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 return dentry_hashtable + hash_32(hash, d_hash_shift);
112 /* Statistics gathering. */
113 struct dentry_stat_t dentry_stat = {
117 static DEFINE_PER_CPU(unsigned int, nr_dentry);
119 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
120 static int get_nr_dentry(void)
124 for_each_possible_cpu(i)
125 sum += per_cpu(nr_dentry, i);
126 return sum < 0 ? 0 : sum;
129 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
130 size_t *lenp, loff_t *ppos)
132 dentry_stat.nr_dentry = get_nr_dentry();
133 return proc_dointvec(table, write, buffer, lenp, ppos);
138 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
139 * The strings are both count bytes long, and count is non-zero.
141 #ifdef CONFIG_DCACHE_WORD_ACCESS
143 #include <asm/word-at-a-time.h>
145 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
146 * aligned allocation for this particular component. We don't
147 * strictly need the load_unaligned_zeropad() safety, but it
148 * doesn't hurt either.
150 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
151 * need the careful unaligned handling.
153 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
155 unsigned long a,b,mask;
158 a = *(unsigned long *)cs;
159 b = load_unaligned_zeropad(ct);
160 if (tcount < sizeof(unsigned long))
162 if (unlikely(a != b))
164 cs += sizeof(unsigned long);
165 ct += sizeof(unsigned long);
166 tcount -= sizeof(unsigned long);
170 mask = ~(~0ul << tcount*8);
171 return unlikely(!!((a ^ b) & mask));
176 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
190 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
192 const unsigned char *cs;
194 * Be careful about RCU walk racing with rename:
195 * use ACCESS_ONCE to fetch the name pointer.
197 * NOTE! Even if a rename will mean that the length
198 * was not loaded atomically, we don't care. The
199 * RCU walk will check the sequence count eventually,
200 * and catch it. And we won't overrun the buffer,
201 * because we're reading the name pointer atomically,
202 * and a dentry name is guaranteed to be properly
203 * terminated with a NUL byte.
205 * End result: even if 'len' is wrong, we'll exit
206 * early because the data cannot match (there can
207 * be no NUL in the ct/tcount data)
209 cs = ACCESS_ONCE(dentry->d_name.name);
210 smp_read_barrier_depends();
211 return dentry_string_cmp(cs, ct, tcount);
214 static void __d_free(struct rcu_head *head)
216 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
218 WARN_ON(!hlist_unhashed(&dentry->d_alias));
219 if (dname_external(dentry))
220 kfree(dentry->d_name.name);
221 kmem_cache_free(dentry_cache, dentry);
227 static void d_free(struct dentry *dentry)
229 BUG_ON(dentry->d_count);
230 this_cpu_dec(nr_dentry);
231 if (dentry->d_op && dentry->d_op->d_release)
232 dentry->d_op->d_release(dentry);
234 /* if dentry was never visible to RCU, immediate free is OK */
235 if (!(dentry->d_flags & DCACHE_RCUACCESS))
236 __d_free(&dentry->d_u.d_rcu);
238 call_rcu(&dentry->d_u.d_rcu, __d_free);
242 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
243 * @dentry: the target dentry
244 * After this call, in-progress rcu-walk path lookup will fail. This
245 * should be called after unhashing, and after changing d_inode (if
246 * the dentry has not already been unhashed).
248 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
250 assert_spin_locked(&dentry->d_lock);
251 /* Go through a barrier */
252 write_seqcount_barrier(&dentry->d_seq);
256 * Release the dentry's inode, using the filesystem
257 * d_iput() operation if defined. Dentry has no refcount
260 static void dentry_iput(struct dentry * dentry)
261 __releases(dentry->d_lock)
262 __releases(dentry->d_inode->i_lock)
264 struct inode *inode = dentry->d_inode;
266 dentry->d_inode = NULL;
267 hlist_del_init(&dentry->d_alias);
268 spin_unlock(&dentry->d_lock);
269 spin_unlock(&inode->i_lock);
271 fsnotify_inoderemove(inode);
272 if (dentry->d_op && dentry->d_op->d_iput)
273 dentry->d_op->d_iput(dentry, inode);
277 spin_unlock(&dentry->d_lock);
282 * Release the dentry's inode, using the filesystem
283 * d_iput() operation if defined. dentry remains in-use.
285 static void dentry_unlink_inode(struct dentry * dentry)
286 __releases(dentry->d_lock)
287 __releases(dentry->d_inode->i_lock)
289 struct inode *inode = dentry->d_inode;
290 dentry->d_inode = NULL;
291 hlist_del_init(&dentry->d_alias);
292 dentry_rcuwalk_barrier(dentry);
293 spin_unlock(&dentry->d_lock);
294 spin_unlock(&inode->i_lock);
296 fsnotify_inoderemove(inode);
297 if (dentry->d_op && dentry->d_op->d_iput)
298 dentry->d_op->d_iput(dentry, inode);
304 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
306 static void dentry_lru_add(struct dentry *dentry)
308 if (list_empty(&dentry->d_lru)) {
309 spin_lock(&dcache_lru_lock);
310 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
311 dentry->d_sb->s_nr_dentry_unused++;
312 dentry_stat.nr_unused++;
313 spin_unlock(&dcache_lru_lock);
317 static void __dentry_lru_del(struct dentry *dentry)
319 list_del_init(&dentry->d_lru);
320 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
321 dentry->d_sb->s_nr_dentry_unused--;
322 dentry_stat.nr_unused--;
326 * Remove a dentry with references from the LRU.
328 static void dentry_lru_del(struct dentry *dentry)
330 if (!list_empty(&dentry->d_lru)) {
331 spin_lock(&dcache_lru_lock);
332 __dentry_lru_del(dentry);
333 spin_unlock(&dcache_lru_lock);
337 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
339 spin_lock(&dcache_lru_lock);
340 if (list_empty(&dentry->d_lru)) {
341 list_add_tail(&dentry->d_lru, list);
342 dentry->d_sb->s_nr_dentry_unused++;
343 dentry_stat.nr_unused++;
345 list_move_tail(&dentry->d_lru, list);
347 spin_unlock(&dcache_lru_lock);
351 * d_kill - kill dentry and return parent
352 * @dentry: dentry to kill
353 * @parent: parent dentry
355 * The dentry must already be unhashed and removed from the LRU.
357 * If this is the root of the dentry tree, return NULL.
359 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
362 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
363 __releases(dentry->d_lock)
364 __releases(parent->d_lock)
365 __releases(dentry->d_inode->i_lock)
367 list_del(&dentry->d_u.d_child);
369 * Inform try_to_ascend() that we are no longer attached to the
372 dentry->d_flags |= DCACHE_DENTRY_KILLED;
374 spin_unlock(&parent->d_lock);
377 * dentry_iput drops the locks, at which point nobody (except
378 * transient RCU lookups) can reach this dentry.
385 * Unhash a dentry without inserting an RCU walk barrier or checking that
386 * dentry->d_lock is locked. The caller must take care of that, if
389 static void __d_shrink(struct dentry *dentry)
391 if (!d_unhashed(dentry)) {
392 struct hlist_bl_head *b;
393 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
394 b = &dentry->d_sb->s_anon;
396 b = d_hash(dentry->d_parent, dentry->d_name.hash);
399 __hlist_bl_del(&dentry->d_hash);
400 dentry->d_hash.pprev = NULL;
406 * d_drop - drop a dentry
407 * @dentry: dentry to drop
409 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
410 * be found through a VFS lookup any more. Note that this is different from
411 * deleting the dentry - d_delete will try to mark the dentry negative if
412 * possible, giving a successful _negative_ lookup, while d_drop will
413 * just make the cache lookup fail.
415 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
416 * reason (NFS timeouts or autofs deletes).
418 * __d_drop requires dentry->d_lock.
420 void __d_drop(struct dentry *dentry)
422 if (!d_unhashed(dentry)) {
424 dentry_rcuwalk_barrier(dentry);
427 EXPORT_SYMBOL(__d_drop);
429 void d_drop(struct dentry *dentry)
431 spin_lock(&dentry->d_lock);
433 spin_unlock(&dentry->d_lock);
435 EXPORT_SYMBOL(d_drop);
438 * Finish off a dentry we've decided to kill.
439 * dentry->d_lock must be held, returns with it unlocked.
440 * If ref is non-zero, then decrement the refcount too.
441 * Returns dentry requiring refcount drop, or NULL if we're done.
443 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
444 __releases(dentry->d_lock)
447 struct dentry *parent;
449 inode = dentry->d_inode;
450 if (inode && !spin_trylock(&inode->i_lock)) {
452 spin_unlock(&dentry->d_lock);
454 return dentry; /* try again with same dentry */
459 parent = dentry->d_parent;
460 if (parent && !spin_trylock(&parent->d_lock)) {
462 spin_unlock(&inode->i_lock);
469 * inform the fs via d_prune that this dentry is about to be
470 * unhashed and destroyed.
472 if (dentry->d_flags & DCACHE_OP_PRUNE)
473 dentry->d_op->d_prune(dentry);
475 dentry_lru_del(dentry);
476 /* if it was on the hash then remove it */
478 return d_kill(dentry, parent);
484 * This is complicated by the fact that we do not want to put
485 * dentries that are no longer on any hash chain on the unused
486 * list: we'd much rather just get rid of them immediately.
488 * However, that implies that we have to traverse the dentry
489 * tree upwards to the parents which might _also_ now be
490 * scheduled for deletion (it may have been only waiting for
491 * its last child to go away).
493 * This tail recursion is done by hand as we don't want to depend
494 * on the compiler to always get this right (gcc generally doesn't).
495 * Real recursion would eat up our stack space.
499 * dput - release a dentry
500 * @dentry: dentry to release
502 * Release a dentry. This will drop the usage count and if appropriate
503 * call the dentry unlink method as well as removing it from the queues and
504 * releasing its resources. If the parent dentries were scheduled for release
505 * they too may now get deleted.
507 void dput(struct dentry *dentry)
513 if (dentry->d_count == 1)
515 spin_lock(&dentry->d_lock);
516 BUG_ON(!dentry->d_count);
517 if (dentry->d_count > 1) {
519 spin_unlock(&dentry->d_lock);
523 if (dentry->d_flags & DCACHE_OP_DELETE) {
524 if (dentry->d_op->d_delete(dentry))
528 /* Unreachable? Get rid of it */
529 if (d_unhashed(dentry))
532 dentry->d_flags |= DCACHE_REFERENCED;
533 dentry_lru_add(dentry);
536 spin_unlock(&dentry->d_lock);
540 dentry = dentry_kill(dentry, 1);
547 * d_invalidate - invalidate a dentry
548 * @dentry: dentry to invalidate
550 * Try to invalidate the dentry if it turns out to be
551 * possible. If there are other dentries that can be
552 * reached through this one we can't delete it and we
553 * return -EBUSY. On success we return 0.
558 int d_invalidate(struct dentry * dentry)
561 * If it's already been dropped, return OK.
563 spin_lock(&dentry->d_lock);
564 if (d_unhashed(dentry)) {
565 spin_unlock(&dentry->d_lock);
569 * Check whether to do a partial shrink_dcache
570 * to get rid of unused child entries.
572 if (!list_empty(&dentry->d_subdirs)) {
573 spin_unlock(&dentry->d_lock);
574 shrink_dcache_parent(dentry);
575 spin_lock(&dentry->d_lock);
579 * Somebody else still using it?
581 * If it's a directory, we can't drop it
582 * for fear of somebody re-populating it
583 * with children (even though dropping it
584 * would make it unreachable from the root,
585 * we might still populate it if it was a
586 * working directory or similar).
587 * We also need to leave mountpoints alone,
590 if (dentry->d_count > 1 && dentry->d_inode) {
591 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
592 spin_unlock(&dentry->d_lock);
598 spin_unlock(&dentry->d_lock);
601 EXPORT_SYMBOL(d_invalidate);
603 /* This must be called with d_lock held */
604 static inline void __dget_dlock(struct dentry *dentry)
609 static inline void __dget(struct dentry *dentry)
611 spin_lock(&dentry->d_lock);
612 __dget_dlock(dentry);
613 spin_unlock(&dentry->d_lock);
616 struct dentry *dget_parent(struct dentry *dentry)
622 * Don't need rcu_dereference because we re-check it was correct under
626 ret = dentry->d_parent;
627 spin_lock(&ret->d_lock);
628 if (unlikely(ret != dentry->d_parent)) {
629 spin_unlock(&ret->d_lock);
634 BUG_ON(!ret->d_count);
636 spin_unlock(&ret->d_lock);
639 EXPORT_SYMBOL(dget_parent);
642 * d_find_alias - grab a hashed alias of inode
643 * @inode: inode in question
644 * @want_discon: flag, used by d_splice_alias, to request
645 * that only a DISCONNECTED alias be returned.
647 * If inode has a hashed alias, or is a directory and has any alias,
648 * acquire the reference to alias and return it. Otherwise return NULL.
649 * Notice that if inode is a directory there can be only one alias and
650 * it can be unhashed only if it has no children, or if it is the root
653 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
654 * any other hashed alias over that one unless @want_discon is set,
655 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
657 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
659 struct dentry *alias, *discon_alias;
663 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
664 spin_lock(&alias->d_lock);
665 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
666 if (IS_ROOT(alias) &&
667 (alias->d_flags & DCACHE_DISCONNECTED)) {
668 discon_alias = alias;
669 } else if (!want_discon) {
671 spin_unlock(&alias->d_lock);
675 spin_unlock(&alias->d_lock);
678 alias = discon_alias;
679 spin_lock(&alias->d_lock);
680 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
681 if (IS_ROOT(alias) &&
682 (alias->d_flags & DCACHE_DISCONNECTED)) {
684 spin_unlock(&alias->d_lock);
688 spin_unlock(&alias->d_lock);
694 struct dentry *d_find_alias(struct inode *inode)
696 struct dentry *de = NULL;
698 if (!hlist_empty(&inode->i_dentry)) {
699 spin_lock(&inode->i_lock);
700 de = __d_find_alias(inode, 0);
701 spin_unlock(&inode->i_lock);
705 EXPORT_SYMBOL(d_find_alias);
708 * Try to kill dentries associated with this inode.
709 * WARNING: you must own a reference to inode.
711 void d_prune_aliases(struct inode *inode)
713 struct dentry *dentry;
715 spin_lock(&inode->i_lock);
716 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
717 spin_lock(&dentry->d_lock);
718 if (!dentry->d_count) {
719 __dget_dlock(dentry);
721 spin_unlock(&dentry->d_lock);
722 spin_unlock(&inode->i_lock);
726 spin_unlock(&dentry->d_lock);
728 spin_unlock(&inode->i_lock);
730 EXPORT_SYMBOL(d_prune_aliases);
733 * Try to throw away a dentry - free the inode, dput the parent.
734 * Requires dentry->d_lock is held, and dentry->d_count == 0.
735 * Releases dentry->d_lock.
737 * This may fail if locks cannot be acquired no problem, just try again.
739 static void try_prune_one_dentry(struct dentry *dentry)
740 __releases(dentry->d_lock)
742 struct dentry *parent;
744 parent = dentry_kill(dentry, 0);
746 * If dentry_kill returns NULL, we have nothing more to do.
747 * if it returns the same dentry, trylocks failed. In either
748 * case, just loop again.
750 * Otherwise, we need to prune ancestors too. This is necessary
751 * to prevent quadratic behavior of shrink_dcache_parent(), but
752 * is also expected to be beneficial in reducing dentry cache
757 if (parent == dentry)
760 /* Prune ancestors. */
763 spin_lock(&dentry->d_lock);
764 if (dentry->d_count > 1) {
766 spin_unlock(&dentry->d_lock);
769 dentry = dentry_kill(dentry, 1);
773 static void shrink_dentry_list(struct list_head *list)
775 struct dentry *dentry;
779 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
780 if (&dentry->d_lru == list)
782 spin_lock(&dentry->d_lock);
783 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
784 spin_unlock(&dentry->d_lock);
789 * We found an inuse dentry which was not removed from
790 * the LRU because of laziness during lookup. Do not free
791 * it - just keep it off the LRU list.
793 if (dentry->d_count) {
794 dentry_lru_del(dentry);
795 spin_unlock(&dentry->d_lock);
801 try_prune_one_dentry(dentry);
809 * prune_dcache_sb - shrink the dcache
811 * @count: number of entries to try to free
813 * Attempt to shrink the superblock dcache LRU by @count entries. This is
814 * done when we need more memory an called from the superblock shrinker
817 * This function may fail to free any resources if all the dentries are in
820 void prune_dcache_sb(struct super_block *sb, int count)
822 struct dentry *dentry;
823 LIST_HEAD(referenced);
827 spin_lock(&dcache_lru_lock);
828 while (!list_empty(&sb->s_dentry_lru)) {
829 dentry = list_entry(sb->s_dentry_lru.prev,
830 struct dentry, d_lru);
831 BUG_ON(dentry->d_sb != sb);
833 if (!spin_trylock(&dentry->d_lock)) {
834 spin_unlock(&dcache_lru_lock);
839 if (dentry->d_flags & DCACHE_REFERENCED) {
840 dentry->d_flags &= ~DCACHE_REFERENCED;
841 list_move(&dentry->d_lru, &referenced);
842 spin_unlock(&dentry->d_lock);
844 list_move_tail(&dentry->d_lru, &tmp);
845 dentry->d_flags |= DCACHE_SHRINK_LIST;
846 spin_unlock(&dentry->d_lock);
850 cond_resched_lock(&dcache_lru_lock);
852 if (!list_empty(&referenced))
853 list_splice(&referenced, &sb->s_dentry_lru);
854 spin_unlock(&dcache_lru_lock);
856 shrink_dentry_list(&tmp);
860 * shrink_dcache_sb - shrink dcache for a superblock
863 * Shrink the dcache for the specified super block. This is used to free
864 * the dcache before unmounting a file system.
866 void shrink_dcache_sb(struct super_block *sb)
870 spin_lock(&dcache_lru_lock);
871 while (!list_empty(&sb->s_dentry_lru)) {
872 list_splice_init(&sb->s_dentry_lru, &tmp);
873 spin_unlock(&dcache_lru_lock);
874 shrink_dentry_list(&tmp);
875 spin_lock(&dcache_lru_lock);
877 spin_unlock(&dcache_lru_lock);
879 EXPORT_SYMBOL(shrink_dcache_sb);
882 * destroy a single subtree of dentries for unmount
883 * - see the comments on shrink_dcache_for_umount() for a description of the
886 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
888 struct dentry *parent;
890 BUG_ON(!IS_ROOT(dentry));
893 /* descend to the first leaf in the current subtree */
894 while (!list_empty(&dentry->d_subdirs))
895 dentry = list_entry(dentry->d_subdirs.next,
896 struct dentry, d_u.d_child);
898 /* consume the dentries from this leaf up through its parents
899 * until we find one with children or run out altogether */
904 * inform the fs that this dentry is about to be
905 * unhashed and destroyed.
907 if (dentry->d_flags & DCACHE_OP_PRUNE)
908 dentry->d_op->d_prune(dentry);
910 dentry_lru_del(dentry);
913 if (dentry->d_count != 0) {
915 "BUG: Dentry %p{i=%lx,n=%s}"
917 " [unmount of %s %s]\n",
920 dentry->d_inode->i_ino : 0UL,
923 dentry->d_sb->s_type->name,
928 if (IS_ROOT(dentry)) {
930 list_del(&dentry->d_u.d_child);
932 parent = dentry->d_parent;
934 list_del(&dentry->d_u.d_child);
937 inode = dentry->d_inode;
939 dentry->d_inode = NULL;
940 hlist_del_init(&dentry->d_alias);
941 if (dentry->d_op && dentry->d_op->d_iput)
942 dentry->d_op->d_iput(dentry, inode);
949 /* finished when we fall off the top of the tree,
950 * otherwise we ascend to the parent and move to the
951 * next sibling if there is one */
955 } while (list_empty(&dentry->d_subdirs));
957 dentry = list_entry(dentry->d_subdirs.next,
958 struct dentry, d_u.d_child);
963 * destroy the dentries attached to a superblock on unmounting
964 * - we don't need to use dentry->d_lock because:
965 * - the superblock is detached from all mountings and open files, so the
966 * dentry trees will not be rearranged by the VFS
967 * - s_umount is write-locked, so the memory pressure shrinker will ignore
968 * any dentries belonging to this superblock that it comes across
969 * - the filesystem itself is no longer permitted to rearrange the dentries
972 void shrink_dcache_for_umount(struct super_block *sb)
974 struct dentry *dentry;
976 if (down_read_trylock(&sb->s_umount))
982 shrink_dcache_for_umount_subtree(dentry);
984 while (!hlist_bl_empty(&sb->s_anon)) {
985 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
986 shrink_dcache_for_umount_subtree(dentry);
991 * This tries to ascend one level of parenthood, but
992 * we can race with renaming, so we need to re-check
993 * the parenthood after dropping the lock and check
994 * that the sequence number still matches.
996 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
998 struct dentry *new = old->d_parent;
1001 spin_unlock(&old->d_lock);
1002 spin_lock(&new->d_lock);
1005 * might go back up the wrong parent if we have had a rename
1008 if (new != old->d_parent ||
1009 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1010 (!locked && read_seqretry(&rename_lock, seq))) {
1011 spin_unlock(&new->d_lock);
1020 * Search for at least 1 mount point in the dentry's subdirs.
1021 * We descend to the next level whenever the d_subdirs
1022 * list is non-empty and continue searching.
1026 * have_submounts - check for mounts over a dentry
1027 * @parent: dentry to check.
1029 * Return true if the parent or its subdirectories contain
1032 int have_submounts(struct dentry *parent)
1034 struct dentry *this_parent;
1035 struct list_head *next;
1039 seq = read_seqbegin(&rename_lock);
1041 this_parent = parent;
1043 if (d_mountpoint(parent))
1045 spin_lock(&this_parent->d_lock);
1047 next = this_parent->d_subdirs.next;
1049 while (next != &this_parent->d_subdirs) {
1050 struct list_head *tmp = next;
1051 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1054 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1055 /* Have we found a mount point ? */
1056 if (d_mountpoint(dentry)) {
1057 spin_unlock(&dentry->d_lock);
1058 spin_unlock(&this_parent->d_lock);
1061 if (!list_empty(&dentry->d_subdirs)) {
1062 spin_unlock(&this_parent->d_lock);
1063 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1064 this_parent = dentry;
1065 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1068 spin_unlock(&dentry->d_lock);
1071 * All done at this level ... ascend and resume the search.
1073 if (this_parent != parent) {
1074 struct dentry *child = this_parent;
1075 this_parent = try_to_ascend(this_parent, locked, seq);
1078 next = child->d_u.d_child.next;
1081 spin_unlock(&this_parent->d_lock);
1082 if (!locked && read_seqretry(&rename_lock, seq))
1085 write_sequnlock(&rename_lock);
1086 return 0; /* No mount points found in tree */
1088 if (!locked && read_seqretry(&rename_lock, seq))
1091 write_sequnlock(&rename_lock);
1098 write_seqlock(&rename_lock);
1101 EXPORT_SYMBOL(have_submounts);
1104 * Search the dentry child list of the specified parent,
1105 * and move any unused dentries to the end of the unused
1106 * list for prune_dcache(). We descend to the next level
1107 * whenever the d_subdirs list is non-empty and continue
1110 * It returns zero iff there are no unused children,
1111 * otherwise it returns the number of children moved to
1112 * the end of the unused list. This may not be the total
1113 * number of unused children, because select_parent can
1114 * drop the lock and return early due to latency
1117 static int select_parent(struct dentry *parent, struct list_head *dispose)
1119 struct dentry *this_parent;
1120 struct list_head *next;
1125 seq = read_seqbegin(&rename_lock);
1127 this_parent = parent;
1128 spin_lock(&this_parent->d_lock);
1130 next = this_parent->d_subdirs.next;
1132 while (next != &this_parent->d_subdirs) {
1133 struct list_head *tmp = next;
1134 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1137 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1140 * move only zero ref count dentries to the dispose list.
1142 * Those which are presently on the shrink list, being processed
1143 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1144 * loop in shrink_dcache_parent() might not make any progress
1147 if (dentry->d_count) {
1148 dentry_lru_del(dentry);
1149 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1150 dentry_lru_move_list(dentry, dispose);
1151 dentry->d_flags |= DCACHE_SHRINK_LIST;
1155 * We can return to the caller if we have found some (this
1156 * ensures forward progress). We'll be coming back to find
1159 if (found && need_resched()) {
1160 spin_unlock(&dentry->d_lock);
1165 * Descend a level if the d_subdirs list is non-empty.
1167 if (!list_empty(&dentry->d_subdirs)) {
1168 spin_unlock(&this_parent->d_lock);
1169 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1170 this_parent = dentry;
1171 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1175 spin_unlock(&dentry->d_lock);
1178 * All done at this level ... ascend and resume the search.
1180 if (this_parent != parent) {
1181 struct dentry *child = this_parent;
1182 this_parent = try_to_ascend(this_parent, locked, seq);
1185 next = child->d_u.d_child.next;
1189 spin_unlock(&this_parent->d_lock);
1190 if (!locked && read_seqretry(&rename_lock, seq))
1193 write_sequnlock(&rename_lock);
1202 write_seqlock(&rename_lock);
1207 * shrink_dcache_parent - prune dcache
1208 * @parent: parent of entries to prune
1210 * Prune the dcache to remove unused children of the parent dentry.
1212 void shrink_dcache_parent(struct dentry * parent)
1217 while ((found = select_parent(parent, &dispose)) != 0) {
1218 shrink_dentry_list(&dispose);
1222 EXPORT_SYMBOL(shrink_dcache_parent);
1225 * __d_alloc - allocate a dcache entry
1226 * @sb: filesystem it will belong to
1227 * @name: qstr of the name
1229 * Allocates a dentry. It returns %NULL if there is insufficient memory
1230 * available. On a success the dentry is returned. The name passed in is
1231 * copied and the copy passed in may be reused after this call.
1234 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1236 struct dentry *dentry;
1239 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1244 * We guarantee that the inline name is always NUL-terminated.
1245 * This way the memcpy() done by the name switching in rename
1246 * will still always have a NUL at the end, even if we might
1247 * be overwriting an internal NUL character
1249 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1250 if (name->len > DNAME_INLINE_LEN-1) {
1251 dname = kmalloc(name->len + 1, GFP_KERNEL);
1253 kmem_cache_free(dentry_cache, dentry);
1257 dname = dentry->d_iname;
1260 dentry->d_name.len = name->len;
1261 dentry->d_name.hash = name->hash;
1262 memcpy(dname, name->name, name->len);
1263 dname[name->len] = 0;
1265 /* Make sure we always see the terminating NUL character */
1267 dentry->d_name.name = dname;
1269 dentry->d_count = 1;
1270 dentry->d_flags = 0;
1271 spin_lock_init(&dentry->d_lock);
1272 seqcount_init(&dentry->d_seq);
1273 dentry->d_inode = NULL;
1274 dentry->d_parent = dentry;
1276 dentry->d_op = NULL;
1277 dentry->d_fsdata = NULL;
1278 INIT_HLIST_BL_NODE(&dentry->d_hash);
1279 INIT_LIST_HEAD(&dentry->d_lru);
1280 INIT_LIST_HEAD(&dentry->d_subdirs);
1281 INIT_HLIST_NODE(&dentry->d_alias);
1282 INIT_LIST_HEAD(&dentry->d_u.d_child);
1283 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1285 this_cpu_inc(nr_dentry);
1291 * d_alloc - allocate a dcache entry
1292 * @parent: parent of entry to allocate
1293 * @name: qstr of the name
1295 * Allocates a dentry. It returns %NULL if there is insufficient memory
1296 * available. On a success the dentry is returned. The name passed in is
1297 * copied and the copy passed in may be reused after this call.
1299 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1301 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1305 spin_lock(&parent->d_lock);
1307 * don't need child lock because it is not subject
1308 * to concurrency here
1310 __dget_dlock(parent);
1311 dentry->d_parent = parent;
1312 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1313 spin_unlock(&parent->d_lock);
1317 EXPORT_SYMBOL(d_alloc);
1319 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1321 struct dentry *dentry = __d_alloc(sb, name);
1323 dentry->d_flags |= DCACHE_DISCONNECTED;
1326 EXPORT_SYMBOL(d_alloc_pseudo);
1328 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1333 q.len = strlen(name);
1334 q.hash = full_name_hash(q.name, q.len);
1335 return d_alloc(parent, &q);
1337 EXPORT_SYMBOL(d_alloc_name);
1339 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1341 WARN_ON_ONCE(dentry->d_op);
1342 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1344 DCACHE_OP_REVALIDATE |
1345 DCACHE_OP_WEAK_REVALIDATE |
1346 DCACHE_OP_DELETE ));
1351 dentry->d_flags |= DCACHE_OP_HASH;
1353 dentry->d_flags |= DCACHE_OP_COMPARE;
1354 if (op->d_revalidate)
1355 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1356 if (op->d_weak_revalidate)
1357 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1359 dentry->d_flags |= DCACHE_OP_DELETE;
1361 dentry->d_flags |= DCACHE_OP_PRUNE;
1364 EXPORT_SYMBOL(d_set_d_op);
1366 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1368 spin_lock(&dentry->d_lock);
1370 if (unlikely(IS_AUTOMOUNT(inode)))
1371 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1372 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1374 dentry->d_inode = inode;
1375 dentry_rcuwalk_barrier(dentry);
1376 spin_unlock(&dentry->d_lock);
1377 fsnotify_d_instantiate(dentry, inode);
1381 * d_instantiate - fill in inode information for a dentry
1382 * @entry: dentry to complete
1383 * @inode: inode to attach to this dentry
1385 * Fill in inode information in the entry.
1387 * This turns negative dentries into productive full members
1390 * NOTE! This assumes that the inode count has been incremented
1391 * (or otherwise set) by the caller to indicate that it is now
1392 * in use by the dcache.
1395 void d_instantiate(struct dentry *entry, struct inode * inode)
1397 BUG_ON(!hlist_unhashed(&entry->d_alias));
1399 spin_lock(&inode->i_lock);
1400 __d_instantiate(entry, inode);
1402 spin_unlock(&inode->i_lock);
1403 security_d_instantiate(entry, inode);
1405 EXPORT_SYMBOL(d_instantiate);
1408 * d_instantiate_unique - instantiate a non-aliased dentry
1409 * @entry: dentry to instantiate
1410 * @inode: inode to attach to this dentry
1412 * Fill in inode information in the entry. On success, it returns NULL.
1413 * If an unhashed alias of "entry" already exists, then we return the
1414 * aliased dentry instead and drop one reference to inode.
1416 * Note that in order to avoid conflicts with rename() etc, the caller
1417 * had better be holding the parent directory semaphore.
1419 * This also assumes that the inode count has been incremented
1420 * (or otherwise set) by the caller to indicate that it is now
1421 * in use by the dcache.
1423 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1424 struct inode *inode)
1426 struct dentry *alias;
1427 int len = entry->d_name.len;
1428 const char *name = entry->d_name.name;
1429 unsigned int hash = entry->d_name.hash;
1432 __d_instantiate(entry, NULL);
1436 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1438 * Don't need alias->d_lock here, because aliases with
1439 * d_parent == entry->d_parent are not subject to name or
1440 * parent changes, because the parent inode i_mutex is held.
1442 if (alias->d_name.hash != hash)
1444 if (alias->d_parent != entry->d_parent)
1446 if (alias->d_name.len != len)
1448 if (dentry_cmp(alias, name, len))
1454 __d_instantiate(entry, inode);
1458 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1460 struct dentry *result;
1462 BUG_ON(!hlist_unhashed(&entry->d_alias));
1465 spin_lock(&inode->i_lock);
1466 result = __d_instantiate_unique(entry, inode);
1468 spin_unlock(&inode->i_lock);
1471 security_d_instantiate(entry, inode);
1475 BUG_ON(!d_unhashed(result));
1480 EXPORT_SYMBOL(d_instantiate_unique);
1482 struct dentry *d_make_root(struct inode *root_inode)
1484 struct dentry *res = NULL;
1487 static const struct qstr name = QSTR_INIT("/", 1);
1489 res = __d_alloc(root_inode->i_sb, &name);
1491 d_instantiate(res, root_inode);
1497 EXPORT_SYMBOL(d_make_root);
1499 static struct dentry * __d_find_any_alias(struct inode *inode)
1501 struct dentry *alias;
1503 if (hlist_empty(&inode->i_dentry))
1505 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1511 * d_find_any_alias - find any alias for a given inode
1512 * @inode: inode to find an alias for
1514 * If any aliases exist for the given inode, take and return a
1515 * reference for one of them. If no aliases exist, return %NULL.
1517 struct dentry *d_find_any_alias(struct inode *inode)
1521 spin_lock(&inode->i_lock);
1522 de = __d_find_any_alias(inode);
1523 spin_unlock(&inode->i_lock);
1526 EXPORT_SYMBOL(d_find_any_alias);
1529 * d_obtain_alias - find or allocate a dentry for a given inode
1530 * @inode: inode to allocate the dentry for
1532 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1533 * similar open by handle operations. The returned dentry may be anonymous,
1534 * or may have a full name (if the inode was already in the cache).
1536 * When called on a directory inode, we must ensure that the inode only ever
1537 * has one dentry. If a dentry is found, that is returned instead of
1538 * allocating a new one.
1540 * On successful return, the reference to the inode has been transferred
1541 * to the dentry. In case of an error the reference on the inode is released.
1542 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1543 * be passed in and will be the error will be propagate to the return value,
1544 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1546 struct dentry *d_obtain_alias(struct inode *inode)
1548 static const struct qstr anonstring = QSTR_INIT("/", 1);
1553 return ERR_PTR(-ESTALE);
1555 return ERR_CAST(inode);
1557 res = d_find_any_alias(inode);
1561 tmp = __d_alloc(inode->i_sb, &anonstring);
1563 res = ERR_PTR(-ENOMEM);
1567 spin_lock(&inode->i_lock);
1568 res = __d_find_any_alias(inode);
1570 spin_unlock(&inode->i_lock);
1575 /* attach a disconnected dentry */
1576 spin_lock(&tmp->d_lock);
1577 tmp->d_inode = inode;
1578 tmp->d_flags |= DCACHE_DISCONNECTED;
1579 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1580 hlist_bl_lock(&tmp->d_sb->s_anon);
1581 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1582 hlist_bl_unlock(&tmp->d_sb->s_anon);
1583 spin_unlock(&tmp->d_lock);
1584 spin_unlock(&inode->i_lock);
1585 security_d_instantiate(tmp, inode);
1590 if (res && !IS_ERR(res))
1591 security_d_instantiate(res, inode);
1595 EXPORT_SYMBOL(d_obtain_alias);
1598 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1599 * @inode: the inode which may have a disconnected dentry
1600 * @dentry: a negative dentry which we want to point to the inode.
1602 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1603 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1604 * and return it, else simply d_add the inode to the dentry and return NULL.
1606 * This is needed in the lookup routine of any filesystem that is exportable
1607 * (via knfsd) so that we can build dcache paths to directories effectively.
1609 * If a dentry was found and moved, then it is returned. Otherwise NULL
1610 * is returned. This matches the expected return value of ->lookup.
1613 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1615 struct dentry *new = NULL;
1618 return ERR_CAST(inode);
1620 if (inode && S_ISDIR(inode->i_mode)) {
1621 spin_lock(&inode->i_lock);
1622 new = __d_find_alias(inode, 1);
1624 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1625 spin_unlock(&inode->i_lock);
1626 security_d_instantiate(new, inode);
1627 d_move(new, dentry);
1630 /* already taking inode->i_lock, so d_add() by hand */
1631 __d_instantiate(dentry, inode);
1632 spin_unlock(&inode->i_lock);
1633 security_d_instantiate(dentry, inode);
1637 d_add(dentry, inode);
1640 EXPORT_SYMBOL(d_splice_alias);
1643 * d_add_ci - lookup or allocate new dentry with case-exact name
1644 * @inode: the inode case-insensitive lookup has found
1645 * @dentry: the negative dentry that was passed to the parent's lookup func
1646 * @name: the case-exact name to be associated with the returned dentry
1648 * This is to avoid filling the dcache with case-insensitive names to the
1649 * same inode, only the actual correct case is stored in the dcache for
1650 * case-insensitive filesystems.
1652 * For a case-insensitive lookup match and if the the case-exact dentry
1653 * already exists in in the dcache, use it and return it.
1655 * If no entry exists with the exact case name, allocate new dentry with
1656 * the exact case, and return the spliced entry.
1658 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1661 struct dentry *found;
1665 * First check if a dentry matching the name already exists,
1666 * if not go ahead and create it now.
1668 found = d_hash_and_lookup(dentry->d_parent, name);
1669 if (unlikely(IS_ERR(found)))
1672 new = d_alloc(dentry->d_parent, name);
1674 found = ERR_PTR(-ENOMEM);
1678 found = d_splice_alias(inode, new);
1687 * If a matching dentry exists, and it's not negative use it.
1689 * Decrement the reference count to balance the iget() done
1692 if (found->d_inode) {
1693 if (unlikely(found->d_inode != inode)) {
1694 /* This can't happen because bad inodes are unhashed. */
1695 BUG_ON(!is_bad_inode(inode));
1696 BUG_ON(!is_bad_inode(found->d_inode));
1703 * Negative dentry: instantiate it unless the inode is a directory and
1704 * already has a dentry.
1706 new = d_splice_alias(inode, found);
1717 EXPORT_SYMBOL(d_add_ci);
1720 * Do the slow-case of the dentry name compare.
1722 * Unlike the dentry_cmp() function, we need to atomically
1723 * load the name, length and inode information, so that the
1724 * filesystem can rely on them, and can use the 'name' and
1725 * 'len' information without worrying about walking off the
1726 * end of memory etc.
1728 * Thus the read_seqcount_retry() and the "duplicate" info
1729 * in arguments (the low-level filesystem should not look
1730 * at the dentry inode or name contents directly, since
1731 * rename can change them while we're in RCU mode).
1733 enum slow_d_compare {
1739 static noinline enum slow_d_compare slow_dentry_cmp(
1740 const struct dentry *parent,
1741 struct inode *inode,
1742 struct dentry *dentry,
1744 const struct qstr *name)
1746 int tlen = dentry->d_name.len;
1747 const char *tname = dentry->d_name.name;
1748 struct inode *i = dentry->d_inode;
1750 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1752 return D_COMP_SEQRETRY;
1754 if (parent->d_op->d_compare(parent, inode,
1757 return D_COMP_NOMATCH;
1762 * __d_lookup_rcu - search for a dentry (racy, store-free)
1763 * @parent: parent dentry
1764 * @name: qstr of name we wish to find
1765 * @seqp: returns d_seq value at the point where the dentry was found
1766 * @inode: returns dentry->d_inode when the inode was found valid.
1767 * Returns: dentry, or NULL
1769 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1770 * resolution (store-free path walking) design described in
1771 * Documentation/filesystems/path-lookup.txt.
1773 * This is not to be used outside core vfs.
1775 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1776 * held, and rcu_read_lock held. The returned dentry must not be stored into
1777 * without taking d_lock and checking d_seq sequence count against @seq
1780 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1783 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1784 * the returned dentry, so long as its parent's seqlock is checked after the
1785 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1786 * is formed, giving integrity down the path walk.
1788 * NOTE! The caller *has* to check the resulting dentry against the sequence
1789 * number we've returned before using any of the resulting dentry state!
1791 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1792 const struct qstr *name,
1793 unsigned *seqp, struct inode *inode)
1795 u64 hashlen = name->hash_len;
1796 const unsigned char *str = name->name;
1797 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1798 struct hlist_bl_node *node;
1799 struct dentry *dentry;
1802 * Note: There is significant duplication with __d_lookup_rcu which is
1803 * required to prevent single threaded performance regressions
1804 * especially on architectures where smp_rmb (in seqcounts) are costly.
1805 * Keep the two functions in sync.
1809 * The hash list is protected using RCU.
1811 * Carefully use d_seq when comparing a candidate dentry, to avoid
1812 * races with d_move().
1814 * It is possible that concurrent renames can mess up our list
1815 * walk here and result in missing our dentry, resulting in the
1816 * false-negative result. d_lookup() protects against concurrent
1817 * renames using rename_lock seqlock.
1819 * See Documentation/filesystems/path-lookup.txt for more details.
1821 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1826 * The dentry sequence count protects us from concurrent
1827 * renames, and thus protects inode, parent and name fields.
1829 * The caller must perform a seqcount check in order
1830 * to do anything useful with the returned dentry,
1831 * including using the 'd_inode' pointer.
1833 * NOTE! We do a "raw" seqcount_begin here. That means that
1834 * we don't wait for the sequence count to stabilize if it
1835 * is in the middle of a sequence change. If we do the slow
1836 * dentry compare, we will do seqretries until it is stable,
1837 * and if we end up with a successful lookup, we actually
1838 * want to exit RCU lookup anyway.
1840 seq = raw_seqcount_begin(&dentry->d_seq);
1841 if (dentry->d_parent != parent)
1843 if (d_unhashed(dentry))
1847 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1848 if (dentry->d_name.hash != hashlen_hash(hashlen))
1850 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1853 case D_COMP_NOMATCH:
1860 if (dentry->d_name.hash_len != hashlen)
1862 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
1869 * d_lookup - search for a dentry
1870 * @parent: parent dentry
1871 * @name: qstr of name we wish to find
1872 * Returns: dentry, or NULL
1874 * d_lookup searches the children of the parent dentry for the name in
1875 * question. If the dentry is found its reference count is incremented and the
1876 * dentry is returned. The caller must use dput to free the entry when it has
1877 * finished using it. %NULL is returned if the dentry does not exist.
1879 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
1881 struct dentry *dentry;
1885 seq = read_seqbegin(&rename_lock);
1886 dentry = __d_lookup(parent, name);
1889 } while (read_seqretry(&rename_lock, seq));
1892 EXPORT_SYMBOL(d_lookup);
1895 * __d_lookup - search for a dentry (racy)
1896 * @parent: parent dentry
1897 * @name: qstr of name we wish to find
1898 * Returns: dentry, or NULL
1900 * __d_lookup is like d_lookup, however it may (rarely) return a
1901 * false-negative result due to unrelated rename activity.
1903 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1904 * however it must be used carefully, eg. with a following d_lookup in
1905 * the case of failure.
1907 * __d_lookup callers must be commented.
1909 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
1911 unsigned int len = name->len;
1912 unsigned int hash = name->hash;
1913 const unsigned char *str = name->name;
1914 struct hlist_bl_head *b = d_hash(parent, hash);
1915 struct hlist_bl_node *node;
1916 struct dentry *found = NULL;
1917 struct dentry *dentry;
1920 * Note: There is significant duplication with __d_lookup_rcu which is
1921 * required to prevent single threaded performance regressions
1922 * especially on architectures where smp_rmb (in seqcounts) are costly.
1923 * Keep the two functions in sync.
1927 * The hash list is protected using RCU.
1929 * Take d_lock when comparing a candidate dentry, to avoid races
1932 * It is possible that concurrent renames can mess up our list
1933 * walk here and result in missing our dentry, resulting in the
1934 * false-negative result. d_lookup() protects against concurrent
1935 * renames using rename_lock seqlock.
1937 * See Documentation/filesystems/path-lookup.txt for more details.
1941 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1943 if (dentry->d_name.hash != hash)
1946 spin_lock(&dentry->d_lock);
1947 if (dentry->d_parent != parent)
1949 if (d_unhashed(dentry))
1953 * It is safe to compare names since d_move() cannot
1954 * change the qstr (protected by d_lock).
1956 if (parent->d_flags & DCACHE_OP_COMPARE) {
1957 int tlen = dentry->d_name.len;
1958 const char *tname = dentry->d_name.name;
1959 if (parent->d_op->d_compare(parent, parent->d_inode,
1960 dentry, dentry->d_inode,
1964 if (dentry->d_name.len != len)
1966 if (dentry_cmp(dentry, str, len))
1972 spin_unlock(&dentry->d_lock);
1975 spin_unlock(&dentry->d_lock);
1983 * d_hash_and_lookup - hash the qstr then search for a dentry
1984 * @dir: Directory to search in
1985 * @name: qstr of name we wish to find
1987 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
1989 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1992 * Check for a fs-specific hash function. Note that we must
1993 * calculate the standard hash first, as the d_op->d_hash()
1994 * routine may choose to leave the hash value unchanged.
1996 name->hash = full_name_hash(name->name, name->len);
1997 if (dir->d_flags & DCACHE_OP_HASH) {
1998 int err = dir->d_op->d_hash(dir, dir->d_inode, name);
1999 if (unlikely(err < 0))
2000 return ERR_PTR(err);
2002 return d_lookup(dir, name);
2004 EXPORT_SYMBOL(d_hash_and_lookup);
2007 * d_validate - verify dentry provided from insecure source (deprecated)
2008 * @dentry: The dentry alleged to be valid child of @dparent
2009 * @dparent: The parent dentry (known to be valid)
2011 * An insecure source has sent us a dentry, here we verify it and dget() it.
2012 * This is used by ncpfs in its readdir implementation.
2013 * Zero is returned in the dentry is invalid.
2015 * This function is slow for big directories, and deprecated, do not use it.
2017 int d_validate(struct dentry *dentry, struct dentry *dparent)
2019 struct dentry *child;
2021 spin_lock(&dparent->d_lock);
2022 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2023 if (dentry == child) {
2024 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2025 __dget_dlock(dentry);
2026 spin_unlock(&dentry->d_lock);
2027 spin_unlock(&dparent->d_lock);
2031 spin_unlock(&dparent->d_lock);
2035 EXPORT_SYMBOL(d_validate);
2038 * When a file is deleted, we have two options:
2039 * - turn this dentry into a negative dentry
2040 * - unhash this dentry and free it.
2042 * Usually, we want to just turn this into
2043 * a negative dentry, but if anybody else is
2044 * currently using the dentry or the inode
2045 * we can't do that and we fall back on removing
2046 * it from the hash queues and waiting for
2047 * it to be deleted later when it has no users
2051 * d_delete - delete a dentry
2052 * @dentry: The dentry to delete
2054 * Turn the dentry into a negative dentry if possible, otherwise
2055 * remove it from the hash queues so it can be deleted later
2058 void d_delete(struct dentry * dentry)
2060 struct inode *inode;
2063 * Are we the only user?
2066 spin_lock(&dentry->d_lock);
2067 inode = dentry->d_inode;
2068 isdir = S_ISDIR(inode->i_mode);
2069 if (dentry->d_count == 1) {
2070 if (!spin_trylock(&inode->i_lock)) {
2071 spin_unlock(&dentry->d_lock);
2075 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2076 dentry_unlink_inode(dentry);
2077 fsnotify_nameremove(dentry, isdir);
2081 if (!d_unhashed(dentry))
2084 spin_unlock(&dentry->d_lock);
2086 fsnotify_nameremove(dentry, isdir);
2088 EXPORT_SYMBOL(d_delete);
2090 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2092 BUG_ON(!d_unhashed(entry));
2094 entry->d_flags |= DCACHE_RCUACCESS;
2095 hlist_bl_add_head_rcu(&entry->d_hash, b);
2099 static void _d_rehash(struct dentry * entry)
2101 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2105 * d_rehash - add an entry back to the hash
2106 * @entry: dentry to add to the hash
2108 * Adds a dentry to the hash according to its name.
2111 void d_rehash(struct dentry * entry)
2113 spin_lock(&entry->d_lock);
2115 spin_unlock(&entry->d_lock);
2117 EXPORT_SYMBOL(d_rehash);
2120 * dentry_update_name_case - update case insensitive dentry with a new name
2121 * @dentry: dentry to be updated
2124 * Update a case insensitive dentry with new case of name.
2126 * dentry must have been returned by d_lookup with name @name. Old and new
2127 * name lengths must match (ie. no d_compare which allows mismatched name
2130 * Parent inode i_mutex must be held over d_lookup and into this call (to
2131 * keep renames and concurrent inserts, and readdir(2) away).
2133 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2135 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2136 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2138 spin_lock(&dentry->d_lock);
2139 write_seqcount_begin(&dentry->d_seq);
2140 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2141 write_seqcount_end(&dentry->d_seq);
2142 spin_unlock(&dentry->d_lock);
2144 EXPORT_SYMBOL(dentry_update_name_case);
2146 static void switch_names(struct dentry *dentry, struct dentry *target)
2148 if (dname_external(target)) {
2149 if (dname_external(dentry)) {
2151 * Both external: swap the pointers
2153 swap(target->d_name.name, dentry->d_name.name);
2156 * dentry:internal, target:external. Steal target's
2157 * storage and make target internal.
2159 memcpy(target->d_iname, dentry->d_name.name,
2160 dentry->d_name.len + 1);
2161 dentry->d_name.name = target->d_name.name;
2162 target->d_name.name = target->d_iname;
2165 if (dname_external(dentry)) {
2167 * dentry:external, target:internal. Give dentry's
2168 * storage to target and make dentry internal
2170 memcpy(dentry->d_iname, target->d_name.name,
2171 target->d_name.len + 1);
2172 target->d_name.name = dentry->d_name.name;
2173 dentry->d_name.name = dentry->d_iname;
2176 * Both are internal. Just copy target to dentry
2178 memcpy(dentry->d_iname, target->d_name.name,
2179 target->d_name.len + 1);
2180 dentry->d_name.len = target->d_name.len;
2184 swap(dentry->d_name.len, target->d_name.len);
2187 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2190 * XXXX: do we really need to take target->d_lock?
2192 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2193 spin_lock(&target->d_parent->d_lock);
2195 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2196 spin_lock(&dentry->d_parent->d_lock);
2197 spin_lock_nested(&target->d_parent->d_lock,
2198 DENTRY_D_LOCK_NESTED);
2200 spin_lock(&target->d_parent->d_lock);
2201 spin_lock_nested(&dentry->d_parent->d_lock,
2202 DENTRY_D_LOCK_NESTED);
2205 if (target < dentry) {
2206 spin_lock_nested(&target->d_lock, 2);
2207 spin_lock_nested(&dentry->d_lock, 3);
2209 spin_lock_nested(&dentry->d_lock, 2);
2210 spin_lock_nested(&target->d_lock, 3);
2214 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2215 struct dentry *target)
2217 if (target->d_parent != dentry->d_parent)
2218 spin_unlock(&dentry->d_parent->d_lock);
2219 if (target->d_parent != target)
2220 spin_unlock(&target->d_parent->d_lock);
2224 * When switching names, the actual string doesn't strictly have to
2225 * be preserved in the target - because we're dropping the target
2226 * anyway. As such, we can just do a simple memcpy() to copy over
2227 * the new name before we switch.
2229 * Note that we have to be a lot more careful about getting the hash
2230 * switched - we have to switch the hash value properly even if it
2231 * then no longer matches the actual (corrupted) string of the target.
2232 * The hash value has to match the hash queue that the dentry is on..
2235 * __d_move - move a dentry
2236 * @dentry: entry to move
2237 * @target: new dentry
2239 * Update the dcache to reflect the move of a file name. Negative
2240 * dcache entries should not be moved in this way. Caller must hold
2241 * rename_lock, the i_mutex of the source and target directories,
2242 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2244 static void __d_move(struct dentry * dentry, struct dentry * target)
2246 if (!dentry->d_inode)
2247 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2249 BUG_ON(d_ancestor(dentry, target));
2250 BUG_ON(d_ancestor(target, dentry));
2252 dentry_lock_for_move(dentry, target);
2254 write_seqcount_begin(&dentry->d_seq);
2255 write_seqcount_begin(&target->d_seq);
2257 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2260 * Move the dentry to the target hash queue. Don't bother checking
2261 * for the same hash queue because of how unlikely it is.
2264 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2266 /* Unhash the target: dput() will then get rid of it */
2269 list_del(&dentry->d_u.d_child);
2270 list_del(&target->d_u.d_child);
2272 /* Switch the names.. */
2273 switch_names(dentry, target);
2274 swap(dentry->d_name.hash, target->d_name.hash);
2276 /* ... and switch the parents */
2277 if (IS_ROOT(dentry)) {
2278 dentry->d_parent = target->d_parent;
2279 target->d_parent = target;
2280 INIT_LIST_HEAD(&target->d_u.d_child);
2282 swap(dentry->d_parent, target->d_parent);
2284 /* And add them back to the (new) parent lists */
2285 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2288 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2290 write_seqcount_end(&target->d_seq);
2291 write_seqcount_end(&dentry->d_seq);
2293 dentry_unlock_parents_for_move(dentry, target);
2294 spin_unlock(&target->d_lock);
2295 fsnotify_d_move(dentry);
2296 spin_unlock(&dentry->d_lock);
2300 * d_move - move a dentry
2301 * @dentry: entry to move
2302 * @target: new dentry
2304 * Update the dcache to reflect the move of a file name. Negative
2305 * dcache entries should not be moved in this way. See the locking
2306 * requirements for __d_move.
2308 void d_move(struct dentry *dentry, struct dentry *target)
2310 write_seqlock(&rename_lock);
2311 __d_move(dentry, target);
2312 write_sequnlock(&rename_lock);
2314 EXPORT_SYMBOL(d_move);
2317 * d_ancestor - search for an ancestor
2318 * @p1: ancestor dentry
2321 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2322 * an ancestor of p2, else NULL.
2324 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2328 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2329 if (p->d_parent == p1)
2336 * This helper attempts to cope with remotely renamed directories
2338 * It assumes that the caller is already holding
2339 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2341 * Note: If ever the locking in lock_rename() changes, then please
2342 * remember to update this too...
2344 static struct dentry *__d_unalias(struct inode *inode,
2345 struct dentry *dentry, struct dentry *alias)
2347 struct mutex *m1 = NULL, *m2 = NULL;
2348 struct dentry *ret = ERR_PTR(-EBUSY);
2350 /* If alias and dentry share a parent, then no extra locks required */
2351 if (alias->d_parent == dentry->d_parent)
2354 /* See lock_rename() */
2355 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2357 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2358 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2360 m2 = &alias->d_parent->d_inode->i_mutex;
2362 if (likely(!d_mountpoint(alias))) {
2363 __d_move(alias, dentry);
2367 spin_unlock(&inode->i_lock);
2376 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2377 * named dentry in place of the dentry to be replaced.
2378 * returns with anon->d_lock held!
2380 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2382 struct dentry *dparent;
2384 dentry_lock_for_move(anon, dentry);
2386 write_seqcount_begin(&dentry->d_seq);
2387 write_seqcount_begin(&anon->d_seq);
2389 dparent = dentry->d_parent;
2391 switch_names(dentry, anon);
2392 swap(dentry->d_name.hash, anon->d_name.hash);
2394 dentry->d_parent = dentry;
2395 list_del_init(&dentry->d_u.d_child);
2396 anon->d_parent = dparent;
2397 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2399 write_seqcount_end(&dentry->d_seq);
2400 write_seqcount_end(&anon->d_seq);
2402 dentry_unlock_parents_for_move(anon, dentry);
2403 spin_unlock(&dentry->d_lock);
2405 /* anon->d_lock still locked, returns locked */
2406 anon->d_flags &= ~DCACHE_DISCONNECTED;
2410 * d_materialise_unique - introduce an inode into the tree
2411 * @dentry: candidate dentry
2412 * @inode: inode to bind to the dentry, to which aliases may be attached
2414 * Introduces an dentry into the tree, substituting an extant disconnected
2415 * root directory alias in its place if there is one. Caller must hold the
2416 * i_mutex of the parent directory.
2418 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2420 struct dentry *actual;
2422 BUG_ON(!d_unhashed(dentry));
2426 __d_instantiate(dentry, NULL);
2431 spin_lock(&inode->i_lock);
2433 if (S_ISDIR(inode->i_mode)) {
2434 struct dentry *alias;
2436 /* Does an aliased dentry already exist? */
2437 alias = __d_find_alias(inode, 0);
2440 write_seqlock(&rename_lock);
2442 if (d_ancestor(alias, dentry)) {
2443 /* Check for loops */
2444 actual = ERR_PTR(-ELOOP);
2445 spin_unlock(&inode->i_lock);
2446 } else if (IS_ROOT(alias)) {
2447 /* Is this an anonymous mountpoint that we
2448 * could splice into our tree? */
2449 __d_materialise_dentry(dentry, alias);
2450 write_sequnlock(&rename_lock);
2454 /* Nope, but we must(!) avoid directory
2455 * aliasing. This drops inode->i_lock */
2456 actual = __d_unalias(inode, dentry, alias);
2458 write_sequnlock(&rename_lock);
2459 if (IS_ERR(actual)) {
2460 if (PTR_ERR(actual) == -ELOOP)
2461 pr_warn_ratelimited(
2462 "VFS: Lookup of '%s' in %s %s"
2463 " would have caused loop\n",
2464 dentry->d_name.name,
2465 inode->i_sb->s_type->name,
2473 /* Add a unique reference */
2474 actual = __d_instantiate_unique(dentry, inode);
2478 BUG_ON(!d_unhashed(actual));
2480 spin_lock(&actual->d_lock);
2483 spin_unlock(&actual->d_lock);
2484 spin_unlock(&inode->i_lock);
2486 if (actual == dentry) {
2487 security_d_instantiate(dentry, inode);
2494 EXPORT_SYMBOL_GPL(d_materialise_unique);
2496 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2500 return -ENAMETOOLONG;
2502 memcpy(*buffer, str, namelen);
2506 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2508 return prepend(buffer, buflen, name->name, name->len);
2512 * prepend_path - Prepend path string to a buffer
2513 * @path: the dentry/vfsmount to report
2514 * @root: root vfsmnt/dentry
2515 * @buffer: pointer to the end of the buffer
2516 * @buflen: pointer to buffer length
2518 * Caller holds the rename_lock.
2520 static int prepend_path(const struct path *path,
2521 const struct path *root,
2522 char **buffer, int *buflen)
2524 struct dentry *dentry = path->dentry;
2525 struct vfsmount *vfsmnt = path->mnt;
2526 struct mount *mnt = real_mount(vfsmnt);
2530 while (dentry != root->dentry || vfsmnt != root->mnt) {
2531 struct dentry * parent;
2533 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2535 if (!mnt_has_parent(mnt))
2537 dentry = mnt->mnt_mountpoint;
2538 mnt = mnt->mnt_parent;
2542 parent = dentry->d_parent;
2544 spin_lock(&dentry->d_lock);
2545 error = prepend_name(buffer, buflen, &dentry->d_name);
2546 spin_unlock(&dentry->d_lock);
2548 error = prepend(buffer, buflen, "/", 1);
2556 if (!error && !slash)
2557 error = prepend(buffer, buflen, "/", 1);
2563 * Filesystems needing to implement special "root names"
2564 * should do so with ->d_dname()
2566 if (IS_ROOT(dentry) &&
2567 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2568 WARN(1, "Root dentry has weird name <%.*s>\n",
2569 (int) dentry->d_name.len, dentry->d_name.name);
2572 error = prepend(buffer, buflen, "/", 1);
2574 error = is_mounted(vfsmnt) ? 1 : 2;
2579 * __d_path - return the path of a dentry
2580 * @path: the dentry/vfsmount to report
2581 * @root: root vfsmnt/dentry
2582 * @buf: buffer to return value in
2583 * @buflen: buffer length
2585 * Convert a dentry into an ASCII path name.
2587 * Returns a pointer into the buffer or an error code if the
2588 * path was too long.
2590 * "buflen" should be positive.
2592 * If the path is not reachable from the supplied root, return %NULL.
2594 char *__d_path(const struct path *path,
2595 const struct path *root,
2596 char *buf, int buflen)
2598 char *res = buf + buflen;
2601 prepend(&res, &buflen, "\0", 1);
2602 br_read_lock(&vfsmount_lock);
2603 write_seqlock(&rename_lock);
2604 error = prepend_path(path, root, &res, &buflen);
2605 write_sequnlock(&rename_lock);
2606 br_read_unlock(&vfsmount_lock);
2609 return ERR_PTR(error);
2615 char *d_absolute_path(const struct path *path,
2616 char *buf, int buflen)
2618 struct path root = {};
2619 char *res = buf + buflen;
2622 prepend(&res, &buflen, "\0", 1);
2623 br_read_lock(&vfsmount_lock);
2624 write_seqlock(&rename_lock);
2625 error = prepend_path(path, &root, &res, &buflen);
2626 write_sequnlock(&rename_lock);
2627 br_read_unlock(&vfsmount_lock);
2632 return ERR_PTR(error);
2637 * same as __d_path but appends "(deleted)" for unlinked files.
2639 static int path_with_deleted(const struct path *path,
2640 const struct path *root,
2641 char **buf, int *buflen)
2643 prepend(buf, buflen, "\0", 1);
2644 if (d_unlinked(path->dentry)) {
2645 int error = prepend(buf, buflen, " (deleted)", 10);
2650 return prepend_path(path, root, buf, buflen);
2653 static int prepend_unreachable(char **buffer, int *buflen)
2655 return prepend(buffer, buflen, "(unreachable)", 13);
2659 * d_path - return the path of a dentry
2660 * @path: path to report
2661 * @buf: buffer to return value in
2662 * @buflen: buffer length
2664 * Convert a dentry into an ASCII path name. If the entry has been deleted
2665 * the string " (deleted)" is appended. Note that this is ambiguous.
2667 * Returns a pointer into the buffer or an error code if the path was
2668 * too long. Note: Callers should use the returned pointer, not the passed
2669 * in buffer, to use the name! The implementation often starts at an offset
2670 * into the buffer, and may leave 0 bytes at the start.
2672 * "buflen" should be positive.
2674 char *d_path(const struct path *path, char *buf, int buflen)
2676 char *res = buf + buflen;
2681 * We have various synthetic filesystems that never get mounted. On
2682 * these filesystems dentries are never used for lookup purposes, and
2683 * thus don't need to be hashed. They also don't need a name until a
2684 * user wants to identify the object in /proc/pid/fd/. The little hack
2685 * below allows us to generate a name for these objects on demand:
2687 * Some pseudo inodes are mountable. When they are mounted
2688 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
2689 * and instead have d_path return the mounted path.
2691 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
2692 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
2693 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2695 get_fs_root(current->fs, &root);
2696 br_read_lock(&vfsmount_lock);
2697 write_seqlock(&rename_lock);
2698 error = path_with_deleted(path, &root, &res, &buflen);
2699 write_sequnlock(&rename_lock);
2700 br_read_unlock(&vfsmount_lock);
2702 res = ERR_PTR(error);
2706 EXPORT_SYMBOL(d_path);
2709 * Helper function for dentry_operations.d_dname() members
2711 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2712 const char *fmt, ...)
2718 va_start(args, fmt);
2719 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2722 if (sz > sizeof(temp) || sz > buflen)
2723 return ERR_PTR(-ENAMETOOLONG);
2725 buffer += buflen - sz;
2726 return memcpy(buffer, temp, sz);
2729 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2731 char *end = buffer + buflen;
2732 /* these dentries are never renamed, so d_lock is not needed */
2733 if (prepend(&end, &buflen, " (deleted)", 11) ||
2734 prepend_name(&end, &buflen, &dentry->d_name) ||
2735 prepend(&end, &buflen, "/", 1))
2736 end = ERR_PTR(-ENAMETOOLONG);
2741 * Write full pathname from the root of the filesystem into the buffer.
2743 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2745 char *end = buf + buflen;
2748 prepend(&end, &buflen, "\0", 1);
2755 while (!IS_ROOT(dentry)) {
2756 struct dentry *parent = dentry->d_parent;
2760 spin_lock(&dentry->d_lock);
2761 error = prepend_name(&end, &buflen, &dentry->d_name);
2762 spin_unlock(&dentry->d_lock);
2763 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2771 return ERR_PTR(-ENAMETOOLONG);
2774 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2778 write_seqlock(&rename_lock);
2779 retval = __dentry_path(dentry, buf, buflen);
2780 write_sequnlock(&rename_lock);
2784 EXPORT_SYMBOL(dentry_path_raw);
2786 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2791 write_seqlock(&rename_lock);
2792 if (d_unlinked(dentry)) {
2794 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2798 retval = __dentry_path(dentry, buf, buflen);
2799 write_sequnlock(&rename_lock);
2800 if (!IS_ERR(retval) && p)
2801 *p = '/'; /* restore '/' overriden with '\0' */
2804 return ERR_PTR(-ENAMETOOLONG);
2808 * NOTE! The user-level library version returns a
2809 * character pointer. The kernel system call just
2810 * returns the length of the buffer filled (which
2811 * includes the ending '\0' character), or a negative
2812 * error value. So libc would do something like
2814 * char *getcwd(char * buf, size_t size)
2818 * retval = sys_getcwd(buf, size);
2825 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2828 struct path pwd, root;
2829 char *page = (char *) __get_free_page(GFP_USER);
2834 get_fs_root_and_pwd(current->fs, &root, &pwd);
2837 br_read_lock(&vfsmount_lock);
2838 write_seqlock(&rename_lock);
2839 if (!d_unlinked(pwd.dentry)) {
2841 char *cwd = page + PAGE_SIZE;
2842 int buflen = PAGE_SIZE;
2844 prepend(&cwd, &buflen, "\0", 1);
2845 error = prepend_path(&pwd, &root, &cwd, &buflen);
2846 write_sequnlock(&rename_lock);
2847 br_read_unlock(&vfsmount_lock);
2852 /* Unreachable from current root */
2854 error = prepend_unreachable(&cwd, &buflen);
2860 len = PAGE_SIZE + page - cwd;
2863 if (copy_to_user(buf, cwd, len))
2867 write_sequnlock(&rename_lock);
2868 br_read_unlock(&vfsmount_lock);
2874 free_page((unsigned long) page);
2879 * Test whether new_dentry is a subdirectory of old_dentry.
2881 * Trivially implemented using the dcache structure
2885 * is_subdir - is new dentry a subdirectory of old_dentry
2886 * @new_dentry: new dentry
2887 * @old_dentry: old dentry
2889 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2890 * Returns 0 otherwise.
2891 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2894 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2899 if (new_dentry == old_dentry)
2903 /* for restarting inner loop in case of seq retry */
2904 seq = read_seqbegin(&rename_lock);
2906 * Need rcu_readlock to protect against the d_parent trashing
2910 if (d_ancestor(old_dentry, new_dentry))
2915 } while (read_seqretry(&rename_lock, seq));
2920 void d_genocide(struct dentry *root)
2922 struct dentry *this_parent;
2923 struct list_head *next;
2927 seq = read_seqbegin(&rename_lock);
2930 spin_lock(&this_parent->d_lock);
2932 next = this_parent->d_subdirs.next;
2934 while (next != &this_parent->d_subdirs) {
2935 struct list_head *tmp = next;
2936 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2939 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2940 if (d_unhashed(dentry) || !dentry->d_inode) {
2941 spin_unlock(&dentry->d_lock);
2944 if (!list_empty(&dentry->d_subdirs)) {
2945 spin_unlock(&this_parent->d_lock);
2946 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2947 this_parent = dentry;
2948 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2951 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2952 dentry->d_flags |= DCACHE_GENOCIDE;
2955 spin_unlock(&dentry->d_lock);
2957 if (this_parent != root) {
2958 struct dentry *child = this_parent;
2959 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2960 this_parent->d_flags |= DCACHE_GENOCIDE;
2961 this_parent->d_count--;
2963 this_parent = try_to_ascend(this_parent, locked, seq);
2966 next = child->d_u.d_child.next;
2969 spin_unlock(&this_parent->d_lock);
2970 if (!locked && read_seqretry(&rename_lock, seq))
2973 write_sequnlock(&rename_lock);
2980 write_seqlock(&rename_lock);
2985 * find_inode_number - check for dentry with name
2986 * @dir: directory to check
2987 * @name: Name to find.
2989 * Check whether a dentry already exists for the given name,
2990 * and return the inode number if it has an inode. Otherwise
2993 * This routine is used to post-process directory listings for
2994 * filesystems using synthetic inode numbers, and is necessary
2995 * to keep getcwd() working.
2998 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3000 struct dentry * dentry;
3003 dentry = d_hash_and_lookup(dir, name);
3004 if (!IS_ERR_OR_NULL(dentry)) {
3005 if (dentry->d_inode)
3006 ino = dentry->d_inode->i_ino;
3011 EXPORT_SYMBOL(find_inode_number);
3013 static __initdata unsigned long dhash_entries;
3014 static int __init set_dhash_entries(char *str)
3018 dhash_entries = simple_strtoul(str, &str, 0);
3021 __setup("dhash_entries=", set_dhash_entries);
3023 static void __init dcache_init_early(void)
3027 /* If hashes are distributed across NUMA nodes, defer
3028 * hash allocation until vmalloc space is available.
3034 alloc_large_system_hash("Dentry cache",
3035 sizeof(struct hlist_bl_head),
3044 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3045 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3048 static void __init dcache_init(void)
3053 * A constructor could be added for stable state like the lists,
3054 * but it is probably not worth it because of the cache nature
3057 dentry_cache = KMEM_CACHE(dentry,
3058 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3060 /* Hash may have been set up in dcache_init_early */
3065 alloc_large_system_hash("Dentry cache",
3066 sizeof(struct hlist_bl_head),
3075 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3076 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3079 /* SLAB cache for __getname() consumers */
3080 struct kmem_cache *names_cachep __read_mostly;
3081 EXPORT_SYMBOL(names_cachep);
3083 EXPORT_SYMBOL(d_genocide);
3085 void __init vfs_caches_init_early(void)
3087 dcache_init_early();
3091 void __init vfs_caches_init(unsigned long mempages)
3093 unsigned long reserve;
3095 /* Base hash sizes on available memory, with a reserve equal to
3096 150% of current kernel size */
3098 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3099 mempages -= reserve;
3101 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3102 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3106 files_init(mempages);