fsnotify_d_instantiate(dentry, inode);
1588 * d_instantiate - fill in inode information for a dentry
1589 * @entry: dentry to complete
1590 * @inode: inode to attach to this dentry
1592 * Fill in inode information in the entry.
1594 * This turns negative dentries into productive full members
1597 * NOTE! This assumes that the inode count has been incremented
1598 * (or otherwise set) by the caller to indicate that it is now
1599 * in use by the dcache.
1602 void d_instantiate(struct dentry *entry, struct inode * inode)
1604 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1606 spin_lock(&inode->i_lock);
1607 __d_instantiate(entry, inode);
1609 spin_unlock(&inode->i_lock);
1610 security_d_instantiate(entry, inode);
1612 EXPORT_SYMBOL(d_instantiate);
1615 * d_instantiate_unique - instantiate a non-aliased dentry
1616 * @entry: dentry to instantiate
1617 * @inode: inode to attach to this dentry
1619 * Fill in inode information in the entry. On success, it returns NULL.
1620 * If an unhashed alias of "entry" already exists, then we return the
1621 * aliased dentry instead and drop one reference to inode.
1623 * Note that in order to avoid conflicts with rename() etc, the caller
1624 * had better be holding the parent directory semaphore.
1626 * This also assumes that the inode count has been incremented
1627 * (or otherwise set) by the caller to indicate that it is now
1628 * in use by the dcache.
1630 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1631 struct inode *inode)
1633 struct dentry *alias;
1634 int len = entry->d_name.len;
1635 const char *name = entry->d_name.name;
1636 unsigned int hash = entry->d_name.hash;
1639 __d_instantiate(entry, NULL);
1643 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1645 * Don't need alias->d_lock here, because aliases with
1646 * d_parent == entry->d_parent are not subject to name or
1647 * parent changes, because the parent inode i_mutex is held.
1649 if (alias->d_name.hash != hash)
1651 if (alias->d_parent != entry->d_parent)
1653 if (alias->d_name.len != len)
1655 if (dentry_cmp(alias, name, len))
1661 __d_instantiate(entry, inode);
1665 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1667 struct dentry *result;
1669 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1672 spin_lock(&inode->i_lock);
1673 result = __d_instantiate_unique(entry, inode);
1675 spin_unlock(&inode->i_lock);
1678 security_d_instantiate(entry, inode);
1682 BUG_ON(!d_unhashed(result));
1687 EXPORT_SYMBOL(d_instantiate_unique);
1690 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1691 * @entry: dentry to complete
1692 * @inode: inode to attach to this dentry
1694 * Fill in inode information in the entry. If a directory alias is found, then
1695 * return an error (and drop inode). Together with d_materialise_unique() this
1696 * guarantees that a directory inode may never have more than one alias.
1698 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1700 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1702 spin_lock(&inode->i_lock);
1703 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1704 spin_unlock(&inode->i_lock);
1708 __d_instantiate(entry, inode);
1709 spin_unlock(&inode->i_lock);
1710 security_d_instantiate(entry, inode);
1714 EXPORT_SYMBOL(d_instantiate_no_diralias);
1716 struct dentry *d_make_root(struct inode *root_inode)
1718 struct dentry *res = NULL;
1721 static const struct qstr name = QSTR_INIT("/", 1);
1723 res = __d_alloc(root_inode->i_sb, &name);
1725 d_instantiate(res, root_inode);
1731 EXPORT_SYMBOL(d_make_root);
1733 static struct dentry * __d_find_any_alias(struct inode *inode)
1735 struct dentry *alias;
1737 if (hlist_empty(&inode->i_dentry))
1739 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1745 * d_find_any_alias - find any alias for a given inode
1746 * @inode: inode to find an alias for
1748 * If any aliases exist for the given inode, take and return a
1749 * reference for one of them. If no aliases exist, return %NULL.
1751 struct dentry *d_find_any_alias(struct inode *inode)
1755 spin_lock(&inode->i_lock);
1756 de = __d_find_any_alias(inode);
1757 spin_unlock(&inode->i_lock);
1760 EXPORT_SYMBOL(d_find_any_alias);
1762 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1764 static const struct qstr anonstring = QSTR_INIT("/", 1);
1770 return ERR_PTR(-ESTALE);
1772 return ERR_CAST(inode);
1774 res = d_find_any_alias(inode);
1778 tmp = __d_alloc(inode->i_sb, &anonstring);
1780 res = ERR_PTR(-ENOMEM);
1784 spin_lock(&inode->i_lock);
1785 res = __d_find_any_alias(inode);
1787 spin_unlock(&inode->i_lock);
1792 /* attach a disconnected dentry */
1793 add_flags = d_flags_for_inode(inode);
1796 add_flags |= DCACHE_DISCONNECTED;
1798 spin_lock(&tmp->d_lock);
1799 tmp->d_inode = inode;
1800 tmp->d_flags |= add_flags;
1801 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1802 hlist_bl_lock(&tmp->d_sb->s_anon);
1803 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1804 hlist_bl_unlock(&tmp->d_sb->s_anon);
1805 spin_unlock(&tmp->d_lock);
1806 spin_unlock(&inode->i_lock);
1807 security_d_instantiate(tmp, inode);
1812 if (res && !IS_ERR(res))
1813 security_d_instantiate(res, inode);
1819 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1820 * @inode: inode to allocate the dentry for
1822 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1823 * similar open by handle operations. The returned dentry may be anonymous,
1824 * or may have a full name (if the inode was already in the cache).
1826 * When called on a directory inode, we must ensure that the inode only ever
1827 * has one dentry. If a dentry is found, that is returned instead of
1828 * allocating a new one.
1830 * On successful return, the reference to the inode has been transferred
1831 * to the dentry. In case of an error the reference on the inode is released.
1832 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1833 * be passed in and the error will be propagated to the return value,
1834 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1836 struct dentry *d_obtain_alias(struct inode *inode)
1838 return __d_obtain_alias(inode, 1);
1840 EXPORT_SYMBOL(d_obtain_alias);
1843 * d_obtain_root - find or allocate a dentry for a given inode
1844 * @inode: inode to allocate the dentry for
1846 * Obtain an IS_ROOT dentry for the root of a filesystem.
1848 * We must ensure that directory inodes only ever have one dentry. If a
1849 * dentry is found, that is returned instead of allocating a new one.
1851 * On successful return, the reference to the inode has been transferred
1852 * to the dentry. In case of an error the reference on the inode is
1853 * released. A %NULL or IS_ERR inode may be passed in and will be the
1854 * error will be propagate to the return value, with a %NULL @inode
1855 * replaced by ERR_PTR(-ESTALE).
1857 struct dentry *d_obtain_root(struct inode *inode)
1859 return __d_obtain_alias(inode, 0);
1861 EXPORT_SYMBOL(d_obtain_root);
1864 * d_add_ci - lookup or allocate new dentry with case-exact name
1865 * @inode: the inode case-insensitive lookup has found
1866 * @dentry: the negative dentry that was passed to the parent's lookup func
1867 * @name: the case-exact name to be associated with the returned dentry
1869 * This is to avoid filling the dcache with case-insensitive names to the
1870 * same inode, only the actual correct case is stored in the dcache for
1871 * case-insensitive filesystems.
1873 * For a case-insensitive lookup match and if the the case-exact dentry
1874 * already exists in in the dcache, use it and return it.
1876 * If no entry exists with the exact case name, allocate new dentry with
1877 * the exact case, and return the spliced entry.
1879 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1882 struct dentry *found;
1886 * First check if a dentry matching the name already exists,
1887 * if not go ahead and create it now.
1889 found = d_hash_and_lookup(dentry->d_parent, name);
1891 new = d_alloc(dentry->d_parent, name);
1893 found = ERR_PTR(-ENOMEM);
1895 found = d_splice_alias(inode, new);
1906 EXPORT_SYMBOL(d_add_ci);
1909 * Do the slow-case of the dentry name compare.
1911 * Unlike the dentry_cmp() function, we need to atomically
1912 * load the name and length information, so that the
1913 * filesystem can rely on them, and can use the 'name' and
1914 * 'len' information without worrying about walking off the
1915 * end of memory etc.
1917 * Thus the read_seqcount_retry() and the "duplicate" info
1918 * in arguments (the low-level filesystem should not look
1919 * at the dentry inode or name contents directly, since
1920 * rename can change them while we're in RCU mode).
1922 enum slow_d_compare {
1928 static noinline enum slow_d_compare slow_dentry_cmp(
1929 const struct dentry *parent,
1930 struct dentry *dentry,
1932 const struct qstr *name)
1934 int tlen = dentry->d_name.len;
1935 const char *tname = dentry->d_name.name;
1937 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1939 return D_COMP_SEQRETRY;
1941 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1942 return D_COMP_NOMATCH;
1947 * __d_lookup_rcu - search for a dentry (racy, store-free)
1948 * @parent: parent dentry
1949 * @name: qstr of name we wish to find
1950 * @seqp: returns d_seq value at the point where the dentry was found
1951 * Returns: dentry, or NULL
1953 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1954 * resolution (store-free path walking) design described in
1955 * Documentation/filesystems/path-lookup.txt.
1957 * This is not to be used outside core vfs.
1959 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1960 * held, and rcu_read_lock held. The returned dentry must not be stored into
1961 * without taking d_lock and checking d_seq sequence count against @seq
1964 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1967 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1968 * the returned dentry, so long as its parent's seqlock is checked after the
1969 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1970 * is formed, giving integrity down the path walk.
1972 * NOTE! The caller *has* to check the resulting dentry against the sequence
1973 * number we've returned before using any of the resulting dentry state!
1975 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1976 const struct qstr *name,
1979 u64 hashlen = name->hash_len;
1980 const unsigned char *str = name->name;
1981 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1982 struct hlist_bl_node *node;
1983 struct dentry *dentry;
1986 * Note: There is significant duplication with __d_lookup_rcu which is
1987 * required to prevent single threaded performance regressions
1988 * especially on architectures where smp_rmb (in seqcounts) are costly.
1989 * Keep the two functions in sync.
1993 * The hash list is protected using RCU.
1995 * Carefully use d_seq when comparing a candidate dentry, to avoid
1996 * races with d_move().
1998 * It is possible that concurrent renames can mess up our list
1999 * walk here and result in missing our dentry, resulting in the
2000 * false-negative result. d_lookup() protects against concurrent
2001 * renames using rename_lock seqlock.
2003 * See Documentation/filesystems/path-lookup.txt for more details.
2005 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2010 * The dentry sequence count protects us from concurrent
2011 * renames, and thus protects parent and name fields.
2013 * The caller must perform a seqcount check in order
2014 * to do anything useful with the returned dentry.
2016 * NOTE! We do a "raw" seqcount_begin here. That means that
2017 * we don't wait for the sequence count to stabilize if it
2018 * is in the middle of a sequence change. If we do the slow
2019 * dentry compare, we will do seqretries until it is stable,
2020 * and if we end up with a successful lookup, we actually
2021 * want to exit RCU lookup anyway.
2023 seq = raw_seqcount_begin(&dentry->d_seq);
2024 if (dentry->d_parent != parent)
2026 if (d_unhashed(dentry))
2029 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2030 if (dentry->d_name.hash != hashlen_hash(hashlen))
2033 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2036 case D_COMP_NOMATCH:
2043 if (dentry->d_name.hash_len != hashlen)
2046 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2053 * d_lookup - search for a dentry
2054 * @parent: parent dentry
2055 * @name: qstr of name we wish to find
2056 * Returns: dentry, or NULL
2058 * d_lookup searches the children of the parent dentry for the name in
2059 * question. If the dentry is found its reference count is incremented and the
2060 * dentry is returned. The caller must use dput to free the entry when it has
2061 * finished using it. %NULL is returned if the dentry does not exist.
2063 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2065 struct dentry *dentry;
2069 seq = read_seqbegin(&rename_lock);
2070 dentry = __d_lookup(parent, name);
2073 } while (read_seqretry(&rename_lock, seq));
2076 EXPORT_SYMBOL(d_lookup);
2079 * __d_lookup - search for a dentry (racy)
2080 * @parent: parent dentry
2081 * @name: qstr of name we wish to find
2082 * Returns: dentry, or NULL
2084 * __d_lookup is like d_lookup, however it may (rarely) return a
2085 * false-negative result due to unrelated rename activity.
2087 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2088 * however it must be used carefully, eg. with a following d_lookup in
2089 * the case of failure.
2091 * __d_lookup callers must be commented.
2093 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2095 unsigned int len = name->len;
2096 unsigned int hash = name->hash;
2097 const unsigned char *str = name->name;
2098 struct hlist_bl_head *b = d_hash(parent, hash);
2099 struct hlist_bl_node *node;
2100 struct dentry *found = NULL;
2101 struct dentry *dentry;
2104 * Note: There is significant duplication with __d_lookup_rcu which is
2105 * required to prevent single threaded performance regressions
2106 * especially on architectures where smp_rmb (in seqcounts) are costly.
2107 * Keep the two functions in sync.
2111 * The hash list is protected using RCU.
2113 * Take d_lock when comparing a candidate dentry, to avoid races
2116 * It is possible that concurrent renames can mess up our list
2117 * walk here and result in missing our dentry, resulting in the
2118 * false-negative result. d_lookup() protects against concurrent
2119 * renames using rename_lock seqlock.
2121 * See Documentation/filesystems/path-lookup.txt for more details.
2125 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2127 if (dentry->d_name.hash != hash)
2130 spin_lock(&dentry->d_lock);
2131 if (dentry->d_parent != parent)
2133 if (d_unhashed(dentry))
2137 * It is safe to compare names since d_move() cannot
2138 * change the qstr (protected by d_lock).
2140 if (parent->d_flags & DCACHE_OP_COMPARE) {
2141 int tlen = dentry->d_name.len;
2142 const char *tname = dentry->d_name.name;
2143 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2146 if (dentry->d_name.len != len)
2148 if (dentry_cmp(dentry, str, len))
2152 dentry->d_lockref.count++;
2154 spin_unlock(&dentry->d_lock);
2157 spin_unlock(&dentry->d_lock);
2165 * d_hash_and_lookup - hash the qstr then search for a dentry
2166 * @dir: Directory to search in
2167 * @name: qstr of name we wish to find
2169 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2171 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2174 * Check for a fs-specific hash function. Note that we must
2175 * calculate the standard hash first, as the d_op->d_hash()
2176 * routine may choose to leave the hash value unchanged.
2178 name->hash = full_name_hash(name->name, name->len);
2179 if (dir->d_flags & DCACHE_OP_HASH) {
2180 int err = dir->d_op->d_hash(dir, name);
2181 if (unlikely(err < 0))
2182 return ERR_PTR(err);
2184 return d_lookup(dir, name);
2186 EXPORT_SYMBOL(d_hash_and_lookup);
2189 * d_validate - verify dentry provided from insecure source (deprecated)
2190 * @dentry: The dentry alleged to be valid child of @dparent
2191 * @dparent: The parent dentry (known to be valid)
2193 * An insecure source has sent us a dentry, here we verify it and dget() it.
2194 * This is used by ncpfs in its readdir implementation.
2195 * Zero is returned in the dentry is invalid.
2197 * This function is slow for big directories, and deprecated, do not use it.
2199 int d_validate(struct dentry *dentry, struct dentry *dparent)
2201 struct dentry *child;
2203 spin_lock(&dparent->d_lock);
2204 list_for_each_entry(child, &dparent->d_subdirs, d_child) {
2205 if (dentry == child) {
2206 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2207 __dget_dlock(dentry);
2208 spin_unlock(&dentry->d_lock);
2209 spin_unlock(&dparent->d_lock);
2213 spin_unlock(&dparent->d_lock);
2217 EXPORT_SYMBOL(d_validate);
2220 * When a file is deleted, we have two options:
2221 * - turn this dentry into a negative dentry
2222 * - unhash this dentry and free it.
2224 * Usually, we want to just turn this into
2225 * a negative dentry, but if anybody else is
2226 * currently using the dentry or the inode
2227 * we can't do that and we fall back on removing
2228 * it from the hash queues and waiting for
2229 * it to be deleted later when it has no users
2233 * d_delete - delete a dentry
2234 * @dentry: The dentry to delete
2236 * Turn the dentry into a negative dentry if possible, otherwise
2237 * remove it from the hash queues so it can be deleted later
2240 void d_delete(struct dentry * dentry)
2242 struct inode *inode;
2245 * Are we the only user?
2248 spin_lock(&dentry->d_lock);
2249 inode = dentry->d_inode;
2250 isdir = S_ISDIR(inode->i_mode);
2251 if (dentry->d_lockref.count == 1) {
2252 if (!spin_trylock(&inode->i_lock)) {
2253 spin_unlock(&dentry->d_lock);
2257 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2258 dentry_unlink_inode(dentry);
2259 fsnotify_nameremove(dentry, isdir);
2263 if (!d_unhashed(dentry))
2266 spin_unlock(&dentry->d_lock);
2268 fsnotify_nameremove(dentry, isdir);
2270 EXPORT_SYMBOL(d_delete);
2272 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2274 BUG_ON(!d_unhashed(entry));
2276 entry->d_flags |= DCACHE_RCUACCESS;
2277 hlist_bl_add_head_rcu(&entry->d_hash, b);
2281 static void _d_rehash(struct dentry * entry)
2283 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2287 * d_rehash - add an entry back to the hash
2288 * @entry: dentry to add to the hash
2290 * Adds a dentry to the hash according to its name.
2293 void d_rehash(struct dentry * entry)
2295 spin_lock(&entry->d_lock);
2297 spin_unlock(&entry->d_lock);
2299 EXPORT_SYMBOL(d_rehash);
2302 * dentry_update_name_case - update case insensitive dentry with a new name
2303 * @dentry: dentry to be updated
2306 * Update a case insensitive dentry with new case of name.
2308 * dentry must have been returned by d_lookup with name @name. Old and new
2309 * name lengths must match (ie. no d_compare which allows mismatched name
2312 * Parent inode i_mutex must be held over d_lookup and into this call (to
2313 * keep renames and concurrent inserts, and readdir(2) away).
2315 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2317 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2318 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2320 spin_lock(&dentry->d_lock);
2321 write_seqcount_begin(&dentry->d_seq);
2322 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2323 write_seqcount_end(&dentry->d_seq);
2324 spin_unlock(&dentry->d_lock);
2326 EXPORT_SYMBOL(dentry_update_name_case);
2328 static void swap_names(struct dentry *dentry, struct dentry *target)
2330 if (unlikely(dname_external(target))) {
2331 if (unlikely(dname_external(dentry))) {
2333 * Both external: swap the pointers
2335 swap(target->d_name.name, dentry->d_name.name);
2338 * dentry:internal, target:external. Steal target's
2339 * storage and make target internal.
2341 memcpy(target->d_iname, dentry->d_name.name,
2342 dentry->d_name.len + 1);
2343 dentry->d_name.name = target->d_name.name;
2344 target->d_name.name = target->d_iname;
2347 if (unlikely(dname_external(dentry))) {
2349 * dentry:external, target:internal. Give dentry's
2350 * storage to target and make dentry internal
2352 memcpy(dentry->d_iname, target->d_name.name,
2353 target->d_name.len + 1);
2354 target->d_name.name = dentry->d_name.name;
2355 dentry->d_name.name = dentry->d_iname;
2358 * Both are internal.
2361 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2362 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2363 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2364 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2365 swap(((long *) &dentry->d_iname)[i],
2366 ((long *) &target->d_iname)[i]);
2370 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2373 static void copy_name(struct dentry *dentry, struct dentry *target)
2375 struct external_name *old_name = NULL;
2376 if (unlikely(dname_external(dentry)))
2377 old_name = external_name(dentry);
2378 if (unlikely(dname_external(target))) {
2379 atomic_inc(&external_name(target)->u.count);
2380 dentry->d_name = target->d_name;
2382 memcpy(dentry->d_iname, target->d_name.name,
2383 target->d_name.len + 1);
2384 dentry->d_name.name = dentry->d_iname;
2385 dentry->d_name.hash_len = target->d_name.hash_len;
2387 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2388 kfree_rcu(old_name, u.head);
2391 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2394 * XXXX: do we really need to take target->d_lock?
2396 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2397 spin_lock(&target->d_parent->d_lock);
2399 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2400 spin_lock(&dentry->d_parent->d_lock);
2401 spin_lock_nested(&target->d_parent->d_lock,
2402 DENTRY_D_LOCK_NESTED);
2404 spin_lock(&target->d_parent->d_lock);
2405 spin_lock_nested(&dentry->d_parent->d_lock,
2406 DENTRY_D_LOCK_NESTED);
2409 if (target < dentry) {
2410 spin_lock_nested(&target->d_lock, 2);
2411 spin_lock_nested(&dentry->d_lock, 3);
2413 spin_lock_nested(&dentry->d_lock, 2);
2414 spin_lock_nested(&target->d_lock, 3);
2418 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2420 if (target->d_parent != dentry->d_parent)
2421 spin_unlock(&dentry->d_parent->d_lock);
2422 if (target->d_parent != target)
2423 spin_unlock(&target->d_parent->d_lock);
2424 spin_unlock(&target->d_lock);
2425 spin_unlock(&dentry->d_lock);
2429 * When switching names, the actual string doesn't strictly have to
2430 * be preserved in the target - because we're dropping the target
2431 * anyway. As such, we can just do a simple memcpy() to copy over
2432 * the new name before we switch, unless we are going to rehash
2433 * it. Note that if we *do* unhash the target, we are not allowed
2434 * to rehash it without giving it a new name/hash key - whether
2435 * we swap or overwrite the names here, resulting name won't match
2436 * the reality in filesystem; it's only there for d_path() purposes.
2437 * Note that all of this is happening under rename_lock, so the
2438 * any hash lookup seeing it in the middle of manipulations will
2439 * be discarded anyway. So we do not care what happens to the hash
2443 * __d_move - move a dentry
2444 * @dentry: entry to move
2445 * @target: new dentry
2446 * @exchange: exchange the two dentries
2448 * Update the dcache to reflect the move of a file name. Negative
2449 * dcache entries should not be moved in this way. Caller must hold
2450 * rename_lock, the i_mutex of the source and target directories,
2451 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2453 static void __d_move(struct dentry *dentry, struct dentry *target,
2456 if (!dentry->d_inode)
2457 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2459 BUG_ON(d_ancestor(dentry, target));
2460 BUG_ON(d_ancestor(target, dentry));
2462 dentry_lock_for_move(dentry, target);
2464 write_seqcount_begin(&dentry->d_seq);
2465 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2467 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2470 * Move the dentry to the target hash queue. Don't bother checking
2471 * for the same hash queue because of how unlikely it is.
2474 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2477 * Unhash the target (d_delete() is not usable here). If exchanging
2478 * the two dentries, then rehash onto the other's hash queue.
2483 d_hash(dentry->d_parent, dentry->d_name.hash));
2486 /* Switch the names.. */
2488 swap_names(dentry, target);
2490 copy_name(dentry, target);
2492 /* ... and switch them in the tree */
2493 if (IS_ROOT(dentry)) {
2494 /* splicing a tree */
2495 dentry->d_parent = target->d_parent;
2496 target->d_parent = target;
2497 list_del_init(&target->d_child);
2498 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2500 /* swapping two dentries */
2501 swap(dentry->d_parent, target->d_parent);
2502 list_move(&target->d_child, &target->d_parent->d_subdirs);
2503 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2505 fsnotify_d_move(target);
2506 fsnotify_d_move(dentry);
2509 write_seqcount_end(&target->d_seq);
2510 write_seqcount_end(&dentry->d_seq);
2512 dentry_unlock_for_move(dentry, target);
2516 * d_move - move a dentry
2517 * @dentry: entry to move
2518 * @target: new dentry
2520 * Update the dcache to reflect the move of a file name. Negative
2521 * dcache entries should not be moved in this way. See the locking
2522 * requirements for __d_move.
2524 void d_move(struct dentry *dentry, struct dentry *target)
2526 write_seqlock(&rename_lock);
2527 __d_move(dentry, target, false);
2528 write_sequnlock(&rename_lock);
2530 EXPORT_SYMBOL(d_move);
2533 * d_exchange - exchange two dentries
2534 * @dentry1: first dentry
2535 * @dentry2: second dentry
2537 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2539 write_seqlock(&rename_lock);
2541 WARN_ON(!dentry1->d_inode);
2542 WARN_ON(!dentry2->d_inode);
2543 WARN_ON(IS_ROOT(dentry1));
2544 WARN_ON(IS_ROOT(dentry2));
2546 __d_move(dentry1, dentry2, true);
2548 write_sequnlock(&rename_lock);
2552 * d_ancestor - search for an ancestor
2553 * @p1: ancestor dentry
2556 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2557 * an ancestor of p2, else NULL.
2559 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2563 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2564 if (p->d_parent == p1)
2571 * This helper attempts to cope with remotely renamed directories
2573 * It assumes that the caller is already holding
2574 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2576 * Note: If ever the locking in lock_rename() changes, then please
2577 * remember to update this too...
2579 static int __d_unalias(struct inode *inode,
2580 struct dentry *dentry, struct dentry *alias)
2582 struct mutex *m1 = NULL, *m2 = NULL;
2585 /* If alias and dentry share a parent, then no extra locks required */
2586 if (alias->d_parent == dentry->d_parent)
2589 /* See lock_rename() */
2590 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2592 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2593 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2595 m2 = &alias->d_parent->d_inode->i_mutex;
2597 __d_move(alias, dentry, false);
2600 spin_unlock(&inode->i_lock);
2609 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2610 * @inode: the inode which may have a disconnected dentry
2611 * @dentry: a negative dentry which we want to point to the inode.
2613 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2614 * place of the given dentry and return it, else simply d_add the inode
2615 * to the dentry and return NULL.
2617 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2618 * we should error out: directories can't have multiple aliases.
2620 * This is needed in the lookup routine of any filesystem that is exportable
2621 * (via knfsd) so that we can build dcache paths to directories effectively.
2623 * If a dentry was found and moved, then it is returned. Otherwise NULL
2624 * is returned. This matches the expected return value of ->lookup.
2626 * Cluster filesystems may call this function with a negative, hashed dentry.
2627 * In that case, we know that the inode will be a regular file, and also this
2628 * will only occur during atomic_open. So we need to check for the dentry
2629 * being already hashed only in the final case.
2631 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2634 return ERR_CAST(inode);
2636 BUG_ON(!d_unhashed(dentry));
2639 __d_instantiate(dentry, NULL);
2642 spin_lock(&inode->i_lock);
2643 if (S_ISDIR(inode->i_mode)) {
2644 struct dentry *new = __d_find_any_alias(inode);
2645 if (unlikely(new)) {
2646 write_seqlock(&rename_lock);
2647 if (unlikely(d_ancestor(new, dentry))) {
2648 write_sequnlock(&rename_lock);
2649 spin_unlock(&inode->i_lock);
2651 new = ERR_PTR(-ELOOP);
2652 pr_warn_ratelimited(
2653 "VFS: Lookup of '%s' in %s %s"
2654 " would have caused loop\n",
2655 dentry->d_name.name,
2656 inode->i_sb->s_type->name,
2658 } else if (!IS_ROOT(new)) {
2659 int err = __d_unalias(inode, dentry, new);
2660 write_sequnlock(&rename_lock);
2666 __d_move(new, dentry, false);
2667 write_sequnlock(&rename_lock);
2668 spin_unlock(&inode->i_lock);
2669 security_d_instantiate(new, inode);
2675 /* already taking inode->i_lock, so d_add() by hand */
2676 __d_instantiate(dentry, inode);
2677 spin_unlock(&inode->i_lock);
2679 security_d_instantiate(dentry, inode);
2683 EXPORT_SYMBOL(d_splice_alias);
2685 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2689 return -ENAMETOOLONG;
2691 memcpy(*buffer, str, namelen);
2696 * prepend_name - prepend a pathname in front of current buffer pointer
2697 * @buffer: buffer pointer
2698 * @buflen: allocated length of the buffer
2699 * @name: name string and length qstr structure
2701 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2702 * make sure that either the old or the new name pointer and length are
2703 * fetched. However, there may be mismatch between length and pointer.
2704 * The length cannot be trusted, we need to copy it byte-by-byte until
2705 * the length is reached or a null byte is found. It also prepends "/" at
2706 * the beginning of the name. The sequence number check at the caller will
2707 * retry it again when a d_move() does happen. So any garbage in the buffer
2708 * due to mismatched pointer and length will be discarded.
2710 * Data dependency barrier is needed to make sure that we see that terminating
2711 * NUL. Alpha strikes again, film at 11...
2713 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2715 const char *dname = ACCESS_ONCE(name->name);
2716 u32 dlen = ACCESS_ONCE(name->len);
2719 smp_read_barrier_depends();
2721 *buflen -= dlen + 1;
2723 return -ENAMETOOLONG;
2724 p = *buffer -= dlen + 1;
2736 * prepend_path - Prepend path string to a buffer
2737 * @path: the dentry/vfsmount to report
2738 * @root: root vfsmnt/dentry
2739 * @buffer: pointer to the end of the buffer
2740 * @buflen: pointer to buffer length
2742 * The function will first try to write out the pathname without taking any
2743 * lock other than the RCU read lock to make sure that dentries won't go away.
2744 * It only checks the sequence number of the global rename_lock as any change
2745 * in the dentry's d_seq will be preceded by changes in the rename_lock
2746 * sequence number. If the sequence number had been changed, it will restart
2747 * the whole pathname back-tracing sequence again by taking the rename_lock.
2748 * In this case, there is no need to take the RCU read lock as the recursive
2749 * parent pointer references will keep the dentry chain alive as long as no
2750 * rename operation is performed.
2752 static int prepend_path(const struct path *path,
2753 const struct path *root,
2754 char **buffer, int *buflen)
2756 struct dentry *dentry;
2757 struct vfsmount *vfsmnt;
2760 unsigned seq, m_seq = 0;
2766 read_seqbegin_or_lock(&mount_lock, &m_seq);
2773 dentry = path->dentry;
2775 mnt = real_mount(vfsmnt);
2776 read_seqbegin_or_lock(&rename_lock, &seq);
2777 while (dentry != root->dentry || vfsmnt != root->mnt) {
2778 struct dentry * parent;
2780 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2781 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2783 if (mnt != parent) {
2784 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2790 * Filesystems needing to implement special "root names"
2791 * should do so with ->d_dname()
2793 if (IS_ROOT(dentry) &&
2794 (dentry->d_name.len != 1 ||
2795 dentry->d_name.name[0] != '/')) {
2796 WARN(1, "Root dentry has weird name <%.*s>\n",
2797 (int) dentry->d_name.len,
2798 dentry->d_name.name);
2801 error = is_mounted(vfsmnt) ? 1 : 2;
2804 parent = dentry->d_parent;
2806 error = prepend_name(&bptr, &blen, &dentry->d_name);
2814 if (need_seqretry(&rename_lock, seq)) {
2818 done_seqretry(&rename_lock, seq);
2822 if (need_seqretry(&mount_lock, m_seq)) {
2826 done_seqretry(&mount_lock, m_seq);
2828 if (error >= 0 && bptr == *buffer) {
2830 error = -ENAMETOOLONG;
2840 * __d_path - return the path of a dentry
2841 * @path: the dentry/vfsmount to report
2842 * @root: root vfsmnt/dentry
2843 * @buf: buffer to return value in
2844 * @buflen: buffer length
2846 * Convert a dentry into an ASCII path name.
2848 * Returns a pointer into the buffer or an error code if the
2849 * path was too long.
2851 * "buflen" should be positive.
2853 * If the path is not reachable from the supplied root, return %NULL.
2855 char *__d_path(const struct path *path,
2856 const struct path *root,
2857 char *buf, int buflen)
2859 char *res = buf + buflen;
2862 prepend(&res, &buflen, "\0", 1);
2863 error = prepend_path(path, root, &res, &buflen);
2866 return ERR_PTR(error);
2872 char *d_absolute_path(const struct path *path,
2873 char *buf, int buflen)
2875 struct path root = {};
2876 char *res = buf + buflen;
2879 prepend(&res, &buflen, "\0", 1);
2880 error = prepend_path(path, &root, &res, &buflen);
2885 return ERR_PTR(error);
2890 * same as __d_path but appends "(deleted)" for unlinked files.
2892 static int path_with_deleted(const struct path *path,
2893 const struct path *root,
2894 char **buf, int *buflen)
2896 prepend(buf, buflen, "\0", 1);
2897 if (d_unlinked(path->dentry)) {
2898 int error = prepend(buf, buflen, " (deleted)", 10);
2903 return prepend_path(path, root, buf, buflen);
2906 static int prepend_unreachable(char **buffer, int *buflen)
2908 return prepend(buffer, buflen, "(unreachable)", 13);
2911 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2916 seq = read_seqcount_begin(&fs->seq);
2918 } while (read_seqcount_retry(&fs->seq, seq));
2922 * d_path - return the path of a dentry
2923 * @path: path to report
2924 * @buf: buffer to return value in
2925 * @buflen: buffer length
2927 * Convert a dentry into an ASCII path name. If the entry has been deleted
2928 * the string " (deleted)" is appended. Note that this is ambiguous.
2930 * Returns a pointer into the buffer or an error code if the path was
2931 * too long. Note: Callers should use the returned pointer, not the passed
2932 * in buffer, to use the name! The implementation often starts at an offset
2933 * into the buffer, and may leave 0 bytes at the start.
2935 * "buflen" should be positive.
2937 char *d_path(const struct path *path, char *buf, int buflen)
2939 char *res = buf + buflen;
2944 * We have various synthetic filesystems that never get mounted. On
2945 * these filesystems dentries are never used for lookup purposes, and
2946 * thus don't need to be hashed. They also don't need a name until a
2947 * user wants to identify the object in /proc/pid/fd/. The little hack
2948 * below allows us to generate a name for these objects on demand:
2950 * Some pseudo inodes are mountable. When they are mounted
2951 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
2952 * and instead have d_path return the mounted path.
2954 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
2955 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
2956 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2959 get_fs_root_rcu(current->fs, &root);
2960 error = path_with_deleted(path, &root, &res, &buflen);
2964 res = ERR_PTR(error);
2967 EXPORT_SYMBOL(d_path);
2970 * Helper function for dentry_operations.d_dname() members
2972 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2973 const char *fmt, ...)
2979 va_start(args, fmt);
2980 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2983 if (sz > sizeof(temp) || sz > buflen)
2984 return ERR_PTR(-ENAMETOOLONG);
2986 buffer += buflen - sz;
2987 return memcpy(buffer, temp, sz);
2990 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2992 char *end = buffer + buflen;
2993 /* these dentries are never renamed, so d_lock is not needed */
2994 if (prepend(&end, &buflen, " (deleted)", 11) ||
2995 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
2996 prepend(&end, &buflen, "/", 1))
2997 end = ERR_PTR(-ENAMETOOLONG);
3000 EXPORT_SYMBOL(simple_dname);
3003 * Write full pathname from the root of the filesystem into the buffer.
3005 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3007 struct dentry *dentry;
3020 prepend(&end, &len, "\0", 1);
3024 read_seqbegin_or_lock(&rename_lock, &seq);
3025 while (!IS_ROOT(dentry)) {
3026 struct dentry *parent = dentry->d_parent;
3029 error = prepend_name(&end, &len, &dentry->d_name);
3038 if (need_seqretry(&rename_lock, seq)) {
3042 done_seqretry(&rename_lock, seq);
3047 return ERR_PTR(-ENAMETOOLONG);
3050 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3052 return __dentry_path(dentry, buf, buflen);
3054 EXPORT_SYMBOL(dentry_path_raw);
3056 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3061 if (d_unlinked(dentry)) {
3063 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3067 retval = __dentry_path(dentry, buf, buflen);
3068 if (!IS_ERR(retval) && p)
3069 *p = '/'; /* restore '/' overriden with '\0' */
3072 return ERR_PTR(-ENAMETOOLONG);
3075 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3081 seq = read_seqcount_begin(&fs->seq);
3084 } while (read_seqcount_retry(&fs->seq, seq));
3088 * NOTE! The user-level library version returns a
3089 * character pointer. The kernel system call just
3090 * returns the length of the buffer filled (which
3091 * includes the ending '\0' character), or a negative
3092 * error value. So libc would do something like
3094 * char *getcwd(char * buf, size_t size)
3098 * retval = sys_getcwd(buf, size);
3105 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3108 struct path pwd, root;
3109 char *page = __getname();
3115 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3118 if (!d_unlinked(pwd.dentry)) {
3120 char *cwd = page + PATH_MAX;
3121 int buflen = PATH_MAX;
3123 prepend(&cwd, &buflen, "\0", 1);
3124 error = prepend_path(&pwd, &root, &cwd, &buflen);
3130 /* Unreachable from current root */
3132 error = prepend_unreachable(&cwd, &buflen);
3138 len = PATH_MAX + page - cwd;
3141 if (copy_to_user(buf, cwd, len))
3154 * Test whether new_dentry is a subdirectory of old_dentry.
3156 * Trivially implemented using the dcache structure
3160 * is_subdir - is new dentry a subdirectory of old_dentry
3161 * @new_dentry: new dentry
3162 * @old_dentry: old dentry
3164 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3165 * Returns 0 otherwise.
3166 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3169 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3174 if (new_dentry == old_dentry)
3178 /* for restarting inner loop in case of seq retry */
3179 seq = read_seqbegin(&rename_lock);
3181 * Need rcu_readlock to protect against the d_parent trashing
3185 if (d_ancestor(old_dentry, new_dentry))
3190 } while (read_seqretry(&rename_lock, seq));
3195 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3197 struct dentry *root = data;
3198 if (dentry != root) {
3199 if (d_unhashed(dentry) || !dentry->d_inode)
3202 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3203 dentry->d_flags |= DCACHE_GENOCIDE;
3204 dentry->d_lockref.count--;
3207 return D_WALK_CONTINUE;
3210 void d_genocide(struct dentry *parent)
3212 d_walk(parent, parent, d_genocide_kill, NULL);
3215 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3217 inode_dec_link_count(inode);
3218 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3219 !hlist_unhashed(&dentry->d_u.d_alias) ||
3220 !d_unlinked(dentry));
3221 spin_lock(&dentry->d_parent->d_lock);
3222 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3223 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3224 (unsigned long long)inode->i_ino);
3225 spin_unlock(&dentry->d_lock);
3226 spin_unlock(&dentry->d_parent->d_lock);
3227 d_instantiate(dentry, inode);
3229 EXPORT_SYMBOL(d_tmpfile);
3231 static __initdata unsigned long dhash_entries;
3232 static int __init set_dhash_entries(char *str)
3236 dhash_entries = simple_strtoul(str, &str, 0);
3239 __setup("dhash_entries=", set_dhash_entries);
3241 static void __init dcache_init_early(void)
3245 /* If hashes are distributed across NUMA nodes, defer
3246 * hash allocation until vmalloc space is available.
3252 alloc_large_system_hash("Dentry cache",
3253 sizeof(struct hlist_bl_head),
3262 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3263 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3266 static void __init dcache_init(void)
3271 * A constructor could be added for stable state like the lists,
3272 * but it is probably not worth it because of the cache nature
3275 dentry_cache = KMEM_CACHE(dentry,
3276 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3278 /* Hash may have been set up in dcache_init_early */
3283 alloc_large_system_hash("Dentry cache",
3284 sizeof(struct hlist_bl_head),
3293 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3294 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3297 /* SLAB cache for __getname() consumers */
3298 struct kmem_cache *names_cachep __read_mostly;
3299 EXPORT_SYMBOL(names_cachep);
3301 EXPORT_SYMBOL(d_genocide);
3303 void __init vfs_caches_init_early(void)
3305 dcache_init_early();
3309 void __init vfs_caches_init(unsigned long mempages)
3311 unsigned long reserve;
3313 /* Base hash sizes on available memory, with a reserve equal to
3314 150% of current kernel size */
3316 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3317 mempages -= reserve;
3319 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3320 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3324 files_init(mempages);