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 <linux/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;
93 const struct qstr empty_name = QSTR_INIT("", 0);
94 EXPORT_SYMBOL(empty_name);
95 const struct qstr slash_name = QSTR_INIT("/", 1);
96 EXPORT_SYMBOL(slash_name);
99 * This is the single most critical data structure when it comes
100 * to the dcache: the hashtable for lookups. Somebody should try
101 * to make this good - I've just made it work.
103 * This hash-function tries to avoid losing too many bits of hash
104 * information, yet avoid using a prime hash-size or similar.
107 static unsigned int d_hash_mask __read_mostly;
108 static unsigned int d_hash_shift __read_mostly;
110 static struct hlist_bl_head *dentry_hashtable __read_mostly;
112 static inline struct hlist_bl_head *d_hash(unsigned int hash)
114 return dentry_hashtable + (hash >> (32 - d_hash_shift));
117 #define IN_LOOKUP_SHIFT 10
118 static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
120 static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
123 hash += (unsigned long) parent / L1_CACHE_BYTES;
124 return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
128 /* Statistics gathering. */
129 struct dentry_stat_t dentry_stat = {
133 static DEFINE_PER_CPU(long, nr_dentry);
134 static DEFINE_PER_CPU(long, nr_dentry_unused);
136 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
139 * Here we resort to our own counters instead of using generic per-cpu counters
140 * for consistency with what the vfs inode code does. We are expected to harvest
141 * better code and performance by having our own specialized counters.
143 * Please note that the loop is done over all possible CPUs, not over all online
144 * CPUs. The reason for this is that we don't want to play games with CPUs going
145 * on and off. If one of them goes off, we will just keep their counters.
147 * glommer: See cffbc8a for details, and if you ever intend to change this,
148 * please update all vfs counters to match.
150 static long get_nr_dentry(void)
154 for_each_possible_cpu(i)
155 sum += per_cpu(nr_dentry, i);
156 return sum < 0 ? 0 : sum;
159 static long get_nr_dentry_unused(void)
163 for_each_possible_cpu(i)
164 sum += per_cpu(nr_dentry_unused, i);
165 return sum < 0 ? 0 : sum;
168 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
169 size_t *lenp, loff_t *ppos)
171 dentry_stat.nr_dentry = get_nr_dentry();
172 dentry_stat.nr_unused = get_nr_dentry_unused();
173 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
178 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
179 * The strings are both count bytes long, and count is non-zero.
181 #ifdef CONFIG_DCACHE_WORD_ACCESS
183 #include <asm/word-at-a-time.h>
185 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
186 * aligned allocation for this particular component. We don't
187 * strictly need the load_unaligned_zeropad() safety, but it
188 * doesn't hurt either.
190 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
191 * need the careful unaligned handling.
193 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
195 unsigned long a,b,mask;
198 a = *(unsigned long *)cs;
199 b = load_unaligned_zeropad(ct);
200 if (tcount < sizeof(unsigned long))
202 if (unlikely(a != b))
204 cs += sizeof(unsigned long);
205 ct += sizeof(unsigned long);
206 tcount -= sizeof(unsigned long);
210 mask = bytemask_from_count(tcount);
211 return unlikely(!!((a ^ b) & mask));
216 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
230 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
233 * Be careful about RCU walk racing with rename:
234 * use 'READ_ONCE' to fetch the name pointer.
236 * NOTE! Even if a rename will mean that the length
237 * was not loaded atomically, we don't care. The
238 * RCU walk will check the sequence count eventually,
239 * and catch it. And we won't overrun the buffer,
240 * because we're reading the name pointer atomically,
241 * and a dentry name is guaranteed to be properly
242 * terminated with a NUL byte.
244 * End result: even if 'len' is wrong, we'll exit
245 * early because the data cannot match (there can
246 * be no NUL in the ct/tcount data)
248 const unsigned char *cs = READ_ONCE(dentry->d_name.name);
250 return dentry_string_cmp(cs, ct, tcount);
253 struct external_name {
256 struct rcu_head head;
258 unsigned char name[];
261 static inline struct external_name *external_name(struct dentry *dentry)
263 return container_of(dentry->d_name.name, struct external_name, name[0]);
266 static void __d_free(struct rcu_head *head)
268 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
270 kmem_cache_free(dentry_cache, dentry);
273 static void __d_free_external(struct rcu_head *head)
275 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
276 kfree(external_name(dentry));
277 kmem_cache_free(dentry_cache, dentry);
280 static inline int dname_external(const struct dentry *dentry)
282 return dentry->d_name.name != dentry->d_iname;
285 void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
287 spin_lock(&dentry->d_lock);
288 if (unlikely(dname_external(dentry))) {
289 struct external_name *p = external_name(dentry);
290 atomic_inc(&p->u.count);
291 spin_unlock(&dentry->d_lock);
292 name->name = p->name;
294 memcpy(name->inline_name, dentry->d_iname,
295 dentry->d_name.len + 1);
296 spin_unlock(&dentry->d_lock);
297 name->name = name->inline_name;
300 EXPORT_SYMBOL(take_dentry_name_snapshot);
302 void release_dentry_name_snapshot(struct name_snapshot *name)
304 if (unlikely(name->name != name->inline_name)) {
305 struct external_name *p;
306 p = container_of(name->name, struct external_name, name[0]);
307 if (unlikely(atomic_dec_and_test(&p->u.count)))
308 kfree_rcu(p, u.head);
311 EXPORT_SYMBOL(release_dentry_name_snapshot);
313 static inline void __d_set_inode_and_type(struct dentry *dentry,
319 dentry->d_inode = inode;
320 flags = READ_ONCE(dentry->d_flags);
321 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
323 WRITE_ONCE(dentry->d_flags, flags);
326 static inline void __d_clear_type_and_inode(struct dentry *dentry)
328 unsigned flags = READ_ONCE(dentry->d_flags);
330 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
331 WRITE_ONCE(dentry->d_flags, flags);
332 dentry->d_inode = NULL;
335 static void dentry_free(struct dentry *dentry)
337 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
338 if (unlikely(dname_external(dentry))) {
339 struct external_name *p = external_name(dentry);
340 if (likely(atomic_dec_and_test(&p->u.count))) {
341 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
345 /* if dentry was never visible to RCU, immediate free is OK */
346 if (!(dentry->d_flags & DCACHE_RCUACCESS))
347 __d_free(&dentry->d_u.d_rcu);
349 call_rcu(&dentry->d_u.d_rcu, __d_free);
353 * Release the dentry's inode, using the filesystem
354 * d_iput() operation if defined.
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)))
452 else if (unlikely(!(dentry->d_flags & DCACHE_REFERENCED)))
453 dentry->d_flags |= DCACHE_REFERENCED;
457 * d_drop - drop a dentry
458 * @dentry: dentry to drop
460 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
461 * be found through a VFS lookup any more. Note that this is different from
462 * deleting the dentry - d_delete will try to mark the dentry negative if
463 * possible, giving a successful _negative_ lookup, while d_drop will
464 * just make the cache lookup fail.
466 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
467 * reason (NFS timeouts or autofs deletes).
469 * __d_drop requires dentry->d_lock
470 * ___d_drop doesn't mark dentry as "unhashed"
471 * (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
473 static void ___d_drop(struct dentry *dentry)
475 if (!d_unhashed(dentry)) {
476 struct hlist_bl_head *b;
478 * Hashed dentries are normally on the dentry hashtable,
479 * with the exception of those newly allocated by
480 * d_obtain_alias, which are always IS_ROOT:
482 if (unlikely(IS_ROOT(dentry)))
483 b = &dentry->d_sb->s_anon;
485 b = d_hash(dentry->d_name.hash);
488 __hlist_bl_del(&dentry->d_hash);
490 /* After this call, in-progress rcu-walk path lookup will fail. */
491 write_seqcount_invalidate(&dentry->d_seq);
495 void __d_drop(struct dentry *dentry)
498 dentry->d_hash.pprev = NULL;
500 EXPORT_SYMBOL(__d_drop);
502 void d_drop(struct dentry *dentry)
504 spin_lock(&dentry->d_lock);
506 spin_unlock(&dentry->d_lock);
508 EXPORT_SYMBOL(d_drop);
510 static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
514 * Inform d_walk() and shrink_dentry_list() that we are no longer
515 * attached to the dentry tree
517 dentry->d_flags |= DCACHE_DENTRY_KILLED;
518 if (unlikely(list_empty(&dentry->d_child)))
520 __list_del_entry(&dentry->d_child);
522 * Cursors can move around the list of children. While we'd been
523 * a normal list member, it didn't matter - ->d_child.next would've
524 * been updated. However, from now on it won't be and for the
525 * things like d_walk() it might end up with a nasty surprise.
526 * Normally d_walk() doesn't care about cursors moving around -
527 * ->d_lock on parent prevents that and since a cursor has no children
528 * of its own, we get through it without ever unlocking the parent.
529 * There is one exception, though - if we ascend from a child that
530 * gets killed as soon as we unlock it, the next sibling is found
531 * using the value left in its ->d_child.next. And if _that_
532 * pointed to a cursor, and cursor got moved (e.g. by lseek())
533 * before d_walk() regains parent->d_lock, we'll end up skipping
534 * everything the cursor had been moved past.
536 * Solution: make sure that the pointer left behind in ->d_child.next
537 * points to something that won't be moving around. I.e. skip the
540 while (dentry->d_child.next != &parent->d_subdirs) {
541 next = list_entry(dentry->d_child.next, struct dentry, d_child);
542 if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
544 dentry->d_child.next = next->d_child.next;
548 static void __dentry_kill(struct dentry *dentry)
550 struct dentry *parent = NULL;
551 bool can_free = true;
552 if (!IS_ROOT(dentry))
553 parent = dentry->d_parent;
556 * The dentry is now unrecoverably dead to the world.
558 lockref_mark_dead(&dentry->d_lockref);
561 * inform the fs via d_prune that this dentry is about to be
562 * unhashed and destroyed.
564 if (dentry->d_flags & DCACHE_OP_PRUNE)
565 dentry->d_op->d_prune(dentry);
567 if (dentry->d_flags & DCACHE_LRU_LIST) {
568 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
571 /* if it was on the hash then remove it */
573 dentry_unlist(dentry, parent);
575 spin_unlock(&parent->d_lock);
577 dentry_unlink_inode(dentry);
579 spin_unlock(&dentry->d_lock);
580 this_cpu_dec(nr_dentry);
581 if (dentry->d_op && dentry->d_op->d_release)
582 dentry->d_op->d_release(dentry);
584 spin_lock(&dentry->d_lock);
585 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
586 dentry->d_flags |= DCACHE_MAY_FREE;
589 spin_unlock(&dentry->d_lock);
590 if (likely(can_free))
595 * Finish off a dentry we've decided to kill.
596 * dentry->d_lock must be held, returns with it unlocked.
597 * If ref is non-zero, then decrement the refcount too.
598 * Returns dentry requiring refcount drop, or NULL if we're done.
600 static struct dentry *dentry_kill(struct dentry *dentry)
601 __releases(dentry->d_lock)
603 struct inode *inode = dentry->d_inode;
604 struct dentry *parent = NULL;
606 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
609 if (!IS_ROOT(dentry)) {
610 parent = dentry->d_parent;
611 if (unlikely(!spin_trylock(&parent->d_lock))) {
613 spin_unlock(&inode->i_lock);
618 __dentry_kill(dentry);
622 spin_unlock(&dentry->d_lock);
623 return dentry; /* try again with same dentry */
626 static inline struct dentry *lock_parent(struct dentry *dentry)
628 struct dentry *parent = dentry->d_parent;
631 if (unlikely(dentry->d_lockref.count < 0))
633 if (likely(spin_trylock(&parent->d_lock)))
636 spin_unlock(&dentry->d_lock);
638 parent = ACCESS_ONCE(dentry->d_parent);
639 spin_lock(&parent->d_lock);
641 * We can't blindly lock dentry until we are sure
642 * that we won't violate the locking order.
643 * Any changes of dentry->d_parent must have
644 * been done with parent->d_lock held, so
645 * spin_lock() above is enough of a barrier
646 * for checking if it's still our child.
648 if (unlikely(parent != dentry->d_parent)) {
649 spin_unlock(&parent->d_lock);
652 if (parent != dentry) {
653 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
654 if (unlikely(dentry->d_lockref.count < 0)) {
655 spin_unlock(&parent->d_lock);
666 * Try to do a lockless dput(), and return whether that was successful.
668 * If unsuccessful, we return false, having already taken the dentry lock.
670 * The caller needs to hold the RCU read lock, so that the dentry is
671 * guaranteed to stay around even if the refcount goes down to zero!
673 static inline bool fast_dput(struct dentry *dentry)
676 unsigned int d_flags;
679 * If we have a d_op->d_delete() operation, we sould not
680 * let the dentry count go to zero, so use "put_or_lock".
682 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
683 return lockref_put_or_lock(&dentry->d_lockref);
686 * .. otherwise, we can try to just decrement the
687 * lockref optimistically.
689 ret = lockref_put_return(&dentry->d_lockref);
692 * If the lockref_put_return() failed due to the lock being held
693 * by somebody else, the fast path has failed. We will need to
694 * get the lock, and then check the count again.
696 if (unlikely(ret < 0)) {
697 spin_lock(&dentry->d_lock);
698 if (dentry->d_lockref.count > 1) {
699 dentry->d_lockref.count--;
700 spin_unlock(&dentry->d_lock);
707 * If we weren't the last ref, we're done.
713 * Careful, careful. The reference count went down
714 * to zero, but we don't hold the dentry lock, so
715 * somebody else could get it again, and do another
716 * dput(), and we need to not race with that.
718 * However, there is a very special and common case
719 * where we don't care, because there is nothing to
720 * do: the dentry is still hashed, it does not have
721 * a 'delete' op, and it's referenced and already on
724 * NOTE! Since we aren't locked, these values are
725 * not "stable". However, it is sufficient that at
726 * some point after we dropped the reference the
727 * dentry was hashed and the flags had the proper
728 * value. Other dentry users may have re-gotten
729 * a reference to the dentry and change that, but
730 * our work is done - we can leave the dentry
731 * around with a zero refcount.
734 d_flags = ACCESS_ONCE(dentry->d_flags);
735 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
737 /* Nothing to do? Dropping the reference was all we needed? */
738 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
742 * Not the fast normal case? Get the lock. We've already decremented
743 * the refcount, but we'll need to re-check the situation after
746 spin_lock(&dentry->d_lock);
749 * Did somebody else grab a reference to it in the meantime, and
750 * we're no longer the last user after all? Alternatively, somebody
751 * else could have killed it and marked it dead. Either way, we
752 * don't need to do anything else.
754 if (dentry->d_lockref.count) {
755 spin_unlock(&dentry->d_lock);
760 * Re-get the reference we optimistically dropped. We hold the
761 * lock, and we just tested that it was zero, so we can just
764 dentry->d_lockref.count = 1;
772 * This is complicated by the fact that we do not want to put
773 * dentries that are no longer on any hash chain on the unused
774 * list: we'd much rather just get rid of them immediately.
776 * However, that implies that we have to traverse the dentry
777 * tree upwards to the parents which might _also_ now be
778 * scheduled for deletion (it may have been only waiting for
779 * its last child to go away).
781 * This tail recursion is done by hand as we don't want to depend
782 * on the compiler to always get this right (gcc generally doesn't).
783 * Real recursion would eat up our stack space.
787 * dput - release a dentry
788 * @dentry: dentry to release
790 * Release a dentry. This will drop the usage count and if appropriate
791 * call the dentry unlink method as well as removing it from the queues and
792 * releasing its resources. If the parent dentries were scheduled for release
793 * they too may now get deleted.
795 void dput(struct dentry *dentry)
797 if (unlikely(!dentry))
804 if (likely(fast_dput(dentry))) {
809 /* Slow case: now with the dentry lock held */
812 WARN_ON(d_in_lookup(dentry));
814 /* Unreachable? Get rid of it */
815 if (unlikely(d_unhashed(dentry)))
818 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
821 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
822 if (dentry->d_op->d_delete(dentry))
826 dentry_lru_add(dentry);
828 dentry->d_lockref.count--;
829 spin_unlock(&dentry->d_lock);
833 dentry = dentry_kill(dentry);
842 /* This must be called with d_lock held */
843 static inline void __dget_dlock(struct dentry *dentry)
845 dentry->d_lockref.count++;
848 static inline void __dget(struct dentry *dentry)
850 lockref_get(&dentry->d_lockref);
853 struct dentry *dget_parent(struct dentry *dentry)
859 * Do optimistic parent lookup without any
863 ret = ACCESS_ONCE(dentry->d_parent);
864 gotref = lockref_get_not_zero(&ret->d_lockref);
866 if (likely(gotref)) {
867 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
874 * Don't need rcu_dereference because we re-check it was correct under
878 ret = dentry->d_parent;
879 spin_lock(&ret->d_lock);
880 if (unlikely(ret != dentry->d_parent)) {
881 spin_unlock(&ret->d_lock);
886 BUG_ON(!ret->d_lockref.count);
887 ret->d_lockref.count++;
888 spin_unlock(&ret->d_lock);
891 EXPORT_SYMBOL(dget_parent);
894 * d_find_alias - grab a hashed alias of inode
895 * @inode: inode in question
897 * If inode has a hashed alias, or is a directory and has any alias,
898 * acquire the reference to alias and return it. Otherwise return NULL.
899 * Notice that if inode is a directory there can be only one alias and
900 * it can be unhashed only if it has no children, or if it is the root
901 * of a filesystem, or if the directory was renamed and d_revalidate
902 * was the first vfs operation to notice.
904 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
905 * any other hashed alias over that one.
907 static struct dentry *__d_find_alias(struct inode *inode)
909 struct dentry *alias, *discon_alias;
913 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
914 spin_lock(&alias->d_lock);
915 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
916 if (IS_ROOT(alias) &&
917 (alias->d_flags & DCACHE_DISCONNECTED)) {
918 discon_alias = alias;
921 spin_unlock(&alias->d_lock);
925 spin_unlock(&alias->d_lock);
928 alias = discon_alias;
929 spin_lock(&alias->d_lock);
930 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
932 spin_unlock(&alias->d_lock);
935 spin_unlock(&alias->d_lock);
941 struct dentry *d_find_alias(struct inode *inode)
943 struct dentry *de = NULL;
945 if (!hlist_empty(&inode->i_dentry)) {
946 spin_lock(&inode->i_lock);
947 de = __d_find_alias(inode);
948 spin_unlock(&inode->i_lock);
952 EXPORT_SYMBOL(d_find_alias);
955 * Try to kill dentries associated with this inode.
956 * WARNING: you must own a reference to inode.
958 void d_prune_aliases(struct inode *inode)
960 struct dentry *dentry;
962 spin_lock(&inode->i_lock);
963 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
964 spin_lock(&dentry->d_lock);
965 if (!dentry->d_lockref.count) {
966 struct dentry *parent = lock_parent(dentry);
967 if (likely(!dentry->d_lockref.count)) {
968 __dentry_kill(dentry);
973 spin_unlock(&parent->d_lock);
975 spin_unlock(&dentry->d_lock);
977 spin_unlock(&inode->i_lock);
979 EXPORT_SYMBOL(d_prune_aliases);
981 static void shrink_dentry_list(struct list_head *list)
983 struct dentry *dentry, *parent;
985 while (!list_empty(list)) {
987 dentry = list_entry(list->prev, struct dentry, d_lru);
988 spin_lock(&dentry->d_lock);
989 parent = lock_parent(dentry);
992 * The dispose list is isolated and dentries are not accounted
993 * to the LRU here, so we can simply remove it from the list
994 * here regardless of whether it is referenced or not.
996 d_shrink_del(dentry);
999 * We found an inuse dentry which was not removed from
1000 * the LRU because of laziness during lookup. Do not free it.
1002 if (dentry->d_lockref.count > 0) {
1003 spin_unlock(&dentry->d_lock);
1005 spin_unlock(&parent->d_lock);
1010 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1011 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1012 spin_unlock(&dentry->d_lock);
1014 spin_unlock(&parent->d_lock);
1016 dentry_free(dentry);
1020 inode = dentry->d_inode;
1021 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1022 d_shrink_add(dentry, list);
1023 spin_unlock(&dentry->d_lock);
1025 spin_unlock(&parent->d_lock);
1029 __dentry_kill(dentry);
1032 * We need to prune ancestors too. This is necessary to prevent
1033 * quadratic behavior of shrink_dcache_parent(), but is also
1034 * expected to be beneficial in reducing dentry cache
1038 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1039 parent = lock_parent(dentry);
1040 if (dentry->d_lockref.count != 1) {
1041 dentry->d_lockref.count--;
1042 spin_unlock(&dentry->d_lock);
1044 spin_unlock(&parent->d_lock);
1047 inode = dentry->d_inode; /* can't be NULL */
1048 if (unlikely(!spin_trylock(&inode->i_lock))) {
1049 spin_unlock(&dentry->d_lock);
1051 spin_unlock(&parent->d_lock);
1055 __dentry_kill(dentry);
1061 static enum lru_status dentry_lru_isolate(struct list_head *item,
1062 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1064 struct list_head *freeable = arg;
1065 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1069 * we are inverting the lru lock/dentry->d_lock here,
1070 * so use a trylock. If we fail to get the lock, just skip
1073 if (!spin_trylock(&dentry->d_lock))
1077 * Referenced dentries are still in use. If they have active
1078 * counts, just remove them from the LRU. Otherwise give them
1079 * another pass through the LRU.
1081 if (dentry->d_lockref.count) {
1082 d_lru_isolate(lru, dentry);
1083 spin_unlock(&dentry->d_lock);
1087 if (dentry->d_flags & DCACHE_REFERENCED) {
1088 dentry->d_flags &= ~DCACHE_REFERENCED;
1089 spin_unlock(&dentry->d_lock);
1092 * The list move itself will be made by the common LRU code. At
1093 * this point, we've dropped the dentry->d_lock but keep the
1094 * lru lock. This is safe to do, since every list movement is
1095 * protected by the lru lock even if both locks are held.
1097 * This is guaranteed by the fact that all LRU management
1098 * functions are intermediated by the LRU API calls like
1099 * list_lru_add and list_lru_del. List movement in this file
1100 * only ever occur through this functions or through callbacks
1101 * like this one, that are called from the LRU API.
1103 * The only exceptions to this are functions like
1104 * shrink_dentry_list, and code that first checks for the
1105 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1106 * operating only with stack provided lists after they are
1107 * properly isolated from the main list. It is thus, always a
1113 d_lru_shrink_move(lru, dentry, freeable);
1114 spin_unlock(&dentry->d_lock);
1120 * prune_dcache_sb - shrink the dcache
1122 * @sc: shrink control, passed to list_lru_shrink_walk()
1124 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1125 * is done when we need more memory and called from the superblock shrinker
1128 * This function may fail to free any resources if all the dentries are in
1131 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1136 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1137 dentry_lru_isolate, &dispose);
1138 shrink_dentry_list(&dispose);
1142 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1143 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1145 struct list_head *freeable = arg;
1146 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1149 * we are inverting the lru lock/dentry->d_lock here,
1150 * so use a trylock. If we fail to get the lock, just skip
1153 if (!spin_trylock(&dentry->d_lock))
1156 d_lru_shrink_move(lru, dentry, freeable);
1157 spin_unlock(&dentry->d_lock);
1164 * shrink_dcache_sb - shrink dcache for a superblock
1167 * Shrink the dcache for the specified super block. This is used to free
1168 * the dcache before unmounting a file system.
1170 void shrink_dcache_sb(struct super_block *sb)
1177 freed = list_lru_walk(&sb->s_dentry_lru,
1178 dentry_lru_isolate_shrink, &dispose, 1024);
1180 this_cpu_sub(nr_dentry_unused, freed);
1181 shrink_dentry_list(&dispose);
1183 } while (list_lru_count(&sb->s_dentry_lru) > 0);
1185 EXPORT_SYMBOL(shrink_dcache_sb);
1188 * enum d_walk_ret - action to talke during tree walk
1189 * @D_WALK_CONTINUE: contrinue walk
1190 * @D_WALK_QUIT: quit walk
1191 * @D_WALK_NORETRY: quit when retry is needed
1192 * @D_WALK_SKIP: skip this dentry and its children
1202 * d_walk - walk the dentry tree
1203 * @parent: start of walk
1204 * @data: data passed to @enter() and @finish()
1205 * @enter: callback when first entering the dentry
1206 * @finish: callback when successfully finished the walk
1208 * The @enter() and @finish() callbacks are called with d_lock held.
1210 static void d_walk(struct dentry *parent, void *data,
1211 enum d_walk_ret (*enter)(void *, struct dentry *),
1212 void (*finish)(void *))
1214 struct dentry *this_parent;
1215 struct list_head *next;
1217 enum d_walk_ret ret;
1221 read_seqbegin_or_lock(&rename_lock, &seq);
1222 this_parent = parent;
1223 spin_lock(&this_parent->d_lock);
1225 ret = enter(data, this_parent);
1227 case D_WALK_CONTINUE:
1232 case D_WALK_NORETRY:
1237 next = this_parent->d_subdirs.next;
1239 while (next != &this_parent->d_subdirs) {
1240 struct list_head *tmp = next;
1241 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1244 if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1247 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1249 ret = enter(data, dentry);
1251 case D_WALK_CONTINUE:
1254 spin_unlock(&dentry->d_lock);
1256 case D_WALK_NORETRY:
1260 spin_unlock(&dentry->d_lock);
1264 if (!list_empty(&dentry->d_subdirs)) {
1265 spin_unlock(&this_parent->d_lock);
1266 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1267 this_parent = dentry;
1268 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1271 spin_unlock(&dentry->d_lock);
1274 * All done at this level ... ascend and resume the search.
1278 if (this_parent != parent) {
1279 struct dentry *child = this_parent;
1280 this_parent = child->d_parent;
1282 spin_unlock(&child->d_lock);
1283 spin_lock(&this_parent->d_lock);
1285 /* might go back up the wrong parent if we have had a rename. */
1286 if (need_seqretry(&rename_lock, seq))
1288 /* go into the first sibling still alive */
1290 next = child->d_child.next;
1291 if (next == &this_parent->d_subdirs)
1293 child = list_entry(next, struct dentry, d_child);
1294 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1298 if (need_seqretry(&rename_lock, seq))
1305 spin_unlock(&this_parent->d_lock);
1306 done_seqretry(&rename_lock, seq);
1310 spin_unlock(&this_parent->d_lock);
1319 struct check_mount {
1320 struct vfsmount *mnt;
1321 unsigned int mounted;
1324 static enum d_walk_ret path_check_mount(void *data, struct dentry *dentry)
1326 struct check_mount *info = data;
1327 struct path path = { .mnt = info->mnt, .dentry = dentry };
1329 if (likely(!d_mountpoint(dentry)))
1330 return D_WALK_CONTINUE;
1331 if (__path_is_mountpoint(&path)) {
1335 return D_WALK_CONTINUE;
1339 * path_has_submounts - check for mounts over a dentry in the
1340 * current namespace.
1341 * @parent: path to check.
1343 * Return true if the parent or its subdirectories contain
1344 * a mount point in the current namespace.
1346 int path_has_submounts(const struct path *parent)
1348 struct check_mount data = { .mnt = parent->mnt, .mounted = 0 };
1350 read_seqlock_excl(&mount_lock);
1351 d_walk(parent->dentry, &data, path_check_mount, NULL);
1352 read_sequnlock_excl(&mount_lock);
1354 return data.mounted;
1356 EXPORT_SYMBOL(path_has_submounts);
1359 * Called by mount code to set a mountpoint and check if the mountpoint is
1360 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1361 * subtree can become unreachable).
1363 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1364 * this reason take rename_lock and d_lock on dentry and ancestors.
1366 int d_set_mounted(struct dentry *dentry)
1370 write_seqlock(&rename_lock);
1371 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1372 /* Need exclusion wrt. d_invalidate() */
1373 spin_lock(&p->d_lock);
1374 if (unlikely(d_unhashed(p))) {
1375 spin_unlock(&p->d_lock);
1378 spin_unlock(&p->d_lock);
1380 spin_lock(&dentry->d_lock);
1381 if (!d_unlinked(dentry)) {
1383 if (!d_mountpoint(dentry)) {
1384 dentry->d_flags |= DCACHE_MOUNTED;
1388 spin_unlock(&dentry->d_lock);
1390 write_sequnlock(&rename_lock);
1395 * Search the dentry child list of the specified parent,
1396 * and move any unused dentries to the end of the unused
1397 * list for prune_dcache(). We descend to the next level
1398 * whenever the d_subdirs list is non-empty and continue
1401 * It returns zero iff there are no unused children,
1402 * otherwise it returns the number of children moved to
1403 * the end of the unused list. This may not be the total
1404 * number of unused children, because select_parent can
1405 * drop the lock and return early due to latency
1409 struct select_data {
1410 struct dentry *start;
1411 struct list_head dispose;
1415 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1417 struct select_data *data = _data;
1418 enum d_walk_ret ret = D_WALK_CONTINUE;
1420 if (data->start == dentry)
1423 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1426 if (dentry->d_flags & DCACHE_LRU_LIST)
1428 if (!dentry->d_lockref.count) {
1429 d_shrink_add(dentry, &data->dispose);
1434 * We can return to the caller if we have found some (this
1435 * ensures forward progress). We'll be coming back to find
1438 if (!list_empty(&data->dispose))
1439 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1445 * shrink_dcache_parent - prune dcache
1446 * @parent: parent of entries to prune
1448 * Prune the dcache to remove unused children of the parent dentry.
1450 void shrink_dcache_parent(struct dentry *parent)
1453 struct select_data data;
1455 INIT_LIST_HEAD(&data.dispose);
1456 data.start = parent;
1459 d_walk(parent, &data, select_collect, NULL);
1463 shrink_dentry_list(&data.dispose);
1467 EXPORT_SYMBOL(shrink_dcache_parent);
1469 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1471 /* it has busy descendents; complain about those instead */
1472 if (!list_empty(&dentry->d_subdirs))
1473 return D_WALK_CONTINUE;
1475 /* root with refcount 1 is fine */
1476 if (dentry == _data && dentry->d_lockref.count == 1)
1477 return D_WALK_CONTINUE;
1479 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1480 " still in use (%d) [unmount of %s %s]\n",
1483 dentry->d_inode->i_ino : 0UL,
1485 dentry->d_lockref.count,
1486 dentry->d_sb->s_type->name,
1487 dentry->d_sb->s_id);
1489 return D_WALK_CONTINUE;
1492 static void do_one_tree(struct dentry *dentry)
1494 shrink_dcache_parent(dentry);
1495 d_walk(dentry, dentry, umount_check, NULL);
1501 * destroy the dentries attached to a superblock on unmounting
1503 void shrink_dcache_for_umount(struct super_block *sb)
1505 struct dentry *dentry;
1507 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1509 dentry = sb->s_root;
1511 do_one_tree(dentry);
1513 while (!hlist_bl_empty(&sb->s_anon)) {
1514 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1515 do_one_tree(dentry);
1519 struct detach_data {
1520 struct select_data select;
1521 struct dentry *mountpoint;
1523 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1525 struct detach_data *data = _data;
1527 if (d_mountpoint(dentry)) {
1528 __dget_dlock(dentry);
1529 data->mountpoint = dentry;
1533 return select_collect(&data->select, dentry);
1536 static void check_and_drop(void *_data)
1538 struct detach_data *data = _data;
1540 if (!data->mountpoint && list_empty(&data->select.dispose))
1541 __d_drop(data->select.start);
1545 * d_invalidate - detach submounts, prune dcache, and drop
1546 * @dentry: dentry to invalidate (aka detach, prune and drop)
1550 * The final d_drop is done as an atomic operation relative to
1551 * rename_lock ensuring there are no races with d_set_mounted. This
1552 * ensures there are no unhashed dentries on the path to a mountpoint.
1554 void d_invalidate(struct dentry *dentry)
1557 * If it's already been dropped, return OK.
1559 spin_lock(&dentry->d_lock);
1560 if (d_unhashed(dentry)) {
1561 spin_unlock(&dentry->d_lock);
1564 spin_unlock(&dentry->d_lock);
1566 /* Negative dentries can be dropped without further checks */
1567 if (!dentry->d_inode) {
1573 struct detach_data data;
1575 data.mountpoint = NULL;
1576 INIT_LIST_HEAD(&data.select.dispose);
1577 data.select.start = dentry;
1578 data.select.found = 0;
1580 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1582 if (!list_empty(&data.select.dispose))
1583 shrink_dentry_list(&data.select.dispose);
1584 else if (!data.mountpoint)
1587 if (data.mountpoint) {
1588 detach_mounts(data.mountpoint);
1589 dput(data.mountpoint);
1594 EXPORT_SYMBOL(d_invalidate);
1597 * __d_alloc - allocate a dcache entry
1598 * @sb: filesystem it will belong to
1599 * @name: qstr of the name
1601 * Allocates a dentry. It returns %NULL if there is insufficient memory
1602 * available. On a success the dentry is returned. The name passed in is
1603 * copied and the copy passed in may be reused after this call.
1606 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1608 struct dentry *dentry;
1612 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1617 * We guarantee that the inline name is always NUL-terminated.
1618 * This way the memcpy() done by the name switching in rename
1619 * will still always have a NUL at the end, even if we might
1620 * be overwriting an internal NUL character
1622 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1623 if (unlikely(!name)) {
1625 dname = dentry->d_iname;
1626 } else if (name->len > DNAME_INLINE_LEN-1) {
1627 size_t size = offsetof(struct external_name, name[1]);
1628 struct external_name *p = kmalloc(size + name->len,
1629 GFP_KERNEL_ACCOUNT);
1631 kmem_cache_free(dentry_cache, dentry);
1634 atomic_set(&p->u.count, 1);
1636 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1637 kasan_unpoison_shadow(dname,
1638 round_up(name->len + 1, sizeof(unsigned long)));
1640 dname = dentry->d_iname;
1643 dentry->d_name.len = name->len;
1644 dentry->d_name.hash = name->hash;
1645 memcpy(dname, name->name, name->len);
1646 dname[name->len] = 0;
1648 /* Make sure we always see the terminating NUL character */
1650 dentry->d_name.name = dname;
1652 dentry->d_lockref.count = 1;
1653 dentry->d_flags = 0;
1654 spin_lock_init(&dentry->d_lock);
1655 seqcount_init(&dentry->d_seq);
1656 dentry->d_inode = NULL;
1657 dentry->d_parent = dentry;
1659 dentry->d_op = NULL;
1660 dentry->d_fsdata = NULL;
1661 INIT_HLIST_BL_NODE(&dentry->d_hash);
1662 INIT_LIST_HEAD(&dentry->d_lru);
1663 INIT_LIST_HEAD(&dentry->d_subdirs);
1664 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1665 INIT_LIST_HEAD(&dentry->d_child);
1666 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1668 if (dentry->d_op && dentry->d_op->d_init) {
1669 err = dentry->d_op->d_init(dentry);
1671 if (dname_external(dentry))
1672 kfree(external_name(dentry));
1673 kmem_cache_free(dentry_cache, dentry);
1678 this_cpu_inc(nr_dentry);
1684 * d_alloc - allocate a dcache entry
1685 * @parent: parent of entry to allocate
1686 * @name: qstr of the name
1688 * Allocates a dentry. It returns %NULL if there is insufficient memory
1689 * available. On a success the dentry is returned. The name passed in is
1690 * copied and the copy passed in may be reused after this call.
1692 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1694 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1697 dentry->d_flags |= DCACHE_RCUACCESS;
1698 spin_lock(&parent->d_lock);
1700 * don't need child lock because it is not subject
1701 * to concurrency here
1703 __dget_dlock(parent);
1704 dentry->d_parent = parent;
1705 list_add(&dentry->d_child, &parent->d_subdirs);
1706 spin_unlock(&parent->d_lock);
1710 EXPORT_SYMBOL(d_alloc);
1712 struct dentry *d_alloc_cursor(struct dentry * parent)
1714 struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1716 dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1717 dentry->d_parent = dget(parent);
1723 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1724 * @sb: the superblock
1725 * @name: qstr of the name
1727 * For a filesystem that just pins its dentries in memory and never
1728 * performs lookups at all, return an unhashed IS_ROOT dentry.
1730 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1732 return __d_alloc(sb, name);
1734 EXPORT_SYMBOL(d_alloc_pseudo);
1736 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1741 q.hash_len = hashlen_string(parent, name);
1742 return d_alloc(parent, &q);
1744 EXPORT_SYMBOL(d_alloc_name);
1746 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1748 WARN_ON_ONCE(dentry->d_op);
1749 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1751 DCACHE_OP_REVALIDATE |
1752 DCACHE_OP_WEAK_REVALIDATE |
1759 dentry->d_flags |= DCACHE_OP_HASH;
1761 dentry->d_flags |= DCACHE_OP_COMPARE;
1762 if (op->d_revalidate)
1763 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1764 if (op->d_weak_revalidate)
1765 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1767 dentry->d_flags |= DCACHE_OP_DELETE;
1769 dentry->d_flags |= DCACHE_OP_PRUNE;
1771 dentry->d_flags |= DCACHE_OP_REAL;
1774 EXPORT_SYMBOL(d_set_d_op);
1778 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1779 * @dentry - The dentry to mark
1781 * Mark a dentry as falling through to the lower layer (as set with
1782 * d_pin_lower()). This flag may be recorded on the medium.
1784 void d_set_fallthru(struct dentry *dentry)
1786 spin_lock(&dentry->d_lock);
1787 dentry->d_flags |= DCACHE_FALLTHRU;
1788 spin_unlock(&dentry->d_lock);
1790 EXPORT_SYMBOL(d_set_fallthru);
1792 static unsigned d_flags_for_inode(struct inode *inode)
1794 unsigned add_flags = DCACHE_REGULAR_TYPE;
1797 return DCACHE_MISS_TYPE;
1799 if (S_ISDIR(inode->i_mode)) {
1800 add_flags = DCACHE_DIRECTORY_TYPE;
1801 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1802 if (unlikely(!inode->i_op->lookup))
1803 add_flags = DCACHE_AUTODIR_TYPE;
1805 inode->i_opflags |= IOP_LOOKUP;
1807 goto type_determined;
1810 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1811 if (unlikely(inode->i_op->get_link)) {
1812 add_flags = DCACHE_SYMLINK_TYPE;
1813 goto type_determined;
1815 inode->i_opflags |= IOP_NOFOLLOW;
1818 if (unlikely(!S_ISREG(inode->i_mode)))
1819 add_flags = DCACHE_SPECIAL_TYPE;
1822 if (unlikely(IS_AUTOMOUNT(inode)))
1823 add_flags |= DCACHE_NEED_AUTOMOUNT;
1827 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1829 unsigned add_flags = d_flags_for_inode(inode);
1830 WARN_ON(d_in_lookup(dentry));
1832 spin_lock(&dentry->d_lock);
1833 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1834 raw_write_seqcount_begin(&dentry->d_seq);
1835 __d_set_inode_and_type(dentry, inode, add_flags);
1836 raw_write_seqcount_end(&dentry->d_seq);
1837 fsnotify_update_flags(dentry);
1838 spin_unlock(&dentry->d_lock);
1842 * d_instantiate - fill in inode information for a dentry
1843 * @entry: dentry to complete
1844 * @inode: inode to attach to this dentry
1846 * Fill in inode information in the entry.
1848 * This turns negative dentries into productive full members
1851 * NOTE! This assumes that the inode count has been incremented
1852 * (or otherwise set) by the caller to indicate that it is now
1853 * in use by the dcache.
1856 void d_instantiate(struct dentry *entry, struct inode * inode)
1858 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1860 security_d_instantiate(entry, inode);
1861 spin_lock(&inode->i_lock);
1862 __d_instantiate(entry, inode);
1863 spin_unlock(&inode->i_lock);
1866 EXPORT_SYMBOL(d_instantiate);
1869 * This should be equivalent to d_instantiate() + unlock_new_inode(),
1870 * with lockdep-related part of unlock_new_inode() done before
1871 * anything else. Use that instead of open-coding d_instantiate()/
1872 * unlock_new_inode() combinations.
1874 void d_instantiate_new(struct dentry *entry, struct inode *inode)
1876 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1878 lockdep_annotate_inode_mutex_key(inode);
1879 security_d_instantiate(entry, inode);
1880 spin_lock(&inode->i_lock);
1881 __d_instantiate(entry, inode);
1882 WARN_ON(!(inode->i_state & I_NEW));
1883 inode->i_state &= ~I_NEW;
1885 wake_up_bit(&inode->i_state, __I_NEW);
1886 spin_unlock(&inode->i_lock);
1888 EXPORT_SYMBOL(d_instantiate_new);
1891 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1892 * @entry: dentry to complete
1893 * @inode: inode to attach to this dentry
1895 * Fill in inode information in the entry. If a directory alias is found, then
1896 * return an error (and drop inode). Together with d_materialise_unique() this
1897 * guarantees that a directory inode may never have more than one alias.
1899 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1901 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1903 security_d_instantiate(entry, inode);
1904 spin_lock(&inode->i_lock);
1905 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1906 spin_unlock(&inode->i_lock);
1910 __d_instantiate(entry, inode);
1911 spin_unlock(&inode->i_lock);
1915 EXPORT_SYMBOL(d_instantiate_no_diralias);
1917 struct dentry *d_make_root(struct inode *root_inode)
1919 struct dentry *res = NULL;
1922 res = __d_alloc(root_inode->i_sb, NULL);
1924 res->d_flags |= DCACHE_RCUACCESS;
1925 d_instantiate(res, root_inode);
1932 EXPORT_SYMBOL(d_make_root);
1934 static struct dentry * __d_find_any_alias(struct inode *inode)
1936 struct dentry *alias;
1938 if (hlist_empty(&inode->i_dentry))
1940 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1946 * d_find_any_alias - find any alias for a given inode
1947 * @inode: inode to find an alias for
1949 * If any aliases exist for the given inode, take and return a
1950 * reference for one of them. If no aliases exist, return %NULL.
1952 struct dentry *d_find_any_alias(struct inode *inode)
1956 spin_lock(&inode->i_lock);
1957 de = __d_find_any_alias(inode);
1958 spin_unlock(&inode->i_lock);
1961 EXPORT_SYMBOL(d_find_any_alias);
1963 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1970 return ERR_PTR(-ESTALE);
1972 return ERR_CAST(inode);
1974 res = d_find_any_alias(inode);
1978 tmp = __d_alloc(inode->i_sb, NULL);
1980 res = ERR_PTR(-ENOMEM);
1984 security_d_instantiate(tmp, inode);
1985 spin_lock(&inode->i_lock);
1986 res = __d_find_any_alias(inode);
1988 spin_unlock(&inode->i_lock);
1993 /* attach a disconnected dentry */
1994 add_flags = d_flags_for_inode(inode);
1997 add_flags |= DCACHE_DISCONNECTED;
1999 spin_lock(&tmp->d_lock);
2000 __d_set_inode_and_type(tmp, inode, add_flags);
2001 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
2002 hlist_bl_lock(&tmp->d_sb->s_anon);
2003 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
2004 hlist_bl_unlock(&tmp->d_sb->s_anon);
2005 spin_unlock(&tmp->d_lock);
2006 spin_unlock(&inode->i_lock);
2016 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2017 * @inode: inode to allocate the dentry for
2019 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2020 * similar open by handle operations. The returned dentry may be anonymous,
2021 * or may have a full name (if the inode was already in the cache).
2023 * When called on a directory inode, we must ensure that the inode only ever
2024 * has one dentry. If a dentry is found, that is returned instead of
2025 * allocating a new one.
2027 * On successful return, the reference to the inode has been transferred
2028 * to the dentry. In case of an error the reference on the inode is released.
2029 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2030 * be passed in and the error will be propagated to the return value,
2031 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2033 struct dentry *d_obtain_alias(struct inode *inode)
2035 return __d_obtain_alias(inode, 1);
2037 EXPORT_SYMBOL(d_obtain_alias);
2040 * d_obtain_root - find or allocate a dentry for a given inode
2041 * @inode: inode to allocate the dentry for
2043 * Obtain an IS_ROOT dentry for the root of a filesystem.
2045 * We must ensure that directory inodes only ever have one dentry. If a
2046 * dentry is found, that is returned instead of allocating a new one.
2048 * On successful return, the reference to the inode has been transferred
2049 * to the dentry. In case of an error the reference on the inode is
2050 * released. A %NULL or IS_ERR inode may be passed in and will be the
2051 * error will be propagate to the return value, with a %NULL @inode
2052 * replaced by ERR_PTR(-ESTALE).
2054 struct dentry *d_obtain_root(struct inode *inode)
2056 return __d_obtain_alias(inode, 0);
2058 EXPORT_SYMBOL(d_obtain_root);
2061 * d_add_ci - lookup or allocate new dentry with case-exact name
2062 * @inode: the inode case-insensitive lookup has found
2063 * @dentry: the negative dentry that was passed to the parent's lookup func
2064 * @name: the case-exact name to be associated with the returned dentry
2066 * This is to avoid filling the dcache with case-insensitive names to the
2067 * same inode, only the actual correct case is stored in the dcache for
2068 * case-insensitive filesystems.
2070 * For a case-insensitive lookup match and if the the case-exact dentry
2071 * already exists in in the dcache, use it and return it.
2073 * If no entry exists with the exact case name, allocate new dentry with
2074 * the exact case, and return the spliced entry.
2076 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2079 struct dentry *found, *res;
2082 * First check if a dentry matching the name already exists,
2083 * if not go ahead and create it now.
2085 found = d_hash_and_lookup(dentry->d_parent, name);
2090 if (d_in_lookup(dentry)) {
2091 found = d_alloc_parallel(dentry->d_parent, name,
2093 if (IS_ERR(found) || !d_in_lookup(found)) {
2098 found = d_alloc(dentry->d_parent, name);
2101 return ERR_PTR(-ENOMEM);
2104 res = d_splice_alias(inode, found);
2111 EXPORT_SYMBOL(d_add_ci);
2114 static inline bool d_same_name(const struct dentry *dentry,
2115 const struct dentry *parent,
2116 const struct qstr *name)
2118 if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2119 if (dentry->d_name.len != name->len)
2121 return dentry_cmp(dentry, name->name, name->len) == 0;
2123 return parent->d_op->d_compare(dentry,
2124 dentry->d_name.len, dentry->d_name.name,
2129 * __d_lookup_rcu - search for a dentry (racy, store-free)
2130 * @parent: parent dentry
2131 * @name: qstr of name we wish to find
2132 * @seqp: returns d_seq value at the point where the dentry was found
2133 * Returns: dentry, or NULL
2135 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2136 * resolution (store-free path walking) design described in
2137 * Documentation/filesystems/path-lookup.txt.
2139 * This is not to be used outside core vfs.
2141 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2142 * held, and rcu_read_lock held. The returned dentry must not be stored into
2143 * without taking d_lock and checking d_seq sequence count against @seq
2146 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2149 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2150 * the returned dentry, so long as its parent's seqlock is checked after the
2151 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2152 * is formed, giving integrity down the path walk.
2154 * NOTE! The caller *has* to check the resulting dentry against the sequence
2155 * number we've returned before using any of the resulting dentry state!
2157 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2158 const struct qstr *name,
2161 u64 hashlen = name->hash_len;
2162 const unsigned char *str = name->name;
2163 struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2164 struct hlist_bl_node *node;
2165 struct dentry *dentry;
2168 * Note: There is significant duplication with __d_lookup_rcu which is
2169 * required to prevent single threaded performance regressions
2170 * especially on architectures where smp_rmb (in seqcounts) are costly.
2171 * Keep the two functions in sync.
2175 * The hash list is protected using RCU.
2177 * Carefully use d_seq when comparing a candidate dentry, to avoid
2178 * races with d_move().
2180 * It is possible that concurrent renames can mess up our list
2181 * walk here and result in missing our dentry, resulting in the
2182 * false-negative result. d_lookup() protects against concurrent
2183 * renames using rename_lock seqlock.
2185 * See Documentation/filesystems/path-lookup.txt for more details.
2187 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2192 * The dentry sequence count protects us from concurrent
2193 * renames, and thus protects parent and name fields.
2195 * The caller must perform a seqcount check in order
2196 * to do anything useful with the returned dentry.
2198 * NOTE! We do a "raw" seqcount_begin here. That means that
2199 * we don't wait for the sequence count to stabilize if it
2200 * is in the middle of a sequence change. If we do the slow
2201 * dentry compare, we will do seqretries until it is stable,
2202 * and if we end up with a successful lookup, we actually
2203 * want to exit RCU lookup anyway.
2205 * Note that raw_seqcount_begin still *does* smp_rmb(), so
2206 * we are still guaranteed NUL-termination of ->d_name.name.
2208 seq = raw_seqcount_begin(&dentry->d_seq);
2209 if (dentry->d_parent != parent)
2211 if (d_unhashed(dentry))
2214 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2217 if (dentry->d_name.hash != hashlen_hash(hashlen))
2219 tlen = dentry->d_name.len;
2220 tname = dentry->d_name.name;
2221 /* we want a consistent (name,len) pair */
2222 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2226 if (parent->d_op->d_compare(dentry,
2227 tlen, tname, name) != 0)
2230 if (dentry->d_name.hash_len != hashlen)
2232 if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
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 hash = name->hash;
2285 struct hlist_bl_head *b = d_hash(hash);
2286 struct hlist_bl_node *node;
2287 struct dentry *found = NULL;
2288 struct dentry *dentry;
2291 * Note: There is significant duplication with __d_lookup_rcu which is
2292 * required to prevent single threaded performance regressions
2293 * especially on architectures where smp_rmb (in seqcounts) are costly.
2294 * Keep the two functions in sync.
2298 * The hash list is protected using RCU.
2300 * Take d_lock when comparing a candidate dentry, to avoid races
2303 * It is possible that concurrent renames can mess up our list
2304 * walk here and result in missing our dentry, resulting in the
2305 * false-negative result. d_lookup() protects against concurrent
2306 * renames using rename_lock seqlock.
2308 * See Documentation/filesystems/path-lookup.txt for more details.
2312 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2314 if (dentry->d_name.hash != hash)
2317 spin_lock(&dentry->d_lock);
2318 if (dentry->d_parent != parent)
2320 if (d_unhashed(dentry))
2323 if (!d_same_name(dentry, parent, name))
2326 dentry->d_lockref.count++;
2328 spin_unlock(&dentry->d_lock);
2331 spin_unlock(&dentry->d_lock);
2339 * d_hash_and_lookup - hash the qstr then search for a dentry
2340 * @dir: Directory to search in
2341 * @name: qstr of name we wish to find
2343 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2345 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2348 * Check for a fs-specific hash function. Note that we must
2349 * calculate the standard hash first, as the d_op->d_hash()
2350 * routine may choose to leave the hash value unchanged.
2352 name->hash = full_name_hash(dir, name->name, name->len);
2353 if (dir->d_flags & DCACHE_OP_HASH) {
2354 int err = dir->d_op->d_hash(dir, name);
2355 if (unlikely(err < 0))
2356 return ERR_PTR(err);
2358 return d_lookup(dir, name);
2360 EXPORT_SYMBOL(d_hash_and_lookup);
2363 * When a file is deleted, we have two options:
2364 * - turn this dentry into a negative dentry
2365 * - unhash this dentry and free it.
2367 * Usually, we want to just turn this into
2368 * a negative dentry, but if anybody else is
2369 * currently using the dentry or the inode
2370 * we can't do that and we fall back on removing
2371 * it from the hash queues and waiting for
2372 * it to be deleted later when it has no users
2376 * d_delete - delete a dentry
2377 * @dentry: The dentry to delete
2379 * Turn the dentry into a negative dentry if possible, otherwise
2380 * remove it from the hash queues so it can be deleted later
2383 void d_delete(struct dentry * dentry)
2385 struct inode *inode;
2388 * Are we the only user?
2391 spin_lock(&dentry->d_lock);
2392 inode = dentry->d_inode;
2393 isdir = S_ISDIR(inode->i_mode);
2394 if (dentry->d_lockref.count == 1) {
2395 if (!spin_trylock(&inode->i_lock)) {
2396 spin_unlock(&dentry->d_lock);
2400 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2401 dentry_unlink_inode(dentry);
2402 fsnotify_nameremove(dentry, isdir);
2406 if (!d_unhashed(dentry))
2409 spin_unlock(&dentry->d_lock);
2411 fsnotify_nameremove(dentry, isdir);
2413 EXPORT_SYMBOL(d_delete);
2415 static void __d_rehash(struct dentry *entry)
2417 struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2420 hlist_bl_add_head_rcu(&entry->d_hash, b);
2425 * d_rehash - add an entry back to the hash
2426 * @entry: dentry to add to the hash
2428 * Adds a dentry to the hash according to its name.
2431 void d_rehash(struct dentry * entry)
2433 spin_lock(&entry->d_lock);
2435 spin_unlock(&entry->d_lock);
2437 EXPORT_SYMBOL(d_rehash);
2439 static inline unsigned start_dir_add(struct inode *dir)
2443 unsigned n = dir->i_dir_seq;
2444 if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2450 static inline void end_dir_add(struct inode *dir, unsigned n)
2452 smp_store_release(&dir->i_dir_seq, n + 2);
2455 static void d_wait_lookup(struct dentry *dentry)
2457 if (d_in_lookup(dentry)) {
2458 DECLARE_WAITQUEUE(wait, current);
2459 add_wait_queue(dentry->d_wait, &wait);
2461 set_current_state(TASK_UNINTERRUPTIBLE);
2462 spin_unlock(&dentry->d_lock);
2464 spin_lock(&dentry->d_lock);
2465 } while (d_in_lookup(dentry));
2469 struct dentry *d_alloc_parallel(struct dentry *parent,
2470 const struct qstr *name,
2471 wait_queue_head_t *wq)
2473 unsigned int hash = name->hash;
2474 struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2475 struct hlist_bl_node *node;
2476 struct dentry *new = d_alloc(parent, name);
2477 struct dentry *dentry;
2478 unsigned seq, r_seq, d_seq;
2481 return ERR_PTR(-ENOMEM);
2485 seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2486 r_seq = read_seqbegin(&rename_lock);
2487 dentry = __d_lookup_rcu(parent, name, &d_seq);
2488 if (unlikely(dentry)) {
2489 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2493 if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2502 if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2507 if (unlikely(seq & 1)) {
2513 if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2519 * No changes for the parent since the beginning of d_lookup().
2520 * Since all removals from the chain happen with hlist_bl_lock(),
2521 * any potential in-lookup matches are going to stay here until
2522 * we unlock the chain. All fields are stable in everything
2525 hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2526 if (dentry->d_name.hash != hash)
2528 if (dentry->d_parent != parent)
2530 if (!d_same_name(dentry, parent, name))
2533 /* now we can try to grab a reference */
2534 if (!lockref_get_not_dead(&dentry->d_lockref)) {
2541 * somebody is likely to be still doing lookup for it;
2542 * wait for them to finish
2544 spin_lock(&dentry->d_lock);
2545 d_wait_lookup(dentry);
2547 * it's not in-lookup anymore; in principle we should repeat
2548 * everything from dcache lookup, but it's likely to be what
2549 * d_lookup() would've found anyway. If it is, just return it;
2550 * otherwise we really have to repeat the whole thing.
2552 if (unlikely(dentry->d_name.hash != hash))
2554 if (unlikely(dentry->d_parent != parent))
2556 if (unlikely(d_unhashed(dentry)))
2558 if (unlikely(!d_same_name(dentry, parent, name)))
2560 /* OK, it *is* a hashed match; return it */
2561 spin_unlock(&dentry->d_lock);
2566 /* we can't take ->d_lock here; it's OK, though. */
2567 new->d_flags |= DCACHE_PAR_LOOKUP;
2569 hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2573 spin_unlock(&dentry->d_lock);
2577 EXPORT_SYMBOL(d_alloc_parallel);
2579 void __d_lookup_done(struct dentry *dentry)
2581 struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2582 dentry->d_name.hash);
2584 dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2585 __hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2586 wake_up_all(dentry->d_wait);
2587 dentry->d_wait = NULL;
2589 INIT_HLIST_NODE(&dentry->d_u.d_alias);
2590 INIT_LIST_HEAD(&dentry->d_lru);
2592 EXPORT_SYMBOL(__d_lookup_done);
2594 /* inode->i_lock held if inode is non-NULL */
2596 static inline void __d_add(struct dentry *dentry, struct inode *inode)
2598 struct inode *dir = NULL;
2600 spin_lock(&dentry->d_lock);
2601 if (unlikely(d_in_lookup(dentry))) {
2602 dir = dentry->d_parent->d_inode;
2603 n = start_dir_add(dir);
2604 __d_lookup_done(dentry);
2607 unsigned add_flags = d_flags_for_inode(inode);
2608 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2609 raw_write_seqcount_begin(&dentry->d_seq);
2610 __d_set_inode_and_type(dentry, inode, add_flags);
2611 raw_write_seqcount_end(&dentry->d_seq);
2612 fsnotify_update_flags(dentry);
2616 end_dir_add(dir, n);
2617 spin_unlock(&dentry->d_lock);
2619 spin_unlock(&inode->i_lock);
2623 * d_add - add dentry to hash queues
2624 * @entry: dentry to add
2625 * @inode: The inode to attach to this dentry
2627 * This adds the entry to the hash queues and initializes @inode.
2628 * The entry was actually filled in earlier during d_alloc().
2631 void d_add(struct dentry *entry, struct inode *inode)
2634 security_d_instantiate(entry, inode);
2635 spin_lock(&inode->i_lock);
2637 __d_add(entry, inode);
2639 EXPORT_SYMBOL(d_add);
2642 * d_exact_alias - find and hash an exact unhashed alias
2643 * @entry: dentry to add
2644 * @inode: The inode to go with this dentry
2646 * If an unhashed dentry with the same name/parent and desired
2647 * inode already exists, hash and return it. Otherwise, return
2650 * Parent directory should be locked.
2652 struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2654 struct dentry *alias;
2655 unsigned int hash = entry->d_name.hash;
2657 spin_lock(&inode->i_lock);
2658 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2660 * Don't need alias->d_lock here, because aliases with
2661 * d_parent == entry->d_parent are not subject to name or
2662 * parent changes, because the parent inode i_mutex is held.
2664 if (alias->d_name.hash != hash)
2666 if (alias->d_parent != entry->d_parent)
2668 if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2670 spin_lock(&alias->d_lock);
2671 if (!d_unhashed(alias)) {
2672 spin_unlock(&alias->d_lock);
2675 __dget_dlock(alias);
2677 spin_unlock(&alias->d_lock);
2679 spin_unlock(&inode->i_lock);
2682 spin_unlock(&inode->i_lock);
2685 EXPORT_SYMBOL(d_exact_alias);
2688 * dentry_update_name_case - update case insensitive dentry with a new name
2689 * @dentry: dentry to be updated
2692 * Update a case insensitive dentry with new case of name.
2694 * dentry must have been returned by d_lookup with name @name. Old and new
2695 * name lengths must match (ie. no d_compare which allows mismatched name
2698 * Parent inode i_mutex must be held over d_lookup and into this call (to
2699 * keep renames and concurrent inserts, and readdir(2) away).
2701 void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
2703 BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
2704 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2706 spin_lock(&dentry->d_lock);
2707 write_seqcount_begin(&dentry->d_seq);
2708 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2709 write_seqcount_end(&dentry->d_seq);
2710 spin_unlock(&dentry->d_lock);
2712 EXPORT_SYMBOL(dentry_update_name_case);
2714 static void swap_names(struct dentry *dentry, struct dentry *target)
2716 if (unlikely(dname_external(target))) {
2717 if (unlikely(dname_external(dentry))) {
2719 * Both external: swap the pointers
2721 swap(target->d_name.name, dentry->d_name.name);
2724 * dentry:internal, target:external. Steal target's
2725 * storage and make target internal.
2727 memcpy(target->d_iname, dentry->d_name.name,
2728 dentry->d_name.len + 1);
2729 dentry->d_name.name = target->d_name.name;
2730 target->d_name.name = target->d_iname;
2733 if (unlikely(dname_external(dentry))) {
2735 * dentry:external, target:internal. Give dentry's
2736 * storage to target and make dentry internal
2738 memcpy(dentry->d_iname, target->d_name.name,
2739 target->d_name.len + 1);
2740 target->d_name.name = dentry->d_name.name;
2741 dentry->d_name.name = dentry->d_iname;
2744 * Both are internal.
2747 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2748 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2749 swap(((long *) &dentry->d_iname)[i],
2750 ((long *) &target->d_iname)[i]);
2754 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2757 static void copy_name(struct dentry *dentry, struct dentry *target)
2759 struct external_name *old_name = NULL;
2760 if (unlikely(dname_external(dentry)))
2761 old_name = external_name(dentry);
2762 if (unlikely(dname_external(target))) {
2763 atomic_inc(&external_name(target)->u.count);
2764 dentry->d_name = target->d_name;
2766 memcpy(dentry->d_iname, target->d_name.name,
2767 target->d_name.len + 1);
2768 dentry->d_name.name = dentry->d_iname;
2769 dentry->d_name.hash_len = target->d_name.hash_len;
2771 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2772 kfree_rcu(old_name, u.head);
2775 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2778 * XXXX: do we really need to take target->d_lock?
2780 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2781 spin_lock(&target->d_parent->d_lock);
2783 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2784 spin_lock(&dentry->d_parent->d_lock);
2785 spin_lock_nested(&target->d_parent->d_lock,
2786 DENTRY_D_LOCK_NESTED);
2788 spin_lock(&target->d_parent->d_lock);
2789 spin_lock_nested(&dentry->d_parent->d_lock,
2790 DENTRY_D_LOCK_NESTED);
2793 if (target < dentry) {
2794 spin_lock_nested(&target->d_lock, 2);
2795 spin_lock_nested(&dentry->d_lock, 3);
2797 spin_lock_nested(&dentry->d_lock, 2);
2798 spin_lock_nested(&target->d_lock, 3);
2802 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2804 if (target->d_parent != dentry->d_parent)
2805 spin_unlock(&dentry->d_parent->d_lock);
2806 if (target->d_parent != target)
2807 spin_unlock(&target->d_parent->d_lock);
2808 spin_unlock(&target->d_lock);
2809 spin_unlock(&dentry->d_lock);
2813 * When switching names, the actual string doesn't strictly have to
2814 * be preserved in the target - because we're dropping the target
2815 * anyway. As such, we can just do a simple memcpy() to copy over
2816 * the new name before we switch, unless we are going to rehash
2817 * it. Note that if we *do* unhash the target, we are not allowed
2818 * to rehash it without giving it a new name/hash key - whether
2819 * we swap or overwrite the names here, resulting name won't match
2820 * the reality in filesystem; it's only there for d_path() purposes.
2821 * Note that all of this is happening under rename_lock, so the
2822 * any hash lookup seeing it in the middle of manipulations will
2823 * be discarded anyway. So we do not care what happens to the hash
2827 * __d_move - move a dentry
2828 * @dentry: entry to move
2829 * @target: new dentry
2830 * @exchange: exchange the two dentries
2832 * Update the dcache to reflect the move of a file name. Negative
2833 * dcache entries should not be moved in this way. Caller must hold
2834 * rename_lock, the i_mutex of the source and target directories,
2835 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2837 static void __d_move(struct dentry *dentry, struct dentry *target,
2840 struct inode *dir = NULL;
2842 if (!dentry->d_inode)
2843 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2845 BUG_ON(d_ancestor(dentry, target));
2846 BUG_ON(d_ancestor(target, dentry));
2848 dentry_lock_for_move(dentry, target);
2849 if (unlikely(d_in_lookup(target))) {
2850 dir = target->d_parent->d_inode;
2851 n = start_dir_add(dir);
2852 __d_lookup_done(target);
2855 write_seqcount_begin(&dentry->d_seq);
2856 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2859 /* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2863 /* Switch the names.. */
2865 swap_names(dentry, target);
2867 copy_name(dentry, target);
2869 /* rehash in new place(s) */
2874 target->d_hash.pprev = NULL;
2876 /* ... and switch them in the tree */
2877 if (IS_ROOT(dentry)) {
2878 /* splicing a tree */
2879 dentry->d_flags |= DCACHE_RCUACCESS;
2880 dentry->d_parent = target->d_parent;
2881 target->d_parent = target;
2882 list_del_init(&target->d_child);
2883 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2885 /* swapping two dentries */
2886 swap(dentry->d_parent, target->d_parent);
2887 list_move(&target->d_child, &target->d_parent->d_subdirs);
2888 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2890 fsnotify_update_flags(target);
2891 fsnotify_update_flags(dentry);
2894 write_seqcount_end(&target->d_seq);
2895 write_seqcount_end(&dentry->d_seq);
2898 end_dir_add(dir, n);
2899 dentry_unlock_for_move(dentry, target);
2903 * d_move - move a dentry
2904 * @dentry: entry to move
2905 * @target: new dentry
2907 * Update the dcache to reflect the move of a file name. Negative
2908 * dcache entries should not be moved in this way. See the locking
2909 * requirements for __d_move.
2911 void d_move(struct dentry *dentry, struct dentry *target)
2913 write_seqlock(&rename_lock);
2914 __d_move(dentry, target, false);
2915 write_sequnlock(&rename_lock);
2917 EXPORT_SYMBOL(d_move);
2920 * d_exchange - exchange two dentries
2921 * @dentry1: first dentry
2922 * @dentry2: second dentry
2924 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2926 write_seqlock(&rename_lock);
2928 WARN_ON(!dentry1->d_inode);
2929 WARN_ON(!dentry2->d_inode);
2930 WARN_ON(IS_ROOT(dentry1));
2931 WARN_ON(IS_ROOT(dentry2));
2933 __d_move(dentry1, dentry2, true);
2935 write_sequnlock(&rename_lock);
2939 * d_ancestor - search for an ancestor
2940 * @p1: ancestor dentry
2943 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2944 * an ancestor of p2, else NULL.
2946 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2950 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2951 if (p->d_parent == p1)
2958 * This helper attempts to cope with remotely renamed directories
2960 * It assumes that the caller is already holding
2961 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2963 * Note: If ever the locking in lock_rename() changes, then please
2964 * remember to update this too...
2966 static int __d_unalias(struct inode *inode,
2967 struct dentry *dentry, struct dentry *alias)
2969 struct mutex *m1 = NULL;
2970 struct rw_semaphore *m2 = NULL;
2973 /* If alias and dentry share a parent, then no extra locks required */
2974 if (alias->d_parent == dentry->d_parent)
2977 /* See lock_rename() */
2978 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2980 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2981 if (!inode_trylock_shared(alias->d_parent->d_inode))
2983 m2 = &alias->d_parent->d_inode->i_rwsem;
2985 __d_move(alias, dentry, false);
2996 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2997 * @inode: the inode which may have a disconnected dentry
2998 * @dentry: a negative dentry which we want to point to the inode.
3000 * If inode is a directory and has an IS_ROOT alias, then d_move that in
3001 * place of the given dentry and return it, else simply d_add the inode
3002 * to the dentry and return NULL.
3004 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
3005 * we should error out: directories can't have multiple aliases.
3007 * This is needed in the lookup routine of any filesystem that is exportable
3008 * (via knfsd) so that we can build dcache paths to directories effectively.
3010 * If a dentry was found and moved, then it is returned. Otherwise NULL
3011 * is returned. This matches the expected return value of ->lookup.
3013 * Cluster filesystems may call this function with a negative, hashed dentry.
3014 * In that case, we know that the inode will be a regular file, and also this
3015 * will only occur during atomic_open. So we need to check for the dentry
3016 * being already hashed only in the final case.
3018 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3021 return ERR_CAST(inode);
3023 BUG_ON(!d_unhashed(dentry));
3028 security_d_instantiate(dentry, inode);
3029 spin_lock(&inode->i_lock);
3030 if (S_ISDIR(inode->i_mode)) {
3031 struct dentry *new = __d_find_any_alias(inode);
3032 if (unlikely(new)) {
3033 /* The reference to new ensures it remains an alias */
3034 spin_unlock(&inode->i_lock);
3035 write_seqlock(&rename_lock);
3036 if (unlikely(d_ancestor(new, dentry))) {
3037 write_sequnlock(&rename_lock);
3039 new = ERR_PTR(-ELOOP);
3040 pr_warn_ratelimited(
3041 "VFS: Lookup of '%s' in %s %s"
3042 " would have caused loop\n",
3043 dentry->d_name.name,
3044 inode->i_sb->s_type->name,
3046 } else if (!IS_ROOT(new)) {
3047 int err = __d_unalias(inode, dentry, new);
3048 write_sequnlock(&rename_lock);
3054 __d_move(new, dentry, false);
3055 write_sequnlock(&rename_lock);
3062 __d_add(dentry, inode);
3065 EXPORT_SYMBOL(d_splice_alias);
3067 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
3071 return -ENAMETOOLONG;
3073 memcpy(*buffer, str, namelen);
3078 * prepend_name - prepend a pathname in front of current buffer pointer
3079 * @buffer: buffer pointer
3080 * @buflen: allocated length of the buffer
3081 * @name: name string and length qstr structure
3083 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3084 * make sure that either the old or the new name pointer and length are
3085 * fetched. However, there may be mismatch between length and pointer.
3086 * The length cannot be trusted, we need to copy it byte-by-byte until
3087 * the length is reached or a null byte is found. It also prepends "/" at
3088 * the beginning of the name. The sequence number check at the caller will
3089 * retry it again when a d_move() does happen. So any garbage in the buffer
3090 * due to mismatched pointer and length will be discarded.
3092 * Data dependency barrier is needed to make sure that we see that terminating
3093 * NUL. Alpha strikes again, film at 11...
3095 static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
3097 const char *dname = ACCESS_ONCE(name->name);
3098 u32 dlen = ACCESS_ONCE(name->len);
3101 smp_read_barrier_depends();
3103 *buflen -= dlen + 1;
3105 return -ENAMETOOLONG;
3106 p = *buffer -= dlen + 1;
3118 * prepend_path - Prepend path string to a buffer
3119 * @path: the dentry/vfsmount to report
3120 * @root: root vfsmnt/dentry
3121 * @buffer: pointer to the end of the buffer
3122 * @buflen: pointer to buffer length
3124 * The function will first try to write out the pathname without taking any
3125 * lock other than the RCU read lock to make sure that dentries won't go away.
3126 * It only checks the sequence number of the global rename_lock as any change
3127 * in the dentry's d_seq will be preceded by changes in the rename_lock
3128 * sequence number. If the sequence number had been changed, it will restart
3129 * the whole pathname back-tracing sequence again by taking the rename_lock.
3130 * In this case, there is no need to take the RCU read lock as the recursive
3131 * parent pointer references will keep the dentry chain alive as long as no
3132 * rename operation is performed.
3134 static int prepend_path(const struct path *path,
3135 const struct path *root,
3136 char **buffer, int *buflen)
3138 struct dentry *dentry;
3139 struct vfsmount *vfsmnt;
3142 unsigned seq, m_seq = 0;
3148 read_seqbegin_or_lock(&mount_lock, &m_seq);
3155 dentry = path->dentry;
3157 mnt = real_mount(vfsmnt);
3158 read_seqbegin_or_lock(&rename_lock, &seq);
3159 while (dentry != root->dentry || vfsmnt != root->mnt) {
3160 struct dentry * parent;
3162 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
3163 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
3165 if (dentry != vfsmnt->mnt_root) {
3172 if (mnt != parent) {
3173 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3179 error = is_mounted(vfsmnt) ? 1 : 2;
3182 parent = dentry->d_parent;
3184 error = prepend_name(&bptr, &blen, &dentry->d_name);
3192 if (need_seqretry(&rename_lock, seq)) {
3196 done_seqretry(&rename_lock, seq);
3200 if (need_seqretry(&mount_lock, m_seq)) {
3204 done_seqretry(&mount_lock, m_seq);
3206 if (error >= 0 && bptr == *buffer) {
3208 error = -ENAMETOOLONG;
3218 * __d_path - return the path of a dentry
3219 * @path: the dentry/vfsmount to report
3220 * @root: root vfsmnt/dentry
3221 * @buf: buffer to return value in
3222 * @buflen: buffer length
3224 * Convert a dentry into an ASCII path name.
3226 * Returns a pointer into the buffer or an error code if the
3227 * path was too long.
3229 * "buflen" should be positive.
3231 * If the path is not reachable from the supplied root, return %NULL.
3233 char *__d_path(const struct path *path,
3234 const struct path *root,
3235 char *buf, int buflen)
3237 char *res = buf + buflen;
3240 prepend(&res, &buflen, "\0", 1);
3241 error = prepend_path(path, root, &res, &buflen);
3244 return ERR_PTR(error);
3250 char *d_absolute_path(const struct path *path,
3251 char *buf, int buflen)
3253 struct path root = {};
3254 char *res = buf + buflen;
3257 prepend(&res, &buflen, "\0", 1);
3258 error = prepend_path(path, &root, &res, &buflen);
3263 return ERR_PTR(error);
3268 * same as __d_path but appends "(deleted)" for unlinked files.
3270 static int path_with_deleted(const struct path *path,
3271 const struct path *root,
3272 char **buf, int *buflen)
3274 prepend(buf, buflen, "\0", 1);
3275 if (d_unlinked(path->dentry)) {
3276 int error = prepend(buf, buflen, " (deleted)", 10);
3281 return prepend_path(path, root, buf, buflen);
3284 static int prepend_unreachable(char **buffer, int *buflen)
3286 return prepend(buffer, buflen, "(unreachable)", 13);
3289 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3294 seq = read_seqcount_begin(&fs->seq);
3296 } while (read_seqcount_retry(&fs->seq, seq));
3300 * d_path - return the path of a dentry
3301 * @path: path to report
3302 * @buf: buffer to return value in
3303 * @buflen: buffer length
3305 * Convert a dentry into an ASCII path name. If the entry has been deleted
3306 * the string " (deleted)" is appended. Note that this is ambiguous.
3308 * Returns a pointer into the buffer or an error code if the path was
3309 * too long. Note: Callers should use the returned pointer, not the passed
3310 * in buffer, to use the name! The implementation often starts at an offset
3311 * into the buffer, and may leave 0 bytes at the start.
3313 * "buflen" should be positive.
3315 char *d_path(const struct path *path, char *buf, int buflen)
3317 char *res = buf + buflen;
3322 * We have various synthetic filesystems that never get mounted. On
3323 * these filesystems dentries are never used for lookup purposes, and
3324 * thus don't need to be hashed. They also don't need a name until a
3325 * user wants to identify the object in /proc/pid/fd/. The little hack
3326 * below allows us to generate a name for these objects on demand:
3328 * Some pseudo inodes are mountable. When they are mounted
3329 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3330 * and instead have d_path return the mounted path.
3332 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3333 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3334 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3337 get_fs_root_rcu(current->fs, &root);
3338 error = path_with_deleted(path, &root, &res, &buflen);
3342 res = ERR_PTR(error);
3345 EXPORT_SYMBOL(d_path);
3348 * Helper function for dentry_operations.d_dname() members
3350 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3351 const char *fmt, ...)
3357 va_start(args, fmt);
3358 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3361 if (sz > sizeof(temp) || sz > buflen)
3362 return ERR_PTR(-ENAMETOOLONG);
3364 buffer += buflen - sz;
3365 return memcpy(buffer, temp, sz);
3368 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3370 char *end = buffer + buflen;
3371 /* these dentries are never renamed, so d_lock is not needed */
3372 if (prepend(&end, &buflen, " (deleted)", 11) ||
3373 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3374 prepend(&end, &buflen, "/", 1))
3375 end = ERR_PTR(-ENAMETOOLONG);
3378 EXPORT_SYMBOL(simple_dname);
3381 * Write full pathname from the root of the filesystem into the buffer.
3383 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3385 struct dentry *dentry;
3398 prepend(&end, &len, "\0", 1);
3402 read_seqbegin_or_lock(&rename_lock, &seq);
3403 while (!IS_ROOT(dentry)) {
3404 struct dentry *parent = dentry->d_parent;
3407 error = prepend_name(&end, &len, &dentry->d_name);
3416 if (need_seqretry(&rename_lock, seq)) {
3420 done_seqretry(&rename_lock, seq);
3425 return ERR_PTR(-ENAMETOOLONG);
3428 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3430 return __dentry_path(dentry, buf, buflen);
3432 EXPORT_SYMBOL(dentry_path_raw);
3434 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3439 if (d_unlinked(dentry)) {
3441 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3445 retval = __dentry_path(dentry, buf, buflen);
3446 if (!IS_ERR(retval) && p)
3447 *p = '/'; /* restore '/' overriden with '\0' */
3450 return ERR_PTR(-ENAMETOOLONG);
3453 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3459 seq = read_seqcount_begin(&fs->seq);
3462 } while (read_seqcount_retry(&fs->seq, seq));
3466 * NOTE! The user-level library version returns a
3467 * character pointer. The kernel system call just
3468 * returns the length of the buffer filled (which
3469 * includes the ending '\0' character), or a negative
3470 * error value. So libc would do something like
3472 * char *getcwd(char * buf, size_t size)
3476 * retval = sys_getcwd(buf, size);
3483 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3486 struct path pwd, root;
3487 char *page = __getname();
3493 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3496 if (!d_unlinked(pwd.dentry)) {
3498 char *cwd = page + PATH_MAX;
3499 int buflen = PATH_MAX;
3501 prepend(&cwd, &buflen, "\0", 1);
3502 error = prepend_path(&pwd, &root, &cwd, &buflen);
3508 /* Unreachable from current root */
3510 error = prepend_unreachable(&cwd, &buflen);
3516 len = PATH_MAX + page - cwd;
3519 if (copy_to_user(buf, cwd, len))
3532 * Test whether new_dentry is a subdirectory of old_dentry.
3534 * Trivially implemented using the dcache structure
3538 * is_subdir - is new dentry a subdirectory of old_dentry
3539 * @new_dentry: new dentry
3540 * @old_dentry: old dentry
3542 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3543 * Returns false otherwise.
3544 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3547 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3552 if (new_dentry == old_dentry)
3556 /* for restarting inner loop in case of seq retry */
3557 seq = read_seqbegin(&rename_lock);
3559 * Need rcu_readlock to protect against the d_parent trashing
3563 if (d_ancestor(old_dentry, new_dentry))
3568 } while (read_seqretry(&rename_lock, seq));
3573 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3575 struct dentry *root = data;
3576 if (dentry != root) {
3577 if (d_unhashed(dentry) || !dentry->d_inode)
3580 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3581 dentry->d_flags |= DCACHE_GENOCIDE;
3582 dentry->d_lockref.count--;
3585 return D_WALK_CONTINUE;
3588 void d_genocide(struct dentry *parent)
3590 d_walk(parent, parent, d_genocide_kill, NULL);
3593 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3595 inode_dec_link_count(inode);
3596 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3597 !hlist_unhashed(&dentry->d_u.d_alias) ||
3598 !d_unlinked(dentry));
3599 spin_lock(&dentry->d_parent->d_lock);
3600 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3601 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3602 (unsigned long long)inode->i_ino);
3603 spin_unlock(&dentry->d_lock);
3604 spin_unlock(&dentry->d_parent->d_lock);
3605 d_instantiate(dentry, inode);
3607 EXPORT_SYMBOL(d_tmpfile);
3609 static __initdata unsigned long dhash_entries;
3610 static int __init set_dhash_entries(char *str)
3614 dhash_entries = simple_strtoul(str, &str, 0);
3617 __setup("dhash_entries=", set_dhash_entries);
3619 static void __init dcache_init_early(void)
3621 /* If hashes are distributed across NUMA nodes, defer
3622 * hash allocation until vmalloc space is available.
3628 alloc_large_system_hash("Dentry cache",
3629 sizeof(struct hlist_bl_head),
3632 HASH_EARLY | HASH_ZERO,
3639 static void __init dcache_init(void)
3642 * A constructor could be added for stable state like the lists,
3643 * but it is probably not worth it because of the cache nature
3646 dentry_cache = KMEM_CACHE(dentry,
3647 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
3649 /* Hash may have been set up in dcache_init_early */
3654 alloc_large_system_hash("Dentry cache",
3655 sizeof(struct hlist_bl_head),
3665 /* SLAB cache for __getname() consumers */
3666 struct kmem_cache *names_cachep __read_mostly;
3667 EXPORT_SYMBOL(names_cachep);
3669 EXPORT_SYMBOL(d_genocide);
3671 void __init vfs_caches_init_early(void)
3675 for (i = 0; i < ARRAY_SIZE(in_lookup_hashtable); i++)
3676 INIT_HLIST_BL_HEAD(&in_lookup_hashtable[i]);
3678 dcache_init_early();
3682 void __init vfs_caches_init(void)
3684 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3685 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3690 files_maxfiles_init();