4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
46 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
48 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
52 * dentry->d_sb->s_dentry_lru_lock protects:
53 * - the dcache lru lists and counters
60 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
65 * dentry->d_inode->i_lock
67 * dentry->d_sb->s_dentry_lru_lock
68 * dcache_hash_bucket lock
71 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
74 * dentry->d_parent->d_lock
77 * If no ancestor relationship:
78 * if (dentry1 < dentry2)
82 int sysctl_vfs_cache_pressure __read_mostly = 100;
83 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
87 EXPORT_SYMBOL(rename_lock);
89 static struct kmem_cache *dentry_cache __read_mostly;
92 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
100 static unsigned int d_hash_mask __read_mostly;
101 static unsigned int d_hash_shift __read_mostly;
103 static struct hlist_bl_head *dentry_hashtable __read_mostly;
105 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
108 hash += (unsigned long) parent / L1_CACHE_BYTES;
109 hash = hash + (hash >> d_hash_shift);
110 return dentry_hashtable + (hash & d_hash_mask);
113 /* Statistics gathering. */
114 struct dentry_stat_t dentry_stat = {
118 static DEFINE_PER_CPU(long, nr_dentry);
119 static DEFINE_PER_CPU(long, nr_dentry_unused);
121 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
124 * Here we resort to our own counters instead of using generic per-cpu counters
125 * for consistency with what the vfs inode code does. We are expected to harvest
126 * better code and performance by having our own specialized counters.
128 * Please note that the loop is done over all possible CPUs, not over all online
129 * CPUs. The reason for this is that we don't want to play games with CPUs going
130 * on and off. If one of them goes off, we will just keep their counters.
132 * glommer: See cffbc8a for details, and if you ever intend to change this,
133 * please update all vfs counters to match.
135 static long get_nr_dentry(void)
139 for_each_possible_cpu(i)
140 sum += per_cpu(nr_dentry, i);
141 return sum < 0 ? 0 : sum;
144 static long get_nr_dentry_unused(void)
148 for_each_possible_cpu(i)
149 sum += per_cpu(nr_dentry_unused, i);
150 return sum < 0 ? 0 : sum;
153 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
154 size_t *lenp, loff_t *ppos)
156 dentry_stat.nr_dentry = get_nr_dentry();
157 dentry_stat.nr_unused = get_nr_dentry_unused();
158 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
163 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
164 * The strings are both count bytes long, and count is non-zero.
166 #ifdef CONFIG_DCACHE_WORD_ACCESS
168 #include <asm/word-at-a-time.h>
170 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
171 * aligned allocation for this particular component. We don't
172 * strictly need the load_unaligned_zeropad() safety, but it
173 * doesn't hurt either.
175 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
176 * need the careful unaligned handling.
178 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
180 unsigned long a,b,mask;
183 a = *(unsigned long *)cs;
184 b = load_unaligned_zeropad(ct);
185 if (tcount < sizeof(unsigned long))
187 if (unlikely(a != b))
189 cs += sizeof(unsigned long);
190 ct += sizeof(unsigned long);
191 tcount -= sizeof(unsigned long);
195 mask = bytemask_from_count(tcount);
196 return unlikely(!!((a ^ b) & mask));
201 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
215 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
217 const unsigned char *cs;
219 * Be careful about RCU walk racing with rename:
220 * use ACCESS_ONCE to fetch the name pointer.
222 * NOTE! Even if a rename will mean that the length
223 * was not loaded atomically, we don't care. The
224 * RCU walk will check the sequence count eventually,
225 * and catch it. And we won't overrun the buffer,
226 * because we're reading the name pointer atomically,
227 * and a dentry name is guaranteed to be properly
228 * terminated with a NUL byte.
230 * End result: even if 'len' is wrong, we'll exit
231 * early because the data cannot match (there can
232 * be no NUL in the ct/tcount data)
234 cs = ACCESS_ONCE(dentry->d_name.name);
235 smp_read_barrier_depends();
236 return dentry_string_cmp(cs, ct, tcount);
239 static void __d_free(struct rcu_head *head)
241 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
243 WARN_ON(!hlist_unhashed(&dentry->d_alias));
244 if (dname_external(dentry))
245 kfree(dentry->d_name.name);
246 kmem_cache_free(dentry_cache, dentry);
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 *
504 dentry_kill(struct dentry *dentry, int unlock_on_failure)
505 __releases(dentry->d_lock)
507 struct inode *inode = dentry->d_inode;
508 struct dentry *parent = NULL;
510 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
513 if (!IS_ROOT(dentry)) {
514 parent = dentry->d_parent;
515 if (unlikely(!spin_trylock(&parent->d_lock))) {
517 spin_unlock(&inode->i_lock);
522 __dentry_kill(dentry);
526 if (unlock_on_failure) {
527 spin_unlock(&dentry->d_lock);
530 return dentry; /* try again with same dentry */
533 static inline struct dentry *lock_parent(struct dentry *dentry)
535 struct dentry *parent = dentry->d_parent;
538 if (likely(spin_trylock(&parent->d_lock)))
540 spin_unlock(&dentry->d_lock);
543 parent = ACCESS_ONCE(dentry->d_parent);
544 spin_lock(&parent->d_lock);
546 * We can't blindly lock dentry until we are sure
547 * that we won't violate the locking order.
548 * Any changes of dentry->d_parent must have
549 * been done with parent->d_lock held, so
550 * spin_lock() above is enough of a barrier
551 * for checking if it's still our child.
553 if (unlikely(parent != dentry->d_parent)) {
554 spin_unlock(&parent->d_lock);
558 if (parent != dentry)
559 spin_lock(&dentry->d_lock);
568 * This is complicated by the fact that we do not want to put
569 * dentries that are no longer on any hash chain on the unused
570 * list: we'd much rather just get rid of them immediately.
572 * However, that implies that we have to traverse the dentry
573 * tree upwards to the parents which might _also_ now be
574 * scheduled for deletion (it may have been only waiting for
575 * its last child to go away).
577 * This tail recursion is done by hand as we don't want to depend
578 * on the compiler to always get this right (gcc generally doesn't).
579 * Real recursion would eat up our stack space.
583 * dput - release a dentry
584 * @dentry: dentry to release
586 * Release a dentry. This will drop the usage count and if appropriate
587 * call the dentry unlink method as well as removing it from the queues and
588 * releasing its resources. If the parent dentries were scheduled for release
589 * they too may now get deleted.
591 void dput(struct dentry *dentry)
593 if (unlikely(!dentry))
597 if (lockref_put_or_lock(&dentry->d_lockref))
600 /* Unreachable? Get rid of it */
601 if (unlikely(d_unhashed(dentry)))
604 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
605 if (dentry->d_op->d_delete(dentry))
609 if (!(dentry->d_flags & DCACHE_REFERENCED))
610 dentry->d_flags |= DCACHE_REFERENCED;
611 dentry_lru_add(dentry);
613 dentry->d_lockref.count--;
614 spin_unlock(&dentry->d_lock);
618 dentry = dentry_kill(dentry, 1);
625 * d_invalidate - invalidate a dentry
626 * @dentry: dentry to invalidate
628 * Try to invalidate the dentry if it turns out to be
629 * possible. If there are other dentries that can be
630 * reached through this one we can't delete it and we
631 * return -EBUSY. On success we return 0.
636 int d_invalidate(struct dentry * dentry)
639 * If it's already been dropped, return OK.
641 spin_lock(&dentry->d_lock);
642 if (d_unhashed(dentry)) {
643 spin_unlock(&dentry->d_lock);
647 * Check whether to do a partial shrink_dcache
648 * to get rid of unused child entries.
650 if (!list_empty(&dentry->d_subdirs)) {
651 spin_unlock(&dentry->d_lock);
652 shrink_dcache_parent(dentry);
653 spin_lock(&dentry->d_lock);
657 * Somebody else still using it?
659 * If it's a directory, we can't drop it
660 * for fear of somebody re-populating it
661 * with children (even though dropping it
662 * would make it unreachable from the root,
663 * we might still populate it if it was a
664 * working directory or similar).
665 * We also need to leave mountpoints alone,
668 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
669 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
670 spin_unlock(&dentry->d_lock);
676 spin_unlock(&dentry->d_lock);
679 EXPORT_SYMBOL(d_invalidate);
681 /* This must be called with d_lock held */
682 static inline void __dget_dlock(struct dentry *dentry)
684 dentry->d_lockref.count++;
687 static inline void __dget(struct dentry *dentry)
689 lockref_get(&dentry->d_lockref);
692 struct dentry *dget_parent(struct dentry *dentry)
698 * Do optimistic parent lookup without any
702 ret = ACCESS_ONCE(dentry->d_parent);
703 gotref = lockref_get_not_zero(&ret->d_lockref);
705 if (likely(gotref)) {
706 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
713 * Don't need rcu_dereference because we re-check it was correct under
717 ret = dentry->d_parent;
718 spin_lock(&ret->d_lock);
719 if (unlikely(ret != dentry->d_parent)) {
720 spin_unlock(&ret->d_lock);
725 BUG_ON(!ret->d_lockref.count);
726 ret->d_lockref.count++;
727 spin_unlock(&ret->d_lock);
730 EXPORT_SYMBOL(dget_parent);
733 * d_find_alias - grab a hashed alias of inode
734 * @inode: inode in question
735 * @want_discon: flag, used by d_splice_alias, to request
736 * that only a DISCONNECTED alias be returned.
738 * If inode has a hashed alias, or is a directory and has any alias,
739 * acquire the reference to alias and return it. Otherwise return NULL.
740 * Notice that if inode is a directory there can be only one alias and
741 * it can be unhashed only if it has no children, or if it is the root
744 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
745 * any other hashed alias over that one unless @want_discon is set,
746 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
748 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
750 struct dentry *alias, *discon_alias;
754 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
755 spin_lock(&alias->d_lock);
756 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
757 if (IS_ROOT(alias) &&
758 (alias->d_flags & DCACHE_DISCONNECTED)) {
759 discon_alias = alias;
760 } else if (!want_discon) {
762 spin_unlock(&alias->d_lock);
766 spin_unlock(&alias->d_lock);
769 alias = discon_alias;
770 spin_lock(&alias->d_lock);
771 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
772 if (IS_ROOT(alias) &&
773 (alias->d_flags & DCACHE_DISCONNECTED)) {
775 spin_unlock(&alias->d_lock);
779 spin_unlock(&alias->d_lock);
785 struct dentry *d_find_alias(struct inode *inode)
787 struct dentry *de = NULL;
789 if (!hlist_empty(&inode->i_dentry)) {
790 spin_lock(&inode->i_lock);
791 de = __d_find_alias(inode, 0);
792 spin_unlock(&inode->i_lock);
796 EXPORT_SYMBOL(d_find_alias);
799 * Try to kill dentries associated with this inode.
800 * WARNING: you must own a reference to inode.
802 void d_prune_aliases(struct inode *inode)
804 struct dentry *dentry;
806 spin_lock(&inode->i_lock);
807 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
808 spin_lock(&dentry->d_lock);
809 if (!dentry->d_lockref.count) {
811 * inform the fs via d_prune that this dentry
812 * is about to be unhashed and destroyed.
814 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
816 dentry->d_op->d_prune(dentry);
818 __dget_dlock(dentry);
820 spin_unlock(&dentry->d_lock);
821 spin_unlock(&inode->i_lock);
825 spin_unlock(&dentry->d_lock);
827 spin_unlock(&inode->i_lock);
829 EXPORT_SYMBOL(d_prune_aliases);
831 static void shrink_dentry_list(struct list_head *list)
833 struct dentry *dentry, *parent;
835 while (!list_empty(list)) {
837 dentry = list_entry(list->prev, struct dentry, d_lru);
838 spin_lock(&dentry->d_lock);
839 parent = lock_parent(dentry);
842 * The dispose list is isolated and dentries are not accounted
843 * to the LRU here, so we can simply remove it from the list
844 * here regardless of whether it is referenced or not.
846 d_shrink_del(dentry);
849 * We found an inuse dentry which was not removed from
850 * the LRU because of laziness during lookup. Do not free it.
852 if ((int)dentry->d_lockref.count > 0) {
853 spin_unlock(&dentry->d_lock);
855 spin_unlock(&parent->d_lock);
860 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
861 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
862 spin_unlock(&dentry->d_lock);
864 spin_unlock(&parent->d_lock);
870 inode = dentry->d_inode;
871 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
872 d_shrink_add(dentry, list);
873 spin_unlock(&dentry->d_lock);
875 spin_unlock(&parent->d_lock);
879 __dentry_kill(dentry);
882 * We need to prune ancestors too. This is necessary to prevent
883 * quadratic behavior of shrink_dcache_parent(), but is also
884 * expected to be beneficial in reducing dentry cache
888 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
889 parent = lock_parent(dentry);
890 if (dentry->d_lockref.count != 1) {
891 dentry->d_lockref.count--;
892 spin_unlock(&dentry->d_lock);
894 spin_unlock(&parent->d_lock);
897 inode = dentry->d_inode; /* can't be NULL */
898 if (unlikely(!spin_trylock(&inode->i_lock))) {
899 spin_unlock(&dentry->d_lock);
901 spin_unlock(&parent->d_lock);
905 __dentry_kill(dentry);
911 static enum lru_status
912 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
914 struct list_head *freeable = arg;
915 struct dentry *dentry = container_of(item, struct dentry, d_lru);
919 * we are inverting the lru lock/dentry->d_lock here,
920 * so use a trylock. If we fail to get the lock, just skip
923 if (!spin_trylock(&dentry->d_lock))
927 * Referenced dentries are still in use. If they have active
928 * counts, just remove them from the LRU. Otherwise give them
929 * another pass through the LRU.
931 if (dentry->d_lockref.count) {
932 d_lru_isolate(dentry);
933 spin_unlock(&dentry->d_lock);
937 if (dentry->d_flags & DCACHE_REFERENCED) {
938 dentry->d_flags &= ~DCACHE_REFERENCED;
939 spin_unlock(&dentry->d_lock);
942 * The list move itself will be made by the common LRU code. At
943 * this point, we've dropped the dentry->d_lock but keep the
944 * lru lock. This is safe to do, since every list movement is
945 * protected by the lru lock even if both locks are held.
947 * This is guaranteed by the fact that all LRU management
948 * functions are intermediated by the LRU API calls like
949 * list_lru_add and list_lru_del. List movement in this file
950 * only ever occur through this functions or through callbacks
951 * like this one, that are called from the LRU API.
953 * The only exceptions to this are functions like
954 * shrink_dentry_list, and code that first checks for the
955 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
956 * operating only with stack provided lists after they are
957 * properly isolated from the main list. It is thus, always a
963 d_lru_shrink_move(dentry, freeable);
964 spin_unlock(&dentry->d_lock);
970 * prune_dcache_sb - shrink the dcache
972 * @nr_to_scan : number of entries to try to free
973 * @nid: which node to scan for freeable entities
975 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
976 * done when we need more memory an called from the superblock shrinker
979 * This function may fail to free any resources if all the dentries are in
982 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
988 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
989 &dispose, &nr_to_scan);
990 shrink_dentry_list(&dispose);
994 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
995 spinlock_t *lru_lock, void *arg)
997 struct list_head *freeable = arg;
998 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1001 * we are inverting the lru lock/dentry->d_lock here,
1002 * so use a trylock. If we fail to get the lock, just skip
1005 if (!spin_trylock(&dentry->d_lock))
1008 d_lru_shrink_move(dentry, freeable);
1009 spin_unlock(&dentry->d_lock);
1016 * shrink_dcache_sb - shrink dcache for a superblock
1019 * Shrink the dcache for the specified super block. This is used to free
1020 * the dcache before unmounting a file system.
1022 void shrink_dcache_sb(struct super_block *sb)
1029 freed = list_lru_walk(&sb->s_dentry_lru,
1030 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1032 this_cpu_sub(nr_dentry_unused, freed);
1033 shrink_dentry_list(&dispose);
1034 } while (freed > 0);
1036 EXPORT_SYMBOL(shrink_dcache_sb);
1039 * enum d_walk_ret - action to talke during tree walk
1040 * @D_WALK_CONTINUE: contrinue walk
1041 * @D_WALK_QUIT: quit walk
1042 * @D_WALK_NORETRY: quit when retry is needed
1043 * @D_WALK_SKIP: skip this dentry and its children
1053 * d_walk - walk the dentry tree
1054 * @parent: start of walk
1055 * @data: data passed to @enter() and @finish()
1056 * @enter: callback when first entering the dentry
1057 * @finish: callback when successfully finished the walk
1059 * The @enter() and @finish() callbacks are called with d_lock held.
1061 static void d_walk(struct dentry *parent, void *data,
1062 enum d_walk_ret (*enter)(void *, struct dentry *),
1063 void (*finish)(void *))
1065 struct dentry *this_parent;
1066 struct list_head *next;
1068 enum d_walk_ret ret;
1072 read_seqbegin_or_lock(&rename_lock, &seq);
1073 this_parent = parent;
1074 spin_lock(&this_parent->d_lock);
1076 ret = enter(data, this_parent);
1078 case D_WALK_CONTINUE:
1083 case D_WALK_NORETRY:
1088 next = this_parent->d_subdirs.next;
1090 while (next != &this_parent->d_subdirs) {
1091 struct list_head *tmp = next;
1092 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1095 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1097 ret = enter(data, dentry);
1099 case D_WALK_CONTINUE:
1102 spin_unlock(&dentry->d_lock);
1104 case D_WALK_NORETRY:
1108 spin_unlock(&dentry->d_lock);
1112 if (!list_empty(&dentry->d_subdirs)) {
1113 spin_unlock(&this_parent->d_lock);
1114 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1115 this_parent = dentry;
1116 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1119 spin_unlock(&dentry->d_lock);
1122 * All done at this level ... ascend and resume the search.
1124 if (this_parent != parent) {
1125 struct dentry *child = this_parent;
1126 this_parent = child->d_parent;
1129 spin_unlock(&child->d_lock);
1130 spin_lock(&this_parent->d_lock);
1133 * might go back up the wrong parent if we have had a rename
1136 if (this_parent != child->d_parent ||
1137 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1138 need_seqretry(&rename_lock, seq)) {
1139 spin_unlock(&this_parent->d_lock);
1144 next = child->d_u.d_child.next;
1147 if (need_seqretry(&rename_lock, seq)) {
1148 spin_unlock(&this_parent->d_lock);
1155 spin_unlock(&this_parent->d_lock);
1156 done_seqretry(&rename_lock, seq);
1167 * Search for at least 1 mount point in the dentry's subdirs.
1168 * We descend to the next level whenever the d_subdirs
1169 * list is non-empty and continue searching.
1172 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1175 if (d_mountpoint(dentry)) {
1179 return D_WALK_CONTINUE;
1183 * have_submounts - check for mounts over a dentry
1184 * @parent: dentry to check.
1186 * Return true if the parent or its subdirectories contain
1189 int have_submounts(struct dentry *parent)
1193 d_walk(parent, &ret, check_mount, NULL);
1197 EXPORT_SYMBOL(have_submounts);
1200 * Called by mount code to set a mountpoint and check if the mountpoint is
1201 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1202 * subtree can become unreachable).
1204 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1205 * this reason take rename_lock and d_lock on dentry and ancestors.
1207 int d_set_mounted(struct dentry *dentry)
1211 write_seqlock(&rename_lock);
1212 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1213 /* Need exclusion wrt. check_submounts_and_drop() */
1214 spin_lock(&p->d_lock);
1215 if (unlikely(d_unhashed(p))) {
1216 spin_unlock(&p->d_lock);
1219 spin_unlock(&p->d_lock);
1221 spin_lock(&dentry->d_lock);
1222 if (!d_unlinked(dentry)) {
1223 dentry->d_flags |= DCACHE_MOUNTED;
1226 spin_unlock(&dentry->d_lock);
1228 write_sequnlock(&rename_lock);
1233 * Search the dentry child list of the specified parent,
1234 * and move any unused dentries to the end of the unused
1235 * list for prune_dcache(). We descend to the next level
1236 * whenever the d_subdirs list is non-empty and continue
1239 * It returns zero iff there are no unused children,
1240 * otherwise it returns the number of children moved to
1241 * the end of the unused list. This may not be the total
1242 * number of unused children, because select_parent can
1243 * drop the lock and return early due to latency
1247 struct select_data {
1248 struct dentry *start;
1249 struct list_head dispose;
1253 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1255 struct select_data *data = _data;
1256 enum d_walk_ret ret = D_WALK_CONTINUE;
1258 if (data->start == dentry)
1261 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1264 if (dentry->d_flags & DCACHE_LRU_LIST)
1266 if (!dentry->d_lockref.count) {
1267 d_shrink_add(dentry, &data->dispose);
1272 * We can return to the caller if we have found some (this
1273 * ensures forward progress). We'll be coming back to find
1276 if (!list_empty(&data->dispose))
1277 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1283 * shrink_dcache_parent - prune dcache
1284 * @parent: parent of entries to prune
1286 * Prune the dcache to remove unused children of the parent dentry.
1288 void shrink_dcache_parent(struct dentry *parent)
1291 struct select_data data;
1293 INIT_LIST_HEAD(&data.dispose);
1294 data.start = parent;
1297 d_walk(parent, &data, select_collect, NULL);
1301 shrink_dentry_list(&data.dispose);
1305 EXPORT_SYMBOL(shrink_dcache_parent);
1307 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1309 /* it has busy descendents; complain about those instead */
1310 if (!list_empty(&dentry->d_subdirs))
1311 return D_WALK_CONTINUE;
1313 /* root with refcount 1 is fine */
1314 if (dentry == _data && dentry->d_lockref.count == 1)
1315 return D_WALK_CONTINUE;
1317 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1318 " still in use (%d) [unmount of %s %s]\n",
1321 dentry->d_inode->i_ino : 0UL,
1323 dentry->d_lockref.count,
1324 dentry->d_sb->s_type->name,
1325 dentry->d_sb->s_id);
1327 return D_WALK_CONTINUE;
1330 static void do_one_tree(struct dentry *dentry)
1332 shrink_dcache_parent(dentry);
1333 d_walk(dentry, dentry, umount_check, NULL);
1339 * destroy the dentries attached to a superblock on unmounting
1341 void shrink_dcache_for_umount(struct super_block *sb)
1343 struct dentry *dentry;
1345 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1347 dentry = sb->s_root;
1349 do_one_tree(dentry);
1351 while (!hlist_bl_empty(&sb->s_anon)) {
1352 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1353 do_one_tree(dentry);
1357 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1359 struct select_data *data = _data;
1361 if (d_mountpoint(dentry)) {
1362 data->found = -EBUSY;
1366 return select_collect(_data, dentry);
1369 static void check_and_drop(void *_data)
1371 struct select_data *data = _data;
1373 if (d_mountpoint(data->start))
1374 data->found = -EBUSY;
1376 __d_drop(data->start);
1380 * check_submounts_and_drop - prune dcache, check for submounts and drop
1382 * All done as a single atomic operation relative to has_unlinked_ancestor().
1383 * Returns 0 if successfully unhashed @parent. If there were submounts then
1386 * @dentry: dentry to prune and drop
1388 int check_submounts_and_drop(struct dentry *dentry)
1392 /* Negative dentries can be dropped without further checks */
1393 if (!dentry->d_inode) {
1399 struct select_data data;
1401 INIT_LIST_HEAD(&data.dispose);
1402 data.start = dentry;
1405 d_walk(dentry, &data, check_and_collect, check_and_drop);
1408 if (!list_empty(&data.dispose))
1409 shrink_dentry_list(&data.dispose);
1420 EXPORT_SYMBOL(check_submounts_and_drop);
1423 * __d_alloc - allocate a dcache entry
1424 * @sb: filesystem it will belong to
1425 * @name: qstr of the name
1427 * Allocates a dentry. It returns %NULL if there is insufficient memory
1428 * available. On a success the dentry is returned. The name passed in is
1429 * copied and the copy passed in may be reused after this call.
1432 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1434 struct dentry *dentry;
1437 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1442 * We guarantee that the inline name is always NUL-terminated.
1443 * This way the memcpy() done by the name switching in rename
1444 * will still always have a NUL at the end, even if we might
1445 * be overwriting an internal NUL character
1447 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1448 if (name->len > DNAME_INLINE_LEN-1) {
1449 dname = kmalloc(name->len + 1, GFP_KERNEL);
1451 kmem_cache_free(dentry_cache, dentry);
1455 dname = dentry->d_iname;
1458 dentry->d_name.len = name->len;
1459 dentry->d_name.hash = name->hash;
1460 memcpy(dname, name->name, name->len);
1461 dname[name->len] = 0;
1463 /* Make sure we always see the terminating NUL character */
1465 dentry->d_name.name = dname;
1467 dentry->d_lockref.count = 1;
1468 dentry->d_flags = 0;
1469 spin_lock_init(&dentry->d_lock);
1470 seqcount_init(&dentry->d_seq);
1471 dentry->d_inode = NULL;
1472 dentry->d_parent = dentry;
1474 dentry->d_op = NULL;
1475 dentry->d_fsdata = NULL;
1476 INIT_HLIST_BL_NODE(&dentry->d_hash);
1477 INIT_LIST_HEAD(&dentry->d_lru);
1478 INIT_LIST_HEAD(&dentry->d_subdirs);
1479 INIT_HLIST_NODE(&dentry->d_alias);
1480 INIT_LIST_HEAD(&dentry->d_u.d_child);
1481 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1483 this_cpu_inc(nr_dentry);
1489 * d_alloc - allocate a dcache entry
1490 * @parent: parent of entry to allocate
1491 * @name: qstr of the name
1493 * Allocates a dentry. It returns %NULL if there is insufficient memory
1494 * available. On a success the dentry is returned. The name passed in is
1495 * copied and the copy passed in may be reused after this call.
1497 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1499 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1503 spin_lock(&parent->d_lock);
1505 * don't need child lock because it is not subject
1506 * to concurrency here
1508 __dget_dlock(parent);
1509 dentry->d_parent = parent;
1510 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1511 spin_unlock(&parent->d_lock);
1515 EXPORT_SYMBOL(d_alloc);
1518 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1519 * @sb: the superblock
1520 * @name: qstr of the name
1522 * For a filesystem that just pins its dentries in memory and never
1523 * performs lookups at all, return an unhashed IS_ROOT dentry.
1525 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1527 return __d_alloc(sb, name);
1529 EXPORT_SYMBOL(d_alloc_pseudo);
1531 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1536 q.len = strlen(name);
1537 q.hash = full_name_hash(q.name, q.len);
1538 return d_alloc(parent, &q);
1540 EXPORT_SYMBOL(d_alloc_name);
1542 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1544 WARN_ON_ONCE(dentry->d_op);
1545 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1547 DCACHE_OP_REVALIDATE |
1548 DCACHE_OP_WEAK_REVALIDATE |
1549 DCACHE_OP_DELETE ));
1554 dentry->d_flags |= DCACHE_OP_HASH;
1556 dentry->d_flags |= DCACHE_OP_COMPARE;
1557 if (op->d_revalidate)
1558 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1559 if (op->d_weak_revalidate)
1560 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1562 dentry->d_flags |= DCACHE_OP_DELETE;
1564 dentry->d_flags |= DCACHE_OP_PRUNE;
1567 EXPORT_SYMBOL(d_set_d_op);
1569 static unsigned d_flags_for_inode(struct inode *inode)
1571 unsigned add_flags = DCACHE_FILE_TYPE;
1574 return DCACHE_MISS_TYPE;
1576 if (S_ISDIR(inode->i_mode)) {
1577 add_flags = DCACHE_DIRECTORY_TYPE;
1578 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1579 if (unlikely(!inode->i_op->lookup))
1580 add_flags = DCACHE_AUTODIR_TYPE;
1582 inode->i_opflags |= IOP_LOOKUP;
1584 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1585 if (unlikely(inode->i_op->follow_link))
1586 add_flags = DCACHE_SYMLINK_TYPE;
1588 inode->i_opflags |= IOP_NOFOLLOW;
1591 if (unlikely(IS_AUTOMOUNT(inode)))
1592 add_flags |= DCACHE_NEED_AUTOMOUNT;
1596 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1598 unsigned add_flags = d_flags_for_inode(inode);
1600 spin_lock(&dentry->d_lock);
1601 __d_set_type(dentry, add_flags);
1603 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1604 dentry->d_inode = inode;
1605 dentry_rcuwalk_barrier(dentry);
1606 spin_unlock(&dentry->d_lock);
1607 fsnotify_d_instantiate(dentry, inode);
1611 * d_instantiate - fill in inode information for a dentry
1612 * @entry: dentry to complete
1613 * @inode: inode to attach to this dentry
1615 * Fill in inode information in the entry.
1617 * This turns negative dentries into productive full members
1620 * NOTE! This assumes that the inode count has been incremented
1621 * (or otherwise set) by the caller to indicate that it is now
1622 * in use by the dcache.
1625 void d_instantiate(struct dentry *entry, struct inode * inode)
1627 BUG_ON(!hlist_unhashed(&entry->d_alias));
1629 spin_lock(&inode->i_lock);
1630 __d_instantiate(entry, inode);
1632 spin_unlock(&inode->i_lock);
1633 security_d_instantiate(entry, inode);
1635 EXPORT_SYMBOL(d_instantiate);
1638 * d_instantiate_unique - instantiate a non-aliased dentry
1639 * @entry: dentry to instantiate
1640 * @inode: inode to attach to this dentry
1642 * Fill in inode information in the entry. On success, it returns NULL.
1643 * If an unhashed alias of "entry" already exists, then we return the
1644 * aliased dentry instead and drop one reference to inode.
1646 * Note that in order to avoid conflicts with rename() etc, the caller
1647 * had better be holding the parent directory semaphore.
1649 * This also assumes that the inode count has been incremented
1650 * (or otherwise set) by the caller to indicate that it is now
1651 * in use by the dcache.
1653 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1654 struct inode *inode)
1656 struct dentry *alias;
1657 int len = entry->d_name.len;
1658 const char *name = entry->d_name.name;
1659 unsigned int hash = entry->d_name.hash;
1662 __d_instantiate(entry, NULL);
1666 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1668 * Don't need alias->d_lock here, because aliases with
1669 * d_parent == entry->d_parent are not subject to name or
1670 * parent changes, because the parent inode i_mutex is held.
1672 if (alias->d_name.hash != hash)
1674 if (alias->d_parent != entry->d_parent)
1676 if (alias->d_name.len != len)
1678 if (dentry_cmp(alias, name, len))
1684 __d_instantiate(entry, inode);
1688 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1690 struct dentry *result;
1692 BUG_ON(!hlist_unhashed(&entry->d_alias));
1695 spin_lock(&inode->i_lock);
1696 result = __d_instantiate_unique(entry, inode);
1698 spin_unlock(&inode->i_lock);
1701 security_d_instantiate(entry, inode);
1705 BUG_ON(!d_unhashed(result));
1710 EXPORT_SYMBOL(d_instantiate_unique);
1713 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1714 * @entry: dentry to complete
1715 * @inode: inode to attach to this dentry
1717 * Fill in inode information in the entry. If a directory alias is found, then
1718 * return an error (and drop inode). Together with d_materialise_unique() this
1719 * guarantees that a directory inode may never have more than one alias.
1721 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1723 BUG_ON(!hlist_unhashed(&entry->d_alias));
1725 spin_lock(&inode->i_lock);
1726 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1727 spin_unlock(&inode->i_lock);
1731 __d_instantiate(entry, inode);
1732 spin_unlock(&inode->i_lock);
1733 security_d_instantiate(entry, inode);
1737 EXPORT_SYMBOL(d_instantiate_no_diralias);
1739 struct dentry *d_make_root(struct inode *root_inode)
1741 struct dentry *res = NULL;
1744 static const struct qstr name = QSTR_INIT("/", 1);
1746 res = __d_alloc(root_inode->i_sb, &name);
1748 d_instantiate(res, root_inode);
1754 EXPORT_SYMBOL(d_make_root);
1756 static struct dentry * __d_find_any_alias(struct inode *inode)
1758 struct dentry *alias;
1760 if (hlist_empty(&inode->i_dentry))
1762 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1768 * d_find_any_alias - find any alias for a given inode
1769 * @inode: inode to find an alias for
1771 * If any aliases exist for the given inode, take and return a
1772 * reference for one of them. If no aliases exist, return %NULL.
1774 struct dentry *d_find_any_alias(struct inode *inode)
1778 spin_lock(&inode->i_lock);
1779 de = __d_find_any_alias(inode);
1780 spin_unlock(&inode->i_lock);
1783 EXPORT_SYMBOL(d_find_any_alias);
1786 * d_obtain_alias - find or allocate a dentry for a given inode
1787 * @inode: inode to allocate the dentry for
1789 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1790 * similar open by handle operations. The returned dentry may be anonymous,
1791 * or may have a full name (if the inode was already in the cache).
1793 * When called on a directory inode, we must ensure that the inode only ever
1794 * has one dentry. If a dentry is found, that is returned instead of
1795 * allocating a new one.
1797 * On successful return, the reference to the inode has been transferred
1798 * to the dentry. In case of an error the reference on the inode is released.
1799 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1800 * be passed in and will be the error will be propagate to the return value,
1801 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1803 struct dentry *d_obtain_alias(struct inode *inode)
1805 static const struct qstr anonstring = QSTR_INIT("/", 1);
1811 return ERR_PTR(-ESTALE);
1813 return ERR_CAST(inode);
1815 res = d_find_any_alias(inode);
1819 tmp = __d_alloc(inode->i_sb, &anonstring);
1821 res = ERR_PTR(-ENOMEM);
1825 spin_lock(&inode->i_lock);
1826 res = __d_find_any_alias(inode);
1828 spin_unlock(&inode->i_lock);
1833 /* attach a disconnected dentry */
1834 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1836 spin_lock(&tmp->d_lock);
1837 tmp->d_inode = inode;
1838 tmp->d_flags |= add_flags;
1839 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1840 hlist_bl_lock(&tmp->d_sb->s_anon);
1841 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1842 hlist_bl_unlock(&tmp->d_sb->s_anon);
1843 spin_unlock(&tmp->d_lock);
1844 spin_unlock(&inode->i_lock);
1845 security_d_instantiate(tmp, inode);
1850 if (res && !IS_ERR(res))
1851 security_d_instantiate(res, inode);
1855 EXPORT_SYMBOL(d_obtain_alias);
1858 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1859 * @inode: the inode which may have a disconnected dentry
1860 * @dentry: a negative dentry which we want to point to the inode.
1862 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1863 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1864 * and return it, else simply d_add the inode to the dentry and return NULL.
1866 * This is needed in the lookup routine of any filesystem that is exportable
1867 * (via knfsd) so that we can build dcache paths to directories effectively.
1869 * If a dentry was found and moved, then it is returned. Otherwise NULL
1870 * is returned. This matches the expected return value of ->lookup.
1872 * Cluster filesystems may call this function with a negative, hashed dentry.
1873 * In that case, we know that the inode will be a regular file, and also this
1874 * will only occur during atomic_open. So we need to check for the dentry
1875 * being already hashed only in the final case.
1877 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1879 struct dentry *new = NULL;
1882 return ERR_CAST(inode);
1884 if (inode && S_ISDIR(inode->i_mode)) {
1885 spin_lock(&inode->i_lock);
1886 new = __d_find_alias(inode, 1);
1888 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1889 spin_unlock(&inode->i_lock);
1890 security_d_instantiate(new, inode);
1891 d_move(new, dentry);
1894 /* already taking inode->i_lock, so d_add() by hand */
1895 __d_instantiate(dentry, inode);
1896 spin_unlock(&inode->i_lock);
1897 security_d_instantiate(dentry, inode);
1901 d_instantiate(dentry, inode);
1902 if (d_unhashed(dentry))
1907 EXPORT_SYMBOL(d_splice_alias);
1910 * d_add_ci - lookup or allocate new dentry with case-exact name
1911 * @inode: the inode case-insensitive lookup has found
1912 * @dentry: the negative dentry that was passed to the parent's lookup func
1913 * @name: the case-exact name to be associated with the returned dentry
1915 * This is to avoid filling the dcache with case-insensitive names to the
1916 * same inode, only the actual correct case is stored in the dcache for
1917 * case-insensitive filesystems.
1919 * For a case-insensitive lookup match and if the the case-exact dentry
1920 * already exists in in the dcache, use it and return it.
1922 * If no entry exists with the exact case name, allocate new dentry with
1923 * the exact case, and return the spliced entry.
1925 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1928 struct dentry *found;
1932 * First check if a dentry matching the name already exists,
1933 * if not go ahead and create it now.
1935 found = d_hash_and_lookup(dentry->d_parent, name);
1936 if (unlikely(IS_ERR(found)))
1939 new = d_alloc(dentry->d_parent, name);
1941 found = ERR_PTR(-ENOMEM);
1945 found = d_splice_alias(inode, new);
1954 * If a matching dentry exists, and it's not negative use it.
1956 * Decrement the reference count to balance the iget() done
1959 if (found->d_inode) {
1960 if (unlikely(found->d_inode != inode)) {
1961 /* This can't happen because bad inodes are unhashed. */
1962 BUG_ON(!is_bad_inode(inode));
1963 BUG_ON(!is_bad_inode(found->d_inode));
1970 * Negative dentry: instantiate it unless the inode is a directory and
1971 * already has a dentry.
1973 new = d_splice_alias(inode, found);
1984 EXPORT_SYMBOL(d_add_ci);
1987 * Do the slow-case of the dentry name compare.
1989 * Unlike the dentry_cmp() function, we need to atomically
1990 * load the name and length information, so that the
1991 * filesystem can rely on them, and can use the 'name' and
1992 * 'len' information without worrying about walking off the
1993 * end of memory etc.
1995 * Thus the read_seqcount_retry() and the "duplicate" info
1996 * in arguments (the low-level filesystem should not look
1997 * at the dentry inode or name contents directly, since
1998 * rename can change them while we're in RCU mode).
2000 enum slow_d_compare {
2006 static noinline enum slow_d_compare slow_dentry_cmp(
2007 const struct dentry *parent,
2008 struct dentry *dentry,
2010 const struct qstr *name)
2012 int tlen = dentry->d_name.len;
2013 const char *tname = dentry->d_name.name;
2015 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2017 return D_COMP_SEQRETRY;
2019 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2020 return D_COMP_NOMATCH;
2025 * __d_lookup_rcu - search for a dentry (racy, store-free)
2026 * @parent: parent dentry
2027 * @name: qstr of name we wish to find
2028 * @seqp: returns d_seq value at the point where the dentry was found
2029 * Returns: dentry, or NULL
2031 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2032 * resolution (store-free path walking) design described in
2033 * Documentation/filesystems/path-lookup.txt.
2035 * This is not to be used outside core vfs.
2037 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2038 * held, and rcu_read_lock held. The returned dentry must not be stored into
2039 * without taking d_lock and checking d_seq sequence count against @seq
2042 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2045 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2046 * the returned dentry, so long as its parent's seqlock is checked after the
2047 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2048 * is formed, giving integrity down the path walk.
2050 * NOTE! The caller *has* to check the resulting dentry against the sequence
2051 * number we've returned before using any of the resulting dentry state!
2053 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2054 const struct qstr *name,
2057 u64 hashlen = name->hash_len;
2058 const unsigned char *str = name->name;
2059 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2060 struct hlist_bl_node *node;
2061 struct dentry *dentry;
2064 * Note: There is significant duplication with __d_lookup_rcu which is
2065 * required to prevent single threaded performance regressions
2066 * especially on architectures where smp_rmb (in seqcounts) are costly.
2067 * Keep the two functions in sync.
2071 * The hash list is protected using RCU.
2073 * Carefully use d_seq when comparing a candidate dentry, to avoid
2074 * races with d_move().
2076 * It is possible that concurrent renames can mess up our list
2077 * walk here and result in missing our dentry, resulting in the
2078 * false-negative result. d_lookup() protects against concurrent
2079 * renames using rename_lock seqlock.
2081 * See Documentation/filesystems/path-lookup.txt for more details.
2083 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2088 * The dentry sequence count protects us from concurrent
2089 * renames, and thus protects parent and name fields.
2091 * The caller must perform a seqcount check in order
2092 * to do anything useful with the returned dentry.
2094 * NOTE! We do a "raw" seqcount_begin here. That means that
2095 * we don't wait for the sequence count to stabilize if it
2096 * is in the middle of a sequence change. If we do the slow
2097 * dentry compare, we will do seqretries until it is stable,
2098 * and if we end up with a successful lookup, we actually
2099 * want to exit RCU lookup anyway.
2101 seq = raw_seqcount_begin(&dentry->d_seq);
2102 if (dentry->d_parent != parent)
2104 if (d_unhashed(dentry))
2107 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2108 if (dentry->d_name.hash != hashlen_hash(hashlen))
2111 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2114 case D_COMP_NOMATCH:
2121 if (dentry->d_name.hash_len != hashlen)
2124 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2131 * d_lookup - search for a dentry
2132 * @parent: parent dentry
2133 * @name: qstr of name we wish to find
2134 * Returns: dentry, or NULL
2136 * d_lookup searches the children of the parent dentry for the name in
2137 * question. If the dentry is found its reference count is incremented and the
2138 * dentry is returned. The caller must use dput to free the entry when it has
2139 * finished using it. %NULL is returned if the dentry does not exist.
2141 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2143 struct dentry *dentry;
2147 seq = read_seqbegin(&rename_lock);
2148 dentry = __d_lookup(parent, name);
2151 } while (read_seqretry(&rename_lock, seq));
2154 EXPORT_SYMBOL(d_lookup);
2157 * __d_lookup - search for a dentry (racy)
2158 * @parent: parent dentry
2159 * @name: qstr of name we wish to find
2160 * Returns: dentry, or NULL
2162 * __d_lookup is like d_lookup, however it may (rarely) return a
2163 * false-negative result due to unrelated rename activity.
2165 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2166 * however it must be used carefully, eg. with a following d_lookup in
2167 * the case of failure.
2169 * __d_lookup callers must be commented.
2171 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2173 unsigned int len = name->len;
2174 unsigned int hash = name->hash;
2175 const unsigned char *str = name->name;
2176 struct hlist_bl_head *b = d_hash(parent, hash);
2177 struct hlist_bl_node *node;
2178 struct dentry *found = NULL;
2179 struct dentry *dentry;
2182 * Note: There is significant duplication with __d_lookup_rcu which is
2183 * required to prevent single threaded performance regressions
2184 * especially on architectures where smp_rmb (in seqcounts) are costly.
2185 * Keep the two functions in sync.
2189 * The hash list is protected using RCU.
2191 * Take d_lock when comparing a candidate dentry, to avoid races
2194 * It is possible that concurrent renames can mess up our list
2195 * walk here and result in missing our dentry, resulting in the
2196 * false-negative result. d_lookup() protects against concurrent
2197 * renames using rename_lock seqlock.
2199 * See Documentation/filesystems/path-lookup.txt for more details.
2203 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2205 if (dentry->d_name.hash != hash)
2208 spin_lock(&dentry->d_lock);
2209 if (dentry->d_parent != parent)
2211 if (d_unhashed(dentry))
2215 * It is safe to compare names since d_move() cannot
2216 * change the qstr (protected by d_lock).
2218 if (parent->d_flags & DCACHE_OP_COMPARE) {
2219 int tlen = dentry->d_name.len;
2220 const char *tname = dentry->d_name.name;
2221 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2224 if (dentry->d_name.len != len)
2226 if (dentry_cmp(dentry, str, len))
2230 dentry->d_lockref.count++;
2232 spin_unlock(&dentry->d_lock);
2235 spin_unlock(&dentry->d_lock);
2243 * d_hash_and_lookup - hash the qstr then search for a dentry
2244 * @dir: Directory to search in
2245 * @name: qstr of name we wish to find
2247 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2249 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2252 * Check for a fs-specific hash function. Note that we must
2253 * calculate the standard hash first, as the d_op->d_hash()
2254 * routine may choose to leave the hash value unchanged.
2256 name->hash = full_name_hash(name->name, name->len);
2257 if (dir->d_flags & DCACHE_OP_HASH) {
2258 int err = dir->d_op->d_hash(dir, name);
2259 if (unlikely(err < 0))
2260 return ERR_PTR(err);
2262 return d_lookup(dir, name);
2264 EXPORT_SYMBOL(d_hash_and_lookup);
2267 * d_validate - verify dentry provided from insecure source (deprecated)
2268 * @dentry: The dentry alleged to be valid child of @dparent
2269 * @dparent: The parent dentry (known to be valid)
2271 * An insecure source has sent us a dentry, here we verify it and dget() it.
2272 * This is used by ncpfs in its readdir implementation.
2273 * Zero is returned in the dentry is invalid.
2275 * This function is slow for big directories, and deprecated, do not use it.
2277 int d_validate(struct dentry *dentry, struct dentry *dparent)
2279 struct dentry *child;
2281 spin_lock(&dparent->d_lock);
2282 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2283 if (dentry == child) {
2284 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2285 __dget_dlock(dentry);
2286 spin_unlock(&dentry->d_lock);
2287 spin_unlock(&dparent->d_lock);
2291 spin_unlock(&dparent->d_lock);
2295 EXPORT_SYMBOL(d_validate);
2298 * When a file is deleted, we have two options:
2299 * - turn this dentry into a negative dentry
2300 * - unhash this dentry and free it.
2302 * Usually, we want to just turn this into
2303 * a negative dentry, but if anybody else is
2304 * currently using the dentry or the inode
2305 * we can't do that and we fall back on removing
2306 * it from the hash queues and waiting for
2307 * it to be deleted later when it has no users
2311 * d_delete - delete a dentry
2312 * @dentry: The dentry to delete
2314 * Turn the dentry into a negative dentry if possible, otherwise
2315 * remove it from the hash queues so it can be deleted later
2318 void d_delete(struct dentry * dentry)
2320 struct inode *inode;
2323 * Are we the only user?
2326 spin_lock(&dentry->d_lock);
2327 inode = dentry->d_inode;
2328 isdir = S_ISDIR(inode->i_mode);
2329 if (dentry->d_lockref.count == 1) {
2330 if (!spin_trylock(&inode->i_lock)) {
2331 spin_unlock(&dentry->d_lock);
2335 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2336 dentry_unlink_inode(dentry);
2337 fsnotify_nameremove(dentry, isdir);
2341 if (!d_unhashed(dentry))
2344 spin_unlock(&dentry->d_lock);
2346 fsnotify_nameremove(dentry, isdir);
2348 EXPORT_SYMBOL(d_delete);
2350 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2352 BUG_ON(!d_unhashed(entry));
2354 entry->d_flags |= DCACHE_RCUACCESS;
2355 hlist_bl_add_head_rcu(&entry->d_hash, b);
2359 static void _d_rehash(struct dentry * entry)
2361 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2365 * d_rehash - add an entry back to the hash
2366 * @entry: dentry to add to the hash
2368 * Adds a dentry to the hash according to its name.
2371 void d_rehash(struct dentry * entry)
2373 spin_lock(&entry->d_lock);
2375 spin_unlock(&entry->d_lock);
2377 EXPORT_SYMBOL(d_rehash);
2380 * dentry_update_name_case - update case insensitive dentry with a new name
2381 * @dentry: dentry to be updated
2384 * Update a case insensitive dentry with new case of name.
2386 * dentry must have been returned by d_lookup with name @name. Old and new
2387 * name lengths must match (ie. no d_compare which allows mismatched name
2390 * Parent inode i_mutex must be held over d_lookup and into this call (to
2391 * keep renames and concurrent inserts, and readdir(2) away).
2393 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2395 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2396 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2398 spin_lock(&dentry->d_lock);
2399 write_seqcount_begin(&dentry->d_seq);
2400 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2401 write_seqcount_end(&dentry->d_seq);
2402 spin_unlock(&dentry->d_lock);
2404 EXPORT_SYMBOL(dentry_update_name_case);
2406 static void switch_names(struct dentry *dentry, struct dentry *target)
2408 if (dname_external(target)) {
2409 if (dname_external(dentry)) {
2411 * Both external: swap the pointers
2413 swap(target->d_name.name, dentry->d_name.name);
2416 * dentry:internal, target:external. Steal target's
2417 * storage and make target internal.
2419 memcpy(target->d_iname, dentry->d_name.name,
2420 dentry->d_name.len + 1);
2421 dentry->d_name.name = target->d_name.name;
2422 target->d_name.name = target->d_iname;
2425 if (dname_external(dentry)) {
2427 * dentry:external, target:internal. Give dentry's
2428 * storage to target and make dentry internal
2430 memcpy(dentry->d_iname, target->d_name.name,
2431 target->d_name.len + 1);
2432 target->d_name.name = dentry->d_name.name;
2433 dentry->d_name.name = dentry->d_iname;
2436 * Both are internal.
2439 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2440 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2441 swap(((long *) &dentry->d_iname)[i],
2442 ((long *) &target->d_iname)[i]);
2446 swap(dentry->d_name.len, target->d_name.len);
2449 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2452 * XXXX: do we really need to take target->d_lock?
2454 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2455 spin_lock(&target->d_parent->d_lock);
2457 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2458 spin_lock(&dentry->d_parent->d_lock);
2459 spin_lock_nested(&target->d_parent->d_lock,
2460 DENTRY_D_LOCK_NESTED);
2462 spin_lock(&target->d_parent->d_lock);
2463 spin_lock_nested(&dentry->d_parent->d_lock,
2464 DENTRY_D_LOCK_NESTED);
2467 if (target < dentry) {
2468 spin_lock_nested(&target->d_lock, 2);
2469 spin_lock_nested(&dentry->d_lock, 3);
2471 spin_lock_nested(&dentry->d_lock, 2);
2472 spin_lock_nested(&target->d_lock, 3);
2476 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2477 struct dentry *target)
2479 if (target->d_parent != dentry->d_parent)
2480 spin_unlock(&dentry->d_parent->d_lock);
2481 if (target->d_parent != target)
2482 spin_unlock(&target->d_parent->d_lock);
2486 * When switching names, the actual string doesn't strictly have to
2487 * be preserved in the target - because we're dropping the target
2488 * anyway. As such, we can just do a simple memcpy() to copy over
2489 * the new name before we switch.
2491 * Note that we have to be a lot more careful about getting the hash
2492 * switched - we have to switch the hash value properly even if it
2493 * then no longer matches the actual (corrupted) string of the target.
2494 * The hash value has to match the hash queue that the dentry is on..
2497 * __d_move - move a dentry
2498 * @dentry: entry to move
2499 * @target: new dentry
2500 * @exchange: exchange the two dentries
2502 * Update the dcache to reflect the move of a file name. Negative
2503 * dcache entries should not be moved in this way. Caller must hold
2504 * rename_lock, the i_mutex of the source and target directories,
2505 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2507 static void __d_move(struct dentry *dentry, struct dentry *target,
2510 if (!dentry->d_inode)
2511 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2513 BUG_ON(d_ancestor(dentry, target));
2514 BUG_ON(d_ancestor(target, dentry));
2516 dentry_lock_for_move(dentry, target);
2518 write_seqcount_begin(&dentry->d_seq);
2519 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2521 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2524 * Move the dentry to the target hash queue. Don't bother checking
2525 * for the same hash queue because of how unlikely it is.
2528 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2531 * Unhash the target (d_delete() is not usable here). If exchanging
2532 * the two dentries, then rehash onto the other's hash queue.
2537 d_hash(dentry->d_parent, dentry->d_name.hash));
2540 list_del(&dentry->d_u.d_child);
2541 list_del(&target->d_u.d_child);
2543 /* Switch the names.. */
2544 switch_names(dentry, target);
2545 swap(dentry->d_name.hash, target->d_name.hash);
2547 /* ... and switch the parents */
2548 if (IS_ROOT(dentry)) {
2549 dentry->d_parent = target->d_parent;
2550 target->d_parent = target;
2551 INIT_LIST_HEAD(&target->d_u.d_child);
2553 swap(dentry->d_parent, target->d_parent);
2555 /* And add them back to the (new) parent lists */
2556 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2559 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2561 write_seqcount_end(&target->d_seq);
2562 write_seqcount_end(&dentry->d_seq);
2564 dentry_unlock_parents_for_move(dentry, target);
2566 fsnotify_d_move(target);
2567 spin_unlock(&target->d_lock);
2568 fsnotify_d_move(dentry);
2569 spin_unlock(&dentry->d_lock);
2573 * d_move - move a dentry
2574 * @dentry: entry to move
2575 * @target: new dentry
2577 * Update the dcache to reflect the move of a file name. Negative
2578 * dcache entries should not be moved in this way. See the locking
2579 * requirements for __d_move.
2581 void d_move(struct dentry *dentry, struct dentry *target)
2583 write_seqlock(&rename_lock);
2584 __d_move(dentry, target, false);
2585 write_sequnlock(&rename_lock);
2587 EXPORT_SYMBOL(d_move);
2590 * d_exchange - exchange two dentries
2591 * @dentry1: first dentry
2592 * @dentry2: second dentry
2594 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2596 write_seqlock(&rename_lock);
2598 WARN_ON(!dentry1->d_inode);
2599 WARN_ON(!dentry2->d_inode);
2600 WARN_ON(IS_ROOT(dentry1));
2601 WARN_ON(IS_ROOT(dentry2));
2603 __d_move(dentry1, dentry2, true);
2605 write_sequnlock(&rename_lock);
2609 * d_ancestor - search for an ancestor
2610 * @p1: ancestor dentry
2613 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2614 * an ancestor of p2, else NULL.
2616 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2620 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2621 if (p->d_parent == p1)
2628 * This helper attempts to cope with remotely renamed directories
2630 * It assumes that the caller is already holding
2631 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2633 * Note: If ever the locking in lock_rename() changes, then please
2634 * remember to update this too...
2636 static struct dentry *__d_unalias(struct inode *inode,
2637 struct dentry *dentry, struct dentry *alias)
2639 struct mutex *m1 = NULL, *m2 = NULL;
2640 struct dentry *ret = ERR_PTR(-EBUSY);
2642 /* If alias and dentry share a parent, then no extra locks required */
2643 if (alias->d_parent == dentry->d_parent)
2646 /* See lock_rename() */
2647 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2649 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2650 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2652 m2 = &alias->d_parent->d_inode->i_mutex;
2654 if (likely(!d_mountpoint(alias))) {
2655 __d_move(alias, dentry, false);
2659 spin_unlock(&inode->i_lock);
2668 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2669 * named dentry in place of the dentry to be replaced.
2670 * returns with anon->d_lock held!
2672 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2674 struct dentry *dparent;
2676 dentry_lock_for_move(anon, dentry);
2678 write_seqcount_begin(&dentry->d_seq);
2679 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2681 dparent = dentry->d_parent;
2683 switch_names(dentry, anon);
2684 swap(dentry->d_name.hash, anon->d_name.hash);
2686 dentry->d_parent = dentry;
2687 list_del_init(&dentry->d_u.d_child);
2688 anon->d_parent = dparent;
2689 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2691 write_seqcount_end(&dentry->d_seq);
2692 write_seqcount_end(&anon->d_seq);
2694 dentry_unlock_parents_for_move(anon, dentry);
2695 spin_unlock(&dentry->d_lock);
2697 /* anon->d_lock still locked, returns locked */
2701 * d_materialise_unique - introduce an inode into the tree
2702 * @dentry: candidate dentry
2703 * @inode: inode to bind to the dentry, to which aliases may be attached
2705 * Introduces an dentry into the tree, substituting an extant disconnected
2706 * root directory alias in its place if there is one. Caller must hold the
2707 * i_mutex of the parent directory.
2709 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2711 struct dentry *actual;
2713 BUG_ON(!d_unhashed(dentry));
2717 __d_instantiate(dentry, NULL);
2722 spin_lock(&inode->i_lock);
2724 if (S_ISDIR(inode->i_mode)) {
2725 struct dentry *alias;
2727 /* Does an aliased dentry already exist? */
2728 alias = __d_find_alias(inode, 0);
2731 write_seqlock(&rename_lock);
2733 if (d_ancestor(alias, dentry)) {
2734 /* Check for loops */
2735 actual = ERR_PTR(-ELOOP);
2736 spin_unlock(&inode->i_lock);
2737 } else if (IS_ROOT(alias)) {
2738 /* Is this an anonymous mountpoint that we
2739 * could splice into our tree? */
2740 __d_materialise_dentry(dentry, alias);
2741 write_sequnlock(&rename_lock);
2745 /* Nope, but we must(!) avoid directory
2746 * aliasing. This drops inode->i_lock */
2747 actual = __d_unalias(inode, dentry, alias);
2749 write_sequnlock(&rename_lock);
2750 if (IS_ERR(actual)) {
2751 if (PTR_ERR(actual) == -ELOOP)
2752 pr_warn_ratelimited(
2753 "VFS: Lookup of '%s' in %s %s"
2754 " would have caused loop\n",
2755 dentry->d_name.name,
2756 inode->i_sb->s_type->name,
2764 /* Add a unique reference */
2765 actual = __d_instantiate_unique(dentry, inode);
2769 BUG_ON(!d_unhashed(actual));
2771 spin_lock(&actual->d_lock);
2774 spin_unlock(&actual->d_lock);
2775 spin_unlock(&inode->i_lock);
2777 if (actual == dentry) {
2778 security_d_instantiate(dentry, inode);
2785 EXPORT_SYMBOL_GPL(d_materialise_unique);
2787 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2791 return -ENAMETOOLONG;
2793 memcpy(*buffer, str, namelen);
2798 * prepend_name - prepend a pathname in front of current buffer pointer
2799 * @buffer: buffer pointer
2800 * @buflen: allocated length of the buffer
2801 * @name: name string and length qstr structure
2803 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2804 * make sure that either the old or the new name pointer and length are
2805 * fetched. However, there may be mismatch between length and pointer.
2806 * The length cannot be trusted, we need to copy it byte-by-byte until
2807 * the length is reached or a null byte is found. It also prepends "/" at
2808 * the beginning of the name. The sequence number check at the caller will
2809 * retry it again when a d_move() does happen. So any garbage in the buffer
2810 * due to mismatched pointer and length will be discarded.
2812 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2814 const char *dname = ACCESS_ONCE(name->name);
2815 u32 dlen = ACCESS_ONCE(name->len);
2818 *buflen -= dlen + 1;
2820 return -ENAMETOOLONG;
2821 p = *buffer -= dlen + 1;
2833 * prepend_path - Prepend path string to a buffer
2834 * @path: the dentry/vfsmount to report
2835 * @root: root vfsmnt/dentry
2836 * @buffer: pointer to the end of the buffer
2837 * @buflen: pointer to buffer length
2839 * The function will first try to write out the pathname without taking any
2840 * lock other than the RCU read lock to make sure that dentries won't go away.
2841 * It only checks the sequence number of the global rename_lock as any change
2842 * in the dentry's d_seq will be preceded by changes in the rename_lock
2843 * sequence number. If the sequence number had been changed, it will restart
2844 * the whole pathname back-tracing sequence again by taking the rename_lock.
2845 * In this case, there is no need to take the RCU read lock as the recursive
2846 * parent pointer references will keep the dentry chain alive as long as no
2847 * rename operation is performed.
2849 static int prepend_path(const struct path *path,
2850 const struct path *root,
2851 char **buffer, int *buflen)
2853 struct dentry *dentry;
2854 struct vfsmount *vfsmnt;
2857 unsigned seq, m_seq = 0;
2863 read_seqbegin_or_lock(&mount_lock, &m_seq);
2870 dentry = path->dentry;
2872 mnt = real_mount(vfsmnt);
2873 read_seqbegin_or_lock(&rename_lock, &seq);
2874 while (dentry != root->dentry || vfsmnt != root->mnt) {
2875 struct dentry * parent;
2877 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2878 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2880 if (mnt != parent) {
2881 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2887 * Filesystems needing to implement special "root names"
2888 * should do so with ->d_dname()
2890 if (IS_ROOT(dentry) &&
2891 (dentry->d_name.len != 1 ||
2892 dentry->d_name.name[0] != '/')) {
2893 WARN(1, "Root dentry has weird name <%.*s>\n",
2894 (int) dentry->d_name.len,
2895 dentry->d_name.name);
2898 error = is_mounted(vfsmnt) ? 1 : 2;
2901 parent = dentry->d_parent;
2903 error = prepend_name(&bptr, &blen, &dentry->d_name);
2911 if (need_seqretry(&rename_lock, seq)) {
2915 done_seqretry(&rename_lock, seq);
2919 if (need_seqretry(&mount_lock, m_seq)) {
2923 done_seqretry(&mount_lock, m_seq);
2925 if (error >= 0 && bptr == *buffer) {
2927 error = -ENAMETOOLONG;
2937 * __d_path - return the path of a dentry
2938 * @path: the dentry/vfsmount to report
2939 * @root: root vfsmnt/dentry
2940 * @buf: buffer to return value in
2941 * @buflen: buffer length
2943 * Convert a dentry into an ASCII path name.
2945 * Returns a pointer into the buffer or an error code if the
2946 * path was too long.
2948 * "buflen" should be positive.
2950 * If the path is not reachable from the supplied root, return %NULL.
2952 char *__d_path(const struct path *path,
2953 const struct path *root,
2954 char *buf, int buflen)
2956 char *res = buf + buflen;
2959 prepend(&res, &buflen, "\0", 1);
2960 error = prepend_path(path, root, &res, &buflen);
2963 return ERR_PTR(error);
2969 char *d_absolute_path(const struct path *path,
2970 char *buf, int buflen)
2972 struct path root = {};
2973 char *res = buf + buflen;
2976 prepend(&res, &buflen, "\0", 1);
2977 error = prepend_path(path, &root, &res, &buflen);
2982 return ERR_PTR(error);
2987 * same as __d_path but appends "(deleted)" for unlinked files.
2989 static int path_with_deleted(const struct path *path,
2990 const struct path *root,
2991 char **buf, int *buflen)
2993 prepend(buf, buflen, "\0", 1);
2994 if (d_unlinked(path->dentry)) {
2995 int error = prepend(buf, buflen, " (deleted)", 10);
3000 return prepend_path(path, root, buf, buflen);
3003 static int prepend_unreachable(char **buffer, int *buflen)
3005 return prepend(buffer, buflen, "(unreachable)", 13);
3008 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3013 seq = read_seqcount_begin(&fs->seq);
3015 } while (read_seqcount_retry(&fs->seq, seq));
3019 * d_path - return the path of a dentry
3020 * @path: path to report
3021 * @buf: buffer to return value in
3022 * @buflen: buffer length
3024 * Convert a dentry into an ASCII path name. If the entry has been deleted
3025 * the string " (deleted)" is appended. Note that this is ambiguous.
3027 * Returns a pointer into the buffer or an error code if the path was
3028 * too long. Note: Callers should use the returned pointer, not the passed
3029 * in buffer, to use the name! The implementation often starts at an offset
3030 * into the buffer, and may leave 0 bytes at the start.
3032 * "buflen" should be positive.
3034 char *d_path(const struct path *path, char *buf, int buflen)
3036 char *res = buf + buflen;
3041 * We have various synthetic filesystems that never get mounted. On
3042 * these filesystems dentries are never used for lookup purposes, and
3043 * thus don't need to be hashed. They also don't need a name until a
3044 * user wants to identify the object in /proc/pid/fd/. The little hack
3045 * below allows us to generate a name for these objects on demand:
3047 * Some pseudo inodes are mountable. When they are mounted
3048 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3049 * and instead have d_path return the mounted path.
3051 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3052 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3053 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3056 get_fs_root_rcu(current->fs, &root);
3057 error = path_with_deleted(path, &root, &res, &buflen);
3061 res = ERR_PTR(error);
3064 EXPORT_SYMBOL(d_path);
3067 * Helper function for dentry_operations.d_dname() members
3069 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3070 const char *fmt, ...)
3076 va_start(args, fmt);
3077 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3080 if (sz > sizeof(temp) || sz > buflen)
3081 return ERR_PTR(-ENAMETOOLONG);
3083 buffer += buflen - sz;
3084 return memcpy(buffer, temp, sz);
3087 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3089 char *end = buffer + buflen;
3090 /* these dentries are never renamed, so d_lock is not needed */
3091 if (prepend(&end, &buflen, " (deleted)", 11) ||
3092 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3093 prepend(&end, &buflen, "/", 1))
3094 end = ERR_PTR(-ENAMETOOLONG);
3097 EXPORT_SYMBOL(simple_dname);
3100 * Write full pathname from the root of the filesystem into the buffer.
3102 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3104 struct dentry *dentry;
3117 prepend(&end, &len, "\0", 1);
3121 read_seqbegin_or_lock(&rename_lock, &seq);
3122 while (!IS_ROOT(dentry)) {
3123 struct dentry *parent = dentry->d_parent;
3126 error = prepend_name(&end, &len, &dentry->d_name);
3135 if (need_seqretry(&rename_lock, seq)) {
3139 done_seqretry(&rename_lock, seq);
3144 return ERR_PTR(-ENAMETOOLONG);
3147 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3149 return __dentry_path(dentry, buf, buflen);
3151 EXPORT_SYMBOL(dentry_path_raw);
3153 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3158 if (d_unlinked(dentry)) {
3160 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3164 retval = __dentry_path(dentry, buf, buflen);
3165 if (!IS_ERR(retval) && p)
3166 *p = '/'; /* restore '/' overriden with '\0' */
3169 return ERR_PTR(-ENAMETOOLONG);
3172 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3178 seq = read_seqcount_begin(&fs->seq);
3181 } while (read_seqcount_retry(&fs->seq, seq));
3185 * NOTE! The user-level library version returns a
3186 * character pointer. The kernel system call just
3187 * returns the length of the buffer filled (which
3188 * includes the ending '\0' character), or a negative
3189 * error value. So libc would do something like
3191 * char *getcwd(char * buf, size_t size)
3195 * retval = sys_getcwd(buf, size);
3202 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3205 struct path pwd, root;
3206 char *page = __getname();
3212 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3215 if (!d_unlinked(pwd.dentry)) {
3217 char *cwd = page + PATH_MAX;
3218 int buflen = PATH_MAX;
3220 prepend(&cwd, &buflen, "\0", 1);
3221 error = prepend_path(&pwd, &root, &cwd, &buflen);
3227 /* Unreachable from current root */
3229 error = prepend_unreachable(&cwd, &buflen);
3235 len = PATH_MAX + page - cwd;
3238 if (copy_to_user(buf, cwd, len))
3251 * Test whether new_dentry is a subdirectory of old_dentry.
3253 * Trivially implemented using the dcache structure
3257 * is_subdir - is new dentry a subdirectory of old_dentry
3258 * @new_dentry: new dentry
3259 * @old_dentry: old dentry
3261 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3262 * Returns 0 otherwise.
3263 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3266 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3271 if (new_dentry == old_dentry)
3275 /* for restarting inner loop in case of seq retry */
3276 seq = read_seqbegin(&rename_lock);
3278 * Need rcu_readlock to protect against the d_parent trashing
3282 if (d_ancestor(old_dentry, new_dentry))
3287 } while (read_seqretry(&rename_lock, seq));
3292 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3294 struct dentry *root = data;
3295 if (dentry != root) {
3296 if (d_unhashed(dentry) || !dentry->d_inode)
3299 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3300 dentry->d_flags |= DCACHE_GENOCIDE;
3301 dentry->d_lockref.count--;
3304 return D_WALK_CONTINUE;
3307 void d_genocide(struct dentry *parent)
3309 d_walk(parent, parent, d_genocide_kill, NULL);
3312 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3314 inode_dec_link_count(inode);
3315 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3316 !hlist_unhashed(&dentry->d_alias) ||
3317 !d_unlinked(dentry));
3318 spin_lock(&dentry->d_parent->d_lock);
3319 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3320 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3321 (unsigned long long)inode->i_ino);
3322 spin_unlock(&dentry->d_lock);
3323 spin_unlock(&dentry->d_parent->d_lock);
3324 d_instantiate(dentry, inode);
3326 EXPORT_SYMBOL(d_tmpfile);
3328 static __initdata unsigned long dhash_entries;
3329 static int __init set_dhash_entries(char *str)
3333 dhash_entries = simple_strtoul(str, &str, 0);
3336 __setup("dhash_entries=", set_dhash_entries);
3338 static void __init dcache_init_early(void)
3342 /* If hashes are distributed across NUMA nodes, defer
3343 * hash allocation until vmalloc space is available.
3349 alloc_large_system_hash("Dentry cache",
3350 sizeof(struct hlist_bl_head),
3359 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3360 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3363 static void __init dcache_init(void)
3368 * A constructor could be added for stable state like the lists,
3369 * but it is probably not worth it because of the cache nature
3372 dentry_cache = KMEM_CACHE(dentry,
3373 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3375 /* Hash may have been set up in dcache_init_early */
3380 alloc_large_system_hash("Dentry cache",
3381 sizeof(struct hlist_bl_head),
3390 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3391 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3394 /* SLAB cache for __getname() consumers */
3395 struct kmem_cache *names_cachep __read_mostly;
3396 EXPORT_SYMBOL(names_cachep);
3398 EXPORT_SYMBOL(d_genocide);
3400 void __init vfs_caches_init_early(void)
3402 dcache_init_early();
3406 void __init vfs_caches_init(unsigned long mempages)
3408 unsigned long reserve;
3410 /* Base hash sizes on available memory, with a reserve equal to
3411 150% of current kernel size */
3413 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3414 mempages -= reserve;
3416 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3417 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3421 files_init(mempages);