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|move_list) 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_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 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
385 * lose our last reference through the parent walk. In this case, we need to
386 * remove ourselves from the shrink list, not the LRU.
388 static void dentry_lru_del(struct dentry *dentry)
390 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
391 list_del_init(&dentry->d_lru);
392 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
396 if (!list_empty(&dentry->d_lru)) {
397 spin_lock(&dentry->d_sb->s_dentry_lru_lock);
398 __dentry_lru_del(dentry);
399 spin_unlock(&dentry->d_sb->s_dentry_lru_lock);
403 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
405 BUG_ON(dentry->d_flags & DCACHE_SHRINK_LIST);
407 spin_lock(&dentry->d_sb->s_dentry_lru_lock);
408 if (list_empty(&dentry->d_lru)) {
409 dentry->d_flags |= DCACHE_LRU_LIST;
410 list_add_tail(&dentry->d_lru, list);
412 list_move_tail(&dentry->d_lru, list);
413 dentry->d_sb->s_nr_dentry_unused--;
414 this_cpu_dec(nr_dentry_unused);
416 spin_unlock(&dentry->d_sb->s_dentry_lru_lock);
420 * d_kill - kill dentry and return parent
421 * @dentry: dentry to kill
422 * @parent: parent dentry
424 * The dentry must already be unhashed and removed from the LRU.
426 * If this is the root of the dentry tree, return NULL.
428 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
431 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
432 __releases(dentry->d_lock)
433 __releases(parent->d_lock)
434 __releases(dentry->d_inode->i_lock)
436 list_del(&dentry->d_u.d_child);
438 * Inform try_to_ascend() that we are no longer attached to the
441 dentry->d_flags |= DCACHE_DENTRY_KILLED;
443 spin_unlock(&parent->d_lock);
446 * dentry_iput drops the locks, at which point nobody (except
447 * transient RCU lookups) can reach this dentry.
454 * Unhash a dentry without inserting an RCU walk barrier or checking that
455 * dentry->d_lock is locked. The caller must take care of that, if
458 static void __d_shrink(struct dentry *dentry)
460 if (!d_unhashed(dentry)) {
461 struct hlist_bl_head *b;
462 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
463 b = &dentry->d_sb->s_anon;
465 b = d_hash(dentry->d_parent, dentry->d_name.hash);
468 __hlist_bl_del(&dentry->d_hash);
469 dentry->d_hash.pprev = NULL;
475 * d_drop - drop a dentry
476 * @dentry: dentry to drop
478 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
479 * be found through a VFS lookup any more. Note that this is different from
480 * deleting the dentry - d_delete will try to mark the dentry negative if
481 * possible, giving a successful _negative_ lookup, while d_drop will
482 * just make the cache lookup fail.
484 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
485 * reason (NFS timeouts or autofs deletes).
487 * __d_drop requires dentry->d_lock.
489 void __d_drop(struct dentry *dentry)
491 if (!d_unhashed(dentry)) {
493 dentry_rcuwalk_barrier(dentry);
496 EXPORT_SYMBOL(__d_drop);
498 void d_drop(struct dentry *dentry)
500 spin_lock(&dentry->d_lock);
502 spin_unlock(&dentry->d_lock);
504 EXPORT_SYMBOL(d_drop);
507 * Finish off a dentry we've decided to kill.
508 * dentry->d_lock must be held, returns with it unlocked.
509 * If ref is non-zero, then decrement the refcount too.
510 * Returns dentry requiring refcount drop, or NULL if we're done.
512 static inline struct dentry *
513 dentry_kill(struct dentry *dentry, int unlock_on_failure)
514 __releases(dentry->d_lock)
517 struct dentry *parent;
519 inode = dentry->d_inode;
520 if (inode && !spin_trylock(&inode->i_lock)) {
522 if (unlock_on_failure) {
523 spin_unlock(&dentry->d_lock);
526 return dentry; /* try again with same dentry */
531 parent = dentry->d_parent;
532 if (parent && !spin_trylock(&parent->d_lock)) {
534 spin_unlock(&inode->i_lock);
539 * The dentry is now unrecoverably dead to the world.
541 lockref_mark_dead(&dentry->d_lockref);
544 * inform the fs via d_prune that this dentry is about to be
545 * unhashed and destroyed.
547 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
548 dentry->d_op->d_prune(dentry);
550 dentry_lru_del(dentry);
551 /* if it was on the hash then remove it */
553 return d_kill(dentry, parent);
559 * This is complicated by the fact that we do not want to put
560 * dentries that are no longer on any hash chain on the unused
561 * list: we'd much rather just get rid of them immediately.
563 * However, that implies that we have to traverse the dentry
564 * tree upwards to the parents which might _also_ now be
565 * scheduled for deletion (it may have been only waiting for
566 * its last child to go away).
568 * This tail recursion is done by hand as we don't want to depend
569 * on the compiler to always get this right (gcc generally doesn't).
570 * Real recursion would eat up our stack space.
574 * dput - release a dentry
575 * @dentry: dentry to release
577 * Release a dentry. This will drop the usage count and if appropriate
578 * call the dentry unlink method as well as removing it from the queues and
579 * releasing its resources. If the parent dentries were scheduled for release
580 * they too may now get deleted.
582 void dput(struct dentry *dentry)
584 if (unlikely(!dentry))
588 if (lockref_put_or_lock(&dentry->d_lockref))
591 /* Unreachable? Get rid of it */
592 if (unlikely(d_unhashed(dentry)))
595 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
596 if (dentry->d_op->d_delete(dentry))
600 dentry->d_flags |= DCACHE_REFERENCED;
601 dentry_lru_add(dentry);
603 dentry->d_lockref.count--;
604 spin_unlock(&dentry->d_lock);
608 dentry = dentry_kill(dentry, 1);
615 * d_invalidate - invalidate a dentry
616 * @dentry: dentry to invalidate
618 * Try to invalidate the dentry if it turns out to be
619 * possible. If there are other dentries that can be
620 * reached through this one we can't delete it and we
621 * return -EBUSY. On success we return 0.
626 int d_invalidate(struct dentry * dentry)
629 * If it's already been dropped, return OK.
631 spin_lock(&dentry->d_lock);
632 if (d_unhashed(dentry)) {
633 spin_unlock(&dentry->d_lock);
637 * Check whether to do a partial shrink_dcache
638 * to get rid of unused child entries.
640 if (!list_empty(&dentry->d_subdirs)) {
641 spin_unlock(&dentry->d_lock);
642 shrink_dcache_parent(dentry);
643 spin_lock(&dentry->d_lock);
647 * Somebody else still using it?
649 * If it's a directory, we can't drop it
650 * for fear of somebody re-populating it
651 * with children (even though dropping it
652 * would make it unreachable from the root,
653 * we might still populate it if it was a
654 * working directory or similar).
655 * We also need to leave mountpoints alone,
658 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
659 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
660 spin_unlock(&dentry->d_lock);
666 spin_unlock(&dentry->d_lock);
669 EXPORT_SYMBOL(d_invalidate);
671 /* This must be called with d_lock held */
672 static inline void __dget_dlock(struct dentry *dentry)
674 dentry->d_lockref.count++;
677 static inline void __dget(struct dentry *dentry)
679 lockref_get(&dentry->d_lockref);
682 struct dentry *dget_parent(struct dentry *dentry)
688 * Do optimistic parent lookup without any
692 ret = ACCESS_ONCE(dentry->d_parent);
693 gotref = lockref_get_not_zero(&ret->d_lockref);
695 if (likely(gotref)) {
696 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
703 * Don't need rcu_dereference because we re-check it was correct under
707 ret = dentry->d_parent;
708 spin_lock(&ret->d_lock);
709 if (unlikely(ret != dentry->d_parent)) {
710 spin_unlock(&ret->d_lock);
715 BUG_ON(!ret->d_lockref.count);
716 ret->d_lockref.count++;
717 spin_unlock(&ret->d_lock);
720 EXPORT_SYMBOL(dget_parent);
723 * d_find_alias - grab a hashed alias of inode
724 * @inode: inode in question
725 * @want_discon: flag, used by d_splice_alias, to request
726 * that only a DISCONNECTED alias be returned.
728 * If inode has a hashed alias, or is a directory and has any alias,
729 * acquire the reference to alias and return it. Otherwise return NULL.
730 * Notice that if inode is a directory there can be only one alias and
731 * it can be unhashed only if it has no children, or if it is the root
734 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
735 * any other hashed alias over that one unless @want_discon is set,
736 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
738 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
740 struct dentry *alias, *discon_alias;
744 hlist_for_each_entry(alias, &inode->i_dentry, d_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)) {
749 discon_alias = alias;
750 } else if (!want_discon) {
752 spin_unlock(&alias->d_lock);
756 spin_unlock(&alias->d_lock);
759 alias = discon_alias;
760 spin_lock(&alias->d_lock);
761 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
762 if (IS_ROOT(alias) &&
763 (alias->d_flags & DCACHE_DISCONNECTED)) {
765 spin_unlock(&alias->d_lock);
769 spin_unlock(&alias->d_lock);
775 struct dentry *d_find_alias(struct inode *inode)
777 struct dentry *de = NULL;
779 if (!hlist_empty(&inode->i_dentry)) {
780 spin_lock(&inode->i_lock);
781 de = __d_find_alias(inode, 0);
782 spin_unlock(&inode->i_lock);
786 EXPORT_SYMBOL(d_find_alias);
789 * Try to kill dentries associated with this inode.
790 * WARNING: you must own a reference to inode.
792 void d_prune_aliases(struct inode *inode)
794 struct dentry *dentry;
796 spin_lock(&inode->i_lock);
797 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
798 spin_lock(&dentry->d_lock);
799 if (!dentry->d_lockref.count) {
801 * inform the fs via d_prune that this dentry
802 * is about to be unhashed and destroyed.
804 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
806 dentry->d_op->d_prune(dentry);
808 __dget_dlock(dentry);
810 spin_unlock(&dentry->d_lock);
811 spin_unlock(&inode->i_lock);
815 spin_unlock(&dentry->d_lock);
817 spin_unlock(&inode->i_lock);
819 EXPORT_SYMBOL(d_prune_aliases);
822 * Try to throw away a dentry - free the inode, dput the parent.
823 * Requires dentry->d_lock is held, and dentry->d_count == 0.
824 * Releases dentry->d_lock.
826 * This may fail if locks cannot be acquired no problem, just try again.
828 static struct dentry * try_prune_one_dentry(struct dentry *dentry)
829 __releases(dentry->d_lock)
831 struct dentry *parent;
833 parent = dentry_kill(dentry, 0);
835 * If dentry_kill returns NULL, we have nothing more to do.
836 * if it returns the same dentry, trylocks failed. In either
837 * case, just loop again.
839 * Otherwise, we need to prune ancestors too. This is necessary
840 * to prevent quadratic behavior of shrink_dcache_parent(), but
841 * is also expected to be beneficial in reducing dentry cache
846 if (parent == dentry)
849 /* Prune ancestors. */
852 if (lockref_put_or_lock(&dentry->d_lockref))
854 dentry = dentry_kill(dentry, 1);
859 static void shrink_dentry_list(struct list_head *list)
861 struct dentry *dentry;
865 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
866 if (&dentry->d_lru == list)
868 spin_lock(&dentry->d_lock);
869 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
870 spin_unlock(&dentry->d_lock);
875 * The dispose list is isolated and dentries are not accounted
876 * to the LRU here, so we can simply remove it from the list
877 * here regardless of whether it is referenced or not.
879 list_del_init(&dentry->d_lru);
880 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
883 * We found an inuse dentry which was not removed from
884 * the LRU because of laziness during lookup. Do not free it.
886 if (dentry->d_lockref.count) {
887 spin_unlock(&dentry->d_lock);
892 dentry = try_prune_one_dentry(dentry);
896 dentry->d_flags |= DCACHE_SHRINK_LIST;
897 list_add(&dentry->d_lru, list);
898 spin_unlock(&dentry->d_lock);
905 * prune_dcache_sb - shrink the dcache
907 * @count: number of entries to try to free
909 * Attempt to shrink the superblock dcache LRU by @count entries. This is
910 * done when we need more memory an called from the superblock shrinker
913 * This function may fail to free any resources if all the dentries are in
916 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan)
918 struct dentry *dentry;
919 LIST_HEAD(referenced);
924 spin_lock(&sb->s_dentry_lru_lock);
925 while (!list_empty(&sb->s_dentry_lru)) {
926 dentry = list_entry(sb->s_dentry_lru.prev,
927 struct dentry, d_lru);
928 BUG_ON(dentry->d_sb != sb);
930 if (!spin_trylock(&dentry->d_lock)) {
931 spin_unlock(&sb->s_dentry_lru_lock);
936 if (dentry->d_flags & DCACHE_REFERENCED) {
937 dentry->d_flags &= ~DCACHE_REFERENCED;
938 list_move(&dentry->d_lru, &referenced);
939 spin_unlock(&dentry->d_lock);
941 list_move(&dentry->d_lru, &tmp);
942 dentry->d_flags |= DCACHE_SHRINK_LIST;
943 this_cpu_dec(nr_dentry_unused);
944 sb->s_nr_dentry_unused--;
945 spin_unlock(&dentry->d_lock);
950 cond_resched_lock(&sb->s_dentry_lru_lock);
952 if (!list_empty(&referenced))
953 list_splice(&referenced, &sb->s_dentry_lru);
954 spin_unlock(&sb->s_dentry_lru_lock);
956 shrink_dentry_list(&tmp);
961 * Mark all the dentries as on being the dispose list so we don't think they are
962 * still on the LRU if we try to kill them from ascending the parent chain in
963 * try_prune_one_dentry() rather than directly from the dispose list.
967 struct list_head *dispose)
969 struct dentry *dentry;
972 list_for_each_entry_rcu(dentry, dispose, d_lru) {
973 spin_lock(&dentry->d_lock);
974 dentry->d_flags |= DCACHE_SHRINK_LIST;
975 spin_unlock(&dentry->d_lock);
978 shrink_dentry_list(dispose);
982 * shrink_dcache_sb - shrink dcache for a superblock
985 * Shrink the dcache for the specified super block. This is used to free
986 * the dcache before unmounting a file system.
988 void shrink_dcache_sb(struct super_block *sb)
992 spin_lock(&sb->s_dentry_lru_lock);
993 while (!list_empty(&sb->s_dentry_lru)) {
995 * account for removal here so we don't need to handle it later
996 * even though the dentry is no longer on the lru list.
998 list_splice_init(&sb->s_dentry_lru, &tmp);
999 this_cpu_sub(nr_dentry_unused, sb->s_nr_dentry_unused);
1000 sb->s_nr_dentry_unused = 0;
1001 spin_unlock(&sb->s_dentry_lru_lock);
1003 shrink_dcache_list(&tmp);
1005 spin_lock(&sb->s_dentry_lru_lock);
1007 spin_unlock(&sb->s_dentry_lru_lock);
1009 EXPORT_SYMBOL(shrink_dcache_sb);
1012 * destroy a single subtree of dentries for unmount
1013 * - see the comments on shrink_dcache_for_umount() for a description of the
1016 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
1018 struct dentry *parent;
1020 BUG_ON(!IS_ROOT(dentry));
1023 /* descend to the first leaf in the current subtree */
1024 while (!list_empty(&dentry->d_subdirs))
1025 dentry = list_entry(dentry->d_subdirs.next,
1026 struct dentry, d_u.d_child);
1028 /* consume the dentries from this leaf up through its parents
1029 * until we find one with children or run out altogether */
1031 struct inode *inode;
1034 * inform the fs that this dentry is about to be
1035 * unhashed and destroyed.
1037 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
1038 !d_unhashed(dentry))
1039 dentry->d_op->d_prune(dentry);
1041 dentry_lru_del(dentry);
1044 if (dentry->d_lockref.count != 0) {
1046 "BUG: Dentry %p{i=%lx,n=%s}"
1047 " still in use (%d)"
1048 " [unmount of %s %s]\n",
1051 dentry->d_inode->i_ino : 0UL,
1052 dentry->d_name.name,
1053 dentry->d_lockref.count,
1054 dentry->d_sb->s_type->name,
1055 dentry->d_sb->s_id);
1059 if (IS_ROOT(dentry)) {
1061 list_del(&dentry->d_u.d_child);
1063 parent = dentry->d_parent;
1064 parent->d_lockref.count--;
1065 list_del(&dentry->d_u.d_child);
1068 inode = dentry->d_inode;
1070 dentry->d_inode = NULL;
1071 hlist_del_init(&dentry->d_alias);
1072 if (dentry->d_op && dentry->d_op->d_iput)
1073 dentry->d_op->d_iput(dentry, inode);
1080 /* finished when we fall off the top of the tree,
1081 * otherwise we ascend to the parent and move to the
1082 * next sibling if there is one */
1086 } while (list_empty(&dentry->d_subdirs));
1088 dentry = list_entry(dentry->d_subdirs.next,
1089 struct dentry, d_u.d_child);
1094 * destroy the dentries attached to a superblock on unmounting
1095 * - we don't need to use dentry->d_lock because:
1096 * - the superblock is detached from all mountings and open files, so the
1097 * dentry trees will not be rearranged by the VFS
1098 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1099 * any dentries belonging to this superblock that it comes across
1100 * - the filesystem itself is no longer permitted to rearrange the dentries
1101 * in this superblock
1103 void shrink_dcache_for_umount(struct super_block *sb)
1105 struct dentry *dentry;
1107 if (down_read_trylock(&sb->s_umount))
1110 dentry = sb->s_root;
1112 dentry->d_lockref.count--;
1113 shrink_dcache_for_umount_subtree(dentry);
1115 while (!hlist_bl_empty(&sb->s_anon)) {
1116 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1117 shrink_dcache_for_umount_subtree(dentry);
1122 * This tries to ascend one level of parenthood, but
1123 * we can race with renaming, so we need to re-check
1124 * the parenthood after dropping the lock and check
1125 * that the sequence number still matches.
1127 static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
1129 struct dentry *new = old->d_parent;
1132 spin_unlock(&old->d_lock);
1133 spin_lock(&new->d_lock);
1136 * might go back up the wrong parent if we have had a rename
1139 if (new != old->d_parent ||
1140 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1141 need_seqretry(&rename_lock, seq)) {
1142 spin_unlock(&new->d_lock);
1150 * enum d_walk_ret - action to talke during tree walk
1151 * @D_WALK_CONTINUE: contrinue walk
1152 * @D_WALK_QUIT: quit walk
1153 * @D_WALK_NORETRY: quit when retry is needed
1154 * @D_WALK_SKIP: skip this dentry and its children
1164 * d_walk - walk the dentry tree
1165 * @parent: start of walk
1166 * @data: data passed to @enter() and @finish()
1167 * @enter: callback when first entering the dentry
1168 * @finish: callback when successfully finished the walk
1170 * The @enter() and @finish() callbacks are called with d_lock held.
1172 static void d_walk(struct dentry *parent, void *data,
1173 enum d_walk_ret (*enter)(void *, struct dentry *),
1174 void (*finish)(void *))
1176 struct dentry *this_parent;
1177 struct list_head *next;
1179 enum d_walk_ret ret;
1183 read_seqbegin_or_lock(&rename_lock, &seq);
1184 this_parent = parent;
1185 spin_lock(&this_parent->d_lock);
1187 ret = enter(data, this_parent);
1189 case D_WALK_CONTINUE:
1194 case D_WALK_NORETRY:
1199 next = this_parent->d_subdirs.next;
1201 while (next != &this_parent->d_subdirs) {
1202 struct list_head *tmp = next;
1203 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1206 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1208 ret = enter(data, dentry);
1210 case D_WALK_CONTINUE:
1213 spin_unlock(&dentry->d_lock);
1215 case D_WALK_NORETRY:
1219 spin_unlock(&dentry->d_lock);
1223 if (!list_empty(&dentry->d_subdirs)) {
1224 spin_unlock(&this_parent->d_lock);
1225 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1226 this_parent = dentry;
1227 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1230 spin_unlock(&dentry->d_lock);
1233 * All done at this level ... ascend and resume the search.
1235 if (this_parent != parent) {
1236 struct dentry *child = this_parent;
1237 this_parent = try_to_ascend(this_parent, seq);
1240 next = child->d_u.d_child.next;
1243 if (need_seqretry(&rename_lock, seq)) {
1244 spin_unlock(&this_parent->d_lock);
1251 spin_unlock(&this_parent->d_lock);
1252 done_seqretry(&rename_lock, seq);
1263 * Search for at least 1 mount point in the dentry's subdirs.
1264 * We descend to the next level whenever the d_subdirs
1265 * list is non-empty and continue searching.
1269 * have_submounts - check for mounts over a dentry
1270 * @parent: dentry to check.
1272 * Return true if the parent or its subdirectories contain
1276 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1279 if (d_mountpoint(dentry)) {
1283 return D_WALK_CONTINUE;
1286 int have_submounts(struct dentry *parent)
1290 d_walk(parent, &ret, check_mount, NULL);
1294 EXPORT_SYMBOL(have_submounts);
1297 * Called by mount code to set a mountpoint and check if the mountpoint is
1298 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1299 * subtree can become unreachable).
1301 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1302 * this reason take rename_lock and d_lock on dentry and ancestors.
1304 int d_set_mounted(struct dentry *dentry)
1308 write_seqlock(&rename_lock);
1309 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1310 /* Need exclusion wrt. check_submounts_and_drop() */
1311 spin_lock(&p->d_lock);
1312 if (unlikely(d_unhashed(p))) {
1313 spin_unlock(&p->d_lock);
1316 spin_unlock(&p->d_lock);
1318 spin_lock(&dentry->d_lock);
1319 if (!d_unlinked(dentry)) {
1320 dentry->d_flags |= DCACHE_MOUNTED;
1323 spin_unlock(&dentry->d_lock);
1325 write_sequnlock(&rename_lock);
1330 * Search the dentry child list of the specified parent,
1331 * and move any unused dentries to the end of the unused
1332 * list for prune_dcache(). We descend to the next level
1333 * whenever the d_subdirs list is non-empty and continue
1336 * It returns zero iff there are no unused children,
1337 * otherwise it returns the number of children moved to
1338 * the end of the unused list. This may not be the total
1339 * number of unused children, because select_parent can
1340 * drop the lock and return early due to latency
1344 struct select_data {
1345 struct dentry *start;
1346 struct list_head dispose;
1350 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1352 struct select_data *data = _data;
1353 enum d_walk_ret ret = D_WALK_CONTINUE;
1355 if (data->start == dentry)
1359 * move only zero ref count dentries to the dispose list.
1361 * Those which are presently on the shrink list, being processed
1362 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1363 * loop in shrink_dcache_parent() might not make any progress
1366 if (dentry->d_lockref.count) {
1367 dentry_lru_del(dentry);
1368 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1369 dentry_lru_move_list(dentry, &data->dispose);
1370 dentry->d_flags |= DCACHE_SHRINK_LIST;
1372 ret = D_WALK_NORETRY;
1375 * We can return to the caller if we have found some (this
1376 * ensures forward progress). We'll be coming back to find
1379 if (data->found && need_resched())
1386 * shrink_dcache_parent - prune dcache
1387 * @parent: parent of entries to prune
1389 * Prune the dcache to remove unused children of the parent dentry.
1391 void shrink_dcache_parent(struct dentry *parent)
1394 struct select_data data;
1396 INIT_LIST_HEAD(&data.dispose);
1397 data.start = parent;
1400 d_walk(parent, &data, select_collect, NULL);
1404 shrink_dentry_list(&data.dispose);
1408 EXPORT_SYMBOL(shrink_dcache_parent);
1410 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1412 struct select_data *data = _data;
1414 if (d_mountpoint(dentry)) {
1415 data->found = -EBUSY;
1419 return select_collect(_data, dentry);
1422 static void check_and_drop(void *_data)
1424 struct select_data *data = _data;
1426 if (d_mountpoint(data->start))
1427 data->found = -EBUSY;
1429 __d_drop(data->start);
1433 * check_submounts_and_drop - prune dcache, check for submounts and drop
1435 * All done as a single atomic operation relative to has_unlinked_ancestor().
1436 * Returns 0 if successfully unhashed @parent. If there were submounts then
1439 * @dentry: dentry to prune and drop
1441 int check_submounts_and_drop(struct dentry *dentry)
1445 /* Negative dentries can be dropped without further checks */
1446 if (!dentry->d_inode) {
1452 struct select_data data;
1454 INIT_LIST_HEAD(&data.dispose);
1455 data.start = dentry;
1458 d_walk(dentry, &data, check_and_collect, check_and_drop);
1461 if (!list_empty(&data.dispose))
1462 shrink_dentry_list(&data.dispose);
1473 EXPORT_SYMBOL(check_submounts_and_drop);
1476 * __d_alloc - allocate a dcache entry
1477 * @sb: filesystem it will belong to
1478 * @name: qstr of the name
1480 * Allocates a dentry. It returns %NULL if there is insufficient memory
1481 * available. On a success the dentry is returned. The name passed in is
1482 * copied and the copy passed in may be reused after this call.
1485 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1487 struct dentry *dentry;
1490 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1495 * We guarantee that the inline name is always NUL-terminated.
1496 * This way the memcpy() done by the name switching in rename
1497 * will still always have a NUL at the end, even if we might
1498 * be overwriting an internal NUL character
1500 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1501 if (name->len > DNAME_INLINE_LEN-1) {
1502 dname = kmalloc(name->len + 1, GFP_KERNEL);
1504 kmem_cache_free(dentry_cache, dentry);
1508 dname = dentry->d_iname;
1511 dentry->d_name.len = name->len;
1512 dentry->d_name.hash = name->hash;
1513 memcpy(dname, name->name, name->len);
1514 dname[name->len] = 0;
1516 /* Make sure we always see the terminating NUL character */
1518 dentry->d_name.name = dname;
1520 dentry->d_lockref.count = 1;
1521 dentry->d_flags = 0;
1522 spin_lock_init(&dentry->d_lock);
1523 seqcount_init(&dentry->d_seq);
1524 dentry->d_inode = NULL;
1525 dentry->d_parent = dentry;
1527 dentry->d_op = NULL;
1528 dentry->d_fsdata = NULL;
1529 INIT_HLIST_BL_NODE(&dentry->d_hash);
1530 INIT_LIST_HEAD(&dentry->d_lru);
1531 INIT_LIST_HEAD(&dentry->d_subdirs);
1532 INIT_HLIST_NODE(&dentry->d_alias);
1533 INIT_LIST_HEAD(&dentry->d_u.d_child);
1534 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1536 this_cpu_inc(nr_dentry);
1542 * d_alloc - allocate a dcache entry
1543 * @parent: parent of entry to allocate
1544 * @name: qstr of the name
1546 * Allocates a dentry. It returns %NULL if there is insufficient memory
1547 * available. On a success the dentry is returned. The name passed in is
1548 * copied and the copy passed in may be reused after this call.
1550 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1552 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1556 spin_lock(&parent->d_lock);
1558 * don't need child lock because it is not subject
1559 * to concurrency here
1561 __dget_dlock(parent);
1562 dentry->d_parent = parent;
1563 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1564 spin_unlock(&parent->d_lock);
1568 EXPORT_SYMBOL(d_alloc);
1570 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1572 struct dentry *dentry = __d_alloc(sb, name);
1574 dentry->d_flags |= DCACHE_DISCONNECTED;
1577 EXPORT_SYMBOL(d_alloc_pseudo);
1579 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1584 q.len = strlen(name);
1585 q.hash = full_name_hash(q.name, q.len);
1586 return d_alloc(parent, &q);
1588 EXPORT_SYMBOL(d_alloc_name);
1590 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1592 WARN_ON_ONCE(dentry->d_op);
1593 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1595 DCACHE_OP_REVALIDATE |
1596 DCACHE_OP_WEAK_REVALIDATE |
1597 DCACHE_OP_DELETE ));
1602 dentry->d_flags |= DCACHE_OP_HASH;
1604 dentry->d_flags |= DCACHE_OP_COMPARE;
1605 if (op->d_revalidate)
1606 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1607 if (op->d_weak_revalidate)
1608 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1610 dentry->d_flags |= DCACHE_OP_DELETE;
1612 dentry->d_flags |= DCACHE_OP_PRUNE;
1615 EXPORT_SYMBOL(d_set_d_op);
1617 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1619 spin_lock(&dentry->d_lock);
1621 if (unlikely(IS_AUTOMOUNT(inode)))
1622 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1623 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1625 dentry->d_inode = inode;
1626 dentry_rcuwalk_barrier(dentry);
1627 spin_unlock(&dentry->d_lock);
1628 fsnotify_d_instantiate(dentry, inode);
1632 * d_instantiate - fill in inode information for a dentry
1633 * @entry: dentry to complete
1634 * @inode: inode to attach to this dentry
1636 * Fill in inode information in the entry.
1638 * This turns negative dentries into productive full members
1641 * NOTE! This assumes that the inode count has been incremented
1642 * (or otherwise set) by the caller to indicate that it is now
1643 * in use by the dcache.
1646 void d_instantiate(struct dentry *entry, struct inode * inode)
1648 BUG_ON(!hlist_unhashed(&entry->d_alias));
1650 spin_lock(&inode->i_lock);
1651 __d_instantiate(entry, inode);
1653 spin_unlock(&inode->i_lock);
1654 security_d_instantiate(entry, inode);
1656 EXPORT_SYMBOL(d_instantiate);
1659 * d_instantiate_unique - instantiate a non-aliased dentry
1660 * @entry: dentry to instantiate
1661 * @inode: inode to attach to this dentry
1663 * Fill in inode information in the entry. On success, it returns NULL.
1664 * If an unhashed alias of "entry" already exists, then we return the
1665 * aliased dentry instead and drop one reference to inode.
1667 * Note that in order to avoid conflicts with rename() etc, the caller
1668 * had better be holding the parent directory semaphore.
1670 * This also assumes that the inode count has been incremented
1671 * (or otherwise set) by the caller to indicate that it is now
1672 * in use by the dcache.
1674 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1675 struct inode *inode)
1677 struct dentry *alias;
1678 int len = entry->d_name.len;
1679 const char *name = entry->d_name.name;
1680 unsigned int hash = entry->d_name.hash;
1683 __d_instantiate(entry, NULL);
1687 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1689 * Don't need alias->d_lock here, because aliases with
1690 * d_parent == entry->d_parent are not subject to name or
1691 * parent changes, because the parent inode i_mutex is held.
1693 if (alias->d_name.hash != hash)
1695 if (alias->d_parent != entry->d_parent)
1697 if (alias->d_name.len != len)
1699 if (dentry_cmp(alias, name, len))
1705 __d_instantiate(entry, inode);
1709 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1711 struct dentry *result;
1713 BUG_ON(!hlist_unhashed(&entry->d_alias));
1716 spin_lock(&inode->i_lock);
1717 result = __d_instantiate_unique(entry, inode);
1719 spin_unlock(&inode->i_lock);
1722 security_d_instantiate(entry, inode);
1726 BUG_ON(!d_unhashed(result));
1731 EXPORT_SYMBOL(d_instantiate_unique);
1733 struct dentry *d_make_root(struct inode *root_inode)
1735 struct dentry *res = NULL;
1738 static const struct qstr name = QSTR_INIT("/", 1);
1740 res = __d_alloc(root_inode->i_sb, &name);
1742 d_instantiate(res, root_inode);
1748 EXPORT_SYMBOL(d_make_root);
1750 static struct dentry * __d_find_any_alias(struct inode *inode)
1752 struct dentry *alias;
1754 if (hlist_empty(&inode->i_dentry))
1756 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1762 * d_find_any_alias - find any alias for a given inode
1763 * @inode: inode to find an alias for
1765 * If any aliases exist for the given inode, take and return a
1766 * reference for one of them. If no aliases exist, return %NULL.
1768 struct dentry *d_find_any_alias(struct inode *inode)
1772 spin_lock(&inode->i_lock);
1773 de = __d_find_any_alias(inode);
1774 spin_unlock(&inode->i_lock);
1777 EXPORT_SYMBOL(d_find_any_alias);
1780 * d_obtain_alias - find or allocate a dentry for a given inode
1781 * @inode: inode to allocate the dentry for
1783 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1784 * similar open by handle operations. The returned dentry may be anonymous,
1785 * or may have a full name (if the inode was already in the cache).
1787 * When called on a directory inode, we must ensure that the inode only ever
1788 * has one dentry. If a dentry is found, that is returned instead of
1789 * allocating a new one.
1791 * On successful return, the reference to the inode has been transferred
1792 * to the dentry. In case of an error the reference on the inode is released.
1793 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1794 * be passed in and will be the error will be propagate to the return value,
1795 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1797 struct dentry *d_obtain_alias(struct inode *inode)
1799 static const struct qstr anonstring = QSTR_INIT("/", 1);
1804 return ERR_PTR(-ESTALE);
1806 return ERR_CAST(inode);
1808 res = d_find_any_alias(inode);
1812 tmp = __d_alloc(inode->i_sb, &anonstring);
1814 res = ERR_PTR(-ENOMEM);
1818 spin_lock(&inode->i_lock);
1819 res = __d_find_any_alias(inode);
1821 spin_unlock(&inode->i_lock);
1826 /* attach a disconnected dentry */
1827 spin_lock(&tmp->d_lock);
1828 tmp->d_inode = inode;
1829 tmp->d_flags |= DCACHE_DISCONNECTED;
1830 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1831 hlist_bl_lock(&tmp->d_sb->s_anon);
1832 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1833 hlist_bl_unlock(&tmp->d_sb->s_anon);
1834 spin_unlock(&tmp->d_lock);
1835 spin_unlock(&inode->i_lock);
1836 security_d_instantiate(tmp, inode);
1841 if (res && !IS_ERR(res))
1842 security_d_instantiate(res, inode);
1846 EXPORT_SYMBOL(d_obtain_alias);
1849 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1850 * @inode: the inode which may have a disconnected dentry
1851 * @dentry: a negative dentry which we want to point to the inode.
1853 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1854 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1855 * and return it, else simply d_add the inode to the dentry and return NULL.
1857 * This is needed in the lookup routine of any filesystem that is exportable
1858 * (via knfsd) so that we can build dcache paths to directories effectively.
1860 * If a dentry was found and moved, then it is returned. Otherwise NULL
1861 * is returned. This matches the expected return value of ->lookup.
1863 * Cluster filesystems may call this function with a negative, hashed dentry.
1864 * In that case, we know that the inode will be a regular file, and also this
1865 * will only occur during atomic_open. So we need to check for the dentry
1866 * being already hashed only in the final case.
1868 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1870 struct dentry *new = NULL;
1873 return ERR_CAST(inode);
1875 if (inode && S_ISDIR(inode->i_mode)) {
1876 spin_lock(&inode->i_lock);
1877 new = __d_find_alias(inode, 1);
1879 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1880 spin_unlock(&inode->i_lock);
1881 security_d_instantiate(new, inode);
1882 d_move(new, dentry);
1885 /* already taking inode->i_lock, so d_add() by hand */
1886 __d_instantiate(dentry, inode);
1887 spin_unlock(&inode->i_lock);
1888 security_d_instantiate(dentry, inode);
1892 d_instantiate(dentry, inode);
1893 if (d_unhashed(dentry))
1898 EXPORT_SYMBOL(d_splice_alias);
1901 * d_add_ci - lookup or allocate new dentry with case-exact name
1902 * @inode: the inode case-insensitive lookup has found
1903 * @dentry: the negative dentry that was passed to the parent's lookup func
1904 * @name: the case-exact name to be associated with the returned dentry
1906 * This is to avoid filling the dcache with case-insensitive names to the
1907 * same inode, only the actual correct case is stored in the dcache for
1908 * case-insensitive filesystems.
1910 * For a case-insensitive lookup match and if the the case-exact dentry
1911 * already exists in in the dcache, use it and return it.
1913 * If no entry exists with the exact case name, allocate new dentry with
1914 * the exact case, and return the spliced entry.
1916 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1919 struct dentry *found;
1923 * First check if a dentry matching the name already exists,
1924 * if not go ahead and create it now.
1926 found = d_hash_and_lookup(dentry->d_parent, name);
1927 if (unlikely(IS_ERR(found)))
1930 new = d_alloc(dentry->d_parent, name);
1932 found = ERR_PTR(-ENOMEM);
1936 found = d_splice_alias(inode, new);
1945 * If a matching dentry exists, and it's not negative use it.
1947 * Decrement the reference count to balance the iget() done
1950 if (found->d_inode) {
1951 if (unlikely(found->d_inode != inode)) {
1952 /* This can't happen because bad inodes are unhashed. */
1953 BUG_ON(!is_bad_inode(inode));
1954 BUG_ON(!is_bad_inode(found->d_inode));
1961 * Negative dentry: instantiate it unless the inode is a directory and
1962 * already has a dentry.
1964 new = d_splice_alias(inode, found);
1975 EXPORT_SYMBOL(d_add_ci);
1978 * Do the slow-case of the dentry name compare.
1980 * Unlike the dentry_cmp() function, we need to atomically
1981 * load the name and length information, so that the
1982 * filesystem can rely on them, and can use the 'name' and
1983 * 'len' information without worrying about walking off the
1984 * end of memory etc.
1986 * Thus the read_seqcount_retry() and the "duplicate" info
1987 * in arguments (the low-level filesystem should not look
1988 * at the dentry inode or name contents directly, since
1989 * rename can change them while we're in RCU mode).
1991 enum slow_d_compare {
1997 static noinline enum slow_d_compare slow_dentry_cmp(
1998 const struct dentry *parent,
1999 struct dentry *dentry,
2001 const struct qstr *name)
2003 int tlen = dentry->d_name.len;
2004 const char *tname = dentry->d_name.name;
2006 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2008 return D_COMP_SEQRETRY;
2010 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2011 return D_COMP_NOMATCH;
2016 * __d_lookup_rcu - search for a dentry (racy, store-free)
2017 * @parent: parent dentry
2018 * @name: qstr of name we wish to find
2019 * @seqp: returns d_seq value at the point where the dentry was found
2020 * Returns: dentry, or NULL
2022 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2023 * resolution (store-free path walking) design described in
2024 * Documentation/filesystems/path-lookup.txt.
2026 * This is not to be used outside core vfs.
2028 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2029 * held, and rcu_read_lock held. The returned dentry must not be stored into
2030 * without taking d_lock and checking d_seq sequence count against @seq
2033 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2036 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2037 * the returned dentry, so long as its parent's seqlock is checked after the
2038 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2039 * is formed, giving integrity down the path walk.
2041 * NOTE! The caller *has* to check the resulting dentry against the sequence
2042 * number we've returned before using any of the resulting dentry state!
2044 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2045 const struct qstr *name,
2048 u64 hashlen = name->hash_len;
2049 const unsigned char *str = name->name;
2050 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2051 struct hlist_bl_node *node;
2052 struct dentry *dentry;
2055 * Note: There is significant duplication with __d_lookup_rcu which is
2056 * required to prevent single threaded performance regressions
2057 * especially on architectures where smp_rmb (in seqcounts) are costly.
2058 * Keep the two functions in sync.
2062 * The hash list is protected using RCU.
2064 * Carefully use d_seq when comparing a candidate dentry, to avoid
2065 * races with d_move().
2067 * It is possible that concurrent renames can mess up our list
2068 * walk here and result in missing our dentry, resulting in the
2069 * false-negative result. d_lookup() protects against concurrent
2070 * renames using rename_lock seqlock.
2072 * See Documentation/filesystems/path-lookup.txt for more details.
2074 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2079 * The dentry sequence count protects us from concurrent
2080 * renames, and thus protects parent and name fields.
2082 * The caller must perform a seqcount check in order
2083 * to do anything useful with the returned dentry.
2085 * NOTE! We do a "raw" seqcount_begin here. That means that
2086 * we don't wait for the sequence count to stabilize if it
2087 * is in the middle of a sequence change. If we do the slow
2088 * dentry compare, we will do seqretries until it is stable,
2089 * and if we end up with a successful lookup, we actually
2090 * want to exit RCU lookup anyway.
2092 seq = raw_seqcount_begin(&dentry->d_seq);
2093 if (dentry->d_parent != parent)
2095 if (d_unhashed(dentry))
2098 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2099 if (dentry->d_name.hash != hashlen_hash(hashlen))
2102 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2105 case D_COMP_NOMATCH:
2112 if (dentry->d_name.hash_len != hashlen)
2115 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2122 * d_lookup - search for a dentry
2123 * @parent: parent dentry
2124 * @name: qstr of name we wish to find
2125 * Returns: dentry, or NULL
2127 * d_lookup searches the children of the parent dentry for the name in
2128 * question. If the dentry is found its reference count is incremented and the
2129 * dentry is returned. The caller must use dput to free the entry when it has
2130 * finished using it. %NULL is returned if the dentry does not exist.
2132 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2134 struct dentry *dentry;
2138 seq = read_seqbegin(&rename_lock);
2139 dentry = __d_lookup(parent, name);
2142 } while (read_seqretry(&rename_lock, seq));
2145 EXPORT_SYMBOL(d_lookup);
2148 * __d_lookup - search for a dentry (racy)
2149 * @parent: parent dentry
2150 * @name: qstr of name we wish to find
2151 * Returns: dentry, or NULL
2153 * __d_lookup is like d_lookup, however it may (rarely) return a
2154 * false-negative result due to unrelated rename activity.
2156 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2157 * however it must be used carefully, eg. with a following d_lookup in
2158 * the case of failure.
2160 * __d_lookup callers must be commented.
2162 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2164 unsigned int len = name->len;
2165 unsigned int hash = name->hash;
2166 const unsigned char *str = name->name;
2167 struct hlist_bl_head *b = d_hash(parent, hash);
2168 struct hlist_bl_node *node;
2169 struct dentry *found = NULL;
2170 struct dentry *dentry;
2173 * Note: There is significant duplication with __d_lookup_rcu which is
2174 * required to prevent single threaded performance regressions
2175 * especially on architectures where smp_rmb (in seqcounts) are costly.
2176 * Keep the two functions in sync.
2180 * The hash list is protected using RCU.
2182 * Take d_lock when comparing a candidate dentry, to avoid races
2185 * It is possible that concurrent renames can mess up our list
2186 * walk here and result in missing our dentry, resulting in the
2187 * false-negative result. d_lookup() protects against concurrent
2188 * renames using rename_lock seqlock.
2190 * See Documentation/filesystems/path-lookup.txt for more details.
2194 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2196 if (dentry->d_name.hash != hash)
2199 spin_lock(&dentry->d_lock);
2200 if (dentry->d_parent != parent)
2202 if (d_unhashed(dentry))
2206 * It is safe to compare names since d_move() cannot
2207 * change the qstr (protected by d_lock).
2209 if (parent->d_flags & DCACHE_OP_COMPARE) {
2210 int tlen = dentry->d_name.len;
2211 const char *tname = dentry->d_name.name;
2212 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2215 if (dentry->d_name.len != len)
2217 if (dentry_cmp(dentry, str, len))
2221 dentry->d_lockref.count++;
2223 spin_unlock(&dentry->d_lock);
2226 spin_unlock(&dentry->d_lock);
2234 * d_hash_and_lookup - hash the qstr then search for a dentry
2235 * @dir: Directory to search in
2236 * @name: qstr of name we wish to find
2238 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2240 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2243 * Check for a fs-specific hash function. Note that we must
2244 * calculate the standard hash first, as the d_op->d_hash()
2245 * routine may choose to leave the hash value unchanged.
2247 name->hash = full_name_hash(name->name, name->len);
2248 if (dir->d_flags & DCACHE_OP_HASH) {
2249 int err = dir->d_op->d_hash(dir, name);
2250 if (unlikely(err < 0))
2251 return ERR_PTR(err);
2253 return d_lookup(dir, name);
2255 EXPORT_SYMBOL(d_hash_and_lookup);
2258 * d_validate - verify dentry provided from insecure source (deprecated)
2259 * @dentry: The dentry alleged to be valid child of @dparent
2260 * @dparent: The parent dentry (known to be valid)
2262 * An insecure source has sent us a dentry, here we verify it and dget() it.
2263 * This is used by ncpfs in its readdir implementation.
2264 * Zero is returned in the dentry is invalid.
2266 * This function is slow for big directories, and deprecated, do not use it.
2268 int d_validate(struct dentry *dentry, struct dentry *dparent)
2270 struct dentry *child;
2272 spin_lock(&dparent->d_lock);
2273 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2274 if (dentry == child) {
2275 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2276 __dget_dlock(dentry);
2277 spin_unlock(&dentry->d_lock);
2278 spin_unlock(&dparent->d_lock);
2282 spin_unlock(&dparent->d_lock);
2286 EXPORT_SYMBOL(d_validate);
2289 * When a file is deleted, we have two options:
2290 * - turn this dentry into a negative dentry
2291 * - unhash this dentry and free it.
2293 * Usually, we want to just turn this into
2294 * a negative dentry, but if anybody else is
2295 * currently using the dentry or the inode
2296 * we can't do that and we fall back on removing
2297 * it from the hash queues and waiting for
2298 * it to be deleted later when it has no users
2302 * d_delete - delete a dentry
2303 * @dentry: The dentry to delete
2305 * Turn the dentry into a negative dentry if possible, otherwise
2306 * remove it from the hash queues so it can be deleted later
2309 void d_delete(struct dentry * dentry)
2311 struct inode *inode;
2314 * Are we the only user?
2317 spin_lock(&dentry->d_lock);
2318 inode = dentry->d_inode;
2319 isdir = S_ISDIR(inode->i_mode);
2320 if (dentry->d_lockref.count == 1) {
2321 if (!spin_trylock(&inode->i_lock)) {
2322 spin_unlock(&dentry->d_lock);
2326 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2327 dentry_unlink_inode(dentry);
2328 fsnotify_nameremove(dentry, isdir);
2332 if (!d_unhashed(dentry))
2335 spin_unlock(&dentry->d_lock);
2337 fsnotify_nameremove(dentry, isdir);
2339 EXPORT_SYMBOL(d_delete);
2341 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2343 BUG_ON(!d_unhashed(entry));
2345 entry->d_flags |= DCACHE_RCUACCESS;
2346 hlist_bl_add_head_rcu(&entry->d_hash, b);
2350 static void _d_rehash(struct dentry * entry)
2352 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2356 * d_rehash - add an entry back to the hash
2357 * @entry: dentry to add to the hash
2359 * Adds a dentry to the hash according to its name.
2362 void d_rehash(struct dentry * entry)
2364 spin_lock(&entry->d_lock);
2366 spin_unlock(&entry->d_lock);
2368 EXPORT_SYMBOL(d_rehash);
2371 * dentry_update_name_case - update case insensitive dentry with a new name
2372 * @dentry: dentry to be updated
2375 * Update a case insensitive dentry with new case of name.
2377 * dentry must have been returned by d_lookup with name @name. Old and new
2378 * name lengths must match (ie. no d_compare which allows mismatched name
2381 * Parent inode i_mutex must be held over d_lookup and into this call (to
2382 * keep renames and concurrent inserts, and readdir(2) away).
2384 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2386 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2387 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2389 spin_lock(&dentry->d_lock);
2390 write_seqcount_begin(&dentry->d_seq);
2391 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2392 write_seqcount_end(&dentry->d_seq);
2393 spin_unlock(&dentry->d_lock);
2395 EXPORT_SYMBOL(dentry_update_name_case);
2397 static void switch_names(struct dentry *dentry, struct dentry *target)
2399 if (dname_external(target)) {
2400 if (dname_external(dentry)) {
2402 * Both external: swap the pointers
2404 swap(target->d_name.name, dentry->d_name.name);
2407 * dentry:internal, target:external. Steal target's
2408 * storage and make target internal.
2410 memcpy(target->d_iname, dentry->d_name.name,
2411 dentry->d_name.len + 1);
2412 dentry->d_name.name = target->d_name.name;
2413 target->d_name.name = target->d_iname;
2416 if (dname_external(dentry)) {
2418 * dentry:external, target:internal. Give dentry's
2419 * storage to target and make dentry internal
2421 memcpy(dentry->d_iname, target->d_name.name,
2422 target->d_name.len + 1);
2423 target->d_name.name = dentry->d_name.name;
2424 dentry->d_name.name = dentry->d_iname;
2427 * Both are internal. Just copy target to dentry
2429 memcpy(dentry->d_iname, target->d_name.name,
2430 target->d_name.len + 1);
2431 dentry->d_name.len = target->d_name.len;
2435 swap(dentry->d_name.len, target->d_name.len);
2438 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2441 * XXXX: do we really need to take target->d_lock?
2443 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2444 spin_lock(&target->d_parent->d_lock);
2446 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2447 spin_lock(&dentry->d_parent->d_lock);
2448 spin_lock_nested(&target->d_parent->d_lock,
2449 DENTRY_D_LOCK_NESTED);
2451 spin_lock(&target->d_parent->d_lock);
2452 spin_lock_nested(&dentry->d_parent->d_lock,
2453 DENTRY_D_LOCK_NESTED);
2456 if (target < dentry) {
2457 spin_lock_nested(&target->d_lock, 2);
2458 spin_lock_nested(&dentry->d_lock, 3);
2460 spin_lock_nested(&dentry->d_lock, 2);
2461 spin_lock_nested(&target->d_lock, 3);
2465 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2466 struct dentry *target)
2468 if (target->d_parent != dentry->d_parent)
2469 spin_unlock(&dentry->d_parent->d_lock);
2470 if (target->d_parent != target)
2471 spin_unlock(&target->d_parent->d_lock);
2475 * When switching names, the actual string doesn't strictly have to
2476 * be preserved in the target - because we're dropping the target
2477 * anyway. As such, we can just do a simple memcpy() to copy over
2478 * the new name before we switch.
2480 * Note that we have to be a lot more careful about getting the hash
2481 * switched - we have to switch the hash value properly even if it
2482 * then no longer matches the actual (corrupted) string of the target.
2483 * The hash value has to match the hash queue that the dentry is on..
2486 * __d_move - move a dentry
2487 * @dentry: entry to move
2488 * @target: new dentry
2490 * Update the dcache to reflect the move of a file name. Negative
2491 * dcache entries should not be moved in this way. Caller must hold
2492 * rename_lock, the i_mutex of the source and target directories,
2493 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2495 static void __d_move(struct dentry * dentry, struct dentry * target)
2497 if (!dentry->d_inode)
2498 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2500 BUG_ON(d_ancestor(dentry, target));
2501 BUG_ON(d_ancestor(target, dentry));
2503 dentry_lock_for_move(dentry, target);
2505 write_seqcount_begin(&dentry->d_seq);
2506 write_seqcount_begin(&target->d_seq);
2508 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2511 * Move the dentry to the target hash queue. Don't bother checking
2512 * for the same hash queue because of how unlikely it is.
2515 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2517 /* Unhash the target: dput() will then get rid of it */
2520 list_del(&dentry->d_u.d_child);
2521 list_del(&target->d_u.d_child);
2523 /* Switch the names.. */
2524 switch_names(dentry, target);
2525 swap(dentry->d_name.hash, target->d_name.hash);
2527 /* ... and switch the parents */
2528 if (IS_ROOT(dentry)) {
2529 dentry->d_parent = target->d_parent;
2530 target->d_parent = target;
2531 INIT_LIST_HEAD(&target->d_u.d_child);
2533 swap(dentry->d_parent, target->d_parent);
2535 /* And add them back to the (new) parent lists */
2536 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2539 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2541 write_seqcount_end(&target->d_seq);
2542 write_seqcount_end(&dentry->d_seq);
2544 dentry_unlock_parents_for_move(dentry, target);
2545 spin_unlock(&target->d_lock);
2546 fsnotify_d_move(dentry);
2547 spin_unlock(&dentry->d_lock);
2551 * d_move - move a dentry
2552 * @dentry: entry to move
2553 * @target: new dentry
2555 * Update the dcache to reflect the move of a file name. Negative
2556 * dcache entries should not be moved in this way. See the locking
2557 * requirements for __d_move.
2559 void d_move(struct dentry *dentry, struct dentry *target)
2561 write_seqlock(&rename_lock);
2562 __d_move(dentry, target);
2563 write_sequnlock(&rename_lock);
2565 EXPORT_SYMBOL(d_move);
2568 * d_ancestor - search for an ancestor
2569 * @p1: ancestor dentry
2572 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2573 * an ancestor of p2, else NULL.
2575 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2579 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2580 if (p->d_parent == p1)
2587 * This helper attempts to cope with remotely renamed directories
2589 * It assumes that the caller is already holding
2590 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2592 * Note: If ever the locking in lock_rename() changes, then please
2593 * remember to update this too...
2595 static struct dentry *__d_unalias(struct inode *inode,
2596 struct dentry *dentry, struct dentry *alias)
2598 struct mutex *m1 = NULL, *m2 = NULL;
2599 struct dentry *ret = ERR_PTR(-EBUSY);
2601 /* If alias and dentry share a parent, then no extra locks required */
2602 if (alias->d_parent == dentry->d_parent)
2605 /* See lock_rename() */
2606 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2608 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2609 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2611 m2 = &alias->d_parent->d_inode->i_mutex;
2613 if (likely(!d_mountpoint(alias))) {
2614 __d_move(alias, dentry);
2618 spin_unlock(&inode->i_lock);
2627 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2628 * named dentry in place of the dentry to be replaced.
2629 * returns with anon->d_lock held!
2631 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2633 struct dentry *dparent;
2635 dentry_lock_for_move(anon, dentry);
2637 write_seqcount_begin(&dentry->d_seq);
2638 write_seqcount_begin(&anon->d_seq);
2640 dparent = dentry->d_parent;
2642 switch_names(dentry, anon);
2643 swap(dentry->d_name.hash, anon->d_name.hash);
2645 dentry->d_parent = dentry;
2646 list_del_init(&dentry->d_u.d_child);
2647 anon->d_parent = dparent;
2648 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2650 write_seqcount_end(&dentry->d_seq);
2651 write_seqcount_end(&anon->d_seq);
2653 dentry_unlock_parents_for_move(anon, dentry);
2654 spin_unlock(&dentry->d_lock);
2656 /* anon->d_lock still locked, returns locked */
2657 anon->d_flags &= ~DCACHE_DISCONNECTED;
2661 * d_materialise_unique - introduce an inode into the tree
2662 * @dentry: candidate dentry
2663 * @inode: inode to bind to the dentry, to which aliases may be attached
2665 * Introduces an dentry into the tree, substituting an extant disconnected
2666 * root directory alias in its place if there is one. Caller must hold the
2667 * i_mutex of the parent directory.
2669 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2671 struct dentry *actual;
2673 BUG_ON(!d_unhashed(dentry));
2677 __d_instantiate(dentry, NULL);
2682 spin_lock(&inode->i_lock);
2684 if (S_ISDIR(inode->i_mode)) {
2685 struct dentry *alias;
2687 /* Does an aliased dentry already exist? */
2688 alias = __d_find_alias(inode, 0);
2691 write_seqlock(&rename_lock);
2693 if (d_ancestor(alias, dentry)) {
2694 /* Check for loops */
2695 actual = ERR_PTR(-ELOOP);
2696 spin_unlock(&inode->i_lock);
2697 } else if (IS_ROOT(alias)) {
2698 /* Is this an anonymous mountpoint that we
2699 * could splice into our tree? */
2700 __d_materialise_dentry(dentry, alias);
2701 write_sequnlock(&rename_lock);
2705 /* Nope, but we must(!) avoid directory
2706 * aliasing. This drops inode->i_lock */
2707 actual = __d_unalias(inode, dentry, alias);
2709 write_sequnlock(&rename_lock);
2710 if (IS_ERR(actual)) {
2711 if (PTR_ERR(actual) == -ELOOP)
2712 pr_warn_ratelimited(
2713 "VFS: Lookup of '%s' in %s %s"
2714 " would have caused loop\n",
2715 dentry->d_name.name,
2716 inode->i_sb->s_type->name,
2724 /* Add a unique reference */
2725 actual = __d_instantiate_unique(dentry, inode);
2729 BUG_ON(!d_unhashed(actual));
2731 spin_lock(&actual->d_lock);
2734 spin_unlock(&actual->d_lock);
2735 spin_unlock(&inode->i_lock);
2737 if (actual == dentry) {
2738 security_d_instantiate(dentry, inode);
2745 EXPORT_SYMBOL_GPL(d_materialise_unique);
2747 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2751 return -ENAMETOOLONG;
2753 memcpy(*buffer, str, namelen);
2758 * prepend_name - prepend a pathname in front of current buffer pointer
2759 * buffer: buffer pointer
2760 * buflen: allocated length of the buffer
2761 * name: name string and length qstr structure
2763 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2764 * make sure that either the old or the new name pointer and length are
2765 * fetched. However, there may be mismatch between length and pointer.
2766 * The length cannot be trusted, we need to copy it byte-by-byte until
2767 * the length is reached or a null byte is found. It also prepends "/" at
2768 * the beginning of the name. The sequence number check at the caller will
2769 * retry it again when a d_move() does happen. So any garbage in the buffer
2770 * due to mismatched pointer and length will be discarded.
2772 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2774 const char *dname = ACCESS_ONCE(name->name);
2775 u32 dlen = ACCESS_ONCE(name->len);
2778 if (*buflen < dlen + 1)
2779 return -ENAMETOOLONG;
2780 *buflen -= dlen + 1;
2781 p = *buffer -= dlen + 1;
2793 * prepend_path - Prepend path string to a buffer
2794 * @path: the dentry/vfsmount to report
2795 * @root: root vfsmnt/dentry
2796 * @buffer: pointer to the end of the buffer
2797 * @buflen: pointer to buffer length
2799 * The function tries to write out the pathname without taking any lock other
2800 * than the RCU read lock to make sure that dentries won't go away. It only
2801 * checks the sequence number of the global rename_lock as any change in the
2802 * dentry's d_seq will be preceded by changes in the rename_lock sequence
2803 * number. If the sequence number had been change, it will restart the whole
2804 * pathname back-tracing sequence again. It performs a total of 3 trials of
2805 * lockless back-tracing sequences before falling back to take the
2808 static int prepend_path(const struct path *path,
2809 const struct path *root,
2810 char **buffer, int *buflen)
2812 struct dentry *dentry = path->dentry;
2813 struct vfsmount *vfsmnt = path->mnt;
2814 struct mount *mnt = real_mount(vfsmnt);
2824 read_seqbegin_or_lock(&rename_lock, &seq);
2825 while (dentry != root->dentry || vfsmnt != root->mnt) {
2826 struct dentry * parent;
2828 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2830 if (mnt_has_parent(mnt)) {
2831 dentry = mnt->mnt_mountpoint;
2832 mnt = mnt->mnt_parent;
2837 * Filesystems needing to implement special "root names"
2838 * should do so with ->d_dname()
2840 if (IS_ROOT(dentry) &&
2841 (dentry->d_name.len != 1 ||
2842 dentry->d_name.name[0] != '/')) {
2843 WARN(1, "Root dentry has weird name <%.*s>\n",
2844 (int) dentry->d_name.len,
2845 dentry->d_name.name);
2848 error = is_mounted(vfsmnt) ? 1 : 2;
2851 parent = dentry->d_parent;
2853 error = prepend_name(&bptr, &blen, &dentry->d_name);
2861 if (need_seqretry(&rename_lock, seq)) {
2865 done_seqretry(&rename_lock, seq);
2867 if (error >= 0 && bptr == *buffer) {
2869 error = -ENAMETOOLONG;
2879 * __d_path - return the path of a dentry
2880 * @path: the dentry/vfsmount to report
2881 * @root: root vfsmnt/dentry
2882 * @buf: buffer to return value in
2883 * @buflen: buffer length
2885 * Convert a dentry into an ASCII path name.
2887 * Returns a pointer into the buffer or an error code if the
2888 * path was too long.
2890 * "buflen" should be positive.
2892 * If the path is not reachable from the supplied root, return %NULL.
2894 char *__d_path(const struct path *path,
2895 const struct path *root,
2896 char *buf, int buflen)
2898 char *res = buf + buflen;
2901 prepend(&res, &buflen, "\0", 1);
2902 br_read_lock(&vfsmount_lock);
2903 error = prepend_path(path, root, &res, &buflen);
2904 br_read_unlock(&vfsmount_lock);
2907 return ERR_PTR(error);
2913 char *d_absolute_path(const struct path *path,
2914 char *buf, int buflen)
2916 struct path root = {};
2917 char *res = buf + buflen;
2920 prepend(&res, &buflen, "\0", 1);
2921 br_read_lock(&vfsmount_lock);
2922 error = prepend_path(path, &root, &res, &buflen);
2923 br_read_unlock(&vfsmount_lock);
2928 return ERR_PTR(error);
2933 * same as __d_path but appends "(deleted)" for unlinked files.
2935 static int path_with_deleted(const struct path *path,
2936 const struct path *root,
2937 char **buf, int *buflen)
2939 prepend(buf, buflen, "\0", 1);
2940 if (d_unlinked(path->dentry)) {
2941 int error = prepend(buf, buflen, " (deleted)", 10);
2946 return prepend_path(path, root, buf, buflen);
2949 static int prepend_unreachable(char **buffer, int *buflen)
2951 return prepend(buffer, buflen, "(unreachable)", 13);
2955 * d_path - return the path of a dentry
2956 * @path: path to report
2957 * @buf: buffer to return value in
2958 * @buflen: buffer length
2960 * Convert a dentry into an ASCII path name. If the entry has been deleted
2961 * the string " (deleted)" is appended. Note that this is ambiguous.
2963 * Returns a pointer into the buffer or an error code if the path was
2964 * too long. Note: Callers should use the returned pointer, not the passed
2965 * in buffer, to use the name! The implementation often starts at an offset
2966 * into the buffer, and may leave 0 bytes at the start.
2968 * "buflen" should be positive.
2970 char *d_path(const struct path *path, char *buf, int buflen)
2972 char *res = buf + buflen;
2977 * We have various synthetic filesystems that never get mounted. On
2978 * these filesystems dentries are never used for lookup purposes, and
2979 * thus don't need to be hashed. They also don't need a name until a
2980 * user wants to identify the object in /proc/pid/fd/. The little hack
2981 * below allows us to generate a name for these objects on demand:
2983 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2984 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2986 get_fs_root(current->fs, &root);
2987 br_read_lock(&vfsmount_lock);
2988 error = path_with_deleted(path, &root, &res, &buflen);
2989 br_read_unlock(&vfsmount_lock);
2991 res = ERR_PTR(error);
2995 EXPORT_SYMBOL(d_path);
2998 * Helper function for dentry_operations.d_dname() members
3000 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3001 const char *fmt, ...)
3007 va_start(args, fmt);
3008 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3011 if (sz > sizeof(temp) || sz > buflen)
3012 return ERR_PTR(-ENAMETOOLONG);
3014 buffer += buflen - sz;
3015 return memcpy(buffer, temp, sz);
3018 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3020 char *end = buffer + buflen;
3021 /* these dentries are never renamed, so d_lock is not needed */
3022 if (prepend(&end, &buflen, " (deleted)", 11) ||
3023 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3024 prepend(&end, &buflen, "/", 1))
3025 end = ERR_PTR(-ENAMETOOLONG);
3030 * Write full pathname from the root of the filesystem into the buffer.
3032 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
3042 prepend(&end, &len, "\0", 1);
3048 read_seqbegin_or_lock(&rename_lock, &seq);
3049 while (!IS_ROOT(dentry)) {
3050 struct dentry *parent = dentry->d_parent;
3054 error = prepend_name(&end, &len, &dentry->d_name);
3063 if (need_seqretry(&rename_lock, seq)) {
3067 done_seqretry(&rename_lock, seq);
3072 return ERR_PTR(-ENAMETOOLONG);
3075 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3077 return __dentry_path(dentry, buf, buflen);
3079 EXPORT_SYMBOL(dentry_path_raw);
3081 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3086 if (d_unlinked(dentry)) {
3088 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3092 retval = __dentry_path(dentry, buf, buflen);
3093 if (!IS_ERR(retval) && p)
3094 *p = '/'; /* restore '/' overriden with '\0' */
3097 return ERR_PTR(-ENAMETOOLONG);
3101 * NOTE! The user-level library version returns a
3102 * character pointer. The kernel system call just
3103 * returns the length of the buffer filled (which
3104 * includes the ending '\0' character), or a negative
3105 * error value. So libc would do something like
3107 * char *getcwd(char * buf, size_t size)
3111 * retval = sys_getcwd(buf, size);
3118 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3121 struct path pwd, root;
3122 char *page = (char *) __get_free_page(GFP_USER);
3127 get_fs_root_and_pwd(current->fs, &root, &pwd);
3130 br_read_lock(&vfsmount_lock);
3131 if (!d_unlinked(pwd.dentry)) {
3133 char *cwd = page + PAGE_SIZE;
3134 int buflen = PAGE_SIZE;
3136 prepend(&cwd, &buflen, "\0", 1);
3137 error = prepend_path(&pwd, &root, &cwd, &buflen);
3138 br_read_unlock(&vfsmount_lock);
3143 /* Unreachable from current root */
3145 error = prepend_unreachable(&cwd, &buflen);
3151 len = PAGE_SIZE + page - cwd;
3154 if (copy_to_user(buf, cwd, len))
3158 br_read_unlock(&vfsmount_lock);
3164 free_page((unsigned long) page);
3169 * Test whether new_dentry is a subdirectory of old_dentry.
3171 * Trivially implemented using the dcache structure
3175 * is_subdir - is new dentry a subdirectory of old_dentry
3176 * @new_dentry: new dentry
3177 * @old_dentry: old dentry
3179 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3180 * Returns 0 otherwise.
3181 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3184 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3189 if (new_dentry == old_dentry)
3193 /* for restarting inner loop in case of seq retry */
3194 seq = read_seqbegin(&rename_lock);
3196 * Need rcu_readlock to protect against the d_parent trashing
3200 if (d_ancestor(old_dentry, new_dentry))
3205 } while (read_seqretry(&rename_lock, seq));
3210 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3212 struct dentry *root = data;
3213 if (dentry != root) {
3214 if (d_unhashed(dentry) || !dentry->d_inode)
3217 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3218 dentry->d_flags |= DCACHE_GENOCIDE;
3219 dentry->d_lockref.count--;
3222 return D_WALK_CONTINUE;
3225 void d_genocide(struct dentry *parent)
3227 d_walk(parent, parent, d_genocide_kill, NULL);
3230 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3232 inode_dec_link_count(inode);
3233 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3234 !hlist_unhashed(&dentry->d_alias) ||
3235 !d_unlinked(dentry));
3236 spin_lock(&dentry->d_parent->d_lock);
3237 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3238 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3239 (unsigned long long)inode->i_ino);
3240 spin_unlock(&dentry->d_lock);
3241 spin_unlock(&dentry->d_parent->d_lock);
3242 d_instantiate(dentry, inode);
3244 EXPORT_SYMBOL(d_tmpfile);
3246 static __initdata unsigned long dhash_entries;
3247 static int __init set_dhash_entries(char *str)
3251 dhash_entries = simple_strtoul(str, &str, 0);
3254 __setup("dhash_entries=", set_dhash_entries);
3256 static void __init dcache_init_early(void)
3260 /* If hashes are distributed across NUMA nodes, defer
3261 * hash allocation until vmalloc space is available.
3267 alloc_large_system_hash("Dentry cache",
3268 sizeof(struct hlist_bl_head),
3277 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3278 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3281 static void __init dcache_init(void)
3286 * A constructor could be added for stable state like the lists,
3287 * but it is probably not worth it because of the cache nature
3290 dentry_cache = KMEM_CACHE(dentry,
3291 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3293 /* Hash may have been set up in dcache_init_early */
3298 alloc_large_system_hash("Dentry cache",
3299 sizeof(struct hlist_bl_head),
3308 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3309 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3312 /* SLAB cache for __getname() consumers */
3313 struct kmem_cache *names_cachep __read_mostly;
3314 EXPORT_SYMBOL(names_cachep);
3316 EXPORT_SYMBOL(d_genocide);
3318 void __init vfs_caches_init_early(void)
3320 dcache_init_early();
3324 void __init vfs_caches_init(unsigned long mempages)
3326 unsigned long reserve;
3328 /* Base hash sizes on available memory, with a reserve equal to
3329 150% of current kernel size */
3331 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3332 mempages -= reserve;
3334 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3335 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3339 files_init(mempages);