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 other dentries that can be
654 * reached through this one we can't delete it and we
655 * return -EBUSY. On success we return 0.
660 int d_invalidate(struct dentry * dentry)
663 * If it's already been dropped, return OK.
665 spin_lock(&dentry->d_lock);
666 if (d_unhashed(dentry)) {
667 spin_unlock(&dentry->d_lock);
671 * Check whether to do a partial shrink_dcache
672 * to get rid of unused child entries.
674 if (!list_empty(&dentry->d_subdirs)) {
675 spin_unlock(&dentry->d_lock);
676 shrink_dcache_parent(dentry);
677 spin_lock(&dentry->d_lock);
681 * Somebody else still using it?
683 * If it's a directory, we can't drop it
684 * for fear of somebody re-populating it
685 * with children (even though dropping it
686 * would make it unreachable from the root,
687 * we might still populate it if it was a
688 * working directory or similar).
689 * We also need to leave mountpoints alone,
692 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
693 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
694 spin_unlock(&dentry->d_lock);
700 spin_unlock(&dentry->d_lock);
703 EXPORT_SYMBOL(d_invalidate);
705 /* This must be called with d_lock held */
706 static inline void __dget_dlock(struct dentry *dentry)
708 dentry->d_lockref.count++;
711 static inline void __dget(struct dentry *dentry)
713 lockref_get(&dentry->d_lockref);
716 struct dentry *dget_parent(struct dentry *dentry)
722 * Do optimistic parent lookup without any
726 ret = ACCESS_ONCE(dentry->d_parent);
727 gotref = lockref_get_not_zero(&ret->d_lockref);
729 if (likely(gotref)) {
730 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
737 * Don't need rcu_dereference because we re-check it was correct under
741 ret = dentry->d_parent;
742 spin_lock(&ret->d_lock);
743 if (unlikely(ret != dentry->d_parent)) {
744 spin_unlock(&ret->d_lock);
749 BUG_ON(!ret->d_lockref.count);
750 ret->d_lockref.count++;
751 spin_unlock(&ret->d_lock);
754 EXPORT_SYMBOL(dget_parent);
757 * d_find_alias - grab a hashed alias of inode
758 * @inode: inode in question
760 * If inode has a hashed alias, or is a directory and has any alias,
761 * acquire the reference to alias and return it. Otherwise return NULL.
762 * Notice that if inode is a directory there can be only one alias and
763 * it can be unhashed only if it has no children, or if it is the root
766 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
767 * any other hashed alias over that one.
769 static struct dentry *__d_find_alias(struct inode *inode)
771 struct dentry *alias, *discon_alias;
775 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
776 spin_lock(&alias->d_lock);
777 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
778 if (IS_ROOT(alias) &&
779 (alias->d_flags & DCACHE_DISCONNECTED)) {
780 discon_alias = alias;
783 spin_unlock(&alias->d_lock);
787 spin_unlock(&alias->d_lock);
790 alias = discon_alias;
791 spin_lock(&alias->d_lock);
792 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
794 spin_unlock(&alias->d_lock);
797 spin_unlock(&alias->d_lock);
803 struct dentry *d_find_alias(struct inode *inode)
805 struct dentry *de = NULL;
807 if (!hlist_empty(&inode->i_dentry)) {
808 spin_lock(&inode->i_lock);
809 de = __d_find_alias(inode);
810 spin_unlock(&inode->i_lock);
814 EXPORT_SYMBOL(d_find_alias);
817 * Try to kill dentries associated with this inode.
818 * WARNING: you must own a reference to inode.
820 void d_prune_aliases(struct inode *inode)
822 struct dentry *dentry;
824 spin_lock(&inode->i_lock);
825 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
826 spin_lock(&dentry->d_lock);
827 if (!dentry->d_lockref.count) {
829 * inform the fs via d_prune that this dentry
830 * is about to be unhashed and destroyed.
832 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
834 dentry->d_op->d_prune(dentry);
836 __dget_dlock(dentry);
838 spin_unlock(&dentry->d_lock);
839 spin_unlock(&inode->i_lock);
843 spin_unlock(&dentry->d_lock);
845 spin_unlock(&inode->i_lock);
847 EXPORT_SYMBOL(d_prune_aliases);
849 static void shrink_dentry_list(struct list_head *list)
851 struct dentry *dentry, *parent;
853 while (!list_empty(list)) {
855 dentry = list_entry(list->prev, struct dentry, d_lru);
856 spin_lock(&dentry->d_lock);
857 parent = lock_parent(dentry);
860 * The dispose list is isolated and dentries are not accounted
861 * to the LRU here, so we can simply remove it from the list
862 * here regardless of whether it is referenced or not.
864 d_shrink_del(dentry);
867 * We found an inuse dentry which was not removed from
868 * the LRU because of laziness during lookup. Do not free it.
870 if ((int)dentry->d_lockref.count > 0) {
871 spin_unlock(&dentry->d_lock);
873 spin_unlock(&parent->d_lock);
878 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
879 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
880 spin_unlock(&dentry->d_lock);
882 spin_unlock(&parent->d_lock);
888 inode = dentry->d_inode;
889 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
890 d_shrink_add(dentry, list);
891 spin_unlock(&dentry->d_lock);
893 spin_unlock(&parent->d_lock);
897 __dentry_kill(dentry);
900 * We need to prune ancestors too. This is necessary to prevent
901 * quadratic behavior of shrink_dcache_parent(), but is also
902 * expected to be beneficial in reducing dentry cache
906 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
907 parent = lock_parent(dentry);
908 if (dentry->d_lockref.count != 1) {
909 dentry->d_lockref.count--;
910 spin_unlock(&dentry->d_lock);
912 spin_unlock(&parent->d_lock);
915 inode = dentry->d_inode; /* can't be NULL */
916 if (unlikely(!spin_trylock(&inode->i_lock))) {
917 spin_unlock(&dentry->d_lock);
919 spin_unlock(&parent->d_lock);
923 __dentry_kill(dentry);
929 static enum lru_status
930 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
932 struct list_head *freeable = arg;
933 struct dentry *dentry = container_of(item, struct dentry, d_lru);
937 * we are inverting the lru lock/dentry->d_lock here,
938 * so use a trylock. If we fail to get the lock, just skip
941 if (!spin_trylock(&dentry->d_lock))
945 * Referenced dentries are still in use. If they have active
946 * counts, just remove them from the LRU. Otherwise give them
947 * another pass through the LRU.
949 if (dentry->d_lockref.count) {
950 d_lru_isolate(dentry);
951 spin_unlock(&dentry->d_lock);
955 if (dentry->d_flags & DCACHE_REFERENCED) {
956 dentry->d_flags &= ~DCACHE_REFERENCED;
957 spin_unlock(&dentry->d_lock);
960 * The list move itself will be made by the common LRU code. At
961 * this point, we've dropped the dentry->d_lock but keep the
962 * lru lock. This is safe to do, since every list movement is
963 * protected by the lru lock even if both locks are held.
965 * This is guaranteed by the fact that all LRU management
966 * functions are intermediated by the LRU API calls like
967 * list_lru_add and list_lru_del. List movement in this file
968 * only ever occur through this functions or through callbacks
969 * like this one, that are called from the LRU API.
971 * The only exceptions to this are functions like
972 * shrink_dentry_list, and code that first checks for the
973 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
974 * operating only with stack provided lists after they are
975 * properly isolated from the main list. It is thus, always a
981 d_lru_shrink_move(dentry, freeable);
982 spin_unlock(&dentry->d_lock);
988 * prune_dcache_sb - shrink the dcache
990 * @nr_to_scan : number of entries to try to free
991 * @nid: which node to scan for freeable entities
993 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
994 * done when we need more memory an called from the superblock shrinker
997 * This function may fail to free any resources if all the dentries are in
1000 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
1006 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
1007 &dispose, &nr_to_scan);
1008 shrink_dentry_list(&dispose);
1012 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1013 spinlock_t *lru_lock, void *arg)
1015 struct list_head *freeable = arg;
1016 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1019 * we are inverting the lru lock/dentry->d_lock here,
1020 * so use a trylock. If we fail to get the lock, just skip
1023 if (!spin_trylock(&dentry->d_lock))
1026 d_lru_shrink_move(dentry, freeable);
1027 spin_unlock(&dentry->d_lock);
1034 * shrink_dcache_sb - shrink dcache for a superblock
1037 * Shrink the dcache for the specified super block. This is used to free
1038 * the dcache before unmounting a file system.
1040 void shrink_dcache_sb(struct super_block *sb)
1047 freed = list_lru_walk(&sb->s_dentry_lru,
1048 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1050 this_cpu_sub(nr_dentry_unused, freed);
1051 shrink_dentry_list(&dispose);
1052 } while (freed > 0);
1054 EXPORT_SYMBOL(shrink_dcache_sb);
1057 * enum d_walk_ret - action to talke during tree walk
1058 * @D_WALK_CONTINUE: contrinue walk
1059 * @D_WALK_QUIT: quit walk
1060 * @D_WALK_NORETRY: quit when retry is needed
1061 * @D_WALK_SKIP: skip this dentry and its children
1071 * d_walk - walk the dentry tree
1072 * @parent: start of walk
1073 * @data: data passed to @enter() and @finish()
1074 * @enter: callback when first entering the dentry
1075 * @finish: callback when successfully finished the walk
1077 * The @enter() and @finish() callbacks are called with d_lock held.
1079 static void d_walk(struct dentry *parent, void *data,
1080 enum d_walk_ret (*enter)(void *, struct dentry *),
1081 void (*finish)(void *))
1083 struct dentry *this_parent;
1084 struct list_head *next;
1086 enum d_walk_ret ret;
1090 read_seqbegin_or_lock(&rename_lock, &seq);
1091 this_parent = parent;
1092 spin_lock(&this_parent->d_lock);
1094 ret = enter(data, this_parent);
1096 case D_WALK_CONTINUE:
1101 case D_WALK_NORETRY:
1106 next = this_parent->d_subdirs.next;
1108 while (next != &this_parent->d_subdirs) {
1109 struct list_head *tmp = next;
1110 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1113 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1115 ret = enter(data, dentry);
1117 case D_WALK_CONTINUE:
1120 spin_unlock(&dentry->d_lock);
1122 case D_WALK_NORETRY:
1126 spin_unlock(&dentry->d_lock);
1130 if (!list_empty(&dentry->d_subdirs)) {
1131 spin_unlock(&this_parent->d_lock);
1132 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1133 this_parent = dentry;
1134 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1137 spin_unlock(&dentry->d_lock);
1140 * All done at this level ... ascend and resume the search.
1142 if (this_parent != parent) {
1143 struct dentry *child = this_parent;
1144 this_parent = child->d_parent;
1147 spin_unlock(&child->d_lock);
1148 spin_lock(&this_parent->d_lock);
1151 * might go back up the wrong parent if we have had a rename
1154 if (this_parent != child->d_parent ||
1155 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1156 need_seqretry(&rename_lock, seq)) {
1157 spin_unlock(&this_parent->d_lock);
1162 next = child->d_u.d_child.next;
1165 if (need_seqretry(&rename_lock, seq)) {
1166 spin_unlock(&this_parent->d_lock);
1173 spin_unlock(&this_parent->d_lock);
1174 done_seqretry(&rename_lock, seq);
1185 * Search for at least 1 mount point in the dentry's subdirs.
1186 * We descend to the next level whenever the d_subdirs
1187 * list is non-empty and continue searching.
1190 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1193 if (d_mountpoint(dentry)) {
1197 return D_WALK_CONTINUE;
1201 * have_submounts - check for mounts over a dentry
1202 * @parent: dentry to check.
1204 * Return true if the parent or its subdirectories contain
1207 int have_submounts(struct dentry *parent)
1211 d_walk(parent, &ret, check_mount, NULL);
1215 EXPORT_SYMBOL(have_submounts);
1218 * Called by mount code to set a mountpoint and check if the mountpoint is
1219 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1220 * subtree can become unreachable).
1222 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1223 * this reason take rename_lock and d_lock on dentry and ancestors.
1225 int d_set_mounted(struct dentry *dentry)
1229 write_seqlock(&rename_lock);
1230 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1231 /* Need exclusion wrt. check_submounts_and_drop() */
1232 spin_lock(&p->d_lock);
1233 if (unlikely(d_unhashed(p))) {
1234 spin_unlock(&p->d_lock);
1237 spin_unlock(&p->d_lock);
1239 spin_lock(&dentry->d_lock);
1240 if (!d_unlinked(dentry)) {
1241 dentry->d_flags |= DCACHE_MOUNTED;
1244 spin_unlock(&dentry->d_lock);
1246 write_sequnlock(&rename_lock);
1251 * Search the dentry child list of the specified parent,
1252 * and move any unused dentries to the end of the unused
1253 * list for prune_dcache(). We descend to the next level
1254 * whenever the d_subdirs list is non-empty and continue
1257 * It returns zero iff there are no unused children,
1258 * otherwise it returns the number of children moved to
1259 * the end of the unused list. This may not be the total
1260 * number of unused children, because select_parent can
1261 * drop the lock and return early due to latency
1265 struct select_data {
1266 struct dentry *start;
1267 struct list_head dispose;
1271 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1273 struct select_data *data = _data;
1274 enum d_walk_ret ret = D_WALK_CONTINUE;
1276 if (data->start == dentry)
1279 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1282 if (dentry->d_flags & DCACHE_LRU_LIST)
1284 if (!dentry->d_lockref.count) {
1285 d_shrink_add(dentry, &data->dispose);
1290 * We can return to the caller if we have found some (this
1291 * ensures forward progress). We'll be coming back to find
1294 if (!list_empty(&data->dispose))
1295 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1301 * shrink_dcache_parent - prune dcache
1302 * @parent: parent of entries to prune
1304 * Prune the dcache to remove unused children of the parent dentry.
1306 void shrink_dcache_parent(struct dentry *parent)
1309 struct select_data data;
1311 INIT_LIST_HEAD(&data.dispose);
1312 data.start = parent;
1315 d_walk(parent, &data, select_collect, NULL);
1319 shrink_dentry_list(&data.dispose);
1323 EXPORT_SYMBOL(shrink_dcache_parent);
1325 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1327 /* it has busy descendents; complain about those instead */
1328 if (!list_empty(&dentry->d_subdirs))
1329 return D_WALK_CONTINUE;
1331 /* root with refcount 1 is fine */
1332 if (dentry == _data && dentry->d_lockref.count == 1)
1333 return D_WALK_CONTINUE;
1335 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1336 " still in use (%d) [unmount of %s %s]\n",
1339 dentry->d_inode->i_ino : 0UL,
1341 dentry->d_lockref.count,
1342 dentry->d_sb->s_type->name,
1343 dentry->d_sb->s_id);
1345 return D_WALK_CONTINUE;
1348 static void do_one_tree(struct dentry *dentry)
1350 shrink_dcache_parent(dentry);
1351 d_walk(dentry, dentry, umount_check, NULL);
1357 * destroy the dentries attached to a superblock on unmounting
1359 void shrink_dcache_for_umount(struct super_block *sb)
1361 struct dentry *dentry;
1363 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1365 dentry = sb->s_root;
1367 do_one_tree(dentry);
1369 while (!hlist_bl_empty(&sb->s_anon)) {
1370 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1371 do_one_tree(dentry);
1375 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1377 struct select_data *data = _data;
1379 if (d_mountpoint(dentry)) {
1380 data->found = -EBUSY;
1384 return select_collect(_data, dentry);
1387 static void check_and_drop(void *_data)
1389 struct select_data *data = _data;
1391 if (d_mountpoint(data->start))
1392 data->found = -EBUSY;
1394 __d_drop(data->start);
1398 * check_submounts_and_drop - prune dcache, check for submounts and drop
1400 * All done as a single atomic operation relative to has_unlinked_ancestor().
1401 * Returns 0 if successfully unhashed @parent. If there were submounts then
1404 * @dentry: dentry to prune and drop
1406 int check_submounts_and_drop(struct dentry *dentry)
1410 /* Negative dentries can be dropped without further checks */
1411 if (!dentry->d_inode) {
1417 struct select_data data;
1419 INIT_LIST_HEAD(&data.dispose);
1420 data.start = dentry;
1423 d_walk(dentry, &data, check_and_collect, check_and_drop);
1426 if (!list_empty(&data.dispose))
1427 shrink_dentry_list(&data.dispose);
1438 EXPORT_SYMBOL(check_submounts_and_drop);
1441 * __d_alloc - allocate a dcache entry
1442 * @sb: filesystem it will belong to
1443 * @name: qstr of the name
1445 * Allocates a dentry. It returns %NULL if there is insufficient memory
1446 * available. On a success the dentry is returned. The name passed in is
1447 * copied and the copy passed in may be reused after this call.
1450 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1452 struct dentry *dentry;
1455 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1460 * We guarantee that the inline name is always NUL-terminated.
1461 * This way the memcpy() done by the name switching in rename
1462 * will still always have a NUL at the end, even if we might
1463 * be overwriting an internal NUL character
1465 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1466 if (name->len > DNAME_INLINE_LEN-1) {
1467 size_t size = offsetof(struct external_name, name[1]);
1468 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1470 kmem_cache_free(dentry_cache, dentry);
1473 atomic_set(&p->u.count, 1);
1476 dname = dentry->d_iname;
1479 dentry->d_name.len = name->len;
1480 dentry->d_name.hash = name->hash;
1481 memcpy(dname, name->name, name->len);
1482 dname[name->len] = 0;
1484 /* Make sure we always see the terminating NUL character */
1486 dentry->d_name.name = dname;
1488 dentry->d_lockref.count = 1;
1489 dentry->d_flags = 0;
1490 spin_lock_init(&dentry->d_lock);
1491 seqcount_init(&dentry->d_seq);
1492 dentry->d_inode = NULL;
1493 dentry->d_parent = dentry;
1495 dentry->d_op = NULL;
1496 dentry->d_fsdata = NULL;
1497 INIT_HLIST_BL_NODE(&dentry->d_hash);
1498 INIT_LIST_HEAD(&dentry->d_lru);
1499 INIT_LIST_HEAD(&dentry->d_subdirs);
1500 INIT_HLIST_NODE(&dentry->d_alias);
1501 INIT_LIST_HEAD(&dentry->d_u.d_child);
1502 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1504 this_cpu_inc(nr_dentry);
1510 * d_alloc - allocate a dcache entry
1511 * @parent: parent of entry to allocate
1512 * @name: qstr of the name
1514 * Allocates a dentry. It returns %NULL if there is insufficient memory
1515 * available. On a success the dentry is returned. The name passed in is
1516 * copied and the copy passed in may be reused after this call.
1518 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1520 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1524 spin_lock(&parent->d_lock);
1526 * don't need child lock because it is not subject
1527 * to concurrency here
1529 __dget_dlock(parent);
1530 dentry->d_parent = parent;
1531 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1532 spin_unlock(&parent->d_lock);
1536 EXPORT_SYMBOL(d_alloc);
1539 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1540 * @sb: the superblock
1541 * @name: qstr of the name
1543 * For a filesystem that just pins its dentries in memory and never
1544 * performs lookups at all, return an unhashed IS_ROOT dentry.
1546 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1548 return __d_alloc(sb, name);
1550 EXPORT_SYMBOL(d_alloc_pseudo);
1552 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1557 q.len = strlen(name);
1558 q.hash = full_name_hash(q.name, q.len);
1559 return d_alloc(parent, &q);
1561 EXPORT_SYMBOL(d_alloc_name);
1563 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1565 WARN_ON_ONCE(dentry->d_op);
1566 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1568 DCACHE_OP_REVALIDATE |
1569 DCACHE_OP_WEAK_REVALIDATE |
1570 DCACHE_OP_DELETE ));
1575 dentry->d_flags |= DCACHE_OP_HASH;
1577 dentry->d_flags |= DCACHE_OP_COMPARE;
1578 if (op->d_revalidate)
1579 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1580 if (op->d_weak_revalidate)
1581 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1583 dentry->d_flags |= DCACHE_OP_DELETE;
1585 dentry->d_flags |= DCACHE_OP_PRUNE;
1588 EXPORT_SYMBOL(d_set_d_op);
1590 static unsigned d_flags_for_inode(struct inode *inode)
1592 unsigned add_flags = DCACHE_FILE_TYPE;
1595 return DCACHE_MISS_TYPE;
1597 if (S_ISDIR(inode->i_mode)) {
1598 add_flags = DCACHE_DIRECTORY_TYPE;
1599 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1600 if (unlikely(!inode->i_op->lookup))
1601 add_flags = DCACHE_AUTODIR_TYPE;
1603 inode->i_opflags |= IOP_LOOKUP;
1605 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1606 if (unlikely(inode->i_op->follow_link))
1607 add_flags = DCACHE_SYMLINK_TYPE;
1609 inode->i_opflags |= IOP_NOFOLLOW;
1612 if (unlikely(IS_AUTOMOUNT(inode)))
1613 add_flags |= DCACHE_NEED_AUTOMOUNT;
1617 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1619 unsigned add_flags = d_flags_for_inode(inode);
1621 spin_lock(&dentry->d_lock);
1622 __d_set_type(dentry, add_flags);
1624 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1625 dentry->d_inode = inode;
1626 dentry_rcuwalk_barrier(dentry);
1627 spin_unlock(&dentry->d_lock);
1628 fsnotify_d_instantiate(dentry, inode);
1632 * d_instantiate - fill in inode information for a dentry
1633 * @entry: dentry to complete
1634 * @inode: inode to attach to this dentry
1636 * Fill in inode information in the entry.
1638 * This turns negative dentries into productive full members
1641 * NOTE! This assumes that the inode count has been incremented
1642 * (or otherwise set) by the caller to indicate that it is now
1643 * in use by the dcache.
1646 void d_instantiate(struct dentry *entry, struct inode * inode)
1648 BUG_ON(!hlist_unhashed(&entry->d_alias));
1650 spin_lock(&inode->i_lock);
1651 __d_instantiate(entry, inode);
1653 spin_unlock(&inode->i_lock);
1654 security_d_instantiate(entry, inode);
1656 EXPORT_SYMBOL(d_instantiate);
1659 * d_instantiate_unique - instantiate a non-aliased dentry
1660 * @entry: dentry to instantiate
1661 * @inode: inode to attach to this dentry
1663 * Fill in inode information in the entry. On success, it returns NULL.
1664 * If an unhashed alias of "entry" already exists, then we return the
1665 * aliased dentry instead and drop one reference to inode.
1667 * Note that in order to avoid conflicts with rename() etc, the caller
1668 * had better be holding the parent directory semaphore.
1670 * This also assumes that the inode count has been incremented
1671 * (or otherwise set) by the caller to indicate that it is now
1672 * in use by the dcache.
1674 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1675 struct inode *inode)
1677 struct dentry *alias;
1678 int len = entry->d_name.len;
1679 const char *name = entry->d_name.name;
1680 unsigned int hash = entry->d_name.hash;
1683 __d_instantiate(entry, NULL);
1687 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1689 * Don't need alias->d_lock here, because aliases with
1690 * d_parent == entry->d_parent are not subject to name or
1691 * parent changes, because the parent inode i_mutex is held.
1693 if (alias->d_name.hash != hash)
1695 if (alias->d_parent != entry->d_parent)
1697 if (alias->d_name.len != len)
1699 if (dentry_cmp(alias, name, len))
1705 __d_instantiate(entry, inode);
1709 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1711 struct dentry *result;
1713 BUG_ON(!hlist_unhashed(&entry->d_alias));
1716 spin_lock(&inode->i_lock);
1717 result = __d_instantiate_unique(entry, inode);
1719 spin_unlock(&inode->i_lock);
1722 security_d_instantiate(entry, inode);
1726 BUG_ON(!d_unhashed(result));
1731 EXPORT_SYMBOL(d_instantiate_unique);
1734 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1735 * @entry: dentry to complete
1736 * @inode: inode to attach to this dentry
1738 * Fill in inode information in the entry. If a directory alias is found, then
1739 * return an error (and drop inode). Together with d_materialise_unique() this
1740 * guarantees that a directory inode may never have more than one alias.
1742 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1744 BUG_ON(!hlist_unhashed(&entry->d_alias));
1746 spin_lock(&inode->i_lock);
1747 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1748 spin_unlock(&inode->i_lock);
1752 __d_instantiate(entry, inode);
1753 spin_unlock(&inode->i_lock);
1754 security_d_instantiate(entry, inode);
1758 EXPORT_SYMBOL(d_instantiate_no_diralias);
1760 struct dentry *d_make_root(struct inode *root_inode)
1762 struct dentry *res = NULL;
1765 static const struct qstr name = QSTR_INIT("/", 1);
1767 res = __d_alloc(root_inode->i_sb, &name);
1769 d_instantiate(res, root_inode);
1775 EXPORT_SYMBOL(d_make_root);
1777 static struct dentry * __d_find_any_alias(struct inode *inode)
1779 struct dentry *alias;
1781 if (hlist_empty(&inode->i_dentry))
1783 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1789 * d_find_any_alias - find any alias for a given inode
1790 * @inode: inode to find an alias for
1792 * If any aliases exist for the given inode, take and return a
1793 * reference for one of them. If no aliases exist, return %NULL.
1795 struct dentry *d_find_any_alias(struct inode *inode)
1799 spin_lock(&inode->i_lock);
1800 de = __d_find_any_alias(inode);
1801 spin_unlock(&inode->i_lock);
1804 EXPORT_SYMBOL(d_find_any_alias);
1806 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1808 static const struct qstr anonstring = QSTR_INIT("/", 1);
1814 return ERR_PTR(-ESTALE);
1816 return ERR_CAST(inode);
1818 res = d_find_any_alias(inode);
1822 tmp = __d_alloc(inode->i_sb, &anonstring);
1824 res = ERR_PTR(-ENOMEM);
1828 spin_lock(&inode->i_lock);
1829 res = __d_find_any_alias(inode);
1831 spin_unlock(&inode->i_lock);
1836 /* attach a disconnected dentry */
1837 add_flags = d_flags_for_inode(inode);
1840 add_flags |= DCACHE_DISCONNECTED;
1842 spin_lock(&tmp->d_lock);
1843 tmp->d_inode = inode;
1844 tmp->d_flags |= add_flags;
1845 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1846 hlist_bl_lock(&tmp->d_sb->s_anon);
1847 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1848 hlist_bl_unlock(&tmp->d_sb->s_anon);
1849 spin_unlock(&tmp->d_lock);
1850 spin_unlock(&inode->i_lock);
1851 security_d_instantiate(tmp, inode);
1856 if (res && !IS_ERR(res))
1857 security_d_instantiate(res, inode);
1863 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1864 * @inode: inode to allocate the dentry for
1866 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1867 * similar open by handle operations. The returned dentry may be anonymous,
1868 * or may have a full name (if the inode was already in the cache).
1870 * When called on a directory inode, we must ensure that the inode only ever
1871 * has one dentry. If a dentry is found, that is returned instead of
1872 * 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 released.
1876 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1877 * be passed in and the error will be propagated to the return value,
1878 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1880 struct dentry *d_obtain_alias(struct inode *inode)
1882 return __d_obtain_alias(inode, 1);
1884 EXPORT_SYMBOL(d_obtain_alias);
1887 * d_obtain_root - find or allocate a dentry for a given inode
1888 * @inode: inode to allocate the dentry for
1890 * Obtain an IS_ROOT dentry for the root of a filesystem.
1892 * We must ensure that directory inodes only ever have one dentry. If a
1893 * dentry is found, that is returned instead of allocating a new one.
1895 * On successful return, the reference to the inode has been transferred
1896 * to the dentry. In case of an error the reference on the inode is
1897 * released. A %NULL or IS_ERR inode may be passed in and will be the
1898 * error will be propagate to the return value, with a %NULL @inode
1899 * replaced by ERR_PTR(-ESTALE).
1901 struct dentry *d_obtain_root(struct inode *inode)
1903 return __d_obtain_alias(inode, 0);
1905 EXPORT_SYMBOL(d_obtain_root);
1908 * d_add_ci - lookup or allocate new dentry with case-exact name
1909 * @inode: the inode case-insensitive lookup has found
1910 * @dentry: the negative dentry that was passed to the parent's lookup func
1911 * @name: the case-exact name to be associated with the returned dentry
1913 * This is to avoid filling the dcache with case-insensitive names to the
1914 * same inode, only the actual correct case is stored in the dcache for
1915 * case-insensitive filesystems.
1917 * For a case-insensitive lookup match and if the the case-exact dentry
1918 * already exists in in the dcache, use it and return it.
1920 * If no entry exists with the exact case name, allocate new dentry with
1921 * the exact case, and return the spliced entry.
1923 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1926 struct dentry *found;
1930 * First check if a dentry matching the name already exists,
1931 * if not go ahead and create it now.
1933 found = d_hash_and_lookup(dentry->d_parent, name);
1934 if (unlikely(IS_ERR(found)))
1937 new = d_alloc(dentry->d_parent, name);
1939 found = ERR_PTR(-ENOMEM);
1943 found = d_splice_alias(inode, new);
1952 * If a matching dentry exists, and it's not negative use it.
1954 * Decrement the reference count to balance the iget() done
1957 if (found->d_inode) {
1958 if (unlikely(found->d_inode != inode)) {
1959 /* This can't happen because bad inodes are unhashed. */
1960 BUG_ON(!is_bad_inode(inode));
1961 BUG_ON(!is_bad_inode(found->d_inode));
1968 * Negative dentry: instantiate it unless the inode is a directory and
1969 * already has a dentry.
1971 new = d_splice_alias(inode, found);
1982 EXPORT_SYMBOL(d_add_ci);
1985 * Do the slow-case of the dentry name compare.
1987 * Unlike the dentry_cmp() function, we need to atomically
1988 * load the name and length information, so that the
1989 * filesystem can rely on them, and can use the 'name' and
1990 * 'len' information without worrying about walking off the
1991 * end of memory etc.
1993 * Thus the read_seqcount_retry() and the "duplicate" info
1994 * in arguments (the low-level filesystem should not look
1995 * at the dentry inode or name contents directly, since
1996 * rename can change them while we're in RCU mode).
1998 enum slow_d_compare {
2004 static noinline enum slow_d_compare slow_dentry_cmp(
2005 const struct dentry *parent,
2006 struct dentry *dentry,
2008 const struct qstr *name)
2010 int tlen = dentry->d_name.len;
2011 const char *tname = dentry->d_name.name;
2013 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2015 return D_COMP_SEQRETRY;
2017 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2018 return D_COMP_NOMATCH;
2023 * __d_lookup_rcu - search for a dentry (racy, store-free)
2024 * @parent: parent dentry
2025 * @name: qstr of name we wish to find
2026 * @seqp: returns d_seq value at the point where the dentry was found
2027 * Returns: dentry, or NULL
2029 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2030 * resolution (store-free path walking) design described in
2031 * Documentation/filesystems/path-lookup.txt.
2033 * This is not to be used outside core vfs.
2035 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2036 * held, and rcu_read_lock held. The returned dentry must not be stored into
2037 * without taking d_lock and checking d_seq sequence count against @seq
2040 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2043 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2044 * the returned dentry, so long as its parent's seqlock is checked after the
2045 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2046 * is formed, giving integrity down the path walk.
2048 * NOTE! The caller *has* to check the resulting dentry against the sequence
2049 * number we've returned before using any of the resulting dentry state!
2051 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2052 const struct qstr *name,
2055 u64 hashlen = name->hash_len;
2056 const unsigned char *str = name->name;
2057 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2058 struct hlist_bl_node *node;
2059 struct dentry *dentry;
2062 * Note: There is significant duplication with __d_lookup_rcu which is
2063 * required to prevent single threaded performance regressions
2064 * especially on architectures where smp_rmb (in seqcounts) are costly.
2065 * Keep the two functions in sync.
2069 * The hash list is protected using RCU.
2071 * Carefully use d_seq when comparing a candidate dentry, to avoid
2072 * races with d_move().
2074 * It is possible that concurrent renames can mess up our list
2075 * walk here and result in missing our dentry, resulting in the
2076 * false-negative result. d_lookup() protects against concurrent
2077 * renames using rename_lock seqlock.
2079 * See Documentation/filesystems/path-lookup.txt for more details.
2081 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2086 * The dentry sequence count protects us from concurrent
2087 * renames, and thus protects parent and name fields.
2089 * The caller must perform a seqcount check in order
2090 * to do anything useful with the returned dentry.
2092 * NOTE! We do a "raw" seqcount_begin here. That means that
2093 * we don't wait for the sequence count to stabilize if it
2094 * is in the middle of a sequence change. If we do the slow
2095 * dentry compare, we will do seqretries until it is stable,
2096 * and if we end up with a successful lookup, we actually
2097 * want to exit RCU lookup anyway.
2099 seq = raw_seqcount_begin(&dentry->d_seq);
2100 if (dentry->d_parent != parent)
2102 if (d_unhashed(dentry))
2105 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2106 if (dentry->d_name.hash != hashlen_hash(hashlen))
2109 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2112 case D_COMP_NOMATCH:
2119 if (dentry->d_name.hash_len != hashlen)
2122 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2129 * d_lookup - search for a dentry
2130 * @parent: parent dentry
2131 * @name: qstr of name we wish to find
2132 * Returns: dentry, or NULL
2134 * d_lookup searches the children of the parent dentry for the name in
2135 * question. If the dentry is found its reference count is incremented and the
2136 * dentry is returned. The caller must use dput to free the entry when it has
2137 * finished using it. %NULL is returned if the dentry does not exist.
2139 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2141 struct dentry *dentry;
2145 seq = read_seqbegin(&rename_lock);
2146 dentry = __d_lookup(parent, name);
2149 } while (read_seqretry(&rename_lock, seq));
2152 EXPORT_SYMBOL(d_lookup);
2155 * __d_lookup - search for a dentry (racy)
2156 * @parent: parent dentry
2157 * @name: qstr of name we wish to find
2158 * Returns: dentry, or NULL
2160 * __d_lookup is like d_lookup, however it may (rarely) return a
2161 * false-negative result due to unrelated rename activity.
2163 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2164 * however it must be used carefully, eg. with a following d_lookup in
2165 * the case of failure.
2167 * __d_lookup callers must be commented.
2169 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2171 unsigned int len = name->len;
2172 unsigned int hash = name->hash;
2173 const unsigned char *str = name->name;
2174 struct hlist_bl_head *b = d_hash(parent, hash);
2175 struct hlist_bl_node *node;
2176 struct dentry *found = NULL;
2177 struct dentry *dentry;
2180 * Note: There is significant duplication with __d_lookup_rcu which is
2181 * required to prevent single threaded performance regressions
2182 * especially on architectures where smp_rmb (in seqcounts) are costly.
2183 * Keep the two functions in sync.
2187 * The hash list is protected using RCU.
2189 * Take d_lock when comparing a candidate dentry, to avoid races
2192 * It is possible that concurrent renames can mess up our list
2193 * walk here and result in missing our dentry, resulting in the
2194 * false-negative result. d_lookup() protects against concurrent
2195 * renames using rename_lock seqlock.
2197 * See Documentation/filesystems/path-lookup.txt for more details.
2201 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2203 if (dentry->d_name.hash != hash)
2206 spin_lock(&dentry->d_lock);
2207 if (dentry->d_parent != parent)
2209 if (d_unhashed(dentry))
2213 * It is safe to compare names since d_move() cannot
2214 * change the qstr (protected by d_lock).
2216 if (parent->d_flags & DCACHE_OP_COMPARE) {
2217 int tlen = dentry->d_name.len;
2218 const char *tname = dentry->d_name.name;
2219 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2222 if (dentry->d_name.len != len)
2224 if (dentry_cmp(dentry, str, len))
2228 dentry->d_lockref.count++;
2230 spin_unlock(&dentry->d_lock);
2233 spin_unlock(&dentry->d_lock);
2241 * d_hash_and_lookup - hash the qstr then search for a dentry
2242 * @dir: Directory to search in
2243 * @name: qstr of name we wish to find
2245 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2247 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2250 * Check for a fs-specific hash function. Note that we must
2251 * calculate the standard hash first, as the d_op->d_hash()
2252 * routine may choose to leave the hash value unchanged.
2254 name->hash = full_name_hash(name->name, name->len);
2255 if (dir->d_flags & DCACHE_OP_HASH) {
2256 int err = dir->d_op->d_hash(dir, name);
2257 if (unlikely(err < 0))
2258 return ERR_PTR(err);
2260 return d_lookup(dir, name);
2262 EXPORT_SYMBOL(d_hash_and_lookup);
2265 * d_validate - verify dentry provided from insecure source (deprecated)
2266 * @dentry: The dentry alleged to be valid child of @dparent
2267 * @dparent: The parent dentry (known to be valid)
2269 * An insecure source has sent us a dentry, here we verify it and dget() it.
2270 * This is used by ncpfs in its readdir implementation.
2271 * Zero is returned in the dentry is invalid.
2273 * This function is slow for big directories, and deprecated, do not use it.
2275 int d_validate(struct dentry *dentry, struct dentry *dparent)
2277 struct dentry *child;
2279 spin_lock(&dparent->d_lock);
2280 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2281 if (dentry == child) {
2282 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2283 __dget_dlock(dentry);
2284 spin_unlock(&dentry->d_lock);
2285 spin_unlock(&dparent->d_lock);
2289 spin_unlock(&dparent->d_lock);
2293 EXPORT_SYMBOL(d_validate);
2296 * When a file is deleted, we have two options:
2297 * - turn this dentry into a negative dentry
2298 * - unhash this dentry and free it.
2300 * Usually, we want to just turn this into
2301 * a negative dentry, but if anybody else is
2302 * currently using the dentry or the inode
2303 * we can't do that and we fall back on removing
2304 * it from the hash queues and waiting for
2305 * it to be deleted later when it has no users
2309 * d_delete - delete a dentry
2310 * @dentry: The dentry to delete
2312 * Turn the dentry into a negative dentry if possible, otherwise
2313 * remove it from the hash queues so it can be deleted later
2316 void d_delete(struct dentry * dentry)
2318 struct inode *inode;
2321 * Are we the only user?
2324 spin_lock(&dentry->d_lock);
2325 inode = dentry->d_inode;
2326 isdir = S_ISDIR(inode->i_mode);
2327 if (dentry->d_lockref.count == 1) {
2328 if (!spin_trylock(&inode->i_lock)) {
2329 spin_unlock(&dentry->d_lock);
2333 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2334 dentry_unlink_inode(dentry);
2335 fsnotify_nameremove(dentry, isdir);
2339 if (!d_unhashed(dentry))
2342 spin_unlock(&dentry->d_lock);
2344 fsnotify_nameremove(dentry, isdir);
2346 EXPORT_SYMBOL(d_delete);
2348 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2350 BUG_ON(!d_unhashed(entry));
2352 entry->d_flags |= DCACHE_RCUACCESS;
2353 hlist_bl_add_head_rcu(&entry->d_hash, b);
2357 static void _d_rehash(struct dentry * entry)
2359 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2363 * d_rehash - add an entry back to the hash
2364 * @entry: dentry to add to the hash
2366 * Adds a dentry to the hash according to its name.
2369 void d_rehash(struct dentry * entry)
2371 spin_lock(&entry->d_lock);
2373 spin_unlock(&entry->d_lock);
2375 EXPORT_SYMBOL(d_rehash);
2378 * dentry_update_name_case - update case insensitive dentry with a new name
2379 * @dentry: dentry to be updated
2382 * Update a case insensitive dentry with new case of name.
2384 * dentry must have been returned by d_lookup with name @name. Old and new
2385 * name lengths must match (ie. no d_compare which allows mismatched name
2388 * Parent inode i_mutex must be held over d_lookup and into this call (to
2389 * keep renames and concurrent inserts, and readdir(2) away).
2391 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2393 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2394 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2396 spin_lock(&dentry->d_lock);
2397 write_seqcount_begin(&dentry->d_seq);
2398 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2399 write_seqcount_end(&dentry->d_seq);
2400 spin_unlock(&dentry->d_lock);
2402 EXPORT_SYMBOL(dentry_update_name_case);
2404 static void swap_names(struct dentry *dentry, struct dentry *target)
2406 if (unlikely(dname_external(target))) {
2407 if (unlikely(dname_external(dentry))) {
2409 * Both external: swap the pointers
2411 swap(target->d_name.name, dentry->d_name.name);
2414 * dentry:internal, target:external. Steal target's
2415 * storage and make target internal.
2417 memcpy(target->d_iname, dentry->d_name.name,
2418 dentry->d_name.len + 1);
2419 dentry->d_name.name = target->d_name.name;
2420 target->d_name.name = target->d_iname;
2423 if (unlikely(dname_external(dentry))) {
2425 * dentry:external, target:internal. Give dentry's
2426 * storage to target and make dentry internal
2428 memcpy(dentry->d_iname, target->d_name.name,
2429 target->d_name.len + 1);
2430 target->d_name.name = dentry->d_name.name;
2431 dentry->d_name.name = dentry->d_iname;
2434 * Both are internal.
2437 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2438 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2439 swap(((long *) &dentry->d_iname)[i],
2440 ((long *) &target->d_iname)[i]);
2444 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2447 static void copy_name(struct dentry *dentry, struct dentry *target)
2449 struct external_name *old_name = NULL;
2450 if (unlikely(dname_external(dentry)))
2451 old_name = external_name(dentry);
2452 if (unlikely(dname_external(target))) {
2453 atomic_inc(&external_name(target)->u.count);
2454 dentry->d_name = target->d_name;
2456 memcpy(dentry->d_iname, target->d_name.name,
2457 target->d_name.len + 1);
2458 dentry->d_name.name = dentry->d_iname;
2459 dentry->d_name.hash_len = target->d_name.hash_len;
2461 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2462 kfree_rcu(old_name, u.head);
2465 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2468 * XXXX: do we really need to take target->d_lock?
2470 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2471 spin_lock(&target->d_parent->d_lock);
2473 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2474 spin_lock(&dentry->d_parent->d_lock);
2475 spin_lock_nested(&target->d_parent->d_lock,
2476 DENTRY_D_LOCK_NESTED);
2478 spin_lock(&target->d_parent->d_lock);
2479 spin_lock_nested(&dentry->d_parent->d_lock,
2480 DENTRY_D_LOCK_NESTED);
2483 if (target < dentry) {
2484 spin_lock_nested(&target->d_lock, 2);
2485 spin_lock_nested(&dentry->d_lock, 3);
2487 spin_lock_nested(&dentry->d_lock, 2);
2488 spin_lock_nested(&target->d_lock, 3);
2492 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2494 if (target->d_parent != dentry->d_parent)
2495 spin_unlock(&dentry->d_parent->d_lock);
2496 if (target->d_parent != target)
2497 spin_unlock(&target->d_parent->d_lock);
2498 spin_unlock(&target->d_lock);
2499 spin_unlock(&dentry->d_lock);
2503 * When switching names, the actual string doesn't strictly have to
2504 * be preserved in the target - because we're dropping the target
2505 * anyway. As such, we can just do a simple memcpy() to copy over
2506 * the new name before we switch, unless we are going to rehash
2507 * it. Note that if we *do* unhash the target, we are not allowed
2508 * to rehash it without giving it a new name/hash key - whether
2509 * we swap or overwrite the names here, resulting name won't match
2510 * the reality in filesystem; it's only there for d_path() purposes.
2511 * Note that all of this is happening under rename_lock, so the
2512 * any hash lookup seeing it in the middle of manipulations will
2513 * be discarded anyway. So we do not care what happens to the hash
2517 * __d_move - move a dentry
2518 * @dentry: entry to move
2519 * @target: new dentry
2520 * @exchange: exchange the two dentries
2522 * Update the dcache to reflect the move of a file name. Negative
2523 * dcache entries should not be moved in this way. Caller must hold
2524 * rename_lock, the i_mutex of the source and target directories,
2525 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2527 static void __d_move(struct dentry *dentry, struct dentry *target,
2530 if (!dentry->d_inode)
2531 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2533 BUG_ON(d_ancestor(dentry, target));
2534 BUG_ON(d_ancestor(target, dentry));
2536 dentry_lock_for_move(dentry, target);
2538 write_seqcount_begin(&dentry->d_seq);
2539 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2541 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2544 * Move the dentry to the target hash queue. Don't bother checking
2545 * for the same hash queue because of how unlikely it is.
2548 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2551 * Unhash the target (d_delete() is not usable here). If exchanging
2552 * the two dentries, then rehash onto the other's hash queue.
2557 d_hash(dentry->d_parent, dentry->d_name.hash));
2560 /* Switch the names.. */
2562 swap_names(dentry, target);
2564 copy_name(dentry, target);
2566 /* ... and switch them in the tree */
2567 if (IS_ROOT(dentry)) {
2568 /* splicing a tree */
2569 dentry->d_parent = target->d_parent;
2570 target->d_parent = target;
2571 list_del_init(&target->d_u.d_child);
2572 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2574 /* swapping two dentries */
2575 swap(dentry->d_parent, target->d_parent);
2576 list_move(&target->d_u.d_child, &target->d_parent->d_subdirs);
2577 list_move(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2579 fsnotify_d_move(target);
2580 fsnotify_d_move(dentry);
2583 write_seqcount_end(&target->d_seq);
2584 write_seqcount_end(&dentry->d_seq);
2586 dentry_unlock_for_move(dentry, target);
2590 * d_move - move a dentry
2591 * @dentry: entry to move
2592 * @target: new dentry
2594 * Update the dcache to reflect the move of a file name. Negative
2595 * dcache entries should not be moved in this way. See the locking
2596 * requirements for __d_move.
2598 void d_move(struct dentry *dentry, struct dentry *target)
2600 write_seqlock(&rename_lock);
2601 __d_move(dentry, target, false);
2602 write_sequnlock(&rename_lock);
2604 EXPORT_SYMBOL(d_move);
2607 * d_exchange - exchange two dentries
2608 * @dentry1: first dentry
2609 * @dentry2: second dentry
2611 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2613 write_seqlock(&rename_lock);
2615 WARN_ON(!dentry1->d_inode);
2616 WARN_ON(!dentry2->d_inode);
2617 WARN_ON(IS_ROOT(dentry1));
2618 WARN_ON(IS_ROOT(dentry2));
2620 __d_move(dentry1, dentry2, true);
2622 write_sequnlock(&rename_lock);
2626 * d_ancestor - search for an ancestor
2627 * @p1: ancestor dentry
2630 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2631 * an ancestor of p2, else NULL.
2633 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2637 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2638 if (p->d_parent == p1)
2645 * This helper attempts to cope with remotely renamed directories
2647 * It assumes that the caller is already holding
2648 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2650 * Note: If ever the locking in lock_rename() changes, then please
2651 * remember to update this too...
2653 static struct dentry *__d_unalias(struct inode *inode,
2654 struct dentry *dentry, struct dentry *alias)
2656 struct mutex *m1 = NULL, *m2 = NULL;
2657 struct dentry *ret = ERR_PTR(-EBUSY);
2659 /* If alias and dentry share a parent, then no extra locks required */
2660 if (alias->d_parent == dentry->d_parent)
2663 /* See lock_rename() */
2664 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2666 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2667 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2669 m2 = &alias->d_parent->d_inode->i_mutex;
2671 if (likely(!d_mountpoint(alias))) {
2672 __d_move(alias, dentry, false);
2676 spin_unlock(&inode->i_lock);
2685 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2686 * @inode: the inode which may have a disconnected dentry
2687 * @dentry: a negative dentry which we want to point to the inode.
2689 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2690 * place of the given dentry and return it, else simply d_add the inode
2691 * to the dentry and return NULL.
2693 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2694 * we should error out: directories can't have multiple aliases.
2696 * This is needed in the lookup routine of any filesystem that is exportable
2697 * (via knfsd) so that we can build dcache paths to directories effectively.
2699 * If a dentry was found and moved, then it is returned. Otherwise NULL
2700 * is returned. This matches the expected return value of ->lookup.
2702 * Cluster filesystems may call this function with a negative, hashed dentry.
2703 * In that case, we know that the inode will be a regular file, and also this
2704 * will only occur during atomic_open. So we need to check for the dentry
2705 * being already hashed only in the final case.
2707 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2709 struct dentry *new = NULL;
2712 return ERR_CAST(inode);
2714 if (inode && S_ISDIR(inode->i_mode)) {
2715 spin_lock(&inode->i_lock);
2716 new = __d_find_any_alias(inode);
2718 if (!IS_ROOT(new)) {
2719 spin_unlock(&inode->i_lock);
2721 return ERR_PTR(-EIO);
2723 if (d_ancestor(new, dentry)) {
2724 spin_unlock(&inode->i_lock);
2726 return ERR_PTR(-EIO);
2728 write_seqlock(&rename_lock);
2729 __d_move(new, dentry, false);
2730 write_sequnlock(&rename_lock);
2731 spin_unlock(&inode->i_lock);
2732 security_d_instantiate(new, inode);
2735 /* already taking inode->i_lock, so d_add() by hand */
2736 __d_instantiate(dentry, inode);
2737 spin_unlock(&inode->i_lock);
2738 security_d_instantiate(dentry, inode);
2742 d_instantiate(dentry, inode);
2743 if (d_unhashed(dentry))
2748 EXPORT_SYMBOL(d_splice_alias);
2751 * d_materialise_unique - introduce an inode into the tree
2752 * @dentry: candidate dentry
2753 * @inode: inode to bind to the dentry, to which aliases may be attached
2755 * Introduces an dentry into the tree, substituting an extant disconnected
2756 * root directory alias in its place if there is one. Caller must hold the
2757 * i_mutex of the parent directory.
2759 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2761 struct dentry *actual;
2763 BUG_ON(!d_unhashed(dentry));
2767 __d_instantiate(dentry, NULL);
2772 spin_lock(&inode->i_lock);
2774 if (S_ISDIR(inode->i_mode)) {
2775 struct dentry *alias;
2777 /* Does an aliased dentry already exist? */
2778 alias = __d_find_alias(inode);
2781 write_seqlock(&rename_lock);
2783 if (d_ancestor(alias, dentry)) {
2784 /* Check for loops */
2785 actual = ERR_PTR(-ELOOP);
2786 spin_unlock(&inode->i_lock);
2787 } else if (IS_ROOT(alias)) {
2788 /* Is this an anonymous mountpoint that we
2789 * could splice into our tree? */
2790 __d_move(alias, dentry, false);
2791 write_sequnlock(&rename_lock);
2794 /* Nope, but we must(!) avoid directory
2795 * aliasing. This drops inode->i_lock */
2796 actual = __d_unalias(inode, dentry, alias);
2798 write_sequnlock(&rename_lock);
2799 if (IS_ERR(actual)) {
2800 if (PTR_ERR(actual) == -ELOOP)
2801 pr_warn_ratelimited(
2802 "VFS: Lookup of '%s' in %s %s"
2803 " would have caused loop\n",
2804 dentry->d_name.name,
2805 inode->i_sb->s_type->name,
2813 /* Add a unique reference */
2814 actual = __d_instantiate_unique(dentry, inode);
2820 spin_unlock(&inode->i_lock);
2822 if (actual == dentry) {
2823 security_d_instantiate(dentry, inode);
2830 EXPORT_SYMBOL_GPL(d_materialise_unique);
2832 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2836 return -ENAMETOOLONG;
2838 memcpy(*buffer, str, namelen);
2843 * prepend_name - prepend a pathname in front of current buffer pointer
2844 * @buffer: buffer pointer
2845 * @buflen: allocated length of the buffer
2846 * @name: name string and length qstr structure
2848 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2849 * make sure that either the old or the new name pointer and length are
2850 * fetched. However, there may be mismatch between length and pointer.
2851 * The length cannot be trusted, we need to copy it byte-by-byte until
2852 * the length is reached or a null byte is found. It also prepends "/" at
2853 * the beginning of the name. The sequence number check at the caller will
2854 * retry it again when a d_move() does happen. So any garbage in the buffer
2855 * due to mismatched pointer and length will be discarded.
2857 * Data dependency barrier is needed to make sure that we see that terminating
2858 * NUL. Alpha strikes again, film at 11...
2860 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2862 const char *dname = ACCESS_ONCE(name->name);
2863 u32 dlen = ACCESS_ONCE(name->len);
2866 smp_read_barrier_depends();
2868 *buflen -= dlen + 1;
2870 return -ENAMETOOLONG;
2871 p = *buffer -= dlen + 1;
2883 * prepend_path - Prepend path string to a buffer
2884 * @path: the dentry/vfsmount to report
2885 * @root: root vfsmnt/dentry
2886 * @buffer: pointer to the end of the buffer
2887 * @buflen: pointer to buffer length
2889 * The function will first try to write out the pathname without taking any
2890 * lock other than the RCU read lock to make sure that dentries won't go away.
2891 * It only checks the sequence number of the global rename_lock as any change
2892 * in the dentry's d_seq will be preceded by changes in the rename_lock
2893 * sequence number. If the sequence number had been changed, it will restart
2894 * the whole pathname back-tracing sequence again by taking the rename_lock.
2895 * In this case, there is no need to take the RCU read lock as the recursive
2896 * parent pointer references will keep the dentry chain alive as long as no
2897 * rename operation is performed.
2899 static int prepend_path(const struct path *path,
2900 const struct path *root,
2901 char **buffer, int *buflen)
2903 struct dentry *dentry;
2904 struct vfsmount *vfsmnt;
2907 unsigned seq, m_seq = 0;
2913 read_seqbegin_or_lock(&mount_lock, &m_seq);
2920 dentry = path->dentry;
2922 mnt = real_mount(vfsmnt);
2923 read_seqbegin_or_lock(&rename_lock, &seq);
2924 while (dentry != root->dentry || vfsmnt != root->mnt) {
2925 struct dentry * parent;
2927 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2928 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2930 if (mnt != parent) {
2931 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2937 * Filesystems needing to implement special "root names"
2938 * should do so with ->d_dname()
2940 if (IS_ROOT(dentry) &&
2941 (dentry->d_name.len != 1 ||
2942 dentry->d_name.name[0] != '/')) {
2943 WARN(1, "Root dentry has weird name <%.*s>\n",
2944 (int) dentry->d_name.len,
2945 dentry->d_name.name);
2948 error = is_mounted(vfsmnt) ? 1 : 2;
2951 parent = dentry->d_parent;
2953 error = prepend_name(&bptr, &blen, &dentry->d_name);
2961 if (need_seqretry(&rename_lock, seq)) {
2965 done_seqretry(&rename_lock, seq);
2969 if (need_seqretry(&mount_lock, m_seq)) {
2973 done_seqretry(&mount_lock, m_seq);
2975 if (error >= 0 && bptr == *buffer) {
2977 error = -ENAMETOOLONG;
2987 * __d_path - return the path of a dentry
2988 * @path: the dentry/vfsmount to report
2989 * @root: root vfsmnt/dentry
2990 * @buf: buffer to return value in
2991 * @buflen: buffer length
2993 * Convert a dentry into an ASCII path name.
2995 * Returns a pointer into the buffer or an error code if the
2996 * path was too long.
2998 * "buflen" should be positive.
3000 * If the path is not reachable from the supplied root, return %NULL.
3002 char *__d_path(const struct path *path,
3003 const struct path *root,
3004 char *buf, int buflen)
3006 char *res = buf + buflen;
3009 prepend(&res, &buflen, "\0", 1);
3010 error = prepend_path(path, root, &res, &buflen);
3013 return ERR_PTR(error);
3019 char *d_absolute_path(const struct path *path,
3020 char *buf, int buflen)
3022 struct path root = {};
3023 char *res = buf + buflen;
3026 prepend(&res, &buflen, "\0", 1);
3027 error = prepend_path(path, &root, &res, &buflen);
3032 return ERR_PTR(error);
3037 * same as __d_path but appends "(deleted)" for unlinked files.
3039 static int path_with_deleted(const struct path *path,
3040 const struct path *root,
3041 char **buf, int *buflen)
3043 prepend(buf, buflen, "\0", 1);
3044 if (d_unlinked(path->dentry)) {
3045 int error = prepend(buf, buflen, " (deleted)", 10);
3050 return prepend_path(path, root, buf, buflen);
3053 static int prepend_unreachable(char **buffer, int *buflen)
3055 return prepend(buffer, buflen, "(unreachable)", 13);
3058 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3063 seq = read_seqcount_begin(&fs->seq);
3065 } while (read_seqcount_retry(&fs->seq, seq));
3069 * d_path - return the path of a dentry
3070 * @path: path to report
3071 * @buf: buffer to return value in
3072 * @buflen: buffer length
3074 * Convert a dentry into an ASCII path name. If the entry has been deleted
3075 * the string " (deleted)" is appended. Note that this is ambiguous.
3077 * Returns a pointer into the buffer or an error code if the path was
3078 * too long. Note: Callers should use the returned pointer, not the passed
3079 * in buffer, to use the name! The implementation often starts at an offset
3080 * into the buffer, and may leave 0 bytes at the start.
3082 * "buflen" should be positive.
3084 char *d_path(const struct path *path, char *buf, int buflen)
3086 char *res = buf + buflen;
3091 * We have various synthetic filesystems that never get mounted. On
3092 * these filesystems dentries are never used for lookup purposes, and
3093 * thus don't need to be hashed. They also don't need a name until a
3094 * user wants to identify the object in /proc/pid/fd/. The little hack
3095 * below allows us to generate a name for these objects on demand:
3097 * Some pseudo inodes are mountable. When they are mounted
3098 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3099 * and instead have d_path return the mounted path.
3101 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3102 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3103 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3106 get_fs_root_rcu(current->fs, &root);
3107 error = path_with_deleted(path, &root, &res, &buflen);
3111 res = ERR_PTR(error);
3114 EXPORT_SYMBOL(d_path);
3117 * Helper function for dentry_operations.d_dname() members
3119 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3120 const char *fmt, ...)
3126 va_start(args, fmt);
3127 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3130 if (sz > sizeof(temp) || sz > buflen)
3131 return ERR_PTR(-ENAMETOOLONG);
3133 buffer += buflen - sz;
3134 return memcpy(buffer, temp, sz);
3137 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3139 char *end = buffer + buflen;
3140 /* these dentries are never renamed, so d_lock is not needed */
3141 if (prepend(&end, &buflen, " (deleted)", 11) ||
3142 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3143 prepend(&end, &buflen, "/", 1))
3144 end = ERR_PTR(-ENAMETOOLONG);
3147 EXPORT_SYMBOL(simple_dname);
3150 * Write full pathname from the root of the filesystem into the buffer.
3152 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3154 struct dentry *dentry;
3167 prepend(&end, &len, "\0", 1);
3171 read_seqbegin_or_lock(&rename_lock, &seq);
3172 while (!IS_ROOT(dentry)) {
3173 struct dentry *parent = dentry->d_parent;
3176 error = prepend_name(&end, &len, &dentry->d_name);
3185 if (need_seqretry(&rename_lock, seq)) {
3189 done_seqretry(&rename_lock, seq);
3194 return ERR_PTR(-ENAMETOOLONG);
3197 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3199 return __dentry_path(dentry, buf, buflen);
3201 EXPORT_SYMBOL(dentry_path_raw);
3203 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3208 if (d_unlinked(dentry)) {
3210 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3214 retval = __dentry_path(dentry, buf, buflen);
3215 if (!IS_ERR(retval) && p)
3216 *p = '/'; /* restore '/' overriden with '\0' */
3219 return ERR_PTR(-ENAMETOOLONG);
3222 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3228 seq = read_seqcount_begin(&fs->seq);
3231 } while (read_seqcount_retry(&fs->seq, seq));
3235 * NOTE! The user-level library version returns a
3236 * character pointer. The kernel system call just
3237 * returns the length of the buffer filled (which
3238 * includes the ending '\0' character), or a negative
3239 * error value. So libc would do something like
3241 * char *getcwd(char * buf, size_t size)
3245 * retval = sys_getcwd(buf, size);
3252 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3255 struct path pwd, root;
3256 char *page = __getname();
3262 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3265 if (!d_unlinked(pwd.dentry)) {
3267 char *cwd = page + PATH_MAX;
3268 int buflen = PATH_MAX;
3270 prepend(&cwd, &buflen, "\0", 1);
3271 error = prepend_path(&pwd, &root, &cwd, &buflen);
3277 /* Unreachable from current root */
3279 error = prepend_unreachable(&cwd, &buflen);
3285 len = PATH_MAX + page - cwd;
3288 if (copy_to_user(buf, cwd, len))
3301 * Test whether new_dentry is a subdirectory of old_dentry.
3303 * Trivially implemented using the dcache structure
3307 * is_subdir - is new dentry a subdirectory of old_dentry
3308 * @new_dentry: new dentry
3309 * @old_dentry: old dentry
3311 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3312 * Returns 0 otherwise.
3313 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3316 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3321 if (new_dentry == old_dentry)
3325 /* for restarting inner loop in case of seq retry */
3326 seq = read_seqbegin(&rename_lock);
3328 * Need rcu_readlock to protect against the d_parent trashing
3332 if (d_ancestor(old_dentry, new_dentry))
3337 } while (read_seqretry(&rename_lock, seq));
3342 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3344 struct dentry *root = data;
3345 if (dentry != root) {
3346 if (d_unhashed(dentry) || !dentry->d_inode)
3349 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3350 dentry->d_flags |= DCACHE_GENOCIDE;
3351 dentry->d_lockref.count--;
3354 return D_WALK_CONTINUE;
3357 void d_genocide(struct dentry *parent)
3359 d_walk(parent, parent, d_genocide_kill, NULL);
3362 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3364 inode_dec_link_count(inode);
3365 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3366 !hlist_unhashed(&dentry->d_alias) ||
3367 !d_unlinked(dentry));
3368 spin_lock(&dentry->d_parent->d_lock);
3369 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3370 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3371 (unsigned long long)inode->i_ino);
3372 spin_unlock(&dentry->d_lock);
3373 spin_unlock(&dentry->d_parent->d_lock);
3374 d_instantiate(dentry, inode);
3376 EXPORT_SYMBOL(d_tmpfile);
3378 static __initdata unsigned long dhash_entries;
3379 static int __init set_dhash_entries(char *str)
3383 dhash_entries = simple_strtoul(str, &str, 0);
3386 __setup("dhash_entries=", set_dhash_entries);
3388 static void __init dcache_init_early(void)
3392 /* If hashes are distributed across NUMA nodes, defer
3393 * hash allocation until vmalloc space is available.
3399 alloc_large_system_hash("Dentry cache",
3400 sizeof(struct hlist_bl_head),
3409 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3410 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3413 static void __init dcache_init(void)
3418 * A constructor could be added for stable state like the lists,
3419 * but it is probably not worth it because of the cache nature
3422 dentry_cache = KMEM_CACHE(dentry,
3423 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3425 /* Hash may have been set up in dcache_init_early */
3430 alloc_large_system_hash("Dentry cache",
3431 sizeof(struct hlist_bl_head),
3440 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3441 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3444 /* SLAB cache for __getname() consumers */
3445 struct kmem_cache *names_cachep __read_mostly;
3446 EXPORT_SYMBOL(names_cachep);
3448 EXPORT_SYMBOL(d_genocide);
3450 void __init vfs_caches_init_early(void)
3452 dcache_init_early();
3456 void __init vfs_caches_init(unsigned long mempages)
3458 unsigned long reserve;
3460 /* Base hash sizes on available memory, with a reserve equal to
3461 150% of current kernel size */
3463 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3464 mempages -= reserve;
3466 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3467 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3471 files_init(mempages);