include/linux/dcache.h: remove semicolons from HASH_LEN_DECLARE
[linux/fpc-iii.git] / fs / dcache.c
blob8d38cd07b207b9234d0f571bb4a945e05a395c7b
1 /*
2 * fs/dcache.c
4 * Complete reimplementation
5 * (C) 1997 Thomas Schoebel-Theuer,
6 * with heavy changes by Linus Torvalds
7 */
9 /*
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>
19 #include <linux/mm.h>
20 #include <linux/fs.h>
21 #include <linux/fsnotify.h>
22 #include <linux/slab.h>
23 #include <linux/init.h>
24 #include <linux/hash.h>
25 #include <linux/cache.h>
26 #include <linux/export.h>
27 #include <linux/mount.h>
28 #include <linux/file.h>
29 #include <asm/uaccess.h>
30 #include <linux/security.h>
31 #include <linux/seqlock.h>
32 #include <linux/swap.h>
33 #include <linux/bootmem.h>
34 #include <linux/fs_struct.h>
35 #include <linux/hardirq.h>
36 #include <linux/bit_spinlock.h>
37 #include <linux/rculist_bl.h>
38 #include <linux/prefetch.h>
39 #include <linux/ratelimit.h>
40 #include <linux/list_lru.h>
41 #include <linux/kasan.h>
43 #include "internal.h"
44 #include "mount.h"
47 * Usage:
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
56 * d_lock protects:
57 * - d_flags
58 * - d_name
59 * - d_lru
60 * - d_count
61 * - d_unhashed()
62 * - d_parent and d_subdirs
63 * - childrens' d_child and d_parent
64 * - d_u.d_alias, d_inode
66 * Ordering:
67 * dentry->d_inode->i_lock
68 * dentry->d_lock
69 * dentry->d_sb->s_dentry_lru_lock
70 * dcache_hash_bucket lock
71 * s_anon lock
73 * If there is an ancestor relationship:
74 * dentry->d_parent->...->d_parent->d_lock
75 * ...
76 * dentry->d_parent->d_lock
77 * dentry->d_lock
79 * If no ancestor relationship:
80 * if (dentry1 < dentry2)
81 * dentry1->d_lock
82 * dentry2->d_lock
84 int sysctl_vfs_cache_pressure __read_mostly = 100;
85 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
87 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
89 EXPORT_SYMBOL(rename_lock);
91 static struct kmem_cache *dentry_cache __read_mostly;
94 * This is the single most critical data structure when it comes
95 * to the dcache: the hashtable for lookups. Somebody should try
96 * to make this good - I've just made it work.
98 * This hash-function tries to avoid losing too many bits of hash
99 * information, yet avoid using a prime hash-size or similar.
102 static unsigned int d_hash_mask __read_mostly;
103 static unsigned int d_hash_shift __read_mostly;
105 static struct hlist_bl_head *dentry_hashtable __read_mostly;
107 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
108 unsigned int hash)
110 hash += (unsigned long) parent / L1_CACHE_BYTES;
111 return dentry_hashtable + hash_32(hash, d_hash_shift);
114 /* Statistics gathering. */
115 struct dentry_stat_t dentry_stat = {
116 .age_limit = 45,
119 static DEFINE_PER_CPU(long, nr_dentry);
120 static DEFINE_PER_CPU(long, nr_dentry_unused);
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
125 * Here we resort to our own counters instead of using generic per-cpu counters
126 * for consistency with what the vfs inode code does. We are expected to harvest
127 * better code and performance by having our own specialized counters.
129 * Please note that the loop is done over all possible CPUs, not over all online
130 * CPUs. The reason for this is that we don't want to play games with CPUs going
131 * on and off. If one of them goes off, we will just keep their counters.
133 * glommer: See cffbc8a for details, and if you ever intend to change this,
134 * please update all vfs counters to match.
136 static long get_nr_dentry(void)
138 int i;
139 long sum = 0;
140 for_each_possible_cpu(i)
141 sum += per_cpu(nr_dentry, i);
142 return sum < 0 ? 0 : sum;
145 static long get_nr_dentry_unused(void)
147 int i;
148 long sum = 0;
149 for_each_possible_cpu(i)
150 sum += per_cpu(nr_dentry_unused, i);
151 return sum < 0 ? 0 : sum;
154 int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
155 size_t *lenp, loff_t *ppos)
157 dentry_stat.nr_dentry = get_nr_dentry();
158 dentry_stat.nr_unused = get_nr_dentry_unused();
159 return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
161 #endif
164 * Compare 2 name strings, return 0 if they match, otherwise non-zero.
165 * The strings are both count bytes long, and count is non-zero.
167 #ifdef CONFIG_DCACHE_WORD_ACCESS
169 #include <asm/word-at-a-time.h>
171 * NOTE! 'cs' and 'scount' come from a dentry, so it has a
172 * aligned allocation for this particular component. We don't
173 * strictly need the load_unaligned_zeropad() safety, but it
174 * doesn't hurt either.
176 * In contrast, 'ct' and 'tcount' can be from a pathname, and do
177 * need the careful unaligned handling.
179 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
181 unsigned long a,b,mask;
183 for (;;) {
184 a = *(unsigned long *)cs;
185 b = load_unaligned_zeropad(ct);
186 if (tcount < sizeof(unsigned long))
187 break;
188 if (unlikely(a != b))
189 return 1;
190 cs += sizeof(unsigned long);
191 ct += sizeof(unsigned long);
192 tcount -= sizeof(unsigned long);
193 if (!tcount)
194 return 0;
196 mask = bytemask_from_count(tcount);
197 return unlikely(!!((a ^ b) & mask));
200 #else
202 static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
204 do {
205 if (*cs != *ct)
206 return 1;
207 cs++;
208 ct++;
209 tcount--;
210 } while (tcount);
211 return 0;
214 #endif
216 static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
218 const unsigned char *cs;
220 * Be careful about RCU walk racing with rename:
221 * use ACCESS_ONCE to fetch the name pointer.
223 * NOTE! Even if a rename will mean that the length
224 * was not loaded atomically, we don't care. The
225 * RCU walk will check the sequence count eventually,
226 * and catch it. And we won't overrun the buffer,
227 * because we're reading the name pointer atomically,
228 * and a dentry name is guaranteed to be properly
229 * terminated with a NUL byte.
231 * End result: even if 'len' is wrong, we'll exit
232 * early because the data cannot match (there can
233 * be no NUL in the ct/tcount data)
235 cs = ACCESS_ONCE(dentry->d_name.name);
236 smp_read_barrier_depends();
237 return dentry_string_cmp(cs, ct, tcount);
240 struct external_name {
241 union {
242 atomic_t count;
243 struct rcu_head head;
244 } u;
245 unsigned char name[];
248 static inline struct external_name *external_name(struct dentry *dentry)
250 return container_of(dentry->d_name.name, struct external_name, name[0]);
253 static void __d_free(struct rcu_head *head)
255 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
257 kmem_cache_free(dentry_cache, dentry);
260 static void __d_free_external(struct rcu_head *head)
262 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
263 kfree(external_name(dentry));
264 kmem_cache_free(dentry_cache, dentry);
267 static inline int dname_external(const struct dentry *dentry)
269 return dentry->d_name.name != dentry->d_iname;
273 * Make sure other CPUs see the inode attached before the type is set.
275 static inline void __d_set_inode_and_type(struct dentry *dentry,
276 struct inode *inode,
277 unsigned type_flags)
279 unsigned flags;
281 dentry->d_inode = inode;
282 smp_wmb();
283 flags = READ_ONCE(dentry->d_flags);
284 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
285 flags |= type_flags;
286 WRITE_ONCE(dentry->d_flags, flags);
290 * Ideally, we want to make sure that other CPUs see the flags cleared before
291 * the inode is detached, but this is really a violation of RCU principles
292 * since the ordering suggests we should always set inode before flags.
294 * We should instead replace or discard the entire dentry - but that sucks
295 * performancewise on mass deletion/rename.
297 static inline void __d_clear_type_and_inode(struct dentry *dentry)
299 unsigned flags = READ_ONCE(dentry->d_flags);
301 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
302 WRITE_ONCE(dentry->d_flags, flags);
303 smp_wmb();
304 dentry->d_inode = NULL;
307 static void dentry_free(struct dentry *dentry)
309 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
310 if (unlikely(dname_external(dentry))) {
311 struct external_name *p = external_name(dentry);
312 if (likely(atomic_dec_and_test(&p->u.count))) {
313 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
314 return;
317 /* if dentry was never visible to RCU, immediate free is OK */
318 if (!(dentry->d_flags & DCACHE_RCUACCESS))
319 __d_free(&dentry->d_u.d_rcu);
320 else
321 call_rcu(&dentry->d_u.d_rcu, __d_free);
325 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
326 * @dentry: the target dentry
327 * After this call, in-progress rcu-walk path lookup will fail. This
328 * should be called after unhashing, and after changing d_inode (if
329 * the dentry has not already been unhashed).
331 static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
333 lockdep_assert_held(&dentry->d_lock);
334 /* Go through am invalidation barrier */
335 write_seqcount_invalidate(&dentry->d_seq);
339 * Release the dentry's inode, using the filesystem
340 * d_iput() operation if defined. Dentry has no refcount
341 * and is unhashed.
343 static void dentry_iput(struct dentry * dentry)
344 __releases(dentry->d_lock)
345 __releases(dentry->d_inode->i_lock)
347 struct inode *inode = dentry->d_inode;
348 if (inode) {
349 __d_clear_type_and_inode(dentry);
350 hlist_del_init(&dentry->d_u.d_alias);
351 spin_unlock(&dentry->d_lock);
352 spin_unlock(&inode->i_lock);
353 if (!inode->i_nlink)
354 fsnotify_inoderemove(inode);
355 if (dentry->d_op && dentry->d_op->d_iput)
356 dentry->d_op->d_iput(dentry, inode);
357 else
358 iput(inode);
359 } else {
360 spin_unlock(&dentry->d_lock);
365 * Release the dentry's inode, using the filesystem
366 * d_iput() operation if defined. dentry remains in-use.
368 static void dentry_unlink_inode(struct dentry * dentry)
369 __releases(dentry->d_lock)
370 __releases(dentry->d_inode->i_lock)
372 struct inode *inode = dentry->d_inode;
373 __d_clear_type_and_inode(dentry);
374 hlist_del_init(&dentry->d_u.d_alias);
375 dentry_rcuwalk_invalidate(dentry);
376 spin_unlock(&dentry->d_lock);
377 spin_unlock(&inode->i_lock);
378 if (!inode->i_nlink)
379 fsnotify_inoderemove(inode);
380 if (dentry->d_op && dentry->d_op->d_iput)
381 dentry->d_op->d_iput(dentry, inode);
382 else
383 iput(inode);
387 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
388 * is in use - which includes both the "real" per-superblock
389 * LRU list _and_ the DCACHE_SHRINK_LIST use.
391 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
392 * on the shrink list (ie not on the superblock LRU list).
394 * The per-cpu "nr_dentry_unused" counters are updated with
395 * the DCACHE_LRU_LIST bit.
397 * These helper functions make sure we always follow the
398 * rules. d_lock must be held by the caller.
400 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
401 static void d_lru_add(struct dentry *dentry)
403 D_FLAG_VERIFY(dentry, 0);
404 dentry->d_flags |= DCACHE_LRU_LIST;
405 this_cpu_inc(nr_dentry_unused);
406 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
409 static void d_lru_del(struct dentry *dentry)
411 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
412 dentry->d_flags &= ~DCACHE_LRU_LIST;
413 this_cpu_dec(nr_dentry_unused);
414 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
417 static void d_shrink_del(struct dentry *dentry)
419 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
420 list_del_init(&dentry->d_lru);
421 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
422 this_cpu_dec(nr_dentry_unused);
425 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
427 D_FLAG_VERIFY(dentry, 0);
428 list_add(&dentry->d_lru, list);
429 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
430 this_cpu_inc(nr_dentry_unused);
434 * These can only be called under the global LRU lock, ie during the
435 * callback for freeing the LRU list. "isolate" removes it from the
436 * LRU lists entirely, while shrink_move moves it to the indicated
437 * private list.
439 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
441 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
442 dentry->d_flags &= ~DCACHE_LRU_LIST;
443 this_cpu_dec(nr_dentry_unused);
444 list_lru_isolate(lru, &dentry->d_lru);
447 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
448 struct list_head *list)
450 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
451 dentry->d_flags |= DCACHE_SHRINK_LIST;
452 list_lru_isolate_move(lru, &dentry->d_lru, list);
456 * dentry_lru_(add|del)_list) must be called with d_lock held.
458 static void dentry_lru_add(struct dentry *dentry)
460 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
461 d_lru_add(dentry);
465 * d_drop - drop a dentry
466 * @dentry: dentry to drop
468 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
469 * be found through a VFS lookup any more. Note that this is different from
470 * deleting the dentry - d_delete will try to mark the dentry negative if
471 * possible, giving a successful _negative_ lookup, while d_drop will
472 * just make the cache lookup fail.
474 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
475 * reason (NFS timeouts or autofs deletes).
477 * __d_drop requires dentry->d_lock.
479 void __d_drop(struct dentry *dentry)
481 if (!d_unhashed(dentry)) {
482 struct hlist_bl_head *b;
484 * Hashed dentries are normally on the dentry hashtable,
485 * with the exception of those newly allocated by
486 * d_obtain_alias, which are always IS_ROOT:
488 if (unlikely(IS_ROOT(dentry)))
489 b = &dentry->d_sb->s_anon;
490 else
491 b = d_hash(dentry->d_parent, dentry->d_name.hash);
493 hlist_bl_lock(b);
494 __hlist_bl_del(&dentry->d_hash);
495 dentry->d_hash.pprev = NULL;
496 hlist_bl_unlock(b);
497 dentry_rcuwalk_invalidate(dentry);
500 EXPORT_SYMBOL(__d_drop);
502 void d_drop(struct dentry *dentry)
504 spin_lock(&dentry->d_lock);
505 __d_drop(dentry);
506 spin_unlock(&dentry->d_lock);
508 EXPORT_SYMBOL(d_drop);
510 static void __dentry_kill(struct dentry *dentry)
512 struct dentry *parent = NULL;
513 bool can_free = true;
514 if (!IS_ROOT(dentry))
515 parent = dentry->d_parent;
518 * The dentry is now unrecoverably dead to the world.
520 lockref_mark_dead(&dentry->d_lockref);
523 * inform the fs via d_prune that this dentry is about to be
524 * unhashed and destroyed.
526 if (dentry->d_flags & DCACHE_OP_PRUNE)
527 dentry->d_op->d_prune(dentry);
529 if (dentry->d_flags & DCACHE_LRU_LIST) {
530 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
531 d_lru_del(dentry);
533 /* if it was on the hash then remove it */
534 __d_drop(dentry);
535 __list_del_entry(&dentry->d_child);
537 * Inform d_walk() that we are no longer attached to the
538 * dentry tree
540 dentry->d_flags |= DCACHE_DENTRY_KILLED;
541 if (parent)
542 spin_unlock(&parent->d_lock);
543 dentry_iput(dentry);
545 * dentry_iput drops the locks, at which point nobody (except
546 * transient RCU lookups) can reach this dentry.
548 BUG_ON(dentry->d_lockref.count > 0);
549 this_cpu_dec(nr_dentry);
550 if (dentry->d_op && dentry->d_op->d_release)
551 dentry->d_op->d_release(dentry);
553 spin_lock(&dentry->d_lock);
554 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
555 dentry->d_flags |= DCACHE_MAY_FREE;
556 can_free = false;
558 spin_unlock(&dentry->d_lock);
559 if (likely(can_free))
560 dentry_free(dentry);
564 * Finish off a dentry we've decided to kill.
565 * dentry->d_lock must be held, returns with it unlocked.
566 * If ref is non-zero, then decrement the refcount too.
567 * Returns dentry requiring refcount drop, or NULL if we're done.
569 static struct dentry *dentry_kill(struct dentry *dentry)
570 __releases(dentry->d_lock)
572 struct inode *inode = dentry->d_inode;
573 struct dentry *parent = NULL;
575 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
576 goto failed;
578 if (!IS_ROOT(dentry)) {
579 parent = dentry->d_parent;
580 if (unlikely(!spin_trylock(&parent->d_lock))) {
581 if (inode)
582 spin_unlock(&inode->i_lock);
583 goto failed;
587 __dentry_kill(dentry);
588 return parent;
590 failed:
591 spin_unlock(&dentry->d_lock);
592 cpu_relax();
593 return dentry; /* try again with same dentry */
596 static inline struct dentry *lock_parent(struct dentry *dentry)
598 struct dentry *parent = dentry->d_parent;
599 if (IS_ROOT(dentry))
600 return NULL;
601 if (unlikely(dentry->d_lockref.count < 0))
602 return NULL;
603 if (likely(spin_trylock(&parent->d_lock)))
604 return parent;
605 rcu_read_lock();
606 spin_unlock(&dentry->d_lock);
607 again:
608 parent = ACCESS_ONCE(dentry->d_parent);
609 spin_lock(&parent->d_lock);
611 * We can't blindly lock dentry until we are sure
612 * that we won't violate the locking order.
613 * Any changes of dentry->d_parent must have
614 * been done with parent->d_lock held, so
615 * spin_lock() above is enough of a barrier
616 * for checking if it's still our child.
618 if (unlikely(parent != dentry->d_parent)) {
619 spin_unlock(&parent->d_lock);
620 goto again;
622 rcu_read_unlock();
623 if (parent != dentry)
624 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
625 else
626 parent = NULL;
627 return parent;
631 * Try to do a lockless dput(), and return whether that was successful.
633 * If unsuccessful, we return false, having already taken the dentry lock.
635 * The caller needs to hold the RCU read lock, so that the dentry is
636 * guaranteed to stay around even if the refcount goes down to zero!
638 static inline bool fast_dput(struct dentry *dentry)
640 int ret;
641 unsigned int d_flags;
644 * If we have a d_op->d_delete() operation, we sould not
645 * let the dentry count go to zero, so use "put_or_lock".
647 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
648 return lockref_put_or_lock(&dentry->d_lockref);
651 * .. otherwise, we can try to just decrement the
652 * lockref optimistically.
654 ret = lockref_put_return(&dentry->d_lockref);
657 * If the lockref_put_return() failed due to the lock being held
658 * by somebody else, the fast path has failed. We will need to
659 * get the lock, and then check the count again.
661 if (unlikely(ret < 0)) {
662 spin_lock(&dentry->d_lock);
663 if (dentry->d_lockref.count > 1) {
664 dentry->d_lockref.count--;
665 spin_unlock(&dentry->d_lock);
666 return 1;
668 return 0;
672 * If we weren't the last ref, we're done.
674 if (ret)
675 return 1;
678 * Careful, careful. The reference count went down
679 * to zero, but we don't hold the dentry lock, so
680 * somebody else could get it again, and do another
681 * dput(), and we need to not race with that.
683 * However, there is a very special and common case
684 * where we don't care, because there is nothing to
685 * do: the dentry is still hashed, it does not have
686 * a 'delete' op, and it's referenced and already on
687 * the LRU list.
689 * NOTE! Since we aren't locked, these values are
690 * not "stable". However, it is sufficient that at
691 * some point after we dropped the reference the
692 * dentry was hashed and the flags had the proper
693 * value. Other dentry users may have re-gotten
694 * a reference to the dentry and change that, but
695 * our work is done - we can leave the dentry
696 * around with a zero refcount.
698 smp_rmb();
699 d_flags = ACCESS_ONCE(dentry->d_flags);
700 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
702 /* Nothing to do? Dropping the reference was all we needed? */
703 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
704 return 1;
707 * Not the fast normal case? Get the lock. We've already decremented
708 * the refcount, but we'll need to re-check the situation after
709 * getting the lock.
711 spin_lock(&dentry->d_lock);
714 * Did somebody else grab a reference to it in the meantime, and
715 * we're no longer the last user after all? Alternatively, somebody
716 * else could have killed it and marked it dead. Either way, we
717 * don't need to do anything else.
719 if (dentry->d_lockref.count) {
720 spin_unlock(&dentry->d_lock);
721 return 1;
725 * Re-get the reference we optimistically dropped. We hold the
726 * lock, and we just tested that it was zero, so we can just
727 * set it to 1.
729 dentry->d_lockref.count = 1;
730 return 0;
735 * This is dput
737 * This is complicated by the fact that we do not want to put
738 * dentries that are no longer on any hash chain on the unused
739 * list: we'd much rather just get rid of them immediately.
741 * However, that implies that we have to traverse the dentry
742 * tree upwards to the parents which might _also_ now be
743 * scheduled for deletion (it may have been only waiting for
744 * its last child to go away).
746 * This tail recursion is done by hand as we don't want to depend
747 * on the compiler to always get this right (gcc generally doesn't).
748 * Real recursion would eat up our stack space.
752 * dput - release a dentry
753 * @dentry: dentry to release
755 * Release a dentry. This will drop the usage count and if appropriate
756 * call the dentry unlink method as well as removing it from the queues and
757 * releasing its resources. If the parent dentries were scheduled for release
758 * they too may now get deleted.
760 void dput(struct dentry *dentry)
762 if (unlikely(!dentry))
763 return;
765 repeat:
766 rcu_read_lock();
767 if (likely(fast_dput(dentry))) {
768 rcu_read_unlock();
769 return;
772 /* Slow case: now with the dentry lock held */
773 rcu_read_unlock();
775 /* Unreachable? Get rid of it */
776 if (unlikely(d_unhashed(dentry)))
777 goto kill_it;
779 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
780 goto kill_it;
782 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
783 if (dentry->d_op->d_delete(dentry))
784 goto kill_it;
787 if (!(dentry->d_flags & DCACHE_REFERENCED))
788 dentry->d_flags |= DCACHE_REFERENCED;
789 dentry_lru_add(dentry);
791 dentry->d_lockref.count--;
792 spin_unlock(&dentry->d_lock);
793 return;
795 kill_it:
796 dentry = dentry_kill(dentry);
797 if (dentry)
798 goto repeat;
800 EXPORT_SYMBOL(dput);
803 /* This must be called with d_lock held */
804 static inline void __dget_dlock(struct dentry *dentry)
806 dentry->d_lockref.count++;
809 static inline void __dget(struct dentry *dentry)
811 lockref_get(&dentry->d_lockref);
814 struct dentry *dget_parent(struct dentry *dentry)
816 int gotref;
817 struct dentry *ret;
820 * Do optimistic parent lookup without any
821 * locking.
823 rcu_read_lock();
824 ret = ACCESS_ONCE(dentry->d_parent);
825 gotref = lockref_get_not_zero(&ret->d_lockref);
826 rcu_read_unlock();
827 if (likely(gotref)) {
828 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
829 return ret;
830 dput(ret);
833 repeat:
835 * Don't need rcu_dereference because we re-check it was correct under
836 * the lock.
838 rcu_read_lock();
839 ret = dentry->d_parent;
840 spin_lock(&ret->d_lock);
841 if (unlikely(ret != dentry->d_parent)) {
842 spin_unlock(&ret->d_lock);
843 rcu_read_unlock();
844 goto repeat;
846 rcu_read_unlock();
847 BUG_ON(!ret->d_lockref.count);
848 ret->d_lockref.count++;
849 spin_unlock(&ret->d_lock);
850 return ret;
852 EXPORT_SYMBOL(dget_parent);
855 * d_find_alias - grab a hashed alias of inode
856 * @inode: inode in question
858 * If inode has a hashed alias, or is a directory and has any alias,
859 * acquire the reference to alias and return it. Otherwise return NULL.
860 * Notice that if inode is a directory there can be only one alias and
861 * it can be unhashed only if it has no children, or if it is the root
862 * of a filesystem, or if the directory was renamed and d_revalidate
863 * was the first vfs operation to notice.
865 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
866 * any other hashed alias over that one.
868 static struct dentry *__d_find_alias(struct inode *inode)
870 struct dentry *alias, *discon_alias;
872 again:
873 discon_alias = NULL;
874 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
875 spin_lock(&alias->d_lock);
876 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
877 if (IS_ROOT(alias) &&
878 (alias->d_flags & DCACHE_DISCONNECTED)) {
879 discon_alias = alias;
880 } else {
881 __dget_dlock(alias);
882 spin_unlock(&alias->d_lock);
883 return alias;
886 spin_unlock(&alias->d_lock);
888 if (discon_alias) {
889 alias = discon_alias;
890 spin_lock(&alias->d_lock);
891 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
892 __dget_dlock(alias);
893 spin_unlock(&alias->d_lock);
894 return alias;
896 spin_unlock(&alias->d_lock);
897 goto again;
899 return NULL;
902 struct dentry *d_find_alias(struct inode *inode)
904 struct dentry *de = NULL;
906 if (!hlist_empty(&inode->i_dentry)) {
907 spin_lock(&inode->i_lock);
908 de = __d_find_alias(inode);
909 spin_unlock(&inode->i_lock);
911 return de;
913 EXPORT_SYMBOL(d_find_alias);
916 * Try to kill dentries associated with this inode.
917 * WARNING: you must own a reference to inode.
919 void d_prune_aliases(struct inode *inode)
921 struct dentry *dentry;
922 restart:
923 spin_lock(&inode->i_lock);
924 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
925 spin_lock(&dentry->d_lock);
926 if (!dentry->d_lockref.count) {
927 struct dentry *parent = lock_parent(dentry);
928 if (likely(!dentry->d_lockref.count)) {
929 __dentry_kill(dentry);
930 dput(parent);
931 goto restart;
933 if (parent)
934 spin_unlock(&parent->d_lock);
936 spin_unlock(&dentry->d_lock);
938 spin_unlock(&inode->i_lock);
940 EXPORT_SYMBOL(d_prune_aliases);
942 static void shrink_dentry_list(struct list_head *list)
944 struct dentry *dentry, *parent;
946 while (!list_empty(list)) {
947 struct inode *inode;
948 dentry = list_entry(list->prev, struct dentry, d_lru);
949 spin_lock(&dentry->d_lock);
950 parent = lock_parent(dentry);
953 * The dispose list is isolated and dentries are not accounted
954 * to the LRU here, so we can simply remove it from the list
955 * here regardless of whether it is referenced or not.
957 d_shrink_del(dentry);
960 * We found an inuse dentry which was not removed from
961 * the LRU because of laziness during lookup. Do not free it.
963 if (dentry->d_lockref.count > 0) {
964 spin_unlock(&dentry->d_lock);
965 if (parent)
966 spin_unlock(&parent->d_lock);
967 continue;
971 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
972 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
973 spin_unlock(&dentry->d_lock);
974 if (parent)
975 spin_unlock(&parent->d_lock);
976 if (can_free)
977 dentry_free(dentry);
978 continue;
981 inode = dentry->d_inode;
982 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
983 d_shrink_add(dentry, list);
984 spin_unlock(&dentry->d_lock);
985 if (parent)
986 spin_unlock(&parent->d_lock);
987 continue;
990 __dentry_kill(dentry);
993 * We need to prune ancestors too. This is necessary to prevent
994 * quadratic behavior of shrink_dcache_parent(), but is also
995 * expected to be beneficial in reducing dentry cache
996 * fragmentation.
998 dentry = parent;
999 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1000 parent = lock_parent(dentry);
1001 if (dentry->d_lockref.count != 1) {
1002 dentry->d_lockref.count--;
1003 spin_unlock(&dentry->d_lock);
1004 if (parent)
1005 spin_unlock(&parent->d_lock);
1006 break;
1008 inode = dentry->d_inode; /* can't be NULL */
1009 if (unlikely(!spin_trylock(&inode->i_lock))) {
1010 spin_unlock(&dentry->d_lock);
1011 if (parent)
1012 spin_unlock(&parent->d_lock);
1013 cpu_relax();
1014 continue;
1016 __dentry_kill(dentry);
1017 dentry = parent;
1022 static enum lru_status dentry_lru_isolate(struct list_head *item,
1023 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1025 struct list_head *freeable = arg;
1026 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1030 * we are inverting the lru lock/dentry->d_lock here,
1031 * so use a trylock. If we fail to get the lock, just skip
1032 * it
1034 if (!spin_trylock(&dentry->d_lock))
1035 return LRU_SKIP;
1038 * Referenced dentries are still in use. If they have active
1039 * counts, just remove them from the LRU. Otherwise give them
1040 * another pass through the LRU.
1042 if (dentry->d_lockref.count) {
1043 d_lru_isolate(lru, dentry);
1044 spin_unlock(&dentry->d_lock);
1045 return LRU_REMOVED;
1048 if (dentry->d_flags & DCACHE_REFERENCED) {
1049 dentry->d_flags &= ~DCACHE_REFERENCED;
1050 spin_unlock(&dentry->d_lock);
1053 * The list move itself will be made by the common LRU code. At
1054 * this point, we've dropped the dentry->d_lock but keep the
1055 * lru lock. This is safe to do, since every list movement is
1056 * protected by the lru lock even if both locks are held.
1058 * This is guaranteed by the fact that all LRU management
1059 * functions are intermediated by the LRU API calls like
1060 * list_lru_add and list_lru_del. List movement in this file
1061 * only ever occur through this functions or through callbacks
1062 * like this one, that are called from the LRU API.
1064 * The only exceptions to this are functions like
1065 * shrink_dentry_list, and code that first checks for the
1066 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1067 * operating only with stack provided lists after they are
1068 * properly isolated from the main list. It is thus, always a
1069 * local access.
1071 return LRU_ROTATE;
1074 d_lru_shrink_move(lru, dentry, freeable);
1075 spin_unlock(&dentry->d_lock);
1077 return LRU_REMOVED;
1081 * prune_dcache_sb - shrink the dcache
1082 * @sb: superblock
1083 * @sc: shrink control, passed to list_lru_shrink_walk()
1085 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1086 * is done when we need more memory and called from the superblock shrinker
1087 * function.
1089 * This function may fail to free any resources if all the dentries are in
1090 * use.
1092 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1094 LIST_HEAD(dispose);
1095 long freed;
1097 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1098 dentry_lru_isolate, &dispose);
1099 shrink_dentry_list(&dispose);
1100 return freed;
1103 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1104 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1106 struct list_head *freeable = arg;
1107 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1110 * we are inverting the lru lock/dentry->d_lock here,
1111 * so use a trylock. If we fail to get the lock, just skip
1112 * it
1114 if (!spin_trylock(&dentry->d_lock))
1115 return LRU_SKIP;
1117 d_lru_shrink_move(lru, dentry, freeable);
1118 spin_unlock(&dentry->d_lock);
1120 return LRU_REMOVED;
1125 * shrink_dcache_sb - shrink dcache for a superblock
1126 * @sb: superblock
1128 * Shrink the dcache for the specified super block. This is used to free
1129 * the dcache before unmounting a file system.
1131 void shrink_dcache_sb(struct super_block *sb)
1133 long freed;
1135 do {
1136 LIST_HEAD(dispose);
1138 freed = list_lru_walk(&sb->s_dentry_lru,
1139 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1141 this_cpu_sub(nr_dentry_unused, freed);
1142 shrink_dentry_list(&dispose);
1143 } while (freed > 0);
1145 EXPORT_SYMBOL(shrink_dcache_sb);
1148 * enum d_walk_ret - action to talke during tree walk
1149 * @D_WALK_CONTINUE: contrinue walk
1150 * @D_WALK_QUIT: quit walk
1151 * @D_WALK_NORETRY: quit when retry is needed
1152 * @D_WALK_SKIP: skip this dentry and its children
1154 enum d_walk_ret {
1155 D_WALK_CONTINUE,
1156 D_WALK_QUIT,
1157 D_WALK_NORETRY,
1158 D_WALK_SKIP,
1162 * d_walk - walk the dentry tree
1163 * @parent: start of walk
1164 * @data: data passed to @enter() and @finish()
1165 * @enter: callback when first entering the dentry
1166 * @finish: callback when successfully finished the walk
1168 * The @enter() and @finish() callbacks are called with d_lock held.
1170 static void d_walk(struct dentry *parent, void *data,
1171 enum d_walk_ret (*enter)(void *, struct dentry *),
1172 void (*finish)(void *))
1174 struct dentry *this_parent;
1175 struct list_head *next;
1176 unsigned seq = 0;
1177 enum d_walk_ret ret;
1178 bool retry = true;
1180 again:
1181 read_seqbegin_or_lock(&rename_lock, &seq);
1182 this_parent = parent;
1183 spin_lock(&this_parent->d_lock);
1185 ret = enter(data, this_parent);
1186 switch (ret) {
1187 case D_WALK_CONTINUE:
1188 break;
1189 case D_WALK_QUIT:
1190 case D_WALK_SKIP:
1191 goto out_unlock;
1192 case D_WALK_NORETRY:
1193 retry = false;
1194 break;
1196 repeat:
1197 next = this_parent->d_subdirs.next;
1198 resume:
1199 while (next != &this_parent->d_subdirs) {
1200 struct list_head *tmp = next;
1201 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1202 next = tmp->next;
1204 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1206 ret = enter(data, dentry);
1207 switch (ret) {
1208 case D_WALK_CONTINUE:
1209 break;
1210 case D_WALK_QUIT:
1211 spin_unlock(&dentry->d_lock);
1212 goto out_unlock;
1213 case D_WALK_NORETRY:
1214 retry = false;
1215 break;
1216 case D_WALK_SKIP:
1217 spin_unlock(&dentry->d_lock);
1218 continue;
1221 if (!list_empty(&dentry->d_subdirs)) {
1222 spin_unlock(&this_parent->d_lock);
1223 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1224 this_parent = dentry;
1225 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1226 goto repeat;
1228 spin_unlock(&dentry->d_lock);
1231 * All done at this level ... ascend and resume the search.
1233 rcu_read_lock();
1234 ascend:
1235 if (this_parent != parent) {
1236 struct dentry *child = this_parent;
1237 this_parent = child->d_parent;
1239 spin_unlock(&child->d_lock);
1240 spin_lock(&this_parent->d_lock);
1242 /* might go back up the wrong parent if we have had a rename. */
1243 if (need_seqretry(&rename_lock, seq))
1244 goto rename_retry;
1245 /* go into the first sibling still alive */
1246 do {
1247 next = child->d_child.next;
1248 if (next == &this_parent->d_subdirs)
1249 goto ascend;
1250 child = list_entry(next, struct dentry, d_child);
1251 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1252 rcu_read_unlock();
1253 goto resume;
1255 if (need_seqretry(&rename_lock, seq))
1256 goto rename_retry;
1257 rcu_read_unlock();
1258 if (finish)
1259 finish(data);
1261 out_unlock:
1262 spin_unlock(&this_parent->d_lock);
1263 done_seqretry(&rename_lock, seq);
1264 return;
1266 rename_retry:
1267 spin_unlock(&this_parent->d_lock);
1268 rcu_read_unlock();
1269 BUG_ON(seq & 1);
1270 if (!retry)
1271 return;
1272 seq = 1;
1273 goto again;
1277 * Search for at least 1 mount point in the dentry's subdirs.
1278 * We descend to the next level whenever the d_subdirs
1279 * list is non-empty and continue searching.
1282 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1284 int *ret = data;
1285 if (d_mountpoint(dentry)) {
1286 *ret = 1;
1287 return D_WALK_QUIT;
1289 return D_WALK_CONTINUE;
1293 * have_submounts - check for mounts over a dentry
1294 * @parent: dentry to check.
1296 * Return true if the parent or its subdirectories contain
1297 * a mount point
1299 int have_submounts(struct dentry *parent)
1301 int ret = 0;
1303 d_walk(parent, &ret, check_mount, NULL);
1305 return ret;
1307 EXPORT_SYMBOL(have_submounts);
1310 * Called by mount code to set a mountpoint and check if the mountpoint is
1311 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1312 * subtree can become unreachable).
1314 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1315 * this reason take rename_lock and d_lock on dentry and ancestors.
1317 int d_set_mounted(struct dentry *dentry)
1319 struct dentry *p;
1320 int ret = -ENOENT;
1321 write_seqlock(&rename_lock);
1322 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1323 /* Need exclusion wrt. d_invalidate() */
1324 spin_lock(&p->d_lock);
1325 if (unlikely(d_unhashed(p))) {
1326 spin_unlock(&p->d_lock);
1327 goto out;
1329 spin_unlock(&p->d_lock);
1331 spin_lock(&dentry->d_lock);
1332 if (!d_unlinked(dentry)) {
1333 dentry->d_flags |= DCACHE_MOUNTED;
1334 ret = 0;
1336 spin_unlock(&dentry->d_lock);
1337 out:
1338 write_sequnlock(&rename_lock);
1339 return ret;
1343 * Search the dentry child list of the specified parent,
1344 * and move any unused dentries to the end of the unused
1345 * list for prune_dcache(). We descend to the next level
1346 * whenever the d_subdirs list is non-empty and continue
1347 * searching.
1349 * It returns zero iff there are no unused children,
1350 * otherwise it returns the number of children moved to
1351 * the end of the unused list. This may not be the total
1352 * number of unused children, because select_parent can
1353 * drop the lock and return early due to latency
1354 * constraints.
1357 struct select_data {
1358 struct dentry *start;
1359 struct list_head dispose;
1360 int found;
1363 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1365 struct select_data *data = _data;
1366 enum d_walk_ret ret = D_WALK_CONTINUE;
1368 if (data->start == dentry)
1369 goto out;
1371 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1372 data->found++;
1373 } else {
1374 if (dentry->d_flags & DCACHE_LRU_LIST)
1375 d_lru_del(dentry);
1376 if (!dentry->d_lockref.count) {
1377 d_shrink_add(dentry, &data->dispose);
1378 data->found++;
1382 * We can return to the caller if we have found some (this
1383 * ensures forward progress). We'll be coming back to find
1384 * the rest.
1386 if (!list_empty(&data->dispose))
1387 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1388 out:
1389 return ret;
1393 * shrink_dcache_parent - prune dcache
1394 * @parent: parent of entries to prune
1396 * Prune the dcache to remove unused children of the parent dentry.
1398 void shrink_dcache_parent(struct dentry *parent)
1400 for (;;) {
1401 struct select_data data;
1403 INIT_LIST_HEAD(&data.dispose);
1404 data.start = parent;
1405 data.found = 0;
1407 d_walk(parent, &data, select_collect, NULL);
1408 if (!data.found)
1409 break;
1411 shrink_dentry_list(&data.dispose);
1412 cond_resched();
1415 EXPORT_SYMBOL(shrink_dcache_parent);
1417 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1419 /* it has busy descendents; complain about those instead */
1420 if (!list_empty(&dentry->d_subdirs))
1421 return D_WALK_CONTINUE;
1423 /* root with refcount 1 is fine */
1424 if (dentry == _data && dentry->d_lockref.count == 1)
1425 return D_WALK_CONTINUE;
1427 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1428 " still in use (%d) [unmount of %s %s]\n",
1429 dentry,
1430 dentry->d_inode ?
1431 dentry->d_inode->i_ino : 0UL,
1432 dentry,
1433 dentry->d_lockref.count,
1434 dentry->d_sb->s_type->name,
1435 dentry->d_sb->s_id);
1436 WARN_ON(1);
1437 return D_WALK_CONTINUE;
1440 static void do_one_tree(struct dentry *dentry)
1442 shrink_dcache_parent(dentry);
1443 d_walk(dentry, dentry, umount_check, NULL);
1444 d_drop(dentry);
1445 dput(dentry);
1449 * destroy the dentries attached to a superblock on unmounting
1451 void shrink_dcache_for_umount(struct super_block *sb)
1453 struct dentry *dentry;
1455 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1457 dentry = sb->s_root;
1458 sb->s_root = NULL;
1459 do_one_tree(dentry);
1461 while (!hlist_bl_empty(&sb->s_anon)) {
1462 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1463 do_one_tree(dentry);
1467 struct detach_data {
1468 struct select_data select;
1469 struct dentry *mountpoint;
1471 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1473 struct detach_data *data = _data;
1475 if (d_mountpoint(dentry)) {
1476 __dget_dlock(dentry);
1477 data->mountpoint = dentry;
1478 return D_WALK_QUIT;
1481 return select_collect(&data->select, dentry);
1484 static void check_and_drop(void *_data)
1486 struct detach_data *data = _data;
1488 if (!data->mountpoint && !data->select.found)
1489 __d_drop(data->select.start);
1493 * d_invalidate - detach submounts, prune dcache, and drop
1494 * @dentry: dentry to invalidate (aka detach, prune and drop)
1496 * no dcache lock.
1498 * The final d_drop is done as an atomic operation relative to
1499 * rename_lock ensuring there are no races with d_set_mounted. This
1500 * ensures there are no unhashed dentries on the path to a mountpoint.
1502 void d_invalidate(struct dentry *dentry)
1505 * If it's already been dropped, return OK.
1507 spin_lock(&dentry->d_lock);
1508 if (d_unhashed(dentry)) {
1509 spin_unlock(&dentry->d_lock);
1510 return;
1512 spin_unlock(&dentry->d_lock);
1514 /* Negative dentries can be dropped without further checks */
1515 if (!dentry->d_inode) {
1516 d_drop(dentry);
1517 return;
1520 for (;;) {
1521 struct detach_data data;
1523 data.mountpoint = NULL;
1524 INIT_LIST_HEAD(&data.select.dispose);
1525 data.select.start = dentry;
1526 data.select.found = 0;
1528 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1530 if (data.select.found)
1531 shrink_dentry_list(&data.select.dispose);
1533 if (data.mountpoint) {
1534 detach_mounts(data.mountpoint);
1535 dput(data.mountpoint);
1538 if (!data.mountpoint && !data.select.found)
1539 break;
1541 cond_resched();
1544 EXPORT_SYMBOL(d_invalidate);
1547 * __d_alloc - allocate a dcache entry
1548 * @sb: filesystem it will belong to
1549 * @name: qstr of the name
1551 * Allocates a dentry. It returns %NULL if there is insufficient memory
1552 * available. On a success the dentry is returned. The name passed in is
1553 * copied and the copy passed in may be reused after this call.
1556 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1558 struct dentry *dentry;
1559 char *dname;
1561 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1562 if (!dentry)
1563 return NULL;
1566 * We guarantee that the inline name is always NUL-terminated.
1567 * This way the memcpy() done by the name switching in rename
1568 * will still always have a NUL at the end, even if we might
1569 * be overwriting an internal NUL character
1571 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1572 if (name->len > DNAME_INLINE_LEN-1) {
1573 size_t size = offsetof(struct external_name, name[1]);
1574 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1575 if (!p) {
1576 kmem_cache_free(dentry_cache, dentry);
1577 return NULL;
1579 atomic_set(&p->u.count, 1);
1580 dname = p->name;
1581 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1582 kasan_unpoison_shadow(dname,
1583 round_up(name->len + 1, sizeof(unsigned long)));
1584 } else {
1585 dname = dentry->d_iname;
1588 dentry->d_name.len = name->len;
1589 dentry->d_name.hash = name->hash;
1590 memcpy(dname, name->name, name->len);
1591 dname[name->len] = 0;
1593 /* Make sure we always see the terminating NUL character */
1594 smp_wmb();
1595 dentry->d_name.name = dname;
1597 dentry->d_lockref.count = 1;
1598 dentry->d_flags = 0;
1599 spin_lock_init(&dentry->d_lock);
1600 seqcount_init(&dentry->d_seq);
1601 dentry->d_inode = NULL;
1602 dentry->d_parent = dentry;
1603 dentry->d_sb = sb;
1604 dentry->d_op = NULL;
1605 dentry->d_fsdata = NULL;
1606 INIT_HLIST_BL_NODE(&dentry->d_hash);
1607 INIT_LIST_HEAD(&dentry->d_lru);
1608 INIT_LIST_HEAD(&dentry->d_subdirs);
1609 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1610 INIT_LIST_HEAD(&dentry->d_child);
1611 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1613 this_cpu_inc(nr_dentry);
1615 return dentry;
1619 * d_alloc - allocate a dcache entry
1620 * @parent: parent of entry to allocate
1621 * @name: qstr of the name
1623 * Allocates a dentry. It returns %NULL if there is insufficient memory
1624 * available. On a success the dentry is returned. The name passed in is
1625 * copied and the copy passed in may be reused after this call.
1627 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1629 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1630 if (!dentry)
1631 return NULL;
1633 spin_lock(&parent->d_lock);
1635 * don't need child lock because it is not subject
1636 * to concurrency here
1638 __dget_dlock(parent);
1639 dentry->d_parent = parent;
1640 list_add(&dentry->d_child, &parent->d_subdirs);
1641 spin_unlock(&parent->d_lock);
1643 return dentry;
1645 EXPORT_SYMBOL(d_alloc);
1648 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1649 * @sb: the superblock
1650 * @name: qstr of the name
1652 * For a filesystem that just pins its dentries in memory and never
1653 * performs lookups at all, return an unhashed IS_ROOT dentry.
1655 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1657 return __d_alloc(sb, name);
1659 EXPORT_SYMBOL(d_alloc_pseudo);
1661 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1663 struct qstr q;
1665 q.name = name;
1666 q.len = strlen(name);
1667 q.hash = full_name_hash(q.name, q.len);
1668 return d_alloc(parent, &q);
1670 EXPORT_SYMBOL(d_alloc_name);
1672 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1674 WARN_ON_ONCE(dentry->d_op);
1675 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1676 DCACHE_OP_COMPARE |
1677 DCACHE_OP_REVALIDATE |
1678 DCACHE_OP_WEAK_REVALIDATE |
1679 DCACHE_OP_DELETE |
1680 DCACHE_OP_SELECT_INODE));
1681 dentry->d_op = op;
1682 if (!op)
1683 return;
1684 if (op->d_hash)
1685 dentry->d_flags |= DCACHE_OP_HASH;
1686 if (op->d_compare)
1687 dentry->d_flags |= DCACHE_OP_COMPARE;
1688 if (op->d_revalidate)
1689 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1690 if (op->d_weak_revalidate)
1691 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1692 if (op->d_delete)
1693 dentry->d_flags |= DCACHE_OP_DELETE;
1694 if (op->d_prune)
1695 dentry->d_flags |= DCACHE_OP_PRUNE;
1696 if (op->d_select_inode)
1697 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
1700 EXPORT_SYMBOL(d_set_d_op);
1704 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1705 * @dentry - The dentry to mark
1707 * Mark a dentry as falling through to the lower layer (as set with
1708 * d_pin_lower()). This flag may be recorded on the medium.
1710 void d_set_fallthru(struct dentry *dentry)
1712 spin_lock(&dentry->d_lock);
1713 dentry->d_flags |= DCACHE_FALLTHRU;
1714 spin_unlock(&dentry->d_lock);
1716 EXPORT_SYMBOL(d_set_fallthru);
1718 static unsigned d_flags_for_inode(struct inode *inode)
1720 unsigned add_flags = DCACHE_REGULAR_TYPE;
1722 if (!inode)
1723 return DCACHE_MISS_TYPE;
1725 if (S_ISDIR(inode->i_mode)) {
1726 add_flags = DCACHE_DIRECTORY_TYPE;
1727 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1728 if (unlikely(!inode->i_op->lookup))
1729 add_flags = DCACHE_AUTODIR_TYPE;
1730 else
1731 inode->i_opflags |= IOP_LOOKUP;
1733 goto type_determined;
1736 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1737 if (unlikely(inode->i_op->get_link)) {
1738 add_flags = DCACHE_SYMLINK_TYPE;
1739 goto type_determined;
1741 inode->i_opflags |= IOP_NOFOLLOW;
1744 if (unlikely(!S_ISREG(inode->i_mode)))
1745 add_flags = DCACHE_SPECIAL_TYPE;
1747 type_determined:
1748 if (unlikely(IS_AUTOMOUNT(inode)))
1749 add_flags |= DCACHE_NEED_AUTOMOUNT;
1750 return add_flags;
1753 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1755 unsigned add_flags = d_flags_for_inode(inode);
1757 spin_lock(&dentry->d_lock);
1758 if (inode)
1759 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1760 __d_set_inode_and_type(dentry, inode, add_flags);
1761 dentry_rcuwalk_invalidate(dentry);
1762 spin_unlock(&dentry->d_lock);
1763 fsnotify_d_instantiate(dentry, inode);
1767 * d_instantiate - fill in inode information for a dentry
1768 * @entry: dentry to complete
1769 * @inode: inode to attach to this dentry
1771 * Fill in inode information in the entry.
1773 * This turns negative dentries into productive full members
1774 * of society.
1776 * NOTE! This assumes that the inode count has been incremented
1777 * (or otherwise set) by the caller to indicate that it is now
1778 * in use by the dcache.
1781 void d_instantiate(struct dentry *entry, struct inode * inode)
1783 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1784 if (inode)
1785 spin_lock(&inode->i_lock);
1786 __d_instantiate(entry, inode);
1787 if (inode)
1788 spin_unlock(&inode->i_lock);
1789 security_d_instantiate(entry, inode);
1791 EXPORT_SYMBOL(d_instantiate);
1794 * d_instantiate_unique - instantiate a non-aliased dentry
1795 * @entry: dentry to instantiate
1796 * @inode: inode to attach to this dentry
1798 * Fill in inode information in the entry. On success, it returns NULL.
1799 * If an unhashed alias of "entry" already exists, then we return the
1800 * aliased dentry instead and drop one reference to inode.
1802 * Note that in order to avoid conflicts with rename() etc, the caller
1803 * had better be holding the parent directory semaphore.
1805 * This also assumes that the inode count has been incremented
1806 * (or otherwise set) by the caller to indicate that it is now
1807 * in use by the dcache.
1809 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1810 struct inode *inode)
1812 struct dentry *alias;
1813 int len = entry->d_name.len;
1814 const char *name = entry->d_name.name;
1815 unsigned int hash = entry->d_name.hash;
1817 if (!inode) {
1818 __d_instantiate(entry, NULL);
1819 return NULL;
1822 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1824 * Don't need alias->d_lock here, because aliases with
1825 * d_parent == entry->d_parent are not subject to name or
1826 * parent changes, because the parent inode i_mutex is held.
1828 if (alias->d_name.hash != hash)
1829 continue;
1830 if (alias->d_parent != entry->d_parent)
1831 continue;
1832 if (alias->d_name.len != len)
1833 continue;
1834 if (dentry_cmp(alias, name, len))
1835 continue;
1836 __dget(alias);
1837 return alias;
1840 __d_instantiate(entry, inode);
1841 return NULL;
1844 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1846 struct dentry *result;
1848 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1850 if (inode)
1851 spin_lock(&inode->i_lock);
1852 result = __d_instantiate_unique(entry, inode);
1853 if (inode)
1854 spin_unlock(&inode->i_lock);
1856 if (!result) {
1857 security_d_instantiate(entry, inode);
1858 return NULL;
1861 BUG_ON(!d_unhashed(result));
1862 iput(inode);
1863 return result;
1866 EXPORT_SYMBOL(d_instantiate_unique);
1869 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1870 * @entry: dentry to complete
1871 * @inode: inode to attach to this dentry
1873 * Fill in inode information in the entry. If a directory alias is found, then
1874 * return an error (and drop inode). Together with d_materialise_unique() this
1875 * guarantees that a directory inode may never have more than one alias.
1877 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1879 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1881 spin_lock(&inode->i_lock);
1882 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1883 spin_unlock(&inode->i_lock);
1884 iput(inode);
1885 return -EBUSY;
1887 __d_instantiate(entry, inode);
1888 spin_unlock(&inode->i_lock);
1889 security_d_instantiate(entry, inode);
1891 return 0;
1893 EXPORT_SYMBOL(d_instantiate_no_diralias);
1895 struct dentry *d_make_root(struct inode *root_inode)
1897 struct dentry *res = NULL;
1899 if (root_inode) {
1900 static const struct qstr name = QSTR_INIT("/", 1);
1902 res = __d_alloc(root_inode->i_sb, &name);
1903 if (res)
1904 d_instantiate(res, root_inode);
1905 else
1906 iput(root_inode);
1908 return res;
1910 EXPORT_SYMBOL(d_make_root);
1912 static struct dentry * __d_find_any_alias(struct inode *inode)
1914 struct dentry *alias;
1916 if (hlist_empty(&inode->i_dentry))
1917 return NULL;
1918 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1919 __dget(alias);
1920 return alias;
1924 * d_find_any_alias - find any alias for a given inode
1925 * @inode: inode to find an alias for
1927 * If any aliases exist for the given inode, take and return a
1928 * reference for one of them. If no aliases exist, return %NULL.
1930 struct dentry *d_find_any_alias(struct inode *inode)
1932 struct dentry *de;
1934 spin_lock(&inode->i_lock);
1935 de = __d_find_any_alias(inode);
1936 spin_unlock(&inode->i_lock);
1937 return de;
1939 EXPORT_SYMBOL(d_find_any_alias);
1941 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1943 static const struct qstr anonstring = QSTR_INIT("/", 1);
1944 struct dentry *tmp;
1945 struct dentry *res;
1946 unsigned add_flags;
1948 if (!inode)
1949 return ERR_PTR(-ESTALE);
1950 if (IS_ERR(inode))
1951 return ERR_CAST(inode);
1953 res = d_find_any_alias(inode);
1954 if (res)
1955 goto out_iput;
1957 tmp = __d_alloc(inode->i_sb, &anonstring);
1958 if (!tmp) {
1959 res = ERR_PTR(-ENOMEM);
1960 goto out_iput;
1963 spin_lock(&inode->i_lock);
1964 res = __d_find_any_alias(inode);
1965 if (res) {
1966 spin_unlock(&inode->i_lock);
1967 dput(tmp);
1968 goto out_iput;
1971 /* attach a disconnected dentry */
1972 add_flags = d_flags_for_inode(inode);
1974 if (disconnected)
1975 add_flags |= DCACHE_DISCONNECTED;
1977 spin_lock(&tmp->d_lock);
1978 __d_set_inode_and_type(tmp, inode, add_flags);
1979 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1980 hlist_bl_lock(&tmp->d_sb->s_anon);
1981 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1982 hlist_bl_unlock(&tmp->d_sb->s_anon);
1983 spin_unlock(&tmp->d_lock);
1984 spin_unlock(&inode->i_lock);
1985 security_d_instantiate(tmp, inode);
1987 return tmp;
1989 out_iput:
1990 if (res && !IS_ERR(res))
1991 security_d_instantiate(res, inode);
1992 iput(inode);
1993 return res;
1997 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1998 * @inode: inode to allocate the dentry for
2000 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
2001 * similar open by handle operations. The returned dentry may be anonymous,
2002 * or may have a full name (if the inode was already in the cache).
2004 * When called on a directory inode, we must ensure that the inode only ever
2005 * has one dentry. If a dentry is found, that is returned instead of
2006 * allocating a new one.
2008 * On successful return, the reference to the inode has been transferred
2009 * to the dentry. In case of an error the reference on the inode is released.
2010 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2011 * be passed in and the error will be propagated to the return value,
2012 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2014 struct dentry *d_obtain_alias(struct inode *inode)
2016 return __d_obtain_alias(inode, 1);
2018 EXPORT_SYMBOL(d_obtain_alias);
2021 * d_obtain_root - find or allocate a dentry for a given inode
2022 * @inode: inode to allocate the dentry for
2024 * Obtain an IS_ROOT dentry for the root of a filesystem.
2026 * We must ensure that directory inodes only ever have one dentry. If a
2027 * dentry is found, that is returned instead of allocating a new one.
2029 * On successful return, the reference to the inode has been transferred
2030 * to the dentry. In case of an error the reference on the inode is
2031 * released. A %NULL or IS_ERR inode may be passed in and will be the
2032 * error will be propagate to the return value, with a %NULL @inode
2033 * replaced by ERR_PTR(-ESTALE).
2035 struct dentry *d_obtain_root(struct inode *inode)
2037 return __d_obtain_alias(inode, 0);
2039 EXPORT_SYMBOL(d_obtain_root);
2042 * d_add_ci - lookup or allocate new dentry with case-exact name
2043 * @inode: the inode case-insensitive lookup has found
2044 * @dentry: the negative dentry that was passed to the parent's lookup func
2045 * @name: the case-exact name to be associated with the returned dentry
2047 * This is to avoid filling the dcache with case-insensitive names to the
2048 * same inode, only the actual correct case is stored in the dcache for
2049 * case-insensitive filesystems.
2051 * For a case-insensitive lookup match and if the the case-exact dentry
2052 * already exists in in the dcache, use it and return it.
2054 * If no entry exists with the exact case name, allocate new dentry with
2055 * the exact case, and return the spliced entry.
2057 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2058 struct qstr *name)
2060 struct dentry *found;
2061 struct dentry *new;
2064 * First check if a dentry matching the name already exists,
2065 * if not go ahead and create it now.
2067 found = d_hash_and_lookup(dentry->d_parent, name);
2068 if (!found) {
2069 new = d_alloc(dentry->d_parent, name);
2070 if (!new) {
2071 found = ERR_PTR(-ENOMEM);
2072 } else {
2073 found = d_splice_alias(inode, new);
2074 if (found) {
2075 dput(new);
2076 return found;
2078 return new;
2081 iput(inode);
2082 return found;
2084 EXPORT_SYMBOL(d_add_ci);
2087 * Do the slow-case of the dentry name compare.
2089 * Unlike the dentry_cmp() function, we need to atomically
2090 * load the name and length information, so that the
2091 * filesystem can rely on them, and can use the 'name' and
2092 * 'len' information without worrying about walking off the
2093 * end of memory etc.
2095 * Thus the read_seqcount_retry() and the "duplicate" info
2096 * in arguments (the low-level filesystem should not look
2097 * at the dentry inode or name contents directly, since
2098 * rename can change them while we're in RCU mode).
2100 enum slow_d_compare {
2101 D_COMP_OK,
2102 D_COMP_NOMATCH,
2103 D_COMP_SEQRETRY,
2106 static noinline enum slow_d_compare slow_dentry_cmp(
2107 const struct dentry *parent,
2108 struct dentry *dentry,
2109 unsigned int seq,
2110 const struct qstr *name)
2112 int tlen = dentry->d_name.len;
2113 const char *tname = dentry->d_name.name;
2115 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2116 cpu_relax();
2117 return D_COMP_SEQRETRY;
2119 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2120 return D_COMP_NOMATCH;
2121 return D_COMP_OK;
2125 * __d_lookup_rcu - search for a dentry (racy, store-free)
2126 * @parent: parent dentry
2127 * @name: qstr of name we wish to find
2128 * @seqp: returns d_seq value at the point where the dentry was found
2129 * Returns: dentry, or NULL
2131 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2132 * resolution (store-free path walking) design described in
2133 * Documentation/filesystems/path-lookup.txt.
2135 * This is not to be used outside core vfs.
2137 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2138 * held, and rcu_read_lock held. The returned dentry must not be stored into
2139 * without taking d_lock and checking d_seq sequence count against @seq
2140 * returned here.
2142 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2143 * function.
2145 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2146 * the returned dentry, so long as its parent's seqlock is checked after the
2147 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2148 * is formed, giving integrity down the path walk.
2150 * NOTE! The caller *has* to check the resulting dentry against the sequence
2151 * number we've returned before using any of the resulting dentry state!
2153 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2154 const struct qstr *name,
2155 unsigned *seqp)
2157 u64 hashlen = name->hash_len;
2158 const unsigned char *str = name->name;
2159 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2160 struct hlist_bl_node *node;
2161 struct dentry *dentry;
2164 * Note: There is significant duplication with __d_lookup_rcu which is
2165 * required to prevent single threaded performance regressions
2166 * especially on architectures where smp_rmb (in seqcounts) are costly.
2167 * Keep the two functions in sync.
2171 * The hash list is protected using RCU.
2173 * Carefully use d_seq when comparing a candidate dentry, to avoid
2174 * races with d_move().
2176 * It is possible that concurrent renames can mess up our list
2177 * walk here and result in missing our dentry, resulting in the
2178 * false-negative result. d_lookup() protects against concurrent
2179 * renames using rename_lock seqlock.
2181 * See Documentation/filesystems/path-lookup.txt for more details.
2183 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2184 unsigned seq;
2186 seqretry:
2188 * The dentry sequence count protects us from concurrent
2189 * renames, and thus protects parent and name fields.
2191 * The caller must perform a seqcount check in order
2192 * to do anything useful with the returned dentry.
2194 * NOTE! We do a "raw" seqcount_begin here. That means that
2195 * we don't wait for the sequence count to stabilize if it
2196 * is in the middle of a sequence change. If we do the slow
2197 * dentry compare, we will do seqretries until it is stable,
2198 * and if we end up with a successful lookup, we actually
2199 * want to exit RCU lookup anyway.
2201 seq = raw_seqcount_begin(&dentry->d_seq);
2202 if (dentry->d_parent != parent)
2203 continue;
2204 if (d_unhashed(dentry))
2205 continue;
2207 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2208 if (dentry->d_name.hash != hashlen_hash(hashlen))
2209 continue;
2210 *seqp = seq;
2211 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2212 case D_COMP_OK:
2213 return dentry;
2214 case D_COMP_NOMATCH:
2215 continue;
2216 default:
2217 goto seqretry;
2221 if (dentry->d_name.hash_len != hashlen)
2222 continue;
2223 *seqp = seq;
2224 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2225 return dentry;
2227 return NULL;
2231 * d_lookup - search for a dentry
2232 * @parent: parent dentry
2233 * @name: qstr of name we wish to find
2234 * Returns: dentry, or NULL
2236 * d_lookup searches the children of the parent dentry for the name in
2237 * question. If the dentry is found its reference count is incremented and the
2238 * dentry is returned. The caller must use dput to free the entry when it has
2239 * finished using it. %NULL is returned if the dentry does not exist.
2241 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2243 struct dentry *dentry;
2244 unsigned seq;
2246 do {
2247 seq = read_seqbegin(&rename_lock);
2248 dentry = __d_lookup(parent, name);
2249 if (dentry)
2250 break;
2251 } while (read_seqretry(&rename_lock, seq));
2252 return dentry;
2254 EXPORT_SYMBOL(d_lookup);
2257 * __d_lookup - search for a dentry (racy)
2258 * @parent: parent dentry
2259 * @name: qstr of name we wish to find
2260 * Returns: dentry, or NULL
2262 * __d_lookup is like d_lookup, however it may (rarely) return a
2263 * false-negative result due to unrelated rename activity.
2265 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2266 * however it must be used carefully, eg. with a following d_lookup in
2267 * the case of failure.
2269 * __d_lookup callers must be commented.
2271 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2273 unsigned int len = name->len;
2274 unsigned int hash = name->hash;
2275 const unsigned char *str = name->name;
2276 struct hlist_bl_head *b = d_hash(parent, hash);
2277 struct hlist_bl_node *node;
2278 struct dentry *found = NULL;
2279 struct dentry *dentry;
2282 * Note: There is significant duplication with __d_lookup_rcu which is
2283 * required to prevent single threaded performance regressions
2284 * especially on architectures where smp_rmb (in seqcounts) are costly.
2285 * Keep the two functions in sync.
2289 * The hash list is protected using RCU.
2291 * Take d_lock when comparing a candidate dentry, to avoid races
2292 * with d_move().
2294 * It is possible that concurrent renames can mess up our list
2295 * walk here and result in missing our dentry, resulting in the
2296 * false-negative result. d_lookup() protects against concurrent
2297 * renames using rename_lock seqlock.
2299 * See Documentation/filesystems/path-lookup.txt for more details.
2301 rcu_read_lock();
2303 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2305 if (dentry->d_name.hash != hash)
2306 continue;
2308 spin_lock(&dentry->d_lock);
2309 if (dentry->d_parent != parent)
2310 goto next;
2311 if (d_unhashed(dentry))
2312 goto next;
2315 * It is safe to compare names since d_move() cannot
2316 * change the qstr (protected by d_lock).
2318 if (parent->d_flags & DCACHE_OP_COMPARE) {
2319 int tlen = dentry->d_name.len;
2320 const char *tname = dentry->d_name.name;
2321 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2322 goto next;
2323 } else {
2324 if (dentry->d_name.len != len)
2325 goto next;
2326 if (dentry_cmp(dentry, str, len))
2327 goto next;
2330 dentry->d_lockref.count++;
2331 found = dentry;
2332 spin_unlock(&dentry->d_lock);
2333 break;
2334 next:
2335 spin_unlock(&dentry->d_lock);
2337 rcu_read_unlock();
2339 return found;
2343 * d_hash_and_lookup - hash the qstr then search for a dentry
2344 * @dir: Directory to search in
2345 * @name: qstr of name we wish to find
2347 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2349 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2352 * Check for a fs-specific hash function. Note that we must
2353 * calculate the standard hash first, as the d_op->d_hash()
2354 * routine may choose to leave the hash value unchanged.
2356 name->hash = full_name_hash(name->name, name->len);
2357 if (dir->d_flags & DCACHE_OP_HASH) {
2358 int err = dir->d_op->d_hash(dir, name);
2359 if (unlikely(err < 0))
2360 return ERR_PTR(err);
2362 return d_lookup(dir, name);
2364 EXPORT_SYMBOL(d_hash_and_lookup);
2367 * When a file is deleted, we have two options:
2368 * - turn this dentry into a negative dentry
2369 * - unhash this dentry and free it.
2371 * Usually, we want to just turn this into
2372 * a negative dentry, but if anybody else is
2373 * currently using the dentry or the inode
2374 * we can't do that and we fall back on removing
2375 * it from the hash queues and waiting for
2376 * it to be deleted later when it has no users
2380 * d_delete - delete a dentry
2381 * @dentry: The dentry to delete
2383 * Turn the dentry into a negative dentry if possible, otherwise
2384 * remove it from the hash queues so it can be deleted later
2387 void d_delete(struct dentry * dentry)
2389 struct inode *inode;
2390 int isdir = 0;
2392 * Are we the only user?
2394 again:
2395 spin_lock(&dentry->d_lock);
2396 inode = dentry->d_inode;
2397 isdir = S_ISDIR(inode->i_mode);
2398 if (dentry->d_lockref.count == 1) {
2399 if (!spin_trylock(&inode->i_lock)) {
2400 spin_unlock(&dentry->d_lock);
2401 cpu_relax();
2402 goto again;
2404 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2405 dentry_unlink_inode(dentry);
2406 fsnotify_nameremove(dentry, isdir);
2407 return;
2410 if (!d_unhashed(dentry))
2411 __d_drop(dentry);
2413 spin_unlock(&dentry->d_lock);
2415 fsnotify_nameremove(dentry, isdir);
2417 EXPORT_SYMBOL(d_delete);
2419 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2421 BUG_ON(!d_unhashed(entry));
2422 hlist_bl_lock(b);
2423 entry->d_flags |= DCACHE_RCUACCESS;
2424 hlist_bl_add_head_rcu(&entry->d_hash, b);
2425 hlist_bl_unlock(b);
2428 static void _d_rehash(struct dentry * entry)
2430 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2434 * d_rehash - add an entry back to the hash
2435 * @entry: dentry to add to the hash
2437 * Adds a dentry to the hash according to its name.
2440 void d_rehash(struct dentry * entry)
2442 spin_lock(&entry->d_lock);
2443 _d_rehash(entry);
2444 spin_unlock(&entry->d_lock);
2446 EXPORT_SYMBOL(d_rehash);
2449 * dentry_update_name_case - update case insensitive dentry with a new name
2450 * @dentry: dentry to be updated
2451 * @name: new name
2453 * Update a case insensitive dentry with new case of name.
2455 * dentry must have been returned by d_lookup with name @name. Old and new
2456 * name lengths must match (ie. no d_compare which allows mismatched name
2457 * lengths).
2459 * Parent inode i_mutex must be held over d_lookup and into this call (to
2460 * keep renames and concurrent inserts, and readdir(2) away).
2462 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2464 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2465 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2467 spin_lock(&dentry->d_lock);
2468 write_seqcount_begin(&dentry->d_seq);
2469 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2470 write_seqcount_end(&dentry->d_seq);
2471 spin_unlock(&dentry->d_lock);
2473 EXPORT_SYMBOL(dentry_update_name_case);
2475 static void swap_names(struct dentry *dentry, struct dentry *target)
2477 if (unlikely(dname_external(target))) {
2478 if (unlikely(dname_external(dentry))) {
2480 * Both external: swap the pointers
2482 swap(target->d_name.name, dentry->d_name.name);
2483 } else {
2485 * dentry:internal, target:external. Steal target's
2486 * storage and make target internal.
2488 memcpy(target->d_iname, dentry->d_name.name,
2489 dentry->d_name.len + 1);
2490 dentry->d_name.name = target->d_name.name;
2491 target->d_name.name = target->d_iname;
2493 } else {
2494 if (unlikely(dname_external(dentry))) {
2496 * dentry:external, target:internal. Give dentry's
2497 * storage to target and make dentry internal
2499 memcpy(dentry->d_iname, target->d_name.name,
2500 target->d_name.len + 1);
2501 target->d_name.name = dentry->d_name.name;
2502 dentry->d_name.name = dentry->d_iname;
2503 } else {
2505 * Both are internal.
2507 unsigned int i;
2508 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2509 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2510 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2511 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2512 swap(((long *) &dentry->d_iname)[i],
2513 ((long *) &target->d_iname)[i]);
2517 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2520 static void copy_name(struct dentry *dentry, struct dentry *target)
2522 struct external_name *old_name = NULL;
2523 if (unlikely(dname_external(dentry)))
2524 old_name = external_name(dentry);
2525 if (unlikely(dname_external(target))) {
2526 atomic_inc(&external_name(target)->u.count);
2527 dentry->d_name = target->d_name;
2528 } else {
2529 memcpy(dentry->d_iname, target->d_name.name,
2530 target->d_name.len + 1);
2531 dentry->d_name.name = dentry->d_iname;
2532 dentry->d_name.hash_len = target->d_name.hash_len;
2534 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2535 kfree_rcu(old_name, u.head);
2538 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2541 * XXXX: do we really need to take target->d_lock?
2543 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2544 spin_lock(&target->d_parent->d_lock);
2545 else {
2546 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2547 spin_lock(&dentry->d_parent->d_lock);
2548 spin_lock_nested(&target->d_parent->d_lock,
2549 DENTRY_D_LOCK_NESTED);
2550 } else {
2551 spin_lock(&target->d_parent->d_lock);
2552 spin_lock_nested(&dentry->d_parent->d_lock,
2553 DENTRY_D_LOCK_NESTED);
2556 if (target < dentry) {
2557 spin_lock_nested(&target->d_lock, 2);
2558 spin_lock_nested(&dentry->d_lock, 3);
2559 } else {
2560 spin_lock_nested(&dentry->d_lock, 2);
2561 spin_lock_nested(&target->d_lock, 3);
2565 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2567 if (target->d_parent != dentry->d_parent)
2568 spin_unlock(&dentry->d_parent->d_lock);
2569 if (target->d_parent != target)
2570 spin_unlock(&target->d_parent->d_lock);
2571 spin_unlock(&target->d_lock);
2572 spin_unlock(&dentry->d_lock);
2576 * When switching names, the actual string doesn't strictly have to
2577 * be preserved in the target - because we're dropping the target
2578 * anyway. As such, we can just do a simple memcpy() to copy over
2579 * the new name before we switch, unless we are going to rehash
2580 * it. Note that if we *do* unhash the target, we are not allowed
2581 * to rehash it without giving it a new name/hash key - whether
2582 * we swap or overwrite the names here, resulting name won't match
2583 * the reality in filesystem; it's only there for d_path() purposes.
2584 * Note that all of this is happening under rename_lock, so the
2585 * any hash lookup seeing it in the middle of manipulations will
2586 * be discarded anyway. So we do not care what happens to the hash
2587 * key in that case.
2590 * __d_move - move a dentry
2591 * @dentry: entry to move
2592 * @target: new dentry
2593 * @exchange: exchange the two dentries
2595 * Update the dcache to reflect the move of a file name. Negative
2596 * dcache entries should not be moved in this way. Caller must hold
2597 * rename_lock, the i_mutex of the source and target directories,
2598 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2600 static void __d_move(struct dentry *dentry, struct dentry *target,
2601 bool exchange)
2603 if (!dentry->d_inode)
2604 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2606 BUG_ON(d_ancestor(dentry, target));
2607 BUG_ON(d_ancestor(target, dentry));
2609 dentry_lock_for_move(dentry, target);
2611 write_seqcount_begin(&dentry->d_seq);
2612 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2614 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2617 * Move the dentry to the target hash queue. Don't bother checking
2618 * for the same hash queue because of how unlikely it is.
2620 __d_drop(dentry);
2621 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2624 * Unhash the target (d_delete() is not usable here). If exchanging
2625 * the two dentries, then rehash onto the other's hash queue.
2627 __d_drop(target);
2628 if (exchange) {
2629 __d_rehash(target,
2630 d_hash(dentry->d_parent, dentry->d_name.hash));
2633 /* Switch the names.. */
2634 if (exchange)
2635 swap_names(dentry, target);
2636 else
2637 copy_name(dentry, target);
2639 /* ... and switch them in the tree */
2640 if (IS_ROOT(dentry)) {
2641 /* splicing a tree */
2642 dentry->d_parent = target->d_parent;
2643 target->d_parent = target;
2644 list_del_init(&target->d_child);
2645 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2646 } else {
2647 /* swapping two dentries */
2648 swap(dentry->d_parent, target->d_parent);
2649 list_move(&target->d_child, &target->d_parent->d_subdirs);
2650 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2651 if (exchange)
2652 fsnotify_d_move(target);
2653 fsnotify_d_move(dentry);
2656 write_seqcount_end(&target->d_seq);
2657 write_seqcount_end(&dentry->d_seq);
2659 dentry_unlock_for_move(dentry, target);
2663 * d_move - move a dentry
2664 * @dentry: entry to move
2665 * @target: new dentry
2667 * Update the dcache to reflect the move of a file name. Negative
2668 * dcache entries should not be moved in this way. See the locking
2669 * requirements for __d_move.
2671 void d_move(struct dentry *dentry, struct dentry *target)
2673 write_seqlock(&rename_lock);
2674 __d_move(dentry, target, false);
2675 write_sequnlock(&rename_lock);
2677 EXPORT_SYMBOL(d_move);
2680 * d_exchange - exchange two dentries
2681 * @dentry1: first dentry
2682 * @dentry2: second dentry
2684 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2686 write_seqlock(&rename_lock);
2688 WARN_ON(!dentry1->d_inode);
2689 WARN_ON(!dentry2->d_inode);
2690 WARN_ON(IS_ROOT(dentry1));
2691 WARN_ON(IS_ROOT(dentry2));
2693 __d_move(dentry1, dentry2, true);
2695 write_sequnlock(&rename_lock);
2699 * d_ancestor - search for an ancestor
2700 * @p1: ancestor dentry
2701 * @p2: child dentry
2703 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2704 * an ancestor of p2, else NULL.
2706 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2708 struct dentry *p;
2710 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2711 if (p->d_parent == p1)
2712 return p;
2714 return NULL;
2718 * This helper attempts to cope with remotely renamed directories
2720 * It assumes that the caller is already holding
2721 * dentry->d_parent->d_inode->i_mutex, and rename_lock
2723 * Note: If ever the locking in lock_rename() changes, then please
2724 * remember to update this too...
2726 static int __d_unalias(struct inode *inode,
2727 struct dentry *dentry, struct dentry *alias)
2729 struct mutex *m1 = NULL, *m2 = NULL;
2730 int ret = -ESTALE;
2732 /* If alias and dentry share a parent, then no extra locks required */
2733 if (alias->d_parent == dentry->d_parent)
2734 goto out_unalias;
2736 /* See lock_rename() */
2737 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2738 goto out_err;
2739 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2740 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2741 goto out_err;
2742 m2 = &alias->d_parent->d_inode->i_mutex;
2743 out_unalias:
2744 __d_move(alias, dentry, false);
2745 ret = 0;
2746 out_err:
2747 if (m2)
2748 mutex_unlock(m2);
2749 if (m1)
2750 mutex_unlock(m1);
2751 return ret;
2755 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2756 * @inode: the inode which may have a disconnected dentry
2757 * @dentry: a negative dentry which we want to point to the inode.
2759 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2760 * place of the given dentry and return it, else simply d_add the inode
2761 * to the dentry and return NULL.
2763 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2764 * we should error out: directories can't have multiple aliases.
2766 * This is needed in the lookup routine of any filesystem that is exportable
2767 * (via knfsd) so that we can build dcache paths to directories effectively.
2769 * If a dentry was found and moved, then it is returned. Otherwise NULL
2770 * is returned. This matches the expected return value of ->lookup.
2772 * Cluster filesystems may call this function with a negative, hashed dentry.
2773 * In that case, we know that the inode will be a regular file, and also this
2774 * will only occur during atomic_open. So we need to check for the dentry
2775 * being already hashed only in the final case.
2777 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2779 if (IS_ERR(inode))
2780 return ERR_CAST(inode);
2782 BUG_ON(!d_unhashed(dentry));
2784 if (!inode) {
2785 __d_instantiate(dentry, NULL);
2786 goto out;
2788 spin_lock(&inode->i_lock);
2789 if (S_ISDIR(inode->i_mode)) {
2790 struct dentry *new = __d_find_any_alias(inode);
2791 if (unlikely(new)) {
2792 /* The reference to new ensures it remains an alias */
2793 spin_unlock(&inode->i_lock);
2794 write_seqlock(&rename_lock);
2795 if (unlikely(d_ancestor(new, dentry))) {
2796 write_sequnlock(&rename_lock);
2797 dput(new);
2798 new = ERR_PTR(-ELOOP);
2799 pr_warn_ratelimited(
2800 "VFS: Lookup of '%s' in %s %s"
2801 " would have caused loop\n",
2802 dentry->d_name.name,
2803 inode->i_sb->s_type->name,
2804 inode->i_sb->s_id);
2805 } else if (!IS_ROOT(new)) {
2806 int err = __d_unalias(inode, dentry, new);
2807 write_sequnlock(&rename_lock);
2808 if (err) {
2809 dput(new);
2810 new = ERR_PTR(err);
2812 } else {
2813 __d_move(new, dentry, false);
2814 write_sequnlock(&rename_lock);
2815 security_d_instantiate(new, inode);
2817 iput(inode);
2818 return new;
2821 /* already taking inode->i_lock, so d_add() by hand */
2822 __d_instantiate(dentry, inode);
2823 spin_unlock(&inode->i_lock);
2824 out:
2825 security_d_instantiate(dentry, inode);
2826 d_rehash(dentry);
2827 return NULL;
2829 EXPORT_SYMBOL(d_splice_alias);
2831 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2833 *buflen -= namelen;
2834 if (*buflen < 0)
2835 return -ENAMETOOLONG;
2836 *buffer -= namelen;
2837 memcpy(*buffer, str, namelen);
2838 return 0;
2842 * prepend_name - prepend a pathname in front of current buffer pointer
2843 * @buffer: buffer pointer
2844 * @buflen: allocated length of the buffer
2845 * @name: name string and length qstr structure
2847 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2848 * make sure that either the old or the new name pointer and length are
2849 * fetched. However, there may be mismatch between length and pointer.
2850 * The length cannot be trusted, we need to copy it byte-by-byte until
2851 * the length is reached or a null byte is found. It also prepends "/" at
2852 * the beginning of the name. The sequence number check at the caller will
2853 * retry it again when a d_move() does happen. So any garbage in the buffer
2854 * due to mismatched pointer and length will be discarded.
2856 * Data dependency barrier is needed to make sure that we see that terminating
2857 * NUL. Alpha strikes again, film at 11...
2859 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2861 const char *dname = ACCESS_ONCE(name->name);
2862 u32 dlen = ACCESS_ONCE(name->len);
2863 char *p;
2865 smp_read_barrier_depends();
2867 *buflen -= dlen + 1;
2868 if (*buflen < 0)
2869 return -ENAMETOOLONG;
2870 p = *buffer -= dlen + 1;
2871 *p++ = '/';
2872 while (dlen--) {
2873 char c = *dname++;
2874 if (!c)
2875 break;
2876 *p++ = c;
2878 return 0;
2882 * prepend_path - Prepend path string to a buffer
2883 * @path: the dentry/vfsmount to report
2884 * @root: root vfsmnt/dentry
2885 * @buffer: pointer to the end of the buffer
2886 * @buflen: pointer to buffer length
2888 * The function will first try to write out the pathname without taking any
2889 * lock other than the RCU read lock to make sure that dentries won't go away.
2890 * It only checks the sequence number of the global rename_lock as any change
2891 * in the dentry's d_seq will be preceded by changes in the rename_lock
2892 * sequence number. If the sequence number had been changed, it will restart
2893 * the whole pathname back-tracing sequence again by taking the rename_lock.
2894 * In this case, there is no need to take the RCU read lock as the recursive
2895 * parent pointer references will keep the dentry chain alive as long as no
2896 * rename operation is performed.
2898 static int prepend_path(const struct path *path,
2899 const struct path *root,
2900 char **buffer, int *buflen)
2902 struct dentry *dentry;
2903 struct vfsmount *vfsmnt;
2904 struct mount *mnt;
2905 int error = 0;
2906 unsigned seq, m_seq = 0;
2907 char *bptr;
2908 int blen;
2910 rcu_read_lock();
2911 restart_mnt:
2912 read_seqbegin_or_lock(&mount_lock, &m_seq);
2913 seq = 0;
2914 rcu_read_lock();
2915 restart:
2916 bptr = *buffer;
2917 blen = *buflen;
2918 error = 0;
2919 dentry = path->dentry;
2920 vfsmnt = path->mnt;
2921 mnt = real_mount(vfsmnt);
2922 read_seqbegin_or_lock(&rename_lock, &seq);
2923 while (dentry != root->dentry || vfsmnt != root->mnt) {
2924 struct dentry * parent;
2926 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2927 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2928 /* Escaped? */
2929 if (dentry != vfsmnt->mnt_root) {
2930 bptr = *buffer;
2931 blen = *buflen;
2932 error = 3;
2933 break;
2935 /* Global root? */
2936 if (mnt != parent) {
2937 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2938 mnt = parent;
2939 vfsmnt = &mnt->mnt;
2940 continue;
2942 if (!error)
2943 error = is_mounted(vfsmnt) ? 1 : 2;
2944 break;
2946 parent = dentry->d_parent;
2947 prefetch(parent);
2948 error = prepend_name(&bptr, &blen, &dentry->d_name);
2949 if (error)
2950 break;
2952 dentry = parent;
2954 if (!(seq & 1))
2955 rcu_read_unlock();
2956 if (need_seqretry(&rename_lock, seq)) {
2957 seq = 1;
2958 goto restart;
2960 done_seqretry(&rename_lock, seq);
2962 if (!(m_seq & 1))
2963 rcu_read_unlock();
2964 if (need_seqretry(&mount_lock, m_seq)) {
2965 m_seq = 1;
2966 goto restart_mnt;
2968 done_seqretry(&mount_lock, m_seq);
2970 if (error >= 0 && bptr == *buffer) {
2971 if (--blen < 0)
2972 error = -ENAMETOOLONG;
2973 else
2974 *--bptr = '/';
2976 *buffer = bptr;
2977 *buflen = blen;
2978 return error;
2982 * __d_path - return the path of a dentry
2983 * @path: the dentry/vfsmount to report
2984 * @root: root vfsmnt/dentry
2985 * @buf: buffer to return value in
2986 * @buflen: buffer length
2988 * Convert a dentry into an ASCII path name.
2990 * Returns a pointer into the buffer or an error code if the
2991 * path was too long.
2993 * "buflen" should be positive.
2995 * If the path is not reachable from the supplied root, return %NULL.
2997 char *__d_path(const struct path *path,
2998 const struct path *root,
2999 char *buf, int buflen)
3001 char *res = buf + buflen;
3002 int error;
3004 prepend(&res, &buflen, "\0", 1);
3005 error = prepend_path(path, root, &res, &buflen);
3007 if (error < 0)
3008 return ERR_PTR(error);
3009 if (error > 0)
3010 return NULL;
3011 return res;
3014 char *d_absolute_path(const struct path *path,
3015 char *buf, int buflen)
3017 struct path root = {};
3018 char *res = buf + buflen;
3019 int error;
3021 prepend(&res, &buflen, "\0", 1);
3022 error = prepend_path(path, &root, &res, &buflen);
3024 if (error > 1)
3025 error = -EINVAL;
3026 if (error < 0)
3027 return ERR_PTR(error);
3028 return res;
3032 * same as __d_path but appends "(deleted)" for unlinked files.
3034 static int path_with_deleted(const struct path *path,
3035 const struct path *root,
3036 char **buf, int *buflen)
3038 prepend(buf, buflen, "\0", 1);
3039 if (d_unlinked(path->dentry)) {
3040 int error = prepend(buf, buflen, " (deleted)", 10);
3041 if (error)
3042 return error;
3045 return prepend_path(path, root, buf, buflen);
3048 static int prepend_unreachable(char **buffer, int *buflen)
3050 return prepend(buffer, buflen, "(unreachable)", 13);
3053 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3055 unsigned seq;
3057 do {
3058 seq = read_seqcount_begin(&fs->seq);
3059 *root = fs->root;
3060 } while (read_seqcount_retry(&fs->seq, seq));
3064 * d_path - return the path of a dentry
3065 * @path: path to report
3066 * @buf: buffer to return value in
3067 * @buflen: buffer length
3069 * Convert a dentry into an ASCII path name. If the entry has been deleted
3070 * the string " (deleted)" is appended. Note that this is ambiguous.
3072 * Returns a pointer into the buffer or an error code if the path was
3073 * too long. Note: Callers should use the returned pointer, not the passed
3074 * in buffer, to use the name! The implementation often starts at an offset
3075 * into the buffer, and may leave 0 bytes at the start.
3077 * "buflen" should be positive.
3079 char *d_path(const struct path *path, char *buf, int buflen)
3081 char *res = buf + buflen;
3082 struct path root;
3083 int error;
3086 * We have various synthetic filesystems that never get mounted. On
3087 * these filesystems dentries are never used for lookup purposes, and
3088 * thus don't need to be hashed. They also don't need a name until a
3089 * user wants to identify the object in /proc/pid/fd/. The little hack
3090 * below allows us to generate a name for these objects on demand:
3092 * Some pseudo inodes are mountable. When they are mounted
3093 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3094 * and instead have d_path return the mounted path.
3096 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3097 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3098 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3100 rcu_read_lock();
3101 get_fs_root_rcu(current->fs, &root);
3102 error = path_with_deleted(path, &root, &res, &buflen);
3103 rcu_read_unlock();
3105 if (error < 0)
3106 res = ERR_PTR(error);
3107 return res;
3109 EXPORT_SYMBOL(d_path);
3112 * Helper function for dentry_operations.d_dname() members
3114 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3115 const char *fmt, ...)
3117 va_list args;
3118 char temp[64];
3119 int sz;
3121 va_start(args, fmt);
3122 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3123 va_end(args);
3125 if (sz > sizeof(temp) || sz > buflen)
3126 return ERR_PTR(-ENAMETOOLONG);
3128 buffer += buflen - sz;
3129 return memcpy(buffer, temp, sz);
3132 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3134 char *end = buffer + buflen;
3135 /* these dentries are never renamed, so d_lock is not needed */
3136 if (prepend(&end, &buflen, " (deleted)", 11) ||
3137 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3138 prepend(&end, &buflen, "/", 1))
3139 end = ERR_PTR(-ENAMETOOLONG);
3140 return end;
3142 EXPORT_SYMBOL(simple_dname);
3145 * Write full pathname from the root of the filesystem into the buffer.
3147 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3149 struct dentry *dentry;
3150 char *end, *retval;
3151 int len, seq = 0;
3152 int error = 0;
3154 if (buflen < 2)
3155 goto Elong;
3157 rcu_read_lock();
3158 restart:
3159 dentry = d;
3160 end = buf + buflen;
3161 len = buflen;
3162 prepend(&end, &len, "\0", 1);
3163 /* Get '/' right */
3164 retval = end-1;
3165 *retval = '/';
3166 read_seqbegin_or_lock(&rename_lock, &seq);
3167 while (!IS_ROOT(dentry)) {
3168 struct dentry *parent = dentry->d_parent;
3170 prefetch(parent);
3171 error = prepend_name(&end, &len, &dentry->d_name);
3172 if (error)
3173 break;
3175 retval = end;
3176 dentry = parent;
3178 if (!(seq & 1))
3179 rcu_read_unlock();
3180 if (need_seqretry(&rename_lock, seq)) {
3181 seq = 1;
3182 goto restart;
3184 done_seqretry(&rename_lock, seq);
3185 if (error)
3186 goto Elong;
3187 return retval;
3188 Elong:
3189 return ERR_PTR(-ENAMETOOLONG);
3192 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3194 return __dentry_path(dentry, buf, buflen);
3196 EXPORT_SYMBOL(dentry_path_raw);
3198 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3200 char *p = NULL;
3201 char *retval;
3203 if (d_unlinked(dentry)) {
3204 p = buf + buflen;
3205 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3206 goto Elong;
3207 buflen++;
3209 retval = __dentry_path(dentry, buf, buflen);
3210 if (!IS_ERR(retval) && p)
3211 *p = '/'; /* restore '/' overriden with '\0' */
3212 return retval;
3213 Elong:
3214 return ERR_PTR(-ENAMETOOLONG);
3217 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3218 struct path *pwd)
3220 unsigned seq;
3222 do {
3223 seq = read_seqcount_begin(&fs->seq);
3224 *root = fs->root;
3225 *pwd = fs->pwd;
3226 } while (read_seqcount_retry(&fs->seq, seq));
3230 * NOTE! The user-level library version returns a
3231 * character pointer. The kernel system call just
3232 * returns the length of the buffer filled (which
3233 * includes the ending '\0' character), or a negative
3234 * error value. So libc would do something like
3236 * char *getcwd(char * buf, size_t size)
3238 * int retval;
3240 * retval = sys_getcwd(buf, size);
3241 * if (retval >= 0)
3242 * return buf;
3243 * errno = -retval;
3244 * return NULL;
3247 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3249 int error;
3250 struct path pwd, root;
3251 char *page = __getname();
3253 if (!page)
3254 return -ENOMEM;
3256 rcu_read_lock();
3257 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3259 error = -ENOENT;
3260 if (!d_unlinked(pwd.dentry)) {
3261 unsigned long len;
3262 char *cwd = page + PATH_MAX;
3263 int buflen = PATH_MAX;
3265 prepend(&cwd, &buflen, "\0", 1);
3266 error = prepend_path(&pwd, &root, &cwd, &buflen);
3267 rcu_read_unlock();
3269 if (error < 0)
3270 goto out;
3272 /* Unreachable from current root */
3273 if (error > 0) {
3274 error = prepend_unreachable(&cwd, &buflen);
3275 if (error)
3276 goto out;
3279 error = -ERANGE;
3280 len = PATH_MAX + page - cwd;
3281 if (len <= size) {
3282 error = len;
3283 if (copy_to_user(buf, cwd, len))
3284 error = -EFAULT;
3286 } else {
3287 rcu_read_unlock();
3290 out:
3291 __putname(page);
3292 return error;
3296 * Test whether new_dentry is a subdirectory of old_dentry.
3298 * Trivially implemented using the dcache structure
3302 * is_subdir - is new dentry a subdirectory of old_dentry
3303 * @new_dentry: new dentry
3304 * @old_dentry: old dentry
3306 * Returns true if new_dentry is a subdirectory of the parent (at any depth).
3307 * Returns false otherwise.
3308 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3311 bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3313 bool result;
3314 unsigned seq;
3316 if (new_dentry == old_dentry)
3317 return true;
3319 do {
3320 /* for restarting inner loop in case of seq retry */
3321 seq = read_seqbegin(&rename_lock);
3323 * Need rcu_readlock to protect against the d_parent trashing
3324 * due to d_move
3326 rcu_read_lock();
3327 if (d_ancestor(old_dentry, new_dentry))
3328 result = true;
3329 else
3330 result = false;
3331 rcu_read_unlock();
3332 } while (read_seqretry(&rename_lock, seq));
3334 return result;
3337 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3339 struct dentry *root = data;
3340 if (dentry != root) {
3341 if (d_unhashed(dentry) || !dentry->d_inode)
3342 return D_WALK_SKIP;
3344 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3345 dentry->d_flags |= DCACHE_GENOCIDE;
3346 dentry->d_lockref.count--;
3349 return D_WALK_CONTINUE;
3352 void d_genocide(struct dentry *parent)
3354 d_walk(parent, parent, d_genocide_kill, NULL);
3357 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3359 inode_dec_link_count(inode);
3360 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3361 !hlist_unhashed(&dentry->d_u.d_alias) ||
3362 !d_unlinked(dentry));
3363 spin_lock(&dentry->d_parent->d_lock);
3364 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3365 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3366 (unsigned long long)inode->i_ino);
3367 spin_unlock(&dentry->d_lock);
3368 spin_unlock(&dentry->d_parent->d_lock);
3369 d_instantiate(dentry, inode);
3371 EXPORT_SYMBOL(d_tmpfile);
3373 static __initdata unsigned long dhash_entries;
3374 static int __init set_dhash_entries(char *str)
3376 if (!str)
3377 return 0;
3378 dhash_entries = simple_strtoul(str, &str, 0);
3379 return 1;
3381 __setup("dhash_entries=", set_dhash_entries);
3383 static void __init dcache_init_early(void)
3385 unsigned int loop;
3387 /* If hashes are distributed across NUMA nodes, defer
3388 * hash allocation until vmalloc space is available.
3390 if (hashdist)
3391 return;
3393 dentry_hashtable =
3394 alloc_large_system_hash("Dentry cache",
3395 sizeof(struct hlist_bl_head),
3396 dhash_entries,
3398 HASH_EARLY,
3399 &d_hash_shift,
3400 &d_hash_mask,
3404 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3405 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3408 static void __init dcache_init(void)
3410 unsigned int loop;
3413 * A constructor could be added for stable state like the lists,
3414 * but it is probably not worth it because of the cache nature
3415 * of the dcache.
3417 dentry_cache = KMEM_CACHE(dentry,
3418 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3420 /* Hash may have been set up in dcache_init_early */
3421 if (!hashdist)
3422 return;
3424 dentry_hashtable =
3425 alloc_large_system_hash("Dentry cache",
3426 sizeof(struct hlist_bl_head),
3427 dhash_entries,
3430 &d_hash_shift,
3431 &d_hash_mask,
3435 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3436 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3439 /* SLAB cache for __getname() consumers */
3440 struct kmem_cache *names_cachep __read_mostly;
3441 EXPORT_SYMBOL(names_cachep);
3443 EXPORT_SYMBOL(d_genocide);
3445 void __init vfs_caches_init_early(void)
3447 dcache_init_early();
3448 inode_init_early();
3451 void __init vfs_caches_init(void)
3453 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3454 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3456 dcache_init();
3457 inode_init();
3458 files_init();
3459 files_maxfiles_init();
3460 mnt_init();
3461 bdev_cache_init();
3462 chrdev_init();