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>
41 #include <linux/kasan.h>
48 * dcache->d_inode->i_lock protects:
49 * - i_dentry, d_u.d_alias, d_inode of aliases
50 * dcache_hash_bucket lock protects:
51 * - the dcache hash table
52 * s_anon bl list spinlock protects:
53 * - the s_anon list (see __d_drop)
54 * dentry->d_sb->s_dentry_lru_lock protects:
55 * - the dcache lru lists and counters
62 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
64 * - d_u.d_alias, d_inode
67 * dentry->d_inode->i_lock
69 * dentry->d_sb->s_dentry_lru_lock
70 * dcache_hash_bucket lock
73 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
76 * dentry->d_parent->d_lock
79 * If no ancestor relationship:
80 * if (dentry1 < dentry2)
84 int sysctl_vfs_cache_pressure __read_mostly = 100;
85 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
87 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
89 EXPORT_SYMBOL(rename_lock);
91 static struct kmem_cache *dentry_cache __read_mostly;
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
102 static unsigned int d_hash_mask __read_mostly;
103 static unsigned int d_hash_shift __read_mostly;
105 static struct hlist_bl_head *dentry_hashtable __read_mostly;
107 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 return dentry_hashtable + hash_32(hash, d_hash_shift);
114 #define IN_LOOKUP_SHIFT 10
115 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
117 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
120 hash += (unsigned long) parent / L1_CACHE_BYTES;
121 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
125 /* Statistics gathering. */
126 struct dentry_stat_t dentry_stat = {
130 static DEFINE_PER_CPU(long, nr_dentry);
131 static DEFINE_PER_CPU(long, nr_dentry_unused);
133 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
136 * Here we resort to our own counters instead of using generic per-cpu counters
137 * for consistency with what the vfs inode code does. We are expected to harvest
138 * better code and performance by having our own specialized counters.
140 * Please note that the loop is done over all possible CPUs, not over all online
141 * CPUs. The reason for this is that we don't want to play games with CPUs going
142 * on and off. If one of them goes off, we will just keep their counters.
144 * glommer: See cffbc8a for details, and if you ever intend to change this,
145 * please update all vfs counters to match.
147 static long get_nr_dentry(void)
151 for_each_possible_cpu(i)
152 sum += per_cpu(nr_dentry, i);
153 return sum < 0 ? 0 : sum;
156 static long get_nr_dentry_unused(void)
160 for_each_possible_cpu(i)
161 sum += per_cpu(nr_dentry_unused, i);
162 return sum < 0 ? 0 : sum;
165 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
166 size_t *lenp, loff_t *ppos)
168 dentry_stat.nr_dentry = get_nr_dentry();
169 dentry_stat.nr_unused = get_nr_dentry_unused();
170 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
175 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
176 * The strings are both count bytes long, and count is non-zero.
178 #ifdef CONFIG_DCACHE_WORD_ACCESS
180 #include <asm/word-at-a-time.h>
182 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
183 * aligned allocation for this particular component. We don't
184 * strictly need the load_unaligned_zeropad() safety, but it
185 * doesn't hurt either.
187 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
188 * need the careful unaligned handling.
190 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
192 unsigned long a,b,mask;
195 a = *(unsigned long *)cs;
196 b = load_unaligned_zeropad(ct);
197 if (tcount < sizeof(unsigned long))
199 if (unlikely(a != b))
201 cs += sizeof(unsigned long);
202 ct += sizeof(unsigned long);
203 tcount -= sizeof(unsigned long);
207 mask = bytemask_from_count(tcount);
208 return unlikely(!!((a ^ b) & mask));
213 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
227 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
230 * Be careful about RCU walk racing with rename:
231 * use 'lockless_dereference' to fetch the name pointer.
233 * NOTE! Even if a rename will mean that the length
234 * was not loaded atomically, we don't care. The
235 * RCU walk will check the sequence count eventually,
236 * and catch it. And we won't overrun the buffer,
237 * because we're reading the name pointer atomically,
238 * and a dentry name is guaranteed to be properly
239 * terminated with a NUL byte.
241 * End result: even if 'len' is wrong, we'll exit
242 * early because the data cannot match (there can
243 * be no NUL in the ct/tcount data)
245 const unsigned char *cs = lockless_dereference(dentry->d_name.name);
247 return dentry_string_cmp(cs, ct, tcount);
250 struct external_name {
253 struct rcu_head head;
255 unsigned char name[];
258 static inline struct external_name *external_name(struct dentry *dentry)
260 return container_of(dentry->d_name.name, struct external_name, name[0]);
263 static void __d_free(struct rcu_head *head)
265 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
267 kmem_cache_free(dentry_cache, dentry);
270 static void __d_free_external(struct rcu_head *head)
272 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
273 kfree(external_name(dentry));
274 kmem_cache_free(dentry_cache, dentry);
277 static inline int dname_external(const struct dentry *dentry)
279 return dentry->d_name.name != dentry->d_iname;
282 static inline void __d_set_inode_and_type(struct dentry *dentry,
288 dentry->d_inode = inode;
289 flags = READ_ONCE(dentry->d_flags);
290 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
292 WRITE_ONCE(dentry->d_flags, flags);
295 static inline void __d_clear_type_and_inode(struct dentry *dentry)
297 unsigned flags = READ_ONCE(dentry->d_flags);
299 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
300 WRITE_ONCE(dentry->d_flags, flags);
301 dentry->d_inode = NULL;
304 static void dentry_free(struct dentry *dentry)
306 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
307 if (unlikely(dname_external(dentry))) {
308 struct external_name *p = external_name(dentry);
309 if (likely(atomic_dec_and_test(&p->u.count))) {
310 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
314 /* if dentry was never visible to RCU, immediate free is OK */
315 if (!(dentry->d_flags & DCACHE_RCUACCESS))
316 __d_free(&dentry->d_u.d_rcu);
318 call_rcu(&dentry->d_u.d_rcu, __d_free);
322 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
323 * @dentry: the target dentry
324 * After this call, in-progress rcu-walk path lookup will fail. This
325 * should be called after unhashing, and after changing d_inode (if
326 * the dentry has not already been unhashed).
328 static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
330 lockdep_assert_held(&dentry->d_lock);
331 /* Go through am invalidation barrier */
332 write_seqcount_invalidate(&dentry->d_seq);
336 * Release the dentry's inode, using the filesystem
337 * d_iput() operation if defined. Dentry has no refcount
340 static void dentry_iput(struct dentry * dentry)
341 __releases(dentry->d_lock)
342 __releases(dentry->d_inode->i_lock)
344 struct inode *inode = dentry->d_inode;
346 __d_clear_type_and_inode(dentry);
347 hlist_del_init(&dentry->d_u.d_alias);
348 spin_unlock(&dentry->d_lock);
349 spin_unlock(&inode->i_lock);
351 fsnotify_inoderemove(inode);
352 if (dentry->d_op && dentry->d_op->d_iput)
353 dentry->d_op->d_iput(dentry, inode);
357 spin_unlock(&dentry->d_lock);
362 * Release the dentry's inode, using the filesystem
363 * d_iput() operation if defined. dentry remains in-use.
365 static void dentry_unlink_inode(struct dentry * dentry)
366 __releases(dentry->d_lock)
367 __releases(dentry->d_inode->i_lock)
369 struct inode *inode = dentry->d_inode;
371 raw_write_seqcount_begin(&dentry->d_seq);
372 __d_clear_type_and_inode(dentry);
373 hlist_del_init(&dentry->d_u.d_alias);
374 raw_write_seqcount_end(&dentry->d_seq);
375 spin_unlock(&dentry->d_lock);
376 spin_unlock(&inode->i_lock);
378 fsnotify_inoderemove(inode);
379 if (dentry->d_op && dentry->d_op->d_iput)
380 dentry->d_op->d_iput(dentry, inode);
386 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
387 * is in use - which includes both the "real" per-superblock
388 * LRU list _and_ the DCACHE_SHRINK_LIST use.
390 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
391 * on the shrink list (ie not on the superblock LRU list).
393 * The per-cpu "nr_dentry_unused" counters are updated with
394 * the DCACHE_LRU_LIST bit.
396 * These helper functions make sure we always follow the
397 * rules. d_lock must be held by the caller.
399 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
400 static void d_lru_add(struct dentry *dentry)
402 D_FLAG_VERIFY(dentry, 0);
403 dentry->d_flags |= DCACHE_LRU_LIST;
404 this_cpu_inc(nr_dentry_unused);
405 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
408 static void d_lru_del(struct dentry *dentry)
410 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
411 dentry->d_flags &= ~DCACHE_LRU_LIST;
412 this_cpu_dec(nr_dentry_unused);
413 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
416 static void d_shrink_del(struct dentry *dentry)
418 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
419 list_del_init(&dentry->d_lru);
420 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
421 this_cpu_dec(nr_dentry_unused);
424 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
426 D_FLAG_VERIFY(dentry, 0);
427 list_add(&dentry->d_lru, list);
428 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
429 this_cpu_inc(nr_dentry_unused);
433 * These can only be called under the global LRU lock, ie during the
434 * callback for freeing the LRU list. "isolate" removes it from the
435 * LRU lists entirely, while shrink_move moves it to the indicated
438 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
440 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
441 dentry->d_flags &= ~DCACHE_LRU_LIST;
442 this_cpu_dec(nr_dentry_unused);
443 list_lru_isolate(lru, &dentry->d_lru);
446 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
447 struct list_head *list)
449 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
450 dentry->d_flags |= DCACHE_SHRINK_LIST;
451 list_lru_isolate_move(lru, &dentry->d_lru, list);
455 * dentry_lru_(add|del)_list) must be called with d_lock held.
457 static void dentry_lru_add(struct dentry *dentry)
459 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
464 * d_drop - drop a dentry
465 * @dentry: dentry to drop
467 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
468 * be found through a VFS lookup any more. Note that this is different from
469 * deleting the dentry - d_delete will try to mark the dentry negative if
470 * possible, giving a successful _negative_ lookup, while d_drop will
471 * just make the cache lookup fail.
473 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
474 * reason (NFS timeouts or autofs deletes).
476 * __d_drop requires dentry->d_lock.
478 void __d_drop(struct dentry *dentry)
480 if (!d_unhashed(dentry)) {
481 struct hlist_bl_head *b;
483 * Hashed dentries are normally on the dentry hashtable,
484 * with the exception of those newly allocated by
485 * d_obtain_alias, which are always IS_ROOT:
487 if (unlikely(IS_ROOT(dentry)))
488 b = &dentry->d_sb->s_anon;
490 b = d_hash(dentry->d_parent, dentry->d_name.hash);
493 __hlist_bl_del(&dentry->d_hash);
494 dentry->d_hash.pprev = NULL;
496 dentry_rcuwalk_invalidate(dentry);
499 EXPORT_SYMBOL(__d_drop);
501 void d_drop(struct dentry *dentry)
503 spin_lock(&dentry->d_lock);
505 spin_unlock(&dentry->d_lock);
507 EXPORT_SYMBOL(d_drop);
509 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
513 * Inform d_walk() and shrink_dentry_list() that we are no longer
514 * attached to the dentry tree
516 dentry->d_flags |= DCACHE_DENTRY_KILLED;
517 if (unlikely(list_empty(&dentry->d_child)))
519 __list_del_entry(&dentry->d_child);
521 * Cursors can move around the list of children. While we'd been
522 * a normal list member, it didn't matter - ->d_child.next would've
523 * been updated. However, from now on it won't be and for the
524 * things like d_walk() it might end up with a nasty surprise.
525 * Normally d_walk() doesn't care about cursors moving around -
526 * ->d_lock on parent prevents that and since a cursor has no children
527 * of its own, we get through it without ever unlocking the parent.
528 * There is one exception, though - if we ascend from a child that
529 * gets killed as soon as we unlock it, the next sibling is found
530 * using the value left in its ->d_child.next. And if _that_
531 * pointed to a cursor, and cursor got moved (e.g. by lseek())
532 * before d_walk() regains parent->d_lock, we'll end up skipping
533 * everything the cursor had been moved past.
535 * Solution: make sure that the pointer left behind in ->d_child.next
536 * points to something that won't be moving around. I.e. skip the
539 while (dentry->d_child.next != &parent->d_subdirs) {
540 next = list_entry(dentry->d_child.next, struct dentry, d_child);
541 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
543 dentry->d_child.next = next->d_child.next;
547 static void __dentry_kill(struct dentry *dentry)
549 struct dentry *parent = NULL;
550 bool can_free = true;
551 if (!IS_ROOT(dentry))
552 parent = dentry->d_parent;
555 * The dentry is now unrecoverably dead to the world.
557 lockref_mark_dead(&dentry->d_lockref);
560 * inform the fs via d_prune that this dentry is about to be
561 * unhashed and destroyed.
563 if (dentry->d_flags & DCACHE_OP_PRUNE)
564 dentry->d_op->d_prune(dentry);
566 if (dentry->d_flags & DCACHE_LRU_LIST) {
567 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
570 /* if it was on the hash then remove it */
572 dentry_unlist(dentry, parent);
574 spin_unlock(&parent->d_lock);
577 * dentry_iput drops the locks, at which point nobody (except
578 * transient RCU lookups) can reach this dentry.
580 BUG_ON(dentry->d_lockref.count > 0);
581 this_cpu_dec(nr_dentry);
582 if (dentry->d_op && dentry->d_op->d_release)
583 dentry->d_op->d_release(dentry);
585 spin_lock(&dentry->d_lock);
586 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
587 dentry->d_flags |= DCACHE_MAY_FREE;
590 spin_unlock(&dentry->d_lock);
591 if (likely(can_free))
596 * Finish off a dentry we've decided to kill.
597 * dentry->d_lock must be held, returns with it unlocked.
598 * If ref is non-zero, then decrement the refcount too.
599 * Returns dentry requiring refcount drop, or NULL if we're done.
601 static struct dentry *dentry_kill(struct dentry *dentry)
602 __releases(dentry->d_lock)
604 struct inode *inode = dentry->d_inode;
605 struct dentry *parent = NULL;
607 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
610 if (!IS_ROOT(dentry)) {
611 parent = dentry->d_parent;
612 if (unlikely(!spin_trylock(&parent->d_lock))) {
614 spin_unlock(&inode->i_lock);
619 __dentry_kill(dentry);
623 spin_unlock(&dentry->d_lock);
625 return dentry; /* try again with same dentry */
628 static inline struct dentry *lock_parent(struct dentry *dentry)
630 struct dentry *parent = dentry->d_parent;
633 if (unlikely(dentry->d_lockref.count < 0))
635 if (likely(spin_trylock(&parent->d_lock)))
638 spin_unlock(&dentry->d_lock);
640 parent = ACCESS_ONCE(dentry->d_parent);
641 spin_lock(&parent->d_lock);
643 * We can't blindly lock dentry until we are sure
644 * that we won't violate the locking order.
645 * Any changes of dentry->d_parent must have
646 * been done with parent->d_lock held, so
647 * spin_lock() above is enough of a barrier
648 * for checking if it's still our child.
650 if (unlikely(parent != dentry->d_parent)) {
651 spin_unlock(&parent->d_lock);
655 if (parent != dentry)
656 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
663 * Try to do a lockless dput(), and return whether that was successful.
665 * If unsuccessful, we return false, having already taken the dentry lock.
667 * The caller needs to hold the RCU read lock, so that the dentry is
668 * guaranteed to stay around even if the refcount goes down to zero!
670 static inline bool fast_dput(struct dentry *dentry)
673 unsigned int d_flags;
676 * If we have a d_op->d_delete() operation, we sould not
677 * let the dentry count go to zero, so use "put_or_lock".
679 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
680 return lockref_put_or_lock(&dentry->d_lockref);
683 * .. otherwise, we can try to just decrement the
684 * lockref optimistically.
686 ret = lockref_put_return(&dentry->d_lockref);
689 * If the lockref_put_return() failed due to the lock being held
690 * by somebody else, the fast path has failed. We will need to
691 * get the lock, and then check the count again.
693 if (unlikely(ret < 0)) {
694 spin_lock(&dentry->d_lock);
695 if (dentry->d_lockref.count > 1) {
696 dentry->d_lockref.count--;
697 spin_unlock(&dentry->d_lock);
704 * If we weren't the last ref, we're done.
710 * Careful, careful. The reference count went down
711 * to zero, but we don't hold the dentry lock, so
712 * somebody else could get it again, and do another
713 * dput(), and we need to not race with that.
715 * However, there is a very special and common case
716 * where we don't care, because there is nothing to
717 * do: the dentry is still hashed, it does not have
718 * a 'delete' op, and it's referenced and already on
721 * NOTE! Since we aren't locked, these values are
722 * not "stable". However, it is sufficient that at
723 * some point after we dropped the reference the
724 * dentry was hashed and the flags had the proper
725 * value. Other dentry users may have re-gotten
726 * a reference to the dentry and change that, but
727 * our work is done - we can leave the dentry
728 * around with a zero refcount.
731 d_flags = ACCESS_ONCE(dentry->d_flags);
732 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
734 /* Nothing to do? Dropping the reference was all we needed? */
735 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
739 * Not the fast normal case? Get the lock. We've already decremented
740 * the refcount, but we'll need to re-check the situation after
743 spin_lock(&dentry->d_lock);
746 * Did somebody else grab a reference to it in the meantime, and
747 * we're no longer the last user after all? Alternatively, somebody
748 * else could have killed it and marked it dead. Either way, we
749 * don't need to do anything else.
751 if (dentry->d_lockref.count) {
752 spin_unlock(&dentry->d_lock);
757 * Re-get the reference we optimistically dropped. We hold the
758 * lock, and we just tested that it was zero, so we can just
761 dentry->d_lockref.count = 1;
769 * This is complicated by the fact that we do not want to put
770 * dentries that are no longer on any hash chain on the unused
771 * list: we'd much rather just get rid of them immediately.
773 * However, that implies that we have to traverse the dentry
774 * tree upwards to the parents which might _also_ now be
775 * scheduled for deletion (it may have been only waiting for
776 * its last child to go away).
778 * This tail recursion is done by hand as we don't want to depend
779 * on the compiler to always get this right (gcc generally doesn't).
780 * Real recursion would eat up our stack space.
784 * dput - release a dentry
785 * @dentry: dentry to release
787 * Release a dentry. This will drop the usage count and if appropriate
788 * call the dentry unlink method as well as removing it from the queues and
789 * releasing its resources. If the parent dentries were scheduled for release
790 * they too may now get deleted.
792 void dput(struct dentry *dentry)
794 if (unlikely(!dentry))
799 if (likely(fast_dput(dentry))) {
804 /* Slow case: now with the dentry lock held */
807 WARN_ON(d_in_lookup(dentry));
809 /* Unreachable? Get rid of it */
810 if (unlikely(d_unhashed(dentry)))
813 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
816 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
817 if (dentry->d_op->d_delete(dentry))
821 if (!(dentry->d_flags & DCACHE_REFERENCED))
822 dentry->d_flags |= DCACHE_REFERENCED;
823 dentry_lru_add(dentry);
825 dentry->d_lockref.count--;
826 spin_unlock(&dentry->d_lock);
830 dentry = dentry_kill(dentry);
837 /* This must be called with d_lock held */
838 static inline void __dget_dlock(struct dentry *dentry)
840 dentry->d_lockref.count++;
843 static inline void __dget(struct dentry *dentry)
845 lockref_get(&dentry->d_lockref);
848 struct dentry *dget_parent(struct dentry *dentry)
854 * Do optimistic parent lookup without any
858 ret = ACCESS_ONCE(dentry->d_parent);
859 gotref = lockref_get_not_zero(&ret->d_lockref);
861 if (likely(gotref)) {
862 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
869 * Don't need rcu_dereference because we re-check it was correct under
873 ret = dentry->d_parent;
874 spin_lock(&ret->d_lock);
875 if (unlikely(ret != dentry->d_parent)) {
876 spin_unlock(&ret->d_lock);
881 BUG_ON(!ret->d_lockref.count);
882 ret->d_lockref.count++;
883 spin_unlock(&ret->d_lock);
886 EXPORT_SYMBOL(dget_parent);
889 * d_find_alias - grab a hashed alias of inode
890 * @inode: inode in question
892 * If inode has a hashed alias, or is a directory and has any alias,
893 * acquire the reference to alias and return it. Otherwise return NULL.
894 * Notice that if inode is a directory there can be only one alias and
895 * it can be unhashed only if it has no children, or if it is the root
896 * of a filesystem, or if the directory was renamed and d_revalidate
897 * was the first vfs operation to notice.
899 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
900 * any other hashed alias over that one.
902 static struct dentry *__d_find_alias(struct inode *inode)
904 struct dentry *alias, *discon_alias;
908 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
909 spin_lock(&alias->d_lock);
910 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
911 if (IS_ROOT(alias) &&
912 (alias->d_flags & DCACHE_DISCONNECTED)) {
913 discon_alias = alias;
916 spin_unlock(&alias->d_lock);
920 spin_unlock(&alias->d_lock);
923 alias = discon_alias;
924 spin_lock(&alias->d_lock);
925 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
927 spin_unlock(&alias->d_lock);
930 spin_unlock(&alias->d_lock);
936 struct dentry *d_find_alias(struct inode *inode)
938 struct dentry *de = NULL;
940 if (!hlist_empty(&inode->i_dentry)) {
941 spin_lock(&inode->i_lock);
942 de = __d_find_alias(inode);
943 spin_unlock(&inode->i_lock);
947 EXPORT_SYMBOL(d_find_alias);
950 * Try to kill dentries associated with this inode.
951 * WARNING: you must own a reference to inode.
953 void d_prune_aliases(struct inode *inode)
955 struct dentry *dentry;
957 spin_lock(&inode->i_lock);
958 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
959 spin_lock(&dentry->d_lock);
960 if (!dentry->d_lockref.count) {
961 struct dentry *parent = lock_parent(dentry);
962 if (likely(!dentry->d_lockref.count)) {
963 __dentry_kill(dentry);
968 spin_unlock(&parent->d_lock);
970 spin_unlock(&dentry->d_lock);
972 spin_unlock(&inode->i_lock);
974 EXPORT_SYMBOL(d_prune_aliases);
976 static void shrink_dentry_list(struct list_head *list)
978 struct dentry *dentry, *parent;
980 while (!list_empty(list)) {
982 dentry = list_entry(list->prev, struct dentry, d_lru);
983 spin_lock(&dentry->d_lock);
984 parent = lock_parent(dentry);
987 * The dispose list is isolated and dentries are not accounted
988 * to the LRU here, so we can simply remove it from the list
989 * here regardless of whether it is referenced or not.
991 d_shrink_del(dentry);
994 * We found an inuse dentry which was not removed from
995 * the LRU because of laziness during lookup. Do not free it.
997 if (dentry->d_lockref.count > 0) {
998 spin_unlock(&dentry->d_lock);
1000 spin_unlock(&parent->d_lock);
1005 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1006 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1007 spin_unlock(&dentry->d_lock);
1009 spin_unlock(&parent->d_lock);
1011 dentry_free(dentry);
1015 inode = dentry->d_inode;
1016 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1017 d_shrink_add(dentry, list);
1018 spin_unlock(&dentry->d_lock);
1020 spin_unlock(&parent->d_lock);
1024 __dentry_kill(dentry);
1027 * We need to prune ancestors too. This is necessary to prevent
1028 * quadratic behavior of shrink_dcache_parent(), but is also
1029 * expected to be beneficial in reducing dentry cache
1033 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1034 parent = lock_parent(dentry);
1035 if (dentry->d_lockref.count != 1) {
1036 dentry->d_lockref.count--;
1037 spin_unlock(&dentry->d_lock);
1039 spin_unlock(&parent->d_lock);
1042 inode = dentry->d_inode; /* can't be NULL */
1043 if (unlikely(!spin_trylock(&inode->i_lock))) {
1044 spin_unlock(&dentry->d_lock);
1046 spin_unlock(&parent->d_lock);
1050 __dentry_kill(dentry);
1056 static enum lru_status dentry_lru_isolate(struct list_head *item,
1057 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1059 struct list_head *freeable = arg;
1060 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1064 * we are inverting the lru lock/dentry->d_lock here,
1065 * so use a trylock. If we fail to get the lock, just skip
1068 if (!spin_trylock(&dentry->d_lock))
1072 * Referenced dentries are still in use. If they have active
1073 * counts, just remove them from the LRU. Otherwise give them
1074 * another pass through the LRU.
1076 if (dentry->d_lockref.count) {
1077 d_lru_isolate(lru, dentry);
1078 spin_unlock(&dentry->d_lock);
1082 if (dentry->d_flags & DCACHE_REFERENCED) {
1083 dentry->d_flags &= ~DCACHE_REFERENCED;
1084 spin_unlock(&dentry->d_lock);
1087 * The list move itself will be made by the common LRU code. At
1088 * this point, we've dropped the dentry->d_lock but keep the
1089 * lru lock. This is safe to do, since every list movement is
1090 * protected by the lru lock even if both locks are held.
1092 * This is guaranteed by the fact that all LRU management
1093 * functions are intermediated by the LRU API calls like
1094 * list_lru_add and list_lru_del. List movement in this file
1095 * only ever occur through this functions or through callbacks
1096 * like this one, that are called from the LRU API.
1098 * The only exceptions to this are functions like
1099 * shrink_dentry_list, and code that first checks for the
1100 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1101 * operating only with stack provided lists after they are
1102 * properly isolated from the main list. It is thus, always a
1108 d_lru_shrink_move(lru, dentry, freeable);
1109 spin_unlock(&dentry->d_lock);
1115 * prune_dcache_sb - shrink the dcache
1117 * @sc: shrink control, passed to list_lru_shrink_walk()
1119 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1120 * is done when we need more memory and called from the superblock shrinker
1123 * This function may fail to free any resources if all the dentries are in
1126 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1131 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1132 dentry_lru_isolate, &dispose);
1133 shrink_dentry_list(&dispose);
1137 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1138 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1140 struct list_head *freeable = arg;
1141 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1144 * we are inverting the lru lock/dentry->d_lock here,
1145 * so use a trylock. If we fail to get the lock, just skip
1148 if (!spin_trylock(&dentry->d_lock))
1151 d_lru_shrink_move(lru, dentry, freeable);
1152 spin_unlock(&dentry->d_lock);
1159 * shrink_dcache_sb - shrink dcache for a superblock
1162 * Shrink the dcache for the specified super block. This is used to free
1163 * the dcache before unmounting a file system.
1165 void shrink_dcache_sb(struct super_block *sb)
1172 freed = list_lru_walk(&sb->s_dentry_lru,
1173 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1175 this_cpu_sub(nr_dentry_unused, freed);
1176 shrink_dentry_list(&dispose);
1177 } while (freed > 0);
1179 EXPORT_SYMBOL(shrink_dcache_sb);
1182 * enum d_walk_ret - action to talke during tree walk
1183 * @D_WALK_CONTINUE: contrinue walk
1184 * @D_WALK_QUIT: quit walk
1185 * @D_WALK_NORETRY: quit when retry is needed
1186 * @D_WALK_SKIP: skip this dentry and its children
1196 * d_walk - walk the dentry tree
1197 * @parent: start of walk
1198 * @data: data passed to @enter() and @finish()
1199 * @enter: callback when first entering the dentry
1200 * @finish: callback when successfully finished the walk
1202 * The @enter() and @finish() callbacks are called with d_lock held.
1204 static void d_walk(struct dentry *parent, void *data,
1205 enum d_walk_ret (*enter)(void *, struct dentry *),
1206 void (*finish)(void *))
1208 struct dentry *this_parent;
1209 struct list_head *next;
1211 enum d_walk_ret ret;
1215 read_seqbegin_or_lock(&rename_lock, &seq);
1216 this_parent = parent;
1217 spin_lock(&this_parent->d_lock);
1219 ret = enter(data, this_parent);
1221 case D_WALK_CONTINUE:
1226 case D_WALK_NORETRY:
1231 next = this_parent->d_subdirs.next;
1233 while (next != &this_parent->d_subdirs) {
1234 struct list_head *tmp = next;
1235 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1238 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1241 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1243 ret = enter(data, dentry);
1245 case D_WALK_CONTINUE:
1248 spin_unlock(&dentry->d_lock);
1250 case D_WALK_NORETRY:
1254 spin_unlock(&dentry->d_lock);
1258 if (!list_empty(&dentry->d_subdirs)) {
1259 spin_unlock(&this_parent->d_lock);
1260 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1261 this_parent = dentry;
1262 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1265 spin_unlock(&dentry->d_lock);
1268 * All done at this level ... ascend and resume the search.
1272 if (this_parent != parent) {
1273 struct dentry *child = this_parent;
1274 this_parent = child->d_parent;
1276 spin_unlock(&child->d_lock);
1277 spin_lock(&this_parent->d_lock);
1279 /* might go back up the wrong parent if we have had a rename. */
1280 if (need_seqretry(&rename_lock, seq))
1282 /* go into the first sibling still alive */
1284 next = child->d_child.next;
1285 if (next == &this_parent->d_subdirs)
1287 child = list_entry(next, struct dentry, d_child);
1288 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1292 if (need_seqretry(&rename_lock, seq))
1299 spin_unlock(&this_parent->d_lock);
1300 done_seqretry(&rename_lock, seq);
1304 spin_unlock(&this_parent->d_lock);
1314 * Search for at least 1 mount point in the dentry's subdirs.
1315 * We descend to the next level whenever the d_subdirs
1316 * list is non-empty and continue searching.
1319 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1322 if (d_mountpoint(dentry)) {
1326 return D_WALK_CONTINUE;
1330 * have_submounts - check for mounts over a dentry
1331 * @parent: dentry to check.
1333 * Return true if the parent or its subdirectories contain
1336 int have_submounts(struct dentry *parent)
1340 d_walk(parent, &ret, check_mount, NULL);
1344 EXPORT_SYMBOL(have_submounts);
1347 * Called by mount code to set a mountpoint and check if the mountpoint is
1348 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1349 * subtree can become unreachable).
1351 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1352 * this reason take rename_lock and d_lock on dentry and ancestors.
1354 int d_set_mounted(struct dentry *dentry)
1358 write_seqlock(&rename_lock);
1359 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1360 /* Need exclusion wrt. d_invalidate() */
1361 spin_lock(&p->d_lock);
1362 if (unlikely(d_unhashed(p))) {
1363 spin_unlock(&p->d_lock);
1366 spin_unlock(&p->d_lock);
1368 spin_lock(&dentry->d_lock);
1369 if (!d_unlinked(dentry)) {
1370 dentry->d_flags |= DCACHE_MOUNTED;
1373 spin_unlock(&dentry->d_lock);
1375 write_sequnlock(&rename_lock);
1380 * Search the dentry child list of the specified parent,
1381 * and move any unused dentries to the end of the unused
1382 * list for prune_dcache(). We descend to the next level
1383 * whenever the d_subdirs list is non-empty and continue
1386 * It returns zero iff there are no unused children,
1387 * otherwise it returns the number of children moved to
1388 * the end of the unused list. This may not be the total
1389 * number of unused children, because select_parent can
1390 * drop the lock and return early due to latency
1394 struct select_data {
1395 struct dentry *start;
1396 struct list_head dispose;
1400 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1402 struct select_data *data = _data;
1403 enum d_walk_ret ret = D_WALK_CONTINUE;
1405 if (data->start == dentry)
1408 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1411 if (dentry->d_flags & DCACHE_LRU_LIST)
1413 if (!dentry->d_lockref.count) {
1414 d_shrink_add(dentry, &data->dispose);
1419 * We can return to the caller if we have found some (this
1420 * ensures forward progress). We'll be coming back to find
1423 if (!list_empty(&data->dispose))
1424 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1430 * shrink_dcache_parent - prune dcache
1431 * @parent: parent of entries to prune
1433 * Prune the dcache to remove unused children of the parent dentry.
1435 void shrink_dcache_parent(struct dentry *parent)
1438 struct select_data data;
1440 INIT_LIST_HEAD(&data.dispose);
1441 data.start = parent;
1444 d_walk(parent, &data, select_collect, NULL);
1448 shrink_dentry_list(&data.dispose);
1452 EXPORT_SYMBOL(shrink_dcache_parent);
1454 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1456 /* it has busy descendents; complain about those instead */
1457 if (!list_empty(&dentry->d_subdirs))
1458 return D_WALK_CONTINUE;
1460 /* root with refcount 1 is fine */
1461 if (dentry == _data && dentry->d_lockref.count == 1)
1462 return D_WALK_CONTINUE;
1464 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1465 " still in use (%d) [unmount of %s %s]\n",
1468 dentry->d_inode->i_ino : 0UL,
1470 dentry->d_lockref.count,
1471 dentry->d_sb->s_type->name,
1472 dentry->d_sb->s_id);
1474 return D_WALK_CONTINUE;
1477 static void do_one_tree(struct dentry *dentry)
1479 shrink_dcache_parent(dentry);
1480 d_walk(dentry, dentry, umount_check, NULL);
1486 * destroy the dentries attached to a superblock on unmounting
1488 void shrink_dcache_for_umount(struct super_block *sb)
1490 struct dentry *dentry;
1492 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1494 dentry = sb->s_root;
1496 do_one_tree(dentry);
1498 while (!hlist_bl_empty(&sb->s_anon)) {
1499 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1500 do_one_tree(dentry);
1504 struct detach_data {
1505 struct select_data select;
1506 struct dentry *mountpoint;
1508 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1510 struct detach_data *data = _data;
1512 if (d_mountpoint(dentry)) {
1513 __dget_dlock(dentry);
1514 data->mountpoint = dentry;
1518 return select_collect(&data->select, dentry);
1521 static void check_and_drop(void *_data)
1523 struct detach_data *data = _data;
1525 if (!data->mountpoint && !data->select.found)
1526 __d_drop(data->select.start);
1530 * d_invalidate - detach submounts, prune dcache, and drop
1531 * @dentry: dentry to invalidate (aka detach, prune and drop)
1535 * The final d_drop is done as an atomic operation relative to
1536 * rename_lock ensuring there are no races with d_set_mounted. This
1537 * ensures there are no unhashed dentries on the path to a mountpoint.
1539 void d_invalidate(struct dentry *dentry)
1542 * If it's already been dropped, return OK.
1544 spin_lock(&dentry->d_lock);
1545 if (d_unhashed(dentry)) {
1546 spin_unlock(&dentry->d_lock);
1549 spin_unlock(&dentry->d_lock);
1551 /* Negative dentries can be dropped without further checks */
1552 if (!dentry->d_inode) {
1558 struct detach_data data;
1560 data.mountpoint = NULL;
1561 INIT_LIST_HEAD(&data.select.dispose);
1562 data.select.start = dentry;
1563 data.select.found = 0;
1565 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1567 if (data.select.found)
1568 shrink_dentry_list(&data.select.dispose);
1570 if (data.mountpoint) {
1571 detach_mounts(data.mountpoint);
1572 dput(data.mountpoint);
1575 if (!data.mountpoint && !data.select.found)
1581 EXPORT_SYMBOL(d_invalidate);
1584 * __d_alloc - allocate a dcache entry
1585 * @sb: filesystem it will belong to
1586 * @name: qstr of the name
1588 * Allocates a dentry. It returns %NULL if there is insufficient memory
1589 * available. On a success the dentry is returned. The name passed in is
1590 * copied and the copy passed in may be reused after this call.
1593 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1595 struct dentry *dentry;
1598 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1603 * We guarantee that the inline name is always NUL-terminated.
1604 * This way the memcpy() done by the name switching in rename
1605 * will still always have a NUL at the end, even if we might
1606 * be overwriting an internal NUL character
1608 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1609 if (unlikely(!name)) {
1610 static const struct qstr anon = QSTR_INIT("/", 1);
1612 dname = dentry->d_iname;
1613 } else if (name->len > DNAME_INLINE_LEN-1) {
1614 size_t size = offsetof(struct external_name, name[1]);
1615 struct external_name *p = kmalloc(size + name->len,
1616 GFP_KERNEL_ACCOUNT);
1618 kmem_cache_free(dentry_cache, dentry);
1621 atomic_set(&p->u.count, 1);
1623 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1624 kasan_unpoison_shadow(dname,
1625 round_up(name->len + 1, sizeof(unsigned long)));
1627 dname = dentry->d_iname;
1630 dentry->d_name.len = name->len;
1631 dentry->d_name.hash = name->hash;
1632 memcpy(dname, name->name, name->len);
1633 dname[name->len] = 0;
1635 /* Make sure we always see the terminating NUL character */
1637 dentry->d_name.name = dname;
1639 dentry->d_lockref.count = 1;
1640 dentry->d_flags = 0;
1641 spin_lock_init(&dentry->d_lock);
1642 seqcount_init(&dentry->d_seq);
1643 dentry->d_inode = NULL;
1644 dentry->d_parent = dentry;
1646 dentry->d_op = NULL;
1647 dentry->d_fsdata = NULL;
1648 INIT_HLIST_BL_NODE(&dentry->d_hash);
1649 INIT_LIST_HEAD(&dentry->d_lru);
1650 INIT_LIST_HEAD(&dentry->d_subdirs);
1651 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1652 INIT_LIST_HEAD(&dentry->d_child);
1653 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1655 this_cpu_inc(nr_dentry);
1661 * d_alloc - allocate a dcache entry
1662 * @parent: parent of entry to allocate
1663 * @name: qstr of the name
1665 * Allocates a dentry. It returns %NULL if there is insufficient memory
1666 * available. On a success the dentry is returned. The name passed in is
1667 * copied and the copy passed in may be reused after this call.
1669 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1671 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1674 dentry->d_flags |= DCACHE_RCUACCESS;
1675 spin_lock(&parent->d_lock);
1677 * don't need child lock because it is not subject
1678 * to concurrency here
1680 __dget_dlock(parent);
1681 dentry->d_parent = parent;
1682 list_add(&dentry->d_child, &parent->d_subdirs);
1683 spin_unlock(&parent->d_lock);
1687 EXPORT_SYMBOL(d_alloc);
1689 struct dentry *d_alloc_cursor(struct dentry * parent)
1691 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1693 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1694 dentry->d_parent = dget(parent);
1700 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1701 * @sb: the superblock
1702 * @name: qstr of the name
1704 * For a filesystem that just pins its dentries in memory and never
1705 * performs lookups at all, return an unhashed IS_ROOT dentry.
1707 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1709 return __d_alloc(sb, name);
1711 EXPORT_SYMBOL(d_alloc_pseudo);
1713 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1718 q.hash_len = hashlen_string(name);
1719 return d_alloc(parent, &q);
1721 EXPORT_SYMBOL(d_alloc_name);
1723 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1725 WARN_ON_ONCE(dentry->d_op);
1726 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1728 DCACHE_OP_REVALIDATE |
1729 DCACHE_OP_WEAK_REVALIDATE |
1731 DCACHE_OP_SELECT_INODE |
1737 dentry->d_flags |= DCACHE_OP_HASH;
1739 dentry->d_flags |= DCACHE_OP_COMPARE;
1740 if (op->d_revalidate)
1741 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1742 if (op->d_weak_revalidate)
1743 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1745 dentry->d_flags |= DCACHE_OP_DELETE;
1747 dentry->d_flags |= DCACHE_OP_PRUNE;
1748 if (op->d_select_inode)
1749 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
1751 dentry->d_flags |= DCACHE_OP_REAL;
1754 EXPORT_SYMBOL(d_set_d_op);
1758 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1759 * @dentry - The dentry to mark
1761 * Mark a dentry as falling through to the lower layer (as set with
1762 * d_pin_lower()). This flag may be recorded on the medium.
1764 void d_set_fallthru(struct dentry *dentry)
1766 spin_lock(&dentry->d_lock);
1767 dentry->d_flags |= DCACHE_FALLTHRU;
1768 spin_unlock(&dentry->d_lock);
1770 EXPORT_SYMBOL(d_set_fallthru);
1772 static unsigned d_flags_for_inode(struct inode *inode)
1774 unsigned add_flags = DCACHE_REGULAR_TYPE;
1777 return DCACHE_MISS_TYPE;
1779 if (S_ISDIR(inode->i_mode)) {
1780 add_flags = DCACHE_DIRECTORY_TYPE;
1781 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1782 if (unlikely(!inode->i_op->lookup))
1783 add_flags = DCACHE_AUTODIR_TYPE;
1785 inode->i_opflags |= IOP_LOOKUP;
1787 goto type_determined;
1790 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1791 if (unlikely(inode->i_op->get_link)) {
1792 add_flags = DCACHE_SYMLINK_TYPE;
1793 goto type_determined;
1795 inode->i_opflags |= IOP_NOFOLLOW;
1798 if (unlikely(!S_ISREG(inode->i_mode)))
1799 add_flags = DCACHE_SPECIAL_TYPE;
1802 if (unlikely(IS_AUTOMOUNT(inode)))
1803 add_flags |= DCACHE_NEED_AUTOMOUNT;
1807 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1809 unsigned add_flags = d_flags_for_inode(inode);
1810 WARN_ON(d_in_lookup(dentry));
1812 spin_lock(&dentry->d_lock);
1813 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1814 raw_write_seqcount_begin(&dentry->d_seq);
1815 __d_set_inode_and_type(dentry, inode, add_flags);
1816 raw_write_seqcount_end(&dentry->d_seq);
1817 fsnotify_update_flags(dentry);
1818 spin_unlock(&dentry->d_lock);
1822 * d_instantiate - fill in inode information for a dentry
1823 * @entry: dentry to complete
1824 * @inode: inode to attach to this dentry
1826 * Fill in inode information in the entry.
1828 * This turns negative dentries into productive full members
1831 * NOTE! This assumes that the inode count has been incremented
1832 * (or otherwise set) by the caller to indicate that it is now
1833 * in use by the dcache.
1836 void d_instantiate(struct dentry *entry, struct inode * inode)
1838 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1840 security_d_instantiate(entry, inode);
1841 spin_lock(&inode->i_lock);
1842 __d_instantiate(entry, inode);
1843 spin_unlock(&inode->i_lock);
1846 EXPORT_SYMBOL(d_instantiate);
1849 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1850 * @entry: dentry to complete
1851 * @inode: inode to attach to this dentry
1853 * Fill in inode information in the entry. If a directory alias is found, then
1854 * return an error (and drop inode). Together with d_materialise_unique() this
1855 * guarantees that a directory inode may never have more than one alias.
1857 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1859 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1861 security_d_instantiate(entry, inode);
1862 spin_lock(&inode->i_lock);
1863 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1864 spin_unlock(&inode->i_lock);
1868 __d_instantiate(entry, inode);
1869 spin_unlock(&inode->i_lock);
1873 EXPORT_SYMBOL(d_instantiate_no_diralias);
1875 struct dentry *d_make_root(struct inode *root_inode)
1877 struct dentry *res = NULL;
1880 res = __d_alloc(root_inode->i_sb, NULL);
1882 d_instantiate(res, root_inode);
1888 EXPORT_SYMBOL(d_make_root);
1890 static struct dentry * __d_find_any_alias(struct inode *inode)
1892 struct dentry *alias;
1894 if (hlist_empty(&inode->i_dentry))
1896 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1902 * d_find_any_alias - find any alias for a given inode
1903 * @inode: inode to find an alias for
1905 * If any aliases exist for the given inode, take and return a
1906 * reference for one of them. If no aliases exist, return %NULL.
1908 struct dentry *d_find_any_alias(struct inode *inode)
1912 spin_lock(&inode->i_lock);
1913 de = __d_find_any_alias(inode);
1914 spin_unlock(&inode->i_lock);
1917 EXPORT_SYMBOL(d_find_any_alias);
1919 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1926 return ERR_PTR(-ESTALE);
1928 return ERR_CAST(inode);
1930 res = d_find_any_alias(inode);
1934 tmp = __d_alloc(inode->i_sb, NULL);
1936 res = ERR_PTR(-ENOMEM);
1940 security_d_instantiate(tmp, inode);
1941 spin_lock(&inode->i_lock);
1942 res = __d_find_any_alias(inode);
1944 spin_unlock(&inode->i_lock);
1949 /* attach a disconnected dentry */
1950 add_flags = d_flags_for_inode(inode);
1953 add_flags |= DCACHE_DISCONNECTED;
1955 spin_lock(&tmp->d_lock);
1956 __d_set_inode_and_type(tmp, inode, add_flags);
1957 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1958 hlist_bl_lock(&tmp->d_sb->s_anon);
1959 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1960 hlist_bl_unlock(&tmp->d_sb->s_anon);
1961 spin_unlock(&tmp->d_lock);
1962 spin_unlock(&inode->i_lock);
1972 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1973 * @inode: inode to allocate the dentry for
1975 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1976 * similar open by handle operations. The returned dentry may be anonymous,
1977 * or may have a full name (if the inode was already in the cache).
1979 * When called on a directory inode, we must ensure that the inode only ever
1980 * has one dentry. If a dentry is found, that is returned instead of
1981 * allocating a new one.
1983 * On successful return, the reference to the inode has been transferred
1984 * to the dentry. In case of an error the reference on the inode is released.
1985 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1986 * be passed in and the error will be propagated to the return value,
1987 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1989 struct dentry *d_obtain_alias(struct inode *inode)
1991 return __d_obtain_alias(inode, 1);
1993 EXPORT_SYMBOL(d_obtain_alias);
1996 * d_obtain_root - find or allocate a dentry for a given inode
1997 * @inode: inode to allocate the dentry for
1999 * Obtain an IS_ROOT dentry for the root of a filesystem.
2001 * We must ensure that directory inodes only ever have one dentry. If a
2002 * dentry is found, that is returned instead of allocating a new one.
2004 * On successful return, the reference to the inode has been transferred
2005 * to the dentry. In case of an error the reference on the inode is
2006 * released. A %NULL or IS_ERR inode may be passed in and will be the
2007 * error will be propagate to the return value, with a %NULL @inode
2008 * replaced by ERR_PTR(-ESTALE).
2010 struct dentry *d_obtain_root(struct inode *inode)
2012 return __d_obtain_alias(inode, 0);
2014 EXPORT_SYMBOL(d_obtain_root);
2017 * d_add_ci - lookup or allocate new dentry with case-exact name
2018 * @inode: the inode case-insensitive lookup has found
2019 * @dentry: the negative dentry that was passed to the parent's lookup func
2020 * @name: the case-exact name to be associated with the returned dentry
2022 * This is to avoid filling the dcache with case-insensitive names to the
2023 * same inode, only the actual correct case is stored in the dcache for
2024 * case-insensitive filesystems.
2026 * For a case-insensitive lookup match and if the the case-exact dentry
2027 * already exists in in the dcache, use it and return it.
2029 * If no entry exists with the exact case name, allocate new dentry with
2030 * the exact case, and return the spliced entry.
2032 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2035 struct dentry *found, *res;
2038 * First check if a dentry matching the name already exists,
2039 * if not go ahead and create it now.
2041 found = d_hash_and_lookup(dentry->d_parent, name);
2046 if (d_in_lookup(dentry)) {
2047 found = d_alloc_parallel(dentry->d_parent, name,
2049 if (IS_ERR(found) || !d_in_lookup(found)) {
2054 found = d_alloc(dentry->d_parent, name);
2057 return ERR_PTR(-ENOMEM);
2060 res = d_splice_alias(inode, found);
2067 EXPORT_SYMBOL(d_add_ci);
2070 * Do the slow-case of the dentry name compare.
2072 * Unlike the dentry_cmp() function, we need to atomically
2073 * load the name and length information, so that the
2074 * filesystem can rely on them, and can use the 'name' and
2075 * 'len' information without worrying about walking off the
2076 * end of memory etc.
2078 * Thus the read_seqcount_retry() and the "duplicate" info
2079 * in arguments (the low-level filesystem should not look
2080 * at the dentry inode or name contents directly, since
2081 * rename can change them while we're in RCU mode).
2083 enum slow_d_compare {
2089 static noinline enum slow_d_compare slow_dentry_cmp(
2090 const struct dentry *parent,
2091 struct dentry *dentry,
2093 const struct qstr *name)
2095 int tlen = dentry->d_name.len;
2096 const char *tname = dentry->d_name.name;
2098 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2100 return D_COMP_SEQRETRY;
2102 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2103 return D_COMP_NOMATCH;
2108 * __d_lookup_rcu - search for a dentry (racy, store-free)
2109 * @parent: parent dentry
2110 * @name: qstr of name we wish to find
2111 * @seqp: returns d_seq value at the point where the dentry was found
2112 * Returns: dentry, or NULL
2114 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2115 * resolution (store-free path walking) design described in
2116 * Documentation/filesystems/path-lookup.txt.
2118 * This is not to be used outside core vfs.
2120 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2121 * held, and rcu_read_lock held. The returned dentry must not be stored into
2122 * without taking d_lock and checking d_seq sequence count against @seq
2125 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2128 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2129 * the returned dentry, so long as its parent's seqlock is checked after the
2130 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2131 * is formed, giving integrity down the path walk.
2133 * NOTE! The caller *has* to check the resulting dentry against the sequence
2134 * number we've returned before using any of the resulting dentry state!
2136 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2137 const struct qstr *name,
2140 u64 hashlen = name->hash_len;
2141 const unsigned char *str = name->name;
2142 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2143 struct hlist_bl_node *node;
2144 struct dentry *dentry;
2147 * Note: There is significant duplication with __d_lookup_rcu which is
2148 * required to prevent single threaded performance regressions
2149 * especially on architectures where smp_rmb (in seqcounts) are costly.
2150 * Keep the two functions in sync.
2154 * The hash list is protected using RCU.
2156 * Carefully use d_seq when comparing a candidate dentry, to avoid
2157 * races with d_move().
2159 * It is possible that concurrent renames can mess up our list
2160 * walk here and result in missing our dentry, resulting in the
2161 * false-negative result. d_lookup() protects against concurrent
2162 * renames using rename_lock seqlock.
2164 * See Documentation/filesystems/path-lookup.txt for more details.
2166 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2171 * The dentry sequence count protects us from concurrent
2172 * renames, and thus protects parent and name fields.
2174 * The caller must perform a seqcount check in order
2175 * to do anything useful with the returned dentry.
2177 * NOTE! We do a "raw" seqcount_begin here. That means that
2178 * we don't wait for the sequence count to stabilize if it
2179 * is in the middle of a sequence change. If we do the slow
2180 * dentry compare, we will do seqretries until it is stable,
2181 * and if we end up with a successful lookup, we actually
2182 * want to exit RCU lookup anyway.
2184 seq = raw_seqcount_begin(&dentry->d_seq);
2185 if (dentry->d_parent != parent)
2187 if (d_unhashed(dentry))
2190 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2191 if (dentry->d_name.hash != hashlen_hash(hashlen))
2194 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2197 case D_COMP_NOMATCH:
2204 if (dentry->d_name.hash_len != hashlen)
2207 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2214 * d_lookup - search for a dentry
2215 * @parent: parent dentry
2216 * @name: qstr of name we wish to find
2217 * Returns: dentry, or NULL
2219 * d_lookup searches the children of the parent dentry for the name in
2220 * question. If the dentry is found its reference count is incremented and the
2221 * dentry is returned. The caller must use dput to free the entry when it has
2222 * finished using it. %NULL is returned if the dentry does not exist.
2224 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2226 struct dentry *dentry;
2230 seq = read_seqbegin(&rename_lock);
2231 dentry = __d_lookup(parent, name);
2234 } while (read_seqretry(&rename_lock, seq));
2237 EXPORT_SYMBOL(d_lookup);
2240 * __d_lookup - search for a dentry (racy)
2241 * @parent: parent dentry
2242 * @name: qstr of name we wish to find
2243 * Returns: dentry, or NULL
2245 * __d_lookup is like d_lookup, however it may (rarely) return a
2246 * false-negative result due to unrelated rename activity.
2248 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2249 * however it must be used carefully, eg. with a following d_lookup in
2250 * the case of failure.
2252 * __d_lookup callers must be commented.
2254 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2256 unsigned int len = name->len;
2257 unsigned int hash = name->hash;
2258 const unsigned char *str = name->name;
2259 struct hlist_bl_head *b = d_hash(parent, hash);
2260 struct hlist_bl_node *node;
2261 struct dentry *found = NULL;
2262 struct dentry *dentry;
2265 * Note: There is significant duplication with __d_lookup_rcu which is
2266 * required to prevent single threaded performance regressions
2267 * especially on architectures where smp_rmb (in seqcounts) are costly.
2268 * Keep the two functions in sync.
2272 * The hash list is protected using RCU.
2274 * Take d_lock when comparing a candidate dentry, to avoid races
2277 * It is possible that concurrent renames can mess up our list
2278 * walk here and result in missing our dentry, resulting in the
2279 * false-negative result. d_lookup() protects against concurrent
2280 * renames using rename_lock seqlock.
2282 * See Documentation/filesystems/path-lookup.txt for more details.
2286 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2288 if (dentry->d_name.hash != hash)
2291 spin_lock(&dentry->d_lock);
2292 if (dentry->d_parent != parent)
2294 if (d_unhashed(dentry))
2298 * It is safe to compare names since d_move() cannot
2299 * change the qstr (protected by d_lock).
2301 if (parent->d_flags & DCACHE_OP_COMPARE) {
2302 int tlen = dentry->d_name.len;
2303 const char *tname = dentry->d_name.name;
2304 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2307 if (dentry->d_name.len != len)
2309 if (dentry_cmp(dentry, str, len))
2313 dentry->d_lockref.count++;
2315 spin_unlock(&dentry->d_lock);
2318 spin_unlock(&dentry->d_lock);
2326 * d_hash_and_lookup - hash the qstr then search for a dentry
2327 * @dir: Directory to search in
2328 * @name: qstr of name we wish to find
2330 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2332 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2335 * Check for a fs-specific hash function. Note that we must
2336 * calculate the standard hash first, as the d_op->d_hash()
2337 * routine may choose to leave the hash value unchanged.
2339 name->hash = full_name_hash(name->name, name->len);
2340 if (dir->d_flags & DCACHE_OP_HASH) {
2341 int err = dir->d_op->d_hash(dir, name);
2342 if (unlikely(err < 0))
2343 return ERR_PTR(err);
2345 return d_lookup(dir, name);
2347 EXPORT_SYMBOL(d_hash_and_lookup);
2350 * When a file is deleted, we have two options:
2351 * - turn this dentry into a negative dentry
2352 * - unhash this dentry and free it.
2354 * Usually, we want to just turn this into
2355 * a negative dentry, but if anybody else is
2356 * currently using the dentry or the inode
2357 * we can't do that and we fall back on removing
2358 * it from the hash queues and waiting for
2359 * it to be deleted later when it has no users
2363 * d_delete - delete a dentry
2364 * @dentry: The dentry to delete
2366 * Turn the dentry into a negative dentry if possible, otherwise
2367 * remove it from the hash queues so it can be deleted later
2370 void d_delete(struct dentry * dentry)
2372 struct inode *inode;
2375 * Are we the only user?
2378 spin_lock(&dentry->d_lock);
2379 inode = dentry->d_inode;
2380 isdir = S_ISDIR(inode->i_mode);
2381 if (dentry->d_lockref.count == 1) {
2382 if (!spin_trylock(&inode->i_lock)) {
2383 spin_unlock(&dentry->d_lock);
2387 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2388 dentry_unlink_inode(dentry);
2389 fsnotify_nameremove(dentry, isdir);
2393 if (!d_unhashed(dentry))
2396 spin_unlock(&dentry->d_lock);
2398 fsnotify_nameremove(dentry, isdir);
2400 EXPORT_SYMBOL(d_delete);
2402 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2404 BUG_ON(!d_unhashed(entry));
2406 hlist_bl_add_head_rcu(&entry->d_hash, b);
2410 static void _d_rehash(struct dentry * entry)
2412 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2416 * d_rehash - add an entry back to the hash
2417 * @entry: dentry to add to the hash
2419 * Adds a dentry to the hash according to its name.
2422 void d_rehash(struct dentry * entry)
2424 spin_lock(&entry->d_lock);
2426 spin_unlock(&entry->d_lock);
2428 EXPORT_SYMBOL(d_rehash);
2430 static inline unsigned start_dir_add(struct inode *dir)
2434 unsigned n = dir->i_dir_seq;
2435 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2441 static inline void end_dir_add(struct inode *dir, unsigned n)
2443 smp_store_release(&dir->i_dir_seq, n + 2);
2446 static void d_wait_lookup(struct dentry *dentry)
2448 if (d_in_lookup(dentry)) {
2449 DECLARE_WAITQUEUE(wait, current);
2450 add_wait_queue(dentry->d_wait, &wait);
2452 set_current_state(TASK_UNINTERRUPTIBLE);
2453 spin_unlock(&dentry->d_lock);
2455 spin_lock(&dentry->d_lock);
2456 } while (d_in_lookup(dentry));
2460 struct dentry *d_alloc_parallel(struct dentry *parent,
2461 const struct qstr *name,
2462 wait_queue_head_t *wq)
2464 unsigned int len = name->len;
2465 unsigned int hash = name->hash;
2466 const unsigned char *str = name->name;
2467 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2468 struct hlist_bl_node *node;
2469 struct dentry *new = d_alloc(parent, name);
2470 struct dentry *dentry;
2471 unsigned seq, r_seq, d_seq;
2474 return ERR_PTR(-ENOMEM);
2478 seq = smp_load_acquire(&parent->d_inode->i_dir_seq) & ~1;
2479 r_seq = read_seqbegin(&rename_lock);
2480 dentry = __d_lookup_rcu(parent, name, &d_seq);
2481 if (unlikely(dentry)) {
2482 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2486 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2495 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2500 if (unlikely(parent->d_inode->i_dir_seq != seq)) {
2506 * No changes for the parent since the beginning of d_lookup().
2507 * Since all removals from the chain happen with hlist_bl_lock(),
2508 * any potential in-lookup matches are going to stay here until
2509 * we unlock the chain. All fields are stable in everything
2512 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2513 if (dentry->d_name.hash != hash)
2515 if (dentry->d_parent != parent)
2517 if (parent->d_flags & DCACHE_OP_COMPARE) {
2518 int tlen = dentry->d_name.len;
2519 const char *tname = dentry->d_name.name;
2520 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2523 if (dentry->d_name.len != len)
2525 if (dentry_cmp(dentry, str, len))
2529 /* now we can try to grab a reference */
2530 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2537 * somebody is likely to be still doing lookup for it;
2538 * wait for them to finish
2540 spin_lock(&dentry->d_lock);
2541 d_wait_lookup(dentry);
2543 * it's not in-lookup anymore; in principle we should repeat
2544 * everything from dcache lookup, but it's likely to be what
2545 * d_lookup() would've found anyway. If it is, just return it;
2546 * otherwise we really have to repeat the whole thing.
2548 if (unlikely(dentry->d_name.hash != hash))
2550 if (unlikely(dentry->d_parent != parent))
2552 if (unlikely(d_unhashed(dentry)))
2554 if (parent->d_flags & DCACHE_OP_COMPARE) {
2555 int tlen = dentry->d_name.len;
2556 const char *tname = dentry->d_name.name;
2557 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2560 if (unlikely(dentry->d_name.len != len))
2562 if (unlikely(dentry_cmp(dentry, str, len)))
2565 /* OK, it *is* a hashed match; return it */
2566 spin_unlock(&dentry->d_lock);
2571 /* we can't take ->d_lock here; it's OK, though. */
2572 new->d_flags |= DCACHE_PAR_LOOKUP;
2574 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2578 spin_unlock(&dentry->d_lock);
2582 EXPORT_SYMBOL(d_alloc_parallel);
2584 void __d_lookup_done(struct dentry *dentry)
2586 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2587 dentry->d_name.hash);
2589 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2590 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2591 wake_up_all(dentry->d_wait);
2592 dentry->d_wait = NULL;
2594 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2595 INIT_LIST_HEAD(&dentry->d_lru);
2597 EXPORT_SYMBOL(__d_lookup_done);
2599 /* inode->i_lock held if inode is non-NULL */
2601 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2603 struct inode *dir = NULL;
2605 spin_lock(&dentry->d_lock);
2606 if (unlikely(d_in_lookup(dentry))) {
2607 dir = dentry->d_parent->d_inode;
2608 n = start_dir_add(dir);
2609 __d_lookup_done(dentry);
2612 unsigned add_flags = d_flags_for_inode(inode);
2613 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2614 raw_write_seqcount_begin(&dentry->d_seq);
2615 __d_set_inode_and_type(dentry, inode, add_flags);
2616 raw_write_seqcount_end(&dentry->d_seq);
2617 fsnotify_update_flags(dentry);
2621 end_dir_add(dir, n);
2622 spin_unlock(&dentry->d_lock);
2624 spin_unlock(&inode->i_lock);
2628 * d_add - add dentry to hash queues
2629 * @entry: dentry to add
2630 * @inode: The inode to attach to this dentry
2632 * This adds the entry to the hash queues and initializes @inode.
2633 * The entry was actually filled in earlier during d_alloc().
2636 void d_add(struct dentry *entry, struct inode *inode)
2639 security_d_instantiate(entry, inode);
2640 spin_lock(&inode->i_lock);
2642 __d_add(entry, inode);
2644 EXPORT_SYMBOL(d_add);
2647 * d_exact_alias - find and hash an exact unhashed alias
2648 * @entry: dentry to add
2649 * @inode: The inode to go with this dentry
2651 * If an unhashed dentry with the same name/parent and desired
2652 * inode already exists, hash and return it. Otherwise, return
2655 * Parent directory should be locked.
2657 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2659 struct dentry *alias;
2660 int len = entry->d_name.len;
2661 const char *name = entry->d_name.name;
2662 unsigned int hash = entry->d_name.hash;
2664 spin_lock(&inode->i_lock);
2665 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2667 * Don't need alias->d_lock here, because aliases with
2668 * d_parent == entry->d_parent are not subject to name or
2669 * parent changes, because the parent inode i_mutex is held.
2671 if (alias->d_name.hash != hash)
2673 if (alias->d_parent != entry->d_parent)
2675 if (alias->d_name.len != len)
2677 if (dentry_cmp(alias, name, len))
2679 spin_lock(&alias->d_lock);
2680 if (!d_unhashed(alias)) {
2681 spin_unlock(&alias->d_lock);
2684 __dget_dlock(alias);
2686 spin_unlock(&alias->d_lock);
2688 spin_unlock(&inode->i_lock);
2691 spin_unlock(&inode->i_lock);
2694 EXPORT_SYMBOL(d_exact_alias);
2697 * dentry_update_name_case - update case insensitive dentry with a new name
2698 * @dentry: dentry to be updated
2701 * Update a case insensitive dentry with new case of name.
2703 * dentry must have been returned by d_lookup with name @name. Old and new
2704 * name lengths must match (ie. no d_compare which allows mismatched name
2707 * Parent inode i_mutex must be held over d_lookup and into this call (to
2708 * keep renames and concurrent inserts, and readdir(2) away).
2710 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2712 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
2713 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2715 spin_lock(&dentry->d_lock);
2716 write_seqcount_begin(&dentry->d_seq);
2717 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2718 write_seqcount_end(&dentry->d_seq);
2719 spin_unlock(&dentry->d_lock);
2721 EXPORT_SYMBOL(dentry_update_name_case);
2723 static void swap_names(struct dentry *dentry, struct dentry *target)
2725 if (unlikely(dname_external(target))) {
2726 if (unlikely(dname_external(dentry))) {
2728 * Both external: swap the pointers
2730 swap(target->d_name.name, dentry->d_name.name);
2733 * dentry:internal, target:external. Steal target's
2734 * storage and make target internal.
2736 memcpy(target->d_iname, dentry->d_name.name,
2737 dentry->d_name.len + 1);
2738 dentry->d_name.name = target->d_name.name;
2739 target->d_name.name = target->d_iname;
2742 if (unlikely(dname_external(dentry))) {
2744 * dentry:external, target:internal. Give dentry's
2745 * storage to target and make dentry internal
2747 memcpy(dentry->d_iname, target->d_name.name,
2748 target->d_name.len + 1);
2749 target->d_name.name = dentry->d_name.name;
2750 dentry->d_name.name = dentry->d_iname;
2753 * Both are internal.
2756 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2757 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2758 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2759 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2760 swap(((long *) &dentry->d_iname)[i],
2761 ((long *) &target->d_iname)[i]);
2765 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2768 static void copy_name(struct dentry *dentry, struct dentry *target)
2770 struct external_name *old_name = NULL;
2771 if (unlikely(dname_external(dentry)))
2772 old_name = external_name(dentry);
2773 if (unlikely(dname_external(target))) {
2774 atomic_inc(&external_name(target)->u.count);
2775 dentry->d_name = target->d_name;
2777 memcpy(dentry->d_iname, target->d_name.name,
2778 target->d_name.len + 1);
2779 dentry->d_name.name = dentry->d_iname;
2780 dentry->d_name.hash_len = target->d_name.hash_len;
2782 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2783 kfree_rcu(old_name, u.head);
2786 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2789 * XXXX: do we really need to take target->d_lock?
2791 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2792 spin_lock(&target->d_parent->d_lock);
2794 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2795 spin_lock(&dentry->d_parent->d_lock);
2796 spin_lock_nested(&target->d_parent->d_lock,
2797 DENTRY_D_LOCK_NESTED);
2799 spin_lock(&target->d_parent->d_lock);
2800 spin_lock_nested(&dentry->d_parent->d_lock,
2801 DENTRY_D_LOCK_NESTED);
2804 if (target < dentry) {
2805 spin_lock_nested(&target->d_lock, 2);
2806 spin_lock_nested(&dentry->d_lock, 3);
2808 spin_lock_nested(&dentry->d_lock, 2);
2809 spin_lock_nested(&target->d_lock, 3);
2813 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2815 if (target->d_parent != dentry->d_parent)
2816 spin_unlock(&dentry->d_parent->d_lock);
2817 if (target->d_parent != target)
2818 spin_unlock(&target->d_parent->d_lock);
2819 spin_unlock(&target->d_lock);
2820 spin_unlock(&dentry->d_lock);
2824 * When switching names, the actual string doesn't strictly have to
2825 * be preserved in the target - because we're dropping the target
2826 * anyway. As such, we can just do a simple memcpy() to copy over
2827 * the new name before we switch, unless we are going to rehash
2828 * it. Note that if we *do* unhash the target, we are not allowed
2829 * to rehash it without giving it a new name/hash key - whether
2830 * we swap or overwrite the names here, resulting name won't match
2831 * the reality in filesystem; it's only there for d_path() purposes.
2832 * Note that all of this is happening under rename_lock, so the
2833 * any hash lookup seeing it in the middle of manipulations will
2834 * be discarded anyway. So we do not care what happens to the hash
2838 * __d_move - move a dentry
2839 * @dentry: entry to move
2840 * @target: new dentry
2841 * @exchange: exchange the two dentries
2843 * Update the dcache to reflect the move of a file name. Negative
2844 * dcache entries should not be moved in this way. Caller must hold
2845 * rename_lock, the i_mutex of the source and target directories,
2846 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2848 static void __d_move(struct dentry *dentry, struct dentry *target,
2851 struct inode *dir = NULL;
2853 if (!dentry->d_inode)
2854 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2856 BUG_ON(d_ancestor(dentry, target));
2857 BUG_ON(d_ancestor(target, dentry));
2859 dentry_lock_for_move(dentry, target);
2860 if (unlikely(d_in_lookup(target))) {
2861 dir = target->d_parent->d_inode;
2862 n = start_dir_add(dir);
2863 __d_lookup_done(target);
2866 write_seqcount_begin(&dentry->d_seq);
2867 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2869 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2872 * Move the dentry to the target hash queue. Don't bother checking
2873 * for the same hash queue because of how unlikely it is.
2876 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2879 * Unhash the target (d_delete() is not usable here). If exchanging
2880 * the two dentries, then rehash onto the other's hash queue.
2885 d_hash(dentry->d_parent, dentry->d_name.hash));
2888 /* Switch the names.. */
2890 swap_names(dentry, target);
2892 copy_name(dentry, target);
2894 /* ... and switch them in the tree */
2895 if (IS_ROOT(dentry)) {
2896 /* splicing a tree */
2897 dentry->d_flags |= DCACHE_RCUACCESS;
2898 dentry->d_parent = target->d_parent;
2899 target->d_parent = target;
2900 list_del_init(&target->d_child);
2901 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2903 /* swapping two dentries */
2904 swap(dentry->d_parent, target->d_parent);
2905 list_move(&target->d_child, &target->d_parent->d_subdirs);
2906 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2908 fsnotify_update_flags(target);
2909 fsnotify_update_flags(dentry);
2912 write_seqcount_end(&target->d_seq);
2913 write_seqcount_end(&dentry->d_seq);
2916 end_dir_add(dir, n);
2917 dentry_unlock_for_move(dentry, target);
2921 * d_move - move a dentry
2922 * @dentry: entry to move
2923 * @target: new dentry
2925 * Update the dcache to reflect the move of a file name. Negative
2926 * dcache entries should not be moved in this way. See the locking
2927 * requirements for __d_move.
2929 void d_move(struct dentry *dentry, struct dentry *target)
2931 write_seqlock(&rename_lock);
2932 __d_move(dentry, target, false);
2933 write_sequnlock(&rename_lock);
2935 EXPORT_SYMBOL(d_move);
2938 * d_exchange - exchange two dentries
2939 * @dentry1: first dentry
2940 * @dentry2: second dentry
2942 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2944 write_seqlock(&rename_lock);
2946 WARN_ON(!dentry1->d_inode);
2947 WARN_ON(!dentry2->d_inode);
2948 WARN_ON(IS_ROOT(dentry1));
2949 WARN_ON(IS_ROOT(dentry2));
2951 __d_move(dentry1, dentry2, true);
2953 write_sequnlock(&rename_lock);
2957 * d_ancestor - search for an ancestor
2958 * @p1: ancestor dentry
2961 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2962 * an ancestor of p2, else NULL.
2964 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2968 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2969 if (p->d_parent == p1)
2976 * This helper attempts to cope with remotely renamed directories
2978 * It assumes that the caller is already holding
2979 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2981 * Note: If ever the locking in lock_rename() changes, then please
2982 * remember to update this too...
2984 static int __d_unalias(struct inode *inode,
2985 struct dentry *dentry, struct dentry *alias)
2987 struct mutex *m1 = NULL;
2988 struct rw_semaphore *m2 = NULL;
2991 /* If alias and dentry share a parent, then no extra locks required */
2992 if (alias->d_parent == dentry->d_parent)
2995 /* See lock_rename() */
2996 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2998 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2999 if (!inode_trylock_shared(alias->d_parent->d_inode))
3001 m2 = &alias->d_parent->d_inode->i_rwsem;
3003 __d_move(alias, dentry, false);
3014 * d_splice_alias - splice a disconnected dentry into the tree if one exists
3015 * @inode: the inode which may have a disconnected dentry
3016 * @dentry: a negative dentry which we want to point to the inode.
3018 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3019 * place of the given dentry and return it, else simply d_add the inode
3020 * to the dentry and return NULL.
3022 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3023 * we should error out: directories can't have multiple aliases.
3025 * This is needed in the lookup routine of any filesystem that is exportable
3026 * (via knfsd) so that we can build dcache paths to directories effectively.
3028 * If a dentry was found and moved, then it is returned. Otherwise NULL
3029 * is returned. This matches the expected return value of ->lookup.
3031 * Cluster filesystems may call this function with a negative, hashed dentry.
3032 * In that case, we know that the inode will be a regular file, and also this
3033 * will only occur during atomic_open. So we need to check for the dentry
3034 * being already hashed only in the final case.
3036 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3039 return ERR_CAST(inode);
3041 BUG_ON(!d_unhashed(dentry));
3046 security_d_instantiate(dentry, inode);
3047 spin_lock(&inode->i_lock);
3048 if (S_ISDIR(inode->i_mode)) {
3049 struct dentry *new = __d_find_any_alias(inode);
3050 if (unlikely(new)) {
3051 /* The reference to new ensures it remains an alias */
3052 spin_unlock(&inode->i_lock);
3053 write_seqlock(&rename_lock);
3054 if (unlikely(d_ancestor(new, dentry))) {
3055 write_sequnlock(&rename_lock);
3057 new = ERR_PTR(-ELOOP);
3058 pr_warn_ratelimited(
3059 "VFS: Lookup of '%s' in %s %s"
3060 " would have caused loop\n",
3061 dentry->d_name.name,
3062 inode->i_sb->s_type->name,
3064 } else if (!IS_ROOT(new)) {
3065 int err = __d_unalias(inode, dentry, new);
3066 write_sequnlock(&rename_lock);
3072 __d_move(new, dentry, false);
3073 write_sequnlock(&rename_lock);
3080 __d_add(dentry, inode);
3083 EXPORT_SYMBOL(d_splice_alias);
3085 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
3089 return -ENAMETOOLONG;
3091 memcpy(*buffer, str, namelen);
3096 * prepend_name - prepend a pathname in front of current buffer pointer
3097 * @buffer: buffer pointer
3098 * @buflen: allocated length of the buffer
3099 * @name: name string and length qstr structure
3101 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3102 * make sure that either the old or the new name pointer and length are
3103 * fetched. However, there may be mismatch between length and pointer.
3104 * The length cannot be trusted, we need to copy it byte-by-byte until
3105 * the length is reached or a null byte is found. It also prepends "/" at
3106 * the beginning of the name. The sequence number check at the caller will
3107 * retry it again when a d_move() does happen. So any garbage in the buffer
3108 * due to mismatched pointer and length will be discarded.
3110 * Data dependency barrier is needed to make sure that we see that terminating
3111 * NUL. Alpha strikes again, film at 11...
3113 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
3115 const char *dname = ACCESS_ONCE(name->name);
3116 u32 dlen = ACCESS_ONCE(name->len);
3119 smp_read_barrier_depends();
3121 *buflen -= dlen + 1;
3123 return -ENAMETOOLONG;
3124 p = *buffer -= dlen + 1;
3136 * prepend_path - Prepend path string to a buffer
3137 * @path: the dentry/vfsmount to report
3138 * @root: root vfsmnt/dentry
3139 * @buffer: pointer to the end of the buffer
3140 * @buflen: pointer to buffer length
3142 * The function will first try to write out the pathname without taking any
3143 * lock other than the RCU read lock to make sure that dentries won't go away.
3144 * It only checks the sequence number of the global rename_lock as any change
3145 * in the dentry's d_seq will be preceded by changes in the rename_lock
3146 * sequence number. If the sequence number had been changed, it will restart
3147 * the whole pathname back-tracing sequence again by taking the rename_lock.
3148 * In this case, there is no need to take the RCU read lock as the recursive
3149 * parent pointer references will keep the dentry chain alive as long as no
3150 * rename operation is performed.
3152 static int prepend_path(const struct path *path,
3153 const struct path *root,
3154 char **buffer, int *buflen)
3156 struct dentry *dentry;
3157 struct vfsmount *vfsmnt;
3160 unsigned seq, m_seq = 0;
3166 read_seqbegin_or_lock(&mount_lock, &m_seq);
3173 dentry = path->dentry;
3175 mnt = real_mount(vfsmnt);
3176 read_seqbegin_or_lock(&rename_lock, &seq);
3177 while (dentry != root->dentry || vfsmnt != root->mnt) {
3178 struct dentry * parent;
3180 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
3181 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
3183 if (dentry != vfsmnt->mnt_root) {
3190 if (mnt != parent) {
3191 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3197 error = is_mounted(vfsmnt) ? 1 : 2;
3200 parent = dentry->d_parent;
3202 error = prepend_name(&bptr, &blen, &dentry->d_name);
3210 if (need_seqretry(&rename_lock, seq)) {
3214 done_seqretry(&rename_lock, seq);
3218 if (need_seqretry(&mount_lock, m_seq)) {
3222 done_seqretry(&mount_lock, m_seq);
3224 if (error >= 0 && bptr == *buffer) {
3226 error = -ENAMETOOLONG;
3236 * __d_path - return the path of a dentry
3237 * @path: the dentry/vfsmount to report
3238 * @root: root vfsmnt/dentry
3239 * @buf: buffer to return value in
3240 * @buflen: buffer length
3242 * Convert a dentry into an ASCII path name.
3244 * Returns a pointer into the buffer or an error code if the
3245 * path was too long.
3247 * "buflen" should be positive.
3249 * If the path is not reachable from the supplied root, return %NULL.
3251 char *__d_path(const struct path *path,
3252 const struct path *root,
3253 char *buf, int buflen)
3255 char *res = buf + buflen;
3258 prepend(&res, &buflen, "\0", 1);
3259 error = prepend_path(path, root, &res, &buflen);
3262 return ERR_PTR(error);
3268 char *d_absolute_path(const struct path *path,
3269 char *buf, int buflen)
3271 struct path root = {};
3272 char *res = buf + buflen;
3275 prepend(&res, &buflen, "\0", 1);
3276 error = prepend_path(path, &root, &res, &buflen);
3281 return ERR_PTR(error);
3286 * same as __d_path but appends "(deleted)" for unlinked files.
3288 static int path_with_deleted(const struct path *path,
3289 const struct path *root,
3290 char **buf, int *buflen)
3292 prepend(buf, buflen, "\0", 1);
3293 if (d_unlinked(path->dentry)) {
3294 int error = prepend(buf, buflen, " (deleted)", 10);
3299 return prepend_path(path, root, buf, buflen);
3302 static int prepend_unreachable(char **buffer, int *buflen)
3304 return prepend(buffer, buflen, "(unreachable)", 13);
3307 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3312 seq = read_seqcount_begin(&fs->seq);
3314 } while (read_seqcount_retry(&fs->seq, seq));
3318 * d_path - return the path of a dentry
3319 * @path: path to report
3320 * @buf: buffer to return value in
3321 * @buflen: buffer length
3323 * Convert a dentry into an ASCII path name. If the entry has been deleted
3324 * the string " (deleted)" is appended. Note that this is ambiguous.
3326 * Returns a pointer into the buffer or an error code if the path was
3327 * too long. Note: Callers should use the returned pointer, not the passed
3328 * in buffer, to use the name! The implementation often starts at an offset
3329 * into the buffer, and may leave 0 bytes at the start.
3331 * "buflen" should be positive.
3333 char *d_path(const struct path *path, char *buf, int buflen)
3335 char *res = buf + buflen;
3340 * We have various synthetic filesystems that never get mounted. On
3341 * these filesystems dentries are never used for lookup purposes, and
3342 * thus don't need to be hashed. They also don't need a name until a
3343 * user wants to identify the object in /proc/pid/fd/. The little hack
3344 * below allows us to generate a name for these objects on demand:
3346 * Some pseudo inodes are mountable. When they are mounted
3347 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3348 * and instead have d_path return the mounted path.
3350 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3351 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3352 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3355 get_fs_root_rcu(current->fs, &root);
3356 error = path_with_deleted(path, &root, &res, &buflen);
3360 res = ERR_PTR(error);
3363 EXPORT_SYMBOL(d_path);
3366 * Helper function for dentry_operations.d_dname() members
3368 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3369 const char *fmt, ...)
3375 va_start(args, fmt);
3376 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3379 if (sz > sizeof(temp) || sz > buflen)
3380 return ERR_PTR(-ENAMETOOLONG);
3382 buffer += buflen - sz;
3383 return memcpy(buffer, temp, sz);
3386 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3388 char *end = buffer + buflen;
3389 /* these dentries are never renamed, so d_lock is not needed */
3390 if (prepend(&end, &buflen, " (deleted)", 11) ||
3391 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3392 prepend(&end, &buflen, "/", 1))
3393 end = ERR_PTR(-ENAMETOOLONG);
3396 EXPORT_SYMBOL(simple_dname);
3399 * Write full pathname from the root of the filesystem into the buffer.
3401 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3403 struct dentry *dentry;
3416 prepend(&end, &len, "\0", 1);
3420 read_seqbegin_or_lock(&rename_lock, &seq);
3421 while (!IS_ROOT(dentry)) {
3422 struct dentry *parent = dentry->d_parent;
3425 error = prepend_name(&end, &len, &dentry->d_name);
3434 if (need_seqretry(&rename_lock, seq)) {
3438 done_seqretry(&rename_lock, seq);
3443 return ERR_PTR(-ENAMETOOLONG);
3446 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3448 return __dentry_path(dentry, buf, buflen);
3450 EXPORT_SYMBOL(dentry_path_raw);
3452 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3457 if (d_unlinked(dentry)) {
3459 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3463 retval = __dentry_path(dentry, buf, buflen);
3464 if (!IS_ERR(retval) && p)
3465 *p = '/'; /* restore '/' overriden with '\0' */
3468 return ERR_PTR(-ENAMETOOLONG);
3471 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3477 seq = read_seqcount_begin(&fs->seq);
3480 } while (read_seqcount_retry(&fs->seq, seq));
3484 * NOTE! The user-level library version returns a
3485 * character pointer. The kernel system call just
3486 * returns the length of the buffer filled (which
3487 * includes the ending '\0' character), or a negative
3488 * error value. So libc would do something like
3490 * char *getcwd(char * buf, size_t size)
3494 * retval = sys_getcwd(buf, size);
3501 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3504 struct path pwd, root;
3505 char *page = __getname();
3511 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3514 if (!d_unlinked(pwd.dentry)) {
3516 char *cwd = page + PATH_MAX;
3517 int buflen = PATH_MAX;
3519 prepend(&cwd, &buflen, "\0", 1);
3520 error = prepend_path(&pwd, &root, &cwd, &buflen);
3526 /* Unreachable from current root */
3528 error = prepend_unreachable(&cwd, &buflen);
3534 len = PATH_MAX + page - cwd;
3537 if (copy_to_user(buf, cwd, len))
3550 * Test whether new_dentry is a subdirectory of old_dentry.
3552 * Trivially implemented using the dcache structure
3556 * is_subdir - is new dentry a subdirectory of old_dentry
3557 * @new_dentry: new dentry
3558 * @old_dentry: old dentry
3560 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3561 * Returns false otherwise.
3562 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3565 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3570 if (new_dentry == old_dentry)
3574 /* for restarting inner loop in case of seq retry */
3575 seq = read_seqbegin(&rename_lock);
3577 * Need rcu_readlock to protect against the d_parent trashing
3581 if (d_ancestor(old_dentry, new_dentry))
3586 } while (read_seqretry(&rename_lock, seq));
3591 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3593 struct dentry *root = data;
3594 if (dentry != root) {
3595 if (d_unhashed(dentry) || !dentry->d_inode)
3598 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3599 dentry->d_flags |= DCACHE_GENOCIDE;
3600 dentry->d_lockref.count--;
3603 return D_WALK_CONTINUE;
3606 void d_genocide(struct dentry *parent)
3608 d_walk(parent, parent, d_genocide_kill, NULL);
3611 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3613 inode_dec_link_count(inode);
3614 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3615 !hlist_unhashed(&dentry->d_u.d_alias) ||
3616 !d_unlinked(dentry));
3617 spin_lock(&dentry->d_parent->d_lock);
3618 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3619 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3620 (unsigned long long)inode->i_ino);
3621 spin_unlock(&dentry->d_lock);
3622 spin_unlock(&dentry->d_parent->d_lock);
3623 d_instantiate(dentry, inode);
3625 EXPORT_SYMBOL(d_tmpfile);
3627 static __initdata unsigned long dhash_entries;
3628 static int __init set_dhash_entries(char *str)
3632 dhash_entries = simple_strtoul(str, &str, 0);
3635 __setup("dhash_entries=", set_dhash_entries);
3637 static void __init dcache_init_early(void)
3641 /* If hashes are distributed across NUMA nodes, defer
3642 * hash allocation until vmalloc space is available.
3648 alloc_large_system_hash("Dentry cache",
3649 sizeof(struct hlist_bl_head),
3658 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3659 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3662 static void __init dcache_init(void)
3667 * A constructor could be added for stable state like the lists,
3668 * but it is probably not worth it because of the cache nature
3671 dentry_cache = KMEM_CACHE(dentry,
3672 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
3674 /* Hash may have been set up in dcache_init_early */
3679 alloc_large_system_hash("Dentry cache",
3680 sizeof(struct hlist_bl_head),
3689 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3690 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3693 /* SLAB cache for __getname() consumers */
3694 struct kmem_cache *names_cachep __read_mostly;
3695 EXPORT_SYMBOL(names_cachep);
3697 EXPORT_SYMBOL(d_genocide);
3699 void __init vfs_caches_init_early(void)
3701 dcache_init_early();
3705 void __init vfs_caches_init(void)
3707 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3708 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3713 files_maxfiles_init();