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 * 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);
445 * Finish off a dentry we've decided to kill.
446 * dentry->d_lock must be held, returns with it unlocked.
447 * If ref is non-zero, then decrement the refcount too.
448 * Returns dentry requiring refcount drop, or NULL if we're done.
450 static struct dentry *
451 dentry_kill(struct dentry *dentry, int unlock_on_failure)
452 __releases(dentry->d_lock)
455 struct dentry *parent = NULL;
456 bool can_free = true;
458 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
459 can_free = dentry->d_flags & DCACHE_MAY_FREE;
460 spin_unlock(&dentry->d_lock);
464 inode = dentry->d_inode;
465 if (inode && !spin_trylock(&inode->i_lock)) {
467 if (unlock_on_failure) {
468 spin_unlock(&dentry->d_lock);
471 return dentry; /* try again with same dentry */
473 if (!IS_ROOT(dentry))
474 parent = dentry->d_parent;
475 if (parent && !spin_trylock(&parent->d_lock)) {
477 spin_unlock(&inode->i_lock);
482 * The dentry is now unrecoverably dead to the world.
484 lockref_mark_dead(&dentry->d_lockref);
487 * inform the fs via d_prune that this dentry is about to be
488 * unhashed and destroyed.
490 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
491 dentry->d_op->d_prune(dentry);
493 if (dentry->d_flags & DCACHE_LRU_LIST) {
494 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
497 /* if it was on the hash then remove it */
499 list_del(&dentry->d_u.d_child);
501 * Inform d_walk() that we are no longer attached to the
504 dentry->d_flags |= DCACHE_DENTRY_KILLED;
506 spin_unlock(&parent->d_lock);
509 * dentry_iput drops the locks, at which point nobody (except
510 * transient RCU lookups) can reach this dentry.
512 BUG_ON((int)dentry->d_lockref.count > 0);
513 this_cpu_dec(nr_dentry);
514 if (dentry->d_op && dentry->d_op->d_release)
515 dentry->d_op->d_release(dentry);
517 spin_lock(&dentry->d_lock);
518 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
519 dentry->d_flags |= DCACHE_MAY_FREE;
522 spin_unlock(&dentry->d_lock);
524 if (likely(can_free))
532 * This is complicated by the fact that we do not want to put
533 * dentries that are no longer on any hash chain on the unused
534 * list: we'd much rather just get rid of them immediately.
536 * However, that implies that we have to traverse the dentry
537 * tree upwards to the parents which might _also_ now be
538 * scheduled for deletion (it may have been only waiting for
539 * its last child to go away).
541 * This tail recursion is done by hand as we don't want to depend
542 * on the compiler to always get this right (gcc generally doesn't).
543 * Real recursion would eat up our stack space.
547 * dput - release a dentry
548 * @dentry: dentry to release
550 * Release a dentry. This will drop the usage count and if appropriate
551 * call the dentry unlink method as well as removing it from the queues and
552 * releasing its resources. If the parent dentries were scheduled for release
553 * they too may now get deleted.
555 void dput(struct dentry *dentry)
557 if (unlikely(!dentry))
561 if (lockref_put_or_lock(&dentry->d_lockref))
564 /* Unreachable? Get rid of it */
565 if (unlikely(d_unhashed(dentry)))
568 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
569 if (dentry->d_op->d_delete(dentry))
573 if (!(dentry->d_flags & DCACHE_REFERENCED))
574 dentry->d_flags |= DCACHE_REFERENCED;
575 dentry_lru_add(dentry);
577 dentry->d_lockref.count--;
578 spin_unlock(&dentry->d_lock);
582 dentry = dentry_kill(dentry, 1);
589 * d_invalidate - invalidate a dentry
590 * @dentry: dentry to invalidate
592 * Try to invalidate the dentry if it turns out to be
593 * possible. If there are other dentries that can be
594 * reached through this one we can't delete it and we
595 * return -EBUSY. On success we return 0.
600 int d_invalidate(struct dentry * dentry)
603 * If it's already been dropped, return OK.
605 spin_lock(&dentry->d_lock);
606 if (d_unhashed(dentry)) {
607 spin_unlock(&dentry->d_lock);
611 * Check whether to do a partial shrink_dcache
612 * to get rid of unused child entries.
614 if (!list_empty(&dentry->d_subdirs)) {
615 spin_unlock(&dentry->d_lock);
616 shrink_dcache_parent(dentry);
617 spin_lock(&dentry->d_lock);
621 * Somebody else still using it?
623 * If it's a directory, we can't drop it
624 * for fear of somebody re-populating it
625 * with children (even though dropping it
626 * would make it unreachable from the root,
627 * we might still populate it if it was a
628 * working directory or similar).
629 * We also need to leave mountpoints alone,
632 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
633 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
634 spin_unlock(&dentry->d_lock);
640 spin_unlock(&dentry->d_lock);
643 EXPORT_SYMBOL(d_invalidate);
645 /* This must be called with d_lock held */
646 static inline void __dget_dlock(struct dentry *dentry)
648 dentry->d_lockref.count++;
651 static inline void __dget(struct dentry *dentry)
653 lockref_get(&dentry->d_lockref);
656 struct dentry *dget_parent(struct dentry *dentry)
662 * Do optimistic parent lookup without any
666 ret = ACCESS_ONCE(dentry->d_parent);
667 gotref = lockref_get_not_zero(&ret->d_lockref);
669 if (likely(gotref)) {
670 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
677 * Don't need rcu_dereference because we re-check it was correct under
681 ret = dentry->d_parent;
682 spin_lock(&ret->d_lock);
683 if (unlikely(ret != dentry->d_parent)) {
684 spin_unlock(&ret->d_lock);
689 BUG_ON(!ret->d_lockref.count);
690 ret->d_lockref.count++;
691 spin_unlock(&ret->d_lock);
694 EXPORT_SYMBOL(dget_parent);
697 * d_find_alias - grab a hashed alias of inode
698 * @inode: inode in question
699 * @want_discon: flag, used by d_splice_alias, to request
700 * that only a DISCONNECTED alias be returned.
702 * If inode has a hashed alias, or is a directory and has any alias,
703 * acquire the reference to alias and return it. Otherwise return NULL.
704 * Notice that if inode is a directory there can be only one alias and
705 * it can be unhashed only if it has no children, or if it is the root
708 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
709 * any other hashed alias over that one unless @want_discon is set,
710 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
712 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
714 struct dentry *alias, *discon_alias;
718 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
719 spin_lock(&alias->d_lock);
720 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
721 if (IS_ROOT(alias) &&
722 (alias->d_flags & DCACHE_DISCONNECTED)) {
723 discon_alias = alias;
724 } else if (!want_discon) {
726 spin_unlock(&alias->d_lock);
730 spin_unlock(&alias->d_lock);
733 alias = discon_alias;
734 spin_lock(&alias->d_lock);
735 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
736 if (IS_ROOT(alias) &&
737 (alias->d_flags & DCACHE_DISCONNECTED)) {
739 spin_unlock(&alias->d_lock);
743 spin_unlock(&alias->d_lock);
749 struct dentry *d_find_alias(struct inode *inode)
751 struct dentry *de = NULL;
753 if (!hlist_empty(&inode->i_dentry)) {
754 spin_lock(&inode->i_lock);
755 de = __d_find_alias(inode, 0);
756 spin_unlock(&inode->i_lock);
760 EXPORT_SYMBOL(d_find_alias);
763 * Try to kill dentries associated with this inode.
764 * WARNING: you must own a reference to inode.
766 void d_prune_aliases(struct inode *inode)
768 struct dentry *dentry;
770 spin_lock(&inode->i_lock);
771 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
772 spin_lock(&dentry->d_lock);
773 if (!dentry->d_lockref.count) {
775 * inform the fs via d_prune that this dentry
776 * is about to be unhashed and destroyed.
778 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
780 dentry->d_op->d_prune(dentry);
782 __dget_dlock(dentry);
784 spin_unlock(&dentry->d_lock);
785 spin_unlock(&inode->i_lock);
789 spin_unlock(&dentry->d_lock);
791 spin_unlock(&inode->i_lock);
793 EXPORT_SYMBOL(d_prune_aliases);
795 static void shrink_dentry_list(struct list_head *list)
797 struct dentry *dentry, *parent;
801 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
802 if (&dentry->d_lru == list)
806 * Get the dentry lock, and re-verify that the dentry is
807 * this on the shrinking list. If it is, we know that
808 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
810 spin_lock(&dentry->d_lock);
811 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
812 spin_unlock(&dentry->d_lock);
817 * The dispose list is isolated and dentries are not accounted
818 * to the LRU here, so we can simply remove it from the list
819 * here regardless of whether it is referenced or not.
821 d_shrink_del(dentry);
824 * We found an inuse dentry which was not removed from
825 * the LRU because of laziness during lookup. Do not free it.
827 if ((int)dentry->d_lockref.count > 0) {
828 spin_unlock(&dentry->d_lock);
833 parent = dentry_kill(dentry, 0);
835 * If dentry_kill returns NULL, we have nothing more to do.
841 if (unlikely(parent == dentry)) {
843 * trylocks have failed and d_lock has been held the
844 * whole time, so it could not have been added to any
845 * other lists. Just add it back to the shrink list.
848 d_shrink_add(dentry, list);
849 spin_unlock(&dentry->d_lock);
853 * We need to prune ancestors too. This is necessary to prevent
854 * quadratic behavior of shrink_dcache_parent(), but is also
855 * expected to be beneficial in reducing dentry cache
859 while (dentry && !lockref_put_or_lock(&dentry->d_lockref))
860 dentry = dentry_kill(dentry, 1);
866 static enum lru_status
867 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
869 struct list_head *freeable = arg;
870 struct dentry *dentry = container_of(item, struct dentry, d_lru);
874 * we are inverting the lru lock/dentry->d_lock here,
875 * so use a trylock. If we fail to get the lock, just skip
878 if (!spin_trylock(&dentry->d_lock))
882 * Referenced dentries are still in use. If they have active
883 * counts, just remove them from the LRU. Otherwise give them
884 * another pass through the LRU.
886 if (dentry->d_lockref.count) {
887 d_lru_isolate(dentry);
888 spin_unlock(&dentry->d_lock);
892 if (dentry->d_flags & DCACHE_REFERENCED) {
893 dentry->d_flags &= ~DCACHE_REFERENCED;
894 spin_unlock(&dentry->d_lock);
897 * The list move itself will be made by the common LRU code. At
898 * this point, we've dropped the dentry->d_lock but keep the
899 * lru lock. This is safe to do, since every list movement is
900 * protected by the lru lock even if both locks are held.
902 * This is guaranteed by the fact that all LRU management
903 * functions are intermediated by the LRU API calls like
904 * list_lru_add and list_lru_del. List movement in this file
905 * only ever occur through this functions or through callbacks
906 * like this one, that are called from the LRU API.
908 * The only exceptions to this are functions like
909 * shrink_dentry_list, and code that first checks for the
910 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
911 * operating only with stack provided lists after they are
912 * properly isolated from the main list. It is thus, always a
918 d_lru_shrink_move(dentry, freeable);
919 spin_unlock(&dentry->d_lock);
925 * prune_dcache_sb - shrink the dcache
927 * @nr_to_scan : number of entries to try to free
928 * @nid: which node to scan for freeable entities
930 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
931 * done when we need more memory an called from the superblock shrinker
934 * This function may fail to free any resources if all the dentries are in
937 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
943 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
944 &dispose, &nr_to_scan);
945 shrink_dentry_list(&dispose);
949 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
950 spinlock_t *lru_lock, void *arg)
952 struct list_head *freeable = arg;
953 struct dentry *dentry = container_of(item, struct dentry, d_lru);
956 * we are inverting the lru lock/dentry->d_lock here,
957 * so use a trylock. If we fail to get the lock, just skip
960 if (!spin_trylock(&dentry->d_lock))
963 d_lru_shrink_move(dentry, freeable);
964 spin_unlock(&dentry->d_lock);
971 * shrink_dcache_sb - shrink dcache for a superblock
974 * Shrink the dcache for the specified super block. This is used to free
975 * the dcache before unmounting a file system.
977 void shrink_dcache_sb(struct super_block *sb)
984 freed = list_lru_walk(&sb->s_dentry_lru,
985 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
987 this_cpu_sub(nr_dentry_unused, freed);
988 shrink_dentry_list(&dispose);
991 EXPORT_SYMBOL(shrink_dcache_sb);
994 * enum d_walk_ret - action to talke during tree walk
995 * @D_WALK_CONTINUE: contrinue walk
996 * @D_WALK_QUIT: quit walk
997 * @D_WALK_NORETRY: quit when retry is needed
998 * @D_WALK_SKIP: skip this dentry and its children
1008 * d_walk - walk the dentry tree
1009 * @parent: start of walk
1010 * @data: data passed to @enter() and @finish()
1011 * @enter: callback when first entering the dentry
1012 * @finish: callback when successfully finished the walk
1014 * The @enter() and @finish() callbacks are called with d_lock held.
1016 static void d_walk(struct dentry *parent, void *data,
1017 enum d_walk_ret (*enter)(void *, struct dentry *),
1018 void (*finish)(void *))
1020 struct dentry *this_parent;
1021 struct list_head *next;
1023 enum d_walk_ret ret;
1027 read_seqbegin_or_lock(&rename_lock, &seq);
1028 this_parent = parent;
1029 spin_lock(&this_parent->d_lock);
1031 ret = enter(data, this_parent);
1033 case D_WALK_CONTINUE:
1038 case D_WALK_NORETRY:
1043 next = this_parent->d_subdirs.next;
1045 while (next != &this_parent->d_subdirs) {
1046 struct list_head *tmp = next;
1047 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1050 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1052 ret = enter(data, dentry);
1054 case D_WALK_CONTINUE:
1057 spin_unlock(&dentry->d_lock);
1059 case D_WALK_NORETRY:
1063 spin_unlock(&dentry->d_lock);
1067 if (!list_empty(&dentry->d_subdirs)) {
1068 spin_unlock(&this_parent->d_lock);
1069 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1070 this_parent = dentry;
1071 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1074 spin_unlock(&dentry->d_lock);
1077 * All done at this level ... ascend and resume the search.
1079 if (this_parent != parent) {
1080 struct dentry *child = this_parent;
1081 this_parent = child->d_parent;
1084 spin_unlock(&child->d_lock);
1085 spin_lock(&this_parent->d_lock);
1088 * might go back up the wrong parent if we have had a rename
1091 if (this_parent != child->d_parent ||
1092 (child->d_flags & DCACHE_DENTRY_KILLED) ||
1093 need_seqretry(&rename_lock, seq)) {
1094 spin_unlock(&this_parent->d_lock);
1099 next = child->d_u.d_child.next;
1102 if (need_seqretry(&rename_lock, seq)) {
1103 spin_unlock(&this_parent->d_lock);
1110 spin_unlock(&this_parent->d_lock);
1111 done_seqretry(&rename_lock, seq);
1122 * Search for at least 1 mount point in the dentry's subdirs.
1123 * We descend to the next level whenever the d_subdirs
1124 * list is non-empty and continue searching.
1127 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1130 if (d_mountpoint(dentry)) {
1134 return D_WALK_CONTINUE;
1138 * have_submounts - check for mounts over a dentry
1139 * @parent: dentry to check.
1141 * Return true if the parent or its subdirectories contain
1144 int have_submounts(struct dentry *parent)
1148 d_walk(parent, &ret, check_mount, NULL);
1152 EXPORT_SYMBOL(have_submounts);
1155 * Called by mount code to set a mountpoint and check if the mountpoint is
1156 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1157 * subtree can become unreachable).
1159 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1160 * this reason take rename_lock and d_lock on dentry and ancestors.
1162 int d_set_mounted(struct dentry *dentry)
1166 write_seqlock(&rename_lock);
1167 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1168 /* Need exclusion wrt. check_submounts_and_drop() */
1169 spin_lock(&p->d_lock);
1170 if (unlikely(d_unhashed(p))) {
1171 spin_unlock(&p->d_lock);
1174 spin_unlock(&p->d_lock);
1176 spin_lock(&dentry->d_lock);
1177 if (!d_unlinked(dentry)) {
1178 dentry->d_flags |= DCACHE_MOUNTED;
1181 spin_unlock(&dentry->d_lock);
1183 write_sequnlock(&rename_lock);
1188 * Search the dentry child list of the specified parent,
1189 * and move any unused dentries to the end of the unused
1190 * list for prune_dcache(). We descend to the next level
1191 * whenever the d_subdirs list is non-empty and continue
1194 * It returns zero iff there are no unused children,
1195 * otherwise it returns the number of children moved to
1196 * the end of the unused list. This may not be the total
1197 * number of unused children, because select_parent can
1198 * drop the lock and return early due to latency
1202 struct select_data {
1203 struct dentry *start;
1204 struct list_head dispose;
1208 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1210 struct select_data *data = _data;
1211 enum d_walk_ret ret = D_WALK_CONTINUE;
1213 if (data->start == dentry)
1216 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1219 if (dentry->d_flags & DCACHE_LRU_LIST)
1221 if (!dentry->d_lockref.count) {
1222 d_shrink_add(dentry, &data->dispose);
1227 * We can return to the caller if we have found some (this
1228 * ensures forward progress). We'll be coming back to find
1231 if (!list_empty(&data->dispose))
1232 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1238 * shrink_dcache_parent - prune dcache
1239 * @parent: parent of entries to prune
1241 * Prune the dcache to remove unused children of the parent dentry.
1243 void shrink_dcache_parent(struct dentry *parent)
1246 struct select_data data;
1248 INIT_LIST_HEAD(&data.dispose);
1249 data.start = parent;
1252 d_walk(parent, &data, select_collect, NULL);
1256 shrink_dentry_list(&data.dispose);
1260 EXPORT_SYMBOL(shrink_dcache_parent);
1262 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1264 /* it has busy descendents; complain about those instead */
1265 if (!list_empty(&dentry->d_subdirs))
1266 return D_WALK_CONTINUE;
1268 /* root with refcount 1 is fine */
1269 if (dentry == _data && dentry->d_lockref.count == 1)
1270 return D_WALK_CONTINUE;
1272 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1273 " still in use (%d) [unmount of %s %s]\n",
1276 dentry->d_inode->i_ino : 0UL,
1278 dentry->d_lockref.count,
1279 dentry->d_sb->s_type->name,
1280 dentry->d_sb->s_id);
1282 return D_WALK_CONTINUE;
1285 static void do_one_tree(struct dentry *dentry)
1287 shrink_dcache_parent(dentry);
1288 d_walk(dentry, dentry, umount_check, NULL);
1294 * destroy the dentries attached to a superblock on unmounting
1296 void shrink_dcache_for_umount(struct super_block *sb)
1298 struct dentry *dentry;
1300 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1302 dentry = sb->s_root;
1304 do_one_tree(dentry);
1306 while (!hlist_bl_empty(&sb->s_anon)) {
1307 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1308 do_one_tree(dentry);
1312 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1314 struct select_data *data = _data;
1316 if (d_mountpoint(dentry)) {
1317 data->found = -EBUSY;
1321 return select_collect(_data, dentry);
1324 static void check_and_drop(void *_data)
1326 struct select_data *data = _data;
1328 if (d_mountpoint(data->start))
1329 data->found = -EBUSY;
1331 __d_drop(data->start);
1335 * check_submounts_and_drop - prune dcache, check for submounts and drop
1337 * All done as a single atomic operation relative to has_unlinked_ancestor().
1338 * Returns 0 if successfully unhashed @parent. If there were submounts then
1341 * @dentry: dentry to prune and drop
1343 int check_submounts_and_drop(struct dentry *dentry)
1347 /* Negative dentries can be dropped without further checks */
1348 if (!dentry->d_inode) {
1354 struct select_data data;
1356 INIT_LIST_HEAD(&data.dispose);
1357 data.start = dentry;
1360 d_walk(dentry, &data, check_and_collect, check_and_drop);
1363 if (!list_empty(&data.dispose))
1364 shrink_dentry_list(&data.dispose);
1375 EXPORT_SYMBOL(check_submounts_and_drop);
1378 * __d_alloc - allocate a dcache entry
1379 * @sb: filesystem it will belong to
1380 * @name: qstr of the name
1382 * Allocates a dentry. It returns %NULL if there is insufficient memory
1383 * available. On a success the dentry is returned. The name passed in is
1384 * copied and the copy passed in may be reused after this call.
1387 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1389 struct dentry *dentry;
1392 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1397 * We guarantee that the inline name is always NUL-terminated.
1398 * This way the memcpy() done by the name switching in rename
1399 * will still always have a NUL at the end, even if we might
1400 * be overwriting an internal NUL character
1402 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1403 if (name->len > DNAME_INLINE_LEN-1) {
1404 dname = kmalloc(name->len + 1, GFP_KERNEL);
1406 kmem_cache_free(dentry_cache, dentry);
1410 dname = dentry->d_iname;
1413 dentry->d_name.len = name->len;
1414 dentry->d_name.hash = name->hash;
1415 memcpy(dname, name->name, name->len);
1416 dname[name->len] = 0;
1418 /* Make sure we always see the terminating NUL character */
1420 dentry->d_name.name = dname;
1422 dentry->d_lockref.count = 1;
1423 dentry->d_flags = 0;
1424 spin_lock_init(&dentry->d_lock);
1425 seqcount_init(&dentry->d_seq);
1426 dentry->d_inode = NULL;
1427 dentry->d_parent = dentry;
1429 dentry->d_op = NULL;
1430 dentry->d_fsdata = NULL;
1431 INIT_HLIST_BL_NODE(&dentry->d_hash);
1432 INIT_LIST_HEAD(&dentry->d_lru);
1433 INIT_LIST_HEAD(&dentry->d_subdirs);
1434 INIT_HLIST_NODE(&dentry->d_alias);
1435 INIT_LIST_HEAD(&dentry->d_u.d_child);
1436 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1438 this_cpu_inc(nr_dentry);
1444 * d_alloc - allocate a dcache entry
1445 * @parent: parent of entry to allocate
1446 * @name: qstr of the name
1448 * Allocates a dentry. It returns %NULL if there is insufficient memory
1449 * available. On a success the dentry is returned. The name passed in is
1450 * copied and the copy passed in may be reused after this call.
1452 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1454 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1458 spin_lock(&parent->d_lock);
1460 * don't need child lock because it is not subject
1461 * to concurrency here
1463 __dget_dlock(parent);
1464 dentry->d_parent = parent;
1465 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1466 spin_unlock(&parent->d_lock);
1470 EXPORT_SYMBOL(d_alloc);
1473 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1474 * @sb: the superblock
1475 * @name: qstr of the name
1477 * For a filesystem that just pins its dentries in memory and never
1478 * performs lookups at all, return an unhashed IS_ROOT dentry.
1480 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1482 return __d_alloc(sb, name);
1484 EXPORT_SYMBOL(d_alloc_pseudo);
1486 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1491 q.len = strlen(name);
1492 q.hash = full_name_hash(q.name, q.len);
1493 return d_alloc(parent, &q);
1495 EXPORT_SYMBOL(d_alloc_name);
1497 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1499 WARN_ON_ONCE(dentry->d_op);
1500 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1502 DCACHE_OP_REVALIDATE |
1503 DCACHE_OP_WEAK_REVALIDATE |
1504 DCACHE_OP_DELETE ));
1509 dentry->d_flags |= DCACHE_OP_HASH;
1511 dentry->d_flags |= DCACHE_OP_COMPARE;
1512 if (op->d_revalidate)
1513 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1514 if (op->d_weak_revalidate)
1515 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1517 dentry->d_flags |= DCACHE_OP_DELETE;
1519 dentry->d_flags |= DCACHE_OP_PRUNE;
1522 EXPORT_SYMBOL(d_set_d_op);
1524 static unsigned d_flags_for_inode(struct inode *inode)
1526 unsigned add_flags = DCACHE_FILE_TYPE;
1529 return DCACHE_MISS_TYPE;
1531 if (S_ISDIR(inode->i_mode)) {
1532 add_flags = DCACHE_DIRECTORY_TYPE;
1533 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1534 if (unlikely(!inode->i_op->lookup))
1535 add_flags = DCACHE_AUTODIR_TYPE;
1537 inode->i_opflags |= IOP_LOOKUP;
1539 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1540 if (unlikely(inode->i_op->follow_link))
1541 add_flags = DCACHE_SYMLINK_TYPE;
1543 inode->i_opflags |= IOP_NOFOLLOW;
1546 if (unlikely(IS_AUTOMOUNT(inode)))
1547 add_flags |= DCACHE_NEED_AUTOMOUNT;
1551 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1553 unsigned add_flags = d_flags_for_inode(inode);
1555 spin_lock(&dentry->d_lock);
1556 __d_set_type(dentry, add_flags);
1558 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1559 dentry->d_inode = inode;
1560 dentry_rcuwalk_barrier(dentry);
1561 spin_unlock(&dentry->d_lock);
1562 fsnotify_d_instantiate(dentry, inode);
1566 * d_instantiate - fill in inode information for a dentry
1567 * @entry: dentry to complete
1568 * @inode: inode to attach to this dentry
1570 * Fill in inode information in the entry.
1572 * This turns negative dentries into productive full members
1575 * NOTE! This assumes that the inode count has been incremented
1576 * (or otherwise set) by the caller to indicate that it is now
1577 * in use by the dcache.
1580 void d_instantiate(struct dentry *entry, struct inode * inode)
1582 BUG_ON(!hlist_unhashed(&entry->d_alias));
1584 spin_lock(&inode->i_lock);
1585 __d_instantiate(entry, inode);
1587 spin_unlock(&inode->i_lock);
1588 security_d_instantiate(entry, inode);
1590 EXPORT_SYMBOL(d_instantiate);
1593 * d_instantiate_unique - instantiate a non-aliased dentry
1594 * @entry: dentry to instantiate
1595 * @inode: inode to attach to this dentry
1597 * Fill in inode information in the entry. On success, it returns NULL.
1598 * If an unhashed alias of "entry" already exists, then we return the
1599 * aliased dentry instead and drop one reference to inode.
1601 * Note that in order to avoid conflicts with rename() etc, the caller
1602 * had better be holding the parent directory semaphore.
1604 * This also assumes that the inode count has been incremented
1605 * (or otherwise set) by the caller to indicate that it is now
1606 * in use by the dcache.
1608 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1609 struct inode *inode)
1611 struct dentry *alias;
1612 int len = entry->d_name.len;
1613 const char *name = entry->d_name.name;
1614 unsigned int hash = entry->d_name.hash;
1617 __d_instantiate(entry, NULL);
1621 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1623 * Don't need alias->d_lock here, because aliases with
1624 * d_parent == entry->d_parent are not subject to name or
1625 * parent changes, because the parent inode i_mutex is held.
1627 if (alias->d_name.hash != hash)
1629 if (alias->d_parent != entry->d_parent)
1631 if (alias->d_name.len != len)
1633 if (dentry_cmp(alias, name, len))
1639 __d_instantiate(entry, inode);
1643 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1645 struct dentry *result;
1647 BUG_ON(!hlist_unhashed(&entry->d_alias));
1650 spin_lock(&inode->i_lock);
1651 result = __d_instantiate_unique(entry, inode);
1653 spin_unlock(&inode->i_lock);
1656 security_d_instantiate(entry, inode);
1660 BUG_ON(!d_unhashed(result));
1665 EXPORT_SYMBOL(d_instantiate_unique);
1668 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1669 * @entry: dentry to complete
1670 * @inode: inode to attach to this dentry
1672 * Fill in inode information in the entry. If a directory alias is found, then
1673 * return an error (and drop inode). Together with d_materialise_unique() this
1674 * guarantees that a directory inode may never have more than one alias.
1676 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1678 BUG_ON(!hlist_unhashed(&entry->d_alias));
1680 spin_lock(&inode->i_lock);
1681 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1682 spin_unlock(&inode->i_lock);
1686 __d_instantiate(entry, inode);
1687 spin_unlock(&inode->i_lock);
1688 security_d_instantiate(entry, inode);
1692 EXPORT_SYMBOL(d_instantiate_no_diralias);
1694 struct dentry *d_make_root(struct inode *root_inode)
1696 struct dentry *res = NULL;
1699 static const struct qstr name = QSTR_INIT("/", 1);
1701 res = __d_alloc(root_inode->i_sb, &name);
1703 d_instantiate(res, root_inode);
1709 EXPORT_SYMBOL(d_make_root);
1711 static struct dentry * __d_find_any_alias(struct inode *inode)
1713 struct dentry *alias;
1715 if (hlist_empty(&inode->i_dentry))
1717 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1723 * d_find_any_alias - find any alias for a given inode
1724 * @inode: inode to find an alias for
1726 * If any aliases exist for the given inode, take and return a
1727 * reference for one of them. If no aliases exist, return %NULL.
1729 struct dentry *d_find_any_alias(struct inode *inode)
1733 spin_lock(&inode->i_lock);
1734 de = __d_find_any_alias(inode);
1735 spin_unlock(&inode->i_lock);
1738 EXPORT_SYMBOL(d_find_any_alias);
1741 * d_obtain_alias - find or allocate a dentry for a given inode
1742 * @inode: inode to allocate the dentry for
1744 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1745 * similar open by handle operations. The returned dentry may be anonymous,
1746 * or may have a full name (if the inode was already in the cache).
1748 * When called on a directory inode, we must ensure that the inode only ever
1749 * has one dentry. If a dentry is found, that is returned instead of
1750 * allocating a new one.
1752 * On successful return, the reference to the inode has been transferred
1753 * to the dentry. In case of an error the reference on the inode is released.
1754 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1755 * be passed in and will be the error will be propagate to the return value,
1756 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1758 struct dentry *d_obtain_alias(struct inode *inode)
1760 static const struct qstr anonstring = QSTR_INIT("/", 1);
1766 return ERR_PTR(-ESTALE);
1768 return ERR_CAST(inode);
1770 res = d_find_any_alias(inode);
1774 tmp = __d_alloc(inode->i_sb, &anonstring);
1776 res = ERR_PTR(-ENOMEM);
1780 spin_lock(&inode->i_lock);
1781 res = __d_find_any_alias(inode);
1783 spin_unlock(&inode->i_lock);
1788 /* attach a disconnected dentry */
1789 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1791 spin_lock(&tmp->d_lock);
1792 tmp->d_inode = inode;
1793 tmp->d_flags |= add_flags;
1794 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1795 hlist_bl_lock(&tmp->d_sb->s_anon);
1796 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1797 hlist_bl_unlock(&tmp->d_sb->s_anon);
1798 spin_unlock(&tmp->d_lock);
1799 spin_unlock(&inode->i_lock);
1800 security_d_instantiate(tmp, inode);
1805 if (res && !IS_ERR(res))
1806 security_d_instantiate(res, inode);
1810 EXPORT_SYMBOL(d_obtain_alias);
1813 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1814 * @inode: the inode which may have a disconnected dentry
1815 * @dentry: a negative dentry which we want to point to the inode.
1817 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1818 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1819 * and return it, else simply d_add the inode to the dentry and return NULL.
1821 * This is needed in the lookup routine of any filesystem that is exportable
1822 * (via knfsd) so that we can build dcache paths to directories effectively.
1824 * If a dentry was found and moved, then it is returned. Otherwise NULL
1825 * is returned. This matches the expected return value of ->lookup.
1827 * Cluster filesystems may call this function with a negative, hashed dentry.
1828 * In that case, we know that the inode will be a regular file, and also this
1829 * will only occur during atomic_open. So we need to check for the dentry
1830 * being already hashed only in the final case.
1832 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1834 struct dentry *new = NULL;
1837 return ERR_CAST(inode);
1839 if (inode && S_ISDIR(inode->i_mode)) {
1840 spin_lock(&inode->i_lock);
1841 new = __d_find_alias(inode, 1);
1843 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1844 spin_unlock(&inode->i_lock);
1845 security_d_instantiate(new, inode);
1846 d_move(new, dentry);
1849 /* already taking inode->i_lock, so d_add() by hand */
1850 __d_instantiate(dentry, inode);
1851 spin_unlock(&inode->i_lock);
1852 security_d_instantiate(dentry, inode);
1856 d_instantiate(dentry, inode);
1857 if (d_unhashed(dentry))
1862 EXPORT_SYMBOL(d_splice_alias);
1865 * d_add_ci - lookup or allocate new dentry with case-exact name
1866 * @inode: the inode case-insensitive lookup has found
1867 * @dentry: the negative dentry that was passed to the parent's lookup func
1868 * @name: the case-exact name to be associated with the returned dentry
1870 * This is to avoid filling the dcache with case-insensitive names to the
1871 * same inode, only the actual correct case is stored in the dcache for
1872 * case-insensitive filesystems.
1874 * For a case-insensitive lookup match and if the the case-exact dentry
1875 * already exists in in the dcache, use it and return it.
1877 * If no entry exists with the exact case name, allocate new dentry with
1878 * the exact case, and return the spliced entry.
1880 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1883 struct dentry *found;
1887 * First check if a dentry matching the name already exists,
1888 * if not go ahead and create it now.
1890 found = d_hash_and_lookup(dentry->d_parent, name);
1891 if (unlikely(IS_ERR(found)))
1894 new = d_alloc(dentry->d_parent, name);
1896 found = ERR_PTR(-ENOMEM);
1900 found = d_splice_alias(inode, new);
1909 * If a matching dentry exists, and it's not negative use it.
1911 * Decrement the reference count to balance the iget() done
1914 if (found->d_inode) {
1915 if (unlikely(found->d_inode != inode)) {
1916 /* This can't happen because bad inodes are unhashed. */
1917 BUG_ON(!is_bad_inode(inode));
1918 BUG_ON(!is_bad_inode(found->d_inode));
1925 * Negative dentry: instantiate it unless the inode is a directory and
1926 * already has a dentry.
1928 new = d_splice_alias(inode, found);
1939 EXPORT_SYMBOL(d_add_ci);
1942 * Do the slow-case of the dentry name compare.
1944 * Unlike the dentry_cmp() function, we need to atomically
1945 * load the name and length information, so that the
1946 * filesystem can rely on them, and can use the 'name' and
1947 * 'len' information without worrying about walking off the
1948 * end of memory etc.
1950 * Thus the read_seqcount_retry() and the "duplicate" info
1951 * in arguments (the low-level filesystem should not look
1952 * at the dentry inode or name contents directly, since
1953 * rename can change them while we're in RCU mode).
1955 enum slow_d_compare {
1961 static noinline enum slow_d_compare slow_dentry_cmp(
1962 const struct dentry *parent,
1963 struct dentry *dentry,
1965 const struct qstr *name)
1967 int tlen = dentry->d_name.len;
1968 const char *tname = dentry->d_name.name;
1970 if (read_seqcount_retry(&dentry->d_seq, seq)) {
1972 return D_COMP_SEQRETRY;
1974 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
1975 return D_COMP_NOMATCH;
1980 * __d_lookup_rcu - search for a dentry (racy, store-free)
1981 * @parent: parent dentry
1982 * @name: qstr of name we wish to find
1983 * @seqp: returns d_seq value at the point where the dentry was found
1984 * Returns: dentry, or NULL
1986 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1987 * resolution (store-free path walking) design described in
1988 * Documentation/filesystems/path-lookup.txt.
1990 * This is not to be used outside core vfs.
1992 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1993 * held, and rcu_read_lock held. The returned dentry must not be stored into
1994 * without taking d_lock and checking d_seq sequence count against @seq
1997 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2000 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2001 * the returned dentry, so long as its parent's seqlock is checked after the
2002 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2003 * is formed, giving integrity down the path walk.
2005 * NOTE! The caller *has* to check the resulting dentry against the sequence
2006 * number we've returned before using any of the resulting dentry state!
2008 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2009 const struct qstr *name,
2012 u64 hashlen = name->hash_len;
2013 const unsigned char *str = name->name;
2014 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2015 struct hlist_bl_node *node;
2016 struct dentry *dentry;
2019 * Note: There is significant duplication with __d_lookup_rcu which is
2020 * required to prevent single threaded performance regressions
2021 * especially on architectures where smp_rmb (in seqcounts) are costly.
2022 * Keep the two functions in sync.
2026 * The hash list is protected using RCU.
2028 * Carefully use d_seq when comparing a candidate dentry, to avoid
2029 * races with d_move().
2031 * It is possible that concurrent renames can mess up our list
2032 * walk here and result in missing our dentry, resulting in the
2033 * false-negative result. d_lookup() protects against concurrent
2034 * renames using rename_lock seqlock.
2036 * See Documentation/filesystems/path-lookup.txt for more details.
2038 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2043 * The dentry sequence count protects us from concurrent
2044 * renames, and thus protects parent and name fields.
2046 * The caller must perform a seqcount check in order
2047 * to do anything useful with the returned dentry.
2049 * NOTE! We do a "raw" seqcount_begin here. That means that
2050 * we don't wait for the sequence count to stabilize if it
2051 * is in the middle of a sequence change. If we do the slow
2052 * dentry compare, we will do seqretries until it is stable,
2053 * and if we end up with a successful lookup, we actually
2054 * want to exit RCU lookup anyway.
2056 seq = raw_seqcount_begin(&dentry->d_seq);
2057 if (dentry->d_parent != parent)
2059 if (d_unhashed(dentry))
2062 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2063 if (dentry->d_name.hash != hashlen_hash(hashlen))
2066 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2069 case D_COMP_NOMATCH:
2076 if (dentry->d_name.hash_len != hashlen)
2079 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2086 * d_lookup - search for a dentry
2087 * @parent: parent dentry
2088 * @name: qstr of name we wish to find
2089 * Returns: dentry, or NULL
2091 * d_lookup searches the children of the parent dentry for the name in
2092 * question. If the dentry is found its reference count is incremented and the
2093 * dentry is returned. The caller must use dput to free the entry when it has
2094 * finished using it. %NULL is returned if the dentry does not exist.
2096 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2098 struct dentry *dentry;
2102 seq = read_seqbegin(&rename_lock);
2103 dentry = __d_lookup(parent, name);
2106 } while (read_seqretry(&rename_lock, seq));
2109 EXPORT_SYMBOL(d_lookup);
2112 * __d_lookup - search for a dentry (racy)
2113 * @parent: parent dentry
2114 * @name: qstr of name we wish to find
2115 * Returns: dentry, or NULL
2117 * __d_lookup is like d_lookup, however it may (rarely) return a
2118 * false-negative result due to unrelated rename activity.
2120 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2121 * however it must be used carefully, eg. with a following d_lookup in
2122 * the case of failure.
2124 * __d_lookup callers must be commented.
2126 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2128 unsigned int len = name->len;
2129 unsigned int hash = name->hash;
2130 const unsigned char *str = name->name;
2131 struct hlist_bl_head *b = d_hash(parent, hash);
2132 struct hlist_bl_node *node;
2133 struct dentry *found = NULL;
2134 struct dentry *dentry;
2137 * Note: There is significant duplication with __d_lookup_rcu which is
2138 * required to prevent single threaded performance regressions
2139 * especially on architectures where smp_rmb (in seqcounts) are costly.
2140 * Keep the two functions in sync.
2144 * The hash list is protected using RCU.
2146 * Take d_lock when comparing a candidate dentry, to avoid races
2149 * It is possible that concurrent renames can mess up our list
2150 * walk here and result in missing our dentry, resulting in the
2151 * false-negative result. d_lookup() protects against concurrent
2152 * renames using rename_lock seqlock.
2154 * See Documentation/filesystems/path-lookup.txt for more details.
2158 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2160 if (dentry->d_name.hash != hash)
2163 spin_lock(&dentry->d_lock);
2164 if (dentry->d_parent != parent)
2166 if (d_unhashed(dentry))
2170 * It is safe to compare names since d_move() cannot
2171 * change the qstr (protected by d_lock).
2173 if (parent->d_flags & DCACHE_OP_COMPARE) {
2174 int tlen = dentry->d_name.len;
2175 const char *tname = dentry->d_name.name;
2176 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2179 if (dentry->d_name.len != len)
2181 if (dentry_cmp(dentry, str, len))
2185 dentry->d_lockref.count++;
2187 spin_unlock(&dentry->d_lock);
2190 spin_unlock(&dentry->d_lock);
2198 * d_hash_and_lookup - hash the qstr then search for a dentry
2199 * @dir: Directory to search in
2200 * @name: qstr of name we wish to find
2202 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2204 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2207 * Check for a fs-specific hash function. Note that we must
2208 * calculate the standard hash first, as the d_op->d_hash()
2209 * routine may choose to leave the hash value unchanged.
2211 name->hash = full_name_hash(name->name, name->len);
2212 if (dir->d_flags & DCACHE_OP_HASH) {
2213 int err = dir->d_op->d_hash(dir, name);
2214 if (unlikely(err < 0))
2215 return ERR_PTR(err);
2217 return d_lookup(dir, name);
2219 EXPORT_SYMBOL(d_hash_and_lookup);
2222 * d_validate - verify dentry provided from insecure source (deprecated)
2223 * @dentry: The dentry alleged to be valid child of @dparent
2224 * @dparent: The parent dentry (known to be valid)
2226 * An insecure source has sent us a dentry, here we verify it and dget() it.
2227 * This is used by ncpfs in its readdir implementation.
2228 * Zero is returned in the dentry is invalid.
2230 * This function is slow for big directories, and deprecated, do not use it.
2232 int d_validate(struct dentry *dentry, struct dentry *dparent)
2234 struct dentry *child;
2236 spin_lock(&dparent->d_lock);
2237 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2238 if (dentry == child) {
2239 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2240 __dget_dlock(dentry);
2241 spin_unlock(&dentry->d_lock);
2242 spin_unlock(&dparent->d_lock);
2246 spin_unlock(&dparent->d_lock);
2250 EXPORT_SYMBOL(d_validate);
2253 * When a file is deleted, we have two options:
2254 * - turn this dentry into a negative dentry
2255 * - unhash this dentry and free it.
2257 * Usually, we want to just turn this into
2258 * a negative dentry, but if anybody else is
2259 * currently using the dentry or the inode
2260 * we can't do that and we fall back on removing
2261 * it from the hash queues and waiting for
2262 * it to be deleted later when it has no users
2266 * d_delete - delete a dentry
2267 * @dentry: The dentry to delete
2269 * Turn the dentry into a negative dentry if possible, otherwise
2270 * remove it from the hash queues so it can be deleted later
2273 void d_delete(struct dentry * dentry)
2275 struct inode *inode;
2278 * Are we the only user?
2281 spin_lock(&dentry->d_lock);
2282 inode = dentry->d_inode;
2283 isdir = S_ISDIR(inode->i_mode);
2284 if (dentry->d_lockref.count == 1) {
2285 if (!spin_trylock(&inode->i_lock)) {
2286 spin_unlock(&dentry->d_lock);
2290 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2291 dentry_unlink_inode(dentry);
2292 fsnotify_nameremove(dentry, isdir);
2296 if (!d_unhashed(dentry))
2299 spin_unlock(&dentry->d_lock);
2301 fsnotify_nameremove(dentry, isdir);
2303 EXPORT_SYMBOL(d_delete);
2305 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2307 BUG_ON(!d_unhashed(entry));
2309 entry->d_flags |= DCACHE_RCUACCESS;
2310 hlist_bl_add_head_rcu(&entry->d_hash, b);
2314 static void _d_rehash(struct dentry * entry)
2316 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2320 * d_rehash - add an entry back to the hash
2321 * @entry: dentry to add to the hash
2323 * Adds a dentry to the hash according to its name.
2326 void d_rehash(struct dentry * entry)
2328 spin_lock(&entry->d_lock);
2330 spin_unlock(&entry->d_lock);
2332 EXPORT_SYMBOL(d_rehash);
2335 * dentry_update_name_case - update case insensitive dentry with a new name
2336 * @dentry: dentry to be updated
2339 * Update a case insensitive dentry with new case of name.
2341 * dentry must have been returned by d_lookup with name @name. Old and new
2342 * name lengths must match (ie. no d_compare which allows mismatched name
2345 * Parent inode i_mutex must be held over d_lookup and into this call (to
2346 * keep renames and concurrent inserts, and readdir(2) away).
2348 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2350 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2351 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2353 spin_lock(&dentry->d_lock);
2354 write_seqcount_begin(&dentry->d_seq);
2355 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2356 write_seqcount_end(&dentry->d_seq);
2357 spin_unlock(&dentry->d_lock);
2359 EXPORT_SYMBOL(dentry_update_name_case);
2361 static void switch_names(struct dentry *dentry, struct dentry *target)
2363 if (dname_external(target)) {
2364 if (dname_external(dentry)) {
2366 * Both external: swap the pointers
2368 swap(target->d_name.name, dentry->d_name.name);
2371 * dentry:internal, target:external. Steal target's
2372 * storage and make target internal.
2374 memcpy(target->d_iname, dentry->d_name.name,
2375 dentry->d_name.len + 1);
2376 dentry->d_name.name = target->d_name.name;
2377 target->d_name.name = target->d_iname;
2380 if (dname_external(dentry)) {
2382 * dentry:external, target:internal. Give dentry's
2383 * storage to target and make dentry internal
2385 memcpy(dentry->d_iname, target->d_name.name,
2386 target->d_name.len + 1);
2387 target->d_name.name = dentry->d_name.name;
2388 dentry->d_name.name = dentry->d_iname;
2391 * Both are internal.
2394 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2395 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2396 swap(((long *) &dentry->d_iname)[i],
2397 ((long *) &target->d_iname)[i]);
2401 swap(dentry->d_name.len, target->d_name.len);
2404 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2407 * XXXX: do we really need to take target->d_lock?
2409 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2410 spin_lock(&target->d_parent->d_lock);
2412 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2413 spin_lock(&dentry->d_parent->d_lock);
2414 spin_lock_nested(&target->d_parent->d_lock,
2415 DENTRY_D_LOCK_NESTED);
2417 spin_lock(&target->d_parent->d_lock);
2418 spin_lock_nested(&dentry->d_parent->d_lock,
2419 DENTRY_D_LOCK_NESTED);
2422 if (target < dentry) {
2423 spin_lock_nested(&target->d_lock, 2);
2424 spin_lock_nested(&dentry->d_lock, 3);
2426 spin_lock_nested(&dentry->d_lock, 2);
2427 spin_lock_nested(&target->d_lock, 3);
2431 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2432 struct dentry *target)
2434 if (target->d_parent != dentry->d_parent)
2435 spin_unlock(&dentry->d_parent->d_lock);
2436 if (target->d_parent != target)
2437 spin_unlock(&target->d_parent->d_lock);
2441 * When switching names, the actual string doesn't strictly have to
2442 * be preserved in the target - because we're dropping the target
2443 * anyway. As such, we can just do a simple memcpy() to copy over
2444 * the new name before we switch.
2446 * Note that we have to be a lot more careful about getting the hash
2447 * switched - we have to switch the hash value properly even if it
2448 * then no longer matches the actual (corrupted) string of the target.
2449 * The hash value has to match the hash queue that the dentry is on..
2452 * __d_move - move a dentry
2453 * @dentry: entry to move
2454 * @target: new dentry
2455 * @exchange: exchange the two dentries
2457 * Update the dcache to reflect the move of a file name. Negative
2458 * dcache entries should not be moved in this way. Caller must hold
2459 * rename_lock, the i_mutex of the source and target directories,
2460 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2462 static void __d_move(struct dentry *dentry, struct dentry *target,
2465 if (!dentry->d_inode)
2466 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2468 BUG_ON(d_ancestor(dentry, target));
2469 BUG_ON(d_ancestor(target, dentry));
2471 dentry_lock_for_move(dentry, target);
2473 write_seqcount_begin(&dentry->d_seq);
2474 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2476 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2479 * Move the dentry to the target hash queue. Don't bother checking
2480 * for the same hash queue because of how unlikely it is.
2483 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2486 * Unhash the target (d_delete() is not usable here). If exchanging
2487 * the two dentries, then rehash onto the other's hash queue.
2492 d_hash(dentry->d_parent, dentry->d_name.hash));
2495 list_del(&dentry->d_u.d_child);
2496 list_del(&target->d_u.d_child);
2498 /* Switch the names.. */
2499 switch_names(dentry, target);
2500 swap(dentry->d_name.hash, target->d_name.hash);
2502 /* ... and switch the parents */
2503 if (IS_ROOT(dentry)) {
2504 dentry->d_parent = target->d_parent;
2505 target->d_parent = target;
2506 INIT_LIST_HEAD(&target->d_u.d_child);
2508 swap(dentry->d_parent, target->d_parent);
2510 /* And add them back to the (new) parent lists */
2511 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2514 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2516 write_seqcount_end(&target->d_seq);
2517 write_seqcount_end(&dentry->d_seq);
2519 dentry_unlock_parents_for_move(dentry, target);
2521 fsnotify_d_move(target);
2522 spin_unlock(&target->d_lock);
2523 fsnotify_d_move(dentry);
2524 spin_unlock(&dentry->d_lock);
2528 * d_move - move a dentry
2529 * @dentry: entry to move
2530 * @target: new dentry
2532 * Update the dcache to reflect the move of a file name. Negative
2533 * dcache entries should not be moved in this way. See the locking
2534 * requirements for __d_move.
2536 void d_move(struct dentry *dentry, struct dentry *target)
2538 write_seqlock(&rename_lock);
2539 __d_move(dentry, target, false);
2540 write_sequnlock(&rename_lock);
2542 EXPORT_SYMBOL(d_move);
2545 * d_exchange - exchange two dentries
2546 * @dentry1: first dentry
2547 * @dentry2: second dentry
2549 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2551 write_seqlock(&rename_lock);
2553 WARN_ON(!dentry1->d_inode);
2554 WARN_ON(!dentry2->d_inode);
2555 WARN_ON(IS_ROOT(dentry1));
2556 WARN_ON(IS_ROOT(dentry2));
2558 __d_move(dentry1, dentry2, true);
2560 write_sequnlock(&rename_lock);
2564 * d_ancestor - search for an ancestor
2565 * @p1: ancestor dentry
2568 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2569 * an ancestor of p2, else NULL.
2571 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2575 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2576 if (p->d_parent == p1)
2583 * This helper attempts to cope with remotely renamed directories
2585 * It assumes that the caller is already holding
2586 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2588 * Note: If ever the locking in lock_rename() changes, then please
2589 * remember to update this too...
2591 static struct dentry *__d_unalias(struct inode *inode,
2592 struct dentry *dentry, struct dentry *alias)
2594 struct mutex *m1 = NULL, *m2 = NULL;
2595 struct dentry *ret = ERR_PTR(-EBUSY);
2597 /* If alias and dentry share a parent, then no extra locks required */
2598 if (alias->d_parent == dentry->d_parent)
2601 /* See lock_rename() */
2602 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2604 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2605 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2607 m2 = &alias->d_parent->d_inode->i_mutex;
2609 if (likely(!d_mountpoint(alias))) {
2610 __d_move(alias, dentry, false);
2614 spin_unlock(&inode->i_lock);
2623 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2624 * named dentry in place of the dentry to be replaced.
2625 * returns with anon->d_lock held!
2627 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2629 struct dentry *dparent;
2631 dentry_lock_for_move(anon, dentry);
2633 write_seqcount_begin(&dentry->d_seq);
2634 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2636 dparent = dentry->d_parent;
2638 switch_names(dentry, anon);
2639 swap(dentry->d_name.hash, anon->d_name.hash);
2641 dentry->d_parent = dentry;
2642 list_del_init(&dentry->d_u.d_child);
2643 anon->d_parent = dparent;
2644 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2646 write_seqcount_end(&dentry->d_seq);
2647 write_seqcount_end(&anon->d_seq);
2649 dentry_unlock_parents_for_move(anon, dentry);
2650 spin_unlock(&dentry->d_lock);
2652 /* anon->d_lock still locked, returns locked */
2656 * d_materialise_unique - introduce an inode into the tree
2657 * @dentry: candidate dentry
2658 * @inode: inode to bind to the dentry, to which aliases may be attached
2660 * Introduces an dentry into the tree, substituting an extant disconnected
2661 * root directory alias in its place if there is one. Caller must hold the
2662 * i_mutex of the parent directory.
2664 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2666 struct dentry *actual;
2668 BUG_ON(!d_unhashed(dentry));
2672 __d_instantiate(dentry, NULL);
2677 spin_lock(&inode->i_lock);
2679 if (S_ISDIR(inode->i_mode)) {
2680 struct dentry *alias;
2682 /* Does an aliased dentry already exist? */
2683 alias = __d_find_alias(inode, 0);
2686 write_seqlock(&rename_lock);
2688 if (d_ancestor(alias, dentry)) {
2689 /* Check for loops */
2690 actual = ERR_PTR(-ELOOP);
2691 spin_unlock(&inode->i_lock);
2692 } else if (IS_ROOT(alias)) {
2693 /* Is this an anonymous mountpoint that we
2694 * could splice into our tree? */
2695 __d_materialise_dentry(dentry, alias);
2696 write_sequnlock(&rename_lock);
2700 /* Nope, but we must(!) avoid directory
2701 * aliasing. This drops inode->i_lock */
2702 actual = __d_unalias(inode, dentry, alias);
2704 write_sequnlock(&rename_lock);
2705 if (IS_ERR(actual)) {
2706 if (PTR_ERR(actual) == -ELOOP)
2707 pr_warn_ratelimited(
2708 "VFS: Lookup of '%s' in %s %s"
2709 " would have caused loop\n",
2710 dentry->d_name.name,
2711 inode->i_sb->s_type->name,
2719 /* Add a unique reference */
2720 actual = __d_instantiate_unique(dentry, inode);
2724 BUG_ON(!d_unhashed(actual));
2726 spin_lock(&actual->d_lock);
2729 spin_unlock(&actual->d_lock);
2730 spin_unlock(&inode->i_lock);
2732 if (actual == dentry) {
2733 security_d_instantiate(dentry, inode);
2740 EXPORT_SYMBOL_GPL(d_materialise_unique);
2742 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2746 return -ENAMETOOLONG;
2748 memcpy(*buffer, str, namelen);
2753 * prepend_name - prepend a pathname in front of current buffer pointer
2754 * @buffer: buffer pointer
2755 * @buflen: allocated length of the buffer
2756 * @name: name string and length qstr structure
2758 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2759 * make sure that either the old or the new name pointer and length are
2760 * fetched. However, there may be mismatch between length and pointer.
2761 * The length cannot be trusted, we need to copy it byte-by-byte until
2762 * the length is reached or a null byte is found. It also prepends "/" at
2763 * the beginning of the name. The sequence number check at the caller will
2764 * retry it again when a d_move() does happen. So any garbage in the buffer
2765 * due to mismatched pointer and length will be discarded.
2767 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2769 const char *dname = ACCESS_ONCE(name->name);
2770 u32 dlen = ACCESS_ONCE(name->len);
2773 *buflen -= dlen + 1;
2775 return -ENAMETOOLONG;
2776 p = *buffer -= dlen + 1;
2788 * prepend_path - Prepend path string to a buffer
2789 * @path: the dentry/vfsmount to report
2790 * @root: root vfsmnt/dentry
2791 * @buffer: pointer to the end of the buffer
2792 * @buflen: pointer to buffer length
2794 * The function will first try to write out the pathname without taking any
2795 * lock other than the RCU read lock to make sure that dentries won't go away.
2796 * It only checks the sequence number of the global rename_lock as any change
2797 * in the dentry's d_seq will be preceded by changes in the rename_lock
2798 * sequence number. If the sequence number had been changed, it will restart
2799 * the whole pathname back-tracing sequence again by taking the rename_lock.
2800 * In this case, there is no need to take the RCU read lock as the recursive
2801 * parent pointer references will keep the dentry chain alive as long as no
2802 * rename operation is performed.
2804 static int prepend_path(const struct path *path,
2805 const struct path *root,
2806 char **buffer, int *buflen)
2808 struct dentry *dentry;
2809 struct vfsmount *vfsmnt;
2812 unsigned seq, m_seq = 0;
2818 read_seqbegin_or_lock(&mount_lock, &m_seq);
2825 dentry = path->dentry;
2827 mnt = real_mount(vfsmnt);
2828 read_seqbegin_or_lock(&rename_lock, &seq);
2829 while (dentry != root->dentry || vfsmnt != root->mnt) {
2830 struct dentry * parent;
2832 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2833 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2835 if (mnt != parent) {
2836 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2842 * Filesystems needing to implement special "root names"
2843 * should do so with ->d_dname()
2845 if (IS_ROOT(dentry) &&
2846 (dentry->d_name.len != 1 ||
2847 dentry->d_name.name[0] != '/')) {
2848 WARN(1, "Root dentry has weird name <%.*s>\n",
2849 (int) dentry->d_name.len,
2850 dentry->d_name.name);
2853 error = is_mounted(vfsmnt) ? 1 : 2;
2856 parent = dentry->d_parent;
2858 error = prepend_name(&bptr, &blen, &dentry->d_name);
2866 if (need_seqretry(&rename_lock, seq)) {
2870 done_seqretry(&rename_lock, seq);
2874 if (need_seqretry(&mount_lock, m_seq)) {
2878 done_seqretry(&mount_lock, m_seq);
2880 if (error >= 0 && bptr == *buffer) {
2882 error = -ENAMETOOLONG;
2892 * __d_path - return the path of a dentry
2893 * @path: the dentry/vfsmount to report
2894 * @root: root vfsmnt/dentry
2895 * @buf: buffer to return value in
2896 * @buflen: buffer length
2898 * Convert a dentry into an ASCII path name.
2900 * Returns a pointer into the buffer or an error code if the
2901 * path was too long.
2903 * "buflen" should be positive.
2905 * If the path is not reachable from the supplied root, return %NULL.
2907 char *__d_path(const struct path *path,
2908 const struct path *root,
2909 char *buf, int buflen)
2911 char *res = buf + buflen;
2914 prepend(&res, &buflen, "\0", 1);
2915 error = prepend_path(path, root, &res, &buflen);
2918 return ERR_PTR(error);
2924 char *d_absolute_path(const struct path *path,
2925 char *buf, int buflen)
2927 struct path root = {};
2928 char *res = buf + buflen;
2931 prepend(&res, &buflen, "\0", 1);
2932 error = prepend_path(path, &root, &res, &buflen);
2937 return ERR_PTR(error);
2942 * same as __d_path but appends "(deleted)" for unlinked files.
2944 static int path_with_deleted(const struct path *path,
2945 const struct path *root,
2946 char **buf, int *buflen)
2948 prepend(buf, buflen, "\0", 1);
2949 if (d_unlinked(path->dentry)) {
2950 int error = prepend(buf, buflen, " (deleted)", 10);
2955 return prepend_path(path, root, buf, buflen);
2958 static int prepend_unreachable(char **buffer, int *buflen)
2960 return prepend(buffer, buflen, "(unreachable)", 13);
2963 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
2968 seq = read_seqcount_begin(&fs->seq);
2970 } while (read_seqcount_retry(&fs->seq, seq));
2974 * d_path - return the path of a dentry
2975 * @path: path to report
2976 * @buf: buffer to return value in
2977 * @buflen: buffer length
2979 * Convert a dentry into an ASCII path name. If the entry has been deleted
2980 * the string " (deleted)" is appended. Note that this is ambiguous.
2982 * Returns a pointer into the buffer or an error code if the path was
2983 * too long. Note: Callers should use the returned pointer, not the passed
2984 * in buffer, to use the name! The implementation often starts at an offset
2985 * into the buffer, and may leave 0 bytes at the start.
2987 * "buflen" should be positive.
2989 char *d_path(const struct path *path, char *buf, int buflen)
2991 char *res = buf + buflen;
2996 * We have various synthetic filesystems that never get mounted. On
2997 * these filesystems dentries are never used for lookup purposes, and
2998 * thus don't need to be hashed. They also don't need a name until a
2999 * user wants to identify the object in /proc/pid/fd/. The little hack
3000 * below allows us to generate a name for these objects on demand:
3002 * Some pseudo inodes are mountable. When they are mounted
3003 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3004 * and instead have d_path return the mounted path.
3006 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3007 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3008 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3011 get_fs_root_rcu(current->fs, &root);
3012 error = path_with_deleted(path, &root, &res, &buflen);
3016 res = ERR_PTR(error);
3019 EXPORT_SYMBOL(d_path);
3022 * Helper function for dentry_operations.d_dname() members
3024 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3025 const char *fmt, ...)
3031 va_start(args, fmt);
3032 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3035 if (sz > sizeof(temp) || sz > buflen)
3036 return ERR_PTR(-ENAMETOOLONG);
3038 buffer += buflen - sz;
3039 return memcpy(buffer, temp, sz);
3042 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3044 char *end = buffer + buflen;
3045 /* these dentries are never renamed, so d_lock is not needed */
3046 if (prepend(&end, &buflen, " (deleted)", 11) ||
3047 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3048 prepend(&end, &buflen, "/", 1))
3049 end = ERR_PTR(-ENAMETOOLONG);
3052 EXPORT_SYMBOL(simple_dname);
3055 * Write full pathname from the root of the filesystem into the buffer.
3057 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3059 struct dentry *dentry;
3072 prepend(&end, &len, "\0", 1);
3076 read_seqbegin_or_lock(&rename_lock, &seq);
3077 while (!IS_ROOT(dentry)) {
3078 struct dentry *parent = dentry->d_parent;
3081 error = prepend_name(&end, &len, &dentry->d_name);
3090 if (need_seqretry(&rename_lock, seq)) {
3094 done_seqretry(&rename_lock, seq);
3099 return ERR_PTR(-ENAMETOOLONG);
3102 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3104 return __dentry_path(dentry, buf, buflen);
3106 EXPORT_SYMBOL(dentry_path_raw);
3108 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3113 if (d_unlinked(dentry)) {
3115 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3119 retval = __dentry_path(dentry, buf, buflen);
3120 if (!IS_ERR(retval) && p)
3121 *p = '/'; /* restore '/' overriden with '\0' */
3124 return ERR_PTR(-ENAMETOOLONG);
3127 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3133 seq = read_seqcount_begin(&fs->seq);
3136 } while (read_seqcount_retry(&fs->seq, seq));
3140 * NOTE! The user-level library version returns a
3141 * character pointer. The kernel system call just
3142 * returns the length of the buffer filled (which
3143 * includes the ending '\0' character), or a negative
3144 * error value. So libc would do something like
3146 * char *getcwd(char * buf, size_t size)
3150 * retval = sys_getcwd(buf, size);
3157 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3160 struct path pwd, root;
3161 char *page = __getname();
3167 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3170 if (!d_unlinked(pwd.dentry)) {
3172 char *cwd = page + PATH_MAX;
3173 int buflen = PATH_MAX;
3175 prepend(&cwd, &buflen, "\0", 1);
3176 error = prepend_path(&pwd, &root, &cwd, &buflen);
3182 /* Unreachable from current root */
3184 error = prepend_unreachable(&cwd, &buflen);
3190 len = PATH_MAX + page - cwd;
3193 if (copy_to_user(buf, cwd, len))
3206 * Test whether new_dentry is a subdirectory of old_dentry.
3208 * Trivially implemented using the dcache structure
3212 * is_subdir - is new dentry a subdirectory of old_dentry
3213 * @new_dentry: new dentry
3214 * @old_dentry: old dentry
3216 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3217 * Returns 0 otherwise.
3218 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3221 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3226 if (new_dentry == old_dentry)
3230 /* for restarting inner loop in case of seq retry */
3231 seq = read_seqbegin(&rename_lock);
3233 * Need rcu_readlock to protect against the d_parent trashing
3237 if (d_ancestor(old_dentry, new_dentry))
3242 } while (read_seqretry(&rename_lock, seq));
3247 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3249 struct dentry *root = data;
3250 if (dentry != root) {
3251 if (d_unhashed(dentry) || !dentry->d_inode)
3254 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3255 dentry->d_flags |= DCACHE_GENOCIDE;
3256 dentry->d_lockref.count--;
3259 return D_WALK_CONTINUE;
3262 void d_genocide(struct dentry *parent)
3264 d_walk(parent, parent, d_genocide_kill, NULL);
3267 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3269 inode_dec_link_count(inode);
3270 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3271 !hlist_unhashed(&dentry->d_alias) ||
3272 !d_unlinked(dentry));
3273 spin_lock(&dentry->d_parent->d_lock);
3274 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3275 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3276 (unsigned long long)inode->i_ino);
3277 spin_unlock(&dentry->d_lock);
3278 spin_unlock(&dentry->d_parent->d_lock);
3279 d_instantiate(dentry, inode);
3281 EXPORT_SYMBOL(d_tmpfile);
3283 static __initdata unsigned long dhash_entries;
3284 static int __init set_dhash_entries(char *str)
3288 dhash_entries = simple_strtoul(str, &str, 0);
3291 __setup("dhash_entries=", set_dhash_entries);
3293 static void __init dcache_init_early(void)
3297 /* If hashes are distributed across NUMA nodes, defer
3298 * hash allocation until vmalloc space is available.
3304 alloc_large_system_hash("Dentry cache",
3305 sizeof(struct hlist_bl_head),
3314 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3315 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3318 static void __init dcache_init(void)
3323 * A constructor could be added for stable state like the lists,
3324 * but it is probably not worth it because of the cache nature
3327 dentry_cache = KMEM_CACHE(dentry,
3328 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3330 /* Hash may have been set up in dcache_init_early */
3335 alloc_large_system_hash("Dentry cache",
3336 sizeof(struct hlist_bl_head),
3345 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3346 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3349 /* SLAB cache for __getname() consumers */
3350 struct kmem_cache *names_cachep __read_mostly;
3351 EXPORT_SYMBOL(names_cachep);
3353 EXPORT_SYMBOL(d_genocide);
3355 void __init vfs_caches_init_early(void)
3357 dcache_init_early();
3361 void __init vfs_caches_init(unsigned long mempages)
3363 unsigned long reserve;
3365 /* Base hash sizes on available memory, with a reserve equal to
3366 150% of current kernel size */
3368 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3369 mempages -= reserve;
3371 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3372 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3376 files_init(mempages);