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(struct 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);
444 static void __dentry_kill(struct dentry *dentry)
446 struct dentry *parent = NULL;
447 bool can_free = true;
448 if (!IS_ROOT(dentry))
449 parent = dentry->d_parent;
452 * The dentry is now unrecoverably dead to the world.
454 lockref_mark_dead(&dentry->d_lockref);
457 * inform the fs via d_prune that this dentry is about to be
458 * unhashed and destroyed.
460 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
461 dentry->d_op->d_prune(dentry);
463 if (dentry->d_flags & DCACHE_LRU_LIST) {
464 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
467 /* if it was on the hash then remove it */
469 list_del(&dentry->d_u.d_child);
471 * Inform d_walk() that we are no longer attached to the
474 dentry->d_flags |= DCACHE_DENTRY_KILLED;
476 spin_unlock(&parent->d_lock);
479 * dentry_iput drops the locks, at which point nobody (except
480 * transient RCU lookups) can reach this dentry.
482 BUG_ON((int)dentry->d_lockref.count > 0);
483 this_cpu_dec(nr_dentry);
484 if (dentry->d_op && dentry->d_op->d_release)
485 dentry->d_op->d_release(dentry);
487 spin_lock(&dentry->d_lock);
488 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
489 dentry->d_flags |= DCACHE_MAY_FREE;
492 spin_unlock(&dentry->d_lock);
493 if (likely(can_free))
498 * Finish off a dentry we've decided to kill.
499 * dentry->d_lock must be held, returns with it unlocked.
500 * If ref is non-zero, then decrement the refcount too.
501 * Returns dentry requiring refcount drop, or NULL if we're done.
503 static struct dentry *dentry_kill(struct dentry *dentry)
504 __releases(dentry->d_lock)
506 struct inode *inode = dentry->d_inode;
507 struct dentry *parent = NULL;
509 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
512 if (!IS_ROOT(dentry)) {
513 parent = dentry->d_parent;
514 if (unlikely(!spin_trylock(&parent->d_lock))) {
516 spin_unlock(&inode->i_lock);
521 __dentry_kill(dentry);
525 spin_unlock(&dentry->d_lock);
527 return dentry; /* try again with same dentry */
530 static inline struct dentry *lock_parent(struct dentry *dentry)
532 struct dentry *parent = dentry->d_parent;
535 if (unlikely((int)dentry->d_lockref.count < 0))
537 if (likely(spin_trylock(&parent->d_lock)))
540 spin_unlock(&dentry->d_lock);
542 parent = ACCESS_ONCE(dentry->d_parent);
543 spin_lock(&parent->d_lock);
545 * We can't blindly lock dentry until we are sure
546 * that we won't violate the locking order.
547 * Any changes of dentry->d_parent must have
548 * been done with parent->d_lock held, so
549 * spin_lock() above is enough of a barrier
550 * for checking if it's still our child.
552 if (unlikely(parent != dentry->d_parent)) {
553 spin_unlock(&parent->d_lock);
557 if (parent != dentry)
558 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
567 * This is complicated by the fact that we do not want to put
568 * dentries that are no longer on any hash chain on the unused
569 * list: we'd much rather just get rid of them immediately.
571 * However, that implies that we have to traverse the dentry
572 * tree upwards to the parents which might _also_ now be
573 * scheduled for deletion (it may have been only waiting for
574 * its last child to go away).
576 * This tail recursion is done by hand as we don't want to depend
577 * on the compiler to always get this right (gcc generally doesn't).
578 * Real recursion would eat up our stack space.
582 * dput - release a dentry
583 * @dentry: dentry to release
585 * Release a dentry. This will drop the usage count and if appropriate
586 * call the dentry unlink method as well as removing it from the queues and
587 * releasing its resources. If the parent dentries were scheduled for release
588 * they too may now get deleted.
590 void dput(struct dentry *dentry)
592 if (unlikely(!dentry))
596 if (lockref_put_or_lock(&dentry->d_lockref))
599 /* Unreachable? Get rid of it */
600 if (unlikely(d_unhashed(dentry)))
603 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
604 if (dentry->d_op->d_delete(dentry))
608 if (!(dentry->d_flags & DCACHE_REFERENCED))
609 dentry->d_flags |= DCACHE_REFERENCED;
610 dentry_lru_add(dentry);
612 dentry->d_lockref.count--;
613 spin_unlock(&dentry->d_lock);
617 dentry = dentry_kill(dentry);
624 * d_invalidate - invalidate a dentry
625 * @dentry: dentry to invalidate
627 * Try to invalidate the dentry if it turns out to be
628 * possible. If there are other dentries that can be
629 * reached through this one we can't delete it and we
630 * return -EBUSY. On success we return 0.
635 int d_invalidate(struct dentry * dentry)
638 * If it's already been dropped, return OK.
640 spin_lock(&dentry->d_lock);
641 if (d_unhashed(dentry)) {
642 spin_unlock(&dentry->d_lock);
646 * Check whether to do a partial shrink_dcache
647 * to get rid of unused child entries.
649 if (!list_empty(&dentry->d_subdirs)) {
650 spin_unlock(&dentry->d_lock);
651 shrink_dcache_parent(dentry);
652 spin_lock(&dentry->d_lock);
656 * Somebody else still using it?
658 * If it's a directory, we can't drop it
659 * for fear of somebody re-populating it
660 * with children (even though dropping it
661 * would make it unreachable from the root,
662 * we might still populate it if it was a
663 * working directory or similar).
664 * We also need to leave mountpoints alone,
667 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
668 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
669 spin_unlock(&dentry->d_lock);
675 spin_unlock(&dentry->d_lock);
678 EXPORT_SYMBOL(d_invalidate);
680 /* This must be called with d_lock held */
681 static inline void __dget_dlock(struct dentry *dentry)
683 dentry->d_lockref.count++;
686 static inline void __dget(struct dentry *dentry)
688 lockref_get(&dentry->d_lockref);
691 struct dentry *dget_parent(struct dentry *dentry)
697 * Do optimistic parent lookup without any
701 ret = ACCESS_ONCE(dentry->d_parent);
702 gotref = lockref_get_not_zero(&ret->d_lockref);
704 if (likely(gotref)) {
705 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
712 * Don't need rcu_dereference because we re-check it was correct under
716 ret = dentry->d_parent;
717 spin_lock(&ret->d_lock);
718 if (unlikely(ret != dentry->d_parent)) {
719 spin_unlock(&ret->d_lock);
724 BUG_ON(!ret->d_lockref.count);
725 ret->d_lockref.count++;
726 spin_unlock(&ret->d_lock);
729 EXPORT_SYMBOL(dget_parent);
732 * d_find_alias - grab a hashed alias of inode
733 * @inode: inode in question
734 * @want_discon: flag, used by d_splice_alias, to request
735 * that only a DISCONNECTED alias be returned.
737 * If inode has a hashed alias, or is a directory and has any alias,
738 * acquire the reference to alias and return it. Otherwise return NULL.
739 * Notice that if inode is a directory there can be only one alias and
740 * it can be unhashed only if it has no children, or if it is the root
743 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
744 * any other hashed alias over that one unless @want_discon is set,
745 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
747 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
749 struct dentry *alias, *discon_alias;
753 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
754 spin_lock(&alias->d_lock);
755 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
756 if (IS_ROOT(alias) &&
757 (alias->d_flags & DCACHE_DISCONNECTED)) {
758 discon_alias = alias;
759 } else if (!want_discon) {
761 spin_unlock(&alias->d_lock);
765 spin_unlock(&alias->d_lock);
768 alias = discon_alias;
769 spin_lock(&alias->d_lock);
770 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
771 if (IS_ROOT(alias) &&
772 (alias->d_flags & DCACHE_DISCONNECTED)) {
774 spin_unlock(&alias->d_lock);
778 spin_unlock(&alias->d_lock);
784 struct dentry *d_find_alias(struct inode *inode)
786 struct dentry *de = NULL;
788 if (!hlist_empty(&inode->i_dentry)) {
789 spin_lock(&inode->i_lock);
790 de = __d_find_alias(inode, 0);
791 spin_unlock(&inode->i_lock);
795 EXPORT_SYMBOL(d_find_alias);
798 * Try to kill dentries associated with this inode.
799 * WARNING: you must own a reference to inode.
801 void d_prune_aliases(struct inode *inode)
803 struct dentry *dentry;
805 spin_lock(&inode->i_lock);
806 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
807 spin_lock(&dentry->d_lock);
808 if (!dentry->d_lockref.count) {
810 * inform the fs via d_prune that this dentry
811 * is about to be unhashed and destroyed.
813 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
815 dentry->d_op->d_prune(dentry);
817 __dget_dlock(dentry);
819 spin_unlock(&dentry->d_lock);
820 spin_unlock(&inode->i_lock);
824 spin_unlock(&dentry->d_lock);
826 spin_unlock(&inode->i_lock);
828 EXPORT_SYMBOL(d_prune_aliases);
830 static void shrink_dentry_list(struct list_head *list)
832 struct dentry *dentry, *parent;
834 while (!list_empty(list)) {
836 dentry = list_entry(list->prev, struct dentry, d_lru);
837 spin_lock(&dentry->d_lock);
838 parent = lock_parent(dentry);
841 * The dispose list is isolated and dentries are not accounted
842 * to the LRU here, so we can simply remove it from the list
843 * here regardless of whether it is referenced or not.
845 d_shrink_del(dentry);
848 * We found an inuse dentry which was not removed from
849 * the LRU because of laziness during lookup. Do not free it.
851 if ((int)dentry->d_lockref.count > 0) {
852 spin_unlock(&dentry->d_lock);
854 spin_unlock(&parent->d_lock);
859 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
860 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
861 spin_unlock(&dentry->d_lock);
863 spin_unlock(&parent->d_lock);
869 inode = dentry->d_inode;
870 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
871 d_shrink_add(dentry, list);
872 spin_unlock(&dentry->d_lock);
874 spin_unlock(&parent->d_lock);
878 __dentry_kill(dentry);
881 * We need to prune ancestors too. This is necessary to prevent
882 * quadratic behavior of shrink_dcache_parent(), but is also
883 * expected to be beneficial in reducing dentry cache
887 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
888 parent = lock_parent(dentry);
889 if (dentry->d_lockref.count != 1) {
890 dentry->d_lockref.count--;
891 spin_unlock(&dentry->d_lock);
893 spin_unlock(&parent->d_lock);
896 inode = dentry->d_inode; /* can't be NULL */
897 if (unlikely(!spin_trylock(&inode->i_lock))) {
898 spin_unlock(&dentry->d_lock);
900 spin_unlock(&parent->d_lock);
904 __dentry_kill(dentry);
910 static enum lru_status
911 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
913 struct list_head *freeable = arg;
914 struct dentry *dentry = container_of(item, struct dentry, d_lru);
918 * we are inverting the lru lock/dentry->d_lock here,
919 * so use a trylock. If we fail to get the lock, just skip
922 if (!spin_trylock(&dentry->d_lock))
926 * Referenced dentries are still in use. If they have active
927 * counts, just remove them from the LRU. Otherwise give them
928 * another pass through the LRU.
930 if (dentry->d_lockref.count) {
931 d_lru_isolate(dentry);
932 spin_unlock(&dentry->d_lock);
936 if (dentry->d_flags & DCACHE_REFERENCED) {
937 dentry->d_flags &= ~DCACHE_REFERENCED;
938 spin_unlock(&dentry->d_lock);
941 * The list move itself will be made by the common LRU code. At
942 * this point, we've dropped the dentry->d_lock but keep the
943 * lru lock. This is safe to do, since every list movement is
944 * protected by the lru lock even if both locks are held.
946 * This is guaranteed by the fact that all LRU management
947 * functions are intermediated by the LRU API calls like
948 * list_lru_add and list_lru_del. List movement in this file
949 * only ever occur through this functions or through callbacks
950 * like this one, that are called from the LRU API.
952 * The only exceptions to this are functions like
953 * shrink_dentry_list, and code that first checks for the
954 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
955 * operating only with stack provided lists after they are
956 * properly isolated from the main list. It is thus, always a
962 d_lru_shrink_move(dentry, freeable);
963 spin_unlock(&dentry->d_lock);
969 * prune_dcache_sb - shrink the dcache
971 * @nr_to_scan : number of entries to try to free
972 * @nid: which node to scan for freeable entities
974 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
975 * done when we need more memory an called from the superblock shrinker
978 * This function may fail to free any resources if all the dentries are in
981 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
987 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
988 &dispose, &nr_to_scan);
989 shrink_dentry_list(&dispose);
993 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
994 spinlock_t *lru_lock, void *arg)
996 struct list_head *freeable = arg;
997 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1000 * we are inverting the lru lock/dentry->d_lock here,
1001 * so use a trylock. If we fail to get the lock, just skip
1004 if (!spin_trylock(&dentry->d_lock))
1007 d_lru_shrink_move(dentry, freeable);
1008 spin_unlock(&dentry->d_lock);
1015 * shrink_dcache_sb - shrink dcache for a superblock
1018 * Shrink the dcache for the specified super block. This is used to free
1019 * the dcache before unmounting a file system.
1021 void shrink_dcache_sb(struct super_block *sb)
1028 freed = list_lru_walk(&sb->s_dentry_lru,
1029 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1031 this_cpu_sub(nr_dentry_unused, freed);
1032 shrink_dentry_list(&dispose);
1033 } while (freed > 0);
1035 EXPORT_SYMBOL(shrink_dcache_sb);
1038 * enum d_walk_ret - action to talke during tree walk
1039 * @D_WALK_CONTINUE: contrinue walk
1040 * @D_WALK_QUIT: quit walk
1041 * @D_WALK_NORETRY: quit when retry is needed
1042 * @D_WALK_SKIP: skip this dentry and its children
1052 * d_walk - walk the dentry tree
1053 * @parent: start of walk
1054 * @data: data passed to @enter() and @finish()
1055 * @enter: callback when first entering the dentry
1056 * @finish: callback when successfully finished the walk
1058 * The @enter() and @finish() callbacks are called with d_lock held.
1060 static void d_walk(struct dentry *parent, void *data,
1061 enum d_walk_ret (*enter)(void *, struct dentry *),
1062 void (*finish)(void *))
1064 struct dentry *this_parent;
1065 struct list_head *next;
1067 enum d_walk_ret ret;
1071 read_seqbegin_or_lock(&rename_lock, &seq);
1072 this_parent = parent;
1073 spin_lock(&this_parent->d_lock);
1075 ret = enter(data, this_parent);
1077 case D_WALK_CONTINUE:
1082 case D_WALK_NORETRY:
1087 next = this_parent->d_subdirs.next;
1089 while (next != &this_parent->d_subdirs) {
1090 struct list_head *tmp = next;
1091 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1094 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1096 ret = enter(data, dentry);
1098 case D_WALK_CONTINUE:
1101 spin_unlock(&dentry->d_lock);
1103 case D_WALK_NORETRY:
1107 spin_unlock(&dentry->d_lock);
1111 if (!list_empty(&dentry->d_subdirs)) {
1112 spin_unlock(&this_parent->d_lock);
1113 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1114 this_parent = dentry;
1115 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1118 spin_unlock(&dentry->d_lock);
1121 * All done at this level ... ascend and resume the search.
1123 if (this_parent != parent) {
1124 struct dentry *child = this_parent;
1125 this_parent = child->d_parent;
1128 spin_unlock(&child->d_lock);
1129 spin_lock(&this_parent->d_lock);
1132 * might go back up the wrong parent if we have had a rename
1135 if (this_parent != child->d_parent ||
1136 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1137 need_seqretry(&rename_lock, seq)) {
1138 spin_unlock(&this_parent->d_lock);
1143 next = child->d_u.d_child.next;
1146 if (need_seqretry(&rename_lock, seq)) {
1147 spin_unlock(&this_parent->d_lock);
1154 spin_unlock(&this_parent->d_lock);
1155 done_seqretry(&rename_lock, seq);
1166 * Search for at least 1 mount point in the dentry's subdirs.
1167 * We descend to the next level whenever the d_subdirs
1168 * list is non-empty and continue searching.
1171 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1174 if (d_mountpoint(dentry)) {
1178 return D_WALK_CONTINUE;
1182 * have_submounts - check for mounts over a dentry
1183 * @parent: dentry to check.
1185 * Return true if the parent or its subdirectories contain
1188 int have_submounts(struct dentry *parent)
1192 d_walk(parent, &ret, check_mount, NULL);
1196 EXPORT_SYMBOL(have_submounts);
1199 * Called by mount code to set a mountpoint and check if the mountpoint is
1200 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1201 * subtree can become unreachable).
1203 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1204 * this reason take rename_lock and d_lock on dentry and ancestors.
1206 int d_set_mounted(struct dentry *dentry)
1210 write_seqlock(&rename_lock);
1211 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1212 /* Need exclusion wrt. check_submounts_and_drop() */
1213 spin_lock(&p->d_lock);
1214 if (unlikely(d_unhashed(p))) {
1215 spin_unlock(&p->d_lock);
1218 spin_unlock(&p->d_lock);
1220 spin_lock(&dentry->d_lock);
1221 if (!d_unlinked(dentry)) {
1222 dentry->d_flags |= DCACHE_MOUNTED;
1225 spin_unlock(&dentry->d_lock);
1227 write_sequnlock(&rename_lock);
1232 * Search the dentry child list of the specified parent,
1233 * and move any unused dentries to the end of the unused
1234 * list for prune_dcache(). We descend to the next level
1235 * whenever the d_subdirs list is non-empty and continue
1238 * It returns zero iff there are no unused children,
1239 * otherwise it returns the number of children moved to
1240 * the end of the unused list. This may not be the total
1241 * number of unused children, because select_parent can
1242 * drop the lock and return early due to latency
1246 struct select_data {
1247 struct dentry *start;
1248 struct list_head dispose;
1252 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1254 struct select_data *data = _data;
1255 enum d_walk_ret ret = D_WALK_CONTINUE;
1257 if (data->start == dentry)
1260 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1263 if (dentry->d_flags & DCACHE_LRU_LIST)
1265 if (!dentry->d_lockref.count) {
1266 d_shrink_add(dentry, &data->dispose);
1271 * We can return to the caller if we have found some (this
1272 * ensures forward progress). We'll be coming back to find
1275 if (!list_empty(&data->dispose))
1276 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1282 * shrink_dcache_parent - prune dcache
1283 * @parent: parent of entries to prune
1285 * Prune the dcache to remove unused children of the parent dentry.
1287 void shrink_dcache_parent(struct dentry *parent)
1290 struct select_data data;
1292 INIT_LIST_HEAD(&data.dispose);
1293 data.start = parent;
1296 d_walk(parent, &data, select_collect, NULL);
1300 shrink_dentry_list(&data.dispose);
1304 EXPORT_SYMBOL(shrink_dcache_parent);
1306 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1308 /* it has busy descendents; complain about those instead */
1309 if (!list_empty(&dentry->d_subdirs))
1310 return D_WALK_CONTINUE;
1312 /* root with refcount 1 is fine */
1313 if (dentry == _data && dentry->d_lockref.count == 1)
1314 return D_WALK_CONTINUE;
1316 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1317 " still in use (%d) [unmount of %s %s]\n",
1320 dentry->d_inode->i_ino : 0UL,
1322 dentry->d_lockref.count,
1323 dentry->d_sb->s_type->name,
1324 dentry->d_sb->s_id);
1326 return D_WALK_CONTINUE;
1329 static void do_one_tree(struct dentry *dentry)
1331 shrink_dcache_parent(dentry);
1332 d_walk(dentry, dentry, umount_check, NULL);
1338 * destroy the dentries attached to a superblock on unmounting
1340 void shrink_dcache_for_umount(struct super_block *sb)
1342 struct dentry *dentry;
1344 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1346 dentry = sb->s_root;
1348 do_one_tree(dentry);
1350 while (!hlist_bl_empty(&sb->s_anon)) {
1351 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1352 do_one_tree(dentry);
1356 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1358 struct select_data *data = _data;
1360 if (d_mountpoint(dentry)) {
1361 data->found = -EBUSY;
1365 return select_collect(_data, dentry);
1368 static void check_and_drop(void *_data)
1370 struct select_data *data = _data;
1372 if (d_mountpoint(data->start))
1373 data->found = -EBUSY;
1375 __d_drop(data->start);
1379 * check_submounts_and_drop - prune dcache, check for submounts and drop
1381 * All done as a single atomic operation relative to has_unlinked_ancestor().
1382 * Returns 0 if successfully unhashed @parent. If there were submounts then
1385 * @dentry: dentry to prune and drop
1387 int check_submounts_and_drop(struct dentry *dentry)
1391 /* Negative dentries can be dropped without further checks */
1392 if (!dentry->d_inode) {
1398 struct select_data data;
1400 INIT_LIST_HEAD(&data.dispose);
1401 data.start = dentry;
1404 d_walk(dentry, &data, check_and_collect, check_and_drop);
1407 if (!list_empty(&data.dispose))
1408 shrink_dentry_list(&data.dispose);
1419 EXPORT_SYMBOL(check_submounts_and_drop);
1422 * __d_alloc - allocate a dcache entry
1423 * @sb: filesystem it will belong to
1424 * @name: qstr of the name
1426 * Allocates a dentry. It returns %NULL if there is insufficient memory
1427 * available. On a success the dentry is returned. The name passed in is
1428 * copied and the copy passed in may be reused after this call.
1431 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1433 struct dentry *dentry;
1436 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1441 * We guarantee that the inline name is always NUL-terminated.
1442 * This way the memcpy() done by the name switching in rename
1443 * will still always have a NUL at the end, even if we might
1444 * be overwriting an internal NUL character
1446 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1447 if (name->len > DNAME_INLINE_LEN-1) {
1448 dname = kmalloc(name->len + 1, GFP_KERNEL);
1450 kmem_cache_free(dentry_cache, dentry);
1454 dname = dentry->d_iname;
1457 dentry->d_name.len = name->len;
1458 dentry->d_name.hash = name->hash;
1459 memcpy(dname, name->name, name->len);
1460 dname[name->len] = 0;
1462 /* Make sure we always see the terminating NUL character */
1464 dentry->d_name.name = dname;
1466 dentry->d_lockref.count = 1;
1467 dentry->d_flags = 0;
1468 spin_lock_init(&dentry->d_lock);
1469 seqcount_init(&dentry->d_seq);
1470 dentry->d_inode = NULL;
1471 dentry->d_parent = dentry;
1473 dentry->d_op = NULL;
1474 dentry->d_fsdata = NULL;
1475 INIT_HLIST_BL_NODE(&dentry->d_hash);
1476 INIT_LIST_HEAD(&dentry->d_lru);
1477 INIT_LIST_HEAD(&dentry->d_subdirs);
1478 INIT_HLIST_NODE(&dentry->d_alias);
1479 INIT_LIST_HEAD(&dentry->d_u.d_child);
1480 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1482 this_cpu_inc(nr_dentry);
1488 * d_alloc - allocate a dcache entry
1489 * @parent: parent of entry to allocate
1490 * @name: qstr of the name
1492 * Allocates a dentry. It returns %NULL if there is insufficient memory
1493 * available. On a success the dentry is returned. The name passed in is
1494 * copied and the copy passed in may be reused after this call.
1496 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1498 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1502 spin_lock(&parent->d_lock);
1504 * don't need child lock because it is not subject
1505 * to concurrency here
1507 __dget_dlock(parent);
1508 dentry->d_parent = parent;
1509 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1510 spin_unlock(&parent->d_lock);
1514 EXPORT_SYMBOL(d_alloc);
1517 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1518 * @sb: the superblock
1519 * @name: qstr of the name
1521 * For a filesystem that just pins its dentries in memory and never
1522 * performs lookups at all, return an unhashed IS_ROOT dentry.
1524 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1526 return __d_alloc(sb, name);
1528 EXPORT_SYMBOL(d_alloc_pseudo);
1530 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1535 q.len = strlen(name);
1536 q.hash = full_name_hash(q.name, q.len);
1537 return d_alloc(parent, &q);
1539 EXPORT_SYMBOL(d_alloc_name);
1541 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1543 WARN_ON_ONCE(dentry->d_op);
1544 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1546 DCACHE_OP_REVALIDATE |
1547 DCACHE_OP_WEAK_REVALIDATE |
1548 DCACHE_OP_DELETE ));
1553 dentry->d_flags |= DCACHE_OP_HASH;
1555 dentry->d_flags |= DCACHE_OP_COMPARE;
1556 if (op->d_revalidate)
1557 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1558 if (op->d_weak_revalidate)
1559 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1561 dentry->d_flags |= DCACHE_OP_DELETE;
1563 dentry->d_flags |= DCACHE_OP_PRUNE;
1566 EXPORT_SYMBOL(d_set_d_op);
1568 static unsigned d_flags_for_inode(struct inode *inode)
1570 unsigned add_flags = DCACHE_FILE_TYPE;
1573 return DCACHE_MISS_TYPE;
1575 if (S_ISDIR(inode->i_mode)) {
1576 add_flags = DCACHE_DIRECTORY_TYPE;
1577 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1578 if (unlikely(!inode->i_op->lookup))
1579 add_flags = DCACHE_AUTODIR_TYPE;
1581 inode->i_opflags |= IOP_LOOKUP;
1583 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1584 if (unlikely(inode->i_op->follow_link))
1585 add_flags = DCACHE_SYMLINK_TYPE;
1587 inode->i_opflags |= IOP_NOFOLLOW;
1590 if (unlikely(IS_AUTOMOUNT(inode)))
1591 add_flags |= DCACHE_NEED_AUTOMOUNT;
1595 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1597 unsigned add_flags = d_flags_for_inode(inode);
1599 spin_lock(&dentry->d_lock);
1600 __d_set_type(dentry, add_flags);
1602 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1603 dentry->d_inode = inode;
1604 dentry_rcuwalk_barrier(dentry);
1605 spin_unlock(&dentry->d_lock);
1606 fsnotify_d_instantiate(dentry, inode);
1610 * d_instantiate - fill in inode information for a dentry
1611 * @entry: dentry to complete
1612 * @inode: inode to attach to this dentry
1614 * Fill in inode information in the entry.
1616 * This turns negative dentries into productive full members
1619 * NOTE! This assumes that the inode count has been incremented
1620 * (or otherwise set) by the caller to indicate that it is now
1621 * in use by the dcache.
1624 void d_instantiate(struct dentry *entry, struct inode * inode)
1626 BUG_ON(!hlist_unhashed(&entry->d_alias));
1628 spin_lock(&inode->i_lock);
1629 __d_instantiate(entry, inode);
1631 spin_unlock(&inode->i_lock);
1632 security_d_instantiate(entry, inode);
1634 EXPORT_SYMBOL(d_instantiate);
1637 * d_instantiate_unique - instantiate a non-aliased dentry
1638 * @entry: dentry to instantiate
1639 * @inode: inode to attach to this dentry
1641 * Fill in inode information in the entry. On success, it returns NULL.
1642 * If an unhashed alias of "entry" already exists, then we return the
1643 * aliased dentry instead and drop one reference to inode.
1645 * Note that in order to avoid conflicts with rename() etc, the caller
1646 * had better be holding the parent directory semaphore.
1648 * This also assumes that the inode count has been incremented
1649 * (or otherwise set) by the caller to indicate that it is now
1650 * in use by the dcache.
1652 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1653 struct inode *inode)
1655 struct dentry *alias;
1656 int len = entry->d_name.len;
1657 const char *name = entry->d_name.name;
1658 unsigned int hash = entry->d_name.hash;
1661 __d_instantiate(entry, NULL);
1665 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1667 * Don't need alias->d_lock here, because aliases with
1668 * d_parent == entry->d_parent are not subject to name or
1669 * parent changes, because the parent inode i_mutex is held.
1671 if (alias->d_name.hash != hash)
1673 if (alias->d_parent != entry->d_parent)
1675 if (alias->d_name.len != len)
1677 if (dentry_cmp(alias, name, len))
1683 __d_instantiate(entry, inode);
1687 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1689 struct dentry *result;
1691 BUG_ON(!hlist_unhashed(&entry->d_alias));
1694 spin_lock(&inode->i_lock);
1695 result = __d_instantiate_unique(entry, inode);
1697 spin_unlock(&inode->i_lock);
1700 security_d_instantiate(entry, inode);
1704 BUG_ON(!d_unhashed(result));
1709 EXPORT_SYMBOL(d_instantiate_unique);
1712 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1713 * @entry: dentry to complete
1714 * @inode: inode to attach to this dentry
1716 * Fill in inode information in the entry. If a directory alias is found, then
1717 * return an error (and drop inode). Together with d_materialise_unique() this
1718 * guarantees that a directory inode may never have more than one alias.
1720 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1722 BUG_ON(!hlist_unhashed(&entry->d_alias));
1724 spin_lock(&inode->i_lock);
1725 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1726 spin_unlock(&inode->i_lock);
1730 __d_instantiate(entry, inode);
1731 spin_unlock(&inode->i_lock);
1732 security_d_instantiate(entry, inode);
1736 EXPORT_SYMBOL(d_instantiate_no_diralias);
1738 struct dentry *d_make_root(struct inode *root_inode)
1740 struct dentry *res = NULL;
1743 static const struct qstr name = QSTR_INIT("/", 1);
1745 res = __d_alloc(root_inode->i_sb, &name);
1747 d_instantiate(res, root_inode);
1753 EXPORT_SYMBOL(d_make_root);
1755 static struct dentry * __d_find_any_alias(struct inode *inode)
1757 struct dentry *alias;
1759 if (hlist_empty(&inode->i_dentry))
1761 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1767 * d_find_any_alias - find any alias for a given inode
1768 * @inode: inode to find an alias for
1770 * If any aliases exist for the given inode, take and return a
1771 * reference for one of them. If no aliases exist, return %NULL.
1773 struct dentry *d_find_any_alias(struct inode *inode)
1777 spin_lock(&inode->i_lock);
1778 de = __d_find_any_alias(inode);
1779 spin_unlock(&inode->i_lock);
1782 EXPORT_SYMBOL(d_find_any_alias);
1784 struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1786 static const struct qstr anonstring = QSTR_INIT("/", 1);
1792 return ERR_PTR(-ESTALE);
1794 return ERR_CAST(inode);
1796 res = d_find_any_alias(inode);
1800 tmp = __d_alloc(inode->i_sb, &anonstring);
1802 res = ERR_PTR(-ENOMEM);
1806 spin_lock(&inode->i_lock);
1807 res = __d_find_any_alias(inode);
1809 spin_unlock(&inode->i_lock);
1814 /* attach a disconnected dentry */
1815 add_flags = d_flags_for_inode(inode);
1818 add_flags |= DCACHE_DISCONNECTED;
1820 spin_lock(&tmp->d_lock);
1821 tmp->d_inode = inode;
1822 tmp->d_flags |= add_flags;
1823 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1824 hlist_bl_lock(&tmp->d_sb->s_anon);
1825 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1826 hlist_bl_unlock(&tmp->d_sb->s_anon);
1827 spin_unlock(&tmp->d_lock);
1828 spin_unlock(&inode->i_lock);
1829 security_d_instantiate(tmp, inode);
1834 if (res && !IS_ERR(res))
1835 security_d_instantiate(res, inode);
1841 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1842 * @inode: inode to allocate the dentry for
1844 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1845 * similar open by handle operations. The returned dentry may be anonymous,
1846 * or may have a full name (if the inode was already in the cache).
1848 * When called on a directory inode, we must ensure that the inode only ever
1849 * has one dentry. If a dentry is found, that is returned instead of
1850 * allocating a new one.
1852 * On successful return, the reference to the inode has been transferred
1853 * to the dentry. In case of an error the reference on the inode is released.
1854 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1855 * be passed in and the error will be propagated to the return value,
1856 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1858 struct dentry *d_obtain_alias(struct inode *inode)
1860 return __d_obtain_alias(inode, 1);
1862 EXPORT_SYMBOL(d_obtain_alias);
1865 * d_obtain_root - find or allocate a dentry for a given inode
1866 * @inode: inode to allocate the dentry for
1868 * Obtain an IS_ROOT dentry for the root of a filesystem.
1870 * We must ensure that directory inodes only ever have one dentry. If a
1871 * dentry is found, that is returned instead of allocating a new one.
1873 * On successful return, the reference to the inode has been transferred
1874 * to the dentry. In case of an error the reference on the inode is
1875 * released. A %NULL or IS_ERR inode may be passed in and will be the
1876 * error will be propagate to the return value, with a %NULL @inode
1877 * replaced by ERR_PTR(-ESTALE).
1879 struct dentry *d_obtain_root(struct inode *inode)
1881 return __d_obtain_alias(inode, 0);
1883 EXPORT_SYMBOL(d_obtain_root);
1886 * d_add_ci - lookup or allocate new dentry with case-exact name
1887 * @inode: the inode case-insensitive lookup has found
1888 * @dentry: the negative dentry that was passed to the parent's lookup func
1889 * @name: the case-exact name to be associated with the returned dentry
1891 * This is to avoid filling the dcache with case-insensitive names to the
1892 * same inode, only the actual correct case is stored in the dcache for
1893 * case-insensitive filesystems.
1895 * For a case-insensitive lookup match and if the the case-exact dentry
1896 * already exists in in the dcache, use it and return it.
1898 * If no entry exists with the exact case name, allocate new dentry with
1899 * the exact case, and return the spliced entry.
1901 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1904 struct dentry *found;
1908 * First check if a dentry matching the name already exists,
1909 * if not go ahead and create it now.
1911 found = d_hash_and_lookup(dentry->d_parent, name);
1912 if (unlikely(IS_ERR(found)))
1915 new = d_alloc(dentry->d_parent, name);
1917 found = ERR_PTR(-ENOMEM);
1921 found = d_splice_alias(inode, new);
1930 * If a matching dentry exists, and it's not negative use it.
1932 * Decrement the reference count to balance the iget() done
1935 if (found->d_inode) {
1936 if (unlikely(found->d_inode != inode)) {
1937 /* This can't happen because bad inodes are unhashed. */
1938 BUG_ON(!is_bad_inode(inode));
1939 BUG_ON(!is_bad_inode(found->d_inode));
1946 * Negative dentry: instantiate it unless the inode is a directory and
1947 * already has a dentry.
1949 new = d_splice_alias(inode, found);
1960 EXPORT_SYMBOL(d_add_ci);
1963 * Do the slow-case of the dentry name compare.
1965 * Unlike the dentry_cmp() function, we need to atomically
1966 * load the name and length information, so that the
1967 * filesystem can rely on them, and can use the 'name' and
1968 * 'len' information without worrying about walking off the
1969 * end of memory etc.
1971 * Thus the read_seqcount_retry() and the "duplicate" info
1972 * in arguments (the low-level filesystem should not look
1973 * at the dentry inode or name contents directly, since
1974 * rename can change them while we're in RCU mode).
1976 enum slow_d_compare {
1982 static noinline enum slow_d_compare slow_dentry_cmp(
1983 const struct dentry *parent,
1984 struct dentry *dentry,
1986 const struct qstr *name)
1988 int tlen = dentry->d_name.len;
1989 const char *tname = dentry->d_name.name;
1991 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1993 return D_COMP_SEQRETRY;
1995 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1996 return D_COMP_NOMATCH;
2001 * __d_lookup_rcu - search for a dentry (racy, store-free)
2002 * @parent: parent dentry
2003 * @name: qstr of name we wish to find
2004 * @seqp: returns d_seq value at the point where the dentry was found
2005 * Returns: dentry, or NULL
2007 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2008 * resolution (store-free path walking) design described in
2009 * Documentation/filesystems/path-lookup.txt.
2011 * This is not to be used outside core vfs.
2013 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2014 * held, and rcu_read_lock held. The returned dentry must not be stored into
2015 * without taking d_lock and checking d_seq sequence count against @seq
2018 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2021 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2022 * the returned dentry, so long as its parent's seqlock is checked after the
2023 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2024 * is formed, giving integrity down the path walk.
2026 * NOTE! The caller *has* to check the resulting dentry against the sequence
2027 * number we've returned before using any of the resulting dentry state!
2029 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2030 const struct qstr *name,
2033 u64 hashlen = name->hash_len;
2034 const unsigned char *str = name->name;
2035 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2036 struct hlist_bl_node *node;
2037 struct dentry *dentry;
2040 * Note: There is significant duplication with __d_lookup_rcu which is
2041 * required to prevent single threaded performance regressions
2042 * especially on architectures where smp_rmb (in seqcounts) are costly.
2043 * Keep the two functions in sync.
2047 * The hash list is protected using RCU.
2049 * Carefully use d_seq when comparing a candidate dentry, to avoid
2050 * races with d_move().
2052 * It is possible that concurrent renames can mess up our list
2053 * walk here and result in missing our dentry, resulting in the
2054 * false-negative result. d_lookup() protects against concurrent
2055 * renames using rename_lock seqlock.
2057 * See Documentation/filesystems/path-lookup.txt for more details.
2059 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2064 * The dentry sequence count protects us from concurrent
2065 * renames, and thus protects parent and name fields.
2067 * The caller must perform a seqcount check in order
2068 * to do anything useful with the returned dentry.
2070 * NOTE! We do a "raw" seqcount_begin here. That means that
2071 * we don't wait for the sequence count to stabilize if it
2072 * is in the middle of a sequence change. If we do the slow
2073 * dentry compare, we will do seqretries until it is stable,
2074 * and if we end up with a successful lookup, we actually
2075 * want to exit RCU lookup anyway.
2077 seq = raw_seqcount_begin(&dentry->d_seq);
2078 if (dentry->d_parent != parent)
2080 if (d_unhashed(dentry))
2083 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2084 if (dentry->d_name.hash != hashlen_hash(hashlen))
2087 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2090 case D_COMP_NOMATCH:
2097 if (dentry->d_name.hash_len != hashlen)
2100 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2107 * d_lookup - search for a dentry
2108 * @parent: parent dentry
2109 * @name: qstr of name we wish to find
2110 * Returns: dentry, or NULL
2112 * d_lookup searches the children of the parent dentry for the name in
2113 * question. If the dentry is found its reference count is incremented and the
2114 * dentry is returned. The caller must use dput to free the entry when it has
2115 * finished using it. %NULL is returned if the dentry does not exist.
2117 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2119 struct dentry *dentry;
2123 seq = read_seqbegin(&rename_lock);
2124 dentry = __d_lookup(parent, name);
2127 } while (read_seqretry(&rename_lock, seq));
2130 EXPORT_SYMBOL(d_lookup);
2133 * __d_lookup - search for a dentry (racy)
2134 * @parent: parent dentry
2135 * @name: qstr of name we wish to find
2136 * Returns: dentry, or NULL
2138 * __d_lookup is like d_lookup, however it may (rarely) return a
2139 * false-negative result due to unrelated rename activity.
2141 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2142 * however it must be used carefully, eg. with a following d_lookup in
2143 * the case of failure.
2145 * __d_lookup callers must be commented.
2147 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2149 unsigned int len = name->len;
2150 unsigned int hash = name->hash;
2151 const unsigned char *str = name->name;
2152 struct hlist_bl_head *b = d_hash(parent, hash);
2153 struct hlist_bl_node *node;
2154 struct dentry *found = NULL;
2155 struct dentry *dentry;
2158 * Note: There is significant duplication with __d_lookup_rcu which is
2159 * required to prevent single threaded performance regressions
2160 * especially on architectures where smp_rmb (in seqcounts) are costly.
2161 * Keep the two functions in sync.
2165 * The hash list is protected using RCU.
2167 * Take d_lock when comparing a candidate dentry, to avoid races
2170 * It is possible that concurrent renames can mess up our list
2171 * walk here and result in missing our dentry, resulting in the
2172 * false-negative result. d_lookup() protects against concurrent
2173 * renames using rename_lock seqlock.
2175 * See Documentation/filesystems/path-lookup.txt for more details.
2179 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2181 if (dentry->d_name.hash != hash)
2184 spin_lock(&dentry->d_lock);
2185 if (dentry->d_parent != parent)
2187 if (d_unhashed(dentry))
2191 * It is safe to compare names since d_move() cannot
2192 * change the qstr (protected by d_lock).
2194 if (parent->d_flags & DCACHE_OP_COMPARE) {
2195 int tlen = dentry->d_name.len;
2196 const char *tname = dentry->d_name.name;
2197 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2200 if (dentry->d_name.len != len)
2202 if (dentry_cmp(dentry, str, len))
2206 dentry->d_lockref.count++;
2208 spin_unlock(&dentry->d_lock);
2211 spin_unlock(&dentry->d_lock);
2219 * d_hash_and_lookup - hash the qstr then search for a dentry
2220 * @dir: Directory to search in
2221 * @name: qstr of name we wish to find
2223 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2225 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2228 * Check for a fs-specific hash function. Note that we must
2229 * calculate the standard hash first, as the d_op->d_hash()
2230 * routine may choose to leave the hash value unchanged.
2232 name->hash = full_name_hash(name->name, name->len);
2233 if (dir->d_flags & DCACHE_OP_HASH) {
2234 int err = dir->d_op->d_hash(dir, name);
2235 if (unlikely(err < 0))
2236 return ERR_PTR(err);
2238 return d_lookup(dir, name);
2240 EXPORT_SYMBOL(d_hash_and_lookup);
2243 * d_validate - verify dentry provided from insecure source (deprecated)
2244 * @dentry: The dentry alleged to be valid child of @dparent
2245 * @dparent: The parent dentry (known to be valid)
2247 * An insecure source has sent us a dentry, here we verify it and dget() it.
2248 * This is used by ncpfs in its readdir implementation.
2249 * Zero is returned in the dentry is invalid.
2251 * This function is slow for big directories, and deprecated, do not use it.
2253 int d_validate(struct dentry *dentry, struct dentry *dparent)
2255 struct dentry *child;
2257 spin_lock(&dparent->d_lock);
2258 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2259 if (dentry == child) {
2260 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2261 __dget_dlock(dentry);
2262 spin_unlock(&dentry->d_lock);
2263 spin_unlock(&dparent->d_lock);
2267 spin_unlock(&dparent->d_lock);
2271 EXPORT_SYMBOL(d_validate);
2274 * When a file is deleted, we have two options:
2275 * - turn this dentry into a negative dentry
2276 * - unhash this dentry and free it.
2278 * Usually, we want to just turn this into
2279 * a negative dentry, but if anybody else is
2280 * currently using the dentry or the inode
2281 * we can't do that and we fall back on removing
2282 * it from the hash queues and waiting for
2283 * it to be deleted later when it has no users
2287 * d_delete - delete a dentry
2288 * @dentry: The dentry to delete
2290 * Turn the dentry into a negative dentry if possible, otherwise
2291 * remove it from the hash queues so it can be deleted later
2294 void d_delete(struct dentry * dentry)
2296 struct inode *inode;
2299 * Are we the only user?
2302 spin_lock(&dentry->d_lock);
2303 inode = dentry->d_inode;
2304 isdir = S_ISDIR(inode->i_mode);
2305 if (dentry->d_lockref.count == 1) {
2306 if (!spin_trylock(&inode->i_lock)) {
2307 spin_unlock(&dentry->d_lock);
2311 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2312 dentry_unlink_inode(dentry);
2313 fsnotify_nameremove(dentry, isdir);
2317 if (!d_unhashed(dentry))
2320 spin_unlock(&dentry->d_lock);
2322 fsnotify_nameremove(dentry, isdir);
2324 EXPORT_SYMBOL(d_delete);
2326 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2328 BUG_ON(!d_unhashed(entry));
2330 entry->d_flags |= DCACHE_RCUACCESS;
2331 hlist_bl_add_head_rcu(&entry->d_hash, b);
2335 static void _d_rehash(struct dentry * entry)
2337 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2341 * d_rehash - add an entry back to the hash
2342 * @entry: dentry to add to the hash
2344 * Adds a dentry to the hash according to its name.
2347 void d_rehash(struct dentry * entry)
2349 spin_lock(&entry->d_lock);
2351 spin_unlock(&entry->d_lock);
2353 EXPORT_SYMBOL(d_rehash);
2356 * dentry_update_name_case - update case insensitive dentry with a new name
2357 * @dentry: dentry to be updated
2360 * Update a case insensitive dentry with new case of name.
2362 * dentry must have been returned by d_lookup with name @name. Old and new
2363 * name lengths must match (ie. no d_compare which allows mismatched name
2366 * Parent inode i_mutex must be held over d_lookup and into this call (to
2367 * keep renames and concurrent inserts, and readdir(2) away).
2369 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2371 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2372 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2374 spin_lock(&dentry->d_lock);
2375 write_seqcount_begin(&dentry->d_seq);
2376 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2377 write_seqcount_end(&dentry->d_seq);
2378 spin_unlock(&dentry->d_lock);
2380 EXPORT_SYMBOL(dentry_update_name_case);
2382 static void switch_names(struct dentry *dentry, struct dentry *target)
2384 if (dname_external(target)) {
2385 if (dname_external(dentry)) {
2387 * Both external: swap the pointers
2389 swap(target->d_name.name, dentry->d_name.name);
2392 * dentry:internal, target:external. Steal target's
2393 * storage and make target internal.
2395 memcpy(target->d_iname, dentry->d_name.name,
2396 dentry->d_name.len + 1);
2397 dentry->d_name.name = target->d_name.name;
2398 target->d_name.name = target->d_iname;
2401 if (dname_external(dentry)) {
2403 * dentry:external, target:internal. Give dentry's
2404 * storage to target and make dentry internal
2406 memcpy(dentry->d_iname, target->d_name.name,
2407 target->d_name.len + 1);
2408 target->d_name.name = dentry->d_name.name;
2409 dentry->d_name.name = dentry->d_iname;
2412 * Both are internal.
2415 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2416 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2417 swap(((long *) &dentry->d_iname)[i],
2418 ((long *) &target->d_iname)[i]);
2422 swap(dentry->d_name.len, target->d_name.len);
2425 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2428 * XXXX: do we really need to take target->d_lock?
2430 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2431 spin_lock(&target->d_parent->d_lock);
2433 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2434 spin_lock(&dentry->d_parent->d_lock);
2435 spin_lock_nested(&target->d_parent->d_lock,
2436 DENTRY_D_LOCK_NESTED);
2438 spin_lock(&target->d_parent->d_lock);
2439 spin_lock_nested(&dentry->d_parent->d_lock,
2440 DENTRY_D_LOCK_NESTED);
2443 if (target < dentry) {
2444 spin_lock_nested(&target->d_lock, 2);
2445 spin_lock_nested(&dentry->d_lock, 3);
2447 spin_lock_nested(&dentry->d_lock, 2);
2448 spin_lock_nested(&target->d_lock, 3);
2452 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2453 struct dentry *target)
2455 if (target->d_parent != dentry->d_parent)
2456 spin_unlock(&dentry->d_parent->d_lock);
2457 if (target->d_parent != target)
2458 spin_unlock(&target->d_parent->d_lock);
2462 * When switching names, the actual string doesn't strictly have to
2463 * be preserved in the target - because we're dropping the target
2464 * anyway. As such, we can just do a simple memcpy() to copy over
2465 * the new name before we switch.
2467 * Note that we have to be a lot more careful about getting the hash
2468 * switched - we have to switch the hash value properly even if it
2469 * then no longer matches the actual (corrupted) string of the target.
2470 * The hash value has to match the hash queue that the dentry is on..
2473 * __d_move - move a dentry
2474 * @dentry: entry to move
2475 * @target: new dentry
2476 * @exchange: exchange the two dentries
2478 * Update the dcache to reflect the move of a file name. Negative
2479 * dcache entries should not be moved in this way. Caller must hold
2480 * rename_lock, the i_mutex of the source and target directories,
2481 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2483 static void __d_move(struct dentry *dentry, struct dentry *target,
2486 if (!dentry->d_inode)
2487 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2489 BUG_ON(d_ancestor(dentry, target));
2490 BUG_ON(d_ancestor(target, dentry));
2492 dentry_lock_for_move(dentry, target);
2494 write_seqcount_begin(&dentry->d_seq);
2495 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2497 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2500 * Move the dentry to the target hash queue. Don't bother checking
2501 * for the same hash queue because of how unlikely it is.
2504 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2507 * Unhash the target (d_delete() is not usable here). If exchanging
2508 * the two dentries, then rehash onto the other's hash queue.
2513 d_hash(dentry->d_parent, dentry->d_name.hash));
2516 list_del(&dentry->d_u.d_child);
2517 list_del(&target->d_u.d_child);
2519 /* Switch the names.. */
2520 switch_names(dentry, target);
2521 swap(dentry->d_name.hash, target->d_name.hash);
2523 /* ... and switch the parents */
2524 if (IS_ROOT(dentry)) {
2525 dentry->d_parent = target->d_parent;
2526 target->d_parent = target;
2527 INIT_LIST_HEAD(&target->d_u.d_child);
2529 swap(dentry->d_parent, target->d_parent);
2531 /* And add them back to the (new) parent lists */
2532 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2535 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2537 write_seqcount_end(&target->d_seq);
2538 write_seqcount_end(&dentry->d_seq);
2540 dentry_unlock_parents_for_move(dentry, target);
2542 fsnotify_d_move(target);
2543 spin_unlock(&target->d_lock);
2544 fsnotify_d_move(dentry);
2545 spin_unlock(&dentry->d_lock);
2549 * d_move - move a dentry
2550 * @dentry: entry to move
2551 * @target: new dentry
2553 * Update the dcache to reflect the move of a file name. Negative
2554 * dcache entries should not be moved in this way. See the locking
2555 * requirements for __d_move.
2557 void d_move(struct dentry *dentry, struct dentry *target)
2559 write_seqlock(&rename_lock);
2560 __d_move(dentry, target, false);
2561 write_sequnlock(&rename_lock);
2563 EXPORT_SYMBOL(d_move);
2566 * d_exchange - exchange two dentries
2567 * @dentry1: first dentry
2568 * @dentry2: second dentry
2570 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2572 write_seqlock(&rename_lock);
2574 WARN_ON(!dentry1->d_inode);
2575 WARN_ON(!dentry2->d_inode);
2576 WARN_ON(IS_ROOT(dentry1));
2577 WARN_ON(IS_ROOT(dentry2));
2579 __d_move(dentry1, dentry2, true);
2581 write_sequnlock(&rename_lock);
2585 * d_ancestor - search for an ancestor
2586 * @p1: ancestor dentry
2589 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2590 * an ancestor of p2, else NULL.
2592 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2596 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2597 if (p->d_parent == p1)
2604 * This helper attempts to cope with remotely renamed directories
2606 * It assumes that the caller is already holding
2607 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2609 * Note: If ever the locking in lock_rename() changes, then please
2610 * remember to update this too...
2612 static struct dentry *__d_unalias(struct inode *inode,
2613 struct dentry *dentry, struct dentry *alias)
2615 struct mutex *m1 = NULL, *m2 = NULL;
2616 struct dentry *ret = ERR_PTR(-EBUSY);
2618 /* If alias and dentry share a parent, then no extra locks required */
2619 if (alias->d_parent == dentry->d_parent)
2622 /* See lock_rename() */
2623 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2625 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2626 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2628 m2 = &alias->d_parent->d_inode->i_mutex;
2630 if (likely(!d_mountpoint(alias))) {
2631 __d_move(alias, dentry, false);
2635 spin_unlock(&inode->i_lock);
2644 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2645 * named dentry in place of the dentry to be replaced.
2646 * returns with anon->d_lock held!
2648 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2650 struct dentry *dparent;
2652 dentry_lock_for_move(anon, dentry);
2654 write_seqcount_begin(&dentry->d_seq);
2655 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2657 dparent = dentry->d_parent;
2659 switch_names(dentry, anon);
2660 swap(dentry->d_name.hash, anon->d_name.hash);
2662 dentry->d_parent = dentry;
2663 list_del_init(&dentry->d_u.d_child);
2664 anon->d_parent = dparent;
2665 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2667 write_seqcount_end(&dentry->d_seq);
2668 write_seqcount_end(&anon->d_seq);
2670 dentry_unlock_parents_for_move(anon, dentry);
2671 spin_unlock(&dentry->d_lock);
2673 /* anon->d_lock still locked, returns locked */
2677 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2678 * @inode: the inode which may have a disconnected dentry
2679 * @dentry: a negative dentry which we want to point to the inode.
2681 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2682 * place of the given dentry and return it, else simply d_add the inode
2683 * to the dentry and return NULL.
2685 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2686 * we should error out: directories can't have multiple aliases.
2688 * This is needed in the lookup routine of any filesystem that is exportable
2689 * (via knfsd) so that we can build dcache paths to directories effectively.
2691 * If a dentry was found and moved, then it is returned. Otherwise NULL
2692 * is returned. This matches the expected return value of ->lookup.
2694 * Cluster filesystems may call this function with a negative, hashed dentry.
2695 * In that case, we know that the inode will be a regular file, and also this
2696 * will only occur during atomic_open. So we need to check for the dentry
2697 * being already hashed only in the final case.
2699 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2701 struct dentry *new = NULL;
2704 return ERR_CAST(inode);
2706 if (inode && S_ISDIR(inode->i_mode)) {
2707 spin_lock(&inode->i_lock);
2708 new = __d_find_any_alias(inode);
2710 if (!IS_ROOT(new)) {
2711 spin_unlock(&inode->i_lock);
2713 return ERR_PTR(-EIO);
2715 write_seqlock(&rename_lock);
2716 __d_materialise_dentry(dentry, new);
2717 write_sequnlock(&rename_lock);
2720 spin_unlock(&new->d_lock);
2721 spin_unlock(&inode->i_lock);
2722 security_d_instantiate(new, inode);
2725 /* already taking inode->i_lock, so d_add() by hand */
2726 __d_instantiate(dentry, inode);
2727 spin_unlock(&inode->i_lock);
2728 security_d_instantiate(dentry, inode);
2732 d_instantiate(dentry, inode);
2733 if (d_unhashed(dentry))
2738 EXPORT_SYMBOL(d_splice_alias);
2741 * d_materialise_unique - introduce an inode into the tree
2742 * @dentry: candidate dentry
2743 * @inode: inode to bind to the dentry, to which aliases may be attached
2745 * Introduces an dentry into the tree, substituting an extant disconnected
2746 * root directory alias in its place if there is one. Caller must hold the
2747 * i_mutex of the parent directory.
2749 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2751 struct dentry *actual;
2753 BUG_ON(!d_unhashed(dentry));
2757 __d_instantiate(dentry, NULL);
2762 spin_lock(&inode->i_lock);
2764 if (S_ISDIR(inode->i_mode)) {
2765 struct dentry *alias;
2767 /* Does an aliased dentry already exist? */
2768 alias = __d_find_alias(inode, 0);
2771 write_seqlock(&rename_lock);
2773 if (d_ancestor(alias, dentry)) {
2774 /* Check for loops */
2775 actual = ERR_PTR(-ELOOP);
2776 spin_unlock(&inode->i_lock);
2777 } else if (IS_ROOT(alias)) {
2778 /* Is this an anonymous mountpoint that we
2779 * could splice into our tree? */
2780 __d_materialise_dentry(dentry, alias);
2781 write_sequnlock(&rename_lock);
2785 /* Nope, but we must(!) avoid directory
2786 * aliasing. This drops inode->i_lock */
2787 actual = __d_unalias(inode, dentry, alias);
2789 write_sequnlock(&rename_lock);
2790 if (IS_ERR(actual)) {
2791 if (PTR_ERR(actual) == -ELOOP)
2792 pr_warn_ratelimited(
2793 "VFS: Lookup of '%s' in %s %s"
2794 " would have caused loop\n",
2795 dentry->d_name.name,
2796 inode->i_sb->s_type->name,
2804 /* Add a unique reference */
2805 actual = __d_instantiate_unique(dentry, inode);
2809 BUG_ON(!d_unhashed(actual));
2811 spin_lock(&actual->d_lock);
2814 spin_unlock(&actual->d_lock);
2815 spin_unlock(&inode->i_lock);
2817 if (actual == dentry) {
2818 security_d_instantiate(dentry, inode);
2825 EXPORT_SYMBOL_GPL(d_materialise_unique);
2827 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2831 return -ENAMETOOLONG;
2833 memcpy(*buffer, str, namelen);
2838 * prepend_name - prepend a pathname in front of current buffer pointer
2839 * @buffer: buffer pointer
2840 * @buflen: allocated length of the buffer
2841 * @name: name string and length qstr structure
2843 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2844 * make sure that either the old or the new name pointer and length are
2845 * fetched. However, there may be mismatch between length and pointer.
2846 * The length cannot be trusted, we need to copy it byte-by-byte until
2847 * the length is reached or a null byte is found. It also prepends "/" at
2848 * the beginning of the name. The sequence number check at the caller will
2849 * retry it again when a d_move() does happen. So any garbage in the buffer
2850 * due to mismatched pointer and length will be discarded.
2852 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2854 const char *dname = ACCESS_ONCE(name->name);
2855 u32 dlen = ACCESS_ONCE(name->len);
2858 *buflen -= dlen + 1;
2860 return -ENAMETOOLONG;
2861 p = *buffer -= dlen + 1;
2873 * prepend_path - Prepend path string to a buffer
2874 * @path: the dentry/vfsmount to report
2875 * @root: root vfsmnt/dentry
2876 * @buffer: pointer to the end of the buffer
2877 * @buflen: pointer to buffer length
2879 * The function will first try to write out the pathname without taking any
2880 * lock other than the RCU read lock to make sure that dentries won't go away.
2881 * It only checks the sequence number of the global rename_lock as any change
2882 * in the dentry's d_seq will be preceded by changes in the rename_lock
2883 * sequence number. If the sequence number had been changed, it will restart
2884 * the whole pathname back-tracing sequence again by taking the rename_lock.
2885 * In this case, there is no need to take the RCU read lock as the recursive
2886 * parent pointer references will keep the dentry chain alive as long as no
2887 * rename operation is performed.
2889 static int prepend_path(const struct path *path,
2890 const struct path *root,
2891 char **buffer, int *buflen)
2893 struct dentry *dentry;
2894 struct vfsmount *vfsmnt;
2897 unsigned seq, m_seq = 0;
2903 read_seqbegin_or_lock(&mount_lock, &m_seq);
2910 dentry = path->dentry;
2912 mnt = real_mount(vfsmnt);
2913 read_seqbegin_or_lock(&rename_lock, &seq);
2914 while (dentry != root->dentry || vfsmnt != root->mnt) {
2915 struct dentry * parent;
2917 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2918 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2920 if (mnt != parent) {
2921 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2927 * Filesystems needing to implement special "root names"
2928 * should do so with ->d_dname()
2930 if (IS_ROOT(dentry) &&
2931 (dentry->d_name.len != 1 ||
2932 dentry->d_name.name[0] != '/')) {
2933 WARN(1, "Root dentry has weird name <%.*s>\n",
2934 (int) dentry->d_name.len,
2935 dentry->d_name.name);
2938 error = is_mounted(vfsmnt) ? 1 : 2;
2941 parent = dentry->d_parent;
2943 error = prepend_name(&bptr, &blen, &dentry->d_name);
2951 if (need_seqretry(&rename_lock, seq)) {
2955 done_seqretry(&rename_lock, seq);
2959 if (need_seqretry(&mount_lock, m_seq)) {
2963 done_seqretry(&mount_lock, m_seq);
2965 if (error >= 0 && bptr == *buffer) {
2967 error = -ENAMETOOLONG;
2977 * __d_path - return the path of a dentry
2978 * @path: the dentry/vfsmount to report
2979 * @root: root vfsmnt/dentry
2980 * @buf: buffer to return value in
2981 * @buflen: buffer length
2983 * Convert a dentry into an ASCII path name.
2985 * Returns a pointer into the buffer or an error code if the
2986 * path was too long.
2988 * "buflen" should be positive.
2990 * If the path is not reachable from the supplied root, return %NULL.
2992 char *__d_path(const struct path *path,
2993 const struct path *root,
2994 char *buf, int buflen)
2996 char *res = buf + buflen;
2999 prepend(&res, &buflen, "\0", 1);
3000 error = prepend_path(path, root, &res, &buflen);
3003 return ERR_PTR(error);
3009 char *d_absolute_path(const struct path *path,
3010 char *buf, int buflen)
3012 struct path root = {};
3013 char *res = buf + buflen;
3016 prepend(&res, &buflen, "\0", 1);
3017 error = prepend_path(path, &root, &res, &buflen);
3022 return ERR_PTR(error);
3027 * same as __d_path but appends "(deleted)" for unlinked files.
3029 static int path_with_deleted(const struct path *path,
3030 const struct path *root,
3031 char **buf, int *buflen)
3033 prepend(buf, buflen, "\0", 1);
3034 if (d_unlinked(path->dentry)) {
3035 int error = prepend(buf, buflen, " (deleted)", 10);
3040 return prepend_path(path, root, buf, buflen);
3043 static int prepend_unreachable(char **buffer, int *buflen)
3045 return prepend(buffer, buflen, "(unreachable)", 13);
3048 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3053 seq = read_seqcount_begin(&fs->seq);
3055 } while (read_seqcount_retry(&fs->seq, seq));
3059 * d_path - return the path of a dentry
3060 * @path: path to report
3061 * @buf: buffer to return value in
3062 * @buflen: buffer length
3064 * Convert a dentry into an ASCII path name. If the entry has been deleted
3065 * the string " (deleted)" is appended. Note that this is ambiguous.
3067 * Returns a pointer into the buffer or an error code if the path was
3068 * too long. Note: Callers should use the returned pointer, not the passed
3069 * in buffer, to use the name! The implementation often starts at an offset
3070 * into the buffer, and may leave 0 bytes at the start.
3072 * "buflen" should be positive.
3074 char *d_path(const struct path *path, char *buf, int buflen)
3076 char *res = buf + buflen;
3081 * We have various synthetic filesystems that never get mounted. On
3082 * these filesystems dentries are never used for lookup purposes, and
3083 * thus don't need to be hashed. They also don't need a name until a
3084 * user wants to identify the object in /proc/pid/fd/. The little hack
3085 * below allows us to generate a name for these objects on demand:
3087 * Some pseudo inodes are mountable. When they are mounted
3088 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3089 * and instead have d_path return the mounted path.
3091 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3092 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3093 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3096 get_fs_root_rcu(current->fs, &root);
3097 error = path_with_deleted(path, &root, &res, &buflen);
3101 res = ERR_PTR(error);
3104 EXPORT_SYMBOL(d_path);
3107 * Helper function for dentry_operations.d_dname() members
3109 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3110 const char *fmt, ...)
3116 va_start(args, fmt);
3117 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3120 if (sz > sizeof(temp) || sz > buflen)
3121 return ERR_PTR(-ENAMETOOLONG);
3123 buffer += buflen - sz;
3124 return memcpy(buffer, temp, sz);
3127 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3129 char *end = buffer + buflen;
3130 /* these dentries are never renamed, so d_lock is not needed */
3131 if (prepend(&end, &buflen, " (deleted)", 11) ||
3132 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3133 prepend(&end, &buflen, "/", 1))
3134 end = ERR_PTR(-ENAMETOOLONG);
3137 EXPORT_SYMBOL(simple_dname);
3140 * Write full pathname from the root of the filesystem into the buffer.
3142 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3144 struct dentry *dentry;
3157 prepend(&end, &len, "\0", 1);
3161 read_seqbegin_or_lock(&rename_lock, &seq);
3162 while (!IS_ROOT(dentry)) {
3163 struct dentry *parent = dentry->d_parent;
3166 error = prepend_name(&end, &len, &dentry->d_name);
3175 if (need_seqretry(&rename_lock, seq)) {
3179 done_seqretry(&rename_lock, seq);
3184 return ERR_PTR(-ENAMETOOLONG);
3187 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3189 return __dentry_path(dentry, buf, buflen);
3191 EXPORT_SYMBOL(dentry_path_raw);
3193 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3198 if (d_unlinked(dentry)) {
3200 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3204 retval = __dentry_path(dentry, buf, buflen);
3205 if (!IS_ERR(retval) && p)
3206 *p = '/'; /* restore '/' overriden with '\0' */
3209 return ERR_PTR(-ENAMETOOLONG);
3212 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3218 seq = read_seqcount_begin(&fs->seq);
3221 } while (read_seqcount_retry(&fs->seq, seq));
3225 * NOTE! The user-level library version returns a
3226 * character pointer. The kernel system call just
3227 * returns the length of the buffer filled (which
3228 * includes the ending '\0' character), or a negative
3229 * error value. So libc would do something like
3231 * char *getcwd(char * buf, size_t size)
3235 * retval = sys_getcwd(buf, size);
3242 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3245 struct path pwd, root;
3246 char *page = __getname();
3252 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3255 if (!d_unlinked(pwd.dentry)) {
3257 char *cwd = page + PATH_MAX;
3258 int buflen = PATH_MAX;
3260 prepend(&cwd, &buflen, "\0", 1);
3261 error = prepend_path(&pwd, &root, &cwd, &buflen);
3267 /* Unreachable from current root */
3269 error = prepend_unreachable(&cwd, &buflen);
3275 len = PATH_MAX + page - cwd;
3278 if (copy_to_user(buf, cwd, len))
3291 * Test whether new_dentry is a subdirectory of old_dentry.
3293 * Trivially implemented using the dcache structure
3297 * is_subdir - is new dentry a subdirectory of old_dentry
3298 * @new_dentry: new dentry
3299 * @old_dentry: old dentry
3301 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3302 * Returns 0 otherwise.
3303 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3306 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3311 if (new_dentry == old_dentry)
3315 /* for restarting inner loop in case of seq retry */
3316 seq = read_seqbegin(&rename_lock);
3318 * Need rcu_readlock to protect against the d_parent trashing
3322 if (d_ancestor(old_dentry, new_dentry))
3327 } while (read_seqretry(&rename_lock, seq));
3332 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3334 struct dentry *root = data;
3335 if (dentry != root) {
3336 if (d_unhashed(dentry) || !dentry->d_inode)
3339 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3340 dentry->d_flags |= DCACHE_GENOCIDE;
3341 dentry->d_lockref.count--;
3344 return D_WALK_CONTINUE;
3347 void d_genocide(struct dentry *parent)
3349 d_walk(parent, parent, d_genocide_kill, NULL);
3352 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3354 inode_dec_link_count(inode);
3355 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3356 !hlist_unhashed(&dentry->d_alias) ||
3357 !d_unlinked(dentry));
3358 spin_lock(&dentry->d_parent->d_lock);
3359 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3360 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3361 (unsigned long long)inode->i_ino);
3362 spin_unlock(&dentry->d_lock);
3363 spin_unlock(&dentry->d_parent->d_lock);
3364 d_instantiate(dentry, inode);
3366 EXPORT_SYMBOL(d_tmpfile);
3368 static __initdata unsigned long dhash_entries;
3369 static int __init set_dhash_entries(char *str)
3373 dhash_entries = simple_strtoul(str, &str, 0);
3376 __setup("dhash_entries=", set_dhash_entries);
3378 static void __init dcache_init_early(void)
3382 /* If hashes are distributed across NUMA nodes, defer
3383 * hash allocation until vmalloc space is available.
3389 alloc_large_system_hash("Dentry cache",
3390 sizeof(struct hlist_bl_head),
3399 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3400 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3403 static void __init dcache_init(void)
3408 * A constructor could be added for stable state like the lists,
3409 * but it is probably not worth it because of the cache nature
3412 dentry_cache = KMEM_CACHE(dentry,
3413 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3415 /* Hash may have been set up in dcache_init_early */
3420 alloc_large_system_hash("Dentry cache",
3421 sizeof(struct hlist_bl_head),
3430 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3431 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3434 /* SLAB cache for __getname() consumers */
3435 struct kmem_cache *names_cachep __read_mostly;
3436 EXPORT_SYMBOL(names_cachep);
3438 EXPORT_SYMBOL(d_genocide);
3440 void __init vfs_caches_init_early(void)
3442 dcache_init_early();
3446 void __init vfs_caches_init(unsigned long mempages)
3448 unsigned long reserve;
3450 /* Base hash sizes on available memory, with a reserve equal to
3451 150% of current kernel size */
3453 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3454 mempages -= reserve;
3456 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3457 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3461 files_init(mempages);