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 * dentry->d_sb->s_dentry_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
66 * dentry->d_sb->s_dentry_lru_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 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
86 EXPORT_SYMBOL(rename_lock);
88 static struct kmem_cache *dentry_cache __read_mostly;
91 * read_seqbegin_or_lock - begin a sequence number check or locking block
93 * seq : sequence number to be checked
95 * First try it once optimistically without taking the lock. If that fails,
96 * take the lock. The sequence number is also used as a marker for deciding
97 * whether to be a reader (even) or writer (odd).
98 * N.B. seq must be initialized to an even number to begin with.
100 static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
102 if (!(*seq & 1)) /* Even */
103 *seq = read_seqbegin(lock);
108 static inline int need_seqretry(seqlock_t *lock, int seq)
110 return !(seq & 1) && read_seqretry(lock, seq);
113 static inline void done_seqretry(seqlock_t *lock, int seq)
116 write_sequnlock(lock);
120 * This is the single most critical data structure when it comes
121 * to the dcache: the hashtable for lookups. Somebody should try
122 * to make this good - I've just made it work.
124 * This hash-function tries to avoid losing too many bits of hash
125 * information, yet avoid using a prime hash-size or similar.
127 #define D_HASHBITS d_hash_shift
128 #define D_HASHMASK d_hash_mask
130 static unsigned int d_hash_mask __read_mostly;
131 static unsigned int d_hash_shift __read_mostly;
133 static struct hlist_bl_head *dentry_hashtable __read_mostly;
135 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
138 hash += (unsigned long) parent / L1_CACHE_BYTES;
139 hash = hash + (hash >> D_HASHBITS);
140 return dentry_hashtable + (hash & D_HASHMASK);
143 /* Statistics gathering. */
144 struct dentry_stat_t dentry_stat = {
148 static DEFINE_PER_CPU(long, nr_dentry);
149 static DEFINE_PER_CPU(long, nr_dentry_unused);
151 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
154 * Here we resort to our own counters instead of using generic per-cpu counters
155 * for consistency with what the vfs inode code does. We are expected to harvest
156 * better code and performance by having our own specialized counters.
158 * Please note that the loop is done over all possible CPUs, not over all online
159 * CPUs. The reason for this is that we don't want to play games with CPUs going
160 * on and off. If one of them goes off, we will just keep their counters.
162 * glommer: See cffbc8a for details, and if you ever intend to change this,
163 * please update all vfs counters to match.
165 static long get_nr_dentry(void)
169 for_each_possible_cpu(i)
170 sum += per_cpu(nr_dentry, i);
171 return sum < 0 ? 0 : sum;
174 static long get_nr_dentry_unused(void)
178 for_each_possible_cpu(i)
179 sum += per_cpu(nr_dentry_unused, i);
180 return sum < 0 ? 0 : sum;
183 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
184 size_t *lenp, loff_t *ppos)
186 dentry_stat.nr_dentry = get_nr_dentry();
187 dentry_stat.nr_unused = get_nr_dentry_unused();
188 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
193 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
194 * The strings are both count bytes long, and count is non-zero.
196 #ifdef CONFIG_DCACHE_WORD_ACCESS
198 #include <asm/word-at-a-time.h>
200 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
201 * aligned allocation for this particular component. We don't
202 * strictly need the load_unaligned_zeropad() safety, but it
203 * doesn't hurt either.
205 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
206 * need the careful unaligned handling.
208 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
210 unsigned long a,b,mask;
213 a = *(unsigned long *)cs;
214 b = load_unaligned_zeropad(ct);
215 if (tcount < sizeof(unsigned long))
217 if (unlikely(a != b))
219 cs += sizeof(unsigned long);
220 ct += sizeof(unsigned long);
221 tcount -= sizeof(unsigned long);
225 mask = ~(~0ul << tcount*8);
226 return unlikely(!!((a ^ b) & mask));
231 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
245 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
247 const unsigned char *cs;
249 * Be careful about RCU walk racing with rename:
250 * use ACCESS_ONCE to fetch the name pointer.
252 * NOTE! Even if a rename will mean that the length
253 * was not loaded atomically, we don't care. The
254 * RCU walk will check the sequence count eventually,
255 * and catch it. And we won't overrun the buffer,
256 * because we're reading the name pointer atomically,
257 * and a dentry name is guaranteed to be properly
258 * terminated with a NUL byte.
260 * End result: even if 'len' is wrong, we'll exit
261 * early because the data cannot match (there can
262 * be no NUL in the ct/tcount data)
264 cs = ACCESS_ONCE(dentry->d_name.name);
265 smp_read_barrier_depends();
266 return dentry_string_cmp(cs, ct, tcount);
269 static void __d_free(struct rcu_head *head)
271 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
273 WARN_ON(!hlist_unhashed(&dentry->d_alias));
274 if (dname_external(dentry))
275 kfree(dentry->d_name.name);
276 kmem_cache_free(dentry_cache, dentry);
282 static void d_free(struct dentry *dentry)
284 BUG_ON((int)dentry->d_lockref.count > 0);
285 this_cpu_dec(nr_dentry);
286 if (dentry->d_op && dentry->d_op->d_release)
287 dentry->d_op->d_release(dentry);
289 /* if dentry was never visible to RCU, immediate free is OK */
290 if (!(dentry->d_flags & DCACHE_RCUACCESS))
291 __d_free(&dentry->d_u.d_rcu);
293 call_rcu(&dentry->d_u.d_rcu, __d_free);
297 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
298 * @dentry: the target dentry
299 * After this call, in-progress rcu-walk path lookup will fail. This
300 * should be called after unhashing, and after changing d_inode (if
301 * the dentry has not already been unhashed).
303 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
305 assert_spin_locked(&dentry->d_lock);
306 /* Go through a barrier */
307 write_seqcount_barrier(&dentry->d_seq);
311 * Release the dentry's inode, using the filesystem
312 * d_iput() operation if defined. Dentry has no refcount
315 static void dentry_iput(struct dentry * dentry)
316 __releases(dentry->d_lock)
317 __releases(dentry->d_inode->i_lock)
319 struct inode *inode = dentry->d_inode;
321 dentry->d_inode = NULL;
322 hlist_del_init(&dentry->d_alias);
323 spin_unlock(&dentry->d_lock);
324 spin_unlock(&inode->i_lock);
326 fsnotify_inoderemove(inode);
327 if (dentry->d_op && dentry->d_op->d_iput)
328 dentry->d_op->d_iput(dentry, inode);
332 spin_unlock(&dentry->d_lock);
337 * Release the dentry's inode, using the filesystem
338 * d_iput() operation if defined. dentry remains in-use.
340 static void dentry_unlink_inode(struct dentry * dentry)
341 __releases(dentry->d_lock)
342 __releases(dentry->d_inode->i_lock)
344 struct inode *inode = dentry->d_inode;
345 dentry->d_inode = NULL;
346 hlist_del_init(&dentry->d_alias);
347 dentry_rcuwalk_barrier(dentry);
348 spin_unlock(&dentry->d_lock);
349 spin_unlock(&inode->i_lock);
351 fsnotify_inoderemove(inode);
352 if (dentry->d_op && dentry->d_op->d_iput)
353 dentry->d_op->d_iput(dentry, inode);
359 * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
361 static void dentry_lru_add(struct dentry *dentry)
363 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) {
364 spin_lock(&dentry->d_sb->s_dentry_lru_lock);
365 dentry->d_flags |= DCACHE_LRU_LIST;
366 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
367 dentry->d_sb->s_nr_dentry_unused++;
368 this_cpu_inc(nr_dentry_unused);
369 spin_unlock(&dentry->d_sb->s_dentry_lru_lock);
373 static void __dentry_lru_del(struct dentry *dentry)
375 list_del_init(&dentry->d_lru);
376 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
377 dentry->d_sb->s_nr_dentry_unused--;
378 this_cpu_dec(nr_dentry_unused);
382 * Remove a dentry with references from the LRU.
384 static void dentry_lru_del(struct dentry *dentry)
386 if (!list_empty(&dentry->d_lru)) {
387 spin_lock(&dentry->d_sb->s_dentry_lru_lock);
388 __dentry_lru_del(dentry);
389 spin_unlock(&dentry->d_sb->s_dentry_lru_lock);
393 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
395 spin_lock(&dentry->d_sb->s_dentry_lru_lock);
396 if (list_empty(&dentry->d_lru)) {
397 dentry->d_flags |= DCACHE_LRU_LIST;
398 list_add_tail(&dentry->d_lru, list);
399 dentry->d_sb->s_nr_dentry_unused++;
400 this_cpu_inc(nr_dentry_unused);
402 list_move_tail(&dentry->d_lru, list);
404 spin_unlock(&dentry->d_sb->s_dentry_lru_lock);
408 * d_kill - kill dentry and return parent
409 * @dentry: dentry to kill
410 * @parent: parent dentry
412 * The dentry must already be unhashed and removed from the LRU.
414 * If this is the root of the dentry tree, return NULL.
416 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
419 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
420 __releases(dentry->d_lock)
421 __releases(parent->d_lock)
422 __releases(dentry->d_inode->i_lock)
424 list_del(&dentry->d_u.d_child);
426 * Inform try_to_ascend() that we are no longer attached to the
429 dentry->d_flags |= DCACHE_DENTRY_KILLED;
431 spin_unlock(&parent->d_lock);
434 * dentry_iput drops the locks, at which point nobody (except
435 * transient RCU lookups) can reach this dentry.
442 * Unhash a dentry without inserting an RCU walk barrier or checking that
443 * dentry->d_lock is locked. The caller must take care of that, if
446 static void __d_shrink(struct dentry *dentry)
448 if (!d_unhashed(dentry)) {
449 struct hlist_bl_head *b;
450 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
451 b = &dentry->d_sb->s_anon;
453 b = d_hash(dentry->d_parent, dentry->d_name.hash);
456 __hlist_bl_del(&dentry->d_hash);
457 dentry->d_hash.pprev = NULL;
463 * d_drop - drop a dentry
464 * @dentry: dentry to drop
466 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
467 * be found through a VFS lookup any more. Note that this is different from
468 * deleting the dentry - d_delete will try to mark the dentry negative if
469 * possible, giving a successful _negative_ lookup, while d_drop will
470 * just make the cache lookup fail.
472 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
473 * reason (NFS timeouts or autofs deletes).
475 * __d_drop requires dentry->d_lock.
477 void __d_drop(struct dentry *dentry)
479 if (!d_unhashed(dentry)) {
481 dentry_rcuwalk_barrier(dentry);
484 EXPORT_SYMBOL(__d_drop);
486 void d_drop(struct dentry *dentry)
488 spin_lock(&dentry->d_lock);
490 spin_unlock(&dentry->d_lock);
492 EXPORT_SYMBOL(d_drop);
495 * Finish off a dentry we've decided to kill.
496 * dentry->d_lock must be held, returns with it unlocked.
497 * If ref is non-zero, then decrement the refcount too.
498 * Returns dentry requiring refcount drop, or NULL if we're done.
500 static inline struct dentry *dentry_kill(struct dentry *dentry)
501 __releases(dentry->d_lock)
504 struct dentry *parent;
506 inode = dentry->d_inode;
507 if (inode && !spin_trylock(&inode->i_lock)) {
509 spin_unlock(&dentry->d_lock);
511 return dentry; /* try again with same dentry */
516 parent = dentry->d_parent;
517 if (parent && !spin_trylock(&parent->d_lock)) {
519 spin_unlock(&inode->i_lock);
524 * The dentry is now unrecoverably dead to the world.
526 lockref_mark_dead(&dentry->d_lockref);
529 * inform the fs via d_prune that this dentry is about to be
530 * unhashed and destroyed.
532 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
533 dentry->d_op->d_prune(dentry);
535 dentry_lru_del(dentry);
536 /* if it was on the hash then remove it */
538 return d_kill(dentry, parent);
544 * This is complicated by the fact that we do not want to put
545 * dentries that are no longer on any hash chain on the unused
546 * list: we'd much rather just get rid of them immediately.
548 * However, that implies that we have to traverse the dentry
549 * tree upwards to the parents which might _also_ now be
550 * scheduled for deletion (it may have been only waiting for
551 * its last child to go away).
553 * This tail recursion is done by hand as we don't want to depend
554 * on the compiler to always get this right (gcc generally doesn't).
555 * Real recursion would eat up our stack space.
559 * dput - release a dentry
560 * @dentry: dentry to release
562 * Release a dentry. This will drop the usage count and if appropriate
563 * call the dentry unlink method as well as removing it from the queues and
564 * releasing its resources. If the parent dentries were scheduled for release
565 * they too may now get deleted.
567 void dput(struct dentry *dentry)
569 if (unlikely(!dentry))
573 if (lockref_put_or_lock(&dentry->d_lockref))
576 /* Unreachable? Get rid of it */
577 if (unlikely(d_unhashed(dentry)))
580 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
581 if (dentry->d_op->d_delete(dentry))
585 dentry->d_flags |= DCACHE_REFERENCED;
586 dentry_lru_add(dentry);
588 dentry->d_lockref.count--;
589 spin_unlock(&dentry->d_lock);
593 dentry = dentry_kill(dentry);
600 * d_invalidate - invalidate a dentry
601 * @dentry: dentry to invalidate
603 * Try to invalidate the dentry if it turns out to be
604 * possible. If there are other dentries that can be
605 * reached through this one we can't delete it and we
606 * return -EBUSY. On success we return 0.
611 int d_invalidate(struct dentry * dentry)
614 * If it's already been dropped, return OK.
616 spin_lock(&dentry->d_lock);
617 if (d_unhashed(dentry)) {
618 spin_unlock(&dentry->d_lock);
622 * Check whether to do a partial shrink_dcache
623 * to get rid of unused child entries.
625 if (!list_empty(&dentry->d_subdirs)) {
626 spin_unlock(&dentry->d_lock);
627 shrink_dcache_parent(dentry);
628 spin_lock(&dentry->d_lock);
632 * Somebody else still using it?
634 * If it's a directory, we can't drop it
635 * for fear of somebody re-populating it
636 * with children (even though dropping it
637 * would make it unreachable from the root,
638 * we might still populate it if it was a
639 * working directory or similar).
640 * We also need to leave mountpoints alone,
643 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
644 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
645 spin_unlock(&dentry->d_lock);
651 spin_unlock(&dentry->d_lock);
654 EXPORT_SYMBOL(d_invalidate);
656 /* This must be called with d_lock held */
657 static inline void __dget_dlock(struct dentry *dentry)
659 dentry->d_lockref.count++;
662 static inline void __dget(struct dentry *dentry)
664 lockref_get(&dentry->d_lockref);
667 struct dentry *dget_parent(struct dentry *dentry)
673 * Do optimistic parent lookup without any
677 ret = ACCESS_ONCE(dentry->d_parent);
678 gotref = lockref_get_not_zero(&ret->d_lockref);
680 if (likely(gotref)) {
681 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
688 * Don't need rcu_dereference because we re-check it was correct under
692 ret = dentry->d_parent;
693 spin_lock(&ret->d_lock);
694 if (unlikely(ret != dentry->d_parent)) {
695 spin_unlock(&ret->d_lock);
700 BUG_ON(!ret->d_lockref.count);
701 ret->d_lockref.count++;
702 spin_unlock(&ret->d_lock);
705 EXPORT_SYMBOL(dget_parent);
708 * d_find_alias - grab a hashed alias of inode
709 * @inode: inode in question
710 * @want_discon: flag, used by d_splice_alias, to request
711 * that only a DISCONNECTED alias be returned.
713 * If inode has a hashed alias, or is a directory and has any alias,
714 * acquire the reference to alias and return it. Otherwise return NULL.
715 * Notice that if inode is a directory there can be only one alias and
716 * it can be unhashed only if it has no children, or if it is the root
719 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
720 * any other hashed alias over that one unless @want_discon is set,
721 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
723 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
725 struct dentry *alias, *discon_alias;
729 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
730 spin_lock(&alias->d_lock);
731 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
732 if (IS_ROOT(alias) &&
733 (alias->d_flags & DCACHE_DISCONNECTED)) {
734 discon_alias = alias;
735 } else if (!want_discon) {
737 spin_unlock(&alias->d_lock);
741 spin_unlock(&alias->d_lock);
744 alias = discon_alias;
745 spin_lock(&alias->d_lock);
746 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
747 if (IS_ROOT(alias) &&
748 (alias->d_flags & DCACHE_DISCONNECTED)) {
750 spin_unlock(&alias->d_lock);
754 spin_unlock(&alias->d_lock);
760 struct dentry *d_find_alias(struct inode *inode)
762 struct dentry *de = NULL;
764 if (!hlist_empty(&inode->i_dentry)) {
765 spin_lock(&inode->i_lock);
766 de = __d_find_alias(inode, 0);
767 spin_unlock(&inode->i_lock);
771 EXPORT_SYMBOL(d_find_alias);
774 * Try to kill dentries associated with this inode.
775 * WARNING: you must own a reference to inode.
777 void d_prune_aliases(struct inode *inode)
779 struct dentry *dentry;
781 spin_lock(&inode->i_lock);
782 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
783 spin_lock(&dentry->d_lock);
784 if (!dentry->d_lockref.count) {
786 * inform the fs via d_prune that this dentry
787 * is about to be unhashed and destroyed.
789 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
791 dentry->d_op->d_prune(dentry);
793 __dget_dlock(dentry);
795 spin_unlock(&dentry->d_lock);
796 spin_unlock(&inode->i_lock);
800 spin_unlock(&dentry->d_lock);
802 spin_unlock(&inode->i_lock);
804 EXPORT_SYMBOL(d_prune_aliases);
807 * Try to throw away a dentry - free the inode, dput the parent.
808 * Requires dentry->d_lock is held, and dentry->d_count == 0.
809 * Releases dentry->d_lock.
811 * This may fail if locks cannot be acquired no problem, just try again.
813 static void try_prune_one_dentry(struct dentry *dentry)
814 __releases(dentry->d_lock)
816 struct dentry *parent;
818 parent = dentry_kill(dentry);
820 * If dentry_kill returns NULL, we have nothing more to do.
821 * if it returns the same dentry, trylocks failed. In either
822 * case, just loop again.
824 * Otherwise, we need to prune ancestors too. This is necessary
825 * to prevent quadratic behavior of shrink_dcache_parent(), but
826 * is also expected to be beneficial in reducing dentry cache
831 if (parent == dentry)
834 /* Prune ancestors. */
837 if (lockref_put_or_lock(&dentry->d_lockref))
839 dentry = dentry_kill(dentry);
843 static void shrink_dentry_list(struct list_head *list)
845 struct dentry *dentry;
849 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
850 if (&dentry->d_lru == list)
852 spin_lock(&dentry->d_lock);
853 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
854 spin_unlock(&dentry->d_lock);
859 * We found an inuse dentry which was not removed from
860 * the LRU because of laziness during lookup. Do not free
861 * it - just keep it off the LRU list.
863 if (dentry->d_lockref.count) {
864 dentry_lru_del(dentry);
865 spin_unlock(&dentry->d_lock);
871 try_prune_one_dentry(dentry);
879 * prune_dcache_sb - shrink the dcache
881 * @count: number of entries to try to free
883 * Attempt to shrink the superblock dcache LRU by @count entries. This is
884 * done when we need more memory an called from the superblock shrinker
887 * This function may fail to free any resources if all the dentries are in
890 void prune_dcache_sb(struct super_block *sb, int count)
892 struct dentry *dentry;
893 LIST_HEAD(referenced);
897 spin_lock(&sb->s_dentry_lru_lock);
898 while (!list_empty(&sb->s_dentry_lru)) {
899 dentry = list_entry(sb->s_dentry_lru.prev,
900 struct dentry, d_lru);
901 BUG_ON(dentry->d_sb != sb);
903 if (!spin_trylock(&dentry->d_lock)) {
904 spin_unlock(&sb->s_dentry_lru_lock);
909 if (dentry->d_flags & DCACHE_REFERENCED) {
910 dentry->d_flags &= ~DCACHE_REFERENCED;
911 list_move(&dentry->d_lru, &referenced);
912 spin_unlock(&dentry->d_lock);
914 list_move_tail(&dentry->d_lru, &tmp);
915 dentry->d_flags |= DCACHE_SHRINK_LIST;
916 spin_unlock(&dentry->d_lock);
920 cond_resched_lock(&sb->s_dentry_lru_lock);
922 if (!list_empty(&referenced))
923 list_splice(&referenced, &sb->s_dentry_lru);
924 spin_unlock(&sb->s_dentry_lru_lock);
926 shrink_dentry_list(&tmp);
930 * shrink_dcache_sb - shrink dcache for a superblock
933 * Shrink the dcache for the specified super block. This is used to free
934 * the dcache before unmounting a file system.
936 void shrink_dcache_sb(struct super_block *sb)
940 spin_lock(&sb->s_dentry_lru_lock);
941 while (!list_empty(&sb->s_dentry_lru)) {
942 list_splice_init(&sb->s_dentry_lru, &tmp);
943 spin_unlock(&sb->s_dentry_lru_lock);
944 shrink_dentry_list(&tmp);
945 spin_lock(&sb->s_dentry_lru_lock);
947 spin_unlock(&sb->s_dentry_lru_lock);
949 EXPORT_SYMBOL(shrink_dcache_sb);
952 * destroy a single subtree of dentries for unmount
953 * - see the comments on shrink_dcache_for_umount() for a description of the
956 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
958 struct dentry *parent;
960 BUG_ON(!IS_ROOT(dentry));
963 /* descend to the first leaf in the current subtree */
964 while (!list_empty(&dentry->d_subdirs))
965 dentry = list_entry(dentry->d_subdirs.next,
966 struct dentry, d_u.d_child);
968 /* consume the dentries from this leaf up through its parents
969 * until we find one with children or run out altogether */
974 * inform the fs that this dentry is about to be
975 * unhashed and destroyed.
977 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
979 dentry->d_op->d_prune(dentry);
981 dentry_lru_del(dentry);
984 if (dentry->d_lockref.count != 0) {
986 "BUG: Dentry %p{i=%lx,n=%s}"
988 " [unmount of %s %s]\n",
991 dentry->d_inode->i_ino : 0UL,
993 dentry->d_lockref.count,
994 dentry->d_sb->s_type->name,
999 if (IS_ROOT(dentry)) {
1001 list_del(&dentry->d_u.d_child);
1003 parent = dentry->d_parent;
1004 parent->d_lockref.count--;
1005 list_del(&dentry->d_u.d_child);
1008 inode = dentry->d_inode;
1010 dentry->d_inode = NULL;
1011 hlist_del_init(&dentry->d_alias);
1012 if (dentry->d_op && dentry->d_op->d_iput)
1013 dentry->d_op->d_iput(dentry, inode);
1020 /* finished when we fall off the top of the tree,
1021 * otherwise we ascend to the parent and move to the
1022 * next sibling if there is one */
1026 } while (list_empty(&dentry->d_subdirs));
1028 dentry = list_entry(dentry->d_subdirs.next,
1029 struct dentry, d_u.d_child);
1034 * destroy the dentries attached to a superblock on unmounting
1035 * - we don't need to use dentry->d_lock because:
1036 * - the superblock is detached from all mountings and open files, so the
1037 * dentry trees will not be rearranged by the VFS
1038 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1039 * any dentries belonging to this superblock that it comes across
1040 * - the filesystem itself is no longer permitted to rearrange the dentries
1041 * in this superblock
1043 void shrink_dcache_for_umount(struct super_block *sb)
1045 struct dentry *dentry;
1047 if (down_read_trylock(&sb->s_umount))
1050 dentry = sb->s_root;
1052 dentry->d_lockref.count--;
1053 shrink_dcache_for_umount_subtree(dentry);
1055 while (!hlist_bl_empty(&sb->s_anon)) {
1056 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1057 shrink_dcache_for_umount_subtree(dentry);
1062 * This tries to ascend one level of parenthood, but
1063 * we can race with renaming, so we need to re-check
1064 * the parenthood after dropping the lock and check
1065 * that the sequence number still matches.
1067 static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
1069 struct dentry *new = old->d_parent;
1072 spin_unlock(&old->d_lock);
1073 spin_lock(&new->d_lock);
1076 * might go back up the wrong parent if we have had a rename
1079 if (new != old->d_parent ||
1080 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1081 need_seqretry(&rename_lock, seq)) {
1082 spin_unlock(&new->d_lock);
1090 * enum d_walk_ret - action to talke during tree walk
1091 * @D_WALK_CONTINUE: contrinue walk
1092 * @D_WALK_QUIT: quit walk
1093 * @D_WALK_NORETRY: quit when retry is needed
1094 * @D_WALK_SKIP: skip this dentry and its children
1104 * d_walk - walk the dentry tree
1105 * @parent: start of walk
1106 * @data: data passed to @enter() and @finish()
1107 * @enter: callback when first entering the dentry
1108 * @finish: callback when successfully finished the walk
1110 * The @enter() and @finish() callbacks are called with d_lock held.
1112 static void d_walk(struct dentry *parent, void *data,
1113 enum d_walk_ret (*enter)(void *, struct dentry *),
1114 void (*finish)(void *))
1116 struct dentry *this_parent;
1117 struct list_head *next;
1119 enum d_walk_ret ret;
1123 read_seqbegin_or_lock(&rename_lock, &seq);
1124 this_parent = parent;
1125 spin_lock(&this_parent->d_lock);
1127 ret = enter(data, this_parent);
1129 case D_WALK_CONTINUE:
1134 case D_WALK_NORETRY:
1139 next = this_parent->d_subdirs.next;
1141 while (next != &this_parent->d_subdirs) {
1142 struct list_head *tmp = next;
1143 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1146 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1148 ret = enter(data, dentry);
1150 case D_WALK_CONTINUE:
1153 spin_unlock(&dentry->d_lock);
1155 case D_WALK_NORETRY:
1159 spin_unlock(&dentry->d_lock);
1163 if (!list_empty(&dentry->d_subdirs)) {
1164 spin_unlock(&this_parent->d_lock);
1165 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1166 this_parent = dentry;
1167 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1170 spin_unlock(&dentry->d_lock);
1173 * All done at this level ... ascend and resume the search.
1175 if (this_parent != parent) {
1176 struct dentry *child = this_parent;
1177 this_parent = try_to_ascend(this_parent, seq);
1180 next = child->d_u.d_child.next;
1183 if (need_seqretry(&rename_lock, seq)) {
1184 spin_unlock(&this_parent->d_lock);
1191 spin_unlock(&this_parent->d_lock);
1192 done_seqretry(&rename_lock, seq);
1203 * Search for at least 1 mount point in the dentry's subdirs.
1204 * We descend to the next level whenever the d_subdirs
1205 * list is non-empty and continue searching.
1209 * have_submounts - check for mounts over a dentry
1210 * @parent: dentry to check.
1212 * Return true if the parent or its subdirectories contain
1216 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1219 if (d_mountpoint(dentry)) {
1223 return D_WALK_CONTINUE;
1226 int have_submounts(struct dentry *parent)
1230 d_walk(parent, &ret, check_mount, NULL);
1234 EXPORT_SYMBOL(have_submounts);
1237 * Called by mount code to set a mountpoint and check if the mountpoint is
1238 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1239 * subtree can become unreachable).
1241 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1242 * this reason take rename_lock and d_lock on dentry and ancestors.
1244 int d_set_mounted(struct dentry *dentry)
1248 write_seqlock(&rename_lock);
1249 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1250 /* Need exclusion wrt. check_submounts_and_drop() */
1251 spin_lock(&p->d_lock);
1252 if (unlikely(d_unhashed(p))) {
1253 spin_unlock(&p->d_lock);
1256 spin_unlock(&p->d_lock);
1258 spin_lock(&dentry->d_lock);
1259 if (!d_unlinked(dentry)) {
1260 dentry->d_flags |= DCACHE_MOUNTED;
1263 spin_unlock(&dentry->d_lock);
1265 write_sequnlock(&rename_lock);
1270 * Search the dentry child list of the specified parent,
1271 * and move any unused dentries to the end of the unused
1272 * list for prune_dcache(). We descend to the next level
1273 * whenever the d_subdirs list is non-empty and continue
1276 * It returns zero iff there are no unused children,
1277 * otherwise it returns the number of children moved to
1278 * the end of the unused list. This may not be the total
1279 * number of unused children, because select_parent can
1280 * drop the lock and return early due to latency
1284 struct select_data {
1285 struct dentry *start;
1286 struct list_head dispose;
1290 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1292 struct select_data *data = _data;
1293 enum d_walk_ret ret = D_WALK_CONTINUE;
1295 if (data->start == dentry)
1299 * move only zero ref count dentries to the dispose list.
1301 * Those which are presently on the shrink list, being processed
1302 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1303 * loop in shrink_dcache_parent() might not make any progress
1306 if (dentry->d_lockref.count) {
1307 dentry_lru_del(dentry);
1308 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1309 dentry_lru_move_list(dentry, &data->dispose);
1310 dentry->d_flags |= DCACHE_SHRINK_LIST;
1312 ret = D_WALK_NORETRY;
1315 * We can return to the caller if we have found some (this
1316 * ensures forward progress). We'll be coming back to find
1319 if (data->found && need_resched())
1326 * shrink_dcache_parent - prune dcache
1327 * @parent: parent of entries to prune
1329 * Prune the dcache to remove unused children of the parent dentry.
1331 void shrink_dcache_parent(struct dentry *parent)
1334 struct select_data data;
1336 INIT_LIST_HEAD(&data.dispose);
1337 data.start = parent;
1340 d_walk(parent, &data, select_collect, NULL);
1344 shrink_dentry_list(&data.dispose);
1348 EXPORT_SYMBOL(shrink_dcache_parent);
1350 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1352 struct select_data *data = _data;
1354 if (d_mountpoint(dentry)) {
1355 data->found = -EBUSY;
1359 return select_collect(_data, dentry);
1362 static void check_and_drop(void *_data)
1364 struct select_data *data = _data;
1366 if (d_mountpoint(data->start))
1367 data->found = -EBUSY;
1369 __d_drop(data->start);
1373 * check_submounts_and_drop - prune dcache, check for submounts and drop
1375 * All done as a single atomic operation relative to has_unlinked_ancestor().
1376 * Returns 0 if successfully unhashed @parent. If there were submounts then
1379 * @dentry: dentry to prune and drop
1381 int check_submounts_and_drop(struct dentry *dentry)
1385 /* Negative dentries can be dropped without further checks */
1386 if (!dentry->d_inode) {
1392 struct select_data data;
1394 INIT_LIST_HEAD(&data.dispose);
1395 data.start = dentry;
1398 d_walk(dentry, &data, check_and_collect, check_and_drop);
1401 if (!list_empty(&data.dispose))
1402 shrink_dentry_list(&data.dispose);
1413 EXPORT_SYMBOL(check_submounts_and_drop);
1416 * __d_alloc - allocate a dcache entry
1417 * @sb: filesystem it will belong to
1418 * @name: qstr of the name
1420 * Allocates a dentry. It returns %NULL if there is insufficient memory
1421 * available. On a success the dentry is returned. The name passed in is
1422 * copied and the copy passed in may be reused after this call.
1425 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1427 struct dentry *dentry;
1430 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1435 * We guarantee that the inline name is always NUL-terminated.
1436 * This way the memcpy() done by the name switching in rename
1437 * will still always have a NUL at the end, even if we might
1438 * be overwriting an internal NUL character
1440 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1441 if (name->len > DNAME_INLINE_LEN-1) {
1442 dname = kmalloc(name->len + 1, GFP_KERNEL);
1444 kmem_cache_free(dentry_cache, dentry);
1448 dname = dentry->d_iname;
1451 dentry->d_name.len = name->len;
1452 dentry->d_name.hash = name->hash;
1453 memcpy(dname, name->name, name->len);
1454 dname[name->len] = 0;
1456 /* Make sure we always see the terminating NUL character */
1458 dentry->d_name.name = dname;
1460 dentry->d_lockref.count = 1;
1461 dentry->d_flags = 0;
1462 spin_lock_init(&dentry->d_lock);
1463 seqcount_init(&dentry->d_seq);
1464 dentry->d_inode = NULL;
1465 dentry->d_parent = dentry;
1467 dentry->d_op = NULL;
1468 dentry->d_fsdata = NULL;
1469 INIT_HLIST_BL_NODE(&dentry->d_hash);
1470 INIT_LIST_HEAD(&dentry->d_lru);
1471 INIT_LIST_HEAD(&dentry->d_subdirs);
1472 INIT_HLIST_NODE(&dentry->d_alias);
1473 INIT_LIST_HEAD(&dentry->d_u.d_child);
1474 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1476 this_cpu_inc(nr_dentry);
1482 * d_alloc - allocate a dcache entry
1483 * @parent: parent of entry to allocate
1484 * @name: qstr of the name
1486 * Allocates a dentry. It returns %NULL if there is insufficient memory
1487 * available. On a success the dentry is returned. The name passed in is
1488 * copied and the copy passed in may be reused after this call.
1490 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1492 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1496 spin_lock(&parent->d_lock);
1498 * don't need child lock because it is not subject
1499 * to concurrency here
1501 __dget_dlock(parent);
1502 dentry->d_parent = parent;
1503 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1504 spin_unlock(&parent->d_lock);
1508 EXPORT_SYMBOL(d_alloc);
1510 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1512 struct dentry *dentry = __d_alloc(sb, name);
1514 dentry->d_flags |= DCACHE_DISCONNECTED;
1517 EXPORT_SYMBOL(d_alloc_pseudo);
1519 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1524 q.len = strlen(name);
1525 q.hash = full_name_hash(q.name, q.len);
1526 return d_alloc(parent, &q);
1528 EXPORT_SYMBOL(d_alloc_name);
1530 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1532 WARN_ON_ONCE(dentry->d_op);
1533 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1535 DCACHE_OP_REVALIDATE |
1536 DCACHE_OP_WEAK_REVALIDATE |
1537 DCACHE_OP_DELETE ));
1542 dentry->d_flags |= DCACHE_OP_HASH;
1544 dentry->d_flags |= DCACHE_OP_COMPARE;
1545 if (op->d_revalidate)
1546 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1547 if (op->d_weak_revalidate)
1548 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1550 dentry->d_flags |= DCACHE_OP_DELETE;
1552 dentry->d_flags |= DCACHE_OP_PRUNE;
1555 EXPORT_SYMBOL(d_set_d_op);
1557 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1559 spin_lock(&dentry->d_lock);
1561 if (unlikely(IS_AUTOMOUNT(inode)))
1562 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1563 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1565 dentry->d_inode = inode;
1566 dentry_rcuwalk_barrier(dentry);
1567 spin_unlock(&dentry->d_lock);
1568 fsnotify_d_instantiate(dentry, inode);
1572 * d_instantiate - fill in inode information for a dentry
1573 * @entry: dentry to complete
1574 * @inode: inode to attach to this dentry
1576 * Fill in inode information in the entry.
1578 * This turns negative dentries into productive full members
1581 * NOTE! This assumes that the inode count has been incremented
1582 * (or otherwise set) by the caller to indicate that it is now
1583 * in use by the dcache.
1586 void d_instantiate(struct dentry *entry, struct inode * inode)
1588 BUG_ON(!hlist_unhashed(&entry->d_alias));
1590 spin_lock(&inode->i_lock);
1591 __d_instantiate(entry, inode);
1593 spin_unlock(&inode->i_lock);
1594 security_d_instantiate(entry, inode);
1596 EXPORT_SYMBOL(d_instantiate);
1599 * d_instantiate_unique - instantiate a non-aliased dentry
1600 * @entry: dentry to instantiate
1601 * @inode: inode to attach to this dentry
1603 * Fill in inode information in the entry. On success, it returns NULL.
1604 * If an unhashed alias of "entry" already exists, then we return the
1605 * aliased dentry instead and drop one reference to inode.
1607 * Note that in order to avoid conflicts with rename() etc, the caller
1608 * had better be holding the parent directory semaphore.
1610 * This also assumes that the inode count has been incremented
1611 * (or otherwise set) by the caller to indicate that it is now
1612 * in use by the dcache.
1614 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1615 struct inode *inode)
1617 struct dentry *alias;
1618 int len = entry->d_name.len;
1619 const char *name = entry->d_name.name;
1620 unsigned int hash = entry->d_name.hash;
1623 __d_instantiate(entry, NULL);
1627 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1629 * Don't need alias->d_lock here, because aliases with
1630 * d_parent == entry->d_parent are not subject to name or
1631 * parent changes, because the parent inode i_mutex is held.
1633 if (alias->d_name.hash != hash)
1635 if (alias->d_parent != entry->d_parent)
1637 if (alias->d_name.len != len)
1639 if (dentry_cmp(alias, name, len))
1645 __d_instantiate(entry, inode);
1649 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1651 struct dentry *result;
1653 BUG_ON(!hlist_unhashed(&entry->d_alias));
1656 spin_lock(&inode->i_lock);
1657 result = __d_instantiate_unique(entry, inode);
1659 spin_unlock(&inode->i_lock);
1662 security_d_instantiate(entry, inode);
1666 BUG_ON(!d_unhashed(result));
1671 EXPORT_SYMBOL(d_instantiate_unique);
1673 struct dentry *d_make_root(struct inode *root_inode)
1675 struct dentry *res = NULL;
1678 static const struct qstr name = QSTR_INIT("/", 1);
1680 res = __d_alloc(root_inode->i_sb, &name);
1682 d_instantiate(res, root_inode);
1688 EXPORT_SYMBOL(d_make_root);
1690 static struct dentry * __d_find_any_alias(struct inode *inode)
1692 struct dentry *alias;
1694 if (hlist_empty(&inode->i_dentry))
1696 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1702 * d_find_any_alias - find any alias for a given inode
1703 * @inode: inode to find an alias for
1705 * If any aliases exist for the given inode, take and return a
1706 * reference for one of them. If no aliases exist, return %NULL.
1708 struct dentry *d_find_any_alias(struct inode *inode)
1712 spin_lock(&inode->i_lock);
1713 de = __d_find_any_alias(inode);
1714 spin_unlock(&inode->i_lock);
1717 EXPORT_SYMBOL(d_find_any_alias);
1720 * d_obtain_alias - find or allocate a dentry for a given inode
1721 * @inode: inode to allocate the dentry for
1723 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1724 * similar open by handle operations. The returned dentry may be anonymous,
1725 * or may have a full name (if the inode was already in the cache).
1727 * When called on a directory inode, we must ensure that the inode only ever
1728 * has one dentry. If a dentry is found, that is returned instead of
1729 * allocating a new one.
1731 * On successful return, the reference to the inode has been transferred
1732 * to the dentry. In case of an error the reference on the inode is released.
1733 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1734 * be passed in and will be the error will be propagate to the return value,
1735 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1737 struct dentry *d_obtain_alias(struct inode *inode)
1739 static const struct qstr anonstring = QSTR_INIT("/", 1);
1744 return ERR_PTR(-ESTALE);
1746 return ERR_CAST(inode);
1748 res = d_find_any_alias(inode);
1752 tmp = __d_alloc(inode->i_sb, &anonstring);
1754 res = ERR_PTR(-ENOMEM);
1758 spin_lock(&inode->i_lock);
1759 res = __d_find_any_alias(inode);
1761 spin_unlock(&inode->i_lock);
1766 /* attach a disconnected dentry */
1767 spin_lock(&tmp->d_lock);
1768 tmp->d_inode = inode;
1769 tmp->d_flags |= DCACHE_DISCONNECTED;
1770 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1771 hlist_bl_lock(&tmp->d_sb->s_anon);
1772 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1773 hlist_bl_unlock(&tmp->d_sb->s_anon);
1774 spin_unlock(&tmp->d_lock);
1775 spin_unlock(&inode->i_lock);
1776 security_d_instantiate(tmp, inode);
1781 if (res && !IS_ERR(res))
1782 security_d_instantiate(res, inode);
1786 EXPORT_SYMBOL(d_obtain_alias);
1789 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1790 * @inode: the inode which may have a disconnected dentry
1791 * @dentry: a negative dentry which we want to point to the inode.
1793 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1794 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1795 * and return it, else simply d_add the inode to the dentry and return NULL.
1797 * This is needed in the lookup routine of any filesystem that is exportable
1798 * (via knfsd) so that we can build dcache paths to directories effectively.
1800 * If a dentry was found and moved, then it is returned. Otherwise NULL
1801 * is returned. This matches the expected return value of ->lookup.
1803 * Cluster filesystems may call this function with a negative, hashed dentry.
1804 * In that case, we know that the inode will be a regular file, and also this
1805 * will only occur during atomic_open. So we need to check for the dentry
1806 * being already hashed only in the final case.
1808 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1810 struct dentry *new = NULL;
1813 return ERR_CAST(inode);
1815 if (inode && S_ISDIR(inode->i_mode)) {
1816 spin_lock(&inode->i_lock);
1817 new = __d_find_alias(inode, 1);
1819 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1820 spin_unlock(&inode->i_lock);
1821 security_d_instantiate(new, inode);
1822 d_move(new, dentry);
1825 /* already taking inode->i_lock, so d_add() by hand */
1826 __d_instantiate(dentry, inode);
1827 spin_unlock(&inode->i_lock);
1828 security_d_instantiate(dentry, inode);
1832 d_instantiate(dentry, inode);
1833 if (d_unhashed(dentry))
1838 EXPORT_SYMBOL(d_splice_alias);
1841 * d_add_ci - lookup or allocate new dentry with case-exact name
1842 * @inode: the inode case-insensitive lookup has found
1843 * @dentry: the negative dentry that was passed to the parent's lookup func
1844 * @name: the case-exact name to be associated with the returned dentry
1846 * This is to avoid filling the dcache with case-insensitive names to the
1847 * same inode, only the actual correct case is stored in the dcache for
1848 * case-insensitive filesystems.
1850 * For a case-insensitive lookup match and if the the case-exact dentry
1851 * already exists in in the dcache, use it and return it.
1853 * If no entry exists with the exact case name, allocate new dentry with
1854 * the exact case, and return the spliced entry.
1856 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1859 struct dentry *found;
1863 * First check if a dentry matching the name already exists,
1864 * if not go ahead and create it now.
1866 found = d_hash_and_lookup(dentry->d_parent, name);
1867 if (unlikely(IS_ERR(found)))
1870 new = d_alloc(dentry->d_parent, name);
1872 found = ERR_PTR(-ENOMEM);
1876 found = d_splice_alias(inode, new);
1885 * If a matching dentry exists, and it's not negative use it.
1887 * Decrement the reference count to balance the iget() done
1890 if (found->d_inode) {
1891 if (unlikely(found->d_inode != inode)) {
1892 /* This can't happen because bad inodes are unhashed. */
1893 BUG_ON(!is_bad_inode(inode));
1894 BUG_ON(!is_bad_inode(found->d_inode));
1901 * Negative dentry: instantiate it unless the inode is a directory and
1902 * already has a dentry.
1904 new = d_splice_alias(inode, found);
1915 EXPORT_SYMBOL(d_add_ci);
1918 * Do the slow-case of the dentry name compare.
1920 * Unlike the dentry_cmp() function, we need to atomically
1921 * load the name and length information, so that the
1922 * filesystem can rely on them, and can use the 'name' and
1923 * 'len' information without worrying about walking off the
1924 * end of memory etc.
1926 * Thus the read_seqcount_retry() and the "duplicate" info
1927 * in arguments (the low-level filesystem should not look
1928 * at the dentry inode or name contents directly, since
1929 * rename can change them while we're in RCU mode).
1931 enum slow_d_compare {
1937 static noinline enum slow_d_compare slow_dentry_cmp(
1938 const struct dentry *parent,
1939 struct dentry *dentry,
1941 const struct qstr *name)
1943 int tlen = dentry->d_name.len;
1944 const char *tname = dentry->d_name.name;
1946 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1948 return D_COMP_SEQRETRY;
1950 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1951 return D_COMP_NOMATCH;
1956 * __d_lookup_rcu - search for a dentry (racy, store-free)
1957 * @parent: parent dentry
1958 * @name: qstr of name we wish to find
1959 * @seqp: returns d_seq value at the point where the dentry was found
1960 * Returns: dentry, or NULL
1962 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1963 * resolution (store-free path walking) design described in
1964 * Documentation/filesystems/path-lookup.txt.
1966 * This is not to be used outside core vfs.
1968 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1969 * held, and rcu_read_lock held. The returned dentry must not be stored into
1970 * without taking d_lock and checking d_seq sequence count against @seq
1973 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
1976 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1977 * the returned dentry, so long as its parent's seqlock is checked after the
1978 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1979 * is formed, giving integrity down the path walk.
1981 * NOTE! The caller *has* to check the resulting dentry against the sequence
1982 * number we've returned before using any of the resulting dentry state!
1984 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1985 const struct qstr *name,
1988 u64 hashlen = name->hash_len;
1989 const unsigned char *str = name->name;
1990 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
1991 struct hlist_bl_node *node;
1992 struct dentry *dentry;
1995 * Note: There is significant duplication with __d_lookup_rcu which is
1996 * required to prevent single threaded performance regressions
1997 * especially on architectures where smp_rmb (in seqcounts) are costly.
1998 * Keep the two functions in sync.
2002 * The hash list is protected using RCU.
2004 * Carefully use d_seq when comparing a candidate dentry, to avoid
2005 * races with d_move().
2007 * It is possible that concurrent renames can mess up our list
2008 * walk here and result in missing our dentry, resulting in the
2009 * false-negative result. d_lookup() protects against concurrent
2010 * renames using rename_lock seqlock.
2012 * See Documentation/filesystems/path-lookup.txt for more details.
2014 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2019 * The dentry sequence count protects us from concurrent
2020 * renames, and thus protects parent and name fields.
2022 * The caller must perform a seqcount check in order
2023 * to do anything useful with the returned dentry.
2025 * NOTE! We do a "raw" seqcount_begin here. That means that
2026 * we don't wait for the sequence count to stabilize if it
2027 * is in the middle of a sequence change. If we do the slow
2028 * dentry compare, we will do seqretries until it is stable,
2029 * and if we end up with a successful lookup, we actually
2030 * want to exit RCU lookup anyway.
2032 seq = raw_seqcount_begin(&dentry->d_seq);
2033 if (dentry->d_parent != parent)
2035 if (d_unhashed(dentry))
2038 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2039 if (dentry->d_name.hash != hashlen_hash(hashlen))
2042 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2045 case D_COMP_NOMATCH:
2052 if (dentry->d_name.hash_len != hashlen)
2055 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2062 * d_lookup - search for a dentry
2063 * @parent: parent dentry
2064 * @name: qstr of name we wish to find
2065 * Returns: dentry, or NULL
2067 * d_lookup searches the children of the parent dentry for the name in
2068 * question. If the dentry is found its reference count is incremented and the
2069 * dentry is returned. The caller must use dput to free the entry when it has
2070 * finished using it. %NULL is returned if the dentry does not exist.
2072 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2074 struct dentry *dentry;
2078 seq = read_seqbegin(&rename_lock);
2079 dentry = __d_lookup(parent, name);
2082 } while (read_seqretry(&rename_lock, seq));
2085 EXPORT_SYMBOL(d_lookup);
2088 * __d_lookup - search for a dentry (racy)
2089 * @parent: parent dentry
2090 * @name: qstr of name we wish to find
2091 * Returns: dentry, or NULL
2093 * __d_lookup is like d_lookup, however it may (rarely) return a
2094 * false-negative result due to unrelated rename activity.
2096 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2097 * however it must be used carefully, eg. with a following d_lookup in
2098 * the case of failure.
2100 * __d_lookup callers must be commented.
2102 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2104 unsigned int len = name->len;
2105 unsigned int hash = name->hash;
2106 const unsigned char *str = name->name;
2107 struct hlist_bl_head *b = d_hash(parent, hash);
2108 struct hlist_bl_node *node;
2109 struct dentry *found = NULL;
2110 struct dentry *dentry;
2113 * Note: There is significant duplication with __d_lookup_rcu which is
2114 * required to prevent single threaded performance regressions
2115 * especially on architectures where smp_rmb (in seqcounts) are costly.
2116 * Keep the two functions in sync.
2120 * The hash list is protected using RCU.
2122 * Take d_lock when comparing a candidate dentry, to avoid races
2125 * It is possible that concurrent renames can mess up our list
2126 * walk here and result in missing our dentry, resulting in the
2127 * false-negative result. d_lookup() protects against concurrent
2128 * renames using rename_lock seqlock.
2130 * See Documentation/filesystems/path-lookup.txt for more details.
2134 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2136 if (dentry->d_name.hash != hash)
2139 spin_lock(&dentry->d_lock);
2140 if (dentry->d_parent != parent)
2142 if (d_unhashed(dentry))
2146 * It is safe to compare names since d_move() cannot
2147 * change the qstr (protected by d_lock).
2149 if (parent->d_flags & DCACHE_OP_COMPARE) {
2150 int tlen = dentry->d_name.len;
2151 const char *tname = dentry->d_name.name;
2152 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2155 if (dentry->d_name.len != len)
2157 if (dentry_cmp(dentry, str, len))
2161 dentry->d_lockref.count++;
2163 spin_unlock(&dentry->d_lock);
2166 spin_unlock(&dentry->d_lock);
2174 * d_hash_and_lookup - hash the qstr then search for a dentry
2175 * @dir: Directory to search in
2176 * @name: qstr of name we wish to find
2178 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2180 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2183 * Check for a fs-specific hash function. Note that we must
2184 * calculate the standard hash first, as the d_op->d_hash()
2185 * routine may choose to leave the hash value unchanged.
2187 name->hash = full_name_hash(name->name, name->len);
2188 if (dir->d_flags & DCACHE_OP_HASH) {
2189 int err = dir->d_op->d_hash(dir, name);
2190 if (unlikely(err < 0))
2191 return ERR_PTR(err);
2193 return d_lookup(dir, name);
2195 EXPORT_SYMBOL(d_hash_and_lookup);
2198 * d_validate - verify dentry provided from insecure source (deprecated)
2199 * @dentry: The dentry alleged to be valid child of @dparent
2200 * @dparent: The parent dentry (known to be valid)
2202 * An insecure source has sent us a dentry, here we verify it and dget() it.
2203 * This is used by ncpfs in its readdir implementation.
2204 * Zero is returned in the dentry is invalid.
2206 * This function is slow for big directories, and deprecated, do not use it.
2208 int d_validate(struct dentry *dentry, struct dentry *dparent)
2210 struct dentry *child;
2212 spin_lock(&dparent->d_lock);
2213 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2214 if (dentry == child) {
2215 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2216 __dget_dlock(dentry);
2217 spin_unlock(&dentry->d_lock);
2218 spin_unlock(&dparent->d_lock);
2222 spin_unlock(&dparent->d_lock);
2226 EXPORT_SYMBOL(d_validate);
2229 * When a file is deleted, we have two options:
2230 * - turn this dentry into a negative dentry
2231 * - unhash this dentry and free it.
2233 * Usually, we want to just turn this into
2234 * a negative dentry, but if anybody else is
2235 * currently using the dentry or the inode
2236 * we can't do that and we fall back on removing
2237 * it from the hash queues and waiting for
2238 * it to be deleted later when it has no users
2242 * d_delete - delete a dentry
2243 * @dentry: The dentry to delete
2245 * Turn the dentry into a negative dentry if possible, otherwise
2246 * remove it from the hash queues so it can be deleted later
2249 void d_delete(struct dentry * dentry)
2251 struct inode *inode;
2254 * Are we the only user?
2257 spin_lock(&dentry->d_lock);
2258 inode = dentry->d_inode;
2259 isdir = S_ISDIR(inode->i_mode);
2260 if (dentry->d_lockref.count == 1) {
2261 if (!spin_trylock(&inode->i_lock)) {
2262 spin_unlock(&dentry->d_lock);
2266 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2267 dentry_unlink_inode(dentry);
2268 fsnotify_nameremove(dentry, isdir);
2272 if (!d_unhashed(dentry))
2275 spin_unlock(&dentry->d_lock);
2277 fsnotify_nameremove(dentry, isdir);
2279 EXPORT_SYMBOL(d_delete);
2281 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2283 BUG_ON(!d_unhashed(entry));
2285 entry->d_flags |= DCACHE_RCUACCESS;
2286 hlist_bl_add_head_rcu(&entry->d_hash, b);
2290 static void _d_rehash(struct dentry * entry)
2292 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2296 * d_rehash - add an entry back to the hash
2297 * @entry: dentry to add to the hash
2299 * Adds a dentry to the hash according to its name.
2302 void d_rehash(struct dentry * entry)
2304 spin_lock(&entry->d_lock);
2306 spin_unlock(&entry->d_lock);
2308 EXPORT_SYMBOL(d_rehash);
2311 * dentry_update_name_case - update case insensitive dentry with a new name
2312 * @dentry: dentry to be updated
2315 * Update a case insensitive dentry with new case of name.
2317 * dentry must have been returned by d_lookup with name @name. Old and new
2318 * name lengths must match (ie. no d_compare which allows mismatched name
2321 * Parent inode i_mutex must be held over d_lookup and into this call (to
2322 * keep renames and concurrent inserts, and readdir(2) away).
2324 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2326 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2327 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2329 spin_lock(&dentry->d_lock);
2330 write_seqcount_begin(&dentry->d_seq);
2331 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2332 write_seqcount_end(&dentry->d_seq);
2333 spin_unlock(&dentry->d_lock);
2335 EXPORT_SYMBOL(dentry_update_name_case);
2337 static void switch_names(struct dentry *dentry, struct dentry *target)
2339 if (dname_external(target)) {
2340 if (dname_external(dentry)) {
2342 * Both external: swap the pointers
2344 swap(target->d_name.name, dentry->d_name.name);
2347 * dentry:internal, target:external. Steal target's
2348 * storage and make target internal.
2350 memcpy(target->d_iname, dentry->d_name.name,
2351 dentry->d_name.len + 1);
2352 dentry->d_name.name = target->d_name.name;
2353 target->d_name.name = target->d_iname;
2356 if (dname_external(dentry)) {
2358 * dentry:external, target:internal. Give dentry's
2359 * storage to target and make dentry internal
2361 memcpy(dentry->d_iname, target->d_name.name,
2362 target->d_name.len + 1);
2363 target->d_name.name = dentry->d_name.name;
2364 dentry->d_name.name = dentry->d_iname;
2367 * Both are internal. Just copy target to dentry
2369 memcpy(dentry->d_iname, target->d_name.name,
2370 target->d_name.len + 1);
2371 dentry->d_name.len = target->d_name.len;
2375 swap(dentry->d_name.len, target->d_name.len);
2378 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2381 * XXXX: do we really need to take target->d_lock?
2383 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2384 spin_lock(&target->d_parent->d_lock);
2386 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2387 spin_lock(&dentry->d_parent->d_lock);
2388 spin_lock_nested(&target->d_parent->d_lock,
2389 DENTRY_D_LOCK_NESTED);
2391 spin_lock(&target->d_parent->d_lock);
2392 spin_lock_nested(&dentry->d_parent->d_lock,
2393 DENTRY_D_LOCK_NESTED);
2396 if (target < dentry) {
2397 spin_lock_nested(&target->d_lock, 2);
2398 spin_lock_nested(&dentry->d_lock, 3);
2400 spin_lock_nested(&dentry->d_lock, 2);
2401 spin_lock_nested(&target->d_lock, 3);
2405 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2406 struct dentry *target)
2408 if (target->d_parent != dentry->d_parent)
2409 spin_unlock(&dentry->d_parent->d_lock);
2410 if (target->d_parent != target)
2411 spin_unlock(&target->d_parent->d_lock);
2415 * When switching names, the actual string doesn't strictly have to
2416 * be preserved in the target - because we're dropping the target
2417 * anyway. As such, we can just do a simple memcpy() to copy over
2418 * the new name before we switch.
2420 * Note that we have to be a lot more careful about getting the hash
2421 * switched - we have to switch the hash value properly even if it
2422 * then no longer matches the actual (corrupted) string of the target.
2423 * The hash value has to match the hash queue that the dentry is on..
2426 * __d_move - move a dentry
2427 * @dentry: entry to move
2428 * @target: new dentry
2430 * Update the dcache to reflect the move of a file name. Negative
2431 * dcache entries should not be moved in this way. Caller must hold
2432 * rename_lock, the i_mutex of the source and target directories,
2433 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2435 static void __d_move(struct dentry * dentry, struct dentry * target)
2437 if (!dentry->d_inode)
2438 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2440 BUG_ON(d_ancestor(dentry, target));
2441 BUG_ON(d_ancestor(target, dentry));
2443 dentry_lock_for_move(dentry, target);
2445 write_seqcount_begin(&dentry->d_seq);
2446 write_seqcount_begin(&target->d_seq);
2448 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2451 * Move the dentry to the target hash queue. Don't bother checking
2452 * for the same hash queue because of how unlikely it is.
2455 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2457 /* Unhash the target: dput() will then get rid of it */
2460 list_del(&dentry->d_u.d_child);
2461 list_del(&target->d_u.d_child);
2463 /* Switch the names.. */
2464 switch_names(dentry, target);
2465 swap(dentry->d_name.hash, target->d_name.hash);
2467 /* ... and switch the parents */
2468 if (IS_ROOT(dentry)) {
2469 dentry->d_parent = target->d_parent;
2470 target->d_parent = target;
2471 INIT_LIST_HEAD(&target->d_u.d_child);
2473 swap(dentry->d_parent, target->d_parent);
2475 /* And add them back to the (new) parent lists */
2476 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2479 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2481 write_seqcount_end(&target->d_seq);
2482 write_seqcount_end(&dentry->d_seq);
2484 dentry_unlock_parents_for_move(dentry, target);
2485 spin_unlock(&target->d_lock);
2486 fsnotify_d_move(dentry);
2487 spin_unlock(&dentry->d_lock);
2491 * d_move - move a dentry
2492 * @dentry: entry to move
2493 * @target: new dentry
2495 * Update the dcache to reflect the move of a file name. Negative
2496 * dcache entries should not be moved in this way. See the locking
2497 * requirements for __d_move.
2499 void d_move(struct dentry *dentry, struct dentry *target)
2501 write_seqlock(&rename_lock);
2502 __d_move(dentry, target);
2503 write_sequnlock(&rename_lock);
2505 EXPORT_SYMBOL(d_move);
2508 * d_ancestor - search for an ancestor
2509 * @p1: ancestor dentry
2512 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2513 * an ancestor of p2, else NULL.
2515 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2519 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2520 if (p->d_parent == p1)
2527 * This helper attempts to cope with remotely renamed directories
2529 * It assumes that the caller is already holding
2530 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2532 * Note: If ever the locking in lock_rename() changes, then please
2533 * remember to update this too...
2535 static struct dentry *__d_unalias(struct inode *inode,
2536 struct dentry *dentry, struct dentry *alias)
2538 struct mutex *m1 = NULL, *m2 = NULL;
2539 struct dentry *ret = ERR_PTR(-EBUSY);
2541 /* If alias and dentry share a parent, then no extra locks required */
2542 if (alias->d_parent == dentry->d_parent)
2545 /* See lock_rename() */
2546 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2548 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2549 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2551 m2 = &alias->d_parent->d_inode->i_mutex;
2553 if (likely(!d_mountpoint(alias))) {
2554 __d_move(alias, dentry);
2558 spin_unlock(&inode->i_lock);
2567 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2568 * named dentry in place of the dentry to be replaced.
2569 * returns with anon->d_lock held!
2571 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2573 struct dentry *dparent;
2575 dentry_lock_for_move(anon, dentry);
2577 write_seqcount_begin(&dentry->d_seq);
2578 write_seqcount_begin(&anon->d_seq);
2580 dparent = dentry->d_parent;
2582 switch_names(dentry, anon);
2583 swap(dentry->d_name.hash, anon->d_name.hash);
2585 dentry->d_parent = dentry;
2586 list_del_init(&dentry->d_u.d_child);
2587 anon->d_parent = dparent;
2588 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2590 write_seqcount_end(&dentry->d_seq);
2591 write_seqcount_end(&anon->d_seq);
2593 dentry_unlock_parents_for_move(anon, dentry);
2594 spin_unlock(&dentry->d_lock);
2596 /* anon->d_lock still locked, returns locked */
2597 anon->d_flags &= ~DCACHE_DISCONNECTED;
2601 * d_materialise_unique - introduce an inode into the tree
2602 * @dentry: candidate dentry
2603 * @inode: inode to bind to the dentry, to which aliases may be attached
2605 * Introduces an dentry into the tree, substituting an extant disconnected
2606 * root directory alias in its place if there is one. Caller must hold the
2607 * i_mutex of the parent directory.
2609 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2611 struct dentry *actual;
2613 BUG_ON(!d_unhashed(dentry));
2617 __d_instantiate(dentry, NULL);
2622 spin_lock(&inode->i_lock);
2624 if (S_ISDIR(inode->i_mode)) {
2625 struct dentry *alias;
2627 /* Does an aliased dentry already exist? */
2628 alias = __d_find_alias(inode, 0);
2631 write_seqlock(&rename_lock);
2633 if (d_ancestor(alias, dentry)) {
2634 /* Check for loops */
2635 actual = ERR_PTR(-ELOOP);
2636 spin_unlock(&inode->i_lock);
2637 } else if (IS_ROOT(alias)) {
2638 /* Is this an anonymous mountpoint that we
2639 * could splice into our tree? */
2640 __d_materialise_dentry(dentry, alias);
2641 write_sequnlock(&rename_lock);
2645 /* Nope, but we must(!) avoid directory
2646 * aliasing. This drops inode->i_lock */
2647 actual = __d_unalias(inode, dentry, alias);
2649 write_sequnlock(&rename_lock);
2650 if (IS_ERR(actual)) {
2651 if (PTR_ERR(actual) == -ELOOP)
2652 pr_warn_ratelimited(
2653 "VFS: Lookup of '%s' in %s %s"
2654 " would have caused loop\n",
2655 dentry->d_name.name,
2656 inode->i_sb->s_type->name,
2664 /* Add a unique reference */
2665 actual = __d_instantiate_unique(dentry, inode);
2669 BUG_ON(!d_unhashed(actual));
2671 spin_lock(&actual->d_lock);
2674 spin_unlock(&actual->d_lock);
2675 spin_unlock(&inode->i_lock);
2677 if (actual == dentry) {
2678 security_d_instantiate(dentry, inode);
2685 EXPORT_SYMBOL_GPL(d_materialise_unique);
2687 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2691 return -ENAMETOOLONG;
2693 memcpy(*buffer, str, namelen);
2698 * prepend_name - prepend a pathname in front of current buffer pointer
2699 * buffer: buffer pointer
2700 * buflen: allocated length of the buffer
2701 * name: name string and length qstr structure
2703 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2704 * make sure that either the old or the new name pointer and length are
2705 * fetched. However, there may be mismatch between length and pointer.
2706 * The length cannot be trusted, we need to copy it byte-by-byte until
2707 * the length is reached or a null byte is found. It also prepends "/" at
2708 * the beginning of the name. The sequence number check at the caller will
2709 * retry it again when a d_move() does happen. So any garbage in the buffer
2710 * due to mismatched pointer and length will be discarded.
2712 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2714 const char *dname = ACCESS_ONCE(name->name);
2715 u32 dlen = ACCESS_ONCE(name->len);
2718 if (*buflen < dlen + 1)
2719 return -ENAMETOOLONG;
2720 *buflen -= dlen + 1;
2721 p = *buffer -= dlen + 1;
2733 * prepend_path - Prepend path string to a buffer
2734 * @path: the dentry/vfsmount to report
2735 * @root: root vfsmnt/dentry
2736 * @buffer: pointer to the end of the buffer
2737 * @buflen: pointer to buffer length
2739 * The function tries to write out the pathname without taking any lock other
2740 * than the RCU read lock to make sure that dentries won't go away. It only
2741 * checks the sequence number of the global rename_lock as any change in the
2742 * dentry's d_seq will be preceded by changes in the rename_lock sequence
2743 * number. If the sequence number had been change, it will restart the whole
2744 * pathname back-tracing sequence again. It performs a total of 3 trials of
2745 * lockless back-tracing sequences before falling back to take the
2748 static int prepend_path(const struct path *path,
2749 const struct path *root,
2750 char **buffer, int *buflen)
2752 struct dentry *dentry = path->dentry;
2753 struct vfsmount *vfsmnt = path->mnt;
2754 struct mount *mnt = real_mount(vfsmnt);
2764 read_seqbegin_or_lock(&rename_lock, &seq);
2765 while (dentry != root->dentry || vfsmnt != root->mnt) {
2766 struct dentry * parent;
2768 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2770 if (mnt_has_parent(mnt)) {
2771 dentry = mnt->mnt_mountpoint;
2772 mnt = mnt->mnt_parent;
2777 * Filesystems needing to implement special "root names"
2778 * should do so with ->d_dname()
2780 if (IS_ROOT(dentry) &&
2781 (dentry->d_name.len != 1 ||
2782 dentry->d_name.name[0] != '/')) {
2783 WARN(1, "Root dentry has weird name <%.*s>\n",
2784 (int) dentry->d_name.len,
2785 dentry->d_name.name);
2788 error = is_mounted(vfsmnt) ? 1 : 2;
2791 parent = dentry->d_parent;
2793 error = prepend_name(&bptr, &blen, &dentry->d_name);
2801 if (need_seqretry(&rename_lock, seq)) {
2805 done_seqretry(&rename_lock, seq);
2807 if (error >= 0 && bptr == *buffer) {
2809 error = -ENAMETOOLONG;
2819 * __d_path - return the path of a dentry
2820 * @path: the dentry/vfsmount to report
2821 * @root: root vfsmnt/dentry
2822 * @buf: buffer to return value in
2823 * @buflen: buffer length
2825 * Convert a dentry into an ASCII path name.
2827 * Returns a pointer into the buffer or an error code if the
2828 * path was too long.
2830 * "buflen" should be positive.
2832 * If the path is not reachable from the supplied root, return %NULL.
2834 char *__d_path(const struct path *path,
2835 const struct path *root,
2836 char *buf, int buflen)
2838 char *res = buf + buflen;
2841 prepend(&res, &buflen, "\0", 1);
2842 br_read_lock(&vfsmount_lock);
2843 error = prepend_path(path, root, &res, &buflen);
2844 br_read_unlock(&vfsmount_lock);
2847 return ERR_PTR(error);
2853 char *d_absolute_path(const struct path *path,
2854 char *buf, int buflen)
2856 struct path root = {};
2857 char *res = buf + buflen;
2860 prepend(&res, &buflen, "\0", 1);
2861 br_read_lock(&vfsmount_lock);
2862 error = prepend_path(path, &root, &res, &buflen);
2863 br_read_unlock(&vfsmount_lock);
2868 return ERR_PTR(error);
2873 * same as __d_path but appends "(deleted)" for unlinked files.
2875 static int path_with_deleted(const struct path *path,
2876 const struct path *root,
2877 char **buf, int *buflen)
2879 prepend(buf, buflen, "\0", 1);
2880 if (d_unlinked(path->dentry)) {
2881 int error = prepend(buf, buflen, " (deleted)", 10);
2886 return prepend_path(path, root, buf, buflen);
2889 static int prepend_unreachable(char **buffer, int *buflen)
2891 return prepend(buffer, buflen, "(unreachable)", 13);
2895 * d_path - return the path of a dentry
2896 * @path: path to report
2897 * @buf: buffer to return value in
2898 * @buflen: buffer length
2900 * Convert a dentry into an ASCII path name. If the entry has been deleted
2901 * the string " (deleted)" is appended. Note that this is ambiguous.
2903 * Returns a pointer into the buffer or an error code if the path was
2904 * too long. Note: Callers should use the returned pointer, not the passed
2905 * in buffer, to use the name! The implementation often starts at an offset
2906 * into the buffer, and may leave 0 bytes at the start.
2908 * "buflen" should be positive.
2910 char *d_path(const struct path *path, char *buf, int buflen)
2912 char *res = buf + buflen;
2917 * We have various synthetic filesystems that never get mounted. On
2918 * these filesystems dentries are never used for lookup purposes, and
2919 * thus don't need to be hashed. They also don't need a name until a
2920 * user wants to identify the object in /proc/pid/fd/. The little hack
2921 * below allows us to generate a name for these objects on demand:
2923 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2924 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2926 get_fs_root(current->fs, &root);
2927 br_read_lock(&vfsmount_lock);
2928 error = path_with_deleted(path, &root, &res, &buflen);
2929 br_read_unlock(&vfsmount_lock);
2931 res = ERR_PTR(error);
2935 EXPORT_SYMBOL(d_path);
2938 * Helper function for dentry_operations.d_dname() members
2940 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2941 const char *fmt, ...)
2947 va_start(args, fmt);
2948 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2951 if (sz > sizeof(temp) || sz > buflen)
2952 return ERR_PTR(-ENAMETOOLONG);
2954 buffer += buflen - sz;
2955 return memcpy(buffer, temp, sz);
2958 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
2960 char *end = buffer + buflen;
2961 /* these dentries are never renamed, so d_lock is not needed */
2962 if (prepend(&end, &buflen, " (deleted)", 11) ||
2963 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
2964 prepend(&end, &buflen, "/", 1))
2965 end = ERR_PTR(-ENAMETOOLONG);
2970 * Write full pathname from the root of the filesystem into the buffer.
2972 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2982 prepend(&end, &len, "\0", 1);
2988 read_seqbegin_or_lock(&rename_lock, &seq);
2989 while (!IS_ROOT(dentry)) {
2990 struct dentry *parent = dentry->d_parent;
2994 error = prepend_name(&end, &len, &dentry->d_name);
3003 if (need_seqretry(&rename_lock, seq)) {
3007 done_seqretry(&rename_lock, seq);
3012 return ERR_PTR(-ENAMETOOLONG);
3015 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3017 return __dentry_path(dentry, buf, buflen);
3019 EXPORT_SYMBOL(dentry_path_raw);
3021 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3026 if (d_unlinked(dentry)) {
3028 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3032 retval = __dentry_path(dentry, buf, buflen);
3033 if (!IS_ERR(retval) && p)
3034 *p = '/'; /* restore '/' overriden with '\0' */
3037 return ERR_PTR(-ENAMETOOLONG);
3041 * NOTE! The user-level library version returns a
3042 * character pointer. The kernel system call just
3043 * returns the length of the buffer filled (which
3044 * includes the ending '\0' character), or a negative
3045 * error value. So libc would do something like
3047 * char *getcwd(char * buf, size_t size)
3051 * retval = sys_getcwd(buf, size);
3058 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3061 struct path pwd, root;
3062 char *page = (char *) __get_free_page(GFP_USER);
3067 get_fs_root_and_pwd(current->fs, &root, &pwd);
3070 br_read_lock(&vfsmount_lock);
3071 if (!d_unlinked(pwd.dentry)) {
3073 char *cwd = page + PAGE_SIZE;
3074 int buflen = PAGE_SIZE;
3076 prepend(&cwd, &buflen, "\0", 1);
3077 error = prepend_path(&pwd, &root, &cwd, &buflen);
3078 br_read_unlock(&vfsmount_lock);
3083 /* Unreachable from current root */
3085 error = prepend_unreachable(&cwd, &buflen);
3091 len = PAGE_SIZE + page - cwd;
3094 if (copy_to_user(buf, cwd, len))
3098 br_read_unlock(&vfsmount_lock);
3104 free_page((unsigned long) page);
3109 * Test whether new_dentry is a subdirectory of old_dentry.
3111 * Trivially implemented using the dcache structure
3115 * is_subdir - is new dentry a subdirectory of old_dentry
3116 * @new_dentry: new dentry
3117 * @old_dentry: old dentry
3119 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3120 * Returns 0 otherwise.
3121 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3124 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3129 if (new_dentry == old_dentry)
3133 /* for restarting inner loop in case of seq retry */
3134 seq = read_seqbegin(&rename_lock);
3136 * Need rcu_readlock to protect against the d_parent trashing
3140 if (d_ancestor(old_dentry, new_dentry))
3145 } while (read_seqretry(&rename_lock, seq));
3150 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3152 struct dentry *root = data;
3153 if (dentry != root) {
3154 if (d_unhashed(dentry) || !dentry->d_inode)
3157 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3158 dentry->d_flags |= DCACHE_GENOCIDE;
3159 dentry->d_lockref.count--;
3162 return D_WALK_CONTINUE;
3165 void d_genocide(struct dentry *parent)
3167 d_walk(parent, parent, d_genocide_kill, NULL);
3170 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3172 inode_dec_link_count(inode);
3173 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3174 !hlist_unhashed(&dentry->d_alias) ||
3175 !d_unlinked(dentry));
3176 spin_lock(&dentry->d_parent->d_lock);
3177 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3178 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3179 (unsigned long long)inode->i_ino);
3180 spin_unlock(&dentry->d_lock);
3181 spin_unlock(&dentry->d_parent->d_lock);
3182 d_instantiate(dentry, inode);
3184 EXPORT_SYMBOL(d_tmpfile);
3186 static __initdata unsigned long dhash_entries;
3187 static int __init set_dhash_entries(char *str)
3191 dhash_entries = simple_strtoul(str, &str, 0);
3194 __setup("dhash_entries=", set_dhash_entries);
3196 static void __init dcache_init_early(void)
3200 /* If hashes are distributed across NUMA nodes, defer
3201 * hash allocation until vmalloc space is available.
3207 alloc_large_system_hash("Dentry cache",
3208 sizeof(struct hlist_bl_head),
3217 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3218 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3221 static void __init dcache_init(void)
3226 * A constructor could be added for stable state like the lists,
3227 * but it is probably not worth it because of the cache nature
3230 dentry_cache = KMEM_CACHE(dentry,
3231 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3233 /* Hash may have been set up in dcache_init_early */
3238 alloc_large_system_hash("Dentry cache",
3239 sizeof(struct hlist_bl_head),
3248 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3249 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3252 /* SLAB cache for __getname() consumers */
3253 struct kmem_cache *names_cachep __read_mostly;
3254 EXPORT_SYMBOL(names_cachep);
3256 EXPORT_SYMBOL(d_genocide);
3258 void __init vfs_caches_init_early(void)
3260 dcache_init_early();
3264 void __init vfs_caches_init(unsigned long mempages)
3266 unsigned long reserve;
3268 /* Base hash sizes on available memory, with a reserve equal to
3269 150% of current kernel size */
3271 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3272 mempages -= reserve;
3274 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3275 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3279 files_init(mempages);