4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
10 * Notes on the allocation strategy:
12 * The dcache is a master of the icache - whenever a dcache entry
13 * exists, the inode will always exist. "iput()" is done either when
14 * the dcache entry is deleted or garbage collected.
17 #include <linux/syscalls.h>
18 #include <linux/string.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
46 * dcache->d_inode->i_lock protects:
47 * - i_dentry, d_alias, d_inode of aliases
48 * dcache_hash_bucket lock protects:
49 * - the dcache hash table
50 * s_anon bl list spinlock protects:
51 * - the s_anon list (see __d_drop)
52 * dentry->d_sb->s_dentry_lru_lock protects:
53 * - the dcache lru lists and counters
60 * - d_parent and d_subdirs
61 * - childrens' d_child and d_parent
65 * dentry->d_inode->i_lock
67 * dentry->d_sb->s_dentry_lru_lock
68 * dcache_hash_bucket lock
71 * If there is an ancestor relationship:
72 * dentry->d_parent->...->d_parent->d_lock
74 * dentry->d_parent->d_lock
77 * If no ancestor relationship:
78 * if (dentry1 < dentry2)
82 int sysctl_vfs_cache_pressure __read_mostly = 100;
83 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
87 EXPORT_SYMBOL(rename_lock);
89 static struct kmem_cache *dentry_cache __read_mostly;
92 * read_seqbegin_or_lock - begin a sequence number check or locking block
93 * @lock: sequence lock
94 * @seq : sequence number to be checked
96 * First try it once optimistically without taking the lock. If that fails,
97 * take the lock. The sequence number is also used as a marker for deciding
98 * whether to be a reader (even) or writer (odd).
99 * N.B. seq must be initialized to an even number to begin with.
101 static inline void read_seqbegin_or_lock(seqlock_t *lock, int *seq)
103 if (!(*seq & 1)) /* Even */
104 *seq = read_seqbegin(lock);
106 read_seqlock_excl(lock);
109 static inline int need_seqretry(seqlock_t *lock, int seq)
111 return !(seq & 1) && read_seqretry(lock, seq);
114 static inline void done_seqretry(seqlock_t *lock, int seq)
117 read_sequnlock_excl(lock);
121 * This is the single most critical data structure when it comes
122 * to the dcache: the hashtable for lookups. Somebody should try
123 * to make this good - I've just made it work.
125 * This hash-function tries to avoid losing too many bits of hash
126 * information, yet avoid using a prime hash-size or similar.
128 #define D_HASHBITS d_hash_shift
129 #define D_HASHMASK d_hash_mask
131 static unsigned int d_hash_mask __read_mostly;
132 static unsigned int d_hash_shift __read_mostly;
134 static struct hlist_bl_head *dentry_hashtable __read_mostly;
136 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
139 hash += (unsigned long) parent / L1_CACHE_BYTES;
140 hash = hash + (hash >> D_HASHBITS);
141 return dentry_hashtable + (hash & D_HASHMASK);
144 /* Statistics gathering. */
145 struct dentry_stat_t dentry_stat = {
149 static DEFINE_PER_CPU(long, nr_dentry);
150 static DEFINE_PER_CPU(long, nr_dentry_unused);
152 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
155 * Here we resort to our own counters instead of using generic per-cpu counters
156 * for consistency with what the vfs inode code does. We are expected to harvest
157 * better code and performance by having our own specialized counters.
159 * Please note that the loop is done over all possible CPUs, not over all online
160 * CPUs. The reason for this is that we don't want to play games with CPUs going
161 * on and off. If one of them goes off, we will just keep their counters.
163 * glommer: See cffbc8a for details, and if you ever intend to change this,
164 * please update all vfs counters to match.
166 static long get_nr_dentry(void)
170 for_each_possible_cpu(i)
171 sum += per_cpu(nr_dentry, i);
172 return sum < 0 ? 0 : sum;
175 static long get_nr_dentry_unused(void)
179 for_each_possible_cpu(i)
180 sum += per_cpu(nr_dentry_unused, i);
181 return sum < 0 ? 0 : sum;
184 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
185 size_t *lenp, loff_t *ppos)
187 dentry_stat.nr_dentry = get_nr_dentry();
188 dentry_stat.nr_unused = get_nr_dentry_unused();
189 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
194 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
195 * The strings are both count bytes long, and count is non-zero.
197 #ifdef CONFIG_DCACHE_WORD_ACCESS
199 #include <asm/word-at-a-time.h>
201 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
202 * aligned allocation for this particular component. We don't
203 * strictly need the load_unaligned_zeropad() safety, but it
204 * doesn't hurt either.
206 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
207 * need the careful unaligned handling.
209 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
211 unsigned long a,b,mask;
214 a = *(unsigned long *)cs;
215 b = load_unaligned_zeropad(ct);
216 if (tcount < sizeof(unsigned long))
218 if (unlikely(a != b))
220 cs += sizeof(unsigned long);
221 ct += sizeof(unsigned long);
222 tcount -= sizeof(unsigned long);
226 mask = ~(~0ul << tcount*8);
227 return unlikely(!!((a ^ b) & mask));
232 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
246 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
248 const unsigned char *cs;
250 * Be careful about RCU walk racing with rename:
251 * use ACCESS_ONCE to fetch the name pointer.
253 * NOTE! Even if a rename will mean that the length
254 * was not loaded atomically, we don't care. The
255 * RCU walk will check the sequence count eventually,
256 * and catch it. And we won't overrun the buffer,
257 * because we're reading the name pointer atomically,
258 * and a dentry name is guaranteed to be properly
259 * terminated with a NUL byte.
261 * End result: even if 'len' is wrong, we'll exit
262 * early because the data cannot match (there can
263 * be no NUL in the ct/tcount data)
265 cs = ACCESS_ONCE(dentry->d_name.name);
266 smp_read_barrier_depends();
267 return dentry_string_cmp(cs, ct, tcount);
270 static void __d_free(struct rcu_head *head)
272 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
274 WARN_ON(!hlist_unhashed(&dentry->d_alias));
275 if (dname_external(dentry))
276 kfree(dentry->d_name.name);
277 kmem_cache_free(dentry_cache, dentry);
283 static void d_free(struct dentry *dentry)
285 BUG_ON((int)dentry->d_lockref.count > 0);
286 this_cpu_dec(nr_dentry);
287 if (dentry->d_op && dentry->d_op->d_release)
288 dentry->d_op->d_release(dentry);
290 /* if dentry was never visible to RCU, immediate free is OK */
291 if (!(dentry->d_flags & DCACHE_RCUACCESS))
292 __d_free(&dentry->d_u.d_rcu);
294 call_rcu(&dentry->d_u.d_rcu, __d_free);
298 * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
299 * @dentry: the target dentry
300 * After this call, in-progress rcu-walk path lookup will fail. This
301 * should be called after unhashing, and after changing d_inode (if
302 * the dentry has not already been unhashed).
304 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
306 assert_spin_locked(&dentry->d_lock);
307 /* Go through a barrier */
308 write_seqcount_barrier(&dentry->d_seq);
312 * Release the dentry's inode, using the filesystem
313 * d_iput() operation if defined. Dentry has no refcount
316 static void dentry_iput(struct dentry * dentry)
317 __releases(dentry->d_lock)
318 __releases(dentry->d_inode->i_lock)
320 struct inode *inode = dentry->d_inode;
322 dentry->d_inode = NULL;
323 hlist_del_init(&dentry->d_alias);
324 spin_unlock(&dentry->d_lock);
325 spin_unlock(&inode->i_lock);
327 fsnotify_inoderemove(inode);
328 if (dentry->d_op && dentry->d_op->d_iput)
329 dentry->d_op->d_iput(dentry, inode);
333 spin_unlock(&dentry->d_lock);
338 * Release the dentry's inode, using the filesystem
339 * d_iput() operation if defined. dentry remains in-use.
341 static void dentry_unlink_inode(struct dentry * dentry)
342 __releases(dentry->d_lock)
343 __releases(dentry->d_inode->i_lock)
345 struct inode *inode = dentry->d_inode;
346 __d_clear_type(dentry);
347 dentry->d_inode = NULL;
348 hlist_del_init(&dentry->d_alias);
349 dentry_rcuwalk_barrier(dentry);
350 spin_unlock(&dentry->d_lock);
351 spin_unlock(&inode->i_lock);
353 fsnotify_inoderemove(inode);
354 if (dentry->d_op && dentry->d_op->d_iput)
355 dentry->d_op->d_iput(dentry, inode);
361 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
362 * is in use - which includes both the "real" per-superblock
363 * LRU list _and_ the DCACHE_SHRINK_LIST use.
365 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
366 * on the shrink list (ie not on the superblock LRU list).
368 * The per-cpu "nr_dentry_unused" counters are updated with
369 * the DCACHE_LRU_LIST bit.
371 * These helper functions make sure we always follow the
372 * rules. d_lock must be held by the caller.
374 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
375 static void d_lru_add(struct dentry *dentry)
377 D_FLAG_VERIFY(dentry, 0);
378 dentry->d_flags |= DCACHE_LRU_LIST;
379 this_cpu_inc(nr_dentry_unused);
380 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
383 static void d_lru_del(struct dentry *dentry)
385 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
386 dentry->d_flags &= ~DCACHE_LRU_LIST;
387 this_cpu_dec(nr_dentry_unused);
388 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
391 static void d_shrink_del(struct dentry *dentry)
393 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
394 list_del_init(&dentry->d_lru);
395 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
396 this_cpu_dec(nr_dentry_unused);
399 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
401 D_FLAG_VERIFY(dentry, 0);
402 list_add(&dentry->d_lru, list);
403 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
404 this_cpu_inc(nr_dentry_unused);
408 * These can only be called under the global LRU lock, ie during the
409 * callback for freeing the LRU list. "isolate" removes it from the
410 * LRU lists entirely, while shrink_move moves it to the indicated
413 static void d_lru_isolate(struct dentry *dentry)
415 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
416 dentry->d_flags &= ~DCACHE_LRU_LIST;
417 this_cpu_dec(nr_dentry_unused);
418 list_del_init(&dentry->d_lru);
421 static void d_lru_shrink_move(struct dentry *dentry, struct list_head *list)
423 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
424 dentry->d_flags |= DCACHE_SHRINK_LIST;
425 list_move_tail(&dentry->d_lru, list);
429 * dentry_lru_(add|del)_list) must be called with d_lock held.
431 static void dentry_lru_add(struct dentry *dentry)
433 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
438 * Remove a dentry with references from the LRU.
440 * If we are on the shrink list, then we can get to try_prune_one_dentry() and
441 * lose our last reference through the parent walk. In this case, we need to
442 * remove ourselves from the shrink list, not the LRU.
444 static void dentry_lru_del(struct dentry *dentry)
446 if (dentry->d_flags & DCACHE_LRU_LIST) {
447 if (dentry->d_flags & DCACHE_SHRINK_LIST)
448 return d_shrink_del(dentry);
454 * d_kill - kill dentry and return parent
455 * @dentry: dentry to kill
456 * @parent: parent dentry
458 * The dentry must already be unhashed and removed from the LRU.
460 * If this is the root of the dentry tree, return NULL.
462 * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
465 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
466 __releases(dentry->d_lock)
467 __releases(parent->d_lock)
468 __releases(dentry->d_inode->i_lock)
470 list_del(&dentry->d_u.d_child);
472 * Inform try_to_ascend() that we are no longer attached to the
475 dentry->d_flags |= DCACHE_DENTRY_KILLED;
477 spin_unlock(&parent->d_lock);
480 * dentry_iput drops the locks, at which point nobody (except
481 * transient RCU lookups) can reach this dentry.
488 * d_drop - drop a dentry
489 * @dentry: dentry to drop
491 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
492 * be found through a VFS lookup any more. Note that this is different from
493 * deleting the dentry - d_delete will try to mark the dentry negative if
494 * possible, giving a successful _negative_ lookup, while d_drop will
495 * just make the cache lookup fail.
497 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
498 * reason (NFS timeouts or autofs deletes).
500 * __d_drop requires dentry->d_lock.
502 void __d_drop(struct dentry *dentry)
504 if (!d_unhashed(dentry)) {
505 struct hlist_bl_head *b;
507 * Hashed dentries are normally on the dentry hashtable,
508 * with the exception of those newly allocated by
509 * d_obtain_alias, which are always IS_ROOT:
511 if (unlikely(IS_ROOT(dentry)))
512 b = &dentry->d_sb->s_anon;
514 b = d_hash(dentry->d_parent, dentry->d_name.hash);
517 __hlist_bl_del(&dentry->d_hash);
518 dentry->d_hash.pprev = NULL;
520 dentry_rcuwalk_barrier(dentry);
523 EXPORT_SYMBOL(__d_drop);
525 void d_drop(struct dentry *dentry)
527 spin_lock(&dentry->d_lock);
529 spin_unlock(&dentry->d_lock);
531 EXPORT_SYMBOL(d_drop);
534 * Finish off a dentry we've decided to kill.
535 * dentry->d_lock must be held, returns with it unlocked.
536 * If ref is non-zero, then decrement the refcount too.
537 * Returns dentry requiring refcount drop, or NULL if we're done.
539 static inline struct dentry *
540 dentry_kill(struct dentry *dentry, int unlock_on_failure)
541 __releases(dentry->d_lock)
544 struct dentry *parent;
546 inode = dentry->d_inode;
547 if (inode && !spin_trylock(&inode->i_lock)) {
549 if (unlock_on_failure) {
550 spin_unlock(&dentry->d_lock);
553 return dentry; /* try again with same dentry */
558 parent = dentry->d_parent;
559 if (parent && !spin_trylock(&parent->d_lock)) {
561 spin_unlock(&inode->i_lock);
566 * The dentry is now unrecoverably dead to the world.
568 lockref_mark_dead(&dentry->d_lockref);
571 * inform the fs via d_prune that this dentry is about to be
572 * unhashed and destroyed.
574 if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry))
575 dentry->d_op->d_prune(dentry);
577 dentry_lru_del(dentry);
578 /* if it was on the hash then remove it */
580 return d_kill(dentry, parent);
586 * This is complicated by the fact that we do not want to put
587 * dentries that are no longer on any hash chain on the unused
588 * list: we'd much rather just get rid of them immediately.
590 * However, that implies that we have to traverse the dentry
591 * tree upwards to the parents which might _also_ now be
592 * scheduled for deletion (it may have been only waiting for
593 * its last child to go away).
595 * This tail recursion is done by hand as we don't want to depend
596 * on the compiler to always get this right (gcc generally doesn't).
597 * Real recursion would eat up our stack space.
601 * dput - release a dentry
602 * @dentry: dentry to release
604 * Release a dentry. This will drop the usage count and if appropriate
605 * call the dentry unlink method as well as removing it from the queues and
606 * releasing its resources. If the parent dentries were scheduled for release
607 * they too may now get deleted.
609 void dput(struct dentry *dentry)
611 if (unlikely(!dentry))
615 if (lockref_put_or_lock(&dentry->d_lockref))
618 /* Unreachable? Get rid of it */
619 if (unlikely(d_unhashed(dentry)))
622 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
623 if (dentry->d_op->d_delete(dentry))
627 dentry->d_flags |= DCACHE_REFERENCED;
628 dentry_lru_add(dentry);
630 dentry->d_lockref.count--;
631 spin_unlock(&dentry->d_lock);
635 dentry = dentry_kill(dentry, 1);
642 * d_invalidate - invalidate a dentry
643 * @dentry: dentry to invalidate
645 * Try to invalidate the dentry if it turns out to be
646 * possible. If there are other dentries that can be
647 * reached through this one we can't delete it and we
648 * return -EBUSY. On success we return 0.
653 int d_invalidate(struct dentry * dentry)
656 * If it's already been dropped, return OK.
658 spin_lock(&dentry->d_lock);
659 if (d_unhashed(dentry)) {
660 spin_unlock(&dentry->d_lock);
664 * Check whether to do a partial shrink_dcache
665 * to get rid of unused child entries.
667 if (!list_empty(&dentry->d_subdirs)) {
668 spin_unlock(&dentry->d_lock);
669 shrink_dcache_parent(dentry);
670 spin_lock(&dentry->d_lock);
674 * Somebody else still using it?
676 * If it's a directory, we can't drop it
677 * for fear of somebody re-populating it
678 * with children (even though dropping it
679 * would make it unreachable from the root,
680 * we might still populate it if it was a
681 * working directory or similar).
682 * We also need to leave mountpoints alone,
685 if (dentry->d_lockref.count > 1 && dentry->d_inode) {
686 if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
687 spin_unlock(&dentry->d_lock);
693 spin_unlock(&dentry->d_lock);
696 EXPORT_SYMBOL(d_invalidate);
698 /* This must be called with d_lock held */
699 static inline void __dget_dlock(struct dentry *dentry)
701 dentry->d_lockref.count++;
704 static inline void __dget(struct dentry *dentry)
706 lockref_get(&dentry->d_lockref);
709 struct dentry *dget_parent(struct dentry *dentry)
715 * Do optimistic parent lookup without any
719 ret = ACCESS_ONCE(dentry->d_parent);
720 gotref = lockref_get_not_zero(&ret->d_lockref);
722 if (likely(gotref)) {
723 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
730 * Don't need rcu_dereference because we re-check it was correct under
734 ret = dentry->d_parent;
735 spin_lock(&ret->d_lock);
736 if (unlikely(ret != dentry->d_parent)) {
737 spin_unlock(&ret->d_lock);
742 BUG_ON(!ret->d_lockref.count);
743 ret->d_lockref.count++;
744 spin_unlock(&ret->d_lock);
747 EXPORT_SYMBOL(dget_parent);
750 * d_find_alias - grab a hashed alias of inode
751 * @inode: inode in question
752 * @want_discon: flag, used by d_splice_alias, to request
753 * that only a DISCONNECTED alias be returned.
755 * If inode has a hashed alias, or is a directory and has any alias,
756 * acquire the reference to alias and return it. Otherwise return NULL.
757 * Notice that if inode is a directory there can be only one alias and
758 * it can be unhashed only if it has no children, or if it is the root
761 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
762 * any other hashed alias over that one unless @want_discon is set,
763 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
765 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
767 struct dentry *alias, *discon_alias;
771 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
772 spin_lock(&alias->d_lock);
773 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
774 if (IS_ROOT(alias) &&
775 (alias->d_flags & DCACHE_DISCONNECTED)) {
776 discon_alias = alias;
777 } else if (!want_discon) {
779 spin_unlock(&alias->d_lock);
783 spin_unlock(&alias->d_lock);
786 alias = discon_alias;
787 spin_lock(&alias->d_lock);
788 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
789 if (IS_ROOT(alias) &&
790 (alias->d_flags & DCACHE_DISCONNECTED)) {
792 spin_unlock(&alias->d_lock);
796 spin_unlock(&alias->d_lock);
802 struct dentry *d_find_alias(struct inode *inode)
804 struct dentry *de = NULL;
806 if (!hlist_empty(&inode->i_dentry)) {
807 spin_lock(&inode->i_lock);
808 de = __d_find_alias(inode, 0);
809 spin_unlock(&inode->i_lock);
813 EXPORT_SYMBOL(d_find_alias);
816 * Try to kill dentries associated with this inode.
817 * WARNING: you must own a reference to inode.
819 void d_prune_aliases(struct inode *inode)
821 struct dentry *dentry;
823 spin_lock(&inode->i_lock);
824 hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) {
825 spin_lock(&dentry->d_lock);
826 if (!dentry->d_lockref.count) {
828 * inform the fs via d_prune that this dentry
829 * is about to be unhashed and destroyed.
831 if ((dentry->d_flags & DCACHE_OP_PRUNE) &&
833 dentry->d_op->d_prune(dentry);
835 __dget_dlock(dentry);
837 spin_unlock(&dentry->d_lock);
838 spin_unlock(&inode->i_lock);
842 spin_unlock(&dentry->d_lock);
844 spin_unlock(&inode->i_lock);
846 EXPORT_SYMBOL(d_prune_aliases);
849 * Try to throw away a dentry - free the inode, dput the parent.
850 * Requires dentry->d_lock is held, and dentry->d_count == 0.
851 * Releases dentry->d_lock.
853 * This may fail if locks cannot be acquired no problem, just try again.
855 static struct dentry * try_prune_one_dentry(struct dentry *dentry)
856 __releases(dentry->d_lock)
858 struct dentry *parent;
860 parent = dentry_kill(dentry, 0);
862 * If dentry_kill returns NULL, we have nothing more to do.
863 * if it returns the same dentry, trylocks failed. In either
864 * case, just loop again.
866 * Otherwise, we need to prune ancestors too. This is necessary
867 * to prevent quadratic behavior of shrink_dcache_parent(), but
868 * is also expected to be beneficial in reducing dentry cache
873 if (parent == dentry)
876 /* Prune ancestors. */
879 if (lockref_put_or_lock(&dentry->d_lockref))
881 dentry = dentry_kill(dentry, 1);
886 static void shrink_dentry_list(struct list_head *list)
888 struct dentry *dentry;
892 dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
893 if (&dentry->d_lru == list)
897 * Get the dentry lock, and re-verify that the dentry is
898 * this on the shrinking list. If it is, we know that
899 * DCACHE_SHRINK_LIST and DCACHE_LRU_LIST are set.
901 spin_lock(&dentry->d_lock);
902 if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
903 spin_unlock(&dentry->d_lock);
908 * The dispose list is isolated and dentries are not accounted
909 * to the LRU here, so we can simply remove it from the list
910 * here regardless of whether it is referenced or not.
912 d_shrink_del(dentry);
915 * We found an inuse dentry which was not removed from
916 * the LRU because of laziness during lookup. Do not free it.
918 if (dentry->d_lockref.count) {
919 spin_unlock(&dentry->d_lock);
925 * If 'try_to_prune()' returns a dentry, it will
926 * be the same one we passed in, and d_lock will
927 * have been held the whole time, so it will not
928 * have been added to any other lists. We failed
929 * to get the inode lock.
931 * We just add it back to the shrink list.
933 dentry = try_prune_one_dentry(dentry);
937 d_shrink_add(dentry, list);
938 spin_unlock(&dentry->d_lock);
944 static enum lru_status
945 dentry_lru_isolate(struct list_head *item, spinlock_t *lru_lock, void *arg)
947 struct list_head *freeable = arg;
948 struct dentry *dentry = container_of(item, struct dentry, d_lru);
952 * we are inverting the lru lock/dentry->d_lock here,
953 * so use a trylock. If we fail to get the lock, just skip
956 if (!spin_trylock(&dentry->d_lock))
960 * Referenced dentries are still in use. If they have active
961 * counts, just remove them from the LRU. Otherwise give them
962 * another pass through the LRU.
964 if (dentry->d_lockref.count) {
965 d_lru_isolate(dentry);
966 spin_unlock(&dentry->d_lock);
970 if (dentry->d_flags & DCACHE_REFERENCED) {
971 dentry->d_flags &= ~DCACHE_REFERENCED;
972 spin_unlock(&dentry->d_lock);
975 * The list move itself will be made by the common LRU code. At
976 * this point, we've dropped the dentry->d_lock but keep the
977 * lru lock. This is safe to do, since every list movement is
978 * protected by the lru lock even if both locks are held.
980 * This is guaranteed by the fact that all LRU management
981 * functions are intermediated by the LRU API calls like
982 * list_lru_add and list_lru_del. List movement in this file
983 * only ever occur through this functions or through callbacks
984 * like this one, that are called from the LRU API.
986 * The only exceptions to this are functions like
987 * shrink_dentry_list, and code that first checks for the
988 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
989 * operating only with stack provided lists after they are
990 * properly isolated from the main list. It is thus, always a
996 d_lru_shrink_move(dentry, freeable);
997 spin_unlock(&dentry->d_lock);
1003 * prune_dcache_sb - shrink the dcache
1005 * @nr_to_scan : number of entries to try to free
1006 * @nid: which node to scan for freeable entities
1008 * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is
1009 * done when we need more memory an called from the superblock shrinker
1012 * This function may fail to free any resources if all the dentries are in
1015 long prune_dcache_sb(struct super_block *sb, unsigned long nr_to_scan,
1021 freed = list_lru_walk_node(&sb->s_dentry_lru, nid, dentry_lru_isolate,
1022 &dispose, &nr_to_scan);
1023 shrink_dentry_list(&dispose);
1027 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1028 spinlock_t *lru_lock, void *arg)
1030 struct list_head *freeable = arg;
1031 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1034 * we are inverting the lru lock/dentry->d_lock here,
1035 * so use a trylock. If we fail to get the lock, just skip
1038 if (!spin_trylock(&dentry->d_lock))
1041 d_lru_shrink_move(dentry, freeable);
1042 spin_unlock(&dentry->d_lock);
1049 * shrink_dcache_sb - shrink dcache for a superblock
1052 * Shrink the dcache for the specified super block. This is used to free
1053 * the dcache before unmounting a file system.
1055 void shrink_dcache_sb(struct super_block *sb)
1062 freed = list_lru_walk(&sb->s_dentry_lru,
1063 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1065 this_cpu_sub(nr_dentry_unused, freed);
1066 shrink_dentry_list(&dispose);
1067 } while (freed > 0);
1069 EXPORT_SYMBOL(shrink_dcache_sb);
1072 * This tries to ascend one level of parenthood, but
1073 * we can race with renaming, so we need to re-check
1074 * the parenthood after dropping the lock and check
1075 * that the sequence number still matches.
1077 static struct dentry *try_to_ascend(struct dentry *old, unsigned seq)
1079 struct dentry *new = old->d_parent;
1082 spin_unlock(&old->d_lock);
1083 spin_lock(&new->d_lock);
1086 * might go back up the wrong parent if we have had a rename
1089 if (new != old->d_parent ||
1090 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1091 need_seqretry(&rename_lock, seq)) {
1092 spin_unlock(&new->d_lock);
1100 * enum d_walk_ret - action to talke during tree walk
1101 * @D_WALK_CONTINUE: contrinue walk
1102 * @D_WALK_QUIT: quit walk
1103 * @D_WALK_NORETRY: quit when retry is needed
1104 * @D_WALK_SKIP: skip this dentry and its children
1114 * d_walk - walk the dentry tree
1115 * @parent: start of walk
1116 * @data: data passed to @enter() and @finish()
1117 * @enter: callback when first entering the dentry
1118 * @finish: callback when successfully finished the walk
1120 * The @enter() and @finish() callbacks are called with d_lock held.
1122 static void d_walk(struct dentry *parent, void *data,
1123 enum d_walk_ret (*enter)(void *, struct dentry *),
1124 void (*finish)(void *))
1126 struct dentry *this_parent;
1127 struct list_head *next;
1129 enum d_walk_ret ret;
1133 read_seqbegin_or_lock(&rename_lock, &seq);
1134 this_parent = parent;
1135 spin_lock(&this_parent->d_lock);
1137 ret = enter(data, this_parent);
1139 case D_WALK_CONTINUE:
1144 case D_WALK_NORETRY:
1149 next = this_parent->d_subdirs.next;
1151 while (next != &this_parent->d_subdirs) {
1152 struct list_head *tmp = next;
1153 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1156 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1158 ret = enter(data, dentry);
1160 case D_WALK_CONTINUE:
1163 spin_unlock(&dentry->d_lock);
1165 case D_WALK_NORETRY:
1169 spin_unlock(&dentry->d_lock);
1173 if (!list_empty(&dentry->d_subdirs)) {
1174 spin_unlock(&this_parent->d_lock);
1175 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1176 this_parent = dentry;
1177 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1180 spin_unlock(&dentry->d_lock);
1183 * All done at this level ... ascend and resume the search.
1185 if (this_parent != parent) {
1186 struct dentry *child = this_parent;
1187 this_parent = try_to_ascend(this_parent, seq);
1190 next = child->d_u.d_child.next;
1193 if (need_seqretry(&rename_lock, seq)) {
1194 spin_unlock(&this_parent->d_lock);
1201 spin_unlock(&this_parent->d_lock);
1202 done_seqretry(&rename_lock, seq);
1213 * Search for at least 1 mount point in the dentry's subdirs.
1214 * We descend to the next level whenever the d_subdirs
1215 * list is non-empty and continue searching.
1219 * have_submounts - check for mounts over a dentry
1220 * @parent: dentry to check.
1222 * Return true if the parent or its subdirectories contain
1226 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1229 if (d_mountpoint(dentry)) {
1233 return D_WALK_CONTINUE;
1236 int have_submounts(struct dentry *parent)
1240 d_walk(parent, &ret, check_mount, NULL);
1244 EXPORT_SYMBOL(have_submounts);
1247 * Called by mount code to set a mountpoint and check if the mountpoint is
1248 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1249 * subtree can become unreachable).
1251 * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For
1252 * this reason take rename_lock and d_lock on dentry and ancestors.
1254 int d_set_mounted(struct dentry *dentry)
1258 write_seqlock(&rename_lock);
1259 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1260 /* Need exclusion wrt. check_submounts_and_drop() */
1261 spin_lock(&p->d_lock);
1262 if (unlikely(d_unhashed(p))) {
1263 spin_unlock(&p->d_lock);
1266 spin_unlock(&p->d_lock);
1268 spin_lock(&dentry->d_lock);
1269 if (!d_unlinked(dentry)) {
1270 dentry->d_flags |= DCACHE_MOUNTED;
1273 spin_unlock(&dentry->d_lock);
1275 write_sequnlock(&rename_lock);
1280 * Search the dentry child list of the specified parent,
1281 * and move any unused dentries to the end of the unused
1282 * list for prune_dcache(). We descend to the next level
1283 * whenever the d_subdirs list is non-empty and continue
1286 * It returns zero iff there are no unused children,
1287 * otherwise it returns the number of children moved to
1288 * the end of the unused list. This may not be the total
1289 * number of unused children, because select_parent can
1290 * drop the lock and return early due to latency
1294 struct select_data {
1295 struct dentry *start;
1296 struct list_head dispose;
1300 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1302 struct select_data *data = _data;
1303 enum d_walk_ret ret = D_WALK_CONTINUE;
1305 if (data->start == dentry)
1309 * move only zero ref count dentries to the dispose list.
1311 * Those which are presently on the shrink list, being processed
1312 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1313 * loop in shrink_dcache_parent() might not make any progress
1316 if (dentry->d_lockref.count) {
1317 dentry_lru_del(dentry);
1318 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1320 * We can't use d_lru_shrink_move() because we
1321 * need to get the global LRU lock and do the
1325 d_shrink_add(dentry, &data->dispose);
1327 ret = D_WALK_NORETRY;
1330 * We can return to the caller if we have found some (this
1331 * ensures forward progress). We'll be coming back to find
1334 if (data->found && need_resched())
1341 * shrink_dcache_parent - prune dcache
1342 * @parent: parent of entries to prune
1344 * Prune the dcache to remove unused children of the parent dentry.
1346 void shrink_dcache_parent(struct dentry *parent)
1349 struct select_data data;
1351 INIT_LIST_HEAD(&data.dispose);
1352 data.start = parent;
1355 d_walk(parent, &data, select_collect, NULL);
1359 shrink_dentry_list(&data.dispose);
1363 EXPORT_SYMBOL(shrink_dcache_parent);
1365 static enum d_walk_ret umount_collect(void *_data, struct dentry *dentry)
1367 struct select_data *data = _data;
1368 enum d_walk_ret ret = D_WALK_CONTINUE;
1370 if (dentry->d_lockref.count) {
1371 dentry_lru_del(dentry);
1372 if (likely(!list_empty(&dentry->d_subdirs)))
1374 if (dentry == data->start && dentry->d_lockref.count == 1)
1377 "BUG: Dentry %p{i=%lx,n=%s}"
1378 " still in use (%d)"
1379 " [unmount of %s %s]\n",
1382 dentry->d_inode->i_ino : 0UL,
1383 dentry->d_name.name,
1384 dentry->d_lockref.count,
1385 dentry->d_sb->s_type->name,
1386 dentry->d_sb->s_id);
1388 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1390 * We can't use d_lru_shrink_move() because we
1391 * need to get the global LRU lock and do the
1394 if (dentry->d_flags & DCACHE_LRU_LIST)
1396 d_shrink_add(dentry, &data->dispose);
1398 ret = D_WALK_NORETRY;
1401 if (data->found && need_resched())
1407 * destroy the dentries attached to a superblock on unmounting
1409 void shrink_dcache_for_umount(struct super_block *sb)
1411 struct dentry *dentry;
1413 if (down_read_trylock(&sb->s_umount))
1416 dentry = sb->s_root;
1419 struct select_data data;
1421 INIT_LIST_HEAD(&data.dispose);
1422 data.start = dentry;
1425 d_walk(dentry, &data, umount_collect, NULL);
1429 shrink_dentry_list(&data.dispose);
1435 while (!hlist_bl_empty(&sb->s_anon)) {
1436 struct select_data data;
1437 dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1439 INIT_LIST_HEAD(&data.dispose);
1443 d_walk(dentry, &data, umount_collect, NULL);
1445 shrink_dentry_list(&data.dispose);
1450 static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry)
1452 struct select_data *data = _data;
1454 if (d_mountpoint(dentry)) {
1455 data->found = -EBUSY;
1459 return select_collect(_data, dentry);
1462 static void check_and_drop(void *_data)
1464 struct select_data *data = _data;
1466 if (d_mountpoint(data->start))
1467 data->found = -EBUSY;
1469 __d_drop(data->start);
1473 * check_submounts_and_drop - prune dcache, check for submounts and drop
1475 * All done as a single atomic operation relative to has_unlinked_ancestor().
1476 * Returns 0 if successfully unhashed @parent. If there were submounts then
1479 * @dentry: dentry to prune and drop
1481 int check_submounts_and_drop(struct dentry *dentry)
1485 /* Negative dentries can be dropped without further checks */
1486 if (!dentry->d_inode) {
1492 struct select_data data;
1494 INIT_LIST_HEAD(&data.dispose);
1495 data.start = dentry;
1498 d_walk(dentry, &data, check_and_collect, check_and_drop);
1501 if (!list_empty(&data.dispose))
1502 shrink_dentry_list(&data.dispose);
1513 EXPORT_SYMBOL(check_submounts_and_drop);
1516 * __d_alloc - allocate a dcache entry
1517 * @sb: filesystem it will belong to
1518 * @name: qstr of the name
1520 * Allocates a dentry. It returns %NULL if there is insufficient memory
1521 * available. On a success the dentry is returned. The name passed in is
1522 * copied and the copy passed in may be reused after this call.
1525 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1527 struct dentry *dentry;
1530 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1535 * We guarantee that the inline name is always NUL-terminated.
1536 * This way the memcpy() done by the name switching in rename
1537 * will still always have a NUL at the end, even if we might
1538 * be overwriting an internal NUL character
1540 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1541 if (name->len > DNAME_INLINE_LEN-1) {
1542 dname = kmalloc(name->len + 1, GFP_KERNEL);
1544 kmem_cache_free(dentry_cache, dentry);
1548 dname = dentry->d_iname;
1551 dentry->d_name.len = name->len;
1552 dentry->d_name.hash = name->hash;
1553 memcpy(dname, name->name, name->len);
1554 dname[name->len] = 0;
1556 /* Make sure we always see the terminating NUL character */
1558 dentry->d_name.name = dname;
1560 dentry->d_lockref.count = 1;
1561 dentry->d_flags = 0;
1562 spin_lock_init(&dentry->d_lock);
1563 seqcount_init(&dentry->d_seq);
1564 dentry->d_inode = NULL;
1565 dentry->d_parent = dentry;
1567 dentry->d_op = NULL;
1568 dentry->d_fsdata = NULL;
1569 INIT_HLIST_BL_NODE(&dentry->d_hash);
1570 INIT_LIST_HEAD(&dentry->d_lru);
1571 INIT_LIST_HEAD(&dentry->d_subdirs);
1572 INIT_HLIST_NODE(&dentry->d_alias);
1573 INIT_LIST_HEAD(&dentry->d_u.d_child);
1574 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1576 this_cpu_inc(nr_dentry);
1582 * d_alloc - allocate a dcache entry
1583 * @parent: parent of entry to allocate
1584 * @name: qstr of the name
1586 * Allocates a dentry. It returns %NULL if there is insufficient memory
1587 * available. On a success the dentry is returned. The name passed in is
1588 * copied and the copy passed in may be reused after this call.
1590 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1592 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1596 spin_lock(&parent->d_lock);
1598 * don't need child lock because it is not subject
1599 * to concurrency here
1601 __dget_dlock(parent);
1602 dentry->d_parent = parent;
1603 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1604 spin_unlock(&parent->d_lock);
1608 EXPORT_SYMBOL(d_alloc);
1610 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1612 struct dentry *dentry = __d_alloc(sb, name);
1614 dentry->d_flags |= DCACHE_DISCONNECTED;
1617 EXPORT_SYMBOL(d_alloc_pseudo);
1619 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1624 q.len = strlen(name);
1625 q.hash = full_name_hash(q.name, q.len);
1626 return d_alloc(parent, &q);
1628 EXPORT_SYMBOL(d_alloc_name);
1630 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1632 WARN_ON_ONCE(dentry->d_op);
1633 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1635 DCACHE_OP_REVALIDATE |
1636 DCACHE_OP_WEAK_REVALIDATE |
1637 DCACHE_OP_DELETE ));
1642 dentry->d_flags |= DCACHE_OP_HASH;
1644 dentry->d_flags |= DCACHE_OP_COMPARE;
1645 if (op->d_revalidate)
1646 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1647 if (op->d_weak_revalidate)
1648 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1650 dentry->d_flags |= DCACHE_OP_DELETE;
1652 dentry->d_flags |= DCACHE_OP_PRUNE;
1655 EXPORT_SYMBOL(d_set_d_op);
1657 static unsigned d_flags_for_inode(struct inode *inode)
1659 unsigned add_flags = DCACHE_FILE_TYPE;
1662 return DCACHE_MISS_TYPE;
1664 if (S_ISDIR(inode->i_mode)) {
1665 add_flags = DCACHE_DIRECTORY_TYPE;
1666 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1667 if (unlikely(!inode->i_op->lookup))
1668 add_flags = DCACHE_AUTODIR_TYPE;
1670 inode->i_opflags |= IOP_LOOKUP;
1672 } else if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1673 if (unlikely(inode->i_op->follow_link))
1674 add_flags = DCACHE_SYMLINK_TYPE;
1676 inode->i_opflags |= IOP_NOFOLLOW;
1679 if (unlikely(IS_AUTOMOUNT(inode)))
1680 add_flags |= DCACHE_NEED_AUTOMOUNT;
1684 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1686 unsigned add_flags = d_flags_for_inode(inode);
1688 spin_lock(&dentry->d_lock);
1689 dentry->d_flags &= ~DCACHE_ENTRY_TYPE;
1690 dentry->d_flags |= add_flags;
1692 hlist_add_head(&dentry->d_alias, &inode->i_dentry);
1693 dentry->d_inode = inode;
1694 dentry_rcuwalk_barrier(dentry);
1695 spin_unlock(&dentry->d_lock);
1696 fsnotify_d_instantiate(dentry, inode);
1700 * d_instantiate - fill in inode information for a dentry
1701 * @entry: dentry to complete
1702 * @inode: inode to attach to this dentry
1704 * Fill in inode information in the entry.
1706 * This turns negative dentries into productive full members
1709 * NOTE! This assumes that the inode count has been incremented
1710 * (or otherwise set) by the caller to indicate that it is now
1711 * in use by the dcache.
1714 void d_instantiate(struct dentry *entry, struct inode * inode)
1716 BUG_ON(!hlist_unhashed(&entry->d_alias));
1718 spin_lock(&inode->i_lock);
1719 __d_instantiate(entry, inode);
1721 spin_unlock(&inode->i_lock);
1722 security_d_instantiate(entry, inode);
1724 EXPORT_SYMBOL(d_instantiate);
1727 * d_instantiate_unique - instantiate a non-aliased dentry
1728 * @entry: dentry to instantiate
1729 * @inode: inode to attach to this dentry
1731 * Fill in inode information in the entry. On success, it returns NULL.
1732 * If an unhashed alias of "entry" already exists, then we return the
1733 * aliased dentry instead and drop one reference to inode.
1735 * Note that in order to avoid conflicts with rename() etc, the caller
1736 * had better be holding the parent directory semaphore.
1738 * This also assumes that the inode count has been incremented
1739 * (or otherwise set) by the caller to indicate that it is now
1740 * in use by the dcache.
1742 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1743 struct inode *inode)
1745 struct dentry *alias;
1746 int len = entry->d_name.len;
1747 const char *name = entry->d_name.name;
1748 unsigned int hash = entry->d_name.hash;
1751 __d_instantiate(entry, NULL);
1755 hlist_for_each_entry(alias, &inode->i_dentry, d_alias) {
1757 * Don't need alias->d_lock here, because aliases with
1758 * d_parent == entry->d_parent are not subject to name or
1759 * parent changes, because the parent inode i_mutex is held.
1761 if (alias->d_name.hash != hash)
1763 if (alias->d_parent != entry->d_parent)
1765 if (alias->d_name.len != len)
1767 if (dentry_cmp(alias, name, len))
1773 __d_instantiate(entry, inode);
1777 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1779 struct dentry *result;
1781 BUG_ON(!hlist_unhashed(&entry->d_alias));
1784 spin_lock(&inode->i_lock);
1785 result = __d_instantiate_unique(entry, inode);
1787 spin_unlock(&inode->i_lock);
1790 security_d_instantiate(entry, inode);
1794 BUG_ON(!d_unhashed(result));
1799 EXPORT_SYMBOL(d_instantiate_unique);
1802 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1803 * @entry: dentry to complete
1804 * @inode: inode to attach to this dentry
1806 * Fill in inode information in the entry. If a directory alias is found, then
1807 * return an error (and drop inode). Together with d_materialise_unique() this
1808 * guarantees that a directory inode may never have more than one alias.
1810 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1812 BUG_ON(!hlist_unhashed(&entry->d_alias));
1814 spin_lock(&inode->i_lock);
1815 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1816 spin_unlock(&inode->i_lock);
1820 __d_instantiate(entry, inode);
1821 spin_unlock(&inode->i_lock);
1822 security_d_instantiate(entry, inode);
1826 EXPORT_SYMBOL(d_instantiate_no_diralias);
1828 struct dentry *d_make_root(struct inode *root_inode)
1830 struct dentry *res = NULL;
1833 static const struct qstr name = QSTR_INIT("/", 1);
1835 res = __d_alloc(root_inode->i_sb, &name);
1837 d_instantiate(res, root_inode);
1843 EXPORT_SYMBOL(d_make_root);
1845 static struct dentry * __d_find_any_alias(struct inode *inode)
1847 struct dentry *alias;
1849 if (hlist_empty(&inode->i_dentry))
1851 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias);
1857 * d_find_any_alias - find any alias for a given inode
1858 * @inode: inode to find an alias for
1860 * If any aliases exist for the given inode, take and return a
1861 * reference for one of them. If no aliases exist, return %NULL.
1863 struct dentry *d_find_any_alias(struct inode *inode)
1867 spin_lock(&inode->i_lock);
1868 de = __d_find_any_alias(inode);
1869 spin_unlock(&inode->i_lock);
1872 EXPORT_SYMBOL(d_find_any_alias);
1875 * d_obtain_alias - find or allocate a dentry for a given inode
1876 * @inode: inode to allocate the dentry for
1878 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1879 * similar open by handle operations. The returned dentry may be anonymous,
1880 * or may have a full name (if the inode was already in the cache).
1882 * When called on a directory inode, we must ensure that the inode only ever
1883 * has one dentry. If a dentry is found, that is returned instead of
1884 * allocating a new one.
1886 * On successful return, the reference to the inode has been transferred
1887 * to the dentry. In case of an error the reference on the inode is released.
1888 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1889 * be passed in and will be the error will be propagate to the return value,
1890 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1892 struct dentry *d_obtain_alias(struct inode *inode)
1894 static const struct qstr anonstring = QSTR_INIT("/", 1);
1900 return ERR_PTR(-ESTALE);
1902 return ERR_CAST(inode);
1904 res = d_find_any_alias(inode);
1908 tmp = __d_alloc(inode->i_sb, &anonstring);
1910 res = ERR_PTR(-ENOMEM);
1914 spin_lock(&inode->i_lock);
1915 res = __d_find_any_alias(inode);
1917 spin_unlock(&inode->i_lock);
1922 /* attach a disconnected dentry */
1923 add_flags = d_flags_for_inode(inode) | DCACHE_DISCONNECTED;
1925 spin_lock(&tmp->d_lock);
1926 tmp->d_inode = inode;
1927 tmp->d_flags |= add_flags;
1928 hlist_add_head(&tmp->d_alias, &inode->i_dentry);
1929 hlist_bl_lock(&tmp->d_sb->s_anon);
1930 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1931 hlist_bl_unlock(&tmp->d_sb->s_anon);
1932 spin_unlock(&tmp->d_lock);
1933 spin_unlock(&inode->i_lock);
1934 security_d_instantiate(tmp, inode);
1939 if (res && !IS_ERR(res))
1940 security_d_instantiate(res, inode);
1944 EXPORT_SYMBOL(d_obtain_alias);
1947 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1948 * @inode: the inode which may have a disconnected dentry
1949 * @dentry: a negative dentry which we want to point to the inode.
1951 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1952 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1953 * and return it, else simply d_add the inode to the dentry and return NULL.
1955 * This is needed in the lookup routine of any filesystem that is exportable
1956 * (via knfsd) so that we can build dcache paths to directories effectively.
1958 * If a dentry was found and moved, then it is returned. Otherwise NULL
1959 * is returned. This matches the expected return value of ->lookup.
1961 * Cluster filesystems may call this function with a negative, hashed dentry.
1962 * In that case, we know that the inode will be a regular file, and also this
1963 * will only occur during atomic_open. So we need to check for the dentry
1964 * being already hashed only in the final case.
1966 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1968 struct dentry *new = NULL;
1971 return ERR_CAST(inode);
1973 if (inode && S_ISDIR(inode->i_mode)) {
1974 spin_lock(&inode->i_lock);
1975 new = __d_find_alias(inode, 1);
1977 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1978 spin_unlock(&inode->i_lock);
1979 security_d_instantiate(new, inode);
1980 d_move(new, dentry);
1983 /* already taking inode->i_lock, so d_add() by hand */
1984 __d_instantiate(dentry, inode);
1985 spin_unlock(&inode->i_lock);
1986 security_d_instantiate(dentry, inode);
1990 d_instantiate(dentry, inode);
1991 if (d_unhashed(dentry))
1996 EXPORT_SYMBOL(d_splice_alias);
1999 * d_add_ci - lookup or allocate new dentry with case-exact name
2000 * @inode: the inode case-insensitive lookup has found
2001 * @dentry: the negative dentry that was passed to the parent's lookup func
2002 * @name: the case-exact name to be associated with the returned dentry
2004 * This is to avoid filling the dcache with case-insensitive names to the
2005 * same inode, only the actual correct case is stored in the dcache for
2006 * case-insensitive filesystems.
2008 * For a case-insensitive lookup match and if the the case-exact dentry
2009 * already exists in in the dcache, use it and return it.
2011 * If no entry exists with the exact case name, allocate new dentry with
2012 * the exact case, and return the spliced entry.
2014 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2017 struct dentry *found;
2021 * First check if a dentry matching the name already exists,
2022 * if not go ahead and create it now.
2024 found = d_hash_and_lookup(dentry->d_parent, name);
2025 if (unlikely(IS_ERR(found)))
2028 new = d_alloc(dentry->d_parent, name);
2030 found = ERR_PTR(-ENOMEM);
2034 found = d_splice_alias(inode, new);
2043 * If a matching dentry exists, and it's not negative use it.
2045 * Decrement the reference count to balance the iget() done
2048 if (found->d_inode) {
2049 if (unlikely(found->d_inode != inode)) {
2050 /* This can't happen because bad inodes are unhashed. */
2051 BUG_ON(!is_bad_inode(inode));
2052 BUG_ON(!is_bad_inode(found->d_inode));
2059 * Negative dentry: instantiate it unless the inode is a directory and
2060 * already has a dentry.
2062 new = d_splice_alias(inode, found);
2073 EXPORT_SYMBOL(d_add_ci);
2076 * Do the slow-case of the dentry name compare.
2078 * Unlike the dentry_cmp() function, we need to atomically
2079 * load the name and length information, so that the
2080 * filesystem can rely on them, and can use the 'name' and
2081 * 'len' information without worrying about walking off the
2082 * end of memory etc.
2084 * Thus the read_seqcount_retry() and the "duplicate" info
2085 * in arguments (the low-level filesystem should not look
2086 * at the dentry inode or name contents directly, since
2087 * rename can change them while we're in RCU mode).
2089 enum slow_d_compare {
2095 static noinline enum slow_d_compare slow_dentry_cmp(
2096 const struct dentry *parent,
2097 struct dentry *dentry,
2099 const struct qstr *name)
2101 int tlen = dentry->d_name.len;
2102 const char *tname = dentry->d_name.name;
2104 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2106 return D_COMP_SEQRETRY;
2108 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2109 return D_COMP_NOMATCH;
2114 * __d_lookup_rcu - search for a dentry (racy, store-free)
2115 * @parent: parent dentry
2116 * @name: qstr of name we wish to find
2117 * @seqp: returns d_seq value at the point where the dentry was found
2118 * Returns: dentry, or NULL
2120 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2121 * resolution (store-free path walking) design described in
2122 * Documentation/filesystems/path-lookup.txt.
2124 * This is not to be used outside core vfs.
2126 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2127 * held, and rcu_read_lock held. The returned dentry must not be stored into
2128 * without taking d_lock and checking d_seq sequence count against @seq
2131 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2134 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2135 * the returned dentry, so long as its parent's seqlock is checked after the
2136 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2137 * is formed, giving integrity down the path walk.
2139 * NOTE! The caller *has* to check the resulting dentry against the sequence
2140 * number we've returned before using any of the resulting dentry state!
2142 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2143 const struct qstr *name,
2146 u64 hashlen = name->hash_len;
2147 const unsigned char *str = name->name;
2148 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2149 struct hlist_bl_node *node;
2150 struct dentry *dentry;
2153 * Note: There is significant duplication with __d_lookup_rcu which is
2154 * required to prevent single threaded performance regressions
2155 * especially on architectures where smp_rmb (in seqcounts) are costly.
2156 * Keep the two functions in sync.
2160 * The hash list is protected using RCU.
2162 * Carefully use d_seq when comparing a candidate dentry, to avoid
2163 * races with d_move().
2165 * It is possible that concurrent renames can mess up our list
2166 * walk here and result in missing our dentry, resulting in the
2167 * false-negative result. d_lookup() protects against concurrent
2168 * renames using rename_lock seqlock.
2170 * See Documentation/filesystems/path-lookup.txt for more details.
2172 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2177 * The dentry sequence count protects us from concurrent
2178 * renames, and thus protects parent and name fields.
2180 * The caller must perform a seqcount check in order
2181 * to do anything useful with the returned dentry.
2183 * NOTE! We do a "raw" seqcount_begin here. That means that
2184 * we don't wait for the sequence count to stabilize if it
2185 * is in the middle of a sequence change. If we do the slow
2186 * dentry compare, we will do seqretries until it is stable,
2187 * and if we end up with a successful lookup, we actually
2188 * want to exit RCU lookup anyway.
2190 seq = raw_seqcount_begin(&dentry->d_seq);
2191 if (dentry->d_parent != parent)
2193 if (d_unhashed(dentry))
2196 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2197 if (dentry->d_name.hash != hashlen_hash(hashlen))
2200 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2203 case D_COMP_NOMATCH:
2210 if (dentry->d_name.hash_len != hashlen)
2213 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2220 * d_lookup - search for a dentry
2221 * @parent: parent dentry
2222 * @name: qstr of name we wish to find
2223 * Returns: dentry, or NULL
2225 * d_lookup searches the children of the parent dentry for the name in
2226 * question. If the dentry is found its reference count is incremented and the
2227 * dentry is returned. The caller must use dput to free the entry when it has
2228 * finished using it. %NULL is returned if the dentry does not exist.
2230 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2232 struct dentry *dentry;
2236 seq = read_seqbegin(&rename_lock);
2237 dentry = __d_lookup(parent, name);
2240 } while (read_seqretry(&rename_lock, seq));
2243 EXPORT_SYMBOL(d_lookup);
2246 * __d_lookup - search for a dentry (racy)
2247 * @parent: parent dentry
2248 * @name: qstr of name we wish to find
2249 * Returns: dentry, or NULL
2251 * __d_lookup is like d_lookup, however it may (rarely) return a
2252 * false-negative result due to unrelated rename activity.
2254 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2255 * however it must be used carefully, eg. with a following d_lookup in
2256 * the case of failure.
2258 * __d_lookup callers must be commented.
2260 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2262 unsigned int len = name->len;
2263 unsigned int hash = name->hash;
2264 const unsigned char *str = name->name;
2265 struct hlist_bl_head *b = d_hash(parent, hash);
2266 struct hlist_bl_node *node;
2267 struct dentry *found = NULL;
2268 struct dentry *dentry;
2271 * Note: There is significant duplication with __d_lookup_rcu which is
2272 * required to prevent single threaded performance regressions
2273 * especially on architectures where smp_rmb (in seqcounts) are costly.
2274 * Keep the two functions in sync.
2278 * The hash list is protected using RCU.
2280 * Take d_lock when comparing a candidate dentry, to avoid races
2283 * It is possible that concurrent renames can mess up our list
2284 * walk here and result in missing our dentry, resulting in the
2285 * false-negative result. d_lookup() protects against concurrent
2286 * renames using rename_lock seqlock.
2288 * See Documentation/filesystems/path-lookup.txt for more details.
2292 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2294 if (dentry->d_name.hash != hash)
2297 spin_lock(&dentry->d_lock);
2298 if (dentry->d_parent != parent)
2300 if (d_unhashed(dentry))
2304 * It is safe to compare names since d_move() cannot
2305 * change the qstr (protected by d_lock).
2307 if (parent->d_flags & DCACHE_OP_COMPARE) {
2308 int tlen = dentry->d_name.len;
2309 const char *tname = dentry->d_name.name;
2310 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2313 if (dentry->d_name.len != len)
2315 if (dentry_cmp(dentry, str, len))
2319 dentry->d_lockref.count++;
2321 spin_unlock(&dentry->d_lock);
2324 spin_unlock(&dentry->d_lock);
2332 * d_hash_and_lookup - hash the qstr then search for a dentry
2333 * @dir: Directory to search in
2334 * @name: qstr of name we wish to find
2336 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2338 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2341 * Check for a fs-specific hash function. Note that we must
2342 * calculate the standard hash first, as the d_op->d_hash()
2343 * routine may choose to leave the hash value unchanged.
2345 name->hash = full_name_hash(name->name, name->len);
2346 if (dir->d_flags & DCACHE_OP_HASH) {
2347 int err = dir->d_op->d_hash(dir, name);
2348 if (unlikely(err < 0))
2349 return ERR_PTR(err);
2351 return d_lookup(dir, name);
2353 EXPORT_SYMBOL(d_hash_and_lookup);
2356 * d_validate - verify dentry provided from insecure source (deprecated)
2357 * @dentry: The dentry alleged to be valid child of @dparent
2358 * @dparent: The parent dentry (known to be valid)
2360 * An insecure source has sent us a dentry, here we verify it and dget() it.
2361 * This is used by ncpfs in its readdir implementation.
2362 * Zero is returned in the dentry is invalid.
2364 * This function is slow for big directories, and deprecated, do not use it.
2366 int d_validate(struct dentry *dentry, struct dentry *dparent)
2368 struct dentry *child;
2370 spin_lock(&dparent->d_lock);
2371 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2372 if (dentry == child) {
2373 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2374 __dget_dlock(dentry);
2375 spin_unlock(&dentry->d_lock);
2376 spin_unlock(&dparent->d_lock);
2380 spin_unlock(&dparent->d_lock);
2384 EXPORT_SYMBOL(d_validate);
2387 * When a file is deleted, we have two options:
2388 * - turn this dentry into a negative dentry
2389 * - unhash this dentry and free it.
2391 * Usually, we want to just turn this into
2392 * a negative dentry, but if anybody else is
2393 * currently using the dentry or the inode
2394 * we can't do that and we fall back on removing
2395 * it from the hash queues and waiting for
2396 * it to be deleted later when it has no users
2400 * d_delete - delete a dentry
2401 * @dentry: The dentry to delete
2403 * Turn the dentry into a negative dentry if possible, otherwise
2404 * remove it from the hash queues so it can be deleted later
2407 void d_delete(struct dentry * dentry)
2409 struct inode *inode;
2412 * Are we the only user?
2415 spin_lock(&dentry->d_lock);
2416 inode = dentry->d_inode;
2417 isdir = S_ISDIR(inode->i_mode);
2418 if (dentry->d_lockref.count == 1) {
2419 if (!spin_trylock(&inode->i_lock)) {
2420 spin_unlock(&dentry->d_lock);
2424 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2425 dentry_unlink_inode(dentry);
2426 fsnotify_nameremove(dentry, isdir);
2430 if (!d_unhashed(dentry))
2433 spin_unlock(&dentry->d_lock);
2435 fsnotify_nameremove(dentry, isdir);
2437 EXPORT_SYMBOL(d_delete);
2439 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2441 BUG_ON(!d_unhashed(entry));
2443 entry->d_flags |= DCACHE_RCUACCESS;
2444 hlist_bl_add_head_rcu(&entry->d_hash, b);
2448 static void _d_rehash(struct dentry * entry)
2450 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2454 * d_rehash - add an entry back to the hash
2455 * @entry: dentry to add to the hash
2457 * Adds a dentry to the hash according to its name.
2460 void d_rehash(struct dentry * entry)
2462 spin_lock(&entry->d_lock);
2464 spin_unlock(&entry->d_lock);
2466 EXPORT_SYMBOL(d_rehash);
2469 * dentry_update_name_case - update case insensitive dentry with a new name
2470 * @dentry: dentry to be updated
2473 * Update a case insensitive dentry with new case of name.
2475 * dentry must have been returned by d_lookup with name @name. Old and new
2476 * name lengths must match (ie. no d_compare which allows mismatched name
2479 * Parent inode i_mutex must be held over d_lookup and into this call (to
2480 * keep renames and concurrent inserts, and readdir(2) away).
2482 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2484 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2485 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2487 spin_lock(&dentry->d_lock);
2488 write_seqcount_begin(&dentry->d_seq);
2489 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2490 write_seqcount_end(&dentry->d_seq);
2491 spin_unlock(&dentry->d_lock);
2493 EXPORT_SYMBOL(dentry_update_name_case);
2495 static void switch_names(struct dentry *dentry, struct dentry *target)
2497 if (dname_external(target)) {
2498 if (dname_external(dentry)) {
2500 * Both external: swap the pointers
2502 swap(target->d_name.name, dentry->d_name.name);
2505 * dentry:internal, target:external. Steal target's
2506 * storage and make target internal.
2508 memcpy(target->d_iname, dentry->d_name.name,
2509 dentry->d_name.len + 1);
2510 dentry->d_name.name = target->d_name.name;
2511 target->d_name.name = target->d_iname;
2514 if (dname_external(dentry)) {
2516 * dentry:external, target:internal. Give dentry's
2517 * storage to target and make dentry internal
2519 memcpy(dentry->d_iname, target->d_name.name,
2520 target->d_name.len + 1);
2521 target->d_name.name = dentry->d_name.name;
2522 dentry->d_name.name = dentry->d_iname;
2525 * Both are internal. Just copy target to dentry
2527 memcpy(dentry->d_iname, target->d_name.name,
2528 target->d_name.len + 1);
2529 dentry->d_name.len = target->d_name.len;
2533 swap(dentry->d_name.len, target->d_name.len);
2536 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2539 * XXXX: do we really need to take target->d_lock?
2541 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2542 spin_lock(&target->d_parent->d_lock);
2544 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2545 spin_lock(&dentry->d_parent->d_lock);
2546 spin_lock_nested(&target->d_parent->d_lock,
2547 DENTRY_D_LOCK_NESTED);
2549 spin_lock(&target->d_parent->d_lock);
2550 spin_lock_nested(&dentry->d_parent->d_lock,
2551 DENTRY_D_LOCK_NESTED);
2554 if (target < dentry) {
2555 spin_lock_nested(&target->d_lock, 2);
2556 spin_lock_nested(&dentry->d_lock, 3);
2558 spin_lock_nested(&dentry->d_lock, 2);
2559 spin_lock_nested(&target->d_lock, 3);
2563 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2564 struct dentry *target)
2566 if (target->d_parent != dentry->d_parent)
2567 spin_unlock(&dentry->d_parent->d_lock);
2568 if (target->d_parent != target)
2569 spin_unlock(&target->d_parent->d_lock);
2573 * When switching names, the actual string doesn't strictly have to
2574 * be preserved in the target - because we're dropping the target
2575 * anyway. As such, we can just do a simple memcpy() to copy over
2576 * the new name before we switch.
2578 * Note that we have to be a lot more careful about getting the hash
2579 * switched - we have to switch the hash value properly even if it
2580 * then no longer matches the actual (corrupted) string of the target.
2581 * The hash value has to match the hash queue that the dentry is on..
2584 * __d_move - move a dentry
2585 * @dentry: entry to move
2586 * @target: new dentry
2588 * Update the dcache to reflect the move of a file name. Negative
2589 * dcache entries should not be moved in this way. Caller must hold
2590 * rename_lock, the i_mutex of the source and target directories,
2591 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2593 static void __d_move(struct dentry * dentry, struct dentry * target)
2595 if (!dentry->d_inode)
2596 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2598 BUG_ON(d_ancestor(dentry, target));
2599 BUG_ON(d_ancestor(target, dentry));
2601 dentry_lock_for_move(dentry, target);
2603 write_seqcount_begin(&dentry->d_seq);
2604 write_seqcount_begin(&target->d_seq);
2606 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2609 * Move the dentry to the target hash queue. Don't bother checking
2610 * for the same hash queue because of how unlikely it is.
2613 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2615 /* Unhash the target: dput() will then get rid of it */
2618 list_del(&dentry->d_u.d_child);
2619 list_del(&target->d_u.d_child);
2621 /* Switch the names.. */
2622 switch_names(dentry, target);
2623 swap(dentry->d_name.hash, target->d_name.hash);
2625 /* ... and switch the parents */
2626 if (IS_ROOT(dentry)) {
2627 dentry->d_parent = target->d_parent;
2628 target->d_parent = target;
2629 INIT_LIST_HEAD(&target->d_u.d_child);
2631 swap(dentry->d_parent, target->d_parent);
2633 /* And add them back to the (new) parent lists */
2634 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2637 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2639 write_seqcount_end(&target->d_seq);
2640 write_seqcount_end(&dentry->d_seq);
2642 dentry_unlock_parents_for_move(dentry, target);
2643 spin_unlock(&target->d_lock);
2644 fsnotify_d_move(dentry);
2645 spin_unlock(&dentry->d_lock);
2649 * d_move - move a dentry
2650 * @dentry: entry to move
2651 * @target: new dentry
2653 * Update the dcache to reflect the move of a file name. Negative
2654 * dcache entries should not be moved in this way. See the locking
2655 * requirements for __d_move.
2657 void d_move(struct dentry *dentry, struct dentry *target)
2659 write_seqlock(&rename_lock);
2660 __d_move(dentry, target);
2661 write_sequnlock(&rename_lock);
2663 EXPORT_SYMBOL(d_move);
2666 * d_ancestor - search for an ancestor
2667 * @p1: ancestor dentry
2670 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2671 * an ancestor of p2, else NULL.
2673 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2677 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2678 if (p->d_parent == p1)
2685 * This helper attempts to cope with remotely renamed directories
2687 * It assumes that the caller is already holding
2688 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2690 * Note: If ever the locking in lock_rename() changes, then please
2691 * remember to update this too...
2693 static struct dentry *__d_unalias(struct inode *inode,
2694 struct dentry *dentry, struct dentry *alias)
2696 struct mutex *m1 = NULL, *m2 = NULL;
2697 struct dentry *ret = ERR_PTR(-EBUSY);
2699 /* If alias and dentry share a parent, then no extra locks required */
2700 if (alias->d_parent == dentry->d_parent)
2703 /* See lock_rename() */
2704 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2706 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2707 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2709 m2 = &alias->d_parent->d_inode->i_mutex;
2711 if (likely(!d_mountpoint(alias))) {
2712 __d_move(alias, dentry);
2716 spin_unlock(&inode->i_lock);
2725 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2726 * named dentry in place of the dentry to be replaced.
2727 * returns with anon->d_lock held!
2729 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2731 struct dentry *dparent;
2733 dentry_lock_for_move(anon, dentry);
2735 write_seqcount_begin(&dentry->d_seq);
2736 write_seqcount_begin(&anon->d_seq);
2738 dparent = dentry->d_parent;
2740 switch_names(dentry, anon);
2741 swap(dentry->d_name.hash, anon->d_name.hash);
2743 dentry->d_parent = dentry;
2744 list_del_init(&dentry->d_u.d_child);
2745 anon->d_parent = dparent;
2746 list_move(&anon->d_u.d_child, &dparent->d_subdirs);
2748 write_seqcount_end(&dentry->d_seq);
2749 write_seqcount_end(&anon->d_seq);
2751 dentry_unlock_parents_for_move(anon, dentry);
2752 spin_unlock(&dentry->d_lock);
2754 /* anon->d_lock still locked, returns locked */
2755 anon->d_flags &= ~DCACHE_DISCONNECTED;
2759 * d_materialise_unique - introduce an inode into the tree
2760 * @dentry: candidate dentry
2761 * @inode: inode to bind to the dentry, to which aliases may be attached
2763 * Introduces an dentry into the tree, substituting an extant disconnected
2764 * root directory alias in its place if there is one. Caller must hold the
2765 * i_mutex of the parent directory.
2767 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2769 struct dentry *actual;
2771 BUG_ON(!d_unhashed(dentry));
2775 __d_instantiate(dentry, NULL);
2780 spin_lock(&inode->i_lock);
2782 if (S_ISDIR(inode->i_mode)) {
2783 struct dentry *alias;
2785 /* Does an aliased dentry already exist? */
2786 alias = __d_find_alias(inode, 0);
2789 write_seqlock(&rename_lock);
2791 if (d_ancestor(alias, dentry)) {
2792 /* Check for loops */
2793 actual = ERR_PTR(-ELOOP);
2794 spin_unlock(&inode->i_lock);
2795 } else if (IS_ROOT(alias)) {
2796 /* Is this an anonymous mountpoint that we
2797 * could splice into our tree? */
2798 __d_materialise_dentry(dentry, alias);
2799 write_sequnlock(&rename_lock);
2803 /* Nope, but we must(!) avoid directory
2804 * aliasing. This drops inode->i_lock */
2805 actual = __d_unalias(inode, dentry, alias);
2807 write_sequnlock(&rename_lock);
2808 if (IS_ERR(actual)) {
2809 if (PTR_ERR(actual) == -ELOOP)
2810 pr_warn_ratelimited(
2811 "VFS: Lookup of '%s' in %s %s"
2812 " would have caused loop\n",
2813 dentry->d_name.name,
2814 inode->i_sb->s_type->name,
2822 /* Add a unique reference */
2823 actual = __d_instantiate_unique(dentry, inode);
2827 BUG_ON(!d_unhashed(actual));
2829 spin_lock(&actual->d_lock);
2832 spin_unlock(&actual->d_lock);
2833 spin_unlock(&inode->i_lock);
2835 if (actual == dentry) {
2836 security_d_instantiate(dentry, inode);
2843 EXPORT_SYMBOL_GPL(d_materialise_unique);
2845 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2849 return -ENAMETOOLONG;
2851 memcpy(*buffer, str, namelen);
2856 * prepend_name - prepend a pathname in front of current buffer pointer
2857 * @buffer: buffer pointer
2858 * @buflen: allocated length of the buffer
2859 * @name: name string and length qstr structure
2861 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2862 * make sure that either the old or the new name pointer and length are
2863 * fetched. However, there may be mismatch between length and pointer.
2864 * The length cannot be trusted, we need to copy it byte-by-byte until
2865 * the length is reached or a null byte is found. It also prepends "/" at
2866 * the beginning of the name. The sequence number check at the caller will
2867 * retry it again when a d_move() does happen. So any garbage in the buffer
2868 * due to mismatched pointer and length will be discarded.
2870 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2872 const char *dname = ACCESS_ONCE(name->name);
2873 u32 dlen = ACCESS_ONCE(name->len);
2876 if (*buflen < dlen + 1)
2877 return -ENAMETOOLONG;
2878 *buflen -= dlen + 1;
2879 p = *buffer -= dlen + 1;
2891 * prepend_path - Prepend path string to a buffer
2892 * @path: the dentry/vfsmount to report
2893 * @root: root vfsmnt/dentry
2894 * @buffer: pointer to the end of the buffer
2895 * @buflen: pointer to buffer length
2897 * The function will first try to write out the pathname without taking any
2898 * lock other than the RCU read lock to make sure that dentries won't go away.
2899 * It only checks the sequence number of the global rename_lock as any change
2900 * in the dentry's d_seq will be preceded by changes in the rename_lock
2901 * sequence number. If the sequence number had been changed, it will restart
2902 * the whole pathname back-tracing sequence again by taking the rename_lock.
2903 * In this case, there is no need to take the RCU read lock as the recursive
2904 * parent pointer references will keep the dentry chain alive as long as no
2905 * rename operation is performed.
2907 static int prepend_path(const struct path *path,
2908 const struct path *root,
2909 char **buffer, int *buflen)
2911 struct dentry *dentry = path->dentry;
2912 struct vfsmount *vfsmnt = path->mnt;
2913 struct mount *mnt = real_mount(vfsmnt);
2915 unsigned seq, m_seq = 0;
2921 read_seqbegin_or_lock(&mount_lock, &m_seq);
2927 read_seqbegin_or_lock(&rename_lock, &seq);
2928 while (dentry != root->dentry || vfsmnt != root->mnt) {
2929 struct dentry * parent;
2931 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2932 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2934 if (mnt != parent) {
2935 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2941 * Filesystems needing to implement special "root names"
2942 * should do so with ->d_dname()
2944 if (IS_ROOT(dentry) &&
2945 (dentry->d_name.len != 1 ||
2946 dentry->d_name.name[0] != '/')) {
2947 WARN(1, "Root dentry has weird name <%.*s>\n",
2948 (int) dentry->d_name.len,
2949 dentry->d_name.name);
2952 error = is_mounted(vfsmnt) ? 1 : 2;
2955 parent = dentry->d_parent;
2957 error = prepend_name(&bptr, &blen, &dentry->d_name);
2965 if (need_seqretry(&rename_lock, seq)) {
2969 done_seqretry(&rename_lock, seq);
2970 if (need_seqretry(&mount_lock, m_seq)) {
2974 done_seqretry(&mount_lock, m_seq);
2976 if (error >= 0 && bptr == *buffer) {
2978 error = -ENAMETOOLONG;
2988 * __d_path - return the path of a dentry
2989 * @path: the dentry/vfsmount to report
2990 * @root: root vfsmnt/dentry
2991 * @buf: buffer to return value in
2992 * @buflen: buffer length
2994 * Convert a dentry into an ASCII path name.
2996 * Returns a pointer into the buffer or an error code if the
2997 * path was too long.
2999 * "buflen" should be positive.
3001 * If the path is not reachable from the supplied root, return %NULL.
3003 char *__d_path(const struct path *path,
3004 const struct path *root,
3005 char *buf, int buflen)
3007 char *res = buf + buflen;
3010 prepend(&res, &buflen, "\0", 1);
3011 error = prepend_path(path, root, &res, &buflen);
3014 return ERR_PTR(error);
3020 char *d_absolute_path(const struct path *path,
3021 char *buf, int buflen)
3023 struct path root = {};
3024 char *res = buf + buflen;
3027 prepend(&res, &buflen, "\0", 1);
3028 error = prepend_path(path, &root, &res, &buflen);
3033 return ERR_PTR(error);
3038 * same as __d_path but appends "(deleted)" for unlinked files.
3040 static int path_with_deleted(const struct path *path,
3041 const struct path *root,
3042 char **buf, int *buflen)
3044 prepend(buf, buflen, "\0", 1);
3045 if (d_unlinked(path->dentry)) {
3046 int error = prepend(buf, buflen, " (deleted)", 10);
3051 return prepend_path(path, root, buf, buflen);
3054 static int prepend_unreachable(char **buffer, int *buflen)
3056 return prepend(buffer, buflen, "(unreachable)", 13);
3059 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3064 seq = read_seqcount_begin(&fs->seq);
3066 } while (read_seqcount_retry(&fs->seq, seq));
3070 * d_path - return the path of a dentry
3071 * @path: path to report
3072 * @buf: buffer to return value in
3073 * @buflen: buffer length
3075 * Convert a dentry into an ASCII path name. If the entry has been deleted
3076 * the string " (deleted)" is appended. Note that this is ambiguous.
3078 * Returns a pointer into the buffer or an error code if the path was
3079 * too long. Note: Callers should use the returned pointer, not the passed
3080 * in buffer, to use the name! The implementation often starts at an offset
3081 * into the buffer, and may leave 0 bytes at the start.
3083 * "buflen" should be positive.
3085 char *d_path(const struct path *path, char *buf, int buflen)
3087 char *res = buf + buflen;
3092 * We have various synthetic filesystems that never get mounted. On
3093 * these filesystems dentries are never used for lookup purposes, and
3094 * thus don't need to be hashed. They also don't need a name until a
3095 * user wants to identify the object in /proc/pid/fd/. The little hack
3096 * below allows us to generate a name for these objects on demand:
3098 if (path->dentry->d_op && path->dentry->d_op->d_dname)
3099 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3102 get_fs_root_rcu(current->fs, &root);
3103 error = path_with_deleted(path, &root, &res, &buflen);
3107 res = ERR_PTR(error);
3110 EXPORT_SYMBOL(d_path);
3113 * Helper function for dentry_operations.d_dname() members
3115 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3116 const char *fmt, ...)
3122 va_start(args, fmt);
3123 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3126 if (sz > sizeof(temp) || sz > buflen)
3127 return ERR_PTR(-ENAMETOOLONG);
3129 buffer += buflen - sz;
3130 return memcpy(buffer, temp, sz);
3133 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3135 char *end = buffer + buflen;
3136 /* these dentries are never renamed, so d_lock is not needed */
3137 if (prepend(&end, &buflen, " (deleted)", 11) ||
3138 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3139 prepend(&end, &buflen, "/", 1))
3140 end = ERR_PTR(-ENAMETOOLONG);
3145 * Write full pathname from the root of the filesystem into the buffer.
3147 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
3157 prepend(&end, &len, "\0", 1);
3163 read_seqbegin_or_lock(&rename_lock, &seq);
3164 while (!IS_ROOT(dentry)) {
3165 struct dentry *parent = dentry->d_parent;
3169 error = prepend_name(&end, &len, &dentry->d_name);
3178 if (need_seqretry(&rename_lock, seq)) {
3182 done_seqretry(&rename_lock, seq);
3187 return ERR_PTR(-ENAMETOOLONG);
3190 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3192 return __dentry_path(dentry, buf, buflen);
3194 EXPORT_SYMBOL(dentry_path_raw);
3196 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3201 if (d_unlinked(dentry)) {
3203 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3207 retval = __dentry_path(dentry, buf, buflen);
3208 if (!IS_ERR(retval) && p)
3209 *p = '/'; /* restore '/' overriden with '\0' */
3212 return ERR_PTR(-ENAMETOOLONG);
3215 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3221 seq = read_seqcount_begin(&fs->seq);
3224 } while (read_seqcount_retry(&fs->seq, seq));
3228 * NOTE! The user-level library version returns a
3229 * character pointer. The kernel system call just
3230 * returns the length of the buffer filled (which
3231 * includes the ending '\0' character), or a negative
3232 * error value. So libc would do something like
3234 * char *getcwd(char * buf, size_t size)
3238 * retval = sys_getcwd(buf, size);
3245 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3248 struct path pwd, root;
3249 char *page = __getname();
3255 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3258 if (!d_unlinked(pwd.dentry)) {
3260 char *cwd = page + PATH_MAX;
3261 int buflen = PATH_MAX;
3263 prepend(&cwd, &buflen, "\0", 1);
3264 error = prepend_path(&pwd, &root, &cwd, &buflen);
3270 /* Unreachable from current root */
3272 error = prepend_unreachable(&cwd, &buflen);
3278 len = PATH_MAX + page - cwd;
3281 if (copy_to_user(buf, cwd, len))
3294 * Test whether new_dentry is a subdirectory of old_dentry.
3296 * Trivially implemented using the dcache structure
3300 * is_subdir - is new dentry a subdirectory of old_dentry
3301 * @new_dentry: new dentry
3302 * @old_dentry: old dentry
3304 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3305 * Returns 0 otherwise.
3306 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3309 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3314 if (new_dentry == old_dentry)
3318 /* for restarting inner loop in case of seq retry */
3319 seq = read_seqbegin(&rename_lock);
3321 * Need rcu_readlock to protect against the d_parent trashing
3325 if (d_ancestor(old_dentry, new_dentry))
3330 } while (read_seqretry(&rename_lock, seq));
3335 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3337 struct dentry *root = data;
3338 if (dentry != root) {
3339 if (d_unhashed(dentry) || !dentry->d_inode)
3342 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3343 dentry->d_flags |= DCACHE_GENOCIDE;
3344 dentry->d_lockref.count--;
3347 return D_WALK_CONTINUE;
3350 void d_genocide(struct dentry *parent)
3352 d_walk(parent, parent, d_genocide_kill, NULL);
3355 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3357 inode_dec_link_count(inode);
3358 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3359 !hlist_unhashed(&dentry->d_alias) ||
3360 !d_unlinked(dentry));
3361 spin_lock(&dentry->d_parent->d_lock);
3362 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3363 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3364 (unsigned long long)inode->i_ino);
3365 spin_unlock(&dentry->d_lock);
3366 spin_unlock(&dentry->d_parent->d_lock);
3367 d_instantiate(dentry, inode);
3369 EXPORT_SYMBOL(d_tmpfile);
3371 static __initdata unsigned long dhash_entries;
3372 static int __init set_dhash_entries(char *str)
3376 dhash_entries = simple_strtoul(str, &str, 0);
3379 __setup("dhash_entries=", set_dhash_entries);
3381 static void __init dcache_init_early(void)
3385 /* If hashes are distributed across NUMA nodes, defer
3386 * hash allocation until vmalloc space is available.
3392 alloc_large_system_hash("Dentry cache",
3393 sizeof(struct hlist_bl_head),
3402 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3403 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3406 static void __init dcache_init(void)
3411 * A constructor could be added for stable state like the lists,
3412 * but it is probably not worth it because of the cache nature
3415 dentry_cache = KMEM_CACHE(dentry,
3416 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3418 /* Hash may have been set up in dcache_init_early */
3423 alloc_large_system_hash("Dentry cache",
3424 sizeof(struct hlist_bl_head),
3433 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3434 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3437 /* SLAB cache for __getname() consumers */
3438 struct kmem_cache *names_cachep __read_mostly;
3439 EXPORT_SYMBOL(names_cachep);
3441 EXPORT_SYMBOL(d_genocide);
3443 void __init vfs_caches_init_early(void)
3445 dcache_init_early();
3449 void __init vfs_caches_init(unsigned long mempages)
3451 unsigned long reserve;
3453 /* Base hash sizes on available memory, with a reserve equal to
3454 150% of current kernel size */
3456 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3457 mempages -= reserve;
3459 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3460 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3464 files_init(mempages);