ALSA: usb-audio: Add a quirk for Plantronics BT600
[linux/fpc-iii.git] / fs / dcache.c
blob11d466bbfb0b03197389e9ecf3546edbe6e5c271
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;
272 static inline void __d_set_inode_and_type(struct dentry *dentry,
273 struct inode *inode,
274 unsigned type_flags)
276 unsigned flags;
278 dentry->d_inode = inode;
279 flags = READ_ONCE(dentry->d_flags);
280 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
281 flags |= type_flags;
282 WRITE_ONCE(dentry->d_flags, flags);
285 static inline void __d_clear_type_and_inode(struct dentry *dentry)
287 unsigned flags = READ_ONCE(dentry->d_flags);
289 flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
290 WRITE_ONCE(dentry->d_flags, flags);
291 dentry->d_inode = NULL;
294 static void dentry_free(struct dentry *dentry)
296 WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
297 if (unlikely(dname_external(dentry))) {
298 struct external_name *p = external_name(dentry);
299 if (likely(atomic_dec_and_test(&p->u.count))) {
300 call_rcu(&dentry->d_u.d_rcu, __d_free_external);
301 return;
304 /* if dentry was never visible to RCU, immediate free is OK */
305 if (!(dentry->d_flags & DCACHE_RCUACCESS))
306 __d_free(&dentry->d_u.d_rcu);
307 else
308 call_rcu(&dentry->d_u.d_rcu, __d_free);
312 * dentry_rcuwalk_invalidate - invalidate in-progress rcu-walk lookups
313 * @dentry: the target dentry
314 * After this call, in-progress rcu-walk path lookup will fail. This
315 * should be called after unhashing, and after changing d_inode (if
316 * the dentry has not already been unhashed).
318 static inline void dentry_rcuwalk_invalidate(struct dentry *dentry)
320 lockdep_assert_held(&dentry->d_lock);
321 /* Go through am invalidation barrier */
322 write_seqcount_invalidate(&dentry->d_seq);
326 * Release the dentry's inode, using the filesystem
327 * d_iput() operation if defined. Dentry has no refcount
328 * and is unhashed.
330 static void dentry_iput(struct dentry * dentry)
331 __releases(dentry->d_lock)
332 __releases(dentry->d_inode->i_lock)
334 struct inode *inode = dentry->d_inode;
335 if (inode) {
336 __d_clear_type_and_inode(dentry);
337 hlist_del_init(&dentry->d_u.d_alias);
338 spin_unlock(&dentry->d_lock);
339 spin_unlock(&inode->i_lock);
340 if (!inode->i_nlink)
341 fsnotify_inoderemove(inode);
342 if (dentry->d_op && dentry->d_op->d_iput)
343 dentry->d_op->d_iput(dentry, inode);
344 else
345 iput(inode);
346 } else {
347 spin_unlock(&dentry->d_lock);
352 * Release the dentry's inode, using the filesystem
353 * d_iput() operation if defined. dentry remains in-use.
355 static void dentry_unlink_inode(struct dentry * dentry)
356 __releases(dentry->d_lock)
357 __releases(dentry->d_inode->i_lock)
359 struct inode *inode = dentry->d_inode;
361 raw_write_seqcount_begin(&dentry->d_seq);
362 __d_clear_type_and_inode(dentry);
363 hlist_del_init(&dentry->d_u.d_alias);
364 raw_write_seqcount_end(&dentry->d_seq);
365 spin_unlock(&dentry->d_lock);
366 spin_unlock(&inode->i_lock);
367 if (!inode->i_nlink)
368 fsnotify_inoderemove(inode);
369 if (dentry->d_op && dentry->d_op->d_iput)
370 dentry->d_op->d_iput(dentry, inode);
371 else
372 iput(inode);
376 * The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
377 * is in use - which includes both the "real" per-superblock
378 * LRU list _and_ the DCACHE_SHRINK_LIST use.
380 * The DCACHE_SHRINK_LIST bit is set whenever the dentry is
381 * on the shrink list (ie not on the superblock LRU list).
383 * The per-cpu "nr_dentry_unused" counters are updated with
384 * the DCACHE_LRU_LIST bit.
386 * These helper functions make sure we always follow the
387 * rules. d_lock must be held by the caller.
389 #define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
390 static void d_lru_add(struct dentry *dentry)
392 D_FLAG_VERIFY(dentry, 0);
393 dentry->d_flags |= DCACHE_LRU_LIST;
394 this_cpu_inc(nr_dentry_unused);
395 WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
398 static void d_lru_del(struct dentry *dentry)
400 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
401 dentry->d_flags &= ~DCACHE_LRU_LIST;
402 this_cpu_dec(nr_dentry_unused);
403 WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
406 static void d_shrink_del(struct dentry *dentry)
408 D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
409 list_del_init(&dentry->d_lru);
410 dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
411 this_cpu_dec(nr_dentry_unused);
414 static void d_shrink_add(struct dentry *dentry, struct list_head *list)
416 D_FLAG_VERIFY(dentry, 0);
417 list_add(&dentry->d_lru, list);
418 dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
419 this_cpu_inc(nr_dentry_unused);
423 * These can only be called under the global LRU lock, ie during the
424 * callback for freeing the LRU list. "isolate" removes it from the
425 * LRU lists entirely, while shrink_move moves it to the indicated
426 * private list.
428 static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
430 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
431 dentry->d_flags &= ~DCACHE_LRU_LIST;
432 this_cpu_dec(nr_dentry_unused);
433 list_lru_isolate(lru, &dentry->d_lru);
436 static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
437 struct list_head *list)
439 D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
440 dentry->d_flags |= DCACHE_SHRINK_LIST;
441 list_lru_isolate_move(lru, &dentry->d_lru, list);
445 * dentry_lru_(add|del)_list) must be called with d_lock held.
447 static void dentry_lru_add(struct dentry *dentry)
449 if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
450 d_lru_add(dentry);
454 * d_drop - drop a dentry
455 * @dentry: dentry to drop
457 * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
458 * be found through a VFS lookup any more. Note that this is different from
459 * deleting the dentry - d_delete will try to mark the dentry negative if
460 * possible, giving a successful _negative_ lookup, while d_drop will
461 * just make the cache lookup fail.
463 * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
464 * reason (NFS timeouts or autofs deletes).
466 * __d_drop requires dentry->d_lock.
468 void __d_drop(struct dentry *dentry)
470 if (!d_unhashed(dentry)) {
471 struct hlist_bl_head *b;
473 * Hashed dentries are normally on the dentry hashtable,
474 * with the exception of those newly allocated by
475 * d_obtain_alias, which are always IS_ROOT:
477 if (unlikely(IS_ROOT(dentry)))
478 b = &dentry->d_sb->s_anon;
479 else
480 b = d_hash(dentry->d_parent, dentry->d_name.hash);
482 hlist_bl_lock(b);
483 __hlist_bl_del(&dentry->d_hash);
484 dentry->d_hash.pprev = NULL;
485 hlist_bl_unlock(b);
486 dentry_rcuwalk_invalidate(dentry);
489 EXPORT_SYMBOL(__d_drop);
491 void d_drop(struct dentry *dentry)
493 spin_lock(&dentry->d_lock);
494 __d_drop(dentry);
495 spin_unlock(&dentry->d_lock);
497 EXPORT_SYMBOL(d_drop);
499 static void __dentry_kill(struct dentry *dentry)
501 struct dentry *parent = NULL;
502 bool can_free = true;
503 if (!IS_ROOT(dentry))
504 parent = dentry->d_parent;
507 * The dentry is now unrecoverably dead to the world.
509 lockref_mark_dead(&dentry->d_lockref);
512 * inform the fs via d_prune that this dentry is about to be
513 * unhashed and destroyed.
515 if (dentry->d_flags & DCACHE_OP_PRUNE)
516 dentry->d_op->d_prune(dentry);
518 if (dentry->d_flags & DCACHE_LRU_LIST) {
519 if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
520 d_lru_del(dentry);
522 /* if it was on the hash then remove it */
523 __d_drop(dentry);
524 __list_del_entry(&dentry->d_child);
526 * Inform d_walk() that we are no longer attached to the
527 * dentry tree
529 dentry->d_flags |= DCACHE_DENTRY_KILLED;
530 if (parent)
531 spin_unlock(&parent->d_lock);
532 dentry_iput(dentry);
534 * dentry_iput drops the locks, at which point nobody (except
535 * transient RCU lookups) can reach this dentry.
537 BUG_ON(dentry->d_lockref.count > 0);
538 this_cpu_dec(nr_dentry);
539 if (dentry->d_op && dentry->d_op->d_release)
540 dentry->d_op->d_release(dentry);
542 spin_lock(&dentry->d_lock);
543 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
544 dentry->d_flags |= DCACHE_MAY_FREE;
545 can_free = false;
547 spin_unlock(&dentry->d_lock);
548 if (likely(can_free))
549 dentry_free(dentry);
553 * Finish off a dentry we've decided to kill.
554 * dentry->d_lock must be held, returns with it unlocked.
555 * If ref is non-zero, then decrement the refcount too.
556 * Returns dentry requiring refcount drop, or NULL if we're done.
558 static struct dentry *dentry_kill(struct dentry *dentry)
559 __releases(dentry->d_lock)
561 struct inode *inode = dentry->d_inode;
562 struct dentry *parent = NULL;
564 if (inode && unlikely(!spin_trylock(&inode->i_lock)))
565 goto failed;
567 if (!IS_ROOT(dentry)) {
568 parent = dentry->d_parent;
569 if (unlikely(!spin_trylock(&parent->d_lock))) {
570 if (inode)
571 spin_unlock(&inode->i_lock);
572 goto failed;
576 __dentry_kill(dentry);
577 return parent;
579 failed:
580 spin_unlock(&dentry->d_lock);
581 return dentry; /* try again with same dentry */
584 static inline struct dentry *lock_parent(struct dentry *dentry)
586 struct dentry *parent = dentry->d_parent;
587 if (IS_ROOT(dentry))
588 return NULL;
589 if (unlikely(dentry->d_lockref.count < 0))
590 return NULL;
591 if (likely(spin_trylock(&parent->d_lock)))
592 return parent;
593 rcu_read_lock();
594 spin_unlock(&dentry->d_lock);
595 again:
596 parent = ACCESS_ONCE(dentry->d_parent);
597 spin_lock(&parent->d_lock);
599 * We can't blindly lock dentry until we are sure
600 * that we won't violate the locking order.
601 * Any changes of dentry->d_parent must have
602 * been done with parent->d_lock held, so
603 * spin_lock() above is enough of a barrier
604 * for checking if it's still our child.
606 if (unlikely(parent != dentry->d_parent)) {
607 spin_unlock(&parent->d_lock);
608 goto again;
610 rcu_read_unlock();
611 if (parent != dentry)
612 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
613 else
614 parent = NULL;
615 return parent;
619 * Try to do a lockless dput(), and return whether that was successful.
621 * If unsuccessful, we return false, having already taken the dentry lock.
623 * The caller needs to hold the RCU read lock, so that the dentry is
624 * guaranteed to stay around even if the refcount goes down to zero!
626 static inline bool fast_dput(struct dentry *dentry)
628 int ret;
629 unsigned int d_flags;
632 * If we have a d_op->d_delete() operation, we sould not
633 * let the dentry count go to zero, so use "put_or_lock".
635 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
636 return lockref_put_or_lock(&dentry->d_lockref);
639 * .. otherwise, we can try to just decrement the
640 * lockref optimistically.
642 ret = lockref_put_return(&dentry->d_lockref);
645 * If the lockref_put_return() failed due to the lock being held
646 * by somebody else, the fast path has failed. We will need to
647 * get the lock, and then check the count again.
649 if (unlikely(ret < 0)) {
650 spin_lock(&dentry->d_lock);
651 if (dentry->d_lockref.count > 1) {
652 dentry->d_lockref.count--;
653 spin_unlock(&dentry->d_lock);
654 return 1;
656 return 0;
660 * If we weren't the last ref, we're done.
662 if (ret)
663 return 1;
666 * Careful, careful. The reference count went down
667 * to zero, but we don't hold the dentry lock, so
668 * somebody else could get it again, and do another
669 * dput(), and we need to not race with that.
671 * However, there is a very special and common case
672 * where we don't care, because there is nothing to
673 * do: the dentry is still hashed, it does not have
674 * a 'delete' op, and it's referenced and already on
675 * the LRU list.
677 * NOTE! Since we aren't locked, these values are
678 * not "stable". However, it is sufficient that at
679 * some point after we dropped the reference the
680 * dentry was hashed and the flags had the proper
681 * value. Other dentry users may have re-gotten
682 * a reference to the dentry and change that, but
683 * our work is done - we can leave the dentry
684 * around with a zero refcount.
686 smp_rmb();
687 d_flags = ACCESS_ONCE(dentry->d_flags);
688 d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
690 /* Nothing to do? Dropping the reference was all we needed? */
691 if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
692 return 1;
695 * Not the fast normal case? Get the lock. We've already decremented
696 * the refcount, but we'll need to re-check the situation after
697 * getting the lock.
699 spin_lock(&dentry->d_lock);
702 * Did somebody else grab a reference to it in the meantime, and
703 * we're no longer the last user after all? Alternatively, somebody
704 * else could have killed it and marked it dead. Either way, we
705 * don't need to do anything else.
707 if (dentry->d_lockref.count) {
708 spin_unlock(&dentry->d_lock);
709 return 1;
713 * Re-get the reference we optimistically dropped. We hold the
714 * lock, and we just tested that it was zero, so we can just
715 * set it to 1.
717 dentry->d_lockref.count = 1;
718 return 0;
723 * This is dput
725 * This is complicated by the fact that we do not want to put
726 * dentries that are no longer on any hash chain on the unused
727 * list: we'd much rather just get rid of them immediately.
729 * However, that implies that we have to traverse the dentry
730 * tree upwards to the parents which might _also_ now be
731 * scheduled for deletion (it may have been only waiting for
732 * its last child to go away).
734 * This tail recursion is done by hand as we don't want to depend
735 * on the compiler to always get this right (gcc generally doesn't).
736 * Real recursion would eat up our stack space.
740 * dput - release a dentry
741 * @dentry: dentry to release
743 * Release a dentry. This will drop the usage count and if appropriate
744 * call the dentry unlink method as well as removing it from the queues and
745 * releasing its resources. If the parent dentries were scheduled for release
746 * they too may now get deleted.
748 void dput(struct dentry *dentry)
750 if (unlikely(!dentry))
751 return;
753 repeat:
754 might_sleep();
756 rcu_read_lock();
757 if (likely(fast_dput(dentry))) {
758 rcu_read_unlock();
759 return;
762 /* Slow case: now with the dentry lock held */
763 rcu_read_unlock();
765 /* Unreachable? Get rid of it */
766 if (unlikely(d_unhashed(dentry)))
767 goto kill_it;
769 if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
770 goto kill_it;
772 if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
773 if (dentry->d_op->d_delete(dentry))
774 goto kill_it;
777 if (!(dentry->d_flags & DCACHE_REFERENCED))
778 dentry->d_flags |= DCACHE_REFERENCED;
779 dentry_lru_add(dentry);
781 dentry->d_lockref.count--;
782 spin_unlock(&dentry->d_lock);
783 return;
785 kill_it:
786 dentry = dentry_kill(dentry);
787 if (dentry) {
788 cond_resched();
789 goto repeat;
792 EXPORT_SYMBOL(dput);
795 /* This must be called with d_lock held */
796 static inline void __dget_dlock(struct dentry *dentry)
798 dentry->d_lockref.count++;
801 static inline void __dget(struct dentry *dentry)
803 lockref_get(&dentry->d_lockref);
806 struct dentry *dget_parent(struct dentry *dentry)
808 int gotref;
809 struct dentry *ret;
812 * Do optimistic parent lookup without any
813 * locking.
815 rcu_read_lock();
816 ret = ACCESS_ONCE(dentry->d_parent);
817 gotref = lockref_get_not_zero(&ret->d_lockref);
818 rcu_read_unlock();
819 if (likely(gotref)) {
820 if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
821 return ret;
822 dput(ret);
825 repeat:
827 * Don't need rcu_dereference because we re-check it was correct under
828 * the lock.
830 rcu_read_lock();
831 ret = dentry->d_parent;
832 spin_lock(&ret->d_lock);
833 if (unlikely(ret != dentry->d_parent)) {
834 spin_unlock(&ret->d_lock);
835 rcu_read_unlock();
836 goto repeat;
838 rcu_read_unlock();
839 BUG_ON(!ret->d_lockref.count);
840 ret->d_lockref.count++;
841 spin_unlock(&ret->d_lock);
842 return ret;
844 EXPORT_SYMBOL(dget_parent);
847 * d_find_alias - grab a hashed alias of inode
848 * @inode: inode in question
850 * If inode has a hashed alias, or is a directory and has any alias,
851 * acquire the reference to alias and return it. Otherwise return NULL.
852 * Notice that if inode is a directory there can be only one alias and
853 * it can be unhashed only if it has no children, or if it is the root
854 * of a filesystem, or if the directory was renamed and d_revalidate
855 * was the first vfs operation to notice.
857 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
858 * any other hashed alias over that one.
860 static struct dentry *__d_find_alias(struct inode *inode)
862 struct dentry *alias, *discon_alias;
864 again:
865 discon_alias = NULL;
866 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
867 spin_lock(&alias->d_lock);
868 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
869 if (IS_ROOT(alias) &&
870 (alias->d_flags & DCACHE_DISCONNECTED)) {
871 discon_alias = alias;
872 } else {
873 __dget_dlock(alias);
874 spin_unlock(&alias->d_lock);
875 return alias;
878 spin_unlock(&alias->d_lock);
880 if (discon_alias) {
881 alias = discon_alias;
882 spin_lock(&alias->d_lock);
883 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
884 __dget_dlock(alias);
885 spin_unlock(&alias->d_lock);
886 return alias;
888 spin_unlock(&alias->d_lock);
889 goto again;
891 return NULL;
894 struct dentry *d_find_alias(struct inode *inode)
896 struct dentry *de = NULL;
898 if (!hlist_empty(&inode->i_dentry)) {
899 spin_lock(&inode->i_lock);
900 de = __d_find_alias(inode);
901 spin_unlock(&inode->i_lock);
903 return de;
905 EXPORT_SYMBOL(d_find_alias);
908 * Try to kill dentries associated with this inode.
909 * WARNING: you must own a reference to inode.
911 void d_prune_aliases(struct inode *inode)
913 struct dentry *dentry;
914 restart:
915 spin_lock(&inode->i_lock);
916 hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
917 spin_lock(&dentry->d_lock);
918 if (!dentry->d_lockref.count) {
919 struct dentry *parent = lock_parent(dentry);
920 if (likely(!dentry->d_lockref.count)) {
921 __dentry_kill(dentry);
922 dput(parent);
923 goto restart;
925 if (parent)
926 spin_unlock(&parent->d_lock);
928 spin_unlock(&dentry->d_lock);
930 spin_unlock(&inode->i_lock);
932 EXPORT_SYMBOL(d_prune_aliases);
934 static void shrink_dentry_list(struct list_head *list)
936 struct dentry *dentry, *parent;
938 while (!list_empty(list)) {
939 struct inode *inode;
940 dentry = list_entry(list->prev, struct dentry, d_lru);
941 spin_lock(&dentry->d_lock);
942 parent = lock_parent(dentry);
945 * The dispose list is isolated and dentries are not accounted
946 * to the LRU here, so we can simply remove it from the list
947 * here regardless of whether it is referenced or not.
949 d_shrink_del(dentry);
952 * We found an inuse dentry which was not removed from
953 * the LRU because of laziness during lookup. Do not free it.
955 if (dentry->d_lockref.count > 0) {
956 spin_unlock(&dentry->d_lock);
957 if (parent)
958 spin_unlock(&parent->d_lock);
959 continue;
963 if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
964 bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
965 spin_unlock(&dentry->d_lock);
966 if (parent)
967 spin_unlock(&parent->d_lock);
968 if (can_free)
969 dentry_free(dentry);
970 continue;
973 inode = dentry->d_inode;
974 if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
975 d_shrink_add(dentry, list);
976 spin_unlock(&dentry->d_lock);
977 if (parent)
978 spin_unlock(&parent->d_lock);
979 continue;
982 __dentry_kill(dentry);
985 * We need to prune ancestors too. This is necessary to prevent
986 * quadratic behavior of shrink_dcache_parent(), but is also
987 * expected to be beneficial in reducing dentry cache
988 * fragmentation.
990 dentry = parent;
991 while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
992 parent = lock_parent(dentry);
993 if (dentry->d_lockref.count != 1) {
994 dentry->d_lockref.count--;
995 spin_unlock(&dentry->d_lock);
996 if (parent)
997 spin_unlock(&parent->d_lock);
998 break;
1000 inode = dentry->d_inode; /* can't be NULL */
1001 if (unlikely(!spin_trylock(&inode->i_lock))) {
1002 spin_unlock(&dentry->d_lock);
1003 if (parent)
1004 spin_unlock(&parent->d_lock);
1005 cpu_relax();
1006 continue;
1008 __dentry_kill(dentry);
1009 dentry = parent;
1014 static enum lru_status dentry_lru_isolate(struct list_head *item,
1015 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1017 struct list_head *freeable = arg;
1018 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1022 * we are inverting the lru lock/dentry->d_lock here,
1023 * so use a trylock. If we fail to get the lock, just skip
1024 * it
1026 if (!spin_trylock(&dentry->d_lock))
1027 return LRU_SKIP;
1030 * Referenced dentries are still in use. If they have active
1031 * counts, just remove them from the LRU. Otherwise give them
1032 * another pass through the LRU.
1034 if (dentry->d_lockref.count) {
1035 d_lru_isolate(lru, dentry);
1036 spin_unlock(&dentry->d_lock);
1037 return LRU_REMOVED;
1040 if (dentry->d_flags & DCACHE_REFERENCED) {
1041 dentry->d_flags &= ~DCACHE_REFERENCED;
1042 spin_unlock(&dentry->d_lock);
1045 * The list move itself will be made by the common LRU code. At
1046 * this point, we've dropped the dentry->d_lock but keep the
1047 * lru lock. This is safe to do, since every list movement is
1048 * protected by the lru lock even if both locks are held.
1050 * This is guaranteed by the fact that all LRU management
1051 * functions are intermediated by the LRU API calls like
1052 * list_lru_add and list_lru_del. List movement in this file
1053 * only ever occur through this functions or through callbacks
1054 * like this one, that are called from the LRU API.
1056 * The only exceptions to this are functions like
1057 * shrink_dentry_list, and code that first checks for the
1058 * DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1059 * operating only with stack provided lists after they are
1060 * properly isolated from the main list. It is thus, always a
1061 * local access.
1063 return LRU_ROTATE;
1066 d_lru_shrink_move(lru, dentry, freeable);
1067 spin_unlock(&dentry->d_lock);
1069 return LRU_REMOVED;
1073 * prune_dcache_sb - shrink the dcache
1074 * @sb: superblock
1075 * @sc: shrink control, passed to list_lru_shrink_walk()
1077 * Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1078 * is done when we need more memory and called from the superblock shrinker
1079 * function.
1081 * This function may fail to free any resources if all the dentries are in
1082 * use.
1084 long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1086 LIST_HEAD(dispose);
1087 long freed;
1089 freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1090 dentry_lru_isolate, &dispose);
1091 shrink_dentry_list(&dispose);
1092 return freed;
1095 static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1096 struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1098 struct list_head *freeable = arg;
1099 struct dentry *dentry = container_of(item, struct dentry, d_lru);
1102 * we are inverting the lru lock/dentry->d_lock here,
1103 * so use a trylock. If we fail to get the lock, just skip
1104 * it
1106 if (!spin_trylock(&dentry->d_lock))
1107 return LRU_SKIP;
1109 d_lru_shrink_move(lru, dentry, freeable);
1110 spin_unlock(&dentry->d_lock);
1112 return LRU_REMOVED;
1117 * shrink_dcache_sb - shrink dcache for a superblock
1118 * @sb: superblock
1120 * Shrink the dcache for the specified super block. This is used to free
1121 * the dcache before unmounting a file system.
1123 void shrink_dcache_sb(struct super_block *sb)
1125 long freed;
1127 do {
1128 LIST_HEAD(dispose);
1130 freed = list_lru_walk(&sb->s_dentry_lru,
1131 dentry_lru_isolate_shrink, &dispose, UINT_MAX);
1133 this_cpu_sub(nr_dentry_unused, freed);
1134 shrink_dentry_list(&dispose);
1135 } while (freed > 0);
1137 EXPORT_SYMBOL(shrink_dcache_sb);
1140 * enum d_walk_ret - action to talke during tree walk
1141 * @D_WALK_CONTINUE: contrinue walk
1142 * @D_WALK_QUIT: quit walk
1143 * @D_WALK_NORETRY: quit when retry is needed
1144 * @D_WALK_SKIP: skip this dentry and its children
1146 enum d_walk_ret {
1147 D_WALK_CONTINUE,
1148 D_WALK_QUIT,
1149 D_WALK_NORETRY,
1150 D_WALK_SKIP,
1154 * d_walk - walk the dentry tree
1155 * @parent: start of walk
1156 * @data: data passed to @enter() and @finish()
1157 * @enter: callback when first entering the dentry
1158 * @finish: callback when successfully finished the walk
1160 * The @enter() and @finish() callbacks are called with d_lock held.
1162 static void d_walk(struct dentry *parent, void *data,
1163 enum d_walk_ret (*enter)(void *, struct dentry *),
1164 void (*finish)(void *))
1166 struct dentry *this_parent;
1167 struct list_head *next;
1168 unsigned seq = 0;
1169 enum d_walk_ret ret;
1170 bool retry = true;
1172 again:
1173 read_seqbegin_or_lock(&rename_lock, &seq);
1174 this_parent = parent;
1175 spin_lock(&this_parent->d_lock);
1177 ret = enter(data, this_parent);
1178 switch (ret) {
1179 case D_WALK_CONTINUE:
1180 break;
1181 case D_WALK_QUIT:
1182 case D_WALK_SKIP:
1183 goto out_unlock;
1184 case D_WALK_NORETRY:
1185 retry = false;
1186 break;
1188 repeat:
1189 next = this_parent->d_subdirs.next;
1190 resume:
1191 while (next != &this_parent->d_subdirs) {
1192 struct list_head *tmp = next;
1193 struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1194 next = tmp->next;
1196 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1198 ret = enter(data, dentry);
1199 switch (ret) {
1200 case D_WALK_CONTINUE:
1201 break;
1202 case D_WALK_QUIT:
1203 spin_unlock(&dentry->d_lock);
1204 goto out_unlock;
1205 case D_WALK_NORETRY:
1206 retry = false;
1207 break;
1208 case D_WALK_SKIP:
1209 spin_unlock(&dentry->d_lock);
1210 continue;
1213 if (!list_empty(&dentry->d_subdirs)) {
1214 spin_unlock(&this_parent->d_lock);
1215 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1216 this_parent = dentry;
1217 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1218 goto repeat;
1220 spin_unlock(&dentry->d_lock);
1223 * All done at this level ... ascend and resume the search.
1225 rcu_read_lock();
1226 ascend:
1227 if (this_parent != parent) {
1228 struct dentry *child = this_parent;
1229 this_parent = child->d_parent;
1231 spin_unlock(&child->d_lock);
1232 spin_lock(&this_parent->d_lock);
1234 /* might go back up the wrong parent if we have had a rename. */
1235 if (need_seqretry(&rename_lock, seq))
1236 goto rename_retry;
1237 /* go into the first sibling still alive */
1238 do {
1239 next = child->d_child.next;
1240 if (next == &this_parent->d_subdirs)
1241 goto ascend;
1242 child = list_entry(next, struct dentry, d_child);
1243 } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1244 rcu_read_unlock();
1245 goto resume;
1247 if (need_seqretry(&rename_lock, seq))
1248 goto rename_retry;
1249 rcu_read_unlock();
1250 if (finish)
1251 finish(data);
1253 out_unlock:
1254 spin_unlock(&this_parent->d_lock);
1255 done_seqretry(&rename_lock, seq);
1256 return;
1258 rename_retry:
1259 spin_unlock(&this_parent->d_lock);
1260 rcu_read_unlock();
1261 BUG_ON(seq & 1);
1262 if (!retry)
1263 return;
1264 seq = 1;
1265 goto again;
1269 * Search for at least 1 mount point in the dentry's subdirs.
1270 * We descend to the next level whenever the d_subdirs
1271 * list is non-empty and continue searching.
1274 static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1276 int *ret = data;
1277 if (d_mountpoint(dentry)) {
1278 *ret = 1;
1279 return D_WALK_QUIT;
1281 return D_WALK_CONTINUE;
1285 * have_submounts - check for mounts over a dentry
1286 * @parent: dentry to check.
1288 * Return true if the parent or its subdirectories contain
1289 * a mount point
1291 int have_submounts(struct dentry *parent)
1293 int ret = 0;
1295 d_walk(parent, &ret, check_mount, NULL);
1297 return ret;
1299 EXPORT_SYMBOL(have_submounts);
1302 * Called by mount code to set a mountpoint and check if the mountpoint is
1303 * reachable (e.g. NFS can unhash a directory dentry and then the complete
1304 * subtree can become unreachable).
1306 * Only one of d_invalidate() and d_set_mounted() must succeed. For
1307 * this reason take rename_lock and d_lock on dentry and ancestors.
1309 int d_set_mounted(struct dentry *dentry)
1311 struct dentry *p;
1312 int ret = -ENOENT;
1313 write_seqlock(&rename_lock);
1314 for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1315 /* Need exclusion wrt. d_invalidate() */
1316 spin_lock(&p->d_lock);
1317 if (unlikely(d_unhashed(p))) {
1318 spin_unlock(&p->d_lock);
1319 goto out;
1321 spin_unlock(&p->d_lock);
1323 spin_lock(&dentry->d_lock);
1324 if (!d_unlinked(dentry)) {
1325 ret = -EBUSY;
1326 if (!d_mountpoint(dentry)) {
1327 dentry->d_flags |= DCACHE_MOUNTED;
1328 ret = 0;
1331 spin_unlock(&dentry->d_lock);
1332 out:
1333 write_sequnlock(&rename_lock);
1334 return ret;
1338 * Search the dentry child list of the specified parent,
1339 * and move any unused dentries to the end of the unused
1340 * list for prune_dcache(). We descend to the next level
1341 * whenever the d_subdirs list is non-empty and continue
1342 * searching.
1344 * It returns zero iff there are no unused children,
1345 * otherwise it returns the number of children moved to
1346 * the end of the unused list. This may not be the total
1347 * number of unused children, because select_parent can
1348 * drop the lock and return early due to latency
1349 * constraints.
1352 struct select_data {
1353 struct dentry *start;
1354 struct list_head dispose;
1355 int found;
1358 static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1360 struct select_data *data = _data;
1361 enum d_walk_ret ret = D_WALK_CONTINUE;
1363 if (data->start == dentry)
1364 goto out;
1366 if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1367 data->found++;
1368 } else {
1369 if (dentry->d_flags & DCACHE_LRU_LIST)
1370 d_lru_del(dentry);
1371 if (!dentry->d_lockref.count) {
1372 d_shrink_add(dentry, &data->dispose);
1373 data->found++;
1377 * We can return to the caller if we have found some (this
1378 * ensures forward progress). We'll be coming back to find
1379 * the rest.
1381 if (!list_empty(&data->dispose))
1382 ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1383 out:
1384 return ret;
1388 * shrink_dcache_parent - prune dcache
1389 * @parent: parent of entries to prune
1391 * Prune the dcache to remove unused children of the parent dentry.
1393 void shrink_dcache_parent(struct dentry *parent)
1395 for (;;) {
1396 struct select_data data;
1398 INIT_LIST_HEAD(&data.dispose);
1399 data.start = parent;
1400 data.found = 0;
1402 d_walk(parent, &data, select_collect, NULL);
1403 if (!data.found)
1404 break;
1406 shrink_dentry_list(&data.dispose);
1407 cond_resched();
1410 EXPORT_SYMBOL(shrink_dcache_parent);
1412 static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1414 /* it has busy descendents; complain about those instead */
1415 if (!list_empty(&dentry->d_subdirs))
1416 return D_WALK_CONTINUE;
1418 /* root with refcount 1 is fine */
1419 if (dentry == _data && dentry->d_lockref.count == 1)
1420 return D_WALK_CONTINUE;
1422 printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1423 " still in use (%d) [unmount of %s %s]\n",
1424 dentry,
1425 dentry->d_inode ?
1426 dentry->d_inode->i_ino : 0UL,
1427 dentry,
1428 dentry->d_lockref.count,
1429 dentry->d_sb->s_type->name,
1430 dentry->d_sb->s_id);
1431 WARN_ON(1);
1432 return D_WALK_CONTINUE;
1435 static void do_one_tree(struct dentry *dentry)
1437 shrink_dcache_parent(dentry);
1438 d_walk(dentry, dentry, umount_check, NULL);
1439 d_drop(dentry);
1440 dput(dentry);
1444 * destroy the dentries attached to a superblock on unmounting
1446 void shrink_dcache_for_umount(struct super_block *sb)
1448 struct dentry *dentry;
1450 WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1452 dentry = sb->s_root;
1453 sb->s_root = NULL;
1454 do_one_tree(dentry);
1456 while (!hlist_bl_empty(&sb->s_anon)) {
1457 dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1458 do_one_tree(dentry);
1462 struct detach_data {
1463 struct select_data select;
1464 struct dentry *mountpoint;
1466 static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1468 struct detach_data *data = _data;
1470 if (d_mountpoint(dentry)) {
1471 __dget_dlock(dentry);
1472 data->mountpoint = dentry;
1473 return D_WALK_QUIT;
1476 return select_collect(&data->select, dentry);
1479 static void check_and_drop(void *_data)
1481 struct detach_data *data = _data;
1483 if (!data->mountpoint && !data->select.found)
1484 __d_drop(data->select.start);
1488 * d_invalidate - detach submounts, prune dcache, and drop
1489 * @dentry: dentry to invalidate (aka detach, prune and drop)
1491 * no dcache lock.
1493 * The final d_drop is done as an atomic operation relative to
1494 * rename_lock ensuring there are no races with d_set_mounted. This
1495 * ensures there are no unhashed dentries on the path to a mountpoint.
1497 void d_invalidate(struct dentry *dentry)
1500 * If it's already been dropped, return OK.
1502 spin_lock(&dentry->d_lock);
1503 if (d_unhashed(dentry)) {
1504 spin_unlock(&dentry->d_lock);
1505 return;
1507 spin_unlock(&dentry->d_lock);
1509 /* Negative dentries can be dropped without further checks */
1510 if (!dentry->d_inode) {
1511 d_drop(dentry);
1512 return;
1515 for (;;) {
1516 struct detach_data data;
1518 data.mountpoint = NULL;
1519 INIT_LIST_HEAD(&data.select.dispose);
1520 data.select.start = dentry;
1521 data.select.found = 0;
1523 d_walk(dentry, &data, detach_and_collect, check_and_drop);
1525 if (data.select.found)
1526 shrink_dentry_list(&data.select.dispose);
1528 if (data.mountpoint) {
1529 detach_mounts(data.mountpoint);
1530 dput(data.mountpoint);
1533 if (!data.mountpoint && !data.select.found)
1534 break;
1536 cond_resched();
1539 EXPORT_SYMBOL(d_invalidate);
1542 * __d_alloc - allocate a dcache entry
1543 * @sb: filesystem it will belong to
1544 * @name: qstr of the name
1546 * Allocates a dentry. It returns %NULL if there is insufficient memory
1547 * available. On a success the dentry is returned. The name passed in is
1548 * copied and the copy passed in may be reused after this call.
1551 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1553 struct dentry *dentry;
1554 char *dname;
1556 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1557 if (!dentry)
1558 return NULL;
1561 * We guarantee that the inline name is always NUL-terminated.
1562 * This way the memcpy() done by the name switching in rename
1563 * will still always have a NUL at the end, even if we might
1564 * be overwriting an internal NUL character
1566 dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1567 if (name->len > DNAME_INLINE_LEN-1) {
1568 size_t size = offsetof(struct external_name, name[1]);
1569 struct external_name *p = kmalloc(size + name->len, GFP_KERNEL);
1570 if (!p) {
1571 kmem_cache_free(dentry_cache, dentry);
1572 return NULL;
1574 atomic_set(&p->u.count, 1);
1575 dname = p->name;
1576 if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1577 kasan_unpoison_shadow(dname,
1578 round_up(name->len + 1, sizeof(unsigned long)));
1579 } else {
1580 dname = dentry->d_iname;
1583 dentry->d_name.len = name->len;
1584 dentry->d_name.hash = name->hash;
1585 memcpy(dname, name->name, name->len);
1586 dname[name->len] = 0;
1588 /* Make sure we always see the terminating NUL character */
1589 smp_wmb();
1590 dentry->d_name.name = dname;
1592 dentry->d_lockref.count = 1;
1593 dentry->d_flags = 0;
1594 spin_lock_init(&dentry->d_lock);
1595 seqcount_init(&dentry->d_seq);
1596 dentry->d_inode = NULL;
1597 dentry->d_parent = dentry;
1598 dentry->d_sb = sb;
1599 dentry->d_op = NULL;
1600 dentry->d_fsdata = NULL;
1601 INIT_HLIST_BL_NODE(&dentry->d_hash);
1602 INIT_LIST_HEAD(&dentry->d_lru);
1603 INIT_LIST_HEAD(&dentry->d_subdirs);
1604 INIT_HLIST_NODE(&dentry->d_u.d_alias);
1605 INIT_LIST_HEAD(&dentry->d_child);
1606 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1608 this_cpu_inc(nr_dentry);
1610 return dentry;
1614 * d_alloc - allocate a dcache entry
1615 * @parent: parent of entry to allocate
1616 * @name: qstr of the name
1618 * Allocates a dentry. It returns %NULL if there is insufficient memory
1619 * available. On a success the dentry is returned. The name passed in is
1620 * copied and the copy passed in may be reused after this call.
1622 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1624 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1625 if (!dentry)
1626 return NULL;
1627 dentry->d_flags |= DCACHE_RCUACCESS;
1628 spin_lock(&parent->d_lock);
1630 * don't need child lock because it is not subject
1631 * to concurrency here
1633 __dget_dlock(parent);
1634 dentry->d_parent = parent;
1635 list_add(&dentry->d_child, &parent->d_subdirs);
1636 spin_unlock(&parent->d_lock);
1638 return dentry;
1640 EXPORT_SYMBOL(d_alloc);
1643 * d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1644 * @sb: the superblock
1645 * @name: qstr of the name
1647 * For a filesystem that just pins its dentries in memory and never
1648 * performs lookups at all, return an unhashed IS_ROOT dentry.
1650 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1652 return __d_alloc(sb, name);
1654 EXPORT_SYMBOL(d_alloc_pseudo);
1656 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1658 struct qstr q;
1660 q.name = name;
1661 q.len = strlen(name);
1662 q.hash = full_name_hash(q.name, q.len);
1663 return d_alloc(parent, &q);
1665 EXPORT_SYMBOL(d_alloc_name);
1667 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1669 WARN_ON_ONCE(dentry->d_op);
1670 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1671 DCACHE_OP_COMPARE |
1672 DCACHE_OP_REVALIDATE |
1673 DCACHE_OP_WEAK_REVALIDATE |
1674 DCACHE_OP_DELETE |
1675 DCACHE_OP_SELECT_INODE));
1676 dentry->d_op = op;
1677 if (!op)
1678 return;
1679 if (op->d_hash)
1680 dentry->d_flags |= DCACHE_OP_HASH;
1681 if (op->d_compare)
1682 dentry->d_flags |= DCACHE_OP_COMPARE;
1683 if (op->d_revalidate)
1684 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1685 if (op->d_weak_revalidate)
1686 dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1687 if (op->d_delete)
1688 dentry->d_flags |= DCACHE_OP_DELETE;
1689 if (op->d_prune)
1690 dentry->d_flags |= DCACHE_OP_PRUNE;
1691 if (op->d_select_inode)
1692 dentry->d_flags |= DCACHE_OP_SELECT_INODE;
1695 EXPORT_SYMBOL(d_set_d_op);
1699 * d_set_fallthru - Mark a dentry as falling through to a lower layer
1700 * @dentry - The dentry to mark
1702 * Mark a dentry as falling through to the lower layer (as set with
1703 * d_pin_lower()). This flag may be recorded on the medium.
1705 void d_set_fallthru(struct dentry *dentry)
1707 spin_lock(&dentry->d_lock);
1708 dentry->d_flags |= DCACHE_FALLTHRU;
1709 spin_unlock(&dentry->d_lock);
1711 EXPORT_SYMBOL(d_set_fallthru);
1713 static unsigned d_flags_for_inode(struct inode *inode)
1715 unsigned add_flags = DCACHE_REGULAR_TYPE;
1717 if (!inode)
1718 return DCACHE_MISS_TYPE;
1720 if (S_ISDIR(inode->i_mode)) {
1721 add_flags = DCACHE_DIRECTORY_TYPE;
1722 if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1723 if (unlikely(!inode->i_op->lookup))
1724 add_flags = DCACHE_AUTODIR_TYPE;
1725 else
1726 inode->i_opflags |= IOP_LOOKUP;
1728 goto type_determined;
1731 if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1732 if (unlikely(inode->i_op->follow_link)) {
1733 add_flags = DCACHE_SYMLINK_TYPE;
1734 goto type_determined;
1736 inode->i_opflags |= IOP_NOFOLLOW;
1739 if (unlikely(!S_ISREG(inode->i_mode)))
1740 add_flags = DCACHE_SPECIAL_TYPE;
1742 type_determined:
1743 if (unlikely(IS_AUTOMOUNT(inode)))
1744 add_flags |= DCACHE_NEED_AUTOMOUNT;
1745 return add_flags;
1748 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1750 unsigned add_flags = d_flags_for_inode(inode);
1752 spin_lock(&dentry->d_lock);
1753 if (inode)
1754 hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1755 raw_write_seqcount_begin(&dentry->d_seq);
1756 __d_set_inode_and_type(dentry, inode, add_flags);
1757 raw_write_seqcount_end(&dentry->d_seq);
1758 spin_unlock(&dentry->d_lock);
1759 fsnotify_d_instantiate(dentry, inode);
1763 * d_instantiate - fill in inode information for a dentry
1764 * @entry: dentry to complete
1765 * @inode: inode to attach to this dentry
1767 * Fill in inode information in the entry.
1769 * This turns negative dentries into productive full members
1770 * of society.
1772 * NOTE! This assumes that the inode count has been incremented
1773 * (or otherwise set) by the caller to indicate that it is now
1774 * in use by the dcache.
1777 void d_instantiate(struct dentry *entry, struct inode * inode)
1779 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1780 if (inode)
1781 spin_lock(&inode->i_lock);
1782 __d_instantiate(entry, inode);
1783 if (inode)
1784 spin_unlock(&inode->i_lock);
1785 security_d_instantiate(entry, inode);
1787 EXPORT_SYMBOL(d_instantiate);
1790 * d_instantiate_unique - instantiate a non-aliased dentry
1791 * @entry: dentry to instantiate
1792 * @inode: inode to attach to this dentry
1794 * Fill in inode information in the entry. On success, it returns NULL.
1795 * If an unhashed alias of "entry" already exists, then we return the
1796 * aliased dentry instead and drop one reference to inode.
1798 * Note that in order to avoid conflicts with rename() etc, the caller
1799 * had better be holding the parent directory semaphore.
1801 * This also assumes that the inode count has been incremented
1802 * (or otherwise set) by the caller to indicate that it is now
1803 * in use by the dcache.
1805 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1806 struct inode *inode)
1808 struct dentry *alias;
1809 int len = entry->d_name.len;
1810 const char *name = entry->d_name.name;
1811 unsigned int hash = entry->d_name.hash;
1813 if (!inode) {
1814 __d_instantiate(entry, NULL);
1815 return NULL;
1818 hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
1820 * Don't need alias->d_lock here, because aliases with
1821 * d_parent == entry->d_parent are not subject to name or
1822 * parent changes, because the parent inode i_mutex is held.
1824 if (alias->d_name.hash != hash)
1825 continue;
1826 if (alias->d_parent != entry->d_parent)
1827 continue;
1828 if (alias->d_name.len != len)
1829 continue;
1830 if (dentry_cmp(alias, name, len))
1831 continue;
1832 __dget(alias);
1833 return alias;
1836 __d_instantiate(entry, inode);
1837 return NULL;
1840 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1842 struct dentry *result;
1844 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1846 if (inode)
1847 spin_lock(&inode->i_lock);
1848 result = __d_instantiate_unique(entry, inode);
1849 if (inode)
1850 spin_unlock(&inode->i_lock);
1852 if (!result) {
1853 security_d_instantiate(entry, inode);
1854 return NULL;
1857 BUG_ON(!d_unhashed(result));
1858 iput(inode);
1859 return result;
1862 EXPORT_SYMBOL(d_instantiate_unique);
1865 * d_instantiate_no_diralias - instantiate a non-aliased dentry
1866 * @entry: dentry to complete
1867 * @inode: inode to attach to this dentry
1869 * Fill in inode information in the entry. If a directory alias is found, then
1870 * return an error (and drop inode). Together with d_materialise_unique() this
1871 * guarantees that a directory inode may never have more than one alias.
1873 int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1875 BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1877 spin_lock(&inode->i_lock);
1878 if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1879 spin_unlock(&inode->i_lock);
1880 iput(inode);
1881 return -EBUSY;
1883 __d_instantiate(entry, inode);
1884 spin_unlock(&inode->i_lock);
1885 security_d_instantiate(entry, inode);
1887 return 0;
1889 EXPORT_SYMBOL(d_instantiate_no_diralias);
1891 struct dentry *d_make_root(struct inode *root_inode)
1893 struct dentry *res = NULL;
1895 if (root_inode) {
1896 static const struct qstr name = QSTR_INIT("/", 1);
1898 res = __d_alloc(root_inode->i_sb, &name);
1899 if (res)
1900 d_instantiate(res, root_inode);
1901 else
1902 iput(root_inode);
1904 return res;
1906 EXPORT_SYMBOL(d_make_root);
1908 static struct dentry * __d_find_any_alias(struct inode *inode)
1910 struct dentry *alias;
1912 if (hlist_empty(&inode->i_dentry))
1913 return NULL;
1914 alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1915 __dget(alias);
1916 return alias;
1920 * d_find_any_alias - find any alias for a given inode
1921 * @inode: inode to find an alias for
1923 * If any aliases exist for the given inode, take and return a
1924 * reference for one of them. If no aliases exist, return %NULL.
1926 struct dentry *d_find_any_alias(struct inode *inode)
1928 struct dentry *de;
1930 spin_lock(&inode->i_lock);
1931 de = __d_find_any_alias(inode);
1932 spin_unlock(&inode->i_lock);
1933 return de;
1935 EXPORT_SYMBOL(d_find_any_alias);
1937 static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1939 static const struct qstr anonstring = QSTR_INIT("/", 1);
1940 struct dentry *tmp;
1941 struct dentry *res;
1942 unsigned add_flags;
1944 if (!inode)
1945 return ERR_PTR(-ESTALE);
1946 if (IS_ERR(inode))
1947 return ERR_CAST(inode);
1949 res = d_find_any_alias(inode);
1950 if (res)
1951 goto out_iput;
1953 tmp = __d_alloc(inode->i_sb, &anonstring);
1954 if (!tmp) {
1955 res = ERR_PTR(-ENOMEM);
1956 goto out_iput;
1959 spin_lock(&inode->i_lock);
1960 res = __d_find_any_alias(inode);
1961 if (res) {
1962 spin_unlock(&inode->i_lock);
1963 dput(tmp);
1964 goto out_iput;
1967 /* attach a disconnected dentry */
1968 add_flags = d_flags_for_inode(inode);
1970 if (disconnected)
1971 add_flags |= DCACHE_DISCONNECTED;
1973 spin_lock(&tmp->d_lock);
1974 __d_set_inode_and_type(tmp, inode, add_flags);
1975 hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1976 hlist_bl_lock(&tmp->d_sb->s_anon);
1977 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1978 hlist_bl_unlock(&tmp->d_sb->s_anon);
1979 spin_unlock(&tmp->d_lock);
1980 spin_unlock(&inode->i_lock);
1981 security_d_instantiate(tmp, inode);
1983 return tmp;
1985 out_iput:
1986 if (res && !IS_ERR(res))
1987 security_d_instantiate(res, inode);
1988 iput(inode);
1989 return res;
1993 * d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
1994 * @inode: inode to allocate the dentry for
1996 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1997 * similar open by handle operations. The returned dentry may be anonymous,
1998 * or may have a full name (if the inode was already in the cache).
2000 * When called on a directory inode, we must ensure that the inode only ever
2001 * has one dentry. If a dentry is found, that is returned instead of
2002 * allocating a new one.
2004 * On successful return, the reference to the inode has been transferred
2005 * to the dentry. In case of an error the reference on the inode is released.
2006 * To make it easier to use in export operations a %NULL or IS_ERR inode may
2007 * be passed in and the error will be propagated to the return value,
2008 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2010 struct dentry *d_obtain_alias(struct inode *inode)
2012 return __d_obtain_alias(inode, 1);
2014 EXPORT_SYMBOL(d_obtain_alias);
2017 * d_obtain_root - find or allocate a dentry for a given inode
2018 * @inode: inode to allocate the dentry for
2020 * Obtain an IS_ROOT dentry for the root of a filesystem.
2022 * We must ensure that directory inodes only ever have one dentry. If a
2023 * dentry is found, that is returned instead of allocating a new one.
2025 * On successful return, the reference to the inode has been transferred
2026 * to the dentry. In case of an error the reference on the inode is
2027 * released. A %NULL or IS_ERR inode may be passed in and will be the
2028 * error will be propagate to the return value, with a %NULL @inode
2029 * replaced by ERR_PTR(-ESTALE).
2031 struct dentry *d_obtain_root(struct inode *inode)
2033 return __d_obtain_alias(inode, 0);
2035 EXPORT_SYMBOL(d_obtain_root);
2038 * d_add_ci - lookup or allocate new dentry with case-exact name
2039 * @inode: the inode case-insensitive lookup has found
2040 * @dentry: the negative dentry that was passed to the parent's lookup func
2041 * @name: the case-exact name to be associated with the returned dentry
2043 * This is to avoid filling the dcache with case-insensitive names to the
2044 * same inode, only the actual correct case is stored in the dcache for
2045 * case-insensitive filesystems.
2047 * For a case-insensitive lookup match and if the the case-exact dentry
2048 * already exists in in the dcache, use it and return it.
2050 * If no entry exists with the exact case name, allocate new dentry with
2051 * the exact case, and return the spliced entry.
2053 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2054 struct qstr *name)
2056 struct dentry *found;
2057 struct dentry *new;
2060 * First check if a dentry matching the name already exists,
2061 * if not go ahead and create it now.
2063 found = d_hash_and_lookup(dentry->d_parent, name);
2064 if (!found) {
2065 new = d_alloc(dentry->d_parent, name);
2066 if (!new) {
2067 found = ERR_PTR(-ENOMEM);
2068 } else {
2069 found = d_splice_alias(inode, new);
2070 if (found) {
2071 dput(new);
2072 return found;
2074 return new;
2077 iput(inode);
2078 return found;
2080 EXPORT_SYMBOL(d_add_ci);
2083 * Do the slow-case of the dentry name compare.
2085 * Unlike the dentry_cmp() function, we need to atomically
2086 * load the name and length information, so that the
2087 * filesystem can rely on them, and can use the 'name' and
2088 * 'len' information without worrying about walking off the
2089 * end of memory etc.
2091 * Thus the read_seqcount_retry() and the "duplicate" info
2092 * in arguments (the low-level filesystem should not look
2093 * at the dentry inode or name contents directly, since
2094 * rename can change them while we're in RCU mode).
2096 enum slow_d_compare {
2097 D_COMP_OK,
2098 D_COMP_NOMATCH,
2099 D_COMP_SEQRETRY,
2102 static noinline enum slow_d_compare slow_dentry_cmp(
2103 const struct dentry *parent,
2104 struct dentry *dentry,
2105 unsigned int seq,
2106 const struct qstr *name)
2108 int tlen = dentry->d_name.len;
2109 const char *tname = dentry->d_name.name;
2111 if (read_seqcount_retry(&dentry->d_seq, seq)) {
2112 cpu_relax();
2113 return D_COMP_SEQRETRY;
2115 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2116 return D_COMP_NOMATCH;
2117 return D_COMP_OK;
2121 * __d_lookup_rcu - search for a dentry (racy, store-free)
2122 * @parent: parent dentry
2123 * @name: qstr of name we wish to find
2124 * @seqp: returns d_seq value at the point where the dentry was found
2125 * Returns: dentry, or NULL
2127 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
2128 * resolution (store-free path walking) design described in
2129 * Documentation/filesystems/path-lookup.txt.
2131 * This is not to be used outside core vfs.
2133 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2134 * held, and rcu_read_lock held. The returned dentry must not be stored into
2135 * without taking d_lock and checking d_seq sequence count against @seq
2136 * returned here.
2138 * A refcount may be taken on the found dentry with the d_rcu_to_refcount
2139 * function.
2141 * Alternatively, __d_lookup_rcu may be called again to look up the child of
2142 * the returned dentry, so long as its parent's seqlock is checked after the
2143 * child is looked up. Thus, an interlocking stepping of sequence lock checks
2144 * is formed, giving integrity down the path walk.
2146 * NOTE! The caller *has* to check the resulting dentry against the sequence
2147 * number we've returned before using any of the resulting dentry state!
2149 struct dentry *__d_lookup_rcu(const struct dentry *parent,
2150 const struct qstr *name,
2151 unsigned *seqp)
2153 u64 hashlen = name->hash_len;
2154 const unsigned char *str = name->name;
2155 struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen));
2156 struct hlist_bl_node *node;
2157 struct dentry *dentry;
2160 * Note: There is significant duplication with __d_lookup_rcu which is
2161 * required to prevent single threaded performance regressions
2162 * especially on architectures where smp_rmb (in seqcounts) are costly.
2163 * Keep the two functions in sync.
2167 * The hash list is protected using RCU.
2169 * Carefully use d_seq when comparing a candidate dentry, to avoid
2170 * races with d_move().
2172 * It is possible that concurrent renames can mess up our list
2173 * walk here and result in missing our dentry, resulting in the
2174 * false-negative result. d_lookup() protects against concurrent
2175 * renames using rename_lock seqlock.
2177 * See Documentation/filesystems/path-lookup.txt for more details.
2179 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2180 unsigned seq;
2182 seqretry:
2184 * The dentry sequence count protects us from concurrent
2185 * renames, and thus protects parent and name fields.
2187 * The caller must perform a seqcount check in order
2188 * to do anything useful with the returned dentry.
2190 * NOTE! We do a "raw" seqcount_begin here. That means that
2191 * we don't wait for the sequence count to stabilize if it
2192 * is in the middle of a sequence change. If we do the slow
2193 * dentry compare, we will do seqretries until it is stable,
2194 * and if we end up with a successful lookup, we actually
2195 * want to exit RCU lookup anyway.
2197 seq = raw_seqcount_begin(&dentry->d_seq);
2198 if (dentry->d_parent != parent)
2199 continue;
2200 if (d_unhashed(dentry))
2201 continue;
2203 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2204 if (dentry->d_name.hash != hashlen_hash(hashlen))
2205 continue;
2206 *seqp = seq;
2207 switch (slow_dentry_cmp(parent, dentry, seq, name)) {
2208 case D_COMP_OK:
2209 return dentry;
2210 case D_COMP_NOMATCH:
2211 continue;
2212 default:
2213 goto seqretry;
2217 if (dentry->d_name.hash_len != hashlen)
2218 continue;
2219 *seqp = seq;
2220 if (!dentry_cmp(dentry, str, hashlen_len(hashlen)))
2221 return dentry;
2223 return NULL;
2227 * d_lookup - search for a dentry
2228 * @parent: parent dentry
2229 * @name: qstr of name we wish to find
2230 * Returns: dentry, or NULL
2232 * d_lookup searches the children of the parent dentry for the name in
2233 * question. If the dentry is found its reference count is incremented and the
2234 * dentry is returned. The caller must use dput to free the entry when it has
2235 * finished using it. %NULL is returned if the dentry does not exist.
2237 struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2239 struct dentry *dentry;
2240 unsigned seq;
2242 do {
2243 seq = read_seqbegin(&rename_lock);
2244 dentry = __d_lookup(parent, name);
2245 if (dentry)
2246 break;
2247 } while (read_seqretry(&rename_lock, seq));
2248 return dentry;
2250 EXPORT_SYMBOL(d_lookup);
2253 * __d_lookup - search for a dentry (racy)
2254 * @parent: parent dentry
2255 * @name: qstr of name we wish to find
2256 * Returns: dentry, or NULL
2258 * __d_lookup is like d_lookup, however it may (rarely) return a
2259 * false-negative result due to unrelated rename activity.
2261 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2262 * however it must be used carefully, eg. with a following d_lookup in
2263 * the case of failure.
2265 * __d_lookup callers must be commented.
2267 struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2269 unsigned int len = name->len;
2270 unsigned int hash = name->hash;
2271 const unsigned char *str = name->name;
2272 struct hlist_bl_head *b = d_hash(parent, hash);
2273 struct hlist_bl_node *node;
2274 struct dentry *found = NULL;
2275 struct dentry *dentry;
2278 * Note: There is significant duplication with __d_lookup_rcu which is
2279 * required to prevent single threaded performance regressions
2280 * especially on architectures where smp_rmb (in seqcounts) are costly.
2281 * Keep the two functions in sync.
2285 * The hash list is protected using RCU.
2287 * Take d_lock when comparing a candidate dentry, to avoid races
2288 * with d_move().
2290 * It is possible that concurrent renames can mess up our list
2291 * walk here and result in missing our dentry, resulting in the
2292 * false-negative result. d_lookup() protects against concurrent
2293 * renames using rename_lock seqlock.
2295 * See Documentation/filesystems/path-lookup.txt for more details.
2297 rcu_read_lock();
2299 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2301 if (dentry->d_name.hash != hash)
2302 continue;
2304 spin_lock(&dentry->d_lock);
2305 if (dentry->d_parent != parent)
2306 goto next;
2307 if (d_unhashed(dentry))
2308 goto next;
2311 * It is safe to compare names since d_move() cannot
2312 * change the qstr (protected by d_lock).
2314 if (parent->d_flags & DCACHE_OP_COMPARE) {
2315 int tlen = dentry->d_name.len;
2316 const char *tname = dentry->d_name.name;
2317 if (parent->d_op->d_compare(parent, dentry, tlen, tname, name))
2318 goto next;
2319 } else {
2320 if (dentry->d_name.len != len)
2321 goto next;
2322 if (dentry_cmp(dentry, str, len))
2323 goto next;
2326 dentry->d_lockref.count++;
2327 found = dentry;
2328 spin_unlock(&dentry->d_lock);
2329 break;
2330 next:
2331 spin_unlock(&dentry->d_lock);
2333 rcu_read_unlock();
2335 return found;
2339 * d_hash_and_lookup - hash the qstr then search for a dentry
2340 * @dir: Directory to search in
2341 * @name: qstr of name we wish to find
2343 * On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2345 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2348 * Check for a fs-specific hash function. Note that we must
2349 * calculate the standard hash first, as the d_op->d_hash()
2350 * routine may choose to leave the hash value unchanged.
2352 name->hash = full_name_hash(name->name, name->len);
2353 if (dir->d_flags & DCACHE_OP_HASH) {
2354 int err = dir->d_op->d_hash(dir, name);
2355 if (unlikely(err < 0))
2356 return ERR_PTR(err);
2358 return d_lookup(dir, name);
2360 EXPORT_SYMBOL(d_hash_and_lookup);
2363 * When a file is deleted, we have two options:
2364 * - turn this dentry into a negative dentry
2365 * - unhash this dentry and free it.
2367 * Usually, we want to just turn this into
2368 * a negative dentry, but if anybody else is
2369 * currently using the dentry or the inode
2370 * we can't do that and we fall back on removing
2371 * it from the hash queues and waiting for
2372 * it to be deleted later when it has no users
2376 * d_delete - delete a dentry
2377 * @dentry: The dentry to delete
2379 * Turn the dentry into a negative dentry if possible, otherwise
2380 * remove it from the hash queues so it can be deleted later
2383 void d_delete(struct dentry * dentry)
2385 struct inode *inode;
2386 int isdir = 0;
2388 * Are we the only user?
2390 again:
2391 spin_lock(&dentry->d_lock);
2392 inode = dentry->d_inode;
2393 isdir = S_ISDIR(inode->i_mode);
2394 if (dentry->d_lockref.count == 1) {
2395 if (!spin_trylock(&inode->i_lock)) {
2396 spin_unlock(&dentry->d_lock);
2397 cpu_relax();
2398 goto again;
2400 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2401 dentry_unlink_inode(dentry);
2402 fsnotify_nameremove(dentry, isdir);
2403 return;
2406 if (!d_unhashed(dentry))
2407 __d_drop(dentry);
2409 spin_unlock(&dentry->d_lock);
2411 fsnotify_nameremove(dentry, isdir);
2413 EXPORT_SYMBOL(d_delete);
2415 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2417 BUG_ON(!d_unhashed(entry));
2418 hlist_bl_lock(b);
2419 hlist_bl_add_head_rcu(&entry->d_hash, b);
2420 hlist_bl_unlock(b);
2423 static void _d_rehash(struct dentry * entry)
2425 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2429 * d_rehash - add an entry back to the hash
2430 * @entry: dentry to add to the hash
2432 * Adds a dentry to the hash according to its name.
2435 void d_rehash(struct dentry * entry)
2437 spin_lock(&entry->d_lock);
2438 _d_rehash(entry);
2439 spin_unlock(&entry->d_lock);
2441 EXPORT_SYMBOL(d_rehash);
2444 * dentry_update_name_case - update case insensitive dentry with a new name
2445 * @dentry: dentry to be updated
2446 * @name: new name
2448 * Update a case insensitive dentry with new case of name.
2450 * dentry must have been returned by d_lookup with name @name. Old and new
2451 * name lengths must match (ie. no d_compare which allows mismatched name
2452 * lengths).
2454 * Parent inode i_mutex must be held over d_lookup and into this call (to
2455 * keep renames and concurrent inserts, and readdir(2) away).
2457 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2459 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2460 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2462 spin_lock(&dentry->d_lock);
2463 write_seqcount_begin(&dentry->d_seq);
2464 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2465 write_seqcount_end(&dentry->d_seq);
2466 spin_unlock(&dentry->d_lock);
2468 EXPORT_SYMBOL(dentry_update_name_case);
2470 static void swap_names(struct dentry *dentry, struct dentry *target)
2472 if (unlikely(dname_external(target))) {
2473 if (unlikely(dname_external(dentry))) {
2475 * Both external: swap the pointers
2477 swap(target->d_name.name, dentry->d_name.name);
2478 } else {
2480 * dentry:internal, target:external. Steal target's
2481 * storage and make target internal.
2483 memcpy(target->d_iname, dentry->d_name.name,
2484 dentry->d_name.len + 1);
2485 dentry->d_name.name = target->d_name.name;
2486 target->d_name.name = target->d_iname;
2488 } else {
2489 if (unlikely(dname_external(dentry))) {
2491 * dentry:external, target:internal. Give dentry's
2492 * storage to target and make dentry internal
2494 memcpy(dentry->d_iname, target->d_name.name,
2495 target->d_name.len + 1);
2496 target->d_name.name = dentry->d_name.name;
2497 dentry->d_name.name = dentry->d_iname;
2498 } else {
2500 * Both are internal.
2502 unsigned int i;
2503 BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2504 kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2505 kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2506 for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2507 swap(((long *) &dentry->d_iname)[i],
2508 ((long *) &target->d_iname)[i]);
2512 swap(dentry->d_name.hash_len, target->d_name.hash_len);
2515 static void copy_name(struct dentry *dentry, struct dentry *target)
2517 struct external_name *old_name = NULL;
2518 if (unlikely(dname_external(dentry)))
2519 old_name = external_name(dentry);
2520 if (unlikely(dname_external(target))) {
2521 atomic_inc(&external_name(target)->u.count);
2522 dentry->d_name = target->d_name;
2523 } else {
2524 memcpy(dentry->d_iname, target->d_name.name,
2525 target->d_name.len + 1);
2526 dentry->d_name.name = dentry->d_iname;
2527 dentry->d_name.hash_len = target->d_name.hash_len;
2529 if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2530 kfree_rcu(old_name, u.head);
2533 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2536 * XXXX: do we really need to take target->d_lock?
2538 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2539 spin_lock(&target->d_parent->d_lock);
2540 else {
2541 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2542 spin_lock(&dentry->d_parent->d_lock);
2543 spin_lock_nested(&target->d_parent->d_lock,
2544 DENTRY_D_LOCK_NESTED);
2545 } else {
2546 spin_lock(&target->d_parent->d_lock);
2547 spin_lock_nested(&dentry->d_parent->d_lock,
2548 DENTRY_D_LOCK_NESTED);
2551 if (target < dentry) {
2552 spin_lock_nested(&target->d_lock, 2);
2553 spin_lock_nested(&dentry->d_lock, 3);
2554 } else {
2555 spin_lock_nested(&dentry->d_lock, 2);
2556 spin_lock_nested(&target->d_lock, 3);
2560 static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2562 if (target->d_parent != dentry->d_parent)
2563 spin_unlock(&dentry->d_parent->d_lock);
2564 if (target->d_parent != target)
2565 spin_unlock(&target->d_parent->d_lock);
2566 spin_unlock(&target->d_lock);
2567 spin_unlock(&dentry->d_lock);
2571 * When switching names, the actual string doesn't strictly have to
2572 * be preserved in the target - because we're dropping the target
2573 * anyway. As such, we can just do a simple memcpy() to copy over
2574 * the new name before we switch, unless we are going to rehash
2575 * it. Note that if we *do* unhash the target, we are not allowed
2576 * to rehash it without giving it a new name/hash key - whether
2577 * we swap or overwrite the names here, resulting name won't match
2578 * the reality in filesystem; it's only there for d_path() purposes.
2579 * Note that all of this is happening under rename_lock, so the
2580 * any hash lookup seeing it in the middle of manipulations will
2581 * be discarded anyway. So we do not care what happens to the hash
2582 * key in that case.
2585 * __d_move - move a dentry
2586 * @dentry: entry to move
2587 * @target: new dentry
2588 * @exchange: exchange the two dentries
2590 * Update the dcache to reflect the move of a file name. Negative
2591 * dcache entries should not be moved in this way. Caller must hold
2592 * rename_lock, the i_mutex of the source and target directories,
2593 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2595 static void __d_move(struct dentry *dentry, struct dentry *target,
2596 bool exchange)
2598 if (!dentry->d_inode)
2599 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2601 BUG_ON(d_ancestor(dentry, target));
2602 BUG_ON(d_ancestor(target, dentry));
2604 dentry_lock_for_move(dentry, target);
2606 write_seqcount_begin(&dentry->d_seq);
2607 write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2609 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2612 * Move the dentry to the target hash queue. Don't bother checking
2613 * for the same hash queue because of how unlikely it is.
2615 __d_drop(dentry);
2616 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2619 * Unhash the target (d_delete() is not usable here). If exchanging
2620 * the two dentries, then rehash onto the other's hash queue.
2622 __d_drop(target);
2623 if (exchange) {
2624 __d_rehash(target,
2625 d_hash(dentry->d_parent, dentry->d_name.hash));
2628 /* Switch the names.. */
2629 if (exchange)
2630 swap_names(dentry, target);
2631 else
2632 copy_name(dentry, target);
2634 /* ... and switch them in the tree */
2635 if (IS_ROOT(dentry)) {
2636 /* splicing a tree */
2637 dentry->d_flags |= DCACHE_RCUACCESS;
2638 dentry->d_parent = target->d_parent;
2639 target->d_parent = target;
2640 list_del_init(&target->d_child);
2641 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2642 } else {
2643 /* swapping two dentries */
2644 swap(dentry->d_parent, target->d_parent);
2645 list_move(&target->d_child, &target->d_parent->d_subdirs);
2646 list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2647 if (exchange)
2648 fsnotify_d_move(target);
2649 fsnotify_d_move(dentry);
2652 write_seqcount_end(&target->d_seq);
2653 write_seqcount_end(&dentry->d_seq);
2655 dentry_unlock_for_move(dentry, target);
2659 * d_move - move a dentry
2660 * @dentry: entry to move
2661 * @target: new dentry
2663 * Update the dcache to reflect the move of a file name. Negative
2664 * dcache entries should not be moved in this way. See the locking
2665 * requirements for __d_move.
2667 void d_move(struct dentry *dentry, struct dentry *target)
2669 write_seqlock(&rename_lock);
2670 __d_move(dentry, target, false);
2671 write_sequnlock(&rename_lock);
2673 EXPORT_SYMBOL(d_move);
2676 * d_exchange - exchange two dentries
2677 * @dentry1: first dentry
2678 * @dentry2: second dentry
2680 void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2682 write_seqlock(&rename_lock);
2684 WARN_ON(!dentry1->d_inode);
2685 WARN_ON(!dentry2->d_inode);
2686 WARN_ON(IS_ROOT(dentry1));
2687 WARN_ON(IS_ROOT(dentry2));
2689 __d_move(dentry1, dentry2, true);
2691 write_sequnlock(&rename_lock);
2695 * d_ancestor - search for an ancestor
2696 * @p1: ancestor dentry
2697 * @p2: child dentry
2699 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2700 * an ancestor of p2, else NULL.
2702 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2704 struct dentry *p;
2706 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2707 if (p->d_parent == p1)
2708 return p;
2710 return NULL;
2714 * This helper attempts to cope with remotely renamed directories
2716 * It assumes that the caller is already holding
2717 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2719 * Note: If ever the locking in lock_rename() changes, then please
2720 * remember to update this too...
2722 static int __d_unalias(struct inode *inode,
2723 struct dentry *dentry, struct dentry *alias)
2725 struct mutex *m1 = NULL, *m2 = NULL;
2726 int ret = -ESTALE;
2728 /* If alias and dentry share a parent, then no extra locks required */
2729 if (alias->d_parent == dentry->d_parent)
2730 goto out_unalias;
2732 /* See lock_rename() */
2733 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2734 goto out_err;
2735 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2736 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2737 goto out_err;
2738 m2 = &alias->d_parent->d_inode->i_mutex;
2739 out_unalias:
2740 __d_move(alias, dentry, false);
2741 ret = 0;
2742 out_err:
2743 spin_unlock(&inode->i_lock);
2744 if (m2)
2745 mutex_unlock(m2);
2746 if (m1)
2747 mutex_unlock(m1);
2748 return ret;
2752 * d_splice_alias - splice a disconnected dentry into the tree if one exists
2753 * @inode: the inode which may have a disconnected dentry
2754 * @dentry: a negative dentry which we want to point to the inode.
2756 * If inode is a directory and has an IS_ROOT alias, then d_move that in
2757 * place of the given dentry and return it, else simply d_add the inode
2758 * to the dentry and return NULL.
2760 * If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2761 * we should error out: directories can't have multiple aliases.
2763 * This is needed in the lookup routine of any filesystem that is exportable
2764 * (via knfsd) so that we can build dcache paths to directories effectively.
2766 * If a dentry was found and moved, then it is returned. Otherwise NULL
2767 * is returned. This matches the expected return value of ->lookup.
2769 * Cluster filesystems may call this function with a negative, hashed dentry.
2770 * In that case, we know that the inode will be a regular file, and also this
2771 * will only occur during atomic_open. So we need to check for the dentry
2772 * being already hashed only in the final case.
2774 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
2776 if (IS_ERR(inode))
2777 return ERR_CAST(inode);
2779 BUG_ON(!d_unhashed(dentry));
2781 if (!inode) {
2782 __d_instantiate(dentry, NULL);
2783 goto out;
2785 spin_lock(&inode->i_lock);
2786 if (S_ISDIR(inode->i_mode)) {
2787 struct dentry *new = __d_find_any_alias(inode);
2788 if (unlikely(new)) {
2789 write_seqlock(&rename_lock);
2790 if (unlikely(d_ancestor(new, dentry))) {
2791 write_sequnlock(&rename_lock);
2792 spin_unlock(&inode->i_lock);
2793 dput(new);
2794 new = ERR_PTR(-ELOOP);
2795 pr_warn_ratelimited(
2796 "VFS: Lookup of '%s' in %s %s"
2797 " would have caused loop\n",
2798 dentry->d_name.name,
2799 inode->i_sb->s_type->name,
2800 inode->i_sb->s_id);
2801 } else if (!IS_ROOT(new)) {
2802 int err = __d_unalias(inode, dentry, new);
2803 write_sequnlock(&rename_lock);
2804 if (err) {
2805 dput(new);
2806 new = ERR_PTR(err);
2808 } else {
2809 __d_move(new, dentry, false);
2810 write_sequnlock(&rename_lock);
2811 spin_unlock(&inode->i_lock);
2812 security_d_instantiate(new, inode);
2814 iput(inode);
2815 return new;
2818 /* already taking inode->i_lock, so d_add() by hand */
2819 __d_instantiate(dentry, inode);
2820 spin_unlock(&inode->i_lock);
2821 out:
2822 security_d_instantiate(dentry, inode);
2823 d_rehash(dentry);
2824 return NULL;
2826 EXPORT_SYMBOL(d_splice_alias);
2828 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2830 *buflen -= namelen;
2831 if (*buflen < 0)
2832 return -ENAMETOOLONG;
2833 *buffer -= namelen;
2834 memcpy(*buffer, str, namelen);
2835 return 0;
2839 * prepend_name - prepend a pathname in front of current buffer pointer
2840 * @buffer: buffer pointer
2841 * @buflen: allocated length of the buffer
2842 * @name: name string and length qstr structure
2844 * With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
2845 * make sure that either the old or the new name pointer and length are
2846 * fetched. However, there may be mismatch between length and pointer.
2847 * The length cannot be trusted, we need to copy it byte-by-byte until
2848 * the length is reached or a null byte is found. It also prepends "/" at
2849 * the beginning of the name. The sequence number check at the caller will
2850 * retry it again when a d_move() does happen. So any garbage in the buffer
2851 * due to mismatched pointer and length will be discarded.
2853 * Data dependency barrier is needed to make sure that we see that terminating
2854 * NUL. Alpha strikes again, film at 11...
2856 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2858 const char *dname = ACCESS_ONCE(name->name);
2859 u32 dlen = ACCESS_ONCE(name->len);
2860 char *p;
2862 smp_read_barrier_depends();
2864 *buflen -= dlen + 1;
2865 if (*buflen < 0)
2866 return -ENAMETOOLONG;
2867 p = *buffer -= dlen + 1;
2868 *p++ = '/';
2869 while (dlen--) {
2870 char c = *dname++;
2871 if (!c)
2872 break;
2873 *p++ = c;
2875 return 0;
2879 * prepend_path - Prepend path string to a buffer
2880 * @path: the dentry/vfsmount to report
2881 * @root: root vfsmnt/dentry
2882 * @buffer: pointer to the end of the buffer
2883 * @buflen: pointer to buffer length
2885 * The function will first try to write out the pathname without taking any
2886 * lock other than the RCU read lock to make sure that dentries won't go away.
2887 * It only checks the sequence number of the global rename_lock as any change
2888 * in the dentry's d_seq will be preceded by changes in the rename_lock
2889 * sequence number. If the sequence number had been changed, it will restart
2890 * the whole pathname back-tracing sequence again by taking the rename_lock.
2891 * In this case, there is no need to take the RCU read lock as the recursive
2892 * parent pointer references will keep the dentry chain alive as long as no
2893 * rename operation is performed.
2895 static int prepend_path(const struct path *path,
2896 const struct path *root,
2897 char **buffer, int *buflen)
2899 struct dentry *dentry;
2900 struct vfsmount *vfsmnt;
2901 struct mount *mnt;
2902 int error = 0;
2903 unsigned seq, m_seq = 0;
2904 char *bptr;
2905 int blen;
2907 rcu_read_lock();
2908 restart_mnt:
2909 read_seqbegin_or_lock(&mount_lock, &m_seq);
2910 seq = 0;
2911 rcu_read_lock();
2912 restart:
2913 bptr = *buffer;
2914 blen = *buflen;
2915 error = 0;
2916 dentry = path->dentry;
2917 vfsmnt = path->mnt;
2918 mnt = real_mount(vfsmnt);
2919 read_seqbegin_or_lock(&rename_lock, &seq);
2920 while (dentry != root->dentry || vfsmnt != root->mnt) {
2921 struct dentry * parent;
2923 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2924 struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
2925 /* Escaped? */
2926 if (dentry != vfsmnt->mnt_root) {
2927 bptr = *buffer;
2928 blen = *buflen;
2929 error = 3;
2930 break;
2932 /* Global root? */
2933 if (mnt != parent) {
2934 dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
2935 mnt = parent;
2936 vfsmnt = &mnt->mnt;
2937 continue;
2939 if (!error)
2940 error = is_mounted(vfsmnt) ? 1 : 2;
2941 break;
2943 parent = dentry->d_parent;
2944 prefetch(parent);
2945 error = prepend_name(&bptr, &blen, &dentry->d_name);
2946 if (error)
2947 break;
2949 dentry = parent;
2951 if (!(seq & 1))
2952 rcu_read_unlock();
2953 if (need_seqretry(&rename_lock, seq)) {
2954 seq = 1;
2955 goto restart;
2957 done_seqretry(&rename_lock, seq);
2959 if (!(m_seq & 1))
2960 rcu_read_unlock();
2961 if (need_seqretry(&mount_lock, m_seq)) {
2962 m_seq = 1;
2963 goto restart_mnt;
2965 done_seqretry(&mount_lock, m_seq);
2967 if (error >= 0 && bptr == *buffer) {
2968 if (--blen < 0)
2969 error = -ENAMETOOLONG;
2970 else
2971 *--bptr = '/';
2973 *buffer = bptr;
2974 *buflen = blen;
2975 return error;
2979 * __d_path - return the path of a dentry
2980 * @path: the dentry/vfsmount to report
2981 * @root: root vfsmnt/dentry
2982 * @buf: buffer to return value in
2983 * @buflen: buffer length
2985 * Convert a dentry into an ASCII path name.
2987 * Returns a pointer into the buffer or an error code if the
2988 * path was too long.
2990 * "buflen" should be positive.
2992 * If the path is not reachable from the supplied root, return %NULL.
2994 char *__d_path(const struct path *path,
2995 const struct path *root,
2996 char *buf, int buflen)
2998 char *res = buf + buflen;
2999 int error;
3001 prepend(&res, &buflen, "\0", 1);
3002 error = prepend_path(path, root, &res, &buflen);
3004 if (error < 0)
3005 return ERR_PTR(error);
3006 if (error > 0)
3007 return NULL;
3008 return res;
3011 char *d_absolute_path(const struct path *path,
3012 char *buf, int buflen)
3014 struct path root = {};
3015 char *res = buf + buflen;
3016 int error;
3018 prepend(&res, &buflen, "\0", 1);
3019 error = prepend_path(path, &root, &res, &buflen);
3021 if (error > 1)
3022 error = -EINVAL;
3023 if (error < 0)
3024 return ERR_PTR(error);
3025 return res;
3029 * same as __d_path but appends "(deleted)" for unlinked files.
3031 static int path_with_deleted(const struct path *path,
3032 const struct path *root,
3033 char **buf, int *buflen)
3035 prepend(buf, buflen, "\0", 1);
3036 if (d_unlinked(path->dentry)) {
3037 int error = prepend(buf, buflen, " (deleted)", 10);
3038 if (error)
3039 return error;
3042 return prepend_path(path, root, buf, buflen);
3045 static int prepend_unreachable(char **buffer, int *buflen)
3047 return prepend(buffer, buflen, "(unreachable)", 13);
3050 static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3052 unsigned seq;
3054 do {
3055 seq = read_seqcount_begin(&fs->seq);
3056 *root = fs->root;
3057 } while (read_seqcount_retry(&fs->seq, seq));
3061 * d_path - return the path of a dentry
3062 * @path: path to report
3063 * @buf: buffer to return value in
3064 * @buflen: buffer length
3066 * Convert a dentry into an ASCII path name. If the entry has been deleted
3067 * the string " (deleted)" is appended. Note that this is ambiguous.
3069 * Returns a pointer into the buffer or an error code if the path was
3070 * too long. Note: Callers should use the returned pointer, not the passed
3071 * in buffer, to use the name! The implementation often starts at an offset
3072 * into the buffer, and may leave 0 bytes at the start.
3074 * "buflen" should be positive.
3076 char *d_path(const struct path *path, char *buf, int buflen)
3078 char *res = buf + buflen;
3079 struct path root;
3080 int error;
3083 * We have various synthetic filesystems that never get mounted. On
3084 * these filesystems dentries are never used for lookup purposes, and
3085 * thus don't need to be hashed. They also don't need a name until a
3086 * user wants to identify the object in /proc/pid/fd/. The little hack
3087 * below allows us to generate a name for these objects on demand:
3089 * Some pseudo inodes are mountable. When they are mounted
3090 * path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3091 * and instead have d_path return the mounted path.
3093 if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3094 (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3095 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3097 rcu_read_lock();
3098 get_fs_root_rcu(current->fs, &root);
3099 error = path_with_deleted(path, &root, &res, &buflen);
3100 rcu_read_unlock();
3102 if (error < 0)
3103 res = ERR_PTR(error);
3104 return res;
3106 EXPORT_SYMBOL(d_path);
3109 * Helper function for dentry_operations.d_dname() members
3111 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3112 const char *fmt, ...)
3114 va_list args;
3115 char temp[64];
3116 int sz;
3118 va_start(args, fmt);
3119 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3120 va_end(args);
3122 if (sz > sizeof(temp) || sz > buflen)
3123 return ERR_PTR(-ENAMETOOLONG);
3125 buffer += buflen - sz;
3126 return memcpy(buffer, temp, sz);
3129 char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3131 char *end = buffer + buflen;
3132 /* these dentries are never renamed, so d_lock is not needed */
3133 if (prepend(&end, &buflen, " (deleted)", 11) ||
3134 prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3135 prepend(&end, &buflen, "/", 1))
3136 end = ERR_PTR(-ENAMETOOLONG);
3137 return end;
3139 EXPORT_SYMBOL(simple_dname);
3142 * Write full pathname from the root of the filesystem into the buffer.
3144 static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3146 struct dentry *dentry;
3147 char *end, *retval;
3148 int len, seq = 0;
3149 int error = 0;
3151 if (buflen < 2)
3152 goto Elong;
3154 rcu_read_lock();
3155 restart:
3156 dentry = d;
3157 end = buf + buflen;
3158 len = buflen;
3159 prepend(&end, &len, "\0", 1);
3160 /* Get '/' right */
3161 retval = end-1;
3162 *retval = '/';
3163 read_seqbegin_or_lock(&rename_lock, &seq);
3164 while (!IS_ROOT(dentry)) {
3165 struct dentry *parent = dentry->d_parent;
3167 prefetch(parent);
3168 error = prepend_name(&end, &len, &dentry->d_name);
3169 if (error)
3170 break;
3172 retval = end;
3173 dentry = parent;
3175 if (!(seq & 1))
3176 rcu_read_unlock();
3177 if (need_seqretry(&rename_lock, seq)) {
3178 seq = 1;
3179 goto restart;
3181 done_seqretry(&rename_lock, seq);
3182 if (error)
3183 goto Elong;
3184 return retval;
3185 Elong:
3186 return ERR_PTR(-ENAMETOOLONG);
3189 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3191 return __dentry_path(dentry, buf, buflen);
3193 EXPORT_SYMBOL(dentry_path_raw);
3195 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3197 char *p = NULL;
3198 char *retval;
3200 if (d_unlinked(dentry)) {
3201 p = buf + buflen;
3202 if (prepend(&p, &buflen, "//deleted", 10) != 0)
3203 goto Elong;
3204 buflen++;
3206 retval = __dentry_path(dentry, buf, buflen);
3207 if (!IS_ERR(retval) && p)
3208 *p = '/'; /* restore '/' overriden with '\0' */
3209 return retval;
3210 Elong:
3211 return ERR_PTR(-ENAMETOOLONG);
3214 static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3215 struct path *pwd)
3217 unsigned seq;
3219 do {
3220 seq = read_seqcount_begin(&fs->seq);
3221 *root = fs->root;
3222 *pwd = fs->pwd;
3223 } while (read_seqcount_retry(&fs->seq, seq));
3227 * NOTE! The user-level library version returns a
3228 * character pointer. The kernel system call just
3229 * returns the length of the buffer filled (which
3230 * includes the ending '\0' character), or a negative
3231 * error value. So libc would do something like
3233 * char *getcwd(char * buf, size_t size)
3235 * int retval;
3237 * retval = sys_getcwd(buf, size);
3238 * if (retval >= 0)
3239 * return buf;
3240 * errno = -retval;
3241 * return NULL;
3244 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3246 int error;
3247 struct path pwd, root;
3248 char *page = __getname();
3250 if (!page)
3251 return -ENOMEM;
3253 rcu_read_lock();
3254 get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3256 error = -ENOENT;
3257 if (!d_unlinked(pwd.dentry)) {
3258 unsigned long len;
3259 char *cwd = page + PATH_MAX;
3260 int buflen = PATH_MAX;
3262 prepend(&cwd, &buflen, "\0", 1);
3263 error = prepend_path(&pwd, &root, &cwd, &buflen);
3264 rcu_read_unlock();
3266 if (error < 0)
3267 goto out;
3269 /* Unreachable from current root */
3270 if (error > 0) {
3271 error = prepend_unreachable(&cwd, &buflen);
3272 if (error)
3273 goto out;
3276 error = -ERANGE;
3277 len = PATH_MAX + page - cwd;
3278 if (len <= size) {
3279 error = len;
3280 if (copy_to_user(buf, cwd, len))
3281 error = -EFAULT;
3283 } else {
3284 rcu_read_unlock();
3287 out:
3288 __putname(page);
3289 return error;
3293 * Test whether new_dentry is a subdirectory of old_dentry.
3295 * Trivially implemented using the dcache structure
3299 * is_subdir - is new dentry a subdirectory of old_dentry
3300 * @new_dentry: new dentry
3301 * @old_dentry: old dentry
3303 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
3304 * Returns 0 otherwise.
3305 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3308 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3310 int result;
3311 unsigned seq;
3313 if (new_dentry == old_dentry)
3314 return 1;
3316 do {
3317 /* for restarting inner loop in case of seq retry */
3318 seq = read_seqbegin(&rename_lock);
3320 * Need rcu_readlock to protect against the d_parent trashing
3321 * due to d_move
3323 rcu_read_lock();
3324 if (d_ancestor(old_dentry, new_dentry))
3325 result = 1;
3326 else
3327 result = 0;
3328 rcu_read_unlock();
3329 } while (read_seqretry(&rename_lock, seq));
3331 return result;
3334 static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3336 struct dentry *root = data;
3337 if (dentry != root) {
3338 if (d_unhashed(dentry) || !dentry->d_inode)
3339 return D_WALK_SKIP;
3341 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3342 dentry->d_flags |= DCACHE_GENOCIDE;
3343 dentry->d_lockref.count--;
3346 return D_WALK_CONTINUE;
3349 void d_genocide(struct dentry *parent)
3351 d_walk(parent, parent, d_genocide_kill, NULL);
3354 void d_tmpfile(struct dentry *dentry, struct inode *inode)
3356 inode_dec_link_count(inode);
3357 BUG_ON(dentry->d_name.name != dentry->d_iname ||
3358 !hlist_unhashed(&dentry->d_u.d_alias) ||
3359 !d_unlinked(dentry));
3360 spin_lock(&dentry->d_parent->d_lock);
3361 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3362 dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3363 (unsigned long long)inode->i_ino);
3364 spin_unlock(&dentry->d_lock);
3365 spin_unlock(&dentry->d_parent->d_lock);
3366 d_instantiate(dentry, inode);
3368 EXPORT_SYMBOL(d_tmpfile);
3370 static __initdata unsigned long dhash_entries;
3371 static int __init set_dhash_entries(char *str)
3373 if (!str)
3374 return 0;
3375 dhash_entries = simple_strtoul(str, &str, 0);
3376 return 1;
3378 __setup("dhash_entries=", set_dhash_entries);
3380 static void __init dcache_init_early(void)
3382 unsigned int loop;
3384 /* If hashes are distributed across NUMA nodes, defer
3385 * hash allocation until vmalloc space is available.
3387 if (hashdist)
3388 return;
3390 dentry_hashtable =
3391 alloc_large_system_hash("Dentry cache",
3392 sizeof(struct hlist_bl_head),
3393 dhash_entries,
3395 HASH_EARLY,
3396 &d_hash_shift,
3397 &d_hash_mask,
3401 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3402 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3405 static void __init dcache_init(void)
3407 unsigned int loop;
3410 * A constructor could be added for stable state like the lists,
3411 * but it is probably not worth it because of the cache nature
3412 * of the dcache.
3414 dentry_cache = KMEM_CACHE(dentry,
3415 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3417 /* Hash may have been set up in dcache_init_early */
3418 if (!hashdist)
3419 return;
3421 dentry_hashtable =
3422 alloc_large_system_hash("Dentry cache",
3423 sizeof(struct hlist_bl_head),
3424 dhash_entries,
3427 &d_hash_shift,
3428 &d_hash_mask,
3432 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3433 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3436 /* SLAB cache for __getname() consumers */
3437 struct kmem_cache *names_cachep __read_mostly;
3438 EXPORT_SYMBOL(names_cachep);
3440 EXPORT_SYMBOL(d_genocide);
3442 void __init vfs_caches_init_early(void)
3444 dcache_init_early();
3445 inode_init_early();
3448 void __init vfs_caches_init(unsigned long mempages)
3450 unsigned long reserve;
3452 /* Base hash sizes on available memory, with a reserve equal to
3453 150% of current kernel size */
3455 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3456 mempages -= reserve;
3458 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3459 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3461 dcache_init();
3462 inode_init();
3463 files_init(mempages);
3464 mnt_init();
3465 bdev_cache_init();
3466 chrdev_init();