block, cfq: restructure io_cq creation path for io_context interface cleanup
[linux-2.6.git] / fs / dcache.c
blob89509b5a090e27320e45b9c0c2f5480e082b1a37
1 /*
2 * fs/dcache.c
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
9 /*
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>
19 #include <linux/mm.h>
20 #include <linux/fs.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/module.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 "internal.h"
43 * Usage:
44 * dcache->d_inode->i_lock protects:
45 * - i_dentry, d_alias, d_inode of aliases
46 * dcache_hash_bucket lock protects:
47 * - the dcache hash table
48 * s_anon bl list spinlock protects:
49 * - the s_anon list (see __d_drop)
50 * dcache_lru_lock protects:
51 * - the dcache lru lists and counters
52 * d_lock protects:
53 * - d_flags
54 * - d_name
55 * - d_lru
56 * - d_count
57 * - d_unhashed()
58 * - d_parent and d_subdirs
59 * - childrens' d_child and d_parent
60 * - d_alias, d_inode
62 * Ordering:
63 * dentry->d_inode->i_lock
64 * dentry->d_lock
65 * dcache_lru_lock
66 * dcache_hash_bucket lock
67 * s_anon lock
69 * If there is an ancestor relationship:
70 * dentry->d_parent->...->d_parent->d_lock
71 * ...
72 * dentry->d_parent->d_lock
73 * dentry->d_lock
75 * If no ancestor relationship:
76 * if (dentry1 < dentry2)
77 * dentry1->d_lock
78 * dentry2->d_lock
80 int sysctl_vfs_cache_pressure __read_mostly = 100;
81 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
83 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
84 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
86 EXPORT_SYMBOL(rename_lock);
88 static struct kmem_cache *dentry_cache __read_mostly;
91 * This is the single most critical data structure when it comes
92 * to the dcache: the hashtable for lookups. Somebody should try
93 * to make this good - I've just made it work.
95 * This hash-function tries to avoid losing too many bits of hash
96 * information, yet avoid using a prime hash-size or similar.
98 #define D_HASHBITS d_hash_shift
99 #define D_HASHMASK d_hash_mask
101 static unsigned int d_hash_mask __read_mostly;
102 static unsigned int d_hash_shift __read_mostly;
104 static struct hlist_bl_head *dentry_hashtable __read_mostly;
106 static inline struct hlist_bl_head *d_hash(struct dentry *parent,
107 unsigned long hash)
109 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
110 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
111 return dentry_hashtable + (hash & D_HASHMASK);
114 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat = {
116 .age_limit = 45,
119 static DEFINE_PER_CPU(unsigned int, nr_dentry);
121 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
122 static int get_nr_dentry(void)
124 int i;
125 int sum = 0;
126 for_each_possible_cpu(i)
127 sum += per_cpu(nr_dentry, i);
128 return sum < 0 ? 0 : sum;
131 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
132 size_t *lenp, loff_t *ppos)
134 dentry_stat.nr_dentry = get_nr_dentry();
135 return proc_dointvec(table, write, buffer, lenp, ppos);
137 #endif
139 static void __d_free(struct rcu_head *head)
141 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
143 WARN_ON(!list_empty(&dentry->d_alias));
144 if (dname_external(dentry))
145 kfree(dentry->d_name.name);
146 kmem_cache_free(dentry_cache, dentry);
150 * no locks, please.
152 static void d_free(struct dentry *dentry)
154 BUG_ON(dentry->d_count);
155 this_cpu_dec(nr_dentry);
156 if (dentry->d_op && dentry->d_op->d_release)
157 dentry->d_op->d_release(dentry);
159 /* if dentry was never visible to RCU, immediate free is OK */
160 if (!(dentry->d_flags & DCACHE_RCUACCESS))
161 __d_free(&dentry->d_u.d_rcu);
162 else
163 call_rcu(&dentry->d_u.d_rcu, __d_free);
167 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
168 * @dentry: the target dentry
169 * After this call, in-progress rcu-walk path lookup will fail. This
170 * should be called after unhashing, and after changing d_inode (if
171 * the dentry has not already been unhashed).
173 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
175 assert_spin_locked(&dentry->d_lock);
176 /* Go through a barrier */
177 write_seqcount_barrier(&dentry->d_seq);
181 * Release the dentry's inode, using the filesystem
182 * d_iput() operation if defined. Dentry has no refcount
183 * and is unhashed.
185 static void dentry_iput(struct dentry * dentry)
186 __releases(dentry->d_lock)
187 __releases(dentry->d_inode->i_lock)
189 struct inode *inode = dentry->d_inode;
190 if (inode) {
191 dentry->d_inode = NULL;
192 list_del_init(&dentry->d_alias);
193 spin_unlock(&dentry->d_lock);
194 spin_unlock(&inode->i_lock);
195 if (!inode->i_nlink)
196 fsnotify_inoderemove(inode);
197 if (dentry->d_op && dentry->d_op->d_iput)
198 dentry->d_op->d_iput(dentry, inode);
199 else
200 iput(inode);
201 } else {
202 spin_unlock(&dentry->d_lock);
207 * Release the dentry's inode, using the filesystem
208 * d_iput() operation if defined. dentry remains in-use.
210 static void dentry_unlink_inode(struct dentry * dentry)
211 __releases(dentry->d_lock)
212 __releases(dentry->d_inode->i_lock)
214 struct inode *inode = dentry->d_inode;
215 dentry->d_inode = NULL;
216 list_del_init(&dentry->d_alias);
217 dentry_rcuwalk_barrier(dentry);
218 spin_unlock(&dentry->d_lock);
219 spin_unlock(&inode->i_lock);
220 if (!inode->i_nlink)
221 fsnotify_inoderemove(inode);
222 if (dentry->d_op && dentry->d_op->d_iput)
223 dentry->d_op->d_iput(dentry, inode);
224 else
225 iput(inode);
229 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
231 static void dentry_lru_add(struct dentry *dentry)
233 if (list_empty(&dentry->d_lru)) {
234 spin_lock(&dcache_lru_lock);
235 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
236 dentry->d_sb->s_nr_dentry_unused++;
237 dentry_stat.nr_unused++;
238 spin_unlock(&dcache_lru_lock);
242 static void __dentry_lru_del(struct dentry *dentry)
244 list_del_init(&dentry->d_lru);
245 dentry->d_sb->s_nr_dentry_unused--;
246 dentry_stat.nr_unused--;
250 * Remove a dentry with references from the LRU.
252 static void dentry_lru_del(struct dentry *dentry)
254 if (!list_empty(&dentry->d_lru)) {
255 spin_lock(&dcache_lru_lock);
256 __dentry_lru_del(dentry);
257 spin_unlock(&dcache_lru_lock);
262 * Remove a dentry that is unreferenced and about to be pruned
263 * (unhashed and destroyed) from the LRU, and inform the file system.
264 * This wrapper should be called _prior_ to unhashing a victim dentry.
266 static void dentry_lru_prune(struct dentry *dentry)
268 if (!list_empty(&dentry->d_lru)) {
269 if (dentry->d_flags & DCACHE_OP_PRUNE)
270 dentry->d_op->d_prune(dentry);
272 spin_lock(&dcache_lru_lock);
273 __dentry_lru_del(dentry);
274 spin_unlock(&dcache_lru_lock);
278 static void dentry_lru_move_tail(struct dentry *dentry)
280 spin_lock(&dcache_lru_lock);
281 if (list_empty(&dentry->d_lru)) {
282 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
283 dentry->d_sb->s_nr_dentry_unused++;
284 dentry_stat.nr_unused++;
285 } else {
286 list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
288 spin_unlock(&dcache_lru_lock);
292 * d_kill - kill dentry and return parent
293 * @dentry: dentry to kill
294 * @parent: parent dentry
296 * The dentry must already be unhashed and removed from the LRU.
298 * If this is the root of the dentry tree, return NULL.
300 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
301 * d_kill.
303 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
304 __releases(dentry->d_lock)
305 __releases(parent->d_lock)
306 __releases(dentry->d_inode->i_lock)
308 list_del(&dentry->d_u.d_child);
310 * Inform try_to_ascend() that we are no longer attached to the
311 * dentry tree
313 dentry->d_flags |= DCACHE_DISCONNECTED;
314 if (parent)
315 spin_unlock(&parent->d_lock);
316 dentry_iput(dentry);
318 * dentry_iput drops the locks, at which point nobody (except
319 * transient RCU lookups) can reach this dentry.
321 d_free(dentry);
322 return parent;
326 * Unhash a dentry without inserting an RCU walk barrier or checking that
327 * dentry->d_lock is locked. The caller must take care of that, if
328 * appropriate.
330 static void __d_shrink(struct dentry *dentry)
332 if (!d_unhashed(dentry)) {
333 struct hlist_bl_head *b;
334 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
335 b = &dentry->d_sb->s_anon;
336 else
337 b = d_hash(dentry->d_parent, dentry->d_name.hash);
339 hlist_bl_lock(b);
340 __hlist_bl_del(&dentry->d_hash);
341 dentry->d_hash.pprev = NULL;
342 hlist_bl_unlock(b);
347 * d_drop - drop a dentry
348 * @dentry: dentry to drop
350 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
351 * be found through a VFS lookup any more. Note that this is different from
352 * deleting the dentry - d_delete will try to mark the dentry negative if
353 * possible, giving a successful _negative_ lookup, while d_drop will
354 * just make the cache lookup fail.
356 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
357 * reason (NFS timeouts or autofs deletes).
359 * __d_drop requires dentry->d_lock.
361 void __d_drop(struct dentry *dentry)
363 if (!d_unhashed(dentry)) {
364 __d_shrink(dentry);
365 dentry_rcuwalk_barrier(dentry);
368 EXPORT_SYMBOL(__d_drop);
370 void d_drop(struct dentry *dentry)
372 spin_lock(&dentry->d_lock);
373 __d_drop(dentry);
374 spin_unlock(&dentry->d_lock);
376 EXPORT_SYMBOL(d_drop);
379 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
380 * @dentry: dentry to drop
382 * This is called when we do a lookup on a placeholder dentry that needed to be
383 * looked up. The dentry should have been hashed in order for it to be found by
384 * the lookup code, but now needs to be unhashed while we do the actual lookup
385 * and clear the DCACHE_NEED_LOOKUP flag.
387 void d_clear_need_lookup(struct dentry *dentry)
389 spin_lock(&dentry->d_lock);
390 __d_drop(dentry);
391 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
392 spin_unlock(&dentry->d_lock);
394 EXPORT_SYMBOL(d_clear_need_lookup);
397 * Finish off a dentry we've decided to kill.
398 * dentry->d_lock must be held, returns with it unlocked.
399 * If ref is non-zero, then decrement the refcount too.
400 * Returns dentry requiring refcount drop, or NULL if we're done.
402 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
403 __releases(dentry->d_lock)
405 struct inode *inode;
406 struct dentry *parent;
408 inode = dentry->d_inode;
409 if (inode && !spin_trylock(&inode->i_lock)) {
410 relock:
411 spin_unlock(&dentry->d_lock);
412 cpu_relax();
413 return dentry; /* try again with same dentry */
415 if (IS_ROOT(dentry))
416 parent = NULL;
417 else
418 parent = dentry->d_parent;
419 if (parent && !spin_trylock(&parent->d_lock)) {
420 if (inode)
421 spin_unlock(&inode->i_lock);
422 goto relock;
425 if (ref)
426 dentry->d_count--;
428 * if dentry was on the d_lru list delete it from there.
429 * inform the fs via d_prune that this dentry is about to be
430 * unhashed and destroyed.
432 dentry_lru_prune(dentry);
433 /* if it was on the hash then remove it */
434 __d_drop(dentry);
435 return d_kill(dentry, parent);
439 * This is dput
441 * This is complicated by the fact that we do not want to put
442 * dentries that are no longer on any hash chain on the unused
443 * list: we'd much rather just get rid of them immediately.
445 * However, that implies that we have to traverse the dentry
446 * tree upwards to the parents which might _also_ now be
447 * scheduled for deletion (it may have been only waiting for
448 * its last child to go away).
450 * This tail recursion is done by hand as we don't want to depend
451 * on the compiler to always get this right (gcc generally doesn't).
452 * Real recursion would eat up our stack space.
456 * dput - release a dentry
457 * @dentry: dentry to release
459 * Release a dentry. This will drop the usage count and if appropriate
460 * call the dentry unlink method as well as removing it from the queues and
461 * releasing its resources. If the parent dentries were scheduled for release
462 * they too may now get deleted.
464 void dput(struct dentry *dentry)
466 if (!dentry)
467 return;
469 repeat:
470 if (dentry->d_count == 1)
471 might_sleep();
472 spin_lock(&dentry->d_lock);
473 BUG_ON(!dentry->d_count);
474 if (dentry->d_count > 1) {
475 dentry->d_count--;
476 spin_unlock(&dentry->d_lock);
477 return;
480 if (dentry->d_flags & DCACHE_OP_DELETE) {
481 if (dentry->d_op->d_delete(dentry))
482 goto kill_it;
485 /* Unreachable? Get rid of it */
486 if (d_unhashed(dentry))
487 goto kill_it;
490 * If this dentry needs lookup, don't set the referenced flag so that it
491 * is more likely to be cleaned up by the dcache shrinker in case of
492 * memory pressure.
494 if (!d_need_lookup(dentry))
495 dentry->d_flags |= DCACHE_REFERENCED;
496 dentry_lru_add(dentry);
498 dentry->d_count--;
499 spin_unlock(&dentry->d_lock);
500 return;
502 kill_it:
503 dentry = dentry_kill(dentry, 1);
504 if (dentry)
505 goto repeat;
507 EXPORT_SYMBOL(dput);
510 * d_invalidate - invalidate a dentry
511 * @dentry: dentry to invalidate
513 * Try to invalidate the dentry if it turns out to be
514 * possible. If there are other dentries that can be
515 * reached through this one we can't delete it and we
516 * return -EBUSY. On success we return 0.
518 * no dcache lock.
521 int d_invalidate(struct dentry * dentry)
524 * If it's already been dropped, return OK.
526 spin_lock(&dentry->d_lock);
527 if (d_unhashed(dentry)) {
528 spin_unlock(&dentry->d_lock);
529 return 0;
532 * Check whether to do a partial shrink_dcache
533 * to get rid of unused child entries.
535 if (!list_empty(&dentry->d_subdirs)) {
536 spin_unlock(&dentry->d_lock);
537 shrink_dcache_parent(dentry);
538 spin_lock(&dentry->d_lock);
542 * Somebody else still using it?
544 * If it's a directory, we can't drop it
545 * for fear of somebody re-populating it
546 * with children (even though dropping it
547 * would make it unreachable from the root,
548 * we might still populate it if it was a
549 * working directory or similar).
550 * We also need to leave mountpoints alone,
551 * directory or not.
553 if (dentry->d_count > 1 && dentry->d_inode) {
554 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
555 spin_unlock(&dentry->d_lock);
556 return -EBUSY;
560 __d_drop(dentry);
561 spin_unlock(&dentry->d_lock);
562 return 0;
564 EXPORT_SYMBOL(d_invalidate);
566 /* This must be called with d_lock held */
567 static inline void __dget_dlock(struct dentry *dentry)
569 dentry->d_count++;
572 static inline void __dget(struct dentry *dentry)
574 spin_lock(&dentry->d_lock);
575 __dget_dlock(dentry);
576 spin_unlock(&dentry->d_lock);
579 struct dentry *dget_parent(struct dentry *dentry)
581 struct dentry *ret;
583 repeat:
585 * Don't need rcu_dereference because we re-check it was correct under
586 * the lock.
588 rcu_read_lock();
589 ret = dentry->d_parent;
590 spin_lock(&ret->d_lock);
591 if (unlikely(ret != dentry->d_parent)) {
592 spin_unlock(&ret->d_lock);
593 rcu_read_unlock();
594 goto repeat;
596 rcu_read_unlock();
597 BUG_ON(!ret->d_count);
598 ret->d_count++;
599 spin_unlock(&ret->d_lock);
600 return ret;
602 EXPORT_SYMBOL(dget_parent);
605 * d_find_alias - grab a hashed alias of inode
606 * @inode: inode in question
607 * @want_discon: flag, used by d_splice_alias, to request
608 * that only a DISCONNECTED alias be returned.
610 * If inode has a hashed alias, or is a directory and has any alias,
611 * acquire the reference to alias and return it. Otherwise return NULL.
612 * Notice that if inode is a directory there can be only one alias and
613 * it can be unhashed only if it has no children, or if it is the root
614 * of a filesystem.
616 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
617 * any other hashed alias over that one unless @want_discon is set,
618 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
620 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
622 struct dentry *alias, *discon_alias;
624 again:
625 discon_alias = NULL;
626 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
627 spin_lock(&alias->d_lock);
628 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
629 if (IS_ROOT(alias) &&
630 (alias->d_flags & DCACHE_DISCONNECTED)) {
631 discon_alias = alias;
632 } else if (!want_discon) {
633 __dget_dlock(alias);
634 spin_unlock(&alias->d_lock);
635 return alias;
638 spin_unlock(&alias->d_lock);
640 if (discon_alias) {
641 alias = discon_alias;
642 spin_lock(&alias->d_lock);
643 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
644 if (IS_ROOT(alias) &&
645 (alias->d_flags & DCACHE_DISCONNECTED)) {
646 __dget_dlock(alias);
647 spin_unlock(&alias->d_lock);
648 return alias;
651 spin_unlock(&alias->d_lock);
652 goto again;
654 return NULL;
657 struct dentry *d_find_alias(struct inode *inode)
659 struct dentry *de = NULL;
661 if (!list_empty(&inode->i_dentry)) {
662 spin_lock(&inode->i_lock);
663 de = __d_find_alias(inode, 0);
664 spin_unlock(&inode->i_lock);
666 return de;
668 EXPORT_SYMBOL(d_find_alias);
671 * Try to kill dentries associated with this inode.
672 * WARNING: you must own a reference to inode.
674 void d_prune_aliases(struct inode *inode)
676 struct dentry *dentry;
677 restart:
678 spin_lock(&inode->i_lock);
679 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
680 spin_lock(&dentry->d_lock);
681 if (!dentry->d_count) {
682 __dget_dlock(dentry);
683 __d_drop(dentry);
684 spin_unlock(&dentry->d_lock);
685 spin_unlock(&inode->i_lock);
686 dput(dentry);
687 goto restart;
689 spin_unlock(&dentry->d_lock);
691 spin_unlock(&inode->i_lock);
693 EXPORT_SYMBOL(d_prune_aliases);
696 * Try to throw away a dentry - free the inode, dput the parent.
697 * Requires dentry->d_lock is held, and dentry->d_count == 0.
698 * Releases dentry->d_lock.
700 * This may fail if locks cannot be acquired no problem, just try again.
702 static void try_prune_one_dentry(struct dentry *dentry)
703 __releases(dentry->d_lock)
705 struct dentry *parent;
707 parent = dentry_kill(dentry, 0);
709 * If dentry_kill returns NULL, we have nothing more to do.
710 * if it returns the same dentry, trylocks failed. In either
711 * case, just loop again.
713 * Otherwise, we need to prune ancestors too. This is necessary
714 * to prevent quadratic behavior of shrink_dcache_parent(), but
715 * is also expected to be beneficial in reducing dentry cache
716 * fragmentation.
718 if (!parent)
719 return;
720 if (parent == dentry)
721 return;
723 /* Prune ancestors. */
724 dentry = parent;
725 while (dentry) {
726 spin_lock(&dentry->d_lock);
727 if (dentry->d_count > 1) {
728 dentry->d_count--;
729 spin_unlock(&dentry->d_lock);
730 return;
732 dentry = dentry_kill(dentry, 1);
736 static void shrink_dentry_list(struct list_head *list)
738 struct dentry *dentry;
740 rcu_read_lock();
741 for (;;) {
742 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
743 if (&dentry->d_lru == list)
744 break; /* empty */
745 spin_lock(&dentry->d_lock);
746 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
747 spin_unlock(&dentry->d_lock);
748 continue;
752 * We found an inuse dentry which was not removed from
753 * the LRU because of laziness during lookup. Do not free
754 * it - just keep it off the LRU list.
756 if (dentry->d_count) {
757 dentry_lru_del(dentry);
758 spin_unlock(&dentry->d_lock);
759 continue;
762 rcu_read_unlock();
764 try_prune_one_dentry(dentry);
766 rcu_read_lock();
768 rcu_read_unlock();
772 * __shrink_dcache_sb - shrink the dentry LRU on a given superblock
773 * @sb: superblock to shrink dentry LRU.
774 * @count: number of entries to prune
775 * @flags: flags to control the dentry processing
777 * If flags contains DCACHE_REFERENCED reference dentries will not be pruned.
779 static void __shrink_dcache_sb(struct super_block *sb, int count, int flags)
781 struct dentry *dentry;
782 LIST_HEAD(referenced);
783 LIST_HEAD(tmp);
785 relock:
786 spin_lock(&dcache_lru_lock);
787 while (!list_empty(&sb->s_dentry_lru)) {
788 dentry = list_entry(sb->s_dentry_lru.prev,
789 struct dentry, d_lru);
790 BUG_ON(dentry->d_sb != sb);
792 if (!spin_trylock(&dentry->d_lock)) {
793 spin_unlock(&dcache_lru_lock);
794 cpu_relax();
795 goto relock;
799 * If we are honouring the DCACHE_REFERENCED flag and the
800 * dentry has this flag set, don't free it. Clear the flag
801 * and put it back on the LRU.
803 if (flags & DCACHE_REFERENCED &&
804 dentry->d_flags & DCACHE_REFERENCED) {
805 dentry->d_flags &= ~DCACHE_REFERENCED;
806 list_move(&dentry->d_lru, &referenced);
807 spin_unlock(&dentry->d_lock);
808 } else {
809 list_move_tail(&dentry->d_lru, &tmp);
810 spin_unlock(&dentry->d_lock);
811 if (!--count)
812 break;
814 cond_resched_lock(&dcache_lru_lock);
816 if (!list_empty(&referenced))
817 list_splice(&referenced, &sb->s_dentry_lru);
818 spin_unlock(&dcache_lru_lock);
820 shrink_dentry_list(&tmp);
824 * prune_dcache_sb - shrink the dcache
825 * @sb: superblock
826 * @nr_to_scan: number of entries to try to free
828 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
829 * done when we need more memory an called from the superblock shrinker
830 * function.
832 * This function may fail to free any resources if all the dentries are in
833 * use.
835 void prune_dcache_sb(struct super_block *sb, int nr_to_scan)
837 __shrink_dcache_sb(sb, nr_to_scan, DCACHE_REFERENCED);
841 * shrink_dcache_sb - shrink dcache for a superblock
842 * @sb: superblock
844 * Shrink the dcache for the specified super block. This is used to free
845 * the dcache before unmounting a file system.
847 void shrink_dcache_sb(struct super_block *sb)
849 LIST_HEAD(tmp);
851 spin_lock(&dcache_lru_lock);
852 while (!list_empty(&sb->s_dentry_lru)) {
853 list_splice_init(&sb->s_dentry_lru, &tmp);
854 spin_unlock(&dcache_lru_lock);
855 shrink_dentry_list(&tmp);
856 spin_lock(&dcache_lru_lock);
858 spin_unlock(&dcache_lru_lock);
860 EXPORT_SYMBOL(shrink_dcache_sb);
863 * destroy a single subtree of dentries for unmount
864 * - see the comments on shrink_dcache_for_umount() for a description of the
865 * locking
867 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
869 struct dentry *parent;
871 BUG_ON(!IS_ROOT(dentry));
873 for (;;) {
874 /* descend to the first leaf in the current subtree */
875 while (!list_empty(&dentry->d_subdirs))
876 dentry = list_entry(dentry->d_subdirs.next,
877 struct dentry, d_u.d_child);
879 /* consume the dentries from this leaf up through its parents
880 * until we find one with children or run out altogether */
881 do {
882 struct inode *inode;
885 * remove the dentry from the lru, and inform
886 * the fs that this dentry is about to be
887 * unhashed and destroyed.
889 dentry_lru_prune(dentry);
890 __d_shrink(dentry);
892 if (dentry->d_count != 0) {
893 printk(KERN_ERR
894 "BUG: Dentry %p{i=%lx,n=%s}"
895 " still in use (%d)"
896 " [unmount of %s %s]\n",
897 dentry,
898 dentry->d_inode ?
899 dentry->d_inode->i_ino : 0UL,
900 dentry->d_name.name,
901 dentry->d_count,
902 dentry->d_sb->s_type->name,
903 dentry->d_sb->s_id);
904 BUG();
907 if (IS_ROOT(dentry)) {
908 parent = NULL;
909 list_del(&dentry->d_u.d_child);
910 } else {
911 parent = dentry->d_parent;
912 parent->d_count--;
913 list_del(&dentry->d_u.d_child);
916 inode = dentry->d_inode;
917 if (inode) {
918 dentry->d_inode = NULL;
919 list_del_init(&dentry->d_alias);
920 if (dentry->d_op && dentry->d_op->d_iput)
921 dentry->d_op->d_iput(dentry, inode);
922 else
923 iput(inode);
926 d_free(dentry);
928 /* finished when we fall off the top of the tree,
929 * otherwise we ascend to the parent and move to the
930 * next sibling if there is one */
931 if (!parent)
932 return;
933 dentry = parent;
934 } while (list_empty(&dentry->d_subdirs));
936 dentry = list_entry(dentry->d_subdirs.next,
937 struct dentry, d_u.d_child);
942 * destroy the dentries attached to a superblock on unmounting
943 * - we don't need to use dentry->d_lock because:
944 * - the superblock is detached from all mountings and open files, so the
945 * dentry trees will not be rearranged by the VFS
946 * - s_umount is write-locked, so the memory pressure shrinker will ignore
947 * any dentries belonging to this superblock that it comes across
948 * - the filesystem itself is no longer permitted to rearrange the dentries
949 * in this superblock
951 void shrink_dcache_for_umount(struct super_block *sb)
953 struct dentry *dentry;
955 if (down_read_trylock(&sb->s_umount))
956 BUG();
958 dentry = sb->s_root;
959 sb->s_root = NULL;
960 dentry->d_count--;
961 shrink_dcache_for_umount_subtree(dentry);
963 while (!hlist_bl_empty(&sb->s_anon)) {
964 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
965 shrink_dcache_for_umount_subtree(dentry);
970 * This tries to ascend one level of parenthood, but
971 * we can race with renaming, so we need to re-check
972 * the parenthood after dropping the lock and check
973 * that the sequence number still matches.
975 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
977 struct dentry *new = old->d_parent;
979 rcu_read_lock();
980 spin_unlock(&old->d_lock);
981 spin_lock(&new->d_lock);
984 * might go back up the wrong parent if we have had a rename
985 * or deletion
987 if (new != old->d_parent ||
988 (old->d_flags & DCACHE_DISCONNECTED) ||
989 (!locked && read_seqretry(&rename_lock, seq))) {
990 spin_unlock(&new->d_lock);
991 new = NULL;
993 rcu_read_unlock();
994 return new;
999 * Search for at least 1 mount point in the dentry's subdirs.
1000 * We descend to the next level whenever the d_subdirs
1001 * list is non-empty and continue searching.
1005 * have_submounts - check for mounts over a dentry
1006 * @parent: dentry to check.
1008 * Return true if the parent or its subdirectories contain
1009 * a mount point
1011 int have_submounts(struct dentry *parent)
1013 struct dentry *this_parent;
1014 struct list_head *next;
1015 unsigned seq;
1016 int locked = 0;
1018 seq = read_seqbegin(&rename_lock);
1019 again:
1020 this_parent = parent;
1022 if (d_mountpoint(parent))
1023 goto positive;
1024 spin_lock(&this_parent->d_lock);
1025 repeat:
1026 next = this_parent->d_subdirs.next;
1027 resume:
1028 while (next != &this_parent->d_subdirs) {
1029 struct list_head *tmp = next;
1030 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1031 next = tmp->next;
1033 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1034 /* Have we found a mount point ? */
1035 if (d_mountpoint(dentry)) {
1036 spin_unlock(&dentry->d_lock);
1037 spin_unlock(&this_parent->d_lock);
1038 goto positive;
1040 if (!list_empty(&dentry->d_subdirs)) {
1041 spin_unlock(&this_parent->d_lock);
1042 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1043 this_parent = dentry;
1044 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1045 goto repeat;
1047 spin_unlock(&dentry->d_lock);
1050 * All done at this level ... ascend and resume the search.
1052 if (this_parent != parent) {
1053 struct dentry *child = this_parent;
1054 this_parent = try_to_ascend(this_parent, locked, seq);
1055 if (!this_parent)
1056 goto rename_retry;
1057 next = child->d_u.d_child.next;
1058 goto resume;
1060 spin_unlock(&this_parent->d_lock);
1061 if (!locked && read_seqretry(&rename_lock, seq))
1062 goto rename_retry;
1063 if (locked)
1064 write_sequnlock(&rename_lock);
1065 return 0; /* No mount points found in tree */
1066 positive:
1067 if (!locked && read_seqretry(&rename_lock, seq))
1068 goto rename_retry;
1069 if (locked)
1070 write_sequnlock(&rename_lock);
1071 return 1;
1073 rename_retry:
1074 locked = 1;
1075 write_seqlock(&rename_lock);
1076 goto again;
1078 EXPORT_SYMBOL(have_submounts);
1081 * Search the dentry child list for the specified parent,
1082 * and move any unused dentries to the end of the unused
1083 * list for prune_dcache(). We descend to the next level
1084 * whenever the d_subdirs list is non-empty and continue
1085 * searching.
1087 * It returns zero iff there are no unused children,
1088 * otherwise it returns the number of children moved to
1089 * the end of the unused list. This may not be the total
1090 * number of unused children, because select_parent can
1091 * drop the lock and return early due to latency
1092 * constraints.
1094 static int select_parent(struct dentry * parent)
1096 struct dentry *this_parent;
1097 struct list_head *next;
1098 unsigned seq;
1099 int found = 0;
1100 int locked = 0;
1102 seq = read_seqbegin(&rename_lock);
1103 again:
1104 this_parent = parent;
1105 spin_lock(&this_parent->d_lock);
1106 repeat:
1107 next = this_parent->d_subdirs.next;
1108 resume:
1109 while (next != &this_parent->d_subdirs) {
1110 struct list_head *tmp = next;
1111 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1112 next = tmp->next;
1114 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1117 * move only zero ref count dentries to the end
1118 * of the unused list for prune_dcache
1120 if (!dentry->d_count) {
1121 dentry_lru_move_tail(dentry);
1122 found++;
1123 } else {
1124 dentry_lru_del(dentry);
1128 * We can return to the caller if we have found some (this
1129 * ensures forward progress). We'll be coming back to find
1130 * the rest.
1132 if (found && need_resched()) {
1133 spin_unlock(&dentry->d_lock);
1134 goto out;
1138 * Descend a level if the d_subdirs list is non-empty.
1140 if (!list_empty(&dentry->d_subdirs)) {
1141 spin_unlock(&this_parent->d_lock);
1142 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1143 this_parent = dentry;
1144 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1145 goto repeat;
1148 spin_unlock(&dentry->d_lock);
1151 * All done at this level ... ascend and resume the search.
1153 if (this_parent != parent) {
1154 struct dentry *child = this_parent;
1155 this_parent = try_to_ascend(this_parent, locked, seq);
1156 if (!this_parent)
1157 goto rename_retry;
1158 next = child->d_u.d_child.next;
1159 goto resume;
1161 out:
1162 spin_unlock(&this_parent->d_lock);
1163 if (!locked && read_seqretry(&rename_lock, seq))
1164 goto rename_retry;
1165 if (locked)
1166 write_sequnlock(&rename_lock);
1167 return found;
1169 rename_retry:
1170 if (found)
1171 return found;
1172 locked = 1;
1173 write_seqlock(&rename_lock);
1174 goto again;
1178 * shrink_dcache_parent - prune dcache
1179 * @parent: parent of entries to prune
1181 * Prune the dcache to remove unused children of the parent dentry.
1184 void shrink_dcache_parent(struct dentry * parent)
1186 struct super_block *sb = parent->d_sb;
1187 int found;
1189 while ((found = select_parent(parent)) != 0)
1190 __shrink_dcache_sb(sb, found, 0);
1192 EXPORT_SYMBOL(shrink_dcache_parent);
1195 * __d_alloc - allocate a dcache entry
1196 * @sb: filesystem it will belong to
1197 * @name: qstr of the name
1199 * Allocates a dentry. It returns %NULL if there is insufficient memory
1200 * available. On a success the dentry is returned. The name passed in is
1201 * copied and the copy passed in may be reused after this call.
1204 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1206 struct dentry *dentry;
1207 char *dname;
1209 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1210 if (!dentry)
1211 return NULL;
1213 if (name->len > DNAME_INLINE_LEN-1) {
1214 dname = kmalloc(name->len + 1, GFP_KERNEL);
1215 if (!dname) {
1216 kmem_cache_free(dentry_cache, dentry);
1217 return NULL;
1219 } else {
1220 dname = dentry->d_iname;
1222 dentry->d_name.name = dname;
1224 dentry->d_name.len = name->len;
1225 dentry->d_name.hash = name->hash;
1226 memcpy(dname, name->name, name->len);
1227 dname[name->len] = 0;
1229 dentry->d_count = 1;
1230 dentry->d_flags = 0;
1231 spin_lock_init(&dentry->d_lock);
1232 seqcount_init(&dentry->d_seq);
1233 dentry->d_inode = NULL;
1234 dentry->d_parent = dentry;
1235 dentry->d_sb = sb;
1236 dentry->d_op = NULL;
1237 dentry->d_fsdata = NULL;
1238 INIT_HLIST_BL_NODE(&dentry->d_hash);
1239 INIT_LIST_HEAD(&dentry->d_lru);
1240 INIT_LIST_HEAD(&dentry->d_subdirs);
1241 INIT_LIST_HEAD(&dentry->d_alias);
1242 INIT_LIST_HEAD(&dentry->d_u.d_child);
1243 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1245 this_cpu_inc(nr_dentry);
1247 return dentry;
1251 * d_alloc - allocate a dcache entry
1252 * @parent: parent of entry to allocate
1253 * @name: qstr of the name
1255 * Allocates a dentry. It returns %NULL if there is insufficient memory
1256 * available. On a success the dentry is returned. The name passed in is
1257 * copied and the copy passed in may be reused after this call.
1259 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1261 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1262 if (!dentry)
1263 return NULL;
1265 spin_lock(&parent->d_lock);
1267 * don't need child lock because it is not subject
1268 * to concurrency here
1270 __dget_dlock(parent);
1271 dentry->d_parent = parent;
1272 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1273 spin_unlock(&parent->d_lock);
1275 return dentry;
1277 EXPORT_SYMBOL(d_alloc);
1279 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1281 struct dentry *dentry = __d_alloc(sb, name);
1282 if (dentry)
1283 dentry->d_flags |= DCACHE_DISCONNECTED;
1284 return dentry;
1286 EXPORT_SYMBOL(d_alloc_pseudo);
1288 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1290 struct qstr q;
1292 q.name = name;
1293 q.len = strlen(name);
1294 q.hash = full_name_hash(q.name, q.len);
1295 return d_alloc(parent, &q);
1297 EXPORT_SYMBOL(d_alloc_name);
1299 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1301 WARN_ON_ONCE(dentry->d_op);
1302 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1303 DCACHE_OP_COMPARE |
1304 DCACHE_OP_REVALIDATE |
1305 DCACHE_OP_DELETE ));
1306 dentry->d_op = op;
1307 if (!op)
1308 return;
1309 if (op->d_hash)
1310 dentry->d_flags |= DCACHE_OP_HASH;
1311 if (op->d_compare)
1312 dentry->d_flags |= DCACHE_OP_COMPARE;
1313 if (op->d_revalidate)
1314 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1315 if (op->d_delete)
1316 dentry->d_flags |= DCACHE_OP_DELETE;
1317 if (op->d_prune)
1318 dentry->d_flags |= DCACHE_OP_PRUNE;
1321 EXPORT_SYMBOL(d_set_d_op);
1323 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1325 spin_lock(&dentry->d_lock);
1326 if (inode) {
1327 if (unlikely(IS_AUTOMOUNT(inode)))
1328 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1329 list_add(&dentry->d_alias, &inode->i_dentry);
1331 dentry->d_inode = inode;
1332 dentry_rcuwalk_barrier(dentry);
1333 spin_unlock(&dentry->d_lock);
1334 fsnotify_d_instantiate(dentry, inode);
1338 * d_instantiate - fill in inode information for a dentry
1339 * @entry: dentry to complete
1340 * @inode: inode to attach to this dentry
1342 * Fill in inode information in the entry.
1344 * This turns negative dentries into productive full members
1345 * of society.
1347 * NOTE! This assumes that the inode count has been incremented
1348 * (or otherwise set) by the caller to indicate that it is now
1349 * in use by the dcache.
1352 void d_instantiate(struct dentry *entry, struct inode * inode)
1354 BUG_ON(!list_empty(&entry->d_alias));
1355 if (inode)
1356 spin_lock(&inode->i_lock);
1357 __d_instantiate(entry, inode);
1358 if (inode)
1359 spin_unlock(&inode->i_lock);
1360 security_d_instantiate(entry, inode);
1362 EXPORT_SYMBOL(d_instantiate);
1365 * d_instantiate_unique - instantiate a non-aliased dentry
1366 * @entry: dentry to instantiate
1367 * @inode: inode to attach to this dentry
1369 * Fill in inode information in the entry. On success, it returns NULL.
1370 * If an unhashed alias of "entry" already exists, then we return the
1371 * aliased dentry instead and drop one reference to inode.
1373 * Note that in order to avoid conflicts with rename() etc, the caller
1374 * had better be holding the parent directory semaphore.
1376 * This also assumes that the inode count has been incremented
1377 * (or otherwise set) by the caller to indicate that it is now
1378 * in use by the dcache.
1380 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1381 struct inode *inode)
1383 struct dentry *alias;
1384 int len = entry->d_name.len;
1385 const char *name = entry->d_name.name;
1386 unsigned int hash = entry->d_name.hash;
1388 if (!inode) {
1389 __d_instantiate(entry, NULL);
1390 return NULL;
1393 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1394 struct qstr *qstr = &alias->d_name;
1397 * Don't need alias->d_lock here, because aliases with
1398 * d_parent == entry->d_parent are not subject to name or
1399 * parent changes, because the parent inode i_mutex is held.
1401 if (qstr->hash != hash)
1402 continue;
1403 if (alias->d_parent != entry->d_parent)
1404 continue;
1405 if (dentry_cmp(qstr->name, qstr->len, name, len))
1406 continue;
1407 __dget(alias);
1408 return alias;
1411 __d_instantiate(entry, inode);
1412 return NULL;
1415 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1417 struct dentry *result;
1419 BUG_ON(!list_empty(&entry->d_alias));
1421 if (inode)
1422 spin_lock(&inode->i_lock);
1423 result = __d_instantiate_unique(entry, inode);
1424 if (inode)
1425 spin_unlock(&inode->i_lock);
1427 if (!result) {
1428 security_d_instantiate(entry, inode);
1429 return NULL;
1432 BUG_ON(!d_unhashed(result));
1433 iput(inode);
1434 return result;
1437 EXPORT_SYMBOL(d_instantiate_unique);
1440 * d_alloc_root - allocate root dentry
1441 * @root_inode: inode to allocate the root for
1443 * Allocate a root ("/") dentry for the inode given. The inode is
1444 * instantiated and returned. %NULL is returned if there is insufficient
1445 * memory or the inode passed is %NULL.
1448 struct dentry * d_alloc_root(struct inode * root_inode)
1450 struct dentry *res = NULL;
1452 if (root_inode) {
1453 static const struct qstr name = { .name = "/", .len = 1 };
1455 res = __d_alloc(root_inode->i_sb, &name);
1456 if (res)
1457 d_instantiate(res, root_inode);
1459 return res;
1461 EXPORT_SYMBOL(d_alloc_root);
1463 static struct dentry * __d_find_any_alias(struct inode *inode)
1465 struct dentry *alias;
1467 if (list_empty(&inode->i_dentry))
1468 return NULL;
1469 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1470 __dget(alias);
1471 return alias;
1474 static struct dentry * d_find_any_alias(struct inode *inode)
1476 struct dentry *de;
1478 spin_lock(&inode->i_lock);
1479 de = __d_find_any_alias(inode);
1480 spin_unlock(&inode->i_lock);
1481 return de;
1486 * d_obtain_alias - find or allocate a dentry for a given inode
1487 * @inode: inode to allocate the dentry for
1489 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1490 * similar open by handle operations. The returned dentry may be anonymous,
1491 * or may have a full name (if the inode was already in the cache).
1493 * When called on a directory inode, we must ensure that the inode only ever
1494 * has one dentry. If a dentry is found, that is returned instead of
1495 * allocating a new one.
1497 * On successful return, the reference to the inode has been transferred
1498 * to the dentry. In case of an error the reference on the inode is released.
1499 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1500 * be passed in and will be the error will be propagate to the return value,
1501 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1503 struct dentry *d_obtain_alias(struct inode *inode)
1505 static const struct qstr anonstring = { .name = "" };
1506 struct dentry *tmp;
1507 struct dentry *res;
1509 if (!inode)
1510 return ERR_PTR(-ESTALE);
1511 if (IS_ERR(inode))
1512 return ERR_CAST(inode);
1514 res = d_find_any_alias(inode);
1515 if (res)
1516 goto out_iput;
1518 tmp = __d_alloc(inode->i_sb, &anonstring);
1519 if (!tmp) {
1520 res = ERR_PTR(-ENOMEM);
1521 goto out_iput;
1524 spin_lock(&inode->i_lock);
1525 res = __d_find_any_alias(inode);
1526 if (res) {
1527 spin_unlock(&inode->i_lock);
1528 dput(tmp);
1529 goto out_iput;
1532 /* attach a disconnected dentry */
1533 spin_lock(&tmp->d_lock);
1534 tmp->d_inode = inode;
1535 tmp->d_flags |= DCACHE_DISCONNECTED;
1536 list_add(&tmp->d_alias, &inode->i_dentry);
1537 hlist_bl_lock(&tmp->d_sb->s_anon);
1538 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1539 hlist_bl_unlock(&tmp->d_sb->s_anon);
1540 spin_unlock(&tmp->d_lock);
1541 spin_unlock(&inode->i_lock);
1542 security_d_instantiate(tmp, inode);
1544 return tmp;
1546 out_iput:
1547 if (res && !IS_ERR(res))
1548 security_d_instantiate(res, inode);
1549 iput(inode);
1550 return res;
1552 EXPORT_SYMBOL(d_obtain_alias);
1555 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1556 * @inode: the inode which may have a disconnected dentry
1557 * @dentry: a negative dentry which we want to point to the inode.
1559 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1560 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1561 * and return it, else simply d_add the inode to the dentry and return NULL.
1563 * This is needed in the lookup routine of any filesystem that is exportable
1564 * (via knfsd) so that we can build dcache paths to directories effectively.
1566 * If a dentry was found and moved, then it is returned. Otherwise NULL
1567 * is returned. This matches the expected return value of ->lookup.
1570 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1572 struct dentry *new = NULL;
1574 if (IS_ERR(inode))
1575 return ERR_CAST(inode);
1577 if (inode && S_ISDIR(inode->i_mode)) {
1578 spin_lock(&inode->i_lock);
1579 new = __d_find_alias(inode, 1);
1580 if (new) {
1581 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1582 spin_unlock(&inode->i_lock);
1583 security_d_instantiate(new, inode);
1584 d_move(new, dentry);
1585 iput(inode);
1586 } else {
1587 /* already taking inode->i_lock, so d_add() by hand */
1588 __d_instantiate(dentry, inode);
1589 spin_unlock(&inode->i_lock);
1590 security_d_instantiate(dentry, inode);
1591 d_rehash(dentry);
1593 } else
1594 d_add(dentry, inode);
1595 return new;
1597 EXPORT_SYMBOL(d_splice_alias);
1600 * d_add_ci - lookup or allocate new dentry with case-exact name
1601 * @inode: the inode case-insensitive lookup has found
1602 * @dentry: the negative dentry that was passed to the parent's lookup func
1603 * @name: the case-exact name to be associated with the returned dentry
1605 * This is to avoid filling the dcache with case-insensitive names to the
1606 * same inode, only the actual correct case is stored in the dcache for
1607 * case-insensitive filesystems.
1609 * For a case-insensitive lookup match and if the the case-exact dentry
1610 * already exists in in the dcache, use it and return it.
1612 * If no entry exists with the exact case name, allocate new dentry with
1613 * the exact case, and return the spliced entry.
1615 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1616 struct qstr *name)
1618 int error;
1619 struct dentry *found;
1620 struct dentry *new;
1623 * First check if a dentry matching the name already exists,
1624 * if not go ahead and create it now.
1626 found = d_hash_and_lookup(dentry->d_parent, name);
1627 if (!found) {
1628 new = d_alloc(dentry->d_parent, name);
1629 if (!new) {
1630 error = -ENOMEM;
1631 goto err_out;
1634 found = d_splice_alias(inode, new);
1635 if (found) {
1636 dput(new);
1637 return found;
1639 return new;
1643 * If a matching dentry exists, and it's not negative use it.
1645 * Decrement the reference count to balance the iget() done
1646 * earlier on.
1648 if (found->d_inode) {
1649 if (unlikely(found->d_inode != inode)) {
1650 /* This can't happen because bad inodes are unhashed. */
1651 BUG_ON(!is_bad_inode(inode));
1652 BUG_ON(!is_bad_inode(found->d_inode));
1654 iput(inode);
1655 return found;
1659 * We are going to instantiate this dentry, unhash it and clear the
1660 * lookup flag so we can do that.
1662 if (unlikely(d_need_lookup(found)))
1663 d_clear_need_lookup(found);
1666 * Negative dentry: instantiate it unless the inode is a directory and
1667 * already has a dentry.
1669 new = d_splice_alias(inode, found);
1670 if (new) {
1671 dput(found);
1672 found = new;
1674 return found;
1676 err_out:
1677 iput(inode);
1678 return ERR_PTR(error);
1680 EXPORT_SYMBOL(d_add_ci);
1683 * __d_lookup_rcu - search for a dentry (racy, store-free)
1684 * @parent: parent dentry
1685 * @name: qstr of name we wish to find
1686 * @seq: returns d_seq value at the point where the dentry was found
1687 * @inode: returns dentry->d_inode when the inode was found valid.
1688 * Returns: dentry, or NULL
1690 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1691 * resolution (store-free path walking) design described in
1692 * Documentation/filesystems/path-lookup.txt.
1694 * This is not to be used outside core vfs.
1696 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1697 * held, and rcu_read_lock held. The returned dentry must not be stored into
1698 * without taking d_lock and checking d_seq sequence count against @seq
1699 * returned here.
1701 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1702 * function.
1704 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1705 * the returned dentry, so long as its parent's seqlock is checked after the
1706 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1707 * is formed, giving integrity down the path walk.
1709 struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name,
1710 unsigned *seq, struct inode **inode)
1712 unsigned int len = name->len;
1713 unsigned int hash = name->hash;
1714 const unsigned char *str = name->name;
1715 struct hlist_bl_head *b = d_hash(parent, hash);
1716 struct hlist_bl_node *node;
1717 struct dentry *dentry;
1720 * Note: There is significant duplication with __d_lookup_rcu which is
1721 * required to prevent single threaded performance regressions
1722 * especially on architectures where smp_rmb (in seqcounts) are costly.
1723 * Keep the two functions in sync.
1727 * The hash list is protected using RCU.
1729 * Carefully use d_seq when comparing a candidate dentry, to avoid
1730 * races with d_move().
1732 * It is possible that concurrent renames can mess up our list
1733 * walk here and result in missing our dentry, resulting in the
1734 * false-negative result. d_lookup() protects against concurrent
1735 * renames using rename_lock seqlock.
1737 * See Documentation/filesystems/path-lookup.txt for more details.
1739 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1740 struct inode *i;
1741 const char *tname;
1742 int tlen;
1744 if (dentry->d_name.hash != hash)
1745 continue;
1747 seqretry:
1748 *seq = read_seqcount_begin(&dentry->d_seq);
1749 if (dentry->d_parent != parent)
1750 continue;
1751 if (d_unhashed(dentry))
1752 continue;
1753 tlen = dentry->d_name.len;
1754 tname = dentry->d_name.name;
1755 i = dentry->d_inode;
1756 prefetch(tname);
1758 * This seqcount check is required to ensure name and
1759 * len are loaded atomically, so as not to walk off the
1760 * edge of memory when walking. If we could load this
1761 * atomically some other way, we could drop this check.
1763 if (read_seqcount_retry(&dentry->d_seq, *seq))
1764 goto seqretry;
1765 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1766 if (parent->d_op->d_compare(parent, *inode,
1767 dentry, i,
1768 tlen, tname, name))
1769 continue;
1770 } else {
1771 if (dentry_cmp(tname, tlen, str, len))
1772 continue;
1775 * No extra seqcount check is required after the name
1776 * compare. The caller must perform a seqcount check in
1777 * order to do anything useful with the returned dentry
1778 * anyway.
1780 *inode = i;
1781 return dentry;
1783 return NULL;
1787 * d_lookup - search for a dentry
1788 * @parent: parent dentry
1789 * @name: qstr of name we wish to find
1790 * Returns: dentry, or NULL
1792 * d_lookup searches the children of the parent dentry for the name in
1793 * question. If the dentry is found its reference count is incremented and the
1794 * dentry is returned. The caller must use dput to free the entry when it has
1795 * finished using it. %NULL is returned if the dentry does not exist.
1797 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1799 struct dentry *dentry;
1800 unsigned seq;
1802 do {
1803 seq = read_seqbegin(&rename_lock);
1804 dentry = __d_lookup(parent, name);
1805 if (dentry)
1806 break;
1807 } while (read_seqretry(&rename_lock, seq));
1808 return dentry;
1810 EXPORT_SYMBOL(d_lookup);
1813 * __d_lookup - search for a dentry (racy)
1814 * @parent: parent dentry
1815 * @name: qstr of name we wish to find
1816 * Returns: dentry, or NULL
1818 * __d_lookup is like d_lookup, however it may (rarely) return a
1819 * false-negative result due to unrelated rename activity.
1821 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1822 * however it must be used carefully, eg. with a following d_lookup in
1823 * the case of failure.
1825 * __d_lookup callers must be commented.
1827 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1829 unsigned int len = name->len;
1830 unsigned int hash = name->hash;
1831 const unsigned char *str = name->name;
1832 struct hlist_bl_head *b = d_hash(parent, hash);
1833 struct hlist_bl_node *node;
1834 struct dentry *found = NULL;
1835 struct dentry *dentry;
1838 * Note: There is significant duplication with __d_lookup_rcu which is
1839 * required to prevent single threaded performance regressions
1840 * especially on architectures where smp_rmb (in seqcounts) are costly.
1841 * Keep the two functions in sync.
1845 * The hash list is protected using RCU.
1847 * Take d_lock when comparing a candidate dentry, to avoid races
1848 * with d_move().
1850 * It is possible that concurrent renames can mess up our list
1851 * walk here and result in missing our dentry, resulting in the
1852 * false-negative result. d_lookup() protects against concurrent
1853 * renames using rename_lock seqlock.
1855 * See Documentation/filesystems/path-lookup.txt for more details.
1857 rcu_read_lock();
1859 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1860 const char *tname;
1861 int tlen;
1863 if (dentry->d_name.hash != hash)
1864 continue;
1866 spin_lock(&dentry->d_lock);
1867 if (dentry->d_parent != parent)
1868 goto next;
1869 if (d_unhashed(dentry))
1870 goto next;
1873 * It is safe to compare names since d_move() cannot
1874 * change the qstr (protected by d_lock).
1876 tlen = dentry->d_name.len;
1877 tname = dentry->d_name.name;
1878 if (parent->d_flags & DCACHE_OP_COMPARE) {
1879 if (parent->d_op->d_compare(parent, parent->d_inode,
1880 dentry, dentry->d_inode,
1881 tlen, tname, name))
1882 goto next;
1883 } else {
1884 if (dentry_cmp(tname, tlen, str, len))
1885 goto next;
1888 dentry->d_count++;
1889 found = dentry;
1890 spin_unlock(&dentry->d_lock);
1891 break;
1892 next:
1893 spin_unlock(&dentry->d_lock);
1895 rcu_read_unlock();
1897 return found;
1901 * d_hash_and_lookup - hash the qstr then search for a dentry
1902 * @dir: Directory to search in
1903 * @name: qstr of name we wish to find
1905 * On hash failure or on lookup failure NULL is returned.
1907 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1909 struct dentry *dentry = NULL;
1912 * Check for a fs-specific hash function. Note that we must
1913 * calculate the standard hash first, as the d_op->d_hash()
1914 * routine may choose to leave the hash value unchanged.
1916 name->hash = full_name_hash(name->name, name->len);
1917 if (dir->d_flags & DCACHE_OP_HASH) {
1918 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1919 goto out;
1921 dentry = d_lookup(dir, name);
1922 out:
1923 return dentry;
1927 * d_validate - verify dentry provided from insecure source (deprecated)
1928 * @dentry: The dentry alleged to be valid child of @dparent
1929 * @dparent: The parent dentry (known to be valid)
1931 * An insecure source has sent us a dentry, here we verify it and dget() it.
1932 * This is used by ncpfs in its readdir implementation.
1933 * Zero is returned in the dentry is invalid.
1935 * This function is slow for big directories, and deprecated, do not use it.
1937 int d_validate(struct dentry *dentry, struct dentry *dparent)
1939 struct dentry *child;
1941 spin_lock(&dparent->d_lock);
1942 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1943 if (dentry == child) {
1944 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1945 __dget_dlock(dentry);
1946 spin_unlock(&dentry->d_lock);
1947 spin_unlock(&dparent->d_lock);
1948 return 1;
1951 spin_unlock(&dparent->d_lock);
1953 return 0;
1955 EXPORT_SYMBOL(d_validate);
1958 * When a file is deleted, we have two options:
1959 * - turn this dentry into a negative dentry
1960 * - unhash this dentry and free it.
1962 * Usually, we want to just turn this into
1963 * a negative dentry, but if anybody else is
1964 * currently using the dentry or the inode
1965 * we can't do that and we fall back on removing
1966 * it from the hash queues and waiting for
1967 * it to be deleted later when it has no users
1971 * d_delete - delete a dentry
1972 * @dentry: The dentry to delete
1974 * Turn the dentry into a negative dentry if possible, otherwise
1975 * remove it from the hash queues so it can be deleted later
1978 void d_delete(struct dentry * dentry)
1980 struct inode *inode;
1981 int isdir = 0;
1983 * Are we the only user?
1985 again:
1986 spin_lock(&dentry->d_lock);
1987 inode = dentry->d_inode;
1988 isdir = S_ISDIR(inode->i_mode);
1989 if (dentry->d_count == 1) {
1990 if (inode && !spin_trylock(&inode->i_lock)) {
1991 spin_unlock(&dentry->d_lock);
1992 cpu_relax();
1993 goto again;
1995 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1996 dentry_unlink_inode(dentry);
1997 fsnotify_nameremove(dentry, isdir);
1998 return;
2001 if (!d_unhashed(dentry))
2002 __d_drop(dentry);
2004 spin_unlock(&dentry->d_lock);
2006 fsnotify_nameremove(dentry, isdir);
2008 EXPORT_SYMBOL(d_delete);
2010 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2012 BUG_ON(!d_unhashed(entry));
2013 hlist_bl_lock(b);
2014 entry->d_flags |= DCACHE_RCUACCESS;
2015 hlist_bl_add_head_rcu(&entry->d_hash, b);
2016 hlist_bl_unlock(b);
2019 static void _d_rehash(struct dentry * entry)
2021 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2025 * d_rehash - add an entry back to the hash
2026 * @entry: dentry to add to the hash
2028 * Adds a dentry to the hash according to its name.
2031 void d_rehash(struct dentry * entry)
2033 spin_lock(&entry->d_lock);
2034 _d_rehash(entry);
2035 spin_unlock(&entry->d_lock);
2037 EXPORT_SYMBOL(d_rehash);
2040 * dentry_update_name_case - update case insensitive dentry with a new name
2041 * @dentry: dentry to be updated
2042 * @name: new name
2044 * Update a case insensitive dentry with new case of name.
2046 * dentry must have been returned by d_lookup with name @name. Old and new
2047 * name lengths must match (ie. no d_compare which allows mismatched name
2048 * lengths).
2050 * Parent inode i_mutex must be held over d_lookup and into this call (to
2051 * keep renames and concurrent inserts, and readdir(2) away).
2053 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2055 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2056 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2058 spin_lock(&dentry->d_lock);
2059 write_seqcount_begin(&dentry->d_seq);
2060 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2061 write_seqcount_end(&dentry->d_seq);
2062 spin_unlock(&dentry->d_lock);
2064 EXPORT_SYMBOL(dentry_update_name_case);
2066 static void switch_names(struct dentry *dentry, struct dentry *target)
2068 if (dname_external(target)) {
2069 if (dname_external(dentry)) {
2071 * Both external: swap the pointers
2073 swap(target->d_name.name, dentry->d_name.name);
2074 } else {
2076 * dentry:internal, target:external. Steal target's
2077 * storage and make target internal.
2079 memcpy(target->d_iname, dentry->d_name.name,
2080 dentry->d_name.len + 1);
2081 dentry->d_name.name = target->d_name.name;
2082 target->d_name.name = target->d_iname;
2084 } else {
2085 if (dname_external(dentry)) {
2087 * dentry:external, target:internal. Give dentry's
2088 * storage to target and make dentry internal
2090 memcpy(dentry->d_iname, target->d_name.name,
2091 target->d_name.len + 1);
2092 target->d_name.name = dentry->d_name.name;
2093 dentry->d_name.name = dentry->d_iname;
2094 } else {
2096 * Both are internal. Just copy target to dentry
2098 memcpy(dentry->d_iname, target->d_name.name,
2099 target->d_name.len + 1);
2100 dentry->d_name.len = target->d_name.len;
2101 return;
2104 swap(dentry->d_name.len, target->d_name.len);
2107 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2110 * XXXX: do we really need to take target->d_lock?
2112 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2113 spin_lock(&target->d_parent->d_lock);
2114 else {
2115 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2116 spin_lock(&dentry->d_parent->d_lock);
2117 spin_lock_nested(&target->d_parent->d_lock,
2118 DENTRY_D_LOCK_NESTED);
2119 } else {
2120 spin_lock(&target->d_parent->d_lock);
2121 spin_lock_nested(&dentry->d_parent->d_lock,
2122 DENTRY_D_LOCK_NESTED);
2125 if (target < dentry) {
2126 spin_lock_nested(&target->d_lock, 2);
2127 spin_lock_nested(&dentry->d_lock, 3);
2128 } else {
2129 spin_lock_nested(&dentry->d_lock, 2);
2130 spin_lock_nested(&target->d_lock, 3);
2134 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2135 struct dentry *target)
2137 if (target->d_parent != dentry->d_parent)
2138 spin_unlock(&dentry->d_parent->d_lock);
2139 if (target->d_parent != target)
2140 spin_unlock(&target->d_parent->d_lock);
2144 * When switching names, the actual string doesn't strictly have to
2145 * be preserved in the target - because we're dropping the target
2146 * anyway. As such, we can just do a simple memcpy() to copy over
2147 * the new name before we switch.
2149 * Note that we have to be a lot more careful about getting the hash
2150 * switched - we have to switch the hash value properly even if it
2151 * then no longer matches the actual (corrupted) string of the target.
2152 * The hash value has to match the hash queue that the dentry is on..
2155 * __d_move - move a dentry
2156 * @dentry: entry to move
2157 * @target: new dentry
2159 * Update the dcache to reflect the move of a file name. Negative
2160 * dcache entries should not be moved in this way. Caller must hold
2161 * rename_lock, the i_mutex of the source and target directories,
2162 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2164 static void __d_move(struct dentry * dentry, struct dentry * target)
2166 if (!dentry->d_inode)
2167 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2169 BUG_ON(d_ancestor(dentry, target));
2170 BUG_ON(d_ancestor(target, dentry));
2172 dentry_lock_for_move(dentry, target);
2174 write_seqcount_begin(&dentry->d_seq);
2175 write_seqcount_begin(&target->d_seq);
2177 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2180 * Move the dentry to the target hash queue. Don't bother checking
2181 * for the same hash queue because of how unlikely it is.
2183 __d_drop(dentry);
2184 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2186 /* Unhash the target: dput() will then get rid of it */
2187 __d_drop(target);
2189 list_del(&dentry->d_u.d_child);
2190 list_del(&target->d_u.d_child);
2192 /* Switch the names.. */
2193 switch_names(dentry, target);
2194 swap(dentry->d_name.hash, target->d_name.hash);
2196 /* ... and switch the parents */
2197 if (IS_ROOT(dentry)) {
2198 dentry->d_parent = target->d_parent;
2199 target->d_parent = target;
2200 INIT_LIST_HEAD(&target->d_u.d_child);
2201 } else {
2202 swap(dentry->d_parent, target->d_parent);
2204 /* And add them back to the (new) parent lists */
2205 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2208 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2210 write_seqcount_end(&target->d_seq);
2211 write_seqcount_end(&dentry->d_seq);
2213 dentry_unlock_parents_for_move(dentry, target);
2214 spin_unlock(&target->d_lock);
2215 fsnotify_d_move(dentry);
2216 spin_unlock(&dentry->d_lock);
2220 * d_move - move a dentry
2221 * @dentry: entry to move
2222 * @target: new dentry
2224 * Update the dcache to reflect the move of a file name. Negative
2225 * dcache entries should not be moved in this way. See the locking
2226 * requirements for __d_move.
2228 void d_move(struct dentry *dentry, struct dentry *target)
2230 write_seqlock(&rename_lock);
2231 __d_move(dentry, target);
2232 write_sequnlock(&rename_lock);
2234 EXPORT_SYMBOL(d_move);
2237 * d_ancestor - search for an ancestor
2238 * @p1: ancestor dentry
2239 * @p2: child dentry
2241 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2242 * an ancestor of p2, else NULL.
2244 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2246 struct dentry *p;
2248 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2249 if (p->d_parent == p1)
2250 return p;
2252 return NULL;
2256 * This helper attempts to cope with remotely renamed directories
2258 * It assumes that the caller is already holding
2259 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2261 * Note: If ever the locking in lock_rename() changes, then please
2262 * remember to update this too...
2264 static struct dentry *__d_unalias(struct inode *inode,
2265 struct dentry *dentry, struct dentry *alias)
2267 struct mutex *m1 = NULL, *m2 = NULL;
2268 struct dentry *ret;
2270 /* If alias and dentry share a parent, then no extra locks required */
2271 if (alias->d_parent == dentry->d_parent)
2272 goto out_unalias;
2274 /* See lock_rename() */
2275 ret = ERR_PTR(-EBUSY);
2276 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2277 goto out_err;
2278 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2279 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2280 goto out_err;
2281 m2 = &alias->d_parent->d_inode->i_mutex;
2282 out_unalias:
2283 __d_move(alias, dentry);
2284 ret = alias;
2285 out_err:
2286 spin_unlock(&inode->i_lock);
2287 if (m2)
2288 mutex_unlock(m2);
2289 if (m1)
2290 mutex_unlock(m1);
2291 return ret;
2295 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2296 * named dentry in place of the dentry to be replaced.
2297 * returns with anon->d_lock held!
2299 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2301 struct dentry *dparent, *aparent;
2303 dentry_lock_for_move(anon, dentry);
2305 write_seqcount_begin(&dentry->d_seq);
2306 write_seqcount_begin(&anon->d_seq);
2308 dparent = dentry->d_parent;
2309 aparent = anon->d_parent;
2311 switch_names(dentry, anon);
2312 swap(dentry->d_name.hash, anon->d_name.hash);
2314 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2315 list_del(&dentry->d_u.d_child);
2316 if (!IS_ROOT(dentry))
2317 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2318 else
2319 INIT_LIST_HEAD(&dentry->d_u.d_child);
2321 anon->d_parent = (dparent == dentry) ? anon : dparent;
2322 list_del(&anon->d_u.d_child);
2323 if (!IS_ROOT(anon))
2324 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2325 else
2326 INIT_LIST_HEAD(&anon->d_u.d_child);
2328 write_seqcount_end(&dentry->d_seq);
2329 write_seqcount_end(&anon->d_seq);
2331 dentry_unlock_parents_for_move(anon, dentry);
2332 spin_unlock(&dentry->d_lock);
2334 /* anon->d_lock still locked, returns locked */
2335 anon->d_flags &= ~DCACHE_DISCONNECTED;
2339 * d_materialise_unique - introduce an inode into the tree
2340 * @dentry: candidate dentry
2341 * @inode: inode to bind to the dentry, to which aliases may be attached
2343 * Introduces an dentry into the tree, substituting an extant disconnected
2344 * root directory alias in its place if there is one. Caller must hold the
2345 * i_mutex of the parent directory.
2347 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2349 struct dentry *actual;
2351 BUG_ON(!d_unhashed(dentry));
2353 if (!inode) {
2354 actual = dentry;
2355 __d_instantiate(dentry, NULL);
2356 d_rehash(actual);
2357 goto out_nolock;
2360 spin_lock(&inode->i_lock);
2362 if (S_ISDIR(inode->i_mode)) {
2363 struct dentry *alias;
2365 /* Does an aliased dentry already exist? */
2366 alias = __d_find_alias(inode, 0);
2367 if (alias) {
2368 actual = alias;
2369 write_seqlock(&rename_lock);
2371 if (d_ancestor(alias, dentry)) {
2372 /* Check for loops */
2373 actual = ERR_PTR(-ELOOP);
2374 } else if (IS_ROOT(alias)) {
2375 /* Is this an anonymous mountpoint that we
2376 * could splice into our tree? */
2377 __d_materialise_dentry(dentry, alias);
2378 write_sequnlock(&rename_lock);
2379 __d_drop(alias);
2380 goto found;
2381 } else {
2382 /* Nope, but we must(!) avoid directory
2383 * aliasing */
2384 actual = __d_unalias(inode, dentry, alias);
2386 write_sequnlock(&rename_lock);
2387 if (IS_ERR(actual)) {
2388 if (PTR_ERR(actual) == -ELOOP)
2389 pr_warn_ratelimited(
2390 "VFS: Lookup of '%s' in %s %s"
2391 " would have caused loop\n",
2392 dentry->d_name.name,
2393 inode->i_sb->s_type->name,
2394 inode->i_sb->s_id);
2395 dput(alias);
2397 goto out_nolock;
2401 /* Add a unique reference */
2402 actual = __d_instantiate_unique(dentry, inode);
2403 if (!actual)
2404 actual = dentry;
2405 else
2406 BUG_ON(!d_unhashed(actual));
2408 spin_lock(&actual->d_lock);
2409 found:
2410 _d_rehash(actual);
2411 spin_unlock(&actual->d_lock);
2412 spin_unlock(&inode->i_lock);
2413 out_nolock:
2414 if (actual == dentry) {
2415 security_d_instantiate(dentry, inode);
2416 return NULL;
2419 iput(inode);
2420 return actual;
2422 EXPORT_SYMBOL_GPL(d_materialise_unique);
2424 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2426 *buflen -= namelen;
2427 if (*buflen < 0)
2428 return -ENAMETOOLONG;
2429 *buffer -= namelen;
2430 memcpy(*buffer, str, namelen);
2431 return 0;
2434 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2436 return prepend(buffer, buflen, name->name, name->len);
2440 * prepend_path - Prepend path string to a buffer
2441 * @path: the dentry/vfsmount to report
2442 * @root: root vfsmnt/dentry
2443 * @buffer: pointer to the end of the buffer
2444 * @buflen: pointer to buffer length
2446 * Caller holds the rename_lock.
2448 static int prepend_path(const struct path *path,
2449 const struct path *root,
2450 char **buffer, int *buflen)
2452 struct dentry *dentry = path->dentry;
2453 struct vfsmount *vfsmnt = path->mnt;
2454 bool slash = false;
2455 int error = 0;
2457 br_read_lock(vfsmount_lock);
2458 while (dentry != root->dentry || vfsmnt != root->mnt) {
2459 struct dentry * parent;
2461 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2462 /* Global root? */
2463 if (vfsmnt->mnt_parent == vfsmnt) {
2464 goto global_root;
2466 dentry = vfsmnt->mnt_mountpoint;
2467 vfsmnt = vfsmnt->mnt_parent;
2468 continue;
2470 parent = dentry->d_parent;
2471 prefetch(parent);
2472 spin_lock(&dentry->d_lock);
2473 error = prepend_name(buffer, buflen, &dentry->d_name);
2474 spin_unlock(&dentry->d_lock);
2475 if (!error)
2476 error = prepend(buffer, buflen, "/", 1);
2477 if (error)
2478 break;
2480 slash = true;
2481 dentry = parent;
2484 if (!error && !slash)
2485 error = prepend(buffer, buflen, "/", 1);
2487 out:
2488 br_read_unlock(vfsmount_lock);
2489 return error;
2491 global_root:
2493 * Filesystems needing to implement special "root names"
2494 * should do so with ->d_dname()
2496 if (IS_ROOT(dentry) &&
2497 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2498 WARN(1, "Root dentry has weird name <%.*s>\n",
2499 (int) dentry->d_name.len, dentry->d_name.name);
2501 if (!slash)
2502 error = prepend(buffer, buflen, "/", 1);
2503 if (!error)
2504 error = vfsmnt->mnt_ns ? 1 : 2;
2505 goto out;
2509 * __d_path - return the path of a dentry
2510 * @path: the dentry/vfsmount to report
2511 * @root: root vfsmnt/dentry
2512 * @buf: buffer to return value in
2513 * @buflen: buffer length
2515 * Convert a dentry into an ASCII path name.
2517 * Returns a pointer into the buffer or an error code if the
2518 * path was too long.
2520 * "buflen" should be positive.
2522 * If the path is not reachable from the supplied root, return %NULL.
2524 char *__d_path(const struct path *path,
2525 const struct path *root,
2526 char *buf, int buflen)
2528 char *res = buf + buflen;
2529 int error;
2531 prepend(&res, &buflen, "\0", 1);
2532 write_seqlock(&rename_lock);
2533 error = prepend_path(path, root, &res, &buflen);
2534 write_sequnlock(&rename_lock);
2536 if (error < 0)
2537 return ERR_PTR(error);
2538 if (error > 0)
2539 return NULL;
2540 return res;
2543 char *d_absolute_path(const struct path *path,
2544 char *buf, int buflen)
2546 struct path root = {};
2547 char *res = buf + buflen;
2548 int error;
2550 prepend(&res, &buflen, "\0", 1);
2551 write_seqlock(&rename_lock);
2552 error = prepend_path(path, &root, &res, &buflen);
2553 write_sequnlock(&rename_lock);
2555 if (error > 1)
2556 error = -EINVAL;
2557 if (error < 0)
2558 return ERR_PTR(error);
2559 return res;
2563 * same as __d_path but appends "(deleted)" for unlinked files.
2565 static int path_with_deleted(const struct path *path,
2566 const struct path *root,
2567 char **buf, int *buflen)
2569 prepend(buf, buflen, "\0", 1);
2570 if (d_unlinked(path->dentry)) {
2571 int error = prepend(buf, buflen, " (deleted)", 10);
2572 if (error)
2573 return error;
2576 return prepend_path(path, root, buf, buflen);
2579 static int prepend_unreachable(char **buffer, int *buflen)
2581 return prepend(buffer, buflen, "(unreachable)", 13);
2585 * d_path - return the path of a dentry
2586 * @path: path to report
2587 * @buf: buffer to return value in
2588 * @buflen: buffer length
2590 * Convert a dentry into an ASCII path name. If the entry has been deleted
2591 * the string " (deleted)" is appended. Note that this is ambiguous.
2593 * Returns a pointer into the buffer or an error code if the path was
2594 * too long. Note: Callers should use the returned pointer, not the passed
2595 * in buffer, to use the name! The implementation often starts at an offset
2596 * into the buffer, and may leave 0 bytes at the start.
2598 * "buflen" should be positive.
2600 char *d_path(const struct path *path, char *buf, int buflen)
2602 char *res = buf + buflen;
2603 struct path root;
2604 int error;
2607 * We have various synthetic filesystems that never get mounted. On
2608 * these filesystems dentries are never used for lookup purposes, and
2609 * thus don't need to be hashed. They also don't need a name until a
2610 * user wants to identify the object in /proc/pid/fd/. The little hack
2611 * below allows us to generate a name for these objects on demand:
2613 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2614 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2616 get_fs_root(current->fs, &root);
2617 write_seqlock(&rename_lock);
2618 error = path_with_deleted(path, &root, &res, &buflen);
2619 if (error < 0)
2620 res = ERR_PTR(error);
2621 write_sequnlock(&rename_lock);
2622 path_put(&root);
2623 return res;
2625 EXPORT_SYMBOL(d_path);
2628 * d_path_with_unreachable - return the path of a dentry
2629 * @path: path to report
2630 * @buf: buffer to return value in
2631 * @buflen: buffer length
2633 * The difference from d_path() is that this prepends "(unreachable)"
2634 * to paths which are unreachable from the current process' root.
2636 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2638 char *res = buf + buflen;
2639 struct path root;
2640 int error;
2642 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2643 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2645 get_fs_root(current->fs, &root);
2646 write_seqlock(&rename_lock);
2647 error = path_with_deleted(path, &root, &res, &buflen);
2648 if (error > 0)
2649 error = prepend_unreachable(&res, &buflen);
2650 write_sequnlock(&rename_lock);
2651 path_put(&root);
2652 if (error)
2653 res = ERR_PTR(error);
2655 return res;
2659 * Helper function for dentry_operations.d_dname() members
2661 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2662 const char *fmt, ...)
2664 va_list args;
2665 char temp[64];
2666 int sz;
2668 va_start(args, fmt);
2669 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2670 va_end(args);
2672 if (sz > sizeof(temp) || sz > buflen)
2673 return ERR_PTR(-ENAMETOOLONG);
2675 buffer += buflen - sz;
2676 return memcpy(buffer, temp, sz);
2680 * Write full pathname from the root of the filesystem into the buffer.
2682 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2684 char *end = buf + buflen;
2685 char *retval;
2687 prepend(&end, &buflen, "\0", 1);
2688 if (buflen < 1)
2689 goto Elong;
2690 /* Get '/' right */
2691 retval = end-1;
2692 *retval = '/';
2694 while (!IS_ROOT(dentry)) {
2695 struct dentry *parent = dentry->d_parent;
2696 int error;
2698 prefetch(parent);
2699 spin_lock(&dentry->d_lock);
2700 error = prepend_name(&end, &buflen, &dentry->d_name);
2701 spin_unlock(&dentry->d_lock);
2702 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2703 goto Elong;
2705 retval = end;
2706 dentry = parent;
2708 return retval;
2709 Elong:
2710 return ERR_PTR(-ENAMETOOLONG);
2713 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2715 char *retval;
2717 write_seqlock(&rename_lock);
2718 retval = __dentry_path(dentry, buf, buflen);
2719 write_sequnlock(&rename_lock);
2721 return retval;
2723 EXPORT_SYMBOL(dentry_path_raw);
2725 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2727 char *p = NULL;
2728 char *retval;
2730 write_seqlock(&rename_lock);
2731 if (d_unlinked(dentry)) {
2732 p = buf + buflen;
2733 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2734 goto Elong;
2735 buflen++;
2737 retval = __dentry_path(dentry, buf, buflen);
2738 write_sequnlock(&rename_lock);
2739 if (!IS_ERR(retval) && p)
2740 *p = '/'; /* restore '/' overriden with '\0' */
2741 return retval;
2742 Elong:
2743 return ERR_PTR(-ENAMETOOLONG);
2747 * NOTE! The user-level library version returns a
2748 * character pointer. The kernel system call just
2749 * returns the length of the buffer filled (which
2750 * includes the ending '\0' character), or a negative
2751 * error value. So libc would do something like
2753 * char *getcwd(char * buf, size_t size)
2755 * int retval;
2757 * retval = sys_getcwd(buf, size);
2758 * if (retval >= 0)
2759 * return buf;
2760 * errno = -retval;
2761 * return NULL;
2764 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2766 int error;
2767 struct path pwd, root;
2768 char *page = (char *) __get_free_page(GFP_USER);
2770 if (!page)
2771 return -ENOMEM;
2773 get_fs_root_and_pwd(current->fs, &root, &pwd);
2775 error = -ENOENT;
2776 write_seqlock(&rename_lock);
2777 if (!d_unlinked(pwd.dentry)) {
2778 unsigned long len;
2779 char *cwd = page + PAGE_SIZE;
2780 int buflen = PAGE_SIZE;
2782 prepend(&cwd, &buflen, "\0", 1);
2783 error = prepend_path(&pwd, &root, &cwd, &buflen);
2784 write_sequnlock(&rename_lock);
2786 if (error < 0)
2787 goto out;
2789 /* Unreachable from current root */
2790 if (error > 0) {
2791 error = prepend_unreachable(&cwd, &buflen);
2792 if (error)
2793 goto out;
2796 error = -ERANGE;
2797 len = PAGE_SIZE + page - cwd;
2798 if (len <= size) {
2799 error = len;
2800 if (copy_to_user(buf, cwd, len))
2801 error = -EFAULT;
2803 } else {
2804 write_sequnlock(&rename_lock);
2807 out:
2808 path_put(&pwd);
2809 path_put(&root);
2810 free_page((unsigned long) page);
2811 return error;
2815 * Test whether new_dentry is a subdirectory of old_dentry.
2817 * Trivially implemented using the dcache structure
2821 * is_subdir - is new dentry a subdirectory of old_dentry
2822 * @new_dentry: new dentry
2823 * @old_dentry: old dentry
2825 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2826 * Returns 0 otherwise.
2827 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2830 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2832 int result;
2833 unsigned seq;
2835 if (new_dentry == old_dentry)
2836 return 1;
2838 do {
2839 /* for restarting inner loop in case of seq retry */
2840 seq = read_seqbegin(&rename_lock);
2842 * Need rcu_readlock to protect against the d_parent trashing
2843 * due to d_move
2845 rcu_read_lock();
2846 if (d_ancestor(old_dentry, new_dentry))
2847 result = 1;
2848 else
2849 result = 0;
2850 rcu_read_unlock();
2851 } while (read_seqretry(&rename_lock, seq));
2853 return result;
2856 int path_is_under(struct path *path1, struct path *path2)
2858 struct vfsmount *mnt = path1->mnt;
2859 struct dentry *dentry = path1->dentry;
2860 int res;
2862 br_read_lock(vfsmount_lock);
2863 if (mnt != path2->mnt) {
2864 for (;;) {
2865 if (mnt->mnt_parent == mnt) {
2866 br_read_unlock(vfsmount_lock);
2867 return 0;
2869 if (mnt->mnt_parent == path2->mnt)
2870 break;
2871 mnt = mnt->mnt_parent;
2873 dentry = mnt->mnt_mountpoint;
2875 res = is_subdir(dentry, path2->dentry);
2876 br_read_unlock(vfsmount_lock);
2877 return res;
2879 EXPORT_SYMBOL(path_is_under);
2881 void d_genocide(struct dentry *root)
2883 struct dentry *this_parent;
2884 struct list_head *next;
2885 unsigned seq;
2886 int locked = 0;
2888 seq = read_seqbegin(&rename_lock);
2889 again:
2890 this_parent = root;
2891 spin_lock(&this_parent->d_lock);
2892 repeat:
2893 next = this_parent->d_subdirs.next;
2894 resume:
2895 while (next != &this_parent->d_subdirs) {
2896 struct list_head *tmp = next;
2897 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2898 next = tmp->next;
2900 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2901 if (d_unhashed(dentry) || !dentry->d_inode) {
2902 spin_unlock(&dentry->d_lock);
2903 continue;
2905 if (!list_empty(&dentry->d_subdirs)) {
2906 spin_unlock(&this_parent->d_lock);
2907 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2908 this_parent = dentry;
2909 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2910 goto repeat;
2912 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2913 dentry->d_flags |= DCACHE_GENOCIDE;
2914 dentry->d_count--;
2916 spin_unlock(&dentry->d_lock);
2918 if (this_parent != root) {
2919 struct dentry *child = this_parent;
2920 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2921 this_parent->d_flags |= DCACHE_GENOCIDE;
2922 this_parent->d_count--;
2924 this_parent = try_to_ascend(this_parent, locked, seq);
2925 if (!this_parent)
2926 goto rename_retry;
2927 next = child->d_u.d_child.next;
2928 goto resume;
2930 spin_unlock(&this_parent->d_lock);
2931 if (!locked && read_seqretry(&rename_lock, seq))
2932 goto rename_retry;
2933 if (locked)
2934 write_sequnlock(&rename_lock);
2935 return;
2937 rename_retry:
2938 locked = 1;
2939 write_seqlock(&rename_lock);
2940 goto again;
2944 * find_inode_number - check for dentry with name
2945 * @dir: directory to check
2946 * @name: Name to find.
2948 * Check whether a dentry already exists for the given name,
2949 * and return the inode number if it has an inode. Otherwise
2950 * 0 is returned.
2952 * This routine is used to post-process directory listings for
2953 * filesystems using synthetic inode numbers, and is necessary
2954 * to keep getcwd() working.
2957 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2959 struct dentry * dentry;
2960 ino_t ino = 0;
2962 dentry = d_hash_and_lookup(dir, name);
2963 if (dentry) {
2964 if (dentry->d_inode)
2965 ino = dentry->d_inode->i_ino;
2966 dput(dentry);
2968 return ino;
2970 EXPORT_SYMBOL(find_inode_number);
2972 static __initdata unsigned long dhash_entries;
2973 static int __init set_dhash_entries(char *str)
2975 if (!str)
2976 return 0;
2977 dhash_entries = simple_strtoul(str, &str, 0);
2978 return 1;
2980 __setup("dhash_entries=", set_dhash_entries);
2982 static void __init dcache_init_early(void)
2984 int loop;
2986 /* If hashes are distributed across NUMA nodes, defer
2987 * hash allocation until vmalloc space is available.
2989 if (hashdist)
2990 return;
2992 dentry_hashtable =
2993 alloc_large_system_hash("Dentry cache",
2994 sizeof(struct hlist_bl_head),
2995 dhash_entries,
2997 HASH_EARLY,
2998 &d_hash_shift,
2999 &d_hash_mask,
3002 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3003 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3006 static void __init dcache_init(void)
3008 int loop;
3011 * A constructor could be added for stable state like the lists,
3012 * but it is probably not worth it because of the cache nature
3013 * of the dcache.
3015 dentry_cache = KMEM_CACHE(dentry,
3016 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3018 /* Hash may have been set up in dcache_init_early */
3019 if (!hashdist)
3020 return;
3022 dentry_hashtable =
3023 alloc_large_system_hash("Dentry cache",
3024 sizeof(struct hlist_bl_head),
3025 dhash_entries,
3028 &d_hash_shift,
3029 &d_hash_mask,
3032 for (loop = 0; loop < (1 << d_hash_shift); loop++)
3033 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3036 /* SLAB cache for __getname() consumers */
3037 struct kmem_cache *names_cachep __read_mostly;
3038 EXPORT_SYMBOL(names_cachep);
3040 EXPORT_SYMBOL(d_genocide);
3042 void __init vfs_caches_init_early(void)
3044 dcache_init_early();
3045 inode_init_early();
3048 void __init vfs_caches_init(unsigned long mempages)
3050 unsigned long reserve;
3052 /* Base hash sizes on available memory, with a reserve equal to
3053 150% of current kernel size */
3055 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3056 mempages -= reserve;
3058 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3059 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3061 dcache_init();
3062 inode_init();
3063 files_init(mempages);
3064 mnt_init();
3065 bdev_cache_init();
3066 chrdev_init();