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>
45 * dcache->d_inode->i_lock protects:
46 * - i_dentry, d_alias, d_inode of aliases
47 * dcache_hash_bucket lock protects:
48 * - the dcache hash table
49 * s_anon bl list spinlock protects:
50 * - the s_anon list (see __d_drop)
51 * dcache_lru_lock protects:
52 * - the dcache lru lists and counters
59 * - d_parent and d_subdirs
60 * - childrens' d_child and d_parent
64 * dentry->d_inode->i_lock
67 * dcache_hash_bucket lock
70 * If there is an ancestor relationship:
71 * dentry->d_parent->...->d_parent->d_lock
73 * dentry->d_parent->d_lock
76 * If no ancestor relationship:
77 * if (dentry1 < dentry2)
81 int sysctl_vfs_cache_pressure __read_mostly = 100;
82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
84 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
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(unsigned int, nr_dentry);
120 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
121 static int get_nr_dentry(void)
125 for_each_possible_cpu(i)
126 sum += per_cpu(nr_dentry, i);
127 return sum < 0 ? 0 : sum;
130 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
131 size_t *lenp, loff_t *ppos)
133 dentry_stat.nr_dentry = get_nr_dentry();
134 return proc_dointvec(table, write, buffer, lenp, ppos);
139 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
140 * The strings are both count bytes long, and count is non-zero.
142 #ifdef CONFIG_DCACHE_WORD_ACCESS
144 #include <asm/word-at-a-time.h>
146 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
147 * aligned allocation for this particular component. We don't
148 * strictly need the load_unaligned_zeropad() safety, but it
149 * doesn't hurt either.
151 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
152 * need the careful unaligned handling.
154 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
156 unsigned long a,b,mask;
159 a = *(unsigned long *)cs;
160 b = load_unaligned_zeropad(ct);
161 if (tcount < sizeof(unsigned long))
163 if (unlikely(a != b))
165 cs += sizeof(unsigned long);
166 ct += sizeof(unsigned long);
167 tcount -= sizeof(unsigned long);
171 mask = ~(~0ul << tcount*8);
172 return unlikely(!!((a ^ b) & mask));
177 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
191 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
193 const unsigned char *cs;
195 * Be careful about RCU walk racing with rename:
196 * use ACCESS_ONCE to fetch the name pointer.
198 * NOTE! Even if a rename will mean that the length
199 * was not loaded atomically, we don't care. The
200 * RCU walk will check the sequence count eventually,
201 * and catch it. And we won't overrun the buffer,
202 * because we're reading the name pointer atomically,
203 * and a dentry name is guaranteed to be properly
204 * terminated with a NUL byte.
206 * End result: even if 'len' is wrong, we'll exit
207 * early because the data cannot match (there can
208 * be no NUL in the ct/tcount data)
210 cs = ACCESS_ONCE(dentry->d_name.name);
211 smp_read_barrier_depends();
212 return dentry_string_cmp(cs, ct, tcount);
215 static void __d_free(struct rcu_head *head)
217 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
219 WARN_ON(!hlist_unhashed(&dentry->d_alias));
220 if (dname_external(dentry))
221 kfree(dentry->d_name.name);
222 kmem_cache_free(dentry_cache, dentry);
228 static void d_free(struct dentry *dentry)
230 BUG_ON(dentry->d_count);
231 this_cpu_dec(nr_dentry);
232 if (dentry->d_op && dentry->d_op->d_release)
233 dentry->d_op->d_release(dentry);
235 /* if dentry was never visible to RCU, immediate free is OK */
236 if (!(dentry->d_flags & DCACHE_RCUACCESS))
237 __d_free(&dentry->d_u.d_rcu);
239 call_rcu(&dentry->d_u.d_rcu, __d_free);
243 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
244 * @dentry: the target dentry
245 * After this call, in-progress rcu-walk path lookup will fail. This
246 * should be called after unhashing, and after changing d_inode (if
247 * the dentry has not already been unhashed).
249 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
251 assert_spin_locked(&dentry->d_lock);
252 /* Go through a barrier */
253 write_seqcount_barrier(&dentry->d_seq);
257 * Release the dentry's inode, using the filesystem
258 * d_iput() operation if defined. Dentry has no refcount
261 static void dentry_iput(struct dentry * dentry)
262 __releases(dentry->d_lock)
263 __releases(dentry->d_inode->i_lock)
265 struct inode *inode = dentry->d_inode;
267 dentry->d_inode = NULL;
268 hlist_del_init(&dentry->d_alias);
269 spin_unlock(&dentry->d_lock);
270 spin_unlock(&inode->i_lock);
272 fsnotify_inoderemove(inode);
273 if (dentry->d_op && dentry->d_op->d_iput)
274 dentry->d_op->d_iput(dentry, inode);
278 spin_unlock(&dentry->d_lock);
283 * Release the dentry's inode, using the filesystem
284 * d_iput() operation if defined. dentry remains in-use.
286 static void dentry_unlink_inode(struct dentry * dentry)
287 __releases(dentry->d_lock)
288 __releases(dentry->d_inode->i_lock)
290 struct inode *inode = dentry->d_inode;
291 dentry->d_inode = NULL;
292 hlist_del_init(&dentry->d_alias);
293 dentry_rcuwalk_barrier(dentry);
294 spin_unlock(&dentry->d_lock);
295 spin_unlock(&inode->i_lock);
297 fsnotify_inoderemove(inode);
298 if (dentry->d_op && dentry->d_op->d_iput)
299 dentry->d_op->d_iput(dentry, inode);
305 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
307 static void dentry_lru_add(struct dentry *dentry)
309 if (list_empty(&dentry->d_lru)) {
310 spin_lock(&dcache_lru_lock);
311 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
312 dentry->d_sb->s_nr_dentry_unused++;
313 dentry_stat.nr_unused++;
314 spin_unlock(&dcache_lru_lock);
318 static void __dentry_lru_del(struct dentry *dentry)
320 list_del_init(&dentry->d_lru);
321 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
322 dentry->d_sb->s_nr_dentry_unused--;
323 dentry_stat.nr_unused--;
327 * Remove a dentry with references from the LRU.
329 static void dentry_lru_del(struct dentry *dentry)
331 if (!list_empty(&dentry->d_lru)) {
332 spin_lock(&dcache_lru_lock);
333 __dentry_lru_del(dentry);
334 spin_unlock(&dcache_lru_lock);
338 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
340 spin_lock(&dcache_lru_lock);
341 if (list_empty(&dentry->d_lru)) {
342 list_add_tail(&dentry->d_lru, list);
343 dentry->d_sb->s_nr_dentry_unused++;
344 dentry_stat.nr_unused++;
346 list_move_tail(&dentry->d_lru, list);
348 spin_unlock(&dcache_lru_lock);
352 * d_kill - kill dentry and return parent
353 * @dentry: dentry to kill
354 * @parent: parent dentry
356 * The dentry must already be unhashed and removed from the LRU.
358 * If this is the root of the dentry tree, return NULL.
360 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
363 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
364 __releases(dentry->d_lock)
365 __releases(parent->d_lock)
366 __releases(dentry->d_inode->i_lock)
368 list_del(&dentry->d_u.d_child);
370 * Inform try_to_ascend() that we are no longer attached to the
373 dentry->d_flags |= DCACHE_DENTRY_KILLED;
375 spin_unlock(&parent->d_lock);
378 * dentry_iput drops the locks, at which point nobody (except
379 * transient RCU lookups) can reach this dentry.
386 * Unhash a dentry without inserting an RCU walk barrier or checking that
387 * dentry->d_lock is locked. The caller must take care of that, if
390 static void __d_shrink(struct dentry *dentry)
392 if (!d_unhashed(dentry)) {
393 struct hlist_bl_head *b;
394 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
395 b = &dentry->d_sb->s_anon;
397 b = d_hash(dentry->d_parent, dentry->d_name.hash);
400 __hlist_bl_del(&dentry->d_hash);
401 dentry->d_hash.pprev = NULL;
407 * d_drop - drop a dentry
408 * @dentry: dentry to drop
410 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
411 * be found through a VFS lookup any more. Note that this is different from
412 * deleting the dentry - d_delete will try to mark the dentry negative if
413 * possible, giving a successful _negative_ lookup, while d_drop will
414 * just make the cache lookup fail.
416 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
417 * reason (NFS timeouts or autofs deletes).
419 * __d_drop requires dentry->d_lock.
421 void __d_drop(struct dentry *dentry)
423 if (!d_unhashed(dentry)) {
425 dentry_rcuwalk_barrier(dentry);
428 EXPORT_SYMBOL(__d_drop);
430 void d_drop(struct dentry *dentry)
432 spin_lock(&dentry->d_lock);
434 spin_unlock(&dentry->d_lock);
436 EXPORT_SYMBOL(d_drop);
439 * Finish off a dentry we've decided to kill.
440 * dentry->d_lock must be held, returns with it unlocked.
441 * If ref is non-zero, then decrement the refcount too.
442 * Returns dentry requiring refcount drop, or NULL if we're done.
444 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
445 __releases(dentry->d_lock)
448 struct dentry *parent;
450 inode = dentry->d_inode;
451 if (inode && !spin_trylock(&inode->i_lock)) {
453 spin_unlock(&dentry->d_lock);
455 return dentry; /* try again with same dentry */
460 parent = dentry->d_parent;
461 if (parent && !spin_trylock(&parent->d_lock)) {
463 spin_unlock(&inode->i_lock);
470 * inform the fs via d_prune that this dentry is about to be
471 * unhashed and destroyed.
473 if (dentry->d_flags & DCACHE_OP_PRUNE)
474 dentry->d_op->d_prune(dentry);
476 dentry_lru_del(dentry);
477 /* if it was on the hash then remove it */
479 return d_kill(dentry, parent);
485 * This is complicated by the fact that we do not want to put
486 * dentries that are no longer on any hash chain on the unused
487 * list: we'd much rather just get rid of them immediately.
489 * However, that implies that we have to traverse the dentry
490 * tree upwards to the parents which might _also_ now be
491 * scheduled for deletion (it may have been only waiting for
492 * its last child to go away).
494 * This tail recursion is done by hand as we don't want to depend
495 * on the compiler to always get this right (gcc generally doesn't).
496 * Real recursion would eat up our stack space.
500 * dput - release a dentry
501 * @dentry: dentry to release
503 * Release a dentry. This will drop the usage count and if appropriate
504 * call the dentry unlink method as well as removing it from the queues and
505 * releasing its resources. If the parent dentries were scheduled for release
506 * they too may now get deleted.
508 void dput(struct dentry *dentry)
514 if (dentry->d_count == 1)
516 spin_lock(&dentry->d_lock);
517 BUG_ON(!dentry->d_count);
518 if (dentry->d_count > 1) {
520 spin_unlock(&dentry->d_lock);
524 if (dentry->d_flags & DCACHE_OP_DELETE) {
525 if (dentry->d_op->d_delete(dentry))
529 /* Unreachable? Get rid of it */
530 if (d_unhashed(dentry))
533 dentry->d_flags |= DCACHE_REFERENCED;
534 dentry_lru_add(dentry);
537 spin_unlock(&dentry->d_lock);
541 dentry = dentry_kill(dentry, 1);
548 * d_invalidate - invalidate a dentry
549 * @dentry: dentry to invalidate
551 * Try to invalidate the dentry if it turns out to be
552 * possible. If there are other dentries that can be
553 * reached through this one we can't delete it and we
554 * return -EBUSY. On success we return 0.
559 int d_invalidate(struct dentry * dentry)
562 * If it's already been dropped, return OK.
564 spin_lock(&dentry->d_lock);
565 if (d_unhashed(dentry)) {
566 spin_unlock(&dentry->d_lock);
570 * Check whether to do a partial shrink_dcache
571 * to get rid of unused child entries.
573 if (!list_empty(&dentry->d_subdirs)) {
574 spin_unlock(&dentry->d_lock);
575 shrink_dcache_parent(dentry);
576 spin_lock(&dentry->d_lock);
580 * Somebody else still using it?
582 * If it's a directory, we can't drop it
583 * for fear of somebody re-populating it
584 * with children (even though dropping it
585 * would make it unreachable from the root,
586 * we might still populate it if it was a
587 * working directory or similar).
588 * We also need to leave mountpoints alone,
591 if (dentry->d_count > 1 && dentry->d_inode) {
592 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
593 spin_unlock(&dentry->d_lock);
599 spin_unlock(&dentry->d_lock);
602 EXPORT_SYMBOL(d_invalidate);
604 /* This must be called with d_lock held */
605 static inline void __dget_dlock(struct dentry *dentry)
610 static inline void __dget(struct dentry *dentry)
612 spin_lock(&dentry->d_lock);
613 __dget_dlock(dentry);
614 spin_unlock(&dentry->d_lock);
617 struct dentry *dget_parent(struct dentry *dentry)
623 * Don't need rcu_dereference because we re-check it was correct under
627 ret = dentry->d_parent;
628 spin_lock(&ret->d_lock);
629 if (unlikely(ret != dentry->d_parent)) {
630 spin_unlock(&ret->d_lock);
635 BUG_ON(!ret->d_count);
637 spin_unlock(&ret->d_lock);
640 EXPORT_SYMBOL(dget_parent);
643 * d_find_alias - grab a hashed alias of inode
644 * @inode: inode in question
645 * @want_discon: flag, used by d_splice_alias, to request
646 * that only a DISCONNECTED alias be returned.
648 * If inode has a hashed alias, or is a directory and has any alias,
649 * acquire the reference to alias and return it. Otherwise return NULL.
650 * Notice that if inode is a directory there can be only one alias and
651 * it can be unhashed only if it has no children, or if it is the root
654 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
655 * any other hashed alias over that one unless @want_discon is set,
656 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
658 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
660 struct dentry *alias, *discon_alias;
664 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
665 spin_lock(&alias->d_lock);
666 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
667 if (IS_ROOT(alias) &&
668 (alias->d_flags & DCACHE_DISCONNECTED)) {
669 discon_alias = alias;
670 } else if (!want_discon) {
672 spin_unlock(&alias->d_lock);
676 spin_unlock(&alias->d_lock);
679 alias = discon_alias;
680 spin_lock(&alias->d_lock);
681 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
682 if (IS_ROOT(alias) &&
683 (alias->d_flags & DCACHE_DISCONNECTED)) {
685 spin_unlock(&alias->d_lock);
689 spin_unlock(&alias->d_lock);
695 struct dentry *d_find_alias(struct inode *inode)
697 struct dentry *de = NULL;
699 if (!hlist_empty(&inode->i_dentry)) {
700 spin_lock(&inode->i_lock);
701 de = __d_find_alias(inode, 0);
702 spin_unlock(&inode->i_lock);
706 EXPORT_SYMBOL(d_find_alias);
709 * Try to kill dentries associated with this inode.
710 * WARNING: you must own a reference to inode.
712 void d_prune_aliases(struct inode *inode)
714 struct dentry *dentry;
716 spin_lock(&inode->i_lock);
717 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
718 spin_lock(&dentry->d_lock);
719 if (!dentry->d_count) {
720 __dget_dlock(dentry);
722 spin_unlock(&dentry->d_lock);
723 spin_unlock(&inode->i_lock);
727 spin_unlock(&dentry->d_lock);
729 spin_unlock(&inode->i_lock);
731 EXPORT_SYMBOL(d_prune_aliases);
734 * Try to throw away a dentry - free the inode, dput the parent.
735 * Requires dentry->d_lock is held, and dentry->d_count == 0.
736 * Releases dentry->d_lock.
738 * This may fail if locks cannot be acquired no problem, just try again.
740 static void try_prune_one_dentry(struct dentry *dentry)
741 __releases(dentry->d_lock)
743 struct dentry *parent;
745 parent = dentry_kill(dentry, 0);
747 * If dentry_kill returns NULL, we have nothing more to do.
748 * if it returns the same dentry, trylocks failed. In either
749 * case, just loop again.
751 * Otherwise, we need to prune ancestors too. This is necessary
752 * to prevent quadratic behavior of shrink_dcache_parent(), but
753 * is also expected to be beneficial in reducing dentry cache
758 if (parent == dentry)
761 /* Prune ancestors. */
764 spin_lock(&dentry->d_lock);
765 if (dentry->d_count > 1) {
767 spin_unlock(&dentry->d_lock);
770 dentry = dentry_kill(dentry, 1);
774 static void shrink_dentry_list(struct list_head *list)
776 struct dentry *dentry;
780 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
781 if (&dentry->d_lru == list)
783 spin_lock(&dentry->d_lock);
784 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
785 spin_unlock(&dentry->d_lock);
790 * We found an inuse dentry which was not removed from
791 * the LRU because of laziness during lookup. Do not free
792 * it - just keep it off the LRU list.
794 if (dentry->d_count) {
795 dentry_lru_del(dentry);
796 spin_unlock(&dentry->d_lock);
802 try_prune_one_dentry(dentry);
810 * prune_dcache_sb - shrink the dcache
812 * @count: number of entries to try to free
814 * Attempt to shrink the superblock dcache LRU by @count entries. This is
815 * done when we need more memory an called from the superblock shrinker
818 * This function may fail to free any resources if all the dentries are in
821 void prune_dcache_sb(struct super_block *sb, int count)
823 struct dentry *dentry;
824 LIST_HEAD(referenced);
828 spin_lock(&dcache_lru_lock);
829 while (!list_empty(&sb->s_dentry_lru)) {
830 dentry = list_entry(sb->s_dentry_lru.prev,
831 struct dentry, d_lru);
832 BUG_ON(dentry->d_sb != sb);
834 if (!spin_trylock(&dentry->d_lock)) {
835 spin_unlock(&dcache_lru_lock);
840 if (dentry->d_flags & DCACHE_REFERENCED) {
841 dentry->d_flags &= ~DCACHE_REFERENCED;
842 list_move(&dentry->d_lru, &referenced);
843 spin_unlock(&dentry->d_lock);
845 list_move_tail(&dentry->d_lru, &tmp);
846 dentry->d_flags |= DCACHE_SHRINK_LIST;
847 spin_unlock(&dentry->d_lock);
851 cond_resched_lock(&dcache_lru_lock);
853 if (!list_empty(&referenced))
854 list_splice(&referenced, &sb->s_dentry_lru);
855 spin_unlock(&dcache_lru_lock);
857 shrink_dentry_list(&tmp);
861 * shrink_dcache_sb - shrink dcache for a superblock
864 * Shrink the dcache for the specified super block. This is used to free
865 * the dcache before unmounting a file system.
867 void shrink_dcache_sb(struct super_block *sb)
871 spin_lock(&dcache_lru_lock);
872 while (!list_empty(&sb->s_dentry_lru)) {
873 list_splice_init(&sb->s_dentry_lru, &tmp);
874 spin_unlock(&dcache_lru_lock);
875 shrink_dentry_list(&tmp);
876 spin_lock(&dcache_lru_lock);
878 spin_unlock(&dcache_lru_lock);
880 EXPORT_SYMBOL(shrink_dcache_sb);
883 * destroy a single subtree of dentries for unmount
884 * - see the comments on shrink_dcache_for_umount() for a description of the
887 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
889 struct dentry *parent;
891 BUG_ON(!IS_ROOT(dentry));
894 /* descend to the first leaf in the current subtree */
895 while (!list_empty(&dentry->d_subdirs))
896 dentry = list_entry(dentry->d_subdirs.next,
897 struct dentry, d_u.d_child);
899 /* consume the dentries from this leaf up through its parents
900 * until we find one with children or run out altogether */
905 * inform the fs that this dentry is about to be
906 * unhashed and destroyed.
908 if (dentry->d_flags & DCACHE_OP_PRUNE)
909 dentry->d_op->d_prune(dentry);
911 dentry_lru_del(dentry);
914 if (dentry->d_count != 0) {
916 "BUG: Dentry %p{i=%lx,n=%s}"
918 " [unmount of %s %s]\n",
921 dentry->d_inode->i_ino : 0UL,
924 dentry->d_sb->s_type->name,
929 if (IS_ROOT(dentry)) {
931 list_del(&dentry->d_u.d_child);
933 parent = dentry->d_parent;
935 list_del(&dentry->d_u.d_child);
938 inode = dentry->d_inode;
940 dentry->d_inode = NULL;
941 hlist_del_init(&dentry->d_alias);
942 if (dentry->d_op && dentry->d_op->d_iput)
943 dentry->d_op->d_iput(dentry, inode);
950 /* finished when we fall off the top of the tree,
951 * otherwise we ascend to the parent and move to the
952 * next sibling if there is one */
956 } while (list_empty(&dentry->d_subdirs));
958 dentry = list_entry(dentry->d_subdirs.next,
959 struct dentry, d_u.d_child);
964 * destroy the dentries attached to a superblock on unmounting
965 * - we don't need to use dentry->d_lock because:
966 * - the superblock is detached from all mountings and open files, so the
967 * dentry trees will not be rearranged by the VFS
968 * - s_umount is write-locked, so the memory pressure shrinker will ignore
969 * any dentries belonging to this superblock that it comes across
970 * - the filesystem itself is no longer permitted to rearrange the dentries
973 void shrink_dcache_for_umount(struct super_block *sb)
975 struct dentry *dentry;
977 if (down_read_trylock(&sb->s_umount))
983 shrink_dcache_for_umount_subtree(dentry);
985 while (!hlist_bl_empty(&sb->s_anon)) {
986 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
987 shrink_dcache_for_umount_subtree(dentry);
992 * This tries to ascend one level of parenthood, but
993 * we can race with renaming, so we need to re-check
994 * the parenthood after dropping the lock and check
995 * that the sequence number still matches.
997 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
999 struct dentry *new = old->d_parent;
1002 spin_unlock(&old->d_lock);
1003 spin_lock(&new->d_lock);
1006 * might go back up the wrong parent if we have had a rename
1009 if (new != old->d_parent ||
1010 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1011 (!locked && read_seqretry(&rename_lock, seq))) {
1012 spin_unlock(&new->d_lock);
1021 * Search for at least 1 mount point in the dentry's subdirs.
1022 * We descend to the next level whenever the d_subdirs
1023 * list is non-empty and continue searching.
1027 * have_submounts - check for mounts over a dentry
1028 * @parent: dentry to check.
1030 * Return true if the parent or its subdirectories contain
1033 int have_submounts(struct dentry *parent)
1035 struct dentry *this_parent;
1036 struct list_head *next;
1040 seq = read_seqbegin(&rename_lock);
1042 this_parent = parent;
1044 if (d_mountpoint(parent))
1046 spin_lock(&this_parent->d_lock);
1048 next = this_parent->d_subdirs.next;
1050 while (next != &this_parent->d_subdirs) {
1051 struct list_head *tmp = next;
1052 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1055 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1056 /* Have we found a mount point ? */
1057 if (d_mountpoint(dentry)) {
1058 spin_unlock(&dentry->d_lock);
1059 spin_unlock(&this_parent->d_lock);
1062 if (!list_empty(&dentry->d_subdirs)) {
1063 spin_unlock(&this_parent->d_lock);
1064 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1065 this_parent = dentry;
1066 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1069 spin_unlock(&dentry->d_lock);
1072 * All done at this level ... ascend and resume the search.
1074 if (this_parent != parent) {
1075 struct dentry *child = this_parent;
1076 this_parent = try_to_ascend(this_parent, locked, seq);
1079 next = child->d_u.d_child.next;
1082 spin_unlock(&this_parent->d_lock);
1083 if (!locked && read_seqretry(&rename_lock, seq))
1086 write_sequnlock(&rename_lock);
1087 return 0; /* No mount points found in tree */
1089 if (!locked && read_seqretry(&rename_lock, seq))
1092 write_sequnlock(&rename_lock);
1099 write_seqlock(&rename_lock);
1102 EXPORT_SYMBOL(have_submounts);
1105 * Search the dentry child list of the specified parent,
1106 * and move any unused dentries to the end of the unused
1107 * list for prune_dcache(). We descend to the next level
1108 * whenever the d_subdirs list is non-empty and continue
1111 * It returns zero iff there are no unused children,
1112 * otherwise it returns the number of children moved to
1113 * the end of the unused list. This may not be the total
1114 * number of unused children, because select_parent can
1115 * drop the lock and return early due to latency
1118 static int select_parent(struct dentry *parent, struct list_head *dispose)
1120 struct dentry *this_parent;
1121 struct list_head *next;
1126 seq = read_seqbegin(&rename_lock);
1128 this_parent = parent;
1129 spin_lock(&this_parent->d_lock);
1131 next = this_parent->d_subdirs.next;
1133 while (next != &this_parent->d_subdirs) {
1134 struct list_head *tmp = next;
1135 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1138 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1141 * move only zero ref count dentries to the dispose list.
1143 * Those which are presently on the shrink list, being processed
1144 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1145 * loop in shrink_dcache_parent() might not make any progress
1148 if (dentry->d_count) {
1149 dentry_lru_del(dentry);
1150 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1151 dentry_lru_move_list(dentry, dispose);
1152 dentry->d_flags |= DCACHE_SHRINK_LIST;
1156 * We can return to the caller if we have found some (this
1157 * ensures forward progress). We'll be coming back to find
1160 if (found && need_resched()) {
1161 spin_unlock(&dentry->d_lock);
1166 * Descend a level if the d_subdirs list is non-empty.
1168 if (!list_empty(&dentry->d_subdirs)) {
1169 spin_unlock(&this_parent->d_lock);
1170 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1171 this_parent = dentry;
1172 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1176 spin_unlock(&dentry->d_lock);
1179 * All done at this level ... ascend and resume the search.
1181 if (this_parent != parent) {
1182 struct dentry *child = this_parent;
1183 this_parent = try_to_ascend(this_parent, locked, seq);
1186 next = child->d_u.d_child.next;
1190 spin_unlock(&this_parent->d_lock);
1191 if (!locked && read_seqretry(&rename_lock, seq))
1194 write_sequnlock(&rename_lock);
1203 write_seqlock(&rename_lock);
1208 * shrink_dcache_parent - prune dcache
1209 * @parent: parent of entries to prune
1211 * Prune the dcache to remove unused children of the parent dentry.
1213 void shrink_dcache_parent(struct dentry * parent)
1218 while ((found = select_parent(parent, &dispose)) != 0) {
1219 shrink_dentry_list(&dispose);
1223 EXPORT_SYMBOL(shrink_dcache_parent);
1226 * __d_alloc - allocate a dcache entry
1227 * @sb: filesystem it will belong to
1228 * @name: qstr of the name
1230 * Allocates a dentry. It returns %NULL if there is insufficient memory
1231 * available. On a success the dentry is returned. The name passed in is
1232 * copied and the copy passed in may be reused after this call.
1235 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1237 struct dentry *dentry;
1240 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1245 * We guarantee that the inline name is always NUL-terminated.
1246 * This way the memcpy() done by the name switching in rename
1247 * will still always have a NUL at the end, even if we might
1248 * be overwriting an internal NUL character
1250 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1251 if (name->len > DNAME_INLINE_LEN-1) {
1252 dname = kmalloc(name->len + 1, GFP_KERNEL);
1254 kmem_cache_free(dentry_cache, dentry);
1258 dname = dentry->d_iname;
1261 dentry->d_name.len = name->len;
1262 dentry->d_name.hash = name->hash;
1263 memcpy(dname, name->name, name->len);
1264 dname[name->len] = 0;
1266 /* Make sure we always see the terminating NUL character */
1268 dentry->d_name.name = dname;
1270 dentry->d_count = 1;
1271 dentry->d_flags = 0;
1272 spin_lock_init(&dentry->d_lock);
1273 seqcount_init(&dentry->d_seq);
1274 dentry->d_inode = NULL;
1275 dentry->d_parent = dentry;
1277 dentry->d_op = NULL;
1278 dentry->d_fsdata = NULL;
1279 INIT_HLIST_BL_NODE(&dentry->d_hash);
1280 INIT_LIST_HEAD(&dentry->d_lru);
1281 INIT_LIST_HEAD(&dentry->d_subdirs);
1282 INIT_HLIST_NODE(&dentry->d_alias);
1283 INIT_LIST_HEAD(&dentry->d_u.d_child);
1284 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1286 this_cpu_inc(nr_dentry);
1292 * d_alloc - allocate a dcache entry
1293 * @parent: parent of entry to allocate
1294 * @name: qstr of the name
1296 * Allocates a dentry. It returns %NULL if there is insufficient memory
1297 * available. On a success the dentry is returned. The name passed in is
1298 * copied and the copy passed in may be reused after this call.
1300 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1302 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1306 spin_lock(&parent->d_lock);
1308 * don't need child lock because it is not subject
1309 * to concurrency here
1311 __dget_dlock(parent);
1312 dentry->d_parent = parent;
1313 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1314 spin_unlock(&parent->d_lock);
1318 EXPORT_SYMBOL(d_alloc);
1320 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1322 struct dentry *dentry = __d_alloc(sb, name);
1324 dentry->d_flags |= DCACHE_DISCONNECTED;
1327 EXPORT_SYMBOL(d_alloc_pseudo);
1329 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1334 q.len = strlen(name);
1335 q.hash = full_name_hash(q.name, q.len);
1336 return d_alloc(parent, &q);
1338 EXPORT_SYMBOL(d_alloc_name);
1340 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1342 WARN_ON_ONCE(dentry->d_op);
1343 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1345 DCACHE_OP_REVALIDATE |
1346 DCACHE_OP_WEAK_REVALIDATE |
1347 DCACHE_OP_DELETE ));
1352 dentry->d_flags |= DCACHE_OP_HASH;
1354 dentry->d_flags |= DCACHE_OP_COMPARE;
1355 if (op->d_revalidate)
1356 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1357 if (op->d_weak_revalidate)
1358 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1360 dentry->d_flags |= DCACHE_OP_DELETE;
1362 dentry->d_flags |= DCACHE_OP_PRUNE;
1365 EXPORT_SYMBOL(d_set_d_op);
1367 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1369 spin_lock(&dentry->d_lock);
1371 if (unlikely(IS_AUTOMOUNT(inode)))
1372 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1373 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1375 dentry->d_inode = inode;
1376 dentry_rcuwalk_barrier(dentry);
1377 spin_unlock(&dentry->d_lock);
1378 fsnotify_d_instantiate(dentry, inode);
1382 * d_instantiate - fill in inode information for a dentry
1383 * @entry: dentry to complete
1384 * @inode: inode to attach to this dentry
1386 * Fill in inode information in the entry.
1388 * This turns negative dentries into productive full members
1391 * NOTE! This assumes that the inode count has been incremented
1392 * (or otherwise set) by the caller to indicate that it is now
1393 * in use by the dcache.
1396 void d_instantiate(struct dentry *entry, struct inode * inode)
1398 BUG_ON(!hlist_unhashed(&entry->d_alias));
1400 spin_lock(&inode->i_lock);
1401 __d_instantiate(entry, inode);
1403 spin_unlock(&inode->i_lock);
1404 security_d_instantiate(entry, inode);
1406 EXPORT_SYMBOL(d_instantiate);
1409 * d_instantiate_unique - instantiate a non-aliased dentry
1410 * @entry: dentry to instantiate
1411 * @inode: inode to attach to this dentry
1413 * Fill in inode information in the entry. On success, it returns NULL.
1414 * If an unhashed alias of "entry" already exists, then we return the
1415 * aliased dentry instead and drop one reference to inode.
1417 * Note that in order to avoid conflicts with rename() etc, the caller
1418 * had better be holding the parent directory semaphore.
1420 * This also assumes that the inode count has been incremented
1421 * (or otherwise set) by the caller to indicate that it is now
1422 * in use by the dcache.
1424 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1425 struct inode *inode)
1427 struct dentry *alias;
1428 int len = entry->d_name.len;
1429 const char *name = entry->d_name.name;
1430 unsigned int hash = entry->d_name.hash;
1433 __d_instantiate(entry, NULL);
1437 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1439 * Don't need alias->d_lock here, because aliases with
1440 * d_parent == entry->d_parent are not subject to name or
1441 * parent changes, because the parent inode i_mutex is held.
1443 if (alias->d_name.hash != hash)
1445 if (alias->d_parent != entry->d_parent)
1447 if (alias->d_name.len != len)
1449 if (dentry_cmp(alias, name, len))
1455 __d_instantiate(entry, inode);
1459 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1461 struct dentry *result;
1463 BUG_ON(!hlist_unhashed(&entry->d_alias));
1466 spin_lock(&inode->i_lock);
1467 result = __d_instantiate_unique(entry, inode);
1469 spin_unlock(&inode->i_lock);
1472 security_d_instantiate(entry, inode);
1476 BUG_ON(!d_unhashed(result));
1481 EXPORT_SYMBOL(d_instantiate_unique);
1483 struct dentry *d_make_root(struct inode *root_inode)
1485 struct dentry *res = NULL;
1488 static const struct qstr name = QSTR_INIT("/", 1);
1490 res = __d_alloc(root_inode->i_sb, &name);
1492 d_instantiate(res, root_inode);
1498 EXPORT_SYMBOL(d_make_root);
1500 static struct dentry * __d_find_any_alias(struct inode *inode)
1502 struct dentry *alias;
1504 if (hlist_empty(&inode->i_dentry))
1506 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1512 * d_find_any_alias - find any alias for a given inode
1513 * @inode: inode to find an alias for
1515 * If any aliases exist for the given inode, take and return a
1516 * reference for one of them. If no aliases exist, return %NULL.
1518 struct dentry *d_find_any_alias(struct inode *inode)
1522 spin_lock(&inode->i_lock);
1523 de = __d_find_any_alias(inode);
1524 spin_unlock(&inode->i_lock);
1527 EXPORT_SYMBOL(d_find_any_alias);
1530 * d_obtain_alias - find or allocate a dentry for a given inode
1531 * @inode: inode to allocate the dentry for
1533 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1534 * similar open by handle operations. The returned dentry may be anonymous,
1535 * or may have a full name (if the inode was already in the cache).
1537 * When called on a directory inode, we must ensure that the inode only ever
1538 * has one dentry. If a dentry is found, that is returned instead of
1539 * allocating a new one.
1541 * On successful return, the reference to the inode has been transferred
1542 * to the dentry. In case of an error the reference on the inode is released.
1543 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1544 * be passed in and will be the error will be propagate to the return value,
1545 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1547 struct dentry *d_obtain_alias(struct inode *inode)
1549 static const struct qstr anonstring = QSTR_INIT("/", 1);
1554 return ERR_PTR(-ESTALE);
1556 return ERR_CAST(inode);
1558 res = d_find_any_alias(inode);
1562 tmp = __d_alloc(inode->i_sb, &anonstring);
1564 res = ERR_PTR(-ENOMEM);
1568 spin_lock(&inode->i_lock);
1569 res = __d_find_any_alias(inode);
1571 spin_unlock(&inode->i_lock);
1576 /* attach a disconnected dentry */
1577 spin_lock(&tmp->d_lock);
1578 tmp->d_inode = inode;
1579 tmp->d_flags |= DCACHE_DISCONNECTED;
1580 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1581 hlist_bl_lock(&tmp->d_sb->s_anon);
1582 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1583 hlist_bl_unlock(&tmp->d_sb->s_anon);
1584 spin_unlock(&tmp->d_lock);
1585 spin_unlock(&inode->i_lock);
1586 security_d_instantiate(tmp, inode);
1591 if (res && !IS_ERR(res))
1592 security_d_instantiate(res, inode);
1596 EXPORT_SYMBOL(d_obtain_alias);
1599 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1600 * @inode: the inode which may have a disconnected dentry
1601 * @dentry: a negative dentry which we want to point to the inode.
1603 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1604 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1605 * and return it, else simply d_add the inode to the dentry and return NULL.
1607 * This is needed in the lookup routine of any filesystem that is exportable
1608 * (via knfsd) so that we can build dcache paths to directories effectively.
1610 * If a dentry was found and moved, then it is returned. Otherwise NULL
1611 * is returned. This matches the expected return value of ->lookup.
1614 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1616 struct dentry *new = NULL;
1619 return ERR_CAST(inode);
1621 if (inode && S_ISDIR(inode->i_mode)) {
1622 spin_lock(&inode->i_lock);
1623 new = __d_find_alias(inode, 1);
1625 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1626 spin_unlock(&inode->i_lock);
1627 security_d_instantiate(new, inode);
1628 d_move(new, dentry);
1631 /* already taking inode->i_lock, so d_add() by hand */
1632 __d_instantiate(dentry, inode);
1633 spin_unlock(&inode->i_lock);
1634 security_d_instantiate(dentry, inode);
1638 d_add(dentry, inode);
1641 EXPORT_SYMBOL(d_splice_alias);
1644 * d_add_ci - lookup or allocate new dentry with case-exact name
1645 * @inode: the inode case-insensitive lookup has found
1646 * @dentry: the negative dentry that was passed to the parent's lookup func
1647 * @name: the case-exact name to be associated with the returned dentry
1649 * This is to avoid filling the dcache with case-insensitive names to the
1650 * same inode, only the actual correct case is stored in the dcache for
1651 * case-insensitive filesystems.
1653 * For a case-insensitive lookup match and if the the case-exact dentry
1654 * already exists in in the dcache, use it and return it.
1656 * If no entry exists with the exact case name, allocate new dentry with
1657 * the exact case, and return the spliced entry.
1659 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1662 struct dentry *found;
1666 * First check if a dentry matching the name already exists,
1667 * if not go ahead and create it now.
1669 found = d_hash_and_lookup(dentry->d_parent, name);
1670 if (unlikely(IS_ERR(found)))
1673 new = d_alloc(dentry->d_parent, name);
1675 found = ERR_PTR(-ENOMEM);
1679 found = d_splice_alias(inode, new);
1688 * If a matching dentry exists, and it's not negative use it.
1690 * Decrement the reference count to balance the iget() done
1693 if (found->d_inode) {
1694 if (unlikely(found->d_inode != inode)) {
1695 /* This can't happen because bad inodes are unhashed. */
1696 BUG_ON(!is_bad_inode(inode));
1697 BUG_ON(!is_bad_inode(found->d_inode));
1704 * Negative dentry: instantiate it unless the inode is a directory and
1705 * already has a dentry.
1707 new = d_splice_alias(inode, found);
1718 EXPORT_SYMBOL(d_add_ci);
1721 * Do the slow-case of the dentry name compare.
1723 * Unlike the dentry_cmp() function, we need to atomically
1724 * load the name, length and inode information, so that the
1725 * filesystem can rely on them, and can use the 'name' and
1726 * 'len' information without worrying about walking off the
1727 * end of memory etc.
1729 * Thus the read_seqcount_retry() and the "duplicate" info
1730 * in arguments (the low-level filesystem should not look
1731 * at the dentry inode or name contents directly, since
1732 * rename can change them while we're in RCU mode).
1734 enum slow_d_compare {
1740 static noinline enum slow_d_compare slow_dentry_cmp(
1741 const struct dentry *parent,
1742 struct inode *inode,
1743 struct dentry *dentry,
1745 const struct qstr *name)
1747 int tlen = dentry->d_name.len;
1748 const char *tname = dentry->d_name.name;
1749 struct inode *i = dentry->d_inode;
1751 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1753 return D_COMP_SEQRETRY;
1755 if (parent->d_op->d_compare(parent, inode,
1758 return D_COMP_NOMATCH;
1763 * __d_lookup_rcu - search for a dentry (racy, store-free)
1764 * @parent: parent dentry
1765 * @name: qstr of name we wish to find
1766 * @seqp: returns d_seq value at the point where the dentry was found
1767 * @inode: returns dentry->d_inode when the inode was found valid.
1768 * Returns: dentry, or NULL
1770 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1771 * resolution (store-free path walking) design described in
1772 * Documentation/filesystems/path-lookup.txt.
1774 * This is not to be used outside core vfs.
1776 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1777 * held, and rcu_read_lock held. The returned dentry must not be stored into
1778 * without taking d_lock and checking d_seq sequence count against @seq
1781 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1784 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1785 * the returned dentry, so long as its parent's seqlock is checked after the
1786 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1787 * is formed, giving integrity down the path walk.
1789 * NOTE! The caller *has* to check the resulting dentry against the sequence
1790 * number we've returned before using any of the resulting dentry state!
1792 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1793 const struct qstr *name,
1794 unsigned *seqp, struct inode *inode)
1796 u64 hashlen = name->hash_len;
1797 const unsigned char *str = name->name;
1798 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1799 struct hlist_bl_node *node;
1800 struct dentry *dentry;
1803 * Note: There is significant duplication with __d_lookup_rcu which is
1804 * required to prevent single threaded performance regressions
1805 * especially on architectures where smp_rmb (in seqcounts) are costly.
1806 * Keep the two functions in sync.
1810 * The hash list is protected using RCU.
1812 * Carefully use d_seq when comparing a candidate dentry, to avoid
1813 * races with d_move().
1815 * It is possible that concurrent renames can mess up our list
1816 * walk here and result in missing our dentry, resulting in the
1817 * false-negative result. d_lookup() protects against concurrent
1818 * renames using rename_lock seqlock.
1820 * See Documentation/filesystems/path-lookup.txt for more details.
1822 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1827 * The dentry sequence count protects us from concurrent
1828 * renames, and thus protects inode, parent and name fields.
1830 * The caller must perform a seqcount check in order
1831 * to do anything useful with the returned dentry,
1832 * including using the 'd_inode' pointer.
1834 * NOTE! We do a "raw" seqcount_begin here. That means that
1835 * we don't wait for the sequence count to stabilize if it
1836 * is in the middle of a sequence change. If we do the slow
1837 * dentry compare, we will do seqretries until it is stable,
1838 * and if we end up with a successful lookup, we actually
1839 * want to exit RCU lookup anyway.
1841 seq = raw_seqcount_begin(&dentry->d_seq);
1842 if (dentry->d_parent != parent)
1844 if (d_unhashed(dentry))
1848 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1849 if (dentry->d_name.hash != hashlen_hash(hashlen))
1851 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1854 case D_COMP_NOMATCH:
1861 if (dentry->d_name.hash_len != hashlen)
1863 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
1870 * d_lookup - search for a dentry
1871 * @parent: parent dentry
1872 * @name: qstr of name we wish to find
1873 * Returns: dentry, or NULL
1875 * d_lookup searches the children of the parent dentry for the name in
1876 * question. If the dentry is found its reference count is incremented and the
1877 * dentry is returned. The caller must use dput to free the entry when it has
1878 * finished using it. %NULL is returned if the dentry does not exist.
1880 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
1882 struct dentry *dentry;
1886 seq = read_seqbegin(&rename_lock);
1887 dentry = __d_lookup(parent, name);
1890 } while (read_seqretry(&rename_lock, seq));
1893 EXPORT_SYMBOL(d_lookup);
1896 * __d_lookup - search for a dentry (racy)
1897 * @parent: parent dentry
1898 * @name: qstr of name we wish to find
1899 * Returns: dentry, or NULL
1901 * __d_lookup is like d_lookup, however it may (rarely) return a
1902 * false-negative result due to unrelated rename activity.
1904 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1905 * however it must be used carefully, eg. with a following d_lookup in
1906 * the case of failure.
1908 * __d_lookup callers must be commented.
1910 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
1912 unsigned int len = name->len;
1913 unsigned int hash = name->hash;
1914 const unsigned char *str = name->name;
1915 struct hlist_bl_head *b = d_hash(parent, hash);
1916 struct hlist_bl_node *node;
1917 struct dentry *found = NULL;
1918 struct dentry *dentry;
1921 * Note: There is significant duplication with __d_lookup_rcu which is
1922 * required to prevent single threaded performance regressions
1923 * especially on architectures where smp_rmb (in seqcounts) are costly.
1924 * Keep the two functions in sync.
1928 * The hash list is protected using RCU.
1930 * Take d_lock when comparing a candidate dentry, to avoid races
1933 * It is possible that concurrent renames can mess up our list
1934 * walk here and result in missing our dentry, resulting in the
1935 * false-negative result. d_lookup() protects against concurrent
1936 * renames using rename_lock seqlock.
1938 * See Documentation/filesystems/path-lookup.txt for more details.
1942 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1944 if (dentry->d_name.hash != hash)
1947 spin_lock(&dentry->d_lock);
1948 if (dentry->d_parent != parent)
1950 if (d_unhashed(dentry))
1954 * It is safe to compare names since d_move() cannot
1955 * change the qstr (protected by d_lock).
1957 if (parent->d_flags & DCACHE_OP_COMPARE) {
1958 int tlen = dentry->d_name.len;
1959 const char *tname = dentry->d_name.name;
1960 if (parent->d_op->d_compare(parent, parent->d_inode,
1961 dentry, dentry->d_inode,
1965 if (dentry->d_name.len != len)
1967 if (dentry_cmp(dentry, str, len))
1973 spin_unlock(&dentry->d_lock);
1976 spin_unlock(&dentry->d_lock);
1984 * d_hash_and_lookup - hash the qstr then search for a dentry
1985 * @dir: Directory to search in
1986 * @name: qstr of name we wish to find
1988 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
1990 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1993 * Check for a fs-specific hash function. Note that we must
1994 * calculate the standard hash first, as the d_op->d_hash()
1995 * routine may choose to leave the hash value unchanged.
1997 name->hash = full_name_hash(name->name, name->len);
1998 if (dir->d_flags & DCACHE_OP_HASH) {
1999 int err = dir->d_op->d_hash(dir, dir->d_inode, name);
2000 if (unlikely(err < 0))
2001 return ERR_PTR(err);
2003 return d_lookup(dir, name);
2005 EXPORT_SYMBOL(d_hash_and_lookup);
2008 * d_validate - verify dentry provided from insecure source (deprecated)
2009 * @dentry: The dentry alleged to be valid child of @dparent
2010 * @dparent: The parent dentry (known to be valid)
2012 * An insecure source has sent us a dentry, here we verify it and dget() it.
2013 * This is used by ncpfs in its readdir implementation.
2014 * Zero is returned in the dentry is invalid.
2016 * This function is slow for big directories, and deprecated, do not use it.
2018 int d_validate(struct dentry *dentry, struct dentry *dparent)
2020 struct dentry *child;
2022 spin_lock(&dparent->d_lock);
2023 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2024 if (dentry == child) {
2025 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2026 __dget_dlock(dentry);
2027 spin_unlock(&dentry->d_lock);
2028 spin_unlock(&dparent->d_lock);
2032 spin_unlock(&dparent->d_lock);
2036 EXPORT_SYMBOL(d_validate);
2039 * When a file is deleted, we have two options:
2040 * - turn this dentry into a negative dentry
2041 * - unhash this dentry and free it.
2043 * Usually, we want to just turn this into
2044 * a negative dentry, but if anybody else is
2045 * currently using the dentry or the inode
2046 * we can't do that and we fall back on removing
2047 * it from the hash queues and waiting for
2048 * it to be deleted later when it has no users
2052 * d_delete - delete a dentry
2053 * @dentry: The dentry to delete
2055 * Turn the dentry into a negative dentry if possible, otherwise
2056 * remove it from the hash queues so it can be deleted later
2059 void d_delete(struct dentry * dentry)
2061 struct inode *inode;
2064 * Are we the only user?
2067 spin_lock(&dentry->d_lock);
2068 inode = dentry->d_inode;
2069 isdir = S_ISDIR(inode->i_mode);
2070 if (dentry->d_count == 1) {
2071 if (!spin_trylock(&inode->i_lock)) {
2072 spin_unlock(&dentry->d_lock);
2076 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2077 dentry_unlink_inode(dentry);
2078 fsnotify_nameremove(dentry, isdir);
2082 if (!d_unhashed(dentry))
2085 spin_unlock(&dentry->d_lock);
2087 fsnotify_nameremove(dentry, isdir);
2089 EXPORT_SYMBOL(d_delete);
2091 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2093 BUG_ON(!d_unhashed(entry));
2095 entry->d_flags |= DCACHE_RCUACCESS;
2096 hlist_bl_add_head_rcu(&entry->d_hash, b);
2100 static void _d_rehash(struct dentry * entry)
2102 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2106 * d_rehash - add an entry back to the hash
2107 * @entry: dentry to add to the hash
2109 * Adds a dentry to the hash according to its name.
2112 void d_rehash(struct dentry * entry)
2114 spin_lock(&entry->d_lock);
2116 spin_unlock(&entry->d_lock);
2118 EXPORT_SYMBOL(d_rehash);
2121 * dentry_update_name_case - update case insensitive dentry with a new name
2122 * @dentry: dentry to be updated
2125 * Update a case insensitive dentry with new case of name.
2127 * dentry must have been returned by d_lookup with name @name. Old and new
2128 * name lengths must match (ie. no d_compare which allows mismatched name
2131 * Parent inode i_mutex must be held over d_lookup and into this call (to
2132 * keep renames and concurrent inserts, and readdir(2) away).
2134 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2136 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2137 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2139 spin_lock(&dentry->d_lock);
2140 write_seqcount_begin(&dentry->d_seq);
2141 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2142 write_seqcount_end(&dentry->d_seq);
2143 spin_unlock(&dentry->d_lock);
2145 EXPORT_SYMBOL(dentry_update_name_case);
2147 static void switch_names(struct dentry *dentry, struct dentry *target)
2149 if (dname_external(target)) {
2150 if (dname_external(dentry)) {
2152 * Both external: swap the pointers
2154 swap(target->d_name.name, dentry->d_name.name);
2157 * dentry:internal, target:external. Steal target's
2158 * storage and make target internal.
2160 memcpy(target->d_iname, dentry->d_name.name,
2161 dentry->d_name.len + 1);
2162 dentry->d_name.name = target->d_name.name;
2163 target->d_name.name = target->d_iname;
2166 if (dname_external(dentry)) {
2168 * dentry:external, target:internal. Give dentry's
2169 * storage to target and make dentry internal
2171 memcpy(dentry->d_iname, target->d_name.name,
2172 target->d_name.len + 1);
2173 target->d_name.name = dentry->d_name.name;
2174 dentry->d_name.name = dentry->d_iname;
2177 * Both are internal. Just copy target to dentry
2179 memcpy(dentry->d_iname, target->d_name.name,
2180 target->d_name.len + 1);
2181 dentry->d_name.len = target->d_name.len;
2185 swap(dentry->d_name.len, target->d_name.len);
2188 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2191 * XXXX: do we really need to take target->d_lock?
2193 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2194 spin_lock(&target->d_parent->d_lock);
2196 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2197 spin_lock(&dentry->d_parent->d_lock);
2198 spin_lock_nested(&target->d_parent->d_lock,
2199 DENTRY_D_LOCK_NESTED);
2201 spin_lock(&target->d_parent->d_lock);
2202 spin_lock_nested(&dentry->d_parent->d_lock,
2203 DENTRY_D_LOCK_NESTED);
2206 if (target < dentry) {
2207 spin_lock_nested(&target->d_lock, 2);
2208 spin_lock_nested(&dentry->d_lock, 3);
2210 spin_lock_nested(&dentry->d_lock, 2);
2211 spin_lock_nested(&target->d_lock, 3);
2215 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2216 struct dentry *target)
2218 if (target->d_parent != dentry->d_parent)
2219 spin_unlock(&dentry->d_parent->d_lock);
2220 if (target->d_parent != target)
2221 spin_unlock(&target->d_parent->d_lock);
2225 * When switching names, the actual string doesn't strictly have to
2226 * be preserved in the target - because we're dropping the target
2227 * anyway. As such, we can just do a simple memcpy() to copy over
2228 * the new name before we switch.
2230 * Note that we have to be a lot more careful about getting the hash
2231 * switched - we have to switch the hash value properly even if it
2232 * then no longer matches the actual (corrupted) string of the target.
2233 * The hash value has to match the hash queue that the dentry is on..
2236 * __d_move - move a dentry
2237 * @dentry: entry to move
2238 * @target: new dentry
2240 * Update the dcache to reflect the move of a file name. Negative
2241 * dcache entries should not be moved in this way. Caller must hold
2242 * rename_lock, the i_mutex of the source and target directories,
2243 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2245 static void __d_move(struct dentry * dentry, struct dentry * target)
2247 if (!dentry->d_inode)
2248 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2250 BUG_ON(d_ancestor(dentry, target));
2251 BUG_ON(d_ancestor(target, dentry));
2253 dentry_lock_for_move(dentry, target);
2255 write_seqcount_begin(&dentry->d_seq);
2256 write_seqcount_begin(&target->d_seq);
2258 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2261 * Move the dentry to the target hash queue. Don't bother checking
2262 * for the same hash queue because of how unlikely it is.
2265 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2267 /* Unhash the target: dput() will then get rid of it */
2270 list_del(&dentry->d_u.d_child);
2271 list_del(&target->d_u.d_child);
2273 /* Switch the names.. */
2274 switch_names(dentry, target);
2275 swap(dentry->d_name.hash, target->d_name.hash);
2277 /* ... and switch the parents */
2278 if (IS_ROOT(dentry)) {
2279 dentry->d_parent = target->d_parent;
2280 target->d_parent = target;
2281 INIT_LIST_HEAD(&target->d_u.d_child);
2283 swap(dentry->d_parent, target->d_parent);
2285 /* And add them back to the (new) parent lists */
2286 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2289 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2291 write_seqcount_end(&target->d_seq);
2292 write_seqcount_end(&dentry->d_seq);
2294 dentry_unlock_parents_for_move(dentry, target);
2295 spin_unlock(&target->d_lock);
2296 fsnotify_d_move(dentry);
2297 spin_unlock(&dentry->d_lock);
2301 * d_move - move a dentry
2302 * @dentry: entry to move
2303 * @target: new dentry
2305 * Update the dcache to reflect the move of a file name. Negative
2306 * dcache entries should not be moved in this way. See the locking
2307 * requirements for __d_move.
2309 void d_move(struct dentry *dentry, struct dentry *target)
2311 write_seqlock(&rename_lock);
2312 __d_move(dentry, target);
2313 write_sequnlock(&rename_lock);
2315 EXPORT_SYMBOL(d_move);
2318 * d_ancestor - search for an ancestor
2319 * @p1: ancestor dentry
2322 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2323 * an ancestor of p2, else NULL.
2325 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2329 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2330 if (p->d_parent == p1)
2337 * This helper attempts to cope with remotely renamed directories
2339 * It assumes that the caller is already holding
2340 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2342 * Note: If ever the locking in lock_rename() changes, then please
2343 * remember to update this too...
2345 static struct dentry *__d_unalias(struct inode *inode,
2346 struct dentry *dentry, struct dentry *alias)
2348 struct mutex *m1 = NULL, *m2 = NULL;
2349 struct dentry *ret = ERR_PTR(-EBUSY);
2351 /* If alias and dentry share a parent, then no extra locks required */
2352 if (alias->d_parent == dentry->d_parent)
2355 /* See lock_rename() */
2356 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2358 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2359 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2361 m2 = &alias->d_parent->d_inode->i_mutex;
2363 if (likely(!d_mountpoint(alias))) {
2364 __d_move(alias, dentry);
2368 spin_unlock(&inode->i_lock);
2377 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2378 * named dentry in place of the dentry to be replaced.
2379 * returns with anon->d_lock held!
2381 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2383 struct dentry *dparent;
2385 dentry_lock_for_move(anon, dentry);
2387 write_seqcount_begin(&dentry->d_seq);
2388 write_seqcount_begin(&anon->d_seq);
2390 dparent = dentry->d_parent;
2392 switch_names(dentry, anon);
2393 swap(dentry->d_name.hash, anon->d_name.hash);
2395 dentry->d_parent = dentry;
2396 list_del_init(&dentry->d_u.d_child);
2397 anon->d_parent = dparent;
2398 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2400 write_seqcount_end(&dentry->d_seq);
2401 write_seqcount_end(&anon->d_seq);
2403 dentry_unlock_parents_for_move(anon, dentry);
2404 spin_unlock(&dentry->d_lock);
2406 /* anon->d_lock still locked, returns locked */
2407 anon->d_flags &= ~DCACHE_DISCONNECTED;
2411 * d_materialise_unique - introduce an inode into the tree
2412 * @dentry: candidate dentry
2413 * @inode: inode to bind to the dentry, to which aliases may be attached
2415 * Introduces an dentry into the tree, substituting an extant disconnected
2416 * root directory alias in its place if there is one. Caller must hold the
2417 * i_mutex of the parent directory.
2419 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2421 struct dentry *actual;
2423 BUG_ON(!d_unhashed(dentry));
2427 __d_instantiate(dentry, NULL);
2432 spin_lock(&inode->i_lock);
2434 if (S_ISDIR(inode->i_mode)) {
2435 struct dentry *alias;
2437 /* Does an aliased dentry already exist? */
2438 alias = __d_find_alias(inode, 0);
2441 write_seqlock(&rename_lock);
2443 if (d_ancestor(alias, dentry)) {
2444 /* Check for loops */
2445 actual = ERR_PTR(-ELOOP);
2446 spin_unlock(&inode->i_lock);
2447 } else if (IS_ROOT(alias)) {
2448 /* Is this an anonymous mountpoint that we
2449 * could splice into our tree? */
2450 __d_materialise_dentry(dentry, alias);
2451 write_sequnlock(&rename_lock);
2455 /* Nope, but we must(!) avoid directory
2456 * aliasing. This drops inode->i_lock */
2457 actual = __d_unalias(inode, dentry, alias);
2459 write_sequnlock(&rename_lock);
2460 if (IS_ERR(actual)) {
2461 if (PTR_ERR(actual) == -ELOOP)
2462 pr_warn_ratelimited(
2463 "VFS: Lookup of '%s' in %s %s"
2464 " would have caused loop\n",
2465 dentry->d_name.name,
2466 inode->i_sb->s_type->name,
2474 /* Add a unique reference */
2475 actual = __d_instantiate_unique(dentry, inode);
2479 BUG_ON(!d_unhashed(actual));
2481 spin_lock(&actual->d_lock);
2484 spin_unlock(&actual->d_lock);
2485 spin_unlock(&inode->i_lock);
2487 if (actual == dentry) {
2488 security_d_instantiate(dentry, inode);
2495 EXPORT_SYMBOL_GPL(d_materialise_unique);
2497 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2501 return -ENAMETOOLONG;
2503 memcpy(*buffer, str, namelen);
2507 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2509 return prepend(buffer, buflen, name->name, name->len);
2513 * prepend_path - Prepend path string to a buffer
2514 * @path: the dentry/vfsmount to report
2515 * @root: root vfsmnt/dentry
2516 * @buffer: pointer to the end of the buffer
2517 * @buflen: pointer to buffer length
2519 * Caller holds the rename_lock.
2521 static int prepend_path(const struct path *path,
2522 const struct path *root,
2523 char **buffer, int *buflen)
2525 struct dentry *dentry = path->dentry;
2526 struct vfsmount *vfsmnt = path->mnt;
2527 struct mount *mnt = real_mount(vfsmnt);
2531 while (dentry != root->dentry || vfsmnt != root->mnt) {
2532 struct dentry * parent;
2534 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2536 if (!mnt_has_parent(mnt))
2538 dentry = mnt->mnt_mountpoint;
2539 mnt = mnt->mnt_parent;
2543 parent = dentry->d_parent;
2545 spin_lock(&dentry->d_lock);
2546 error = prepend_name(buffer, buflen, &dentry->d_name);
2547 spin_unlock(&dentry->d_lock);
2549 error = prepend(buffer, buflen, "/", 1);
2557 if (!error && !slash)
2558 error = prepend(buffer, buflen, "/", 1);
2564 * Filesystems needing to implement special "root names"
2565 * should do so with ->d_dname()
2567 if (IS_ROOT(dentry) &&
2568 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2569 WARN(1, "Root dentry has weird name <%.*s>\n",
2570 (int) dentry->d_name.len, dentry->d_name.name);
2573 error = prepend(buffer, buflen, "/", 1);
2575 error = is_mounted(vfsmnt) ? 1 : 2;
2580 * __d_path - return the path of a dentry
2581 * @path: the dentry/vfsmount to report
2582 * @root: root vfsmnt/dentry
2583 * @buf: buffer to return value in
2584 * @buflen: buffer length
2586 * Convert a dentry into an ASCII path name.
2588 * Returns a pointer into the buffer or an error code if the
2589 * path was too long.
2591 * "buflen" should be positive.
2593 * If the path is not reachable from the supplied root, return %NULL.
2595 char *__d_path(const struct path *path,
2596 const struct path *root,
2597 char *buf, int buflen)
2599 char *res = buf + buflen;
2602 prepend(&res, &buflen, "\0", 1);
2603 br_read_lock(&vfsmount_lock);
2604 write_seqlock(&rename_lock);
2605 error = prepend_path(path, root, &res, &buflen);
2606 write_sequnlock(&rename_lock);
2607 br_read_unlock(&vfsmount_lock);
2610 return ERR_PTR(error);
2616 char *d_absolute_path(const struct path *path,
2617 char *buf, int buflen)
2619 struct path root = {};
2620 char *res = buf + buflen;
2623 prepend(&res, &buflen, "\0", 1);
2624 br_read_lock(&vfsmount_lock);
2625 write_seqlock(&rename_lock);
2626 error = prepend_path(path, &root, &res, &buflen);
2627 write_sequnlock(&rename_lock);
2628 br_read_unlock(&vfsmount_lock);
2633 return ERR_PTR(error);
2638 * same as __d_path but appends "(deleted)" for unlinked files.
2640 static int path_with_deleted(const struct path *path,
2641 const struct path *root,
2642 char **buf, int *buflen)
2644 prepend(buf, buflen, "\0", 1);
2645 if (d_unlinked(path->dentry)) {
2646 int error = prepend(buf, buflen, " (deleted)", 10);
2651 return prepend_path(path, root, buf, buflen);
2654 static int prepend_unreachable(char **buffer, int *buflen)
2656 return prepend(buffer, buflen, "(unreachable)", 13);
2660 * d_path - return the path of a dentry
2661 * @path: path to report
2662 * @buf: buffer to return value in
2663 * @buflen: buffer length
2665 * Convert a dentry into an ASCII path name. If the entry has been deleted
2666 * the string " (deleted)" is appended. Note that this is ambiguous.
2668 * Returns a pointer into the buffer or an error code if the path was
2669 * too long. Note: Callers should use the returned pointer, not the passed
2670 * in buffer, to use the name! The implementation often starts at an offset
2671 * into the buffer, and may leave 0 bytes at the start.
2673 * "buflen" should be positive.
2675 char *d_path(const struct path *path, char *buf, int buflen)
2677 char *res = buf + buflen;
2682 * We have various synthetic filesystems that never get mounted. On
2683 * these filesystems dentries are never used for lookup purposes, and
2684 * thus don't need to be hashed. They also don't need a name until a
2685 * user wants to identify the object in /proc/pid/fd/. The little hack
2686 * below allows us to generate a name for these objects on demand:
2688 * Some pseudo inodes are mountable. When they are mounted
2689 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
2690 * and instead have d_path return the mounted path.
2692 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
2693 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
2694 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2696 get_fs_root(current->fs, &root);
2697 br_read_lock(&vfsmount_lock);
2698 write_seqlock(&rename_lock);
2699 error = path_with_deleted(path, &root, &res, &buflen);
2700 write_sequnlock(&rename_lock);
2701 br_read_unlock(&vfsmount_lock);
2703 res = ERR_PTR(error);
2707 EXPORT_SYMBOL(d_path);
2710 * Helper function for dentry_operations.d_dname() members
2712 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2713 const char *fmt, ...)
2719 va_start(args, fmt);
2720 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2723 if (sz > sizeof(temp) || sz > buflen)
2724 return ERR_PTR(-ENAMETOOLONG);
2726 buffer += buflen - sz;
2727 return memcpy(buffer, temp, sz);
2730 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2732 char *end = buffer + buflen;
2733 /* these dentries are never renamed, so d_lock is not needed */
2734 if (prepend(&end, &buflen, " (deleted)", 11) ||
2735 prepend_name(&end, &buflen, &dentry->d_name) ||
2736 prepend(&end, &buflen, "/", 1))
2737 end = ERR_PTR(-ENAMETOOLONG);
2742 * Write full pathname from the root of the filesystem into the buffer.
2744 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2746 char *end = buf + buflen;
2749 prepend(&end, &buflen, "\0", 1);
2756 while (!IS_ROOT(dentry)) {
2757 struct dentry *parent = dentry->d_parent;
2761 spin_lock(&dentry->d_lock);
2762 error = prepend_name(&end, &buflen, &dentry->d_name);
2763 spin_unlock(&dentry->d_lock);
2764 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2772 return ERR_PTR(-ENAMETOOLONG);
2775 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2779 write_seqlock(&rename_lock);
2780 retval = __dentry_path(dentry, buf, buflen);
2781 write_sequnlock(&rename_lock);
2785 EXPORT_SYMBOL(dentry_path_raw);
2787 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2792 write_seqlock(&rename_lock);
2793 if (d_unlinked(dentry)) {
2795 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2799 retval = __dentry_path(dentry, buf, buflen);
2800 write_sequnlock(&rename_lock);
2801 if (!IS_ERR(retval) && p)
2802 *p = '/'; /* restore '/' overriden with '\0' */
2805 return ERR_PTR(-ENAMETOOLONG);
2809 * NOTE! The user-level library version returns a
2810 * character pointer. The kernel system call just
2811 * returns the length of the buffer filled (which
2812 * includes the ending '\0' character), or a negative
2813 * error value. So libc would do something like
2815 * char *getcwd(char * buf, size_t size)
2819 * retval = sys_getcwd(buf, size);
2826 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2829 struct path pwd, root;
2830 char *page = (char *) __get_free_page(GFP_USER);
2835 get_fs_root_and_pwd(current->fs, &root, &pwd);
2838 br_read_lock(&vfsmount_lock);
2839 write_seqlock(&rename_lock);
2840 if (!d_unlinked(pwd.dentry)) {
2842 char *cwd = page + PAGE_SIZE;
2843 int buflen = PAGE_SIZE;
2845 prepend(&cwd, &buflen, "\0", 1);
2846 error = prepend_path(&pwd, &root, &cwd, &buflen);
2847 write_sequnlock(&rename_lock);
2848 br_read_unlock(&vfsmount_lock);
2853 /* Unreachable from current root */
2855 error = prepend_unreachable(&cwd, &buflen);
2861 len = PAGE_SIZE + page - cwd;
2864 if (copy_to_user(buf, cwd, len))
2868 write_sequnlock(&rename_lock);
2869 br_read_unlock(&vfsmount_lock);
2875 free_page((unsigned long) page);
2880 * Test whether new_dentry is a subdirectory of old_dentry.
2882 * Trivially implemented using the dcache structure
2886 * is_subdir - is new dentry a subdirectory of old_dentry
2887 * @new_dentry: new dentry
2888 * @old_dentry: old dentry
2890 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2891 * Returns 0 otherwise.
2892 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2895 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2900 if (new_dentry == old_dentry)
2904 /* for restarting inner loop in case of seq retry */
2905 seq = read_seqbegin(&rename_lock);
2907 * Need rcu_readlock to protect against the d_parent trashing
2911 if (d_ancestor(old_dentry, new_dentry))
2916 } while (read_seqretry(&rename_lock, seq));
2921 void d_genocide(struct dentry *root)
2923 struct dentry *this_parent;
2924 struct list_head *next;
2928 seq = read_seqbegin(&rename_lock);
2931 spin_lock(&this_parent->d_lock);
2933 next = this_parent->d_subdirs.next;
2935 while (next != &this_parent->d_subdirs) {
2936 struct list_head *tmp = next;
2937 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2940 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2941 if (d_unhashed(dentry) || !dentry->d_inode) {
2942 spin_unlock(&dentry->d_lock);
2945 if (!list_empty(&dentry->d_subdirs)) {
2946 spin_unlock(&this_parent->d_lock);
2947 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2948 this_parent = dentry;
2949 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2952 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2953 dentry->d_flags |= DCACHE_GENOCIDE;
2956 spin_unlock(&dentry->d_lock);
2958 if (this_parent != root) {
2959 struct dentry *child = this_parent;
2960 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2961 this_parent->d_flags |= DCACHE_GENOCIDE;
2962 this_parent->d_count--;
2964 this_parent = try_to_ascend(this_parent, locked, seq);
2967 next = child->d_u.d_child.next;
2970 spin_unlock(&this_parent->d_lock);
2971 if (!locked && read_seqretry(&rename_lock, seq))
2974 write_sequnlock(&rename_lock);
2981 write_seqlock(&rename_lock);
2986 * find_inode_number - check for dentry with name
2987 * @dir: directory to check
2988 * @name: Name to find.
2990 * Check whether a dentry already exists for the given name,
2991 * and return the inode number if it has an inode. Otherwise
2994 * This routine is used to post-process directory listings for
2995 * filesystems using synthetic inode numbers, and is necessary
2996 * to keep getcwd() working.
2999 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3001 struct dentry * dentry;
3004 dentry = d_hash_and_lookup(dir, name);
3005 if (!IS_ERR_OR_NULL(dentry)) {
3006 if (dentry->d_inode)
3007 ino = dentry->d_inode->i_ino;
3012 EXPORT_SYMBOL(find_inode_number);
3014 static __initdata unsigned long dhash_entries;
3015 static int __init set_dhash_entries(char *str)
3019 dhash_entries = simple_strtoul(str, &str, 0);
3022 __setup("dhash_entries=", set_dhash_entries);
3024 static void __init dcache_init_early(void)
3028 /* If hashes are distributed across NUMA nodes, defer
3029 * hash allocation until vmalloc space is available.
3035 alloc_large_system_hash("Dentry cache",
3036 sizeof(struct hlist_bl_head),
3045 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3046 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3049 static void __init dcache_init(void)
3054 * A constructor could be added for stable state like the lists,
3055 * but it is probably not worth it because of the cache nature
3058 dentry_cache = KMEM_CACHE(dentry,
3059 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3061 /* Hash may have been set up in dcache_init_early */
3066 alloc_large_system_hash("Dentry cache",
3067 sizeof(struct hlist_bl_head),
3076 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3077 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3080 /* SLAB cache for __getname() consumers */
3081 struct kmem_cache *names_cachep __read_mostly;
3082 EXPORT_SYMBOL(names_cachep);
3084 EXPORT_SYMBOL(d_genocide);
3086 void __init vfs_caches_init_early(void)
3088 dcache_init_early();
3092 void __init vfs_caches_init(unsigned long mempages)
3094 unsigned long reserve;
3096 /* Base hash sizes on available memory, with a reserve equal to
3097 150% of current kernel size */
3099 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3100 mempages -= reserve;
3102 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3103 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3107 files_init(mempages);