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 * read_seqbegin_or_lock - begin a sequence number check or locking block
94 * seq : sequence number to be checked
96 * First try it once optimistically without taking the lock. If that fails,
97 * take the lock. The sequence number is also used as a marker for deciding
98 * whether to be a reader (even) or writer (odd).
99 * N.B. seq must be initialized to an even number to begin with.
101 static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
103 if (!(*seq & 1)) { /* Even */
104 *seq = read_seqbegin(lock);
111 * read_seqretry_or_unlock - end a seqretry or lock block & return retry status
112 * lock : sequence lock
113 * seq : sequence number
114 * Return: 1 to retry operation again, 0 to continue
116 static inline int read_seqretry_or_unlock(seqlock_t *lock, int *seq)
118 if (!(*seq & 1)) { /* Even */
120 if (read_seqretry(lock, *seq)) {
121 (*seq)++; /* Take writer lock */
125 write_sequnlock(lock);
130 * This is the single most critical data structure when it comes
131 * to the dcache: the hashtable for lookups. Somebody should try
132 * to make this good - I've just made it work.
134 * This hash-function tries to avoid losing too many bits of hash
135 * information, yet avoid using a prime hash-size or similar.
137 #define D_HASHBITS d_hash_shift
138 #define D_HASHMASK d_hash_mask
140 static unsigned int d_hash_mask __read_mostly;
141 static unsigned int d_hash_shift __read_mostly;
143 static struct hlist_bl_head *dentry_hashtable __read_mostly;
145 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
148 hash += (unsigned long) parent / L1_CACHE_BYTES;
149 hash = hash + (hash >> D_HASHBITS);
150 return dentry_hashtable + (hash & D_HASHMASK);
153 /* Statistics gathering. */
154 struct dentry_stat_t dentry_stat = {
158 static DEFINE_PER_CPU(unsigned int, nr_dentry);
160 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
161 static int get_nr_dentry(void)
165 for_each_possible_cpu(i)
166 sum += per_cpu(nr_dentry, i);
167 return sum < 0 ? 0 : sum;
170 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
171 size_t *lenp, loff_t *ppos)
173 dentry_stat.nr_dentry = get_nr_dentry();
174 return proc_dointvec(table, write, buffer, lenp, ppos);
179 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
180 * The strings are both count bytes long, and count is non-zero.
182 #ifdef CONFIG_DCACHE_WORD_ACCESS
184 #include <asm/word-at-a-time.h>
186 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
187 * aligned allocation for this particular component. We don't
188 * strictly need the load_unaligned_zeropad() safety, but it
189 * doesn't hurt either.
191 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
192 * need the careful unaligned handling.
194 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
196 unsigned long a,b,mask;
199 a = *(unsigned long *)cs;
200 b = load_unaligned_zeropad(ct);
201 if (tcount < sizeof(unsigned long))
203 if (unlikely(a != b))
205 cs += sizeof(unsigned long);
206 ct += sizeof(unsigned long);
207 tcount -= sizeof(unsigned long);
211 mask = ~(~0ul << tcount*8);
212 return unlikely(!!((a ^ b) & mask));
217 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
231 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
233 const unsigned char *cs;
235 * Be careful about RCU walk racing with rename:
236 * use ACCESS_ONCE to fetch the name pointer.
238 * NOTE! Even if a rename will mean that the length
239 * was not loaded atomically, we don't care. The
240 * RCU walk will check the sequence count eventually,
241 * and catch it. And we won't overrun the buffer,
242 * because we're reading the name pointer atomically,
243 * and a dentry name is guaranteed to be properly
244 * terminated with a NUL byte.
246 * End result: even if 'len' is wrong, we'll exit
247 * early because the data cannot match (there can
248 * be no NUL in the ct/tcount data)
250 cs = ACCESS_ONCE(dentry->d_name.name);
251 smp_read_barrier_depends();
252 return dentry_string_cmp(cs, ct, tcount);
255 static void __d_free(struct rcu_head *head)
257 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
259 WARN_ON(!hlist_unhashed(&dentry->d_alias));
260 if (dname_external(dentry))
261 kfree(dentry->d_name.name);
262 kmem_cache_free(dentry_cache, dentry);
268 static void d_free(struct dentry *dentry)
270 BUG_ON(dentry->d_lockref.count);
271 this_cpu_dec(nr_dentry);
272 if (dentry->d_op && dentry->d_op->d_release)
273 dentry->d_op->d_release(dentry);
275 /* if dentry was never visible to RCU, immediate free is OK */
276 if (!(dentry->d_flags & DCACHE_RCUACCESS))
277 __d_free(&dentry->d_u.d_rcu);
279 call_rcu(&dentry->d_u.d_rcu, __d_free);
283 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
284 * @dentry: the target dentry
285 * After this call, in-progress rcu-walk path lookup will fail. This
286 * should be called after unhashing, and after changing d_inode (if
287 * the dentry has not already been unhashed).
289 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
291 assert_spin_locked(&dentry->d_lock);
292 /* Go through a barrier */
293 write_seqcount_barrier(&dentry->d_seq);
297 * Release the dentry's inode, using the filesystem
298 * d_iput() operation if defined. Dentry has no refcount
301 static void dentry_iput(struct dentry * dentry)
302 __releases(dentry->d_lock)
303 __releases(dentry->d_inode->i_lock)
305 struct inode *inode = dentry->d_inode;
307 dentry->d_inode = NULL;
308 hlist_del_init(&dentry->d_alias);
309 spin_unlock(&dentry->d_lock);
310 spin_unlock(&inode->i_lock);
312 fsnotify_inoderemove(inode);
313 if (dentry->d_op && dentry->d_op->d_iput)
314 dentry->d_op->d_iput(dentry, inode);
318 spin_unlock(&dentry->d_lock);
323 * Release the dentry's inode, using the filesystem
324 * d_iput() operation if defined. dentry remains in-use.
326 static void dentry_unlink_inode(struct dentry * dentry)
327 __releases(dentry->d_lock)
328 __releases(dentry->d_inode->i_lock)
330 struct inode *inode = dentry->d_inode;
331 dentry->d_inode = NULL;
332 hlist_del_init(&dentry->d_alias);
333 dentry_rcuwalk_barrier(dentry);
334 spin_unlock(&dentry->d_lock);
335 spin_unlock(&inode->i_lock);
337 fsnotify_inoderemove(inode);
338 if (dentry->d_op && dentry->d_op->d_iput)
339 dentry->d_op->d_iput(dentry, inode);
345 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
347 static void dentry_lru_add(struct dentry *dentry)
349 if (list_empty(&dentry->d_lru)) {
350 spin_lock(&dcache_lru_lock);
351 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
352 dentry->d_sb->s_nr_dentry_unused++;
353 dentry_stat.nr_unused++;
354 spin_unlock(&dcache_lru_lock);
358 static void __dentry_lru_del(struct dentry *dentry)
360 list_del_init(&dentry->d_lru);
361 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
362 dentry->d_sb->s_nr_dentry_unused--;
363 dentry_stat.nr_unused--;
367 * Remove a dentry with references from the LRU.
369 static void dentry_lru_del(struct dentry *dentry)
371 if (!list_empty(&dentry->d_lru)) {
372 spin_lock(&dcache_lru_lock);
373 __dentry_lru_del(dentry);
374 spin_unlock(&dcache_lru_lock);
378 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
380 spin_lock(&dcache_lru_lock);
381 if (list_empty(&dentry->d_lru)) {
382 list_add_tail(&dentry->d_lru, list);
383 dentry->d_sb->s_nr_dentry_unused++;
384 dentry_stat.nr_unused++;
386 list_move_tail(&dentry->d_lru, list);
388 spin_unlock(&dcache_lru_lock);
392 * d_kill - kill dentry and return parent
393 * @dentry: dentry to kill
394 * @parent: parent dentry
396 * The dentry must already be unhashed and removed from the LRU.
398 * If this is the root of the dentry tree, return NULL.
400 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
403 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
404 __releases(dentry->d_lock)
405 __releases(parent->d_lock)
406 __releases(dentry->d_inode->i_lock)
408 list_del(&dentry->d_u.d_child);
410 * Inform try_to_ascend() that we are no longer attached to the
413 dentry->d_flags |= DCACHE_DENTRY_KILLED;
415 spin_unlock(&parent->d_lock);
418 * dentry_iput drops the locks, at which point nobody (except
419 * transient RCU lookups) can reach this dentry.
426 * Unhash a dentry without inserting an RCU walk barrier or checking that
427 * dentry->d_lock is locked. The caller must take care of that, if
430 static void __d_shrink(struct dentry *dentry)
432 if (!d_unhashed(dentry)) {
433 struct hlist_bl_head *b;
434 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
435 b = &dentry->d_sb->s_anon;
437 b = d_hash(dentry->d_parent, dentry->d_name.hash);
440 __hlist_bl_del(&dentry->d_hash);
441 dentry->d_hash.pprev = NULL;
447 * d_drop - drop a dentry
448 * @dentry: dentry to drop
450 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
451 * be found through a VFS lookup any more. Note that this is different from
452 * deleting the dentry - d_delete will try to mark the dentry negative if
453 * possible, giving a successful _negative_ lookup, while d_drop will
454 * just make the cache lookup fail.
456 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
457 * reason (NFS timeouts or autofs deletes).
459 * __d_drop requires dentry->d_lock.
461 void __d_drop(struct dentry *dentry)
463 if (!d_unhashed(dentry)) {
465 dentry_rcuwalk_barrier(dentry);
468 EXPORT_SYMBOL(__d_drop);
470 void d_drop(struct dentry *dentry)
472 spin_lock(&dentry->d_lock);
474 spin_unlock(&dentry->d_lock);
476 EXPORT_SYMBOL(d_drop);
479 * Finish off a dentry we've decided to kill.
480 * dentry->d_lock must be held, returns with it unlocked.
481 * If ref is non-zero, then decrement the refcount too.
482 * Returns dentry requiring refcount drop, or NULL if we're done.
484 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
485 __releases(dentry->d_lock)
488 struct dentry *parent;
490 inode = dentry->d_inode;
491 if (inode && !spin_trylock(&inode->i_lock)) {
493 spin_unlock(&dentry->d_lock);
495 return dentry; /* try again with same dentry */
500 parent = dentry->d_parent;
501 if (parent && !spin_trylock(&parent->d_lock)) {
503 spin_unlock(&inode->i_lock);
508 dentry->d_lockref.count--;
510 * inform the fs via d_prune that this dentry is about to be
511 * unhashed and destroyed.
513 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
514 dentry->d_op->d_prune(dentry);
516 dentry_lru_del(dentry);
517 /* if it was on the hash then remove it */
519 return d_kill(dentry, parent);
525 * This is complicated by the fact that we do not want to put
526 * dentries that are no longer on any hash chain on the unused
527 * list: we'd much rather just get rid of them immediately.
529 * However, that implies that we have to traverse the dentry
530 * tree upwards to the parents which might _also_ now be
531 * scheduled for deletion (it may have been only waiting for
532 * its last child to go away).
534 * This tail recursion is done by hand as we don't want to depend
535 * on the compiler to always get this right (gcc generally doesn't).
536 * Real recursion would eat up our stack space.
540 * dput - release a dentry
541 * @dentry: dentry to release
543 * Release a dentry. This will drop the usage count and if appropriate
544 * call the dentry unlink method as well as removing it from the queues and
545 * releasing its resources. If the parent dentries were scheduled for release
546 * they too may now get deleted.
548 void dput(struct dentry *dentry)
554 if (dentry->d_lockref.count == 1)
556 if (lockref_put_or_lock(&dentry->d_lockref))
559 if (dentry->d_flags & DCACHE_OP_DELETE) {
560 if (dentry->d_op->d_delete(dentry))
564 /* Unreachable? Get rid of it */
565 if (d_unhashed(dentry))
568 dentry->d_flags |= DCACHE_REFERENCED;
569 dentry_lru_add(dentry);
571 dentry->d_lockref.count--;
572 spin_unlock(&dentry->d_lock);
576 dentry = dentry_kill(dentry, 1);
583 * d_invalidate - invalidate a dentry
584 * @dentry: dentry to invalidate
586 * Try to invalidate the dentry if it turns out to be
587 * possible. If there are other dentries that can be
588 * reached through this one we can't delete it and we
589 * return -EBUSY. On success we return 0.
594 int d_invalidate(struct dentry * dentry)
597 * If it's already been dropped, return OK.
599 spin_lock(&dentry->d_lock);
600 if (d_unhashed(dentry)) {
601 spin_unlock(&dentry->d_lock);
605 * Check whether to do a partial shrink_dcache
606 * to get rid of unused child entries.
608 if (!list_empty(&dentry->d_subdirs)) {
609 spin_unlock(&dentry->d_lock);
610 shrink_dcache_parent(dentry);
611 spin_lock(&dentry->d_lock);
615 * Somebody else still using it?
617 * If it's a directory, we can't drop it
618 * for fear of somebody re-populating it
619 * with children (even though dropping it
620 * would make it unreachable from the root,
621 * we might still populate it if it was a
622 * working directory or similar).
623 * We also need to leave mountpoints alone,
626 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
627 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
628 spin_unlock(&dentry->d_lock);
634 spin_unlock(&dentry->d_lock);
637 EXPORT_SYMBOL(d_invalidate);
639 /* This must be called with d_lock held */
640 static inline void __dget_dlock(struct dentry *dentry)
642 dentry->d_lockref.count++;
645 static inline void __dget(struct dentry *dentry)
647 lockref_get(&dentry->d_lockref);
650 struct dentry *dget_parent(struct dentry *dentry)
656 * Do optimistic parent lookup without any
660 ret = ACCESS_ONCE(dentry->d_parent);
661 gotref = lockref_get_not_zero(&ret->d_lockref);
663 if (likely(gotref)) {
664 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
671 * Don't need rcu_dereference because we re-check it was correct under
675 ret = dentry->d_parent;
676 spin_lock(&ret->d_lock);
677 if (unlikely(ret != dentry->d_parent)) {
678 spin_unlock(&ret->d_lock);
683 BUG_ON(!ret->d_lockref.count);
684 ret->d_lockref.count++;
685 spin_unlock(&ret->d_lock);
688 EXPORT_SYMBOL(dget_parent);
691 * d_find_alias - grab a hashed alias of inode
692 * @inode: inode in question
693 * @want_discon: flag, used by d_splice_alias, to request
694 * that only a DISCONNECTED alias be returned.
696 * If inode has a hashed alias, or is a directory and has any alias,
697 * acquire the reference to alias and return it. Otherwise return NULL.
698 * Notice that if inode is a directory there can be only one alias and
699 * it can be unhashed only if it has no children, or if it is the root
702 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
703 * any other hashed alias over that one unless @want_discon is set,
704 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
706 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
708 struct dentry *alias, *discon_alias;
712 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
713 spin_lock(&alias->d_lock);
714 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
715 if (IS_ROOT(alias) &&
716 (alias->d_flags & DCACHE_DISCONNECTED)) {
717 discon_alias = alias;
718 } else if (!want_discon) {
720 spin_unlock(&alias->d_lock);
724 spin_unlock(&alias->d_lock);
727 alias = discon_alias;
728 spin_lock(&alias->d_lock);
729 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
730 if (IS_ROOT(alias) &&
731 (alias->d_flags & DCACHE_DISCONNECTED)) {
733 spin_unlock(&alias->d_lock);
737 spin_unlock(&alias->d_lock);
743 struct dentry *d_find_alias(struct inode *inode)
745 struct dentry *de = NULL;
747 if (!hlist_empty(&inode->i_dentry)) {
748 spin_lock(&inode->i_lock);
749 de = __d_find_alias(inode, 0);
750 spin_unlock(&inode->i_lock);
754 EXPORT_SYMBOL(d_find_alias);
757 * Try to kill dentries associated with this inode.
758 * WARNING: you must own a reference to inode.
760 void d_prune_aliases(struct inode *inode)
762 struct dentry *dentry;
764 spin_lock(&inode->i_lock);
765 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
766 spin_lock(&dentry->d_lock);
767 if (!dentry->d_lockref.count) {
769 * inform the fs via d_prune that this dentry
770 * is about to be unhashed and destroyed.
772 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
774 dentry->d_op->d_prune(dentry);
776 __dget_dlock(dentry);
778 spin_unlock(&dentry->d_lock);
779 spin_unlock(&inode->i_lock);
783 spin_unlock(&dentry->d_lock);
785 spin_unlock(&inode->i_lock);
787 EXPORT_SYMBOL(d_prune_aliases);
790 * Try to throw away a dentry - free the inode, dput the parent.
791 * Requires dentry->d_lock is held, and dentry->d_count == 0.
792 * Releases dentry->d_lock.
794 * This may fail if locks cannot be acquired no problem, just try again.
796 static void try_prune_one_dentry(struct dentry *dentry)
797 __releases(dentry->d_lock)
799 struct dentry *parent;
801 parent = dentry_kill(dentry, 0);
803 * If dentry_kill returns NULL, we have nothing more to do.
804 * if it returns the same dentry, trylocks failed. In either
805 * case, just loop again.
807 * Otherwise, we need to prune ancestors too. This is necessary
808 * to prevent quadratic behavior of shrink_dcache_parent(), but
809 * is also expected to be beneficial in reducing dentry cache
814 if (parent == dentry)
817 /* Prune ancestors. */
820 if (lockref_put_or_lock(&dentry->d_lockref))
822 dentry = dentry_kill(dentry, 1);
826 static void shrink_dentry_list(struct list_head *list)
828 struct dentry *dentry;
832 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
833 if (&dentry->d_lru == list)
835 spin_lock(&dentry->d_lock);
836 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
837 spin_unlock(&dentry->d_lock);
842 * We found an inuse dentry which was not removed from
843 * the LRU because of laziness during lookup. Do not free
844 * it - just keep it off the LRU list.
846 if (dentry->d_lockref.count) {
847 dentry_lru_del(dentry);
848 spin_unlock(&dentry->d_lock);
854 try_prune_one_dentry(dentry);
862 * prune_dcache_sb - shrink the dcache
864 * @count: number of entries to try to free
866 * Attempt to shrink the superblock dcache LRU by @count entries. This is
867 * done when we need more memory an called from the superblock shrinker
870 * This function may fail to free any resources if all the dentries are in
873 void prune_dcache_sb(struct super_block *sb, int count)
875 struct dentry *dentry;
876 LIST_HEAD(referenced);
880 spin_lock(&dcache_lru_lock);
881 while (!list_empty(&sb->s_dentry_lru)) {
882 dentry = list_entry(sb->s_dentry_lru.prev,
883 struct dentry, d_lru);
884 BUG_ON(dentry->d_sb != sb);
886 if (!spin_trylock(&dentry->d_lock)) {
887 spin_unlock(&dcache_lru_lock);
892 if (dentry->d_flags & DCACHE_REFERENCED) {
893 dentry->d_flags &= ~DCACHE_REFERENCED;
894 list_move(&dentry->d_lru, &referenced);
895 spin_unlock(&dentry->d_lock);
897 list_move_tail(&dentry->d_lru, &tmp);
898 dentry->d_flags |= DCACHE_SHRINK_LIST;
899 spin_unlock(&dentry->d_lock);
903 cond_resched_lock(&dcache_lru_lock);
905 if (!list_empty(&referenced))
906 list_splice(&referenced, &sb->s_dentry_lru);
907 spin_unlock(&dcache_lru_lock);
909 shrink_dentry_list(&tmp);
913 * shrink_dcache_sb - shrink dcache for a superblock
916 * Shrink the dcache for the specified super block. This is used to free
917 * the dcache before unmounting a file system.
919 void shrink_dcache_sb(struct super_block *sb)
923 spin_lock(&dcache_lru_lock);
924 while (!list_empty(&sb->s_dentry_lru)) {
925 list_splice_init(&sb->s_dentry_lru, &tmp);
926 spin_unlock(&dcache_lru_lock);
927 shrink_dentry_list(&tmp);
928 spin_lock(&dcache_lru_lock);
930 spin_unlock(&dcache_lru_lock);
932 EXPORT_SYMBOL(shrink_dcache_sb);
935 * destroy a single subtree of dentries for unmount
936 * - see the comments on shrink_dcache_for_umount() for a description of the
939 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
941 struct dentry *parent;
943 BUG_ON(!IS_ROOT(dentry));
946 /* descend to the first leaf in the current subtree */
947 while (!list_empty(&dentry->d_subdirs))
948 dentry = list_entry(dentry->d_subdirs.next,
949 struct dentry, d_u.d_child);
951 /* consume the dentries from this leaf up through its parents
952 * until we find one with children or run out altogether */
957 * inform the fs that this dentry is about to be
958 * unhashed and destroyed.
960 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
962 dentry->d_op->d_prune(dentry);
964 dentry_lru_del(dentry);
967 if (dentry->d_lockref.count != 0) {
969 "BUG: Dentry %p{i=%lx,n=%s}"
971 " [unmount of %s %s]\n",
974 dentry->d_inode->i_ino : 0UL,
976 dentry->d_lockref.count,
977 dentry->d_sb->s_type->name,
982 if (IS_ROOT(dentry)) {
984 list_del(&dentry->d_u.d_child);
986 parent = dentry->d_parent;
987 parent->d_lockref.count--;
988 list_del(&dentry->d_u.d_child);
991 inode = dentry->d_inode;
993 dentry->d_inode = NULL;
994 hlist_del_init(&dentry->d_alias);
995 if (dentry->d_op && dentry->d_op->d_iput)
996 dentry->d_op->d_iput(dentry, inode);
1003 /* finished when we fall off the top of the tree,
1004 * otherwise we ascend to the parent and move to the
1005 * next sibling if there is one */
1009 } while (list_empty(&dentry->d_subdirs));
1011 dentry = list_entry(dentry->d_subdirs.next,
1012 struct dentry, d_u.d_child);
1017 * destroy the dentries attached to a superblock on unmounting
1018 * - we don't need to use dentry->d_lock because:
1019 * - the superblock is detached from all mountings and open files, so the
1020 * dentry trees will not be rearranged by the VFS
1021 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1022 * any dentries belonging to this superblock that it comes across
1023 * - the filesystem itself is no longer permitted to rearrange the dentries
1024 * in this superblock
1026 void shrink_dcache_for_umount(struct super_block *sb)
1028 struct dentry *dentry;
1030 if (down_read_trylock(&sb->s_umount))
1033 dentry = sb->s_root;
1035 dentry->d_lockref.count--;
1036 shrink_dcache_for_umount_subtree(dentry);
1038 while (!hlist_bl_empty(&sb->s_anon)) {
1039 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1040 shrink_dcache_for_umount_subtree(dentry);
1045 * This tries to ascend one level of parenthood, but
1046 * we can race with renaming, so we need to re-check
1047 * the parenthood after dropping the lock and check
1048 * that the sequence number still matches.
1050 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1052 struct dentry *new = old->d_parent;
1055 spin_unlock(&old->d_lock);
1056 spin_lock(&new->d_lock);
1059 * might go back up the wrong parent if we have had a rename
1062 if (new != old->d_parent ||
1063 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1064 (!locked && read_seqretry(&rename_lock, seq))) {
1065 spin_unlock(&new->d_lock);
1073 * enum d_walk_ret - action to talke during tree walk
1074 * @D_WALK_CONTINUE: contrinue walk
1075 * @D_WALK_QUIT: quit walk
1076 * @D_WALK_NORETRY: quit when retry is needed
1077 * @D_WALK_SKIP: skip this dentry and its children
1087 * d_walk - walk the dentry tree
1088 * @parent: start of walk
1089 * @data: data passed to @enter() and @finish()
1090 * @enter: callback when first entering the dentry
1091 * @finish: callback when successfully finished the walk
1093 * The @enter() and @finish() callbacks are called with d_lock held.
1095 static void d_walk(struct dentry *parent, void *data,
1096 enum d_walk_ret (*enter)(void *, struct dentry *),
1097 void (*finish)(void *))
1099 struct dentry *this_parent;
1100 struct list_head *next;
1103 enum d_walk_ret ret;
1106 seq = read_seqbegin(&rename_lock);
1108 this_parent = parent;
1109 spin_lock(&this_parent->d_lock);
1111 ret = enter(data, this_parent);
1113 case D_WALK_CONTINUE:
1118 case D_WALK_NORETRY:
1123 next = this_parent->d_subdirs.next;
1125 while (next != &this_parent->d_subdirs) {
1126 struct list_head *tmp = next;
1127 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1130 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1132 ret = enter(data, dentry);
1134 case D_WALK_CONTINUE:
1137 spin_unlock(&dentry->d_lock);
1139 case D_WALK_NORETRY:
1143 spin_unlock(&dentry->d_lock);
1147 if (!list_empty(&dentry->d_subdirs)) {
1148 spin_unlock(&this_parent->d_lock);
1149 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1150 this_parent = dentry;
1151 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1154 spin_unlock(&dentry->d_lock);
1157 * All done at this level ... ascend and resume the search.
1159 if (this_parent != parent) {
1160 struct dentry *child = this_parent;
1161 this_parent = try_to_ascend(this_parent, locked, seq);
1164 next = child->d_u.d_child.next;
1167 if (!locked && read_seqretry(&rename_lock, seq)) {
1168 spin_unlock(&this_parent->d_lock);
1175 spin_unlock(&this_parent->d_lock);
1177 write_sequnlock(&rename_lock);
1186 write_seqlock(&rename_lock);
1191 * Search for at least 1 mount point in the dentry's subdirs.
1192 * We descend to the next level whenever the d_subdirs
1193 * list is non-empty and continue searching.
1197 * have_submounts - check for mounts over a dentry
1198 * @parent: dentry to check.
1200 * Return true if the parent or its subdirectories contain
1204 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1207 if (d_mountpoint(dentry)) {
1211 return D_WALK_CONTINUE;
1214 int have_submounts(struct dentry *parent)
1218 d_walk(parent, &ret, check_mount, NULL);
1222 EXPORT_SYMBOL(have_submounts);
1225 * Called by mount code to set a mountpoint and check if the mountpoint is
1226 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1227 * subtree can become unreachable).
1229 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1230 * this reason take rename_lock and d_lock on dentry and ancestors.
1232 int d_set_mounted(struct dentry *dentry)
1236 write_seqlock(&rename_lock);
1237 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1238 /* Need exclusion wrt. check_submounts_and_drop() */
1239 spin_lock(&p->d_lock);
1240 if (unlikely(d_unhashed(p))) {
1241 spin_unlock(&p->d_lock);
1244 spin_unlock(&p->d_lock);
1246 spin_lock(&dentry->d_lock);
1247 if (!d_unlinked(dentry)) {
1248 dentry->d_flags |= DCACHE_MOUNTED;
1251 spin_unlock(&dentry->d_lock);
1253 write_sequnlock(&rename_lock);
1258 * Search the dentry child list of the specified parent,
1259 * and move any unused dentries to the end of the unused
1260 * list for prune_dcache(). We descend to the next level
1261 * whenever the d_subdirs list is non-empty and continue
1264 * It returns zero iff there are no unused children,
1265 * otherwise it returns the number of children moved to
1266 * the end of the unused list. This may not be the total
1267 * number of unused children, because select_parent can
1268 * drop the lock and return early due to latency
1272 struct select_data {
1273 struct dentry *start;
1274 struct list_head dispose;
1278 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1280 struct select_data *data = _data;
1281 enum d_walk_ret ret = D_WALK_CONTINUE;
1283 if (data->start == dentry)
1287 * move only zero ref count dentries to the dispose list.
1289 * Those which are presently on the shrink list, being processed
1290 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1291 * loop in shrink_dcache_parent() might not make any progress
1294 if (dentry->d_lockref.count) {
1295 dentry_lru_del(dentry);
1296 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1297 dentry_lru_move_list(dentry, &data->dispose);
1298 dentry->d_flags |= DCACHE_SHRINK_LIST;
1300 ret = D_WALK_NORETRY;
1303 * We can return to the caller if we have found some (this
1304 * ensures forward progress). We'll be coming back to find
1307 if (data->found && need_resched())
1314 * shrink_dcache_parent - prune dcache
1315 * @parent: parent of entries to prune
1317 * Prune the dcache to remove unused children of the parent dentry.
1319 void shrink_dcache_parent(struct dentry *parent)
1322 struct select_data data;
1324 INIT_LIST_HEAD(&data.dispose);
1325 data.start = parent;
1328 d_walk(parent, &data, select_collect, NULL);
1332 shrink_dentry_list(&data.dispose);
1336 EXPORT_SYMBOL(shrink_dcache_parent);
1338 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1340 struct select_data *data = _data;
1342 if (d_mountpoint(dentry)) {
1343 data->found = -EBUSY;
1347 return select_collect(_data, dentry);
1350 static void check_and_drop(void *_data)
1352 struct select_data *data = _data;
1354 if (d_mountpoint(data->start))
1355 data->found = -EBUSY;
1357 __d_drop(data->start);
1361 * check_submounts_and_drop - prune dcache, check for submounts and drop
1363 * All done as a single atomic operation relative to has_unlinked_ancestor().
1364 * Returns 0 if successfully unhashed @parent. If there were submounts then
1367 * @dentry: dentry to prune and drop
1369 int check_submounts_and_drop(struct dentry *dentry)
1373 /* Negative dentries can be dropped without further checks */
1374 if (!dentry->d_inode) {
1380 struct select_data data;
1382 INIT_LIST_HEAD(&data.dispose);
1383 data.start = dentry;
1386 d_walk(dentry, &data, check_and_collect, check_and_drop);
1389 if (!list_empty(&data.dispose))
1390 shrink_dentry_list(&data.dispose);
1401 EXPORT_SYMBOL(check_submounts_and_drop);
1404 * __d_alloc - allocate a dcache entry
1405 * @sb: filesystem it will belong to
1406 * @name: qstr of the name
1408 * Allocates a dentry. It returns %NULL if there is insufficient memory
1409 * available. On a success the dentry is returned. The name passed in is
1410 * copied and the copy passed in may be reused after this call.
1413 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1415 struct dentry *dentry;
1418 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1423 * We guarantee that the inline name is always NUL-terminated.
1424 * This way the memcpy() done by the name switching in rename
1425 * will still always have a NUL at the end, even if we might
1426 * be overwriting an internal NUL character
1428 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1429 if (name->len > DNAME_INLINE_LEN-1) {
1430 dname = kmalloc(name->len + 1, GFP_KERNEL);
1432 kmem_cache_free(dentry_cache, dentry);
1436 dname = dentry->d_iname;
1439 dentry->d_name.len = name->len;
1440 dentry->d_name.hash = name->hash;
1441 memcpy(dname, name->name, name->len);
1442 dname[name->len] = 0;
1444 /* Make sure we always see the terminating NUL character */
1446 dentry->d_name.name = dname;
1448 dentry->d_lockref.count = 1;
1449 dentry->d_flags = 0;
1450 spin_lock_init(&dentry->d_lock);
1451 seqcount_init(&dentry->d_seq);
1452 dentry->d_inode = NULL;
1453 dentry->d_parent = dentry;
1455 dentry->d_op = NULL;
1456 dentry->d_fsdata = NULL;
1457 INIT_HLIST_BL_NODE(&dentry->d_hash);
1458 INIT_LIST_HEAD(&dentry->d_lru);
1459 INIT_LIST_HEAD(&dentry->d_subdirs);
1460 INIT_HLIST_NODE(&dentry->d_alias);
1461 INIT_LIST_HEAD(&dentry->d_u.d_child);
1462 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1464 this_cpu_inc(nr_dentry);
1470 * d_alloc - allocate a dcache entry
1471 * @parent: parent of entry to allocate
1472 * @name: qstr of the name
1474 * Allocates a dentry. It returns %NULL if there is insufficient memory
1475 * available. On a success the dentry is returned. The name passed in is
1476 * copied and the copy passed in may be reused after this call.
1478 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1480 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1484 spin_lock(&parent->d_lock);
1486 * don't need child lock because it is not subject
1487 * to concurrency here
1489 __dget_dlock(parent);
1490 dentry->d_parent = parent;
1491 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1492 spin_unlock(&parent->d_lock);
1496 EXPORT_SYMBOL(d_alloc);
1498 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1500 struct dentry *dentry = __d_alloc(sb, name);
1502 dentry->d_flags |= DCACHE_DISCONNECTED;
1505 EXPORT_SYMBOL(d_alloc_pseudo);
1507 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1512 q.len = strlen(name);
1513 q.hash = full_name_hash(q.name, q.len);
1514 return d_alloc(parent, &q);
1516 EXPORT_SYMBOL(d_alloc_name);
1518 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1520 WARN_ON_ONCE(dentry->d_op);
1521 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1523 DCACHE_OP_REVALIDATE |
1524 DCACHE_OP_WEAK_REVALIDATE |
1525 DCACHE_OP_DELETE ));
1530 dentry->d_flags |= DCACHE_OP_HASH;
1532 dentry->d_flags |= DCACHE_OP_COMPARE;
1533 if (op->d_revalidate)
1534 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1535 if (op->d_weak_revalidate)
1536 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1538 dentry->d_flags |= DCACHE_OP_DELETE;
1540 dentry->d_flags |= DCACHE_OP_PRUNE;
1543 EXPORT_SYMBOL(d_set_d_op);
1545 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1547 spin_lock(&dentry->d_lock);
1549 if (unlikely(IS_AUTOMOUNT(inode)))
1550 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1551 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1553 dentry->d_inode = inode;
1554 dentry_rcuwalk_barrier(dentry);
1555 spin_unlock(&dentry->d_lock);
1556 fsnotify_d_instantiate(dentry, inode);
1560 * d_instantiate - fill in inode information for a dentry
1561 * @entry: dentry to complete
1562 * @inode: inode to attach to this dentry
1564 * Fill in inode information in the entry.
1566 * This turns negative dentries into productive full members
1569 * NOTE! This assumes that the inode count has been incremented
1570 * (or otherwise set) by the caller to indicate that it is now
1571 * in use by the dcache.
1574 void d_instantiate(struct dentry *entry, struct inode * inode)
1576 BUG_ON(!hlist_unhashed(&entry->d_alias));
1578 spin_lock(&inode->i_lock);
1579 __d_instantiate(entry, inode);
1581 spin_unlock(&inode->i_lock);
1582 security_d_instantiate(entry, inode);
1584 EXPORT_SYMBOL(d_instantiate);
1587 * d_instantiate_unique - instantiate a non-aliased dentry
1588 * @entry: dentry to instantiate
1589 * @inode: inode to attach to this dentry
1591 * Fill in inode information in the entry. On success, it returns NULL.
1592 * If an unhashed alias of "entry" already exists, then we return the
1593 * aliased dentry instead and drop one reference to inode.
1595 * Note that in order to avoid conflicts with rename() etc, the caller
1596 * had better be holding the parent directory semaphore.
1598 * This also assumes that the inode count has been incremented
1599 * (or otherwise set) by the caller to indicate that it is now
1600 * in use by the dcache.
1602 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1603 struct inode *inode)
1605 struct dentry *alias;
1606 int len = entry->d_name.len;
1607 const char *name = entry->d_name.name;
1608 unsigned int hash = entry->d_name.hash;
1611 __d_instantiate(entry, NULL);
1615 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1617 * Don't need alias->d_lock here, because aliases with
1618 * d_parent == entry->d_parent are not subject to name or
1619 * parent changes, because the parent inode i_mutex is held.
1621 if (alias->d_name.hash != hash)
1623 if (alias->d_parent != entry->d_parent)
1625 if (alias->d_name.len != len)
1627 if (dentry_cmp(alias, name, len))
1633 __d_instantiate(entry, inode);
1637 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1639 struct dentry *result;
1641 BUG_ON(!hlist_unhashed(&entry->d_alias));
1644 spin_lock(&inode->i_lock);
1645 result = __d_instantiate_unique(entry, inode);
1647 spin_unlock(&inode->i_lock);
1650 security_d_instantiate(entry, inode);
1654 BUG_ON(!d_unhashed(result));
1659 EXPORT_SYMBOL(d_instantiate_unique);
1661 struct dentry *d_make_root(struct inode *root_inode)
1663 struct dentry *res = NULL;
1666 static const struct qstr name = QSTR_INIT("/", 1);
1668 res = __d_alloc(root_inode->i_sb, &name);
1670 d_instantiate(res, root_inode);
1676 EXPORT_SYMBOL(d_make_root);
1678 static struct dentry * __d_find_any_alias(struct inode *inode)
1680 struct dentry *alias;
1682 if (hlist_empty(&inode->i_dentry))
1684 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1690 * d_find_any_alias - find any alias for a given inode
1691 * @inode: inode to find an alias for
1693 * If any aliases exist for the given inode, take and return a
1694 * reference for one of them. If no aliases exist, return %NULL.
1696 struct dentry *d_find_any_alias(struct inode *inode)
1700 spin_lock(&inode->i_lock);
1701 de = __d_find_any_alias(inode);
1702 spin_unlock(&inode->i_lock);
1705 EXPORT_SYMBOL(d_find_any_alias);
1708 * d_obtain_alias - find or allocate a dentry for a given inode
1709 * @inode: inode to allocate the dentry for
1711 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1712 * similar open by handle operations. The returned dentry may be anonymous,
1713 * or may have a full name (if the inode was already in the cache).
1715 * When called on a directory inode, we must ensure that the inode only ever
1716 * has one dentry. If a dentry is found, that is returned instead of
1717 * allocating a new one.
1719 * On successful return, the reference to the inode has been transferred
1720 * to the dentry. In case of an error the reference on the inode is released.
1721 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1722 * be passed in and will be the error will be propagate to the return value,
1723 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1725 struct dentry *d_obtain_alias(struct inode *inode)
1727 static const struct qstr anonstring = QSTR_INIT("/", 1);
1732 return ERR_PTR(-ESTALE);
1734 return ERR_CAST(inode);
1736 res = d_find_any_alias(inode);
1740 tmp = __d_alloc(inode->i_sb, &anonstring);
1742 res = ERR_PTR(-ENOMEM);
1746 spin_lock(&inode->i_lock);
1747 res = __d_find_any_alias(inode);
1749 spin_unlock(&inode->i_lock);
1754 /* attach a disconnected dentry */
1755 spin_lock(&tmp->d_lock);
1756 tmp->d_inode = inode;
1757 tmp->d_flags |= DCACHE_DISCONNECTED;
1758 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1759 hlist_bl_lock(&tmp->d_sb->s_anon);
1760 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1761 hlist_bl_unlock(&tmp->d_sb->s_anon);
1762 spin_unlock(&tmp->d_lock);
1763 spin_unlock(&inode->i_lock);
1764 security_d_instantiate(tmp, inode);
1769 if (res && !IS_ERR(res))
1770 security_d_instantiate(res, inode);
1774 EXPORT_SYMBOL(d_obtain_alias);
1777 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1778 * @inode: the inode which may have a disconnected dentry
1779 * @dentry: a negative dentry which we want to point to the inode.
1781 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1782 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1783 * and return it, else simply d_add the inode to the dentry and return NULL.
1785 * This is needed in the lookup routine of any filesystem that is exportable
1786 * (via knfsd) so that we can build dcache paths to directories effectively.
1788 * If a dentry was found and moved, then it is returned. Otherwise NULL
1789 * is returned. This matches the expected return value of ->lookup.
1791 * Cluster filesystems may call this function with a negative, hashed dentry.
1792 * In that case, we know that the inode will be a regular file, and also this
1793 * will only occur during atomic_open. So we need to check for the dentry
1794 * being already hashed only in the final case.
1796 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1798 struct dentry *new = NULL;
1801 return ERR_CAST(inode);
1803 if (inode && S_ISDIR(inode->i_mode)) {
1804 spin_lock(&inode->i_lock);
1805 new = __d_find_alias(inode, 1);
1807 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1808 spin_unlock(&inode->i_lock);
1809 security_d_instantiate(new, inode);
1810 d_move(new, dentry);
1813 /* already taking inode->i_lock, so d_add() by hand */
1814 __d_instantiate(dentry, inode);
1815 spin_unlock(&inode->i_lock);
1816 security_d_instantiate(dentry, inode);
1820 d_instantiate(dentry, inode);
1821 if (d_unhashed(dentry))
1826 EXPORT_SYMBOL(d_splice_alias);
1829 * d_add_ci - lookup or allocate new dentry with case-exact name
1830 * @inode: the inode case-insensitive lookup has found
1831 * @dentry: the negative dentry that was passed to the parent's lookup func
1832 * @name: the case-exact name to be associated with the returned dentry
1834 * This is to avoid filling the dcache with case-insensitive names to the
1835 * same inode, only the actual correct case is stored in the dcache for
1836 * case-insensitive filesystems.
1838 * For a case-insensitive lookup match and if the the case-exact dentry
1839 * already exists in in the dcache, use it and return it.
1841 * If no entry exists with the exact case name, allocate new dentry with
1842 * the exact case, and return the spliced entry.
1844 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1847 struct dentry *found;
1851 * First check if a dentry matching the name already exists,
1852 * if not go ahead and create it now.
1854 found = d_hash_and_lookup(dentry->d_parent, name);
1855 if (unlikely(IS_ERR(found)))
1858 new = d_alloc(dentry->d_parent, name);
1860 found = ERR_PTR(-ENOMEM);
1864 found = d_splice_alias(inode, new);
1873 * If a matching dentry exists, and it's not negative use it.
1875 * Decrement the reference count to balance the iget() done
1878 if (found->d_inode) {
1879 if (unlikely(found->d_inode != inode)) {
1880 /* This can't happen because bad inodes are unhashed. */
1881 BUG_ON(!is_bad_inode(inode));
1882 BUG_ON(!is_bad_inode(found->d_inode));
1889 * Negative dentry: instantiate it unless the inode is a directory and
1890 * already has a dentry.
1892 new = d_splice_alias(inode, found);
1903 EXPORT_SYMBOL(d_add_ci);
1906 * Do the slow-case of the dentry name compare.
1908 * Unlike the dentry_cmp() function, we need to atomically
1909 * load the name and length information, so that the
1910 * filesystem can rely on them, and can use the 'name' and
1911 * 'len' information without worrying about walking off the
1912 * end of memory etc.
1914 * Thus the read_seqcount_retry() and the "duplicate" info
1915 * in arguments (the low-level filesystem should not look
1916 * at the dentry inode or name contents directly, since
1917 * rename can change them while we're in RCU mode).
1919 enum slow_d_compare {
1925 static noinline enum slow_d_compare slow_dentry_cmp(
1926 const struct dentry *parent,
1927 struct dentry *dentry,
1929 const struct qstr *name)
1931 int tlen = dentry->d_name.len;
1932 const char *tname = dentry->d_name.name;
1934 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1936 return D_COMP_SEQRETRY;
1938 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1939 return D_COMP_NOMATCH;
1944 * __d_lookup_rcu - search for a dentry (racy, store-free)
1945 * @parent: parent dentry
1946 * @name: qstr of name we wish to find
1947 * @seqp: returns d_seq value at the point where the dentry was found
1948 * Returns: dentry, or NULL
1950 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1951 * resolution (store-free path walking) design described in
1952 * Documentation/filesystems/path-lookup.txt.
1954 * This is not to be used outside core vfs.
1956 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1957 * held, and rcu_read_lock held. The returned dentry must not be stored into
1958 * without taking d_lock and checking d_seq sequence count against @seq
1961 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1964 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1965 * the returned dentry, so long as its parent's seqlock is checked after the
1966 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1967 * is formed, giving integrity down the path walk.
1969 * NOTE! The caller *has* to check the resulting dentry against the sequence
1970 * number we've returned before using any of the resulting dentry state!
1972 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1973 const struct qstr *name,
1976 u64 hashlen = name->hash_len;
1977 const unsigned char *str = name->name;
1978 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1979 struct hlist_bl_node *node;
1980 struct dentry *dentry;
1983 * Note: There is significant duplication with __d_lookup_rcu which is
1984 * required to prevent single threaded performance regressions
1985 * especially on architectures where smp_rmb (in seqcounts) are costly.
1986 * Keep the two functions in sync.
1990 * The hash list is protected using RCU.
1992 * Carefully use d_seq when comparing a candidate dentry, to avoid
1993 * races with d_move().
1995 * It is possible that concurrent renames can mess up our list
1996 * walk here and result in missing our dentry, resulting in the
1997 * false-negative result. d_lookup() protects against concurrent
1998 * renames using rename_lock seqlock.
2000 * See Documentation/filesystems/path-lookup.txt for more details.
2002 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2007 * The dentry sequence count protects us from concurrent
2008 * renames, and thus protects parent and name fields.
2010 * The caller must perform a seqcount check in order
2011 * to do anything useful with the returned dentry.
2013 * NOTE! We do a "raw" seqcount_begin here. That means that
2014 * we don't wait for the sequence count to stabilize if it
2015 * is in the middle of a sequence change. If we do the slow
2016 * dentry compare, we will do seqretries until it is stable,
2017 * and if we end up with a successful lookup, we actually
2018 * want to exit RCU lookup anyway.
2020 seq = raw_seqcount_begin(&dentry->d_seq);
2021 if (dentry->d_parent != parent)
2023 if (d_unhashed(dentry))
2026 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2027 if (dentry->d_name.hash != hashlen_hash(hashlen))
2030 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2033 case D_COMP_NOMATCH:
2040 if (dentry->d_name.hash_len != hashlen)
2043 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2050 * d_lookup - search for a dentry
2051 * @parent: parent dentry
2052 * @name: qstr of name we wish to find
2053 * Returns: dentry, or NULL
2055 * d_lookup searches the children of the parent dentry for the name in
2056 * question. If the dentry is found its reference count is incremented and the
2057 * dentry is returned. The caller must use dput to free the entry when it has
2058 * finished using it. %NULL is returned if the dentry does not exist.
2060 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2062 struct dentry *dentry;
2066 seq = read_seqbegin(&rename_lock);
2067 dentry = __d_lookup(parent, name);
2070 } while (read_seqretry(&rename_lock, seq));
2073 EXPORT_SYMBOL(d_lookup);
2076 * __d_lookup - search for a dentry (racy)
2077 * @parent: parent dentry
2078 * @name: qstr of name we wish to find
2079 * Returns: dentry, or NULL
2081 * __d_lookup is like d_lookup, however it may (rarely) return a
2082 * false-negative result due to unrelated rename activity.
2084 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2085 * however it must be used carefully, eg. with a following d_lookup in
2086 * the case of failure.
2088 * __d_lookup callers must be commented.
2090 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2092 unsigned int len = name->len;
2093 unsigned int hash = name->hash;
2094 const unsigned char *str = name->name;
2095 struct hlist_bl_head *b = d_hash(parent, hash);
2096 struct hlist_bl_node *node;
2097 struct dentry *found = NULL;
2098 struct dentry *dentry;
2101 * Note: There is significant duplication with __d_lookup_rcu which is
2102 * required to prevent single threaded performance regressions
2103 * especially on architectures where smp_rmb (in seqcounts) are costly.
2104 * Keep the two functions in sync.
2108 * The hash list is protected using RCU.
2110 * Take d_lock when comparing a candidate dentry, to avoid races
2113 * It is possible that concurrent renames can mess up our list
2114 * walk here and result in missing our dentry, resulting in the
2115 * false-negative result. d_lookup() protects against concurrent
2116 * renames using rename_lock seqlock.
2118 * See Documentation/filesystems/path-lookup.txt for more details.
2122 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2124 if (dentry->d_name.hash != hash)
2127 spin_lock(&dentry->d_lock);
2128 if (dentry->d_parent != parent)
2130 if (d_unhashed(dentry))
2134 * It is safe to compare names since d_move() cannot
2135 * change the qstr (protected by d_lock).
2137 if (parent->d_flags & DCACHE_OP_COMPARE) {
2138 int tlen = dentry->d_name.len;
2139 const char *tname = dentry->d_name.name;
2140 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2143 if (dentry->d_name.len != len)
2145 if (dentry_cmp(dentry, str, len))
2149 dentry->d_lockref.count++;
2151 spin_unlock(&dentry->d_lock);
2154 spin_unlock(&dentry->d_lock);
2162 * d_hash_and_lookup - hash the qstr then search for a dentry
2163 * @dir: Directory to search in
2164 * @name: qstr of name we wish to find
2166 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2168 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2171 * Check for a fs-specific hash function. Note that we must
2172 * calculate the standard hash first, as the d_op->d_hash()
2173 * routine may choose to leave the hash value unchanged.
2175 name->hash = full_name_hash(name->name, name->len);
2176 if (dir->d_flags & DCACHE_OP_HASH) {
2177 int err = dir->d_op->d_hash(dir, name);
2178 if (unlikely(err < 0))
2179 return ERR_PTR(err);
2181 return d_lookup(dir, name);
2183 EXPORT_SYMBOL(d_hash_and_lookup);
2186 * d_validate - verify dentry provided from insecure source (deprecated)
2187 * @dentry: The dentry alleged to be valid child of @dparent
2188 * @dparent: The parent dentry (known to be valid)
2190 * An insecure source has sent us a dentry, here we verify it and dget() it.
2191 * This is used by ncpfs in its readdir implementation.
2192 * Zero is returned in the dentry is invalid.
2194 * This function is slow for big directories, and deprecated, do not use it.
2196 int d_validate(struct dentry *dentry, struct dentry *dparent)
2198 struct dentry *child;
2200 spin_lock(&dparent->d_lock);
2201 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2202 if (dentry == child) {
2203 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2204 __dget_dlock(dentry);
2205 spin_unlock(&dentry->d_lock);
2206 spin_unlock(&dparent->d_lock);
2210 spin_unlock(&dparent->d_lock);
2214 EXPORT_SYMBOL(d_validate);
2217 * When a file is deleted, we have two options:
2218 * - turn this dentry into a negative dentry
2219 * - unhash this dentry and free it.
2221 * Usually, we want to just turn this into
2222 * a negative dentry, but if anybody else is
2223 * currently using the dentry or the inode
2224 * we can't do that and we fall back on removing
2225 * it from the hash queues and waiting for
2226 * it to be deleted later when it has no users
2230 * d_delete - delete a dentry
2231 * @dentry: The dentry to delete
2233 * Turn the dentry into a negative dentry if possible, otherwise
2234 * remove it from the hash queues so it can be deleted later
2237 void d_delete(struct dentry * dentry)
2239 struct inode *inode;
2242 * Are we the only user?
2245 spin_lock(&dentry->d_lock);
2246 inode = dentry->d_inode;
2247 isdir = S_ISDIR(inode->i_mode);
2248 if (dentry->d_lockref.count == 1) {
2249 if (!spin_trylock(&inode->i_lock)) {
2250 spin_unlock(&dentry->d_lock);
2254 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2255 dentry_unlink_inode(dentry);
2256 fsnotify_nameremove(dentry, isdir);
2260 if (!d_unhashed(dentry))
2263 spin_unlock(&dentry->d_lock);
2265 fsnotify_nameremove(dentry, isdir);
2267 EXPORT_SYMBOL(d_delete);
2269 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2271 BUG_ON(!d_unhashed(entry));
2273 entry->d_flags |= DCACHE_RCUACCESS;
2274 hlist_bl_add_head_rcu(&entry->d_hash, b);
2278 static void _d_rehash(struct dentry * entry)
2280 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2284 * d_rehash - add an entry back to the hash
2285 * @entry: dentry to add to the hash
2287 * Adds a dentry to the hash according to its name.
2290 void d_rehash(struct dentry * entry)
2292 spin_lock(&entry->d_lock);
2294 spin_unlock(&entry->d_lock);
2296 EXPORT_SYMBOL(d_rehash);
2299 * dentry_update_name_case - update case insensitive dentry with a new name
2300 * @dentry: dentry to be updated
2303 * Update a case insensitive dentry with new case of name.
2305 * dentry must have been returned by d_lookup with name @name. Old and new
2306 * name lengths must match (ie. no d_compare which allows mismatched name
2309 * Parent inode i_mutex must be held over d_lookup and into this call (to
2310 * keep renames and concurrent inserts, and readdir(2) away).
2312 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2314 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2315 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2317 spin_lock(&dentry->d_lock);
2318 write_seqcount_begin(&dentry->d_seq);
2319 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2320 write_seqcount_end(&dentry->d_seq);
2321 spin_unlock(&dentry->d_lock);
2323 EXPORT_SYMBOL(dentry_update_name_case);
2325 static void switch_names(struct dentry *dentry, struct dentry *target)
2327 if (dname_external(target)) {
2328 if (dname_external(dentry)) {
2330 * Both external: swap the pointers
2332 swap(target->d_name.name, dentry->d_name.name);
2335 * dentry:internal, target:external. Steal target's
2336 * storage and make target internal.
2338 memcpy(target->d_iname, dentry->d_name.name,
2339 dentry->d_name.len + 1);
2340 dentry->d_name.name = target->d_name.name;
2341 target->d_name.name = target->d_iname;
2344 if (dname_external(dentry)) {
2346 * dentry:external, target:internal. Give dentry's
2347 * storage to target and make dentry internal
2349 memcpy(dentry->d_iname, target->d_name.name,
2350 target->d_name.len + 1);
2351 target->d_name.name = dentry->d_name.name;
2352 dentry->d_name.name = dentry->d_iname;
2355 * Both are internal. Just copy target to dentry
2357 memcpy(dentry->d_iname, target->d_name.name,
2358 target->d_name.len + 1);
2359 dentry->d_name.len = target->d_name.len;
2363 swap(dentry->d_name.len, target->d_name.len);
2366 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2369 * XXXX: do we really need to take target->d_lock?
2371 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2372 spin_lock(&target->d_parent->d_lock);
2374 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2375 spin_lock(&dentry->d_parent->d_lock);
2376 spin_lock_nested(&target->d_parent->d_lock,
2377 DENTRY_D_LOCK_NESTED);
2379 spin_lock(&target->d_parent->d_lock);
2380 spin_lock_nested(&dentry->d_parent->d_lock,
2381 DENTRY_D_LOCK_NESTED);
2384 if (target < dentry) {
2385 spin_lock_nested(&target->d_lock, 2);
2386 spin_lock_nested(&dentry->d_lock, 3);
2388 spin_lock_nested(&dentry->d_lock, 2);
2389 spin_lock_nested(&target->d_lock, 3);
2393 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2394 struct dentry *target)
2396 if (target->d_parent != dentry->d_parent)
2397 spin_unlock(&dentry->d_parent->d_lock);
2398 if (target->d_parent != target)
2399 spin_unlock(&target->d_parent->d_lock);
2403 * When switching names, the actual string doesn't strictly have to
2404 * be preserved in the target - because we're dropping the target
2405 * anyway. As such, we can just do a simple memcpy() to copy over
2406 * the new name before we switch.
2408 * Note that we have to be a lot more careful about getting the hash
2409 * switched - we have to switch the hash value properly even if it
2410 * then no longer matches the actual (corrupted) string of the target.
2411 * The hash value has to match the hash queue that the dentry is on..
2414 * __d_move - move a dentry
2415 * @dentry: entry to move
2416 * @target: new dentry
2418 * Update the dcache to reflect the move of a file name. Negative
2419 * dcache entries should not be moved in this way. Caller must hold
2420 * rename_lock, the i_mutex of the source and target directories,
2421 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2423 static void __d_move(struct dentry * dentry, struct dentry * target)
2425 if (!dentry->d_inode)
2426 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2428 BUG_ON(d_ancestor(dentry, target));
2429 BUG_ON(d_ancestor(target, dentry));
2431 dentry_lock_for_move(dentry, target);
2433 write_seqcount_begin(&dentry->d_seq);
2434 write_seqcount_begin(&target->d_seq);
2436 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2439 * Move the dentry to the target hash queue. Don't bother checking
2440 * for the same hash queue because of how unlikely it is.
2443 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2445 /* Unhash the target: dput() will then get rid of it */
2448 list_del(&dentry->d_u.d_child);
2449 list_del(&target->d_u.d_child);
2451 /* Switch the names.. */
2452 switch_names(dentry, target);
2453 swap(dentry->d_name.hash, target->d_name.hash);
2455 /* ... and switch the parents */
2456 if (IS_ROOT(dentry)) {
2457 dentry->d_parent = target->d_parent;
2458 target->d_parent = target;
2459 INIT_LIST_HEAD(&target->d_u.d_child);
2461 swap(dentry->d_parent, target->d_parent);
2463 /* And add them back to the (new) parent lists */
2464 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2467 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2469 write_seqcount_end(&target->d_seq);
2470 write_seqcount_end(&dentry->d_seq);
2472 dentry_unlock_parents_for_move(dentry, target);
2473 spin_unlock(&target->d_lock);
2474 fsnotify_d_move(dentry);
2475 spin_unlock(&dentry->d_lock);
2479 * d_move - move a dentry
2480 * @dentry: entry to move
2481 * @target: new dentry
2483 * Update the dcache to reflect the move of a file name. Negative
2484 * dcache entries should not be moved in this way. See the locking
2485 * requirements for __d_move.
2487 void d_move(struct dentry *dentry, struct dentry *target)
2489 write_seqlock(&rename_lock);
2490 __d_move(dentry, target);
2491 write_sequnlock(&rename_lock);
2493 EXPORT_SYMBOL(d_move);
2496 * d_ancestor - search for an ancestor
2497 * @p1: ancestor dentry
2500 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2501 * an ancestor of p2, else NULL.
2503 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2507 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2508 if (p->d_parent == p1)
2515 * This helper attempts to cope with remotely renamed directories
2517 * It assumes that the caller is already holding
2518 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2520 * Note: If ever the locking in lock_rename() changes, then please
2521 * remember to update this too...
2523 static struct dentry *__d_unalias(struct inode *inode,
2524 struct dentry *dentry, struct dentry *alias)
2526 struct mutex *m1 = NULL, *m2 = NULL;
2527 struct dentry *ret = ERR_PTR(-EBUSY);
2529 /* If alias and dentry share a parent, then no extra locks required */
2530 if (alias->d_parent == dentry->d_parent)
2533 /* See lock_rename() */
2534 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2536 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2537 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2539 m2 = &alias->d_parent->d_inode->i_mutex;
2541 if (likely(!d_mountpoint(alias))) {
2542 __d_move(alias, dentry);
2546 spin_unlock(&inode->i_lock);
2555 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2556 * named dentry in place of the dentry to be replaced.
2557 * returns with anon->d_lock held!
2559 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2561 struct dentry *dparent;
2563 dentry_lock_for_move(anon, dentry);
2565 write_seqcount_begin(&dentry->d_seq);
2566 write_seqcount_begin(&anon->d_seq);
2568 dparent = dentry->d_parent;
2570 switch_names(dentry, anon);
2571 swap(dentry->d_name.hash, anon->d_name.hash);
2573 dentry->d_parent = dentry;
2574 list_del_init(&dentry->d_u.d_child);
2575 anon->d_parent = dparent;
2576 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2578 write_seqcount_end(&dentry->d_seq);
2579 write_seqcount_end(&anon->d_seq);
2581 dentry_unlock_parents_for_move(anon, dentry);
2582 spin_unlock(&dentry->d_lock);
2584 /* anon->d_lock still locked, returns locked */
2585 anon->d_flags &= ~DCACHE_DISCONNECTED;
2589 * d_materialise_unique - introduce an inode into the tree
2590 * @dentry: candidate dentry
2591 * @inode: inode to bind to the dentry, to which aliases may be attached
2593 * Introduces an dentry into the tree, substituting an extant disconnected
2594 * root directory alias in its place if there is one. Caller must hold the
2595 * i_mutex of the parent directory.
2597 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2599 struct dentry *actual;
2601 BUG_ON(!d_unhashed(dentry));
2605 __d_instantiate(dentry, NULL);
2610 spin_lock(&inode->i_lock);
2612 if (S_ISDIR(inode->i_mode)) {
2613 struct dentry *alias;
2615 /* Does an aliased dentry already exist? */
2616 alias = __d_find_alias(inode, 0);
2619 write_seqlock(&rename_lock);
2621 if (d_ancestor(alias, dentry)) {
2622 /* Check for loops */
2623 actual = ERR_PTR(-ELOOP);
2624 spin_unlock(&inode->i_lock);
2625 } else if (IS_ROOT(alias)) {
2626 /* Is this an anonymous mountpoint that we
2627 * could splice into our tree? */
2628 __d_materialise_dentry(dentry, alias);
2629 write_sequnlock(&rename_lock);
2633 /* Nope, but we must(!) avoid directory
2634 * aliasing. This drops inode->i_lock */
2635 actual = __d_unalias(inode, dentry, alias);
2637 write_sequnlock(&rename_lock);
2638 if (IS_ERR(actual)) {
2639 if (PTR_ERR(actual) == -ELOOP)
2640 pr_warn_ratelimited(
2641 "VFS: Lookup of '%s' in %s %s"
2642 " would have caused loop\n",
2643 dentry->d_name.name,
2644 inode->i_sb->s_type->name,
2652 /* Add a unique reference */
2653 actual = __d_instantiate_unique(dentry, inode);
2657 BUG_ON(!d_unhashed(actual));
2659 spin_lock(&actual->d_lock);
2662 spin_unlock(&actual->d_lock);
2663 spin_unlock(&inode->i_lock);
2665 if (actual == dentry) {
2666 security_d_instantiate(dentry, inode);
2673 EXPORT_SYMBOL_GPL(d_materialise_unique);
2675 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2679 return -ENAMETOOLONG;
2681 memcpy(*buffer, str, namelen);
2686 * prepend_name - prepend a pathname in front of current buffer pointer
2687 * buffer: buffer pointer
2688 * buflen: allocated length of the buffer
2689 * name: name string and length qstr structure
2691 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2692 * make sure that either the old or the new name pointer and length are
2693 * fetched. However, there may be mismatch between length and pointer.
2694 * The length cannot be trusted, we need to copy it byte-by-byte until
2695 * the length is reached or a null byte is found. It also prepends "/" at
2696 * the beginning of the name. The sequence number check at the caller will
2697 * retry it again when a d_move() does happen. So any garbage in the buffer
2698 * due to mismatched pointer and length will be discarded.
2700 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2702 const char *dname = ACCESS_ONCE(name->name);
2703 u32 dlen = ACCESS_ONCE(name->len);
2706 if (*buflen < dlen + 1)
2707 return -ENAMETOOLONG;
2708 *buflen -= dlen + 1;
2709 p = *buffer -= dlen + 1;
2721 * prepend_path - Prepend path string to a buffer
2722 * @path: the dentry/vfsmount to report
2723 * @root: root vfsmnt/dentry
2724 * @buffer: pointer to the end of the buffer
2725 * @buflen: pointer to buffer length
2727 * The function tries to write out the pathname without taking any lock other
2728 * than the RCU read lock to make sure that dentries won't go away. It only
2729 * checks the sequence number of the global rename_lock as any change in the
2730 * dentry's d_seq will be preceded by changes in the rename_lock sequence
2731 * number. If the sequence number had been change, it will restart the whole
2732 * pathname back-tracing sequence again. It performs a total of 3 trials of
2733 * lockless back-tracing sequences before falling back to take the
2736 static int prepend_path(const struct path *path,
2737 const struct path *root,
2738 char **buffer, int *buflen)
2740 struct dentry *dentry = path->dentry;
2741 struct vfsmount *vfsmnt = path->mnt;
2742 struct mount *mnt = real_mount(vfsmnt);
2751 read_seqbegin_or_lock(&rename_lock, &seq);
2752 while (dentry != root->dentry || vfsmnt != root->mnt) {
2753 struct dentry * parent;
2755 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2757 if (mnt_has_parent(mnt)) {
2758 dentry = mnt->mnt_mountpoint;
2759 mnt = mnt->mnt_parent;
2764 * Filesystems needing to implement special "root names"
2765 * should do so with ->d_dname()
2767 if (IS_ROOT(dentry) &&
2768 (dentry->d_name.len != 1 ||
2769 dentry->d_name.name[0] != '/')) {
2770 WARN(1, "Root dentry has weird name <%.*s>\n",
2771 (int) dentry->d_name.len,
2772 dentry->d_name.name);
2775 error = is_mounted(vfsmnt) ? 1 : 2;
2778 parent = dentry->d_parent;
2780 error = prepend_name(&bptr, &blen, &dentry->d_name);
2786 if (read_seqretry_or_unlock(&rename_lock, &seq))
2789 if (error >= 0 && bptr == *buffer) {
2791 error = -ENAMETOOLONG;
2801 * __d_path - return the path of a dentry
2802 * @path: the dentry/vfsmount to report
2803 * @root: root vfsmnt/dentry
2804 * @buf: buffer to return value in
2805 * @buflen: buffer length
2807 * Convert a dentry into an ASCII path name.
2809 * Returns a pointer into the buffer or an error code if the
2810 * path was too long.
2812 * "buflen" should be positive.
2814 * If the path is not reachable from the supplied root, return %NULL.
2816 char *__d_path(const struct path *path,
2817 const struct path *root,
2818 char *buf, int buflen)
2820 char *res = buf + buflen;
2823 prepend(&res, &buflen, "\0", 1);
2824 br_read_lock(&vfsmount_lock);
2825 error = prepend_path(path, root, &res, &buflen);
2826 br_read_unlock(&vfsmount_lock);
2829 return ERR_PTR(error);
2835 char *d_absolute_path(const struct path *path,
2836 char *buf, int buflen)
2838 struct path root = {};
2839 char *res = buf + buflen;
2842 prepend(&res, &buflen, "\0", 1);
2843 br_read_lock(&vfsmount_lock);
2844 error = prepend_path(path, &root, &res, &buflen);
2845 br_read_unlock(&vfsmount_lock);
2850 return ERR_PTR(error);
2855 * same as __d_path but appends "(deleted)" for unlinked files.
2857 static int path_with_deleted(const struct path *path,
2858 const struct path *root,
2859 char **buf, int *buflen)
2861 prepend(buf, buflen, "\0", 1);
2862 if (d_unlinked(path->dentry)) {
2863 int error = prepend(buf, buflen, " (deleted)", 10);
2868 return prepend_path(path, root, buf, buflen);
2871 static int prepend_unreachable(char **buffer, int *buflen)
2873 return prepend(buffer, buflen, "(unreachable)", 13);
2877 * d_path - return the path of a dentry
2878 * @path: path to report
2879 * @buf: buffer to return value in
2880 * @buflen: buffer length
2882 * Convert a dentry into an ASCII path name. If the entry has been deleted
2883 * the string " (deleted)" is appended. Note that this is ambiguous.
2885 * Returns a pointer into the buffer or an error code if the path was
2886 * too long. Note: Callers should use the returned pointer, not the passed
2887 * in buffer, to use the name! The implementation often starts at an offset
2888 * into the buffer, and may leave 0 bytes at the start.
2890 * "buflen" should be positive.
2892 char *d_path(const struct path *path, char *buf, int buflen)
2894 char *res = buf + buflen;
2899 * We have various synthetic filesystems that never get mounted. On
2900 * these filesystems dentries are never used for lookup purposes, and
2901 * thus don't need to be hashed. They also don't need a name until a
2902 * user wants to identify the object in /proc/pid/fd/. The little hack
2903 * below allows us to generate a name for these objects on demand:
2905 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2906 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2908 get_fs_root(current->fs, &root);
2909 br_read_lock(&vfsmount_lock);
2910 error = path_with_deleted(path, &root, &res, &buflen);
2911 br_read_unlock(&vfsmount_lock);
2913 res = ERR_PTR(error);
2917 EXPORT_SYMBOL(d_path);
2920 * Helper function for dentry_operations.d_dname() members
2922 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2923 const char *fmt, ...)
2929 va_start(args, fmt);
2930 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2933 if (sz > sizeof(temp) || sz > buflen)
2934 return ERR_PTR(-ENAMETOOLONG);
2936 buffer += buflen - sz;
2937 return memcpy(buffer, temp, sz);
2940 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2942 char *end = buffer + buflen;
2943 /* these dentries are never renamed, so d_lock is not needed */
2944 if (prepend(&end, &buflen, " (deleted)", 11) ||
2945 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
2946 prepend(&end, &buflen, "/", 1))
2947 end = ERR_PTR(-ENAMETOOLONG);
2952 * Write full pathname from the root of the filesystem into the buffer.
2954 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2963 prepend(&end, &len, "\0", 1);
2969 read_seqbegin_or_lock(&rename_lock, &seq);
2970 while (!IS_ROOT(dentry)) {
2971 struct dentry *parent = dentry->d_parent;
2975 error = prepend_name(&end, &len, &dentry->d_name);
2982 if (read_seqretry_or_unlock(&rename_lock, &seq))
2988 return ERR_PTR(-ENAMETOOLONG);
2991 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2993 return __dentry_path(dentry, buf, buflen);
2995 EXPORT_SYMBOL(dentry_path_raw);
2997 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3002 if (d_unlinked(dentry)) {
3004 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3008 retval = __dentry_path(dentry, buf, buflen);
3009 if (!IS_ERR(retval) && p)
3010 *p = '/'; /* restore '/' overriden with '\0' */
3013 return ERR_PTR(-ENAMETOOLONG);
3017 * NOTE! The user-level library version returns a
3018 * character pointer. The kernel system call just
3019 * returns the length of the buffer filled (which
3020 * includes the ending '\0' character), or a negative
3021 * error value. So libc would do something like
3023 * char *getcwd(char * buf, size_t size)
3027 * retval = sys_getcwd(buf, size);
3034 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3037 struct path pwd, root;
3038 char *page = (char *) __get_free_page(GFP_USER);
3043 get_fs_root_and_pwd(current->fs, &root, &pwd);
3046 br_read_lock(&vfsmount_lock);
3047 if (!d_unlinked(pwd.dentry)) {
3049 char *cwd = page + PAGE_SIZE;
3050 int buflen = PAGE_SIZE;
3052 prepend(&cwd, &buflen, "\0", 1);
3053 error = prepend_path(&pwd, &root, &cwd, &buflen);
3054 br_read_unlock(&vfsmount_lock);
3059 /* Unreachable from current root */
3061 error = prepend_unreachable(&cwd, &buflen);
3067 len = PAGE_SIZE + page - cwd;
3070 if (copy_to_user(buf, cwd, len))
3074 br_read_unlock(&vfsmount_lock);
3080 free_page((unsigned long) page);
3085 * Test whether new_dentry is a subdirectory of old_dentry.
3087 * Trivially implemented using the dcache structure
3091 * is_subdir - is new dentry a subdirectory of old_dentry
3092 * @new_dentry: new dentry
3093 * @old_dentry: old dentry
3095 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3096 * Returns 0 otherwise.
3097 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3100 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3105 if (new_dentry == old_dentry)
3109 /* for restarting inner loop in case of seq retry */
3110 seq = read_seqbegin(&rename_lock);
3112 * Need rcu_readlock to protect against the d_parent trashing
3116 if (d_ancestor(old_dentry, new_dentry))
3121 } while (read_seqretry(&rename_lock, seq));
3126 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3128 struct dentry *root = data;
3129 if (dentry != root) {
3130 if (d_unhashed(dentry) || !dentry->d_inode)
3133 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3134 dentry->d_flags |= DCACHE_GENOCIDE;
3135 dentry->d_lockref.count--;
3138 return D_WALK_CONTINUE;
3141 void d_genocide(struct dentry *parent)
3143 d_walk(parent, parent, d_genocide_kill, NULL);
3146 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3148 inode_dec_link_count(inode);
3149 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3150 !hlist_unhashed(&dentry->d_alias) ||
3151 !d_unlinked(dentry));
3152 spin_lock(&dentry->d_parent->d_lock);
3153 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3154 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3155 (unsigned long long)inode->i_ino);
3156 spin_unlock(&dentry->d_lock);
3157 spin_unlock(&dentry->d_parent->d_lock);
3158 d_instantiate(dentry, inode);
3160 EXPORT_SYMBOL(d_tmpfile);
3162 static __initdata unsigned long dhash_entries;
3163 static int __init set_dhash_entries(char *str)
3167 dhash_entries = simple_strtoul(str, &str, 0);
3170 __setup("dhash_entries=", set_dhash_entries);
3172 static void __init dcache_init_early(void)
3176 /* If hashes are distributed across NUMA nodes, defer
3177 * hash allocation until vmalloc space is available.
3183 alloc_large_system_hash("Dentry cache",
3184 sizeof(struct hlist_bl_head),
3193 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3194 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3197 static void __init dcache_init(void)
3202 * A constructor could be added for stable state like the lists,
3203 * but it is probably not worth it because of the cache nature
3206 dentry_cache = KMEM_CACHE(dentry,
3207 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3209 /* Hash may have been set up in dcache_init_early */
3214 alloc_large_system_hash("Dentry cache",
3215 sizeof(struct hlist_bl_head),
3224 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3225 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3228 /* SLAB cache for __getname() consumers */
3229 struct kmem_cache *names_cachep __read_mostly;
3230 EXPORT_SYMBOL(names_cachep);
3232 EXPORT_SYMBOL(d_genocide);
3234 void __init vfs_caches_init_early(void)
3236 dcache_init_early();
3240 void __init vfs_caches_init(unsigned long mempages)
3242 unsigned long reserve;
3244 /* Base hash sizes on available memory, with a reserve equal to
3245 150% of current kernel size */
3247 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3248 mempages -= reserve;
3250 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3251 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3255 files_init(mempages);