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.
99 #define D_HASHBITS d_hash_shift
100 #define D_HASHMASK d_hash_mask
102 static unsigned int d_hash_mask __read_mostly;
103 static unsigned int d_hash_shift __read_mostly;
105 static struct hlist_bl_head *dentry_hashtable __read_mostly;
107 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 hash = hash + (hash >> D_HASHBITS);
112 return dentry_hashtable + (hash & D_HASHMASK);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat = {
120 static DEFINE_PER_CPU(unsigned int, nr_dentry);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
123 static int get_nr_dentry(void)
127 for_each_possible_cpu(i)
128 sum += per_cpu(nr_dentry, i);
129 return sum < 0 ? 0 : sum;
132 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
133 size_t *lenp, loff_t *ppos)
135 dentry_stat.nr_dentry = get_nr_dentry();
136 return proc_dointvec(table, write, buffer, lenp, ppos);
141 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142 * The strings are both count bytes long, and count is non-zero.
144 #ifdef CONFIG_DCACHE_WORD_ACCESS
146 #include <asm/word-at-a-time.h>
148 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149 * aligned allocation for this particular component. We don't
150 * strictly need the load_unaligned_zeropad() safety, but it
151 * doesn't hurt either.
153 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154 * need the careful unaligned handling.
156 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
158 unsigned long a,b,mask;
161 a = *(unsigned long *)cs;
162 b = load_unaligned_zeropad(ct);
163 if (tcount < sizeof(unsigned long))
165 if (unlikely(a != b))
167 cs += sizeof(unsigned long);
168 ct += sizeof(unsigned long);
169 tcount -= sizeof(unsigned long);
173 mask = ~(~0ul << tcount*8);
174 return unlikely(!!((a ^ b) & mask));
179 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
193 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
195 const unsigned char *cs;
197 * Be careful about RCU walk racing with rename:
198 * use ACCESS_ONCE to fetch the name pointer.
200 * NOTE! Even if a rename will mean that the length
201 * was not loaded atomically, we don't care. The
202 * RCU walk will check the sequence count eventually,
203 * and catch it. And we won't overrun the buffer,
204 * because we're reading the name pointer atomically,
205 * and a dentry name is guaranteed to be properly
206 * terminated with a NUL byte.
208 * End result: even if 'len' is wrong, we'll exit
209 * early because the data cannot match (there can
210 * be no NUL in the ct/tcount data)
212 cs = ACCESS_ONCE(dentry->d_name.name);
213 smp_read_barrier_depends();
214 return dentry_string_cmp(cs, ct, tcount);
217 static void __d_free(struct rcu_head *head)
219 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
221 WARN_ON(!list_empty(&dentry->d_alias));
222 if (dname_external(dentry))
223 kfree(dentry->d_name.name);
224 kmem_cache_free(dentry_cache, dentry);
230 static void d_free(struct dentry *dentry)
232 BUG_ON(dentry->d_count);
233 this_cpu_dec(nr_dentry);
234 if (dentry->d_op && dentry->d_op->d_release)
235 dentry->d_op->d_release(dentry);
237 /* if dentry was never visible to RCU, immediate free is OK */
238 if (!(dentry->d_flags & DCACHE_RCUACCESS))
239 __d_free(&dentry->d_u.d_rcu);
241 call_rcu(&dentry->d_u.d_rcu, __d_free);
245 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
246 * @dentry: the target dentry
247 * After this call, in-progress rcu-walk path lookup will fail. This
248 * should be called after unhashing, and after changing d_inode (if
249 * the dentry has not already been unhashed).
251 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
253 assert_spin_locked(&dentry->d_lock);
254 /* Go through a barrier */
255 write_seqcount_barrier(&dentry->d_seq);
259 * Release the dentry's inode, using the filesystem
260 * d_iput() operation if defined. Dentry has no refcount
263 static void dentry_iput(struct dentry * dentry)
264 __releases(dentry->d_lock)
265 __releases(dentry->d_inode->i_lock)
267 struct inode *inode = dentry->d_inode;
269 dentry->d_inode = NULL;
270 list_del_init(&dentry->d_alias);
271 spin_unlock(&dentry->d_lock);
272 spin_unlock(&inode->i_lock);
274 fsnotify_inoderemove(inode);
275 if (dentry->d_op && dentry->d_op->d_iput)
276 dentry->d_op->d_iput(dentry, inode);
280 spin_unlock(&dentry->d_lock);
285 * Release the dentry's inode, using the filesystem
286 * d_iput() operation if defined. dentry remains in-use.
288 static void dentry_unlink_inode(struct dentry * dentry)
289 __releases(dentry->d_lock)
290 __releases(dentry->d_inode->i_lock)
292 struct inode *inode = dentry->d_inode;
293 dentry->d_inode = NULL;
294 list_del_init(&dentry->d_alias);
295 dentry_rcuwalk_barrier(dentry);
296 spin_unlock(&dentry->d_lock);
297 spin_unlock(&inode->i_lock);
299 fsnotify_inoderemove(inode);
300 if (dentry->d_op && dentry->d_op->d_iput)
301 dentry->d_op->d_iput(dentry, inode);
307 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
309 static void dentry_lru_add(struct dentry *dentry)
311 if (list_empty(&dentry->d_lru)) {
312 spin_lock(&dcache_lru_lock);
313 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
314 dentry->d_sb->s_nr_dentry_unused++;
315 dentry_stat.nr_unused++;
316 spin_unlock(&dcache_lru_lock);
320 static void __dentry_lru_del(struct dentry *dentry)
322 list_del_init(&dentry->d_lru);
323 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
324 dentry->d_sb->s_nr_dentry_unused--;
325 dentry_stat.nr_unused--;
329 * Remove a dentry with references from the LRU.
331 static void dentry_lru_del(struct dentry *dentry)
333 if (!list_empty(&dentry->d_lru)) {
334 spin_lock(&dcache_lru_lock);
335 __dentry_lru_del(dentry);
336 spin_unlock(&dcache_lru_lock);
341 * Remove a dentry that is unreferenced and about to be pruned
342 * (unhashed and destroyed) from the LRU, and inform the file system.
343 * This wrapper should be called _prior_ to unhashing a victim dentry.
345 static void dentry_lru_prune(struct dentry *dentry)
347 if (!list_empty(&dentry->d_lru)) {
348 if (dentry->d_flags & DCACHE_OP_PRUNE)
349 dentry->d_op->d_prune(dentry);
351 spin_lock(&dcache_lru_lock);
352 __dentry_lru_del(dentry);
353 spin_unlock(&dcache_lru_lock);
357 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
359 spin_lock(&dcache_lru_lock);
360 if (list_empty(&dentry->d_lru)) {
361 list_add_tail(&dentry->d_lru, list);
362 dentry->d_sb->s_nr_dentry_unused++;
363 dentry_stat.nr_unused++;
365 list_move_tail(&dentry->d_lru, list);
367 spin_unlock(&dcache_lru_lock);
371 * d_kill - kill dentry and return parent
372 * @dentry: dentry to kill
373 * @parent: parent dentry
375 * The dentry must already be unhashed and removed from the LRU.
377 * If this is the root of the dentry tree, return NULL.
379 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
382 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
383 __releases(dentry->d_lock)
384 __releases(parent->d_lock)
385 __releases(dentry->d_inode->i_lock)
387 list_del(&dentry->d_u.d_child);
389 * Inform try_to_ascend() that we are no longer attached to the
392 dentry->d_flags |= DCACHE_DISCONNECTED;
394 spin_unlock(&parent->d_lock);
397 * dentry_iput drops the locks, at which point nobody (except
398 * transient RCU lookups) can reach this dentry.
405 * Unhash a dentry without inserting an RCU walk barrier or checking that
406 * dentry->d_lock is locked. The caller must take care of that, if
409 static void __d_shrink(struct dentry *dentry)
411 if (!d_unhashed(dentry)) {
412 struct hlist_bl_head *b;
413 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
414 b = &dentry->d_sb->s_anon;
416 b = d_hash(dentry->d_parent, dentry->d_name.hash);
419 __hlist_bl_del(&dentry->d_hash);
420 dentry->d_hash.pprev = NULL;
426 * d_drop - drop a dentry
427 * @dentry: dentry to drop
429 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
430 * be found through a VFS lookup any more. Note that this is different from
431 * deleting the dentry - d_delete will try to mark the dentry negative if
432 * possible, giving a successful _negative_ lookup, while d_drop will
433 * just make the cache lookup fail.
435 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
436 * reason (NFS timeouts or autofs deletes).
438 * __d_drop requires dentry->d_lock.
440 void __d_drop(struct dentry *dentry)
442 if (!d_unhashed(dentry)) {
444 dentry_rcuwalk_barrier(dentry);
447 EXPORT_SYMBOL(__d_drop);
449 void d_drop(struct dentry *dentry)
451 spin_lock(&dentry->d_lock);
453 spin_unlock(&dentry->d_lock);
455 EXPORT_SYMBOL(d_drop);
458 * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
459 * @dentry: dentry to drop
461 * This is called when we do a lookup on a placeholder dentry that needed to be
462 * looked up. The dentry should have been hashed in order for it to be found by
463 * the lookup code, but now needs to be unhashed while we do the actual lookup
464 * and clear the DCACHE_NEED_LOOKUP flag.
466 void d_clear_need_lookup(struct dentry *dentry)
468 spin_lock(&dentry->d_lock);
470 dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
471 spin_unlock(&dentry->d_lock);
473 EXPORT_SYMBOL(d_clear_need_lookup);
476 * Finish off a dentry we've decided to kill.
477 * dentry->d_lock must be held, returns with it unlocked.
478 * If ref is non-zero, then decrement the refcount too.
479 * Returns dentry requiring refcount drop, or NULL if we're done.
481 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
482 __releases(dentry->d_lock)
485 struct dentry *parent;
487 inode = dentry->d_inode;
488 if (inode && !spin_trylock(&inode->i_lock)) {
490 spin_unlock(&dentry->d_lock);
492 return dentry; /* try again with same dentry */
497 parent = dentry->d_parent;
498 if (parent && !spin_trylock(&parent->d_lock)) {
500 spin_unlock(&inode->i_lock);
507 * if dentry was on the d_lru list delete it from there.
508 * inform the fs via d_prune that this dentry is about to be
509 * unhashed and destroyed.
511 dentry_lru_prune(dentry);
512 /* if it was on the hash then remove it */
514 return d_kill(dentry, parent);
520 * This is complicated by the fact that we do not want to put
521 * dentries that are no longer on any hash chain on the unused
522 * list: we'd much rather just get rid of them immediately.
524 * However, that implies that we have to traverse the dentry
525 * tree upwards to the parents which might _also_ now be
526 * scheduled for deletion (it may have been only waiting for
527 * its last child to go away).
529 * This tail recursion is done by hand as we don't want to depend
530 * on the compiler to always get this right (gcc generally doesn't).
531 * Real recursion would eat up our stack space.
535 * dput - release a dentry
536 * @dentry: dentry to release
538 * Release a dentry. This will drop the usage count and if appropriate
539 * call the dentry unlink method as well as removing it from the queues and
540 * releasing its resources. If the parent dentries were scheduled for release
541 * they too may now get deleted.
543 void dput(struct dentry *dentry)
549 if (dentry->d_count == 1)
551 spin_lock(&dentry->d_lock);
552 BUG_ON(!dentry->d_count);
553 if (dentry->d_count > 1) {
555 spin_unlock(&dentry->d_lock);
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))
569 * If this dentry needs lookup, don't set the referenced flag so that it
570 * is more likely to be cleaned up by the dcache shrinker in case of
573 if (!d_need_lookup(dentry))
574 dentry->d_flags |= DCACHE_REFERENCED;
575 dentry_lru_add(dentry);
578 spin_unlock(&dentry->d_lock);
582 dentry = dentry_kill(dentry, 1);
589 * d_invalidate - invalidate a dentry
590 * @dentry: dentry to invalidate
592 * Try to invalidate the dentry if it turns out to be
593 * possible. If there are other dentries that can be
594 * reached through this one we can't delete it and we
595 * return -EBUSY. On success we return 0.
600 int d_invalidate(struct dentry * dentry)
603 * If it's already been dropped, return OK.
605 spin_lock(&dentry->d_lock);
606 if (d_unhashed(dentry)) {
607 spin_unlock(&dentry->d_lock);
611 * Check whether to do a partial shrink_dcache
612 * to get rid of unused child entries.
614 if (!list_empty(&dentry->d_subdirs)) {
615 spin_unlock(&dentry->d_lock);
616 shrink_dcache_parent(dentry);
617 spin_lock(&dentry->d_lock);
621 * Somebody else still using it?
623 * If it's a directory, we can't drop it
624 * for fear of somebody re-populating it
625 * with children (even though dropping it
626 * would make it unreachable from the root,
627 * we might still populate it if it was a
628 * working directory or similar).
629 * We also need to leave mountpoints alone,
632 if (dentry->d_count > 1 && dentry->d_inode) {
633 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
634 spin_unlock(&dentry->d_lock);
640 spin_unlock(&dentry->d_lock);
643 EXPORT_SYMBOL(d_invalidate);
645 /* This must be called with d_lock held */
646 static inline void __dget_dlock(struct dentry *dentry)
651 static inline void __dget(struct dentry *dentry)
653 spin_lock(&dentry->d_lock);
654 __dget_dlock(dentry);
655 spin_unlock(&dentry->d_lock);
658 struct dentry *dget_parent(struct dentry *dentry)
664 * Don't need rcu_dereference because we re-check it was correct under
668 ret = dentry->d_parent;
669 spin_lock(&ret->d_lock);
670 if (unlikely(ret != dentry->d_parent)) {
671 spin_unlock(&ret->d_lock);
676 BUG_ON(!ret->d_count);
678 spin_unlock(&ret->d_lock);
681 EXPORT_SYMBOL(dget_parent);
684 * d_find_alias - grab a hashed alias of inode
685 * @inode: inode in question
686 * @want_discon: flag, used by d_splice_alias, to request
687 * that only a DISCONNECTED alias be returned.
689 * If inode has a hashed alias, or is a directory and has any alias,
690 * acquire the reference to alias and return it. Otherwise return NULL.
691 * Notice that if inode is a directory there can be only one alias and
692 * it can be unhashed only if it has no children, or if it is the root
695 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
696 * any other hashed alias over that one unless @want_discon is set,
697 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
699 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
701 struct dentry *alias, *discon_alias;
705 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
706 spin_lock(&alias->d_lock);
707 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
708 if (IS_ROOT(alias) &&
709 (alias->d_flags & DCACHE_DISCONNECTED)) {
710 discon_alias = alias;
711 } else if (!want_discon) {
713 spin_unlock(&alias->d_lock);
717 spin_unlock(&alias->d_lock);
720 alias = discon_alias;
721 spin_lock(&alias->d_lock);
722 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
723 if (IS_ROOT(alias) &&
724 (alias->d_flags & DCACHE_DISCONNECTED)) {
726 spin_unlock(&alias->d_lock);
730 spin_unlock(&alias->d_lock);
736 struct dentry *d_find_alias(struct inode *inode)
738 struct dentry *de = NULL;
740 if (!list_empty(&inode->i_dentry)) {
741 spin_lock(&inode->i_lock);
742 de = __d_find_alias(inode, 0);
743 spin_unlock(&inode->i_lock);
747 EXPORT_SYMBOL(d_find_alias);
750 * Try to kill dentries associated with this inode.
751 * WARNING: you must own a reference to inode.
753 void d_prune_aliases(struct inode *inode)
755 struct dentry *dentry;
757 spin_lock(&inode->i_lock);
758 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
759 spin_lock(&dentry->d_lock);
760 if (!dentry->d_count) {
761 __dget_dlock(dentry);
763 spin_unlock(&dentry->d_lock);
764 spin_unlock(&inode->i_lock);
768 spin_unlock(&dentry->d_lock);
770 spin_unlock(&inode->i_lock);
772 EXPORT_SYMBOL(d_prune_aliases);
775 * Try to throw away a dentry - free the inode, dput the parent.
776 * Requires dentry->d_lock is held, and dentry->d_count == 0.
777 * Releases dentry->d_lock.
779 * This may fail if locks cannot be acquired no problem, just try again.
781 static void try_prune_one_dentry(struct dentry *dentry)
782 __releases(dentry->d_lock)
784 struct dentry *parent;
786 parent = dentry_kill(dentry, 0);
788 * If dentry_kill returns NULL, we have nothing more to do.
789 * if it returns the same dentry, trylocks failed. In either
790 * case, just loop again.
792 * Otherwise, we need to prune ancestors too. This is necessary
793 * to prevent quadratic behavior of shrink_dcache_parent(), but
794 * is also expected to be beneficial in reducing dentry cache
799 if (parent == dentry)
802 /* Prune ancestors. */
805 spin_lock(&dentry->d_lock);
806 if (dentry->d_count > 1) {
808 spin_unlock(&dentry->d_lock);
811 dentry = dentry_kill(dentry, 1);
815 static void shrink_dentry_list(struct list_head *list)
817 struct dentry *dentry;
821 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
822 if (&dentry->d_lru == list)
824 spin_lock(&dentry->d_lock);
825 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
826 spin_unlock(&dentry->d_lock);
831 * We found an inuse dentry which was not removed from
832 * the LRU because of laziness during lookup. Do not free
833 * it - just keep it off the LRU list.
835 if (dentry->d_count) {
836 dentry_lru_del(dentry);
837 spin_unlock(&dentry->d_lock);
843 try_prune_one_dentry(dentry);
851 * prune_dcache_sb - shrink the dcache
853 * @count: number of entries to try to free
855 * Attempt to shrink the superblock dcache LRU by @count entries. This is
856 * done when we need more memory an called from the superblock shrinker
859 * This function may fail to free any resources if all the dentries are in
862 void prune_dcache_sb(struct super_block *sb, int count)
864 struct dentry *dentry;
865 LIST_HEAD(referenced);
869 spin_lock(&dcache_lru_lock);
870 while (!list_empty(&sb->s_dentry_lru)) {
871 dentry = list_entry(sb->s_dentry_lru.prev,
872 struct dentry, d_lru);
873 BUG_ON(dentry->d_sb != sb);
875 if (!spin_trylock(&dentry->d_lock)) {
876 spin_unlock(&dcache_lru_lock);
881 if (dentry->d_flags & DCACHE_REFERENCED) {
882 dentry->d_flags &= ~DCACHE_REFERENCED;
883 list_move(&dentry->d_lru, &referenced);
884 spin_unlock(&dentry->d_lock);
886 list_move_tail(&dentry->d_lru, &tmp);
887 dentry->d_flags |= DCACHE_SHRINK_LIST;
888 spin_unlock(&dentry->d_lock);
892 cond_resched_lock(&dcache_lru_lock);
894 if (!list_empty(&referenced))
895 list_splice(&referenced, &sb->s_dentry_lru);
896 spin_unlock(&dcache_lru_lock);
898 shrink_dentry_list(&tmp);
902 * shrink_dcache_sb - shrink dcache for a superblock
905 * Shrink the dcache for the specified super block. This is used to free
906 * the dcache before unmounting a file system.
908 void shrink_dcache_sb(struct super_block *sb)
912 spin_lock(&dcache_lru_lock);
913 while (!list_empty(&sb->s_dentry_lru)) {
914 list_splice_init(&sb->s_dentry_lru, &tmp);
915 spin_unlock(&dcache_lru_lock);
916 shrink_dentry_list(&tmp);
917 spin_lock(&dcache_lru_lock);
919 spin_unlock(&dcache_lru_lock);
921 EXPORT_SYMBOL(shrink_dcache_sb);
924 * destroy a single subtree of dentries for unmount
925 * - see the comments on shrink_dcache_for_umount() for a description of the
928 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
930 struct dentry *parent;
932 BUG_ON(!IS_ROOT(dentry));
935 /* descend to the first leaf in the current subtree */
936 while (!list_empty(&dentry->d_subdirs))
937 dentry = list_entry(dentry->d_subdirs.next,
938 struct dentry, d_u.d_child);
940 /* consume the dentries from this leaf up through its parents
941 * until we find one with children or run out altogether */
946 * remove the dentry from the lru, and inform
947 * the fs that this dentry is about to be
948 * unhashed and destroyed.
950 dentry_lru_prune(dentry);
953 if (dentry->d_count != 0) {
955 "BUG: Dentry %p{i=%lx,n=%s}"
957 " [unmount of %s %s]\n",
960 dentry->d_inode->i_ino : 0UL,
963 dentry->d_sb->s_type->name,
968 if (IS_ROOT(dentry)) {
970 list_del(&dentry->d_u.d_child);
972 parent = dentry->d_parent;
974 list_del(&dentry->d_u.d_child);
977 inode = dentry->d_inode;
979 dentry->d_inode = NULL;
980 list_del_init(&dentry->d_alias);
981 if (dentry->d_op && dentry->d_op->d_iput)
982 dentry->d_op->d_iput(dentry, inode);
989 /* finished when we fall off the top of the tree,
990 * otherwise we ascend to the parent and move to the
991 * next sibling if there is one */
995 } while (list_empty(&dentry->d_subdirs));
997 dentry = list_entry(dentry->d_subdirs.next,
998 struct dentry, d_u.d_child);
1003 * destroy the dentries attached to a superblock on unmounting
1004 * - we don't need to use dentry->d_lock because:
1005 * - the superblock is detached from all mountings and open files, so the
1006 * dentry trees will not be rearranged by the VFS
1007 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1008 * any dentries belonging to this superblock that it comes across
1009 * - the filesystem itself is no longer permitted to rearrange the dentries
1010 * in this superblock
1012 void shrink_dcache_for_umount(struct super_block *sb)
1014 struct dentry *dentry;
1016 if (down_read_trylock(&sb->s_umount))
1019 dentry = sb->s_root;
1022 shrink_dcache_for_umount_subtree(dentry);
1024 while (!hlist_bl_empty(&sb->s_anon)) {
1025 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1026 shrink_dcache_for_umount_subtree(dentry);
1031 * This tries to ascend one level of parenthood, but
1032 * we can race with renaming, so we need to re-check
1033 * the parenthood after dropping the lock and check
1034 * that the sequence number still matches.
1036 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1038 struct dentry *new = old->d_parent;
1041 spin_unlock(&old->d_lock);
1042 spin_lock(&new->d_lock);
1045 * might go back up the wrong parent if we have had a rename
1048 if (new != old->d_parent ||
1049 (old->d_flags & DCACHE_DISCONNECTED) ||
1050 (!locked && read_seqretry(&rename_lock, seq))) {
1051 spin_unlock(&new->d_lock);
1060 * Search for at least 1 mount point in the dentry's subdirs.
1061 * We descend to the next level whenever the d_subdirs
1062 * list is non-empty and continue searching.
1066 * have_submounts - check for mounts over a dentry
1067 * @parent: dentry to check.
1069 * Return true if the parent or its subdirectories contain
1072 int have_submounts(struct dentry *parent)
1074 struct dentry *this_parent;
1075 struct list_head *next;
1079 seq = read_seqbegin(&rename_lock);
1081 this_parent = parent;
1083 if (d_mountpoint(parent))
1085 spin_lock(&this_parent->d_lock);
1087 next = this_parent->d_subdirs.next;
1089 while (next != &this_parent->d_subdirs) {
1090 struct list_head *tmp = next;
1091 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1094 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1095 /* Have we found a mount point ? */
1096 if (d_mountpoint(dentry)) {
1097 spin_unlock(&dentry->d_lock);
1098 spin_unlock(&this_parent->d_lock);
1101 if (!list_empty(&dentry->d_subdirs)) {
1102 spin_unlock(&this_parent->d_lock);
1103 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1104 this_parent = dentry;
1105 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1108 spin_unlock(&dentry->d_lock);
1111 * All done at this level ... ascend and resume the search.
1113 if (this_parent != parent) {
1114 struct dentry *child = this_parent;
1115 this_parent = try_to_ascend(this_parent, locked, seq);
1118 next = child->d_u.d_child.next;
1121 spin_unlock(&this_parent->d_lock);
1122 if (!locked && read_seqretry(&rename_lock, seq))
1125 write_sequnlock(&rename_lock);
1126 return 0; /* No mount points found in tree */
1128 if (!locked && read_seqretry(&rename_lock, seq))
1131 write_sequnlock(&rename_lock);
1136 write_seqlock(&rename_lock);
1139 EXPORT_SYMBOL(have_submounts);
1142 * Search the dentry child list for the specified parent,
1143 * and move any unused dentries to the end of the unused
1144 * list for prune_dcache(). We descend to the next level
1145 * whenever the d_subdirs list is non-empty and continue
1148 * It returns zero iff there are no unused children,
1149 * otherwise it returns the number of children moved to
1150 * the end of the unused list. This may not be the total
1151 * number of unused children, because select_parent can
1152 * drop the lock and return early due to latency
1155 static int select_parent(struct dentry *parent, struct list_head *dispose)
1157 struct dentry *this_parent;
1158 struct list_head *next;
1163 seq = read_seqbegin(&rename_lock);
1165 this_parent = parent;
1166 spin_lock(&this_parent->d_lock);
1168 next = this_parent->d_subdirs.next;
1170 while (next != &this_parent->d_subdirs) {
1171 struct list_head *tmp = next;
1172 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1175 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1178 * move only zero ref count dentries to the dispose list.
1180 * Those which are presently on the shrink list, being processed
1181 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1182 * loop in shrink_dcache_parent() might not make any progress
1185 if (dentry->d_count) {
1186 dentry_lru_del(dentry);
1187 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1188 dentry_lru_move_list(dentry, dispose);
1189 dentry->d_flags |= DCACHE_SHRINK_LIST;
1193 * We can return to the caller if we have found some (this
1194 * ensures forward progress). We'll be coming back to find
1197 if (found && need_resched()) {
1198 spin_unlock(&dentry->d_lock);
1203 * Descend a level if the d_subdirs list is non-empty.
1205 if (!list_empty(&dentry->d_subdirs)) {
1206 spin_unlock(&this_parent->d_lock);
1207 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1208 this_parent = dentry;
1209 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1213 spin_unlock(&dentry->d_lock);
1216 * All done at this level ... ascend and resume the search.
1218 if (this_parent != parent) {
1219 struct dentry *child = this_parent;
1220 this_parent = try_to_ascend(this_parent, locked, seq);
1223 next = child->d_u.d_child.next;
1227 spin_unlock(&this_parent->d_lock);
1228 if (!locked && read_seqretry(&rename_lock, seq))
1231 write_sequnlock(&rename_lock);
1238 write_seqlock(&rename_lock);
1243 * shrink_dcache_parent - prune dcache
1244 * @parent: parent of entries to prune
1246 * Prune the dcache to remove unused children of the parent dentry.
1248 void shrink_dcache_parent(struct dentry * parent)
1253 while ((found = select_parent(parent, &dispose)) != 0)
1254 shrink_dentry_list(&dispose);
1256 EXPORT_SYMBOL(shrink_dcache_parent);
1259 * __d_alloc - allocate a dcache entry
1260 * @sb: filesystem it will belong to
1261 * @name: qstr of the name
1263 * Allocates a dentry. It returns %NULL if there is insufficient memory
1264 * available. On a success the dentry is returned. The name passed in is
1265 * copied and the copy passed in may be reused after this call.
1268 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1270 struct dentry *dentry;
1273 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1278 * We guarantee that the inline name is always NUL-terminated.
1279 * This way the memcpy() done by the name switching in rename
1280 * will still always have a NUL at the end, even if we might
1281 * be overwriting an internal NUL character
1283 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1284 if (name->len > DNAME_INLINE_LEN-1) {
1285 dname = kmalloc(name->len + 1, GFP_KERNEL);
1287 kmem_cache_free(dentry_cache, dentry);
1291 dname = dentry->d_iname;
1294 dentry->d_name.len = name->len;
1295 dentry->d_name.hash = name->hash;
1296 memcpy(dname, name->name, name->len);
1297 dname[name->len] = 0;
1299 /* Make sure we always see the terminating NUL character */
1301 dentry->d_name.name = dname;
1303 dentry->d_count = 1;
1304 dentry->d_flags = 0;
1305 spin_lock_init(&dentry->d_lock);
1306 seqcount_init(&dentry->d_seq);
1307 dentry->d_inode = NULL;
1308 dentry->d_parent = dentry;
1310 dentry->d_op = NULL;
1311 dentry->d_fsdata = NULL;
1312 INIT_HLIST_BL_NODE(&dentry->d_hash);
1313 INIT_LIST_HEAD(&dentry->d_lru);
1314 INIT_LIST_HEAD(&dentry->d_subdirs);
1315 INIT_LIST_HEAD(&dentry->d_alias);
1316 INIT_LIST_HEAD(&dentry->d_u.d_child);
1317 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1319 this_cpu_inc(nr_dentry);
1325 * d_alloc - allocate a dcache entry
1326 * @parent: parent of entry to allocate
1327 * @name: qstr of the name
1329 * Allocates a dentry. It returns %NULL if there is insufficient memory
1330 * available. On a success the dentry is returned. The name passed in is
1331 * copied and the copy passed in may be reused after this call.
1333 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1335 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1339 spin_lock(&parent->d_lock);
1341 * don't need child lock because it is not subject
1342 * to concurrency here
1344 __dget_dlock(parent);
1345 dentry->d_parent = parent;
1346 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1347 spin_unlock(&parent->d_lock);
1351 EXPORT_SYMBOL(d_alloc);
1353 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1355 struct dentry *dentry = __d_alloc(sb, name);
1357 dentry->d_flags |= DCACHE_DISCONNECTED;
1360 EXPORT_SYMBOL(d_alloc_pseudo);
1362 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1367 q.len = strlen(name);
1368 q.hash = full_name_hash(q.name, q.len);
1369 return d_alloc(parent, &q);
1371 EXPORT_SYMBOL(d_alloc_name);
1373 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1375 WARN_ON_ONCE(dentry->d_op);
1376 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1378 DCACHE_OP_REVALIDATE |
1379 DCACHE_OP_DELETE ));
1384 dentry->d_flags |= DCACHE_OP_HASH;
1386 dentry->d_flags |= DCACHE_OP_COMPARE;
1387 if (op->d_revalidate)
1388 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1390 dentry->d_flags |= DCACHE_OP_DELETE;
1392 dentry->d_flags |= DCACHE_OP_PRUNE;
1395 EXPORT_SYMBOL(d_set_d_op);
1397 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1399 spin_lock(&dentry->d_lock);
1401 if (unlikely(IS_AUTOMOUNT(inode)))
1402 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1403 list_add(&dentry->d_alias, &inode->i_dentry);
1405 dentry->d_inode = inode;
1406 dentry_rcuwalk_barrier(dentry);
1407 spin_unlock(&dentry->d_lock);
1408 fsnotify_d_instantiate(dentry, inode);
1412 * d_instantiate - fill in inode information for a dentry
1413 * @entry: dentry to complete
1414 * @inode: inode to attach to this dentry
1416 * Fill in inode information in the entry.
1418 * This turns negative dentries into productive full members
1421 * NOTE! This assumes that the inode count has been incremented
1422 * (or otherwise set) by the caller to indicate that it is now
1423 * in use by the dcache.
1426 void d_instantiate(struct dentry *entry, struct inode * inode)
1428 BUG_ON(!list_empty(&entry->d_alias));
1430 spin_lock(&inode->i_lock);
1431 __d_instantiate(entry, inode);
1433 spin_unlock(&inode->i_lock);
1434 security_d_instantiate(entry, inode);
1436 EXPORT_SYMBOL(d_instantiate);
1439 * d_instantiate_unique - instantiate a non-aliased dentry
1440 * @entry: dentry to instantiate
1441 * @inode: inode to attach to this dentry
1443 * Fill in inode information in the entry. On success, it returns NULL.
1444 * If an unhashed alias of "entry" already exists, then we return the
1445 * aliased dentry instead and drop one reference to inode.
1447 * Note that in order to avoid conflicts with rename() etc, the caller
1448 * had better be holding the parent directory semaphore.
1450 * This also assumes that the inode count has been incremented
1451 * (or otherwise set) by the caller to indicate that it is now
1452 * in use by the dcache.
1454 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1455 struct inode *inode)
1457 struct dentry *alias;
1458 int len = entry->d_name.len;
1459 const char *name = entry->d_name.name;
1460 unsigned int hash = entry->d_name.hash;
1463 __d_instantiate(entry, NULL);
1467 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1469 * Don't need alias->d_lock here, because aliases with
1470 * d_parent == entry->d_parent are not subject to name or
1471 * parent changes, because the parent inode i_mutex is held.
1473 if (alias->d_name.hash != hash)
1475 if (alias->d_parent != entry->d_parent)
1477 if (alias->d_name.len != len)
1479 if (dentry_cmp(alias, name, len))
1485 __d_instantiate(entry, inode);
1489 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1491 struct dentry *result;
1493 BUG_ON(!list_empty(&entry->d_alias));
1496 spin_lock(&inode->i_lock);
1497 result = __d_instantiate_unique(entry, inode);
1499 spin_unlock(&inode->i_lock);
1502 security_d_instantiate(entry, inode);
1506 BUG_ON(!d_unhashed(result));
1511 EXPORT_SYMBOL(d_instantiate_unique);
1513 struct dentry *d_make_root(struct inode *root_inode)
1515 struct dentry *res = NULL;
1518 static const struct qstr name = QSTR_INIT("/", 1);
1520 res = __d_alloc(root_inode->i_sb, &name);
1522 d_instantiate(res, root_inode);
1528 EXPORT_SYMBOL(d_make_root);
1530 static struct dentry * __d_find_any_alias(struct inode *inode)
1532 struct dentry *alias;
1534 if (list_empty(&inode->i_dentry))
1536 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1542 * d_find_any_alias - find any alias for a given inode
1543 * @inode: inode to find an alias for
1545 * If any aliases exist for the given inode, take and return a
1546 * reference for one of them. If no aliases exist, return %NULL.
1548 struct dentry *d_find_any_alias(struct inode *inode)
1552 spin_lock(&inode->i_lock);
1553 de = __d_find_any_alias(inode);
1554 spin_unlock(&inode->i_lock);
1557 EXPORT_SYMBOL(d_find_any_alias);
1560 * d_obtain_alias - find or allocate a dentry for a given inode
1561 * @inode: inode to allocate the dentry for
1563 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1564 * similar open by handle operations. The returned dentry may be anonymous,
1565 * or may have a full name (if the inode was already in the cache).
1567 * When called on a directory inode, we must ensure that the inode only ever
1568 * has one dentry. If a dentry is found, that is returned instead of
1569 * allocating a new one.
1571 * On successful return, the reference to the inode has been transferred
1572 * to the dentry. In case of an error the reference on the inode is released.
1573 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1574 * be passed in and will be the error will be propagate to the return value,
1575 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1577 struct dentry *d_obtain_alias(struct inode *inode)
1579 static const struct qstr anonstring = { .name = "" };
1584 return ERR_PTR(-ESTALE);
1586 return ERR_CAST(inode);
1588 res = d_find_any_alias(inode);
1592 tmp = __d_alloc(inode->i_sb, &anonstring);
1594 res = ERR_PTR(-ENOMEM);
1598 spin_lock(&inode->i_lock);
1599 res = __d_find_any_alias(inode);
1601 spin_unlock(&inode->i_lock);
1606 /* attach a disconnected dentry */
1607 spin_lock(&tmp->d_lock);
1608 tmp->d_inode = inode;
1609 tmp->d_flags |= DCACHE_DISCONNECTED;
1610 list_add(&tmp->d_alias, &inode->i_dentry);
1611 hlist_bl_lock(&tmp->d_sb->s_anon);
1612 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1613 hlist_bl_unlock(&tmp->d_sb->s_anon);
1614 spin_unlock(&tmp->d_lock);
1615 spin_unlock(&inode->i_lock);
1616 security_d_instantiate(tmp, inode);
1621 if (res && !IS_ERR(res))
1622 security_d_instantiate(res, inode);
1626 EXPORT_SYMBOL(d_obtain_alias);
1629 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1630 * @inode: the inode which may have a disconnected dentry
1631 * @dentry: a negative dentry which we want to point to the inode.
1633 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1634 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1635 * and return it, else simply d_add the inode to the dentry and return NULL.
1637 * This is needed in the lookup routine of any filesystem that is exportable
1638 * (via knfsd) so that we can build dcache paths to directories effectively.
1640 * If a dentry was found and moved, then it is returned. Otherwise NULL
1641 * is returned. This matches the expected return value of ->lookup.
1644 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1646 struct dentry *new = NULL;
1649 return ERR_CAST(inode);
1651 if (inode && S_ISDIR(inode->i_mode)) {
1652 spin_lock(&inode->i_lock);
1653 new = __d_find_any_alias(inode);
1655 spin_unlock(&inode->i_lock);
1656 security_d_instantiate(new, inode);
1657 d_move(new, dentry);
1660 /* already taking inode->i_lock, so d_add() by hand */
1661 __d_instantiate(dentry, inode);
1662 spin_unlock(&inode->i_lock);
1663 security_d_instantiate(dentry, inode);
1667 d_add(dentry, inode);
1670 EXPORT_SYMBOL(d_splice_alias);
1673 * d_add_ci - lookup or allocate new dentry with case-exact name
1674 * @inode: the inode case-insensitive lookup has found
1675 * @dentry: the negative dentry that was passed to the parent's lookup func
1676 * @name: the case-exact name to be associated with the returned dentry
1678 * This is to avoid filling the dcache with case-insensitive names to the
1679 * same inode, only the actual correct case is stored in the dcache for
1680 * case-insensitive filesystems.
1682 * For a case-insensitive lookup match and if the the case-exact dentry
1683 * already exists in in the dcache, use it and return it.
1685 * If no entry exists with the exact case name, allocate new dentry with
1686 * the exact case, and return the spliced entry.
1688 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1692 struct dentry *found;
1696 * First check if a dentry matching the name already exists,
1697 * if not go ahead and create it now.
1699 found = d_hash_and_lookup(dentry->d_parent, name);
1701 new = d_alloc(dentry->d_parent, name);
1707 found = d_splice_alias(inode, new);
1716 * If a matching dentry exists, and it's not negative use it.
1718 * Decrement the reference count to balance the iget() done
1721 if (found->d_inode) {
1722 if (unlikely(found->d_inode != inode)) {
1723 /* This can't happen because bad inodes are unhashed. */
1724 BUG_ON(!is_bad_inode(inode));
1725 BUG_ON(!is_bad_inode(found->d_inode));
1732 * We are going to instantiate this dentry, unhash it and clear the
1733 * lookup flag so we can do that.
1735 if (unlikely(d_need_lookup(found)))
1736 d_clear_need_lookup(found);
1739 * Negative dentry: instantiate it unless the inode is a directory and
1740 * already has a dentry.
1742 new = d_splice_alias(inode, found);
1751 return ERR_PTR(error);
1753 EXPORT_SYMBOL(d_add_ci);
1756 * Do the slow-case of the dentry name compare.
1758 * Unlike the dentry_cmp() function, we need to atomically
1759 * load the name, length and inode information, so that the
1760 * filesystem can rely on them, and can use the 'name' and
1761 * 'len' information without worrying about walking off the
1762 * end of memory etc.
1764 * Thus the read_seqcount_retry() and the "duplicate" info
1765 * in arguments (the low-level filesystem should not look
1766 * at the dentry inode or name contents directly, since
1767 * rename can change them while we're in RCU mode).
1769 enum slow_d_compare {
1775 static noinline enum slow_d_compare slow_dentry_cmp(
1776 const struct dentry *parent,
1777 struct inode *inode,
1778 struct dentry *dentry,
1780 const struct qstr *name)
1782 int tlen = dentry->d_name.len;
1783 const char *tname = dentry->d_name.name;
1784 struct inode *i = dentry->d_inode;
1786 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1788 return D_COMP_SEQRETRY;
1790 if (parent->d_op->d_compare(parent, inode,
1793 return D_COMP_NOMATCH;
1798 * __d_lookup_rcu - search for a dentry (racy, store-free)
1799 * @parent: parent dentry
1800 * @name: qstr of name we wish to find
1801 * @seqp: returns d_seq value at the point where the dentry was found
1802 * @inode: returns dentry->d_inode when the inode was found valid.
1803 * Returns: dentry, or NULL
1805 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1806 * resolution (store-free path walking) design described in
1807 * Documentation/filesystems/path-lookup.txt.
1809 * This is not to be used outside core vfs.
1811 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1812 * held, and rcu_read_lock held. The returned dentry must not be stored into
1813 * without taking d_lock and checking d_seq sequence count against @seq
1816 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1819 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1820 * the returned dentry, so long as its parent's seqlock is checked after the
1821 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1822 * is formed, giving integrity down the path walk.
1824 * NOTE! The caller *has* to check the resulting dentry against the sequence
1825 * number we've returned before using any of the resulting dentry state!
1827 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1828 const struct qstr *name,
1829 unsigned *seqp, struct inode *inode)
1831 u64 hashlen = name->hash_len;
1832 const unsigned char *str = name->name;
1833 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1834 struct hlist_bl_node *node;
1835 struct dentry *dentry;
1838 * Note: There is significant duplication with __d_lookup_rcu which is
1839 * required to prevent single threaded performance regressions
1840 * especially on architectures where smp_rmb (in seqcounts) are costly.
1841 * Keep the two functions in sync.
1845 * The hash list is protected using RCU.
1847 * Carefully use d_seq when comparing a candidate dentry, to avoid
1848 * races with d_move().
1850 * It is possible that concurrent renames can mess up our list
1851 * walk here and result in missing our dentry, resulting in the
1852 * false-negative result. d_lookup() protects against concurrent
1853 * renames using rename_lock seqlock.
1855 * See Documentation/filesystems/path-lookup.txt for more details.
1857 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1862 * The dentry sequence count protects us from concurrent
1863 * renames, and thus protects inode, parent and name fields.
1865 * The caller must perform a seqcount check in order
1866 * to do anything useful with the returned dentry,
1867 * including using the 'd_inode' pointer.
1869 * NOTE! We do a "raw" seqcount_begin here. That means that
1870 * we don't wait for the sequence count to stabilize if it
1871 * is in the middle of a sequence change. If we do the slow
1872 * dentry compare, we will do seqretries until it is stable,
1873 * and if we end up with a successful lookup, we actually
1874 * want to exit RCU lookup anyway.
1876 seq = raw_seqcount_begin(&dentry->d_seq);
1877 if (dentry->d_parent != parent)
1879 if (d_unhashed(dentry))
1883 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1884 if (dentry->d_name.hash != hashlen_hash(hashlen))
1886 switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) {
1889 case D_COMP_NOMATCH:
1896 if (dentry->d_name.hash_len != hashlen)
1898 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
1905 * d_lookup - search for a dentry
1906 * @parent: parent dentry
1907 * @name: qstr of name we wish to find
1908 * Returns: dentry, or NULL
1910 * d_lookup searches the children of the parent dentry for the name in
1911 * question. If the dentry is found its reference count is incremented and the
1912 * dentry is returned. The caller must use dput to free the entry when it has
1913 * finished using it. %NULL is returned if the dentry does not exist.
1915 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1917 struct dentry *dentry;
1921 seq = read_seqbegin(&rename_lock);
1922 dentry = __d_lookup(parent, name);
1925 } while (read_seqretry(&rename_lock, seq));
1928 EXPORT_SYMBOL(d_lookup);
1931 * __d_lookup - search for a dentry (racy)
1932 * @parent: parent dentry
1933 * @name: qstr of name we wish to find
1934 * Returns: dentry, or NULL
1936 * __d_lookup is like d_lookup, however it may (rarely) return a
1937 * false-negative result due to unrelated rename activity.
1939 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1940 * however it must be used carefully, eg. with a following d_lookup in
1941 * the case of failure.
1943 * __d_lookup callers must be commented.
1945 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1947 unsigned int len = name->len;
1948 unsigned int hash = name->hash;
1949 const unsigned char *str = name->name;
1950 struct hlist_bl_head *b = d_hash(parent, hash);
1951 struct hlist_bl_node *node;
1952 struct dentry *found = NULL;
1953 struct dentry *dentry;
1956 * Note: There is significant duplication with __d_lookup_rcu which is
1957 * required to prevent single threaded performance regressions
1958 * especially on architectures where smp_rmb (in seqcounts) are costly.
1959 * Keep the two functions in sync.
1963 * The hash list is protected using RCU.
1965 * Take d_lock when comparing a candidate dentry, to avoid races
1968 * It is possible that concurrent renames can mess up our list
1969 * walk here and result in missing our dentry, resulting in the
1970 * false-negative result. d_lookup() protects against concurrent
1971 * renames using rename_lock seqlock.
1973 * See Documentation/filesystems/path-lookup.txt for more details.
1977 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1979 if (dentry->d_name.hash != hash)
1982 spin_lock(&dentry->d_lock);
1983 if (dentry->d_parent != parent)
1985 if (d_unhashed(dentry))
1989 * It is safe to compare names since d_move() cannot
1990 * change the qstr (protected by d_lock).
1992 if (parent->d_flags & DCACHE_OP_COMPARE) {
1993 int tlen = dentry->d_name.len;
1994 const char *tname = dentry->d_name.name;
1995 if (parent->d_op->d_compare(parent, parent->d_inode,
1996 dentry, dentry->d_inode,
2000 if (dentry->d_name.len != len)
2002 if (dentry_cmp(dentry, str, len))
2008 spin_unlock(&dentry->d_lock);
2011 spin_unlock(&dentry->d_lock);
2019 * d_hash_and_lookup - hash the qstr then search for a dentry
2020 * @dir: Directory to search in
2021 * @name: qstr of name we wish to find
2023 * On hash failure or on lookup failure NULL is returned.
2025 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2027 struct dentry *dentry = NULL;
2030 * Check for a fs-specific hash function. Note that we must
2031 * calculate the standard hash first, as the d_op->d_hash()
2032 * routine may choose to leave the hash value unchanged.
2034 name->hash = full_name_hash(name->name, name->len);
2035 if (dir->d_flags & DCACHE_OP_HASH) {
2036 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
2039 dentry = d_lookup(dir, name);
2045 * d_validate - verify dentry provided from insecure source (deprecated)
2046 * @dentry: The dentry alleged to be valid child of @dparent
2047 * @dparent: The parent dentry (known to be valid)
2049 * An insecure source has sent us a dentry, here we verify it and dget() it.
2050 * This is used by ncpfs in its readdir implementation.
2051 * Zero is returned in the dentry is invalid.
2053 * This function is slow for big directories, and deprecated, do not use it.
2055 int d_validate(struct dentry *dentry, struct dentry *dparent)
2057 struct dentry *child;
2059 spin_lock(&dparent->d_lock);
2060 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2061 if (dentry == child) {
2062 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2063 __dget_dlock(dentry);
2064 spin_unlock(&dentry->d_lock);
2065 spin_unlock(&dparent->d_lock);
2069 spin_unlock(&dparent->d_lock);
2073 EXPORT_SYMBOL(d_validate);
2076 * When a file is deleted, we have two options:
2077 * - turn this dentry into a negative dentry
2078 * - unhash this dentry and free it.
2080 * Usually, we want to just turn this into
2081 * a negative dentry, but if anybody else is
2082 * currently using the dentry or the inode
2083 * we can't do that and we fall back on removing
2084 * it from the hash queues and waiting for
2085 * it to be deleted later when it has no users
2089 * d_delete - delete a dentry
2090 * @dentry: The dentry to delete
2092 * Turn the dentry into a negative dentry if possible, otherwise
2093 * remove it from the hash queues so it can be deleted later
2096 void d_delete(struct dentry * dentry)
2098 struct inode *inode;
2101 * Are we the only user?
2104 spin_lock(&dentry->d_lock);
2105 inode = dentry->d_inode;
2106 isdir = S_ISDIR(inode->i_mode);
2107 if (dentry->d_count == 1) {
2108 if (inode && !spin_trylock(&inode->i_lock)) {
2109 spin_unlock(&dentry->d_lock);
2113 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2114 dentry_unlink_inode(dentry);
2115 fsnotify_nameremove(dentry, isdir);
2119 if (!d_unhashed(dentry))
2122 spin_unlock(&dentry->d_lock);
2124 fsnotify_nameremove(dentry, isdir);
2126 EXPORT_SYMBOL(d_delete);
2128 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2130 BUG_ON(!d_unhashed(entry));
2132 entry->d_flags |= DCACHE_RCUACCESS;
2133 hlist_bl_add_head_rcu(&entry->d_hash, b);
2137 static void _d_rehash(struct dentry * entry)
2139 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2143 * d_rehash - add an entry back to the hash
2144 * @entry: dentry to add to the hash
2146 * Adds a dentry to the hash according to its name.
2149 void d_rehash(struct dentry * entry)
2151 spin_lock(&entry->d_lock);
2153 spin_unlock(&entry->d_lock);
2155 EXPORT_SYMBOL(d_rehash);
2158 * dentry_update_name_case - update case insensitive dentry with a new name
2159 * @dentry: dentry to be updated
2162 * Update a case insensitive dentry with new case of name.
2164 * dentry must have been returned by d_lookup with name @name. Old and new
2165 * name lengths must match (ie. no d_compare which allows mismatched name
2168 * Parent inode i_mutex must be held over d_lookup and into this call (to
2169 * keep renames and concurrent inserts, and readdir(2) away).
2171 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2173 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2174 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2176 spin_lock(&dentry->d_lock);
2177 write_seqcount_begin(&dentry->d_seq);
2178 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2179 write_seqcount_end(&dentry->d_seq);
2180 spin_unlock(&dentry->d_lock);
2182 EXPORT_SYMBOL(dentry_update_name_case);
2184 static void switch_names(struct dentry *dentry, struct dentry *target)
2186 if (dname_external(target)) {
2187 if (dname_external(dentry)) {
2189 * Both external: swap the pointers
2191 swap(target->d_name.name, dentry->d_name.name);
2194 * dentry:internal, target:external. Steal target's
2195 * storage and make target internal.
2197 memcpy(target->d_iname, dentry->d_name.name,
2198 dentry->d_name.len + 1);
2199 dentry->d_name.name = target->d_name.name;
2200 target->d_name.name = target->d_iname;
2203 if (dname_external(dentry)) {
2205 * dentry:external, target:internal. Give dentry's
2206 * storage to target and make dentry internal
2208 memcpy(dentry->d_iname, target->d_name.name,
2209 target->d_name.len + 1);
2210 target->d_name.name = dentry->d_name.name;
2211 dentry->d_name.name = dentry->d_iname;
2214 * Both are internal. Just copy target to dentry
2216 memcpy(dentry->d_iname, target->d_name.name,
2217 target->d_name.len + 1);
2218 dentry->d_name.len = target->d_name.len;
2222 swap(dentry->d_name.len, target->d_name.len);
2225 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2228 * XXXX: do we really need to take target->d_lock?
2230 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2231 spin_lock(&target->d_parent->d_lock);
2233 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2234 spin_lock(&dentry->d_parent->d_lock);
2235 spin_lock_nested(&target->d_parent->d_lock,
2236 DENTRY_D_LOCK_NESTED);
2238 spin_lock(&target->d_parent->d_lock);
2239 spin_lock_nested(&dentry->d_parent->d_lock,
2240 DENTRY_D_LOCK_NESTED);
2243 if (target < dentry) {
2244 spin_lock_nested(&target->d_lock, 2);
2245 spin_lock_nested(&dentry->d_lock, 3);
2247 spin_lock_nested(&dentry->d_lock, 2);
2248 spin_lock_nested(&target->d_lock, 3);
2252 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2253 struct dentry *target)
2255 if (target->d_parent != dentry->d_parent)
2256 spin_unlock(&dentry->d_parent->d_lock);
2257 if (target->d_parent != target)
2258 spin_unlock(&target->d_parent->d_lock);
2262 * When switching names, the actual string doesn't strictly have to
2263 * be preserved in the target - because we're dropping the target
2264 * anyway. As such, we can just do a simple memcpy() to copy over
2265 * the new name before we switch.
2267 * Note that we have to be a lot more careful about getting the hash
2268 * switched - we have to switch the hash value properly even if it
2269 * then no longer matches the actual (corrupted) string of the target.
2270 * The hash value has to match the hash queue that the dentry is on..
2273 * __d_move - move a dentry
2274 * @dentry: entry to move
2275 * @target: new dentry
2277 * Update the dcache to reflect the move of a file name. Negative
2278 * dcache entries should not be moved in this way. Caller must hold
2279 * rename_lock, the i_mutex of the source and target directories,
2280 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2282 static void __d_move(struct dentry * dentry, struct dentry * target)
2284 if (!dentry->d_inode)
2285 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2287 BUG_ON(d_ancestor(dentry, target));
2288 BUG_ON(d_ancestor(target, dentry));
2290 dentry_lock_for_move(dentry, target);
2292 write_seqcount_begin(&dentry->d_seq);
2293 write_seqcount_begin(&target->d_seq);
2295 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2298 * Move the dentry to the target hash queue. Don't bother checking
2299 * for the same hash queue because of how unlikely it is.
2302 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2304 /* Unhash the target: dput() will then get rid of it */
2307 list_del(&dentry->d_u.d_child);
2308 list_del(&target->d_u.d_child);
2310 /* Switch the names.. */
2311 switch_names(dentry, target);
2312 swap(dentry->d_name.hash, target->d_name.hash);
2314 /* ... and switch the parents */
2315 if (IS_ROOT(dentry)) {
2316 dentry->d_parent = target->d_parent;
2317 target->d_parent = target;
2318 INIT_LIST_HEAD(&target->d_u.d_child);
2320 swap(dentry->d_parent, target->d_parent);
2322 /* And add them back to the (new) parent lists */
2323 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2326 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2328 write_seqcount_end(&target->d_seq);
2329 write_seqcount_end(&dentry->d_seq);
2331 dentry_unlock_parents_for_move(dentry, target);
2332 spin_unlock(&target->d_lock);
2333 fsnotify_d_move(dentry);
2334 spin_unlock(&dentry->d_lock);
2338 * d_move - move a dentry
2339 * @dentry: entry to move
2340 * @target: new dentry
2342 * Update the dcache to reflect the move of a file name. Negative
2343 * dcache entries should not be moved in this way. See the locking
2344 * requirements for __d_move.
2346 void d_move(struct dentry *dentry, struct dentry *target)
2348 write_seqlock(&rename_lock);
2349 __d_move(dentry, target);
2350 write_sequnlock(&rename_lock);
2352 EXPORT_SYMBOL(d_move);
2355 * d_ancestor - search for an ancestor
2356 * @p1: ancestor dentry
2359 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2360 * an ancestor of p2, else NULL.
2362 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2366 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2367 if (p->d_parent == p1)
2374 * This helper attempts to cope with remotely renamed directories
2376 * It assumes that the caller is already holding
2377 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2379 * Note: If ever the locking in lock_rename() changes, then please
2380 * remember to update this too...
2382 static struct dentry *__d_unalias(struct inode *inode,
2383 struct dentry *dentry, struct dentry *alias)
2385 struct mutex *m1 = NULL, *m2 = NULL;
2388 /* If alias and dentry share a parent, then no extra locks required */
2389 if (alias->d_parent == dentry->d_parent)
2392 /* See lock_rename() */
2393 ret = ERR_PTR(-EBUSY);
2394 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2396 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2397 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2399 m2 = &alias->d_parent->d_inode->i_mutex;
2401 __d_move(alias, dentry);
2404 spin_unlock(&inode->i_lock);
2413 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2414 * named dentry in place of the dentry to be replaced.
2415 * returns with anon->d_lock held!
2417 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2419 struct dentry *dparent, *aparent;
2421 dentry_lock_for_move(anon, dentry);
2423 write_seqcount_begin(&dentry->d_seq);
2424 write_seqcount_begin(&anon->d_seq);
2426 dparent = dentry->d_parent;
2427 aparent = anon->d_parent;
2429 switch_names(dentry, anon);
2430 swap(dentry->d_name.hash, anon->d_name.hash);
2432 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2433 list_del(&dentry->d_u.d_child);
2434 if (!IS_ROOT(dentry))
2435 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2437 INIT_LIST_HEAD(&dentry->d_u.d_child);
2439 anon->d_parent = (dparent == dentry) ? anon : dparent;
2440 list_del(&anon->d_u.d_child);
2442 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2444 INIT_LIST_HEAD(&anon->d_u.d_child);
2446 write_seqcount_end(&dentry->d_seq);
2447 write_seqcount_end(&anon->d_seq);
2449 dentry_unlock_parents_for_move(anon, dentry);
2450 spin_unlock(&dentry->d_lock);
2452 /* anon->d_lock still locked, returns locked */
2453 anon->d_flags &= ~DCACHE_DISCONNECTED;
2457 * d_materialise_unique - introduce an inode into the tree
2458 * @dentry: candidate dentry
2459 * @inode: inode to bind to the dentry, to which aliases may be attached
2461 * Introduces an dentry into the tree, substituting an extant disconnected
2462 * root directory alias in its place if there is one. Caller must hold the
2463 * i_mutex of the parent directory.
2465 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2467 struct dentry *actual;
2469 BUG_ON(!d_unhashed(dentry));
2473 __d_instantiate(dentry, NULL);
2478 spin_lock(&inode->i_lock);
2480 if (S_ISDIR(inode->i_mode)) {
2481 struct dentry *alias;
2483 /* Does an aliased dentry already exist? */
2484 alias = __d_find_alias(inode, 0);
2487 write_seqlock(&rename_lock);
2489 if (d_ancestor(alias, dentry)) {
2490 /* Check for loops */
2491 actual = ERR_PTR(-ELOOP);
2492 spin_unlock(&inode->i_lock);
2493 } else if (IS_ROOT(alias)) {
2494 /* Is this an anonymous mountpoint that we
2495 * could splice into our tree? */
2496 __d_materialise_dentry(dentry, alias);
2497 write_sequnlock(&rename_lock);
2501 /* Nope, but we must(!) avoid directory
2502 * aliasing. This drops inode->i_lock */
2503 actual = __d_unalias(inode, dentry, alias);
2505 write_sequnlock(&rename_lock);
2506 if (IS_ERR(actual)) {
2507 if (PTR_ERR(actual) == -ELOOP)
2508 pr_warn_ratelimited(
2509 "VFS: Lookup of '%s' in %s %s"
2510 " would have caused loop\n",
2511 dentry->d_name.name,
2512 inode->i_sb->s_type->name,
2520 /* Add a unique reference */
2521 actual = __d_instantiate_unique(dentry, inode);
2525 BUG_ON(!d_unhashed(actual));
2527 spin_lock(&actual->d_lock);
2530 spin_unlock(&actual->d_lock);
2531 spin_unlock(&inode->i_lock);
2533 if (actual == dentry) {
2534 security_d_instantiate(dentry, inode);
2541 EXPORT_SYMBOL_GPL(d_materialise_unique);
2543 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2547 return -ENAMETOOLONG;
2549 memcpy(*buffer, str, namelen);
2553 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2555 return prepend(buffer, buflen, name->name, name->len);
2559 * prepend_path - Prepend path string to a buffer
2560 * @path: the dentry/vfsmount to report
2561 * @root: root vfsmnt/dentry
2562 * @buffer: pointer to the end of the buffer
2563 * @buflen: pointer to buffer length
2565 * Caller holds the rename_lock.
2567 static int prepend_path(const struct path *path,
2568 const struct path *root,
2569 char **buffer, int *buflen)
2571 struct dentry *dentry = path->dentry;
2572 struct vfsmount *vfsmnt = path->mnt;
2573 struct mount *mnt = real_mount(vfsmnt);
2577 br_read_lock(&vfsmount_lock);
2578 while (dentry != root->dentry || vfsmnt != root->mnt) {
2579 struct dentry * parent;
2581 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2583 if (!mnt_has_parent(mnt))
2585 dentry = mnt->mnt_mountpoint;
2586 mnt = mnt->mnt_parent;
2590 parent = dentry->d_parent;
2592 spin_lock(&dentry->d_lock);
2593 error = prepend_name(buffer, buflen, &dentry->d_name);
2594 spin_unlock(&dentry->d_lock);
2596 error = prepend(buffer, buflen, "/", 1);
2604 if (!error && !slash)
2605 error = prepend(buffer, buflen, "/", 1);
2608 br_read_unlock(&vfsmount_lock);
2613 * Filesystems needing to implement special "root names"
2614 * should do so with ->d_dname()
2616 if (IS_ROOT(dentry) &&
2617 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2618 WARN(1, "Root dentry has weird name <%.*s>\n",
2619 (int) dentry->d_name.len, dentry->d_name.name);
2622 error = prepend(buffer, buflen, "/", 1);
2624 error = real_mount(vfsmnt)->mnt_ns ? 1 : 2;
2629 * __d_path - return the path of a dentry
2630 * @path: the dentry/vfsmount to report
2631 * @root: root vfsmnt/dentry
2632 * @buf: buffer to return value in
2633 * @buflen: buffer length
2635 * Convert a dentry into an ASCII path name.
2637 * Returns a pointer into the buffer or an error code if the
2638 * path was too long.
2640 * "buflen" should be positive.
2642 * If the path is not reachable from the supplied root, return %NULL.
2644 char *__d_path(const struct path *path,
2645 const struct path *root,
2646 char *buf, int buflen)
2648 char *res = buf + buflen;
2651 prepend(&res, &buflen, "\0", 1);
2652 write_seqlock(&rename_lock);
2653 error = prepend_path(path, root, &res, &buflen);
2654 write_sequnlock(&rename_lock);
2657 return ERR_PTR(error);
2663 char *d_absolute_path(const struct path *path,
2664 char *buf, int buflen)
2666 struct path root = {};
2667 char *res = buf + buflen;
2670 prepend(&res, &buflen, "\0", 1);
2671 write_seqlock(&rename_lock);
2672 error = prepend_path(path, &root, &res, &buflen);
2673 write_sequnlock(&rename_lock);
2678 return ERR_PTR(error);
2683 * same as __d_path but appends "(deleted)" for unlinked files.
2685 static int path_with_deleted(const struct path *path,
2686 const struct path *root,
2687 char **buf, int *buflen)
2689 prepend(buf, buflen, "\0", 1);
2690 if (d_unlinked(path->dentry)) {
2691 int error = prepend(buf, buflen, " (deleted)", 10);
2696 return prepend_path(path, root, buf, buflen);
2699 static int prepend_unreachable(char **buffer, int *buflen)
2701 return prepend(buffer, buflen, "(unreachable)", 13);
2705 * d_path - return the path of a dentry
2706 * @path: path to report
2707 * @buf: buffer to return value in
2708 * @buflen: buffer length
2710 * Convert a dentry into an ASCII path name. If the entry has been deleted
2711 * the string " (deleted)" is appended. Note that this is ambiguous.
2713 * Returns a pointer into the buffer or an error code if the path was
2714 * too long. Note: Callers should use the returned pointer, not the passed
2715 * in buffer, to use the name! The implementation often starts at an offset
2716 * into the buffer, and may leave 0 bytes at the start.
2718 * "buflen" should be positive.
2720 char *d_path(const struct path *path, char *buf, int buflen)
2722 char *res = buf + buflen;
2727 * We have various synthetic filesystems that never get mounted. On
2728 * these filesystems dentries are never used for lookup purposes, and
2729 * thus don't need to be hashed. They also don't need a name until a
2730 * user wants to identify the object in /proc/pid/fd/. The little hack
2731 * below allows us to generate a name for these objects on demand:
2733 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2734 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2736 get_fs_root(current->fs, &root);
2737 write_seqlock(&rename_lock);
2738 error = path_with_deleted(path, &root, &res, &buflen);
2740 res = ERR_PTR(error);
2741 write_sequnlock(&rename_lock);
2745 EXPORT_SYMBOL(d_path);
2748 * d_path_with_unreachable - return the path of a dentry
2749 * @path: path to report
2750 * @buf: buffer to return value in
2751 * @buflen: buffer length
2753 * The difference from d_path() is that this prepends "(unreachable)"
2754 * to paths which are unreachable from the current process' root.
2756 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2758 char *res = buf + buflen;
2762 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2763 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2765 get_fs_root(current->fs, &root);
2766 write_seqlock(&rename_lock);
2767 error = path_with_deleted(path, &root, &res, &buflen);
2769 error = prepend_unreachable(&res, &buflen);
2770 write_sequnlock(&rename_lock);
2773 res = ERR_PTR(error);
2779 * Helper function for dentry_operations.d_dname() members
2781 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2782 const char *fmt, ...)
2788 va_start(args, fmt);
2789 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2792 if (sz > sizeof(temp) || sz > buflen)
2793 return ERR_PTR(-ENAMETOOLONG);
2795 buffer += buflen - sz;
2796 return memcpy(buffer, temp, sz);
2800 * Write full pathname from the root of the filesystem into the buffer.
2802 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2804 char *end = buf + buflen;
2807 prepend(&end, &buflen, "\0", 1);
2814 while (!IS_ROOT(dentry)) {
2815 struct dentry *parent = dentry->d_parent;
2819 spin_lock(&dentry->d_lock);
2820 error = prepend_name(&end, &buflen, &dentry->d_name);
2821 spin_unlock(&dentry->d_lock);
2822 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2830 return ERR_PTR(-ENAMETOOLONG);
2833 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2837 write_seqlock(&rename_lock);
2838 retval = __dentry_path(dentry, buf, buflen);
2839 write_sequnlock(&rename_lock);
2843 EXPORT_SYMBOL(dentry_path_raw);
2845 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2850 write_seqlock(&rename_lock);
2851 if (d_unlinked(dentry)) {
2853 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2857 retval = __dentry_path(dentry, buf, buflen);
2858 write_sequnlock(&rename_lock);
2859 if (!IS_ERR(retval) && p)
2860 *p = '/'; /* restore '/' overriden with '\0' */
2863 return ERR_PTR(-ENAMETOOLONG);
2867 * NOTE! The user-level library version returns a
2868 * character pointer. The kernel system call just
2869 * returns the length of the buffer filled (which
2870 * includes the ending '\0' character), or a negative
2871 * error value. So libc would do something like
2873 * char *getcwd(char * buf, size_t size)
2877 * retval = sys_getcwd(buf, size);
2884 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2887 struct path pwd, root;
2888 char *page = (char *) __get_free_page(GFP_USER);
2893 get_fs_root_and_pwd(current->fs, &root, &pwd);
2896 write_seqlock(&rename_lock);
2897 if (!d_unlinked(pwd.dentry)) {
2899 char *cwd = page + PAGE_SIZE;
2900 int buflen = PAGE_SIZE;
2902 prepend(&cwd, &buflen, "\0", 1);
2903 error = prepend_path(&pwd, &root, &cwd, &buflen);
2904 write_sequnlock(&rename_lock);
2909 /* Unreachable from current root */
2911 error = prepend_unreachable(&cwd, &buflen);
2917 len = PAGE_SIZE + page - cwd;
2920 if (copy_to_user(buf, cwd, len))
2924 write_sequnlock(&rename_lock);
2930 free_page((unsigned long) page);
2935 * Test whether new_dentry is a subdirectory of old_dentry.
2937 * Trivially implemented using the dcache structure
2941 * is_subdir - is new dentry a subdirectory of old_dentry
2942 * @new_dentry: new dentry
2943 * @old_dentry: old dentry
2945 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2946 * Returns 0 otherwise.
2947 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2950 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2955 if (new_dentry == old_dentry)
2959 /* for restarting inner loop in case of seq retry */
2960 seq = read_seqbegin(&rename_lock);
2962 * Need rcu_readlock to protect against the d_parent trashing
2966 if (d_ancestor(old_dentry, new_dentry))
2971 } while (read_seqretry(&rename_lock, seq));
2976 void d_genocide(struct dentry *root)
2978 struct dentry *this_parent;
2979 struct list_head *next;
2983 seq = read_seqbegin(&rename_lock);
2986 spin_lock(&this_parent->d_lock);
2988 next = this_parent->d_subdirs.next;
2990 while (next != &this_parent->d_subdirs) {
2991 struct list_head *tmp = next;
2992 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2995 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2996 if (d_unhashed(dentry) || !dentry->d_inode) {
2997 spin_unlock(&dentry->d_lock);
3000 if (!list_empty(&dentry->d_subdirs)) {
3001 spin_unlock(&this_parent->d_lock);
3002 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
3003 this_parent = dentry;
3004 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
3007 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3008 dentry->d_flags |= DCACHE_GENOCIDE;
3011 spin_unlock(&dentry->d_lock);
3013 if (this_parent != root) {
3014 struct dentry *child = this_parent;
3015 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
3016 this_parent->d_flags |= DCACHE_GENOCIDE;
3017 this_parent->d_count--;
3019 this_parent = try_to_ascend(this_parent, locked, seq);
3022 next = child->d_u.d_child.next;
3025 spin_unlock(&this_parent->d_lock);
3026 if (!locked && read_seqretry(&rename_lock, seq))
3029 write_sequnlock(&rename_lock);
3034 write_seqlock(&rename_lock);
3039 * find_inode_number - check for dentry with name
3040 * @dir: directory to check
3041 * @name: Name to find.
3043 * Check whether a dentry already exists for the given name,
3044 * and return the inode number if it has an inode. Otherwise
3047 * This routine is used to post-process directory listings for
3048 * filesystems using synthetic inode numbers, and is necessary
3049 * to keep getcwd() working.
3052 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
3054 struct dentry * dentry;
3057 dentry = d_hash_and_lookup(dir, name);
3059 if (dentry->d_inode)
3060 ino = dentry->d_inode->i_ino;
3065 EXPORT_SYMBOL(find_inode_number);
3067 static __initdata unsigned long dhash_entries;
3068 static int __init set_dhash_entries(char *str)
3072 dhash_entries = simple_strtoul(str, &str, 0);
3075 __setup("dhash_entries=", set_dhash_entries);
3077 static void __init dcache_init_early(void)
3081 /* If hashes are distributed across NUMA nodes, defer
3082 * hash allocation until vmalloc space is available.
3088 alloc_large_system_hash("Dentry cache",
3089 sizeof(struct hlist_bl_head),
3098 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3099 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3102 static void __init dcache_init(void)
3107 * A constructor could be added for stable state like the lists,
3108 * but it is probably not worth it because of the cache nature
3111 dentry_cache = KMEM_CACHE(dentry,
3112 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3114 /* Hash may have been set up in dcache_init_early */
3119 alloc_large_system_hash("Dentry cache",
3120 sizeof(struct hlist_bl_head),
3129 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3130 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3133 /* SLAB cache for __getname() consumers */
3134 struct kmem_cache *names_cachep __read_mostly;
3135 EXPORT_SYMBOL(names_cachep);
3137 EXPORT_SYMBOL(d_genocide);
3139 void __init vfs_caches_init_early(void)
3141 dcache_init_early();
3145 void __init vfs_caches_init(unsigned long mempages)
3147 unsigned long reserve;
3149 /* Base hash sizes on available memory, with a reserve equal to
3150 150% of current kernel size */
3152 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3153 mempages -= reserve;
3155 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3156 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3160 files_init(mempages);