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_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
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 hash = hash + (hash >> d_hash_shift);
110 return dentry_hashtable + (hash & d_hash_mask);
113 /* Statistics gathering. */
114 struct dentry_stat_t dentry_stat = {
118 static DEFINE_PER_CPU(long, nr_dentry);
119 static DEFINE_PER_CPU(long, nr_dentry_unused);
121 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
124 * Here we resort to our own counters instead of using generic per-cpu counters
125 * for consistency with what the vfs inode code does. We are expected to harvest
126 * better code and performance by having our own specialized counters.
128 * Please note that the loop is done over all possible CPUs, not over all online
129 * CPUs. The reason for this is that we don't want to play games with CPUs going
130 * on and off. If one of them goes off, we will just keep their counters.
132 * glommer: See cffbc8a for details, and if you ever intend to change this,
133 * please update all vfs counters to match.
135 static long get_nr_dentry(void)
139 for_each_possible_cpu(i)
140 sum += per_cpu(nr_dentry, i);
141 return sum < 0 ? 0 : sum;
144 static long get_nr_dentry_unused(void)
148 for_each_possible_cpu(i)
149 sum += per_cpu(nr_dentry_unused, i);
150 return sum < 0 ? 0 : sum;
153 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
154 size_t *lenp, loff_t *ppos)
156 dentry_stat.nr_dentry = get_nr_dentry();
157 dentry_stat.nr_unused = get_nr_dentry_unused();
158 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
163 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
164 * The strings are both count bytes long, and count is non-zero.
166 #ifdef CONFIG_DCACHE_WORD_ACCESS
168 #include <asm/word-at-a-time.h>
170 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
171 * aligned allocation for this particular component. We don't
172 * strictly need the load_unaligned_zeropad() safety, but it
173 * doesn't hurt either.
175 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
176 * need the careful unaligned handling.
178 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
180 unsigned long a,b,mask;
183 a = *(unsigned long *)cs;
184 b = load_unaligned_zeropad(ct);
185 if (tcount < sizeof(unsigned long))
187 if (unlikely(a != b))
189 cs += sizeof(unsigned long);
190 ct += sizeof(unsigned long);
191 tcount -= sizeof(unsigned long);
195 mask = bytemask_from_count(tcount);
196 return unlikely(!!((a ^ b) & mask));
201 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
215 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
217 const unsigned char *cs;
219 * Be careful about RCU walk racing with rename:
220 * use ACCESS_ONCE to fetch the name pointer.
222 * NOTE! Even if a rename will mean that the length
223 * was not loaded atomically, we don't care. The
224 * RCU walk will check the sequence count eventually,
225 * and catch it. And we won't overrun the buffer,
226 * because we're reading the name pointer atomically,
227 * and a dentry name is guaranteed to be properly
228 * terminated with a NUL byte.
230 * End result: even if 'len' is wrong, we'll exit
231 * early because the data cannot match (there can
232 * be no NUL in the ct/tcount data)
234 cs = ACCESS_ONCE(dentry->d_name.name);
235 smp_read_barrier_depends();
236 return dentry_string_cmp(cs, ct, tcount);
239 static void __d_free(struct rcu_head *head)
241 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
243 WARN_ON(!hlist_unhashed(&dentry->d_alias));
244 if (dname_external(dentry))
245 kfree(dentry->d_name.name);
246 kmem_cache_free(dentry_cache, dentry);
249 static void dentry_free(struct dentry *dentry)
251 /* if dentry was never visible to RCU, immediate free is OK */
252 if (!(dentry->d_flags & DCACHE_RCUACCESS))
253 __d_free(&dentry->d_u.d_rcu);
255 call_rcu(&dentry->d_u.d_rcu, __d_free);
259 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
260 * @dentry: the target dentry
261 * After this call, in-progress rcu-walk path lookup will fail. This
262 * should be called after unhashing, and after changing d_inode (if
263 * the dentry has not already been unhashed).
265 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
267 assert_spin_locked(&dentry->d_lock);
268 /* Go through a barrier */
269 write_seqcount_barrier(&dentry->d_seq);
273 * Release the dentry's inode, using the filesystem
274 * d_iput() operation if defined. Dentry has no refcount
277 static void dentry_iput(struct dentry * dentry)
278 __releases(dentry->d_lock)
279 __releases(dentry->d_inode->i_lock)
281 struct inode *inode = dentry->d_inode;
283 dentry->d_inode = NULL;
284 hlist_del_init(&dentry->d_alias);
285 spin_unlock(&dentry->d_lock);
286 spin_unlock(&inode->i_lock);
288 fsnotify_inoderemove(inode);
289 if (dentry->d_op && dentry->d_op->d_iput)
290 dentry->d_op->d_iput(dentry, inode);
294 spin_unlock(&dentry->d_lock);
299 * Release the dentry's inode, using the filesystem
300 * d_iput() operation if defined. dentry remains in-use.
302 static void dentry_unlink_inode(struct dentry * dentry)
303 __releases(dentry->d_lock)
304 __releases(dentry->d_inode->i_lock)
306 struct inode *inode = dentry->d_inode;
307 __d_clear_type(dentry);
308 dentry->d_inode = NULL;
309 hlist_del_init(&dentry->d_alias);
310 dentry_rcuwalk_barrier(dentry);
311 spin_unlock(&dentry->d_lock);
312 spin_unlock(&inode->i_lock);
314 fsnotify_inoderemove(inode);
315 if (dentry->d_op && dentry->d_op->d_iput)
316 dentry->d_op->d_iput(dentry, inode);
322 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
323 * is in use - which includes both the "real" per-superblock
324 * LRU list _and_ the DCACHE_SHRINK_LIST use.
326 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
327 * on the shrink list (ie not on the superblock LRU list).
329 * The per-cpu "nr_dentry_unused" counters are updated with
330 * the DCACHE_LRU_LIST bit.
332 * These helper functions make sure we always follow the
333 * rules. d_lock must be held by the caller.
335 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
336 static void d_lru_add(struct dentry *dentry)
338 D_FLAG_VERIFY(dentry, 0);
339 dentry->d_flags |= DCACHE_LRU_LIST;
340 this_cpu_inc(nr_dentry_unused);
341 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
344 static void d_lru_del(struct dentry *dentry)
346 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
347 dentry->d_flags &= ~DCACHE_LRU_LIST;
348 this_cpu_dec(nr_dentry_unused);
349 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
352 static void d_shrink_del(struct dentry *dentry)
354 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
355 list_del_init(&dentry->d_lru);
356 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
357 this_cpu_dec(nr_dentry_unused);
360 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
362 D_FLAG_VERIFY(dentry, 0);
363 list_add(&dentry->d_lru, list);
364 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
365 this_cpu_inc(nr_dentry_unused);
369 * These can only be called under the global LRU lock, ie during the
370 * callback for freeing the LRU list. "isolate" removes it from the
371 * LRU lists entirely, while shrink_move moves it to the indicated
374 static void d_lru_isolate(struct dentry *dentry)
376 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
377 dentry->d_flags &= ~DCACHE_LRU_LIST;
378 this_cpu_dec(nr_dentry_unused);
379 list_del_init(&dentry->d_lru);
382 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
384 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
385 dentry->d_flags |= DCACHE_SHRINK_LIST;
386 list_move_tail(&dentry->d_lru, list);
390 * dentry_lru_(add|del)_list) must be called with d_lock held.
392 static void dentry_lru_add(struct dentry *dentry)
394 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
399 * d_drop - drop a dentry
400 * @dentry: dentry to drop
402 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
403 * be found through a VFS lookup any more. Note that this is different from
404 * deleting the dentry - d_delete will try to mark the dentry negative if
405 * possible, giving a successful _negative_ lookup, while d_drop will
406 * just make the cache lookup fail.
408 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
409 * reason (NFS timeouts or autofs deletes).
411 * __d_drop requires dentry->d_lock.
413 void __d_drop(struct dentry *dentry)
415 if (!d_unhashed(dentry)) {
416 struct hlist_bl_head *b;
418 * Hashed dentries are normally on the dentry hashtable,
419 * with the exception of those newly allocated by
420 * d_obtain_alias, which are always IS_ROOT:
422 if (unlikely(IS_ROOT(dentry)))
423 b = &dentry->d_sb->s_anon;
425 b = d_hash(dentry->d_parent, dentry->d_name.hash);
428 __hlist_bl_del(&dentry->d_hash);
429 dentry->d_hash.pprev = NULL;
431 dentry_rcuwalk_barrier(dentry);
434 EXPORT_SYMBOL(__d_drop);
436 void d_drop(struct dentry *dentry)
438 spin_lock(&dentry->d_lock);
440 spin_unlock(&dentry->d_lock);
442 EXPORT_SYMBOL(d_drop);
445 * Finish off a dentry we've decided to kill.
446 * dentry->d_lock must be held, returns with it unlocked.
447 * If ref is non-zero, then decrement the refcount too.
448 * Returns dentry requiring refcount drop, or NULL if we're done.
450 static struct dentry *
451 dentry_kill(struct dentry *dentry, int unlock_on_failure)
452 __releases(dentry->d_lock)
455 struct dentry *parent = NULL;
456 bool can_free = true;
458 inode = dentry->d_inode;
459 if (inode && !spin_trylock(&inode->i_lock)) {
461 if (unlock_on_failure) {
462 spin_unlock(&dentry->d_lock);
465 return dentry; /* try again with same dentry */
467 if (!IS_ROOT(dentry))
468 parent = dentry->d_parent;
469 if (parent && !spin_trylock(&parent->d_lock)) {
471 spin_unlock(&inode->i_lock);
476 * The dentry is now unrecoverably dead to the world.
478 lockref_mark_dead(&dentry->d_lockref);
481 * inform the fs via d_prune that this dentry is about to be
482 * unhashed and destroyed.
484 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
485 dentry->d_op->d_prune(dentry);
487 if (dentry->d_flags & DCACHE_LRU_LIST) {
488 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
491 /* if it was on the hash then remove it */
493 list_del(&dentry->d_u.d_child);
495 * Inform d_walk() that we are no longer attached to the
498 dentry->d_flags |= DCACHE_DENTRY_KILLED;
500 spin_unlock(&parent->d_lock);
503 * dentry_iput drops the locks, at which point nobody (except
504 * transient RCU lookups) can reach this dentry.
506 BUG_ON((int)dentry->d_lockref.count > 0);
507 this_cpu_dec(nr_dentry);
508 if (dentry->d_op && dentry->d_op->d_release)
509 dentry->d_op->d_release(dentry);
511 spin_lock(&dentry->d_lock);
512 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
513 dentry->d_flags |= DCACHE_MAY_FREE;
516 spin_unlock(&dentry->d_lock);
518 if (likely(can_free))
526 * This is complicated by the fact that we do not want to put
527 * dentries that are no longer on any hash chain on the unused
528 * list: we'd much rather just get rid of them immediately.
530 * However, that implies that we have to traverse the dentry
531 * tree upwards to the parents which might _also_ now be
532 * scheduled for deletion (it may have been only waiting for
533 * its last child to go away).
535 * This tail recursion is done by hand as we don't want to depend
536 * on the compiler to always get this right (gcc generally doesn't).
537 * Real recursion would eat up our stack space.
541 * dput - release a dentry
542 * @dentry: dentry to release
544 * Release a dentry. This will drop the usage count and if appropriate
545 * call the dentry unlink method as well as removing it from the queues and
546 * releasing its resources. If the parent dentries were scheduled for release
547 * they too may now get deleted.
549 void dput(struct dentry *dentry)
551 if (unlikely(!dentry))
555 if (lockref_put_or_lock(&dentry->d_lockref))
558 /* Unreachable? Get rid of it */
559 if (unlikely(d_unhashed(dentry)))
562 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
563 if (dentry->d_op->d_delete(dentry))
567 if (!(dentry->d_flags & DCACHE_REFERENCED))
568 dentry->d_flags |= DCACHE_REFERENCED;
569 dentry_lru_add(dentry);
571 dentry->d_lockref.count--;
572 spin_unlock(&dentry->d_lock);
576 dentry = dentry_kill(dentry, 1);
583 * d_invalidate - invalidate a dentry
584 * @dentry: dentry to invalidate
586 * Try to invalidate the dentry if it turns out to be
587 * possible. If there are other dentries that can be
588 * reached through this one we can't delete it and we
589 * return -EBUSY. On success we return 0.
594 int d_invalidate(struct dentry * dentry)
597 * If it's already been dropped, return OK.
599 spin_lock(&dentry->d_lock);
600 if (d_unhashed(dentry)) {
601 spin_unlock(&dentry->d_lock);
605 * Check whether to do a partial shrink_dcache
606 * to get rid of unused child entries.
608 if (!list_empty(&dentry->d_subdirs)) {
609 spin_unlock(&dentry->d_lock);
610 shrink_dcache_parent(dentry);
611 spin_lock(&dentry->d_lock);
615 * Somebody else still using it?
617 * If it's a directory, we can't drop it
618 * for fear of somebody re-populating it
619 * with children (even though dropping it
620 * would make it unreachable from the root,
621 * we might still populate it if it was a
622 * working directory or similar).
623 * We also need to leave mountpoints alone,
626 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
627 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
628 spin_unlock(&dentry->d_lock);
634 spin_unlock(&dentry->d_lock);
637 EXPORT_SYMBOL(d_invalidate);
639 /* This must be called with d_lock held */
640 static inline void __dget_dlock(struct dentry *dentry)
642 dentry->d_lockref.count++;
645 static inline void __dget(struct dentry *dentry)
647 lockref_get(&dentry->d_lockref);
650 struct dentry *dget_parent(struct dentry *dentry)
656 * Do optimistic parent lookup without any
660 ret = ACCESS_ONCE(dentry->d_parent);
661 gotref = lockref_get_not_zero(&ret->d_lockref);
663 if (likely(gotref)) {
664 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
671 * Don't need rcu_dereference because we re-check it was correct under
675 ret = dentry->d_parent;
676 spin_lock(&ret->d_lock);
677 if (unlikely(ret != dentry->d_parent)) {
678 spin_unlock(&ret->d_lock);
683 BUG_ON(!ret->d_lockref.count);
684 ret->d_lockref.count++;
685 spin_unlock(&ret->d_lock);
688 EXPORT_SYMBOL(dget_parent);
691 * d_find_alias - grab a hashed alias of inode
692 * @inode: inode in question
693 * @want_discon: flag, used by d_splice_alias, to request
694 * that only a DISCONNECTED alias be returned.
696 * If inode has a hashed alias, or is a directory and has any alias,
697 * acquire the reference to alias and return it. Otherwise return NULL.
698 * Notice that if inode is a directory there can be only one alias and
699 * it can be unhashed only if it has no children, or if it is the root
702 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
703 * any other hashed alias over that one unless @want_discon is set,
704 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
706 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
708 struct dentry *alias, *discon_alias;
712 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
713 spin_lock(&alias->d_lock);
714 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
715 if (IS_ROOT(alias) &&
716 (alias->d_flags & DCACHE_DISCONNECTED)) {
717 discon_alias = alias;
718 } else if (!want_discon) {
720 spin_unlock(&alias->d_lock);
724 spin_unlock(&alias->d_lock);
727 alias = discon_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)) {
733 spin_unlock(&alias->d_lock);
737 spin_unlock(&alias->d_lock);
743 struct dentry *d_find_alias(struct inode *inode)
745 struct dentry *de = NULL;
747 if (!hlist_empty(&inode->i_dentry)) {
748 spin_lock(&inode->i_lock);
749 de = __d_find_alias(inode, 0);
750 spin_unlock(&inode->i_lock);
754 EXPORT_SYMBOL(d_find_alias);
757 * Try to kill dentries associated with this inode.
758 * WARNING: you must own a reference to inode.
760 void d_prune_aliases(struct inode *inode)
762 struct dentry *dentry;
764 spin_lock(&inode->i_lock);
765 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
766 spin_lock(&dentry->d_lock);
767 if (!dentry->d_lockref.count) {
769 * inform the fs via d_prune that this dentry
770 * is about to be unhashed and destroyed.
772 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
774 dentry->d_op->d_prune(dentry);
776 __dget_dlock(dentry);
778 spin_unlock(&dentry->d_lock);
779 spin_unlock(&inode->i_lock);
783 spin_unlock(&dentry->d_lock);
785 spin_unlock(&inode->i_lock);
787 EXPORT_SYMBOL(d_prune_aliases);
789 static void shrink_dentry_list(struct list_head *list)
791 struct dentry *dentry, *parent;
793 while (!list_empty(list)) {
794 dentry = list_entry(list->prev, struct dentry, d_lru);
795 spin_lock(&dentry->d_lock);
797 * The dispose list is isolated and dentries are not accounted
798 * to the LRU here, so we can simply remove it from the list
799 * here regardless of whether it is referenced or not.
801 d_shrink_del(dentry);
804 * We found an inuse dentry which was not removed from
805 * the LRU because of laziness during lookup. Do not free it.
807 if ((int)dentry->d_lockref.count > 0) {
808 spin_unlock(&dentry->d_lock);
813 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
814 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
815 spin_unlock(&dentry->d_lock);
821 parent = dentry_kill(dentry, 0);
823 * If dentry_kill returns NULL, we have nothing more to do.
828 if (unlikely(parent == dentry)) {
830 * trylocks have failed and d_lock has been held the
831 * whole time, so it could not have been added to any
832 * other lists. Just add it back to the shrink list.
834 d_shrink_add(dentry, list);
835 spin_unlock(&dentry->d_lock);
839 * We need to prune ancestors too. This is necessary to prevent
840 * quadratic behavior of shrink_dcache_parent(), but is also
841 * expected to be beneficial in reducing dentry cache
845 while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
846 dentry = dentry_kill(dentry, 1);
850 static enum lru_status
851 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
853 struct list_head *freeable = arg;
854 struct dentry *dentry = container_of(item, struct dentry, d_lru);
858 * we are inverting the lru lock/dentry->d_lock here,
859 * so use a trylock. If we fail to get the lock, just skip
862 if (!spin_trylock(&dentry->d_lock))
866 * Referenced dentries are still in use. If they have active
867 * counts, just remove them from the LRU. Otherwise give them
868 * another pass through the LRU.
870 if (dentry->d_lockref.count) {
871 d_lru_isolate(dentry);
872 spin_unlock(&dentry->d_lock);
876 if (dentry->d_flags & DCACHE_REFERENCED) {
877 dentry->d_flags &= ~DCACHE_REFERENCED;
878 spin_unlock(&dentry->d_lock);
881 * The list move itself will be made by the common LRU code. At
882 * this point, we've dropped the dentry->d_lock but keep the
883 * lru lock. This is safe to do, since every list movement is
884 * protected by the lru lock even if both locks are held.
886 * This is guaranteed by the fact that all LRU management
887 * functions are intermediated by the LRU API calls like
888 * list_lru_add and list_lru_del. List movement in this file
889 * only ever occur through this functions or through callbacks
890 * like this one, that are called from the LRU API.
892 * The only exceptions to this are functions like
893 * shrink_dentry_list, and code that first checks for the
894 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
895 * operating only with stack provided lists after they are
896 * properly isolated from the main list. It is thus, always a
902 d_lru_shrink_move(dentry, freeable);
903 spin_unlock(&dentry->d_lock);
909 * prune_dcache_sb - shrink the dcache
911 * @nr_to_scan : number of entries to try to free
912 * @nid: which node to scan for freeable entities
914 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
915 * done when we need more memory an called from the superblock shrinker
918 * This function may fail to free any resources if all the dentries are in
921 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
927 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
928 &dispose, &nr_to_scan);
929 shrink_dentry_list(&dispose);
933 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
934 spinlock_t *lru_lock, void *arg)
936 struct list_head *freeable = arg;
937 struct dentry *dentry = container_of(item, struct dentry, d_lru);
940 * we are inverting the lru lock/dentry->d_lock here,
941 * so use a trylock. If we fail to get the lock, just skip
944 if (!spin_trylock(&dentry->d_lock))
947 d_lru_shrink_move(dentry, freeable);
948 spin_unlock(&dentry->d_lock);
955 * shrink_dcache_sb - shrink dcache for a superblock
958 * Shrink the dcache for the specified super block. This is used to free
959 * the dcache before unmounting a file system.
961 void shrink_dcache_sb(struct super_block *sb)
968 freed = list_lru_walk(&sb->s_dentry_lru,
969 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
971 this_cpu_sub(nr_dentry_unused, freed);
972 shrink_dentry_list(&dispose);
975 EXPORT_SYMBOL(shrink_dcache_sb);
978 * enum d_walk_ret - action to talke during tree walk
979 * @D_WALK_CONTINUE: contrinue walk
980 * @D_WALK_QUIT: quit walk
981 * @D_WALK_NORETRY: quit when retry is needed
982 * @D_WALK_SKIP: skip this dentry and its children
992 * d_walk - walk the dentry tree
993 * @parent: start of walk
994 * @data: data passed to @enter() and @finish()
995 * @enter: callback when first entering the dentry
996 * @finish: callback when successfully finished the walk
998 * The @enter() and @finish() callbacks are called with d_lock held.
1000 static void d_walk(struct dentry *parent, void *data,
1001 enum d_walk_ret (*enter)(void *, struct dentry *),
1002 void (*finish)(void *))
1004 struct dentry *this_parent;
1005 struct list_head *next;
1007 enum d_walk_ret ret;
1011 read_seqbegin_or_lock(&rename_lock, &seq);
1012 this_parent = parent;
1013 spin_lock(&this_parent->d_lock);
1015 ret = enter(data, this_parent);
1017 case D_WALK_CONTINUE:
1022 case D_WALK_NORETRY:
1027 next = this_parent->d_subdirs.next;
1029 while (next != &this_parent->d_subdirs) {
1030 struct list_head *tmp = next;
1031 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1034 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1036 ret = enter(data, dentry);
1038 case D_WALK_CONTINUE:
1041 spin_unlock(&dentry->d_lock);
1043 case D_WALK_NORETRY:
1047 spin_unlock(&dentry->d_lock);
1051 if (!list_empty(&dentry->d_subdirs)) {
1052 spin_unlock(&this_parent->d_lock);
1053 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1054 this_parent = dentry;
1055 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1058 spin_unlock(&dentry->d_lock);
1061 * All done at this level ... ascend and resume the search.
1063 if (this_parent != parent) {
1064 struct dentry *child = this_parent;
1065 this_parent = child->d_parent;
1068 spin_unlock(&child->d_lock);
1069 spin_lock(&this_parent->d_lock);
1072 * might go back up the wrong parent if we have had a rename
1075 if (this_parent != child->d_parent ||
1076 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1077 need_seqretry(&rename_lock, seq)) {
1078 spin_unlock(&this_parent->d_lock);
1083 next = child->d_u.d_child.next;
1086 if (need_seqretry(&rename_lock, seq)) {
1087 spin_unlock(&this_parent->d_lock);
1094 spin_unlock(&this_parent->d_lock);
1095 done_seqretry(&rename_lock, seq);
1106 * Search for at least 1 mount point in the dentry's subdirs.
1107 * We descend to the next level whenever the d_subdirs
1108 * list is non-empty and continue searching.
1111 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1114 if (d_mountpoint(dentry)) {
1118 return D_WALK_CONTINUE;
1122 * have_submounts - check for mounts over a dentry
1123 * @parent: dentry to check.
1125 * Return true if the parent or its subdirectories contain
1128 int have_submounts(struct dentry *parent)
1132 d_walk(parent, &ret, check_mount, NULL);
1136 EXPORT_SYMBOL(have_submounts);
1139 * Called by mount code to set a mountpoint and check if the mountpoint is
1140 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1141 * subtree can become unreachable).
1143 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1144 * this reason take rename_lock and d_lock on dentry and ancestors.
1146 int d_set_mounted(struct dentry *dentry)
1150 write_seqlock(&rename_lock);
1151 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1152 /* Need exclusion wrt. check_submounts_and_drop() */
1153 spin_lock(&p->d_lock);
1154 if (unlikely(d_unhashed(p))) {
1155 spin_unlock(&p->d_lock);
1158 spin_unlock(&p->d_lock);
1160 spin_lock(&dentry->d_lock);
1161 if (!d_unlinked(dentry)) {
1162 dentry->d_flags |= DCACHE_MOUNTED;
1165 spin_unlock(&dentry->d_lock);
1167 write_sequnlock(&rename_lock);
1172 * Search the dentry child list of the specified parent,
1173 * and move any unused dentries to the end of the unused
1174 * list for prune_dcache(). We descend to the next level
1175 * whenever the d_subdirs list is non-empty and continue
1178 * It returns zero iff there are no unused children,
1179 * otherwise it returns the number of children moved to
1180 * the end of the unused list. This may not be the total
1181 * number of unused children, because select_parent can
1182 * drop the lock and return early due to latency
1186 struct select_data {
1187 struct dentry *start;
1188 struct list_head dispose;
1192 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1194 struct select_data *data = _data;
1195 enum d_walk_ret ret = D_WALK_CONTINUE;
1197 if (data->start == dentry)
1200 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1203 if (dentry->d_flags & DCACHE_LRU_LIST)
1205 if (!dentry->d_lockref.count) {
1206 d_shrink_add(dentry, &data->dispose);
1211 * We can return to the caller if we have found some (this
1212 * ensures forward progress). We'll be coming back to find
1215 if (!list_empty(&data->dispose))
1216 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1222 * shrink_dcache_parent - prune dcache
1223 * @parent: parent of entries to prune
1225 * Prune the dcache to remove unused children of the parent dentry.
1227 void shrink_dcache_parent(struct dentry *parent)
1230 struct select_data data;
1232 INIT_LIST_HEAD(&data.dispose);
1233 data.start = parent;
1236 d_walk(parent, &data, select_collect, NULL);
1240 shrink_dentry_list(&data.dispose);
1244 EXPORT_SYMBOL(shrink_dcache_parent);
1246 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1248 /* it has busy descendents; complain about those instead */
1249 if (!list_empty(&dentry->d_subdirs))
1250 return D_WALK_CONTINUE;
1252 /* root with refcount 1 is fine */
1253 if (dentry == _data && dentry->d_lockref.count == 1)
1254 return D_WALK_CONTINUE;
1256 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1257 " still in use (%d) [unmount of %s %s]\n",
1260 dentry->d_inode->i_ino : 0UL,
1262 dentry->d_lockref.count,
1263 dentry->d_sb->s_type->name,
1264 dentry->d_sb->s_id);
1266 return D_WALK_CONTINUE;
1269 static void do_one_tree(struct dentry *dentry)
1271 shrink_dcache_parent(dentry);
1272 d_walk(dentry, dentry, umount_check, NULL);
1278 * destroy the dentries attached to a superblock on unmounting
1280 void shrink_dcache_for_umount(struct super_block *sb)
1282 struct dentry *dentry;
1284 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1286 dentry = sb->s_root;
1288 do_one_tree(dentry);
1290 while (!hlist_bl_empty(&sb->s_anon)) {
1291 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1292 do_one_tree(dentry);
1296 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1298 struct select_data *data = _data;
1300 if (d_mountpoint(dentry)) {
1301 data->found = -EBUSY;
1305 return select_collect(_data, dentry);
1308 static void check_and_drop(void *_data)
1310 struct select_data *data = _data;
1312 if (d_mountpoint(data->start))
1313 data->found = -EBUSY;
1315 __d_drop(data->start);
1319 * check_submounts_and_drop - prune dcache, check for submounts and drop
1321 * All done as a single atomic operation relative to has_unlinked_ancestor().
1322 * Returns 0 if successfully unhashed @parent. If there were submounts then
1325 * @dentry: dentry to prune and drop
1327 int check_submounts_and_drop(struct dentry *dentry)
1331 /* Negative dentries can be dropped without further checks */
1332 if (!dentry->d_inode) {
1338 struct select_data data;
1340 INIT_LIST_HEAD(&data.dispose);
1341 data.start = dentry;
1344 d_walk(dentry, &data, check_and_collect, check_and_drop);
1347 if (!list_empty(&data.dispose))
1348 shrink_dentry_list(&data.dispose);
1359 EXPORT_SYMBOL(check_submounts_and_drop);
1362 * __d_alloc - allocate a dcache entry
1363 * @sb: filesystem it will belong to
1364 * @name: qstr of the name
1366 * Allocates a dentry. It returns %NULL if there is insufficient memory
1367 * available. On a success the dentry is returned. The name passed in is
1368 * copied and the copy passed in may be reused after this call.
1371 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1373 struct dentry *dentry;
1376 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1381 * We guarantee that the inline name is always NUL-terminated.
1382 * This way the memcpy() done by the name switching in rename
1383 * will still always have a NUL at the end, even if we might
1384 * be overwriting an internal NUL character
1386 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1387 if (name->len > DNAME_INLINE_LEN-1) {
1388 dname = kmalloc(name->len + 1, GFP_KERNEL);
1390 kmem_cache_free(dentry_cache, dentry);
1394 dname = dentry->d_iname;
1397 dentry->d_name.len = name->len;
1398 dentry->d_name.hash = name->hash;
1399 memcpy(dname, name->name, name->len);
1400 dname[name->len] = 0;
1402 /* Make sure we always see the terminating NUL character */
1404 dentry->d_name.name = dname;
1406 dentry->d_lockref.count = 1;
1407 dentry->d_flags = 0;
1408 spin_lock_init(&dentry->d_lock);
1409 seqcount_init(&dentry->d_seq);
1410 dentry->d_inode = NULL;
1411 dentry->d_parent = dentry;
1413 dentry->d_op = NULL;
1414 dentry->d_fsdata = NULL;
1415 INIT_HLIST_BL_NODE(&dentry->d_hash);
1416 INIT_LIST_HEAD(&dentry->d_lru);
1417 INIT_LIST_HEAD(&dentry->d_subdirs);
1418 INIT_HLIST_NODE(&dentry->d_alias);
1419 INIT_LIST_HEAD(&dentry->d_u.d_child);
1420 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1422 this_cpu_inc(nr_dentry);
1428 * d_alloc - allocate a dcache entry
1429 * @parent: parent of entry to allocate
1430 * @name: qstr of the name
1432 * Allocates a dentry. It returns %NULL if there is insufficient memory
1433 * available. On a success the dentry is returned. The name passed in is
1434 * copied and the copy passed in may be reused after this call.
1436 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1438 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1442 spin_lock(&parent->d_lock);
1444 * don't need child lock because it is not subject
1445 * to concurrency here
1447 __dget_dlock(parent);
1448 dentry->d_parent = parent;
1449 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1450 spin_unlock(&parent->d_lock);
1454 EXPORT_SYMBOL(d_alloc);
1457 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1458 * @sb: the superblock
1459 * @name: qstr of the name
1461 * For a filesystem that just pins its dentries in memory and never
1462 * performs lookups at all, return an unhashed IS_ROOT dentry.
1464 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1466 return __d_alloc(sb, name);
1468 EXPORT_SYMBOL(d_alloc_pseudo);
1470 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1475 q.len = strlen(name);
1476 q.hash = full_name_hash(q.name, q.len);
1477 return d_alloc(parent, &q);
1479 EXPORT_SYMBOL(d_alloc_name);
1481 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1483 WARN_ON_ONCE(dentry->d_op);
1484 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1486 DCACHE_OP_REVALIDATE |
1487 DCACHE_OP_WEAK_REVALIDATE |
1488 DCACHE_OP_DELETE ));
1493 dentry->d_flags |= DCACHE_OP_HASH;
1495 dentry->d_flags |= DCACHE_OP_COMPARE;
1496 if (op->d_revalidate)
1497 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1498 if (op->d_weak_revalidate)
1499 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1501 dentry->d_flags |= DCACHE_OP_DELETE;
1503 dentry->d_flags |= DCACHE_OP_PRUNE;
1506 EXPORT_SYMBOL(d_set_d_op);
1508 static unsigned d_flags_for_inode(struct inode *inode)
1510 unsigned add_flags = DCACHE_FILE_TYPE;
1513 return DCACHE_MISS_TYPE;
1515 if (S_ISDIR(inode->i_mode)) {
1516 add_flags = DCACHE_DIRECTORY_TYPE;
1517 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1518 if (unlikely(!inode->i_op->lookup))
1519 add_flags = DCACHE_AUTODIR_TYPE;
1521 inode->i_opflags |= IOP_LOOKUP;
1523 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1524 if (unlikely(inode->i_op->follow_link))
1525 add_flags = DCACHE_SYMLINK_TYPE;
1527 inode->i_opflags |= IOP_NOFOLLOW;
1530 if (unlikely(IS_AUTOMOUNT(inode)))
1531 add_flags |= DCACHE_NEED_AUTOMOUNT;
1535 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1537 unsigned add_flags = d_flags_for_inode(inode);
1539 spin_lock(&dentry->d_lock);
1540 __d_set_type(dentry, add_flags);
1542 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1543 dentry->d_inode = inode;
1544 dentry_rcuwalk_barrier(dentry);
1545 spin_unlock(&dentry->d_lock);
1546 fsnotify_d_instantiate(dentry, inode);
1550 * d_instantiate - fill in inode information for a dentry
1551 * @entry: dentry to complete
1552 * @inode: inode to attach to this dentry
1554 * Fill in inode information in the entry.
1556 * This turns negative dentries into productive full members
1559 * NOTE! This assumes that the inode count has been incremented
1560 * (or otherwise set) by the caller to indicate that it is now
1561 * in use by the dcache.
1564 void d_instantiate(struct dentry *entry, struct inode * inode)
1566 BUG_ON(!hlist_unhashed(&entry->d_alias));
1568 spin_lock(&inode->i_lock);
1569 __d_instantiate(entry, inode);
1571 spin_unlock(&inode->i_lock);
1572 security_d_instantiate(entry, inode);
1574 EXPORT_SYMBOL(d_instantiate);
1577 * d_instantiate_unique - instantiate a non-aliased dentry
1578 * @entry: dentry to instantiate
1579 * @inode: inode to attach to this dentry
1581 * Fill in inode information in the entry. On success, it returns NULL.
1582 * If an unhashed alias of "entry" already exists, then we return the
1583 * aliased dentry instead and drop one reference to inode.
1585 * Note that in order to avoid conflicts with rename() etc, the caller
1586 * had better be holding the parent directory semaphore.
1588 * This also assumes that the inode count has been incremented
1589 * (or otherwise set) by the caller to indicate that it is now
1590 * in use by the dcache.
1592 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1593 struct inode *inode)
1595 struct dentry *alias;
1596 int len = entry->d_name.len;
1597 const char *name = entry->d_name.name;
1598 unsigned int hash = entry->d_name.hash;
1601 __d_instantiate(entry, NULL);
1605 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1607 * Don't need alias->d_lock here, because aliases with
1608 * d_parent == entry->d_parent are not subject to name or
1609 * parent changes, because the parent inode i_mutex is held.
1611 if (alias->d_name.hash != hash)
1613 if (alias->d_parent != entry->d_parent)
1615 if (alias->d_name.len != len)
1617 if (dentry_cmp(alias, name, len))
1623 __d_instantiate(entry, inode);
1627 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1629 struct dentry *result;
1631 BUG_ON(!hlist_unhashed(&entry->d_alias));
1634 spin_lock(&inode->i_lock);
1635 result = __d_instantiate_unique(entry, inode);
1637 spin_unlock(&inode->i_lock);
1640 security_d_instantiate(entry, inode);
1644 BUG_ON(!d_unhashed(result));
1649 EXPORT_SYMBOL(d_instantiate_unique);
1652 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1653 * @entry: dentry to complete
1654 * @inode: inode to attach to this dentry
1656 * Fill in inode information in the entry. If a directory alias is found, then
1657 * return an error (and drop inode). Together with d_materialise_unique() this
1658 * guarantees that a directory inode may never have more than one alias.
1660 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1662 BUG_ON(!hlist_unhashed(&entry->d_alias));
1664 spin_lock(&inode->i_lock);
1665 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1666 spin_unlock(&inode->i_lock);
1670 __d_instantiate(entry, inode);
1671 spin_unlock(&inode->i_lock);
1672 security_d_instantiate(entry, inode);
1676 EXPORT_SYMBOL(d_instantiate_no_diralias);
1678 struct dentry *d_make_root(struct inode *root_inode)
1680 struct dentry *res = NULL;
1683 static const struct qstr name = QSTR_INIT("/", 1);
1685 res = __d_alloc(root_inode->i_sb, &name);
1687 d_instantiate(res, root_inode);
1693 EXPORT_SYMBOL(d_make_root);
1695 static struct dentry * __d_find_any_alias(struct inode *inode)
1697 struct dentry *alias;
1699 if (hlist_empty(&inode->i_dentry))
1701 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1707 * d_find_any_alias - find any alias for a given inode
1708 * @inode: inode to find an alias for
1710 * If any aliases exist for the given inode, take and return a
1711 * reference for one of them. If no aliases exist, return %NULL.
1713 struct dentry *d_find_any_alias(struct inode *inode)
1717 spin_lock(&inode->i_lock);
1718 de = __d_find_any_alias(inode);
1719 spin_unlock(&inode->i_lock);
1722 EXPORT_SYMBOL(d_find_any_alias);
1725 * d_obtain_alias - find or allocate a dentry for a given inode
1726 * @inode: inode to allocate the dentry for
1728 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1729 * similar open by handle operations. The returned dentry may be anonymous,
1730 * or may have a full name (if the inode was already in the cache).
1732 * When called on a directory inode, we must ensure that the inode only ever
1733 * has one dentry. If a dentry is found, that is returned instead of
1734 * allocating a new one.
1736 * On successful return, the reference to the inode has been transferred
1737 * to the dentry. In case of an error the reference on the inode is released.
1738 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1739 * be passed in and will be the error will be propagate to the return value,
1740 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1742 struct dentry *d_obtain_alias(struct inode *inode)
1744 static const struct qstr anonstring = QSTR_INIT("/", 1);
1750 return ERR_PTR(-ESTALE);
1752 return ERR_CAST(inode);
1754 res = d_find_any_alias(inode);
1758 tmp = __d_alloc(inode->i_sb, &anonstring);
1760 res = ERR_PTR(-ENOMEM);
1764 spin_lock(&inode->i_lock);
1765 res = __d_find_any_alias(inode);
1767 spin_unlock(&inode->i_lock);
1772 /* attach a disconnected dentry */
1773 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1775 spin_lock(&tmp->d_lock);
1776 tmp->d_inode = inode;
1777 tmp->d_flags |= add_flags;
1778 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1779 hlist_bl_lock(&tmp->d_sb->s_anon);
1780 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1781 hlist_bl_unlock(&tmp->d_sb->s_anon);
1782 spin_unlock(&tmp->d_lock);
1783 spin_unlock(&inode->i_lock);
1784 security_d_instantiate(tmp, inode);
1789 if (res && !IS_ERR(res))
1790 security_d_instantiate(res, inode);
1794 EXPORT_SYMBOL(d_obtain_alias);
1797 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1798 * @inode: the inode which may have a disconnected dentry
1799 * @dentry: a negative dentry which we want to point to the inode.
1801 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1802 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1803 * and return it, else simply d_add the inode to the dentry and return NULL.
1805 * This is needed in the lookup routine of any filesystem that is exportable
1806 * (via knfsd) so that we can build dcache paths to directories effectively.
1808 * If a dentry was found and moved, then it is returned. Otherwise NULL
1809 * is returned. This matches the expected return value of ->lookup.
1811 * Cluster filesystems may call this function with a negative, hashed dentry.
1812 * In that case, we know that the inode will be a regular file, and also this
1813 * will only occur during atomic_open. So we need to check for the dentry
1814 * being already hashed only in the final case.
1816 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1818 struct dentry *new = NULL;
1821 return ERR_CAST(inode);
1823 if (inode && S_ISDIR(inode->i_mode)) {
1824 spin_lock(&inode->i_lock);
1825 new = __d_find_alias(inode, 1);
1827 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1828 spin_unlock(&inode->i_lock);
1829 security_d_instantiate(new, inode);
1830 d_move(new, dentry);
1833 /* already taking inode->i_lock, so d_add() by hand */
1834 __d_instantiate(dentry, inode);
1835 spin_unlock(&inode->i_lock);
1836 security_d_instantiate(dentry, inode);
1840 d_instantiate(dentry, inode);
1841 if (d_unhashed(dentry))
1846 EXPORT_SYMBOL(d_splice_alias);
1849 * d_add_ci - lookup or allocate new dentry with case-exact name
1850 * @inode: the inode case-insensitive lookup has found
1851 * @dentry: the negative dentry that was passed to the parent's lookup func
1852 * @name: the case-exact name to be associated with the returned dentry
1854 * This is to avoid filling the dcache with case-insensitive names to the
1855 * same inode, only the actual correct case is stored in the dcache for
1856 * case-insensitive filesystems.
1858 * For a case-insensitive lookup match and if the the case-exact dentry
1859 * already exists in in the dcache, use it and return it.
1861 * If no entry exists with the exact case name, allocate new dentry with
1862 * the exact case, and return the spliced entry.
1864 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1867 struct dentry *found;
1871 * First check if a dentry matching the name already exists,
1872 * if not go ahead and create it now.
1874 found = d_hash_and_lookup(dentry->d_parent, name);
1875 if (unlikely(IS_ERR(found)))
1878 new = d_alloc(dentry->d_parent, name);
1880 found = ERR_PTR(-ENOMEM);
1884 found = d_splice_alias(inode, new);
1893 * If a matching dentry exists, and it's not negative use it.
1895 * Decrement the reference count to balance the iget() done
1898 if (found->d_inode) {
1899 if (unlikely(found->d_inode != inode)) {
1900 /* This can't happen because bad inodes are unhashed. */
1901 BUG_ON(!is_bad_inode(inode));
1902 BUG_ON(!is_bad_inode(found->d_inode));
1909 * Negative dentry: instantiate it unless the inode is a directory and
1910 * already has a dentry.
1912 new = d_splice_alias(inode, found);
1923 EXPORT_SYMBOL(d_add_ci);
1926 * Do the slow-case of the dentry name compare.
1928 * Unlike the dentry_cmp() function, we need to atomically
1929 * load the name and length information, so that the
1930 * filesystem can rely on them, and can use the 'name' and
1931 * 'len' information without worrying about walking off the
1932 * end of memory etc.
1934 * Thus the read_seqcount_retry() and the "duplicate" info
1935 * in arguments (the low-level filesystem should not look
1936 * at the dentry inode or name contents directly, since
1937 * rename can change them while we're in RCU mode).
1939 enum slow_d_compare {
1945 static noinline enum slow_d_compare slow_dentry_cmp(
1946 const struct dentry *parent,
1947 struct dentry *dentry,
1949 const struct qstr *name)
1951 int tlen = dentry->d_name.len;
1952 const char *tname = dentry->d_name.name;
1954 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1956 return D_COMP_SEQRETRY;
1958 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1959 return D_COMP_NOMATCH;
1964 * __d_lookup_rcu - search for a dentry (racy, store-free)
1965 * @parent: parent dentry
1966 * @name: qstr of name we wish to find
1967 * @seqp: returns d_seq value at the point where the dentry was found
1968 * Returns: dentry, or NULL
1970 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1971 * resolution (store-free path walking) design described in
1972 * Documentation/filesystems/path-lookup.txt.
1974 * This is not to be used outside core vfs.
1976 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1977 * held, and rcu_read_lock held. The returned dentry must not be stored into
1978 * without taking d_lock and checking d_seq sequence count against @seq
1981 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1984 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1985 * the returned dentry, so long as its parent's seqlock is checked after the
1986 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1987 * is formed, giving integrity down the path walk.
1989 * NOTE! The caller *has* to check the resulting dentry against the sequence
1990 * number we've returned before using any of the resulting dentry state!
1992 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1993 const struct qstr *name,
1996 u64 hashlen = name->hash_len;
1997 const unsigned char *str = name->name;
1998 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1999 struct hlist_bl_node *node;
2000 struct dentry *dentry;
2003 * Note: There is significant duplication with __d_lookup_rcu which is
2004 * required to prevent single threaded performance regressions
2005 * especially on architectures where smp_rmb (in seqcounts) are costly.
2006 * Keep the two functions in sync.
2010 * The hash list is protected using RCU.
2012 * Carefully use d_seq when comparing a candidate dentry, to avoid
2013 * races with d_move().
2015 * It is possible that concurrent renames can mess up our list
2016 * walk here and result in missing our dentry, resulting in the
2017 * false-negative result. d_lookup() protects against concurrent
2018 * renames using rename_lock seqlock.
2020 * See Documentation/filesystems/path-lookup.txt for more details.
2022 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2027 * The dentry sequence count protects us from concurrent
2028 * renames, and thus protects parent and name fields.
2030 * The caller must perform a seqcount check in order
2031 * to do anything useful with the returned dentry.
2033 * NOTE! We do a "raw" seqcount_begin here. That means that
2034 * we don't wait for the sequence count to stabilize if it
2035 * is in the middle of a sequence change. If we do the slow
2036 * dentry compare, we will do seqretries until it is stable,
2037 * and if we end up with a successful lookup, we actually
2038 * want to exit RCU lookup anyway.
2040 seq = raw_seqcount_begin(&dentry->d_seq);
2041 if (dentry->d_parent != parent)
2043 if (d_unhashed(dentry))
2046 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2047 if (dentry->d_name.hash != hashlen_hash(hashlen))
2050 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2053 case D_COMP_NOMATCH:
2060 if (dentry->d_name.hash_len != hashlen)
2063 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2070 * d_lookup - search for a dentry
2071 * @parent: parent dentry
2072 * @name: qstr of name we wish to find
2073 * Returns: dentry, or NULL
2075 * d_lookup searches the children of the parent dentry for the name in
2076 * question. If the dentry is found its reference count is incremented and the
2077 * dentry is returned. The caller must use dput to free the entry when it has
2078 * finished using it. %NULL is returned if the dentry does not exist.
2080 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2082 struct dentry *dentry;
2086 seq = read_seqbegin(&rename_lock);
2087 dentry = __d_lookup(parent, name);
2090 } while (read_seqretry(&rename_lock, seq));
2093 EXPORT_SYMBOL(d_lookup);
2096 * __d_lookup - search for a dentry (racy)
2097 * @parent: parent dentry
2098 * @name: qstr of name we wish to find
2099 * Returns: dentry, or NULL
2101 * __d_lookup is like d_lookup, however it may (rarely) return a
2102 * false-negative result due to unrelated rename activity.
2104 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2105 * however it must be used carefully, eg. with a following d_lookup in
2106 * the case of failure.
2108 * __d_lookup callers must be commented.
2110 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2112 unsigned int len = name->len;
2113 unsigned int hash = name->hash;
2114 const unsigned char *str = name->name;
2115 struct hlist_bl_head *b = d_hash(parent, hash);
2116 struct hlist_bl_node *node;
2117 struct dentry *found = NULL;
2118 struct dentry *dentry;
2121 * Note: There is significant duplication with __d_lookup_rcu which is
2122 * required to prevent single threaded performance regressions
2123 * especially on architectures where smp_rmb (in seqcounts) are costly.
2124 * Keep the two functions in sync.
2128 * The hash list is protected using RCU.
2130 * Take d_lock when comparing a candidate dentry, to avoid races
2133 * It is possible that concurrent renames can mess up our list
2134 * walk here and result in missing our dentry, resulting in the
2135 * false-negative result. d_lookup() protects against concurrent
2136 * renames using rename_lock seqlock.
2138 * See Documentation/filesystems/path-lookup.txt for more details.
2142 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2144 if (dentry->d_name.hash != hash)
2147 spin_lock(&dentry->d_lock);
2148 if (dentry->d_parent != parent)
2150 if (d_unhashed(dentry))
2154 * It is safe to compare names since d_move() cannot
2155 * change the qstr (protected by d_lock).
2157 if (parent->d_flags & DCACHE_OP_COMPARE) {
2158 int tlen = dentry->d_name.len;
2159 const char *tname = dentry->d_name.name;
2160 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2163 if (dentry->d_name.len != len)
2165 if (dentry_cmp(dentry, str, len))
2169 dentry->d_lockref.count++;
2171 spin_unlock(&dentry->d_lock);
2174 spin_unlock(&dentry->d_lock);
2182 * d_hash_and_lookup - hash the qstr then search for a dentry
2183 * @dir: Directory to search in
2184 * @name: qstr of name we wish to find
2186 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2188 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2191 * Check for a fs-specific hash function. Note that we must
2192 * calculate the standard hash first, as the d_op->d_hash()
2193 * routine may choose to leave the hash value unchanged.
2195 name->hash = full_name_hash(name->name, name->len);
2196 if (dir->d_flags & DCACHE_OP_HASH) {
2197 int err = dir->d_op->d_hash(dir, name);
2198 if (unlikely(err < 0))
2199 return ERR_PTR(err);
2201 return d_lookup(dir, name);
2203 EXPORT_SYMBOL(d_hash_and_lookup);
2206 * d_validate - verify dentry provided from insecure source (deprecated)
2207 * @dentry: The dentry alleged to be valid child of @dparent
2208 * @dparent: The parent dentry (known to be valid)
2210 * An insecure source has sent us a dentry, here we verify it and dget() it.
2211 * This is used by ncpfs in its readdir implementation.
2212 * Zero is returned in the dentry is invalid.
2214 * This function is slow for big directories, and deprecated, do not use it.
2216 int d_validate(struct dentry *dentry, struct dentry *dparent)
2218 struct dentry *child;
2220 spin_lock(&dparent->d_lock);
2221 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2222 if (dentry == child) {
2223 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2224 __dget_dlock(dentry);
2225 spin_unlock(&dentry->d_lock);
2226 spin_unlock(&dparent->d_lock);
2230 spin_unlock(&dparent->d_lock);
2234 EXPORT_SYMBOL(d_validate);
2237 * When a file is deleted, we have two options:
2238 * - turn this dentry into a negative dentry
2239 * - unhash this dentry and free it.
2241 * Usually, we want to just turn this into
2242 * a negative dentry, but if anybody else is
2243 * currently using the dentry or the inode
2244 * we can't do that and we fall back on removing
2245 * it from the hash queues and waiting for
2246 * it to be deleted later when it has no users
2250 * d_delete - delete a dentry
2251 * @dentry: The dentry to delete
2253 * Turn the dentry into a negative dentry if possible, otherwise
2254 * remove it from the hash queues so it can be deleted later
2257 void d_delete(struct dentry * dentry)
2259 struct inode *inode;
2262 * Are we the only user?
2265 spin_lock(&dentry->d_lock);
2266 inode = dentry->d_inode;
2267 isdir = S_ISDIR(inode->i_mode);
2268 if (dentry->d_lockref.count == 1) {
2269 if (!spin_trylock(&inode->i_lock)) {
2270 spin_unlock(&dentry->d_lock);
2274 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2275 dentry_unlink_inode(dentry);
2276 fsnotify_nameremove(dentry, isdir);
2280 if (!d_unhashed(dentry))
2283 spin_unlock(&dentry->d_lock);
2285 fsnotify_nameremove(dentry, isdir);
2287 EXPORT_SYMBOL(d_delete);
2289 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2291 BUG_ON(!d_unhashed(entry));
2293 entry->d_flags |= DCACHE_RCUACCESS;
2294 hlist_bl_add_head_rcu(&entry->d_hash, b);
2298 static void _d_rehash(struct dentry * entry)
2300 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2304 * d_rehash - add an entry back to the hash
2305 * @entry: dentry to add to the hash
2307 * Adds a dentry to the hash according to its name.
2310 void d_rehash(struct dentry * entry)
2312 spin_lock(&entry->d_lock);
2314 spin_unlock(&entry->d_lock);
2316 EXPORT_SYMBOL(d_rehash);
2319 * dentry_update_name_case - update case insensitive dentry with a new name
2320 * @dentry: dentry to be updated
2323 * Update a case insensitive dentry with new case of name.
2325 * dentry must have been returned by d_lookup with name @name. Old and new
2326 * name lengths must match (ie. no d_compare which allows mismatched name
2329 * Parent inode i_mutex must be held over d_lookup and into this call (to
2330 * keep renames and concurrent inserts, and readdir(2) away).
2332 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2334 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2335 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2337 spin_lock(&dentry->d_lock);
2338 write_seqcount_begin(&dentry->d_seq);
2339 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2340 write_seqcount_end(&dentry->d_seq);
2341 spin_unlock(&dentry->d_lock);
2343 EXPORT_SYMBOL(dentry_update_name_case);
2345 static void switch_names(struct dentry *dentry, struct dentry *target)
2347 if (dname_external(target)) {
2348 if (dname_external(dentry)) {
2350 * Both external: swap the pointers
2352 swap(target->d_name.name, dentry->d_name.name);
2355 * dentry:internal, target:external. Steal target's
2356 * storage and make target internal.
2358 memcpy(target->d_iname, dentry->d_name.name,
2359 dentry->d_name.len + 1);
2360 dentry->d_name.name = target->d_name.name;
2361 target->d_name.name = target->d_iname;
2364 if (dname_external(dentry)) {
2366 * dentry:external, target:internal. Give dentry's
2367 * storage to target and make dentry internal
2369 memcpy(dentry->d_iname, target->d_name.name,
2370 target->d_name.len + 1);
2371 target->d_name.name = dentry->d_name.name;
2372 dentry->d_name.name = dentry->d_iname;
2375 * Both are internal.
2378 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2379 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2380 swap(((long *) &dentry->d_iname)[i],
2381 ((long *) &target->d_iname)[i]);
2385 swap(dentry->d_name.len, target->d_name.len);
2388 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2391 * XXXX: do we really need to take target->d_lock?
2393 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2394 spin_lock(&target->d_parent->d_lock);
2396 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2397 spin_lock(&dentry->d_parent->d_lock);
2398 spin_lock_nested(&target->d_parent->d_lock,
2399 DENTRY_D_LOCK_NESTED);
2401 spin_lock(&target->d_parent->d_lock);
2402 spin_lock_nested(&dentry->d_parent->d_lock,
2403 DENTRY_D_LOCK_NESTED);
2406 if (target < dentry) {
2407 spin_lock_nested(&target->d_lock, 2);
2408 spin_lock_nested(&dentry->d_lock, 3);
2410 spin_lock_nested(&dentry->d_lock, 2);
2411 spin_lock_nested(&target->d_lock, 3);
2415 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2416 struct dentry *target)
2418 if (target->d_parent != dentry->d_parent)
2419 spin_unlock(&dentry->d_parent->d_lock);
2420 if (target->d_parent != target)
2421 spin_unlock(&target->d_parent->d_lock);
2425 * When switching names, the actual string doesn't strictly have to
2426 * be preserved in the target - because we're dropping the target
2427 * anyway. As such, we can just do a simple memcpy() to copy over
2428 * the new name before we switch.
2430 * Note that we have to be a lot more careful about getting the hash
2431 * switched - we have to switch the hash value properly even if it
2432 * then no longer matches the actual (corrupted) string of the target.
2433 * The hash value has to match the hash queue that the dentry is on..
2436 * __d_move - move a dentry
2437 * @dentry: entry to move
2438 * @target: new dentry
2439 * @exchange: exchange the two dentries
2441 * Update the dcache to reflect the move of a file name. Negative
2442 * dcache entries should not be moved in this way. Caller must hold
2443 * rename_lock, the i_mutex of the source and target directories,
2444 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2446 static void __d_move(struct dentry *dentry, struct dentry *target,
2449 if (!dentry->d_inode)
2450 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2452 BUG_ON(d_ancestor(dentry, target));
2453 BUG_ON(d_ancestor(target, dentry));
2455 dentry_lock_for_move(dentry, target);
2457 write_seqcount_begin(&dentry->d_seq);
2458 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2460 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2463 * Move the dentry to the target hash queue. Don't bother checking
2464 * for the same hash queue because of how unlikely it is.
2467 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2470 * Unhash the target (d_delete() is not usable here). If exchanging
2471 * the two dentries, then rehash onto the other's hash queue.
2476 d_hash(dentry->d_parent, dentry->d_name.hash));
2479 list_del(&dentry->d_u.d_child);
2480 list_del(&target->d_u.d_child);
2482 /* Switch the names.. */
2483 switch_names(dentry, target);
2484 swap(dentry->d_name.hash, target->d_name.hash);
2486 /* ... and switch the parents */
2487 if (IS_ROOT(dentry)) {
2488 dentry->d_parent = target->d_parent;
2489 target->d_parent = target;
2490 INIT_LIST_HEAD(&target->d_u.d_child);
2492 swap(dentry->d_parent, target->d_parent);
2494 /* And add them back to the (new) parent lists */
2495 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2498 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2500 write_seqcount_end(&target->d_seq);
2501 write_seqcount_end(&dentry->d_seq);
2503 dentry_unlock_parents_for_move(dentry, target);
2505 fsnotify_d_move(target);
2506 spin_unlock(&target->d_lock);
2507 fsnotify_d_move(dentry);
2508 spin_unlock(&dentry->d_lock);
2512 * d_move - move a dentry
2513 * @dentry: entry to move
2514 * @target: new dentry
2516 * Update the dcache to reflect the move of a file name. Negative
2517 * dcache entries should not be moved in this way. See the locking
2518 * requirements for __d_move.
2520 void d_move(struct dentry *dentry, struct dentry *target)
2522 write_seqlock(&rename_lock);
2523 __d_move(dentry, target, false);
2524 write_sequnlock(&rename_lock);
2526 EXPORT_SYMBOL(d_move);
2529 * d_exchange - exchange two dentries
2530 * @dentry1: first dentry
2531 * @dentry2: second dentry
2533 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2535 write_seqlock(&rename_lock);
2537 WARN_ON(!dentry1->d_inode);
2538 WARN_ON(!dentry2->d_inode);
2539 WARN_ON(IS_ROOT(dentry1));
2540 WARN_ON(IS_ROOT(dentry2));
2542 __d_move(dentry1, dentry2, true);
2544 write_sequnlock(&rename_lock);
2548 * d_ancestor - search for an ancestor
2549 * @p1: ancestor dentry
2552 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2553 * an ancestor of p2, else NULL.
2555 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2559 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2560 if (p->d_parent == p1)
2567 * This helper attempts to cope with remotely renamed directories
2569 * It assumes that the caller is already holding
2570 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2572 * Note: If ever the locking in lock_rename() changes, then please
2573 * remember to update this too...
2575 static struct dentry *__d_unalias(struct inode *inode,
2576 struct dentry *dentry, struct dentry *alias)
2578 struct mutex *m1 = NULL, *m2 = NULL;
2579 struct dentry *ret = ERR_PTR(-EBUSY);
2581 /* If alias and dentry share a parent, then no extra locks required */
2582 if (alias->d_parent == dentry->d_parent)
2585 /* See lock_rename() */
2586 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2588 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2589 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2591 m2 = &alias->d_parent->d_inode->i_mutex;
2593 if (likely(!d_mountpoint(alias))) {
2594 __d_move(alias, dentry, false);
2598 spin_unlock(&inode->i_lock);
2607 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2608 * named dentry in place of the dentry to be replaced.
2609 * returns with anon->d_lock held!
2611 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2613 struct dentry *dparent;
2615 dentry_lock_for_move(anon, dentry);
2617 write_seqcount_begin(&dentry->d_seq);
2618 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2620 dparent = dentry->d_parent;
2622 switch_names(dentry, anon);
2623 swap(dentry->d_name.hash, anon->d_name.hash);
2625 dentry->d_parent = dentry;
2626 list_del_init(&dentry->d_u.d_child);
2627 anon->d_parent = dparent;
2628 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2630 write_seqcount_end(&dentry->d_seq);
2631 write_seqcount_end(&anon->d_seq);
2633 dentry_unlock_parents_for_move(anon, dentry);
2634 spin_unlock(&dentry->d_lock);
2636 /* anon->d_lock still locked, returns locked */
2640 * d_materialise_unique - introduce an inode into the tree
2641 * @dentry: candidate dentry
2642 * @inode: inode to bind to the dentry, to which aliases may be attached
2644 * Introduces an dentry into the tree, substituting an extant disconnected
2645 * root directory alias in its place if there is one. Caller must hold the
2646 * i_mutex of the parent directory.
2648 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2650 struct dentry *actual;
2652 BUG_ON(!d_unhashed(dentry));
2656 __d_instantiate(dentry, NULL);
2661 spin_lock(&inode->i_lock);
2663 if (S_ISDIR(inode->i_mode)) {
2664 struct dentry *alias;
2666 /* Does an aliased dentry already exist? */
2667 alias = __d_find_alias(inode, 0);
2670 write_seqlock(&rename_lock);
2672 if (d_ancestor(alias, dentry)) {
2673 /* Check for loops */
2674 actual = ERR_PTR(-ELOOP);
2675 spin_unlock(&inode->i_lock);
2676 } else if (IS_ROOT(alias)) {
2677 /* Is this an anonymous mountpoint that we
2678 * could splice into our tree? */
2679 __d_materialise_dentry(dentry, alias);
2680 write_sequnlock(&rename_lock);
2684 /* Nope, but we must(!) avoid directory
2685 * aliasing. This drops inode->i_lock */
2686 actual = __d_unalias(inode, dentry, alias);
2688 write_sequnlock(&rename_lock);
2689 if (IS_ERR(actual)) {
2690 if (PTR_ERR(actual) == -ELOOP)
2691 pr_warn_ratelimited(
2692 "VFS: Lookup of '%s' in %s %s"
2693 " would have caused loop\n",
2694 dentry->d_name.name,
2695 inode->i_sb->s_type->name,
2703 /* Add a unique reference */
2704 actual = __d_instantiate_unique(dentry, inode);
2708 BUG_ON(!d_unhashed(actual));
2710 spin_lock(&actual->d_lock);
2713 spin_unlock(&actual->d_lock);
2714 spin_unlock(&inode->i_lock);
2716 if (actual == dentry) {
2717 security_d_instantiate(dentry, inode);
2724 EXPORT_SYMBOL_GPL(d_materialise_unique);
2726 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2730 return -ENAMETOOLONG;
2732 memcpy(*buffer, str, namelen);
2737 * prepend_name - prepend a pathname in front of current buffer pointer
2738 * @buffer: buffer pointer
2739 * @buflen: allocated length of the buffer
2740 * @name: name string and length qstr structure
2742 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2743 * make sure that either the old or the new name pointer and length are
2744 * fetched. However, there may be mismatch between length and pointer.
2745 * The length cannot be trusted, we need to copy it byte-by-byte until
2746 * the length is reached or a null byte is found. It also prepends "/" at
2747 * the beginning of the name. The sequence number check at the caller will
2748 * retry it again when a d_move() does happen. So any garbage in the buffer
2749 * due to mismatched pointer and length will be discarded.
2751 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2753 const char *dname = ACCESS_ONCE(name->name);
2754 u32 dlen = ACCESS_ONCE(name->len);
2757 *buflen -= dlen + 1;
2759 return -ENAMETOOLONG;
2760 p = *buffer -= dlen + 1;
2772 * prepend_path - Prepend path string to a buffer
2773 * @path: the dentry/vfsmount to report
2774 * @root: root vfsmnt/dentry
2775 * @buffer: pointer to the end of the buffer
2776 * @buflen: pointer to buffer length
2778 * The function will first try to write out the pathname without taking any
2779 * lock other than the RCU read lock to make sure that dentries won't go away.
2780 * It only checks the sequence number of the global rename_lock as any change
2781 * in the dentry's d_seq will be preceded by changes in the rename_lock
2782 * sequence number. If the sequence number had been changed, it will restart
2783 * the whole pathname back-tracing sequence again by taking the rename_lock.
2784 * In this case, there is no need to take the RCU read lock as the recursive
2785 * parent pointer references will keep the dentry chain alive as long as no
2786 * rename operation is performed.
2788 static int prepend_path(const struct path *path,
2789 const struct path *root,
2790 char **buffer, int *buflen)
2792 struct dentry *dentry;
2793 struct vfsmount *vfsmnt;
2796 unsigned seq, m_seq = 0;
2802 read_seqbegin_or_lock(&mount_lock, &m_seq);
2809 dentry = path->dentry;
2811 mnt = real_mount(vfsmnt);
2812 read_seqbegin_or_lock(&rename_lock, &seq);
2813 while (dentry != root->dentry || vfsmnt != root->mnt) {
2814 struct dentry * parent;
2816 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2817 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2819 if (mnt != parent) {
2820 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2826 * Filesystems needing to implement special "root names"
2827 * should do so with ->d_dname()
2829 if (IS_ROOT(dentry) &&
2830 (dentry->d_name.len != 1 ||
2831 dentry->d_name.name[0] != '/')) {
2832 WARN(1, "Root dentry has weird name <%.*s>\n",
2833 (int) dentry->d_name.len,
2834 dentry->d_name.name);
2837 error = is_mounted(vfsmnt) ? 1 : 2;
2840 parent = dentry->d_parent;
2842 error = prepend_name(&bptr, &blen, &dentry->d_name);
2850 if (need_seqretry(&rename_lock, seq)) {
2854 done_seqretry(&rename_lock, seq);
2858 if (need_seqretry(&mount_lock, m_seq)) {
2862 done_seqretry(&mount_lock, m_seq);
2864 if (error >= 0 && bptr == *buffer) {
2866 error = -ENAMETOOLONG;
2876 * __d_path - return the path of a dentry
2877 * @path: the dentry/vfsmount to report
2878 * @root: root vfsmnt/dentry
2879 * @buf: buffer to return value in
2880 * @buflen: buffer length
2882 * Convert a dentry into an ASCII path name.
2884 * Returns a pointer into the buffer or an error code if the
2885 * path was too long.
2887 * "buflen" should be positive.
2889 * If the path is not reachable from the supplied root, return %NULL.
2891 char *__d_path(const struct path *path,
2892 const struct path *root,
2893 char *buf, int buflen)
2895 char *res = buf + buflen;
2898 prepend(&res, &buflen, "\0", 1);
2899 error = prepend_path(path, root, &res, &buflen);
2902 return ERR_PTR(error);
2908 char *d_absolute_path(const struct path *path,
2909 char *buf, int buflen)
2911 struct path root = {};
2912 char *res = buf + buflen;
2915 prepend(&res, &buflen, "\0", 1);
2916 error = prepend_path(path, &root, &res, &buflen);
2921 return ERR_PTR(error);
2926 * same as __d_path but appends "(deleted)" for unlinked files.
2928 static int path_with_deleted(const struct path *path,
2929 const struct path *root,
2930 char **buf, int *buflen)
2932 prepend(buf, buflen, "\0", 1);
2933 if (d_unlinked(path->dentry)) {
2934 int error = prepend(buf, buflen, " (deleted)", 10);
2939 return prepend_path(path, root, buf, buflen);
2942 static int prepend_unreachable(char **buffer, int *buflen)
2944 return prepend(buffer, buflen, "(unreachable)", 13);
2947 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2952 seq = read_seqcount_begin(&fs->seq);
2954 } while (read_seqcount_retry(&fs->seq, seq));
2958 * d_path - return the path of a dentry
2959 * @path: path to report
2960 * @buf: buffer to return value in
2961 * @buflen: buffer length
2963 * Convert a dentry into an ASCII path name. If the entry has been deleted
2964 * the string " (deleted)" is appended. Note that this is ambiguous.
2966 * Returns a pointer into the buffer or an error code if the path was
2967 * too long. Note: Callers should use the returned pointer, not the passed
2968 * in buffer, to use the name! The implementation often starts at an offset
2969 * into the buffer, and may leave 0 bytes at the start.
2971 * "buflen" should be positive.
2973 char *d_path(const struct path *path, char *buf, int buflen)
2975 char *res = buf + buflen;
2980 * We have various synthetic filesystems that never get mounted. On
2981 * these filesystems dentries are never used for lookup purposes, and
2982 * thus don't need to be hashed. They also don't need a name until a
2983 * user wants to identify the object in /proc/pid/fd/. The little hack
2984 * below allows us to generate a name for these objects on demand:
2986 * Some pseudo inodes are mountable. When they are mounted
2987 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
2988 * and instead have d_path return the mounted path.
2990 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
2991 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
2992 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2995 get_fs_root_rcu(current->fs, &root);
2996 error = path_with_deleted(path, &root, &res, &buflen);
3000 res = ERR_PTR(error);
3003 EXPORT_SYMBOL(d_path);
3006 * Helper function for dentry_operations.d_dname() members
3008 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3009 const char *fmt, ...)
3015 va_start(args, fmt);
3016 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3019 if (sz > sizeof(temp) || sz > buflen)
3020 return ERR_PTR(-ENAMETOOLONG);
3022 buffer += buflen - sz;
3023 return memcpy(buffer, temp, sz);
3026 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3028 char *end = buffer + buflen;
3029 /* these dentries are never renamed, so d_lock is not needed */
3030 if (prepend(&end, &buflen, " (deleted)", 11) ||
3031 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3032 prepend(&end, &buflen, "/", 1))
3033 end = ERR_PTR(-ENAMETOOLONG);
3036 EXPORT_SYMBOL(simple_dname);
3039 * Write full pathname from the root of the filesystem into the buffer.
3041 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3043 struct dentry *dentry;
3056 prepend(&end, &len, "\0", 1);
3060 read_seqbegin_or_lock(&rename_lock, &seq);
3061 while (!IS_ROOT(dentry)) {
3062 struct dentry *parent = dentry->d_parent;
3065 error = prepend_name(&end, &len, &dentry->d_name);
3074 if (need_seqretry(&rename_lock, seq)) {
3078 done_seqretry(&rename_lock, seq);
3083 return ERR_PTR(-ENAMETOOLONG);
3086 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3088 return __dentry_path(dentry, buf, buflen);
3090 EXPORT_SYMBOL(dentry_path_raw);
3092 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3097 if (d_unlinked(dentry)) {
3099 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3103 retval = __dentry_path(dentry, buf, buflen);
3104 if (!IS_ERR(retval) && p)
3105 *p = '/'; /* restore '/' overriden with '\0' */
3108 return ERR_PTR(-ENAMETOOLONG);
3111 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3117 seq = read_seqcount_begin(&fs->seq);
3120 } while (read_seqcount_retry(&fs->seq, seq));
3124 * NOTE! The user-level library version returns a
3125 * character pointer. The kernel system call just
3126 * returns the length of the buffer filled (which
3127 * includes the ending '\0' character), or a negative
3128 * error value. So libc would do something like
3130 * char *getcwd(char * buf, size_t size)
3134 * retval = sys_getcwd(buf, size);
3141 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3144 struct path pwd, root;
3145 char *page = __getname();
3151 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3154 if (!d_unlinked(pwd.dentry)) {
3156 char *cwd = page + PATH_MAX;
3157 int buflen = PATH_MAX;
3159 prepend(&cwd, &buflen, "\0", 1);
3160 error = prepend_path(&pwd, &root, &cwd, &buflen);
3166 /* Unreachable from current root */
3168 error = prepend_unreachable(&cwd, &buflen);
3174 len = PATH_MAX + page - cwd;
3177 if (copy_to_user(buf, cwd, len))
3190 * Test whether new_dentry is a subdirectory of old_dentry.
3192 * Trivially implemented using the dcache structure
3196 * is_subdir - is new dentry a subdirectory of old_dentry
3197 * @new_dentry: new dentry
3198 * @old_dentry: old dentry
3200 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3201 * Returns 0 otherwise.
3202 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3205 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3210 if (new_dentry == old_dentry)
3214 /* for restarting inner loop in case of seq retry */
3215 seq = read_seqbegin(&rename_lock);
3217 * Need rcu_readlock to protect against the d_parent trashing
3221 if (d_ancestor(old_dentry, new_dentry))
3226 } while (read_seqretry(&rename_lock, seq));
3231 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3233 struct dentry *root = data;
3234 if (dentry != root) {
3235 if (d_unhashed(dentry) || !dentry->d_inode)
3238 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3239 dentry->d_flags |= DCACHE_GENOCIDE;
3240 dentry->d_lockref.count--;
3243 return D_WALK_CONTINUE;
3246 void d_genocide(struct dentry *parent)
3248 d_walk(parent, parent, d_genocide_kill, NULL);
3251 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3253 inode_dec_link_count(inode);
3254 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3255 !hlist_unhashed(&dentry->d_alias) ||
3256 !d_unlinked(dentry));
3257 spin_lock(&dentry->d_parent->d_lock);
3258 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3259 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3260 (unsigned long long)inode->i_ino);
3261 spin_unlock(&dentry->d_lock);
3262 spin_unlock(&dentry->d_parent->d_lock);
3263 d_instantiate(dentry, inode);
3265 EXPORT_SYMBOL(d_tmpfile);
3267 static __initdata unsigned long dhash_entries;
3268 static int __init set_dhash_entries(char *str)
3272 dhash_entries = simple_strtoul(str, &str, 0);
3275 __setup("dhash_entries=", set_dhash_entries);
3277 static void __init dcache_init_early(void)
3281 /* If hashes are distributed across NUMA nodes, defer
3282 * hash allocation until vmalloc space is available.
3288 alloc_large_system_hash("Dentry cache",
3289 sizeof(struct hlist_bl_head),
3298 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3299 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3302 static void __init dcache_init(void)
3307 * A constructor could be added for stable state like the lists,
3308 * but it is probably not worth it because of the cache nature
3311 dentry_cache = KMEM_CACHE(dentry,
3312 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3314 /* Hash may have been set up in dcache_init_early */
3319 alloc_large_system_hash("Dentry cache",
3320 sizeof(struct hlist_bl_head),
3329 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3330 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3333 /* SLAB cache for __getname() consumers */
3334 struct kmem_cache *names_cachep __read_mostly;
3335 EXPORT_SYMBOL(names_cachep);
3337 EXPORT_SYMBOL(d_genocide);
3339 void __init vfs_caches_init_early(void)
3341 dcache_init_early();
3345 void __init vfs_caches_init(unsigned long mempages)
3347 unsigned long reserve;
3349 /* Base hash sizes on available memory, with a reserve equal to
3350 150% of current kernel size */
3352 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3353 mempages -= reserve;
3355 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3356 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3360 files_init(mempages);