4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
46 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
48 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
52 * dentry->d_sb->s_dentry_lru_lock protects:
53 * - the dcache lru lists and counters
60 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
65 * dentry->d_inode->i_lock
67 * dentry->d_sb->s_dentry_lru_lock
68 * dcache_hash_bucket lock
71 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
74 * dentry->d_parent->d_lock
77 * If no ancestor relationship:
78 * if (dentry1 < dentry2)
82 int sysctl_vfs_cache_pressure __read_mostly = 100;
83 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
87 EXPORT_SYMBOL(rename_lock);
89 static struct kmem_cache *dentry_cache __read_mostly;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
100 static unsigned int d_hash_mask __read_mostly;
101 static unsigned int d_hash_shift __read_mostly;
103 static struct hlist_bl_head *dentry_hashtable __read_mostly;
105 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
108 hash += (unsigned long) parent / L1_CACHE_BYTES;
109 hash = hash + (hash >> d_hash_shift);
110 return dentry_hashtable + (hash & d_hash_mask);
113 /* Statistics gathering. */
114 struct dentry_stat_t dentry_stat = {
118 static DEFINE_PER_CPU(long, nr_dentry);
119 static DEFINE_PER_CPU(long, nr_dentry_unused);
121 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
124 * Here we resort to our own counters instead of using generic per-cpu counters
125 * for consistency with what the vfs inode code does. We are expected to harvest
126 * better code and performance by having our own specialized counters.
128 * Please note that the loop is done over all possible CPUs, not over all online
129 * CPUs. The reason for this is that we don't want to play games with CPUs going
130 * on and off. If one of them goes off, we will just keep their counters.
132 * glommer: See cffbc8a for details, and if you ever intend to change this,
133 * please update all vfs counters to match.
135 static long get_nr_dentry(void)
139 for_each_possible_cpu(i)
140 sum += per_cpu(nr_dentry, i);
141 return sum < 0 ? 0 : sum;
144 static long get_nr_dentry_unused(void)
148 for_each_possible_cpu(i)
149 sum += per_cpu(nr_dentry_unused, i);
150 return sum < 0 ? 0 : sum;
153 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
154 size_t *lenp, loff_t *ppos)
156 dentry_stat.nr_dentry = get_nr_dentry();
157 dentry_stat.nr_unused = get_nr_dentry_unused();
158 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
163 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
164 * The strings are both count bytes long, and count is non-zero.
166 #ifdef CONFIG_DCACHE_WORD_ACCESS
168 #include <asm/word-at-a-time.h>
170 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
171 * aligned allocation for this particular component. We don't
172 * strictly need the load_unaligned_zeropad() safety, but it
173 * doesn't hurt either.
175 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
176 * need the careful unaligned handling.
178 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
180 unsigned long a,b,mask;
183 a = *(unsigned long *)cs;
184 b = load_unaligned_zeropad(ct);
185 if (tcount < sizeof(unsigned long))
187 if (unlikely(a != b))
189 cs += sizeof(unsigned long);
190 ct += sizeof(unsigned long);
191 tcount -= sizeof(unsigned long);
195 mask = bytemask_from_count(tcount);
196 return unlikely(!!((a ^ b) & mask));
201 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
215 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
217 const unsigned char *cs;
219 * Be careful about RCU walk racing with rename:
220 * use ACCESS_ONCE to fetch the name pointer.
222 * NOTE! Even if a rename will mean that the length
223 * was not loaded atomically, we don't care. The
224 * RCU walk will check the sequence count eventually,
225 * and catch it. And we won't overrun the buffer,
226 * because we're reading the name pointer atomically,
227 * and a dentry name is guaranteed to be properly
228 * terminated with a NUL byte.
230 * End result: even if 'len' is wrong, we'll exit
231 * early because the data cannot match (there can
232 * be no NUL in the ct/tcount data)
234 cs = ACCESS_ONCE(dentry->d_name.name);
235 smp_read_barrier_depends();
236 return dentry_string_cmp(cs, ct, tcount);
239 static void __d_free(struct rcu_head *head)
241 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
243 WARN_ON(!hlist_unhashed(&dentry->d_alias));
244 if (dname_external(dentry))
245 kfree(dentry->d_name.name);
246 kmem_cache_free(dentry_cache, dentry);
249 static void dentry_free(struct dentry *dentry)
251 /* if dentry was never visible to RCU, immediate free is OK */
252 if (!(dentry->d_flags & DCACHE_RCUACCESS))
253 __d_free(&dentry->d_u.d_rcu);
255 call_rcu(&dentry->d_u.d_rcu, __d_free);
259 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
260 * @dentry: the target dentry
261 * After this call, in-progress rcu-walk path lookup will fail. This
262 * should be called after unhashing, and after changing d_inode (if
263 * the dentry has not already been unhashed).
265 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
267 assert_spin_locked(&dentry->d_lock);
268 /* Go through a barrier */
269 write_seqcount_barrier(&dentry->d_seq);
273 * Release the dentry's inode, using the filesystem
274 * d_iput() operation if defined. Dentry has no refcount
277 static void dentry_iput(struct dentry * dentry)
278 __releases(dentry->d_lock)
279 __releases(dentry->d_inode->i_lock)
281 struct inode *inode = dentry->d_inode;
283 dentry->d_inode = NULL;
284 hlist_del_init(&dentry->d_alias);
285 spin_unlock(&dentry->d_lock);
286 spin_unlock(&inode->i_lock);
288 fsnotify_inoderemove(inode);
289 if (dentry->d_op && dentry->d_op->d_iput)
290 dentry->d_op->d_iput(dentry, inode);
294 spin_unlock(&dentry->d_lock);
299 * Release the dentry's inode, using the filesystem
300 * d_iput() operation if defined. dentry remains in-use.
302 static void dentry_unlink_inode(struct dentry * dentry)
303 __releases(dentry->d_lock)
304 __releases(dentry->d_inode->i_lock)
306 struct inode *inode = dentry->d_inode;
307 __d_clear_type(dentry);
308 dentry->d_inode = NULL;
309 hlist_del_init(&dentry->d_alias);
310 dentry_rcuwalk_barrier(dentry);
311 spin_unlock(&dentry->d_lock);
312 spin_unlock(&inode->i_lock);
314 fsnotify_inoderemove(inode);
315 if (dentry->d_op && dentry->d_op->d_iput)
316 dentry->d_op->d_iput(dentry, inode);
322 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
323 * is in use - which includes both the "real" per-superblock
324 * LRU list _and_ the DCACHE_SHRINK_LIST use.
326 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
327 * on the shrink list (ie not on the superblock LRU list).
329 * The per-cpu "nr_dentry_unused" counters are updated with
330 * the DCACHE_LRU_LIST bit.
332 * These helper functions make sure we always follow the
333 * rules. d_lock must be held by the caller.
335 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
336 static void d_lru_add(struct dentry *dentry)
338 D_FLAG_VERIFY(dentry, 0);
339 dentry->d_flags |= DCACHE_LRU_LIST;
340 this_cpu_inc(nr_dentry_unused);
341 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
344 static void d_lru_del(struct dentry *dentry)
346 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
347 dentry->d_flags &= ~DCACHE_LRU_LIST;
348 this_cpu_dec(nr_dentry_unused);
349 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
352 static void d_shrink_del(struct dentry *dentry)
354 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
355 list_del_init(&dentry->d_lru);
356 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
357 this_cpu_dec(nr_dentry_unused);
360 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
362 D_FLAG_VERIFY(dentry, 0);
363 list_add(&dentry->d_lru, list);
364 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
365 this_cpu_inc(nr_dentry_unused);
369 * These can only be called under the global LRU lock, ie during the
370 * callback for freeing the LRU list. "isolate" removes it from the
371 * LRU lists entirely, while shrink_move moves it to the indicated
374 static void d_lru_isolate(struct dentry *dentry)
376 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
377 dentry->d_flags &= ~DCACHE_LRU_LIST;
378 this_cpu_dec(nr_dentry_unused);
379 list_del_init(&dentry->d_lru);
382 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
384 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
385 dentry->d_flags |= DCACHE_SHRINK_LIST;
386 list_move_tail(&dentry->d_lru, list);
390 * dentry_lru_(add|del)_list) must be called with d_lock held.
392 static void dentry_lru_add(struct dentry *dentry)
394 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
399 * d_drop - drop a dentry
400 * @dentry: dentry to drop
402 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
403 * be found through a VFS lookup any more. Note that this is different from
404 * deleting the dentry - d_delete will try to mark the dentry negative if
405 * possible, giving a successful _negative_ lookup, while d_drop will
406 * just make the cache lookup fail.
408 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
409 * reason (NFS timeouts or autofs deletes).
411 * __d_drop requires dentry->d_lock.
413 void __d_drop(struct dentry *dentry)
415 if (!d_unhashed(dentry)) {
416 struct hlist_bl_head *b;
418 * Hashed dentries are normally on the dentry hashtable,
419 * with the exception of those newly allocated by
420 * d_obtain_alias, which are always IS_ROOT:
422 if (unlikely(IS_ROOT(dentry)))
423 b = &dentry->d_sb->s_anon;
425 b = d_hash(dentry->d_parent, dentry->d_name.hash);
428 __hlist_bl_del(&dentry->d_hash);
429 dentry->d_hash.pprev = NULL;
431 dentry_rcuwalk_barrier(dentry);
434 EXPORT_SYMBOL(__d_drop);
436 void d_drop(struct dentry *dentry)
438 spin_lock(&dentry->d_lock);
440 spin_unlock(&dentry->d_lock);
442 EXPORT_SYMBOL(d_drop);
444 static void __dentry_kill(struct dentry *dentry)
446 struct dentry *parent = NULL;
447 bool can_free = true;
448 if (!IS_ROOT(dentry))
449 parent = dentry->d_parent;
452 * The dentry is now unrecoverably dead to the world.
454 lockref_mark_dead(&dentry->d_lockref);
457 * inform the fs via d_prune that this dentry is about to be
458 * unhashed and destroyed.
460 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
461 dentry->d_op->d_prune(dentry);
463 if (dentry->d_flags & DCACHE_LRU_LIST) {
464 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
467 /* if it was on the hash then remove it */
469 list_del(&dentry->d_u.d_child);
471 * Inform d_walk() that we are no longer attached to the
474 dentry->d_flags |= DCACHE_DENTRY_KILLED;
476 spin_unlock(&parent->d_lock);
479 * dentry_iput drops the locks, at which point nobody (except
480 * transient RCU lookups) can reach this dentry.
482 BUG_ON((int)dentry->d_lockref.count > 0);
483 this_cpu_dec(nr_dentry);
484 if (dentry->d_op && dentry->d_op->d_release)
485 dentry->d_op->d_release(dentry);
487 spin_lock(&dentry->d_lock);
488 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
489 dentry->d_flags |= DCACHE_MAY_FREE;
492 spin_unlock(&dentry->d_lock);
493 if (likely(can_free))
498 * Finish off a dentry we've decided to kill.
499 * dentry->d_lock must be held, returns with it unlocked.
500 * If ref is non-zero, then decrement the refcount too.
501 * Returns dentry requiring refcount drop, or NULL if we're done.
503 static struct dentry *dentry_kill(struct dentry *dentry)
504 __releases(dentry->d_lock)
506 struct inode *inode = dentry->d_inode;
507 struct dentry *parent = NULL;
509 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
512 if (!IS_ROOT(dentry)) {
513 parent = dentry->d_parent;
514 if (unlikely(!spin_trylock(&parent->d_lock))) {
516 spin_unlock(&inode->i_lock);
521 __dentry_kill(dentry);
525 spin_unlock(&dentry->d_lock);
527 return dentry; /* try again with same dentry */
530 static inline struct dentry *lock_parent(struct dentry *dentry)
532 struct dentry *parent = dentry->d_parent;
535 if (unlikely((int)dentry->d_lockref.count < 0))
537 if (likely(spin_trylock(&parent->d_lock)))
540 spin_unlock(&dentry->d_lock);
542 parent = ACCESS_ONCE(dentry->d_parent);
543 spin_lock(&parent->d_lock);
545 * We can't blindly lock dentry until we are sure
546 * that we won't violate the locking order.
547 * Any changes of dentry->d_parent must have
548 * been done with parent->d_lock held, so
549 * spin_lock() above is enough of a barrier
550 * for checking if it's still our child.
552 if (unlikely(parent != dentry->d_parent)) {
553 spin_unlock(&parent->d_lock);
557 if (parent != dentry)
558 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
567 * This is complicated by the fact that we do not want to put
568 * dentries that are no longer on any hash chain on the unused
569 * list: we'd much rather just get rid of them immediately.
571 * However, that implies that we have to traverse the dentry
572 * tree upwards to the parents which might _also_ now be
573 * scheduled for deletion (it may have been only waiting for
574 * its last child to go away).
576 * This tail recursion is done by hand as we don't want to depend
577 * on the compiler to always get this right (gcc generally doesn't).
578 * Real recursion would eat up our stack space.
582 * dput - release a dentry
583 * @dentry: dentry to release
585 * Release a dentry. This will drop the usage count and if appropriate
586 * call the dentry unlink method as well as removing it from the queues and
587 * releasing its resources. If the parent dentries were scheduled for release
588 * they too may now get deleted.
590 void dput(struct dentry *dentry)
592 if (unlikely(!dentry))
596 if (lockref_put_or_lock(&dentry->d_lockref))
599 /* Unreachable? Get rid of it */
600 if (unlikely(d_unhashed(dentry)))
603 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
604 if (dentry->d_op->d_delete(dentry))
608 if (!(dentry->d_flags & DCACHE_REFERENCED))
609 dentry->d_flags |= DCACHE_REFERENCED;
610 dentry_lru_add(dentry);
612 dentry->d_lockref.count--;
613 spin_unlock(&dentry->d_lock);
617 dentry = dentry_kill(dentry);
624 * d_invalidate - invalidate a dentry
625 * @dentry: dentry to invalidate
627 * Try to invalidate the dentry if it turns out to be
628 * possible. If there are other dentries that can be
629 * reached through this one we can't delete it and we
630 * return -EBUSY. On success we return 0.
635 int d_invalidate(struct dentry * dentry)
638 * If it's already been dropped, return OK.
640 spin_lock(&dentry->d_lock);
641 if (d_unhashed(dentry)) {
642 spin_unlock(&dentry->d_lock);
646 * Check whether to do a partial shrink_dcache
647 * to get rid of unused child entries.
649 if (!list_empty(&dentry->d_subdirs)) {
650 spin_unlock(&dentry->d_lock);
651 shrink_dcache_parent(dentry);
652 spin_lock(&dentry->d_lock);
656 * Somebody else still using it?
658 * If it's a directory, we can't drop it
659 * for fear of somebody re-populating it
660 * with children (even though dropping it
661 * would make it unreachable from the root,
662 * we might still populate it if it was a
663 * working directory or similar).
664 * We also need to leave mountpoints alone,
667 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
668 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
669 spin_unlock(&dentry->d_lock);
675 spin_unlock(&dentry->d_lock);
678 EXPORT_SYMBOL(d_invalidate);
680 /* This must be called with d_lock held */
681 static inline void __dget_dlock(struct dentry *dentry)
683 dentry->d_lockref.count++;
686 static inline void __dget(struct dentry *dentry)
688 lockref_get(&dentry->d_lockref);
691 struct dentry *dget_parent(struct dentry *dentry)
697 * Do optimistic parent lookup without any
701 ret = ACCESS_ONCE(dentry->d_parent);
702 gotref = lockref_get_not_zero(&ret->d_lockref);
704 if (likely(gotref)) {
705 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
712 * Don't need rcu_dereference because we re-check it was correct under
716 ret = dentry->d_parent;
717 spin_lock(&ret->d_lock);
718 if (unlikely(ret != dentry->d_parent)) {
719 spin_unlock(&ret->d_lock);
724 BUG_ON(!ret->d_lockref.count);
725 ret->d_lockref.count++;
726 spin_unlock(&ret->d_lock);
729 EXPORT_SYMBOL(dget_parent);
732 * d_find_alias - grab a hashed alias of inode
733 * @inode: inode in question
735 * If inode has a hashed alias, or is a directory and has any alias,
736 * acquire the reference to alias and return it. Otherwise return NULL.
737 * Notice that if inode is a directory there can be only one alias and
738 * it can be unhashed only if it has no children, or if it is the root
741 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
742 * any other hashed alias over that one.
744 static struct dentry *__d_find_alias(struct inode *inode)
746 struct dentry *alias, *discon_alias;
750 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
751 spin_lock(&alias->d_lock);
752 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
753 if (IS_ROOT(alias) &&
754 (alias->d_flags & DCACHE_DISCONNECTED)) {
755 discon_alias = alias;
758 spin_unlock(&alias->d_lock);
762 spin_unlock(&alias->d_lock);
765 alias = discon_alias;
766 spin_lock(&alias->d_lock);
767 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
768 if (IS_ROOT(alias) &&
769 (alias->d_flags & DCACHE_DISCONNECTED)) {
771 spin_unlock(&alias->d_lock);
775 spin_unlock(&alias->d_lock);
781 struct dentry *d_find_alias(struct inode *inode)
783 struct dentry *de = NULL;
785 if (!hlist_empty(&inode->i_dentry)) {
786 spin_lock(&inode->i_lock);
787 de = __d_find_alias(inode);
788 spin_unlock(&inode->i_lock);
792 EXPORT_SYMBOL(d_find_alias);
795 * Try to kill dentries associated with this inode.
796 * WARNING: you must own a reference to inode.
798 void d_prune_aliases(struct inode *inode)
800 struct dentry *dentry;
802 spin_lock(&inode->i_lock);
803 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
804 spin_lock(&dentry->d_lock);
805 if (!dentry->d_lockref.count) {
807 * inform the fs via d_prune that this dentry
808 * is about to be unhashed and destroyed.
810 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
812 dentry->d_op->d_prune(dentry);
814 __dget_dlock(dentry);
816 spin_unlock(&dentry->d_lock);
817 spin_unlock(&inode->i_lock);
821 spin_unlock(&dentry->d_lock);
823 spin_unlock(&inode->i_lock);
825 EXPORT_SYMBOL(d_prune_aliases);
827 static void shrink_dentry_list(struct list_head *list)
829 struct dentry *dentry, *parent;
831 while (!list_empty(list)) {
833 dentry = list_entry(list->prev, struct dentry, d_lru);
834 spin_lock(&dentry->d_lock);
835 parent = lock_parent(dentry);
838 * The dispose list is isolated and dentries are not accounted
839 * to the LRU here, so we can simply remove it from the list
840 * here regardless of whether it is referenced or not.
842 d_shrink_del(dentry);
845 * We found an inuse dentry which was not removed from
846 * the LRU because of laziness during lookup. Do not free it.
848 if ((int)dentry->d_lockref.count > 0) {
849 spin_unlock(&dentry->d_lock);
851 spin_unlock(&parent->d_lock);
856 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
857 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
858 spin_unlock(&dentry->d_lock);
860 spin_unlock(&parent->d_lock);
866 inode = dentry->d_inode;
867 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
868 d_shrink_add(dentry, list);
869 spin_unlock(&dentry->d_lock);
871 spin_unlock(&parent->d_lock);
875 __dentry_kill(dentry);
878 * We need to prune ancestors too. This is necessary to prevent
879 * quadratic behavior of shrink_dcache_parent(), but is also
880 * expected to be beneficial in reducing dentry cache
884 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
885 parent = lock_parent(dentry);
886 if (dentry->d_lockref.count != 1) {
887 dentry->d_lockref.count--;
888 spin_unlock(&dentry->d_lock);
890 spin_unlock(&parent->d_lock);
893 inode = dentry->d_inode; /* can't be NULL */
894 if (unlikely(!spin_trylock(&inode->i_lock))) {
895 spin_unlock(&dentry->d_lock);
897 spin_unlock(&parent->d_lock);
901 __dentry_kill(dentry);
907 static enum lru_status
908 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
910 struct list_head *freeable = arg;
911 struct dentry *dentry = container_of(item, struct dentry, d_lru);
915 * we are inverting the lru lock/dentry->d_lock here,
916 * so use a trylock. If we fail to get the lock, just skip
919 if (!spin_trylock(&dentry->d_lock))
923 * Referenced dentries are still in use. If they have active
924 * counts, just remove them from the LRU. Otherwise give them
925 * another pass through the LRU.
927 if (dentry->d_lockref.count) {
928 d_lru_isolate(dentry);
929 spin_unlock(&dentry->d_lock);
933 if (dentry->d_flags & DCACHE_REFERENCED) {
934 dentry->d_flags &= ~DCACHE_REFERENCED;
935 spin_unlock(&dentry->d_lock);
938 * The list move itself will be made by the common LRU code. At
939 * this point, we've dropped the dentry->d_lock but keep the
940 * lru lock. This is safe to do, since every list movement is
941 * protected by the lru lock even if both locks are held.
943 * This is guaranteed by the fact that all LRU management
944 * functions are intermediated by the LRU API calls like
945 * list_lru_add and list_lru_del. List movement in this file
946 * only ever occur through this functions or through callbacks
947 * like this one, that are called from the LRU API.
949 * The only exceptions to this are functions like
950 * shrink_dentry_list, and code that first checks for the
951 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
952 * operating only with stack provided lists after they are
953 * properly isolated from the main list. It is thus, always a
959 d_lru_shrink_move(dentry, freeable);
960 spin_unlock(&dentry->d_lock);
966 * prune_dcache_sb - shrink the dcache
968 * @nr_to_scan : number of entries to try to free
969 * @nid: which node to scan for freeable entities
971 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
972 * done when we need more memory an called from the superblock shrinker
975 * This function may fail to free any resources if all the dentries are in
978 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
984 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
985 &dispose, &nr_to_scan);
986 shrink_dentry_list(&dispose);
990 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
991 spinlock_t *lru_lock, void *arg)
993 struct list_head *freeable = arg;
994 struct dentry *dentry = container_of(item, struct dentry, d_lru);
997 * we are inverting the lru lock/dentry->d_lock here,
998 * so use a trylock. If we fail to get the lock, just skip
1001 if (!spin_trylock(&dentry->d_lock))
1004 d_lru_shrink_move(dentry, freeable);
1005 spin_unlock(&dentry->d_lock);
1012 * shrink_dcache_sb - shrink dcache for a superblock
1015 * Shrink the dcache for the specified super block. This is used to free
1016 * the dcache before unmounting a file system.
1018 void shrink_dcache_sb(struct super_block *sb)
1025 freed = list_lru_walk(&sb->s_dentry_lru,
1026 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1028 this_cpu_sub(nr_dentry_unused, freed);
1029 shrink_dentry_list(&dispose);
1030 } while (freed > 0);
1032 EXPORT_SYMBOL(shrink_dcache_sb);
1035 * enum d_walk_ret - action to talke during tree walk
1036 * @D_WALK_CONTINUE: contrinue walk
1037 * @D_WALK_QUIT: quit walk
1038 * @D_WALK_NORETRY: quit when retry is needed
1039 * @D_WALK_SKIP: skip this dentry and its children
1049 * d_walk - walk the dentry tree
1050 * @parent: start of walk
1051 * @data: data passed to @enter() and @finish()
1052 * @enter: callback when first entering the dentry
1053 * @finish: callback when successfully finished the walk
1055 * The @enter() and @finish() callbacks are called with d_lock held.
1057 static void d_walk(struct dentry *parent, void *data,
1058 enum d_walk_ret (*enter)(void *, struct dentry *),
1059 void (*finish)(void *))
1061 struct dentry *this_parent;
1062 struct list_head *next;
1064 enum d_walk_ret ret;
1068 read_seqbegin_or_lock(&rename_lock, &seq);
1069 this_parent = parent;
1070 spin_lock(&this_parent->d_lock);
1072 ret = enter(data, this_parent);
1074 case D_WALK_CONTINUE:
1079 case D_WALK_NORETRY:
1084 next = this_parent->d_subdirs.next;
1086 while (next != &this_parent->d_subdirs) {
1087 struct list_head *tmp = next;
1088 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1091 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1093 ret = enter(data, dentry);
1095 case D_WALK_CONTINUE:
1098 spin_unlock(&dentry->d_lock);
1100 case D_WALK_NORETRY:
1104 spin_unlock(&dentry->d_lock);
1108 if (!list_empty(&dentry->d_subdirs)) {
1109 spin_unlock(&this_parent->d_lock);
1110 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1111 this_parent = dentry;
1112 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1115 spin_unlock(&dentry->d_lock);
1118 * All done at this level ... ascend and resume the search.
1120 if (this_parent != parent) {
1121 struct dentry *child = this_parent;
1122 this_parent = child->d_parent;
1125 spin_unlock(&child->d_lock);
1126 spin_lock(&this_parent->d_lock);
1129 * might go back up the wrong parent if we have had a rename
1132 if (this_parent != child->d_parent ||
1133 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1134 need_seqretry(&rename_lock, seq)) {
1135 spin_unlock(&this_parent->d_lock);
1140 next = child->d_u.d_child.next;
1143 if (need_seqretry(&rename_lock, seq)) {
1144 spin_unlock(&this_parent->d_lock);
1151 spin_unlock(&this_parent->d_lock);
1152 done_seqretry(&rename_lock, seq);
1163 * Search for at least 1 mount point in the dentry's subdirs.
1164 * We descend to the next level whenever the d_subdirs
1165 * list is non-empty and continue searching.
1168 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1171 if (d_mountpoint(dentry)) {
1175 return D_WALK_CONTINUE;
1179 * have_submounts - check for mounts over a dentry
1180 * @parent: dentry to check.
1182 * Return true if the parent or its subdirectories contain
1185 int have_submounts(struct dentry *parent)
1189 d_walk(parent, &ret, check_mount, NULL);
1193 EXPORT_SYMBOL(have_submounts);
1196 * Called by mount code to set a mountpoint and check if the mountpoint is
1197 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1198 * subtree can become unreachable).
1200 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1201 * this reason take rename_lock and d_lock on dentry and ancestors.
1203 int d_set_mounted(struct dentry *dentry)
1207 write_seqlock(&rename_lock);
1208 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1209 /* Need exclusion wrt. check_submounts_and_drop() */
1210 spin_lock(&p->d_lock);
1211 if (unlikely(d_unhashed(p))) {
1212 spin_unlock(&p->d_lock);
1215 spin_unlock(&p->d_lock);
1217 spin_lock(&dentry->d_lock);
1218 if (!d_unlinked(dentry)) {
1219 dentry->d_flags |= DCACHE_MOUNTED;
1222 spin_unlock(&dentry->d_lock);
1224 write_sequnlock(&rename_lock);
1229 * Search the dentry child list of the specified parent,
1230 * and move any unused dentries to the end of the unused
1231 * list for prune_dcache(). We descend to the next level
1232 * whenever the d_subdirs list is non-empty and continue
1235 * It returns zero iff there are no unused children,
1236 * otherwise it returns the number of children moved to
1237 * the end of the unused list. This may not be the total
1238 * number of unused children, because select_parent can
1239 * drop the lock and return early due to latency
1243 struct select_data {
1244 struct dentry *start;
1245 struct list_head dispose;
1249 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1251 struct select_data *data = _data;
1252 enum d_walk_ret ret = D_WALK_CONTINUE;
1254 if (data->start == dentry)
1257 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1260 if (dentry->d_flags & DCACHE_LRU_LIST)
1262 if (!dentry->d_lockref.count) {
1263 d_shrink_add(dentry, &data->dispose);
1268 * We can return to the caller if we have found some (this
1269 * ensures forward progress). We'll be coming back to find
1272 if (!list_empty(&data->dispose))
1273 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1279 * shrink_dcache_parent - prune dcache
1280 * @parent: parent of entries to prune
1282 * Prune the dcache to remove unused children of the parent dentry.
1284 void shrink_dcache_parent(struct dentry *parent)
1287 struct select_data data;
1289 INIT_LIST_HEAD(&data.dispose);
1290 data.start = parent;
1293 d_walk(parent, &data, select_collect, NULL);
1297 shrink_dentry_list(&data.dispose);
1301 EXPORT_SYMBOL(shrink_dcache_parent);
1303 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1305 /* it has busy descendents; complain about those instead */
1306 if (!list_empty(&dentry->d_subdirs))
1307 return D_WALK_CONTINUE;
1309 /* root with refcount 1 is fine */
1310 if (dentry == _data && dentry->d_lockref.count == 1)
1311 return D_WALK_CONTINUE;
1313 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1314 " still in use (%d) [unmount of %s %s]\n",
1317 dentry->d_inode->i_ino : 0UL,
1319 dentry->d_lockref.count,
1320 dentry->d_sb->s_type->name,
1321 dentry->d_sb->s_id);
1323 return D_WALK_CONTINUE;
1326 static void do_one_tree(struct dentry *dentry)
1328 shrink_dcache_parent(dentry);
1329 d_walk(dentry, dentry, umount_check, NULL);
1335 * destroy the dentries attached to a superblock on unmounting
1337 void shrink_dcache_for_umount(struct super_block *sb)
1339 struct dentry *dentry;
1341 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1343 dentry = sb->s_root;
1345 do_one_tree(dentry);
1347 while (!hlist_bl_empty(&sb->s_anon)) {
1348 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1349 do_one_tree(dentry);
1353 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1355 struct select_data *data = _data;
1357 if (d_mountpoint(dentry)) {
1358 data->found = -EBUSY;
1362 return select_collect(_data, dentry);
1365 static void check_and_drop(void *_data)
1367 struct select_data *data = _data;
1369 if (d_mountpoint(data->start))
1370 data->found = -EBUSY;
1372 __d_drop(data->start);
1376 * check_submounts_and_drop - prune dcache, check for submounts and drop
1378 * All done as a single atomic operation relative to has_unlinked_ancestor().
1379 * Returns 0 if successfully unhashed @parent. If there were submounts then
1382 * @dentry: dentry to prune and drop
1384 int check_submounts_and_drop(struct dentry *dentry)
1388 /* Negative dentries can be dropped without further checks */
1389 if (!dentry->d_inode) {
1395 struct select_data data;
1397 INIT_LIST_HEAD(&data.dispose);
1398 data.start = dentry;
1401 d_walk(dentry, &data, check_and_collect, check_and_drop);
1404 if (!list_empty(&data.dispose))
1405 shrink_dentry_list(&data.dispose);
1416 EXPORT_SYMBOL(check_submounts_and_drop);
1419 * __d_alloc - allocate a dcache entry
1420 * @sb: filesystem it will belong to
1421 * @name: qstr of the name
1423 * Allocates a dentry. It returns %NULL if there is insufficient memory
1424 * available. On a success the dentry is returned. The name passed in is
1425 * copied and the copy passed in may be reused after this call.
1428 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1430 struct dentry *dentry;
1433 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1438 * We guarantee that the inline name is always NUL-terminated.
1439 * This way the memcpy() done by the name switching in rename
1440 * will still always have a NUL at the end, even if we might
1441 * be overwriting an internal NUL character
1443 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1444 if (name->len > DNAME_INLINE_LEN-1) {
1445 dname = kmalloc(name->len + 1, GFP_KERNEL);
1447 kmem_cache_free(dentry_cache, dentry);
1451 dname = dentry->d_iname;
1454 dentry->d_name.len = name->len;
1455 dentry->d_name.hash = name->hash;
1456 memcpy(dname, name->name, name->len);
1457 dname[name->len] = 0;
1459 /* Make sure we always see the terminating NUL character */
1461 dentry->d_name.name = dname;
1463 dentry->d_lockref.count = 1;
1464 dentry->d_flags = 0;
1465 spin_lock_init(&dentry->d_lock);
1466 seqcount_init(&dentry->d_seq);
1467 dentry->d_inode = NULL;
1468 dentry->d_parent = dentry;
1470 dentry->d_op = NULL;
1471 dentry->d_fsdata = NULL;
1472 INIT_HLIST_BL_NODE(&dentry->d_hash);
1473 INIT_LIST_HEAD(&dentry->d_lru);
1474 INIT_LIST_HEAD(&dentry->d_subdirs);
1475 INIT_HLIST_NODE(&dentry->d_alias);
1476 INIT_LIST_HEAD(&dentry->d_u.d_child);
1477 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1479 this_cpu_inc(nr_dentry);
1485 * d_alloc - allocate a dcache entry
1486 * @parent: parent of entry to allocate
1487 * @name: qstr of the name
1489 * Allocates a dentry. It returns %NULL if there is insufficient memory
1490 * available. On a success the dentry is returned. The name passed in is
1491 * copied and the copy passed in may be reused after this call.
1493 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1495 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1499 spin_lock(&parent->d_lock);
1501 * don't need child lock because it is not subject
1502 * to concurrency here
1504 __dget_dlock(parent);
1505 dentry->d_parent = parent;
1506 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1507 spin_unlock(&parent->d_lock);
1511 EXPORT_SYMBOL(d_alloc);
1514 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1515 * @sb: the superblock
1516 * @name: qstr of the name
1518 * For a filesystem that just pins its dentries in memory and never
1519 * performs lookups at all, return an unhashed IS_ROOT dentry.
1521 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1523 return __d_alloc(sb, name);
1525 EXPORT_SYMBOL(d_alloc_pseudo);
1527 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1532 q.len = strlen(name);
1533 q.hash = full_name_hash(q.name, q.len);
1534 return d_alloc(parent, &q);
1536 EXPORT_SYMBOL(d_alloc_name);
1538 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1540 WARN_ON_ONCE(dentry->d_op);
1541 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1543 DCACHE_OP_REVALIDATE |
1544 DCACHE_OP_WEAK_REVALIDATE |
1545 DCACHE_OP_DELETE ));
1550 dentry->d_flags |= DCACHE_OP_HASH;
1552 dentry->d_flags |= DCACHE_OP_COMPARE;
1553 if (op->d_revalidate)
1554 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1555 if (op->d_weak_revalidate)
1556 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1558 dentry->d_flags |= DCACHE_OP_DELETE;
1560 dentry->d_flags |= DCACHE_OP_PRUNE;
1563 EXPORT_SYMBOL(d_set_d_op);
1565 static unsigned d_flags_for_inode(struct inode *inode)
1567 unsigned add_flags = DCACHE_FILE_TYPE;
1570 return DCACHE_MISS_TYPE;
1572 if (S_ISDIR(inode->i_mode)) {
1573 add_flags = DCACHE_DIRECTORY_TYPE;
1574 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1575 if (unlikely(!inode->i_op->lookup))
1576 add_flags = DCACHE_AUTODIR_TYPE;
1578 inode->i_opflags |= IOP_LOOKUP;
1580 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1581 if (unlikely(inode->i_op->follow_link))
1582 add_flags = DCACHE_SYMLINK_TYPE;
1584 inode->i_opflags |= IOP_NOFOLLOW;
1587 if (unlikely(IS_AUTOMOUNT(inode)))
1588 add_flags |= DCACHE_NEED_AUTOMOUNT;
1592 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1594 unsigned add_flags = d_flags_for_inode(inode);
1596 spin_lock(&dentry->d_lock);
1597 __d_set_type(dentry, add_flags);
1599 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1600 dentry->d_inode = inode;
1601 dentry_rcuwalk_barrier(dentry);
1602 spin_unlock(&dentry->d_lock);
1603 fsnotify_d_instantiate(dentry, inode);
1607 * d_instantiate - fill in inode information for a dentry
1608 * @entry: dentry to complete
1609 * @inode: inode to attach to this dentry
1611 * Fill in inode information in the entry.
1613 * This turns negative dentries into productive full members
1616 * NOTE! This assumes that the inode count has been incremented
1617 * (or otherwise set) by the caller to indicate that it is now
1618 * in use by the dcache.
1621 void d_instantiate(struct dentry *entry, struct inode * inode)
1623 BUG_ON(!hlist_unhashed(&entry->d_alias));
1625 spin_lock(&inode->i_lock);
1626 __d_instantiate(entry, inode);
1628 spin_unlock(&inode->i_lock);
1629 security_d_instantiate(entry, inode);
1631 EXPORT_SYMBOL(d_instantiate);
1634 * d_instantiate_unique - instantiate a non-aliased dentry
1635 * @entry: dentry to instantiate
1636 * @inode: inode to attach to this dentry
1638 * Fill in inode information in the entry. On success, it returns NULL.
1639 * If an unhashed alias of "entry" already exists, then we return the
1640 * aliased dentry instead and drop one reference to inode.
1642 * Note that in order to avoid conflicts with rename() etc, the caller
1643 * had better be holding the parent directory semaphore.
1645 * This also assumes that the inode count has been incremented
1646 * (or otherwise set) by the caller to indicate that it is now
1647 * in use by the dcache.
1649 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1650 struct inode *inode)
1652 struct dentry *alias;
1653 int len = entry->d_name.len;
1654 const char *name = entry->d_name.name;
1655 unsigned int hash = entry->d_name.hash;
1658 __d_instantiate(entry, NULL);
1662 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1664 * Don't need alias->d_lock here, because aliases with
1665 * d_parent == entry->d_parent are not subject to name or
1666 * parent changes, because the parent inode i_mutex is held.
1668 if (alias->d_name.hash != hash)
1670 if (alias->d_parent != entry->d_parent)
1672 if (alias->d_name.len != len)
1674 if (dentry_cmp(alias, name, len))
1680 __d_instantiate(entry, inode);
1684 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1686 struct dentry *result;
1688 BUG_ON(!hlist_unhashed(&entry->d_alias));
1691 spin_lock(&inode->i_lock);
1692 result = __d_instantiate_unique(entry, inode);
1694 spin_unlock(&inode->i_lock);
1697 security_d_instantiate(entry, inode);
1701 BUG_ON(!d_unhashed(result));
1706 EXPORT_SYMBOL(d_instantiate_unique);
1709 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1710 * @entry: dentry to complete
1711 * @inode: inode to attach to this dentry
1713 * Fill in inode information in the entry. If a directory alias is found, then
1714 * return an error (and drop inode). Together with d_materialise_unique() this
1715 * guarantees that a directory inode may never have more than one alias.
1717 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1719 BUG_ON(!hlist_unhashed(&entry->d_alias));
1721 spin_lock(&inode->i_lock);
1722 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1723 spin_unlock(&inode->i_lock);
1727 __d_instantiate(entry, inode);
1728 spin_unlock(&inode->i_lock);
1729 security_d_instantiate(entry, inode);
1733 EXPORT_SYMBOL(d_instantiate_no_diralias);
1735 struct dentry *d_make_root(struct inode *root_inode)
1737 struct dentry *res = NULL;
1740 static const struct qstr name = QSTR_INIT("/", 1);
1742 res = __d_alloc(root_inode->i_sb, &name);
1744 d_instantiate(res, root_inode);
1750 EXPORT_SYMBOL(d_make_root);
1752 static struct dentry * __d_find_any_alias(struct inode *inode)
1754 struct dentry *alias;
1756 if (hlist_empty(&inode->i_dentry))
1758 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1764 * d_find_any_alias - find any alias for a given inode
1765 * @inode: inode to find an alias for
1767 * If any aliases exist for the given inode, take and return a
1768 * reference for one of them. If no aliases exist, return %NULL.
1770 struct dentry *d_find_any_alias(struct inode *inode)
1774 spin_lock(&inode->i_lock);
1775 de = __d_find_any_alias(inode);
1776 spin_unlock(&inode->i_lock);
1779 EXPORT_SYMBOL(d_find_any_alias);
1781 struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1783 static const struct qstr anonstring = QSTR_INIT("/", 1);
1789 return ERR_PTR(-ESTALE);
1791 return ERR_CAST(inode);
1793 res = d_find_any_alias(inode);
1797 tmp = __d_alloc(inode->i_sb, &anonstring);
1799 res = ERR_PTR(-ENOMEM);
1803 spin_lock(&inode->i_lock);
1804 res = __d_find_any_alias(inode);
1806 spin_unlock(&inode->i_lock);
1811 /* attach a disconnected dentry */
1812 add_flags = d_flags_for_inode(inode);
1815 add_flags |= DCACHE_DISCONNECTED;
1817 spin_lock(&tmp->d_lock);
1818 tmp->d_inode = inode;
1819 tmp->d_flags |= add_flags;
1820 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1821 hlist_bl_lock(&tmp->d_sb->s_anon);
1822 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1823 hlist_bl_unlock(&tmp->d_sb->s_anon);
1824 spin_unlock(&tmp->d_lock);
1825 spin_unlock(&inode->i_lock);
1826 security_d_instantiate(tmp, inode);
1831 if (res && !IS_ERR(res))
1832 security_d_instantiate(res, inode);
1838 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1839 * @inode: inode to allocate the dentry for
1841 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1842 * similar open by handle operations. The returned dentry may be anonymous,
1843 * or may have a full name (if the inode was already in the cache).
1845 * When called on a directory inode, we must ensure that the inode only ever
1846 * has one dentry. If a dentry is found, that is returned instead of
1847 * allocating a new one.
1849 * On successful return, the reference to the inode has been transferred
1850 * to the dentry. In case of an error the reference on the inode is released.
1851 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1852 * be passed in and the error will be propagated to the return value,
1853 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1855 struct dentry *d_obtain_alias(struct inode *inode)
1857 return __d_obtain_alias(inode, 1);
1859 EXPORT_SYMBOL(d_obtain_alias);
1862 * d_obtain_root - find or allocate a dentry for a given inode
1863 * @inode: inode to allocate the dentry for
1865 * Obtain an IS_ROOT dentry for the root of a filesystem.
1867 * We must ensure that directory inodes only ever have one dentry. If a
1868 * dentry is found, that is returned instead of allocating a new one.
1870 * On successful return, the reference to the inode has been transferred
1871 * to the dentry. In case of an error the reference on the inode is
1872 * released. A %NULL or IS_ERR inode may be passed in and will be the
1873 * error will be propagate to the return value, with a %NULL @inode
1874 * replaced by ERR_PTR(-ESTALE).
1876 struct dentry *d_obtain_root(struct inode *inode)
1878 return __d_obtain_alias(inode, 0);
1880 EXPORT_SYMBOL(d_obtain_root);
1883 * d_add_ci - lookup or allocate new dentry with case-exact name
1884 * @inode: the inode case-insensitive lookup has found
1885 * @dentry: the negative dentry that was passed to the parent's lookup func
1886 * @name: the case-exact name to be associated with the returned dentry
1888 * This is to avoid filling the dcache with case-insensitive names to the
1889 * same inode, only the actual correct case is stored in the dcache for
1890 * case-insensitive filesystems.
1892 * For a case-insensitive lookup match and if the the case-exact dentry
1893 * already exists in in the dcache, use it and return it.
1895 * If no entry exists with the exact case name, allocate new dentry with
1896 * the exact case, and return the spliced entry.
1898 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1901 struct dentry *found;
1905 * First check if a dentry matching the name already exists,
1906 * if not go ahead and create it now.
1908 found = d_hash_and_lookup(dentry->d_parent, name);
1909 if (unlikely(IS_ERR(found)))
1912 new = d_alloc(dentry->d_parent, name);
1914 found = ERR_PTR(-ENOMEM);
1918 found = d_splice_alias(inode, new);
1927 * If a matching dentry exists, and it's not negative use it.
1929 * Decrement the reference count to balance the iget() done
1932 if (found->d_inode) {
1933 if (unlikely(found->d_inode != inode)) {
1934 /* This can't happen because bad inodes are unhashed. */
1935 BUG_ON(!is_bad_inode(inode));
1936 BUG_ON(!is_bad_inode(found->d_inode));
1943 * Negative dentry: instantiate it unless the inode is a directory and
1944 * already has a dentry.
1946 new = d_splice_alias(inode, found);
1957 EXPORT_SYMBOL(d_add_ci);
1960 * Do the slow-case of the dentry name compare.
1962 * Unlike the dentry_cmp() function, we need to atomically
1963 * load the name and length information, so that the
1964 * filesystem can rely on them, and can use the 'name' and
1965 * 'len' information without worrying about walking off the
1966 * end of memory etc.
1968 * Thus the read_seqcount_retry() and the "duplicate" info
1969 * in arguments (the low-level filesystem should not look
1970 * at the dentry inode or name contents directly, since
1971 * rename can change them while we're in RCU mode).
1973 enum slow_d_compare {
1979 static noinline enum slow_d_compare slow_dentry_cmp(
1980 const struct dentry *parent,
1981 struct dentry *dentry,
1983 const struct qstr *name)
1985 int tlen = dentry->d_name.len;
1986 const char *tname = dentry->d_name.name;
1988 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1990 return D_COMP_SEQRETRY;
1992 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1993 return D_COMP_NOMATCH;
1998 * __d_lookup_rcu - search for a dentry (racy, store-free)
1999 * @parent: parent dentry
2000 * @name: qstr of name we wish to find
2001 * @seqp: returns d_seq value at the point where the dentry was found
2002 * Returns: dentry, or NULL
2004 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2005 * resolution (store-free path walking) design described in
2006 * Documentation/filesystems/path-lookup.txt.
2008 * This is not to be used outside core vfs.
2010 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2011 * held, and rcu_read_lock held. The returned dentry must not be stored into
2012 * without taking d_lock and checking d_seq sequence count against @seq
2015 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2018 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2019 * the returned dentry, so long as its parent's seqlock is checked after the
2020 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2021 * is formed, giving integrity down the path walk.
2023 * NOTE! The caller *has* to check the resulting dentry against the sequence
2024 * number we've returned before using any of the resulting dentry state!
2026 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2027 const struct qstr *name,
2030 u64 hashlen = name->hash_len;
2031 const unsigned char *str = name->name;
2032 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2033 struct hlist_bl_node *node;
2034 struct dentry *dentry;
2037 * Note: There is significant duplication with __d_lookup_rcu which is
2038 * required to prevent single threaded performance regressions
2039 * especially on architectures where smp_rmb (in seqcounts) are costly.
2040 * Keep the two functions in sync.
2044 * The hash list is protected using RCU.
2046 * Carefully use d_seq when comparing a candidate dentry, to avoid
2047 * races with d_move().
2049 * It is possible that concurrent renames can mess up our list
2050 * walk here and result in missing our dentry, resulting in the
2051 * false-negative result. d_lookup() protects against concurrent
2052 * renames using rename_lock seqlock.
2054 * See Documentation/filesystems/path-lookup.txt for more details.
2056 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2061 * The dentry sequence count protects us from concurrent
2062 * renames, and thus protects parent and name fields.
2064 * The caller must perform a seqcount check in order
2065 * to do anything useful with the returned dentry.
2067 * NOTE! We do a "raw" seqcount_begin here. That means that
2068 * we don't wait for the sequence count to stabilize if it
2069 * is in the middle of a sequence change. If we do the slow
2070 * dentry compare, we will do seqretries until it is stable,
2071 * and if we end up with a successful lookup, we actually
2072 * want to exit RCU lookup anyway.
2074 seq = raw_seqcount_begin(&dentry->d_seq);
2075 if (dentry->d_parent != parent)
2077 if (d_unhashed(dentry))
2080 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2081 if (dentry->d_name.hash != hashlen_hash(hashlen))
2084 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2087 case D_COMP_NOMATCH:
2094 if (dentry->d_name.hash_len != hashlen)
2097 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2104 * d_lookup - search for a dentry
2105 * @parent: parent dentry
2106 * @name: qstr of name we wish to find
2107 * Returns: dentry, or NULL
2109 * d_lookup searches the children of the parent dentry for the name in
2110 * question. If the dentry is found its reference count is incremented and the
2111 * dentry is returned. The caller must use dput to free the entry when it has
2112 * finished using it. %NULL is returned if the dentry does not exist.
2114 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2116 struct dentry *dentry;
2120 seq = read_seqbegin(&rename_lock);
2121 dentry = __d_lookup(parent, name);
2124 } while (read_seqretry(&rename_lock, seq));
2127 EXPORT_SYMBOL(d_lookup);
2130 * __d_lookup - search for a dentry (racy)
2131 * @parent: parent dentry
2132 * @name: qstr of name we wish to find
2133 * Returns: dentry, or NULL
2135 * __d_lookup is like d_lookup, however it may (rarely) return a
2136 * false-negative result due to unrelated rename activity.
2138 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2139 * however it must be used carefully, eg. with a following d_lookup in
2140 * the case of failure.
2142 * __d_lookup callers must be commented.
2144 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2146 unsigned int len = name->len;
2147 unsigned int hash = name->hash;
2148 const unsigned char *str = name->name;
2149 struct hlist_bl_head *b = d_hash(parent, hash);
2150 struct hlist_bl_node *node;
2151 struct dentry *found = NULL;
2152 struct dentry *dentry;
2155 * Note: There is significant duplication with __d_lookup_rcu which is
2156 * required to prevent single threaded performance regressions
2157 * especially on architectures where smp_rmb (in seqcounts) are costly.
2158 * Keep the two functions in sync.
2162 * The hash list is protected using RCU.
2164 * Take d_lock when comparing a candidate dentry, to avoid races
2167 * It is possible that concurrent renames can mess up our list
2168 * walk here and result in missing our dentry, resulting in the
2169 * false-negative result. d_lookup() protects against concurrent
2170 * renames using rename_lock seqlock.
2172 * See Documentation/filesystems/path-lookup.txt for more details.
2176 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2178 if (dentry->d_name.hash != hash)
2181 spin_lock(&dentry->d_lock);
2182 if (dentry->d_parent != parent)
2184 if (d_unhashed(dentry))
2188 * It is safe to compare names since d_move() cannot
2189 * change the qstr (protected by d_lock).
2191 if (parent->d_flags & DCACHE_OP_COMPARE) {
2192 int tlen = dentry->d_name.len;
2193 const char *tname = dentry->d_name.name;
2194 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2197 if (dentry->d_name.len != len)
2199 if (dentry_cmp(dentry, str, len))
2203 dentry->d_lockref.count++;
2205 spin_unlock(&dentry->d_lock);
2208 spin_unlock(&dentry->d_lock);
2216 * d_hash_and_lookup - hash the qstr then search for a dentry
2217 * @dir: Directory to search in
2218 * @name: qstr of name we wish to find
2220 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2222 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2225 * Check for a fs-specific hash function. Note that we must
2226 * calculate the standard hash first, as the d_op->d_hash()
2227 * routine may choose to leave the hash value unchanged.
2229 name->hash = full_name_hash(name->name, name->len);
2230 if (dir->d_flags & DCACHE_OP_HASH) {
2231 int err = dir->d_op->d_hash(dir, name);
2232 if (unlikely(err < 0))
2233 return ERR_PTR(err);
2235 return d_lookup(dir, name);
2237 EXPORT_SYMBOL(d_hash_and_lookup);
2240 * d_validate - verify dentry provided from insecure source (deprecated)
2241 * @dentry: The dentry alleged to be valid child of @dparent
2242 * @dparent: The parent dentry (known to be valid)
2244 * An insecure source has sent us a dentry, here we verify it and dget() it.
2245 * This is used by ncpfs in its readdir implementation.
2246 * Zero is returned in the dentry is invalid.
2248 * This function is slow for big directories, and deprecated, do not use it.
2250 int d_validate(struct dentry *dentry, struct dentry *dparent)
2252 struct dentry *child;
2254 spin_lock(&dparent->d_lock);
2255 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2256 if (dentry == child) {
2257 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2258 __dget_dlock(dentry);
2259 spin_unlock(&dentry->d_lock);
2260 spin_unlock(&dparent->d_lock);
2264 spin_unlock(&dparent->d_lock);
2268 EXPORT_SYMBOL(d_validate);
2271 * When a file is deleted, we have two options:
2272 * - turn this dentry into a negative dentry
2273 * - unhash this dentry and free it.
2275 * Usually, we want to just turn this into
2276 * a negative dentry, but if anybody else is
2277 * currently using the dentry or the inode
2278 * we can't do that and we fall back on removing
2279 * it from the hash queues and waiting for
2280 * it to be deleted later when it has no users
2284 * d_delete - delete a dentry
2285 * @dentry: The dentry to delete
2287 * Turn the dentry into a negative dentry if possible, otherwise
2288 * remove it from the hash queues so it can be deleted later
2291 void d_delete(struct dentry * dentry)
2293 struct inode *inode;
2296 * Are we the only user?
2299 spin_lock(&dentry->d_lock);
2300 inode = dentry->d_inode;
2301 isdir = S_ISDIR(inode->i_mode);
2302 if (dentry->d_lockref.count == 1) {
2303 if (!spin_trylock(&inode->i_lock)) {
2304 spin_unlock(&dentry->d_lock);
2308 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2309 dentry_unlink_inode(dentry);
2310 fsnotify_nameremove(dentry, isdir);
2314 if (!d_unhashed(dentry))
2317 spin_unlock(&dentry->d_lock);
2319 fsnotify_nameremove(dentry, isdir);
2321 EXPORT_SYMBOL(d_delete);
2323 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2325 BUG_ON(!d_unhashed(entry));
2327 entry->d_flags |= DCACHE_RCUACCESS;
2328 hlist_bl_add_head_rcu(&entry->d_hash, b);
2332 static void _d_rehash(struct dentry * entry)
2334 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2338 * d_rehash - add an entry back to the hash
2339 * @entry: dentry to add to the hash
2341 * Adds a dentry to the hash according to its name.
2344 void d_rehash(struct dentry * entry)
2346 spin_lock(&entry->d_lock);
2348 spin_unlock(&entry->d_lock);
2350 EXPORT_SYMBOL(d_rehash);
2353 * dentry_update_name_case - update case insensitive dentry with a new name
2354 * @dentry: dentry to be updated
2357 * Update a case insensitive dentry with new case of name.
2359 * dentry must have been returned by d_lookup with name @name. Old and new
2360 * name lengths must match (ie. no d_compare which allows mismatched name
2363 * Parent inode i_mutex must be held over d_lookup and into this call (to
2364 * keep renames and concurrent inserts, and readdir(2) away).
2366 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2368 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2369 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2371 spin_lock(&dentry->d_lock);
2372 write_seqcount_begin(&dentry->d_seq);
2373 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2374 write_seqcount_end(&dentry->d_seq);
2375 spin_unlock(&dentry->d_lock);
2377 EXPORT_SYMBOL(dentry_update_name_case);
2379 static void switch_names(struct dentry *dentry, struct dentry *target)
2381 if (dname_external(target)) {
2382 if (dname_external(dentry)) {
2384 * Both external: swap the pointers
2386 swap(target->d_name.name, dentry->d_name.name);
2389 * dentry:internal, target:external. Steal target's
2390 * storage and make target internal.
2392 memcpy(target->d_iname, dentry->d_name.name,
2393 dentry->d_name.len + 1);
2394 dentry->d_name.name = target->d_name.name;
2395 target->d_name.name = target->d_iname;
2398 if (dname_external(dentry)) {
2400 * dentry:external, target:internal. Give dentry's
2401 * storage to target and make dentry internal
2403 memcpy(dentry->d_iname, target->d_name.name,
2404 target->d_name.len + 1);
2405 target->d_name.name = dentry->d_name.name;
2406 dentry->d_name.name = dentry->d_iname;
2409 * Both are internal.
2412 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2413 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2414 swap(((long *) &dentry->d_iname)[i],
2415 ((long *) &target->d_iname)[i]);
2419 swap(dentry->d_name.len, target->d_name.len);
2422 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2425 * XXXX: do we really need to take target->d_lock?
2427 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2428 spin_lock(&target->d_parent->d_lock);
2430 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2431 spin_lock(&dentry->d_parent->d_lock);
2432 spin_lock_nested(&target->d_parent->d_lock,
2433 DENTRY_D_LOCK_NESTED);
2435 spin_lock(&target->d_parent->d_lock);
2436 spin_lock_nested(&dentry->d_parent->d_lock,
2437 DENTRY_D_LOCK_NESTED);
2440 if (target < dentry) {
2441 spin_lock_nested(&target->d_lock, 2);
2442 spin_lock_nested(&dentry->d_lock, 3);
2444 spin_lock_nested(&dentry->d_lock, 2);
2445 spin_lock_nested(&target->d_lock, 3);
2449 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2450 struct dentry *target)
2452 if (target->d_parent != dentry->d_parent)
2453 spin_unlock(&dentry->d_parent->d_lock);
2454 if (target->d_parent != target)
2455 spin_unlock(&target->d_parent->d_lock);
2459 * When switching names, the actual string doesn't strictly have to
2460 * be preserved in the target - because we're dropping the target
2461 * anyway. As such, we can just do a simple memcpy() to copy over
2462 * the new name before we switch.
2464 * Note that we have to be a lot more careful about getting the hash
2465 * switched - we have to switch the hash value properly even if it
2466 * then no longer matches the actual (corrupted) string of the target.
2467 * The hash value has to match the hash queue that the dentry is on..
2470 * __d_move - move a dentry
2471 * @dentry: entry to move
2472 * @target: new dentry
2473 * @exchange: exchange the two dentries
2475 * Update the dcache to reflect the move of a file name. Negative
2476 * dcache entries should not be moved in this way. Caller must hold
2477 * rename_lock, the i_mutex of the source and target directories,
2478 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2480 static void __d_move(struct dentry *dentry, struct dentry *target,
2483 if (!dentry->d_inode)
2484 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2486 BUG_ON(d_ancestor(dentry, target));
2487 BUG_ON(d_ancestor(target, dentry));
2489 dentry_lock_for_move(dentry, target);
2491 write_seqcount_begin(&dentry->d_seq);
2492 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2494 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2497 * Move the dentry to the target hash queue. Don't bother checking
2498 * for the same hash queue because of how unlikely it is.
2501 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2504 * Unhash the target (d_delete() is not usable here). If exchanging
2505 * the two dentries, then rehash onto the other's hash queue.
2510 d_hash(dentry->d_parent, dentry->d_name.hash));
2513 list_del(&dentry->d_u.d_child);
2514 list_del(&target->d_u.d_child);
2516 /* Switch the names.. */
2517 switch_names(dentry, target);
2518 swap(dentry->d_name.hash, target->d_name.hash);
2520 /* ... and switch the parents */
2521 if (IS_ROOT(dentry)) {
2522 dentry->d_parent = target->d_parent;
2523 target->d_parent = target;
2524 INIT_LIST_HEAD(&target->d_u.d_child);
2526 swap(dentry->d_parent, target->d_parent);
2528 /* And add them back to the (new) parent lists */
2529 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2532 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2534 write_seqcount_end(&target->d_seq);
2535 write_seqcount_end(&dentry->d_seq);
2537 dentry_unlock_parents_for_move(dentry, target);
2539 fsnotify_d_move(target);
2540 spin_unlock(&target->d_lock);
2541 fsnotify_d_move(dentry);
2542 spin_unlock(&dentry->d_lock);
2546 * d_move - move a dentry
2547 * @dentry: entry to move
2548 * @target: new dentry
2550 * Update the dcache to reflect the move of a file name. Negative
2551 * dcache entries should not be moved in this way. See the locking
2552 * requirements for __d_move.
2554 void d_move(struct dentry *dentry, struct dentry *target)
2556 write_seqlock(&rename_lock);
2557 __d_move(dentry, target, false);
2558 write_sequnlock(&rename_lock);
2560 EXPORT_SYMBOL(d_move);
2563 * d_exchange - exchange two dentries
2564 * @dentry1: first dentry
2565 * @dentry2: second dentry
2567 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2569 write_seqlock(&rename_lock);
2571 WARN_ON(!dentry1->d_inode);
2572 WARN_ON(!dentry2->d_inode);
2573 WARN_ON(IS_ROOT(dentry1));
2574 WARN_ON(IS_ROOT(dentry2));
2576 __d_move(dentry1, dentry2, true);
2578 write_sequnlock(&rename_lock);
2582 * d_ancestor - search for an ancestor
2583 * @p1: ancestor dentry
2586 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2587 * an ancestor of p2, else NULL.
2589 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2593 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2594 if (p->d_parent == p1)
2601 * This helper attempts to cope with remotely renamed directories
2603 * It assumes that the caller is already holding
2604 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2606 * Note: If ever the locking in lock_rename() changes, then please
2607 * remember to update this too...
2609 static struct dentry *__d_unalias(struct inode *inode,
2610 struct dentry *dentry, struct dentry *alias)
2612 struct mutex *m1 = NULL, *m2 = NULL;
2613 struct dentry *ret = ERR_PTR(-EBUSY);
2615 /* If alias and dentry share a parent, then no extra locks required */
2616 if (alias->d_parent == dentry->d_parent)
2619 /* See lock_rename() */
2620 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2622 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2623 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2625 m2 = &alias->d_parent->d_inode->i_mutex;
2627 if (likely(!d_mountpoint(alias))) {
2628 __d_move(alias, dentry, false);
2632 spin_unlock(&inode->i_lock);
2641 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2642 * named dentry in place of the dentry to be replaced.
2643 * returns with anon->d_lock held!
2645 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2647 struct dentry *dparent;
2649 dentry_lock_for_move(anon, dentry);
2651 write_seqcount_begin(&dentry->d_seq);
2652 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2654 dparent = dentry->d_parent;
2656 switch_names(dentry, anon);
2657 swap(dentry->d_name.hash, anon->d_name.hash);
2659 dentry->d_parent = dentry;
2660 list_del_init(&dentry->d_u.d_child);
2661 anon->d_parent = dparent;
2662 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2664 write_seqcount_end(&dentry->d_seq);
2665 write_seqcount_end(&anon->d_seq);
2667 dentry_unlock_parents_for_move(anon, dentry);
2668 spin_unlock(&dentry->d_lock);
2670 /* anon->d_lock still locked, returns locked */
2674 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2675 * @inode: the inode which may have a disconnected dentry
2676 * @dentry: a negative dentry which we want to point to the inode.
2678 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2679 * place of the given dentry and return it, else simply d_add the inode
2680 * to the dentry and return NULL.
2682 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2683 * we should error out: directories can't have multiple aliases.
2685 * This is needed in the lookup routine of any filesystem that is exportable
2686 * (via knfsd) so that we can build dcache paths to directories effectively.
2688 * If a dentry was found and moved, then it is returned. Otherwise NULL
2689 * is returned. This matches the expected return value of ->lookup.
2691 * Cluster filesystems may call this function with a negative, hashed dentry.
2692 * In that case, we know that the inode will be a regular file, and also this
2693 * will only occur during atomic_open. So we need to check for the dentry
2694 * being already hashed only in the final case.
2696 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2698 struct dentry *new = NULL;
2701 return ERR_CAST(inode);
2703 if (inode && S_ISDIR(inode->i_mode)) {
2704 spin_lock(&inode->i_lock);
2705 new = __d_find_any_alias(inode);
2707 if (!IS_ROOT(new)) {
2708 spin_unlock(&inode->i_lock);
2710 return ERR_PTR(-EIO);
2712 write_seqlock(&rename_lock);
2713 __d_materialise_dentry(dentry, new);
2714 write_sequnlock(&rename_lock);
2717 spin_unlock(&new->d_lock);
2718 spin_unlock(&inode->i_lock);
2719 security_d_instantiate(new, inode);
2722 /* already taking inode->i_lock, so d_add() by hand */
2723 __d_instantiate(dentry, inode);
2724 spin_unlock(&inode->i_lock);
2725 security_d_instantiate(dentry, inode);
2729 d_instantiate(dentry, inode);
2730 if (d_unhashed(dentry))
2735 EXPORT_SYMBOL(d_splice_alias);
2738 * d_materialise_unique - introduce an inode into the tree
2739 * @dentry: candidate dentry
2740 * @inode: inode to bind to the dentry, to which aliases may be attached
2742 * Introduces an dentry into the tree, substituting an extant disconnected
2743 * root directory alias in its place if there is one. Caller must hold the
2744 * i_mutex of the parent directory.
2746 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2748 struct dentry *actual;
2750 BUG_ON(!d_unhashed(dentry));
2754 __d_instantiate(dentry, NULL);
2759 spin_lock(&inode->i_lock);
2761 if (S_ISDIR(inode->i_mode)) {
2762 struct dentry *alias;
2764 /* Does an aliased dentry already exist? */
2765 alias = __d_find_alias(inode);
2768 write_seqlock(&rename_lock);
2770 if (d_ancestor(alias, dentry)) {
2771 /* Check for loops */
2772 actual = ERR_PTR(-ELOOP);
2773 spin_unlock(&inode->i_lock);
2774 } else if (IS_ROOT(alias)) {
2775 /* Is this an anonymous mountpoint that we
2776 * could splice into our tree? */
2777 __d_materialise_dentry(dentry, alias);
2778 write_sequnlock(&rename_lock);
2782 /* Nope, but we must(!) avoid directory
2783 * aliasing. This drops inode->i_lock */
2784 actual = __d_unalias(inode, dentry, alias);
2786 write_sequnlock(&rename_lock);
2787 if (IS_ERR(actual)) {
2788 if (PTR_ERR(actual) == -ELOOP)
2789 pr_warn_ratelimited(
2790 "VFS: Lookup of '%s' in %s %s"
2791 " would have caused loop\n",
2792 dentry->d_name.name,
2793 inode->i_sb->s_type->name,
2801 /* Add a unique reference */
2802 actual = __d_instantiate_unique(dentry, inode);
2806 BUG_ON(!d_unhashed(actual));
2808 spin_lock(&actual->d_lock);
2811 spin_unlock(&actual->d_lock);
2812 spin_unlock(&inode->i_lock);
2814 if (actual == dentry) {
2815 security_d_instantiate(dentry, inode);
2822 EXPORT_SYMBOL_GPL(d_materialise_unique);
2824 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2828 return -ENAMETOOLONG;
2830 memcpy(*buffer, str, namelen);
2835 * prepend_name - prepend a pathname in front of current buffer pointer
2836 * @buffer: buffer pointer
2837 * @buflen: allocated length of the buffer
2838 * @name: name string and length qstr structure
2840 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2841 * make sure that either the old or the new name pointer and length are
2842 * fetched. However, there may be mismatch between length and pointer.
2843 * The length cannot be trusted, we need to copy it byte-by-byte until
2844 * the length is reached or a null byte is found. It also prepends "/" at
2845 * the beginning of the name. The sequence number check at the caller will
2846 * retry it again when a d_move() does happen. So any garbage in the buffer
2847 * due to mismatched pointer and length will be discarded.
2849 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2851 const char *dname = ACCESS_ONCE(name->name);
2852 u32 dlen = ACCESS_ONCE(name->len);
2855 *buflen -= dlen + 1;
2857 return -ENAMETOOLONG;
2858 p = *buffer -= dlen + 1;
2870 * prepend_path - Prepend path string to a buffer
2871 * @path: the dentry/vfsmount to report
2872 * @root: root vfsmnt/dentry
2873 * @buffer: pointer to the end of the buffer
2874 * @buflen: pointer to buffer length
2876 * The function will first try to write out the pathname without taking any
2877 * lock other than the RCU read lock to make sure that dentries won't go away.
2878 * It only checks the sequence number of the global rename_lock as any change
2879 * in the dentry's d_seq will be preceded by changes in the rename_lock
2880 * sequence number. If the sequence number had been changed, it will restart
2881 * the whole pathname back-tracing sequence again by taking the rename_lock.
2882 * In this case, there is no need to take the RCU read lock as the recursive
2883 * parent pointer references will keep the dentry chain alive as long as no
2884 * rename operation is performed.
2886 static int prepend_path(const struct path *path,
2887 const struct path *root,
2888 char **buffer, int *buflen)
2890 struct dentry *dentry;
2891 struct vfsmount *vfsmnt;
2894 unsigned seq, m_seq = 0;
2900 read_seqbegin_or_lock(&mount_lock, &m_seq);
2907 dentry = path->dentry;
2909 mnt = real_mount(vfsmnt);
2910 read_seqbegin_or_lock(&rename_lock, &seq);
2911 while (dentry != root->dentry || vfsmnt != root->mnt) {
2912 struct dentry * parent;
2914 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2915 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2917 if (mnt != parent) {
2918 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2924 * Filesystems needing to implement special "root names"
2925 * should do so with ->d_dname()
2927 if (IS_ROOT(dentry) &&
2928 (dentry->d_name.len != 1 ||
2929 dentry->d_name.name[0] != '/')) {
2930 WARN(1, "Root dentry has weird name <%.*s>\n",
2931 (int) dentry->d_name.len,
2932 dentry->d_name.name);
2935 error = is_mounted(vfsmnt) ? 1 : 2;
2938 parent = dentry->d_parent;
2940 error = prepend_name(&bptr, &blen, &dentry->d_name);
2948 if (need_seqretry(&rename_lock, seq)) {
2952 done_seqretry(&rename_lock, seq);
2956 if (need_seqretry(&mount_lock, m_seq)) {
2960 done_seqretry(&mount_lock, m_seq);
2962 if (error >= 0 && bptr == *buffer) {
2964 error = -ENAMETOOLONG;
2974 * __d_path - return the path of a dentry
2975 * @path: the dentry/vfsmount to report
2976 * @root: root vfsmnt/dentry
2977 * @buf: buffer to return value in
2978 * @buflen: buffer length
2980 * Convert a dentry into an ASCII path name.
2982 * Returns a pointer into the buffer or an error code if the
2983 * path was too long.
2985 * "buflen" should be positive.
2987 * If the path is not reachable from the supplied root, return %NULL.
2989 char *__d_path(const struct path *path,
2990 const struct path *root,
2991 char *buf, int buflen)
2993 char *res = buf + buflen;
2996 prepend(&res, &buflen, "\0", 1);
2997 error = prepend_path(path, root, &res, &buflen);
3000 return ERR_PTR(error);
3006 char *d_absolute_path(const struct path *path,
3007 char *buf, int buflen)
3009 struct path root = {};
3010 char *res = buf + buflen;
3013 prepend(&res, &buflen, "\0", 1);
3014 error = prepend_path(path, &root, &res, &buflen);
3019 return ERR_PTR(error);
3024 * same as __d_path but appends "(deleted)" for unlinked files.
3026 static int path_with_deleted(const struct path *path,
3027 const struct path *root,
3028 char **buf, int *buflen)
3030 prepend(buf, buflen, "\0", 1);
3031 if (d_unlinked(path->dentry)) {
3032 int error = prepend(buf, buflen, " (deleted)", 10);
3037 return prepend_path(path, root, buf, buflen);
3040 static int prepend_unreachable(char **buffer, int *buflen)
3042 return prepend(buffer, buflen, "(unreachable)", 13);
3045 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3050 seq = read_seqcount_begin(&fs->seq);
3052 } while (read_seqcount_retry(&fs->seq, seq));
3056 * d_path - return the path of a dentry
3057 * @path: path to report
3058 * @buf: buffer to return value in
3059 * @buflen: buffer length
3061 * Convert a dentry into an ASCII path name. If the entry has been deleted
3062 * the string " (deleted)" is appended. Note that this is ambiguous.
3064 * Returns a pointer into the buffer or an error code if the path was
3065 * too long. Note: Callers should use the returned pointer, not the passed
3066 * in buffer, to use the name! The implementation often starts at an offset
3067 * into the buffer, and may leave 0 bytes at the start.
3069 * "buflen" should be positive.
3071 char *d_path(const struct path *path, char *buf, int buflen)
3073 char *res = buf + buflen;
3078 * We have various synthetic filesystems that never get mounted. On
3079 * these filesystems dentries are never used for lookup purposes, and
3080 * thus don't need to be hashed. They also don't need a name until a
3081 * user wants to identify the object in /proc/pid/fd/. The little hack
3082 * below allows us to generate a name for these objects on demand:
3084 * Some pseudo inodes are mountable. When they are mounted
3085 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3086 * and instead have d_path return the mounted path.
3088 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3089 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3090 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3093 get_fs_root_rcu(current->fs, &root);
3094 error = path_with_deleted(path, &root, &res, &buflen);
3098 res = ERR_PTR(error);
3101 EXPORT_SYMBOL(d_path);
3104 * Helper function for dentry_operations.d_dname() members
3106 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3107 const char *fmt, ...)
3113 va_start(args, fmt);
3114 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3117 if (sz > sizeof(temp) || sz > buflen)
3118 return ERR_PTR(-ENAMETOOLONG);
3120 buffer += buflen - sz;
3121 return memcpy(buffer, temp, sz);
3124 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3126 char *end = buffer + buflen;
3127 /* these dentries are never renamed, so d_lock is not needed */
3128 if (prepend(&end, &buflen, " (deleted)", 11) ||
3129 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3130 prepend(&end, &buflen, "/", 1))
3131 end = ERR_PTR(-ENAMETOOLONG);
3134 EXPORT_SYMBOL(simple_dname);
3137 * Write full pathname from the root of the filesystem into the buffer.
3139 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3141 struct dentry *dentry;
3154 prepend(&end, &len, "\0", 1);
3158 read_seqbegin_or_lock(&rename_lock, &seq);
3159 while (!IS_ROOT(dentry)) {
3160 struct dentry *parent = dentry->d_parent;
3163 error = prepend_name(&end, &len, &dentry->d_name);
3172 if (need_seqretry(&rename_lock, seq)) {
3176 done_seqretry(&rename_lock, seq);
3181 return ERR_PTR(-ENAMETOOLONG);
3184 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3186 return __dentry_path(dentry, buf, buflen);
3188 EXPORT_SYMBOL(dentry_path_raw);
3190 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3195 if (d_unlinked(dentry)) {
3197 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3201 retval = __dentry_path(dentry, buf, buflen);
3202 if (!IS_ERR(retval) && p)
3203 *p = '/'; /* restore '/' overriden with '\0' */
3206 return ERR_PTR(-ENAMETOOLONG);
3209 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3215 seq = read_seqcount_begin(&fs->seq);
3218 } while (read_seqcount_retry(&fs->seq, seq));
3222 * NOTE! The user-level library version returns a
3223 * character pointer. The kernel system call just
3224 * returns the length of the buffer filled (which
3225 * includes the ending '\0' character), or a negative
3226 * error value. So libc would do something like
3228 * char *getcwd(char * buf, size_t size)
3232 * retval = sys_getcwd(buf, size);
3239 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3242 struct path pwd, root;
3243 char *page = __getname();
3249 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3252 if (!d_unlinked(pwd.dentry)) {
3254 char *cwd = page + PATH_MAX;
3255 int buflen = PATH_MAX;
3257 prepend(&cwd, &buflen, "\0", 1);
3258 error = prepend_path(&pwd, &root, &cwd, &buflen);
3264 /* Unreachable from current root */
3266 error = prepend_unreachable(&cwd, &buflen);
3272 len = PATH_MAX + page - cwd;
3275 if (copy_to_user(buf, cwd, len))
3288 * Test whether new_dentry is a subdirectory of old_dentry.
3290 * Trivially implemented using the dcache structure
3294 * is_subdir - is new dentry a subdirectory of old_dentry
3295 * @new_dentry: new dentry
3296 * @old_dentry: old dentry
3298 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3299 * Returns 0 otherwise.
3300 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3303 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3308 if (new_dentry == old_dentry)
3312 /* for restarting inner loop in case of seq retry */
3313 seq = read_seqbegin(&rename_lock);
3315 * Need rcu_readlock to protect against the d_parent trashing
3319 if (d_ancestor(old_dentry, new_dentry))
3324 } while (read_seqretry(&rename_lock, seq));
3329 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3331 struct dentry *root = data;
3332 if (dentry != root) {
3333 if (d_unhashed(dentry) || !dentry->d_inode)
3336 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3337 dentry->d_flags |= DCACHE_GENOCIDE;
3338 dentry->d_lockref.count--;
3341 return D_WALK_CONTINUE;
3344 void d_genocide(struct dentry *parent)
3346 d_walk(parent, parent, d_genocide_kill, NULL);
3349 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3351 inode_dec_link_count(inode);
3352 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3353 !hlist_unhashed(&dentry->d_alias) ||
3354 !d_unlinked(dentry));
3355 spin_lock(&dentry->d_parent->d_lock);
3356 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3357 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3358 (unsigned long long)inode->i_ino);
3359 spin_unlock(&dentry->d_lock);
3360 spin_unlock(&dentry->d_parent->d_lock);
3361 d_instantiate(dentry, inode);
3363 EXPORT_SYMBOL(d_tmpfile);
3365 static __initdata unsigned long dhash_entries;
3366 static int __init set_dhash_entries(char *str)
3370 dhash_entries = simple_strtoul(str, &str, 0);
3373 __setup("dhash_entries=", set_dhash_entries);
3375 static void __init dcache_init_early(void)
3379 /* If hashes are distributed across NUMA nodes, defer
3380 * hash allocation until vmalloc space is available.
3386 alloc_large_system_hash("Dentry cache",
3387 sizeof(struct hlist_bl_head),
3396 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3397 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3400 static void __init dcache_init(void)
3405 * A constructor could be added for stable state like the lists,
3406 * but it is probably not worth it because of the cache nature
3409 dentry_cache = KMEM_CACHE(dentry,
3410 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3412 /* Hash may have been set up in dcache_init_early */
3417 alloc_large_system_hash("Dentry cache",
3418 sizeof(struct hlist_bl_head),
3427 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3428 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3431 /* SLAB cache for __getname() consumers */
3432 struct kmem_cache *names_cachep __read_mostly;
3433 EXPORT_SYMBOL(names_cachep);
3435 EXPORT_SYMBOL(d_genocide);
3437 void __init vfs_caches_init_early(void)
3439 dcache_init_early();
3443 void __init vfs_caches_init(unsigned long mempages)
3445 unsigned long reserve;
3447 /* Base hash sizes on available memory, with a reserve equal to
3448 150% of current kernel size */
3450 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3451 mempages -= reserve;
3453 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3454 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3458 files_init(mempages);