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 return dentry_hashtable + hash_32(hash, d_hash_shift);
112 /* Statistics gathering. */
113 struct dentry_stat_t dentry_stat = {
117 static DEFINE_PER_CPU(long, nr_dentry);
118 static DEFINE_PER_CPU(long, nr_dentry_unused);
120 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 * Here we resort to our own counters instead of using generic per-cpu counters
124 * for consistency with what the vfs inode code does. We are expected to harvest
125 * better code and performance by having our own specialized counters.
127 * Please note that the loop is done over all possible CPUs, not over all online
128 * CPUs. The reason for this is that we don't want to play games with CPUs going
129 * on and off. If one of them goes off, we will just keep their counters.
131 * glommer: See cffbc8a for details, and if you ever intend to change this,
132 * please update all vfs counters to match.
134 static long get_nr_dentry(void)
138 for_each_possible_cpu(i)
139 sum += per_cpu(nr_dentry, i);
140 return sum < 0 ? 0 : sum;
143 static long get_nr_dentry_unused(void)
147 for_each_possible_cpu(i)
148 sum += per_cpu(nr_dentry_unused, i);
149 return sum < 0 ? 0 : sum;
152 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
153 size_t *lenp, loff_t *ppos)
155 dentry_stat.nr_dentry = get_nr_dentry();
156 dentry_stat.nr_unused = get_nr_dentry_unused();
157 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
162 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
163 * The strings are both count bytes long, and count is non-zero.
165 #ifdef CONFIG_DCACHE_WORD_ACCESS
167 #include <asm/word-at-a-time.h>
169 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
170 * aligned allocation for this particular component. We don't
171 * strictly need the load_unaligned_zeropad() safety, but it
172 * doesn't hurt either.
174 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
175 * need the careful unaligned handling.
177 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
179 unsigned long a,b,mask;
182 a = *(unsigned long *)cs;
183 b = load_unaligned_zeropad(ct);
184 if (tcount < sizeof(unsigned long))
186 if (unlikely(a != b))
188 cs += sizeof(unsigned long);
189 ct += sizeof(unsigned long);
190 tcount -= sizeof(unsigned long);
194 mask = bytemask_from_count(tcount);
195 return unlikely(!!((a ^ b) & mask));
200 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
214 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
216 const unsigned char *cs;
218 * Be careful about RCU walk racing with rename:
219 * use ACCESS_ONCE to fetch the name pointer.
221 * NOTE! Even if a rename will mean that the length
222 * was not loaded atomically, we don't care. The
223 * RCU walk will check the sequence count eventually,
224 * and catch it. And we won't overrun the buffer,
225 * because we're reading the name pointer atomically,
226 * and a dentry name is guaranteed to be properly
227 * terminated with a NUL byte.
229 * End result: even if 'len' is wrong, we'll exit
230 * early because the data cannot match (there can
231 * be no NUL in the ct/tcount data)
233 cs = ACCESS_ONCE(dentry->d_name.name);
234 smp_read_barrier_depends();
235 return dentry_string_cmp(cs, ct, tcount);
238 struct external_name {
241 struct rcu_head head;
243 unsigned char name[];
246 static inline struct external_name *external_name(struct dentry *dentry)
248 return container_of(dentry->d_name.name, struct external_name, name[0]);
251 static void __d_free(struct rcu_head *head)
253 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
255 WARN_ON(!hlist_unhashed(&dentry->d_alias));
256 kmem_cache_free(dentry_cache, dentry);
259 static void __d_free_external(struct rcu_head *head)
261 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
262 WARN_ON(!hlist_unhashed(&dentry->d_alias));
263 kfree(external_name(dentry));
264 kmem_cache_free(dentry_cache, dentry);
267 static void dentry_free(struct dentry *dentry)
269 if (unlikely(dname_external(dentry))) {
270 struct external_name *p = external_name(dentry);
271 if (likely(atomic_dec_and_test(&p->u.count))) {
272 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
276 /* if dentry was never visible to RCU, immediate free is OK */
277 if (!(dentry->d_flags & DCACHE_RCUACCESS))
278 __d_free(&dentry->d_u.d_rcu);
280 call_rcu(&dentry->d_u.d_rcu, __d_free);
284 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
285 * @dentry: the target dentry
286 * After this call, in-progress rcu-walk path lookup will fail. This
287 * should be called after unhashing, and after changing d_inode (if
288 * the dentry has not already been unhashed).
290 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
292 assert_spin_locked(&dentry->d_lock);
293 /* Go through a barrier */
294 write_seqcount_barrier(&dentry->d_seq);
298 * Release the dentry's inode, using the filesystem
299 * d_iput() operation if defined. Dentry has no refcount
302 static void dentry_iput(struct dentry * dentry)
303 __releases(dentry->d_lock)
304 __releases(dentry->d_inode->i_lock)
306 struct inode *inode = dentry->d_inode;
308 dentry->d_inode = NULL;
309 hlist_del_init(&dentry->d_alias);
310 spin_unlock(&dentry->d_lock);
311 spin_unlock(&inode->i_lock);
313 fsnotify_inoderemove(inode);
314 if (dentry->d_op && dentry->d_op->d_iput)
315 dentry->d_op->d_iput(dentry, inode);
319 spin_unlock(&dentry->d_lock);
324 * Release the dentry's inode, using the filesystem
325 * d_iput() operation if defined. dentry remains in-use.
327 static void dentry_unlink_inode(struct dentry * dentry)
328 __releases(dentry->d_lock)
329 __releases(dentry->d_inode->i_lock)
331 struct inode *inode = dentry->d_inode;
332 __d_clear_type(dentry);
333 dentry->d_inode = NULL;
334 hlist_del_init(&dentry->d_alias);
335 dentry_rcuwalk_barrier(dentry);
336 spin_unlock(&dentry->d_lock);
337 spin_unlock(&inode->i_lock);
339 fsnotify_inoderemove(inode);
340 if (dentry->d_op && dentry->d_op->d_iput)
341 dentry->d_op->d_iput(dentry, inode);
347 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
348 * is in use - which includes both the "real" per-superblock
349 * LRU list _and_ the DCACHE_SHRINK_LIST use.
351 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
352 * on the shrink list (ie not on the superblock LRU list).
354 * The per-cpu "nr_dentry_unused" counters are updated with
355 * the DCACHE_LRU_LIST bit.
357 * These helper functions make sure we always follow the
358 * rules. d_lock must be held by the caller.
360 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
361 static void d_lru_add(struct dentry *dentry)
363 D_FLAG_VERIFY(dentry, 0);
364 dentry->d_flags |= DCACHE_LRU_LIST;
365 this_cpu_inc(nr_dentry_unused);
366 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
369 static void d_lru_del(struct dentry *dentry)
371 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
372 dentry->d_flags &= ~DCACHE_LRU_LIST;
373 this_cpu_dec(nr_dentry_unused);
374 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
377 static void d_shrink_del(struct dentry *dentry)
379 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
380 list_del_init(&dentry->d_lru);
381 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
382 this_cpu_dec(nr_dentry_unused);
385 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
387 D_FLAG_VERIFY(dentry, 0);
388 list_add(&dentry->d_lru, list);
389 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
390 this_cpu_inc(nr_dentry_unused);
394 * These can only be called under the global LRU lock, ie during the
395 * callback for freeing the LRU list. "isolate" removes it from the
396 * LRU lists entirely, while shrink_move moves it to the indicated
399 static void d_lru_isolate(struct dentry *dentry)
401 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
402 dentry->d_flags &= ~DCACHE_LRU_LIST;
403 this_cpu_dec(nr_dentry_unused);
404 list_del_init(&dentry->d_lru);
407 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
409 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
410 dentry->d_flags |= DCACHE_SHRINK_LIST;
411 list_move_tail(&dentry->d_lru, list);
415 * dentry_lru_(add|del)_list) must be called with d_lock held.
417 static void dentry_lru_add(struct dentry *dentry)
419 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
424 * d_drop - drop a dentry
425 * @dentry: dentry to drop
427 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
428 * be found through a VFS lookup any more. Note that this is different from
429 * deleting the dentry - d_delete will try to mark the dentry negative if
430 * possible, giving a successful _negative_ lookup, while d_drop will
431 * just make the cache lookup fail.
433 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
434 * reason (NFS timeouts or autofs deletes).
436 * __d_drop requires dentry->d_lock.
438 void __d_drop(struct dentry *dentry)
440 if (!d_unhashed(dentry)) {
441 struct hlist_bl_head *b;
443 * Hashed dentries are normally on the dentry hashtable,
444 * with the exception of those newly allocated by
445 * d_obtain_alias, which are always IS_ROOT:
447 if (unlikely(IS_ROOT(dentry)))
448 b = &dentry->d_sb->s_anon;
450 b = d_hash(dentry->d_parent, dentry->d_name.hash);
453 __hlist_bl_del(&dentry->d_hash);
454 dentry->d_hash.pprev = NULL;
456 dentry_rcuwalk_barrier(dentry);
459 EXPORT_SYMBOL(__d_drop);
461 void d_drop(struct dentry *dentry)
463 spin_lock(&dentry->d_lock);
465 spin_unlock(&dentry->d_lock);
467 EXPORT_SYMBOL(d_drop);
469 static void __dentry_kill(struct dentry *dentry)
471 struct dentry *parent = NULL;
472 bool can_free = true;
473 if (!IS_ROOT(dentry))
474 parent = dentry->d_parent;
477 * The dentry is now unrecoverably dead to the world.
479 lockref_mark_dead(&dentry->d_lockref);
482 * inform the fs via d_prune that this dentry is about to be
483 * unhashed and destroyed.
485 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
486 dentry->d_op->d_prune(dentry);
488 if (dentry->d_flags & DCACHE_LRU_LIST) {
489 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
492 /* if it was on the hash then remove it */
494 list_del(&dentry->d_u.d_child);
496 * Inform d_walk() that we are no longer attached to the
499 dentry->d_flags |= DCACHE_DENTRY_KILLED;
501 spin_unlock(&parent->d_lock);
504 * dentry_iput drops the locks, at which point nobody (except
505 * transient RCU lookups) can reach this dentry.
507 BUG_ON((int)dentry->d_lockref.count > 0);
508 this_cpu_dec(nr_dentry);
509 if (dentry->d_op && dentry->d_op->d_release)
510 dentry->d_op->d_release(dentry);
512 spin_lock(&dentry->d_lock);
513 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
514 dentry->d_flags |= DCACHE_MAY_FREE;
517 spin_unlock(&dentry->d_lock);
518 if (likely(can_free))
523 * Finish off a dentry we've decided to kill.
524 * dentry->d_lock must be held, returns with it unlocked.
525 * If ref is non-zero, then decrement the refcount too.
526 * Returns dentry requiring refcount drop, or NULL if we're done.
528 static struct dentry *dentry_kill(struct dentry *dentry)
529 __releases(dentry->d_lock)
531 struct inode *inode = dentry->d_inode;
532 struct dentry *parent = NULL;
534 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
537 if (!IS_ROOT(dentry)) {
538 parent = dentry->d_parent;
539 if (unlikely(!spin_trylock(&parent->d_lock))) {
541 spin_unlock(&inode->i_lock);
546 __dentry_kill(dentry);
550 spin_unlock(&dentry->d_lock);
552 return dentry; /* try again with same dentry */
555 static inline struct dentry *lock_parent(struct dentry *dentry)
557 struct dentry *parent = dentry->d_parent;
560 if (unlikely((int)dentry->d_lockref.count < 0))
562 if (likely(spin_trylock(&parent->d_lock)))
565 spin_unlock(&dentry->d_lock);
567 parent = ACCESS_ONCE(dentry->d_parent);
568 spin_lock(&parent->d_lock);
570 * We can't blindly lock dentry until we are sure
571 * that we won't violate the locking order.
572 * Any changes of dentry->d_parent must have
573 * been done with parent->d_lock held, so
574 * spin_lock() above is enough of a barrier
575 * for checking if it's still our child.
577 if (unlikely(parent != dentry->d_parent)) {
578 spin_unlock(&parent->d_lock);
582 if (parent != dentry)
583 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
592 * This is complicated by the fact that we do not want to put
593 * dentries that are no longer on any hash chain on the unused
594 * list: we'd much rather just get rid of them immediately.
596 * However, that implies that we have to traverse the dentry
597 * tree upwards to the parents which might _also_ now be
598 * scheduled for deletion (it may have been only waiting for
599 * its last child to go away).
601 * This tail recursion is done by hand as we don't want to depend
602 * on the compiler to always get this right (gcc generally doesn't).
603 * Real recursion would eat up our stack space.
607 * dput - release a dentry
608 * @dentry: dentry to release
610 * Release a dentry. This will drop the usage count and if appropriate
611 * call the dentry unlink method as well as removing it from the queues and
612 * releasing its resources. If the parent dentries were scheduled for release
613 * they too may now get deleted.
615 void dput(struct dentry *dentry)
617 if (unlikely(!dentry))
621 if (lockref_put_or_lock(&dentry->d_lockref))
624 /* Unreachable? Get rid of it */
625 if (unlikely(d_unhashed(dentry)))
628 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
629 if (dentry->d_op->d_delete(dentry))
633 if (!(dentry->d_flags & DCACHE_REFERENCED))
634 dentry->d_flags |= DCACHE_REFERENCED;
635 dentry_lru_add(dentry);
637 dentry->d_lockref.count--;
638 spin_unlock(&dentry->d_lock);
642 dentry = dentry_kill(dentry);
649 * d_invalidate - invalidate a dentry
650 * @dentry: dentry to invalidate
652 * Try to invalidate the dentry if it turns out to be
653 * possible. If there are reasons not to delete it
654 * return -EBUSY. On success return 0.
659 int d_invalidate(struct dentry * dentry)
662 * If it's already been dropped, return OK.
664 spin_lock(&dentry->d_lock);
665 if (d_unhashed(dentry)) {
666 spin_unlock(&dentry->d_lock);
669 spin_unlock(&dentry->d_lock);
671 return check_submounts_and_drop(dentry);
673 EXPORT_SYMBOL(d_invalidate);
675 /* This must be called with d_lock held */
676 static inline void __dget_dlock(struct dentry *dentry)
678 dentry->d_lockref.count++;
681 static inline void __dget(struct dentry *dentry)
683 lockref_get(&dentry->d_lockref);
686 struct dentry *dget_parent(struct dentry *dentry)
692 * Do optimistic parent lookup without any
696 ret = ACCESS_ONCE(dentry->d_parent);
697 gotref = lockref_get_not_zero(&ret->d_lockref);
699 if (likely(gotref)) {
700 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
707 * Don't need rcu_dereference because we re-check it was correct under
711 ret = dentry->d_parent;
712 spin_lock(&ret->d_lock);
713 if (unlikely(ret != dentry->d_parent)) {
714 spin_unlock(&ret->d_lock);
719 BUG_ON(!ret->d_lockref.count);
720 ret->d_lockref.count++;
721 spin_unlock(&ret->d_lock);
724 EXPORT_SYMBOL(dget_parent);
727 * d_find_alias - grab a hashed alias of inode
728 * @inode: inode in question
730 * If inode has a hashed alias, or is a directory and has any alias,
731 * acquire the reference to alias and return it. Otherwise return NULL.
732 * Notice that if inode is a directory there can be only one alias and
733 * it can be unhashed only if it has no children, or if it is the root
734 * of a filesystem, or if the directory was renamed and d_revalidate
735 * was the first vfs operation to notice.
737 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
738 * any other hashed alias over that one.
740 static struct dentry *__d_find_alias(struct inode *inode)
742 struct dentry *alias, *discon_alias;
746 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
747 spin_lock(&alias->d_lock);
748 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
749 if (IS_ROOT(alias) &&
750 (alias->d_flags & DCACHE_DISCONNECTED)) {
751 discon_alias = alias;
754 spin_unlock(&alias->d_lock);
758 spin_unlock(&alias->d_lock);
761 alias = discon_alias;
762 spin_lock(&alias->d_lock);
763 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
765 spin_unlock(&alias->d_lock);
768 spin_unlock(&alias->d_lock);
774 struct dentry *d_find_alias(struct inode *inode)
776 struct dentry *de = NULL;
778 if (!hlist_empty(&inode->i_dentry)) {
779 spin_lock(&inode->i_lock);
780 de = __d_find_alias(inode);
781 spin_unlock(&inode->i_lock);
785 EXPORT_SYMBOL(d_find_alias);
788 * Try to kill dentries associated with this inode.
789 * WARNING: you must own a reference to inode.
791 void d_prune_aliases(struct inode *inode)
793 struct dentry *dentry;
795 spin_lock(&inode->i_lock);
796 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
797 spin_lock(&dentry->d_lock);
798 if (!dentry->d_lockref.count) {
800 * inform the fs via d_prune that this dentry
801 * is about to be unhashed and destroyed.
803 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
805 dentry->d_op->d_prune(dentry);
807 __dget_dlock(dentry);
809 spin_unlock(&dentry->d_lock);
810 spin_unlock(&inode->i_lock);
814 spin_unlock(&dentry->d_lock);
816 spin_unlock(&inode->i_lock);
818 EXPORT_SYMBOL(d_prune_aliases);
820 static void shrink_dentry_list(struct list_head *list)
822 struct dentry *dentry, *parent;
824 while (!list_empty(list)) {
826 dentry = list_entry(list->prev, struct dentry, d_lru);
827 spin_lock(&dentry->d_lock);
828 parent = lock_parent(dentry);
831 * The dispose list is isolated and dentries are not accounted
832 * to the LRU here, so we can simply remove it from the list
833 * here regardless of whether it is referenced or not.
835 d_shrink_del(dentry);
838 * We found an inuse dentry which was not removed from
839 * the LRU because of laziness during lookup. Do not free it.
841 if ((int)dentry->d_lockref.count > 0) {
842 spin_unlock(&dentry->d_lock);
844 spin_unlock(&parent->d_lock);
849 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
850 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
851 spin_unlock(&dentry->d_lock);
853 spin_unlock(&parent->d_lock);
859 inode = dentry->d_inode;
860 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
861 d_shrink_add(dentry, list);
862 spin_unlock(&dentry->d_lock);
864 spin_unlock(&parent->d_lock);
868 __dentry_kill(dentry);
871 * We need to prune ancestors too. This is necessary to prevent
872 * quadratic behavior of shrink_dcache_parent(), but is also
873 * expected to be beneficial in reducing dentry cache
877 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
878 parent = lock_parent(dentry);
879 if (dentry->d_lockref.count != 1) {
880 dentry->d_lockref.count--;
881 spin_unlock(&dentry->d_lock);
883 spin_unlock(&parent->d_lock);
886 inode = dentry->d_inode; /* can't be NULL */
887 if (unlikely(!spin_trylock(&inode->i_lock))) {
888 spin_unlock(&dentry->d_lock);
890 spin_unlock(&parent->d_lock);
894 __dentry_kill(dentry);
900 static enum lru_status
901 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
903 struct list_head *freeable = arg;
904 struct dentry *dentry = container_of(item, struct dentry, d_lru);
908 * we are inverting the lru lock/dentry->d_lock here,
909 * so use a trylock. If we fail to get the lock, just skip
912 if (!spin_trylock(&dentry->d_lock))
916 * Referenced dentries are still in use. If they have active
917 * counts, just remove them from the LRU. Otherwise give them
918 * another pass through the LRU.
920 if (dentry->d_lockref.count) {
921 d_lru_isolate(dentry);
922 spin_unlock(&dentry->d_lock);
926 if (dentry->d_flags & DCACHE_REFERENCED) {
927 dentry->d_flags &= ~DCACHE_REFERENCED;
928 spin_unlock(&dentry->d_lock);
931 * The list move itself will be made by the common LRU code. At
932 * this point, we've dropped the dentry->d_lock but keep the
933 * lru lock. This is safe to do, since every list movement is
934 * protected by the lru lock even if both locks are held.
936 * This is guaranteed by the fact that all LRU management
937 * functions are intermediated by the LRU API calls like
938 * list_lru_add and list_lru_del. List movement in this file
939 * only ever occur through this functions or through callbacks
940 * like this one, that are called from the LRU API.
942 * The only exceptions to this are functions like
943 * shrink_dentry_list, and code that first checks for the
944 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
945 * operating only with stack provided lists after they are
946 * properly isolated from the main list. It is thus, always a
952 d_lru_shrink_move(dentry, freeable);
953 spin_unlock(&dentry->d_lock);
959 * prune_dcache_sb - shrink the dcache
961 * @nr_to_scan : number of entries to try to free
962 * @nid: which node to scan for freeable entities
964 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
965 * done when we need more memory an called from the superblock shrinker
968 * This function may fail to free any resources if all the dentries are in
971 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
977 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
978 &dispose, &nr_to_scan);
979 shrink_dentry_list(&dispose);
983 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
984 spinlock_t *lru_lock, void *arg)
986 struct list_head *freeable = arg;
987 struct dentry *dentry = container_of(item, struct dentry, d_lru);
990 * we are inverting the lru lock/dentry->d_lock here,
991 * so use a trylock. If we fail to get the lock, just skip
994 if (!spin_trylock(&dentry->d_lock))
997 d_lru_shrink_move(dentry, freeable);
998 spin_unlock(&dentry->d_lock);
1005 * shrink_dcache_sb - shrink dcache for a superblock
1008 * Shrink the dcache for the specified super block. This is used to free
1009 * the dcache before unmounting a file system.
1011 void shrink_dcache_sb(struct super_block *sb)
1018 freed = list_lru_walk(&sb->s_dentry_lru,
1019 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1021 this_cpu_sub(nr_dentry_unused, freed);
1022 shrink_dentry_list(&dispose);
1023 } while (freed > 0);
1025 EXPORT_SYMBOL(shrink_dcache_sb);
1028 * enum d_walk_ret - action to talke during tree walk
1029 * @D_WALK_CONTINUE: contrinue walk
1030 * @D_WALK_QUIT: quit walk
1031 * @D_WALK_NORETRY: quit when retry is needed
1032 * @D_WALK_SKIP: skip this dentry and its children
1042 * d_walk - walk the dentry tree
1043 * @parent: start of walk
1044 * @data: data passed to @enter() and @finish()
1045 * @enter: callback when first entering the dentry
1046 * @finish: callback when successfully finished the walk
1048 * The @enter() and @finish() callbacks are called with d_lock held.
1050 static void d_walk(struct dentry *parent, void *data,
1051 enum d_walk_ret (*enter)(void *, struct dentry *),
1052 void (*finish)(void *))
1054 struct dentry *this_parent;
1055 struct list_head *next;
1057 enum d_walk_ret ret;
1061 read_seqbegin_or_lock(&rename_lock, &seq);
1062 this_parent = parent;
1063 spin_lock(&this_parent->d_lock);
1065 ret = enter(data, this_parent);
1067 case D_WALK_CONTINUE:
1072 case D_WALK_NORETRY:
1077 next = this_parent->d_subdirs.next;
1079 while (next != &this_parent->d_subdirs) {
1080 struct list_head *tmp = next;
1081 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1084 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1086 ret = enter(data, dentry);
1088 case D_WALK_CONTINUE:
1091 spin_unlock(&dentry->d_lock);
1093 case D_WALK_NORETRY:
1097 spin_unlock(&dentry->d_lock);
1101 if (!list_empty(&dentry->d_subdirs)) {
1102 spin_unlock(&this_parent->d_lock);
1103 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1104 this_parent = dentry;
1105 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1108 spin_unlock(&dentry->d_lock);
1111 * All done at this level ... ascend and resume the search.
1113 if (this_parent != parent) {
1114 struct dentry *child = this_parent;
1115 this_parent = child->d_parent;
1118 spin_unlock(&child->d_lock);
1119 spin_lock(&this_parent->d_lock);
1122 * might go back up the wrong parent if we have had a rename
1125 if (this_parent != child->d_parent ||
1126 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1127 need_seqretry(&rename_lock, seq)) {
1128 spin_unlock(&this_parent->d_lock);
1133 next = child->d_u.d_child.next;
1136 if (need_seqretry(&rename_lock, seq)) {
1137 spin_unlock(&this_parent->d_lock);
1144 spin_unlock(&this_parent->d_lock);
1145 done_seqretry(&rename_lock, seq);
1156 * Search for at least 1 mount point in the dentry's subdirs.
1157 * We descend to the next level whenever the d_subdirs
1158 * list is non-empty and continue searching.
1161 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1164 if (d_mountpoint(dentry)) {
1168 return D_WALK_CONTINUE;
1172 * have_submounts - check for mounts over a dentry
1173 * @parent: dentry to check.
1175 * Return true if the parent or its subdirectories contain
1178 int have_submounts(struct dentry *parent)
1182 d_walk(parent, &ret, check_mount, NULL);
1186 EXPORT_SYMBOL(have_submounts);
1189 * Called by mount code to set a mountpoint and check if the mountpoint is
1190 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1191 * subtree can become unreachable).
1193 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1194 * this reason take rename_lock and d_lock on dentry and ancestors.
1196 int d_set_mounted(struct dentry *dentry)
1200 write_seqlock(&rename_lock);
1201 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1202 /* Need exclusion wrt. check_submounts_and_drop() */
1203 spin_lock(&p->d_lock);
1204 if (unlikely(d_unhashed(p))) {
1205 spin_unlock(&p->d_lock);
1208 spin_unlock(&p->d_lock);
1210 spin_lock(&dentry->d_lock);
1211 if (!d_unlinked(dentry)) {
1212 dentry->d_flags |= DCACHE_MOUNTED;
1215 spin_unlock(&dentry->d_lock);
1217 write_sequnlock(&rename_lock);
1222 * Search the dentry child list of the specified parent,
1223 * and move any unused dentries to the end of the unused
1224 * list for prune_dcache(). We descend to the next level
1225 * whenever the d_subdirs list is non-empty and continue
1228 * It returns zero iff there are no unused children,
1229 * otherwise it returns the number of children moved to
1230 * the end of the unused list. This may not be the total
1231 * number of unused children, because select_parent can
1232 * drop the lock and return early due to latency
1236 struct select_data {
1237 struct dentry *start;
1238 struct list_head dispose;
1242 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1244 struct select_data *data = _data;
1245 enum d_walk_ret ret = D_WALK_CONTINUE;
1247 if (data->start == dentry)
1250 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1253 if (dentry->d_flags & DCACHE_LRU_LIST)
1255 if (!dentry->d_lockref.count) {
1256 d_shrink_add(dentry, &data->dispose);
1261 * We can return to the caller if we have found some (this
1262 * ensures forward progress). We'll be coming back to find
1265 if (!list_empty(&data->dispose))
1266 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1272 * shrink_dcache_parent - prune dcache
1273 * @parent: parent of entries to prune
1275 * Prune the dcache to remove unused children of the parent dentry.
1277 void shrink_dcache_parent(struct dentry *parent)
1280 struct select_data data;
1282 INIT_LIST_HEAD(&data.dispose);
1283 data.start = parent;
1286 d_walk(parent, &data, select_collect, NULL);
1290 shrink_dentry_list(&data.dispose);
1294 EXPORT_SYMBOL(shrink_dcache_parent);
1296 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1298 /* it has busy descendents; complain about those instead */
1299 if (!list_empty(&dentry->d_subdirs))
1300 return D_WALK_CONTINUE;
1302 /* root with refcount 1 is fine */
1303 if (dentry == _data && dentry->d_lockref.count == 1)
1304 return D_WALK_CONTINUE;
1306 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1307 " still in use (%d) [unmount of %s %s]\n",
1310 dentry->d_inode->i_ino : 0UL,
1312 dentry->d_lockref.count,
1313 dentry->d_sb->s_type->name,
1314 dentry->d_sb->s_id);
1316 return D_WALK_CONTINUE;
1319 static void do_one_tree(struct dentry *dentry)
1321 shrink_dcache_parent(dentry);
1322 d_walk(dentry, dentry, umount_check, NULL);
1328 * destroy the dentries attached to a superblock on unmounting
1330 void shrink_dcache_for_umount(struct super_block *sb)
1332 struct dentry *dentry;
1334 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1336 dentry = sb->s_root;
1338 do_one_tree(dentry);
1340 while (!hlist_bl_empty(&sb->s_anon)) {
1341 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1342 do_one_tree(dentry);
1346 struct detach_data {
1347 struct select_data select;
1348 struct dentry *mountpoint;
1350 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1352 struct detach_data *data = _data;
1354 if (d_mountpoint(dentry)) {
1355 __dget_dlock(dentry);
1356 data->mountpoint = dentry;
1360 return select_collect(&data->select, dentry);
1363 static void check_and_drop(void *_data)
1365 struct detach_data *data = _data;
1367 if (!data->mountpoint && !data->select.found)
1368 __d_drop(data->select.start);
1372 * check_submounts_and_drop - detach submounts, prune dcache, and drop
1374 * The final d_drop is done as an atomic operation relative to
1375 * rename_lock ensuring there are no races with d_set_mounted. This
1376 * ensures there are no unhashed dentries on the path to a mountpoint.
1378 * @dentry: dentry to detach, prune and drop
1380 int check_submounts_and_drop(struct dentry *dentry)
1384 /* Negative dentries can be dropped without further checks */
1385 if (!dentry->d_inode) {
1391 struct detach_data data;
1393 data.mountpoint = NULL;
1394 INIT_LIST_HEAD(&data.select.dispose);
1395 data.select.start = dentry;
1396 data.select.found = 0;
1398 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1400 if (data.select.found)
1401 shrink_dentry_list(&data.select.dispose);
1403 if (data.mountpoint) {
1404 detach_mounts(data.mountpoint);
1405 dput(data.mountpoint);
1408 if (!data.mountpoint && !data.select.found)
1417 EXPORT_SYMBOL(check_submounts_and_drop);
1420 * __d_alloc - allocate a dcache entry
1421 * @sb: filesystem it will belong to
1422 * @name: qstr of the name
1424 * Allocates a dentry. It returns %NULL if there is insufficient memory
1425 * available. On a success the dentry is returned. The name passed in is
1426 * copied and the copy passed in may be reused after this call.
1429 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1431 struct dentry *dentry;
1434 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1439 * We guarantee that the inline name is always NUL-terminated.
1440 * This way the memcpy() done by the name switching in rename
1441 * will still always have a NUL at the end, even if we might
1442 * be overwriting an internal NUL character
1444 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1445 if (name->len > DNAME_INLINE_LEN-1) {
1446 size_t size = offsetof(struct external_name, name[1]);
1447 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1449 kmem_cache_free(dentry_cache, dentry);
1452 atomic_set(&p->u.count, 1);
1455 dname = dentry->d_iname;
1458 dentry->d_name.len = name->len;
1459 dentry->d_name.hash = name->hash;
1460 memcpy(dname, name->name, name->len);
1461 dname[name->len] = 0;
1463 /* Make sure we always see the terminating NUL character */
1465 dentry->d_name.name = dname;
1467 dentry->d_lockref.count = 1;
1468 dentry->d_flags = 0;
1469 spin_lock_init(&dentry->d_lock);
1470 seqcount_init(&dentry->d_seq);
1471 dentry->d_inode = NULL;
1472 dentry->d_parent = dentry;
1474 dentry->d_op = NULL;
1475 dentry->d_fsdata = NULL;
1476 INIT_HLIST_BL_NODE(&dentry->d_hash);
1477 INIT_LIST_HEAD(&dentry->d_lru);
1478 INIT_LIST_HEAD(&dentry->d_subdirs);
1479 INIT_HLIST_NODE(&dentry->d_alias);
1480 INIT_LIST_HEAD(&dentry->d_u.d_child);
1481 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1483 this_cpu_inc(nr_dentry);
1489 * d_alloc - allocate a dcache entry
1490 * @parent: parent of entry to allocate
1491 * @name: qstr of the name
1493 * Allocates a dentry. It returns %NULL if there is insufficient memory
1494 * available. On a success the dentry is returned. The name passed in is
1495 * copied and the copy passed in may be reused after this call.
1497 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1499 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1503 spin_lock(&parent->d_lock);
1505 * don't need child lock because it is not subject
1506 * to concurrency here
1508 __dget_dlock(parent);
1509 dentry->d_parent = parent;
1510 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1511 spin_unlock(&parent->d_lock);
1515 EXPORT_SYMBOL(d_alloc);
1518 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1519 * @sb: the superblock
1520 * @name: qstr of the name
1522 * For a filesystem that just pins its dentries in memory and never
1523 * performs lookups at all, return an unhashed IS_ROOT dentry.
1525 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1527 return __d_alloc(sb, name);
1529 EXPORT_SYMBOL(d_alloc_pseudo);
1531 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1536 q.len = strlen(name);
1537 q.hash = full_name_hash(q.name, q.len);
1538 return d_alloc(parent, &q);
1540 EXPORT_SYMBOL(d_alloc_name);
1542 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1544 WARN_ON_ONCE(dentry->d_op);
1545 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1547 DCACHE_OP_REVALIDATE |
1548 DCACHE_OP_WEAK_REVALIDATE |
1549 DCACHE_OP_DELETE ));
1554 dentry->d_flags |= DCACHE_OP_HASH;
1556 dentry->d_flags |= DCACHE_OP_COMPARE;
1557 if (op->d_revalidate)
1558 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1559 if (op->d_weak_revalidate)
1560 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1562 dentry->d_flags |= DCACHE_OP_DELETE;
1564 dentry->d_flags |= DCACHE_OP_PRUNE;
1567 EXPORT_SYMBOL(d_set_d_op);
1569 static unsigned d_flags_for_inode(struct inode *inode)
1571 unsigned add_flags = DCACHE_FILE_TYPE;
1574 return DCACHE_MISS_TYPE;
1576 if (S_ISDIR(inode->i_mode)) {
1577 add_flags = DCACHE_DIRECTORY_TYPE;
1578 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1579 if (unlikely(!inode->i_op->lookup))
1580 add_flags = DCACHE_AUTODIR_TYPE;
1582 inode->i_opflags |= IOP_LOOKUP;
1584 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1585 if (unlikely(inode->i_op->follow_link))
1586 add_flags = DCACHE_SYMLINK_TYPE;
1588 inode->i_opflags |= IOP_NOFOLLOW;
1591 if (unlikely(IS_AUTOMOUNT(inode)))
1592 add_flags |= DCACHE_NEED_AUTOMOUNT;
1596 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1598 unsigned add_flags = d_flags_for_inode(inode);
1600 spin_lock(&dentry->d_lock);
1601 __d_set_type(dentry, add_flags);
1603 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1604 dentry->d_inode = inode;
1605 dentry_rcuwalk_barrier(dentry);
1606 spin_unlock(&dentry->d_lock);
1607 fsnotify_d_instantiate(dentry, inode);
1611 * d_instantiate - fill in inode information for a dentry
1612 * @entry: dentry to complete
1613 * @inode: inode to attach to this dentry
1615 * Fill in inode information in the entry.
1617 * This turns negative dentries into productive full members
1620 * NOTE! This assumes that the inode count has been incremented
1621 * (or otherwise set) by the caller to indicate that it is now
1622 * in use by the dcache.
1625 void d_instantiate(struct dentry *entry, struct inode * inode)
1627 BUG_ON(!hlist_unhashed(&entry->d_alias));
1629 spin_lock(&inode->i_lock);
1630 __d_instantiate(entry, inode);
1632 spin_unlock(&inode->i_lock);
1633 security_d_instantiate(entry, inode);
1635 EXPORT_SYMBOL(d_instantiate);
1638 * d_instantiate_unique - instantiate a non-aliased dentry
1639 * @entry: dentry to instantiate
1640 * @inode: inode to attach to this dentry
1642 * Fill in inode information in the entry. On success, it returns NULL.
1643 * If an unhashed alias of "entry" already exists, then we return the
1644 * aliased dentry instead and drop one reference to inode.
1646 * Note that in order to avoid conflicts with rename() etc, the caller
1647 * had better be holding the parent directory semaphore.
1649 * This also assumes that the inode count has been incremented
1650 * (or otherwise set) by the caller to indicate that it is now
1651 * in use by the dcache.
1653 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1654 struct inode *inode)
1656 struct dentry *alias;
1657 int len = entry->d_name.len;
1658 const char *name = entry->d_name.name;
1659 unsigned int hash = entry->d_name.hash;
1662 __d_instantiate(entry, NULL);
1666 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1668 * Don't need alias->d_lock here, because aliases with
1669 * d_parent == entry->d_parent are not subject to name or
1670 * parent changes, because the parent inode i_mutex is held.
1672 if (alias->d_name.hash != hash)
1674 if (alias->d_parent != entry->d_parent)
1676 if (alias->d_name.len != len)
1678 if (dentry_cmp(alias, name, len))
1684 __d_instantiate(entry, inode);
1688 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1690 struct dentry *result;
1692 BUG_ON(!hlist_unhashed(&entry->d_alias));
1695 spin_lock(&inode->i_lock);
1696 result = __d_instantiate_unique(entry, inode);
1698 spin_unlock(&inode->i_lock);
1701 security_d_instantiate(entry, inode);
1705 BUG_ON(!d_unhashed(result));
1710 EXPORT_SYMBOL(d_instantiate_unique);
1713 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1714 * @entry: dentry to complete
1715 * @inode: inode to attach to this dentry
1717 * Fill in inode information in the entry. If a directory alias is found, then
1718 * return an error (and drop inode). Together with d_materialise_unique() this
1719 * guarantees that a directory inode may never have more than one alias.
1721 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1723 BUG_ON(!hlist_unhashed(&entry->d_alias));
1725 spin_lock(&inode->i_lock);
1726 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1727 spin_unlock(&inode->i_lock);
1731 __d_instantiate(entry, inode);
1732 spin_unlock(&inode->i_lock);
1733 security_d_instantiate(entry, inode);
1737 EXPORT_SYMBOL(d_instantiate_no_diralias);
1739 struct dentry *d_make_root(struct inode *root_inode)
1741 struct dentry *res = NULL;
1744 static const struct qstr name = QSTR_INIT("/", 1);
1746 res = __d_alloc(root_inode->i_sb, &name);
1748 d_instantiate(res, root_inode);
1754 EXPORT_SYMBOL(d_make_root);
1756 static struct dentry * __d_find_any_alias(struct inode *inode)
1758 struct dentry *alias;
1760 if (hlist_empty(&inode->i_dentry))
1762 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1768 * d_find_any_alias - find any alias for a given inode
1769 * @inode: inode to find an alias for
1771 * If any aliases exist for the given inode, take and return a
1772 * reference for one of them. If no aliases exist, return %NULL.
1774 struct dentry *d_find_any_alias(struct inode *inode)
1778 spin_lock(&inode->i_lock);
1779 de = __d_find_any_alias(inode);
1780 spin_unlock(&inode->i_lock);
1783 EXPORT_SYMBOL(d_find_any_alias);
1785 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1787 static const struct qstr anonstring = QSTR_INIT("/", 1);
1793 return ERR_PTR(-ESTALE);
1795 return ERR_CAST(inode);
1797 res = d_find_any_alias(inode);
1801 tmp = __d_alloc(inode->i_sb, &anonstring);
1803 res = ERR_PTR(-ENOMEM);
1807 spin_lock(&inode->i_lock);
1808 res = __d_find_any_alias(inode);
1810 spin_unlock(&inode->i_lock);
1815 /* attach a disconnected dentry */
1816 add_flags = d_flags_for_inode(inode);
1819 add_flags |= DCACHE_DISCONNECTED;
1821 spin_lock(&tmp->d_lock);
1822 tmp->d_inode = inode;
1823 tmp->d_flags |= add_flags;
1824 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1825 hlist_bl_lock(&tmp->d_sb->s_anon);
1826 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1827 hlist_bl_unlock(&tmp->d_sb->s_anon);
1828 spin_unlock(&tmp->d_lock);
1829 spin_unlock(&inode->i_lock);
1830 security_d_instantiate(tmp, inode);
1835 if (res && !IS_ERR(res))
1836 security_d_instantiate(res, inode);
1842 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1843 * @inode: inode to allocate the dentry for
1845 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1846 * similar open by handle operations. The returned dentry may be anonymous,
1847 * or may have a full name (if the inode was already in the cache).
1849 * When called on a directory inode, we must ensure that the inode only ever
1850 * has one dentry. If a dentry is found, that is returned instead of
1851 * allocating a new one.
1853 * On successful return, the reference to the inode has been transferred
1854 * to the dentry. In case of an error the reference on the inode is released.
1855 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1856 * be passed in and the error will be propagated to the return value,
1857 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1859 struct dentry *d_obtain_alias(struct inode *inode)
1861 return __d_obtain_alias(inode, 1);
1863 EXPORT_SYMBOL(d_obtain_alias);
1866 * d_obtain_root - find or allocate a dentry for a given inode
1867 * @inode: inode to allocate the dentry for
1869 * Obtain an IS_ROOT dentry for the root of a filesystem.
1871 * We must ensure that directory inodes only ever have one dentry. If a
1872 * dentry is found, that is returned instead of allocating a new one.
1874 * On successful return, the reference to the inode has been transferred
1875 * to the dentry. In case of an error the reference on the inode is
1876 * released. A %NULL or IS_ERR inode may be passed in and will be the
1877 * error will be propagate to the return value, with a %NULL @inode
1878 * replaced by ERR_PTR(-ESTALE).
1880 struct dentry *d_obtain_root(struct inode *inode)
1882 return __d_obtain_alias(inode, 0);
1884 EXPORT_SYMBOL(d_obtain_root);
1887 * d_add_ci - lookup or allocate new dentry with case-exact name
1888 * @inode: the inode case-insensitive lookup has found
1889 * @dentry: the negative dentry that was passed to the parent's lookup func
1890 * @name: the case-exact name to be associated with the returned dentry
1892 * This is to avoid filling the dcache with case-insensitive names to the
1893 * same inode, only the actual correct case is stored in the dcache for
1894 * case-insensitive filesystems.
1896 * For a case-insensitive lookup match and if the the case-exact dentry
1897 * already exists in in the dcache, use it and return it.
1899 * If no entry exists with the exact case name, allocate new dentry with
1900 * the exact case, and return the spliced entry.
1902 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1905 struct dentry *found;
1909 * First check if a dentry matching the name already exists,
1910 * if not go ahead and create it now.
1912 found = d_hash_and_lookup(dentry->d_parent, name);
1913 if (unlikely(IS_ERR(found)))
1916 new = d_alloc(dentry->d_parent, name);
1918 found = ERR_PTR(-ENOMEM);
1922 found = d_splice_alias(inode, new);
1931 * If a matching dentry exists, and it's not negative use it.
1933 * Decrement the reference count to balance the iget() done
1936 if (found->d_inode) {
1937 if (unlikely(found->d_inode != inode)) {
1938 /* This can't happen because bad inodes are unhashed. */
1939 BUG_ON(!is_bad_inode(inode));
1940 BUG_ON(!is_bad_inode(found->d_inode));
1947 * Negative dentry: instantiate it unless the inode is a directory and
1948 * already has a dentry.
1950 new = d_splice_alias(inode, found);
1961 EXPORT_SYMBOL(d_add_ci);
1964 * Do the slow-case of the dentry name compare.
1966 * Unlike the dentry_cmp() function, we need to atomically
1967 * load the name and length information, so that the
1968 * filesystem can rely on them, and can use the 'name' and
1969 * 'len' information without worrying about walking off the
1970 * end of memory etc.
1972 * Thus the read_seqcount_retry() and the "duplicate" info
1973 * in arguments (the low-level filesystem should not look
1974 * at the dentry inode or name contents directly, since
1975 * rename can change them while we're in RCU mode).
1977 enum slow_d_compare {
1983 static noinline enum slow_d_compare slow_dentry_cmp(
1984 const struct dentry *parent,
1985 struct dentry *dentry,
1987 const struct qstr *name)
1989 int tlen = dentry->d_name.len;
1990 const char *tname = dentry->d_name.name;
1992 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1994 return D_COMP_SEQRETRY;
1996 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1997 return D_COMP_NOMATCH;
2002 * __d_lookup_rcu - search for a dentry (racy, store-free)
2003 * @parent: parent dentry
2004 * @name: qstr of name we wish to find
2005 * @seqp: returns d_seq value at the point where the dentry was found
2006 * Returns: dentry, or NULL
2008 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2009 * resolution (store-free path walking) design described in
2010 * Documentation/filesystems/path-lookup.txt.
2012 * This is not to be used outside core vfs.
2014 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2015 * held, and rcu_read_lock held. The returned dentry must not be stored into
2016 * without taking d_lock and checking d_seq sequence count against @seq
2019 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2022 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2023 * the returned dentry, so long as its parent's seqlock is checked after the
2024 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2025 * is formed, giving integrity down the path walk.
2027 * NOTE! The caller *has* to check the resulting dentry against the sequence
2028 * number we've returned before using any of the resulting dentry state!
2030 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2031 const struct qstr *name,
2034 u64 hashlen = name->hash_len;
2035 const unsigned char *str = name->name;
2036 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2037 struct hlist_bl_node *node;
2038 struct dentry *dentry;
2041 * Note: There is significant duplication with __d_lookup_rcu which is
2042 * required to prevent single threaded performance regressions
2043 * especially on architectures where smp_rmb (in seqcounts) are costly.
2044 * Keep the two functions in sync.
2048 * The hash list is protected using RCU.
2050 * Carefully use d_seq when comparing a candidate dentry, to avoid
2051 * races with d_move().
2053 * It is possible that concurrent renames can mess up our list
2054 * walk here and result in missing our dentry, resulting in the
2055 * false-negative result. d_lookup() protects against concurrent
2056 * renames using rename_lock seqlock.
2058 * See Documentation/filesystems/path-lookup.txt for more details.
2060 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2065 * The dentry sequence count protects us from concurrent
2066 * renames, and thus protects parent and name fields.
2068 * The caller must perform a seqcount check in order
2069 * to do anything useful with the returned dentry.
2071 * NOTE! We do a "raw" seqcount_begin here. That means that
2072 * we don't wait for the sequence count to stabilize if it
2073 * is in the middle of a sequence change. If we do the slow
2074 * dentry compare, we will do seqretries until it is stable,
2075 * and if we end up with a successful lookup, we actually
2076 * want to exit RCU lookup anyway.
2078 seq = raw_seqcount_begin(&dentry->d_seq);
2079 if (dentry->d_parent != parent)
2081 if (d_unhashed(dentry))
2084 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2085 if (dentry->d_name.hash != hashlen_hash(hashlen))
2088 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2091 case D_COMP_NOMATCH:
2098 if (dentry->d_name.hash_len != hashlen)
2101 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2108 * d_lookup - search for a dentry
2109 * @parent: parent dentry
2110 * @name: qstr of name we wish to find
2111 * Returns: dentry, or NULL
2113 * d_lookup searches the children of the parent dentry for the name in
2114 * question. If the dentry is found its reference count is incremented and the
2115 * dentry is returned. The caller must use dput to free the entry when it has
2116 * finished using it. %NULL is returned if the dentry does not exist.
2118 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2120 struct dentry *dentry;
2124 seq = read_seqbegin(&rename_lock);
2125 dentry = __d_lookup(parent, name);
2128 } while (read_seqretry(&rename_lock, seq));
2131 EXPORT_SYMBOL(d_lookup);
2134 * __d_lookup - search for a dentry (racy)
2135 * @parent: parent dentry
2136 * @name: qstr of name we wish to find
2137 * Returns: dentry, or NULL
2139 * __d_lookup is like d_lookup, however it may (rarely) return a
2140 * false-negative result due to unrelated rename activity.
2142 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2143 * however it must be used carefully, eg. with a following d_lookup in
2144 * the case of failure.
2146 * __d_lookup callers must be commented.
2148 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2150 unsigned int len = name->len;
2151 unsigned int hash = name->hash;
2152 const unsigned char *str = name->name;
2153 struct hlist_bl_head *b = d_hash(parent, hash);
2154 struct hlist_bl_node *node;
2155 struct dentry *found = NULL;
2156 struct dentry *dentry;
2159 * Note: There is significant duplication with __d_lookup_rcu which is
2160 * required to prevent single threaded performance regressions
2161 * especially on architectures where smp_rmb (in seqcounts) are costly.
2162 * Keep the two functions in sync.
2166 * The hash list is protected using RCU.
2168 * Take d_lock when comparing a candidate dentry, to avoid races
2171 * It is possible that concurrent renames can mess up our list
2172 * walk here and result in missing our dentry, resulting in the
2173 * false-negative result. d_lookup() protects against concurrent
2174 * renames using rename_lock seqlock.
2176 * See Documentation/filesystems/path-lookup.txt for more details.
2180 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2182 if (dentry->d_name.hash != hash)
2185 spin_lock(&dentry->d_lock);
2186 if (dentry->d_parent != parent)
2188 if (d_unhashed(dentry))
2192 * It is safe to compare names since d_move() cannot
2193 * change the qstr (protected by d_lock).
2195 if (parent->d_flags & DCACHE_OP_COMPARE) {
2196 int tlen = dentry->d_name.len;
2197 const char *tname = dentry->d_name.name;
2198 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2201 if (dentry->d_name.len != len)
2203 if (dentry_cmp(dentry, str, len))
2207 dentry->d_lockref.count++;
2209 spin_unlock(&dentry->d_lock);
2212 spin_unlock(&dentry->d_lock);
2220 * d_hash_and_lookup - hash the qstr then search for a dentry
2221 * @dir: Directory to search in
2222 * @name: qstr of name we wish to find
2224 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2226 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2229 * Check for a fs-specific hash function. Note that we must
2230 * calculate the standard hash first, as the d_op->d_hash()
2231 * routine may choose to leave the hash value unchanged.
2233 name->hash = full_name_hash(name->name, name->len);
2234 if (dir->d_flags & DCACHE_OP_HASH) {
2235 int err = dir->d_op->d_hash(dir, name);
2236 if (unlikely(err < 0))
2237 return ERR_PTR(err);
2239 return d_lookup(dir, name);
2241 EXPORT_SYMBOL(d_hash_and_lookup);
2244 * d_validate - verify dentry provided from insecure source (deprecated)
2245 * @dentry: The dentry alleged to be valid child of @dparent
2246 * @dparent: The parent dentry (known to be valid)
2248 * An insecure source has sent us a dentry, here we verify it and dget() it.
2249 * This is used by ncpfs in its readdir implementation.
2250 * Zero is returned in the dentry is invalid.
2252 * This function is slow for big directories, and deprecated, do not use it.
2254 int d_validate(struct dentry *dentry, struct dentry *dparent)
2256 struct dentry *child;
2258 spin_lock(&dparent->d_lock);
2259 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2260 if (dentry == child) {
2261 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2262 __dget_dlock(dentry);
2263 spin_unlock(&dentry->d_lock);
2264 spin_unlock(&dparent->d_lock);
2268 spin_unlock(&dparent->d_lock);
2272 EXPORT_SYMBOL(d_validate);
2275 * When a file is deleted, we have two options:
2276 * - turn this dentry into a negative dentry
2277 * - unhash this dentry and free it.
2279 * Usually, we want to just turn this into
2280 * a negative dentry, but if anybody else is
2281 * currently using the dentry or the inode
2282 * we can't do that and we fall back on removing
2283 * it from the hash queues and waiting for
2284 * it to be deleted later when it has no users
2288 * d_delete - delete a dentry
2289 * @dentry: The dentry to delete
2291 * Turn the dentry into a negative dentry if possible, otherwise
2292 * remove it from the hash queues so it can be deleted later
2295 void d_delete(struct dentry * dentry)
2297 struct inode *inode;
2300 * Are we the only user?
2303 spin_lock(&dentry->d_lock);
2304 inode = dentry->d_inode;
2305 isdir = S_ISDIR(inode->i_mode);
2306 if (dentry->d_lockref.count == 1) {
2307 if (!spin_trylock(&inode->i_lock)) {
2308 spin_unlock(&dentry->d_lock);
2312 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2313 dentry_unlink_inode(dentry);
2314 fsnotify_nameremove(dentry, isdir);
2318 if (!d_unhashed(dentry))
2321 spin_unlock(&dentry->d_lock);
2323 fsnotify_nameremove(dentry, isdir);
2325 EXPORT_SYMBOL(d_delete);
2327 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2329 BUG_ON(!d_unhashed(entry));
2331 entry->d_flags |= DCACHE_RCUACCESS;
2332 hlist_bl_add_head_rcu(&entry->d_hash, b);
2336 static void _d_rehash(struct dentry * entry)
2338 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2342 * d_rehash - add an entry back to the hash
2343 * @entry: dentry to add to the hash
2345 * Adds a dentry to the hash according to its name.
2348 void d_rehash(struct dentry * entry)
2350 spin_lock(&entry->d_lock);
2352 spin_unlock(&entry->d_lock);
2354 EXPORT_SYMBOL(d_rehash);
2357 * dentry_update_name_case - update case insensitive dentry with a new name
2358 * @dentry: dentry to be updated
2361 * Update a case insensitive dentry with new case of name.
2363 * dentry must have been returned by d_lookup with name @name. Old and new
2364 * name lengths must match (ie. no d_compare which allows mismatched name
2367 * Parent inode i_mutex must be held over d_lookup and into this call (to
2368 * keep renames and concurrent inserts, and readdir(2) away).
2370 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2372 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2373 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2375 spin_lock(&dentry->d_lock);
2376 write_seqcount_begin(&dentry->d_seq);
2377 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2378 write_seqcount_end(&dentry->d_seq);
2379 spin_unlock(&dentry->d_lock);
2381 EXPORT_SYMBOL(dentry_update_name_case);
2383 static void swap_names(struct dentry *dentry, struct dentry *target)
2385 if (unlikely(dname_external(target))) {
2386 if (unlikely(dname_external(dentry))) {
2388 * Both external: swap the pointers
2390 swap(target->d_name.name, dentry->d_name.name);
2393 * dentry:internal, target:external. Steal target's
2394 * storage and make target internal.
2396 memcpy(target->d_iname, dentry->d_name.name,
2397 dentry->d_name.len + 1);
2398 dentry->d_name.name = target->d_name.name;
2399 target->d_name.name = target->d_iname;
2402 if (unlikely(dname_external(dentry))) {
2404 * dentry:external, target:internal. Give dentry's
2405 * storage to target and make dentry internal
2407 memcpy(dentry->d_iname, target->d_name.name,
2408 target->d_name.len + 1);
2409 target->d_name.name = dentry->d_name.name;
2410 dentry->d_name.name = dentry->d_iname;
2413 * Both are internal.
2416 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2417 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2418 swap(((long *) &dentry->d_iname)[i],
2419 ((long *) &target->d_iname)[i]);
2423 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2426 static void copy_name(struct dentry *dentry, struct dentry *target)
2428 struct external_name *old_name = NULL;
2429 if (unlikely(dname_external(dentry)))
2430 old_name = external_name(dentry);
2431 if (unlikely(dname_external(target))) {
2432 atomic_inc(&external_name(target)->u.count);
2433 dentry->d_name = target->d_name;
2435 memcpy(dentry->d_iname, target->d_name.name,
2436 target->d_name.len + 1);
2437 dentry->d_name.name = dentry->d_iname;
2438 dentry->d_name.hash_len = target->d_name.hash_len;
2440 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2441 kfree_rcu(old_name, u.head);
2444 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2447 * XXXX: do we really need to take target->d_lock?
2449 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2450 spin_lock(&target->d_parent->d_lock);
2452 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2453 spin_lock(&dentry->d_parent->d_lock);
2454 spin_lock_nested(&target->d_parent->d_lock,
2455 DENTRY_D_LOCK_NESTED);
2457 spin_lock(&target->d_parent->d_lock);
2458 spin_lock_nested(&dentry->d_parent->d_lock,
2459 DENTRY_D_LOCK_NESTED);
2462 if (target < dentry) {
2463 spin_lock_nested(&target->d_lock, 2);
2464 spin_lock_nested(&dentry->d_lock, 3);
2466 spin_lock_nested(&dentry->d_lock, 2);
2467 spin_lock_nested(&target->d_lock, 3);
2471 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2473 if (target->d_parent != dentry->d_parent)
2474 spin_unlock(&dentry->d_parent->d_lock);
2475 if (target->d_parent != target)
2476 spin_unlock(&target->d_parent->d_lock);
2477 spin_unlock(&target->d_lock);
2478 spin_unlock(&dentry->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, unless we are going to rehash
2486 * it. Note that if we *do* unhash the target, we are not allowed
2487 * to rehash it without giving it a new name/hash key - whether
2488 * we swap or overwrite the names here, resulting name won't match
2489 * the reality in filesystem; it's only there for d_path() purposes.
2490 * Note that all of this is happening under rename_lock, so the
2491 * any hash lookup seeing it in the middle of manipulations will
2492 * be discarded anyway. So we do not care what happens to the hash
2496 * __d_move - move a dentry
2497 * @dentry: entry to move
2498 * @target: new dentry
2499 * @exchange: exchange the two dentries
2501 * Update the dcache to reflect the move of a file name. Negative
2502 * dcache entries should not be moved in this way. Caller must hold
2503 * rename_lock, the i_mutex of the source and target directories,
2504 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2506 static void __d_move(struct dentry *dentry, struct dentry *target,
2509 if (!dentry->d_inode)
2510 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2512 BUG_ON(d_ancestor(dentry, target));
2513 BUG_ON(d_ancestor(target, dentry));
2515 dentry_lock_for_move(dentry, target);
2517 write_seqcount_begin(&dentry->d_seq);
2518 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2520 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2523 * Move the dentry to the target hash queue. Don't bother checking
2524 * for the same hash queue because of how unlikely it is.
2527 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2530 * Unhash the target (d_delete() is not usable here). If exchanging
2531 * the two dentries, then rehash onto the other's hash queue.
2536 d_hash(dentry->d_parent, dentry->d_name.hash));
2539 /* Switch the names.. */
2541 swap_names(dentry, target);
2543 copy_name(dentry, target);
2545 /* ... and switch them in the tree */
2546 if (IS_ROOT(dentry)) {
2547 /* splicing a tree */
2548 dentry->d_parent = target->d_parent;
2549 target->d_parent = target;
2550 list_del_init(&target->d_u.d_child);
2551 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2553 /* swapping two dentries */
2554 swap(dentry->d_parent, target->d_parent);
2555 list_move(&target->d_u.d_child, &target->d_parent->d_subdirs);
2556 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2558 fsnotify_d_move(target);
2559 fsnotify_d_move(dentry);
2562 write_seqcount_end(&target->d_seq);
2563 write_seqcount_end(&dentry->d_seq);
2565 dentry_unlock_for_move(dentry, target);
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 __d_move(alias, dentry, false);
2653 spin_unlock(&inode->i_lock);
2662 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2663 * @inode: the inode which may have a disconnected dentry
2664 * @dentry: a negative dentry which we want to point to the inode.
2666 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2667 * place of the given dentry and return it, else simply d_add the inode
2668 * to the dentry and return NULL.
2670 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2671 * we should error out: directories can't have multiple aliases.
2673 * This is needed in the lookup routine of any filesystem that is exportable
2674 * (via knfsd) so that we can build dcache paths to directories effectively.
2676 * If a dentry was found and moved, then it is returned. Otherwise NULL
2677 * is returned. This matches the expected return value of ->lookup.
2679 * Cluster filesystems may call this function with a negative, hashed dentry.
2680 * In that case, we know that the inode will be a regular file, and also this
2681 * will only occur during atomic_open. So we need to check for the dentry
2682 * being already hashed only in the final case.
2684 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2686 struct dentry *new = NULL;
2689 return ERR_CAST(inode);
2691 if (inode && S_ISDIR(inode->i_mode)) {
2692 spin_lock(&inode->i_lock);
2693 new = __d_find_any_alias(inode);
2695 if (!IS_ROOT(new)) {
2696 spin_unlock(&inode->i_lock);
2698 return ERR_PTR(-EIO);
2700 if (d_ancestor(new, dentry)) {
2701 spin_unlock(&inode->i_lock);
2703 return ERR_PTR(-EIO);
2705 write_seqlock(&rename_lock);
2706 __d_move(new, dentry, false);
2707 write_sequnlock(&rename_lock);
2708 spin_unlock(&inode->i_lock);
2709 security_d_instantiate(new, inode);
2712 /* already taking inode->i_lock, so d_add() by hand */
2713 __d_instantiate(dentry, inode);
2714 spin_unlock(&inode->i_lock);
2715 security_d_instantiate(dentry, inode);
2719 d_instantiate(dentry, inode);
2720 if (d_unhashed(dentry))
2725 EXPORT_SYMBOL(d_splice_alias);
2728 * d_materialise_unique - introduce an inode into the tree
2729 * @dentry: candidate dentry
2730 * @inode: inode to bind to the dentry, to which aliases may be attached
2732 * Introduces an dentry into the tree, substituting an extant disconnected
2733 * root directory alias in its place if there is one. Caller must hold the
2734 * i_mutex of the parent directory.
2736 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2738 struct dentry *actual;
2740 BUG_ON(!d_unhashed(dentry));
2744 __d_instantiate(dentry, NULL);
2749 spin_lock(&inode->i_lock);
2751 if (S_ISDIR(inode->i_mode)) {
2752 struct dentry *alias;
2754 /* Does an aliased dentry already exist? */
2755 alias = __d_find_alias(inode);
2758 write_seqlock(&rename_lock);
2760 if (d_ancestor(alias, dentry)) {
2761 /* Check for loops */
2762 actual = ERR_PTR(-ELOOP);
2763 spin_unlock(&inode->i_lock);
2764 } else if (IS_ROOT(alias)) {
2765 /* Is this an anonymous mountpoint that we
2766 * could splice into our tree? */
2767 __d_move(alias, dentry, false);
2768 write_sequnlock(&rename_lock);
2771 /* Nope, but we must(!) avoid directory
2772 * aliasing. This drops inode->i_lock */
2773 actual = __d_unalias(inode, dentry, alias);
2775 write_sequnlock(&rename_lock);
2776 if (IS_ERR(actual)) {
2777 if (PTR_ERR(actual) == -ELOOP)
2778 pr_warn_ratelimited(
2779 "VFS: Lookup of '%s' in %s %s"
2780 " would have caused loop\n",
2781 dentry->d_name.name,
2782 inode->i_sb->s_type->name,
2790 /* Add a unique reference */
2791 actual = __d_instantiate_unique(dentry, inode);
2797 spin_unlock(&inode->i_lock);
2799 if (actual == dentry) {
2800 security_d_instantiate(dentry, inode);
2807 EXPORT_SYMBOL_GPL(d_materialise_unique);
2809 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2813 return -ENAMETOOLONG;
2815 memcpy(*buffer, str, namelen);
2820 * prepend_name - prepend a pathname in front of current buffer pointer
2821 * @buffer: buffer pointer
2822 * @buflen: allocated length of the buffer
2823 * @name: name string and length qstr structure
2825 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2826 * make sure that either the old or the new name pointer and length are
2827 * fetched. However, there may be mismatch between length and pointer.
2828 * The length cannot be trusted, we need to copy it byte-by-byte until
2829 * the length is reached or a null byte is found. It also prepends "/" at
2830 * the beginning of the name. The sequence number check at the caller will
2831 * retry it again when a d_move() does happen. So any garbage in the buffer
2832 * due to mismatched pointer and length will be discarded.
2834 * Data dependency barrier is needed to make sure that we see that terminating
2835 * NUL. Alpha strikes again, film at 11...
2837 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2839 const char *dname = ACCESS_ONCE(name->name);
2840 u32 dlen = ACCESS_ONCE(name->len);
2843 smp_read_barrier_depends();
2845 *buflen -= dlen + 1;
2847 return -ENAMETOOLONG;
2848 p = *buffer -= dlen + 1;
2860 * prepend_path - Prepend path string to a buffer
2861 * @path: the dentry/vfsmount to report
2862 * @root: root vfsmnt/dentry
2863 * @buffer: pointer to the end of the buffer
2864 * @buflen: pointer to buffer length
2866 * The function will first try to write out the pathname without taking any
2867 * lock other than the RCU read lock to make sure that dentries won't go away.
2868 * It only checks the sequence number of the global rename_lock as any change
2869 * in the dentry's d_seq will be preceded by changes in the rename_lock
2870 * sequence number. If the sequence number had been changed, it will restart
2871 * the whole pathname back-tracing sequence again by taking the rename_lock.
2872 * In this case, there is no need to take the RCU read lock as the recursive
2873 * parent pointer references will keep the dentry chain alive as long as no
2874 * rename operation is performed.
2876 static int prepend_path(const struct path *path,
2877 const struct path *root,
2878 char **buffer, int *buflen)
2880 struct dentry *dentry;
2881 struct vfsmount *vfsmnt;
2884 unsigned seq, m_seq = 0;
2890 read_seqbegin_or_lock(&mount_lock, &m_seq);
2897 dentry = path->dentry;
2899 mnt = real_mount(vfsmnt);
2900 read_seqbegin_or_lock(&rename_lock, &seq);
2901 while (dentry != root->dentry || vfsmnt != root->mnt) {
2902 struct dentry * parent;
2904 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2905 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2907 if (mnt != parent) {
2908 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2914 * Filesystems needing to implement special "root names"
2915 * should do so with ->d_dname()
2917 if (IS_ROOT(dentry) &&
2918 (dentry->d_name.len != 1 ||
2919 dentry->d_name.name[0] != '/')) {
2920 WARN(1, "Root dentry has weird name <%.*s>\n",
2921 (int) dentry->d_name.len,
2922 dentry->d_name.name);
2925 error = is_mounted(vfsmnt) ? 1 : 2;
2928 parent = dentry->d_parent;
2930 error = prepend_name(&bptr, &blen, &dentry->d_name);
2938 if (need_seqretry(&rename_lock, seq)) {
2942 done_seqretry(&rename_lock, seq);
2946 if (need_seqretry(&mount_lock, m_seq)) {
2950 done_seqretry(&mount_lock, m_seq);
2952 if (error >= 0 && bptr == *buffer) {
2954 error = -ENAMETOOLONG;
2964 * __d_path - return the path of a dentry
2965 * @path: the dentry/vfsmount to report
2966 * @root: root vfsmnt/dentry
2967 * @buf: buffer to return value in
2968 * @buflen: buffer length
2970 * Convert a dentry into an ASCII path name.
2972 * Returns a pointer into the buffer or an error code if the
2973 * path was too long.
2975 * "buflen" should be positive.
2977 * If the path is not reachable from the supplied root, return %NULL.
2979 char *__d_path(const struct path *path,
2980 const struct path *root,
2981 char *buf, int buflen)
2983 char *res = buf + buflen;
2986 prepend(&res, &buflen, "\0", 1);
2987 error = prepend_path(path, root, &res, &buflen);
2990 return ERR_PTR(error);
2996 char *d_absolute_path(const struct path *path,
2997 char *buf, int buflen)
2999 struct path root = {};
3000 char *res = buf + buflen;
3003 prepend(&res, &buflen, "\0", 1);
3004 error = prepend_path(path, &root, &res, &buflen);
3009 return ERR_PTR(error);
3014 * same as __d_path but appends "(deleted)" for unlinked files.
3016 static int path_with_deleted(const struct path *path,
3017 const struct path *root,
3018 char **buf, int *buflen)
3020 prepend(buf, buflen, "\0", 1);
3021 if (d_unlinked(path->dentry)) {
3022 int error = prepend(buf, buflen, " (deleted)", 10);
3027 return prepend_path(path, root, buf, buflen);
3030 static int prepend_unreachable(char **buffer, int *buflen)
3032 return prepend(buffer, buflen, "(unreachable)", 13);
3035 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3040 seq = read_seqcount_begin(&fs->seq);
3042 } while (read_seqcount_retry(&fs->seq, seq));
3046 * d_path - return the path of a dentry
3047 * @path: path to report
3048 * @buf: buffer to return value in
3049 * @buflen: buffer length
3051 * Convert a dentry into an ASCII path name. If the entry has been deleted
3052 * the string " (deleted)" is appended. Note that this is ambiguous.
3054 * Returns a pointer into the buffer or an error code if the path was
3055 * too long. Note: Callers should use the returned pointer, not the passed
3056 * in buffer, to use the name! The implementation often starts at an offset
3057 * into the buffer, and may leave 0 bytes at the start.
3059 * "buflen" should be positive.
3061 char *d_path(const struct path *path, char *buf, int buflen)
3063 char *res = buf + buflen;
3068 * We have various synthetic filesystems that never get mounted. On
3069 * these filesystems dentries are never used for lookup purposes, and
3070 * thus don't need to be hashed. They also don't need a name until a
3071 * user wants to identify the object in /proc/pid/fd/. The little hack
3072 * below allows us to generate a name for these objects on demand:
3074 * Some pseudo inodes are mountable. When they are mounted
3075 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3076 * and instead have d_path return the mounted path.
3078 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3079 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3080 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3083 get_fs_root_rcu(current->fs, &root);
3084 error = path_with_deleted(path, &root, &res, &buflen);
3088 res = ERR_PTR(error);
3091 EXPORT_SYMBOL(d_path);
3094 * Helper function for dentry_operations.d_dname() members
3096 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3097 const char *fmt, ...)
3103 va_start(args, fmt);
3104 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3107 if (sz > sizeof(temp) || sz > buflen)
3108 return ERR_PTR(-ENAMETOOLONG);
3110 buffer += buflen - sz;
3111 return memcpy(buffer, temp, sz);
3114 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3116 char *end = buffer + buflen;
3117 /* these dentries are never renamed, so d_lock is not needed */
3118 if (prepend(&end, &buflen, " (deleted)", 11) ||
3119 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3120 prepend(&end, &buflen, "/", 1))
3121 end = ERR_PTR(-ENAMETOOLONG);
3124 EXPORT_SYMBOL(simple_dname);
3127 * Write full pathname from the root of the filesystem into the buffer.
3129 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3131 struct dentry *dentry;
3144 prepend(&end, &len, "\0", 1);
3148 read_seqbegin_or_lock(&rename_lock, &seq);
3149 while (!IS_ROOT(dentry)) {
3150 struct dentry *parent = dentry->d_parent;
3153 error = prepend_name(&end, &len, &dentry->d_name);
3162 if (need_seqretry(&rename_lock, seq)) {
3166 done_seqretry(&rename_lock, seq);
3171 return ERR_PTR(-ENAMETOOLONG);
3174 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3176 return __dentry_path(dentry, buf, buflen);
3178 EXPORT_SYMBOL(dentry_path_raw);
3180 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3185 if (d_unlinked(dentry)) {
3187 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3191 retval = __dentry_path(dentry, buf, buflen);
3192 if (!IS_ERR(retval) && p)
3193 *p = '/'; /* restore '/' overriden with '\0' */
3196 return ERR_PTR(-ENAMETOOLONG);
3199 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3205 seq = read_seqcount_begin(&fs->seq);
3208 } while (read_seqcount_retry(&fs->seq, seq));
3212 * NOTE! The user-level library version returns a
3213 * character pointer. The kernel system call just
3214 * returns the length of the buffer filled (which
3215 * includes the ending '\0' character), or a negative
3216 * error value. So libc would do something like
3218 * char *getcwd(char * buf, size_t size)
3222 * retval = sys_getcwd(buf, size);
3229 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3232 struct path pwd, root;
3233 char *page = __getname();
3239 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3242 if (!d_unlinked(pwd.dentry)) {
3244 char *cwd = page + PATH_MAX;
3245 int buflen = PATH_MAX;
3247 prepend(&cwd, &buflen, "\0", 1);
3248 error = prepend_path(&pwd, &root, &cwd, &buflen);
3254 /* Unreachable from current root */
3256 error = prepend_unreachable(&cwd, &buflen);
3262 len = PATH_MAX + page - cwd;
3265 if (copy_to_user(buf, cwd, len))
3278 * Test whether new_dentry is a subdirectory of old_dentry.
3280 * Trivially implemented using the dcache structure
3284 * is_subdir - is new dentry a subdirectory of old_dentry
3285 * @new_dentry: new dentry
3286 * @old_dentry: old dentry
3288 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3289 * Returns 0 otherwise.
3290 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3293 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3298 if (new_dentry == old_dentry)
3302 /* for restarting inner loop in case of seq retry */
3303 seq = read_seqbegin(&rename_lock);
3305 * Need rcu_readlock to protect against the d_parent trashing
3309 if (d_ancestor(old_dentry, new_dentry))
3314 } while (read_seqretry(&rename_lock, seq));
3319 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3321 struct dentry *root = data;
3322 if (dentry != root) {
3323 if (d_unhashed(dentry) || !dentry->d_inode)
3326 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3327 dentry->d_flags |= DCACHE_GENOCIDE;
3328 dentry->d_lockref.count--;
3331 return D_WALK_CONTINUE;
3334 void d_genocide(struct dentry *parent)
3336 d_walk(parent, parent, d_genocide_kill, NULL);
3339 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3341 inode_dec_link_count(inode);
3342 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3343 !hlist_unhashed(&dentry->d_alias) ||
3344 !d_unlinked(dentry));
3345 spin_lock(&dentry->d_parent->d_lock);
3346 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3347 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3348 (unsigned long long)inode->i_ino);
3349 spin_unlock(&dentry->d_lock);
3350 spin_unlock(&dentry->d_parent->d_lock);
3351 d_instantiate(dentry, inode);
3353 EXPORT_SYMBOL(d_tmpfile);
3355 static __initdata unsigned long dhash_entries;
3356 static int __init set_dhash_entries(char *str)
3360 dhash_entries = simple_strtoul(str, &str, 0);
3363 __setup("dhash_entries=", set_dhash_entries);
3365 static void __init dcache_init_early(void)
3369 /* If hashes are distributed across NUMA nodes, defer
3370 * hash allocation until vmalloc space is available.
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 static void __init dcache_init(void)
3395 * A constructor could be added for stable state like the lists,
3396 * but it is probably not worth it because of the cache nature
3399 dentry_cache = KMEM_CACHE(dentry,
3400 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3402 /* Hash may have been set up in dcache_init_early */
3407 alloc_large_system_hash("Dentry cache",
3408 sizeof(struct hlist_bl_head),
3417 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3418 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3421 /* SLAB cache for __getname() consumers */
3422 struct kmem_cache *names_cachep __read_mostly;
3423 EXPORT_SYMBOL(names_cachep);
3425 EXPORT_SYMBOL(d_genocide);
3427 void __init vfs_caches_init_early(void)
3429 dcache_init_early();
3433 void __init vfs_caches_init(unsigned long mempages)
3435 unsigned long reserve;
3437 /* Base hash sizes on available memory, with a reserve equal to
3438 150% of current kernel size */
3440 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3441 mempages -= reserve;
3443 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3444 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3448 files_init(mempages);