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/delay.h>
23 #include <linux/slab.h>
24 #include <linux/init.h>
25 #include <linux/hash.h>
26 #include <linux/cache.h>
27 #include <linux/export.h>
28 #include <linux/mount.h>
29 #include <linux/file.h>
30 #include <asm/uaccess.h>
31 #include <linux/security.h>
32 #include <linux/seqlock.h>
33 #include <linux/swap.h>
34 #include <linux/bootmem.h>
35 #include <linux/fs_struct.h>
36 #include <linux/hardirq.h>
37 #include <linux/bit_spinlock.h>
38 #include <linux/rculist_bl.h>
39 #include <linux/prefetch.h>
40 #include <linux/ratelimit.h>
41 #include <linux/list_lru.h>
42 #include <linux/kasan.h>
49 * dcache->d_inode->i_lock protects:
50 * - i_dentry, d_u.d_alias, d_inode of aliases
51 * dcache_hash_bucket lock protects:
52 * - the dcache hash table
53 * s_anon bl list spinlock protects:
54 * - the s_anon list (see __d_drop)
55 * dentry->d_sb->s_dentry_lru_lock protects:
56 * - the dcache lru lists and counters
63 * - d_parent and d_subdirs
64 * - childrens' d_child and d_parent
65 * - d_u.d_alias, d_inode
68 * dentry->d_inode->i_lock
70 * dentry->d_sb->s_dentry_lru_lock
71 * dcache_hash_bucket lock
74 * If there is an ancestor relationship:
75 * dentry->d_parent->...->d_parent->d_lock
77 * dentry->d_parent->d_lock
80 * If no ancestor relationship:
81 * if (dentry1 < dentry2)
85 int sysctl_vfs_cache_pressure __read_mostly = 100;
86 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
88 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
90 EXPORT_SYMBOL(rename_lock);
92 static struct kmem_cache *dentry_cache __read_mostly;
95 * This is the single most critical data structure when it comes
96 * to the dcache: the hashtable for lookups. Somebody should try
97 * to make this good - I've just made it work.
99 * This hash-function tries to avoid losing too many bits of hash
100 * information, yet avoid using a prime hash-size or similar.
103 static unsigned int d_hash_mask __read_mostly;
104 static unsigned int d_hash_shift __read_mostly;
106 static struct hlist_bl_head *dentry_hashtable __read_mostly;
108 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
111 hash += (unsigned long) parent / L1_CACHE_BYTES;
112 return dentry_hashtable + hash_32(hash, d_hash_shift);
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat = {
120 static DEFINE_PER_CPU(long, nr_dentry);
121 static DEFINE_PER_CPU(long, nr_dentry_unused);
123 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
126 * Here we resort to our own counters instead of using generic per-cpu counters
127 * for consistency with what the vfs inode code does. We are expected to harvest
128 * better code and performance by having our own specialized counters.
130 * Please note that the loop is done over all possible CPUs, not over all online
131 * CPUs. The reason for this is that we don't want to play games with CPUs going
132 * on and off. If one of them goes off, we will just keep their counters.
134 * glommer: See cffbc8a for details, and if you ever intend to change this,
135 * please update all vfs counters to match.
137 static long get_nr_dentry(void)
141 for_each_possible_cpu(i)
142 sum += per_cpu(nr_dentry, i);
143 return sum < 0 ? 0 : sum;
146 static long get_nr_dentry_unused(void)
150 for_each_possible_cpu(i)
151 sum += per_cpu(nr_dentry_unused, i);
152 return sum < 0 ? 0 : sum;
155 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
156 size_t *lenp, loff_t *ppos)
158 dentry_stat.nr_dentry = get_nr_dentry();
159 dentry_stat.nr_unused = get_nr_dentry_unused();
160 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
165 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
166 * The strings are both count bytes long, and count is non-zero.
168 #ifdef CONFIG_DCACHE_WORD_ACCESS
170 #include <asm/word-at-a-time.h>
172 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
173 * aligned allocation for this particular component. We don't
174 * strictly need the load_unaligned_zeropad() safety, but it
175 * doesn't hurt either.
177 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
178 * need the careful unaligned handling.
180 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
182 unsigned long a,b,mask;
185 a = *(unsigned long *)cs;
186 b = load_unaligned_zeropad(ct);
187 if (tcount < sizeof(unsigned long))
189 if (unlikely(a != b))
191 cs += sizeof(unsigned long);
192 ct += sizeof(unsigned long);
193 tcount -= sizeof(unsigned long);
197 mask = bytemask_from_count(tcount);
198 return unlikely(!!((a ^ b) & mask));
203 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
217 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
219 const unsigned char *cs;
221 * Be careful about RCU walk racing with rename:
222 * use ACCESS_ONCE to fetch the name pointer.
224 * NOTE! Even if a rename will mean that the length
225 * was not loaded atomically, we don't care. The
226 * RCU walk will check the sequence count eventually,
227 * and catch it. And we won't overrun the buffer,
228 * because we're reading the name pointer atomically,
229 * and a dentry name is guaranteed to be properly
230 * terminated with a NUL byte.
232 * End result: even if 'len' is wrong, we'll exit
233 * early because the data cannot match (there can
234 * be no NUL in the ct/tcount data)
236 cs = ACCESS_ONCE(dentry->d_name.name);
237 smp_read_barrier_depends();
238 return dentry_string_cmp(cs, ct, tcount);
241 struct external_name {
244 struct rcu_head head;
246 unsigned char name[];
249 static inline struct external_name *external_name(struct dentry *dentry)
251 return container_of(dentry->d_name.name, struct external_name, name[0]);
254 static void __d_free(struct rcu_head *head)
256 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
258 kmem_cache_free(dentry_cache, dentry);
261 static void __d_free_external(struct rcu_head *head)
263 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
264 kfree(external_name(dentry));
265 kmem_cache_free(dentry_cache, dentry);
268 static inline int dname_external(const struct dentry *dentry)
270 return dentry->d_name.name != dentry->d_iname;
273 static inline void __d_set_inode_and_type(struct dentry *dentry,
279 dentry->d_inode = inode;
280 flags = READ_ONCE(dentry->d_flags);
281 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
283 WRITE_ONCE(dentry->d_flags, flags);
286 static inline void __d_clear_type_and_inode(struct dentry *dentry)
288 unsigned flags = READ_ONCE(dentry->d_flags);
290 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
291 WRITE_ONCE(dentry->d_flags, flags);
292 dentry->d_inode = NULL;
295 static void dentry_free(struct dentry *dentry)
297 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
298 if (unlikely(dname_external(dentry))) {
299 struct external_name *p = external_name(dentry);
300 if (likely(atomic_dec_and_test(&p->u.count))) {
301 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
305 /* if dentry was never visible to RCU, immediate free is OK */
306 if (!(dentry->d_flags & DCACHE_RCUACCESS))
307 __d_free(&dentry->d_u.d_rcu);
309 call_rcu(&dentry->d_u.d_rcu, __d_free);
313 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
314 * @dentry: the target dentry
315 * After this call, in-progress rcu-walk path lookup will fail. This
316 * should be called after unhashing, and after changing d_inode (if
317 * the dentry has not already been unhashed).
319 static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
321 lockdep_assert_held(&dentry->d_lock);
322 /* Go through am invalidation barrier */
323 write_seqcount_invalidate(&dentry->d_seq);
327 * Release the dentry's inode, using the filesystem
328 * d_iput() operation if defined. Dentry has no refcount
331 static void dentry_iput(struct dentry * dentry)
332 __releases(dentry->d_lock)
333 __releases(dentry->d_inode->i_lock)
335 struct inode *inode = dentry->d_inode;
337 __d_clear_type_and_inode(dentry);
338 hlist_del_init(&dentry->d_u.d_alias);
339 spin_unlock(&dentry->d_lock);
340 spin_unlock(&inode->i_lock);
342 fsnotify_inoderemove(inode);
343 if (dentry->d_op && dentry->d_op->d_iput)
344 dentry->d_op->d_iput(dentry, inode);
348 spin_unlock(&dentry->d_lock);
353 * Release the dentry's inode, using the filesystem
354 * d_iput() operation if defined. dentry remains in-use.
356 static void dentry_unlink_inode(struct dentry * dentry)
357 __releases(dentry->d_lock)
358 __releases(dentry->d_inode->i_lock)
360 struct inode *inode = dentry->d_inode;
362 raw_write_seqcount_begin(&dentry->d_seq);
363 __d_clear_type_and_inode(dentry);
364 hlist_del_init(&dentry->d_u.d_alias);
365 raw_write_seqcount_end(&dentry->d_seq);
366 spin_unlock(&dentry->d_lock);
367 spin_unlock(&inode->i_lock);
369 fsnotify_inoderemove(inode);
370 if (dentry->d_op && dentry->d_op->d_iput)
371 dentry->d_op->d_iput(dentry, inode);
377 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
378 * is in use - which includes both the "real" per-superblock
379 * LRU list _and_ the DCACHE_SHRINK_LIST use.
381 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
382 * on the shrink list (ie not on the superblock LRU list).
384 * The per-cpu "nr_dentry_unused" counters are updated with
385 * the DCACHE_LRU_LIST bit.
387 * These helper functions make sure we always follow the
388 * rules. d_lock must be held by the caller.
390 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
391 static void d_lru_add(struct dentry *dentry)
393 D_FLAG_VERIFY(dentry, 0);
394 dentry->d_flags |= DCACHE_LRU_LIST;
395 this_cpu_inc(nr_dentry_unused);
396 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
399 static void d_lru_del(struct dentry *dentry)
401 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
402 dentry->d_flags &= ~DCACHE_LRU_LIST;
403 this_cpu_dec(nr_dentry_unused);
404 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
407 static void d_shrink_del(struct dentry *dentry)
409 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
410 list_del_init(&dentry->d_lru);
411 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
412 this_cpu_dec(nr_dentry_unused);
415 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
417 D_FLAG_VERIFY(dentry, 0);
418 list_add(&dentry->d_lru, list);
419 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
420 this_cpu_inc(nr_dentry_unused);
424 * These can only be called under the global LRU lock, ie during the
425 * callback for freeing the LRU list. "isolate" removes it from the
426 * LRU lists entirely, while shrink_move moves it to the indicated
429 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
431 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
432 dentry->d_flags &= ~DCACHE_LRU_LIST;
433 this_cpu_dec(nr_dentry_unused);
434 list_lru_isolate(lru, &dentry->d_lru);
437 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
438 struct list_head *list)
440 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
441 dentry->d_flags |= DCACHE_SHRINK_LIST;
442 list_lru_isolate_move(lru, &dentry->d_lru, list);
446 * dentry_lru_(add|del)_list) must be called with d_lock held.
448 static void dentry_lru_add(struct dentry *dentry)
450 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
455 * d_drop - drop a dentry
456 * @dentry: dentry to drop
458 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
459 * be found through a VFS lookup any more. Note that this is different from
460 * deleting the dentry - d_delete will try to mark the dentry negative if
461 * possible, giving a successful _negative_ lookup, while d_drop will
462 * just make the cache lookup fail.
464 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
465 * reason (NFS timeouts or autofs deletes).
467 * __d_drop requires dentry->d_lock.
469 void __d_drop(struct dentry *dentry)
471 if (!d_unhashed(dentry)) {
472 struct hlist_bl_head *b;
474 * Hashed dentries are normally on the dentry hashtable,
475 * with the exception of those newly allocated by
476 * d_obtain_alias, which are always IS_ROOT:
478 if (unlikely(IS_ROOT(dentry)))
479 b = &dentry->d_sb->s_anon;
481 b = d_hash(dentry->d_parent, dentry->d_name.hash);
484 __hlist_bl_del(&dentry->d_hash);
485 dentry->d_hash.pprev = NULL;
487 dentry_rcuwalk_invalidate(dentry);
490 EXPORT_SYMBOL(__d_drop);
492 void d_drop(struct dentry *dentry)
494 spin_lock(&dentry->d_lock);
496 spin_unlock(&dentry->d_lock);
498 EXPORT_SYMBOL(d_drop);
500 static void __dentry_kill(struct dentry *dentry)
502 struct dentry *parent = NULL;
503 bool can_free = true;
504 if (!IS_ROOT(dentry))
505 parent = dentry->d_parent;
508 * The dentry is now unrecoverably dead to the world.
510 lockref_mark_dead(&dentry->d_lockref);
513 * inform the fs via d_prune that this dentry is about to be
514 * unhashed and destroyed.
516 if (dentry->d_flags & DCACHE_OP_PRUNE)
517 dentry->d_op->d_prune(dentry);
519 if (dentry->d_flags & DCACHE_LRU_LIST) {
520 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
523 /* if it was on the hash then remove it */
525 __list_del_entry(&dentry->d_child);
527 * Inform d_walk() that we are no longer attached to the
530 dentry->d_flags |= DCACHE_DENTRY_KILLED;
532 spin_unlock(&parent->d_lock);
535 * dentry_iput drops the locks, at which point nobody (except
536 * transient RCU lookups) can reach this dentry.
538 BUG_ON(dentry->d_lockref.count > 0);
539 this_cpu_dec(nr_dentry);
540 if (dentry->d_op && dentry->d_op->d_release)
541 dentry->d_op->d_release(dentry);
543 spin_lock(&dentry->d_lock);
544 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
545 dentry->d_flags |= DCACHE_MAY_FREE;
548 spin_unlock(&dentry->d_lock);
549 if (likely(can_free))
554 * Finish off a dentry we've decided to kill.
555 * dentry->d_lock must be held, returns with it unlocked.
556 * If ref is non-zero, then decrement the refcount too.
557 * Returns dentry requiring refcount drop, or NULL if we're done.
559 static struct dentry *dentry_kill(struct dentry *dentry)
560 __releases(dentry->d_lock)
562 struct inode *inode = dentry->d_inode;
563 struct dentry *parent = NULL;
565 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
568 if (!IS_ROOT(dentry)) {
569 parent = dentry->d_parent;
570 if (unlikely(!spin_trylock(&parent->d_lock))) {
572 spin_unlock(&inode->i_lock);
577 __dentry_kill(dentry);
581 spin_unlock(&dentry->d_lock);
582 return dentry; /* try again with same dentry */
585 static inline struct dentry *lock_parent(struct dentry *dentry)
587 struct dentry *parent = dentry->d_parent;
590 if (unlikely(dentry->d_lockref.count < 0))
592 if (likely(spin_trylock(&parent->d_lock)))
595 spin_unlock(&dentry->d_lock);
597 parent = ACCESS_ONCE(dentry->d_parent);
598 spin_lock(&parent->d_lock);
600 * We can't blindly lock dentry until we are sure
601 * that we won't violate the locking order.
602 * Any changes of dentry->d_parent must have
603 * been done with parent->d_lock held, so
604 * spin_lock() above is enough of a barrier
605 * for checking if it's still our child.
607 if (unlikely(parent != dentry->d_parent)) {
608 spin_unlock(&parent->d_lock);
612 if (parent != dentry)
613 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
620 * Try to do a lockless dput(), and return whether that was successful.
622 * If unsuccessful, we return false, having already taken the dentry lock.
624 * The caller needs to hold the RCU read lock, so that the dentry is
625 * guaranteed to stay around even if the refcount goes down to zero!
627 static inline bool fast_dput(struct dentry *dentry)
630 unsigned int d_flags;
633 * If we have a d_op->d_delete() operation, we sould not
634 * let the dentry count go to zero, so use "put_or_lock".
636 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
637 return lockref_put_or_lock(&dentry->d_lockref);
640 * .. otherwise, we can try to just decrement the
641 * lockref optimistically.
643 ret = lockref_put_return(&dentry->d_lockref);
646 * If the lockref_put_return() failed due to the lock being held
647 * by somebody else, the fast path has failed. We will need to
648 * get the lock, and then check the count again.
650 if (unlikely(ret < 0)) {
651 spin_lock(&dentry->d_lock);
652 if (dentry->d_lockref.count > 1) {
653 dentry->d_lockref.count--;
654 spin_unlock(&dentry->d_lock);
661 * If we weren't the last ref, we're done.
667 * Careful, careful. The reference count went down
668 * to zero, but we don't hold the dentry lock, so
669 * somebody else could get it again, and do another
670 * dput(), and we need to not race with that.
672 * However, there is a very special and common case
673 * where we don't care, because there is nothing to
674 * do: the dentry is still hashed, it does not have
675 * a 'delete' op, and it's referenced and already on
678 * NOTE! Since we aren't locked, these values are
679 * not "stable". However, it is sufficient that at
680 * some point after we dropped the reference the
681 * dentry was hashed and the flags had the proper
682 * value. Other dentry users may have re-gotten
683 * a reference to the dentry and change that, but
684 * our work is done - we can leave the dentry
685 * around with a zero refcount.
688 d_flags = ACCESS_ONCE(dentry->d_flags);
689 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
691 /* Nothing to do? Dropping the reference was all we needed? */
692 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
696 * Not the fast normal case? Get the lock. We've already decremented
697 * the refcount, but we'll need to re-check the situation after
700 spin_lock(&dentry->d_lock);
703 * Did somebody else grab a reference to it in the meantime, and
704 * we're no longer the last user after all? Alternatively, somebody
705 * else could have killed it and marked it dead. Either way, we
706 * don't need to do anything else.
708 if (dentry->d_lockref.count) {
709 spin_unlock(&dentry->d_lock);
714 * Re-get the reference we optimistically dropped. We hold the
715 * lock, and we just tested that it was zero, so we can just
718 dentry->d_lockref.count = 1;
726 * This is complicated by the fact that we do not want to put
727 * dentries that are no longer on any hash chain on the unused
728 * list: we'd much rather just get rid of them immediately.
730 * However, that implies that we have to traverse the dentry
731 * tree upwards to the parents which might _also_ now be
732 * scheduled for deletion (it may have been only waiting for
733 * its last child to go away).
735 * This tail recursion is done by hand as we don't want to depend
736 * on the compiler to always get this right (gcc generally doesn't).
737 * Real recursion would eat up our stack space.
741 * dput - release a dentry
742 * @dentry: dentry to release
744 * Release a dentry. This will drop the usage count and if appropriate
745 * call the dentry unlink method as well as removing it from the queues and
746 * releasing its resources. If the parent dentries were scheduled for release
747 * they too may now get deleted.
749 void dput(struct dentry *dentry)
751 struct dentry *parent;
753 if (unlikely(!dentry))
760 if (likely(fast_dput(dentry))) {
765 /* Slow case: now with the dentry lock held */
768 /* Unreachable? Get rid of it */
769 if (unlikely(d_unhashed(dentry)))
772 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
775 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
776 if (dentry->d_op->d_delete(dentry))
780 if (!(dentry->d_flags & DCACHE_REFERENCED))
781 dentry->d_flags |= DCACHE_REFERENCED;
782 dentry_lru_add(dentry);
784 dentry->d_lockref.count--;
785 spin_unlock(&dentry->d_lock);
789 parent = dentry_kill(dentry);
793 if (parent == dentry) {
794 /* the task with the highest priority won't schedule */
807 /* This must be called with d_lock held */
808 static inline void __dget_dlock(struct dentry *dentry)
810 dentry->d_lockref.count++;
813 static inline void __dget(struct dentry *dentry)
815 lockref_get(&dentry->d_lockref);
818 struct dentry *dget_parent(struct dentry *dentry)
824 * Do optimistic parent lookup without any
828 ret = ACCESS_ONCE(dentry->d_parent);
829 gotref = lockref_get_not_zero(&ret->d_lockref);
831 if (likely(gotref)) {
832 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
839 * Don't need rcu_dereference because we re-check it was correct under
843 ret = dentry->d_parent;
844 spin_lock(&ret->d_lock);
845 if (unlikely(ret != dentry->d_parent)) {
846 spin_unlock(&ret->d_lock);
851 BUG_ON(!ret->d_lockref.count);
852 ret->d_lockref.count++;
853 spin_unlock(&ret->d_lock);
856 EXPORT_SYMBOL(dget_parent);
859 * d_find_alias - grab a hashed alias of inode
860 * @inode: inode in question
862 * If inode has a hashed alias, or is a directory and has any alias,
863 * acquire the reference to alias and return it. Otherwise return NULL.
864 * Notice that if inode is a directory there can be only one alias and
865 * it can be unhashed only if it has no children, or if it is the root
866 * of a filesystem, or if the directory was renamed and d_revalidate
867 * was the first vfs operation to notice.
869 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
870 * any other hashed alias over that one.
872 static struct dentry *__d_find_alias(struct inode *inode)
874 struct dentry *alias, *discon_alias;
878 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
879 spin_lock(&alias->d_lock);
880 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
881 if (IS_ROOT(alias) &&
882 (alias->d_flags & DCACHE_DISCONNECTED)) {
883 discon_alias = alias;
886 spin_unlock(&alias->d_lock);
890 spin_unlock(&alias->d_lock);
893 alias = discon_alias;
894 spin_lock(&alias->d_lock);
895 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
897 spin_unlock(&alias->d_lock);
900 spin_unlock(&alias->d_lock);
906 struct dentry *d_find_alias(struct inode *inode)
908 struct dentry *de = NULL;
910 if (!hlist_empty(&inode->i_dentry)) {
911 spin_lock(&inode->i_lock);
912 de = __d_find_alias(inode);
913 spin_unlock(&inode->i_lock);
917 EXPORT_SYMBOL(d_find_alias);
920 * Try to kill dentries associated with this inode.
921 * WARNING: you must own a reference to inode.
923 void d_prune_aliases(struct inode *inode)
925 struct dentry *dentry;
927 spin_lock(&inode->i_lock);
928 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
929 spin_lock(&dentry->d_lock);
930 if (!dentry->d_lockref.count) {
931 struct dentry *parent = lock_parent(dentry);
932 if (likely(!dentry->d_lockref.count)) {
933 __dentry_kill(dentry);
938 spin_unlock(&parent->d_lock);
940 spin_unlock(&dentry->d_lock);
942 spin_unlock(&inode->i_lock);
944 EXPORT_SYMBOL(d_prune_aliases);
946 static void shrink_dentry_list(struct list_head *list)
948 struct dentry *dentry, *parent;
950 while (!list_empty(list)) {
952 dentry = list_entry(list->prev, struct dentry, d_lru);
953 spin_lock(&dentry->d_lock);
954 parent = lock_parent(dentry);
957 * The dispose list is isolated and dentries are not accounted
958 * to the LRU here, so we can simply remove it from the list
959 * here regardless of whether it is referenced or not.
961 d_shrink_del(dentry);
964 * We found an inuse dentry which was not removed from
965 * the LRU because of laziness during lookup. Do not free it.
967 if (dentry->d_lockref.count > 0) {
968 spin_unlock(&dentry->d_lock);
970 spin_unlock(&parent->d_lock);
975 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
976 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
977 spin_unlock(&dentry->d_lock);
979 spin_unlock(&parent->d_lock);
985 inode = dentry->d_inode;
986 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
987 d_shrink_add(dentry, list);
988 spin_unlock(&dentry->d_lock);
990 spin_unlock(&parent->d_lock);
994 __dentry_kill(dentry);
997 * We need to prune ancestors too. This is necessary to prevent
998 * quadratic behavior of shrink_dcache_parent(), but is also
999 * expected to be beneficial in reducing dentry cache
1003 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1004 parent = lock_parent(dentry);
1005 if (dentry->d_lockref.count != 1) {
1006 dentry->d_lockref.count--;
1007 spin_unlock(&dentry->d_lock);
1009 spin_unlock(&parent->d_lock);
1012 inode = dentry->d_inode; /* can't be NULL */
1013 if (unlikely(!spin_trylock(&inode->i_lock))) {
1014 spin_unlock(&dentry->d_lock);
1016 spin_unlock(&parent->d_lock);
1020 __dentry_kill(dentry);
1026 static enum lru_status dentry_lru_isolate(struct list_head *item,
1027 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1029 struct list_head *freeable = arg;
1030 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1034 * we are inverting the lru lock/dentry->d_lock here,
1035 * so use a trylock. If we fail to get the lock, just skip
1038 if (!spin_trylock(&dentry->d_lock))
1042 * Referenced dentries are still in use. If they have active
1043 * counts, just remove them from the LRU. Otherwise give them
1044 * another pass through the LRU.
1046 if (dentry->d_lockref.count) {
1047 d_lru_isolate(lru, dentry);
1048 spin_unlock(&dentry->d_lock);
1052 if (dentry->d_flags & DCACHE_REFERENCED) {
1053 dentry->d_flags &= ~DCACHE_REFERENCED;
1054 spin_unlock(&dentry->d_lock);
1057 * The list move itself will be made by the common LRU code. At
1058 * this point, we've dropped the dentry->d_lock but keep the
1059 * lru lock. This is safe to do, since every list movement is
1060 * protected by the lru lock even if both locks are held.
1062 * This is guaranteed by the fact that all LRU management
1063 * functions are intermediated by the LRU API calls like
1064 * list_lru_add and list_lru_del. List movement in this file
1065 * only ever occur through this functions or through callbacks
1066 * like this one, that are called from the LRU API.
1068 * The only exceptions to this are functions like
1069 * shrink_dentry_list, and code that first checks for the
1070 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1071 * operating only with stack provided lists after they are
1072 * properly isolated from the main list. It is thus, always a
1078 d_lru_shrink_move(lru, dentry, freeable);
1079 spin_unlock(&dentry->d_lock);
1085 * prune_dcache_sb - shrink the dcache
1087 * @sc: shrink control, passed to list_lru_shrink_walk()
1089 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1090 * is done when we need more memory and called from the superblock shrinker
1093 * This function may fail to free any resources if all the dentries are in
1096 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1101 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1102 dentry_lru_isolate, &dispose);
1103 shrink_dentry_list(&dispose);
1107 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1108 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1110 struct list_head *freeable = arg;
1111 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1114 * we are inverting the lru lock/dentry->d_lock here,
1115 * so use a trylock. If we fail to get the lock, just skip
1118 if (!spin_trylock(&dentry->d_lock))
1121 d_lru_shrink_move(lru, dentry, freeable);
1122 spin_unlock(&dentry->d_lock);
1129 * shrink_dcache_sb - shrink dcache for a superblock
1132 * Shrink the dcache for the specified super block. This is used to free
1133 * the dcache before unmounting a file system.
1135 void shrink_dcache_sb(struct super_block *sb)
1142 freed = list_lru_walk(&sb->s_dentry_lru,
1143 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1145 this_cpu_sub(nr_dentry_unused, freed);
1146 shrink_dentry_list(&dispose);
1147 } while (freed > 0);
1149 EXPORT_SYMBOL(shrink_dcache_sb);
1152 * enum d_walk_ret - action to talke during tree walk
1153 * @D_WALK_CONTINUE: contrinue walk
1154 * @D_WALK_QUIT: quit walk
1155 * @D_WALK_NORETRY: quit when retry is needed
1156 * @D_WALK_SKIP: skip this dentry and its children
1166 * d_walk - walk the dentry tree
1167 * @parent: start of walk
1168 * @data: data passed to @enter() and @finish()
1169 * @enter: callback when first entering the dentry
1170 * @finish: callback when successfully finished the walk
1172 * The @enter() and @finish() callbacks are called with d_lock held.
1174 static void d_walk(struct dentry *parent, void *data,
1175 enum d_walk_ret (*enter)(void *, struct dentry *),
1176 void (*finish)(void *))
1178 struct dentry *this_parent;
1179 struct list_head *next;
1181 enum d_walk_ret ret;
1185 read_seqbegin_or_lock(&rename_lock, &seq);
1186 this_parent = parent;
1187 spin_lock(&this_parent->d_lock);
1189 ret = enter(data, this_parent);
1191 case D_WALK_CONTINUE:
1196 case D_WALK_NORETRY:
1201 next = this_parent->d_subdirs.next;
1203 while (next != &this_parent->d_subdirs) {
1204 struct list_head *tmp = next;
1205 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1208 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1210 ret = enter(data, dentry);
1212 case D_WALK_CONTINUE:
1215 spin_unlock(&dentry->d_lock);
1217 case D_WALK_NORETRY:
1221 spin_unlock(&dentry->d_lock);
1225 if (!list_empty(&dentry->d_subdirs)) {
1226 spin_unlock(&this_parent->d_lock);
1227 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1228 this_parent = dentry;
1229 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1232 spin_unlock(&dentry->d_lock);
1235 * All done at this level ... ascend and resume the search.
1239 if (this_parent != parent) {
1240 struct dentry *child = this_parent;
1241 this_parent = child->d_parent;
1243 spin_unlock(&child->d_lock);
1244 spin_lock(&this_parent->d_lock);
1246 /* might go back up the wrong parent if we have had a rename. */
1247 if (need_seqretry(&rename_lock, seq))
1249 /* go into the first sibling still alive */
1251 next = child->d_child.next;
1252 if (next == &this_parent->d_subdirs)
1254 child = list_entry(next, struct dentry, d_child);
1255 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1259 if (need_seqretry(&rename_lock, seq))
1266 spin_unlock(&this_parent->d_lock);
1267 done_seqretry(&rename_lock, seq);
1271 spin_unlock(&this_parent->d_lock);
1281 * Search for at least 1 mount point in the dentry's subdirs.
1282 * We descend to the next level whenever the d_subdirs
1283 * list is non-empty and continue searching.
1286 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1289 if (d_mountpoint(dentry)) {
1293 return D_WALK_CONTINUE;
1297 * have_submounts - check for mounts over a dentry
1298 * @parent: dentry to check.
1300 * Return true if the parent or its subdirectories contain
1303 int have_submounts(struct dentry *parent)
1307 d_walk(parent, &ret, check_mount, NULL);
1311 EXPORT_SYMBOL(have_submounts);
1314 * Called by mount code to set a mountpoint and check if the mountpoint is
1315 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1316 * subtree can become unreachable).
1318 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1319 * this reason take rename_lock and d_lock on dentry and ancestors.
1321 int d_set_mounted(struct dentry *dentry)
1325 write_seqlock(&rename_lock);
1326 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1327 /* Need exclusion wrt. d_invalidate() */
1328 spin_lock(&p->d_lock);
1329 if (unlikely(d_unhashed(p))) {
1330 spin_unlock(&p->d_lock);
1333 spin_unlock(&p->d_lock);
1335 spin_lock(&dentry->d_lock);
1336 if (!d_unlinked(dentry)) {
1338 if (!d_mountpoint(dentry)) {
1339 dentry->d_flags |= DCACHE_MOUNTED;
1343 spin_unlock(&dentry->d_lock);
1345 write_sequnlock(&rename_lock);
1350 * Search the dentry child list of the specified parent,
1351 * and move any unused dentries to the end of the unused
1352 * list for prune_dcache(). We descend to the next level
1353 * whenever the d_subdirs list is non-empty and continue
1356 * It returns zero iff there are no unused children,
1357 * otherwise it returns the number of children moved to
1358 * the end of the unused list. This may not be the total
1359 * number of unused children, because select_parent can
1360 * drop the lock and return early due to latency
1364 struct select_data {
1365 struct dentry *start;
1366 struct list_head dispose;
1370 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1372 struct select_data *data = _data;
1373 enum d_walk_ret ret = D_WALK_CONTINUE;
1375 if (data->start == dentry)
1378 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1381 if (dentry->d_flags & DCACHE_LRU_LIST)
1383 if (!dentry->d_lockref.count) {
1384 d_shrink_add(dentry, &data->dispose);
1389 * We can return to the caller if we have found some (this
1390 * ensures forward progress). We'll be coming back to find
1393 if (!list_empty(&data->dispose))
1394 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1400 * shrink_dcache_parent - prune dcache
1401 * @parent: parent of entries to prune
1403 * Prune the dcache to remove unused children of the parent dentry.
1405 void shrink_dcache_parent(struct dentry *parent)
1408 struct select_data data;
1410 INIT_LIST_HEAD(&data.dispose);
1411 data.start = parent;
1414 d_walk(parent, &data, select_collect, NULL);
1418 shrink_dentry_list(&data.dispose);
1422 EXPORT_SYMBOL(shrink_dcache_parent);
1424 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1426 /* it has busy descendents; complain about those instead */
1427 if (!list_empty(&dentry->d_subdirs))
1428 return D_WALK_CONTINUE;
1430 /* root with refcount 1 is fine */
1431 if (dentry == _data && dentry->d_lockref.count == 1)
1432 return D_WALK_CONTINUE;
1434 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1435 " still in use (%d) [unmount of %s %s]\n",
1438 dentry->d_inode->i_ino : 0UL,
1440 dentry->d_lockref.count,
1441 dentry->d_sb->s_type->name,
1442 dentry->d_sb->s_id);
1444 return D_WALK_CONTINUE;
1447 static void do_one_tree(struct dentry *dentry)
1449 shrink_dcache_parent(dentry);
1450 d_walk(dentry, dentry, umount_check, NULL);
1456 * destroy the dentries attached to a superblock on unmounting
1458 void shrink_dcache_for_umount(struct super_block *sb)
1460 struct dentry *dentry;
1462 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1464 dentry = sb->s_root;
1466 do_one_tree(dentry);
1468 while (!hlist_bl_empty(&sb->s_anon)) {
1469 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1470 do_one_tree(dentry);
1474 struct detach_data {
1475 struct select_data select;
1476 struct dentry *mountpoint;
1478 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1480 struct detach_data *data = _data;
1482 if (d_mountpoint(dentry)) {
1483 __dget_dlock(dentry);
1484 data->mountpoint = dentry;
1488 return select_collect(&data->select, dentry);
1491 static void check_and_drop(void *_data)
1493 struct detach_data *data = _data;
1495 if (!data->mountpoint && !data->select.found)
1496 __d_drop(data->select.start);
1500 * d_invalidate - detach submounts, prune dcache, and drop
1501 * @dentry: dentry to invalidate (aka detach, prune and drop)
1505 * The final d_drop is done as an atomic operation relative to
1506 * rename_lock ensuring there are no races with d_set_mounted. This
1507 * ensures there are no unhashed dentries on the path to a mountpoint.
1509 void d_invalidate(struct dentry *dentry)
1512 * If it's already been dropped, return OK.
1514 spin_lock(&dentry->d_lock);
1515 if (d_unhashed(dentry)) {
1516 spin_unlock(&dentry->d_lock);
1519 spin_unlock(&dentry->d_lock);
1521 /* Negative dentries can be dropped without further checks */
1522 if (!dentry->d_inode) {
1528 struct detach_data data;
1530 data.mountpoint = NULL;
1531 INIT_LIST_HEAD(&data.select.dispose);
1532 data.select.start = dentry;
1533 data.select.found = 0;
1535 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1537 if (data.select.found)
1538 shrink_dentry_list(&data.select.dispose);
1540 if (data.mountpoint) {
1541 detach_mounts(data.mountpoint);
1542 dput(data.mountpoint);
1545 if (!data.mountpoint && !data.select.found)
1551 EXPORT_SYMBOL(d_invalidate);
1554 * __d_alloc - allocate a dcache entry
1555 * @sb: filesystem it will belong to
1556 * @name: qstr of the name
1558 * Allocates a dentry. It returns %NULL if there is insufficient memory
1559 * available. On a success the dentry is returned. The name passed in is
1560 * copied and the copy passed in may be reused after this call.
1563 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1565 struct dentry *dentry;
1568 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1573 * We guarantee that the inline name is always NUL-terminated.
1574 * This way the memcpy() done by the name switching in rename
1575 * will still always have a NUL at the end, even if we might
1576 * be overwriting an internal NUL character
1578 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1579 if (name->len > DNAME_INLINE_LEN-1) {
1580 size_t size = offsetof(struct external_name, name[1]);
1581 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1583 kmem_cache_free(dentry_cache, dentry);
1586 atomic_set(&p->u.count, 1);
1588 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1589 kasan_unpoison_shadow(dname,
1590 round_up(name->len + 1, sizeof(unsigned long)));
1592 dname = dentry->d_iname;
1595 dentry->d_name.len = name->len;
1596 dentry->d_name.hash = name->hash;
1597 memcpy(dname, name->name, name->len);
1598 dname[name->len] = 0;
1600 /* Make sure we always see the terminating NUL character */
1602 dentry->d_name.name = dname;
1604 dentry->d_lockref.count = 1;
1605 dentry->d_flags = 0;
1606 spin_lock_init(&dentry->d_lock);
1607 seqcount_init(&dentry->d_seq);
1608 dentry->d_inode = NULL;
1609 dentry->d_parent = dentry;
1611 dentry->d_op = NULL;
1612 dentry->d_fsdata = NULL;
1613 INIT_HLIST_BL_NODE(&dentry->d_hash);
1614 INIT_LIST_HEAD(&dentry->d_lru);
1615 INIT_LIST_HEAD(&dentry->d_subdirs);
1616 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1617 INIT_LIST_HEAD(&dentry->d_child);
1618 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1620 this_cpu_inc(nr_dentry);
1626 * d_alloc - allocate a dcache entry
1627 * @parent: parent of entry to allocate
1628 * @name: qstr of the name
1630 * Allocates a dentry. It returns %NULL if there is insufficient memory
1631 * available. On a success the dentry is returned. The name passed in is
1632 * copied and the copy passed in may be reused after this call.
1634 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1636 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1639 dentry->d_flags |= DCACHE_RCUACCESS;
1640 spin_lock(&parent->d_lock);
1642 * don't need child lock because it is not subject
1643 * to concurrency here
1645 __dget_dlock(parent);
1646 dentry->d_parent = parent;
1647 list_add(&dentry->d_child, &parent->d_subdirs);
1648 spin_unlock(&parent->d_lock);
1652 EXPORT_SYMBOL(d_alloc);
1655 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1656 * @sb: the superblock
1657 * @name: qstr of the name
1659 * For a filesystem that just pins its dentries in memory and never
1660 * performs lookups at all, return an unhashed IS_ROOT dentry.
1662 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1664 return __d_alloc(sb, name);
1666 EXPORT_SYMBOL(d_alloc_pseudo);
1668 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1673 q.len = strlen(name);
1674 q.hash = full_name_hash(q.name, q.len);
1675 return d_alloc(parent, &q);
1677 EXPORT_SYMBOL(d_alloc_name);
1679 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1681 WARN_ON_ONCE(dentry->d_op);
1682 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1684 DCACHE_OP_REVALIDATE |
1685 DCACHE_OP_WEAK_REVALIDATE |
1687 DCACHE_OP_SELECT_INODE |
1693 dentry->d_flags |= DCACHE_OP_HASH;
1695 dentry->d_flags |= DCACHE_OP_COMPARE;
1696 if (op->d_revalidate)
1697 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1698 if (op->d_weak_revalidate)
1699 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1701 dentry->d_flags |= DCACHE_OP_DELETE;
1703 dentry->d_flags |= DCACHE_OP_PRUNE;
1704 if (op->d_select_inode)
1705 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
1707 dentry->d_flags |= DCACHE_OP_REAL;
1710 EXPORT_SYMBOL(d_set_d_op);
1714 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1715 * @dentry - The dentry to mark
1717 * Mark a dentry as falling through to the lower layer (as set with
1718 * d_pin_lower()). This flag may be recorded on the medium.
1720 void d_set_fallthru(struct dentry *dentry)
1722 spin_lock(&dentry->d_lock);
1723 dentry->d_flags |= DCACHE_FALLTHRU;
1724 spin_unlock(&dentry->d_lock);
1726 EXPORT_SYMBOL(d_set_fallthru);
1728 static unsigned d_flags_for_inode(struct inode *inode)
1730 unsigned add_flags = DCACHE_REGULAR_TYPE;
1733 return DCACHE_MISS_TYPE;
1735 if (S_ISDIR(inode->i_mode)) {
1736 add_flags = DCACHE_DIRECTORY_TYPE;
1737 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1738 if (unlikely(!inode->i_op->lookup))
1739 add_flags = DCACHE_AUTODIR_TYPE;
1741 inode->i_opflags |= IOP_LOOKUP;
1743 goto type_determined;
1746 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1747 if (unlikely(inode->i_op->follow_link)) {
1748 add_flags = DCACHE_SYMLINK_TYPE;
1749 goto type_determined;
1751 inode->i_opflags |= IOP_NOFOLLOW;
1754 if (unlikely(!S_ISREG(inode->i_mode)))
1755 add_flags = DCACHE_SPECIAL_TYPE;
1758 if (unlikely(IS_AUTOMOUNT(inode)))
1759 add_flags |= DCACHE_NEED_AUTOMOUNT;
1763 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1765 unsigned add_flags = d_flags_for_inode(inode);
1767 spin_lock(&dentry->d_lock);
1769 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1770 raw_write_seqcount_begin(&dentry->d_seq);
1771 __d_set_inode_and_type(dentry, inode, add_flags);
1772 raw_write_seqcount_end(&dentry->d_seq);
1773 spin_unlock(&dentry->d_lock);
1774 fsnotify_d_instantiate(dentry, inode);
1778 * d_instantiate - fill in inode information for a dentry
1779 * @entry: dentry to complete
1780 * @inode: inode to attach to this dentry
1782 * Fill in inode information in the entry.
1784 * This turns negative dentries into productive full members
1787 * NOTE! This assumes that the inode count has been incremented
1788 * (or otherwise set) by the caller to indicate that it is now
1789 * in use by the dcache.
1792 void d_instantiate(struct dentry *entry, struct inode * inode)
1794 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1796 spin_lock(&inode->i_lock);
1797 __d_instantiate(entry, inode);
1799 spin_unlock(&inode->i_lock);
1800 security_d_instantiate(entry, inode);
1802 EXPORT_SYMBOL(d_instantiate);
1805 * d_instantiate_unique - instantiate a non-aliased dentry
1806 * @entry: dentry to instantiate
1807 * @inode: inode to attach to this dentry
1809 * Fill in inode information in the entry. On success, it returns NULL.
1810 * If an unhashed alias of "entry" already exists, then we return the
1811 * aliased dentry instead and drop one reference to inode.
1813 * Note that in order to avoid conflicts with rename() etc, the caller
1814 * had better be holding the parent directory semaphore.
1816 * This also assumes that the inode count has been incremented
1817 * (or otherwise set) by the caller to indicate that it is now
1818 * in use by the dcache.
1820 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1821 struct inode *inode)
1823 struct dentry *alias;
1824 int len = entry->d_name.len;
1825 const char *name = entry->d_name.name;
1826 unsigned int hash = entry->d_name.hash;
1829 __d_instantiate(entry, NULL);
1833 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1835 * Don't need alias->d_lock here, because aliases with
1836 * d_parent == entry->d_parent are not subject to name or
1837 * parent changes, because the parent inode i_mutex is held.
1839 if (alias->d_name.hash != hash)
1841 if (alias->d_parent != entry->d_parent)
1843 if (alias->d_name.len != len)
1845 if (dentry_cmp(alias, name, len))
1851 __d_instantiate(entry, inode);
1855 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1857 struct dentry *result;
1859 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1862 spin_lock(&inode->i_lock);
1863 result = __d_instantiate_unique(entry, inode);
1865 spin_unlock(&inode->i_lock);
1868 security_d_instantiate(entry, inode);
1872 BUG_ON(!d_unhashed(result));
1877 EXPORT_SYMBOL(d_instantiate_unique);
1880 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1881 * @entry: dentry to complete
1882 * @inode: inode to attach to this dentry
1884 * Fill in inode information in the entry. If a directory alias is found, then
1885 * return an error (and drop inode). Together with d_materialise_unique() this
1886 * guarantees that a directory inode may never have more than one alias.
1888 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1890 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1892 spin_lock(&inode->i_lock);
1893 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1894 spin_unlock(&inode->i_lock);
1898 __d_instantiate(entry, inode);
1899 spin_unlock(&inode->i_lock);
1900 security_d_instantiate(entry, inode);
1904 EXPORT_SYMBOL(d_instantiate_no_diralias);
1906 struct dentry *d_make_root(struct inode *root_inode)
1908 struct dentry *res = NULL;
1911 static const struct qstr name = QSTR_INIT("/", 1);
1913 res = __d_alloc(root_inode->i_sb, &name);
1915 d_instantiate(res, root_inode);
1921 EXPORT_SYMBOL(d_make_root);
1923 static struct dentry * __d_find_any_alias(struct inode *inode)
1925 struct dentry *alias;
1927 if (hlist_empty(&inode->i_dentry))
1929 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1935 * d_find_any_alias - find any alias for a given inode
1936 * @inode: inode to find an alias for
1938 * If any aliases exist for the given inode, take and return a
1939 * reference for one of them. If no aliases exist, return %NULL.
1941 struct dentry *d_find_any_alias(struct inode *inode)
1945 spin_lock(&inode->i_lock);
1946 de = __d_find_any_alias(inode);
1947 spin_unlock(&inode->i_lock);
1950 EXPORT_SYMBOL(d_find_any_alias);
1952 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1954 static const struct qstr anonstring = QSTR_INIT("/", 1);
1960 return ERR_PTR(-ESTALE);
1962 return ERR_CAST(inode);
1964 res = d_find_any_alias(inode);
1968 tmp = __d_alloc(inode->i_sb, &anonstring);
1970 res = ERR_PTR(-ENOMEM);
1974 spin_lock(&inode->i_lock);
1975 res = __d_find_any_alias(inode);
1977 spin_unlock(&inode->i_lock);
1982 /* attach a disconnected dentry */
1983 add_flags = d_flags_for_inode(inode);
1986 add_flags |= DCACHE_DISCONNECTED;
1988 spin_lock(&tmp->d_lock);
1989 __d_set_inode_and_type(tmp, inode, add_flags);
1990 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1991 hlist_bl_lock(&tmp->d_sb->s_anon);
1992 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1993 hlist_bl_unlock(&tmp->d_sb->s_anon);
1994 spin_unlock(&tmp->d_lock);
1995 spin_unlock(&inode->i_lock);
1996 security_d_instantiate(tmp, inode);
2001 if (res && !IS_ERR(res))
2002 security_d_instantiate(res, inode);
2008 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2009 * @inode: inode to allocate the dentry for
2011 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2012 * similar open by handle operations. The returned dentry may be anonymous,
2013 * or may have a full name (if the inode was already in the cache).
2015 * When called on a directory inode, we must ensure that the inode only ever
2016 * has one dentry. If a dentry is found, that is returned instead of
2017 * allocating a new one.
2019 * On successful return, the reference to the inode has been transferred
2020 * to the dentry. In case of an error the reference on the inode is released.
2021 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2022 * be passed in and the error will be propagated to the return value,
2023 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2025 struct dentry *d_obtain_alias(struct inode *inode)
2027 return __d_obtain_alias(inode, 1);
2029 EXPORT_SYMBOL(d_obtain_alias);
2032 * d_obtain_root - find or allocate a dentry for a given inode
2033 * @inode: inode to allocate the dentry for
2035 * Obtain an IS_ROOT dentry for the root of a filesystem.
2037 * We must ensure that directory inodes only ever have one dentry. If a
2038 * dentry is found, that is returned instead of allocating a new one.
2040 * On successful return, the reference to the inode has been transferred
2041 * to the dentry. In case of an error the reference on the inode is
2042 * released. A %NULL or IS_ERR inode may be passed in and will be the
2043 * error will be propagate to the return value, with a %NULL @inode
2044 * replaced by ERR_PTR(-ESTALE).
2046 struct dentry *d_obtain_root(struct inode *inode)
2048 return __d_obtain_alias(inode, 0);
2050 EXPORT_SYMBOL(d_obtain_root);
2053 * d_add_ci - lookup or allocate new dentry with case-exact name
2054 * @inode: the inode case-insensitive lookup has found
2055 * @dentry: the negative dentry that was passed to the parent's lookup func
2056 * @name: the case-exact name to be associated with the returned dentry
2058 * This is to avoid filling the dcache with case-insensitive names to the
2059 * same inode, only the actual correct case is stored in the dcache for
2060 * case-insensitive filesystems.
2062 * For a case-insensitive lookup match and if the the case-exact dentry
2063 * already exists in in the dcache, use it and return it.
2065 * If no entry exists with the exact case name, allocate new dentry with
2066 * the exact case, and return the spliced entry.
2068 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2071 struct dentry *found;
2075 * First check if a dentry matching the name already exists,
2076 * if not go ahead and create it now.
2078 found = d_hash_and_lookup(dentry->d_parent, name);
2080 new = d_alloc(dentry->d_parent, name);
2082 found = ERR_PTR(-ENOMEM);
2084 found = d_splice_alias(inode, new);
2095 EXPORT_SYMBOL(d_add_ci);
2098 * Do the slow-case of the dentry name compare.
2100 * Unlike the dentry_cmp() function, we need to atomically
2101 * load the name and length information, so that the
2102 * filesystem can rely on them, and can use the 'name' and
2103 * 'len' information without worrying about walking off the
2104 * end of memory etc.
2106 * Thus the read_seqcount_retry() and the "duplicate" info
2107 * in arguments (the low-level filesystem should not look
2108 * at the dentry inode or name contents directly, since
2109 * rename can change them while we're in RCU mode).
2111 enum slow_d_compare {
2117 static noinline enum slow_d_compare slow_dentry_cmp(
2118 const struct dentry *parent,
2119 struct dentry *dentry,
2121 const struct qstr *name)
2123 int tlen = dentry->d_name.len;
2124 const char *tname = dentry->d_name.name;
2126 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2128 return D_COMP_SEQRETRY;
2130 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2131 return D_COMP_NOMATCH;
2136 * __d_lookup_rcu - search for a dentry (racy, store-free)
2137 * @parent: parent dentry
2138 * @name: qstr of name we wish to find
2139 * @seqp: returns d_seq value at the point where the dentry was found
2140 * Returns: dentry, or NULL
2142 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2143 * resolution (store-free path walking) design described in
2144 * Documentation/filesystems/path-lookup.txt.
2146 * This is not to be used outside core vfs.
2148 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2149 * held, and rcu_read_lock held. The returned dentry must not be stored into
2150 * without taking d_lock and checking d_seq sequence count against @seq
2153 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2156 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2157 * the returned dentry, so long as its parent's seqlock is checked after the
2158 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2159 * is formed, giving integrity down the path walk.
2161 * NOTE! The caller *has* to check the resulting dentry against the sequence
2162 * number we've returned before using any of the resulting dentry state!
2164 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2165 const struct qstr *name,
2168 u64 hashlen = name->hash_len;
2169 const unsigned char *str = name->name;
2170 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2171 struct hlist_bl_node *node;
2172 struct dentry *dentry;
2175 * Note: There is significant duplication with __d_lookup_rcu which is
2176 * required to prevent single threaded performance regressions
2177 * especially on architectures where smp_rmb (in seqcounts) are costly.
2178 * Keep the two functions in sync.
2182 * The hash list is protected using RCU.
2184 * Carefully use d_seq when comparing a candidate dentry, to avoid
2185 * races with d_move().
2187 * It is possible that concurrent renames can mess up our list
2188 * walk here and result in missing our dentry, resulting in the
2189 * false-negative result. d_lookup() protects against concurrent
2190 * renames using rename_lock seqlock.
2192 * See Documentation/filesystems/path-lookup.txt for more details.
2194 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2199 * The dentry sequence count protects us from concurrent
2200 * renames, and thus protects parent and name fields.
2202 * The caller must perform a seqcount check in order
2203 * to do anything useful with the returned dentry.
2205 * NOTE! We do a "raw" seqcount_begin here. That means that
2206 * we don't wait for the sequence count to stabilize if it
2207 * is in the middle of a sequence change. If we do the slow
2208 * dentry compare, we will do seqretries until it is stable,
2209 * and if we end up with a successful lookup, we actually
2210 * want to exit RCU lookup anyway.
2212 seq = raw_seqcount_begin(&dentry->d_seq);
2213 if (dentry->d_parent != parent)
2215 if (d_unhashed(dentry))
2218 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2219 if (dentry->d_name.hash != hashlen_hash(hashlen))
2222 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2225 case D_COMP_NOMATCH:
2232 if (dentry->d_name.hash_len != hashlen)
2235 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2242 * d_lookup - search for a dentry
2243 * @parent: parent dentry
2244 * @name: qstr of name we wish to find
2245 * Returns: dentry, or NULL
2247 * d_lookup searches the children of the parent dentry for the name in
2248 * question. If the dentry is found its reference count is incremented and the
2249 * dentry is returned. The caller must use dput to free the entry when it has
2250 * finished using it. %NULL is returned if the dentry does not exist.
2252 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2254 struct dentry *dentry;
2258 seq = read_seqbegin(&rename_lock);
2259 dentry = __d_lookup(parent, name);
2262 } while (read_seqretry(&rename_lock, seq));
2265 EXPORT_SYMBOL(d_lookup);
2268 * __d_lookup - search for a dentry (racy)
2269 * @parent: parent dentry
2270 * @name: qstr of name we wish to find
2271 * Returns: dentry, or NULL
2273 * __d_lookup is like d_lookup, however it may (rarely) return a
2274 * false-negative result due to unrelated rename activity.
2276 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2277 * however it must be used carefully, eg. with a following d_lookup in
2278 * the case of failure.
2280 * __d_lookup callers must be commented.
2282 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2284 unsigned int len = name->len;
2285 unsigned int hash = name->hash;
2286 const unsigned char *str = name->name;
2287 struct hlist_bl_head *b = d_hash(parent, hash);
2288 struct hlist_bl_node *node;
2289 struct dentry *found = NULL;
2290 struct dentry *dentry;
2293 * Note: There is significant duplication with __d_lookup_rcu which is
2294 * required to prevent single threaded performance regressions
2295 * especially on architectures where smp_rmb (in seqcounts) are costly.
2296 * Keep the two functions in sync.
2300 * The hash list is protected using RCU.
2302 * Take d_lock when comparing a candidate dentry, to avoid races
2305 * It is possible that concurrent renames can mess up our list
2306 * walk here and result in missing our dentry, resulting in the
2307 * false-negative result. d_lookup() protects against concurrent
2308 * renames using rename_lock seqlock.
2310 * See Documentation/filesystems/path-lookup.txt for more details.
2314 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2316 if (dentry->d_name.hash != hash)
2319 spin_lock(&dentry->d_lock);
2320 if (dentry->d_parent != parent)
2322 if (d_unhashed(dentry))
2326 * It is safe to compare names since d_move() cannot
2327 * change the qstr (protected by d_lock).
2329 if (parent->d_flags & DCACHE_OP_COMPARE) {
2330 int tlen = dentry->d_name.len;
2331 const char *tname = dentry->d_name.name;
2332 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2335 if (dentry->d_name.len != len)
2337 if (dentry_cmp(dentry, str, len))
2341 dentry->d_lockref.count++;
2343 spin_unlock(&dentry->d_lock);
2346 spin_unlock(&dentry->d_lock);
2354 * d_hash_and_lookup - hash the qstr then search for a dentry
2355 * @dir: Directory to search in
2356 * @name: qstr of name we wish to find
2358 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2360 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2363 * Check for a fs-specific hash function. Note that we must
2364 * calculate the standard hash first, as the d_op->d_hash()
2365 * routine may choose to leave the hash value unchanged.
2367 name->hash = full_name_hash(name->name, name->len);
2368 if (dir->d_flags & DCACHE_OP_HASH) {
2369 int err = dir->d_op->d_hash(dir, name);
2370 if (unlikely(err < 0))
2371 return ERR_PTR(err);
2373 return d_lookup(dir, name);
2375 EXPORT_SYMBOL(d_hash_and_lookup);
2378 * When a file is deleted, we have two options:
2379 * - turn this dentry into a negative dentry
2380 * - unhash this dentry and free it.
2382 * Usually, we want to just turn this into
2383 * a negative dentry, but if anybody else is
2384 * currently using the dentry or the inode
2385 * we can't do that and we fall back on removing
2386 * it from the hash queues and waiting for
2387 * it to be deleted later when it has no users
2391 * d_delete - delete a dentry
2392 * @dentry: The dentry to delete
2394 * Turn the dentry into a negative dentry if possible, otherwise
2395 * remove it from the hash queues so it can be deleted later
2398 void d_delete(struct dentry * dentry)
2400 struct inode *inode;
2403 * Are we the only user?
2406 spin_lock(&dentry->d_lock);
2407 inode = dentry->d_inode;
2408 isdir = S_ISDIR(inode->i_mode);
2409 if (dentry->d_lockref.count == 1) {
2410 if (!spin_trylock(&inode->i_lock)) {
2411 spin_unlock(&dentry->d_lock);
2415 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2416 dentry_unlink_inode(dentry);
2417 fsnotify_nameremove(dentry, isdir);
2421 if (!d_unhashed(dentry))
2424 spin_unlock(&dentry->d_lock);
2426 fsnotify_nameremove(dentry, isdir);
2428 EXPORT_SYMBOL(d_delete);
2430 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2432 BUG_ON(!d_unhashed(entry));
2434 hlist_bl_add_head_rcu(&entry->d_hash, b);
2438 static void _d_rehash(struct dentry * entry)
2440 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2444 * d_rehash - add an entry back to the hash
2445 * @entry: dentry to add to the hash
2447 * Adds a dentry to the hash according to its name.
2450 void d_rehash(struct dentry * entry)
2452 spin_lock(&entry->d_lock);
2454 spin_unlock(&entry->d_lock);
2456 EXPORT_SYMBOL(d_rehash);
2459 * dentry_update_name_case - update case insensitive dentry with a new name
2460 * @dentry: dentry to be updated
2463 * Update a case insensitive dentry with new case of name.
2465 * dentry must have been returned by d_lookup with name @name. Old and new
2466 * name lengths must match (ie. no d_compare which allows mismatched name
2469 * Parent inode i_mutex must be held over d_lookup and into this call (to
2470 * keep renames and concurrent inserts, and readdir(2) away).
2472 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2474 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2475 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2477 spin_lock(&dentry->d_lock);
2478 write_seqcount_begin(&dentry->d_seq);
2479 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2480 write_seqcount_end(&dentry->d_seq);
2481 spin_unlock(&dentry->d_lock);
2483 EXPORT_SYMBOL(dentry_update_name_case);
2485 static void swap_names(struct dentry *dentry, struct dentry *target)
2487 if (unlikely(dname_external(target))) {
2488 if (unlikely(dname_external(dentry))) {
2490 * Both external: swap the pointers
2492 swap(target->d_name.name, dentry->d_name.name);
2495 * dentry:internal, target:external. Steal target's
2496 * storage and make target internal.
2498 memcpy(target->d_iname, dentry->d_name.name,
2499 dentry->d_name.len + 1);
2500 dentry->d_name.name = target->d_name.name;
2501 target->d_name.name = target->d_iname;
2504 if (unlikely(dname_external(dentry))) {
2506 * dentry:external, target:internal. Give dentry's
2507 * storage to target and make dentry internal
2509 memcpy(dentry->d_iname, target->d_name.name,
2510 target->d_name.len + 1);
2511 target->d_name.name = dentry->d_name.name;
2512 dentry->d_name.name = dentry->d_iname;
2515 * Both are internal.
2518 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2519 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2520 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2521 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2522 swap(((long *) &dentry->d_iname)[i],
2523 ((long *) &target->d_iname)[i]);
2527 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2530 static void copy_name(struct dentry *dentry, struct dentry *target)
2532 struct external_name *old_name = NULL;
2533 if (unlikely(dname_external(dentry)))
2534 old_name = external_name(dentry);
2535 if (unlikely(dname_external(target))) {
2536 atomic_inc(&external_name(target)->u.count);
2537 dentry->d_name = target->d_name;
2539 memcpy(dentry->d_iname, target->d_name.name,
2540 target->d_name.len + 1);
2541 dentry->d_name.name = dentry->d_iname;
2542 dentry->d_name.hash_len = target->d_name.hash_len;
2544 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2545 kfree_rcu(old_name, u.head);
2548 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2551 * XXXX: do we really need to take target->d_lock?
2553 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2554 spin_lock(&target->d_parent->d_lock);
2556 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2557 spin_lock(&dentry->d_parent->d_lock);
2558 spin_lock_nested(&target->d_parent->d_lock,
2559 DENTRY_D_LOCK_NESTED);
2561 spin_lock(&target->d_parent->d_lock);
2562 spin_lock_nested(&dentry->d_parent->d_lock,
2563 DENTRY_D_LOCK_NESTED);
2566 if (target < dentry) {
2567 spin_lock_nested(&target->d_lock, 2);
2568 spin_lock_nested(&dentry->d_lock, 3);
2570 spin_lock_nested(&dentry->d_lock, 2);
2571 spin_lock_nested(&target->d_lock, 3);
2575 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2577 if (target->d_parent != dentry->d_parent)
2578 spin_unlock(&dentry->d_parent->d_lock);
2579 if (target->d_parent != target)
2580 spin_unlock(&target->d_parent->d_lock);
2581 spin_unlock(&target->d_lock);
2582 spin_unlock(&dentry->d_lock);
2586 * When switching names, the actual string doesn't strictly have to
2587 * be preserved in the target - because we're dropping the target
2588 * anyway. As such, we can just do a simple memcpy() to copy over
2589 * the new name before we switch, unless we are going to rehash
2590 * it. Note that if we *do* unhash the target, we are not allowed
2591 * to rehash it without giving it a new name/hash key - whether
2592 * we swap or overwrite the names here, resulting name won't match
2593 * the reality in filesystem; it's only there for d_path() purposes.
2594 * Note that all of this is happening under rename_lock, so the
2595 * any hash lookup seeing it in the middle of manipulations will
2596 * be discarded anyway. So we do not care what happens to the hash
2600 * __d_move - move a dentry
2601 * @dentry: entry to move
2602 * @target: new dentry
2603 * @exchange: exchange the two dentries
2605 * Update the dcache to reflect the move of a file name. Negative
2606 * dcache entries should not be moved in this way. Caller must hold
2607 * rename_lock, the i_mutex of the source and target directories,
2608 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2610 static void __d_move(struct dentry *dentry, struct dentry *target,
2613 if (!dentry->d_inode)
2614 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2616 BUG_ON(d_ancestor(dentry, target));
2617 BUG_ON(d_ancestor(target, dentry));
2619 dentry_lock_for_move(dentry, target);
2621 write_seqcount_begin(&dentry->d_seq);
2622 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2624 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2627 * Move the dentry to the target hash queue. Don't bother checking
2628 * for the same hash queue because of how unlikely it is.
2631 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2634 * Unhash the target (d_delete() is not usable here). If exchanging
2635 * the two dentries, then rehash onto the other's hash queue.
2640 d_hash(dentry->d_parent, dentry->d_name.hash));
2643 /* Switch the names.. */
2645 swap_names(dentry, target);
2647 copy_name(dentry, target);
2649 /* ... and switch them in the tree */
2650 if (IS_ROOT(dentry)) {
2651 /* splicing a tree */
2652 dentry->d_flags |= DCACHE_RCUACCESS;
2653 dentry->d_parent = target->d_parent;
2654 target->d_parent = target;
2655 list_del_init(&target->d_child);
2656 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2658 /* swapping two dentries */
2659 swap(dentry->d_parent, target->d_parent);
2660 list_move(&target->d_child, &target->d_parent->d_subdirs);
2661 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2663 fsnotify_d_move(target);
2664 fsnotify_d_move(dentry);
2667 write_seqcount_end(&target->d_seq);
2668 write_seqcount_end(&dentry->d_seq);
2670 dentry_unlock_for_move(dentry, target);
2674 * d_move - move a dentry
2675 * @dentry: entry to move
2676 * @target: new dentry
2678 * Update the dcache to reflect the move of a file name. Negative
2679 * dcache entries should not be moved in this way. See the locking
2680 * requirements for __d_move.
2682 void d_move(struct dentry *dentry, struct dentry *target)
2684 write_seqlock(&rename_lock);
2685 __d_move(dentry, target, false);
2686 write_sequnlock(&rename_lock);
2688 EXPORT_SYMBOL(d_move);
2691 * d_exchange - exchange two dentries
2692 * @dentry1: first dentry
2693 * @dentry2: second dentry
2695 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2697 write_seqlock(&rename_lock);
2699 WARN_ON(!dentry1->d_inode);
2700 WARN_ON(!dentry2->d_inode);
2701 WARN_ON(IS_ROOT(dentry1));
2702 WARN_ON(IS_ROOT(dentry2));
2704 __d_move(dentry1, dentry2, true);
2706 write_sequnlock(&rename_lock);
2710 * d_ancestor - search for an ancestor
2711 * @p1: ancestor dentry
2714 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2715 * an ancestor of p2, else NULL.
2717 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2721 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2722 if (p->d_parent == p1)
2729 * This helper attempts to cope with remotely renamed directories
2731 * It assumes that the caller is already holding
2732 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2734 * Note: If ever the locking in lock_rename() changes, then please
2735 * remember to update this too...
2737 static int __d_unalias(struct inode *inode,
2738 struct dentry *dentry, struct dentry *alias)
2740 struct mutex *m1 = NULL, *m2 = NULL;
2743 /* If alias and dentry share a parent, then no extra locks required */
2744 if (alias->d_parent == dentry->d_parent)
2747 /* See lock_rename() */
2748 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2750 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2751 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2753 m2 = &alias->d_parent->d_inode->i_mutex;
2755 __d_move(alias, dentry, false);
2766 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2767 * @inode: the inode which may have a disconnected dentry
2768 * @dentry: a negative dentry which we want to point to the inode.
2770 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2771 * place of the given dentry and return it, else simply d_add the inode
2772 * to the dentry and return NULL.
2774 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2775 * we should error out: directories can't have multiple aliases.
2777 * This is needed in the lookup routine of any filesystem that is exportable
2778 * (via knfsd) so that we can build dcache paths to directories effectively.
2780 * If a dentry was found and moved, then it is returned. Otherwise NULL
2781 * is returned. This matches the expected return value of ->lookup.
2783 * Cluster filesystems may call this function with a negative, hashed dentry.
2784 * In that case, we know that the inode will be a regular file, and also this
2785 * will only occur during atomic_open. So we need to check for the dentry
2786 * being already hashed only in the final case.
2788 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2791 return ERR_CAST(inode);
2793 BUG_ON(!d_unhashed(dentry));
2796 __d_instantiate(dentry, NULL);
2799 spin_lock(&inode->i_lock);
2800 if (S_ISDIR(inode->i_mode)) {
2801 struct dentry *new = __d_find_any_alias(inode);
2802 if (unlikely(new)) {
2803 /* The reference to new ensures it remains an alias */
2804 spin_unlock(&inode->i_lock);
2805 write_seqlock(&rename_lock);
2806 if (unlikely(d_ancestor(new, dentry))) {
2807 write_sequnlock(&rename_lock);
2809 new = ERR_PTR(-ELOOP);
2810 pr_warn_ratelimited(
2811 "VFS: Lookup of '%s' in %s %s"
2812 " would have caused loop\n",
2813 dentry->d_name.name,
2814 inode->i_sb->s_type->name,
2816 } else if (!IS_ROOT(new)) {
2817 int err = __d_unalias(inode, dentry, new);
2818 write_sequnlock(&rename_lock);
2824 __d_move(new, dentry, false);
2825 write_sequnlock(&rename_lock);
2826 security_d_instantiate(new, inode);
2832 /* already taking inode->i_lock, so d_add() by hand */
2833 __d_instantiate(dentry, inode);
2834 spin_unlock(&inode->i_lock);
2836 security_d_instantiate(dentry, inode);
2840 EXPORT_SYMBOL(d_splice_alias);
2842 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2846 return -ENAMETOOLONG;
2848 memcpy(*buffer, str, namelen);
2853 * prepend_name - prepend a pathname in front of current buffer pointer
2854 * @buffer: buffer pointer
2855 * @buflen: allocated length of the buffer
2856 * @name: name string and length qstr structure
2858 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2859 * make sure that either the old or the new name pointer and length are
2860 * fetched. However, there may be mismatch between length and pointer.
2861 * The length cannot be trusted, we need to copy it byte-by-byte until
2862 * the length is reached or a null byte is found. It also prepends "/" at
2863 * the beginning of the name. The sequence number check at the caller will
2864 * retry it again when a d_move() does happen. So any garbage in the buffer
2865 * due to mismatched pointer and length will be discarded.
2867 * Data dependency barrier is needed to make sure that we see that terminating
2868 * NUL. Alpha strikes again, film at 11...
2870 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2872 const char *dname = ACCESS_ONCE(name->name);
2873 u32 dlen = ACCESS_ONCE(name->len);
2876 smp_read_barrier_depends();
2878 *buflen -= dlen + 1;
2880 return -ENAMETOOLONG;
2881 p = *buffer -= dlen + 1;
2893 * prepend_path - Prepend path string to a buffer
2894 * @path: the dentry/vfsmount to report
2895 * @root: root vfsmnt/dentry
2896 * @buffer: pointer to the end of the buffer
2897 * @buflen: pointer to buffer length
2899 * The function will first try to write out the pathname without taking any
2900 * lock other than the RCU read lock to make sure that dentries won't go away.
2901 * It only checks the sequence number of the global rename_lock as any change
2902 * in the dentry's d_seq will be preceded by changes in the rename_lock
2903 * sequence number. If the sequence number had been changed, it will restart
2904 * the whole pathname back-tracing sequence again by taking the rename_lock.
2905 * In this case, there is no need to take the RCU read lock as the recursive
2906 * parent pointer references will keep the dentry chain alive as long as no
2907 * rename operation is performed.
2909 static int prepend_path(const struct path *path,
2910 const struct path *root,
2911 char **buffer, int *buflen)
2913 struct dentry *dentry;
2914 struct vfsmount *vfsmnt;
2917 unsigned seq, m_seq = 0;
2923 read_seqbegin_or_lock(&mount_lock, &m_seq);
2930 dentry = path->dentry;
2932 mnt = real_mount(vfsmnt);
2933 read_seqbegin_or_lock(&rename_lock, &seq);
2934 while (dentry != root->dentry || vfsmnt != root->mnt) {
2935 struct dentry * parent;
2937 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2938 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2940 if (dentry != vfsmnt->mnt_root) {
2947 if (mnt != parent) {
2948 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2954 error = is_mounted(vfsmnt) ? 1 : 2;
2957 parent = dentry->d_parent;
2959 error = prepend_name(&bptr, &blen, &dentry->d_name);
2967 if (need_seqretry(&rename_lock, seq)) {
2971 done_seqretry(&rename_lock, seq);
2975 if (need_seqretry(&mount_lock, m_seq)) {
2979 done_seqretry(&mount_lock, m_seq);
2981 if (error >= 0 && bptr == *buffer) {
2983 error = -ENAMETOOLONG;
2993 * __d_path - return the path of a dentry
2994 * @path: the dentry/vfsmount to report
2995 * @root: root vfsmnt/dentry
2996 * @buf: buffer to return value in
2997 * @buflen: buffer length
2999 * Convert a dentry into an ASCII path name.
3001 * Returns a pointer into the buffer or an error code if the
3002 * path was too long.
3004 * "buflen" should be positive.
3006 * If the path is not reachable from the supplied root, return %NULL.
3008 char *__d_path(const struct path *path,
3009 const struct path *root,
3010 char *buf, int buflen)
3012 char *res = buf + buflen;
3015 prepend(&res, &buflen, "\0", 1);
3016 error = prepend_path(path, root, &res, &buflen);
3019 return ERR_PTR(error);
3025 char *d_absolute_path(const struct path *path,
3026 char *buf, int buflen)
3028 struct path root = {};
3029 char *res = buf + buflen;
3032 prepend(&res, &buflen, "\0", 1);
3033 error = prepend_path(path, &root, &res, &buflen);
3038 return ERR_PTR(error);
3043 * same as __d_path but appends "(deleted)" for unlinked files.
3045 static int path_with_deleted(const struct path *path,
3046 const struct path *root,
3047 char **buf, int *buflen)
3049 prepend(buf, buflen, "\0", 1);
3050 if (d_unlinked(path->dentry)) {
3051 int error = prepend(buf, buflen, " (deleted)", 10);
3056 return prepend_path(path, root, buf, buflen);
3059 static int prepend_unreachable(char **buffer, int *buflen)
3061 return prepend(buffer, buflen, "(unreachable)", 13);
3064 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3069 seq = read_seqcount_begin(&fs->seq);
3071 } while (read_seqcount_retry(&fs->seq, seq));
3075 * d_path - return the path of a dentry
3076 * @path: path to report
3077 * @buf: buffer to return value in
3078 * @buflen: buffer length
3080 * Convert a dentry into an ASCII path name. If the entry has been deleted
3081 * the string " (deleted)" is appended. Note that this is ambiguous.
3083 * Returns a pointer into the buffer or an error code if the path was
3084 * too long. Note: Callers should use the returned pointer, not the passed
3085 * in buffer, to use the name! The implementation often starts at an offset
3086 * into the buffer, and may leave 0 bytes at the start.
3088 * "buflen" should be positive.
3090 char *d_path(const struct path *path, char *buf, int buflen)
3092 char *res = buf + buflen;
3097 * We have various synthetic filesystems that never get mounted. On
3098 * these filesystems dentries are never used for lookup purposes, and
3099 * thus don't need to be hashed. They also don't need a name until a
3100 * user wants to identify the object in /proc/pid/fd/. The little hack
3101 * below allows us to generate a name for these objects on demand:
3103 * Some pseudo inodes are mountable. When they are mounted
3104 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3105 * and instead have d_path return the mounted path.
3107 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3108 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3109 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3112 get_fs_root_rcu(current->fs, &root);
3113 error = path_with_deleted(path, &root, &res, &buflen);
3117 res = ERR_PTR(error);
3120 EXPORT_SYMBOL(d_path);
3123 * Helper function for dentry_operations.d_dname() members
3125 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3126 const char *fmt, ...)
3132 va_start(args, fmt);
3133 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3136 if (sz > sizeof(temp) || sz > buflen)
3137 return ERR_PTR(-ENAMETOOLONG);
3139 buffer += buflen - sz;
3140 return memcpy(buffer, temp, sz);
3143 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3145 char *end = buffer + buflen;
3146 /* these dentries are never renamed, so d_lock is not needed */
3147 if (prepend(&end, &buflen, " (deleted)", 11) ||
3148 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3149 prepend(&end, &buflen, "/", 1))
3150 end = ERR_PTR(-ENAMETOOLONG);
3153 EXPORT_SYMBOL(simple_dname);
3156 * Write full pathname from the root of the filesystem into the buffer.
3158 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3160 struct dentry *dentry;
3173 prepend(&end, &len, "\0", 1);
3177 read_seqbegin_or_lock(&rename_lock, &seq);
3178 while (!IS_ROOT(dentry)) {
3179 struct dentry *parent = dentry->d_parent;
3182 error = prepend_name(&end, &len, &dentry->d_name);
3191 if (need_seqretry(&rename_lock, seq)) {
3195 done_seqretry(&rename_lock, seq);
3200 return ERR_PTR(-ENAMETOOLONG);
3203 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3205 return __dentry_path(dentry, buf, buflen);
3207 EXPORT_SYMBOL(dentry_path_raw);
3209 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3214 if (d_unlinked(dentry)) {
3216 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3220 retval = __dentry_path(dentry, buf, buflen);
3221 if (!IS_ERR(retval) && p)
3222 *p = '/'; /* restore '/' overriden with '\0' */
3225 return ERR_PTR(-ENAMETOOLONG);
3228 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3234 seq = read_seqcount_begin(&fs->seq);
3237 } while (read_seqcount_retry(&fs->seq, seq));
3241 * NOTE! The user-level library version returns a
3242 * character pointer. The kernel system call just
3243 * returns the length of the buffer filled (which
3244 * includes the ending '\0' character), or a negative
3245 * error value. So libc would do something like
3247 * char *getcwd(char * buf, size_t size)
3251 * retval = sys_getcwd(buf, size);
3258 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3261 struct path pwd, root;
3262 char *page = __getname();
3268 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3271 if (!d_unlinked(pwd.dentry)) {
3273 char *cwd = page + PATH_MAX;
3274 int buflen = PATH_MAX;
3276 prepend(&cwd, &buflen, "\0", 1);
3277 error = prepend_path(&pwd, &root, &cwd, &buflen);
3283 /* Unreachable from current root */
3285 error = prepend_unreachable(&cwd, &buflen);
3291 len = PATH_MAX + page - cwd;
3294 if (copy_to_user(buf, cwd, len))
3307 * Test whether new_dentry is a subdirectory of old_dentry.
3309 * Trivially implemented using the dcache structure
3313 * is_subdir - is new dentry a subdirectory of old_dentry
3314 * @new_dentry: new dentry
3315 * @old_dentry: old dentry
3317 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3318 * Returns 0 otherwise.
3319 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3322 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3327 if (new_dentry == old_dentry)
3331 /* for restarting inner loop in case of seq retry */
3332 seq = read_seqbegin(&rename_lock);
3334 * Need rcu_readlock to protect against the d_parent trashing
3338 if (d_ancestor(old_dentry, new_dentry))
3343 } while (read_seqretry(&rename_lock, seq));
3348 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3350 struct dentry *root = data;
3351 if (dentry != root) {
3352 if (d_unhashed(dentry) || !dentry->d_inode)
3355 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3356 dentry->d_flags |= DCACHE_GENOCIDE;
3357 dentry->d_lockref.count--;
3360 return D_WALK_CONTINUE;
3363 void d_genocide(struct dentry *parent)
3365 d_walk(parent, parent, d_genocide_kill, NULL);
3368 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3370 inode_dec_link_count(inode);
3371 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3372 !hlist_unhashed(&dentry->d_u.d_alias) ||
3373 !d_unlinked(dentry));
3374 spin_lock(&dentry->d_parent->d_lock);
3375 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3376 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3377 (unsigned long long)inode->i_ino);
3378 spin_unlock(&dentry->d_lock);
3379 spin_unlock(&dentry->d_parent->d_lock);
3380 d_instantiate(dentry, inode);
3382 EXPORT_SYMBOL(d_tmpfile);
3384 static __initdata unsigned long dhash_entries;
3385 static int __init set_dhash_entries(char *str)
3389 dhash_entries = simple_strtoul(str, &str, 0);
3392 __setup("dhash_entries=", set_dhash_entries);
3394 static void __init dcache_init_early(void)
3398 /* If hashes are distributed across NUMA nodes, defer
3399 * hash allocation until vmalloc space is available.
3405 alloc_large_system_hash("Dentry cache",
3406 sizeof(struct hlist_bl_head),
3415 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3416 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3419 static void __init dcache_init(void)
3424 * A constructor could be added for stable state like the lists,
3425 * but it is probably not worth it because of the cache nature
3428 dentry_cache = KMEM_CACHE(dentry,
3429 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3431 /* Hash may have been set up in dcache_init_early */
3436 alloc_large_system_hash("Dentry cache",
3437 sizeof(struct hlist_bl_head),
3446 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3447 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3450 /* SLAB cache for __getname() consumers */
3451 struct kmem_cache *names_cachep __read_mostly;
3452 EXPORT_SYMBOL(names_cachep);
3454 EXPORT_SYMBOL(d_genocide);
3456 void __init vfs_caches_init_early(void)
3458 dcache_init_early();
3462 void __init vfs_caches_init(void)
3464 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3465 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3470 files_maxfiles_init();