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);
250 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
251 * @dentry: the target dentry
252 * After this call, in-progress rcu-walk path lookup will fail. This
253 * should be called after unhashing, and after changing d_inode (if
254 * the dentry has not already been unhashed).
256 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
258 assert_spin_locked(&dentry->d_lock);
259 /* Go through a barrier */
260 write_seqcount_barrier(&dentry->d_seq);
264 * Release the dentry's inode, using the filesystem
265 * d_iput() operation if defined. Dentry has no refcount
268 static void dentry_iput(struct dentry * dentry)
269 __releases(dentry->d_lock)
270 __releases(dentry->d_inode->i_lock)
272 struct inode *inode = dentry->d_inode;
274 dentry->d_inode = NULL;
275 hlist_del_init(&dentry->d_alias);
276 spin_unlock(&dentry->d_lock);
277 spin_unlock(&inode->i_lock);
279 fsnotify_inoderemove(inode);
280 if (dentry->d_op && dentry->d_op->d_iput)
281 dentry->d_op->d_iput(dentry, inode);
285 spin_unlock(&dentry->d_lock);
290 * Release the dentry's inode, using the filesystem
291 * d_iput() operation if defined. dentry remains in-use.
293 static void dentry_unlink_inode(struct dentry * dentry)
294 __releases(dentry->d_lock)
295 __releases(dentry->d_inode->i_lock)
297 struct inode *inode = dentry->d_inode;
298 __d_clear_type(dentry);
299 dentry->d_inode = NULL;
300 hlist_del_init(&dentry->d_alias);
301 dentry_rcuwalk_barrier(dentry);
302 spin_unlock(&dentry->d_lock);
303 spin_unlock(&inode->i_lock);
305 fsnotify_inoderemove(inode);
306 if (dentry->d_op && dentry->d_op->d_iput)
307 dentry->d_op->d_iput(dentry, inode);
313 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
314 * is in use - which includes both the "real" per-superblock
315 * LRU list _and_ the DCACHE_SHRINK_LIST use.
317 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
318 * on the shrink list (ie not on the superblock LRU list).
320 * The per-cpu "nr_dentry_unused" counters are updated with
321 * the DCACHE_LRU_LIST bit.
323 * These helper functions make sure we always follow the
324 * rules. d_lock must be held by the caller.
326 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
327 static void d_lru_add(struct dentry *dentry)
329 D_FLAG_VERIFY(dentry, 0);
330 dentry->d_flags |= DCACHE_LRU_LIST;
331 this_cpu_inc(nr_dentry_unused);
332 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
335 static void d_lru_del(struct dentry *dentry)
337 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
338 dentry->d_flags &= ~DCACHE_LRU_LIST;
339 this_cpu_dec(nr_dentry_unused);
340 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
343 static void d_shrink_del(struct dentry *dentry)
345 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
346 list_del_init(&dentry->d_lru);
347 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
348 this_cpu_dec(nr_dentry_unused);
351 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
353 D_FLAG_VERIFY(dentry, 0);
354 list_add(&dentry->d_lru, list);
355 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
356 this_cpu_inc(nr_dentry_unused);
360 * These can only be called under the global LRU lock, ie during the
361 * callback for freeing the LRU list. "isolate" removes it from the
362 * LRU lists entirely, while shrink_move moves it to the indicated
365 static void d_lru_isolate(struct dentry *dentry)
367 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
368 dentry->d_flags &= ~DCACHE_LRU_LIST;
369 this_cpu_dec(nr_dentry_unused);
370 list_del_init(&dentry->d_lru);
373 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
375 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
376 dentry->d_flags |= DCACHE_SHRINK_LIST;
377 list_move_tail(&dentry->d_lru, list);
381 * dentry_lru_(add|del)_list) must be called with d_lock held.
383 static void dentry_lru_add(struct dentry *dentry)
385 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
390 * Remove a dentry with references from the LRU.
392 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
393 * lose our last reference through the parent walk. In this case, we need to
394 * remove ourselves from the shrink list, not the LRU.
396 static void dentry_lru_del(struct dentry *dentry)
398 if (dentry->d_flags & DCACHE_LRU_LIST) {
399 if (dentry->d_flags & DCACHE_SHRINK_LIST)
400 return d_shrink_del(dentry);
406 * d_drop - drop a dentry
407 * @dentry: dentry to drop
409 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
410 * be found through a VFS lookup any more. Note that this is different from
411 * deleting the dentry - d_delete will try to mark the dentry negative if
412 * possible, giving a successful _negative_ lookup, while d_drop will
413 * just make the cache lookup fail.
415 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
416 * reason (NFS timeouts or autofs deletes).
418 * __d_drop requires dentry->d_lock.
420 void __d_drop(struct dentry *dentry)
422 if (!d_unhashed(dentry)) {
423 struct hlist_bl_head *b;
425 * Hashed dentries are normally on the dentry hashtable,
426 * with the exception of those newly allocated by
427 * d_obtain_alias, which are always IS_ROOT:
429 if (unlikely(IS_ROOT(dentry)))
430 b = &dentry->d_sb->s_anon;
432 b = d_hash(dentry->d_parent, dentry->d_name.hash);
435 __hlist_bl_del(&dentry->d_hash);
436 dentry->d_hash.pprev = NULL;
438 dentry_rcuwalk_barrier(dentry);
441 EXPORT_SYMBOL(__d_drop);
443 void d_drop(struct dentry *dentry)
445 spin_lock(&dentry->d_lock);
447 spin_unlock(&dentry->d_lock);
449 EXPORT_SYMBOL(d_drop);
452 * Finish off a dentry we've decided to kill.
453 * dentry->d_lock must be held, returns with it unlocked.
454 * If ref is non-zero, then decrement the refcount too.
455 * Returns dentry requiring refcount drop, or NULL if we're done.
457 static struct dentry *
458 dentry_kill(struct dentry *dentry, int unlock_on_failure)
459 __releases(dentry->d_lock)
462 struct dentry *parent;
464 inode = dentry->d_inode;
465 if (inode && !spin_trylock(&inode->i_lock)) {
467 if (unlock_on_failure) {
468 spin_unlock(&dentry->d_lock);
471 return dentry; /* try again with same dentry */
476 parent = dentry->d_parent;
477 if (parent && !spin_trylock(&parent->d_lock)) {
479 spin_unlock(&inode->i_lock);
484 * The dentry is now unrecoverably dead to the world.
486 lockref_mark_dead(&dentry->d_lockref);
489 * inform the fs via d_prune that this dentry is about to be
490 * unhashed and destroyed.
492 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
493 dentry->d_op->d_prune(dentry);
495 dentry_lru_del(dentry);
496 /* if it was on the hash then remove it */
498 list_del(&dentry->d_u.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 /* if dentry was never visible to RCU, immediate free is OK */
517 if (!(dentry->d_flags & DCACHE_RCUACCESS))
518 __d_free(&dentry->d_u.d_rcu);
520 call_rcu(&dentry->d_u.d_rcu, __d_free);
527 * This is complicated by the fact that we do not want to put
528 * dentries that are no longer on any hash chain on the unused
529 * list: we'd much rather just get rid of them immediately.
531 * However, that implies that we have to traverse the dentry
532 * tree upwards to the parents which might _also_ now be
533 * scheduled for deletion (it may have been only waiting for
534 * its last child to go away).
536 * This tail recursion is done by hand as we don't want to depend
537 * on the compiler to always get this right (gcc generally doesn't).
538 * Real recursion would eat up our stack space.
542 * dput - release a dentry
543 * @dentry: dentry to release
545 * Release a dentry. This will drop the usage count and if appropriate
546 * call the dentry unlink method as well as removing it from the queues and
547 * releasing its resources. If the parent dentries were scheduled for release
548 * they too may now get deleted.
550 void dput(struct dentry *dentry)
552 if (unlikely(!dentry))
556 if (lockref_put_or_lock(&dentry->d_lockref))
559 /* Unreachable? Get rid of it */
560 if (unlikely(d_unhashed(dentry)))
563 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
564 if (dentry->d_op->d_delete(dentry))
568 if (!(dentry->d_flags & DCACHE_REFERENCED))
569 dentry->d_flags |= DCACHE_REFERENCED;
570 dentry_lru_add(dentry);
572 dentry->d_lockref.count--;
573 spin_unlock(&dentry->d_lock);
577 dentry = dentry_kill(dentry, 1);
584 * d_invalidate - invalidate a dentry
585 * @dentry: dentry to invalidate
587 * Try to invalidate the dentry if it turns out to be
588 * possible. If there are other dentries that can be
589 * reached through this one we can't delete it and we
590 * return -EBUSY. On success we return 0.
595 int d_invalidate(struct dentry * dentry)
598 * If it's already been dropped, return OK.
600 spin_lock(&dentry->d_lock);
601 if (d_unhashed(dentry)) {
602 spin_unlock(&dentry->d_lock);
606 * Check whether to do a partial shrink_dcache
607 * to get rid of unused child entries.
609 if (!list_empty(&dentry->d_subdirs)) {
610 spin_unlock(&dentry->d_lock);
611 shrink_dcache_parent(dentry);
612 spin_lock(&dentry->d_lock);
616 * Somebody else still using it?
618 * If it's a directory, we can't drop it
619 * for fear of somebody re-populating it
620 * with children (even though dropping it
621 * would make it unreachable from the root,
622 * we might still populate it if it was a
623 * working directory or similar).
624 * We also need to leave mountpoints alone,
627 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
628 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
629 spin_unlock(&dentry->d_lock);
635 spin_unlock(&dentry->d_lock);
638 EXPORT_SYMBOL(d_invalidate);
640 /* This must be called with d_lock held */
641 static inline void __dget_dlock(struct dentry *dentry)
643 dentry->d_lockref.count++;
646 static inline void __dget(struct dentry *dentry)
648 lockref_get(&dentry->d_lockref);
651 struct dentry *dget_parent(struct dentry *dentry)
657 * Do optimistic parent lookup without any
661 ret = ACCESS_ONCE(dentry->d_parent);
662 gotref = lockref_get_not_zero(&ret->d_lockref);
664 if (likely(gotref)) {
665 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
672 * Don't need rcu_dereference because we re-check it was correct under
676 ret = dentry->d_parent;
677 spin_lock(&ret->d_lock);
678 if (unlikely(ret != dentry->d_parent)) {
679 spin_unlock(&ret->d_lock);
684 BUG_ON(!ret->d_lockref.count);
685 ret->d_lockref.count++;
686 spin_unlock(&ret->d_lock);
689 EXPORT_SYMBOL(dget_parent);
692 * d_find_alias - grab a hashed alias of inode
693 * @inode: inode in question
694 * @want_discon: flag, used by d_splice_alias, to request
695 * that only a DISCONNECTED alias be returned.
697 * If inode has a hashed alias, or is a directory and has any alias,
698 * acquire the reference to alias and return it. Otherwise return NULL.
699 * Notice that if inode is a directory there can be only one alias and
700 * it can be unhashed only if it has no children, or if it is the root
703 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
704 * any other hashed alias over that one unless @want_discon is set,
705 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
707 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
709 struct dentry *alias, *discon_alias;
713 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
714 spin_lock(&alias->d_lock);
715 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
716 if (IS_ROOT(alias) &&
717 (alias->d_flags & DCACHE_DISCONNECTED)) {
718 discon_alias = alias;
719 } else if (!want_discon) {
721 spin_unlock(&alias->d_lock);
725 spin_unlock(&alias->d_lock);
728 alias = discon_alias;
729 spin_lock(&alias->d_lock);
730 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
731 if (IS_ROOT(alias) &&
732 (alias->d_flags & DCACHE_DISCONNECTED)) {
734 spin_unlock(&alias->d_lock);
738 spin_unlock(&alias->d_lock);
744 struct dentry *d_find_alias(struct inode *inode)
746 struct dentry *de = NULL;
748 if (!hlist_empty(&inode->i_dentry)) {
749 spin_lock(&inode->i_lock);
750 de = __d_find_alias(inode, 0);
751 spin_unlock(&inode->i_lock);
755 EXPORT_SYMBOL(d_find_alias);
758 * Try to kill dentries associated with this inode.
759 * WARNING: you must own a reference to inode.
761 void d_prune_aliases(struct inode *inode)
763 struct dentry *dentry;
765 spin_lock(&inode->i_lock);
766 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
767 spin_lock(&dentry->d_lock);
768 if (!dentry->d_lockref.count) {
770 * inform the fs via d_prune that this dentry
771 * is about to be unhashed and destroyed.
773 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
775 dentry->d_op->d_prune(dentry);
777 __dget_dlock(dentry);
779 spin_unlock(&dentry->d_lock);
780 spin_unlock(&inode->i_lock);
784 spin_unlock(&dentry->d_lock);
786 spin_unlock(&inode->i_lock);
788 EXPORT_SYMBOL(d_prune_aliases);
790 static void shrink_dentry_list(struct list_head *list)
792 struct dentry *dentry, *parent;
796 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
797 if (&dentry->d_lru == list)
801 * Get the dentry lock, and re-verify that the dentry is
802 * this on the shrinking list. If it is, we know that
803 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
805 spin_lock(&dentry->d_lock);
806 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
807 spin_unlock(&dentry->d_lock);
812 * The dispose list is isolated and dentries are not accounted
813 * to the LRU here, so we can simply remove it from the list
814 * here regardless of whether it is referenced or not.
816 d_shrink_del(dentry);
819 * We found an inuse dentry which was not removed from
820 * the LRU because of laziness during lookup. Do not free it.
822 if (dentry->d_lockref.count) {
823 spin_unlock(&dentry->d_lock);
828 parent = dentry_kill(dentry, 0);
830 * If dentry_kill returns NULL, we have nothing more to do.
836 if (unlikely(parent == dentry)) {
838 * trylocks have failed and d_lock has been held the
839 * whole time, so it could not have been added to any
840 * other lists. Just add it back to the shrink list.
843 d_shrink_add(dentry, list);
844 spin_unlock(&dentry->d_lock);
848 * We need to prune ancestors too. This is necessary to prevent
849 * quadratic behavior of shrink_dcache_parent(), but is also
850 * expected to be beneficial in reducing dentry cache
854 while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
855 dentry = dentry_kill(dentry, 1);
861 static enum lru_status
862 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
864 struct list_head *freeable = arg;
865 struct dentry *dentry = container_of(item, struct dentry, d_lru);
869 * we are inverting the lru lock/dentry->d_lock here,
870 * so use a trylock. If we fail to get the lock, just skip
873 if (!spin_trylock(&dentry->d_lock))
877 * Referenced dentries are still in use. If they have active
878 * counts, just remove them from the LRU. Otherwise give them
879 * another pass through the LRU.
881 if (dentry->d_lockref.count) {
882 d_lru_isolate(dentry);
883 spin_unlock(&dentry->d_lock);
887 if (dentry->d_flags & DCACHE_REFERENCED) {
888 dentry->d_flags &= ~DCACHE_REFERENCED;
889 spin_unlock(&dentry->d_lock);
892 * The list move itself will be made by the common LRU code. At
893 * this point, we've dropped the dentry->d_lock but keep the
894 * lru lock. This is safe to do, since every list movement is
895 * protected by the lru lock even if both locks are held.
897 * This is guaranteed by the fact that all LRU management
898 * functions are intermediated by the LRU API calls like
899 * list_lru_add and list_lru_del. List movement in this file
900 * only ever occur through this functions or through callbacks
901 * like this one, that are called from the LRU API.
903 * The only exceptions to this are functions like
904 * shrink_dentry_list, and code that first checks for the
905 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
906 * operating only with stack provided lists after they are
907 * properly isolated from the main list. It is thus, always a
913 d_lru_shrink_move(dentry, freeable);
914 spin_unlock(&dentry->d_lock);
920 * prune_dcache_sb - shrink the dcache
922 * @nr_to_scan : number of entries to try to free
923 * @nid: which node to scan for freeable entities
925 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
926 * done when we need more memory an called from the superblock shrinker
929 * This function may fail to free any resources if all the dentries are in
932 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
938 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
939 &dispose, &nr_to_scan);
940 shrink_dentry_list(&dispose);
944 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
945 spinlock_t *lru_lock, void *arg)
947 struct list_head *freeable = arg;
948 struct dentry *dentry = container_of(item, struct dentry, d_lru);
951 * we are inverting the lru lock/dentry->d_lock here,
952 * so use a trylock. If we fail to get the lock, just skip
955 if (!spin_trylock(&dentry->d_lock))
958 d_lru_shrink_move(dentry, freeable);
959 spin_unlock(&dentry->d_lock);
966 * shrink_dcache_sb - shrink dcache for a superblock
969 * Shrink the dcache for the specified super block. This is used to free
970 * the dcache before unmounting a file system.
972 void shrink_dcache_sb(struct super_block *sb)
979 freed = list_lru_walk(&sb->s_dentry_lru,
980 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
982 this_cpu_sub(nr_dentry_unused, freed);
983 shrink_dentry_list(&dispose);
986 EXPORT_SYMBOL(shrink_dcache_sb);
989 * enum d_walk_ret - action to talke during tree walk
990 * @D_WALK_CONTINUE: contrinue walk
991 * @D_WALK_QUIT: quit walk
992 * @D_WALK_NORETRY: quit when retry is needed
993 * @D_WALK_SKIP: skip this dentry and its children
1003 * d_walk - walk the dentry tree
1004 * @parent: start of walk
1005 * @data: data passed to @enter() and @finish()
1006 * @enter: callback when first entering the dentry
1007 * @finish: callback when successfully finished the walk
1009 * The @enter() and @finish() callbacks are called with d_lock held.
1011 static void d_walk(struct dentry *parent, void *data,
1012 enum d_walk_ret (*enter)(void *, struct dentry *),
1013 void (*finish)(void *))
1015 struct dentry *this_parent;
1016 struct list_head *next;
1018 enum d_walk_ret ret;
1022 read_seqbegin_or_lock(&rename_lock, &seq);
1023 this_parent = parent;
1024 spin_lock(&this_parent->d_lock);
1026 ret = enter(data, this_parent);
1028 case D_WALK_CONTINUE:
1033 case D_WALK_NORETRY:
1038 next = this_parent->d_subdirs.next;
1040 while (next != &this_parent->d_subdirs) {
1041 struct list_head *tmp = next;
1042 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1045 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1047 ret = enter(data, dentry);
1049 case D_WALK_CONTINUE:
1052 spin_unlock(&dentry->d_lock);
1054 case D_WALK_NORETRY:
1058 spin_unlock(&dentry->d_lock);
1062 if (!list_empty(&dentry->d_subdirs)) {
1063 spin_unlock(&this_parent->d_lock);
1064 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1065 this_parent = dentry;
1066 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1069 spin_unlock(&dentry->d_lock);
1072 * All done at this level ... ascend and resume the search.
1074 if (this_parent != parent) {
1075 struct dentry *child = this_parent;
1076 this_parent = child->d_parent;
1079 spin_unlock(&child->d_lock);
1080 spin_lock(&this_parent->d_lock);
1083 * might go back up the wrong parent if we have had a rename
1086 if (this_parent != child->d_parent ||
1087 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1088 need_seqretry(&rename_lock, seq)) {
1089 spin_unlock(&this_parent->d_lock);
1094 next = child->d_u.d_child.next;
1097 if (need_seqretry(&rename_lock, seq)) {
1098 spin_unlock(&this_parent->d_lock);
1105 spin_unlock(&this_parent->d_lock);
1106 done_seqretry(&rename_lock, seq);
1117 * Search for at least 1 mount point in the dentry's subdirs.
1118 * We descend to the next level whenever the d_subdirs
1119 * list is non-empty and continue searching.
1122 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1125 if (d_mountpoint(dentry)) {
1129 return D_WALK_CONTINUE;
1133 * have_submounts - check for mounts over a dentry
1134 * @parent: dentry to check.
1136 * Return true if the parent or its subdirectories contain
1139 int have_submounts(struct dentry *parent)
1143 d_walk(parent, &ret, check_mount, NULL);
1147 EXPORT_SYMBOL(have_submounts);
1150 * Called by mount code to set a mountpoint and check if the mountpoint is
1151 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1152 * subtree can become unreachable).
1154 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1155 * this reason take rename_lock and d_lock on dentry and ancestors.
1157 int d_set_mounted(struct dentry *dentry)
1161 write_seqlock(&rename_lock);
1162 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1163 /* Need exclusion wrt. check_submounts_and_drop() */
1164 spin_lock(&p->d_lock);
1165 if (unlikely(d_unhashed(p))) {
1166 spin_unlock(&p->d_lock);
1169 spin_unlock(&p->d_lock);
1171 spin_lock(&dentry->d_lock);
1172 if (!d_unlinked(dentry)) {
1173 dentry->d_flags |= DCACHE_MOUNTED;
1176 spin_unlock(&dentry->d_lock);
1178 write_sequnlock(&rename_lock);
1183 * Search the dentry child list of the specified parent,
1184 * and move any unused dentries to the end of the unused
1185 * list for prune_dcache(). We descend to the next level
1186 * whenever the d_subdirs list is non-empty and continue
1189 * It returns zero iff there are no unused children,
1190 * otherwise it returns the number of children moved to
1191 * the end of the unused list. This may not be the total
1192 * number of unused children, because select_parent can
1193 * drop the lock and return early due to latency
1197 struct select_data {
1198 struct dentry *start;
1199 struct list_head dispose;
1203 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1205 struct select_data *data = _data;
1206 enum d_walk_ret ret = D_WALK_CONTINUE;
1208 if (data->start == dentry)
1212 * move only zero ref count dentries to the dispose list.
1214 * Those which are presently on the shrink list, being processed
1215 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1216 * loop in shrink_dcache_parent() might not make any progress
1219 if (dentry->d_lockref.count) {
1220 dentry_lru_del(dentry);
1221 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1223 * We can't use d_lru_shrink_move() because we
1224 * need to get the global LRU lock and do the
1228 d_shrink_add(dentry, &data->dispose);
1230 ret = D_WALK_NORETRY;
1233 * We can return to the caller if we have found some (this
1234 * ensures forward progress). We'll be coming back to find
1237 if (data->found && need_resched())
1244 * shrink_dcache_parent - prune dcache
1245 * @parent: parent of entries to prune
1247 * Prune the dcache to remove unused children of the parent dentry.
1249 void shrink_dcache_parent(struct dentry *parent)
1252 struct select_data data;
1254 INIT_LIST_HEAD(&data.dispose);
1255 data.start = parent;
1258 d_walk(parent, &data, select_collect, NULL);
1262 shrink_dentry_list(&data.dispose);
1266 EXPORT_SYMBOL(shrink_dcache_parent);
1268 static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry)
1270 struct select_data *data = _data;
1271 enum d_walk_ret ret = D_WALK_CONTINUE;
1273 if (dentry->d_lockref.count) {
1274 dentry_lru_del(dentry);
1275 if (likely(!list_empty(&dentry->d_subdirs)))
1277 if (dentry == data->start && dentry->d_lockref.count == 1)
1280 "BUG: Dentry %p{i=%lx,n=%s}"
1281 " still in use (%d)"
1282 " [unmount of %s %s]\n",
1285 dentry->d_inode->i_ino : 0UL,
1286 dentry->d_name.name,
1287 dentry->d_lockref.count,
1288 dentry->d_sb->s_type->name,
1289 dentry->d_sb->s_id);
1291 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1293 * We can't use d_lru_shrink_move() because we
1294 * need to get the global LRU lock and do the
1297 if (dentry->d_flags & DCACHE_LRU_LIST)
1299 d_shrink_add(dentry, &data->dispose);
1301 ret = D_WALK_NORETRY;
1304 if (data->found && need_resched())
1310 * destroy the dentries attached to a superblock on unmounting
1312 void shrink_dcache_for_umount(struct super_block *sb)
1314 struct dentry *dentry;
1316 if (down_read_trylock(&sb->s_umount))
1319 dentry = sb->s_root;
1322 struct select_data data;
1324 INIT_LIST_HEAD(&data.dispose);
1325 data.start = dentry;
1328 d_walk(dentry, &data, umount_collect, NULL);
1332 shrink_dentry_list(&data.dispose);
1338 while (!hlist_bl_empty(&sb->s_anon)) {
1339 struct select_data data;
1340 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1342 INIT_LIST_HEAD(&data.dispose);
1346 d_walk(dentry, &data, umount_collect, NULL);
1348 shrink_dentry_list(&data.dispose);
1353 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1355 struct select_data *data = _data;
1357 if (d_mountpoint(dentry)) {
1358 data->found = -EBUSY;
1362 return select_collect(_data, dentry);
1365 static void check_and_drop(void *_data)
1367 struct select_data *data = _data;
1369 if (d_mountpoint(data->start))
1370 data->found = -EBUSY;
1372 __d_drop(data->start);
1376 * check_submounts_and_drop - prune dcache, check for submounts and drop
1378 * All done as a single atomic operation relative to has_unlinked_ancestor().
1379 * Returns 0 if successfully unhashed @parent. If there were submounts then
1382 * @dentry: dentry to prune and drop
1384 int check_submounts_and_drop(struct dentry *dentry)
1388 /* Negative dentries can be dropped without further checks */
1389 if (!dentry->d_inode) {
1395 struct select_data data;
1397 INIT_LIST_HEAD(&data.dispose);
1398 data.start = dentry;
1401 d_walk(dentry, &data, check_and_collect, check_and_drop);
1404 if (!list_empty(&data.dispose))
1405 shrink_dentry_list(&data.dispose);
1416 EXPORT_SYMBOL(check_submounts_and_drop);
1419 * __d_alloc - allocate a dcache entry
1420 * @sb: filesystem it will belong to
1421 * @name: qstr of the name
1423 * Allocates a dentry. It returns %NULL if there is insufficient memory
1424 * available. On a success the dentry is returned. The name passed in is
1425 * copied and the copy passed in may be reused after this call.
1428 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1430 struct dentry *dentry;
1433 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1438 * We guarantee that the inline name is always NUL-terminated.
1439 * This way the memcpy() done by the name switching in rename
1440 * will still always have a NUL at the end, even if we might
1441 * be overwriting an internal NUL character
1443 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1444 if (name->len > DNAME_INLINE_LEN-1) {
1445 dname = kmalloc(name->len + 1, GFP_KERNEL);
1447 kmem_cache_free(dentry_cache, dentry);
1451 dname = dentry->d_iname;
1454 dentry->d_name.len = name->len;
1455 dentry->d_name.hash = name->hash;
1456 memcpy(dname, name->name, name->len);
1457 dname[name->len] = 0;
1459 /* Make sure we always see the terminating NUL character */
1461 dentry->d_name.name = dname;
1463 dentry->d_lockref.count = 1;
1464 dentry->d_flags = 0;
1465 spin_lock_init(&dentry->d_lock);
1466 seqcount_init(&dentry->d_seq);
1467 dentry->d_inode = NULL;
1468 dentry->d_parent = dentry;
1470 dentry->d_op = NULL;
1471 dentry->d_fsdata = NULL;
1472 INIT_HLIST_BL_NODE(&dentry->d_hash);
1473 INIT_LIST_HEAD(&dentry->d_lru);
1474 INIT_LIST_HEAD(&dentry->d_subdirs);
1475 INIT_HLIST_NODE(&dentry->d_alias);
1476 INIT_LIST_HEAD(&dentry->d_u.d_child);
1477 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1479 this_cpu_inc(nr_dentry);
1485 * d_alloc - allocate a dcache entry
1486 * @parent: parent of entry to allocate
1487 * @name: qstr of the name
1489 * Allocates a dentry. It returns %NULL if there is insufficient memory
1490 * available. On a success the dentry is returned. The name passed in is
1491 * copied and the copy passed in may be reused after this call.
1493 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1495 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1499 spin_lock(&parent->d_lock);
1501 * don't need child lock because it is not subject
1502 * to concurrency here
1504 __dget_dlock(parent);
1505 dentry->d_parent = parent;
1506 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1507 spin_unlock(&parent->d_lock);
1511 EXPORT_SYMBOL(d_alloc);
1514 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1515 * @sb: the superblock
1516 * @name: qstr of the name
1518 * For a filesystem that just pins its dentries in memory and never
1519 * performs lookups at all, return an unhashed IS_ROOT dentry.
1521 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1523 return __d_alloc(sb, name);
1525 EXPORT_SYMBOL(d_alloc_pseudo);
1527 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1532 q.len = strlen(name);
1533 q.hash = full_name_hash(q.name, q.len);
1534 return d_alloc(parent, &q);
1536 EXPORT_SYMBOL(d_alloc_name);
1538 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1540 WARN_ON_ONCE(dentry->d_op);
1541 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1543 DCACHE_OP_REVALIDATE |
1544 DCACHE_OP_WEAK_REVALIDATE |
1545 DCACHE_OP_DELETE ));
1550 dentry->d_flags |= DCACHE_OP_HASH;
1552 dentry->d_flags |= DCACHE_OP_COMPARE;
1553 if (op->d_revalidate)
1554 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1555 if (op->d_weak_revalidate)
1556 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1558 dentry->d_flags |= DCACHE_OP_DELETE;
1560 dentry->d_flags |= DCACHE_OP_PRUNE;
1563 EXPORT_SYMBOL(d_set_d_op);
1565 static unsigned d_flags_for_inode(struct inode *inode)
1567 unsigned add_flags = DCACHE_FILE_TYPE;
1570 return DCACHE_MISS_TYPE;
1572 if (S_ISDIR(inode->i_mode)) {
1573 add_flags = DCACHE_DIRECTORY_TYPE;
1574 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1575 if (unlikely(!inode->i_op->lookup))
1576 add_flags = DCACHE_AUTODIR_TYPE;
1578 inode->i_opflags |= IOP_LOOKUP;
1580 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1581 if (unlikely(inode->i_op->follow_link))
1582 add_flags = DCACHE_SYMLINK_TYPE;
1584 inode->i_opflags |= IOP_NOFOLLOW;
1587 if (unlikely(IS_AUTOMOUNT(inode)))
1588 add_flags |= DCACHE_NEED_AUTOMOUNT;
1592 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1594 unsigned add_flags = d_flags_for_inode(inode);
1596 spin_lock(&dentry->d_lock);
1597 __d_set_type(dentry, add_flags);
1599 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1600 dentry->d_inode = inode;
1601 dentry_rcuwalk_barrier(dentry);
1602 spin_unlock(&dentry->d_lock);
1603 fsnotify_d_instantiate(dentry, inode);
1607 * d_instantiate - fill in inode information for a dentry
1608 * @entry: dentry to complete
1609 * @inode: inode to attach to this dentry
1611 * Fill in inode information in the entry.
1613 * This turns negative dentries into productive full members
1616 * NOTE! This assumes that the inode count has been incremented
1617 * (or otherwise set) by the caller to indicate that it is now
1618 * in use by the dcache.
1621 void d_instantiate(struct dentry *entry, struct inode * inode)
1623 BUG_ON(!hlist_unhashed(&entry->d_alias));
1625 spin_lock(&inode->i_lock);
1626 __d_instantiate(entry, inode);
1628 spin_unlock(&inode->i_lock);
1629 security_d_instantiate(entry, inode);
1631 EXPORT_SYMBOL(d_instantiate);
1634 * d_instantiate_unique - instantiate a non-aliased dentry
1635 * @entry: dentry to instantiate
1636 * @inode: inode to attach to this dentry
1638 * Fill in inode information in the entry. On success, it returns NULL.
1639 * If an unhashed alias of "entry" already exists, then we return the
1640 * aliased dentry instead and drop one reference to inode.
1642 * Note that in order to avoid conflicts with rename() etc, the caller
1643 * had better be holding the parent directory semaphore.
1645 * This also assumes that the inode count has been incremented
1646 * (or otherwise set) by the caller to indicate that it is now
1647 * in use by the dcache.
1649 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1650 struct inode *inode)
1652 struct dentry *alias;
1653 int len = entry->d_name.len;
1654 const char *name = entry->d_name.name;
1655 unsigned int hash = entry->d_name.hash;
1658 __d_instantiate(entry, NULL);
1662 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1664 * Don't need alias->d_lock here, because aliases with
1665 * d_parent == entry->d_parent are not subject to name or
1666 * parent changes, because the parent inode i_mutex is held.
1668 if (alias->d_name.hash != hash)
1670 if (alias->d_parent != entry->d_parent)
1672 if (alias->d_name.len != len)
1674 if (dentry_cmp(alias, name, len))
1680 __d_instantiate(entry, inode);
1684 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1686 struct dentry *result;
1688 BUG_ON(!hlist_unhashed(&entry->d_alias));
1691 spin_lock(&inode->i_lock);
1692 result = __d_instantiate_unique(entry, inode);
1694 spin_unlock(&inode->i_lock);
1697 security_d_instantiate(entry, inode);
1701 BUG_ON(!d_unhashed(result));
1706 EXPORT_SYMBOL(d_instantiate_unique);
1709 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1710 * @entry: dentry to complete
1711 * @inode: inode to attach to this dentry
1713 * Fill in inode information in the entry. If a directory alias is found, then
1714 * return an error (and drop inode). Together with d_materialise_unique() this
1715 * guarantees that a directory inode may never have more than one alias.
1717 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1719 BUG_ON(!hlist_unhashed(&entry->d_alias));
1721 spin_lock(&inode->i_lock);
1722 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1723 spin_unlock(&inode->i_lock);
1727 __d_instantiate(entry, inode);
1728 spin_unlock(&inode->i_lock);
1729 security_d_instantiate(entry, inode);
1733 EXPORT_SYMBOL(d_instantiate_no_diralias);
1735 struct dentry *d_make_root(struct inode *root_inode)
1737 struct dentry *res = NULL;
1740 static const struct qstr name = QSTR_INIT("/", 1);
1742 res = __d_alloc(root_inode->i_sb, &name);
1744 d_instantiate(res, root_inode);
1750 EXPORT_SYMBOL(d_make_root);
1752 static struct dentry * __d_find_any_alias(struct inode *inode)
1754 struct dentry *alias;
1756 if (hlist_empty(&inode->i_dentry))
1758 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1764 * d_find_any_alias - find any alias for a given inode
1765 * @inode: inode to find an alias for
1767 * If any aliases exist for the given inode, take and return a
1768 * reference for one of them. If no aliases exist, return %NULL.
1770 struct dentry *d_find_any_alias(struct inode *inode)
1774 spin_lock(&inode->i_lock);
1775 de = __d_find_any_alias(inode);
1776 spin_unlock(&inode->i_lock);
1779 EXPORT_SYMBOL(d_find_any_alias);
1782 * d_obtain_alias - find or allocate a dentry for a given inode
1783 * @inode: inode to allocate the dentry for
1785 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1786 * similar open by handle operations. The returned dentry may be anonymous,
1787 * or may have a full name (if the inode was already in the cache).
1789 * When called on a directory inode, we must ensure that the inode only ever
1790 * has one dentry. If a dentry is found, that is returned instead of
1791 * allocating a new one.
1793 * On successful return, the reference to the inode has been transferred
1794 * to the dentry. In case of an error the reference on the inode is released.
1795 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1796 * be passed in and will be the error will be propagate to the return value,
1797 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1799 struct dentry *d_obtain_alias(struct inode *inode)
1801 static const struct qstr anonstring = QSTR_INIT("/", 1);
1807 return ERR_PTR(-ESTALE);
1809 return ERR_CAST(inode);
1811 res = d_find_any_alias(inode);
1815 tmp = __d_alloc(inode->i_sb, &anonstring);
1817 res = ERR_PTR(-ENOMEM);
1821 spin_lock(&inode->i_lock);
1822 res = __d_find_any_alias(inode);
1824 spin_unlock(&inode->i_lock);
1829 /* attach a disconnected dentry */
1830 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1832 spin_lock(&tmp->d_lock);
1833 tmp->d_inode = inode;
1834 tmp->d_flags |= add_flags;
1835 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1836 hlist_bl_lock(&tmp->d_sb->s_anon);
1837 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1838 hlist_bl_unlock(&tmp->d_sb->s_anon);
1839 spin_unlock(&tmp->d_lock);
1840 spin_unlock(&inode->i_lock);
1841 security_d_instantiate(tmp, inode);
1846 if (res && !IS_ERR(res))
1847 security_d_instantiate(res, inode);
1851 EXPORT_SYMBOL(d_obtain_alias);
1854 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1855 * @inode: the inode which may have a disconnected dentry
1856 * @dentry: a negative dentry which we want to point to the inode.
1858 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1859 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1860 * and return it, else simply d_add the inode to the dentry and return NULL.
1862 * This is needed in the lookup routine of any filesystem that is exportable
1863 * (via knfsd) so that we can build dcache paths to directories effectively.
1865 * If a dentry was found and moved, then it is returned. Otherwise NULL
1866 * is returned. This matches the expected return value of ->lookup.
1868 * Cluster filesystems may call this function with a negative, hashed dentry.
1869 * In that case, we know that the inode will be a regular file, and also this
1870 * will only occur during atomic_open. So we need to check for the dentry
1871 * being already hashed only in the final case.
1873 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1875 struct dentry *new = NULL;
1878 return ERR_CAST(inode);
1880 if (inode && S_ISDIR(inode->i_mode)) {
1881 spin_lock(&inode->i_lock);
1882 new = __d_find_alias(inode, 1);
1884 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1885 spin_unlock(&inode->i_lock);
1886 security_d_instantiate(new, inode);
1887 d_move(new, dentry);
1890 /* already taking inode->i_lock, so d_add() by hand */
1891 __d_instantiate(dentry, inode);
1892 spin_unlock(&inode->i_lock);
1893 security_d_instantiate(dentry, inode);
1897 d_instantiate(dentry, inode);
1898 if (d_unhashed(dentry))
1903 EXPORT_SYMBOL(d_splice_alias);
1906 * d_add_ci - lookup or allocate new dentry with case-exact name
1907 * @inode: the inode case-insensitive lookup has found
1908 * @dentry: the negative dentry that was passed to the parent's lookup func
1909 * @name: the case-exact name to be associated with the returned dentry
1911 * This is to avoid filling the dcache with case-insensitive names to the
1912 * same inode, only the actual correct case is stored in the dcache for
1913 * case-insensitive filesystems.
1915 * For a case-insensitive lookup match and if the the case-exact dentry
1916 * already exists in in the dcache, use it and return it.
1918 * If no entry exists with the exact case name, allocate new dentry with
1919 * the exact case, and return the spliced entry.
1921 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1924 struct dentry *found;
1928 * First check if a dentry matching the name already exists,
1929 * if not go ahead and create it now.
1931 found = d_hash_and_lookup(dentry->d_parent, name);
1932 if (unlikely(IS_ERR(found)))
1935 new = d_alloc(dentry->d_parent, name);
1937 found = ERR_PTR(-ENOMEM);
1941 found = d_splice_alias(inode, new);
1950 * If a matching dentry exists, and it's not negative use it.
1952 * Decrement the reference count to balance the iget() done
1955 if (found->d_inode) {
1956 if (unlikely(found->d_inode != inode)) {
1957 /* This can't happen because bad inodes are unhashed. */
1958 BUG_ON(!is_bad_inode(inode));
1959 BUG_ON(!is_bad_inode(found->d_inode));
1966 * Negative dentry: instantiate it unless the inode is a directory and
1967 * already has a dentry.
1969 new = d_splice_alias(inode, found);
1980 EXPORT_SYMBOL(d_add_ci);
1983 * Do the slow-case of the dentry name compare.
1985 * Unlike the dentry_cmp() function, we need to atomically
1986 * load the name and length information, so that the
1987 * filesystem can rely on them, and can use the 'name' and
1988 * 'len' information without worrying about walking off the
1989 * end of memory etc.
1991 * Thus the read_seqcount_retry() and the "duplicate" info
1992 * in arguments (the low-level filesystem should not look
1993 * at the dentry inode or name contents directly, since
1994 * rename can change them while we're in RCU mode).
1996 enum slow_d_compare {
2002 static noinline enum slow_d_compare slow_dentry_cmp(
2003 const struct dentry *parent,
2004 struct dentry *dentry,
2006 const struct qstr *name)
2008 int tlen = dentry->d_name.len;
2009 const char *tname = dentry->d_name.name;
2011 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2013 return D_COMP_SEQRETRY;
2015 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2016 return D_COMP_NOMATCH;
2021 * __d_lookup_rcu - search for a dentry (racy, store-free)
2022 * @parent: parent dentry
2023 * @name: qstr of name we wish to find
2024 * @seqp: returns d_seq value at the point where the dentry was found
2025 * Returns: dentry, or NULL
2027 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2028 * resolution (store-free path walking) design described in
2029 * Documentation/filesystems/path-lookup.txt.
2031 * This is not to be used outside core vfs.
2033 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2034 * held, and rcu_read_lock held. The returned dentry must not be stored into
2035 * without taking d_lock and checking d_seq sequence count against @seq
2038 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2041 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2042 * the returned dentry, so long as its parent's seqlock is checked after the
2043 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2044 * is formed, giving integrity down the path walk.
2046 * NOTE! The caller *has* to check the resulting dentry against the sequence
2047 * number we've returned before using any of the resulting dentry state!
2049 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2050 const struct qstr *name,
2053 u64 hashlen = name->hash_len;
2054 const unsigned char *str = name->name;
2055 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2056 struct hlist_bl_node *node;
2057 struct dentry *dentry;
2060 * Note: There is significant duplication with __d_lookup_rcu which is
2061 * required to prevent single threaded performance regressions
2062 * especially on architectures where smp_rmb (in seqcounts) are costly.
2063 * Keep the two functions in sync.
2067 * The hash list is protected using RCU.
2069 * Carefully use d_seq when comparing a candidate dentry, to avoid
2070 * races with d_move().
2072 * It is possible that concurrent renames can mess up our list
2073 * walk here and result in missing our dentry, resulting in the
2074 * false-negative result. d_lookup() protects against concurrent
2075 * renames using rename_lock seqlock.
2077 * See Documentation/filesystems/path-lookup.txt for more details.
2079 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2084 * The dentry sequence count protects us from concurrent
2085 * renames, and thus protects parent and name fields.
2087 * The caller must perform a seqcount check in order
2088 * to do anything useful with the returned dentry.
2090 * NOTE! We do a "raw" seqcount_begin here. That means that
2091 * we don't wait for the sequence count to stabilize if it
2092 * is in the middle of a sequence change. If we do the slow
2093 * dentry compare, we will do seqretries until it is stable,
2094 * and if we end up with a successful lookup, we actually
2095 * want to exit RCU lookup anyway.
2097 seq = raw_seqcount_begin(&dentry->d_seq);
2098 if (dentry->d_parent != parent)
2100 if (d_unhashed(dentry))
2103 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2104 if (dentry->d_name.hash != hashlen_hash(hashlen))
2107 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2110 case D_COMP_NOMATCH:
2117 if (dentry->d_name.hash_len != hashlen)
2120 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2127 * d_lookup - search for a dentry
2128 * @parent: parent dentry
2129 * @name: qstr of name we wish to find
2130 * Returns: dentry, or NULL
2132 * d_lookup searches the children of the parent dentry for the name in
2133 * question. If the dentry is found its reference count is incremented and the
2134 * dentry is returned. The caller must use dput to free the entry when it has
2135 * finished using it. %NULL is returned if the dentry does not exist.
2137 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2139 struct dentry *dentry;
2143 seq = read_seqbegin(&rename_lock);
2144 dentry = __d_lookup(parent, name);
2147 } while (read_seqretry(&rename_lock, seq));
2150 EXPORT_SYMBOL(d_lookup);
2153 * __d_lookup - search for a dentry (racy)
2154 * @parent: parent dentry
2155 * @name: qstr of name we wish to find
2156 * Returns: dentry, or NULL
2158 * __d_lookup is like d_lookup, however it may (rarely) return a
2159 * false-negative result due to unrelated rename activity.
2161 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2162 * however it must be used carefully, eg. with a following d_lookup in
2163 * the case of failure.
2165 * __d_lookup callers must be commented.
2167 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2169 unsigned int len = name->len;
2170 unsigned int hash = name->hash;
2171 const unsigned char *str = name->name;
2172 struct hlist_bl_head *b = d_hash(parent, hash);
2173 struct hlist_bl_node *node;
2174 struct dentry *found = NULL;
2175 struct dentry *dentry;
2178 * Note: There is significant duplication with __d_lookup_rcu which is
2179 * required to prevent single threaded performance regressions
2180 * especially on architectures where smp_rmb (in seqcounts) are costly.
2181 * Keep the two functions in sync.
2185 * The hash list is protected using RCU.
2187 * Take d_lock when comparing a candidate dentry, to avoid races
2190 * It is possible that concurrent renames can mess up our list
2191 * walk here and result in missing our dentry, resulting in the
2192 * false-negative result. d_lookup() protects against concurrent
2193 * renames using rename_lock seqlock.
2195 * See Documentation/filesystems/path-lookup.txt for more details.
2199 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2201 if (dentry->d_name.hash != hash)
2204 spin_lock(&dentry->d_lock);
2205 if (dentry->d_parent != parent)
2207 if (d_unhashed(dentry))
2211 * It is safe to compare names since d_move() cannot
2212 * change the qstr (protected by d_lock).
2214 if (parent->d_flags & DCACHE_OP_COMPARE) {
2215 int tlen = dentry->d_name.len;
2216 const char *tname = dentry->d_name.name;
2217 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2220 if (dentry->d_name.len != len)
2222 if (dentry_cmp(dentry, str, len))
2226 dentry->d_lockref.count++;
2228 spin_unlock(&dentry->d_lock);
2231 spin_unlock(&dentry->d_lock);
2239 * d_hash_and_lookup - hash the qstr then search for a dentry
2240 * @dir: Directory to search in
2241 * @name: qstr of name we wish to find
2243 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2245 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2248 * Check for a fs-specific hash function. Note that we must
2249 * calculate the standard hash first, as the d_op->d_hash()
2250 * routine may choose to leave the hash value unchanged.
2252 name->hash = full_name_hash(name->name, name->len);
2253 if (dir->d_flags & DCACHE_OP_HASH) {
2254 int err = dir->d_op->d_hash(dir, name);
2255 if (unlikely(err < 0))
2256 return ERR_PTR(err);
2258 return d_lookup(dir, name);
2260 EXPORT_SYMBOL(d_hash_and_lookup);
2263 * d_validate - verify dentry provided from insecure source (deprecated)
2264 * @dentry: The dentry alleged to be valid child of @dparent
2265 * @dparent: The parent dentry (known to be valid)
2267 * An insecure source has sent us a dentry, here we verify it and dget() it.
2268 * This is used by ncpfs in its readdir implementation.
2269 * Zero is returned in the dentry is invalid.
2271 * This function is slow for big directories, and deprecated, do not use it.
2273 int d_validate(struct dentry *dentry, struct dentry *dparent)
2275 struct dentry *child;
2277 spin_lock(&dparent->d_lock);
2278 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2279 if (dentry == child) {
2280 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2281 __dget_dlock(dentry);
2282 spin_unlock(&dentry->d_lock);
2283 spin_unlock(&dparent->d_lock);
2287 spin_unlock(&dparent->d_lock);
2291 EXPORT_SYMBOL(d_validate);
2294 * When a file is deleted, we have two options:
2295 * - turn this dentry into a negative dentry
2296 * - unhash this dentry and free it.
2298 * Usually, we want to just turn this into
2299 * a negative dentry, but if anybody else is
2300 * currently using the dentry or the inode
2301 * we can't do that and we fall back on removing
2302 * it from the hash queues and waiting for
2303 * it to be deleted later when it has no users
2307 * d_delete - delete a dentry
2308 * @dentry: The dentry to delete
2310 * Turn the dentry into a negative dentry if possible, otherwise
2311 * remove it from the hash queues so it can be deleted later
2314 void d_delete(struct dentry * dentry)
2316 struct inode *inode;
2319 * Are we the only user?
2322 spin_lock(&dentry->d_lock);
2323 inode = dentry->d_inode;
2324 isdir = S_ISDIR(inode->i_mode);
2325 if (dentry->d_lockref.count == 1) {
2326 if (!spin_trylock(&inode->i_lock)) {
2327 spin_unlock(&dentry->d_lock);
2331 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2332 dentry_unlink_inode(dentry);
2333 fsnotify_nameremove(dentry, isdir);
2337 if (!d_unhashed(dentry))
2340 spin_unlock(&dentry->d_lock);
2342 fsnotify_nameremove(dentry, isdir);
2344 EXPORT_SYMBOL(d_delete);
2346 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2348 BUG_ON(!d_unhashed(entry));
2350 entry->d_flags |= DCACHE_RCUACCESS;
2351 hlist_bl_add_head_rcu(&entry->d_hash, b);
2355 static void _d_rehash(struct dentry * entry)
2357 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2361 * d_rehash - add an entry back to the hash
2362 * @entry: dentry to add to the hash
2364 * Adds a dentry to the hash according to its name.
2367 void d_rehash(struct dentry * entry)
2369 spin_lock(&entry->d_lock);
2371 spin_unlock(&entry->d_lock);
2373 EXPORT_SYMBOL(d_rehash);
2376 * dentry_update_name_case - update case insensitive dentry with a new name
2377 * @dentry: dentry to be updated
2380 * Update a case insensitive dentry with new case of name.
2382 * dentry must have been returned by d_lookup with name @name. Old and new
2383 * name lengths must match (ie. no d_compare which allows mismatched name
2386 * Parent inode i_mutex must be held over d_lookup and into this call (to
2387 * keep renames and concurrent inserts, and readdir(2) away).
2389 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2391 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2392 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2394 spin_lock(&dentry->d_lock);
2395 write_seqcount_begin(&dentry->d_seq);
2396 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2397 write_seqcount_end(&dentry->d_seq);
2398 spin_unlock(&dentry->d_lock);
2400 EXPORT_SYMBOL(dentry_update_name_case);
2402 static void switch_names(struct dentry *dentry, struct dentry *target)
2404 if (dname_external(target)) {
2405 if (dname_external(dentry)) {
2407 * Both external: swap the pointers
2409 swap(target->d_name.name, dentry->d_name.name);
2412 * dentry:internal, target:external. Steal target's
2413 * storage and make target internal.
2415 memcpy(target->d_iname, dentry->d_name.name,
2416 dentry->d_name.len + 1);
2417 dentry->d_name.name = target->d_name.name;
2418 target->d_name.name = target->d_iname;
2421 if (dname_external(dentry)) {
2423 * dentry:external, target:internal. Give dentry's
2424 * storage to target and make dentry internal
2426 memcpy(dentry->d_iname, target->d_name.name,
2427 target->d_name.len + 1);
2428 target->d_name.name = dentry->d_name.name;
2429 dentry->d_name.name = dentry->d_iname;
2432 * Both are internal.
2435 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2436 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2437 swap(((long *) &dentry->d_iname)[i],
2438 ((long *) &target->d_iname)[i]);
2442 swap(dentry->d_name.len, target->d_name.len);
2445 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2448 * XXXX: do we really need to take target->d_lock?
2450 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2451 spin_lock(&target->d_parent->d_lock);
2453 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2454 spin_lock(&dentry->d_parent->d_lock);
2455 spin_lock_nested(&target->d_parent->d_lock,
2456 DENTRY_D_LOCK_NESTED);
2458 spin_lock(&target->d_parent->d_lock);
2459 spin_lock_nested(&dentry->d_parent->d_lock,
2460 DENTRY_D_LOCK_NESTED);
2463 if (target < dentry) {
2464 spin_lock_nested(&target->d_lock, 2);
2465 spin_lock_nested(&dentry->d_lock, 3);
2467 spin_lock_nested(&dentry->d_lock, 2);
2468 spin_lock_nested(&target->d_lock, 3);
2472 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2473 struct dentry *target)
2475 if (target->d_parent != dentry->d_parent)
2476 spin_unlock(&dentry->d_parent->d_lock);
2477 if (target->d_parent != target)
2478 spin_unlock(&target->d_parent->d_lock);
2482 * When switching names, the actual string doesn't strictly have to
2483 * be preserved in the target - because we're dropping the target
2484 * anyway. As such, we can just do a simple memcpy() to copy over
2485 * the new name before we switch.
2487 * Note that we have to be a lot more careful about getting the hash
2488 * switched - we have to switch the hash value properly even if it
2489 * then no longer matches the actual (corrupted) string of the target.
2490 * The hash value has to match the hash queue that the dentry is on..
2493 * __d_move - move a dentry
2494 * @dentry: entry to move
2495 * @target: new dentry
2496 * @exchange: exchange the two dentries
2498 * Update the dcache to reflect the move of a file name. Negative
2499 * dcache entries should not be moved in this way. Caller must hold
2500 * rename_lock, the i_mutex of the source and target directories,
2501 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2503 static void __d_move(struct dentry *dentry, struct dentry *target,
2506 if (!dentry->d_inode)
2507 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2509 BUG_ON(d_ancestor(dentry, target));
2510 BUG_ON(d_ancestor(target, dentry));
2512 dentry_lock_for_move(dentry, target);
2514 write_seqcount_begin(&dentry->d_seq);
2515 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2517 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2520 * Move the dentry to the target hash queue. Don't bother checking
2521 * for the same hash queue because of how unlikely it is.
2524 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2527 * Unhash the target (d_delete() is not usable here). If exchanging
2528 * the two dentries, then rehash onto the other's hash queue.
2533 d_hash(dentry->d_parent, dentry->d_name.hash));
2536 list_del(&dentry->d_u.d_child);
2537 list_del(&target->d_u.d_child);
2539 /* Switch the names.. */
2540 switch_names(dentry, target);
2541 swap(dentry->d_name.hash, target->d_name.hash);
2543 /* ... and switch the parents */
2544 if (IS_ROOT(dentry)) {
2545 dentry->d_parent = target->d_parent;
2546 target->d_parent = target;
2547 INIT_LIST_HEAD(&target->d_u.d_child);
2549 swap(dentry->d_parent, target->d_parent);
2551 /* And add them back to the (new) parent lists */
2552 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2555 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2557 write_seqcount_end(&target->d_seq);
2558 write_seqcount_end(&dentry->d_seq);
2560 dentry_unlock_parents_for_move(dentry, target);
2562 fsnotify_d_move(target);
2563 spin_unlock(&target->d_lock);
2564 fsnotify_d_move(dentry);
2565 spin_unlock(&dentry->d_lock);
2569 * d_move - move a dentry
2570 * @dentry: entry to move
2571 * @target: new dentry
2573 * Update the dcache to reflect the move of a file name. Negative
2574 * dcache entries should not be moved in this way. See the locking
2575 * requirements for __d_move.
2577 void d_move(struct dentry *dentry, struct dentry *target)
2579 write_seqlock(&rename_lock);
2580 __d_move(dentry, target, false);
2581 write_sequnlock(&rename_lock);
2583 EXPORT_SYMBOL(d_move);
2586 * d_exchange - exchange two dentries
2587 * @dentry1: first dentry
2588 * @dentry2: second dentry
2590 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2592 write_seqlock(&rename_lock);
2594 WARN_ON(!dentry1->d_inode);
2595 WARN_ON(!dentry2->d_inode);
2596 WARN_ON(IS_ROOT(dentry1));
2597 WARN_ON(IS_ROOT(dentry2));
2599 __d_move(dentry1, dentry2, true);
2601 write_sequnlock(&rename_lock);
2605 * d_ancestor - search for an ancestor
2606 * @p1: ancestor dentry
2609 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2610 * an ancestor of p2, else NULL.
2612 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2616 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2617 if (p->d_parent == p1)
2624 * This helper attempts to cope with remotely renamed directories
2626 * It assumes that the caller is already holding
2627 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2629 * Note: If ever the locking in lock_rename() changes, then please
2630 * remember to update this too...
2632 static struct dentry *__d_unalias(struct inode *inode,
2633 struct dentry *dentry, struct dentry *alias)
2635 struct mutex *m1 = NULL, *m2 = NULL;
2636 struct dentry *ret = ERR_PTR(-EBUSY);
2638 /* If alias and dentry share a parent, then no extra locks required */
2639 if (alias->d_parent == dentry->d_parent)
2642 /* See lock_rename() */
2643 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2645 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2646 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2648 m2 = &alias->d_parent->d_inode->i_mutex;
2650 if (likely(!d_mountpoint(alias))) {
2651 __d_move(alias, dentry, false);
2655 spin_unlock(&inode->i_lock);
2664 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2665 * named dentry in place of the dentry to be replaced.
2666 * returns with anon->d_lock held!
2668 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2670 struct dentry *dparent;
2672 dentry_lock_for_move(anon, dentry);
2674 write_seqcount_begin(&dentry->d_seq);
2675 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2677 dparent = dentry->d_parent;
2679 switch_names(dentry, anon);
2680 swap(dentry->d_name.hash, anon->d_name.hash);
2682 dentry->d_parent = dentry;
2683 list_del_init(&dentry->d_u.d_child);
2684 anon->d_parent = dparent;
2685 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2687 write_seqcount_end(&dentry->d_seq);
2688 write_seqcount_end(&anon->d_seq);
2690 dentry_unlock_parents_for_move(anon, dentry);
2691 spin_unlock(&dentry->d_lock);
2693 /* anon->d_lock still locked, returns locked */
2697 * d_materialise_unique - introduce an inode into the tree
2698 * @dentry: candidate dentry
2699 * @inode: inode to bind to the dentry, to which aliases may be attached
2701 * Introduces an dentry into the tree, substituting an extant disconnected
2702 * root directory alias in its place if there is one. Caller must hold the
2703 * i_mutex of the parent directory.
2705 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2707 struct dentry *actual;
2709 BUG_ON(!d_unhashed(dentry));
2713 __d_instantiate(dentry, NULL);
2718 spin_lock(&inode->i_lock);
2720 if (S_ISDIR(inode->i_mode)) {
2721 struct dentry *alias;
2723 /* Does an aliased dentry already exist? */
2724 alias = __d_find_alias(inode, 0);
2727 write_seqlock(&rename_lock);
2729 if (d_ancestor(alias, dentry)) {
2730 /* Check for loops */
2731 actual = ERR_PTR(-ELOOP);
2732 spin_unlock(&inode->i_lock);
2733 } else if (IS_ROOT(alias)) {
2734 /* Is this an anonymous mountpoint that we
2735 * could splice into our tree? */
2736 __d_materialise_dentry(dentry, alias);
2737 write_sequnlock(&rename_lock);
2741 /* Nope, but we must(!) avoid directory
2742 * aliasing. This drops inode->i_lock */
2743 actual = __d_unalias(inode, dentry, alias);
2745 write_sequnlock(&rename_lock);
2746 if (IS_ERR(actual)) {
2747 if (PTR_ERR(actual) == -ELOOP)
2748 pr_warn_ratelimited(
2749 "VFS: Lookup of '%s' in %s %s"
2750 " would have caused loop\n",
2751 dentry->d_name.name,
2752 inode->i_sb->s_type->name,
2760 /* Add a unique reference */
2761 actual = __d_instantiate_unique(dentry, inode);
2765 BUG_ON(!d_unhashed(actual));
2767 spin_lock(&actual->d_lock);
2770 spin_unlock(&actual->d_lock);
2771 spin_unlock(&inode->i_lock);
2773 if (actual == dentry) {
2774 security_d_instantiate(dentry, inode);
2781 EXPORT_SYMBOL_GPL(d_materialise_unique);
2783 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2787 return -ENAMETOOLONG;
2789 memcpy(*buffer, str, namelen);
2794 * prepend_name - prepend a pathname in front of current buffer pointer
2795 * @buffer: buffer pointer
2796 * @buflen: allocated length of the buffer
2797 * @name: name string and length qstr structure
2799 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2800 * make sure that either the old or the new name pointer and length are
2801 * fetched. However, there may be mismatch between length and pointer.
2802 * The length cannot be trusted, we need to copy it byte-by-byte until
2803 * the length is reached or a null byte is found. It also prepends "/" at
2804 * the beginning of the name. The sequence number check at the caller will
2805 * retry it again when a d_move() does happen. So any garbage in the buffer
2806 * due to mismatched pointer and length will be discarded.
2808 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2810 const char *dname = ACCESS_ONCE(name->name);
2811 u32 dlen = ACCESS_ONCE(name->len);
2814 *buflen -= dlen + 1;
2816 return -ENAMETOOLONG;
2817 p = *buffer -= dlen + 1;
2829 * prepend_path - Prepend path string to a buffer
2830 * @path: the dentry/vfsmount to report
2831 * @root: root vfsmnt/dentry
2832 * @buffer: pointer to the end of the buffer
2833 * @buflen: pointer to buffer length
2835 * The function will first try to write out the pathname without taking any
2836 * lock other than the RCU read lock to make sure that dentries won't go away.
2837 * It only checks the sequence number of the global rename_lock as any change
2838 * in the dentry's d_seq will be preceded by changes in the rename_lock
2839 * sequence number. If the sequence number had been changed, it will restart
2840 * the whole pathname back-tracing sequence again by taking the rename_lock.
2841 * In this case, there is no need to take the RCU read lock as the recursive
2842 * parent pointer references will keep the dentry chain alive as long as no
2843 * rename operation is performed.
2845 static int prepend_path(const struct path *path,
2846 const struct path *root,
2847 char **buffer, int *buflen)
2849 struct dentry *dentry;
2850 struct vfsmount *vfsmnt;
2853 unsigned seq, m_seq = 0;
2859 read_seqbegin_or_lock(&mount_lock, &m_seq);
2866 dentry = path->dentry;
2868 mnt = real_mount(vfsmnt);
2869 read_seqbegin_or_lock(&rename_lock, &seq);
2870 while (dentry != root->dentry || vfsmnt != root->mnt) {
2871 struct dentry * parent;
2873 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2874 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2876 if (mnt != parent) {
2877 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2883 * Filesystems needing to implement special "root names"
2884 * should do so with ->d_dname()
2886 if (IS_ROOT(dentry) &&
2887 (dentry->d_name.len != 1 ||
2888 dentry->d_name.name[0] != '/')) {
2889 WARN(1, "Root dentry has weird name <%.*s>\n",
2890 (int) dentry->d_name.len,
2891 dentry->d_name.name);
2894 error = is_mounted(vfsmnt) ? 1 : 2;
2897 parent = dentry->d_parent;
2899 error = prepend_name(&bptr, &blen, &dentry->d_name);
2907 if (need_seqretry(&rename_lock, seq)) {
2911 done_seqretry(&rename_lock, seq);
2915 if (need_seqretry(&mount_lock, m_seq)) {
2919 done_seqretry(&mount_lock, m_seq);
2921 if (error >= 0 && bptr == *buffer) {
2923 error = -ENAMETOOLONG;
2933 * __d_path - return the path of a dentry
2934 * @path: the dentry/vfsmount to report
2935 * @root: root vfsmnt/dentry
2936 * @buf: buffer to return value in
2937 * @buflen: buffer length
2939 * Convert a dentry into an ASCII path name.
2941 * Returns a pointer into the buffer or an error code if the
2942 * path was too long.
2944 * "buflen" should be positive.
2946 * If the path is not reachable from the supplied root, return %NULL.
2948 char *__d_path(const struct path *path,
2949 const struct path *root,
2950 char *buf, int buflen)
2952 char *res = buf + buflen;
2955 prepend(&res, &buflen, "\0", 1);
2956 error = prepend_path(path, root, &res, &buflen);
2959 return ERR_PTR(error);
2965 char *d_absolute_path(const struct path *path,
2966 char *buf, int buflen)
2968 struct path root = {};
2969 char *res = buf + buflen;
2972 prepend(&res, &buflen, "\0", 1);
2973 error = prepend_path(path, &root, &res, &buflen);
2978 return ERR_PTR(error);
2983 * same as __d_path but appends "(deleted)" for unlinked files.
2985 static int path_with_deleted(const struct path *path,
2986 const struct path *root,
2987 char **buf, int *buflen)
2989 prepend(buf, buflen, "\0", 1);
2990 if (d_unlinked(path->dentry)) {
2991 int error = prepend(buf, buflen, " (deleted)", 10);
2996 return prepend_path(path, root, buf, buflen);
2999 static int prepend_unreachable(char **buffer, int *buflen)
3001 return prepend(buffer, buflen, "(unreachable)", 13);
3004 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3009 seq = read_seqcount_begin(&fs->seq);
3011 } while (read_seqcount_retry(&fs->seq, seq));
3015 * d_path - return the path of a dentry
3016 * @path: path to report
3017 * @buf: buffer to return value in
3018 * @buflen: buffer length
3020 * Convert a dentry into an ASCII path name. If the entry has been deleted
3021 * the string " (deleted)" is appended. Note that this is ambiguous.
3023 * Returns a pointer into the buffer or an error code if the path was
3024 * too long. Note: Callers should use the returned pointer, not the passed
3025 * in buffer, to use the name! The implementation often starts at an offset
3026 * into the buffer, and may leave 0 bytes at the start.
3028 * "buflen" should be positive.
3030 char *d_path(const struct path *path, char *buf, int buflen)
3032 char *res = buf + buflen;
3037 * We have various synthetic filesystems that never get mounted. On
3038 * these filesystems dentries are never used for lookup purposes, and
3039 * thus don't need to be hashed. They also don't need a name until a
3040 * user wants to identify the object in /proc/pid/fd/. The little hack
3041 * below allows us to generate a name for these objects on demand:
3043 * Some pseudo inodes are mountable. When they are mounted
3044 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3045 * and instead have d_path return the mounted path.
3047 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3048 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3049 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3052 get_fs_root_rcu(current->fs, &root);
3053 error = path_with_deleted(path, &root, &res, &buflen);
3057 res = ERR_PTR(error);
3060 EXPORT_SYMBOL(d_path);
3063 * Helper function for dentry_operations.d_dname() members
3065 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3066 const char *fmt, ...)
3072 va_start(args, fmt);
3073 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3076 if (sz > sizeof(temp) || sz > buflen)
3077 return ERR_PTR(-ENAMETOOLONG);
3079 buffer += buflen - sz;
3080 return memcpy(buffer, temp, sz);
3083 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3085 char *end = buffer + buflen;
3086 /* these dentries are never renamed, so d_lock is not needed */
3087 if (prepend(&end, &buflen, " (deleted)", 11) ||
3088 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3089 prepend(&end, &buflen, "/", 1))
3090 end = ERR_PTR(-ENAMETOOLONG);
3093 EXPORT_SYMBOL(simple_dname);
3096 * Write full pathname from the root of the filesystem into the buffer.
3098 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3100 struct dentry *dentry;
3113 prepend(&end, &len, "\0", 1);
3117 read_seqbegin_or_lock(&rename_lock, &seq);
3118 while (!IS_ROOT(dentry)) {
3119 struct dentry *parent = dentry->d_parent;
3122 error = prepend_name(&end, &len, &dentry->d_name);
3131 if (need_seqretry(&rename_lock, seq)) {
3135 done_seqretry(&rename_lock, seq);
3140 return ERR_PTR(-ENAMETOOLONG);
3143 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3145 return __dentry_path(dentry, buf, buflen);
3147 EXPORT_SYMBOL(dentry_path_raw);
3149 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3154 if (d_unlinked(dentry)) {
3156 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3160 retval = __dentry_path(dentry, buf, buflen);
3161 if (!IS_ERR(retval) && p)
3162 *p = '/'; /* restore '/' overriden with '\0' */
3165 return ERR_PTR(-ENAMETOOLONG);
3168 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3174 seq = read_seqcount_begin(&fs->seq);
3177 } while (read_seqcount_retry(&fs->seq, seq));
3181 * NOTE! The user-level library version returns a
3182 * character pointer. The kernel system call just
3183 * returns the length of the buffer filled (which
3184 * includes the ending '\0' character), or a negative
3185 * error value. So libc would do something like
3187 * char *getcwd(char * buf, size_t size)
3191 * retval = sys_getcwd(buf, size);
3198 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3201 struct path pwd, root;
3202 char *page = __getname();
3208 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3211 if (!d_unlinked(pwd.dentry)) {
3213 char *cwd = page + PATH_MAX;
3214 int buflen = PATH_MAX;
3216 prepend(&cwd, &buflen, "\0", 1);
3217 error = prepend_path(&pwd, &root, &cwd, &buflen);
3223 /* Unreachable from current root */
3225 error = prepend_unreachable(&cwd, &buflen);
3231 len = PATH_MAX + page - cwd;
3234 if (copy_to_user(buf, cwd, len))
3247 * Test whether new_dentry is a subdirectory of old_dentry.
3249 * Trivially implemented using the dcache structure
3253 * is_subdir - is new dentry a subdirectory of old_dentry
3254 * @new_dentry: new dentry
3255 * @old_dentry: old dentry
3257 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3258 * Returns 0 otherwise.
3259 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3262 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3267 if (new_dentry == old_dentry)
3271 /* for restarting inner loop in case of seq retry */
3272 seq = read_seqbegin(&rename_lock);
3274 * Need rcu_readlock to protect against the d_parent trashing
3278 if (d_ancestor(old_dentry, new_dentry))
3283 } while (read_seqretry(&rename_lock, seq));
3288 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3290 struct dentry *root = data;
3291 if (dentry != root) {
3292 if (d_unhashed(dentry) || !dentry->d_inode)
3295 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3296 dentry->d_flags |= DCACHE_GENOCIDE;
3297 dentry->d_lockref.count--;
3300 return D_WALK_CONTINUE;
3303 void d_genocide(struct dentry *parent)
3305 d_walk(parent, parent, d_genocide_kill, NULL);
3308 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3310 inode_dec_link_count(inode);
3311 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3312 !hlist_unhashed(&dentry->d_alias) ||
3313 !d_unlinked(dentry));
3314 spin_lock(&dentry->d_parent->d_lock);
3315 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3316 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3317 (unsigned long long)inode->i_ino);
3318 spin_unlock(&dentry->d_lock);
3319 spin_unlock(&dentry->d_parent->d_lock);
3320 d_instantiate(dentry, inode);
3322 EXPORT_SYMBOL(d_tmpfile);
3324 static __initdata unsigned long dhash_entries;
3325 static int __init set_dhash_entries(char *str)
3329 dhash_entries = simple_strtoul(str, &str, 0);
3332 __setup("dhash_entries=", set_dhash_entries);
3334 static void __init dcache_init_early(void)
3338 /* If hashes are distributed across NUMA nodes, defer
3339 * hash allocation until vmalloc space is available.
3345 alloc_large_system_hash("Dentry cache",
3346 sizeof(struct hlist_bl_head),
3355 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3356 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3359 static void __init dcache_init(void)
3364 * A constructor could be added for stable state like the lists,
3365 * but it is probably not worth it because of the cache nature
3368 dentry_cache = KMEM_CACHE(dentry,
3369 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3371 /* Hash may have been set up in dcache_init_early */
3376 alloc_large_system_hash("Dentry cache",
3377 sizeof(struct hlist_bl_head),
3386 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3387 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3390 /* SLAB cache for __getname() consumers */
3391 struct kmem_cache *names_cachep __read_mostly;
3392 EXPORT_SYMBOL(names_cachep);
3394 EXPORT_SYMBOL(d_genocide);
3396 void __init vfs_caches_init_early(void)
3398 dcache_init_early();
3402 void __init vfs_caches_init(unsigned long mempages)
3404 unsigned long reserve;
3406 /* Base hash sizes on available memory, with a reserve equal to
3407 150% of current kernel size */
3409 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3410 mempages -= reserve;
3412 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3413 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3417 files_init(mempages);