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>
40 #include <linux/list_lru.h>
46 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
48 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
52 * dentry->d_sb->s_dentry_lru_lock protects:
53 * - the dcache lru lists and counters
60 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
65 * dentry->d_inode->i_lock
67 * dentry->d_sb->s_dentry_lru_lock
68 * dcache_hash_bucket lock
71 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
74 * dentry->d_parent->d_lock
77 * If no ancestor relationship:
78 * if (dentry1 < dentry2)
82 int sysctl_vfs_cache_pressure __read_mostly = 100;
83 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
87 EXPORT_SYMBOL(rename_lock);
89 static struct kmem_cache *dentry_cache __read_mostly;
92 * read_seqbegin_or_lock - begin a sequence number check or locking block
94 * seq : sequence number to be checked
96 * First try it once optimistically without taking the lock. If that fails,
97 * take the lock. The sequence number is also used as a marker for deciding
98 * whether to be a reader (even) or writer (odd).
99 * N.B. seq must be initialized to an even number to begin with.
101 static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
103 if (!(*seq & 1)) /* Even */
104 *seq = read_seqbegin(lock);
109 static inline int need_seqretry(seqlock_t *lock, int seq)
111 return !(seq & 1) && read_seqretry(lock, seq);
114 static inline void done_seqretry(seqlock_t *lock, int seq)
117 write_sequnlock(lock);
121 * This is the single most critical data structure when it comes
122 * to the dcache: the hashtable for lookups. Somebody should try
123 * to make this good - I've just made it work.
125 * This hash-function tries to avoid losing too many bits of hash
126 * information, yet avoid using a prime hash-size or similar.
128 #define D_HASHBITS d_hash_shift
129 #define D_HASHMASK d_hash_mask
131 static unsigned int d_hash_mask __read_mostly;
132 static unsigned int d_hash_shift __read_mostly;
134 static struct hlist_bl_head *dentry_hashtable __read_mostly;
136 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
139 hash += (unsigned long) parent / L1_CACHE_BYTES;
140 hash = hash + (hash >> D_HASHBITS);
141 return dentry_hashtable + (hash & D_HASHMASK);
144 /* Statistics gathering. */
145 struct dentry_stat_t dentry_stat = {
149 static DEFINE_PER_CPU(long, nr_dentry);
150 static DEFINE_PER_CPU(long, nr_dentry_unused);
152 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
155 * Here we resort to our own counters instead of using generic per-cpu counters
156 * for consistency with what the vfs inode code does. We are expected to harvest
157 * better code and performance by having our own specialized counters.
159 * Please note that the loop is done over all possible CPUs, not over all online
160 * CPUs. The reason for this is that we don't want to play games with CPUs going
161 * on and off. If one of them goes off, we will just keep their counters.
163 * glommer: See cffbc8a for details, and if you ever intend to change this,
164 * please update all vfs counters to match.
166 static long get_nr_dentry(void)
170 for_each_possible_cpu(i)
171 sum += per_cpu(nr_dentry, i);
172 return sum < 0 ? 0 : sum;
175 static long get_nr_dentry_unused(void)
179 for_each_possible_cpu(i)
180 sum += per_cpu(nr_dentry_unused, i);
181 return sum < 0 ? 0 : sum;
184 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
185 size_t *lenp, loff_t *ppos)
187 dentry_stat.nr_dentry = get_nr_dentry();
188 dentry_stat.nr_unused = get_nr_dentry_unused();
189 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
194 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
195 * The strings are both count bytes long, and count is non-zero.
197 #ifdef CONFIG_DCACHE_WORD_ACCESS
199 #include <asm/word-at-a-time.h>
201 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
202 * aligned allocation for this particular component. We don't
203 * strictly need the load_unaligned_zeropad() safety, but it
204 * doesn't hurt either.
206 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
207 * need the careful unaligned handling.
209 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
211 unsigned long a,b,mask;
214 a = *(unsigned long *)cs;
215 b = load_unaligned_zeropad(ct);
216 if (tcount < sizeof(unsigned long))
218 if (unlikely(a != b))
220 cs += sizeof(unsigned long);
221 ct += sizeof(unsigned long);
222 tcount -= sizeof(unsigned long);
226 mask = ~(~0ul << tcount*8);
227 return unlikely(!!((a ^ b) & mask));
232 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
246 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
248 const unsigned char *cs;
250 * Be careful about RCU walk racing with rename:
251 * use ACCESS_ONCE to fetch the name pointer.
253 * NOTE! Even if a rename will mean that the length
254 * was not loaded atomically, we don't care. The
255 * RCU walk will check the sequence count eventually,
256 * and catch it. And we won't overrun the buffer,
257 * because we're reading the name pointer atomically,
258 * and a dentry name is guaranteed to be properly
259 * terminated with a NUL byte.
261 * End result: even if 'len' is wrong, we'll exit
262 * early because the data cannot match (there can
263 * be no NUL in the ct/tcount data)
265 cs = ACCESS_ONCE(dentry->d_name.name);
266 smp_read_barrier_depends();
267 return dentry_string_cmp(cs, ct, tcount);
270 static void __d_free(struct rcu_head *head)
272 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
274 WARN_ON(!hlist_unhashed(&dentry->d_alias));
275 if (dname_external(dentry))
276 kfree(dentry->d_name.name);
277 kmem_cache_free(dentry_cache, dentry);
283 static void d_free(struct dentry *dentry)
285 BUG_ON((int)dentry->d_lockref.count > 0);
286 this_cpu_dec(nr_dentry);
287 if (dentry->d_op && dentry->d_op->d_release)
288 dentry->d_op->d_release(dentry);
290 /* if dentry was never visible to RCU, immediate free is OK */
291 if (!(dentry->d_flags & DCACHE_RCUACCESS))
292 __d_free(&dentry->d_u.d_rcu);
294 call_rcu(&dentry->d_u.d_rcu, __d_free);
298 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
299 * @dentry: the target dentry
300 * After this call, in-progress rcu-walk path lookup will fail. This
301 * should be called after unhashing, and after changing d_inode (if
302 * the dentry has not already been unhashed).
304 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
306 assert_spin_locked(&dentry->d_lock);
307 /* Go through a barrier */
308 write_seqcount_barrier(&dentry->d_seq);
312 * Release the dentry's inode, using the filesystem
313 * d_iput() operation if defined. Dentry has no refcount
316 static void dentry_iput(struct dentry * dentry)
317 __releases(dentry->d_lock)
318 __releases(dentry->d_inode->i_lock)
320 struct inode *inode = dentry->d_inode;
322 dentry->d_inode = NULL;
323 hlist_del_init(&dentry->d_alias);
324 spin_unlock(&dentry->d_lock);
325 spin_unlock(&inode->i_lock);
327 fsnotify_inoderemove(inode);
328 if (dentry->d_op && dentry->d_op->d_iput)
329 dentry->d_op->d_iput(dentry, inode);
333 spin_unlock(&dentry->d_lock);
338 * Release the dentry's inode, using the filesystem
339 * d_iput() operation if defined. dentry remains in-use.
341 static void dentry_unlink_inode(struct dentry * dentry)
342 __releases(dentry->d_lock)
343 __releases(dentry->d_inode->i_lock)
345 struct inode *inode = dentry->d_inode;
346 dentry->d_inode = NULL;
347 hlist_del_init(&dentry->d_alias);
348 dentry_rcuwalk_barrier(dentry);
349 spin_unlock(&dentry->d_lock);
350 spin_unlock(&inode->i_lock);
352 fsnotify_inoderemove(inode);
353 if (dentry->d_op && dentry->d_op->d_iput)
354 dentry->d_op->d_iput(dentry, inode);
360 * dentry_lru_(add|del)_list) must be called with d_lock held.
362 static void dentry_lru_add(struct dentry *dentry)
364 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST))) {
365 if (list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru))
366 this_cpu_inc(nr_dentry_unused);
367 dentry->d_flags |= DCACHE_LRU_LIST;
372 * Remove a dentry with references from the LRU.
374 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
375 * lose our last reference through the parent walk. In this case, we need to
376 * remove ourselves from the shrink list, not the LRU.
378 static void dentry_lru_del(struct dentry *dentry)
380 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
381 list_del_init(&dentry->d_lru);
382 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
386 if (list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru))
387 this_cpu_dec(nr_dentry_unused);
388 dentry->d_flags &= ~DCACHE_LRU_LIST;
392 * d_kill - kill dentry and return parent
393 * @dentry: dentry to kill
394 * @parent: parent dentry
396 * The dentry must already be unhashed and removed from the LRU.
398 * If this is the root of the dentry tree, return NULL.
400 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
403 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
404 __releases(dentry->d_lock)
405 __releases(parent->d_lock)
406 __releases(dentry->d_inode->i_lock)
408 list_del(&dentry->d_u.d_child);
410 * Inform try_to_ascend() that we are no longer attached to the
413 dentry->d_flags |= DCACHE_DENTRY_KILLED;
415 spin_unlock(&parent->d_lock);
418 * dentry_iput drops the locks, at which point nobody (except
419 * transient RCU lookups) can reach this dentry.
426 * Unhash a dentry without inserting an RCU walk barrier or checking that
427 * dentry->d_lock is locked. The caller must take care of that, if
430 static void __d_shrink(struct dentry *dentry)
432 if (!d_unhashed(dentry)) {
433 struct hlist_bl_head *b;
434 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
435 b = &dentry->d_sb->s_anon;
437 b = d_hash(dentry->d_parent, dentry->d_name.hash);
440 __hlist_bl_del(&dentry->d_hash);
441 dentry->d_hash.pprev = NULL;
447 * d_drop - drop a dentry
448 * @dentry: dentry to drop
450 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
451 * be found through a VFS lookup any more. Note that this is different from
452 * deleting the dentry - d_delete will try to mark the dentry negative if
453 * possible, giving a successful _negative_ lookup, while d_drop will
454 * just make the cache lookup fail.
456 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
457 * reason (NFS timeouts or autofs deletes).
459 * __d_drop requires dentry->d_lock.
461 void __d_drop(struct dentry *dentry)
463 if (!d_unhashed(dentry)) {
465 dentry_rcuwalk_barrier(dentry);
468 EXPORT_SYMBOL(__d_drop);
470 void d_drop(struct dentry *dentry)
472 spin_lock(&dentry->d_lock);
474 spin_unlock(&dentry->d_lock);
476 EXPORT_SYMBOL(d_drop);
479 * Finish off a dentry we've decided to kill.
480 * dentry->d_lock must be held, returns with it unlocked.
481 * If ref is non-zero, then decrement the refcount too.
482 * Returns dentry requiring refcount drop, or NULL if we're done.
484 static inline struct dentry *
485 dentry_kill(struct dentry *dentry, int unlock_on_failure)
486 __releases(dentry->d_lock)
489 struct dentry *parent;
491 inode = dentry->d_inode;
492 if (inode && !spin_trylock(&inode->i_lock)) {
494 if (unlock_on_failure) {
495 spin_unlock(&dentry->d_lock);
498 return dentry; /* try again with same dentry */
503 parent = dentry->d_parent;
504 if (parent && !spin_trylock(&parent->d_lock)) {
506 spin_unlock(&inode->i_lock);
511 * The dentry is now unrecoverably dead to the world.
513 lockref_mark_dead(&dentry->d_lockref);
516 * inform the fs via d_prune that this dentry is about to be
517 * unhashed and destroyed.
519 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
520 dentry->d_op->d_prune(dentry);
522 dentry_lru_del(dentry);
523 /* if it was on the hash then remove it */
525 return d_kill(dentry, parent);
531 * This is complicated by the fact that we do not want to put
532 * dentries that are no longer on any hash chain on the unused
533 * list: we'd much rather just get rid of them immediately.
535 * However, that implies that we have to traverse the dentry
536 * tree upwards to the parents which might _also_ now be
537 * scheduled for deletion (it may have been only waiting for
538 * its last child to go away).
540 * This tail recursion is done by hand as we don't want to depend
541 * on the compiler to always get this right (gcc generally doesn't).
542 * Real recursion would eat up our stack space.
546 * dput - release a dentry
547 * @dentry: dentry to release
549 * Release a dentry. This will drop the usage count and if appropriate
550 * call the dentry unlink method as well as removing it from the queues and
551 * releasing its resources. If the parent dentries were scheduled for release
552 * they too may now get deleted.
554 void dput(struct dentry *dentry)
556 if (unlikely(!dentry))
560 if (lockref_put_or_lock(&dentry->d_lockref))
563 /* Unreachable? Get rid of it */
564 if (unlikely(d_unhashed(dentry)))
567 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
568 if (dentry->d_op->d_delete(dentry))
572 dentry->d_flags |= DCACHE_REFERENCED;
573 dentry_lru_add(dentry);
575 dentry->d_lockref.count--;
576 spin_unlock(&dentry->d_lock);
580 dentry = dentry_kill(dentry, 1);
587 * d_invalidate - invalidate a dentry
588 * @dentry: dentry to invalidate
590 * Try to invalidate the dentry if it turns out to be
591 * possible. If there are other dentries that can be
592 * reached through this one we can't delete it and we
593 * return -EBUSY. On success we return 0.
598 int d_invalidate(struct dentry * dentry)
601 * If it's already been dropped, return OK.
603 spin_lock(&dentry->d_lock);
604 if (d_unhashed(dentry)) {
605 spin_unlock(&dentry->d_lock);
609 * Check whether to do a partial shrink_dcache
610 * to get rid of unused child entries.
612 if (!list_empty(&dentry->d_subdirs)) {
613 spin_unlock(&dentry->d_lock);
614 shrink_dcache_parent(dentry);
615 spin_lock(&dentry->d_lock);
619 * Somebody else still using it?
621 * If it's a directory, we can't drop it
622 * for fear of somebody re-populating it
623 * with children (even though dropping it
624 * would make it unreachable from the root,
625 * we might still populate it if it was a
626 * working directory or similar).
627 * We also need to leave mountpoints alone,
630 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
631 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
632 spin_unlock(&dentry->d_lock);
638 spin_unlock(&dentry->d_lock);
641 EXPORT_SYMBOL(d_invalidate);
643 /* This must be called with d_lock held */
644 static inline void __dget_dlock(struct dentry *dentry)
646 dentry->d_lockref.count++;
649 static inline void __dget(struct dentry *dentry)
651 lockref_get(&dentry->d_lockref);
654 struct dentry *dget_parent(struct dentry *dentry)
660 * Do optimistic parent lookup without any
664 ret = ACCESS_ONCE(dentry->d_parent);
665 gotref = lockref_get_not_zero(&ret->d_lockref);
667 if (likely(gotref)) {
668 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
675 * Don't need rcu_dereference because we re-check it was correct under
679 ret = dentry->d_parent;
680 spin_lock(&ret->d_lock);
681 if (unlikely(ret != dentry->d_parent)) {
682 spin_unlock(&ret->d_lock);
687 BUG_ON(!ret->d_lockref.count);
688 ret->d_lockref.count++;
689 spin_unlock(&ret->d_lock);
692 EXPORT_SYMBOL(dget_parent);
695 * d_find_alias - grab a hashed alias of inode
696 * @inode: inode in question
697 * @want_discon: flag, used by d_splice_alias, to request
698 * that only a DISCONNECTED alias be returned.
700 * If inode has a hashed alias, or is a directory and has any alias,
701 * acquire the reference to alias and return it. Otherwise return NULL.
702 * Notice that if inode is a directory there can be only one alias and
703 * it can be unhashed only if it has no children, or if it is the root
706 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
707 * any other hashed alias over that one unless @want_discon is set,
708 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
710 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
712 struct dentry *alias, *discon_alias;
716 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
717 spin_lock(&alias->d_lock);
718 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
719 if (IS_ROOT(alias) &&
720 (alias->d_flags & DCACHE_DISCONNECTED)) {
721 discon_alias = alias;
722 } else if (!want_discon) {
724 spin_unlock(&alias->d_lock);
728 spin_unlock(&alias->d_lock);
731 alias = discon_alias;
732 spin_lock(&alias->d_lock);
733 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
734 if (IS_ROOT(alias) &&
735 (alias->d_flags & DCACHE_DISCONNECTED)) {
737 spin_unlock(&alias->d_lock);
741 spin_unlock(&alias->d_lock);
747 struct dentry *d_find_alias(struct inode *inode)
749 struct dentry *de = NULL;
751 if (!hlist_empty(&inode->i_dentry)) {
752 spin_lock(&inode->i_lock);
753 de = __d_find_alias(inode, 0);
754 spin_unlock(&inode->i_lock);
758 EXPORT_SYMBOL(d_find_alias);
761 * Try to kill dentries associated with this inode.
762 * WARNING: you must own a reference to inode.
764 void d_prune_aliases(struct inode *inode)
766 struct dentry *dentry;
768 spin_lock(&inode->i_lock);
769 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
770 spin_lock(&dentry->d_lock);
771 if (!dentry->d_lockref.count) {
773 * inform the fs via d_prune that this dentry
774 * is about to be unhashed and destroyed.
776 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
778 dentry->d_op->d_prune(dentry);
780 __dget_dlock(dentry);
782 spin_unlock(&dentry->d_lock);
783 spin_unlock(&inode->i_lock);
787 spin_unlock(&dentry->d_lock);
789 spin_unlock(&inode->i_lock);
791 EXPORT_SYMBOL(d_prune_aliases);
794 * Try to throw away a dentry - free the inode, dput the parent.
795 * Requires dentry->d_lock is held, and dentry->d_count == 0.
796 * Releases dentry->d_lock.
798 * This may fail if locks cannot be acquired no problem, just try again.
800 static struct dentry * try_prune_one_dentry(struct dentry *dentry)
801 __releases(dentry->d_lock)
803 struct dentry *parent;
805 parent = dentry_kill(dentry, 0);
807 * If dentry_kill returns NULL, we have nothing more to do.
808 * if it returns the same dentry, trylocks failed. In either
809 * case, just loop again.
811 * Otherwise, we need to prune ancestors too. This is necessary
812 * to prevent quadratic behavior of shrink_dcache_parent(), but
813 * is also expected to be beneficial in reducing dentry cache
818 if (parent == dentry)
821 /* Prune ancestors. */
824 if (lockref_put_or_lock(&dentry->d_lockref))
826 dentry = dentry_kill(dentry, 1);
831 static void shrink_dentry_list(struct list_head *list)
833 struct dentry *dentry;
837 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
838 if (&dentry->d_lru == list)
840 spin_lock(&dentry->d_lock);
841 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
842 spin_unlock(&dentry->d_lock);
847 * The dispose list is isolated and dentries are not accounted
848 * to the LRU here, so we can simply remove it from the list
849 * here regardless of whether it is referenced or not.
851 list_del_init(&dentry->d_lru);
852 dentry->d_flags &= ~DCACHE_SHRINK_LIST;
855 * We found an inuse dentry which was not removed from
856 * the LRU because of laziness during lookup. Do not free it.
858 if (dentry->d_lockref.count) {
859 spin_unlock(&dentry->d_lock);
864 dentry = try_prune_one_dentry(dentry);
868 dentry->d_flags |= DCACHE_SHRINK_LIST;
869 list_add(&dentry->d_lru, list);
870 spin_unlock(&dentry->d_lock);
876 static enum lru_status
877 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
879 struct list_head *freeable = arg;
880 struct dentry *dentry = container_of(item, struct dentry, d_lru);
884 * we are inverting the lru lock/dentry->d_lock here,
885 * so use a trylock. If we fail to get the lock, just skip
888 if (!spin_trylock(&dentry->d_lock))
892 * Referenced dentries are still in use. If they have active
893 * counts, just remove them from the LRU. Otherwise give them
894 * another pass through the LRU.
896 if (dentry->d_lockref.count) {
897 list_del_init(&dentry->d_lru);
898 spin_unlock(&dentry->d_lock);
902 if (dentry->d_flags & DCACHE_REFERENCED) {
903 dentry->d_flags &= ~DCACHE_REFERENCED;
904 spin_unlock(&dentry->d_lock);
907 * The list move itself will be made by the common LRU code. At
908 * this point, we've dropped the dentry->d_lock but keep the
909 * lru lock. This is safe to do, since every list movement is
910 * protected by the lru lock even if both locks are held.
912 * This is guaranteed by the fact that all LRU management
913 * functions are intermediated by the LRU API calls like
914 * list_lru_add and list_lru_del. List movement in this file
915 * only ever occur through this functions or through callbacks
916 * like this one, that are called from the LRU API.
918 * The only exceptions to this are functions like
919 * shrink_dentry_list, and code that first checks for the
920 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
921 * operating only with stack provided lists after they are
922 * properly isolated from the main list. It is thus, always a
928 dentry->d_flags |= DCACHE_SHRINK_LIST;
929 list_move_tail(&dentry->d_lru, freeable);
930 this_cpu_dec(nr_dentry_unused);
931 spin_unlock(&dentry->d_lock);
937 * prune_dcache_sb - shrink the dcache
939 * @nr_to_scan : number of entries to try to free
941 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
942 * done when we need more memory an called from the superblock shrinker
945 * This function may fail to free any resources if all the dentries are in
948 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan)
953 freed = list_lru_walk(&sb->s_dentry_lru, dentry_lru_isolate,
954 &dispose, nr_to_scan);
955 shrink_dentry_list(&dispose);
960 * Mark all the dentries as on being the dispose list so we don't think they are
961 * still on the LRU if we try to kill them from ascending the parent chain in
962 * try_prune_one_dentry() rather than directly from the dispose list.
966 struct list_head *dispose)
968 struct dentry *dentry;
971 list_for_each_entry_rcu(dentry, dispose, d_lru) {
972 spin_lock(&dentry->d_lock);
973 dentry->d_flags |= DCACHE_SHRINK_LIST;
974 spin_unlock(&dentry->d_lock);
977 shrink_dentry_list(dispose);
981 * shrink_dcache_sb - shrink dcache for a superblock
984 * Shrink the dcache for the specified super block. This is used to free
985 * the dcache before unmounting a file system.
987 void shrink_dcache_sb(struct super_block *sb)
991 disposed = list_lru_dispose_all(&sb->s_dentry_lru, shrink_dcache_list);
992 this_cpu_sub(nr_dentry_unused, disposed);
994 EXPORT_SYMBOL(shrink_dcache_sb);
997 * destroy a single subtree of dentries for unmount
998 * - see the comments on shrink_dcache_for_umount() for a description of the
1001 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
1003 struct dentry *parent;
1005 BUG_ON(!IS_ROOT(dentry));
1008 /* descend to the first leaf in the current subtree */
1009 while (!list_empty(&dentry->d_subdirs))
1010 dentry = list_entry(dentry->d_subdirs.next,
1011 struct dentry, d_u.d_child);
1013 /* consume the dentries from this leaf up through its parents
1014 * until we find one with children or run out altogether */
1016 struct inode *inode;
1019 * inform the fs that this dentry is about to be
1020 * unhashed and destroyed.
1022 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
1023 !d_unhashed(dentry))
1024 dentry->d_op->d_prune(dentry);
1026 dentry_lru_del(dentry);
1029 if (dentry->d_lockref.count != 0) {
1031 "BUG: Dentry %p{i=%lx,n=%s}"
1032 " still in use (%d)"
1033 " [unmount of %s %s]\n",
1036 dentry->d_inode->i_ino : 0UL,
1037 dentry->d_name.name,
1038 dentry->d_lockref.count,
1039 dentry->d_sb->s_type->name,
1040 dentry->d_sb->s_id);
1044 if (IS_ROOT(dentry)) {
1046 list_del(&dentry->d_u.d_child);
1048 parent = dentry->d_parent;
1049 parent->d_lockref.count--;
1050 list_del(&dentry->d_u.d_child);
1053 inode = dentry->d_inode;
1055 dentry->d_inode = NULL;
1056 hlist_del_init(&dentry->d_alias);
1057 if (dentry->d_op && dentry->d_op->d_iput)
1058 dentry->d_op->d_iput(dentry, inode);
1065 /* finished when we fall off the top of the tree,
1066 * otherwise we ascend to the parent and move to the
1067 * next sibling if there is one */
1071 } while (list_empty(&dentry->d_subdirs));
1073 dentry = list_entry(dentry->d_subdirs.next,
1074 struct dentry, d_u.d_child);
1079 * destroy the dentries attached to a superblock on unmounting
1080 * - we don't need to use dentry->d_lock because:
1081 * - the superblock is detached from all mountings and open files, so the
1082 * dentry trees will not be rearranged by the VFS
1083 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1084 * any dentries belonging to this superblock that it comes across
1085 * - the filesystem itself is no longer permitted to rearrange the dentries
1086 * in this superblock
1088 void shrink_dcache_for_umount(struct super_block *sb)
1090 struct dentry *dentry;
1092 if (down_read_trylock(&sb->s_umount))
1095 dentry = sb->s_root;
1097 dentry->d_lockref.count--;
1098 shrink_dcache_for_umount_subtree(dentry);
1100 while (!hlist_bl_empty(&sb->s_anon)) {
1101 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1102 shrink_dcache_for_umount_subtree(dentry);
1107 * This tries to ascend one level of parenthood, but
1108 * we can race with renaming, so we need to re-check
1109 * the parenthood after dropping the lock and check
1110 * that the sequence number still matches.
1112 static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
1114 struct dentry *new = old->d_parent;
1117 spin_unlock(&old->d_lock);
1118 spin_lock(&new->d_lock);
1121 * might go back up the wrong parent if we have had a rename
1124 if (new != old->d_parent ||
1125 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1126 need_seqretry(&rename_lock, seq)) {
1127 spin_unlock(&new->d_lock);
1135 * enum d_walk_ret - action to talke during tree walk
1136 * @D_WALK_CONTINUE: contrinue walk
1137 * @D_WALK_QUIT: quit walk
1138 * @D_WALK_NORETRY: quit when retry is needed
1139 * @D_WALK_SKIP: skip this dentry and its children
1149 * d_walk - walk the dentry tree
1150 * @parent: start of walk
1151 * @data: data passed to @enter() and @finish()
1152 * @enter: callback when first entering the dentry
1153 * @finish: callback when successfully finished the walk
1155 * The @enter() and @finish() callbacks are called with d_lock held.
1157 static void d_walk(struct dentry *parent, void *data,
1158 enum d_walk_ret (*enter)(void *, struct dentry *),
1159 void (*finish)(void *))
1161 struct dentry *this_parent;
1162 struct list_head *next;
1164 enum d_walk_ret ret;
1168 read_seqbegin_or_lock(&rename_lock, &seq);
1169 this_parent = parent;
1170 spin_lock(&this_parent->d_lock);
1172 ret = enter(data, this_parent);
1174 case D_WALK_CONTINUE:
1179 case D_WALK_NORETRY:
1184 next = this_parent->d_subdirs.next;
1186 while (next != &this_parent->d_subdirs) {
1187 struct list_head *tmp = next;
1188 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1191 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1193 ret = enter(data, dentry);
1195 case D_WALK_CONTINUE:
1198 spin_unlock(&dentry->d_lock);
1200 case D_WALK_NORETRY:
1204 spin_unlock(&dentry->d_lock);
1208 if (!list_empty(&dentry->d_subdirs)) {
1209 spin_unlock(&this_parent->d_lock);
1210 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1211 this_parent = dentry;
1212 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1215 spin_unlock(&dentry->d_lock);
1218 * All done at this level ... ascend and resume the search.
1220 if (this_parent != parent) {
1221 struct dentry *child = this_parent;
1222 this_parent = try_to_ascend(this_parent, seq);
1225 next = child->d_u.d_child.next;
1228 if (need_seqretry(&rename_lock, seq)) {
1229 spin_unlock(&this_parent->d_lock);
1236 spin_unlock(&this_parent->d_lock);
1237 done_seqretry(&rename_lock, seq);
1248 * Search for at least 1 mount point in the dentry's subdirs.
1249 * We descend to the next level whenever the d_subdirs
1250 * list is non-empty and continue searching.
1254 * have_submounts - check for mounts over a dentry
1255 * @parent: dentry to check.
1257 * Return true if the parent or its subdirectories contain
1261 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1264 if (d_mountpoint(dentry)) {
1268 return D_WALK_CONTINUE;
1271 int have_submounts(struct dentry *parent)
1275 d_walk(parent, &ret, check_mount, NULL);
1279 EXPORT_SYMBOL(have_submounts);
1282 * Called by mount code to set a mountpoint and check if the mountpoint is
1283 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1284 * subtree can become unreachable).
1286 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1287 * this reason take rename_lock and d_lock on dentry and ancestors.
1289 int d_set_mounted(struct dentry *dentry)
1293 write_seqlock(&rename_lock);
1294 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1295 /* Need exclusion wrt. check_submounts_and_drop() */
1296 spin_lock(&p->d_lock);
1297 if (unlikely(d_unhashed(p))) {
1298 spin_unlock(&p->d_lock);
1301 spin_unlock(&p->d_lock);
1303 spin_lock(&dentry->d_lock);
1304 if (!d_unlinked(dentry)) {
1305 dentry->d_flags |= DCACHE_MOUNTED;
1308 spin_unlock(&dentry->d_lock);
1310 write_sequnlock(&rename_lock);
1315 * Search the dentry child list of the specified parent,
1316 * and move any unused dentries to the end of the unused
1317 * list for prune_dcache(). We descend to the next level
1318 * whenever the d_subdirs list is non-empty and continue
1321 * It returns zero iff there are no unused children,
1322 * otherwise it returns the number of children moved to
1323 * the end of the unused list. This may not be the total
1324 * number of unused children, because select_parent can
1325 * drop the lock and return early due to latency
1329 struct select_data {
1330 struct dentry *start;
1331 struct list_head dispose;
1335 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1337 struct select_data *data = _data;
1338 enum d_walk_ret ret = D_WALK_CONTINUE;
1340 if (data->start == dentry)
1344 * move only zero ref count dentries to the dispose list.
1346 * Those which are presently on the shrink list, being processed
1347 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1348 * loop in shrink_dcache_parent() might not make any progress
1351 if (dentry->d_lockref.count) {
1352 dentry_lru_del(dentry);
1353 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1354 dentry_lru_del(dentry);
1355 list_add_tail(&dentry->d_lru, &data->dispose);
1356 dentry->d_flags |= DCACHE_SHRINK_LIST;
1358 ret = D_WALK_NORETRY;
1361 * We can return to the caller if we have found some (this
1362 * ensures forward progress). We'll be coming back to find
1365 if (data->found && need_resched())
1372 * shrink_dcache_parent - prune dcache
1373 * @parent: parent of entries to prune
1375 * Prune the dcache to remove unused children of the parent dentry.
1377 void shrink_dcache_parent(struct dentry *parent)
1380 struct select_data data;
1382 INIT_LIST_HEAD(&data.dispose);
1383 data.start = parent;
1386 d_walk(parent, &data, select_collect, NULL);
1390 shrink_dentry_list(&data.dispose);
1394 EXPORT_SYMBOL(shrink_dcache_parent);
1396 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1398 struct select_data *data = _data;
1400 if (d_mountpoint(dentry)) {
1401 data->found = -EBUSY;
1405 return select_collect(_data, dentry);
1408 static void check_and_drop(void *_data)
1410 struct select_data *data = _data;
1412 if (d_mountpoint(data->start))
1413 data->found = -EBUSY;
1415 __d_drop(data->start);
1419 * check_submounts_and_drop - prune dcache, check for submounts and drop
1421 * All done as a single atomic operation relative to has_unlinked_ancestor().
1422 * Returns 0 if successfully unhashed @parent. If there were submounts then
1425 * @dentry: dentry to prune and drop
1427 int check_submounts_and_drop(struct dentry *dentry)
1431 /* Negative dentries can be dropped without further checks */
1432 if (!dentry->d_inode) {
1438 struct select_data data;
1440 INIT_LIST_HEAD(&data.dispose);
1441 data.start = dentry;
1444 d_walk(dentry, &data, check_and_collect, check_and_drop);
1447 if (!list_empty(&data.dispose))
1448 shrink_dentry_list(&data.dispose);
1459 EXPORT_SYMBOL(check_submounts_and_drop);
1462 * __d_alloc - allocate a dcache entry
1463 * @sb: filesystem it will belong to
1464 * @name: qstr of the name
1466 * Allocates a dentry. It returns %NULL if there is insufficient memory
1467 * available. On a success the dentry is returned. The name passed in is
1468 * copied and the copy passed in may be reused after this call.
1471 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1473 struct dentry *dentry;
1476 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1481 * We guarantee that the inline name is always NUL-terminated.
1482 * This way the memcpy() done by the name switching in rename
1483 * will still always have a NUL at the end, even if we might
1484 * be overwriting an internal NUL character
1486 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1487 if (name->len > DNAME_INLINE_LEN-1) {
1488 dname = kmalloc(name->len + 1, GFP_KERNEL);
1490 kmem_cache_free(dentry_cache, dentry);
1494 dname = dentry->d_iname;
1497 dentry->d_name.len = name->len;
1498 dentry->d_name.hash = name->hash;
1499 memcpy(dname, name->name, name->len);
1500 dname[name->len] = 0;
1502 /* Make sure we always see the terminating NUL character */
1504 dentry->d_name.name = dname;
1506 dentry->d_lockref.count = 1;
1507 dentry->d_flags = 0;
1508 spin_lock_init(&dentry->d_lock);
1509 seqcount_init(&dentry->d_seq);
1510 dentry->d_inode = NULL;
1511 dentry->d_parent = dentry;
1513 dentry->d_op = NULL;
1514 dentry->d_fsdata = NULL;
1515 INIT_HLIST_BL_NODE(&dentry->d_hash);
1516 INIT_LIST_HEAD(&dentry->d_lru);
1517 INIT_LIST_HEAD(&dentry->d_subdirs);
1518 INIT_HLIST_NODE(&dentry->d_alias);
1519 INIT_LIST_HEAD(&dentry->d_u.d_child);
1520 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1522 this_cpu_inc(nr_dentry);
1528 * d_alloc - allocate a dcache entry
1529 * @parent: parent of entry to allocate
1530 * @name: qstr of the name
1532 * Allocates a dentry. It returns %NULL if there is insufficient memory
1533 * available. On a success the dentry is returned. The name passed in is
1534 * copied and the copy passed in may be reused after this call.
1536 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1538 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1542 spin_lock(&parent->d_lock);
1544 * don't need child lock because it is not subject
1545 * to concurrency here
1547 __dget_dlock(parent);
1548 dentry->d_parent = parent;
1549 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1550 spin_unlock(&parent->d_lock);
1554 EXPORT_SYMBOL(d_alloc);
1556 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1558 struct dentry *dentry = __d_alloc(sb, name);
1560 dentry->d_flags |= DCACHE_DISCONNECTED;
1563 EXPORT_SYMBOL(d_alloc_pseudo);
1565 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1570 q.len = strlen(name);
1571 q.hash = full_name_hash(q.name, q.len);
1572 return d_alloc(parent, &q);
1574 EXPORT_SYMBOL(d_alloc_name);
1576 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1578 WARN_ON_ONCE(dentry->d_op);
1579 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1581 DCACHE_OP_REVALIDATE |
1582 DCACHE_OP_WEAK_REVALIDATE |
1583 DCACHE_OP_DELETE ));
1588 dentry->d_flags |= DCACHE_OP_HASH;
1590 dentry->d_flags |= DCACHE_OP_COMPARE;
1591 if (op->d_revalidate)
1592 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1593 if (op->d_weak_revalidate)
1594 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1596 dentry->d_flags |= DCACHE_OP_DELETE;
1598 dentry->d_flags |= DCACHE_OP_PRUNE;
1601 EXPORT_SYMBOL(d_set_d_op);
1603 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1605 spin_lock(&dentry->d_lock);
1607 if (unlikely(IS_AUTOMOUNT(inode)))
1608 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1609 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1611 dentry->d_inode = inode;
1612 dentry_rcuwalk_barrier(dentry);
1613 spin_unlock(&dentry->d_lock);
1614 fsnotify_d_instantiate(dentry, inode);
1618 * d_instantiate - fill in inode information for a dentry
1619 * @entry: dentry to complete
1620 * @inode: inode to attach to this dentry
1622 * Fill in inode information in the entry.
1624 * This turns negative dentries into productive full members
1627 * NOTE! This assumes that the inode count has been incremented
1628 * (or otherwise set) by the caller to indicate that it is now
1629 * in use by the dcache.
1632 void d_instantiate(struct dentry *entry, struct inode * inode)
1634 BUG_ON(!hlist_unhashed(&entry->d_alias));
1636 spin_lock(&inode->i_lock);
1637 __d_instantiate(entry, inode);
1639 spin_unlock(&inode->i_lock);
1640 security_d_instantiate(entry, inode);
1642 EXPORT_SYMBOL(d_instantiate);
1645 * d_instantiate_unique - instantiate a non-aliased dentry
1646 * @entry: dentry to instantiate
1647 * @inode: inode to attach to this dentry
1649 * Fill in inode information in the entry. On success, it returns NULL.
1650 * If an unhashed alias of "entry" already exists, then we return the
1651 * aliased dentry instead and drop one reference to inode.
1653 * Note that in order to avoid conflicts with rename() etc, the caller
1654 * had better be holding the parent directory semaphore.
1656 * This also assumes that the inode count has been incremented
1657 * (or otherwise set) by the caller to indicate that it is now
1658 * in use by the dcache.
1660 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1661 struct inode *inode)
1663 struct dentry *alias;
1664 int len = entry->d_name.len;
1665 const char *name = entry->d_name.name;
1666 unsigned int hash = entry->d_name.hash;
1669 __d_instantiate(entry, NULL);
1673 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1675 * Don't need alias->d_lock here, because aliases with
1676 * d_parent == entry->d_parent are not subject to name or
1677 * parent changes, because the parent inode i_mutex is held.
1679 if (alias->d_name.hash != hash)
1681 if (alias->d_parent != entry->d_parent)
1683 if (alias->d_name.len != len)
1685 if (dentry_cmp(alias, name, len))
1691 __d_instantiate(entry, inode);
1695 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1697 struct dentry *result;
1699 BUG_ON(!hlist_unhashed(&entry->d_alias));
1702 spin_lock(&inode->i_lock);
1703 result = __d_instantiate_unique(entry, inode);
1705 spin_unlock(&inode->i_lock);
1708 security_d_instantiate(entry, inode);
1712 BUG_ON(!d_unhashed(result));
1717 EXPORT_SYMBOL(d_instantiate_unique);
1719 struct dentry *d_make_root(struct inode *root_inode)
1721 struct dentry *res = NULL;
1724 static const struct qstr name = QSTR_INIT("/", 1);
1726 res = __d_alloc(root_inode->i_sb, &name);
1728 d_instantiate(res, root_inode);
1734 EXPORT_SYMBOL(d_make_root);
1736 static struct dentry * __d_find_any_alias(struct inode *inode)
1738 struct dentry *alias;
1740 if (hlist_empty(&inode->i_dentry))
1742 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1748 * d_find_any_alias - find any alias for a given inode
1749 * @inode: inode to find an alias for
1751 * If any aliases exist for the given inode, take and return a
1752 * reference for one of them. If no aliases exist, return %NULL.
1754 struct dentry *d_find_any_alias(struct inode *inode)
1758 spin_lock(&inode->i_lock);
1759 de = __d_find_any_alias(inode);
1760 spin_unlock(&inode->i_lock);
1763 EXPORT_SYMBOL(d_find_any_alias);
1766 * d_obtain_alias - find or allocate a dentry for a given inode
1767 * @inode: inode to allocate the dentry for
1769 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1770 * similar open by handle operations. The returned dentry may be anonymous,
1771 * or may have a full name (if the inode was already in the cache).
1773 * When called on a directory inode, we must ensure that the inode only ever
1774 * has one dentry. If a dentry is found, that is returned instead of
1775 * allocating a new one.
1777 * On successful return, the reference to the inode has been transferred
1778 * to the dentry. In case of an error the reference on the inode is released.
1779 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1780 * be passed in and will be the error will be propagate to the return value,
1781 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1783 struct dentry *d_obtain_alias(struct inode *inode)
1785 static const struct qstr anonstring = QSTR_INIT("/", 1);
1790 return ERR_PTR(-ESTALE);
1792 return ERR_CAST(inode);
1794 res = d_find_any_alias(inode);
1798 tmp = __d_alloc(inode->i_sb, &anonstring);
1800 res = ERR_PTR(-ENOMEM);
1804 spin_lock(&inode->i_lock);
1805 res = __d_find_any_alias(inode);
1807 spin_unlock(&inode->i_lock);
1812 /* attach a disconnected dentry */
1813 spin_lock(&tmp->d_lock);
1814 tmp->d_inode = inode;
1815 tmp->d_flags |= DCACHE_DISCONNECTED;
1816 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1817 hlist_bl_lock(&tmp->d_sb->s_anon);
1818 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1819 hlist_bl_unlock(&tmp->d_sb->s_anon);
1820 spin_unlock(&tmp->d_lock);
1821 spin_unlock(&inode->i_lock);
1822 security_d_instantiate(tmp, inode);
1827 if (res && !IS_ERR(res))
1828 security_d_instantiate(res, inode);
1832 EXPORT_SYMBOL(d_obtain_alias);
1835 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1836 * @inode: the inode which may have a disconnected dentry
1837 * @dentry: a negative dentry which we want to point to the inode.
1839 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1840 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1841 * and return it, else simply d_add the inode to the dentry and return NULL.
1843 * This is needed in the lookup routine of any filesystem that is exportable
1844 * (via knfsd) so that we can build dcache paths to directories effectively.
1846 * If a dentry was found and moved, then it is returned. Otherwise NULL
1847 * is returned. This matches the expected return value of ->lookup.
1849 * Cluster filesystems may call this function with a negative, hashed dentry.
1850 * In that case, we know that the inode will be a regular file, and also this
1851 * will only occur during atomic_open. So we need to check for the dentry
1852 * being already hashed only in the final case.
1854 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1856 struct dentry *new = NULL;
1859 return ERR_CAST(inode);
1861 if (inode && S_ISDIR(inode->i_mode)) {
1862 spin_lock(&inode->i_lock);
1863 new = __d_find_alias(inode, 1);
1865 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1866 spin_unlock(&inode->i_lock);
1867 security_d_instantiate(new, inode);
1868 d_move(new, dentry);
1871 /* already taking inode->i_lock, so d_add() by hand */
1872 __d_instantiate(dentry, inode);
1873 spin_unlock(&inode->i_lock);
1874 security_d_instantiate(dentry, inode);
1878 d_instantiate(dentry, inode);
1879 if (d_unhashed(dentry))
1884 EXPORT_SYMBOL(d_splice_alias);
1887 * d_add_ci - lookup or allocate new dentry with case-exact name
1888 * @inode: the inode case-insensitive lookup has found
1889 * @dentry: the negative dentry that was passed to the parent's lookup func
1890 * @name: the case-exact name to be associated with the returned dentry
1892 * This is to avoid filling the dcache with case-insensitive names to the
1893 * same inode, only the actual correct case is stored in the dcache for
1894 * case-insensitive filesystems.
1896 * For a case-insensitive lookup match and if the the case-exact dentry
1897 * already exists in in the dcache, use it and return it.
1899 * If no entry exists with the exact case name, allocate new dentry with
1900 * the exact case, and return the spliced entry.
1902 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1905 struct dentry *found;
1909 * First check if a dentry matching the name already exists,
1910 * if not go ahead and create it now.
1912 found = d_hash_and_lookup(dentry->d_parent, name);
1913 if (unlikely(IS_ERR(found)))
1916 new = d_alloc(dentry->d_parent, name);
1918 found = ERR_PTR(-ENOMEM);
1922 found = d_splice_alias(inode, new);
1931 * If a matching dentry exists, and it's not negative use it.
1933 * Decrement the reference count to balance the iget() done
1936 if (found->d_inode) {
1937 if (unlikely(found->d_inode != inode)) {
1938 /* This can't happen because bad inodes are unhashed. */
1939 BUG_ON(!is_bad_inode(inode));
1940 BUG_ON(!is_bad_inode(found->d_inode));
1947 * Negative dentry: instantiate it unless the inode is a directory and
1948 * already has a dentry.
1950 new = d_splice_alias(inode, found);
1961 EXPORT_SYMBOL(d_add_ci);
1964 * Do the slow-case of the dentry name compare.
1966 * Unlike the dentry_cmp() function, we need to atomically
1967 * load the name and length information, so that the
1968 * filesystem can rely on them, and can use the 'name' and
1969 * 'len' information without worrying about walking off the
1970 * end of memory etc.
1972 * Thus the read_seqcount_retry() and the "duplicate" info
1973 * in arguments (the low-level filesystem should not look
1974 * at the dentry inode or name contents directly, since
1975 * rename can change them while we're in RCU mode).
1977 enum slow_d_compare {
1983 static noinline enum slow_d_compare slow_dentry_cmp(
1984 const struct dentry *parent,
1985 struct dentry *dentry,
1987 const struct qstr *name)
1989 int tlen = dentry->d_name.len;
1990 const char *tname = dentry->d_name.name;
1992 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1994 return D_COMP_SEQRETRY;
1996 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1997 return D_COMP_NOMATCH;
2002 * __d_lookup_rcu - search for a dentry (racy, store-free)
2003 * @parent: parent dentry
2004 * @name: qstr of name we wish to find
2005 * @seqp: returns d_seq value at the point where the dentry was found
2006 * Returns: dentry, or NULL
2008 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2009 * resolution (store-free path walking) design described in
2010 * Documentation/filesystems/path-lookup.txt.
2012 * This is not to be used outside core vfs.
2014 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2015 * held, and rcu_read_lock held. The returned dentry must not be stored into
2016 * without taking d_lock and checking d_seq sequence count against @seq
2019 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2022 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2023 * the returned dentry, so long as its parent's seqlock is checked after the
2024 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2025 * is formed, giving integrity down the path walk.
2027 * NOTE! The caller *has* to check the resulting dentry against the sequence
2028 * number we've returned before using any of the resulting dentry state!
2030 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2031 const struct qstr *name,
2034 u64 hashlen = name->hash_len;
2035 const unsigned char *str = name->name;
2036 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2037 struct hlist_bl_node *node;
2038 struct dentry *dentry;
2041 * Note: There is significant duplication with __d_lookup_rcu which is
2042 * required to prevent single threaded performance regressions
2043 * especially on architectures where smp_rmb (in seqcounts) are costly.
2044 * Keep the two functions in sync.
2048 * The hash list is protected using RCU.
2050 * Carefully use d_seq when comparing a candidate dentry, to avoid
2051 * races with d_move().
2053 * It is possible that concurrent renames can mess up our list
2054 * walk here and result in missing our dentry, resulting in the
2055 * false-negative result. d_lookup() protects against concurrent
2056 * renames using rename_lock seqlock.
2058 * See Documentation/filesystems/path-lookup.txt for more details.
2060 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2065 * The dentry sequence count protects us from concurrent
2066 * renames, and thus protects parent and name fields.
2068 * The caller must perform a seqcount check in order
2069 * to do anything useful with the returned dentry.
2071 * NOTE! We do a "raw" seqcount_begin here. That means that
2072 * we don't wait for the sequence count to stabilize if it
2073 * is in the middle of a sequence change. If we do the slow
2074 * dentry compare, we will do seqretries until it is stable,
2075 * and if we end up with a successful lookup, we actually
2076 * want to exit RCU lookup anyway.
2078 seq = raw_seqcount_begin(&dentry->d_seq);
2079 if (dentry->d_parent != parent)
2081 if (d_unhashed(dentry))
2084 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2085 if (dentry->d_name.hash != hashlen_hash(hashlen))
2088 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2091 case D_COMP_NOMATCH:
2098 if (dentry->d_name.hash_len != hashlen)
2101 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2108 * d_lookup - search for a dentry
2109 * @parent: parent dentry
2110 * @name: qstr of name we wish to find
2111 * Returns: dentry, or NULL
2113 * d_lookup searches the children of the parent dentry for the name in
2114 * question. If the dentry is found its reference count is incremented and the
2115 * dentry is returned. The caller must use dput to free the entry when it has
2116 * finished using it. %NULL is returned if the dentry does not exist.
2118 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2120 struct dentry *dentry;
2124 seq = read_seqbegin(&rename_lock);
2125 dentry = __d_lookup(parent, name);
2128 } while (read_seqretry(&rename_lock, seq));
2131 EXPORT_SYMBOL(d_lookup);
2134 * __d_lookup - search for a dentry (racy)
2135 * @parent: parent dentry
2136 * @name: qstr of name we wish to find
2137 * Returns: dentry, or NULL
2139 * __d_lookup is like d_lookup, however it may (rarely) return a
2140 * false-negative result due to unrelated rename activity.
2142 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2143 * however it must be used carefully, eg. with a following d_lookup in
2144 * the case of failure.
2146 * __d_lookup callers must be commented.
2148 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2150 unsigned int len = name->len;
2151 unsigned int hash = name->hash;
2152 const unsigned char *str = name->name;
2153 struct hlist_bl_head *b = d_hash(parent, hash);
2154 struct hlist_bl_node *node;
2155 struct dentry *found = NULL;
2156 struct dentry *dentry;
2159 * Note: There is significant duplication with __d_lookup_rcu which is
2160 * required to prevent single threaded performance regressions
2161 * especially on architectures where smp_rmb (in seqcounts) are costly.
2162 * Keep the two functions in sync.
2166 * The hash list is protected using RCU.
2168 * Take d_lock when comparing a candidate dentry, to avoid races
2171 * It is possible that concurrent renames can mess up our list
2172 * walk here and result in missing our dentry, resulting in the
2173 * false-negative result. d_lookup() protects against concurrent
2174 * renames using rename_lock seqlock.
2176 * See Documentation/filesystems/path-lookup.txt for more details.
2180 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2182 if (dentry->d_name.hash != hash)
2185 spin_lock(&dentry->d_lock);
2186 if (dentry->d_parent != parent)
2188 if (d_unhashed(dentry))
2192 * It is safe to compare names since d_move() cannot
2193 * change the qstr (protected by d_lock).
2195 if (parent->d_flags & DCACHE_OP_COMPARE) {
2196 int tlen = dentry->d_name.len;
2197 const char *tname = dentry->d_name.name;
2198 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2201 if (dentry->d_name.len != len)
2203 if (dentry_cmp(dentry, str, len))
2207 dentry->d_lockref.count++;
2209 spin_unlock(&dentry->d_lock);
2212 spin_unlock(&dentry->d_lock);
2220 * d_hash_and_lookup - hash the qstr then search for a dentry
2221 * @dir: Directory to search in
2222 * @name: qstr of name we wish to find
2224 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2226 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2229 * Check for a fs-specific hash function. Note that we must
2230 * calculate the standard hash first, as the d_op->d_hash()
2231 * routine may choose to leave the hash value unchanged.
2233 name->hash = full_name_hash(name->name, name->len);
2234 if (dir->d_flags & DCACHE_OP_HASH) {
2235 int err = dir->d_op->d_hash(dir, name);
2236 if (unlikely(err < 0))
2237 return ERR_PTR(err);
2239 return d_lookup(dir, name);
2241 EXPORT_SYMBOL(d_hash_and_lookup);
2244 * d_validate - verify dentry provided from insecure source (deprecated)
2245 * @dentry: The dentry alleged to be valid child of @dparent
2246 * @dparent: The parent dentry (known to be valid)
2248 * An insecure source has sent us a dentry, here we verify it and dget() it.
2249 * This is used by ncpfs in its readdir implementation.
2250 * Zero is returned in the dentry is invalid.
2252 * This function is slow for big directories, and deprecated, do not use it.
2254 int d_validate(struct dentry *dentry, struct dentry *dparent)
2256 struct dentry *child;
2258 spin_lock(&dparent->d_lock);
2259 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2260 if (dentry == child) {
2261 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2262 __dget_dlock(dentry);
2263 spin_unlock(&dentry->d_lock);
2264 spin_unlock(&dparent->d_lock);
2268 spin_unlock(&dparent->d_lock);
2272 EXPORT_SYMBOL(d_validate);
2275 * When a file is deleted, we have two options:
2276 * - turn this dentry into a negative dentry
2277 * - unhash this dentry and free it.
2279 * Usually, we want to just turn this into
2280 * a negative dentry, but if anybody else is
2281 * currently using the dentry or the inode
2282 * we can't do that and we fall back on removing
2283 * it from the hash queues and waiting for
2284 * it to be deleted later when it has no users
2288 * d_delete - delete a dentry
2289 * @dentry: The dentry to delete
2291 * Turn the dentry into a negative dentry if possible, otherwise
2292 * remove it from the hash queues so it can be deleted later
2295 void d_delete(struct dentry * dentry)
2297 struct inode *inode;
2300 * Are we the only user?
2303 spin_lock(&dentry->d_lock);
2304 inode = dentry->d_inode;
2305 isdir = S_ISDIR(inode->i_mode);
2306 if (dentry->d_lockref.count == 1) {
2307 if (!spin_trylock(&inode->i_lock)) {
2308 spin_unlock(&dentry->d_lock);
2312 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2313 dentry_unlink_inode(dentry);
2314 fsnotify_nameremove(dentry, isdir);
2318 if (!d_unhashed(dentry))
2321 spin_unlock(&dentry->d_lock);
2323 fsnotify_nameremove(dentry, isdir);
2325 EXPORT_SYMBOL(d_delete);
2327 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2329 BUG_ON(!d_unhashed(entry));
2331 entry->d_flags |= DCACHE_RCUACCESS;
2332 hlist_bl_add_head_rcu(&entry->d_hash, b);
2336 static void _d_rehash(struct dentry * entry)
2338 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2342 * d_rehash - add an entry back to the hash
2343 * @entry: dentry to add to the hash
2345 * Adds a dentry to the hash according to its name.
2348 void d_rehash(struct dentry * entry)
2350 spin_lock(&entry->d_lock);
2352 spin_unlock(&entry->d_lock);
2354 EXPORT_SYMBOL(d_rehash);
2357 * dentry_update_name_case - update case insensitive dentry with a new name
2358 * @dentry: dentry to be updated
2361 * Update a case insensitive dentry with new case of name.
2363 * dentry must have been returned by d_lookup with name @name. Old and new
2364 * name lengths must match (ie. no d_compare which allows mismatched name
2367 * Parent inode i_mutex must be held over d_lookup and into this call (to
2368 * keep renames and concurrent inserts, and readdir(2) away).
2370 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2372 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2373 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2375 spin_lock(&dentry->d_lock);
2376 write_seqcount_begin(&dentry->d_seq);
2377 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2378 write_seqcount_end(&dentry->d_seq);
2379 spin_unlock(&dentry->d_lock);
2381 EXPORT_SYMBOL(dentry_update_name_case);
2383 static void switch_names(struct dentry *dentry, struct dentry *target)
2385 if (dname_external(target)) {
2386 if (dname_external(dentry)) {
2388 * Both external: swap the pointers
2390 swap(target->d_name.name, dentry->d_name.name);
2393 * dentry:internal, target:external. Steal target's
2394 * storage and make target internal.
2396 memcpy(target->d_iname, dentry->d_name.name,
2397 dentry->d_name.len + 1);
2398 dentry->d_name.name = target->d_name.name;
2399 target->d_name.name = target->d_iname;
2402 if (dname_external(dentry)) {
2404 * dentry:external, target:internal. Give dentry's
2405 * storage to target and make dentry internal
2407 memcpy(dentry->d_iname, target->d_name.name,
2408 target->d_name.len + 1);
2409 target->d_name.name = dentry->d_name.name;
2410 dentry->d_name.name = dentry->d_iname;
2413 * Both are internal. Just copy target to dentry
2415 memcpy(dentry->d_iname, target->d_name.name,
2416 target->d_name.len + 1);
2417 dentry->d_name.len = target->d_name.len;
2421 swap(dentry->d_name.len, target->d_name.len);
2424 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2427 * XXXX: do we really need to take target->d_lock?
2429 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2430 spin_lock(&target->d_parent->d_lock);
2432 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2433 spin_lock(&dentry->d_parent->d_lock);
2434 spin_lock_nested(&target->d_parent->d_lock,
2435 DENTRY_D_LOCK_NESTED);
2437 spin_lock(&target->d_parent->d_lock);
2438 spin_lock_nested(&dentry->d_parent->d_lock,
2439 DENTRY_D_LOCK_NESTED);
2442 if (target < dentry) {
2443 spin_lock_nested(&target->d_lock, 2);
2444 spin_lock_nested(&dentry->d_lock, 3);
2446 spin_lock_nested(&dentry->d_lock, 2);
2447 spin_lock_nested(&target->d_lock, 3);
2451 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2452 struct dentry *target)
2454 if (target->d_parent != dentry->d_parent)
2455 spin_unlock(&dentry->d_parent->d_lock);
2456 if (target->d_parent != target)
2457 spin_unlock(&target->d_parent->d_lock);
2461 * When switching names, the actual string doesn't strictly have to
2462 * be preserved in the target - because we're dropping the target
2463 * anyway. As such, we can just do a simple memcpy() to copy over
2464 * the new name before we switch.
2466 * Note that we have to be a lot more careful about getting the hash
2467 * switched - we have to switch the hash value properly even if it
2468 * then no longer matches the actual (corrupted) string of the target.
2469 * The hash value has to match the hash queue that the dentry is on..
2472 * __d_move - move a dentry
2473 * @dentry: entry to move
2474 * @target: new dentry
2476 * Update the dcache to reflect the move of a file name. Negative
2477 * dcache entries should not be moved in this way. Caller must hold
2478 * rename_lock, the i_mutex of the source and target directories,
2479 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2481 static void __d_move(struct dentry * dentry, struct dentry * target)
2483 if (!dentry->d_inode)
2484 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2486 BUG_ON(d_ancestor(dentry, target));
2487 BUG_ON(d_ancestor(target, dentry));
2489 dentry_lock_for_move(dentry, target);
2491 write_seqcount_begin(&dentry->d_seq);
2492 write_seqcount_begin(&target->d_seq);
2494 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2497 * Move the dentry to the target hash queue. Don't bother checking
2498 * for the same hash queue because of how unlikely it is.
2501 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2503 /* Unhash the target: dput() will then get rid of it */
2506 list_del(&dentry->d_u.d_child);
2507 list_del(&target->d_u.d_child);
2509 /* Switch the names.. */
2510 switch_names(dentry, target);
2511 swap(dentry->d_name.hash, target->d_name.hash);
2513 /* ... and switch the parents */
2514 if (IS_ROOT(dentry)) {
2515 dentry->d_parent = target->d_parent;
2516 target->d_parent = target;
2517 INIT_LIST_HEAD(&target->d_u.d_child);
2519 swap(dentry->d_parent, target->d_parent);
2521 /* And add them back to the (new) parent lists */
2522 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2525 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2527 write_seqcount_end(&target->d_seq);
2528 write_seqcount_end(&dentry->d_seq);
2530 dentry_unlock_parents_for_move(dentry, target);
2531 spin_unlock(&target->d_lock);
2532 fsnotify_d_move(dentry);
2533 spin_unlock(&dentry->d_lock);
2537 * d_move - move a dentry
2538 * @dentry: entry to move
2539 * @target: new dentry
2541 * Update the dcache to reflect the move of a file name. Negative
2542 * dcache entries should not be moved in this way. See the locking
2543 * requirements for __d_move.
2545 void d_move(struct dentry *dentry, struct dentry *target)
2547 write_seqlock(&rename_lock);
2548 __d_move(dentry, target);
2549 write_sequnlock(&rename_lock);
2551 EXPORT_SYMBOL(d_move);
2554 * d_ancestor - search for an ancestor
2555 * @p1: ancestor dentry
2558 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2559 * an ancestor of p2, else NULL.
2561 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2565 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2566 if (p->d_parent == p1)
2573 * This helper attempts to cope with remotely renamed directories
2575 * It assumes that the caller is already holding
2576 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2578 * Note: If ever the locking in lock_rename() changes, then please
2579 * remember to update this too...
2581 static struct dentry *__d_unalias(struct inode *inode,
2582 struct dentry *dentry, struct dentry *alias)
2584 struct mutex *m1 = NULL, *m2 = NULL;
2585 struct dentry *ret = ERR_PTR(-EBUSY);
2587 /* If alias and dentry share a parent, then no extra locks required */
2588 if (alias->d_parent == dentry->d_parent)
2591 /* See lock_rename() */
2592 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2594 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2595 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2597 m2 = &alias->d_parent->d_inode->i_mutex;
2599 if (likely(!d_mountpoint(alias))) {
2600 __d_move(alias, dentry);
2604 spin_unlock(&inode->i_lock);
2613 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2614 * named dentry in place of the dentry to be replaced.
2615 * returns with anon->d_lock held!
2617 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2619 struct dentry *dparent;
2621 dentry_lock_for_move(anon, dentry);
2623 write_seqcount_begin(&dentry->d_seq);
2624 write_seqcount_begin(&anon->d_seq);
2626 dparent = dentry->d_parent;
2628 switch_names(dentry, anon);
2629 swap(dentry->d_name.hash, anon->d_name.hash);
2631 dentry->d_parent = dentry;
2632 list_del_init(&dentry->d_u.d_child);
2633 anon->d_parent = dparent;
2634 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2636 write_seqcount_end(&dentry->d_seq);
2637 write_seqcount_end(&anon->d_seq);
2639 dentry_unlock_parents_for_move(anon, dentry);
2640 spin_unlock(&dentry->d_lock);
2642 /* anon->d_lock still locked, returns locked */
2643 anon->d_flags &= ~DCACHE_DISCONNECTED;
2647 * d_materialise_unique - introduce an inode into the tree
2648 * @dentry: candidate dentry
2649 * @inode: inode to bind to the dentry, to which aliases may be attached
2651 * Introduces an dentry into the tree, substituting an extant disconnected
2652 * root directory alias in its place if there is one. Caller must hold the
2653 * i_mutex of the parent directory.
2655 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2657 struct dentry *actual;
2659 BUG_ON(!d_unhashed(dentry));
2663 __d_instantiate(dentry, NULL);
2668 spin_lock(&inode->i_lock);
2670 if (S_ISDIR(inode->i_mode)) {
2671 struct dentry *alias;
2673 /* Does an aliased dentry already exist? */
2674 alias = __d_find_alias(inode, 0);
2677 write_seqlock(&rename_lock);
2679 if (d_ancestor(alias, dentry)) {
2680 /* Check for loops */
2681 actual = ERR_PTR(-ELOOP);
2682 spin_unlock(&inode->i_lock);
2683 } else if (IS_ROOT(alias)) {
2684 /* Is this an anonymous mountpoint that we
2685 * could splice into our tree? */
2686 __d_materialise_dentry(dentry, alias);
2687 write_sequnlock(&rename_lock);
2691 /* Nope, but we must(!) avoid directory
2692 * aliasing. This drops inode->i_lock */
2693 actual = __d_unalias(inode, dentry, alias);
2695 write_sequnlock(&rename_lock);
2696 if (IS_ERR(actual)) {
2697 if (PTR_ERR(actual) == -ELOOP)
2698 pr_warn_ratelimited(
2699 "VFS: Lookup of '%s' in %s %s"
2700 " would have caused loop\n",
2701 dentry->d_name.name,
2702 inode->i_sb->s_type->name,
2710 /* Add a unique reference */
2711 actual = __d_instantiate_unique(dentry, inode);
2715 BUG_ON(!d_unhashed(actual));
2717 spin_lock(&actual->d_lock);
2720 spin_unlock(&actual->d_lock);
2721 spin_unlock(&inode->i_lock);
2723 if (actual == dentry) {
2724 security_d_instantiate(dentry, inode);
2731 EXPORT_SYMBOL_GPL(d_materialise_unique);
2733 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2737 return -ENAMETOOLONG;
2739 memcpy(*buffer, str, namelen);
2744 * prepend_name - prepend a pathname in front of current buffer pointer
2745 * buffer: buffer pointer
2746 * buflen: allocated length of the buffer
2747 * name: name string and length qstr structure
2749 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2750 * make sure that either the old or the new name pointer and length are
2751 * fetched. However, there may be mismatch between length and pointer.
2752 * The length cannot be trusted, we need to copy it byte-by-byte until
2753 * the length is reached or a null byte is found. It also prepends "/" at
2754 * the beginning of the name. The sequence number check at the caller will
2755 * retry it again when a d_move() does happen. So any garbage in the buffer
2756 * due to mismatched pointer and length will be discarded.
2758 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2760 const char *dname = ACCESS_ONCE(name->name);
2761 u32 dlen = ACCESS_ONCE(name->len);
2764 if (*buflen < dlen + 1)
2765 return -ENAMETOOLONG;
2766 *buflen -= dlen + 1;
2767 p = *buffer -= dlen + 1;
2779 * prepend_path - Prepend path string to a buffer
2780 * @path: the dentry/vfsmount to report
2781 * @root: root vfsmnt/dentry
2782 * @buffer: pointer to the end of the buffer
2783 * @buflen: pointer to buffer length
2785 * The function tries to write out the pathname without taking any lock other
2786 * than the RCU read lock to make sure that dentries won't go away. It only
2787 * checks the sequence number of the global rename_lock as any change in the
2788 * dentry's d_seq will be preceded by changes in the rename_lock sequence
2789 * number. If the sequence number had been change, it will restart the whole
2790 * pathname back-tracing sequence again. It performs a total of 3 trials of
2791 * lockless back-tracing sequences before falling back to take the
2794 static int prepend_path(const struct path *path,
2795 const struct path *root,
2796 char **buffer, int *buflen)
2798 struct dentry *dentry = path->dentry;
2799 struct vfsmount *vfsmnt = path->mnt;
2800 struct mount *mnt = real_mount(vfsmnt);
2810 read_seqbegin_or_lock(&rename_lock, &seq);
2811 while (dentry != root->dentry || vfsmnt != root->mnt) {
2812 struct dentry * parent;
2814 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2816 if (mnt_has_parent(mnt)) {
2817 dentry = mnt->mnt_mountpoint;
2818 mnt = mnt->mnt_parent;
2823 * Filesystems needing to implement special "root names"
2824 * should do so with ->d_dname()
2826 if (IS_ROOT(dentry) &&
2827 (dentry->d_name.len != 1 ||
2828 dentry->d_name.name[0] != '/')) {
2829 WARN(1, "Root dentry has weird name <%.*s>\n",
2830 (int) dentry->d_name.len,
2831 dentry->d_name.name);
2834 error = is_mounted(vfsmnt) ? 1 : 2;
2837 parent = dentry->d_parent;
2839 error = prepend_name(&bptr, &blen, &dentry->d_name);
2847 if (need_seqretry(&rename_lock, seq)) {
2851 done_seqretry(&rename_lock, seq);
2853 if (error >= 0 && bptr == *buffer) {
2855 error = -ENAMETOOLONG;
2865 * __d_path - return the path of a dentry
2866 * @path: the dentry/vfsmount to report
2867 * @root: root vfsmnt/dentry
2868 * @buf: buffer to return value in
2869 * @buflen: buffer length
2871 * Convert a dentry into an ASCII path name.
2873 * Returns a pointer into the buffer or an error code if the
2874 * path was too long.
2876 * "buflen" should be positive.
2878 * If the path is not reachable from the supplied root, return %NULL.
2880 char *__d_path(const struct path *path,
2881 const struct path *root,
2882 char *buf, int buflen)
2884 char *res = buf + buflen;
2887 prepend(&res, &buflen, "\0", 1);
2888 br_read_lock(&vfsmount_lock);
2889 error = prepend_path(path, root, &res, &buflen);
2890 br_read_unlock(&vfsmount_lock);
2893 return ERR_PTR(error);
2899 char *d_absolute_path(const struct path *path,
2900 char *buf, int buflen)
2902 struct path root = {};
2903 char *res = buf + buflen;
2906 prepend(&res, &buflen, "\0", 1);
2907 br_read_lock(&vfsmount_lock);
2908 error = prepend_path(path, &root, &res, &buflen);
2909 br_read_unlock(&vfsmount_lock);
2914 return ERR_PTR(error);
2919 * same as __d_path but appends "(deleted)" for unlinked files.
2921 static int path_with_deleted(const struct path *path,
2922 const struct path *root,
2923 char **buf, int *buflen)
2925 prepend(buf, buflen, "\0", 1);
2926 if (d_unlinked(path->dentry)) {
2927 int error = prepend(buf, buflen, " (deleted)", 10);
2932 return prepend_path(path, root, buf, buflen);
2935 static int prepend_unreachable(char **buffer, int *buflen)
2937 return prepend(buffer, buflen, "(unreachable)", 13);
2941 * d_path - return the path of a dentry
2942 * @path: path to report
2943 * @buf: buffer to return value in
2944 * @buflen: buffer length
2946 * Convert a dentry into an ASCII path name. If the entry has been deleted
2947 * the string " (deleted)" is appended. Note that this is ambiguous.
2949 * Returns a pointer into the buffer or an error code if the path was
2950 * too long. Note: Callers should use the returned pointer, not the passed
2951 * in buffer, to use the name! The implementation often starts at an offset
2952 * into the buffer, and may leave 0 bytes at the start.
2954 * "buflen" should be positive.
2956 char *d_path(const struct path *path, char *buf, int buflen)
2958 char *res = buf + buflen;
2963 * We have various synthetic filesystems that never get mounted. On
2964 * these filesystems dentries are never used for lookup purposes, and
2965 * thus don't need to be hashed. They also don't need a name until a
2966 * user wants to identify the object in /proc/pid/fd/. The little hack
2967 * below allows us to generate a name for these objects on demand:
2969 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2970 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2972 get_fs_root(current->fs, &root);
2973 br_read_lock(&vfsmount_lock);
2974 error = path_with_deleted(path, &root, &res, &buflen);
2975 br_read_unlock(&vfsmount_lock);
2977 res = ERR_PTR(error);
2981 EXPORT_SYMBOL(d_path);
2984 * Helper function for dentry_operations.d_dname() members
2986 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2987 const char *fmt, ...)
2993 va_start(args, fmt);
2994 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2997 if (sz > sizeof(temp) || sz > buflen)
2998 return ERR_PTR(-ENAMETOOLONG);
3000 buffer += buflen - sz;
3001 return memcpy(buffer, temp, sz);
3004 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3006 char *end = buffer + buflen;
3007 /* these dentries are never renamed, so d_lock is not needed */
3008 if (prepend(&end, &buflen, " (deleted)", 11) ||
3009 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3010 prepend(&end, &buflen, "/", 1))
3011 end = ERR_PTR(-ENAMETOOLONG);
3016 * Write full pathname from the root of the filesystem into the buffer.
3018 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
3028 prepend(&end, &len, "\0", 1);
3034 read_seqbegin_or_lock(&rename_lock, &seq);
3035 while (!IS_ROOT(dentry)) {
3036 struct dentry *parent = dentry->d_parent;
3040 error = prepend_name(&end, &len, &dentry->d_name);
3049 if (need_seqretry(&rename_lock, seq)) {
3053 done_seqretry(&rename_lock, seq);
3058 return ERR_PTR(-ENAMETOOLONG);
3061 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3063 return __dentry_path(dentry, buf, buflen);
3065 EXPORT_SYMBOL(dentry_path_raw);
3067 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3072 if (d_unlinked(dentry)) {
3074 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3078 retval = __dentry_path(dentry, buf, buflen);
3079 if (!IS_ERR(retval) && p)
3080 *p = '/'; /* restore '/' overriden with '\0' */
3083 return ERR_PTR(-ENAMETOOLONG);
3087 * NOTE! The user-level library version returns a
3088 * character pointer. The kernel system call just
3089 * returns the length of the buffer filled (which
3090 * includes the ending '\0' character), or a negative
3091 * error value. So libc would do something like
3093 * char *getcwd(char * buf, size_t size)
3097 * retval = sys_getcwd(buf, size);
3104 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3107 struct path pwd, root;
3108 char *page = (char *) __get_free_page(GFP_USER);
3113 get_fs_root_and_pwd(current->fs, &root, &pwd);
3116 br_read_lock(&vfsmount_lock);
3117 if (!d_unlinked(pwd.dentry)) {
3119 char *cwd = page + PAGE_SIZE;
3120 int buflen = PAGE_SIZE;
3122 prepend(&cwd, &buflen, "\0", 1);
3123 error = prepend_path(&pwd, &root, &cwd, &buflen);
3124 br_read_unlock(&vfsmount_lock);
3129 /* Unreachable from current root */
3131 error = prepend_unreachable(&cwd, &buflen);
3137 len = PAGE_SIZE + page - cwd;
3140 if (copy_to_user(buf, cwd, len))
3144 br_read_unlock(&vfsmount_lock);
3150 free_page((unsigned long) page);
3155 * Test whether new_dentry is a subdirectory of old_dentry.
3157 * Trivially implemented using the dcache structure
3161 * is_subdir - is new dentry a subdirectory of old_dentry
3162 * @new_dentry: new dentry
3163 * @old_dentry: old dentry
3165 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3166 * Returns 0 otherwise.
3167 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3170 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3175 if (new_dentry == old_dentry)
3179 /* for restarting inner loop in case of seq retry */
3180 seq = read_seqbegin(&rename_lock);
3182 * Need rcu_readlock to protect against the d_parent trashing
3186 if (d_ancestor(old_dentry, new_dentry))
3191 } while (read_seqretry(&rename_lock, seq));
3196 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3198 struct dentry *root = data;
3199 if (dentry != root) {
3200 if (d_unhashed(dentry) || !dentry->d_inode)
3203 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3204 dentry->d_flags |= DCACHE_GENOCIDE;
3205 dentry->d_lockref.count--;
3208 return D_WALK_CONTINUE;
3211 void d_genocide(struct dentry *parent)
3213 d_walk(parent, parent, d_genocide_kill, NULL);
3216 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3218 inode_dec_link_count(inode);
3219 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3220 !hlist_unhashed(&dentry->d_alias) ||
3221 !d_unlinked(dentry));
3222 spin_lock(&dentry->d_parent->d_lock);
3223 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3224 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3225 (unsigned long long)inode->i_ino);
3226 spin_unlock(&dentry->d_lock);
3227 spin_unlock(&dentry->d_parent->d_lock);
3228 d_instantiate(dentry, inode);
3230 EXPORT_SYMBOL(d_tmpfile);
3232 static __initdata unsigned long dhash_entries;
3233 static int __init set_dhash_entries(char *str)
3237 dhash_entries = simple_strtoul(str, &str, 0);
3240 __setup("dhash_entries=", set_dhash_entries);
3242 static void __init dcache_init_early(void)
3246 /* If hashes are distributed across NUMA nodes, defer
3247 * hash allocation until vmalloc space is available.
3253 alloc_large_system_hash("Dentry cache",
3254 sizeof(struct hlist_bl_head),
3263 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3264 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3267 static void __init dcache_init(void)
3272 * A constructor could be added for stable state like the lists,
3273 * but it is probably not worth it because of the cache nature
3276 dentry_cache = KMEM_CACHE(dentry,
3277 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3279 /* Hash may have been set up in dcache_init_early */
3284 alloc_large_system_hash("Dentry cache",
3285 sizeof(struct hlist_bl_head),
3294 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3295 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3298 /* SLAB cache for __getname() consumers */
3299 struct kmem_cache *names_cachep __read_mostly;
3300 EXPORT_SYMBOL(names_cachep);
3302 EXPORT_SYMBOL(d_genocide);
3304 void __init vfs_caches_init_early(void)
3306 dcache_init_early();
3310 void __init vfs_caches_init(unsigned long mempages)
3312 unsigned long reserve;
3314 /* Base hash sizes on available memory, with a reserve equal to
3315 150% of current kernel size */
3317 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3318 mempages -= reserve;
3320 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3321 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3325 files_init(mempages);