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(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 * Remove a dentry with references from the LRU.
401 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
402 * lose our last reference through the parent walk. In this case, we need to
403 * remove ourselves from the shrink list, not the LRU.
405 static void dentry_lru_del(struct dentry *dentry)
407 if (dentry->d_flags & DCACHE_LRU_LIST) {
408 if (dentry->d_flags & DCACHE_SHRINK_LIST)
409 return d_shrink_del(dentry);
415 * d_drop - drop a dentry
416 * @dentry: dentry to drop
418 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
419 * be found through a VFS lookup any more. Note that this is different from
420 * deleting the dentry - d_delete will try to mark the dentry negative if
421 * possible, giving a successful _negative_ lookup, while d_drop will
422 * just make the cache lookup fail.
424 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
425 * reason (NFS timeouts or autofs deletes).
427 * __d_drop requires dentry->d_lock.
429 void __d_drop(struct dentry *dentry)
431 if (!d_unhashed(dentry)) {
432 struct hlist_bl_head *b;
434 * Hashed dentries are normally on the dentry hashtable,
435 * with the exception of those newly allocated by
436 * d_obtain_alias, which are always IS_ROOT:
438 if (unlikely(IS_ROOT(dentry)))
439 b = &dentry->d_sb->s_anon;
441 b = d_hash(dentry->d_parent, dentry->d_name.hash);
444 __hlist_bl_del(&dentry->d_hash);
445 dentry->d_hash.pprev = NULL;
447 dentry_rcuwalk_barrier(dentry);
450 EXPORT_SYMBOL(__d_drop);
452 void d_drop(struct dentry *dentry)
454 spin_lock(&dentry->d_lock);
456 spin_unlock(&dentry->d_lock);
458 EXPORT_SYMBOL(d_drop);
461 * Finish off a dentry we've decided to kill.
462 * dentry->d_lock must be held, returns with it unlocked.
463 * If ref is non-zero, then decrement the refcount too.
464 * Returns dentry requiring refcount drop, or NULL if we're done.
466 static struct dentry *
467 dentry_kill(struct dentry *dentry, int unlock_on_failure)
468 __releases(dentry->d_lock)
471 struct dentry *parent;
473 inode = dentry->d_inode;
474 if (inode && !spin_trylock(&inode->i_lock)) {
476 if (unlock_on_failure) {
477 spin_unlock(&dentry->d_lock);
480 return dentry; /* try again with same dentry */
485 parent = dentry->d_parent;
486 if (parent && !spin_trylock(&parent->d_lock)) {
488 spin_unlock(&inode->i_lock);
493 * The dentry is now unrecoverably dead to the world.
495 lockref_mark_dead(&dentry->d_lockref);
498 * inform the fs via d_prune that this dentry is about to be
499 * unhashed and destroyed.
501 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
502 dentry->d_op->d_prune(dentry);
504 if (dentry->d_flags & DCACHE_LRU_LIST) {
505 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
508 d_shrink_del(dentry);
510 /* if it was on the hash then remove it */
512 list_del(&dentry->d_u.d_child);
514 * Inform d_walk() that we are no longer attached to the
517 dentry->d_flags |= DCACHE_DENTRY_KILLED;
519 spin_unlock(&parent->d_lock);
522 * dentry_iput drops the locks, at which point nobody (except
523 * transient RCU lookups) can reach this dentry.
525 BUG_ON((int)dentry->d_lockref.count > 0);
526 this_cpu_dec(nr_dentry);
527 if (dentry->d_op && dentry->d_op->d_release)
528 dentry->d_op->d_release(dentry);
537 * This is complicated by the fact that we do not want to put
538 * dentries that are no longer on any hash chain on the unused
539 * list: we'd much rather just get rid of them immediately.
541 * However, that implies that we have to traverse the dentry
542 * tree upwards to the parents which might _also_ now be
543 * scheduled for deletion (it may have been only waiting for
544 * its last child to go away).
546 * This tail recursion is done by hand as we don't want to depend
547 * on the compiler to always get this right (gcc generally doesn't).
548 * Real recursion would eat up our stack space.
552 * dput - release a dentry
553 * @dentry: dentry to release
555 * Release a dentry. This will drop the usage count and if appropriate
556 * call the dentry unlink method as well as removing it from the queues and
557 * releasing its resources. If the parent dentries were scheduled for release
558 * they too may now get deleted.
560 void dput(struct dentry *dentry)
562 if (unlikely(!dentry))
566 if (lockref_put_or_lock(&dentry->d_lockref))
569 /* Unreachable? Get rid of it */
570 if (unlikely(d_unhashed(dentry)))
573 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
574 if (dentry->d_op->d_delete(dentry))
578 if (!(dentry->d_flags & DCACHE_REFERENCED))
579 dentry->d_flags |= DCACHE_REFERENCED;
580 dentry_lru_add(dentry);
582 dentry->d_lockref.count--;
583 spin_unlock(&dentry->d_lock);
587 dentry = dentry_kill(dentry, 1);
594 * d_invalidate - invalidate a dentry
595 * @dentry: dentry to invalidate
597 * Try to invalidate the dentry if it turns out to be
598 * possible. If there are other dentries that can be
599 * reached through this one we can't delete it and we
600 * return -EBUSY. On success we return 0.
605 int d_invalidate(struct dentry * dentry)
608 * If it's already been dropped, return OK.
610 spin_lock(&dentry->d_lock);
611 if (d_unhashed(dentry)) {
612 spin_unlock(&dentry->d_lock);
616 * Check whether to do a partial shrink_dcache
617 * to get rid of unused child entries.
619 if (!list_empty(&dentry->d_subdirs)) {
620 spin_unlock(&dentry->d_lock);
621 shrink_dcache_parent(dentry);
622 spin_lock(&dentry->d_lock);
626 * Somebody else still using it?
628 * If it's a directory, we can't drop it
629 * for fear of somebody re-populating it
630 * with children (even though dropping it
631 * would make it unreachable from the root,
632 * we might still populate it if it was a
633 * working directory or similar).
634 * We also need to leave mountpoints alone,
637 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
638 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
639 spin_unlock(&dentry->d_lock);
645 spin_unlock(&dentry->d_lock);
648 EXPORT_SYMBOL(d_invalidate);
650 /* This must be called with d_lock held */
651 static inline void __dget_dlock(struct dentry *dentry)
653 dentry->d_lockref.count++;
656 static inline void __dget(struct dentry *dentry)
658 lockref_get(&dentry->d_lockref);
661 struct dentry *dget_parent(struct dentry *dentry)
667 * Do optimistic parent lookup without any
671 ret = ACCESS_ONCE(dentry->d_parent);
672 gotref = lockref_get_not_zero(&ret->d_lockref);
674 if (likely(gotref)) {
675 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
682 * Don't need rcu_dereference because we re-check it was correct under
686 ret = dentry->d_parent;
687 spin_lock(&ret->d_lock);
688 if (unlikely(ret != dentry->d_parent)) {
689 spin_unlock(&ret->d_lock);
694 BUG_ON(!ret->d_lockref.count);
695 ret->d_lockref.count++;
696 spin_unlock(&ret->d_lock);
699 EXPORT_SYMBOL(dget_parent);
702 * d_find_alias - grab a hashed alias of inode
703 * @inode: inode in question
704 * @want_discon: flag, used by d_splice_alias, to request
705 * that only a DISCONNECTED alias be returned.
707 * If inode has a hashed alias, or is a directory and has any alias,
708 * acquire the reference to alias and return it. Otherwise return NULL.
709 * Notice that if inode is a directory there can be only one alias and
710 * it can be unhashed only if it has no children, or if it is the root
713 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
714 * any other hashed alias over that one unless @want_discon is set,
715 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
717 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
719 struct dentry *alias, *discon_alias;
723 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
724 spin_lock(&alias->d_lock);
725 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
726 if (IS_ROOT(alias) &&
727 (alias->d_flags & DCACHE_DISCONNECTED)) {
728 discon_alias = alias;
729 } else if (!want_discon) {
731 spin_unlock(&alias->d_lock);
735 spin_unlock(&alias->d_lock);
738 alias = discon_alias;
739 spin_lock(&alias->d_lock);
740 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
741 if (IS_ROOT(alias) &&
742 (alias->d_flags & DCACHE_DISCONNECTED)) {
744 spin_unlock(&alias->d_lock);
748 spin_unlock(&alias->d_lock);
754 struct dentry *d_find_alias(struct inode *inode)
756 struct dentry *de = NULL;
758 if (!hlist_empty(&inode->i_dentry)) {
759 spin_lock(&inode->i_lock);
760 de = __d_find_alias(inode, 0);
761 spin_unlock(&inode->i_lock);
765 EXPORT_SYMBOL(d_find_alias);
768 * Try to kill dentries associated with this inode.
769 * WARNING: you must own a reference to inode.
771 void d_prune_aliases(struct inode *inode)
773 struct dentry *dentry;
775 spin_lock(&inode->i_lock);
776 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
777 spin_lock(&dentry->d_lock);
778 if (!dentry->d_lockref.count) {
780 * inform the fs via d_prune that this dentry
781 * is about to be unhashed and destroyed.
783 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
785 dentry->d_op->d_prune(dentry);
787 __dget_dlock(dentry);
789 spin_unlock(&dentry->d_lock);
790 spin_unlock(&inode->i_lock);
794 spin_unlock(&dentry->d_lock);
796 spin_unlock(&inode->i_lock);
798 EXPORT_SYMBOL(d_prune_aliases);
800 static void shrink_dentry_list(struct list_head *list)
802 struct dentry *dentry, *parent;
806 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
807 if (&dentry->d_lru == list)
811 * Get the dentry lock, and re-verify that the dentry is
812 * this on the shrinking list. If it is, we know that
813 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
815 spin_lock(&dentry->d_lock);
816 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
817 spin_unlock(&dentry->d_lock);
822 * The dispose list is isolated and dentries are not accounted
823 * to the LRU here, so we can simply remove it from the list
824 * here regardless of whether it is referenced or not.
826 d_shrink_del(dentry);
829 * We found an inuse dentry which was not removed from
830 * the LRU because of laziness during lookup. Do not free it.
832 if (dentry->d_lockref.count) {
833 spin_unlock(&dentry->d_lock);
838 parent = dentry_kill(dentry, 0);
840 * If dentry_kill returns NULL, we have nothing more to do.
846 if (unlikely(parent == dentry)) {
848 * trylocks have failed and d_lock has been held the
849 * whole time, so it could not have been added to any
850 * other lists. Just add it back to the shrink list.
853 d_shrink_add(dentry, list);
854 spin_unlock(&dentry->d_lock);
858 * We need to prune ancestors too. This is necessary to prevent
859 * quadratic behavior of shrink_dcache_parent(), but is also
860 * expected to be beneficial in reducing dentry cache
864 while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
865 dentry = dentry_kill(dentry, 1);
871 static enum lru_status
872 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
874 struct list_head *freeable = arg;
875 struct dentry *dentry = container_of(item, struct dentry, d_lru);
879 * we are inverting the lru lock/dentry->d_lock here,
880 * so use a trylock. If we fail to get the lock, just skip
883 if (!spin_trylock(&dentry->d_lock))
887 * Referenced dentries are still in use. If they have active
888 * counts, just remove them from the LRU. Otherwise give them
889 * another pass through the LRU.
891 if (dentry->d_lockref.count) {
892 d_lru_isolate(dentry);
893 spin_unlock(&dentry->d_lock);
897 if (dentry->d_flags & DCACHE_REFERENCED) {
898 dentry->d_flags &= ~DCACHE_REFERENCED;
899 spin_unlock(&dentry->d_lock);
902 * The list move itself will be made by the common LRU code. At
903 * this point, we've dropped the dentry->d_lock but keep the
904 * lru lock. This is safe to do, since every list movement is
905 * protected by the lru lock even if both locks are held.
907 * This is guaranteed by the fact that all LRU management
908 * functions are intermediated by the LRU API calls like
909 * list_lru_add and list_lru_del. List movement in this file
910 * only ever occur through this functions or through callbacks
911 * like this one, that are called from the LRU API.
913 * The only exceptions to this are functions like
914 * shrink_dentry_list, and code that first checks for the
915 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
916 * operating only with stack provided lists after they are
917 * properly isolated from the main list. It is thus, always a
923 d_lru_shrink_move(dentry, freeable);
924 spin_unlock(&dentry->d_lock);
930 * prune_dcache_sb - shrink the dcache
932 * @nr_to_scan : number of entries to try to free
933 * @nid: which node to scan for freeable entities
935 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
936 * done when we need more memory an called from the superblock shrinker
939 * This function may fail to free any resources if all the dentries are in
942 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
948 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
949 &dispose, &nr_to_scan);
950 shrink_dentry_list(&dispose);
954 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
955 spinlock_t *lru_lock, void *arg)
957 struct list_head *freeable = arg;
958 struct dentry *dentry = container_of(item, struct dentry, d_lru);
961 * we are inverting the lru lock/dentry->d_lock here,
962 * so use a trylock. If we fail to get the lock, just skip
965 if (!spin_trylock(&dentry->d_lock))
968 d_lru_shrink_move(dentry, freeable);
969 spin_unlock(&dentry->d_lock);
976 * shrink_dcache_sb - shrink dcache for a superblock
979 * Shrink the dcache for the specified super block. This is used to free
980 * the dcache before unmounting a file system.
982 void shrink_dcache_sb(struct super_block *sb)
989 freed = list_lru_walk(&sb->s_dentry_lru,
990 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
992 this_cpu_sub(nr_dentry_unused, freed);
993 shrink_dentry_list(&dispose);
996 EXPORT_SYMBOL(shrink_dcache_sb);
999 * enum d_walk_ret - action to talke during tree walk
1000 * @D_WALK_CONTINUE: contrinue walk
1001 * @D_WALK_QUIT: quit walk
1002 * @D_WALK_NORETRY: quit when retry is needed
1003 * @D_WALK_SKIP: skip this dentry and its children
1013 * d_walk - walk the dentry tree
1014 * @parent: start of walk
1015 * @data: data passed to @enter() and @finish()
1016 * @enter: callback when first entering the dentry
1017 * @finish: callback when successfully finished the walk
1019 * The @enter() and @finish() callbacks are called with d_lock held.
1021 static void d_walk(struct dentry *parent, void *data,
1022 enum d_walk_ret (*enter)(void *, struct dentry *),
1023 void (*finish)(void *))
1025 struct dentry *this_parent;
1026 struct list_head *next;
1028 enum d_walk_ret ret;
1032 read_seqbegin_or_lock(&rename_lock, &seq);
1033 this_parent = parent;
1034 spin_lock(&this_parent->d_lock);
1036 ret = enter(data, this_parent);
1038 case D_WALK_CONTINUE:
1043 case D_WALK_NORETRY:
1048 next = this_parent->d_subdirs.next;
1050 while (next != &this_parent->d_subdirs) {
1051 struct list_head *tmp = next;
1052 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1055 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1057 ret = enter(data, dentry);
1059 case D_WALK_CONTINUE:
1062 spin_unlock(&dentry->d_lock);
1064 case D_WALK_NORETRY:
1068 spin_unlock(&dentry->d_lock);
1072 if (!list_empty(&dentry->d_subdirs)) {
1073 spin_unlock(&this_parent->d_lock);
1074 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1075 this_parent = dentry;
1076 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1079 spin_unlock(&dentry->d_lock);
1082 * All done at this level ... ascend and resume the search.
1084 if (this_parent != parent) {
1085 struct dentry *child = this_parent;
1086 this_parent = child->d_parent;
1089 spin_unlock(&child->d_lock);
1090 spin_lock(&this_parent->d_lock);
1093 * might go back up the wrong parent if we have had a rename
1096 if (this_parent != child->d_parent ||
1097 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1098 need_seqretry(&rename_lock, seq)) {
1099 spin_unlock(&this_parent->d_lock);
1104 next = child->d_u.d_child.next;
1107 if (need_seqretry(&rename_lock, seq)) {
1108 spin_unlock(&this_parent->d_lock);
1115 spin_unlock(&this_parent->d_lock);
1116 done_seqretry(&rename_lock, seq);
1127 * Search for at least 1 mount point in the dentry's subdirs.
1128 * We descend to the next level whenever the d_subdirs
1129 * list is non-empty and continue searching.
1132 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1135 if (d_mountpoint(dentry)) {
1139 return D_WALK_CONTINUE;
1143 * have_submounts - check for mounts over a dentry
1144 * @parent: dentry to check.
1146 * Return true if the parent or its subdirectories contain
1149 int have_submounts(struct dentry *parent)
1153 d_walk(parent, &ret, check_mount, NULL);
1157 EXPORT_SYMBOL(have_submounts);
1160 * Called by mount code to set a mountpoint and check if the mountpoint is
1161 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1162 * subtree can become unreachable).
1164 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1165 * this reason take rename_lock and d_lock on dentry and ancestors.
1167 int d_set_mounted(struct dentry *dentry)
1171 write_seqlock(&rename_lock);
1172 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1173 /* Need exclusion wrt. check_submounts_and_drop() */
1174 spin_lock(&p->d_lock);
1175 if (unlikely(d_unhashed(p))) {
1176 spin_unlock(&p->d_lock);
1179 spin_unlock(&p->d_lock);
1181 spin_lock(&dentry->d_lock);
1182 if (!d_unlinked(dentry)) {
1183 dentry->d_flags |= DCACHE_MOUNTED;
1186 spin_unlock(&dentry->d_lock);
1188 write_sequnlock(&rename_lock);
1193 * Search the dentry child list of the specified parent,
1194 * and move any unused dentries to the end of the unused
1195 * list for prune_dcache(). We descend to the next level
1196 * whenever the d_subdirs list is non-empty and continue
1199 * It returns zero iff there are no unused children,
1200 * otherwise it returns the number of children moved to
1201 * the end of the unused list. This may not be the total
1202 * number of unused children, because select_parent can
1203 * drop the lock and return early due to latency
1207 struct select_data {
1208 struct dentry *start;
1209 struct list_head dispose;
1213 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1215 struct select_data *data = _data;
1216 enum d_walk_ret ret = D_WALK_CONTINUE;
1218 if (data->start == dentry)
1222 * move only zero ref count dentries to the dispose list.
1224 * Those which are presently on the shrink list, being processed
1225 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1226 * loop in shrink_dcache_parent() might not make any progress
1229 if (dentry->d_lockref.count) {
1230 dentry_lru_del(dentry);
1231 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1233 * We can't use d_lru_shrink_move() because we
1234 * need to get the global LRU lock and do the
1238 d_shrink_add(dentry, &data->dispose);
1240 ret = D_WALK_NORETRY;
1243 * We can return to the caller if we have found some (this
1244 * ensures forward progress). We'll be coming back to find
1247 if (data->found && need_resched())
1254 * shrink_dcache_parent - prune dcache
1255 * @parent: parent of entries to prune
1257 * Prune the dcache to remove unused children of the parent dentry.
1259 void shrink_dcache_parent(struct dentry *parent)
1262 struct select_data data;
1264 INIT_LIST_HEAD(&data.dispose);
1265 data.start = parent;
1268 d_walk(parent, &data, select_collect, NULL);
1272 shrink_dentry_list(&data.dispose);
1276 EXPORT_SYMBOL(shrink_dcache_parent);
1278 static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry)
1280 struct select_data *data = _data;
1281 enum d_walk_ret ret = D_WALK_CONTINUE;
1283 if (dentry->d_lockref.count) {
1284 dentry_lru_del(dentry);
1285 if (likely(!list_empty(&dentry->d_subdirs)))
1287 if (dentry == data->start && dentry->d_lockref.count == 1)
1290 "BUG: Dentry %p{i=%lx,n=%s}"
1291 " still in use (%d)"
1292 " [unmount of %s %s]\n",
1295 dentry->d_inode->i_ino : 0UL,
1296 dentry->d_name.name,
1297 dentry->d_lockref.count,
1298 dentry->d_sb->s_type->name,
1299 dentry->d_sb->s_id);
1301 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1303 * We can't use d_lru_shrink_move() because we
1304 * need to get the global LRU lock and do the
1307 if (dentry->d_flags & DCACHE_LRU_LIST)
1309 d_shrink_add(dentry, &data->dispose);
1311 ret = D_WALK_NORETRY;
1314 if (data->found && need_resched())
1320 * destroy the dentries attached to a superblock on unmounting
1322 void shrink_dcache_for_umount(struct super_block *sb)
1324 struct dentry *dentry;
1326 if (down_read_trylock(&sb->s_umount))
1329 dentry = sb->s_root;
1332 struct select_data data;
1334 INIT_LIST_HEAD(&data.dispose);
1335 data.start = dentry;
1338 d_walk(dentry, &data, umount_collect, NULL);
1342 shrink_dentry_list(&data.dispose);
1348 while (!hlist_bl_empty(&sb->s_anon)) {
1349 struct select_data data;
1350 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1352 INIT_LIST_HEAD(&data.dispose);
1356 d_walk(dentry, &data, umount_collect, NULL);
1358 shrink_dentry_list(&data.dispose);
1363 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1365 struct select_data *data = _data;
1367 if (d_mountpoint(dentry)) {
1368 data->found = -EBUSY;
1372 return select_collect(_data, dentry);
1375 static void check_and_drop(void *_data)
1377 struct select_data *data = _data;
1379 if (d_mountpoint(data->start))
1380 data->found = -EBUSY;
1382 __d_drop(data->start);
1386 * check_submounts_and_drop - prune dcache, check for submounts and drop
1388 * All done as a single atomic operation relative to has_unlinked_ancestor().
1389 * Returns 0 if successfully unhashed @parent. If there were submounts then
1392 * @dentry: dentry to prune and drop
1394 int check_submounts_and_drop(struct dentry *dentry)
1398 /* Negative dentries can be dropped without further checks */
1399 if (!dentry->d_inode) {
1405 struct select_data data;
1407 INIT_LIST_HEAD(&data.dispose);
1408 data.start = dentry;
1411 d_walk(dentry, &data, check_and_collect, check_and_drop);
1414 if (!list_empty(&data.dispose))
1415 shrink_dentry_list(&data.dispose);
1426 EXPORT_SYMBOL(check_submounts_and_drop);
1429 * __d_alloc - allocate a dcache entry
1430 * @sb: filesystem it will belong to
1431 * @name: qstr of the name
1433 * Allocates a dentry. It returns %NULL if there is insufficient memory
1434 * available. On a success the dentry is returned. The name passed in is
1435 * copied and the copy passed in may be reused after this call.
1438 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1440 struct dentry *dentry;
1443 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1448 * We guarantee that the inline name is always NUL-terminated.
1449 * This way the memcpy() done by the name switching in rename
1450 * will still always have a NUL at the end, even if we might
1451 * be overwriting an internal NUL character
1453 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1454 if (name->len > DNAME_INLINE_LEN-1) {
1455 dname = kmalloc(name->len + 1, GFP_KERNEL);
1457 kmem_cache_free(dentry_cache, dentry);
1461 dname = dentry->d_iname;
1464 dentry->d_name.len = name->len;
1465 dentry->d_name.hash = name->hash;
1466 memcpy(dname, name->name, name->len);
1467 dname[name->len] = 0;
1469 /* Make sure we always see the terminating NUL character */
1471 dentry->d_name.name = dname;
1473 dentry->d_lockref.count = 1;
1474 dentry->d_flags = 0;
1475 spin_lock_init(&dentry->d_lock);
1476 seqcount_init(&dentry->d_seq);
1477 dentry->d_inode = NULL;
1478 dentry->d_parent = dentry;
1480 dentry->d_op = NULL;
1481 dentry->d_fsdata = NULL;
1482 INIT_HLIST_BL_NODE(&dentry->d_hash);
1483 INIT_LIST_HEAD(&dentry->d_lru);
1484 INIT_LIST_HEAD(&dentry->d_subdirs);
1485 INIT_HLIST_NODE(&dentry->d_alias);
1486 INIT_LIST_HEAD(&dentry->d_u.d_child);
1487 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1489 this_cpu_inc(nr_dentry);
1495 * d_alloc - allocate a dcache entry
1496 * @parent: parent of entry to allocate
1497 * @name: qstr of the name
1499 * Allocates a dentry. It returns %NULL if there is insufficient memory
1500 * available. On a success the dentry is returned. The name passed in is
1501 * copied and the copy passed in may be reused after this call.
1503 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1505 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1509 spin_lock(&parent->d_lock);
1511 * don't need child lock because it is not subject
1512 * to concurrency here
1514 __dget_dlock(parent);
1515 dentry->d_parent = parent;
1516 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1517 spin_unlock(&parent->d_lock);
1521 EXPORT_SYMBOL(d_alloc);
1524 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1525 * @sb: the superblock
1526 * @name: qstr of the name
1528 * For a filesystem that just pins its dentries in memory and never
1529 * performs lookups at all, return an unhashed IS_ROOT dentry.
1531 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1533 return __d_alloc(sb, name);
1535 EXPORT_SYMBOL(d_alloc_pseudo);
1537 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1542 q.len = strlen(name);
1543 q.hash = full_name_hash(q.name, q.len);
1544 return d_alloc(parent, &q);
1546 EXPORT_SYMBOL(d_alloc_name);
1548 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1550 WARN_ON_ONCE(dentry->d_op);
1551 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1553 DCACHE_OP_REVALIDATE |
1554 DCACHE_OP_WEAK_REVALIDATE |
1555 DCACHE_OP_DELETE ));
1560 dentry->d_flags |= DCACHE_OP_HASH;
1562 dentry->d_flags |= DCACHE_OP_COMPARE;
1563 if (op->d_revalidate)
1564 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1565 if (op->d_weak_revalidate)
1566 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1568 dentry->d_flags |= DCACHE_OP_DELETE;
1570 dentry->d_flags |= DCACHE_OP_PRUNE;
1573 EXPORT_SYMBOL(d_set_d_op);
1575 static unsigned d_flags_for_inode(struct inode *inode)
1577 unsigned add_flags = DCACHE_FILE_TYPE;
1580 return DCACHE_MISS_TYPE;
1582 if (S_ISDIR(inode->i_mode)) {
1583 add_flags = DCACHE_DIRECTORY_TYPE;
1584 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1585 if (unlikely(!inode->i_op->lookup))
1586 add_flags = DCACHE_AUTODIR_TYPE;
1588 inode->i_opflags |= IOP_LOOKUP;
1590 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1591 if (unlikely(inode->i_op->follow_link))
1592 add_flags = DCACHE_SYMLINK_TYPE;
1594 inode->i_opflags |= IOP_NOFOLLOW;
1597 if (unlikely(IS_AUTOMOUNT(inode)))
1598 add_flags |= DCACHE_NEED_AUTOMOUNT;
1602 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1604 unsigned add_flags = d_flags_for_inode(inode);
1606 spin_lock(&dentry->d_lock);
1607 __d_set_type(dentry, add_flags);
1609 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1610 dentry->d_inode = inode;
1611 dentry_rcuwalk_barrier(dentry);
1612 spin_unlock(&dentry->d_lock);
1613 fsnotify_d_instantiate(dentry, inode);
1617 * d_instantiate - fill in inode information for a dentry
1618 * @entry: dentry to complete
1619 * @inode: inode to attach to this dentry
1621 * Fill in inode information in the entry.
1623 * This turns negative dentries into productive full members
1626 * NOTE! This assumes that the inode count has been incremented
1627 * (or otherwise set) by the caller to indicate that it is now
1628 * in use by the dcache.
1631 void d_instantiate(struct dentry *entry, struct inode * inode)
1633 BUG_ON(!hlist_unhashed(&entry->d_alias));
1635 spin_lock(&inode->i_lock);
1636 __d_instantiate(entry, inode);
1638 spin_unlock(&inode->i_lock);
1639 security_d_instantiate(entry, inode);
1641 EXPORT_SYMBOL(d_instantiate);
1644 * d_instantiate_unique - instantiate a non-aliased dentry
1645 * @entry: dentry to instantiate
1646 * @inode: inode to attach to this dentry
1648 * Fill in inode information in the entry. On success, it returns NULL.
1649 * If an unhashed alias of "entry" already exists, then we return the
1650 * aliased dentry instead and drop one reference to inode.
1652 * Note that in order to avoid conflicts with rename() etc, the caller
1653 * had better be holding the parent directory semaphore.
1655 * This also assumes that the inode count has been incremented
1656 * (or otherwise set) by the caller to indicate that it is now
1657 * in use by the dcache.
1659 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1660 struct inode *inode)
1662 struct dentry *alias;
1663 int len = entry->d_name.len;
1664 const char *name = entry->d_name.name;
1665 unsigned int hash = entry->d_name.hash;
1668 __d_instantiate(entry, NULL);
1672 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1674 * Don't need alias->d_lock here, because aliases with
1675 * d_parent == entry->d_parent are not subject to name or
1676 * parent changes, because the parent inode i_mutex is held.
1678 if (alias->d_name.hash != hash)
1680 if (alias->d_parent != entry->d_parent)
1682 if (alias->d_name.len != len)
1684 if (dentry_cmp(alias, name, len))
1690 __d_instantiate(entry, inode);
1694 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1696 struct dentry *result;
1698 BUG_ON(!hlist_unhashed(&entry->d_alias));
1701 spin_lock(&inode->i_lock);
1702 result = __d_instantiate_unique(entry, inode);
1704 spin_unlock(&inode->i_lock);
1707 security_d_instantiate(entry, inode);
1711 BUG_ON(!d_unhashed(result));
1716 EXPORT_SYMBOL(d_instantiate_unique);
1719 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1720 * @entry: dentry to complete
1721 * @inode: inode to attach to this dentry
1723 * Fill in inode information in the entry. If a directory alias is found, then
1724 * return an error (and drop inode). Together with d_materialise_unique() this
1725 * guarantees that a directory inode may never have more than one alias.
1727 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1729 BUG_ON(!hlist_unhashed(&entry->d_alias));
1731 spin_lock(&inode->i_lock);
1732 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1733 spin_unlock(&inode->i_lock);
1737 __d_instantiate(entry, inode);
1738 spin_unlock(&inode->i_lock);
1739 security_d_instantiate(entry, inode);
1743 EXPORT_SYMBOL(d_instantiate_no_diralias);
1745 struct dentry *d_make_root(struct inode *root_inode)
1747 struct dentry *res = NULL;
1750 static const struct qstr name = QSTR_INIT("/", 1);
1752 res = __d_alloc(root_inode->i_sb, &name);
1754 d_instantiate(res, root_inode);
1760 EXPORT_SYMBOL(d_make_root);
1762 static struct dentry * __d_find_any_alias(struct inode *inode)
1764 struct dentry *alias;
1766 if (hlist_empty(&inode->i_dentry))
1768 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1774 * d_find_any_alias - find any alias for a given inode
1775 * @inode: inode to find an alias for
1777 * If any aliases exist for the given inode, take and return a
1778 * reference for one of them. If no aliases exist, return %NULL.
1780 struct dentry *d_find_any_alias(struct inode *inode)
1784 spin_lock(&inode->i_lock);
1785 de = __d_find_any_alias(inode);
1786 spin_unlock(&inode->i_lock);
1789 EXPORT_SYMBOL(d_find_any_alias);
1792 * d_obtain_alias - find or allocate a dentry for a given inode
1793 * @inode: inode to allocate the dentry for
1795 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1796 * similar open by handle operations. The returned dentry may be anonymous,
1797 * or may have a full name (if the inode was already in the cache).
1799 * When called on a directory inode, we must ensure that the inode only ever
1800 * has one dentry. If a dentry is found, that is returned instead of
1801 * allocating a new one.
1803 * On successful return, the reference to the inode has been transferred
1804 * to the dentry. In case of an error the reference on the inode is released.
1805 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1806 * be passed in and will be the error will be propagate to the return value,
1807 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1809 struct dentry *d_obtain_alias(struct inode *inode)
1811 static const struct qstr anonstring = QSTR_INIT("/", 1);
1817 return ERR_PTR(-ESTALE);
1819 return ERR_CAST(inode);
1821 res = d_find_any_alias(inode);
1825 tmp = __d_alloc(inode->i_sb, &anonstring);
1827 res = ERR_PTR(-ENOMEM);
1831 spin_lock(&inode->i_lock);
1832 res = __d_find_any_alias(inode);
1834 spin_unlock(&inode->i_lock);
1839 /* attach a disconnected dentry */
1840 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1842 spin_lock(&tmp->d_lock);
1843 tmp->d_inode = inode;
1844 tmp->d_flags |= add_flags;
1845 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1846 hlist_bl_lock(&tmp->d_sb->s_anon);
1847 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1848 hlist_bl_unlock(&tmp->d_sb->s_anon);
1849 spin_unlock(&tmp->d_lock);
1850 spin_unlock(&inode->i_lock);
1851 security_d_instantiate(tmp, inode);
1856 if (res && !IS_ERR(res))
1857 security_d_instantiate(res, inode);
1861 EXPORT_SYMBOL(d_obtain_alias);
1864 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1865 * @inode: the inode which may have a disconnected dentry
1866 * @dentry: a negative dentry which we want to point to the inode.
1868 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1869 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1870 * and return it, else simply d_add the inode to the dentry and return NULL.
1872 * This is needed in the lookup routine of any filesystem that is exportable
1873 * (via knfsd) so that we can build dcache paths to directories effectively.
1875 * If a dentry was found and moved, then it is returned. Otherwise NULL
1876 * is returned. This matches the expected return value of ->lookup.
1878 * Cluster filesystems may call this function with a negative, hashed dentry.
1879 * In that case, we know that the inode will be a regular file, and also this
1880 * will only occur during atomic_open. So we need to check for the dentry
1881 * being already hashed only in the final case.
1883 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1885 struct dentry *new = NULL;
1888 return ERR_CAST(inode);
1890 if (inode && S_ISDIR(inode->i_mode)) {
1891 spin_lock(&inode->i_lock);
1892 new = __d_find_alias(inode, 1);
1894 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1895 spin_unlock(&inode->i_lock);
1896 security_d_instantiate(new, inode);
1897 d_move(new, dentry);
1900 /* already taking inode->i_lock, so d_add() by hand */
1901 __d_instantiate(dentry, inode);
1902 spin_unlock(&inode->i_lock);
1903 security_d_instantiate(dentry, inode);
1907 d_instantiate(dentry, inode);
1908 if (d_unhashed(dentry))
1913 EXPORT_SYMBOL(d_splice_alias);
1916 * d_add_ci - lookup or allocate new dentry with case-exact name
1917 * @inode: the inode case-insensitive lookup has found
1918 * @dentry: the negative dentry that was passed to the parent's lookup func
1919 * @name: the case-exact name to be associated with the returned dentry
1921 * This is to avoid filling the dcache with case-insensitive names to the
1922 * same inode, only the actual correct case is stored in the dcache for
1923 * case-insensitive filesystems.
1925 * For a case-insensitive lookup match and if the the case-exact dentry
1926 * already exists in in the dcache, use it and return it.
1928 * If no entry exists with the exact case name, allocate new dentry with
1929 * the exact case, and return the spliced entry.
1931 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1934 struct dentry *found;
1938 * First check if a dentry matching the name already exists,
1939 * if not go ahead and create it now.
1941 found = d_hash_and_lookup(dentry->d_parent, name);
1942 if (unlikely(IS_ERR(found)))
1945 new = d_alloc(dentry->d_parent, name);
1947 found = ERR_PTR(-ENOMEM);
1951 found = d_splice_alias(inode, new);
1960 * If a matching dentry exists, and it's not negative use it.
1962 * Decrement the reference count to balance the iget() done
1965 if (found->d_inode) {
1966 if (unlikely(found->d_inode != inode)) {
1967 /* This can't happen because bad inodes are unhashed. */
1968 BUG_ON(!is_bad_inode(inode));
1969 BUG_ON(!is_bad_inode(found->d_inode));
1976 * Negative dentry: instantiate it unless the inode is a directory and
1977 * already has a dentry.
1979 new = d_splice_alias(inode, found);
1990 EXPORT_SYMBOL(d_add_ci);
1993 * Do the slow-case of the dentry name compare.
1995 * Unlike the dentry_cmp() function, we need to atomically
1996 * load the name and length information, so that the
1997 * filesystem can rely on them, and can use the 'name' and
1998 * 'len' information without worrying about walking off the
1999 * end of memory etc.
2001 * Thus the read_seqcount_retry() and the "duplicate" info
2002 * in arguments (the low-level filesystem should not look
2003 * at the dentry inode or name contents directly, since
2004 * rename can change them while we're in RCU mode).
2006 enum slow_d_compare {
2012 static noinline enum slow_d_compare slow_dentry_cmp(
2013 const struct dentry *parent,
2014 struct dentry *dentry,
2016 const struct qstr *name)
2018 int tlen = dentry->d_name.len;
2019 const char *tname = dentry->d_name.name;
2021 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2023 return D_COMP_SEQRETRY;
2025 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2026 return D_COMP_NOMATCH;
2031 * __d_lookup_rcu - search for a dentry (racy, store-free)
2032 * @parent: parent dentry
2033 * @name: qstr of name we wish to find
2034 * @seqp: returns d_seq value at the point where the dentry was found
2035 * Returns: dentry, or NULL
2037 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2038 * resolution (store-free path walking) design described in
2039 * Documentation/filesystems/path-lookup.txt.
2041 * This is not to be used outside core vfs.
2043 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2044 * held, and rcu_read_lock held. The returned dentry must not be stored into
2045 * without taking d_lock and checking d_seq sequence count against @seq
2048 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2051 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2052 * the returned dentry, so long as its parent's seqlock is checked after the
2053 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2054 * is formed, giving integrity down the path walk.
2056 * NOTE! The caller *has* to check the resulting dentry against the sequence
2057 * number we've returned before using any of the resulting dentry state!
2059 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2060 const struct qstr *name,
2063 u64 hashlen = name->hash_len;
2064 const unsigned char *str = name->name;
2065 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2066 struct hlist_bl_node *node;
2067 struct dentry *dentry;
2070 * Note: There is significant duplication with __d_lookup_rcu which is
2071 * required to prevent single threaded performance regressions
2072 * especially on architectures where smp_rmb (in seqcounts) are costly.
2073 * Keep the two functions in sync.
2077 * The hash list is protected using RCU.
2079 * Carefully use d_seq when comparing a candidate dentry, to avoid
2080 * races with d_move().
2082 * It is possible that concurrent renames can mess up our list
2083 * walk here and result in missing our dentry, resulting in the
2084 * false-negative result. d_lookup() protects against concurrent
2085 * renames using rename_lock seqlock.
2087 * See Documentation/filesystems/path-lookup.txt for more details.
2089 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2094 * The dentry sequence count protects us from concurrent
2095 * renames, and thus protects parent and name fields.
2097 * The caller must perform a seqcount check in order
2098 * to do anything useful with the returned dentry.
2100 * NOTE! We do a "raw" seqcount_begin here. That means that
2101 * we don't wait for the sequence count to stabilize if it
2102 * is in the middle of a sequence change. If we do the slow
2103 * dentry compare, we will do seqretries until it is stable,
2104 * and if we end up with a successful lookup, we actually
2105 * want to exit RCU lookup anyway.
2107 seq = raw_seqcount_begin(&dentry->d_seq);
2108 if (dentry->d_parent != parent)
2110 if (d_unhashed(dentry))
2113 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2114 if (dentry->d_name.hash != hashlen_hash(hashlen))
2117 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2120 case D_COMP_NOMATCH:
2127 if (dentry->d_name.hash_len != hashlen)
2130 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2137 * d_lookup - search for a dentry
2138 * @parent: parent dentry
2139 * @name: qstr of name we wish to find
2140 * Returns: dentry, or NULL
2142 * d_lookup searches the children of the parent dentry for the name in
2143 * question. If the dentry is found its reference count is incremented and the
2144 * dentry is returned. The caller must use dput to free the entry when it has
2145 * finished using it. %NULL is returned if the dentry does not exist.
2147 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2149 struct dentry *dentry;
2153 seq = read_seqbegin(&rename_lock);
2154 dentry = __d_lookup(parent, name);
2157 } while (read_seqretry(&rename_lock, seq));
2160 EXPORT_SYMBOL(d_lookup);
2163 * __d_lookup - search for a dentry (racy)
2164 * @parent: parent dentry
2165 * @name: qstr of name we wish to find
2166 * Returns: dentry, or NULL
2168 * __d_lookup is like d_lookup, however it may (rarely) return a
2169 * false-negative result due to unrelated rename activity.
2171 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2172 * however it must be used carefully, eg. with a following d_lookup in
2173 * the case of failure.
2175 * __d_lookup callers must be commented.
2177 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2179 unsigned int len = name->len;
2180 unsigned int hash = name->hash;
2181 const unsigned char *str = name->name;
2182 struct hlist_bl_head *b = d_hash(parent, hash);
2183 struct hlist_bl_node *node;
2184 struct dentry *found = NULL;
2185 struct dentry *dentry;
2188 * Note: There is significant duplication with __d_lookup_rcu which is
2189 * required to prevent single threaded performance regressions
2190 * especially on architectures where smp_rmb (in seqcounts) are costly.
2191 * Keep the two functions in sync.
2195 * The hash list is protected using RCU.
2197 * Take d_lock when comparing a candidate dentry, to avoid races
2200 * It is possible that concurrent renames can mess up our list
2201 * walk here and result in missing our dentry, resulting in the
2202 * false-negative result. d_lookup() protects against concurrent
2203 * renames using rename_lock seqlock.
2205 * See Documentation/filesystems/path-lookup.txt for more details.
2209 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2211 if (dentry->d_name.hash != hash)
2214 spin_lock(&dentry->d_lock);
2215 if (dentry->d_parent != parent)
2217 if (d_unhashed(dentry))
2221 * It is safe to compare names since d_move() cannot
2222 * change the qstr (protected by d_lock).
2224 if (parent->d_flags & DCACHE_OP_COMPARE) {
2225 int tlen = dentry->d_name.len;
2226 const char *tname = dentry->d_name.name;
2227 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2230 if (dentry->d_name.len != len)
2232 if (dentry_cmp(dentry, str, len))
2236 dentry->d_lockref.count++;
2238 spin_unlock(&dentry->d_lock);
2241 spin_unlock(&dentry->d_lock);
2249 * d_hash_and_lookup - hash the qstr then search for a dentry
2250 * @dir: Directory to search in
2251 * @name: qstr of name we wish to find
2253 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2255 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2258 * Check for a fs-specific hash function. Note that we must
2259 * calculate the standard hash first, as the d_op->d_hash()
2260 * routine may choose to leave the hash value unchanged.
2262 name->hash = full_name_hash(name->name, name->len);
2263 if (dir->d_flags & DCACHE_OP_HASH) {
2264 int err = dir->d_op->d_hash(dir, name);
2265 if (unlikely(err < 0))
2266 return ERR_PTR(err);
2268 return d_lookup(dir, name);
2270 EXPORT_SYMBOL(d_hash_and_lookup);
2273 * d_validate - verify dentry provided from insecure source (deprecated)
2274 * @dentry: The dentry alleged to be valid child of @dparent
2275 * @dparent: The parent dentry (known to be valid)
2277 * An insecure source has sent us a dentry, here we verify it and dget() it.
2278 * This is used by ncpfs in its readdir implementation.
2279 * Zero is returned in the dentry is invalid.
2281 * This function is slow for big directories, and deprecated, do not use it.
2283 int d_validate(struct dentry *dentry, struct dentry *dparent)
2285 struct dentry *child;
2287 spin_lock(&dparent->d_lock);
2288 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2289 if (dentry == child) {
2290 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2291 __dget_dlock(dentry);
2292 spin_unlock(&dentry->d_lock);
2293 spin_unlock(&dparent->d_lock);
2297 spin_unlock(&dparent->d_lock);
2301 EXPORT_SYMBOL(d_validate);
2304 * When a file is deleted, we have two options:
2305 * - turn this dentry into a negative dentry
2306 * - unhash this dentry and free it.
2308 * Usually, we want to just turn this into
2309 * a negative dentry, but if anybody else is
2310 * currently using the dentry or the inode
2311 * we can't do that and we fall back on removing
2312 * it from the hash queues and waiting for
2313 * it to be deleted later when it has no users
2317 * d_delete - delete a dentry
2318 * @dentry: The dentry to delete
2320 * Turn the dentry into a negative dentry if possible, otherwise
2321 * remove it from the hash queues so it can be deleted later
2324 void d_delete(struct dentry * dentry)
2326 struct inode *inode;
2329 * Are we the only user?
2332 spin_lock(&dentry->d_lock);
2333 inode = dentry->d_inode;
2334 isdir = S_ISDIR(inode->i_mode);
2335 if (dentry->d_lockref.count == 1) {
2336 if (!spin_trylock(&inode->i_lock)) {
2337 spin_unlock(&dentry->d_lock);
2341 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2342 dentry_unlink_inode(dentry);
2343 fsnotify_nameremove(dentry, isdir);
2347 if (!d_unhashed(dentry))
2350 spin_unlock(&dentry->d_lock);
2352 fsnotify_nameremove(dentry, isdir);
2354 EXPORT_SYMBOL(d_delete);
2356 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2358 BUG_ON(!d_unhashed(entry));
2360 entry->d_flags |= DCACHE_RCUACCESS;
2361 hlist_bl_add_head_rcu(&entry->d_hash, b);
2365 static void _d_rehash(struct dentry * entry)
2367 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2371 * d_rehash - add an entry back to the hash
2372 * @entry: dentry to add to the hash
2374 * Adds a dentry to the hash according to its name.
2377 void d_rehash(struct dentry * entry)
2379 spin_lock(&entry->d_lock);
2381 spin_unlock(&entry->d_lock);
2383 EXPORT_SYMBOL(d_rehash);
2386 * dentry_update_name_case - update case insensitive dentry with a new name
2387 * @dentry: dentry to be updated
2390 * Update a case insensitive dentry with new case of name.
2392 * dentry must have been returned by d_lookup with name @name. Old and new
2393 * name lengths must match (ie. no d_compare which allows mismatched name
2396 * Parent inode i_mutex must be held over d_lookup and into this call (to
2397 * keep renames and concurrent inserts, and readdir(2) away).
2399 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2401 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2402 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2404 spin_lock(&dentry->d_lock);
2405 write_seqcount_begin(&dentry->d_seq);
2406 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2407 write_seqcount_end(&dentry->d_seq);
2408 spin_unlock(&dentry->d_lock);
2410 EXPORT_SYMBOL(dentry_update_name_case);
2412 static void switch_names(struct dentry *dentry, struct dentry *target)
2414 if (dname_external(target)) {
2415 if (dname_external(dentry)) {
2417 * Both external: swap the pointers
2419 swap(target->d_name.name, dentry->d_name.name);
2422 * dentry:internal, target:external. Steal target's
2423 * storage and make target internal.
2425 memcpy(target->d_iname, dentry->d_name.name,
2426 dentry->d_name.len + 1);
2427 dentry->d_name.name = target->d_name.name;
2428 target->d_name.name = target->d_iname;
2431 if (dname_external(dentry)) {
2433 * dentry:external, target:internal. Give dentry's
2434 * storage to target and make dentry internal
2436 memcpy(dentry->d_iname, target->d_name.name,
2437 target->d_name.len + 1);
2438 target->d_name.name = dentry->d_name.name;
2439 dentry->d_name.name = dentry->d_iname;
2442 * Both are internal.
2445 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2446 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2447 swap(((long *) &dentry->d_iname)[i],
2448 ((long *) &target->d_iname)[i]);
2452 swap(dentry->d_name.len, target->d_name.len);
2455 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2458 * XXXX: do we really need to take target->d_lock?
2460 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2461 spin_lock(&target->d_parent->d_lock);
2463 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2464 spin_lock(&dentry->d_parent->d_lock);
2465 spin_lock_nested(&target->d_parent->d_lock,
2466 DENTRY_D_LOCK_NESTED);
2468 spin_lock(&target->d_parent->d_lock);
2469 spin_lock_nested(&dentry->d_parent->d_lock,
2470 DENTRY_D_LOCK_NESTED);
2473 if (target < dentry) {
2474 spin_lock_nested(&target->d_lock, 2);
2475 spin_lock_nested(&dentry->d_lock, 3);
2477 spin_lock_nested(&dentry->d_lock, 2);
2478 spin_lock_nested(&target->d_lock, 3);
2482 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2483 struct dentry *target)
2485 if (target->d_parent != dentry->d_parent)
2486 spin_unlock(&dentry->d_parent->d_lock);
2487 if (target->d_parent != target)
2488 spin_unlock(&target->d_parent->d_lock);
2492 * When switching names, the actual string doesn't strictly have to
2493 * be preserved in the target - because we're dropping the target
2494 * anyway. As such, we can just do a simple memcpy() to copy over
2495 * the new name before we switch.
2497 * Note that we have to be a lot more careful about getting the hash
2498 * switched - we have to switch the hash value properly even if it
2499 * then no longer matches the actual (corrupted) string of the target.
2500 * The hash value has to match the hash queue that the dentry is on..
2503 * __d_move - move a dentry
2504 * @dentry: entry to move
2505 * @target: new dentry
2506 * @exchange: exchange the two dentries
2508 * Update the dcache to reflect the move of a file name. Negative
2509 * dcache entries should not be moved in this way. Caller must hold
2510 * rename_lock, the i_mutex of the source and target directories,
2511 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2513 static void __d_move(struct dentry *dentry, struct dentry *target,
2516 if (!dentry->d_inode)
2517 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2519 BUG_ON(d_ancestor(dentry, target));
2520 BUG_ON(d_ancestor(target, dentry));
2522 dentry_lock_for_move(dentry, target);
2524 write_seqcount_begin(&dentry->d_seq);
2525 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2527 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2530 * Move the dentry to the target hash queue. Don't bother checking
2531 * for the same hash queue because of how unlikely it is.
2534 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2537 * Unhash the target (d_delete() is not usable here). If exchanging
2538 * the two dentries, then rehash onto the other's hash queue.
2543 d_hash(dentry->d_parent, dentry->d_name.hash));
2546 list_del(&dentry->d_u.d_child);
2547 list_del(&target->d_u.d_child);
2549 /* Switch the names.. */
2550 switch_names(dentry, target);
2551 swap(dentry->d_name.hash, target->d_name.hash);
2553 /* ... and switch the parents */
2554 if (IS_ROOT(dentry)) {
2555 dentry->d_parent = target->d_parent;
2556 target->d_parent = target;
2557 INIT_LIST_HEAD(&target->d_u.d_child);
2559 swap(dentry->d_parent, target->d_parent);
2561 /* And add them back to the (new) parent lists */
2562 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2565 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2567 write_seqcount_end(&target->d_seq);
2568 write_seqcount_end(&dentry->d_seq);
2570 dentry_unlock_parents_for_move(dentry, target);
2572 fsnotify_d_move(target);
2573 spin_unlock(&target->d_lock);
2574 fsnotify_d_move(dentry);
2575 spin_unlock(&dentry->d_lock);
2579 * d_move - move a dentry
2580 * @dentry: entry to move
2581 * @target: new dentry
2583 * Update the dcache to reflect the move of a file name. Negative
2584 * dcache entries should not be moved in this way. See the locking
2585 * requirements for __d_move.
2587 void d_move(struct dentry *dentry, struct dentry *target)
2589 write_seqlock(&rename_lock);
2590 __d_move(dentry, target, false);
2591 write_sequnlock(&rename_lock);
2593 EXPORT_SYMBOL(d_move);
2596 * d_exchange - exchange two dentries
2597 * @dentry1: first dentry
2598 * @dentry2: second dentry
2600 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2602 write_seqlock(&rename_lock);
2604 WARN_ON(!dentry1->d_inode);
2605 WARN_ON(!dentry2->d_inode);
2606 WARN_ON(IS_ROOT(dentry1));
2607 WARN_ON(IS_ROOT(dentry2));
2609 __d_move(dentry1, dentry2, true);
2611 write_sequnlock(&rename_lock);
2615 * d_ancestor - search for an ancestor
2616 * @p1: ancestor dentry
2619 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2620 * an ancestor of p2, else NULL.
2622 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2626 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2627 if (p->d_parent == p1)
2634 * This helper attempts to cope with remotely renamed directories
2636 * It assumes that the caller is already holding
2637 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2639 * Note: If ever the locking in lock_rename() changes, then please
2640 * remember to update this too...
2642 static struct dentry *__d_unalias(struct inode *inode,
2643 struct dentry *dentry, struct dentry *alias)
2645 struct mutex *m1 = NULL, *m2 = NULL;
2646 struct dentry *ret = ERR_PTR(-EBUSY);
2648 /* If alias and dentry share a parent, then no extra locks required */
2649 if (alias->d_parent == dentry->d_parent)
2652 /* See lock_rename() */
2653 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2655 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2656 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2658 m2 = &alias->d_parent->d_inode->i_mutex;
2660 if (likely(!d_mountpoint(alias))) {
2661 __d_move(alias, dentry, false);
2665 spin_unlock(&inode->i_lock);
2674 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2675 * named dentry in place of the dentry to be replaced.
2676 * returns with anon->d_lock held!
2678 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2680 struct dentry *dparent;
2682 dentry_lock_for_move(anon, dentry);
2684 write_seqcount_begin(&dentry->d_seq);
2685 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2687 dparent = dentry->d_parent;
2689 switch_names(dentry, anon);
2690 swap(dentry->d_name.hash, anon->d_name.hash);
2692 dentry->d_parent = dentry;
2693 list_del_init(&dentry->d_u.d_child);
2694 anon->d_parent = dparent;
2695 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2697 write_seqcount_end(&dentry->d_seq);
2698 write_seqcount_end(&anon->d_seq);
2700 dentry_unlock_parents_for_move(anon, dentry);
2701 spin_unlock(&dentry->d_lock);
2703 /* anon->d_lock still locked, returns locked */
2707 * d_materialise_unique - introduce an inode into the tree
2708 * @dentry: candidate dentry
2709 * @inode: inode to bind to the dentry, to which aliases may be attached
2711 * Introduces an dentry into the tree, substituting an extant disconnected
2712 * root directory alias in its place if there is one. Caller must hold the
2713 * i_mutex of the parent directory.
2715 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2717 struct dentry *actual;
2719 BUG_ON(!d_unhashed(dentry));
2723 __d_instantiate(dentry, NULL);
2728 spin_lock(&inode->i_lock);
2730 if (S_ISDIR(inode->i_mode)) {
2731 struct dentry *alias;
2733 /* Does an aliased dentry already exist? */
2734 alias = __d_find_alias(inode, 0);
2737 write_seqlock(&rename_lock);
2739 if (d_ancestor(alias, dentry)) {
2740 /* Check for loops */
2741 actual = ERR_PTR(-ELOOP);
2742 spin_unlock(&inode->i_lock);
2743 } else if (IS_ROOT(alias)) {
2744 /* Is this an anonymous mountpoint that we
2745 * could splice into our tree? */
2746 __d_materialise_dentry(dentry, alias);
2747 write_sequnlock(&rename_lock);
2751 /* Nope, but we must(!) avoid directory
2752 * aliasing. This drops inode->i_lock */
2753 actual = __d_unalias(inode, dentry, alias);
2755 write_sequnlock(&rename_lock);
2756 if (IS_ERR(actual)) {
2757 if (PTR_ERR(actual) == -ELOOP)
2758 pr_warn_ratelimited(
2759 "VFS: Lookup of '%s' in %s %s"
2760 " would have caused loop\n",
2761 dentry->d_name.name,
2762 inode->i_sb->s_type->name,
2770 /* Add a unique reference */
2771 actual = __d_instantiate_unique(dentry, inode);
2775 BUG_ON(!d_unhashed(actual));
2777 spin_lock(&actual->d_lock);
2780 spin_unlock(&actual->d_lock);
2781 spin_unlock(&inode->i_lock);
2783 if (actual == dentry) {
2784 security_d_instantiate(dentry, inode);
2791 EXPORT_SYMBOL_GPL(d_materialise_unique);
2793 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2797 return -ENAMETOOLONG;
2799 memcpy(*buffer, str, namelen);
2804 * prepend_name - prepend a pathname in front of current buffer pointer
2805 * @buffer: buffer pointer
2806 * @buflen: allocated length of the buffer
2807 * @name: name string and length qstr structure
2809 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2810 * make sure that either the old or the new name pointer and length are
2811 * fetched. However, there may be mismatch between length and pointer.
2812 * The length cannot be trusted, we need to copy it byte-by-byte until
2813 * the length is reached or a null byte is found. It also prepends "/" at
2814 * the beginning of the name. The sequence number check at the caller will
2815 * retry it again when a d_move() does happen. So any garbage in the buffer
2816 * due to mismatched pointer and length will be discarded.
2818 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2820 const char *dname = ACCESS_ONCE(name->name);
2821 u32 dlen = ACCESS_ONCE(name->len);
2824 *buflen -= dlen + 1;
2826 return -ENAMETOOLONG;
2827 p = *buffer -= dlen + 1;
2839 * prepend_path - Prepend path string to a buffer
2840 * @path: the dentry/vfsmount to report
2841 * @root: root vfsmnt/dentry
2842 * @buffer: pointer to the end of the buffer
2843 * @buflen: pointer to buffer length
2845 * The function will first try to write out the pathname without taking any
2846 * lock other than the RCU read lock to make sure that dentries won't go away.
2847 * It only checks the sequence number of the global rename_lock as any change
2848 * in the dentry's d_seq will be preceded by changes in the rename_lock
2849 * sequence number. If the sequence number had been changed, it will restart
2850 * the whole pathname back-tracing sequence again by taking the rename_lock.
2851 * In this case, there is no need to take the RCU read lock as the recursive
2852 * parent pointer references will keep the dentry chain alive as long as no
2853 * rename operation is performed.
2855 static int prepend_path(const struct path *path,
2856 const struct path *root,
2857 char **buffer, int *buflen)
2859 struct dentry *dentry;
2860 struct vfsmount *vfsmnt;
2863 unsigned seq, m_seq = 0;
2869 read_seqbegin_or_lock(&mount_lock, &m_seq);
2876 dentry = path->dentry;
2878 mnt = real_mount(vfsmnt);
2879 read_seqbegin_or_lock(&rename_lock, &seq);
2880 while (dentry != root->dentry || vfsmnt != root->mnt) {
2881 struct dentry * parent;
2883 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2884 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2886 if (mnt != parent) {
2887 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2893 * Filesystems needing to implement special "root names"
2894 * should do so with ->d_dname()
2896 if (IS_ROOT(dentry) &&
2897 (dentry->d_name.len != 1 ||
2898 dentry->d_name.name[0] != '/')) {
2899 WARN(1, "Root dentry has weird name <%.*s>\n",
2900 (int) dentry->d_name.len,
2901 dentry->d_name.name);
2904 error = is_mounted(vfsmnt) ? 1 : 2;
2907 parent = dentry->d_parent;
2909 error = prepend_name(&bptr, &blen, &dentry->d_name);
2917 if (need_seqretry(&rename_lock, seq)) {
2921 done_seqretry(&rename_lock, seq);
2925 if (need_seqretry(&mount_lock, m_seq)) {
2929 done_seqretry(&mount_lock, m_seq);
2931 if (error >= 0 && bptr == *buffer) {
2933 error = -ENAMETOOLONG;
2943 * __d_path - return the path of a dentry
2944 * @path: the dentry/vfsmount to report
2945 * @root: root vfsmnt/dentry
2946 * @buf: buffer to return value in
2947 * @buflen: buffer length
2949 * Convert a dentry into an ASCII path name.
2951 * Returns a pointer into the buffer or an error code if the
2952 * path was too long.
2954 * "buflen" should be positive.
2956 * If the path is not reachable from the supplied root, return %NULL.
2958 char *__d_path(const struct path *path,
2959 const struct path *root,
2960 char *buf, int buflen)
2962 char *res = buf + buflen;
2965 prepend(&res, &buflen, "\0", 1);
2966 error = prepend_path(path, root, &res, &buflen);
2969 return ERR_PTR(error);
2975 char *d_absolute_path(const struct path *path,
2976 char *buf, int buflen)
2978 struct path root = {};
2979 char *res = buf + buflen;
2982 prepend(&res, &buflen, "\0", 1);
2983 error = prepend_path(path, &root, &res, &buflen);
2988 return ERR_PTR(error);
2993 * same as __d_path but appends "(deleted)" for unlinked files.
2995 static int path_with_deleted(const struct path *path,
2996 const struct path *root,
2997 char **buf, int *buflen)
2999 prepend(buf, buflen, "\0", 1);
3000 if (d_unlinked(path->dentry)) {
3001 int error = prepend(buf, buflen, " (deleted)", 10);
3006 return prepend_path(path, root, buf, buflen);
3009 static int prepend_unreachable(char **buffer, int *buflen)
3011 return prepend(buffer, buflen, "(unreachable)", 13);
3014 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3019 seq = read_seqcount_begin(&fs->seq);
3021 } while (read_seqcount_retry(&fs->seq, seq));
3025 * d_path - return the path of a dentry
3026 * @path: path to report
3027 * @buf: buffer to return value in
3028 * @buflen: buffer length
3030 * Convert a dentry into an ASCII path name. If the entry has been deleted
3031 * the string " (deleted)" is appended. Note that this is ambiguous.
3033 * Returns a pointer into the buffer or an error code if the path was
3034 * too long. Note: Callers should use the returned pointer, not the passed
3035 * in buffer, to use the name! The implementation often starts at an offset
3036 * into the buffer, and may leave 0 bytes at the start.
3038 * "buflen" should be positive.
3040 char *d_path(const struct path *path, char *buf, int buflen)
3042 char *res = buf + buflen;
3047 * We have various synthetic filesystems that never get mounted. On
3048 * these filesystems dentries are never used for lookup purposes, and
3049 * thus don't need to be hashed. They also don't need a name until a
3050 * user wants to identify the object in /proc/pid/fd/. The little hack
3051 * below allows us to generate a name for these objects on demand:
3053 * Some pseudo inodes are mountable. When they are mounted
3054 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3055 * and instead have d_path return the mounted path.
3057 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3058 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3059 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3062 get_fs_root_rcu(current->fs, &root);
3063 error = path_with_deleted(path, &root, &res, &buflen);
3067 res = ERR_PTR(error);
3070 EXPORT_SYMBOL(d_path);
3073 * Helper function for dentry_operations.d_dname() members
3075 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3076 const char *fmt, ...)
3082 va_start(args, fmt);
3083 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3086 if (sz > sizeof(temp) || sz > buflen)
3087 return ERR_PTR(-ENAMETOOLONG);
3089 buffer += buflen - sz;
3090 return memcpy(buffer, temp, sz);
3093 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3095 char *end = buffer + buflen;
3096 /* these dentries are never renamed, so d_lock is not needed */
3097 if (prepend(&end, &buflen, " (deleted)", 11) ||
3098 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3099 prepend(&end, &buflen, "/", 1))
3100 end = ERR_PTR(-ENAMETOOLONG);
3103 EXPORT_SYMBOL(simple_dname);
3106 * Write full pathname from the root of the filesystem into the buffer.
3108 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3110 struct dentry *dentry;
3123 prepend(&end, &len, "\0", 1);
3127 read_seqbegin_or_lock(&rename_lock, &seq);
3128 while (!IS_ROOT(dentry)) {
3129 struct dentry *parent = dentry->d_parent;
3132 error = prepend_name(&end, &len, &dentry->d_name);
3141 if (need_seqretry(&rename_lock, seq)) {
3145 done_seqretry(&rename_lock, seq);
3150 return ERR_PTR(-ENAMETOOLONG);
3153 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3155 return __dentry_path(dentry, buf, buflen);
3157 EXPORT_SYMBOL(dentry_path_raw);
3159 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3164 if (d_unlinked(dentry)) {
3166 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3170 retval = __dentry_path(dentry, buf, buflen);
3171 if (!IS_ERR(retval) && p)
3172 *p = '/'; /* restore '/' overriden with '\0' */
3175 return ERR_PTR(-ENAMETOOLONG);
3178 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3184 seq = read_seqcount_begin(&fs->seq);
3187 } while (read_seqcount_retry(&fs->seq, seq));
3191 * NOTE! The user-level library version returns a
3192 * character pointer. The kernel system call just
3193 * returns the length of the buffer filled (which
3194 * includes the ending '\0' character), or a negative
3195 * error value. So libc would do something like
3197 * char *getcwd(char * buf, size_t size)
3201 * retval = sys_getcwd(buf, size);
3208 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3211 struct path pwd, root;
3212 char *page = __getname();
3218 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3221 if (!d_unlinked(pwd.dentry)) {
3223 char *cwd = page + PATH_MAX;
3224 int buflen = PATH_MAX;
3226 prepend(&cwd, &buflen, "\0", 1);
3227 error = prepend_path(&pwd, &root, &cwd, &buflen);
3233 /* Unreachable from current root */
3235 error = prepend_unreachable(&cwd, &buflen);
3241 len = PATH_MAX + page - cwd;
3244 if (copy_to_user(buf, cwd, len))
3257 * Test whether new_dentry is a subdirectory of old_dentry.
3259 * Trivially implemented using the dcache structure
3263 * is_subdir - is new dentry a subdirectory of old_dentry
3264 * @new_dentry: new dentry
3265 * @old_dentry: old dentry
3267 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3268 * Returns 0 otherwise.
3269 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3272 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3277 if (new_dentry == old_dentry)
3281 /* for restarting inner loop in case of seq retry */
3282 seq = read_seqbegin(&rename_lock);
3284 * Need rcu_readlock to protect against the d_parent trashing
3288 if (d_ancestor(old_dentry, new_dentry))
3293 } while (read_seqretry(&rename_lock, seq));
3298 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3300 struct dentry *root = data;
3301 if (dentry != root) {
3302 if (d_unhashed(dentry) || !dentry->d_inode)
3305 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3306 dentry->d_flags |= DCACHE_GENOCIDE;
3307 dentry->d_lockref.count--;
3310 return D_WALK_CONTINUE;
3313 void d_genocide(struct dentry *parent)
3315 d_walk(parent, parent, d_genocide_kill, NULL);
3318 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3320 inode_dec_link_count(inode);
3321 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3322 !hlist_unhashed(&dentry->d_alias) ||
3323 !d_unlinked(dentry));
3324 spin_lock(&dentry->d_parent->d_lock);
3325 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3326 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3327 (unsigned long long)inode->i_ino);
3328 spin_unlock(&dentry->d_lock);
3329 spin_unlock(&dentry->d_parent->d_lock);
3330 d_instantiate(dentry, inode);
3332 EXPORT_SYMBOL(d_tmpfile);
3334 static __initdata unsigned long dhash_entries;
3335 static int __init set_dhash_entries(char *str)
3339 dhash_entries = simple_strtoul(str, &str, 0);
3342 __setup("dhash_entries=", set_dhash_entries);
3344 static void __init dcache_init_early(void)
3348 /* If hashes are distributed across NUMA nodes, defer
3349 * hash allocation until vmalloc space is available.
3355 alloc_large_system_hash("Dentry cache",
3356 sizeof(struct hlist_bl_head),
3365 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3366 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3369 static void __init dcache_init(void)
3374 * A constructor could be added for stable state like the lists,
3375 * but it is probably not worth it because of the cache nature
3378 dentry_cache = KMEM_CACHE(dentry,
3379 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3381 /* Hash may have been set up in dcache_init_early */
3386 alloc_large_system_hash("Dentry cache",
3387 sizeof(struct hlist_bl_head),
3396 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3397 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3400 /* SLAB cache for __getname() consumers */
3401 struct kmem_cache *names_cachep __read_mostly;
3402 EXPORT_SYMBOL(names_cachep);
3404 EXPORT_SYMBOL(d_genocide);
3406 void __init vfs_caches_init_early(void)
3408 dcache_init_early();
3412 void __init vfs_caches_init(unsigned long mempages)
3414 unsigned long reserve;
3416 /* Base hash sizes on available memory, with a reserve equal to
3417 150% of current kernel size */
3419 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3420 mempages -= reserve;
3422 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3423 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3427 files_init(mempages);