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);
959 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
960 spinlock_t *lru_lock, void *arg)
962 struct list_head *freeable = arg;
963 struct dentry *dentry = container_of(item, struct dentry, d_lru);
966 * we are inverting the lru lock/dentry->d_lock here,
967 * so use a trylock. If we fail to get the lock, just skip
970 if (!spin_trylock(&dentry->d_lock))
973 dentry->d_flags |= DCACHE_SHRINK_LIST;
974 list_move_tail(&dentry->d_lru, freeable);
975 this_cpu_dec(nr_dentry_unused);
976 spin_unlock(&dentry->d_lock);
983 * shrink_dcache_sb - shrink dcache for a superblock
986 * Shrink the dcache for the specified super block. This is used to free
987 * the dcache before unmounting a file system.
989 void shrink_dcache_sb(struct super_block *sb)
996 freed = list_lru_walk(&sb->s_dentry_lru,
997 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
999 this_cpu_sub(nr_dentry_unused, freed);
1000 shrink_dentry_list(&dispose);
1001 } while (freed > 0);
1003 EXPORT_SYMBOL(shrink_dcache_sb);
1006 * destroy a single subtree of dentries for unmount
1007 * - see the comments on shrink_dcache_for_umount() for a description of the
1010 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
1012 struct dentry *parent;
1014 BUG_ON(!IS_ROOT(dentry));
1017 /* descend to the first leaf in the current subtree */
1018 while (!list_empty(&dentry->d_subdirs))
1019 dentry = list_entry(dentry->d_subdirs.next,
1020 struct dentry, d_u.d_child);
1022 /* consume the dentries from this leaf up through its parents
1023 * until we find one with children or run out altogether */
1025 struct inode *inode;
1028 * inform the fs that this dentry is about to be
1029 * unhashed and destroyed.
1031 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
1032 !d_unhashed(dentry))
1033 dentry->d_op->d_prune(dentry);
1035 dentry_lru_del(dentry);
1038 if (dentry->d_lockref.count != 0) {
1040 "BUG: Dentry %p{i=%lx,n=%s}"
1041 " still in use (%d)"
1042 " [unmount of %s %s]\n",
1045 dentry->d_inode->i_ino : 0UL,
1046 dentry->d_name.name,
1047 dentry->d_lockref.count,
1048 dentry->d_sb->s_type->name,
1049 dentry->d_sb->s_id);
1053 if (IS_ROOT(dentry)) {
1055 list_del(&dentry->d_u.d_child);
1057 parent = dentry->d_parent;
1058 parent->d_lockref.count--;
1059 list_del(&dentry->d_u.d_child);
1062 inode = dentry->d_inode;
1064 dentry->d_inode = NULL;
1065 hlist_del_init(&dentry->d_alias);
1066 if (dentry->d_op && dentry->d_op->d_iput)
1067 dentry->d_op->d_iput(dentry, inode);
1074 /* finished when we fall off the top of the tree,
1075 * otherwise we ascend to the parent and move to the
1076 * next sibling if there is one */
1080 } while (list_empty(&dentry->d_subdirs));
1082 dentry = list_entry(dentry->d_subdirs.next,
1083 struct dentry, d_u.d_child);
1088 * destroy the dentries attached to a superblock on unmounting
1089 * - we don't need to use dentry->d_lock because:
1090 * - the superblock is detached from all mountings and open files, so the
1091 * dentry trees will not be rearranged by the VFS
1092 * - s_umount is write-locked, so the memory pressure shrinker will ignore
1093 * any dentries belonging to this superblock that it comes across
1094 * - the filesystem itself is no longer permitted to rearrange the dentries
1095 * in this superblock
1097 void shrink_dcache_for_umount(struct super_block *sb)
1099 struct dentry *dentry;
1101 if (down_read_trylock(&sb->s_umount))
1104 dentry = sb->s_root;
1106 dentry->d_lockref.count--;
1107 shrink_dcache_for_umount_subtree(dentry);
1109 while (!hlist_bl_empty(&sb->s_anon)) {
1110 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1111 shrink_dcache_for_umount_subtree(dentry);
1116 * This tries to ascend one level of parenthood, but
1117 * we can race with renaming, so we need to re-check
1118 * the parenthood after dropping the lock and check
1119 * that the sequence number still matches.
1121 static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
1123 struct dentry *new = old->d_parent;
1126 spin_unlock(&old->d_lock);
1127 spin_lock(&new->d_lock);
1130 * might go back up the wrong parent if we have had a rename
1133 if (new != old->d_parent ||
1134 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1135 need_seqretry(&rename_lock, seq)) {
1136 spin_unlock(&new->d_lock);
1144 * enum d_walk_ret - action to talke during tree walk
1145 * @D_WALK_CONTINUE: contrinue walk
1146 * @D_WALK_QUIT: quit walk
1147 * @D_WALK_NORETRY: quit when retry is needed
1148 * @D_WALK_SKIP: skip this dentry and its children
1158 * d_walk - walk the dentry tree
1159 * @parent: start of walk
1160 * @data: data passed to @enter() and @finish()
1161 * @enter: callback when first entering the dentry
1162 * @finish: callback when successfully finished the walk
1164 * The @enter() and @finish() callbacks are called with d_lock held.
1166 static void d_walk(struct dentry *parent, void *data,
1167 enum d_walk_ret (*enter)(void *, struct dentry *),
1168 void (*finish)(void *))
1170 struct dentry *this_parent;
1171 struct list_head *next;
1173 enum d_walk_ret ret;
1177 read_seqbegin_or_lock(&rename_lock, &seq);
1178 this_parent = parent;
1179 spin_lock(&this_parent->d_lock);
1181 ret = enter(data, this_parent);
1183 case D_WALK_CONTINUE:
1188 case D_WALK_NORETRY:
1193 next = this_parent->d_subdirs.next;
1195 while (next != &this_parent->d_subdirs) {
1196 struct list_head *tmp = next;
1197 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1200 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1202 ret = enter(data, dentry);
1204 case D_WALK_CONTINUE:
1207 spin_unlock(&dentry->d_lock);
1209 case D_WALK_NORETRY:
1213 spin_unlock(&dentry->d_lock);
1217 if (!list_empty(&dentry->d_subdirs)) {
1218 spin_unlock(&this_parent->d_lock);
1219 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1220 this_parent = dentry;
1221 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1224 spin_unlock(&dentry->d_lock);
1227 * All done at this level ... ascend and resume the search.
1229 if (this_parent != parent) {
1230 struct dentry *child = this_parent;
1231 this_parent = try_to_ascend(this_parent, seq);
1234 next = child->d_u.d_child.next;
1237 if (need_seqretry(&rename_lock, seq)) {
1238 spin_unlock(&this_parent->d_lock);
1245 spin_unlock(&this_parent->d_lock);
1246 done_seqretry(&rename_lock, seq);
1257 * Search for at least 1 mount point in the dentry's subdirs.
1258 * We descend to the next level whenever the d_subdirs
1259 * list is non-empty and continue searching.
1263 * have_submounts - check for mounts over a dentry
1264 * @parent: dentry to check.
1266 * Return true if the parent or its subdirectories contain
1270 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1273 if (d_mountpoint(dentry)) {
1277 return D_WALK_CONTINUE;
1280 int have_submounts(struct dentry *parent)
1284 d_walk(parent, &ret, check_mount, NULL);
1288 EXPORT_SYMBOL(have_submounts);
1291 * Called by mount code to set a mountpoint and check if the mountpoint is
1292 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1293 * subtree can become unreachable).
1295 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1296 * this reason take rename_lock and d_lock on dentry and ancestors.
1298 int d_set_mounted(struct dentry *dentry)
1302 write_seqlock(&rename_lock);
1303 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1304 /* Need exclusion wrt. check_submounts_and_drop() */
1305 spin_lock(&p->d_lock);
1306 if (unlikely(d_unhashed(p))) {
1307 spin_unlock(&p->d_lock);
1310 spin_unlock(&p->d_lock);
1312 spin_lock(&dentry->d_lock);
1313 if (!d_unlinked(dentry)) {
1314 dentry->d_flags |= DCACHE_MOUNTED;
1317 spin_unlock(&dentry->d_lock);
1319 write_sequnlock(&rename_lock);
1324 * Search the dentry child list of the specified parent,
1325 * and move any unused dentries to the end of the unused
1326 * list for prune_dcache(). We descend to the next level
1327 * whenever the d_subdirs list is non-empty and continue
1330 * It returns zero iff there are no unused children,
1331 * otherwise it returns the number of children moved to
1332 * the end of the unused list. This may not be the total
1333 * number of unused children, because select_parent can
1334 * drop the lock and return early due to latency
1338 struct select_data {
1339 struct dentry *start;
1340 struct list_head dispose;
1344 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1346 struct select_data *data = _data;
1347 enum d_walk_ret ret = D_WALK_CONTINUE;
1349 if (data->start == dentry)
1353 * move only zero ref count dentries to the dispose list.
1355 * Those which are presently on the shrink list, being processed
1356 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1357 * loop in shrink_dcache_parent() might not make any progress
1360 if (dentry->d_lockref.count) {
1361 dentry_lru_del(dentry);
1362 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1363 dentry_lru_del(dentry);
1364 list_add_tail(&dentry->d_lru, &data->dispose);
1365 dentry->d_flags |= DCACHE_SHRINK_LIST;
1367 ret = D_WALK_NORETRY;
1370 * We can return to the caller if we have found some (this
1371 * ensures forward progress). We'll be coming back to find
1374 if (data->found && need_resched())
1381 * shrink_dcache_parent - prune dcache
1382 * @parent: parent of entries to prune
1384 * Prune the dcache to remove unused children of the parent dentry.
1386 void shrink_dcache_parent(struct dentry *parent)
1389 struct select_data data;
1391 INIT_LIST_HEAD(&data.dispose);
1392 data.start = parent;
1395 d_walk(parent, &data, select_collect, NULL);
1399 shrink_dentry_list(&data.dispose);
1403 EXPORT_SYMBOL(shrink_dcache_parent);
1405 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1407 struct select_data *data = _data;
1409 if (d_mountpoint(dentry)) {
1410 data->found = -EBUSY;
1414 return select_collect(_data, dentry);
1417 static void check_and_drop(void *_data)
1419 struct select_data *data = _data;
1421 if (d_mountpoint(data->start))
1422 data->found = -EBUSY;
1424 __d_drop(data->start);
1428 * check_submounts_and_drop - prune dcache, check for submounts and drop
1430 * All done as a single atomic operation relative to has_unlinked_ancestor().
1431 * Returns 0 if successfully unhashed @parent. If there were submounts then
1434 * @dentry: dentry to prune and drop
1436 int check_submounts_and_drop(struct dentry *dentry)
1440 /* Negative dentries can be dropped without further checks */
1441 if (!dentry->d_inode) {
1447 struct select_data data;
1449 INIT_LIST_HEAD(&data.dispose);
1450 data.start = dentry;
1453 d_walk(dentry, &data, check_and_collect, check_and_drop);
1456 if (!list_empty(&data.dispose))
1457 shrink_dentry_list(&data.dispose);
1468 EXPORT_SYMBOL(check_submounts_and_drop);
1471 * __d_alloc - allocate a dcache entry
1472 * @sb: filesystem it will belong to
1473 * @name: qstr of the name
1475 * Allocates a dentry. It returns %NULL if there is insufficient memory
1476 * available. On a success the dentry is returned. The name passed in is
1477 * copied and the copy passed in may be reused after this call.
1480 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1482 struct dentry *dentry;
1485 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1490 * We guarantee that the inline name is always NUL-terminated.
1491 * This way the memcpy() done by the name switching in rename
1492 * will still always have a NUL at the end, even if we might
1493 * be overwriting an internal NUL character
1495 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1496 if (name->len > DNAME_INLINE_LEN-1) {
1497 dname = kmalloc(name->len + 1, GFP_KERNEL);
1499 kmem_cache_free(dentry_cache, dentry);
1503 dname = dentry->d_iname;
1506 dentry->d_name.len = name->len;
1507 dentry->d_name.hash = name->hash;
1508 memcpy(dname, name->name, name->len);
1509 dname[name->len] = 0;
1511 /* Make sure we always see the terminating NUL character */
1513 dentry->d_name.name = dname;
1515 dentry->d_lockref.count = 1;
1516 dentry->d_flags = 0;
1517 spin_lock_init(&dentry->d_lock);
1518 seqcount_init(&dentry->d_seq);
1519 dentry->d_inode = NULL;
1520 dentry->d_parent = dentry;
1522 dentry->d_op = NULL;
1523 dentry->d_fsdata = NULL;
1524 INIT_HLIST_BL_NODE(&dentry->d_hash);
1525 INIT_LIST_HEAD(&dentry->d_lru);
1526 INIT_LIST_HEAD(&dentry->d_subdirs);
1527 INIT_HLIST_NODE(&dentry->d_alias);
1528 INIT_LIST_HEAD(&dentry->d_u.d_child);
1529 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1531 this_cpu_inc(nr_dentry);
1537 * d_alloc - allocate a dcache entry
1538 * @parent: parent of entry to allocate
1539 * @name: qstr of the name
1541 * Allocates a dentry. It returns %NULL if there is insufficient memory
1542 * available. On a success the dentry is returned. The name passed in is
1543 * copied and the copy passed in may be reused after this call.
1545 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1547 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1551 spin_lock(&parent->d_lock);
1553 * don't need child lock because it is not subject
1554 * to concurrency here
1556 __dget_dlock(parent);
1557 dentry->d_parent = parent;
1558 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1559 spin_unlock(&parent->d_lock);
1563 EXPORT_SYMBOL(d_alloc);
1565 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1567 struct dentry *dentry = __d_alloc(sb, name);
1569 dentry->d_flags |= DCACHE_DISCONNECTED;
1572 EXPORT_SYMBOL(d_alloc_pseudo);
1574 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1579 q.len = strlen(name);
1580 q.hash = full_name_hash(q.name, q.len);
1581 return d_alloc(parent, &q);
1583 EXPORT_SYMBOL(d_alloc_name);
1585 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1587 WARN_ON_ONCE(dentry->d_op);
1588 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1590 DCACHE_OP_REVALIDATE |
1591 DCACHE_OP_WEAK_REVALIDATE |
1592 DCACHE_OP_DELETE ));
1597 dentry->d_flags |= DCACHE_OP_HASH;
1599 dentry->d_flags |= DCACHE_OP_COMPARE;
1600 if (op->d_revalidate)
1601 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1602 if (op->d_weak_revalidate)
1603 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1605 dentry->d_flags |= DCACHE_OP_DELETE;
1607 dentry->d_flags |= DCACHE_OP_PRUNE;
1610 EXPORT_SYMBOL(d_set_d_op);
1612 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1614 spin_lock(&dentry->d_lock);
1616 if (unlikely(IS_AUTOMOUNT(inode)))
1617 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1618 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1620 dentry->d_inode = inode;
1621 dentry_rcuwalk_barrier(dentry);
1622 spin_unlock(&dentry->d_lock);
1623 fsnotify_d_instantiate(dentry, inode);
1627 * d_instantiate - fill in inode information for a dentry
1628 * @entry: dentry to complete
1629 * @inode: inode to attach to this dentry
1631 * Fill in inode information in the entry.
1633 * This turns negative dentries into productive full members
1636 * NOTE! This assumes that the inode count has been incremented
1637 * (or otherwise set) by the caller to indicate that it is now
1638 * in use by the dcache.
1641 void d_instantiate(struct dentry *entry, struct inode * inode)
1643 BUG_ON(!hlist_unhashed(&entry->d_alias));
1645 spin_lock(&inode->i_lock);
1646 __d_instantiate(entry, inode);
1648 spin_unlock(&inode->i_lock);
1649 security_d_instantiate(entry, inode);
1651 EXPORT_SYMBOL(d_instantiate);
1654 * d_instantiate_unique - instantiate a non-aliased dentry
1655 * @entry: dentry to instantiate
1656 * @inode: inode to attach to this dentry
1658 * Fill in inode information in the entry. On success, it returns NULL.
1659 * If an unhashed alias of "entry" already exists, then we return the
1660 * aliased dentry instead and drop one reference to inode.
1662 * Note that in order to avoid conflicts with rename() etc, the caller
1663 * had better be holding the parent directory semaphore.
1665 * This also assumes that the inode count has been incremented
1666 * (or otherwise set) by the caller to indicate that it is now
1667 * in use by the dcache.
1669 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1670 struct inode *inode)
1672 struct dentry *alias;
1673 int len = entry->d_name.len;
1674 const char *name = entry->d_name.name;
1675 unsigned int hash = entry->d_name.hash;
1678 __d_instantiate(entry, NULL);
1682 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1684 * Don't need alias->d_lock here, because aliases with
1685 * d_parent == entry->d_parent are not subject to name or
1686 * parent changes, because the parent inode i_mutex is held.
1688 if (alias->d_name.hash != hash)
1690 if (alias->d_parent != entry->d_parent)
1692 if (alias->d_name.len != len)
1694 if (dentry_cmp(alias, name, len))
1700 __d_instantiate(entry, inode);
1704 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1706 struct dentry *result;
1708 BUG_ON(!hlist_unhashed(&entry->d_alias));
1711 spin_lock(&inode->i_lock);
1712 result = __d_instantiate_unique(entry, inode);
1714 spin_unlock(&inode->i_lock);
1717 security_d_instantiate(entry, inode);
1721 BUG_ON(!d_unhashed(result));
1726 EXPORT_SYMBOL(d_instantiate_unique);
1728 struct dentry *d_make_root(struct inode *root_inode)
1730 struct dentry *res = NULL;
1733 static const struct qstr name = QSTR_INIT("/", 1);
1735 res = __d_alloc(root_inode->i_sb, &name);
1737 d_instantiate(res, root_inode);
1743 EXPORT_SYMBOL(d_make_root);
1745 static struct dentry * __d_find_any_alias(struct inode *inode)
1747 struct dentry *alias;
1749 if (hlist_empty(&inode->i_dentry))
1751 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1757 * d_find_any_alias - find any alias for a given inode
1758 * @inode: inode to find an alias for
1760 * If any aliases exist for the given inode, take and return a
1761 * reference for one of them. If no aliases exist, return %NULL.
1763 struct dentry *d_find_any_alias(struct inode *inode)
1767 spin_lock(&inode->i_lock);
1768 de = __d_find_any_alias(inode);
1769 spin_unlock(&inode->i_lock);
1772 EXPORT_SYMBOL(d_find_any_alias);
1775 * d_obtain_alias - find or allocate a dentry for a given inode
1776 * @inode: inode to allocate the dentry for
1778 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1779 * similar open by handle operations. The returned dentry may be anonymous,
1780 * or may have a full name (if the inode was already in the cache).
1782 * When called on a directory inode, we must ensure that the inode only ever
1783 * has one dentry. If a dentry is found, that is returned instead of
1784 * allocating a new one.
1786 * On successful return, the reference to the inode has been transferred
1787 * to the dentry. In case of an error the reference on the inode is released.
1788 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1789 * be passed in and will be the error will be propagate to the return value,
1790 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1792 struct dentry *d_obtain_alias(struct inode *inode)
1794 static const struct qstr anonstring = QSTR_INIT("/", 1);
1799 return ERR_PTR(-ESTALE);
1801 return ERR_CAST(inode);
1803 res = d_find_any_alias(inode);
1807 tmp = __d_alloc(inode->i_sb, &anonstring);
1809 res = ERR_PTR(-ENOMEM);
1813 spin_lock(&inode->i_lock);
1814 res = __d_find_any_alias(inode);
1816 spin_unlock(&inode->i_lock);
1821 /* attach a disconnected dentry */
1822 spin_lock(&tmp->d_lock);
1823 tmp->d_inode = inode;
1824 tmp->d_flags |= DCACHE_DISCONNECTED;
1825 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1826 hlist_bl_lock(&tmp->d_sb->s_anon);
1827 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1828 hlist_bl_unlock(&tmp->d_sb->s_anon);
1829 spin_unlock(&tmp->d_lock);
1830 spin_unlock(&inode->i_lock);
1831 security_d_instantiate(tmp, inode);
1836 if (res && !IS_ERR(res))
1837 security_d_instantiate(res, inode);
1841 EXPORT_SYMBOL(d_obtain_alias);
1844 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1845 * @inode: the inode which may have a disconnected dentry
1846 * @dentry: a negative dentry which we want to point to the inode.
1848 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1849 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1850 * and return it, else simply d_add the inode to the dentry and return NULL.
1852 * This is needed in the lookup routine of any filesystem that is exportable
1853 * (via knfsd) so that we can build dcache paths to directories effectively.
1855 * If a dentry was found and moved, then it is returned. Otherwise NULL
1856 * is returned. This matches the expected return value of ->lookup.
1858 * Cluster filesystems may call this function with a negative, hashed dentry.
1859 * In that case, we know that the inode will be a regular file, and also this
1860 * will only occur during atomic_open. So we need to check for the dentry
1861 * being already hashed only in the final case.
1863 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1865 struct dentry *new = NULL;
1868 return ERR_CAST(inode);
1870 if (inode && S_ISDIR(inode->i_mode)) {
1871 spin_lock(&inode->i_lock);
1872 new = __d_find_alias(inode, 1);
1874 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1875 spin_unlock(&inode->i_lock);
1876 security_d_instantiate(new, inode);
1877 d_move(new, dentry);
1880 /* already taking inode->i_lock, so d_add() by hand */
1881 __d_instantiate(dentry, inode);
1882 spin_unlock(&inode->i_lock);
1883 security_d_instantiate(dentry, inode);
1887 d_instantiate(dentry, inode);
1888 if (d_unhashed(dentry))
1893 EXPORT_SYMBOL(d_splice_alias);
1896 * d_add_ci - lookup or allocate new dentry with case-exact name
1897 * @inode: the inode case-insensitive lookup has found
1898 * @dentry: the negative dentry that was passed to the parent's lookup func
1899 * @name: the case-exact name to be associated with the returned dentry
1901 * This is to avoid filling the dcache with case-insensitive names to the
1902 * same inode, only the actual correct case is stored in the dcache for
1903 * case-insensitive filesystems.
1905 * For a case-insensitive lookup match and if the the case-exact dentry
1906 * already exists in in the dcache, use it and return it.
1908 * If no entry exists with the exact case name, allocate new dentry with
1909 * the exact case, and return the spliced entry.
1911 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1914 struct dentry *found;
1918 * First check if a dentry matching the name already exists,
1919 * if not go ahead and create it now.
1921 found = d_hash_and_lookup(dentry->d_parent, name);
1922 if (unlikely(IS_ERR(found)))
1925 new = d_alloc(dentry->d_parent, name);
1927 found = ERR_PTR(-ENOMEM);
1931 found = d_splice_alias(inode, new);
1940 * If a matching dentry exists, and it's not negative use it.
1942 * Decrement the reference count to balance the iget() done
1945 if (found->d_inode) {
1946 if (unlikely(found->d_inode != inode)) {
1947 /* This can't happen because bad inodes are unhashed. */
1948 BUG_ON(!is_bad_inode(inode));
1949 BUG_ON(!is_bad_inode(found->d_inode));
1956 * Negative dentry: instantiate it unless the inode is a directory and
1957 * already has a dentry.
1959 new = d_splice_alias(inode, found);
1970 EXPORT_SYMBOL(d_add_ci);
1973 * Do the slow-case of the dentry name compare.
1975 * Unlike the dentry_cmp() function, we need to atomically
1976 * load the name and length information, so that the
1977 * filesystem can rely on them, and can use the 'name' and
1978 * 'len' information without worrying about walking off the
1979 * end of memory etc.
1981 * Thus the read_seqcount_retry() and the "duplicate" info
1982 * in arguments (the low-level filesystem should not look
1983 * at the dentry inode or name contents directly, since
1984 * rename can change them while we're in RCU mode).
1986 enum slow_d_compare {
1992 static noinline enum slow_d_compare slow_dentry_cmp(
1993 const struct dentry *parent,
1994 struct dentry *dentry,
1996 const struct qstr *name)
1998 int tlen = dentry->d_name.len;
1999 const char *tname = dentry->d_name.name;
2001 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2003 return D_COMP_SEQRETRY;
2005 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2006 return D_COMP_NOMATCH;
2011 * __d_lookup_rcu - search for a dentry (racy, store-free)
2012 * @parent: parent dentry
2013 * @name: qstr of name we wish to find
2014 * @seqp: returns d_seq value at the point where the dentry was found
2015 * Returns: dentry, or NULL
2017 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2018 * resolution (store-free path walking) design described in
2019 * Documentation/filesystems/path-lookup.txt.
2021 * This is not to be used outside core vfs.
2023 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2024 * held, and rcu_read_lock held. The returned dentry must not be stored into
2025 * without taking d_lock and checking d_seq sequence count against @seq
2028 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2031 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2032 * the returned dentry, so long as its parent's seqlock is checked after the
2033 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2034 * is formed, giving integrity down the path walk.
2036 * NOTE! The caller *has* to check the resulting dentry against the sequence
2037 * number we've returned before using any of the resulting dentry state!
2039 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2040 const struct qstr *name,
2043 u64 hashlen = name->hash_len;
2044 const unsigned char *str = name->name;
2045 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2046 struct hlist_bl_node *node;
2047 struct dentry *dentry;
2050 * Note: There is significant duplication with __d_lookup_rcu which is
2051 * required to prevent single threaded performance regressions
2052 * especially on architectures where smp_rmb (in seqcounts) are costly.
2053 * Keep the two functions in sync.
2057 * The hash list is protected using RCU.
2059 * Carefully use d_seq when comparing a candidate dentry, to avoid
2060 * races with d_move().
2062 * It is possible that concurrent renames can mess up our list
2063 * walk here and result in missing our dentry, resulting in the
2064 * false-negative result. d_lookup() protects against concurrent
2065 * renames using rename_lock seqlock.
2067 * See Documentation/filesystems/path-lookup.txt for more details.
2069 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2074 * The dentry sequence count protects us from concurrent
2075 * renames, and thus protects parent and name fields.
2077 * The caller must perform a seqcount check in order
2078 * to do anything useful with the returned dentry.
2080 * NOTE! We do a "raw" seqcount_begin here. That means that
2081 * we don't wait for the sequence count to stabilize if it
2082 * is in the middle of a sequence change. If we do the slow
2083 * dentry compare, we will do seqretries until it is stable,
2084 * and if we end up with a successful lookup, we actually
2085 * want to exit RCU lookup anyway.
2087 seq = raw_seqcount_begin(&dentry->d_seq);
2088 if (dentry->d_parent != parent)
2090 if (d_unhashed(dentry))
2093 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2094 if (dentry->d_name.hash != hashlen_hash(hashlen))
2097 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2100 case D_COMP_NOMATCH:
2107 if (dentry->d_name.hash_len != hashlen)
2110 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2117 * d_lookup - search for a dentry
2118 * @parent: parent dentry
2119 * @name: qstr of name we wish to find
2120 * Returns: dentry, or NULL
2122 * d_lookup searches the children of the parent dentry for the name in
2123 * question. If the dentry is found its reference count is incremented and the
2124 * dentry is returned. The caller must use dput to free the entry when it has
2125 * finished using it. %NULL is returned if the dentry does not exist.
2127 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2129 struct dentry *dentry;
2133 seq = read_seqbegin(&rename_lock);
2134 dentry = __d_lookup(parent, name);
2137 } while (read_seqretry(&rename_lock, seq));
2140 EXPORT_SYMBOL(d_lookup);
2143 * __d_lookup - search for a dentry (racy)
2144 * @parent: parent dentry
2145 * @name: qstr of name we wish to find
2146 * Returns: dentry, or NULL
2148 * __d_lookup is like d_lookup, however it may (rarely) return a
2149 * false-negative result due to unrelated rename activity.
2151 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2152 * however it must be used carefully, eg. with a following d_lookup in
2153 * the case of failure.
2155 * __d_lookup callers must be commented.
2157 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2159 unsigned int len = name->len;
2160 unsigned int hash = name->hash;
2161 const unsigned char *str = name->name;
2162 struct hlist_bl_head *b = d_hash(parent, hash);
2163 struct hlist_bl_node *node;
2164 struct dentry *found = NULL;
2165 struct dentry *dentry;
2168 * Note: There is significant duplication with __d_lookup_rcu which is
2169 * required to prevent single threaded performance regressions
2170 * especially on architectures where smp_rmb (in seqcounts) are costly.
2171 * Keep the two functions in sync.
2175 * The hash list is protected using RCU.
2177 * Take d_lock when comparing a candidate dentry, to avoid races
2180 * It is possible that concurrent renames can mess up our list
2181 * walk here and result in missing our dentry, resulting in the
2182 * false-negative result. d_lookup() protects against concurrent
2183 * renames using rename_lock seqlock.
2185 * See Documentation/filesystems/path-lookup.txt for more details.
2189 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2191 if (dentry->d_name.hash != hash)
2194 spin_lock(&dentry->d_lock);
2195 if (dentry->d_parent != parent)
2197 if (d_unhashed(dentry))
2201 * It is safe to compare names since d_move() cannot
2202 * change the qstr (protected by d_lock).
2204 if (parent->d_flags & DCACHE_OP_COMPARE) {
2205 int tlen = dentry->d_name.len;
2206 const char *tname = dentry->d_name.name;
2207 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2210 if (dentry->d_name.len != len)
2212 if (dentry_cmp(dentry, str, len))
2216 dentry->d_lockref.count++;
2218 spin_unlock(&dentry->d_lock);
2221 spin_unlock(&dentry->d_lock);
2229 * d_hash_and_lookup - hash the qstr then search for a dentry
2230 * @dir: Directory to search in
2231 * @name: qstr of name we wish to find
2233 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2235 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2238 * Check for a fs-specific hash function. Note that we must
2239 * calculate the standard hash first, as the d_op->d_hash()
2240 * routine may choose to leave the hash value unchanged.
2242 name->hash = full_name_hash(name->name, name->len);
2243 if (dir->d_flags & DCACHE_OP_HASH) {
2244 int err = dir->d_op->d_hash(dir, name);
2245 if (unlikely(err < 0))
2246 return ERR_PTR(err);
2248 return d_lookup(dir, name);
2250 EXPORT_SYMBOL(d_hash_and_lookup);
2253 * d_validate - verify dentry provided from insecure source (deprecated)
2254 * @dentry: The dentry alleged to be valid child of @dparent
2255 * @dparent: The parent dentry (known to be valid)
2257 * An insecure source has sent us a dentry, here we verify it and dget() it.
2258 * This is used by ncpfs in its readdir implementation.
2259 * Zero is returned in the dentry is invalid.
2261 * This function is slow for big directories, and deprecated, do not use it.
2263 int d_validate(struct dentry *dentry, struct dentry *dparent)
2265 struct dentry *child;
2267 spin_lock(&dparent->d_lock);
2268 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2269 if (dentry == child) {
2270 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2271 __dget_dlock(dentry);
2272 spin_unlock(&dentry->d_lock);
2273 spin_unlock(&dparent->d_lock);
2277 spin_unlock(&dparent->d_lock);
2281 EXPORT_SYMBOL(d_validate);
2284 * When a file is deleted, we have two options:
2285 * - turn this dentry into a negative dentry
2286 * - unhash this dentry and free it.
2288 * Usually, we want to just turn this into
2289 * a negative dentry, but if anybody else is
2290 * currently using the dentry or the inode
2291 * we can't do that and we fall back on removing
2292 * it from the hash queues and waiting for
2293 * it to be deleted later when it has no users
2297 * d_delete - delete a dentry
2298 * @dentry: The dentry to delete
2300 * Turn the dentry into a negative dentry if possible, otherwise
2301 * remove it from the hash queues so it can be deleted later
2304 void d_delete(struct dentry * dentry)
2306 struct inode *inode;
2309 * Are we the only user?
2312 spin_lock(&dentry->d_lock);
2313 inode = dentry->d_inode;
2314 isdir = S_ISDIR(inode->i_mode);
2315 if (dentry->d_lockref.count == 1) {
2316 if (!spin_trylock(&inode->i_lock)) {
2317 spin_unlock(&dentry->d_lock);
2321 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2322 dentry_unlink_inode(dentry);
2323 fsnotify_nameremove(dentry, isdir);
2327 if (!d_unhashed(dentry))
2330 spin_unlock(&dentry->d_lock);
2332 fsnotify_nameremove(dentry, isdir);
2334 EXPORT_SYMBOL(d_delete);
2336 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2338 BUG_ON(!d_unhashed(entry));
2340 entry->d_flags |= DCACHE_RCUACCESS;
2341 hlist_bl_add_head_rcu(&entry->d_hash, b);
2345 static void _d_rehash(struct dentry * entry)
2347 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2351 * d_rehash - add an entry back to the hash
2352 * @entry: dentry to add to the hash
2354 * Adds a dentry to the hash according to its name.
2357 void d_rehash(struct dentry * entry)
2359 spin_lock(&entry->d_lock);
2361 spin_unlock(&entry->d_lock);
2363 EXPORT_SYMBOL(d_rehash);
2366 * dentry_update_name_case - update case insensitive dentry with a new name
2367 * @dentry: dentry to be updated
2370 * Update a case insensitive dentry with new case of name.
2372 * dentry must have been returned by d_lookup with name @name. Old and new
2373 * name lengths must match (ie. no d_compare which allows mismatched name
2376 * Parent inode i_mutex must be held over d_lookup and into this call (to
2377 * keep renames and concurrent inserts, and readdir(2) away).
2379 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2381 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2382 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2384 spin_lock(&dentry->d_lock);
2385 write_seqcount_begin(&dentry->d_seq);
2386 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2387 write_seqcount_end(&dentry->d_seq);
2388 spin_unlock(&dentry->d_lock);
2390 EXPORT_SYMBOL(dentry_update_name_case);
2392 static void switch_names(struct dentry *dentry, struct dentry *target)
2394 if (dname_external(target)) {
2395 if (dname_external(dentry)) {
2397 * Both external: swap the pointers
2399 swap(target->d_name.name, dentry->d_name.name);
2402 * dentry:internal, target:external. Steal target's
2403 * storage and make target internal.
2405 memcpy(target->d_iname, dentry->d_name.name,
2406 dentry->d_name.len + 1);
2407 dentry->d_name.name = target->d_name.name;
2408 target->d_name.name = target->d_iname;
2411 if (dname_external(dentry)) {
2413 * dentry:external, target:internal. Give dentry's
2414 * storage to target and make dentry internal
2416 memcpy(dentry->d_iname, target->d_name.name,
2417 target->d_name.len + 1);
2418 target->d_name.name = dentry->d_name.name;
2419 dentry->d_name.name = dentry->d_iname;
2422 * Both are internal. Just copy target to dentry
2424 memcpy(dentry->d_iname, target->d_name.name,
2425 target->d_name.len + 1);
2426 dentry->d_name.len = target->d_name.len;
2430 swap(dentry->d_name.len, target->d_name.len);
2433 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2436 * XXXX: do we really need to take target->d_lock?
2438 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2439 spin_lock(&target->d_parent->d_lock);
2441 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2442 spin_lock(&dentry->d_parent->d_lock);
2443 spin_lock_nested(&target->d_parent->d_lock,
2444 DENTRY_D_LOCK_NESTED);
2446 spin_lock(&target->d_parent->d_lock);
2447 spin_lock_nested(&dentry->d_parent->d_lock,
2448 DENTRY_D_LOCK_NESTED);
2451 if (target < dentry) {
2452 spin_lock_nested(&target->d_lock, 2);
2453 spin_lock_nested(&dentry->d_lock, 3);
2455 spin_lock_nested(&dentry->d_lock, 2);
2456 spin_lock_nested(&target->d_lock, 3);
2460 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2461 struct dentry *target)
2463 if (target->d_parent != dentry->d_parent)
2464 spin_unlock(&dentry->d_parent->d_lock);
2465 if (target->d_parent != target)
2466 spin_unlock(&target->d_parent->d_lock);
2470 * When switching names, the actual string doesn't strictly have to
2471 * be preserved in the target - because we're dropping the target
2472 * anyway. As such, we can just do a simple memcpy() to copy over
2473 * the new name before we switch.
2475 * Note that we have to be a lot more careful about getting the hash
2476 * switched - we have to switch the hash value properly even if it
2477 * then no longer matches the actual (corrupted) string of the target.
2478 * The hash value has to match the hash queue that the dentry is on..
2481 * __d_move - move a dentry
2482 * @dentry: entry to move
2483 * @target: new dentry
2485 * Update the dcache to reflect the move of a file name. Negative
2486 * dcache entries should not be moved in this way. Caller must hold
2487 * rename_lock, the i_mutex of the source and target directories,
2488 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2490 static void __d_move(struct dentry * dentry, struct dentry * target)
2492 if (!dentry->d_inode)
2493 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2495 BUG_ON(d_ancestor(dentry, target));
2496 BUG_ON(d_ancestor(target, dentry));
2498 dentry_lock_for_move(dentry, target);
2500 write_seqcount_begin(&dentry->d_seq);
2501 write_seqcount_begin(&target->d_seq);
2503 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2506 * Move the dentry to the target hash queue. Don't bother checking
2507 * for the same hash queue because of how unlikely it is.
2510 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2512 /* Unhash the target: dput() will then get rid of it */
2515 list_del(&dentry->d_u.d_child);
2516 list_del(&target->d_u.d_child);
2518 /* Switch the names.. */
2519 switch_names(dentry, target);
2520 swap(dentry->d_name.hash, target->d_name.hash);
2522 /* ... and switch the parents */
2523 if (IS_ROOT(dentry)) {
2524 dentry->d_parent = target->d_parent;
2525 target->d_parent = target;
2526 INIT_LIST_HEAD(&target->d_u.d_child);
2528 swap(dentry->d_parent, target->d_parent);
2530 /* And add them back to the (new) parent lists */
2531 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2534 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2536 write_seqcount_end(&target->d_seq);
2537 write_seqcount_end(&dentry->d_seq);
2539 dentry_unlock_parents_for_move(dentry, target);
2540 spin_unlock(&target->d_lock);
2541 fsnotify_d_move(dentry);
2542 spin_unlock(&dentry->d_lock);
2546 * d_move - move a dentry
2547 * @dentry: entry to move
2548 * @target: new dentry
2550 * Update the dcache to reflect the move of a file name. Negative
2551 * dcache entries should not be moved in this way. See the locking
2552 * requirements for __d_move.
2554 void d_move(struct dentry *dentry, struct dentry *target)
2556 write_seqlock(&rename_lock);
2557 __d_move(dentry, target);
2558 write_sequnlock(&rename_lock);
2560 EXPORT_SYMBOL(d_move);
2563 * d_ancestor - search for an ancestor
2564 * @p1: ancestor dentry
2567 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2568 * an ancestor of p2, else NULL.
2570 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2574 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2575 if (p->d_parent == p1)
2582 * This helper attempts to cope with remotely renamed directories
2584 * It assumes that the caller is already holding
2585 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2587 * Note: If ever the locking in lock_rename() changes, then please
2588 * remember to update this too...
2590 static struct dentry *__d_unalias(struct inode *inode,
2591 struct dentry *dentry, struct dentry *alias)
2593 struct mutex *m1 = NULL, *m2 = NULL;
2594 struct dentry *ret = ERR_PTR(-EBUSY);
2596 /* If alias and dentry share a parent, then no extra locks required */
2597 if (alias->d_parent == dentry->d_parent)
2600 /* See lock_rename() */
2601 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2603 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2604 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2606 m2 = &alias->d_parent->d_inode->i_mutex;
2608 if (likely(!d_mountpoint(alias))) {
2609 __d_move(alias, dentry);
2613 spin_unlock(&inode->i_lock);
2622 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2623 * named dentry in place of the dentry to be replaced.
2624 * returns with anon->d_lock held!
2626 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2628 struct dentry *dparent;
2630 dentry_lock_for_move(anon, dentry);
2632 write_seqcount_begin(&dentry->d_seq);
2633 write_seqcount_begin(&anon->d_seq);
2635 dparent = dentry->d_parent;
2637 switch_names(dentry, anon);
2638 swap(dentry->d_name.hash, anon->d_name.hash);
2640 dentry->d_parent = dentry;
2641 list_del_init(&dentry->d_u.d_child);
2642 anon->d_parent = dparent;
2643 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2645 write_seqcount_end(&dentry->d_seq);
2646 write_seqcount_end(&anon->d_seq);
2648 dentry_unlock_parents_for_move(anon, dentry);
2649 spin_unlock(&dentry->d_lock);
2651 /* anon->d_lock still locked, returns locked */
2652 anon->d_flags &= ~DCACHE_DISCONNECTED;
2656 * d_materialise_unique - introduce an inode into the tree
2657 * @dentry: candidate dentry
2658 * @inode: inode to bind to the dentry, to which aliases may be attached
2660 * Introduces an dentry into the tree, substituting an extant disconnected
2661 * root directory alias in its place if there is one. Caller must hold the
2662 * i_mutex of the parent directory.
2664 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2666 struct dentry *actual;
2668 BUG_ON(!d_unhashed(dentry));
2672 __d_instantiate(dentry, NULL);
2677 spin_lock(&inode->i_lock);
2679 if (S_ISDIR(inode->i_mode)) {
2680 struct dentry *alias;
2682 /* Does an aliased dentry already exist? */
2683 alias = __d_find_alias(inode, 0);
2686 write_seqlock(&rename_lock);
2688 if (d_ancestor(alias, dentry)) {
2689 /* Check for loops */
2690 actual = ERR_PTR(-ELOOP);
2691 spin_unlock(&inode->i_lock);
2692 } else if (IS_ROOT(alias)) {
2693 /* Is this an anonymous mountpoint that we
2694 * could splice into our tree? */
2695 __d_materialise_dentry(dentry, alias);
2696 write_sequnlock(&rename_lock);
2700 /* Nope, but we must(!) avoid directory
2701 * aliasing. This drops inode->i_lock */
2702 actual = __d_unalias(inode, dentry, alias);
2704 write_sequnlock(&rename_lock);
2705 if (IS_ERR(actual)) {
2706 if (PTR_ERR(actual) == -ELOOP)
2707 pr_warn_ratelimited(
2708 "VFS: Lookup of '%s' in %s %s"
2709 " would have caused loop\n",
2710 dentry->d_name.name,
2711 inode->i_sb->s_type->name,
2719 /* Add a unique reference */
2720 actual = __d_instantiate_unique(dentry, inode);
2724 BUG_ON(!d_unhashed(actual));
2726 spin_lock(&actual->d_lock);
2729 spin_unlock(&actual->d_lock);
2730 spin_unlock(&inode->i_lock);
2732 if (actual == dentry) {
2733 security_d_instantiate(dentry, inode);
2740 EXPORT_SYMBOL_GPL(d_materialise_unique);
2742 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2746 return -ENAMETOOLONG;
2748 memcpy(*buffer, str, namelen);
2753 * prepend_name - prepend a pathname in front of current buffer pointer
2754 * buffer: buffer pointer
2755 * buflen: allocated length of the buffer
2756 * name: name string and length qstr structure
2758 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2759 * make sure that either the old or the new name pointer and length are
2760 * fetched. However, there may be mismatch between length and pointer.
2761 * The length cannot be trusted, we need to copy it byte-by-byte until
2762 * the length is reached or a null byte is found. It also prepends "/" at
2763 * the beginning of the name. The sequence number check at the caller will
2764 * retry it again when a d_move() does happen. So any garbage in the buffer
2765 * due to mismatched pointer and length will be discarded.
2767 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2769 const char *dname = ACCESS_ONCE(name->name);
2770 u32 dlen = ACCESS_ONCE(name->len);
2773 if (*buflen < dlen + 1)
2774 return -ENAMETOOLONG;
2775 *buflen -= dlen + 1;
2776 p = *buffer -= dlen + 1;
2788 * prepend_path - Prepend path string to a buffer
2789 * @path: the dentry/vfsmount to report
2790 * @root: root vfsmnt/dentry
2791 * @buffer: pointer to the end of the buffer
2792 * @buflen: pointer to buffer length
2794 * The function tries to write out the pathname without taking any lock other
2795 * than the RCU read lock to make sure that dentries won't go away. It only
2796 * checks the sequence number of the global rename_lock as any change in the
2797 * dentry's d_seq will be preceded by changes in the rename_lock sequence
2798 * number. If the sequence number had been change, it will restart the whole
2799 * pathname back-tracing sequence again. It performs a total of 3 trials of
2800 * lockless back-tracing sequences before falling back to take the
2803 static int prepend_path(const struct path *path,
2804 const struct path *root,
2805 char **buffer, int *buflen)
2807 struct dentry *dentry = path->dentry;
2808 struct vfsmount *vfsmnt = path->mnt;
2809 struct mount *mnt = real_mount(vfsmnt);
2819 read_seqbegin_or_lock(&rename_lock, &seq);
2820 while (dentry != root->dentry || vfsmnt != root->mnt) {
2821 struct dentry * parent;
2823 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2825 if (mnt_has_parent(mnt)) {
2826 dentry = mnt->mnt_mountpoint;
2827 mnt = mnt->mnt_parent;
2832 * Filesystems needing to implement special "root names"
2833 * should do so with ->d_dname()
2835 if (IS_ROOT(dentry) &&
2836 (dentry->d_name.len != 1 ||
2837 dentry->d_name.name[0] != '/')) {
2838 WARN(1, "Root dentry has weird name <%.*s>\n",
2839 (int) dentry->d_name.len,
2840 dentry->d_name.name);
2843 error = is_mounted(vfsmnt) ? 1 : 2;
2846 parent = dentry->d_parent;
2848 error = prepend_name(&bptr, &blen, &dentry->d_name);
2856 if (need_seqretry(&rename_lock, seq)) {
2860 done_seqretry(&rename_lock, seq);
2862 if (error >= 0 && bptr == *buffer) {
2864 error = -ENAMETOOLONG;
2874 * __d_path - return the path of a dentry
2875 * @path: the dentry/vfsmount to report
2876 * @root: root vfsmnt/dentry
2877 * @buf: buffer to return value in
2878 * @buflen: buffer length
2880 * Convert a dentry into an ASCII path name.
2882 * Returns a pointer into the buffer or an error code if the
2883 * path was too long.
2885 * "buflen" should be positive.
2887 * If the path is not reachable from the supplied root, return %NULL.
2889 char *__d_path(const struct path *path,
2890 const struct path *root,
2891 char *buf, int buflen)
2893 char *res = buf + buflen;
2896 prepend(&res, &buflen, "\0", 1);
2897 br_read_lock(&vfsmount_lock);
2898 error = prepend_path(path, root, &res, &buflen);
2899 br_read_unlock(&vfsmount_lock);
2902 return ERR_PTR(error);
2908 char *d_absolute_path(const struct path *path,
2909 char *buf, int buflen)
2911 struct path root = {};
2912 char *res = buf + buflen;
2915 prepend(&res, &buflen, "\0", 1);
2916 br_read_lock(&vfsmount_lock);
2917 error = prepend_path(path, &root, &res, &buflen);
2918 br_read_unlock(&vfsmount_lock);
2923 return ERR_PTR(error);
2928 * same as __d_path but appends "(deleted)" for unlinked files.
2930 static int path_with_deleted(const struct path *path,
2931 const struct path *root,
2932 char **buf, int *buflen)
2934 prepend(buf, buflen, "\0", 1);
2935 if (d_unlinked(path->dentry)) {
2936 int error = prepend(buf, buflen, " (deleted)", 10);
2941 return prepend_path(path, root, buf, buflen);
2944 static int prepend_unreachable(char **buffer, int *buflen)
2946 return prepend(buffer, buflen, "(unreachable)", 13);
2950 * d_path - return the path of a dentry
2951 * @path: path to report
2952 * @buf: buffer to return value in
2953 * @buflen: buffer length
2955 * Convert a dentry into an ASCII path name. If the entry has been deleted
2956 * the string " (deleted)" is appended. Note that this is ambiguous.
2958 * Returns a pointer into the buffer or an error code if the path was
2959 * too long. Note: Callers should use the returned pointer, not the passed
2960 * in buffer, to use the name! The implementation often starts at an offset
2961 * into the buffer, and may leave 0 bytes at the start.
2963 * "buflen" should be positive.
2965 char *d_path(const struct path *path, char *buf, int buflen)
2967 char *res = buf + buflen;
2972 * We have various synthetic filesystems that never get mounted. On
2973 * these filesystems dentries are never used for lookup purposes, and
2974 * thus don't need to be hashed. They also don't need a name until a
2975 * user wants to identify the object in /proc/pid/fd/. The little hack
2976 * below allows us to generate a name for these objects on demand:
2978 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2979 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2981 get_fs_root(current->fs, &root);
2982 br_read_lock(&vfsmount_lock);
2983 error = path_with_deleted(path, &root, &res, &buflen);
2984 br_read_unlock(&vfsmount_lock);
2986 res = ERR_PTR(error);
2990 EXPORT_SYMBOL(d_path);
2993 * Helper function for dentry_operations.d_dname() members
2995 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2996 const char *fmt, ...)
3002 va_start(args, fmt);
3003 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3006 if (sz > sizeof(temp) || sz > buflen)
3007 return ERR_PTR(-ENAMETOOLONG);
3009 buffer += buflen - sz;
3010 return memcpy(buffer, temp, sz);
3013 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3015 char *end = buffer + buflen;
3016 /* these dentries are never renamed, so d_lock is not needed */
3017 if (prepend(&end, &buflen, " (deleted)", 11) ||
3018 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3019 prepend(&end, &buflen, "/", 1))
3020 end = ERR_PTR(-ENAMETOOLONG);
3025 * Write full pathname from the root of the filesystem into the buffer.
3027 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
3037 prepend(&end, &len, "\0", 1);
3043 read_seqbegin_or_lock(&rename_lock, &seq);
3044 while (!IS_ROOT(dentry)) {
3045 struct dentry *parent = dentry->d_parent;
3049 error = prepend_name(&end, &len, &dentry->d_name);
3058 if (need_seqretry(&rename_lock, seq)) {
3062 done_seqretry(&rename_lock, seq);
3067 return ERR_PTR(-ENAMETOOLONG);
3070 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3072 return __dentry_path(dentry, buf, buflen);
3074 EXPORT_SYMBOL(dentry_path_raw);
3076 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3081 if (d_unlinked(dentry)) {
3083 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3087 retval = __dentry_path(dentry, buf, buflen);
3088 if (!IS_ERR(retval) && p)
3089 *p = '/'; /* restore '/' overriden with '\0' */
3092 return ERR_PTR(-ENAMETOOLONG);
3096 * NOTE! The user-level library version returns a
3097 * character pointer. The kernel system call just
3098 * returns the length of the buffer filled (which
3099 * includes the ending '\0' character), or a negative
3100 * error value. So libc would do something like
3102 * char *getcwd(char * buf, size_t size)
3106 * retval = sys_getcwd(buf, size);
3113 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3116 struct path pwd, root;
3117 char *page = (char *) __get_free_page(GFP_USER);
3122 get_fs_root_and_pwd(current->fs, &root, &pwd);
3125 br_read_lock(&vfsmount_lock);
3126 if (!d_unlinked(pwd.dentry)) {
3128 char *cwd = page + PAGE_SIZE;
3129 int buflen = PAGE_SIZE;
3131 prepend(&cwd, &buflen, "\0", 1);
3132 error = prepend_path(&pwd, &root, &cwd, &buflen);
3133 br_read_unlock(&vfsmount_lock);
3138 /* Unreachable from current root */
3140 error = prepend_unreachable(&cwd, &buflen);
3146 len = PAGE_SIZE + page - cwd;
3149 if (copy_to_user(buf, cwd, len))
3153 br_read_unlock(&vfsmount_lock);
3159 free_page((unsigned long) page);
3164 * Test whether new_dentry is a subdirectory of old_dentry.
3166 * Trivially implemented using the dcache structure
3170 * is_subdir - is new dentry a subdirectory of old_dentry
3171 * @new_dentry: new dentry
3172 * @old_dentry: old dentry
3174 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3175 * Returns 0 otherwise.
3176 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3179 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3184 if (new_dentry == old_dentry)
3188 /* for restarting inner loop in case of seq retry */
3189 seq = read_seqbegin(&rename_lock);
3191 * Need rcu_readlock to protect against the d_parent trashing
3195 if (d_ancestor(old_dentry, new_dentry))
3200 } while (read_seqretry(&rename_lock, seq));
3205 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3207 struct dentry *root = data;
3208 if (dentry != root) {
3209 if (d_unhashed(dentry) || !dentry->d_inode)
3212 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3213 dentry->d_flags |= DCACHE_GENOCIDE;
3214 dentry->d_lockref.count--;
3217 return D_WALK_CONTINUE;
3220 void d_genocide(struct dentry *parent)
3222 d_walk(parent, parent, d_genocide_kill, NULL);
3225 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3227 inode_dec_link_count(inode);
3228 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3229 !hlist_unhashed(&dentry->d_alias) ||
3230 !d_unlinked(dentry));
3231 spin_lock(&dentry->d_parent->d_lock);
3232 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3233 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3234 (unsigned long long)inode->i_ino);
3235 spin_unlock(&dentry->d_lock);
3236 spin_unlock(&dentry->d_parent->d_lock);
3237 d_instantiate(dentry, inode);
3239 EXPORT_SYMBOL(d_tmpfile);
3241 static __initdata unsigned long dhash_entries;
3242 static int __init set_dhash_entries(char *str)
3246 dhash_entries = simple_strtoul(str, &str, 0);
3249 __setup("dhash_entries=", set_dhash_entries);
3251 static void __init dcache_init_early(void)
3255 /* If hashes are distributed across NUMA nodes, defer
3256 * hash allocation until vmalloc space is available.
3262 alloc_large_system_hash("Dentry cache",
3263 sizeof(struct hlist_bl_head),
3272 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3273 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3276 static void __init dcache_init(void)
3281 * A constructor could be added for stable state like the lists,
3282 * but it is probably not worth it because of the cache nature
3285 dentry_cache = KMEM_CACHE(dentry,
3286 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3288 /* Hash may have been set up in dcache_init_early */
3293 alloc_large_system_hash("Dentry cache",
3294 sizeof(struct hlist_bl_head),
3303 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3304 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3307 /* SLAB cache for __getname() consumers */
3308 struct kmem_cache *names_cachep __read_mostly;
3309 EXPORT_SYMBOL(names_cachep);
3311 EXPORT_SYMBOL(d_genocide);
3313 void __init vfs_caches_init_early(void)
3315 dcache_init_early();
3319 void __init vfs_caches_init(unsigned long mempages)
3321 unsigned long reserve;
3323 /* Base hash sizes on available memory, with a reserve equal to
3324 150% of current kernel size */
3326 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3327 mempages -= reserve;
3329 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3330 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3334 files_init(mempages);