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 = ~(~0ul << tcount*8);
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);
252 static void d_free(struct dentry *dentry)
254 BUG_ON((int)dentry->d_lockref.count > 0);
255 this_cpu_dec(nr_dentry);
256 if (dentry->d_op && dentry->d_op->d_release)
257 dentry->d_op->d_release(dentry);
259 /* if dentry was never visible to RCU, immediate free is OK */
260 if (!(dentry->d_flags & DCACHE_RCUACCESS))
261 __d_free(&dentry->d_u.d_rcu);
263 call_rcu(&dentry->d_u.d_rcu, __d_free);
267 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
268 * @dentry: the target dentry
269 * After this call, in-progress rcu-walk path lookup will fail. This
270 * should be called after unhashing, and after changing d_inode (if
271 * the dentry has not already been unhashed).
273 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
275 assert_spin_locked(&dentry->d_lock);
276 /* Go through a barrier */
277 write_seqcount_barrier(&dentry->d_seq);
281 * Release the dentry's inode, using the filesystem
282 * d_iput() operation if defined. Dentry has no refcount
285 static void dentry_iput(struct dentry * dentry)
286 __releases(dentry->d_lock)
287 __releases(dentry->d_inode->i_lock)
289 struct inode *inode = dentry->d_inode;
291 dentry->d_inode = NULL;
292 hlist_del_init(&dentry->d_alias);
293 spin_unlock(&dentry->d_lock);
294 spin_unlock(&inode->i_lock);
296 fsnotify_inoderemove(inode);
297 if (dentry->d_op && dentry->d_op->d_iput)
298 dentry->d_op->d_iput(dentry, inode);
302 spin_unlock(&dentry->d_lock);
307 * Release the dentry's inode, using the filesystem
308 * d_iput() operation if defined. dentry remains in-use.
310 static void dentry_unlink_inode(struct dentry * dentry)
311 __releases(dentry->d_lock)
312 __releases(dentry->d_inode->i_lock)
314 struct inode *inode = dentry->d_inode;
315 __d_clear_type(dentry);
316 dentry->d_inode = NULL;
317 hlist_del_init(&dentry->d_alias);
318 dentry_rcuwalk_barrier(dentry);
319 spin_unlock(&dentry->d_lock);
320 spin_unlock(&inode->i_lock);
322 fsnotify_inoderemove(inode);
323 if (dentry->d_op && dentry->d_op->d_iput)
324 dentry->d_op->d_iput(dentry, inode);
330 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
331 * is in use - which includes both the "real" per-superblock
332 * LRU list _and_ the DCACHE_SHRINK_LIST use.
334 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
335 * on the shrink list (ie not on the superblock LRU list).
337 * The per-cpu "nr_dentry_unused" counters are updated with
338 * the DCACHE_LRU_LIST bit.
340 * These helper functions make sure we always follow the
341 * rules. d_lock must be held by the caller.
343 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
344 static void d_lru_add(struct dentry *dentry)
346 D_FLAG_VERIFY(dentry, 0);
347 dentry->d_flags |= DCACHE_LRU_LIST;
348 this_cpu_inc(nr_dentry_unused);
349 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
352 static void d_lru_del(struct dentry *dentry)
354 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
355 dentry->d_flags &= ~DCACHE_LRU_LIST;
356 this_cpu_dec(nr_dentry_unused);
357 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
360 static void d_shrink_del(struct dentry *dentry)
362 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
363 list_del_init(&dentry->d_lru);
364 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
365 this_cpu_dec(nr_dentry_unused);
368 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
370 D_FLAG_VERIFY(dentry, 0);
371 list_add(&dentry->d_lru, list);
372 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
373 this_cpu_inc(nr_dentry_unused);
377 * These can only be called under the global LRU lock, ie during the
378 * callback for freeing the LRU list. "isolate" removes it from the
379 * LRU lists entirely, while shrink_move moves it to the indicated
382 static void d_lru_isolate(struct dentry *dentry)
384 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
385 dentry->d_flags &= ~DCACHE_LRU_LIST;
386 this_cpu_dec(nr_dentry_unused);
387 list_del_init(&dentry->d_lru);
390 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
392 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
393 dentry->d_flags |= DCACHE_SHRINK_LIST;
394 list_move_tail(&dentry->d_lru, list);
398 * dentry_lru_(add|del)_list) must be called with d_lock held.
400 static void dentry_lru_add(struct dentry *dentry)
402 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
407 * Remove a dentry with references from the LRU.
409 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
410 * lose our last reference through the parent walk. In this case, we need to
411 * remove ourselves from the shrink list, not the LRU.
413 static void dentry_lru_del(struct dentry *dentry)
415 if (dentry->d_flags & DCACHE_LRU_LIST) {
416 if (dentry->d_flags & DCACHE_SHRINK_LIST)
417 return d_shrink_del(dentry);
423 * d_kill - kill dentry and return parent
424 * @dentry: dentry to kill
425 * @parent: parent dentry
427 * The dentry must already be unhashed and removed from the LRU.
429 * If this is the root of the dentry tree, return NULL.
431 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
434 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
435 __releases(dentry->d_lock)
436 __releases(parent->d_lock)
437 __releases(dentry->d_inode->i_lock)
439 list_del(&dentry->d_u.d_child);
441 * Inform try_to_ascend() that we are no longer attached to the
444 dentry->d_flags |= DCACHE_DENTRY_KILLED;
446 spin_unlock(&parent->d_lock);
449 * dentry_iput drops the locks, at which point nobody (except
450 * transient RCU lookups) can reach this dentry.
457 * d_drop - drop a dentry
458 * @dentry: dentry to drop
460 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
461 * be found through a VFS lookup any more. Note that this is different from
462 * deleting the dentry - d_delete will try to mark the dentry negative if
463 * possible, giving a successful _negative_ lookup, while d_drop will
464 * just make the cache lookup fail.
466 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
467 * reason (NFS timeouts or autofs deletes).
469 * __d_drop requires dentry->d_lock.
471 void __d_drop(struct dentry *dentry)
473 if (!d_unhashed(dentry)) {
474 struct hlist_bl_head *b;
476 * Hashed dentries are normally on the dentry hashtable,
477 * with the exception of those newly allocated by
478 * d_obtain_alias, which are always IS_ROOT:
480 if (unlikely(IS_ROOT(dentry)))
481 b = &dentry->d_sb->s_anon;
483 b = d_hash(dentry->d_parent, dentry->d_name.hash);
486 __hlist_bl_del(&dentry->d_hash);
487 dentry->d_hash.pprev = NULL;
489 dentry_rcuwalk_barrier(dentry);
492 EXPORT_SYMBOL(__d_drop);
494 void d_drop(struct dentry *dentry)
496 spin_lock(&dentry->d_lock);
498 spin_unlock(&dentry->d_lock);
500 EXPORT_SYMBOL(d_drop);
503 * Finish off a dentry we've decided to kill.
504 * dentry->d_lock must be held, returns with it unlocked.
505 * If ref is non-zero, then decrement the refcount too.
506 * Returns dentry requiring refcount drop, or NULL if we're done.
508 static struct dentry *
509 dentry_kill(struct dentry *dentry, int unlock_on_failure)
510 __releases(dentry->d_lock)
513 struct dentry *parent;
515 inode = dentry->d_inode;
516 if (inode && !spin_trylock(&inode->i_lock)) {
518 if (unlock_on_failure) {
519 spin_unlock(&dentry->d_lock);
522 return dentry; /* try again with same dentry */
527 parent = dentry->d_parent;
528 if (parent && !spin_trylock(&parent->d_lock)) {
530 spin_unlock(&inode->i_lock);
535 * The dentry is now unrecoverably dead to the world.
537 lockref_mark_dead(&dentry->d_lockref);
540 * inform the fs via d_prune that this dentry is about to be
541 * unhashed and destroyed.
543 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
544 dentry->d_op->d_prune(dentry);
546 dentry_lru_del(dentry);
547 /* if it was on the hash then remove it */
549 return d_kill(dentry, parent);
555 * This is complicated by the fact that we do not want to put
556 * dentries that are no longer on any hash chain on the unused
557 * list: we'd much rather just get rid of them immediately.
559 * However, that implies that we have to traverse the dentry
560 * tree upwards to the parents which might _also_ now be
561 * scheduled for deletion (it may have been only waiting for
562 * its last child to go away).
564 * This tail recursion is done by hand as we don't want to depend
565 * on the compiler to always get this right (gcc generally doesn't).
566 * Real recursion would eat up our stack space.
570 * dput - release a dentry
571 * @dentry: dentry to release
573 * Release a dentry. This will drop the usage count and if appropriate
574 * call the dentry unlink method as well as removing it from the queues and
575 * releasing its resources. If the parent dentries were scheduled for release
576 * they too may now get deleted.
578 void dput(struct dentry *dentry)
580 if (unlikely(!dentry))
584 if (lockref_put_or_lock(&dentry->d_lockref))
587 /* Unreachable? Get rid of it */
588 if (unlikely(d_unhashed(dentry)))
591 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
592 if (dentry->d_op->d_delete(dentry))
596 if (!(dentry->d_flags & DCACHE_REFERENCED))
597 dentry->d_flags |= DCACHE_REFERENCED;
598 dentry_lru_add(dentry);
600 dentry->d_lockref.count--;
601 spin_unlock(&dentry->d_lock);
605 dentry = dentry_kill(dentry, 1);
612 * d_invalidate - invalidate a dentry
613 * @dentry: dentry to invalidate
615 * Try to invalidate the dentry if it turns out to be
616 * possible. If there are other dentries that can be
617 * reached through this one we can't delete it and we
618 * return -EBUSY. On success we return 0.
623 int d_invalidate(struct dentry * dentry)
626 * If it's already been dropped, return OK.
628 spin_lock(&dentry->d_lock);
629 if (d_unhashed(dentry)) {
630 spin_unlock(&dentry->d_lock);
634 * Check whether to do a partial shrink_dcache
635 * to get rid of unused child entries.
637 if (!list_empty(&dentry->d_subdirs)) {
638 spin_unlock(&dentry->d_lock);
639 shrink_dcache_parent(dentry);
640 spin_lock(&dentry->d_lock);
644 * Somebody else still using it?
646 * If it's a directory, we can't drop it
647 * for fear of somebody re-populating it
648 * with children (even though dropping it
649 * would make it unreachable from the root,
650 * we might still populate it if it was a
651 * working directory or similar).
652 * We also need to leave mountpoints alone,
655 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
656 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
657 spin_unlock(&dentry->d_lock);
663 spin_unlock(&dentry->d_lock);
666 EXPORT_SYMBOL(d_invalidate);
668 /* This must be called with d_lock held */
669 static inline void __dget_dlock(struct dentry *dentry)
671 dentry->d_lockref.count++;
674 static inline void __dget(struct dentry *dentry)
676 lockref_get(&dentry->d_lockref);
679 struct dentry *dget_parent(struct dentry *dentry)
685 * Do optimistic parent lookup without any
689 ret = ACCESS_ONCE(dentry->d_parent);
690 gotref = lockref_get_not_zero(&ret->d_lockref);
692 if (likely(gotref)) {
693 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
700 * Don't need rcu_dereference because we re-check it was correct under
704 ret = dentry->d_parent;
705 spin_lock(&ret->d_lock);
706 if (unlikely(ret != dentry->d_parent)) {
707 spin_unlock(&ret->d_lock);
712 BUG_ON(!ret->d_lockref.count);
713 ret->d_lockref.count++;
714 spin_unlock(&ret->d_lock);
717 EXPORT_SYMBOL(dget_parent);
720 * d_find_alias - grab a hashed alias of inode
721 * @inode: inode in question
722 * @want_discon: flag, used by d_splice_alias, to request
723 * that only a DISCONNECTED alias be returned.
725 * If inode has a hashed alias, or is a directory and has any alias,
726 * acquire the reference to alias and return it. Otherwise return NULL.
727 * Notice that if inode is a directory there can be only one alias and
728 * it can be unhashed only if it has no children, or if it is the root
731 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
732 * any other hashed alias over that one unless @want_discon is set,
733 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
735 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
737 struct dentry *alias, *discon_alias;
741 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
742 spin_lock(&alias->d_lock);
743 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
744 if (IS_ROOT(alias) &&
745 (alias->d_flags & DCACHE_DISCONNECTED)) {
746 discon_alias = alias;
747 } else if (!want_discon) {
749 spin_unlock(&alias->d_lock);
753 spin_unlock(&alias->d_lock);
756 alias = discon_alias;
757 spin_lock(&alias->d_lock);
758 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
759 if (IS_ROOT(alias) &&
760 (alias->d_flags & DCACHE_DISCONNECTED)) {
762 spin_unlock(&alias->d_lock);
766 spin_unlock(&alias->d_lock);
772 struct dentry *d_find_alias(struct inode *inode)
774 struct dentry *de = NULL;
776 if (!hlist_empty(&inode->i_dentry)) {
777 spin_lock(&inode->i_lock);
778 de = __d_find_alias(inode, 0);
779 spin_unlock(&inode->i_lock);
783 EXPORT_SYMBOL(d_find_alias);
786 * Try to kill dentries associated with this inode.
787 * WARNING: you must own a reference to inode.
789 void d_prune_aliases(struct inode *inode)
791 struct dentry *dentry;
793 spin_lock(&inode->i_lock);
794 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
795 spin_lock(&dentry->d_lock);
796 if (!dentry->d_lockref.count) {
798 * inform the fs via d_prune that this dentry
799 * is about to be unhashed and destroyed.
801 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
803 dentry->d_op->d_prune(dentry);
805 __dget_dlock(dentry);
807 spin_unlock(&dentry->d_lock);
808 spin_unlock(&inode->i_lock);
812 spin_unlock(&dentry->d_lock);
814 spin_unlock(&inode->i_lock);
816 EXPORT_SYMBOL(d_prune_aliases);
819 * Try to throw away a dentry - free the inode, dput the parent.
820 * Requires dentry->d_lock is held, and dentry->d_count == 0.
821 * Releases dentry->d_lock.
823 * This may fail if locks cannot be acquired no problem, just try again.
825 static struct dentry * try_prune_one_dentry(struct dentry *dentry)
826 __releases(dentry->d_lock)
828 struct dentry *parent;
830 parent = dentry_kill(dentry, 0);
832 * If dentry_kill returns NULL, we have nothing more to do.
833 * if it returns the same dentry, trylocks failed. In either
834 * case, just loop again.
836 * Otherwise, we need to prune ancestors too. This is necessary
837 * to prevent quadratic behavior of shrink_dcache_parent(), but
838 * is also expected to be beneficial in reducing dentry cache
843 if (parent == dentry)
846 /* Prune ancestors. */
849 if (lockref_put_or_lock(&dentry->d_lockref))
851 dentry = dentry_kill(dentry, 1);
856 static void shrink_dentry_list(struct list_head *list)
858 struct dentry *dentry;
862 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
863 if (&dentry->d_lru == list)
867 * Get the dentry lock, and re-verify that the dentry is
868 * this on the shrinking list. If it is, we know that
869 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
871 spin_lock(&dentry->d_lock);
872 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
873 spin_unlock(&dentry->d_lock);
878 * The dispose list is isolated and dentries are not accounted
879 * to the LRU here, so we can simply remove it from the list
880 * here regardless of whether it is referenced or not.
882 d_shrink_del(dentry);
885 * We found an inuse dentry which was not removed from
886 * the LRU because of laziness during lookup. Do not free it.
888 if (dentry->d_lockref.count) {
889 spin_unlock(&dentry->d_lock);
895 * If 'try_to_prune()' returns a dentry, it will
896 * be the same one we passed in, and d_lock will
897 * have been held the whole time, so it will not
898 * have been added to any other lists. We failed
899 * to get the inode lock.
901 * We just add it back to the shrink list.
903 dentry = try_prune_one_dentry(dentry);
907 d_shrink_add(dentry, list);
908 spin_unlock(&dentry->d_lock);
914 static enum lru_status
915 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
917 struct list_head *freeable = arg;
918 struct dentry *dentry = container_of(item, struct dentry, d_lru);
922 * we are inverting the lru lock/dentry->d_lock here,
923 * so use a trylock. If we fail to get the lock, just skip
926 if (!spin_trylock(&dentry->d_lock))
930 * Referenced dentries are still in use. If they have active
931 * counts, just remove them from the LRU. Otherwise give them
932 * another pass through the LRU.
934 if (dentry->d_lockref.count) {
935 d_lru_isolate(dentry);
936 spin_unlock(&dentry->d_lock);
940 if (dentry->d_flags & DCACHE_REFERENCED) {
941 dentry->d_flags &= ~DCACHE_REFERENCED;
942 spin_unlock(&dentry->d_lock);
945 * The list move itself will be made by the common LRU code. At
946 * this point, we've dropped the dentry->d_lock but keep the
947 * lru lock. This is safe to do, since every list movement is
948 * protected by the lru lock even if both locks are held.
950 * This is guaranteed by the fact that all LRU management
951 * functions are intermediated by the LRU API calls like
952 * list_lru_add and list_lru_del. List movement in this file
953 * only ever occur through this functions or through callbacks
954 * like this one, that are called from the LRU API.
956 * The only exceptions to this are functions like
957 * shrink_dentry_list, and code that first checks for the
958 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
959 * operating only with stack provided lists after they are
960 * properly isolated from the main list. It is thus, always a
966 d_lru_shrink_move(dentry, freeable);
967 spin_unlock(&dentry->d_lock);
973 * prune_dcache_sb - shrink the dcache
975 * @nr_to_scan : number of entries to try to free
976 * @nid: which node to scan for freeable entities
978 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
979 * done when we need more memory an called from the superblock shrinker
982 * This function may fail to free any resources if all the dentries are in
985 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
991 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
992 &dispose, &nr_to_scan);
993 shrink_dentry_list(&dispose);
997 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
998 spinlock_t *lru_lock, void *arg)
1000 struct list_head *freeable = arg;
1001 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1004 * we are inverting the lru lock/dentry->d_lock here,
1005 * so use a trylock. If we fail to get the lock, just skip
1008 if (!spin_trylock(&dentry->d_lock))
1011 d_lru_shrink_move(dentry, freeable);
1012 spin_unlock(&dentry->d_lock);
1019 * shrink_dcache_sb - shrink dcache for a superblock
1022 * Shrink the dcache for the specified super block. This is used to free
1023 * the dcache before unmounting a file system.
1025 void shrink_dcache_sb(struct super_block *sb)
1032 freed = list_lru_walk(&sb->s_dentry_lru,
1033 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1035 this_cpu_sub(nr_dentry_unused, freed);
1036 shrink_dentry_list(&dispose);
1037 } while (freed > 0);
1039 EXPORT_SYMBOL(shrink_dcache_sb);
1042 * This tries to ascend one level of parenthood, but
1043 * we can race with renaming, so we need to re-check
1044 * the parenthood after dropping the lock and check
1045 * that the sequence number still matches.
1047 static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
1049 struct dentry *new = old->d_parent;
1052 spin_unlock(&old->d_lock);
1053 spin_lock(&new->d_lock);
1056 * might go back up the wrong parent if we have had a rename
1059 if (new != old->d_parent ||
1060 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1061 need_seqretry(&rename_lock, seq)) {
1062 spin_unlock(&new->d_lock);
1070 * enum d_walk_ret - action to talke during tree walk
1071 * @D_WALK_CONTINUE: contrinue walk
1072 * @D_WALK_QUIT: quit walk
1073 * @D_WALK_NORETRY: quit when retry is needed
1074 * @D_WALK_SKIP: skip this dentry and its children
1084 * d_walk - walk the dentry tree
1085 * @parent: start of walk
1086 * @data: data passed to @enter() and @finish()
1087 * @enter: callback when first entering the dentry
1088 * @finish: callback when successfully finished the walk
1090 * The @enter() and @finish() callbacks are called with d_lock held.
1092 static void d_walk(struct dentry *parent, void *data,
1093 enum d_walk_ret (*enter)(void *, struct dentry *),
1094 void (*finish)(void *))
1096 struct dentry *this_parent;
1097 struct list_head *next;
1099 enum d_walk_ret ret;
1103 read_seqbegin_or_lock(&rename_lock, &seq);
1104 this_parent = parent;
1105 spin_lock(&this_parent->d_lock);
1107 ret = enter(data, this_parent);
1109 case D_WALK_CONTINUE:
1114 case D_WALK_NORETRY:
1119 next = this_parent->d_subdirs.next;
1121 while (next != &this_parent->d_subdirs) {
1122 struct list_head *tmp = next;
1123 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1126 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1128 ret = enter(data, dentry);
1130 case D_WALK_CONTINUE:
1133 spin_unlock(&dentry->d_lock);
1135 case D_WALK_NORETRY:
1139 spin_unlock(&dentry->d_lock);
1143 if (!list_empty(&dentry->d_subdirs)) {
1144 spin_unlock(&this_parent->d_lock);
1145 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1146 this_parent = dentry;
1147 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1150 spin_unlock(&dentry->d_lock);
1153 * All done at this level ... ascend and resume the search.
1155 if (this_parent != parent) {
1156 struct dentry *child = this_parent;
1157 this_parent = try_to_ascend(this_parent, seq);
1160 next = child->d_u.d_child.next;
1163 if (need_seqretry(&rename_lock, seq)) {
1164 spin_unlock(&this_parent->d_lock);
1171 spin_unlock(&this_parent->d_lock);
1172 done_seqretry(&rename_lock, seq);
1183 * Search for at least 1 mount point in the dentry's subdirs.
1184 * We descend to the next level whenever the d_subdirs
1185 * list is non-empty and continue searching.
1188 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1191 if (d_mountpoint(dentry)) {
1195 return D_WALK_CONTINUE;
1199 * have_submounts - check for mounts over a dentry
1200 * @parent: dentry to check.
1202 * Return true if the parent or its subdirectories contain
1205 int have_submounts(struct dentry *parent)
1209 d_walk(parent, &ret, check_mount, NULL);
1213 EXPORT_SYMBOL(have_submounts);
1216 * Called by mount code to set a mountpoint and check if the mountpoint is
1217 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1218 * subtree can become unreachable).
1220 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1221 * this reason take rename_lock and d_lock on dentry and ancestors.
1223 int d_set_mounted(struct dentry *dentry)
1227 write_seqlock(&rename_lock);
1228 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1229 /* Need exclusion wrt. check_submounts_and_drop() */
1230 spin_lock(&p->d_lock);
1231 if (unlikely(d_unhashed(p))) {
1232 spin_unlock(&p->d_lock);
1235 spin_unlock(&p->d_lock);
1237 spin_lock(&dentry->d_lock);
1238 if (!d_unlinked(dentry)) {
1239 dentry->d_flags |= DCACHE_MOUNTED;
1242 spin_unlock(&dentry->d_lock);
1244 write_sequnlock(&rename_lock);
1249 * Search the dentry child list of the specified parent,
1250 * and move any unused dentries to the end of the unused
1251 * list for prune_dcache(). We descend to the next level
1252 * whenever the d_subdirs list is non-empty and continue
1255 * It returns zero iff there are no unused children,
1256 * otherwise it returns the number of children moved to
1257 * the end of the unused list. This may not be the total
1258 * number of unused children, because select_parent can
1259 * drop the lock and return early due to latency
1263 struct select_data {
1264 struct dentry *start;
1265 struct list_head dispose;
1269 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1271 struct select_data *data = _data;
1272 enum d_walk_ret ret = D_WALK_CONTINUE;
1274 if (data->start == dentry)
1278 * move only zero ref count dentries to the dispose list.
1280 * Those which are presently on the shrink list, being processed
1281 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1282 * loop in shrink_dcache_parent() might not make any progress
1285 if (dentry->d_lockref.count) {
1286 dentry_lru_del(dentry);
1287 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1289 * We can't use d_lru_shrink_move() because we
1290 * need to get the global LRU lock and do the
1294 d_shrink_add(dentry, &data->dispose);
1296 ret = D_WALK_NORETRY;
1299 * We can return to the caller if we have found some (this
1300 * ensures forward progress). We'll be coming back to find
1303 if (data->found && need_resched())
1310 * shrink_dcache_parent - prune dcache
1311 * @parent: parent of entries to prune
1313 * Prune the dcache to remove unused children of the parent dentry.
1315 void shrink_dcache_parent(struct dentry *parent)
1318 struct select_data data;
1320 INIT_LIST_HEAD(&data.dispose);
1321 data.start = parent;
1324 d_walk(parent, &data, select_collect, NULL);
1328 shrink_dentry_list(&data.dispose);
1332 EXPORT_SYMBOL(shrink_dcache_parent);
1334 static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry)
1336 struct select_data *data = _data;
1337 enum d_walk_ret ret = D_WALK_CONTINUE;
1339 if (dentry->d_lockref.count) {
1340 dentry_lru_del(dentry);
1341 if (likely(!list_empty(&dentry->d_subdirs)))
1343 if (dentry == data->start && dentry->d_lockref.count == 1)
1346 "BUG: Dentry %p{i=%lx,n=%s}"
1347 " still in use (%d)"
1348 " [unmount of %s %s]\n",
1351 dentry->d_inode->i_ino : 0UL,
1352 dentry->d_name.name,
1353 dentry->d_lockref.count,
1354 dentry->d_sb->s_type->name,
1355 dentry->d_sb->s_id);
1357 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1359 * We can't use d_lru_shrink_move() because we
1360 * need to get the global LRU lock and do the
1363 if (dentry->d_flags & DCACHE_LRU_LIST)
1365 d_shrink_add(dentry, &data->dispose);
1367 ret = D_WALK_NORETRY;
1370 if (data->found && need_resched())
1376 * destroy the dentries attached to a superblock on unmounting
1378 void shrink_dcache_for_umount(struct super_block *sb)
1380 struct dentry *dentry;
1382 if (down_read_trylock(&sb->s_umount))
1385 dentry = sb->s_root;
1388 struct select_data data;
1390 INIT_LIST_HEAD(&data.dispose);
1391 data.start = dentry;
1394 d_walk(dentry, &data, umount_collect, NULL);
1398 shrink_dentry_list(&data.dispose);
1404 while (!hlist_bl_empty(&sb->s_anon)) {
1405 struct select_data data;
1406 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1408 INIT_LIST_HEAD(&data.dispose);
1412 d_walk(dentry, &data, umount_collect, NULL);
1414 shrink_dentry_list(&data.dispose);
1419 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1421 struct select_data *data = _data;
1423 if (d_mountpoint(dentry)) {
1424 data->found = -EBUSY;
1428 return select_collect(_data, dentry);
1431 static void check_and_drop(void *_data)
1433 struct select_data *data = _data;
1435 if (d_mountpoint(data->start))
1436 data->found = -EBUSY;
1438 __d_drop(data->start);
1442 * check_submounts_and_drop - prune dcache, check for submounts and drop
1444 * All done as a single atomic operation relative to has_unlinked_ancestor().
1445 * Returns 0 if successfully unhashed @parent. If there were submounts then
1448 * @dentry: dentry to prune and drop
1450 int check_submounts_and_drop(struct dentry *dentry)
1454 /* Negative dentries can be dropped without further checks */
1455 if (!dentry->d_inode) {
1461 struct select_data data;
1463 INIT_LIST_HEAD(&data.dispose);
1464 data.start = dentry;
1467 d_walk(dentry, &data, check_and_collect, check_and_drop);
1470 if (!list_empty(&data.dispose))
1471 shrink_dentry_list(&data.dispose);
1482 EXPORT_SYMBOL(check_submounts_and_drop);
1485 * __d_alloc - allocate a dcache entry
1486 * @sb: filesystem it will belong to
1487 * @name: qstr of the name
1489 * Allocates a dentry. It returns %NULL if there is insufficient memory
1490 * available. On a success the dentry is returned. The name passed in is
1491 * copied and the copy passed in may be reused after this call.
1494 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1496 struct dentry *dentry;
1499 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1504 * We guarantee that the inline name is always NUL-terminated.
1505 * This way the memcpy() done by the name switching in rename
1506 * will still always have a NUL at the end, even if we might
1507 * be overwriting an internal NUL character
1509 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1510 if (name->len > DNAME_INLINE_LEN-1) {
1511 dname = kmalloc(name->len + 1, GFP_KERNEL);
1513 kmem_cache_free(dentry_cache, dentry);
1517 dname = dentry->d_iname;
1520 dentry->d_name.len = name->len;
1521 dentry->d_name.hash = name->hash;
1522 memcpy(dname, name->name, name->len);
1523 dname[name->len] = 0;
1525 /* Make sure we always see the terminating NUL character */
1527 dentry->d_name.name = dname;
1529 dentry->d_lockref.count = 1;
1530 dentry->d_flags = 0;
1531 spin_lock_init(&dentry->d_lock);
1532 seqcount_init(&dentry->d_seq);
1533 dentry->d_inode = NULL;
1534 dentry->d_parent = dentry;
1536 dentry->d_op = NULL;
1537 dentry->d_fsdata = NULL;
1538 INIT_HLIST_BL_NODE(&dentry->d_hash);
1539 INIT_LIST_HEAD(&dentry->d_lru);
1540 INIT_LIST_HEAD(&dentry->d_subdirs);
1541 INIT_HLIST_NODE(&dentry->d_alias);
1542 INIT_LIST_HEAD(&dentry->d_u.d_child);
1543 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1545 this_cpu_inc(nr_dentry);
1551 * d_alloc - allocate a dcache entry
1552 * @parent: parent of entry to allocate
1553 * @name: qstr of the name
1555 * Allocates a dentry. It returns %NULL if there is insufficient memory
1556 * available. On a success the dentry is returned. The name passed in is
1557 * copied and the copy passed in may be reused after this call.
1559 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1561 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1565 spin_lock(&parent->d_lock);
1567 * don't need child lock because it is not subject
1568 * to concurrency here
1570 __dget_dlock(parent);
1571 dentry->d_parent = parent;
1572 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1573 spin_unlock(&parent->d_lock);
1577 EXPORT_SYMBOL(d_alloc);
1580 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1581 * @sb: the superblock
1582 * @name: qstr of the name
1584 * For a filesystem that just pins its dentries in memory and never
1585 * performs lookups at all, return an unhashed IS_ROOT dentry.
1587 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1589 return __d_alloc(sb, name);
1591 EXPORT_SYMBOL(d_alloc_pseudo);
1593 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1598 q.len = strlen(name);
1599 q.hash = full_name_hash(q.name, q.len);
1600 return d_alloc(parent, &q);
1602 EXPORT_SYMBOL(d_alloc_name);
1604 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1606 WARN_ON_ONCE(dentry->d_op);
1607 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1609 DCACHE_OP_REVALIDATE |
1610 DCACHE_OP_WEAK_REVALIDATE |
1611 DCACHE_OP_DELETE ));
1616 dentry->d_flags |= DCACHE_OP_HASH;
1618 dentry->d_flags |= DCACHE_OP_COMPARE;
1619 if (op->d_revalidate)
1620 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1621 if (op->d_weak_revalidate)
1622 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1624 dentry->d_flags |= DCACHE_OP_DELETE;
1626 dentry->d_flags |= DCACHE_OP_PRUNE;
1629 EXPORT_SYMBOL(d_set_d_op);
1631 static unsigned d_flags_for_inode(struct inode *inode)
1633 unsigned add_flags = DCACHE_FILE_TYPE;
1636 return DCACHE_MISS_TYPE;
1638 if (S_ISDIR(inode->i_mode)) {
1639 add_flags = DCACHE_DIRECTORY_TYPE;
1640 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1641 if (unlikely(!inode->i_op->lookup))
1642 add_flags = DCACHE_AUTODIR_TYPE;
1644 inode->i_opflags |= IOP_LOOKUP;
1646 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1647 if (unlikely(inode->i_op->follow_link))
1648 add_flags = DCACHE_SYMLINK_TYPE;
1650 inode->i_opflags |= IOP_NOFOLLOW;
1653 if (unlikely(IS_AUTOMOUNT(inode)))
1654 add_flags |= DCACHE_NEED_AUTOMOUNT;
1658 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1660 unsigned add_flags = d_flags_for_inode(inode);
1662 spin_lock(&dentry->d_lock);
1663 dentry->d_flags &= ~DCACHE_ENTRY_TYPE;
1664 dentry->d_flags |= add_flags;
1666 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1667 dentry->d_inode = inode;
1668 dentry_rcuwalk_barrier(dentry);
1669 spin_unlock(&dentry->d_lock);
1670 fsnotify_d_instantiate(dentry, inode);
1674 * d_instantiate - fill in inode information for a dentry
1675 * @entry: dentry to complete
1676 * @inode: inode to attach to this dentry
1678 * Fill in inode information in the entry.
1680 * This turns negative dentries into productive full members
1683 * NOTE! This assumes that the inode count has been incremented
1684 * (or otherwise set) by the caller to indicate that it is now
1685 * in use by the dcache.
1688 void d_instantiate(struct dentry *entry, struct inode * inode)
1690 BUG_ON(!hlist_unhashed(&entry->d_alias));
1692 spin_lock(&inode->i_lock);
1693 __d_instantiate(entry, inode);
1695 spin_unlock(&inode->i_lock);
1696 security_d_instantiate(entry, inode);
1698 EXPORT_SYMBOL(d_instantiate);
1701 * d_instantiate_unique - instantiate a non-aliased dentry
1702 * @entry: dentry to instantiate
1703 * @inode: inode to attach to this dentry
1705 * Fill in inode information in the entry. On success, it returns NULL.
1706 * If an unhashed alias of "entry" already exists, then we return the
1707 * aliased dentry instead and drop one reference to inode.
1709 * Note that in order to avoid conflicts with rename() etc, the caller
1710 * had better be holding the parent directory semaphore.
1712 * This also assumes that the inode count has been incremented
1713 * (or otherwise set) by the caller to indicate that it is now
1714 * in use by the dcache.
1716 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1717 struct inode *inode)
1719 struct dentry *alias;
1720 int len = entry->d_name.len;
1721 const char *name = entry->d_name.name;
1722 unsigned int hash = entry->d_name.hash;
1725 __d_instantiate(entry, NULL);
1729 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1731 * Don't need alias->d_lock here, because aliases with
1732 * d_parent == entry->d_parent are not subject to name or
1733 * parent changes, because the parent inode i_mutex is held.
1735 if (alias->d_name.hash != hash)
1737 if (alias->d_parent != entry->d_parent)
1739 if (alias->d_name.len != len)
1741 if (dentry_cmp(alias, name, len))
1747 __d_instantiate(entry, inode);
1751 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1753 struct dentry *result;
1755 BUG_ON(!hlist_unhashed(&entry->d_alias));
1758 spin_lock(&inode->i_lock);
1759 result = __d_instantiate_unique(entry, inode);
1761 spin_unlock(&inode->i_lock);
1764 security_d_instantiate(entry, inode);
1768 BUG_ON(!d_unhashed(result));
1773 EXPORT_SYMBOL(d_instantiate_unique);
1776 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1777 * @entry: dentry to complete
1778 * @inode: inode to attach to this dentry
1780 * Fill in inode information in the entry. If a directory alias is found, then
1781 * return an error (and drop inode). Together with d_materialise_unique() this
1782 * guarantees that a directory inode may never have more than one alias.
1784 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1786 BUG_ON(!hlist_unhashed(&entry->d_alias));
1788 spin_lock(&inode->i_lock);
1789 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1790 spin_unlock(&inode->i_lock);
1794 __d_instantiate(entry, inode);
1795 spin_unlock(&inode->i_lock);
1796 security_d_instantiate(entry, inode);
1800 EXPORT_SYMBOL(d_instantiate_no_diralias);
1802 struct dentry *d_make_root(struct inode *root_inode)
1804 struct dentry *res = NULL;
1807 static const struct qstr name = QSTR_INIT("/", 1);
1809 res = __d_alloc(root_inode->i_sb, &name);
1811 d_instantiate(res, root_inode);
1817 EXPORT_SYMBOL(d_make_root);
1819 static struct dentry * __d_find_any_alias(struct inode *inode)
1821 struct dentry *alias;
1823 if (hlist_empty(&inode->i_dentry))
1825 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1831 * d_find_any_alias - find any alias for a given inode
1832 * @inode: inode to find an alias for
1834 * If any aliases exist for the given inode, take and return a
1835 * reference for one of them. If no aliases exist, return %NULL.
1837 struct dentry *d_find_any_alias(struct inode *inode)
1841 spin_lock(&inode->i_lock);
1842 de = __d_find_any_alias(inode);
1843 spin_unlock(&inode->i_lock);
1846 EXPORT_SYMBOL(d_find_any_alias);
1849 * d_obtain_alias - find or allocate a dentry for a given inode
1850 * @inode: inode to allocate the dentry for
1852 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1853 * similar open by handle operations. The returned dentry may be anonymous,
1854 * or may have a full name (if the inode was already in the cache).
1856 * When called on a directory inode, we must ensure that the inode only ever
1857 * has one dentry. If a dentry is found, that is returned instead of
1858 * allocating a new one.
1860 * On successful return, the reference to the inode has been transferred
1861 * to the dentry. In case of an error the reference on the inode is released.
1862 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1863 * be passed in and will be the error will be propagate to the return value,
1864 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1866 struct dentry *d_obtain_alias(struct inode *inode)
1868 static const struct qstr anonstring = QSTR_INIT("/", 1);
1874 return ERR_PTR(-ESTALE);
1876 return ERR_CAST(inode);
1878 res = d_find_any_alias(inode);
1882 tmp = __d_alloc(inode->i_sb, &anonstring);
1884 res = ERR_PTR(-ENOMEM);
1888 spin_lock(&inode->i_lock);
1889 res = __d_find_any_alias(inode);
1891 spin_unlock(&inode->i_lock);
1896 /* attach a disconnected dentry */
1897 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1899 spin_lock(&tmp->d_lock);
1900 tmp->d_inode = inode;
1901 tmp->d_flags |= add_flags;
1902 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1903 hlist_bl_lock(&tmp->d_sb->s_anon);
1904 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1905 hlist_bl_unlock(&tmp->d_sb->s_anon);
1906 spin_unlock(&tmp->d_lock);
1907 spin_unlock(&inode->i_lock);
1908 security_d_instantiate(tmp, inode);
1913 if (res && !IS_ERR(res))
1914 security_d_instantiate(res, inode);
1918 EXPORT_SYMBOL(d_obtain_alias);
1921 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1922 * @inode: the inode which may have a disconnected dentry
1923 * @dentry: a negative dentry which we want to point to the inode.
1925 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1926 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1927 * and return it, else simply d_add the inode to the dentry and return NULL.
1929 * This is needed in the lookup routine of any filesystem that is exportable
1930 * (via knfsd) so that we can build dcache paths to directories effectively.
1932 * If a dentry was found and moved, then it is returned. Otherwise NULL
1933 * is returned. This matches the expected return value of ->lookup.
1935 * Cluster filesystems may call this function with a negative, hashed dentry.
1936 * In that case, we know that the inode will be a regular file, and also this
1937 * will only occur during atomic_open. So we need to check for the dentry
1938 * being already hashed only in the final case.
1940 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1942 struct dentry *new = NULL;
1945 return ERR_CAST(inode);
1947 if (inode && S_ISDIR(inode->i_mode)) {
1948 spin_lock(&inode->i_lock);
1949 new = __d_find_alias(inode, 1);
1951 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1952 spin_unlock(&inode->i_lock);
1953 security_d_instantiate(new, inode);
1954 d_move(new, dentry);
1957 /* already taking inode->i_lock, so d_add() by hand */
1958 __d_instantiate(dentry, inode);
1959 spin_unlock(&inode->i_lock);
1960 security_d_instantiate(dentry, inode);
1964 d_instantiate(dentry, inode);
1965 if (d_unhashed(dentry))
1970 EXPORT_SYMBOL(d_splice_alias);
1973 * d_add_ci - lookup or allocate new dentry with case-exact name
1974 * @inode: the inode case-insensitive lookup has found
1975 * @dentry: the negative dentry that was passed to the parent's lookup func
1976 * @name: the case-exact name to be associated with the returned dentry
1978 * This is to avoid filling the dcache with case-insensitive names to the
1979 * same inode, only the actual correct case is stored in the dcache for
1980 * case-insensitive filesystems.
1982 * For a case-insensitive lookup match and if the the case-exact dentry
1983 * already exists in in the dcache, use it and return it.
1985 * If no entry exists with the exact case name, allocate new dentry with
1986 * the exact case, and return the spliced entry.
1988 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1991 struct dentry *found;
1995 * First check if a dentry matching the name already exists,
1996 * if not go ahead and create it now.
1998 found = d_hash_and_lookup(dentry->d_parent, name);
1999 if (unlikely(IS_ERR(found)))
2002 new = d_alloc(dentry->d_parent, name);
2004 found = ERR_PTR(-ENOMEM);
2008 found = d_splice_alias(inode, new);
2017 * If a matching dentry exists, and it's not negative use it.
2019 * Decrement the reference count to balance the iget() done
2022 if (found->d_inode) {
2023 if (unlikely(found->d_inode != inode)) {
2024 /* This can't happen because bad inodes are unhashed. */
2025 BUG_ON(!is_bad_inode(inode));
2026 BUG_ON(!is_bad_inode(found->d_inode));
2033 * Negative dentry: instantiate it unless the inode is a directory and
2034 * already has a dentry.
2036 new = d_splice_alias(inode, found);
2047 EXPORT_SYMBOL(d_add_ci);
2050 * Do the slow-case of the dentry name compare.
2052 * Unlike the dentry_cmp() function, we need to atomically
2053 * load the name and length information, so that the
2054 * filesystem can rely on them, and can use the 'name' and
2055 * 'len' information without worrying about walking off the
2056 * end of memory etc.
2058 * Thus the read_seqcount_retry() and the "duplicate" info
2059 * in arguments (the low-level filesystem should not look
2060 * at the dentry inode or name contents directly, since
2061 * rename can change them while we're in RCU mode).
2063 enum slow_d_compare {
2069 static noinline enum slow_d_compare slow_dentry_cmp(
2070 const struct dentry *parent,
2071 struct dentry *dentry,
2073 const struct qstr *name)
2075 int tlen = dentry->d_name.len;
2076 const char *tname = dentry->d_name.name;
2078 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2080 return D_COMP_SEQRETRY;
2082 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2083 return D_COMP_NOMATCH;
2088 * __d_lookup_rcu - search for a dentry (racy, store-free)
2089 * @parent: parent dentry
2090 * @name: qstr of name we wish to find
2091 * @seqp: returns d_seq value at the point where the dentry was found
2092 * Returns: dentry, or NULL
2094 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2095 * resolution (store-free path walking) design described in
2096 * Documentation/filesystems/path-lookup.txt.
2098 * This is not to be used outside core vfs.
2100 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2101 * held, and rcu_read_lock held. The returned dentry must not be stored into
2102 * without taking d_lock and checking d_seq sequence count against @seq
2105 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2108 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2109 * the returned dentry, so long as its parent's seqlock is checked after the
2110 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2111 * is formed, giving integrity down the path walk.
2113 * NOTE! The caller *has* to check the resulting dentry against the sequence
2114 * number we've returned before using any of the resulting dentry state!
2116 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2117 const struct qstr *name,
2120 u64 hashlen = name->hash_len;
2121 const unsigned char *str = name->name;
2122 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2123 struct hlist_bl_node *node;
2124 struct dentry *dentry;
2127 * Note: There is significant duplication with __d_lookup_rcu which is
2128 * required to prevent single threaded performance regressions
2129 * especially on architectures where smp_rmb (in seqcounts) are costly.
2130 * Keep the two functions in sync.
2134 * The hash list is protected using RCU.
2136 * Carefully use d_seq when comparing a candidate dentry, to avoid
2137 * races with d_move().
2139 * It is possible that concurrent renames can mess up our list
2140 * walk here and result in missing our dentry, resulting in the
2141 * false-negative result. d_lookup() protects against concurrent
2142 * renames using rename_lock seqlock.
2144 * See Documentation/filesystems/path-lookup.txt for more details.
2146 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2151 * The dentry sequence count protects us from concurrent
2152 * renames, and thus protects parent and name fields.
2154 * The caller must perform a seqcount check in order
2155 * to do anything useful with the returned dentry.
2157 * NOTE! We do a "raw" seqcount_begin here. That means that
2158 * we don't wait for the sequence count to stabilize if it
2159 * is in the middle of a sequence change. If we do the slow
2160 * dentry compare, we will do seqretries until it is stable,
2161 * and if we end up with a successful lookup, we actually
2162 * want to exit RCU lookup anyway.
2164 seq = raw_seqcount_begin(&dentry->d_seq);
2165 if (dentry->d_parent != parent)
2167 if (d_unhashed(dentry))
2170 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2171 if (dentry->d_name.hash != hashlen_hash(hashlen))
2174 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2177 case D_COMP_NOMATCH:
2184 if (dentry->d_name.hash_len != hashlen)
2187 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2194 * d_lookup - search for a dentry
2195 * @parent: parent dentry
2196 * @name: qstr of name we wish to find
2197 * Returns: dentry, or NULL
2199 * d_lookup searches the children of the parent dentry for the name in
2200 * question. If the dentry is found its reference count is incremented and the
2201 * dentry is returned. The caller must use dput to free the entry when it has
2202 * finished using it. %NULL is returned if the dentry does not exist.
2204 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2206 struct dentry *dentry;
2210 seq = read_seqbegin(&rename_lock);
2211 dentry = __d_lookup(parent, name);
2214 } while (read_seqretry(&rename_lock, seq));
2217 EXPORT_SYMBOL(d_lookup);
2220 * __d_lookup - search for a dentry (racy)
2221 * @parent: parent dentry
2222 * @name: qstr of name we wish to find
2223 * Returns: dentry, or NULL
2225 * __d_lookup is like d_lookup, however it may (rarely) return a
2226 * false-negative result due to unrelated rename activity.
2228 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2229 * however it must be used carefully, eg. with a following d_lookup in
2230 * the case of failure.
2232 * __d_lookup callers must be commented.
2234 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2236 unsigned int len = name->len;
2237 unsigned int hash = name->hash;
2238 const unsigned char *str = name->name;
2239 struct hlist_bl_head *b = d_hash(parent, hash);
2240 struct hlist_bl_node *node;
2241 struct dentry *found = NULL;
2242 struct dentry *dentry;
2245 * Note: There is significant duplication with __d_lookup_rcu which is
2246 * required to prevent single threaded performance regressions
2247 * especially on architectures where smp_rmb (in seqcounts) are costly.
2248 * Keep the two functions in sync.
2252 * The hash list is protected using RCU.
2254 * Take d_lock when comparing a candidate dentry, to avoid races
2257 * It is possible that concurrent renames can mess up our list
2258 * walk here and result in missing our dentry, resulting in the
2259 * false-negative result. d_lookup() protects against concurrent
2260 * renames using rename_lock seqlock.
2262 * See Documentation/filesystems/path-lookup.txt for more details.
2266 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2268 if (dentry->d_name.hash != hash)
2271 spin_lock(&dentry->d_lock);
2272 if (dentry->d_parent != parent)
2274 if (d_unhashed(dentry))
2278 * It is safe to compare names since d_move() cannot
2279 * change the qstr (protected by d_lock).
2281 if (parent->d_flags & DCACHE_OP_COMPARE) {
2282 int tlen = dentry->d_name.len;
2283 const char *tname = dentry->d_name.name;
2284 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2287 if (dentry->d_name.len != len)
2289 if (dentry_cmp(dentry, str, len))
2293 dentry->d_lockref.count++;
2295 spin_unlock(&dentry->d_lock);
2298 spin_unlock(&dentry->d_lock);
2306 * d_hash_and_lookup - hash the qstr then search for a dentry
2307 * @dir: Directory to search in
2308 * @name: qstr of name we wish to find
2310 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2312 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2315 * Check for a fs-specific hash function. Note that we must
2316 * calculate the standard hash first, as the d_op->d_hash()
2317 * routine may choose to leave the hash value unchanged.
2319 name->hash = full_name_hash(name->name, name->len);
2320 if (dir->d_flags & DCACHE_OP_HASH) {
2321 int err = dir->d_op->d_hash(dir, name);
2322 if (unlikely(err < 0))
2323 return ERR_PTR(err);
2325 return d_lookup(dir, name);
2327 EXPORT_SYMBOL(d_hash_and_lookup);
2330 * d_validate - verify dentry provided from insecure source (deprecated)
2331 * @dentry: The dentry alleged to be valid child of @dparent
2332 * @dparent: The parent dentry (known to be valid)
2334 * An insecure source has sent us a dentry, here we verify it and dget() it.
2335 * This is used by ncpfs in its readdir implementation.
2336 * Zero is returned in the dentry is invalid.
2338 * This function is slow for big directories, and deprecated, do not use it.
2340 int d_validate(struct dentry *dentry, struct dentry *dparent)
2342 struct dentry *child;
2344 spin_lock(&dparent->d_lock);
2345 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2346 if (dentry == child) {
2347 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2348 __dget_dlock(dentry);
2349 spin_unlock(&dentry->d_lock);
2350 spin_unlock(&dparent->d_lock);
2354 spin_unlock(&dparent->d_lock);
2358 EXPORT_SYMBOL(d_validate);
2361 * When a file is deleted, we have two options:
2362 * - turn this dentry into a negative dentry
2363 * - unhash this dentry and free it.
2365 * Usually, we want to just turn this into
2366 * a negative dentry, but if anybody else is
2367 * currently using the dentry or the inode
2368 * we can't do that and we fall back on removing
2369 * it from the hash queues and waiting for
2370 * it to be deleted later when it has no users
2374 * d_delete - delete a dentry
2375 * @dentry: The dentry to delete
2377 * Turn the dentry into a negative dentry if possible, otherwise
2378 * remove it from the hash queues so it can be deleted later
2381 void d_delete(struct dentry * dentry)
2383 struct inode *inode;
2386 * Are we the only user?
2389 spin_lock(&dentry->d_lock);
2390 inode = dentry->d_inode;
2391 isdir = S_ISDIR(inode->i_mode);
2392 if (dentry->d_lockref.count == 1) {
2393 if (!spin_trylock(&inode->i_lock)) {
2394 spin_unlock(&dentry->d_lock);
2398 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2399 dentry_unlink_inode(dentry);
2400 fsnotify_nameremove(dentry, isdir);
2404 if (!d_unhashed(dentry))
2407 spin_unlock(&dentry->d_lock);
2409 fsnotify_nameremove(dentry, isdir);
2411 EXPORT_SYMBOL(d_delete);
2413 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2415 BUG_ON(!d_unhashed(entry));
2417 entry->d_flags |= DCACHE_RCUACCESS;
2418 hlist_bl_add_head_rcu(&entry->d_hash, b);
2422 static void _d_rehash(struct dentry * entry)
2424 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2428 * d_rehash - add an entry back to the hash
2429 * @entry: dentry to add to the hash
2431 * Adds a dentry to the hash according to its name.
2434 void d_rehash(struct dentry * entry)
2436 spin_lock(&entry->d_lock);
2438 spin_unlock(&entry->d_lock);
2440 EXPORT_SYMBOL(d_rehash);
2443 * dentry_update_name_case - update case insensitive dentry with a new name
2444 * @dentry: dentry to be updated
2447 * Update a case insensitive dentry with new case of name.
2449 * dentry must have been returned by d_lookup with name @name. Old and new
2450 * name lengths must match (ie. no d_compare which allows mismatched name
2453 * Parent inode i_mutex must be held over d_lookup and into this call (to
2454 * keep renames and concurrent inserts, and readdir(2) away).
2456 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2458 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2459 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2461 spin_lock(&dentry->d_lock);
2462 write_seqcount_begin(&dentry->d_seq);
2463 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2464 write_seqcount_end(&dentry->d_seq);
2465 spin_unlock(&dentry->d_lock);
2467 EXPORT_SYMBOL(dentry_update_name_case);
2469 static void switch_names(struct dentry *dentry, struct dentry *target)
2471 if (dname_external(target)) {
2472 if (dname_external(dentry)) {
2474 * Both external: swap the pointers
2476 swap(target->d_name.name, dentry->d_name.name);
2479 * dentry:internal, target:external. Steal target's
2480 * storage and make target internal.
2482 memcpy(target->d_iname, dentry->d_name.name,
2483 dentry->d_name.len + 1);
2484 dentry->d_name.name = target->d_name.name;
2485 target->d_name.name = target->d_iname;
2488 if (dname_external(dentry)) {
2490 * dentry:external, target:internal. Give dentry's
2491 * storage to target and make dentry internal
2493 memcpy(dentry->d_iname, target->d_name.name,
2494 target->d_name.len + 1);
2495 target->d_name.name = dentry->d_name.name;
2496 dentry->d_name.name = dentry->d_iname;
2499 * Both are internal. Just copy target to dentry
2501 memcpy(dentry->d_iname, target->d_name.name,
2502 target->d_name.len + 1);
2503 dentry->d_name.len = target->d_name.len;
2507 swap(dentry->d_name.len, target->d_name.len);
2510 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2513 * XXXX: do we really need to take target->d_lock?
2515 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2516 spin_lock(&target->d_parent->d_lock);
2518 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2519 spin_lock(&dentry->d_parent->d_lock);
2520 spin_lock_nested(&target->d_parent->d_lock,
2521 DENTRY_D_LOCK_NESTED);
2523 spin_lock(&target->d_parent->d_lock);
2524 spin_lock_nested(&dentry->d_parent->d_lock,
2525 DENTRY_D_LOCK_NESTED);
2528 if (target < dentry) {
2529 spin_lock_nested(&target->d_lock, 2);
2530 spin_lock_nested(&dentry->d_lock, 3);
2532 spin_lock_nested(&dentry->d_lock, 2);
2533 spin_lock_nested(&target->d_lock, 3);
2537 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2538 struct dentry *target)
2540 if (target->d_parent != dentry->d_parent)
2541 spin_unlock(&dentry->d_parent->d_lock);
2542 if (target->d_parent != target)
2543 spin_unlock(&target->d_parent->d_lock);
2547 * When switching names, the actual string doesn't strictly have to
2548 * be preserved in the target - because we're dropping the target
2549 * anyway. As such, we can just do a simple memcpy() to copy over
2550 * the new name before we switch.
2552 * Note that we have to be a lot more careful about getting the hash
2553 * switched - we have to switch the hash value properly even if it
2554 * then no longer matches the actual (corrupted) string of the target.
2555 * The hash value has to match the hash queue that the dentry is on..
2558 * __d_move - move a dentry
2559 * @dentry: entry to move
2560 * @target: new dentry
2562 * Update the dcache to reflect the move of a file name. Negative
2563 * dcache entries should not be moved in this way. Caller must hold
2564 * rename_lock, the i_mutex of the source and target directories,
2565 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2567 static void __d_move(struct dentry * dentry, struct dentry * target)
2569 if (!dentry->d_inode)
2570 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2572 BUG_ON(d_ancestor(dentry, target));
2573 BUG_ON(d_ancestor(target, dentry));
2575 dentry_lock_for_move(dentry, target);
2577 write_seqcount_begin(&dentry->d_seq);
2578 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2580 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2583 * Move the dentry to the target hash queue. Don't bother checking
2584 * for the same hash queue because of how unlikely it is.
2587 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2589 /* Unhash the target: dput() will then get rid of it */
2592 list_del(&dentry->d_u.d_child);
2593 list_del(&target->d_u.d_child);
2595 /* Switch the names.. */
2596 switch_names(dentry, target);
2597 swap(dentry->d_name.hash, target->d_name.hash);
2599 /* ... and switch the parents */
2600 if (IS_ROOT(dentry)) {
2601 dentry->d_parent = target->d_parent;
2602 target->d_parent = target;
2603 INIT_LIST_HEAD(&target->d_u.d_child);
2605 swap(dentry->d_parent, target->d_parent);
2607 /* And add them back to the (new) parent lists */
2608 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2611 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2613 write_seqcount_end(&target->d_seq);
2614 write_seqcount_end(&dentry->d_seq);
2616 dentry_unlock_parents_for_move(dentry, target);
2617 spin_unlock(&target->d_lock);
2618 fsnotify_d_move(dentry);
2619 spin_unlock(&dentry->d_lock);
2623 * d_move - move a dentry
2624 * @dentry: entry to move
2625 * @target: new dentry
2627 * Update the dcache to reflect the move of a file name. Negative
2628 * dcache entries should not be moved in this way. See the locking
2629 * requirements for __d_move.
2631 void d_move(struct dentry *dentry, struct dentry *target)
2633 write_seqlock(&rename_lock);
2634 __d_move(dentry, target);
2635 write_sequnlock(&rename_lock);
2637 EXPORT_SYMBOL(d_move);
2640 * d_ancestor - search for an ancestor
2641 * @p1: ancestor dentry
2644 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2645 * an ancestor of p2, else NULL.
2647 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2651 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2652 if (p->d_parent == p1)
2659 * This helper attempts to cope with remotely renamed directories
2661 * It assumes that the caller is already holding
2662 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2664 * Note: If ever the locking in lock_rename() changes, then please
2665 * remember to update this too...
2667 static struct dentry *__d_unalias(struct inode *inode,
2668 struct dentry *dentry, struct dentry *alias)
2670 struct mutex *m1 = NULL, *m2 = NULL;
2671 struct dentry *ret = ERR_PTR(-EBUSY);
2673 /* If alias and dentry share a parent, then no extra locks required */
2674 if (alias->d_parent == dentry->d_parent)
2677 /* See lock_rename() */
2678 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2680 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2681 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2683 m2 = &alias->d_parent->d_inode->i_mutex;
2685 if (likely(!d_mountpoint(alias))) {
2686 __d_move(alias, dentry);
2690 spin_unlock(&inode->i_lock);
2699 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2700 * named dentry in place of the dentry to be replaced.
2701 * returns with anon->d_lock held!
2703 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2705 struct dentry *dparent;
2707 dentry_lock_for_move(anon, dentry);
2709 write_seqcount_begin(&dentry->d_seq);
2710 write_seqcount_begin_nested(&anon->d_seq, DENTRY_D_LOCK_NESTED);
2712 dparent = dentry->d_parent;
2714 switch_names(dentry, anon);
2715 swap(dentry->d_name.hash, anon->d_name.hash);
2717 dentry->d_parent = dentry;
2718 list_del_init(&dentry->d_u.d_child);
2719 anon->d_parent = dparent;
2720 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2722 write_seqcount_end(&dentry->d_seq);
2723 write_seqcount_end(&anon->d_seq);
2725 dentry_unlock_parents_for_move(anon, dentry);
2726 spin_unlock(&dentry->d_lock);
2728 /* anon->d_lock still locked, returns locked */
2732 * d_materialise_unique - introduce an inode into the tree
2733 * @dentry: candidate dentry
2734 * @inode: inode to bind to the dentry, to which aliases may be attached
2736 * Introduces an dentry into the tree, substituting an extant disconnected
2737 * root directory alias in its place if there is one. Caller must hold the
2738 * i_mutex of the parent directory.
2740 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2742 struct dentry *actual;
2744 BUG_ON(!d_unhashed(dentry));
2748 __d_instantiate(dentry, NULL);
2753 spin_lock(&inode->i_lock);
2755 if (S_ISDIR(inode->i_mode)) {
2756 struct dentry *alias;
2758 /* Does an aliased dentry already exist? */
2759 alias = __d_find_alias(inode, 0);
2762 write_seqlock(&rename_lock);
2764 if (d_ancestor(alias, dentry)) {
2765 /* Check for loops */
2766 actual = ERR_PTR(-ELOOP);
2767 spin_unlock(&inode->i_lock);
2768 } else if (IS_ROOT(alias)) {
2769 /* Is this an anonymous mountpoint that we
2770 * could splice into our tree? */
2771 __d_materialise_dentry(dentry, alias);
2772 write_sequnlock(&rename_lock);
2776 /* Nope, but we must(!) avoid directory
2777 * aliasing. This drops inode->i_lock */
2778 actual = __d_unalias(inode, dentry, alias);
2780 write_sequnlock(&rename_lock);
2781 if (IS_ERR(actual)) {
2782 if (PTR_ERR(actual) == -ELOOP)
2783 pr_warn_ratelimited(
2784 "VFS: Lookup of '%s' in %s %s"
2785 " would have caused loop\n",
2786 dentry->d_name.name,
2787 inode->i_sb->s_type->name,
2795 /* Add a unique reference */
2796 actual = __d_instantiate_unique(dentry, inode);
2800 BUG_ON(!d_unhashed(actual));
2802 spin_lock(&actual->d_lock);
2805 spin_unlock(&actual->d_lock);
2806 spin_unlock(&inode->i_lock);
2808 if (actual == dentry) {
2809 security_d_instantiate(dentry, inode);
2816 EXPORT_SYMBOL_GPL(d_materialise_unique);
2818 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2822 return -ENAMETOOLONG;
2824 memcpy(*buffer, str, namelen);
2829 * prepend_name - prepend a pathname in front of current buffer pointer
2830 * @buffer: buffer pointer
2831 * @buflen: allocated length of the buffer
2832 * @name: name string and length qstr structure
2834 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2835 * make sure that either the old or the new name pointer and length are
2836 * fetched. However, there may be mismatch between length and pointer.
2837 * The length cannot be trusted, we need to copy it byte-by-byte until
2838 * the length is reached or a null byte is found. It also prepends "/" at
2839 * the beginning of the name. The sequence number check at the caller will
2840 * retry it again when a d_move() does happen. So any garbage in the buffer
2841 * due to mismatched pointer and length will be discarded.
2843 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2845 const char *dname = ACCESS_ONCE(name->name);
2846 u32 dlen = ACCESS_ONCE(name->len);
2849 if (*buflen < dlen + 1)
2850 return -ENAMETOOLONG;
2851 *buflen -= dlen + 1;
2852 p = *buffer -= dlen + 1;
2864 * prepend_path - Prepend path string to a buffer
2865 * @path: the dentry/vfsmount to report
2866 * @root: root vfsmnt/dentry
2867 * @buffer: pointer to the end of the buffer
2868 * @buflen: pointer to buffer length
2870 * The function will first try to write out the pathname without taking any
2871 * lock other than the RCU read lock to make sure that dentries won't go away.
2872 * It only checks the sequence number of the global rename_lock as any change
2873 * in the dentry's d_seq will be preceded by changes in the rename_lock
2874 * sequence number. If the sequence number had been changed, it will restart
2875 * the whole pathname back-tracing sequence again by taking the rename_lock.
2876 * In this case, there is no need to take the RCU read lock as the recursive
2877 * parent pointer references will keep the dentry chain alive as long as no
2878 * rename operation is performed.
2880 static int prepend_path(const struct path *path,
2881 const struct path *root,
2882 char **buffer, int *buflen)
2884 struct dentry *dentry;
2885 struct vfsmount *vfsmnt;
2888 unsigned seq, m_seq = 0;
2894 read_seqbegin_or_lock(&mount_lock, &m_seq);
2901 dentry = path->dentry;
2903 mnt = real_mount(vfsmnt);
2904 read_seqbegin_or_lock(&rename_lock, &seq);
2905 while (dentry != root->dentry || vfsmnt != root->mnt) {
2906 struct dentry * parent;
2908 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2909 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2911 if (mnt != parent) {
2912 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2918 * Filesystems needing to implement special "root names"
2919 * should do so with ->d_dname()
2921 if (IS_ROOT(dentry) &&
2922 (dentry->d_name.len != 1 ||
2923 dentry->d_name.name[0] != '/')) {
2924 WARN(1, "Root dentry has weird name <%.*s>\n",
2925 (int) dentry->d_name.len,
2926 dentry->d_name.name);
2929 error = is_mounted(vfsmnt) ? 1 : 2;
2932 parent = dentry->d_parent;
2934 error = prepend_name(&bptr, &blen, &dentry->d_name);
2942 if (need_seqretry(&rename_lock, seq)) {
2946 done_seqretry(&rename_lock, seq);
2950 if (need_seqretry(&mount_lock, m_seq)) {
2954 done_seqretry(&mount_lock, m_seq);
2956 if (error >= 0 && bptr == *buffer) {
2958 error = -ENAMETOOLONG;
2968 * __d_path - return the path of a dentry
2969 * @path: the dentry/vfsmount to report
2970 * @root: root vfsmnt/dentry
2971 * @buf: buffer to return value in
2972 * @buflen: buffer length
2974 * Convert a dentry into an ASCII path name.
2976 * Returns a pointer into the buffer or an error code if the
2977 * path was too long.
2979 * "buflen" should be positive.
2981 * If the path is not reachable from the supplied root, return %NULL.
2983 char *__d_path(const struct path *path,
2984 const struct path *root,
2985 char *buf, int buflen)
2987 char *res = buf + buflen;
2990 prepend(&res, &buflen, "\0", 1);
2991 error = prepend_path(path, root, &res, &buflen);
2994 return ERR_PTR(error);
3000 char *d_absolute_path(const struct path *path,
3001 char *buf, int buflen)
3003 struct path root = {};
3004 char *res = buf + buflen;
3007 prepend(&res, &buflen, "\0", 1);
3008 error = prepend_path(path, &root, &res, &buflen);
3013 return ERR_PTR(error);
3018 * same as __d_path but appends "(deleted)" for unlinked files.
3020 static int path_with_deleted(const struct path *path,
3021 const struct path *root,
3022 char **buf, int *buflen)
3024 prepend(buf, buflen, "\0", 1);
3025 if (d_unlinked(path->dentry)) {
3026 int error = prepend(buf, buflen, " (deleted)", 10);
3031 return prepend_path(path, root, buf, buflen);
3034 static int prepend_unreachable(char **buffer, int *buflen)
3036 return prepend(buffer, buflen, "(unreachable)", 13);
3039 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3044 seq = read_seqcount_begin(&fs->seq);
3046 } while (read_seqcount_retry(&fs->seq, seq));
3050 * d_path - return the path of a dentry
3051 * @path: path to report
3052 * @buf: buffer to return value in
3053 * @buflen: buffer length
3055 * Convert a dentry into an ASCII path name. If the entry has been deleted
3056 * the string " (deleted)" is appended. Note that this is ambiguous.
3058 * Returns a pointer into the buffer or an error code if the path was
3059 * too long. Note: Callers should use the returned pointer, not the passed
3060 * in buffer, to use the name! The implementation often starts at an offset
3061 * into the buffer, and may leave 0 bytes at the start.
3063 * "buflen" should be positive.
3065 char *d_path(const struct path *path, char *buf, int buflen)
3067 char *res = buf + buflen;
3072 * We have various synthetic filesystems that never get mounted. On
3073 * these filesystems dentries are never used for lookup purposes, and
3074 * thus don't need to be hashed. They also don't need a name until a
3075 * user wants to identify the object in /proc/pid/fd/. The little hack
3076 * below allows us to generate a name for these objects on demand:
3078 if (path->dentry->d_op && path->dentry->d_op->d_dname)
3079 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3082 get_fs_root_rcu(current->fs, &root);
3083 error = path_with_deleted(path, &root, &res, &buflen);
3087 res = ERR_PTR(error);
3090 EXPORT_SYMBOL(d_path);
3093 * Helper function for dentry_operations.d_dname() members
3095 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3096 const char *fmt, ...)
3102 va_start(args, fmt);
3103 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3106 if (sz > sizeof(temp) || sz > buflen)
3107 return ERR_PTR(-ENAMETOOLONG);
3109 buffer += buflen - sz;
3110 return memcpy(buffer, temp, sz);
3113 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3115 char *end = buffer + buflen;
3116 /* these dentries are never renamed, so d_lock is not needed */
3117 if (prepend(&end, &buflen, " (deleted)", 11) ||
3118 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3119 prepend(&end, &buflen, "/", 1))
3120 end = ERR_PTR(-ENAMETOOLONG);
3125 * Write full pathname from the root of the filesystem into the buffer.
3127 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
3137 prepend(&end, &len, "\0", 1);
3143 read_seqbegin_or_lock(&rename_lock, &seq);
3144 while (!IS_ROOT(dentry)) {
3145 struct dentry *parent = dentry->d_parent;
3149 error = prepend_name(&end, &len, &dentry->d_name);
3158 if (need_seqretry(&rename_lock, seq)) {
3162 done_seqretry(&rename_lock, seq);
3167 return ERR_PTR(-ENAMETOOLONG);
3170 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3172 return __dentry_path(dentry, buf, buflen);
3174 EXPORT_SYMBOL(dentry_path_raw);
3176 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3181 if (d_unlinked(dentry)) {
3183 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3187 retval = __dentry_path(dentry, buf, buflen);
3188 if (!IS_ERR(retval) && p)
3189 *p = '/'; /* restore '/' overriden with '\0' */
3192 return ERR_PTR(-ENAMETOOLONG);
3195 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3201 seq = read_seqcount_begin(&fs->seq);
3204 } while (read_seqcount_retry(&fs->seq, seq));
3208 * NOTE! The user-level library version returns a
3209 * character pointer. The kernel system call just
3210 * returns the length of the buffer filled (which
3211 * includes the ending '\0' character), or a negative
3212 * error value. So libc would do something like
3214 * char *getcwd(char * buf, size_t size)
3218 * retval = sys_getcwd(buf, size);
3225 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3228 struct path pwd, root;
3229 char *page = __getname();
3235 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3238 if (!d_unlinked(pwd.dentry)) {
3240 char *cwd = page + PATH_MAX;
3241 int buflen = PATH_MAX;
3243 prepend(&cwd, &buflen, "\0", 1);
3244 error = prepend_path(&pwd, &root, &cwd, &buflen);
3250 /* Unreachable from current root */
3252 error = prepend_unreachable(&cwd, &buflen);
3258 len = PATH_MAX + page - cwd;
3261 if (copy_to_user(buf, cwd, len))
3274 * Test whether new_dentry is a subdirectory of old_dentry.
3276 * Trivially implemented using the dcache structure
3280 * is_subdir - is new dentry a subdirectory of old_dentry
3281 * @new_dentry: new dentry
3282 * @old_dentry: old dentry
3284 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3285 * Returns 0 otherwise.
3286 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3289 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3294 if (new_dentry == old_dentry)
3298 /* for restarting inner loop in case of seq retry */
3299 seq = read_seqbegin(&rename_lock);
3301 * Need rcu_readlock to protect against the d_parent trashing
3305 if (d_ancestor(old_dentry, new_dentry))
3310 } while (read_seqretry(&rename_lock, seq));
3315 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3317 struct dentry *root = data;
3318 if (dentry != root) {
3319 if (d_unhashed(dentry) || !dentry->d_inode)
3322 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3323 dentry->d_flags |= DCACHE_GENOCIDE;
3324 dentry->d_lockref.count--;
3327 return D_WALK_CONTINUE;
3330 void d_genocide(struct dentry *parent)
3332 d_walk(parent, parent, d_genocide_kill, NULL);
3335 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3337 inode_dec_link_count(inode);
3338 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3339 !hlist_unhashed(&dentry->d_alias) ||
3340 !d_unlinked(dentry));
3341 spin_lock(&dentry->d_parent->d_lock);
3342 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3343 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3344 (unsigned long long)inode->i_ino);
3345 spin_unlock(&dentry->d_lock);
3346 spin_unlock(&dentry->d_parent->d_lock);
3347 d_instantiate(dentry, inode);
3349 EXPORT_SYMBOL(d_tmpfile);
3351 static __initdata unsigned long dhash_entries;
3352 static int __init set_dhash_entries(char *str)
3356 dhash_entries = simple_strtoul(str, &str, 0);
3359 __setup("dhash_entries=", set_dhash_entries);
3361 static void __init dcache_init_early(void)
3365 /* If hashes are distributed across NUMA nodes, defer
3366 * hash allocation until vmalloc space is available.
3372 alloc_large_system_hash("Dentry cache",
3373 sizeof(struct hlist_bl_head),
3382 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3383 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3386 static void __init dcache_init(void)
3391 * A constructor could be added for stable state like the lists,
3392 * but it is probably not worth it because of the cache nature
3395 dentry_cache = KMEM_CACHE(dentry,
3396 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3398 /* Hash may have been set up in dcache_init_early */
3403 alloc_large_system_hash("Dentry cache",
3404 sizeof(struct hlist_bl_head),
3413 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3414 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3417 /* SLAB cache for __getname() consumers */
3418 struct kmem_cache *names_cachep __read_mostly;
3419 EXPORT_SYMBOL(names_cachep);
3421 EXPORT_SYMBOL(d_genocide);
3423 void __init vfs_caches_init_early(void)
3425 dcache_init_early();
3429 void __init vfs_caches_init(unsigned long mempages)
3431 unsigned long reserve;
3433 /* Base hash sizes on available memory, with a reserve equal to
3434 150% of current kernel size */
3436 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3437 mempages -= reserve;
3439 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3440 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3444 files_init(mempages);