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 * This is the single most critical data structure when it comes
93 * to the dcache: the hashtable for lookups. Somebody should try
94 * to make this good - I've just made it work.
96 * This hash-function tries to avoid losing too many bits of hash
97 * information, yet avoid using a prime hash-size or similar.
100 static unsigned int d_hash_mask __read_mostly;
101 static unsigned int d_hash_shift __read_mostly;
103 static struct hlist_bl_head *dentry_hashtable __read_mostly;
105 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
108 hash += (unsigned long) parent / L1_CACHE_BYTES;
109 hash = hash + (hash >> d_hash_shift);
110 return dentry_hashtable + (hash & d_hash_mask);
113 /* Statistics gathering. */
114 struct dentry_stat_t dentry_stat = {
118 static DEFINE_PER_CPU(long, nr_dentry);
119 static DEFINE_PER_CPU(long, nr_dentry_unused);
121 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
124 * Here we resort to our own counters instead of using generic per-cpu counters
125 * for consistency with what the vfs inode code does. We are expected to harvest
126 * better code and performance by having our own specialized counters.
128 * Please note that the loop is done over all possible CPUs, not over all online
129 * CPUs. The reason for this is that we don't want to play games with CPUs going
130 * on and off. If one of them goes off, we will just keep their counters.
132 * glommer: See cffbc8a for details, and if you ever intend to change this,
133 * please update all vfs counters to match.
135 static long get_nr_dentry(void)
139 for_each_possible_cpu(i)
140 sum += per_cpu(nr_dentry, i);
141 return sum < 0 ? 0 : sum;
144 static long get_nr_dentry_unused(void)
148 for_each_possible_cpu(i)
149 sum += per_cpu(nr_dentry_unused, i);
150 return sum < 0 ? 0 : sum;
153 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
154 size_t *lenp, loff_t *ppos)
156 dentry_stat.nr_dentry = get_nr_dentry();
157 dentry_stat.nr_unused = get_nr_dentry_unused();
158 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
163 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
164 * The strings are both count bytes long, and count is non-zero.
166 #ifdef CONFIG_DCACHE_WORD_ACCESS
168 #include <asm/word-at-a-time.h>
170 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
171 * aligned allocation for this particular component. We don't
172 * strictly need the load_unaligned_zeropad() safety, but it
173 * doesn't hurt either.
175 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
176 * need the careful unaligned handling.
178 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
180 unsigned long a,b,mask;
183 a = *(unsigned long *)cs;
184 b = load_unaligned_zeropad(ct);
185 if (tcount < sizeof(unsigned long))
187 if (unlikely(a != b))
189 cs += sizeof(unsigned long);
190 ct += sizeof(unsigned long);
191 tcount -= sizeof(unsigned long);
195 mask = bytemask_from_count(tcount);
196 return unlikely(!!((a ^ b) & mask));
201 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
215 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
217 const unsigned char *cs;
219 * Be careful about RCU walk racing with rename:
220 * use ACCESS_ONCE to fetch the name pointer.
222 * NOTE! Even if a rename will mean that the length
223 * was not loaded atomically, we don't care. The
224 * RCU walk will check the sequence count eventually,
225 * and catch it. And we won't overrun the buffer,
226 * because we're reading the name pointer atomically,
227 * and a dentry name is guaranteed to be properly
228 * terminated with a NUL byte.
230 * End result: even if 'len' is wrong, we'll exit
231 * early because the data cannot match (there can
232 * be no NUL in the ct/tcount data)
234 cs = ACCESS_ONCE(dentry->d_name.name);
235 smp_read_barrier_depends();
236 return dentry_string_cmp(cs, ct, tcount);
239 static void __d_free(struct rcu_head *head)
241 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
243 WARN_ON(!hlist_unhashed(&dentry->d_alias));
244 if (dname_external(dentry))
245 kfree(dentry->d_name.name);
246 kmem_cache_free(dentry_cache, dentry);
249 static void dentry_free(struct dentry *dentry)
251 /* if dentry was never visible to RCU, immediate free is OK */
252 if (!(dentry->d_flags & DCACHE_RCUACCESS))
253 __d_free(&dentry->d_u.d_rcu);
255 call_rcu(&dentry->d_u.d_rcu, __d_free);
259 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
260 * @dentry: the target dentry
261 * After this call, in-progress rcu-walk path lookup will fail. This
262 * should be called after unhashing, and after changing d_inode (if
263 * the dentry has not already been unhashed).
265 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
267 assert_spin_locked(&dentry->d_lock);
268 /* Go through a barrier */
269 write_seqcount_barrier(&dentry->d_seq);
273 * Release the dentry's inode, using the filesystem
274 * d_iput() operation if defined. Dentry has no refcount
277 static void dentry_iput(struct dentry * dentry)
278 __releases(dentry->d_lock)
279 __releases(dentry->d_inode->i_lock)
281 struct inode *inode = dentry->d_inode;
283 dentry->d_inode = NULL;
284 hlist_del_init(&dentry->d_alias);
285 spin_unlock(&dentry->d_lock);
286 spin_unlock(&inode->i_lock);
288 fsnotify_inoderemove(inode);
289 if (dentry->d_op && dentry->d_op->d_iput)
290 dentry->d_op->d_iput(dentry, inode);
294 spin_unlock(&dentry->d_lock);
299 * Release the dentry's inode, using the filesystem
300 * d_iput() operation if defined. dentry remains in-use.
302 static void dentry_unlink_inode(struct dentry * dentry)
303 __releases(dentry->d_lock)
304 __releases(dentry->d_inode->i_lock)
306 struct inode *inode = dentry->d_inode;
307 __d_clear_type(dentry);
308 dentry->d_inode = NULL;
309 hlist_del_init(&dentry->d_alias);
310 dentry_rcuwalk_barrier(dentry);
311 spin_unlock(&dentry->d_lock);
312 spin_unlock(&inode->i_lock);
314 fsnotify_inoderemove(inode);
315 if (dentry->d_op && dentry->d_op->d_iput)
316 dentry->d_op->d_iput(dentry, inode);
322 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
323 * is in use - which includes both the "real" per-superblock
324 * LRU list _and_ the DCACHE_SHRINK_LIST use.
326 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
327 * on the shrink list (ie not on the superblock LRU list).
329 * The per-cpu "nr_dentry_unused" counters are updated with
330 * the DCACHE_LRU_LIST bit.
332 * These helper functions make sure we always follow the
333 * rules. d_lock must be held by the caller.
335 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
336 static void d_lru_add(struct dentry *dentry)
338 D_FLAG_VERIFY(dentry, 0);
339 dentry->d_flags |= DCACHE_LRU_LIST;
340 this_cpu_inc(nr_dentry_unused);
341 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
344 static void d_lru_del(struct dentry *dentry)
346 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
347 dentry->d_flags &= ~DCACHE_LRU_LIST;
348 this_cpu_dec(nr_dentry_unused);
349 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
352 static void d_shrink_del(struct dentry *dentry)
354 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
355 list_del_init(&dentry->d_lru);
356 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
357 this_cpu_dec(nr_dentry_unused);
360 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
362 D_FLAG_VERIFY(dentry, 0);
363 list_add(&dentry->d_lru, list);
364 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
365 this_cpu_inc(nr_dentry_unused);
369 * These can only be called under the global LRU lock, ie during the
370 * callback for freeing the LRU list. "isolate" removes it from the
371 * LRU lists entirely, while shrink_move moves it to the indicated
374 static void d_lru_isolate(struct dentry *dentry)
376 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
377 dentry->d_flags &= ~DCACHE_LRU_LIST;
378 this_cpu_dec(nr_dentry_unused);
379 list_del_init(&dentry->d_lru);
382 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
384 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
385 dentry->d_flags |= DCACHE_SHRINK_LIST;
386 list_move_tail(&dentry->d_lru, list);
390 * dentry_lru_(add|del)_list) must be called with d_lock held.
392 static void dentry_lru_add(struct dentry *dentry)
394 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
399 * d_drop - drop a dentry
400 * @dentry: dentry to drop
402 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
403 * be found through a VFS lookup any more. Note that this is different from
404 * deleting the dentry - d_delete will try to mark the dentry negative if
405 * possible, giving a successful _negative_ lookup, while d_drop will
406 * just make the cache lookup fail.
408 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
409 * reason (NFS timeouts or autofs deletes).
411 * __d_drop requires dentry->d_lock.
413 void __d_drop(struct dentry *dentry)
415 if (!d_unhashed(dentry)) {
416 struct hlist_bl_head *b;
418 * Hashed dentries are normally on the dentry hashtable,
419 * with the exception of those newly allocated by
420 * d_obtain_alias, which are always IS_ROOT:
422 if (unlikely(IS_ROOT(dentry)))
423 b = &dentry->d_sb->s_anon;
425 b = d_hash(dentry->d_parent, dentry->d_name.hash);
428 __hlist_bl_del(&dentry->d_hash);
429 dentry->d_hash.pprev = NULL;
431 dentry_rcuwalk_barrier(dentry);
434 EXPORT_SYMBOL(__d_drop);
436 void d_drop(struct dentry *dentry)
438 spin_lock(&dentry->d_lock);
440 spin_unlock(&dentry->d_lock);
442 EXPORT_SYMBOL(d_drop);
444 static void __dentry_kill(struct dentry *dentry)
446 struct dentry *parent = NULL;
447 bool can_free = true;
448 if (!IS_ROOT(dentry))
449 parent = dentry->d_parent;
452 * The dentry is now unrecoverably dead to the world.
454 lockref_mark_dead(&dentry->d_lockref);
457 * inform the fs via d_prune that this dentry is about to be
458 * unhashed and destroyed.
460 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
461 dentry->d_op->d_prune(dentry);
463 if (dentry->d_flags & DCACHE_LRU_LIST) {
464 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
467 /* if it was on the hash then remove it */
469 list_del(&dentry->d_u.d_child);
471 * Inform d_walk() that we are no longer attached to the
474 dentry->d_flags |= DCACHE_DENTRY_KILLED;
476 spin_unlock(&parent->d_lock);
479 * dentry_iput drops the locks, at which point nobody (except
480 * transient RCU lookups) can reach this dentry.
482 BUG_ON((int)dentry->d_lockref.count > 0);
483 this_cpu_dec(nr_dentry);
484 if (dentry->d_op && dentry->d_op->d_release)
485 dentry->d_op->d_release(dentry);
487 spin_lock(&dentry->d_lock);
488 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
489 dentry->d_flags |= DCACHE_MAY_FREE;
492 spin_unlock(&dentry->d_lock);
493 if (likely(can_free))
498 * Finish off a dentry we've decided to kill.
499 * dentry->d_lock must be held, returns with it unlocked.
500 * If ref is non-zero, then decrement the refcount too.
501 * Returns dentry requiring refcount drop, or NULL if we're done.
503 static struct dentry *dentry_kill(struct dentry *dentry)
504 __releases(dentry->d_lock)
506 struct inode *inode = dentry->d_inode;
507 struct dentry *parent = NULL;
509 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
512 if (!IS_ROOT(dentry)) {
513 parent = dentry->d_parent;
514 if (unlikely(!spin_trylock(&parent->d_lock))) {
516 spin_unlock(&inode->i_lock);
521 __dentry_kill(dentry);
525 spin_unlock(&dentry->d_lock);
527 return dentry; /* try again with same dentry */
530 static inline struct dentry *lock_parent(struct dentry *dentry)
532 struct dentry *parent = dentry->d_parent;
535 if (unlikely((int)dentry->d_lockref.count < 0))
537 if (likely(spin_trylock(&parent->d_lock)))
540 spin_unlock(&dentry->d_lock);
542 parent = ACCESS_ONCE(dentry->d_parent);
543 spin_lock(&parent->d_lock);
545 * We can't blindly lock dentry until we are sure
546 * that we won't violate the locking order.
547 * Any changes of dentry->d_parent must have
548 * been done with parent->d_lock held, so
549 * spin_lock() above is enough of a barrier
550 * for checking if it's still our child.
552 if (unlikely(parent != dentry->d_parent)) {
553 spin_unlock(&parent->d_lock);
557 if (parent != dentry)
558 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
567 * This is complicated by the fact that we do not want to put
568 * dentries that are no longer on any hash chain on the unused
569 * list: we'd much rather just get rid of them immediately.
571 * However, that implies that we have to traverse the dentry
572 * tree upwards to the parents which might _also_ now be
573 * scheduled for deletion (it may have been only waiting for
574 * its last child to go away).
576 * This tail recursion is done by hand as we don't want to depend
577 * on the compiler to always get this right (gcc generally doesn't).
578 * Real recursion would eat up our stack space.
582 * dput - release a dentry
583 * @dentry: dentry to release
585 * Release a dentry. This will drop the usage count and if appropriate
586 * call the dentry unlink method as well as removing it from the queues and
587 * releasing its resources. If the parent dentries were scheduled for release
588 * they too may now get deleted.
590 void dput(struct dentry *dentry)
592 if (unlikely(!dentry))
596 if (lockref_put_or_lock(&dentry->d_lockref))
599 /* Unreachable? Get rid of it */
600 if (unlikely(d_unhashed(dentry)))
603 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
604 if (dentry->d_op->d_delete(dentry))
608 if (!(dentry->d_flags & DCACHE_REFERENCED))
609 dentry->d_flags |= DCACHE_REFERENCED;
610 dentry_lru_add(dentry);
612 dentry->d_lockref.count--;
613 spin_unlock(&dentry->d_lock);
617 dentry = dentry_kill(dentry);
624 /* This must be called with d_lock held */
625 static inline void __dget_dlock(struct dentry *dentry)
627 dentry->d_lockref.count++;
630 static inline void __dget(struct dentry *dentry)
632 lockref_get(&dentry->d_lockref);
635 struct dentry *dget_parent(struct dentry *dentry)
641 * Do optimistic parent lookup without any
645 ret = ACCESS_ONCE(dentry->d_parent);
646 gotref = lockref_get_not_zero(&ret->d_lockref);
648 if (likely(gotref)) {
649 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
656 * Don't need rcu_dereference because we re-check it was correct under
660 ret = dentry->d_parent;
661 spin_lock(&ret->d_lock);
662 if (unlikely(ret != dentry->d_parent)) {
663 spin_unlock(&ret->d_lock);
668 BUG_ON(!ret->d_lockref.count);
669 ret->d_lockref.count++;
670 spin_unlock(&ret->d_lock);
673 EXPORT_SYMBOL(dget_parent);
676 * d_find_alias - grab a hashed alias of inode
677 * @inode: inode in question
678 * @want_discon: flag, used by d_splice_alias, to request
679 * that only a DISCONNECTED alias be returned.
681 * If inode has a hashed alias, or is a directory and has any alias,
682 * acquire the reference to alias and return it. Otherwise return NULL.
683 * Notice that if inode is a directory there can be only one alias and
684 * it can be unhashed only if it has no children, or if it is the root
685 * of a filesystem, or if the directory was renamed and d_revalidate
686 * was the first vfs operation to notice.
688 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
689 * any other hashed alias over that one unless @want_discon is set,
690 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
692 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
694 struct dentry *alias, *discon_alias;
698 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
699 spin_lock(&alias->d_lock);
700 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
701 if (IS_ROOT(alias) &&
702 (alias->d_flags & DCACHE_DISCONNECTED)) {
703 discon_alias = alias;
704 } else if (!want_discon) {
706 spin_unlock(&alias->d_lock);
710 spin_unlock(&alias->d_lock);
713 alias = discon_alias;
714 spin_lock(&alias->d_lock);
715 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
716 if (IS_ROOT(alias) &&
717 (alias->d_flags & DCACHE_DISCONNECTED)) {
719 spin_unlock(&alias->d_lock);
723 spin_unlock(&alias->d_lock);
729 struct dentry *d_find_alias(struct inode *inode)
731 struct dentry *de = NULL;
733 if (!hlist_empty(&inode->i_dentry)) {
734 spin_lock(&inode->i_lock);
735 de = __d_find_alias(inode, 0);
736 spin_unlock(&inode->i_lock);
740 EXPORT_SYMBOL(d_find_alias);
743 * Try to kill dentries associated with this inode.
744 * WARNING: you must own a reference to inode.
746 void d_prune_aliases(struct inode *inode)
748 struct dentry *dentry;
750 spin_lock(&inode->i_lock);
751 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
752 spin_lock(&dentry->d_lock);
753 if (!dentry->d_lockref.count) {
755 * inform the fs via d_prune that this dentry
756 * is about to be unhashed and destroyed.
758 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
760 dentry->d_op->d_prune(dentry);
762 __dget_dlock(dentry);
764 spin_unlock(&dentry->d_lock);
765 spin_unlock(&inode->i_lock);
769 spin_unlock(&dentry->d_lock);
771 spin_unlock(&inode->i_lock);
773 EXPORT_SYMBOL(d_prune_aliases);
775 static void shrink_dentry_list(struct list_head *list)
777 struct dentry *dentry, *parent;
779 while (!list_empty(list)) {
781 dentry = list_entry(list->prev, struct dentry, d_lru);
782 spin_lock(&dentry->d_lock);
783 parent = lock_parent(dentry);
786 * The dispose list is isolated and dentries are not accounted
787 * to the LRU here, so we can simply remove it from the list
788 * here regardless of whether it is referenced or not.
790 d_shrink_del(dentry);
793 * We found an inuse dentry which was not removed from
794 * the LRU because of laziness during lookup. Do not free it.
796 if ((int)dentry->d_lockref.count > 0) {
797 spin_unlock(&dentry->d_lock);
799 spin_unlock(&parent->d_lock);
804 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
805 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
806 spin_unlock(&dentry->d_lock);
808 spin_unlock(&parent->d_lock);
814 inode = dentry->d_inode;
815 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
816 d_shrink_add(dentry, list);
817 spin_unlock(&dentry->d_lock);
819 spin_unlock(&parent->d_lock);
823 __dentry_kill(dentry);
826 * We need to prune ancestors too. This is necessary to prevent
827 * quadratic behavior of shrink_dcache_parent(), but is also
828 * expected to be beneficial in reducing dentry cache
832 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
833 parent = lock_parent(dentry);
834 if (dentry->d_lockref.count != 1) {
835 dentry->d_lockref.count--;
836 spin_unlock(&dentry->d_lock);
838 spin_unlock(&parent->d_lock);
841 inode = dentry->d_inode; /* can't be NULL */
842 if (unlikely(!spin_trylock(&inode->i_lock))) {
843 spin_unlock(&dentry->d_lock);
845 spin_unlock(&parent->d_lock);
849 __dentry_kill(dentry);
855 static enum lru_status
856 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
858 struct list_head *freeable = arg;
859 struct dentry *dentry = container_of(item, struct dentry, d_lru);
863 * we are inverting the lru lock/dentry->d_lock here,
864 * so use a trylock. If we fail to get the lock, just skip
867 if (!spin_trylock(&dentry->d_lock))
871 * Referenced dentries are still in use. If they have active
872 * counts, just remove them from the LRU. Otherwise give them
873 * another pass through the LRU.
875 if (dentry->d_lockref.count) {
876 d_lru_isolate(dentry);
877 spin_unlock(&dentry->d_lock);
881 if (dentry->d_flags & DCACHE_REFERENCED) {
882 dentry->d_flags &= ~DCACHE_REFERENCED;
883 spin_unlock(&dentry->d_lock);
886 * The list move itself will be made by the common LRU code. At
887 * this point, we've dropped the dentry->d_lock but keep the
888 * lru lock. This is safe to do, since every list movement is
889 * protected by the lru lock even if both locks are held.
891 * This is guaranteed by the fact that all LRU management
892 * functions are intermediated by the LRU API calls like
893 * list_lru_add and list_lru_del. List movement in this file
894 * only ever occur through this functions or through callbacks
895 * like this one, that are called from the LRU API.
897 * The only exceptions to this are functions like
898 * shrink_dentry_list, and code that first checks for the
899 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
900 * operating only with stack provided lists after they are
901 * properly isolated from the main list. It is thus, always a
907 d_lru_shrink_move(dentry, freeable);
908 spin_unlock(&dentry->d_lock);
914 * prune_dcache_sb - shrink the dcache
916 * @nr_to_scan : number of entries to try to free
917 * @nid: which node to scan for freeable entities
919 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
920 * done when we need more memory an called from the superblock shrinker
923 * This function may fail to free any resources if all the dentries are in
926 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
932 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
933 &dispose, &nr_to_scan);
934 shrink_dentry_list(&dispose);
938 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
939 spinlock_t *lru_lock, void *arg)
941 struct list_head *freeable = arg;
942 struct dentry *dentry = container_of(item, struct dentry, d_lru);
945 * we are inverting the lru lock/dentry->d_lock here,
946 * so use a trylock. If we fail to get the lock, just skip
949 if (!spin_trylock(&dentry->d_lock))
952 d_lru_shrink_move(dentry, freeable);
953 spin_unlock(&dentry->d_lock);
960 * shrink_dcache_sb - shrink dcache for a superblock
963 * Shrink the dcache for the specified super block. This is used to free
964 * the dcache before unmounting a file system.
966 void shrink_dcache_sb(struct super_block *sb)
973 freed = list_lru_walk(&sb->s_dentry_lru,
974 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
976 this_cpu_sub(nr_dentry_unused, freed);
977 shrink_dentry_list(&dispose);
980 EXPORT_SYMBOL(shrink_dcache_sb);
983 * enum d_walk_ret - action to talke during tree walk
984 * @D_WALK_CONTINUE: contrinue walk
985 * @D_WALK_QUIT: quit walk
986 * @D_WALK_NORETRY: quit when retry is needed
987 * @D_WALK_SKIP: skip this dentry and its children
997 * d_walk - walk the dentry tree
998 * @parent: start of walk
999 * @data: data passed to @enter() and @finish()
1000 * @enter: callback when first entering the dentry
1001 * @finish: callback when successfully finished the walk
1003 * The @enter() and @finish() callbacks are called with d_lock held.
1005 static void d_walk(struct dentry *parent, void *data,
1006 enum d_walk_ret (*enter)(void *, struct dentry *),
1007 void (*finish)(void *))
1009 struct dentry *this_parent;
1010 struct list_head *next;
1012 enum d_walk_ret ret;
1016 read_seqbegin_or_lock(&rename_lock, &seq);
1017 this_parent = parent;
1018 spin_lock(&this_parent->d_lock);
1020 ret = enter(data, this_parent);
1022 case D_WALK_CONTINUE:
1027 case D_WALK_NORETRY:
1032 next = this_parent->d_subdirs.next;
1034 while (next != &this_parent->d_subdirs) {
1035 struct list_head *tmp = next;
1036 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1039 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1041 ret = enter(data, dentry);
1043 case D_WALK_CONTINUE:
1046 spin_unlock(&dentry->d_lock);
1048 case D_WALK_NORETRY:
1052 spin_unlock(&dentry->d_lock);
1056 if (!list_empty(&dentry->d_subdirs)) {
1057 spin_unlock(&this_parent->d_lock);
1058 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1059 this_parent = dentry;
1060 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1063 spin_unlock(&dentry->d_lock);
1066 * All done at this level ... ascend and resume the search.
1068 if (this_parent != parent) {
1069 struct dentry *child = this_parent;
1070 this_parent = child->d_parent;
1073 spin_unlock(&child->d_lock);
1074 spin_lock(&this_parent->d_lock);
1077 * might go back up the wrong parent if we have had a rename
1080 if (this_parent != child->d_parent ||
1081 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1082 need_seqretry(&rename_lock, seq)) {
1083 spin_unlock(&this_parent->d_lock);
1088 next = child->d_u.d_child.next;
1091 if (need_seqretry(&rename_lock, seq)) {
1092 spin_unlock(&this_parent->d_lock);
1099 spin_unlock(&this_parent->d_lock);
1100 done_seqretry(&rename_lock, seq);
1111 * Search for at least 1 mount point in the dentry's subdirs.
1112 * We descend to the next level whenever the d_subdirs
1113 * list is non-empty and continue searching.
1116 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1119 if (d_mountpoint(dentry)) {
1123 return D_WALK_CONTINUE;
1127 * have_submounts - check for mounts over a dentry
1128 * @parent: dentry to check.
1130 * Return true if the parent or its subdirectories contain
1133 int have_submounts(struct dentry *parent)
1137 d_walk(parent, &ret, check_mount, NULL);
1141 EXPORT_SYMBOL(have_submounts);
1144 * Called by mount code to set a mountpoint and check if the mountpoint is
1145 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1146 * subtree can become unreachable).
1148 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1149 * this reason take rename_lock and d_lock on dentry and ancestors.
1151 int d_set_mounted(struct dentry *dentry)
1155 write_seqlock(&rename_lock);
1156 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1157 /* Need exclusion wrt. d_invalidate() */
1158 spin_lock(&p->d_lock);
1159 if (unlikely(d_unhashed(p))) {
1160 spin_unlock(&p->d_lock);
1163 spin_unlock(&p->d_lock);
1165 spin_lock(&dentry->d_lock);
1166 if (!d_unlinked(dentry)) {
1167 dentry->d_flags |= DCACHE_MOUNTED;
1170 spin_unlock(&dentry->d_lock);
1172 write_sequnlock(&rename_lock);
1177 * Search the dentry child list of the specified parent,
1178 * and move any unused dentries to the end of the unused
1179 * list for prune_dcache(). We descend to the next level
1180 * whenever the d_subdirs list is non-empty and continue
1183 * It returns zero iff there are no unused children,
1184 * otherwise it returns the number of children moved to
1185 * the end of the unused list. This may not be the total
1186 * number of unused children, because select_parent can
1187 * drop the lock and return early due to latency
1191 struct select_data {
1192 struct dentry *start;
1193 struct list_head dispose;
1197 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1199 struct select_data *data = _data;
1200 enum d_walk_ret ret = D_WALK_CONTINUE;
1202 if (data->start == dentry)
1205 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1208 if (dentry->d_flags & DCACHE_LRU_LIST)
1210 if (!dentry->d_lockref.count) {
1211 d_shrink_add(dentry, &data->dispose);
1216 * We can return to the caller if we have found some (this
1217 * ensures forward progress). We'll be coming back to find
1220 if (!list_empty(&data->dispose))
1221 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1227 * shrink_dcache_parent - prune dcache
1228 * @parent: parent of entries to prune
1230 * Prune the dcache to remove unused children of the parent dentry.
1232 void shrink_dcache_parent(struct dentry *parent)
1235 struct select_data data;
1237 INIT_LIST_HEAD(&data.dispose);
1238 data.start = parent;
1241 d_walk(parent, &data, select_collect, NULL);
1245 shrink_dentry_list(&data.dispose);
1249 EXPORT_SYMBOL(shrink_dcache_parent);
1251 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1253 /* it has busy descendents; complain about those instead */
1254 if (!list_empty(&dentry->d_subdirs))
1255 return D_WALK_CONTINUE;
1257 /* root with refcount 1 is fine */
1258 if (dentry == _data && dentry->d_lockref.count == 1)
1259 return D_WALK_CONTINUE;
1261 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1262 " still in use (%d) [unmount of %s %s]\n",
1265 dentry->d_inode->i_ino : 0UL,
1267 dentry->d_lockref.count,
1268 dentry->d_sb->s_type->name,
1269 dentry->d_sb->s_id);
1271 return D_WALK_CONTINUE;
1274 static void do_one_tree(struct dentry *dentry)
1276 shrink_dcache_parent(dentry);
1277 d_walk(dentry, dentry, umount_check, NULL);
1283 * destroy the dentries attached to a superblock on unmounting
1285 void shrink_dcache_for_umount(struct super_block *sb)
1287 struct dentry *dentry;
1289 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1291 dentry = sb->s_root;
1293 do_one_tree(dentry);
1295 while (!hlist_bl_empty(&sb->s_anon)) {
1296 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1297 do_one_tree(dentry);
1301 struct detach_data {
1302 struct select_data select;
1303 struct dentry *mountpoint;
1305 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1307 struct detach_data *data = _data;
1309 if (d_mountpoint(dentry)) {
1310 __dget_dlock(dentry);
1311 data->mountpoint = dentry;
1315 return select_collect(&data->select, dentry);
1318 static void check_and_drop(void *_data)
1320 struct detach_data *data = _data;
1322 if (!data->mountpoint && !data->select.found)
1323 __d_drop(data->select.start);
1327 * d_invalidate - detach submounts, prune dcache, and drop
1328 * @dentry: dentry to invalidate (aka detach, prune and drop)
1332 * The final d_drop is done as an atomic operation relative to
1333 * rename_lock ensuring there are no races with d_set_mounted. This
1334 * ensures there are no unhashed dentries on the path to a mountpoint.
1336 void d_invalidate(struct dentry *dentry)
1339 * If it's already been dropped, return OK.
1341 spin_lock(&dentry->d_lock);
1342 if (d_unhashed(dentry)) {
1343 spin_unlock(&dentry->d_lock);
1346 spin_unlock(&dentry->d_lock);
1348 /* Negative dentries can be dropped without further checks */
1349 if (!dentry->d_inode) {
1355 struct detach_data data;
1357 data.mountpoint = NULL;
1358 INIT_LIST_HEAD(&data.select.dispose);
1359 data.select.start = dentry;
1360 data.select.found = 0;
1362 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1364 if (data.select.found)
1365 shrink_dentry_list(&data.select.dispose);
1367 if (data.mountpoint) {
1368 detach_mounts(data.mountpoint);
1369 dput(data.mountpoint);
1372 if (!data.mountpoint && !data.select.found)
1381 EXPORT_SYMBOL(d_invalidate);
1384 * __d_alloc - allocate a dcache entry
1385 * @sb: filesystem it will belong to
1386 * @name: qstr of the name
1388 * Allocates a dentry. It returns %NULL if there is insufficient memory
1389 * available. On a success the dentry is returned. The name passed in is
1390 * copied and the copy passed in may be reused after this call.
1393 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1395 struct dentry *dentry;
1398 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1403 * We guarantee that the inline name is always NUL-terminated.
1404 * This way the memcpy() done by the name switching in rename
1405 * will still always have a NUL at the end, even if we might
1406 * be overwriting an internal NUL character
1408 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1409 if (name->len > DNAME_INLINE_LEN-1) {
1410 dname = kmalloc(name->len + 1, GFP_KERNEL);
1412 kmem_cache_free(dentry_cache, dentry);
1416 dname = dentry->d_iname;
1419 dentry->d_name.len = name->len;
1420 dentry->d_name.hash = name->hash;
1421 memcpy(dname, name->name, name->len);
1422 dname[name->len] = 0;
1424 /* Make sure we always see the terminating NUL character */
1426 dentry->d_name.name = dname;
1428 dentry->d_lockref.count = 1;
1429 dentry->d_flags = 0;
1430 spin_lock_init(&dentry->d_lock);
1431 seqcount_init(&dentry->d_seq);
1432 dentry->d_inode = NULL;
1433 dentry->d_parent = dentry;
1435 dentry->d_op = NULL;
1436 dentry->d_fsdata = NULL;
1437 INIT_HLIST_BL_NODE(&dentry->d_hash);
1438 INIT_LIST_HEAD(&dentry->d_lru);
1439 INIT_LIST_HEAD(&dentry->d_subdirs);
1440 INIT_HLIST_NODE(&dentry->d_alias);
1441 INIT_LIST_HEAD(&dentry->d_u.d_child);
1442 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1444 this_cpu_inc(nr_dentry);
1450 * d_alloc - allocate a dcache entry
1451 * @parent: parent of entry to allocate
1452 * @name: qstr of the name
1454 * Allocates a dentry. It returns %NULL if there is insufficient memory
1455 * available. On a success the dentry is returned. The name passed in is
1456 * copied and the copy passed in may be reused after this call.
1458 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1460 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1464 spin_lock(&parent->d_lock);
1466 * don't need child lock because it is not subject
1467 * to concurrency here
1469 __dget_dlock(parent);
1470 dentry->d_parent = parent;
1471 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1472 spin_unlock(&parent->d_lock);
1476 EXPORT_SYMBOL(d_alloc);
1479 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1480 * @sb: the superblock
1481 * @name: qstr of the name
1483 * For a filesystem that just pins its dentries in memory and never
1484 * performs lookups at all, return an unhashed IS_ROOT dentry.
1486 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1488 return __d_alloc(sb, name);
1490 EXPORT_SYMBOL(d_alloc_pseudo);
1492 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1497 q.len = strlen(name);
1498 q.hash = full_name_hash(q.name, q.len);
1499 return d_alloc(parent, &q);
1501 EXPORT_SYMBOL(d_alloc_name);
1503 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1505 WARN_ON_ONCE(dentry->d_op);
1506 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1508 DCACHE_OP_REVALIDATE |
1509 DCACHE_OP_WEAK_REVALIDATE |
1510 DCACHE_OP_DELETE ));
1515 dentry->d_flags |= DCACHE_OP_HASH;
1517 dentry->d_flags |= DCACHE_OP_COMPARE;
1518 if (op->d_revalidate)
1519 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1520 if (op->d_weak_revalidate)
1521 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1523 dentry->d_flags |= DCACHE_OP_DELETE;
1525 dentry->d_flags |= DCACHE_OP_PRUNE;
1528 EXPORT_SYMBOL(d_set_d_op);
1530 static unsigned d_flags_for_inode(struct inode *inode)
1532 unsigned add_flags = DCACHE_FILE_TYPE;
1535 return DCACHE_MISS_TYPE;
1537 if (S_ISDIR(inode->i_mode)) {
1538 add_flags = DCACHE_DIRECTORY_TYPE;
1539 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1540 if (unlikely(!inode->i_op->lookup))
1541 add_flags = DCACHE_AUTODIR_TYPE;
1543 inode->i_opflags |= IOP_LOOKUP;
1545 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1546 if (unlikely(inode->i_op->follow_link))
1547 add_flags = DCACHE_SYMLINK_TYPE;
1549 inode->i_opflags |= IOP_NOFOLLOW;
1552 if (unlikely(IS_AUTOMOUNT(inode)))
1553 add_flags |= DCACHE_NEED_AUTOMOUNT;
1557 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1559 unsigned add_flags = d_flags_for_inode(inode);
1561 spin_lock(&dentry->d_lock);
1562 __d_set_type(dentry, add_flags);
1564 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1565 dentry->d_inode = inode;
1566 dentry_rcuwalk_barrier(dentry);
1567 spin_unlock(&dentry->d_lock);
1568 fsnotify_d_instantiate(dentry, inode);
1572 * d_instantiate - fill in inode information for a dentry
1573 * @entry: dentry to complete
1574 * @inode: inode to attach to this dentry
1576 * Fill in inode information in the entry.
1578 * This turns negative dentries into productive full members
1581 * NOTE! This assumes that the inode count has been incremented
1582 * (or otherwise set) by the caller to indicate that it is now
1583 * in use by the dcache.
1586 void d_instantiate(struct dentry *entry, struct inode * inode)
1588 BUG_ON(!hlist_unhashed(&entry->d_alias));
1590 spin_lock(&inode->i_lock);
1591 __d_instantiate(entry, inode);
1593 spin_unlock(&inode->i_lock);
1594 security_d_instantiate(entry, inode);
1596 EXPORT_SYMBOL(d_instantiate);
1599 * d_instantiate_unique - instantiate a non-aliased dentry
1600 * @entry: dentry to instantiate
1601 * @inode: inode to attach to this dentry
1603 * Fill in inode information in the entry. On success, it returns NULL.
1604 * If an unhashed alias of "entry" already exists, then we return the
1605 * aliased dentry instead and drop one reference to inode.
1607 * Note that in order to avoid conflicts with rename() etc, the caller
1608 * had better be holding the parent directory semaphore.
1610 * This also assumes that the inode count has been incremented
1611 * (or otherwise set) by the caller to indicate that it is now
1612 * in use by the dcache.
1614 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1615 struct inode *inode)
1617 struct dentry *alias;
1618 int len = entry->d_name.len;
1619 const char *name = entry->d_name.name;
1620 unsigned int hash = entry->d_name.hash;
1623 __d_instantiate(entry, NULL);
1627 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1629 * Don't need alias->d_lock here, because aliases with
1630 * d_parent == entry->d_parent are not subject to name or
1631 * parent changes, because the parent inode i_mutex is held.
1633 if (alias->d_name.hash != hash)
1635 if (alias->d_parent != entry->d_parent)
1637 if (alias->d_name.len != len)
1639 if (dentry_cmp(alias, name, len))
1645 __d_instantiate(entry, inode);
1649 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1651 struct dentry *result;
1653 BUG_ON(!hlist_unhashed(&entry->d_alias));
1656 spin_lock(&inode->i_lock);
1657 result = __d_instantiate_unique(entry, inode);
1659 spin_unlock(&inode->i_lock);
1662 security_d_instantiate(entry, inode);
1666 BUG_ON(!d_unhashed(result));
1671 EXPORT_SYMBOL(d_instantiate_unique);
1674 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1675 * @entry: dentry to complete
1676 * @inode: inode to attach to this dentry
1678 * Fill in inode information in the entry. If a directory alias is found, then
1679 * return an error (and drop inode). Together with d_materialise_unique() this
1680 * guarantees that a directory inode may never have more than one alias.
1682 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1684 BUG_ON(!hlist_unhashed(&entry->d_alias));
1686 spin_lock(&inode->i_lock);
1687 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1688 spin_unlock(&inode->i_lock);
1692 __d_instantiate(entry, inode);
1693 spin_unlock(&inode->i_lock);
1694 security_d_instantiate(entry, inode);
1698 EXPORT_SYMBOL(d_instantiate_no_diralias);
1700 struct dentry *d_make_root(struct inode *root_inode)
1702 struct dentry *res = NULL;
1705 static const struct qstr name = QSTR_INIT("/", 1);
1707 res = __d_alloc(root_inode->i_sb, &name);
1709 d_instantiate(res, root_inode);
1715 EXPORT_SYMBOL(d_make_root);
1717 static struct dentry * __d_find_any_alias(struct inode *inode)
1719 struct dentry *alias;
1721 if (hlist_empty(&inode->i_dentry))
1723 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1729 * d_find_any_alias - find any alias for a given inode
1730 * @inode: inode to find an alias for
1732 * If any aliases exist for the given inode, take and return a
1733 * reference for one of them. If no aliases exist, return %NULL.
1735 struct dentry *d_find_any_alias(struct inode *inode)
1739 spin_lock(&inode->i_lock);
1740 de = __d_find_any_alias(inode);
1741 spin_unlock(&inode->i_lock);
1744 EXPORT_SYMBOL(d_find_any_alias);
1747 * d_obtain_alias - find or allocate a dentry for a given inode
1748 * @inode: inode to allocate the dentry for
1750 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1751 * similar open by handle operations. The returned dentry may be anonymous,
1752 * or may have a full name (if the inode was already in the cache).
1754 * When called on a directory inode, we must ensure that the inode only ever
1755 * has one dentry. If a dentry is found, that is returned instead of
1756 * allocating a new one.
1758 * On successful return, the reference to the inode has been transferred
1759 * to the dentry. In case of an error the reference on the inode is released.
1760 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1761 * be passed in and will be the error will be propagate to the return value,
1762 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1764 struct dentry *d_obtain_alias(struct inode *inode)
1766 static const struct qstr anonstring = QSTR_INIT("/", 1);
1772 return ERR_PTR(-ESTALE);
1774 return ERR_CAST(inode);
1776 res = d_find_any_alias(inode);
1780 tmp = __d_alloc(inode->i_sb, &anonstring);
1782 res = ERR_PTR(-ENOMEM);
1786 spin_lock(&inode->i_lock);
1787 res = __d_find_any_alias(inode);
1789 spin_unlock(&inode->i_lock);
1794 /* attach a disconnected dentry */
1795 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1797 spin_lock(&tmp->d_lock);
1798 tmp->d_inode = inode;
1799 tmp->d_flags |= add_flags;
1800 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1801 hlist_bl_lock(&tmp->d_sb->s_anon);
1802 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1803 hlist_bl_unlock(&tmp->d_sb->s_anon);
1804 spin_unlock(&tmp->d_lock);
1805 spin_unlock(&inode->i_lock);
1806 security_d_instantiate(tmp, inode);
1811 if (res && !IS_ERR(res))
1812 security_d_instantiate(res, inode);
1816 EXPORT_SYMBOL(d_obtain_alias);
1819 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1820 * @inode: the inode which may have a disconnected dentry
1821 * @dentry: a negative dentry which we want to point to the inode.
1823 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1824 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1825 * and return it, else simply d_add the inode to the dentry and return NULL.
1827 * This is needed in the lookup routine of any filesystem that is exportable
1828 * (via knfsd) so that we can build dcache paths to directories effectively.
1830 * If a dentry was found and moved, then it is returned. Otherwise NULL
1831 * is returned. This matches the expected return value of ->lookup.
1833 * Cluster filesystems may call this function with a negative, hashed dentry.
1834 * In that case, we know that the inode will be a regular file, and also this
1835 * will only occur during atomic_open. So we need to check for the dentry
1836 * being already hashed only in the final case.
1838 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1840 struct dentry *new = NULL;
1843 return ERR_CAST(inode);
1845 if (inode && S_ISDIR(inode->i_mode)) {
1846 spin_lock(&inode->i_lock);
1847 new = __d_find_alias(inode, 1);
1849 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1850 spin_unlock(&inode->i_lock);
1851 security_d_instantiate(new, inode);
1852 d_move(new, dentry);
1855 /* already taking inode->i_lock, so d_add() by hand */
1856 __d_instantiate(dentry, inode);
1857 spin_unlock(&inode->i_lock);
1858 security_d_instantiate(dentry, inode);
1862 d_instantiate(dentry, inode);
1863 if (d_unhashed(dentry))
1868 EXPORT_SYMBOL(d_splice_alias);
1871 * d_add_ci - lookup or allocate new dentry with case-exact name
1872 * @inode: the inode case-insensitive lookup has found
1873 * @dentry: the negative dentry that was passed to the parent's lookup func
1874 * @name: the case-exact name to be associated with the returned dentry
1876 * This is to avoid filling the dcache with case-insensitive names to the
1877 * same inode, only the actual correct case is stored in the dcache for
1878 * case-insensitive filesystems.
1880 * For a case-insensitive lookup match and if the the case-exact dentry
1881 * already exists in in the dcache, use it and return it.
1883 * If no entry exists with the exact case name, allocate new dentry with
1884 * the exact case, and return the spliced entry.
1886 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1889 struct dentry *found;
1893 * First check if a dentry matching the name already exists,
1894 * if not go ahead and create it now.
1896 found = d_hash_and_lookup(dentry->d_parent, name);
1897 if (unlikely(IS_ERR(found)))
1900 new = d_alloc(dentry->d_parent, name);
1902 found = ERR_PTR(-ENOMEM);
1906 found = d_splice_alias(inode, new);
1915 * If a matching dentry exists, and it's not negative use it.
1917 * Decrement the reference count to balance the iget() done
1920 if (found->d_inode) {
1921 if (unlikely(found->d_inode != inode)) {
1922 /* This can't happen because bad inodes are unhashed. */
1923 BUG_ON(!is_bad_inode(inode));
1924 BUG_ON(!is_bad_inode(found->d_inode));
1931 * Negative dentry: instantiate it unless the inode is a directory and
1932 * already has a dentry.
1934 new = d_splice_alias(inode, found);
1945 EXPORT_SYMBOL(d_add_ci);
1948 * Do the slow-case of the dentry name compare.
1950 * Unlike the dentry_cmp() function, we need to atomically
1951 * load the name and length information, so that the
1952 * filesystem can rely on them, and can use the 'name' and
1953 * 'len' information without worrying about walking off the
1954 * end of memory etc.
1956 * Thus the read_seqcount_retry() and the "duplicate" info
1957 * in arguments (the low-level filesystem should not look
1958 * at the dentry inode or name contents directly, since
1959 * rename can change them while we're in RCU mode).
1961 enum slow_d_compare {
1967 static noinline enum slow_d_compare slow_dentry_cmp(
1968 const struct dentry *parent,
1969 struct dentry *dentry,
1971 const struct qstr *name)
1973 int tlen = dentry->d_name.len;
1974 const char *tname = dentry->d_name.name;
1976 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1978 return D_COMP_SEQRETRY;
1980 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1981 return D_COMP_NOMATCH;
1986 * __d_lookup_rcu - search for a dentry (racy, store-free)
1987 * @parent: parent dentry
1988 * @name: qstr of name we wish to find
1989 * @seqp: returns d_seq value at the point where the dentry was found
1990 * Returns: dentry, or NULL
1992 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1993 * resolution (store-free path walking) design described in
1994 * Documentation/filesystems/path-lookup.txt.
1996 * This is not to be used outside core vfs.
1998 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1999 * held, and rcu_read_lock held. The returned dentry must not be stored into
2000 * without taking d_lock and checking d_seq sequence count against @seq
2003 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2006 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2007 * the returned dentry, so long as its parent's seqlock is checked after the
2008 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2009 * is formed, giving integrity down the path walk.
2011 * NOTE! The caller *has* to check the resulting dentry against the sequence
2012 * number we've returned before using any of the resulting dentry state!
2014 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2015 const struct qstr *name,
2018 u64 hashlen = name->hash_len;
2019 const unsigned char *str = name->name;
2020 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2021 struct hlist_bl_node *node;
2022 struct dentry *dentry;
2025 * Note: There is significant duplication with __d_lookup_rcu which is
2026 * required to prevent single threaded performance regressions
2027 * especially on architectures where smp_rmb (in seqcounts) are costly.
2028 * Keep the two functions in sync.
2032 * The hash list is protected using RCU.
2034 * Carefully use d_seq when comparing a candidate dentry, to avoid
2035 * races with d_move().
2037 * It is possible that concurrent renames can mess up our list
2038 * walk here and result in missing our dentry, resulting in the
2039 * false-negative result. d_lookup() protects against concurrent
2040 * renames using rename_lock seqlock.
2042 * See Documentation/filesystems/path-lookup.txt for more details.
2044 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2049 * The dentry sequence count protects us from concurrent
2050 * renames, and thus protects parent and name fields.
2052 * The caller must perform a seqcount check in order
2053 * to do anything useful with the returned dentry.
2055 * NOTE! We do a "raw" seqcount_begin here. That means that
2056 * we don't wait for the sequence count to stabilize if it
2057 * is in the middle of a sequence change. If we do the slow
2058 * dentry compare, we will do seqretries until it is stable,
2059 * and if we end up with a successful lookup, we actually
2060 * want to exit RCU lookup anyway.
2062 seq = raw_seqcount_begin(&dentry->d_seq);
2063 if (dentry->d_parent != parent)
2065 if (d_unhashed(dentry))
2068 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2069 if (dentry->d_name.hash != hashlen_hash(hashlen))
2072 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2075 case D_COMP_NOMATCH:
2082 if (dentry->d_name.hash_len != hashlen)
2085 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2092 * d_lookup - search for a dentry
2093 * @parent: parent dentry
2094 * @name: qstr of name we wish to find
2095 * Returns: dentry, or NULL
2097 * d_lookup searches the children of the parent dentry for the name in
2098 * question. If the dentry is found its reference count is incremented and the
2099 * dentry is returned. The caller must use dput to free the entry when it has
2100 * finished using it. %NULL is returned if the dentry does not exist.
2102 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2104 struct dentry *dentry;
2108 seq = read_seqbegin(&rename_lock);
2109 dentry = __d_lookup(parent, name);
2112 } while (read_seqretry(&rename_lock, seq));
2115 EXPORT_SYMBOL(d_lookup);
2118 * __d_lookup - search for a dentry (racy)
2119 * @parent: parent dentry
2120 * @name: qstr of name we wish to find
2121 * Returns: dentry, or NULL
2123 * __d_lookup is like d_lookup, however it may (rarely) return a
2124 * false-negative result due to unrelated rename activity.
2126 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2127 * however it must be used carefully, eg. with a following d_lookup in
2128 * the case of failure.
2130 * __d_lookup callers must be commented.
2132 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2134 unsigned int len = name->len;
2135 unsigned int hash = name->hash;
2136 const unsigned char *str = name->name;
2137 struct hlist_bl_head *b = d_hash(parent, hash);
2138 struct hlist_bl_node *node;
2139 struct dentry *found = NULL;
2140 struct dentry *dentry;
2143 * Note: There is significant duplication with __d_lookup_rcu which is
2144 * required to prevent single threaded performance regressions
2145 * especially on architectures where smp_rmb (in seqcounts) are costly.
2146 * Keep the two functions in sync.
2150 * The hash list is protected using RCU.
2152 * Take d_lock when comparing a candidate dentry, to avoid races
2155 * It is possible that concurrent renames can mess up our list
2156 * walk here and result in missing our dentry, resulting in the
2157 * false-negative result. d_lookup() protects against concurrent
2158 * renames using rename_lock seqlock.
2160 * See Documentation/filesystems/path-lookup.txt for more details.
2164 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2166 if (dentry->d_name.hash != hash)
2169 spin_lock(&dentry->d_lock);
2170 if (dentry->d_parent != parent)
2172 if (d_unhashed(dentry))
2176 * It is safe to compare names since d_move() cannot
2177 * change the qstr (protected by d_lock).
2179 if (parent->d_flags & DCACHE_OP_COMPARE) {
2180 int tlen = dentry->d_name.len;
2181 const char *tname = dentry->d_name.name;
2182 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2185 if (dentry->d_name.len != len)
2187 if (dentry_cmp(dentry, str, len))
2191 dentry->d_lockref.count++;
2193 spin_unlock(&dentry->d_lock);
2196 spin_unlock(&dentry->d_lock);
2204 * d_hash_and_lookup - hash the qstr then search for a dentry
2205 * @dir: Directory to search in
2206 * @name: qstr of name we wish to find
2208 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2210 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2213 * Check for a fs-specific hash function. Note that we must
2214 * calculate the standard hash first, as the d_op->d_hash()
2215 * routine may choose to leave the hash value unchanged.
2217 name->hash = full_name_hash(name->name, name->len);
2218 if (dir->d_flags & DCACHE_OP_HASH) {
2219 int err = dir->d_op->d_hash(dir, name);
2220 if (unlikely(err < 0))
2221 return ERR_PTR(err);
2223 return d_lookup(dir, name);
2225 EXPORT_SYMBOL(d_hash_and_lookup);
2228 * d_validate - verify dentry provided from insecure source (deprecated)
2229 * @dentry: The dentry alleged to be valid child of @dparent
2230 * @dparent: The parent dentry (known to be valid)
2232 * An insecure source has sent us a dentry, here we verify it and dget() it.
2233 * This is used by ncpfs in its readdir implementation.
2234 * Zero is returned in the dentry is invalid.
2236 * This function is slow for big directories, and deprecated, do not use it.
2238 int d_validate(struct dentry *dentry, struct dentry *dparent)
2240 struct dentry *child;
2242 spin_lock(&dparent->d_lock);
2243 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2244 if (dentry == child) {
2245 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2246 __dget_dlock(dentry);
2247 spin_unlock(&dentry->d_lock);
2248 spin_unlock(&dparent->d_lock);
2252 spin_unlock(&dparent->d_lock);
2256 EXPORT_SYMBOL(d_validate);
2259 * When a file is deleted, we have two options:
2260 * - turn this dentry into a negative dentry
2261 * - unhash this dentry and free it.
2263 * Usually, we want to just turn this into
2264 * a negative dentry, but if anybody else is
2265 * currently using the dentry or the inode
2266 * we can't do that and we fall back on removing
2267 * it from the hash queues and waiting for
2268 * it to be deleted later when it has no users
2272 * d_delete - delete a dentry
2273 * @dentry: The dentry to delete
2275 * Turn the dentry into a negative dentry if possible, otherwise
2276 * remove it from the hash queues so it can be deleted later
2279 void d_delete(struct dentry * dentry)
2281 struct inode *inode;
2284 * Are we the only user?
2287 spin_lock(&dentry->d_lock);
2288 inode = dentry->d_inode;
2289 isdir = S_ISDIR(inode->i_mode);
2290 if (dentry->d_lockref.count == 1) {
2291 if (!spin_trylock(&inode->i_lock)) {
2292 spin_unlock(&dentry->d_lock);
2296 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2297 dentry_unlink_inode(dentry);
2298 fsnotify_nameremove(dentry, isdir);
2302 if (!d_unhashed(dentry))
2305 spin_unlock(&dentry->d_lock);
2307 fsnotify_nameremove(dentry, isdir);
2309 EXPORT_SYMBOL(d_delete);
2311 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2313 BUG_ON(!d_unhashed(entry));
2315 entry->d_flags |= DCACHE_RCUACCESS;
2316 hlist_bl_add_head_rcu(&entry->d_hash, b);
2320 static void _d_rehash(struct dentry * entry)
2322 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2326 * d_rehash - add an entry back to the hash
2327 * @entry: dentry to add to the hash
2329 * Adds a dentry to the hash according to its name.
2332 void d_rehash(struct dentry * entry)
2334 spin_lock(&entry->d_lock);
2336 spin_unlock(&entry->d_lock);
2338 EXPORT_SYMBOL(d_rehash);
2341 * dentry_update_name_case - update case insensitive dentry with a new name
2342 * @dentry: dentry to be updated
2345 * Update a case insensitive dentry with new case of name.
2347 * dentry must have been returned by d_lookup with name @name. Old and new
2348 * name lengths must match (ie. no d_compare which allows mismatched name
2351 * Parent inode i_mutex must be held over d_lookup and into this call (to
2352 * keep renames and concurrent inserts, and readdir(2) away).
2354 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2356 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2357 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2359 spin_lock(&dentry->d_lock);
2360 write_seqcount_begin(&dentry->d_seq);
2361 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2362 write_seqcount_end(&dentry->d_seq);
2363 spin_unlock(&dentry->d_lock);
2365 EXPORT_SYMBOL(dentry_update_name_case);
2367 static void switch_names(struct dentry *dentry, struct dentry *target)
2369 if (dname_external(target)) {
2370 if (dname_external(dentry)) {
2372 * Both external: swap the pointers
2374 swap(target->d_name.name, dentry->d_name.name);
2377 * dentry:internal, target:external. Steal target's
2378 * storage and make target internal.
2380 memcpy(target->d_iname, dentry->d_name.name,
2381 dentry->d_name.len + 1);
2382 dentry->d_name.name = target->d_name.name;
2383 target->d_name.name = target->d_iname;
2386 if (dname_external(dentry)) {
2388 * dentry:external, target:internal. Give dentry's
2389 * storage to target and make dentry internal
2391 memcpy(dentry->d_iname, target->d_name.name,
2392 target->d_name.len + 1);
2393 target->d_name.name = dentry->d_name.name;
2394 dentry->d_name.name = dentry->d_iname;
2397 * Both are internal.
2400 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2401 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2402 swap(((long *) &dentry->d_iname)[i],
2403 ((long *) &target->d_iname)[i]);
2407 swap(dentry->d_name.len, target->d_name.len);
2410 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2413 * XXXX: do we really need to take target->d_lock?
2415 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2416 spin_lock(&target->d_parent->d_lock);
2418 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2419 spin_lock(&dentry->d_parent->d_lock);
2420 spin_lock_nested(&target->d_parent->d_lock,
2421 DENTRY_D_LOCK_NESTED);
2423 spin_lock(&target->d_parent->d_lock);
2424 spin_lock_nested(&dentry->d_parent->d_lock,
2425 DENTRY_D_LOCK_NESTED);
2428 if (target < dentry) {
2429 spin_lock_nested(&target->d_lock, 2);
2430 spin_lock_nested(&dentry->d_lock, 3);
2432 spin_lock_nested(&dentry->d_lock, 2);
2433 spin_lock_nested(&target->d_lock, 3);
2437 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2438 struct dentry *target)
2440 if (target->d_parent != dentry->d_parent)
2441 spin_unlock(&dentry->d_parent->d_lock);
2442 if (target->d_parent != target)
2443 spin_unlock(&target->d_parent->d_lock);
2447 * When switching names, the actual string doesn't strictly have to
2448 * be preserved in the target - because we're dropping the target
2449 * anyway. As such, we can just do a simple memcpy() to copy over
2450 * the new name before we switch.
2452 * Note that we have to be a lot more careful about getting the hash
2453 * switched - we have to switch the hash value properly even if it
2454 * then no longer matches the actual (corrupted) string of the target.
2455 * The hash value has to match the hash queue that the dentry is on..
2458 * __d_move - move a dentry
2459 * @dentry: entry to move
2460 * @target: new dentry
2461 * @exchange: exchange the two dentries
2463 * Update the dcache to reflect the move of a file name. Negative
2464 * dcache entries should not be moved in this way. Caller must hold
2465 * rename_lock, the i_mutex of the source and target directories,
2466 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2468 static void __d_move(struct dentry *dentry, struct dentry *target,
2471 if (!dentry->d_inode)
2472 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2474 BUG_ON(d_ancestor(dentry, target));
2475 BUG_ON(d_ancestor(target, dentry));
2477 dentry_lock_for_move(dentry, target);
2479 write_seqcount_begin(&dentry->d_seq);
2480 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2482 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2485 * Move the dentry to the target hash queue. Don't bother checking
2486 * for the same hash queue because of how unlikely it is.
2489 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2492 * Unhash the target (d_delete() is not usable here). If exchanging
2493 * the two dentries, then rehash onto the other's hash queue.
2498 d_hash(dentry->d_parent, dentry->d_name.hash));
2501 list_del(&dentry->d_u.d_child);
2502 list_del(&target->d_u.d_child);
2504 /* Switch the names.. */
2505 switch_names(dentry, target);
2506 swap(dentry->d_name.hash, target->d_name.hash);
2508 /* ... and switch the parents */
2509 if (IS_ROOT(dentry)) {
2510 dentry->d_parent = target->d_parent;
2511 target->d_parent = target;
2512 INIT_LIST_HEAD(&target->d_u.d_child);
2514 swap(dentry->d_parent, target->d_parent);
2516 /* And add them back to the (new) parent lists */
2517 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2520 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2522 write_seqcount_end(&target->d_seq);
2523 write_seqcount_end(&dentry->d_seq);
2525 dentry_unlock_parents_for_move(dentry, target);
2527 fsnotify_d_move(target);
2528 spin_unlock(&target->d_lock);
2529 fsnotify_d_move(dentry);
2530 spin_unlock(&dentry->d_lock);
2534 * d_move - move a dentry
2535 * @dentry: entry to move
2536 * @target: new dentry
2538 * Update the dcache to reflect the move of a file name. Negative
2539 * dcache entries should not be moved in this way. See the locking
2540 * requirements for __d_move.
2542 void d_move(struct dentry *dentry, struct dentry *target)
2544 write_seqlock(&rename_lock);
2545 __d_move(dentry, target, false);
2546 write_sequnlock(&rename_lock);
2548 EXPORT_SYMBOL(d_move);
2551 * d_exchange - exchange two dentries
2552 * @dentry1: first dentry
2553 * @dentry2: second dentry
2555 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2557 write_seqlock(&rename_lock);
2559 WARN_ON(!dentry1->d_inode);
2560 WARN_ON(!dentry2->d_inode);
2561 WARN_ON(IS_ROOT(dentry1));
2562 WARN_ON(IS_ROOT(dentry2));
2564 __d_move(dentry1, dentry2, true);
2566 write_sequnlock(&rename_lock);
2570 * d_ancestor - search for an ancestor
2571 * @p1: ancestor dentry
2574 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2575 * an ancestor of p2, else NULL.
2577 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2581 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2582 if (p->d_parent == p1)
2589 * This helper attempts to cope with remotely renamed directories
2591 * It assumes that the caller is already holding
2592 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2594 * Note: If ever the locking in lock_rename() changes, then please
2595 * remember to update this too...
2597 static struct dentry *__d_unalias(struct inode *inode,
2598 struct dentry *dentry, struct dentry *alias)
2600 struct mutex *m1 = NULL, *m2 = NULL;
2601 struct dentry *ret = ERR_PTR(-EBUSY);
2603 /* If alias and dentry share a parent, then no extra locks required */
2604 if (alias->d_parent == dentry->d_parent)
2607 /* See lock_rename() */
2608 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2610 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2611 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2613 m2 = &alias->d_parent->d_inode->i_mutex;
2615 __d_move(alias, dentry, false);
2618 spin_unlock(&inode->i_lock);
2627 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2628 * named dentry in place of the dentry to be replaced.
2629 * returns with anon->d_lock held!
2631 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2633 struct dentry *dparent;
2635 dentry_lock_for_move(anon, dentry);
2637 write_seqcount_begin(&dentry->d_seq);
2638 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2640 dparent = dentry->d_parent;
2642 switch_names(anon, dentry);
2643 swap(dentry->d_name.hash, anon->d_name.hash);
2645 dentry->d_parent = dentry;
2646 list_del_init(&dentry->d_u.d_child);
2647 anon->d_parent = dparent;
2648 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2650 write_seqcount_end(&dentry->d_seq);
2651 write_seqcount_end(&anon->d_seq);
2653 dentry_unlock_parents_for_move(anon, dentry);
2654 spin_unlock(&dentry->d_lock);
2656 /* anon->d_lock still locked, returns locked */
2660 * d_materialise_unique - introduce an inode into the tree
2661 * @dentry: candidate dentry
2662 * @inode: inode to bind to the dentry, to which aliases may be attached
2664 * Introduces an dentry into the tree, substituting an extant disconnected
2665 * root directory alias in its place if there is one. Caller must hold the
2666 * i_mutex of the parent directory.
2668 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2670 struct dentry *actual;
2672 BUG_ON(!d_unhashed(dentry));
2676 __d_instantiate(dentry, NULL);
2681 spin_lock(&inode->i_lock);
2683 if (S_ISDIR(inode->i_mode)) {
2684 struct dentry *alias;
2686 /* Does an aliased dentry already exist? */
2687 alias = __d_find_alias(inode, 0);
2690 write_seqlock(&rename_lock);
2692 if (d_ancestor(alias, dentry)) {
2693 /* Check for loops */
2694 actual = ERR_PTR(-ELOOP);
2695 spin_unlock(&inode->i_lock);
2696 } else if (IS_ROOT(alias)) {
2697 /* Is this an anonymous mountpoint that we
2698 * could splice into our tree? */
2699 __d_materialise_dentry(dentry, alias);
2700 write_sequnlock(&rename_lock);
2704 /* Nope, but we must(!) avoid directory
2705 * aliasing. This drops inode->i_lock */
2706 actual = __d_unalias(inode, dentry, alias);
2708 write_sequnlock(&rename_lock);
2709 if (IS_ERR(actual)) {
2710 if (PTR_ERR(actual) == -ELOOP)
2711 pr_warn_ratelimited(
2712 "VFS: Lookup of '%s' in %s %s"
2713 " would have caused loop\n",
2714 dentry->d_name.name,
2715 inode->i_sb->s_type->name,
2723 /* Add a unique reference */
2724 actual = __d_instantiate_unique(dentry, inode);
2728 BUG_ON(!d_unhashed(actual));
2730 spin_lock(&actual->d_lock);
2733 spin_unlock(&actual->d_lock);
2734 spin_unlock(&inode->i_lock);
2736 if (actual == dentry) {
2737 security_d_instantiate(dentry, inode);
2744 EXPORT_SYMBOL_GPL(d_materialise_unique);
2746 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2750 return -ENAMETOOLONG;
2752 memcpy(*buffer, str, namelen);
2757 * prepend_name - prepend a pathname in front of current buffer pointer
2758 * @buffer: buffer pointer
2759 * @buflen: allocated length of the buffer
2760 * @name: name string and length qstr structure
2762 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2763 * make sure that either the old or the new name pointer and length are
2764 * fetched. However, there may be mismatch between length and pointer.
2765 * The length cannot be trusted, we need to copy it byte-by-byte until
2766 * the length is reached or a null byte is found. It also prepends "/" at
2767 * the beginning of the name. The sequence number check at the caller will
2768 * retry it again when a d_move() does happen. So any garbage in the buffer
2769 * due to mismatched pointer and length will be discarded.
2771 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2773 const char *dname = ACCESS_ONCE(name->name);
2774 u32 dlen = ACCESS_ONCE(name->len);
2777 *buflen -= dlen + 1;
2779 return -ENAMETOOLONG;
2780 p = *buffer -= dlen + 1;
2792 * prepend_path - Prepend path string to a buffer
2793 * @path: the dentry/vfsmount to report
2794 * @root: root vfsmnt/dentry
2795 * @buffer: pointer to the end of the buffer
2796 * @buflen: pointer to buffer length
2798 * The function will first try to write out the pathname without taking any
2799 * lock other than the RCU read lock to make sure that dentries won't go away.
2800 * It only checks the sequence number of the global rename_lock as any change
2801 * in the dentry's d_seq will be preceded by changes in the rename_lock
2802 * sequence number. If the sequence number had been changed, it will restart
2803 * the whole pathname back-tracing sequence again by taking the rename_lock.
2804 * In this case, there is no need to take the RCU read lock as the recursive
2805 * parent pointer references will keep the dentry chain alive as long as no
2806 * rename operation is performed.
2808 static int prepend_path(const struct path *path,
2809 const struct path *root,
2810 char **buffer, int *buflen)
2812 struct dentry *dentry;
2813 struct vfsmount *vfsmnt;
2816 unsigned seq, m_seq = 0;
2822 read_seqbegin_or_lock(&mount_lock, &m_seq);
2829 dentry = path->dentry;
2831 mnt = real_mount(vfsmnt);
2832 read_seqbegin_or_lock(&rename_lock, &seq);
2833 while (dentry != root->dentry || vfsmnt != root->mnt) {
2834 struct dentry * parent;
2836 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2837 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2839 if (mnt != parent) {
2840 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2846 * Filesystems needing to implement special "root names"
2847 * should do so with ->d_dname()
2849 if (IS_ROOT(dentry) &&
2850 (dentry->d_name.len != 1 ||
2851 dentry->d_name.name[0] != '/')) {
2852 WARN(1, "Root dentry has weird name <%.*s>\n",
2853 (int) dentry->d_name.len,
2854 dentry->d_name.name);
2857 error = is_mounted(vfsmnt) ? 1 : 2;
2860 parent = dentry->d_parent;
2862 error = prepend_name(&bptr, &blen, &dentry->d_name);
2870 if (need_seqretry(&rename_lock, seq)) {
2874 done_seqretry(&rename_lock, seq);
2878 if (need_seqretry(&mount_lock, m_seq)) {
2882 done_seqretry(&mount_lock, m_seq);
2884 if (error >= 0 && bptr == *buffer) {
2886 error = -ENAMETOOLONG;
2896 * __d_path - return the path of a dentry
2897 * @path: the dentry/vfsmount to report
2898 * @root: root vfsmnt/dentry
2899 * @buf: buffer to return value in
2900 * @buflen: buffer length
2902 * Convert a dentry into an ASCII path name.
2904 * Returns a pointer into the buffer or an error code if the
2905 * path was too long.
2907 * "buflen" should be positive.
2909 * If the path is not reachable from the supplied root, return %NULL.
2911 char *__d_path(const struct path *path,
2912 const struct path *root,
2913 char *buf, int buflen)
2915 char *res = buf + buflen;
2918 prepend(&res, &buflen, "\0", 1);
2919 error = prepend_path(path, root, &res, &buflen);
2922 return ERR_PTR(error);
2928 char *d_absolute_path(const struct path *path,
2929 char *buf, int buflen)
2931 struct path root = {};
2932 char *res = buf + buflen;
2935 prepend(&res, &buflen, "\0", 1);
2936 error = prepend_path(path, &root, &res, &buflen);
2941 return ERR_PTR(error);
2946 * same as __d_path but appends "(deleted)" for unlinked files.
2948 static int path_with_deleted(const struct path *path,
2949 const struct path *root,
2950 char **buf, int *buflen)
2952 prepend(buf, buflen, "\0", 1);
2953 if (d_unlinked(path->dentry)) {
2954 int error = prepend(buf, buflen, " (deleted)", 10);
2959 return prepend_path(path, root, buf, buflen);
2962 static int prepend_unreachable(char **buffer, int *buflen)
2964 return prepend(buffer, buflen, "(unreachable)", 13);
2967 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2972 seq = read_seqcount_begin(&fs->seq);
2974 } while (read_seqcount_retry(&fs->seq, seq));
2978 * d_path - return the path of a dentry
2979 * @path: path to report
2980 * @buf: buffer to return value in
2981 * @buflen: buffer length
2983 * Convert a dentry into an ASCII path name. If the entry has been deleted
2984 * the string " (deleted)" is appended. Note that this is ambiguous.
2986 * Returns a pointer into the buffer or an error code if the path was
2987 * too long. Note: Callers should use the returned pointer, not the passed
2988 * in buffer, to use the name! The implementation often starts at an offset
2989 * into the buffer, and may leave 0 bytes at the start.
2991 * "buflen" should be positive.
2993 char *d_path(const struct path *path, char *buf, int buflen)
2995 char *res = buf + buflen;
3000 * We have various synthetic filesystems that never get mounted. On
3001 * these filesystems dentries are never used for lookup purposes, and
3002 * thus don't need to be hashed. They also don't need a name until a
3003 * user wants to identify the object in /proc/pid/fd/. The little hack
3004 * below allows us to generate a name for these objects on demand:
3006 * Some pseudo inodes are mountable. When they are mounted
3007 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3008 * and instead have d_path return the mounted path.
3010 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3011 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3012 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3015 get_fs_root_rcu(current->fs, &root);
3016 error = path_with_deleted(path, &root, &res, &buflen);
3020 res = ERR_PTR(error);
3023 EXPORT_SYMBOL(d_path);
3026 * Helper function for dentry_operations.d_dname() members
3028 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3029 const char *fmt, ...)
3035 va_start(args, fmt);
3036 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3039 if (sz > sizeof(temp) || sz > buflen)
3040 return ERR_PTR(-ENAMETOOLONG);
3042 buffer += buflen - sz;
3043 return memcpy(buffer, temp, sz);
3046 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3048 char *end = buffer + buflen;
3049 /* these dentries are never renamed, so d_lock is not needed */
3050 if (prepend(&end, &buflen, " (deleted)", 11) ||
3051 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3052 prepend(&end, &buflen, "/", 1))
3053 end = ERR_PTR(-ENAMETOOLONG);
3056 EXPORT_SYMBOL(simple_dname);
3059 * Write full pathname from the root of the filesystem into the buffer.
3061 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3063 struct dentry *dentry;
3076 prepend(&end, &len, "\0", 1);
3080 read_seqbegin_or_lock(&rename_lock, &seq);
3081 while (!IS_ROOT(dentry)) {
3082 struct dentry *parent = dentry->d_parent;
3085 error = prepend_name(&end, &len, &dentry->d_name);
3094 if (need_seqretry(&rename_lock, seq)) {
3098 done_seqretry(&rename_lock, seq);
3103 return ERR_PTR(-ENAMETOOLONG);
3106 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3108 return __dentry_path(dentry, buf, buflen);
3110 EXPORT_SYMBOL(dentry_path_raw);
3112 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3117 if (d_unlinked(dentry)) {
3119 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3123 retval = __dentry_path(dentry, buf, buflen);
3124 if (!IS_ERR(retval) && p)
3125 *p = '/'; /* restore '/' overriden with '\0' */
3128 return ERR_PTR(-ENAMETOOLONG);
3131 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3137 seq = read_seqcount_begin(&fs->seq);
3140 } while (read_seqcount_retry(&fs->seq, seq));
3144 * NOTE! The user-level library version returns a
3145 * character pointer. The kernel system call just
3146 * returns the length of the buffer filled (which
3147 * includes the ending '\0' character), or a negative
3148 * error value. So libc would do something like
3150 * char *getcwd(char * buf, size_t size)
3154 * retval = sys_getcwd(buf, size);
3161 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3164 struct path pwd, root;
3165 char *page = __getname();
3171 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3174 if (!d_unlinked(pwd.dentry)) {
3176 char *cwd = page + PATH_MAX;
3177 int buflen = PATH_MAX;
3179 prepend(&cwd, &buflen, "\0", 1);
3180 error = prepend_path(&pwd, &root, &cwd, &buflen);
3186 /* Unreachable from current root */
3188 error = prepend_unreachable(&cwd, &buflen);
3194 len = PATH_MAX + page - cwd;
3197 if (copy_to_user(buf, cwd, len))
3210 * Test whether new_dentry is a subdirectory of old_dentry.
3212 * Trivially implemented using the dcache structure
3216 * is_subdir - is new dentry a subdirectory of old_dentry
3217 * @new_dentry: new dentry
3218 * @old_dentry: old dentry
3220 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3221 * Returns 0 otherwise.
3222 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3225 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3230 if (new_dentry == old_dentry)
3234 /* for restarting inner loop in case of seq retry */
3235 seq = read_seqbegin(&rename_lock);
3237 * Need rcu_readlock to protect against the d_parent trashing
3241 if (d_ancestor(old_dentry, new_dentry))
3246 } while (read_seqretry(&rename_lock, seq));
3251 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3253 struct dentry *root = data;
3254 if (dentry != root) {
3255 if (d_unhashed(dentry) || !dentry->d_inode)
3258 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3259 dentry->d_flags |= DCACHE_GENOCIDE;
3260 dentry->d_lockref.count--;
3263 return D_WALK_CONTINUE;
3266 void d_genocide(struct dentry *parent)
3268 d_walk(parent, parent, d_genocide_kill, NULL);
3271 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3273 inode_dec_link_count(inode);
3274 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3275 !hlist_unhashed(&dentry->d_alias) ||
3276 !d_unlinked(dentry));
3277 spin_lock(&dentry->d_parent->d_lock);
3278 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3279 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3280 (unsigned long long)inode->i_ino);
3281 spin_unlock(&dentry->d_lock);
3282 spin_unlock(&dentry->d_parent->d_lock);
3283 d_instantiate(dentry, inode);
3285 EXPORT_SYMBOL(d_tmpfile);
3287 static __initdata unsigned long dhash_entries;
3288 static int __init set_dhash_entries(char *str)
3292 dhash_entries = simple_strtoul(str, &str, 0);
3295 __setup("dhash_entries=", set_dhash_entries);
3297 static void __init dcache_init_early(void)
3301 /* If hashes are distributed across NUMA nodes, defer
3302 * hash allocation until vmalloc space is available.
3308 alloc_large_system_hash("Dentry cache",
3309 sizeof(struct hlist_bl_head),
3318 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3319 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3322 static void __init dcache_init(void)
3327 * A constructor could be added for stable state like the lists,
3328 * but it is probably not worth it because of the cache nature
3331 dentry_cache = KMEM_CACHE(dentry,
3332 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3334 /* Hash may have been set up in dcache_init_early */
3339 alloc_large_system_hash("Dentry cache",
3340 sizeof(struct hlist_bl_head),
3349 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3350 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3353 /* SLAB cache for __getname() consumers */
3354 struct kmem_cache *names_cachep __read_mostly;
3355 EXPORT_SYMBOL(names_cachep);
3357 EXPORT_SYMBOL(d_genocide);
3359 void __init vfs_caches_init_early(void)
3361 dcache_init_early();
3365 void __init vfs_caches_init(unsigned long mempages)
3367 unsigned long reserve;
3369 /* Base hash sizes on available memory, with a reserve equal to
3370 150% of current kernel size */
3372 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3373 mempages -= reserve;
3375 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3376 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3380 files_init(mempages);