path: root/fs/dcache.c (plain)
blob: 426fe9bbec660eaf21fbb05c3a01410a59db3b10

1	/*
2	* fs/dcache.c
3	*
4	* Complete reimplementation
5	* (C) 1997 Thomas Schoebel-Theuer,
6	* with heavy changes by Linus Torvalds
7	*/
8
9	/*
10	* Notes on the allocation strategy:
11	*
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.
15	*/
16
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>
42
43	#include "internal.h"
44	#include "mount.h"
45
46	/*
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
65	*
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
72	*
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
78	*
79	* If no ancestor relationship:
80	* if (dentry1 < dentry2)
81	* dentry1->d_lock
82	* dentry2->d_lock
83	*/
84	int sysctl_vfs_cache_pressure __read_mostly = 100;
85	EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
86
87	__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
88
89	EXPORT_SYMBOL(rename_lock);
90
91	static struct kmem_cache *dentry_cache __read_mostly;
92
93	/*
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.
97	*
98	* This hash-function tries to avoid losing too many bits of hash
99	* information, yet avoid using a prime hash-size or similar.
100	*/
101
102	static unsigned int d_hash_mask __read_mostly;
103	static unsigned int d_hash_shift __read_mostly;
104
105	static struct hlist_bl_head *dentry_hashtable __read_mostly;
106
107	static inline struct hlist_bl_head *d_hash(unsigned int hash)
108	{
109	return dentry_hashtable + (hash >> (32 - d_hash_shift));
110	}
111
112	#define IN_LOOKUP_SHIFT 10
113	static struct hlist_bl_head in_lookup_hashtable[1 << IN_LOOKUP_SHIFT];
114
115	static inline struct hlist_bl_head *in_lookup_hash(const struct dentry *parent,
116	unsigned int hash)
117	{
118	hash += (unsigned long) parent / L1_CACHE_BYTES;
119	return in_lookup_hashtable + hash_32(hash, IN_LOOKUP_SHIFT);
120	}
121
122
123	/* Statistics gathering. */
124	struct dentry_stat_t dentry_stat = {
125	.age_limit = 45,
126	};
127
128	static DEFINE_PER_CPU(long, nr_dentry);
129	static DEFINE_PER_CPU(long, nr_dentry_unused);
130
131	#if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
132
133	/*
134	* Here we resort to our own counters instead of using generic per-cpu counters
135	* for consistency with what the vfs inode code does. We are expected to harvest
136	* better code and performance by having our own specialized counters.
137	*
138	* Please note that the loop is done over all possible CPUs, not over all online
139	* CPUs. The reason for this is that we don't want to play games with CPUs going
140	* on and off. If one of them goes off, we will just keep their counters.
141	*
142	* glommer: See cffbc8a for details, and if you ever intend to change this,
143	* please update all vfs counters to match.
144	*/
145	static long get_nr_dentry(void)
146	{
147	int i;
148	long sum = 0;
149	for_each_possible_cpu(i)
150	sum += per_cpu(nr_dentry, i);
151	return sum < 0 ? 0 : sum;
152	}
153
154	static long get_nr_dentry_unused(void)
155	{
156	int i;
157	long sum = 0;
158	for_each_possible_cpu(i)
159	sum += per_cpu(nr_dentry_unused, i);
160	return sum < 0 ? 0 : sum;
161	}
162
163	int proc_nr_dentry(struct ctl_table *table, int write, void __user *buffer,
164	size_t *lenp, loff_t *ppos)
165	{
166	dentry_stat.nr_dentry = get_nr_dentry();
167	dentry_stat.nr_unused = get_nr_dentry_unused();
168	return proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
169	}
170	#endif
171
172	/*
173	* Compare 2 name strings, return 0 if they match, otherwise non-zero.
174	* The strings are both count bytes long, and count is non-zero.
175	*/
176	#ifdef CONFIG_DCACHE_WORD_ACCESS
177
178	#include <asm/word-at-a-time.h>
179	/*
180	* NOTE! 'cs' and 'scount' come from a dentry, so it has a
181	* aligned allocation for this particular component. We don't
182	* strictly need the load_unaligned_zeropad() safety, but it
183	* doesn't hurt either.
184	*
185	* In contrast, 'ct' and 'tcount' can be from a pathname, and do
186	* need the careful unaligned handling.
187	*/
188	static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
189	{
190	unsigned long a,b,mask;
191
192	for (;;) {
193	a = *(unsigned long *)cs;
194	b = load_unaligned_zeropad(ct);
195	if (tcount < sizeof(unsigned long))
196	break;
197	if (unlikely(a != b))
198	return 1;
199	cs += sizeof(unsigned long);
200	ct += sizeof(unsigned long);
201	tcount -= sizeof(unsigned long);
202	if (!tcount)
203	return 0;
204	}
205	mask = bytemask_from_count(tcount);
206	return unlikely(!!((a ^ b) & mask));
207	}
208
209	#else
210
211	static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount)
212	{
213	do {
214	if (*cs != *ct)
215	return 1;
216	cs++;
217	ct++;
218	tcount--;
219	} while (tcount);
220	return 0;
221	}
222
223	#endif
224
225	static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount)
226	{
227	/*
228	* Be careful about RCU walk racing with rename:
229	* use 'lockless_dereference' to fetch the name pointer.
230	*
231	* NOTE! Even if a rename will mean that the length
232	* was not loaded atomically, we don't care. The
233	* RCU walk will check the sequence count eventually,
234	* and catch it. And we won't overrun the buffer,
235	* because we're reading the name pointer atomically,
236	* and a dentry name is guaranteed to be properly
237	* terminated with a NUL byte.
238	*
239	* End result: even if 'len' is wrong, we'll exit
240	* early because the data cannot match (there can
241	* be no NUL in the ct/tcount data)
242	*/
243	const unsigned char *cs = lockless_dereference(dentry->d_name.name);
244
245	return dentry_string_cmp(cs, ct, tcount);
246	}
247
248	struct external_name {
249	union {
250	atomic_t count;
251	struct rcu_head head;
252	} u;
253	unsigned char name[];
254	};
255
256	static inline struct external_name *external_name(struct dentry *dentry)
257	{
258	return container_of(dentry->d_name.name, struct external_name, name[0]);
259	}
260
261	static void __d_free(struct rcu_head *head)
262	{
263	struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
264
265	kmem_cache_free(dentry_cache, dentry);
266	}
267
268	static void __d_free_external(struct rcu_head *head)
269	{
270	struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
271	kfree(external_name(dentry));
272	kmem_cache_free(dentry_cache, dentry);
273	}
274
275	static inline int dname_external(const struct dentry *dentry)
276	{
277	return dentry->d_name.name != dentry->d_iname;
278	}
279
280	void take_dentry_name_snapshot(struct name_snapshot *name, struct dentry *dentry)
281	{
282	spin_lock(&dentry->d_lock);
283	if (unlikely(dname_external(dentry))) {
284	struct external_name *p = external_name(dentry);
285	atomic_inc(&p->u.count);
286	spin_unlock(&dentry->d_lock);
287	name->name = p->name;
288	} else {
289	memcpy(name->inline_name, dentry->d_iname,
290	dentry->d_name.len + 1);
291	spin_unlock(&dentry->d_lock);
292	name->name = name->inline_name;
293	}
294	}
295	EXPORT_SYMBOL(take_dentry_name_snapshot);
296
297	void release_dentry_name_snapshot(struct name_snapshot *name)
298	{
299	if (unlikely(name->name != name->inline_name)) {
300	struct external_name *p;
301	p = container_of(name->name, struct external_name, name[0]);
302	if (unlikely(atomic_dec_and_test(&p->u.count)))
303	kfree_rcu(p, u.head);
304	}
305	}
306	EXPORT_SYMBOL(release_dentry_name_snapshot);
307
308	static inline void __d_set_inode_and_type(struct dentry *dentry,
309	struct inode *inode,
310	unsigned type_flags)
311	{
312	unsigned flags;
313
314	dentry->d_inode = inode;
315	flags = READ_ONCE(dentry->d_flags);
316	flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
317	flags |= type_flags;
318	WRITE_ONCE(dentry->d_flags, flags);
319	}
320
321	static inline void __d_clear_type_and_inode(struct dentry *dentry)
322	{
323	unsigned flags = READ_ONCE(dentry->d_flags);
324
325	flags &= ~(DCACHE_ENTRY_TYPE | DCACHE_FALLTHRU);
326	WRITE_ONCE(dentry->d_flags, flags);
327	dentry->d_inode = NULL;
328	}
329
330	static void dentry_free(struct dentry *dentry)
331	{
332	WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias));
333	if (unlikely(dname_external(dentry))) {
334	struct external_name *p = external_name(dentry);
335	if (likely(atomic_dec_and_test(&p->u.count))) {
336	call_rcu(&dentry->d_u.d_rcu, __d_free_external);
337	return;
338	}
339	}
340	/* if dentry was never visible to RCU, immediate free is OK */
341	if (!(dentry->d_flags & DCACHE_RCUACCESS))
342	__d_free(&dentry->d_u.d_rcu);
343	else
344	call_rcu(&dentry->d_u.d_rcu, __d_free);
345	}
346
347	/*
348	* Release the dentry's inode, using the filesystem
349	* d_iput() operation if defined.
350	*/
351	static void dentry_unlink_inode(struct dentry * dentry)
352	__releases(dentry->d_lock)
353	__releases(dentry->d_inode->i_lock)
354	{
355	struct inode *inode = dentry->d_inode;
356
357	raw_write_seqcount_begin(&dentry->d_seq);
358	__d_clear_type_and_inode(dentry);
359	hlist_del_init(&dentry->d_u.d_alias);
360	raw_write_seqcount_end(&dentry->d_seq);
361	spin_unlock(&dentry->d_lock);
362	spin_unlock(&inode->i_lock);
363	if (!inode->i_nlink)
364	fsnotify_inoderemove(inode);
365	if (dentry->d_op && dentry->d_op->d_iput)
366	dentry->d_op->d_iput(dentry, inode);
367	else
368	iput(inode);
369	}
370
371	/*
372	* The DCACHE_LRU_LIST bit is set whenever the 'd_lru' entry
373	* is in use - which includes both the "real" per-superblock
374	* LRU list _and_ the DCACHE_SHRINK_LIST use.
375	*
376	* The DCACHE_SHRINK_LIST bit is set whenever the dentry is
377	* on the shrink list (ie not on the superblock LRU list).
378	*
379	* The per-cpu "nr_dentry_unused" counters are updated with
380	* the DCACHE_LRU_LIST bit.
381	*
382	* These helper functions make sure we always follow the
383	* rules. d_lock must be held by the caller.
384	*/
385	#define D_FLAG_VERIFY(dentry,x) WARN_ON_ONCE(((dentry)->d_flags & (DCACHE_LRU_LIST | DCACHE_SHRINK_LIST)) != (x))
386	static void d_lru_add(struct dentry *dentry)
387	{
388	D_FLAG_VERIFY(dentry, 0);
389	dentry->d_flags |= DCACHE_LRU_LIST;
390	this_cpu_inc(nr_dentry_unused);
391	WARN_ON_ONCE(!list_lru_add(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
392	}
393
394	static void d_lru_del(struct dentry *dentry)
395	{
396	D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
397	dentry->d_flags &= ~DCACHE_LRU_LIST;
398	this_cpu_dec(nr_dentry_unused);
399	WARN_ON_ONCE(!list_lru_del(&dentry->d_sb->s_dentry_lru, &dentry->d_lru));
400	}
401
402	static void d_shrink_del(struct dentry *dentry)
403	{
404	D_FLAG_VERIFY(dentry, DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
405	list_del_init(&dentry->d_lru);
406	dentry->d_flags &= ~(DCACHE_SHRINK_LIST | DCACHE_LRU_LIST);
407	this_cpu_dec(nr_dentry_unused);
408	}
409
410	static void d_shrink_add(struct dentry *dentry, struct list_head *list)
411	{
412	D_FLAG_VERIFY(dentry, 0);
413	list_add(&dentry->d_lru, list);
414	dentry->d_flags |= DCACHE_SHRINK_LIST | DCACHE_LRU_LIST;
415	this_cpu_inc(nr_dentry_unused);
416	}
417
418	/*
419	* These can only be called under the global LRU lock, ie during the
420	* callback for freeing the LRU list. "isolate" removes it from the
421	* LRU lists entirely, while shrink_move moves it to the indicated
422	* private list.
423	*/
424	static void d_lru_isolate(struct list_lru_one *lru, struct dentry *dentry)
425	{
426	D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
427	dentry->d_flags &= ~DCACHE_LRU_LIST;
428	this_cpu_dec(nr_dentry_unused);
429	list_lru_isolate(lru, &dentry->d_lru);
430	}
431
432	static void d_lru_shrink_move(struct list_lru_one *lru, struct dentry *dentry,
433	struct list_head *list)
434	{
435	D_FLAG_VERIFY(dentry, DCACHE_LRU_LIST);
436	dentry->d_flags |= DCACHE_SHRINK_LIST;
437	list_lru_isolate_move(lru, &dentry->d_lru, list);
438	}
439
440	/*
441	* dentry_lru_(add|del)_list) must be called with d_lock held.
442	*/
443	static void dentry_lru_add(struct dentry *dentry)
444	{
445	if (unlikely(!(dentry->d_flags & DCACHE_LRU_LIST)))
446	d_lru_add(dentry);
447	}
448
449	/**
450	* d_drop - drop a dentry
451	* @dentry: dentry to drop
452	*
453	* d_drop() unhashes the entry from the parent dentry hashes, so that it won't
454	* be found through a VFS lookup any more. Note that this is different from
455	* deleting the dentry - d_delete will try to mark the dentry negative if
456	* possible, giving a successful _negative_ lookup, while d_drop will
457	* just make the cache lookup fail.
458	*
459	* d_drop() is used mainly for stuff that wants to invalidate a dentry for some
460	* reason (NFS timeouts or autofs deletes).
461	*
462	* __d_drop requires dentry->d_lock
463	* ___d_drop doesn't mark dentry as "unhashed"
464	* (dentry->d_hash.pprev will be LIST_POISON2, not NULL).
465	*/
466	static void ___d_drop(struct dentry *dentry)
467	{
468	if (!d_unhashed(dentry)) {
469	struct hlist_bl_head *b;
470	/*
471	* Hashed dentries are normally on the dentry hashtable,
472	* with the exception of those newly allocated by
473	* d_obtain_alias, which are always IS_ROOT:
474	*/
475	if (unlikely(IS_ROOT(dentry)))
476	b = &dentry->d_sb->s_anon;
477	else
478	b = d_hash(dentry->d_name.hash);
479
480	hlist_bl_lock(b);
481	__hlist_bl_del(&dentry->d_hash);
482	hlist_bl_unlock(b);
483	/* After this call, in-progress rcu-walk path lookup will fail. */
484	write_seqcount_invalidate(&dentry->d_seq);
485	}
486	}
487
488	void __d_drop(struct dentry *dentry)
489	{
490	___d_drop(dentry);
491	dentry->d_hash.pprev = NULL;
492	}
493	EXPORT_SYMBOL(__d_drop);
494
495	void d_drop(struct dentry *dentry)
496	{
497	spin_lock(&dentry->d_lock);
498	__d_drop(dentry);
499	spin_unlock(&dentry->d_lock);
500	}
501	EXPORT_SYMBOL(d_drop);
502
503	static inline void dentry_unlist(struct dentry *dentry, struct dentry *parent)
504	{
505	struct dentry *next;
506	/*
507	* Inform d_walk() and shrink_dentry_list() that we are no longer
508	* attached to the dentry tree
509	*/
510	dentry->d_flags |= DCACHE_DENTRY_KILLED;
511	if (unlikely(list_empty(&dentry->d_child)))
512	return;
513	__list_del_entry(&dentry->d_child);
514	/*
515	* Cursors can move around the list of children. While we'd been
516	* a normal list member, it didn't matter - ->d_child.next would've
517	* been updated. However, from now on it won't be and for the
518	* things like d_walk() it might end up with a nasty surprise.
519	* Normally d_walk() doesn't care about cursors moving around -
520	* ->d_lock on parent prevents that and since a cursor has no children
521	* of its own, we get through it without ever unlocking the parent.
522	* There is one exception, though - if we ascend from a child that
523	* gets killed as soon as we unlock it, the next sibling is found
524	* using the value left in its ->d_child.next. And if _that_
525	* pointed to a cursor, and cursor got moved (e.g. by lseek())
526	* before d_walk() regains parent->d_lock, we'll end up skipping
527	* everything the cursor had been moved past.
528	*
529	* Solution: make sure that the pointer left behind in ->d_child.next
530	* points to something that won't be moving around. I.e. skip the
531	* cursors.
532	*/
533	while (dentry->d_child.next != &parent->d_subdirs) {
534	next = list_entry(dentry->d_child.next, struct dentry, d_child);
535	if (likely(!(next->d_flags & DCACHE_DENTRY_CURSOR)))
536	break;
537	dentry->d_child.next = next->d_child.next;
538	}
539	}
540
541	static void __dentry_kill(struct dentry *dentry)
542	{
543	struct dentry *parent = NULL;
544	bool can_free = true;
545	if (!IS_ROOT(dentry))
546	parent = dentry->d_parent;
547
548	/*
549	* The dentry is now unrecoverably dead to the world.
550	*/
551	lockref_mark_dead(&dentry->d_lockref);
552
553	/*
554	* inform the fs via d_prune that this dentry is about to be
555	* unhashed and destroyed.
556	*/
557	if (dentry->d_flags & DCACHE_OP_PRUNE)
558	dentry->d_op->d_prune(dentry);
559
560	if (dentry->d_flags & DCACHE_LRU_LIST) {
561	if (!(dentry->d_flags & DCACHE_SHRINK_LIST))
562	d_lru_del(dentry);
563	}
564	/* if it was on the hash then remove it */
565	__d_drop(dentry);
566	dentry_unlist(dentry, parent);
567	if (parent)
568	spin_unlock(&parent->d_lock);
569	if (dentry->d_inode)
570	dentry_unlink_inode(dentry);
571	else
572	spin_unlock(&dentry->d_lock);
573	this_cpu_dec(nr_dentry);
574	if (dentry->d_op && dentry->d_op->d_release)
575	dentry->d_op->d_release(dentry);
576
577	spin_lock(&dentry->d_lock);
578	if (dentry->d_flags & DCACHE_SHRINK_LIST) {
579	dentry->d_flags |= DCACHE_MAY_FREE;
580	can_free = false;
581	}
582	spin_unlock(&dentry->d_lock);
583	if (likely(can_free))
584	dentry_free(dentry);
585	}
586
587	/*
588	* Finish off a dentry we've decided to kill.
589	* dentry->d_lock must be held, returns with it unlocked.
590	* If ref is non-zero, then decrement the refcount too.
591	* Returns dentry requiring refcount drop, or NULL if we're done.
592	*/
593	static struct dentry *dentry_kill(struct dentry *dentry)
594	__releases(dentry->d_lock)
595	{
596	struct inode *inode = dentry->d_inode;
597	struct dentry *parent = NULL;
598
599	if (inode && unlikely(!spin_trylock(&inode->i_lock)))
600	goto failed;
601
602	if (!IS_ROOT(dentry)) {
603	parent = dentry->d_parent;
604	if (unlikely(!spin_trylock(&parent->d_lock))) {
605	if (inode)
606	spin_unlock(&inode->i_lock);
607	goto failed;
608	}
609	}
610
611	__dentry_kill(dentry);
612	return parent;
613
614	failed:
615	spin_unlock(&dentry->d_lock);
616	return dentry; /* try again with same dentry */
617	}
618
619	static inline struct dentry *lock_parent(struct dentry *dentry)
620	{
621	struct dentry *parent = dentry->d_parent;
622	if (IS_ROOT(dentry))
623	return NULL;
624	if (unlikely(dentry->d_lockref.count < 0))
625	return NULL;
626	if (likely(spin_trylock(&parent->d_lock)))
627	return parent;
628	rcu_read_lock();
629	spin_unlock(&dentry->d_lock);
630	again:
631	parent = ACCESS_ONCE(dentry->d_parent);
632	spin_lock(&parent->d_lock);
633	/*
634	* We can't blindly lock dentry until we are sure
635	* that we won't violate the locking order.
636	* Any changes of dentry->d_parent must have
637	* been done with parent->d_lock held, so
638	* spin_lock() above is enough of a barrier
639	* for checking if it's still our child.
640	*/
641	if (unlikely(parent != dentry->d_parent)) {
642	spin_unlock(&parent->d_lock);
643	goto again;
644	}
645	if (parent != dentry) {
646	spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
647	if (unlikely(dentry->d_lockref.count < 0)) {
648	spin_unlock(&parent->d_lock);
649	parent = NULL;
650	}
651	} else {
652	parent = NULL;
653	}
654	rcu_read_unlock();
655	return parent;
656	}
657
658	/*
659	* Try to do a lockless dput(), and return whether that was successful.
660	*
661	* If unsuccessful, we return false, having already taken the dentry lock.
662	*
663	* The caller needs to hold the RCU read lock, so that the dentry is
664	* guaranteed to stay around even if the refcount goes down to zero!
665	*/
666	static inline bool fast_dput(struct dentry *dentry)
667	{
668	int ret;
669	unsigned int d_flags;
670
671	/*
672	* If we have a d_op->d_delete() operation, we sould not
673	* let the dentry count go to zero, so use "put_or_lock".
674	*/
675	if (unlikely(dentry->d_flags & DCACHE_OP_DELETE))
676	return lockref_put_or_lock(&dentry->d_lockref);
677
678	/*
679	* .. otherwise, we can try to just decrement the
680	* lockref optimistically.
681	*/
682	ret = lockref_put_return(&dentry->d_lockref);
683
684	/*
685	* If the lockref_put_return() failed due to the lock being held
686	* by somebody else, the fast path has failed. We will need to
687	* get the lock, and then check the count again.
688	*/
689	if (unlikely(ret < 0)) {
690	spin_lock(&dentry->d_lock);
691	if (dentry->d_lockref.count > 1) {
692	dentry->d_lockref.count--;
693	spin_unlock(&dentry->d_lock);
694	return 1;
695	}
696	return 0;
697	}
698
699	/*
700	* If we weren't the last ref, we're done.
701	*/
702	if (ret)
703	return 1;
704
705	/*
706	* Careful, careful. The reference count went down
707	* to zero, but we don't hold the dentry lock, so
708	* somebody else could get it again, and do another
709	* dput(), and we need to not race with that.
710	*
711	* However, there is a very special and common case
712	* where we don't care, because there is nothing to
713	* do: the dentry is still hashed, it does not have
714	* a 'delete' op, and it's referenced and already on
715	* the LRU list.
716	*
717	* NOTE! Since we aren't locked, these values are
718	* not "stable". However, it is sufficient that at
719	* some point after we dropped the reference the
720	* dentry was hashed and the flags had the proper
721	* value. Other dentry users may have re-gotten
722	* a reference to the dentry and change that, but
723	* our work is done - we can leave the dentry
724	* around with a zero refcount.
725	*/
726	smp_rmb();
727	d_flags = ACCESS_ONCE(dentry->d_flags);
728	d_flags &= DCACHE_REFERENCED | DCACHE_LRU_LIST | DCACHE_DISCONNECTED;
729
730	/* Nothing to do? Dropping the reference was all we needed? */
731	if (d_flags == (DCACHE_REFERENCED | DCACHE_LRU_LIST) && !d_unhashed(dentry))
732	return 1;
733
734	/*
735	* Not the fast normal case? Get the lock. We've already decremented
736	* the refcount, but we'll need to re-check the situation after
737	* getting the lock.
738	*/
739	spin_lock(&dentry->d_lock);
740
741	/*
742	* Did somebody else grab a reference to it in the meantime, and
743	* we're no longer the last user after all? Alternatively, somebody
744	* else could have killed it and marked it dead. Either way, we
745	* don't need to do anything else.
746	*/
747	if (dentry->d_lockref.count) {
748	spin_unlock(&dentry->d_lock);
749	return 1;
750	}
751
752	/*
753	* Re-get the reference we optimistically dropped. We hold the
754	* lock, and we just tested that it was zero, so we can just
755	* set it to 1.
756	*/
757	dentry->d_lockref.count = 1;
758	return 0;
759	}
760
761
762	/*
763	* This is dput
764	*
765	* This is complicated by the fact that we do not want to put
766	* dentries that are no longer on any hash chain on the unused
767	* list: we'd much rather just get rid of them immediately.
768	*
769	* However, that implies that we have to traverse the dentry
770	* tree upwards to the parents which might _also_ now be
771	* scheduled for deletion (it may have been only waiting for
772	* its last child to go away).
773	*
774	* This tail recursion is done by hand as we don't want to depend
775	* on the compiler to always get this right (gcc generally doesn't).
776	* Real recursion would eat up our stack space.
777	*/
778
779	/*
780	* dput - release a dentry
781	* @dentry: dentry to release
782	*
783	* Release a dentry. This will drop the usage count and if appropriate
784	* call the dentry unlink method as well as removing it from the queues and
785	* releasing its resources. If the parent dentries were scheduled for release
786	* they too may now get deleted.
787	*/
788	void dput(struct dentry *dentry)
789	{
790	if (unlikely(!dentry))
791	return;
792
793	repeat:
794	might_sleep();
795
796	rcu_read_lock();
797	if (likely(fast_dput(dentry))) {
798	rcu_read_unlock();
799	return;
800	}
801
802	/* Slow case: now with the dentry lock held */
803	rcu_read_unlock();
804
805	WARN_ON(d_in_lookup(dentry));
806
807	/* Unreachable? Get rid of it */
808	if (unlikely(d_unhashed(dentry)))
809	goto kill_it;
810
811	if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
812	goto kill_it;
813
814	if (unlikely(dentry->d_flags & DCACHE_OP_DELETE)) {
815	if (dentry->d_op->d_delete(dentry))
816	goto kill_it;
817	}
818
819	if (!(dentry->d_flags & DCACHE_REFERENCED))
820	dentry->d_flags |= DCACHE_REFERENCED;
821	dentry_lru_add(dentry);
822
823	dentry->d_lockref.count--;
824	spin_unlock(&dentry->d_lock);
825	return;
826
827	kill_it:
828	dentry = dentry_kill(dentry);
829	if (dentry) {
830	cond_resched();
831	goto repeat;
832	}
833	}
834	EXPORT_SYMBOL(dput);
835
836
837	/* This must be called with d_lock held */
838	static inline void __dget_dlock(struct dentry *dentry)
839	{
840	dentry->d_lockref.count++;
841	}
842
843	static inline void __dget(struct dentry *dentry)
844	{
845	lockref_get(&dentry->d_lockref);
846	}
847
848	struct dentry *dget_parent(struct dentry *dentry)
849	{
850	int gotref;
851	struct dentry *ret;
852
853	/*
854	* Do optimistic parent lookup without any
855	* locking.
856	*/
857	rcu_read_lock();
858	ret = ACCESS_ONCE(dentry->d_parent);
859	gotref = lockref_get_not_zero(&ret->d_lockref);
860	rcu_read_unlock();
861	if (likely(gotref)) {
862	if (likely(ret == ACCESS_ONCE(dentry->d_parent)))
863	return ret;
864	dput(ret);
865	}
866
867	repeat:
868	/*
869	* Don't need rcu_dereference because we re-check it was correct under
870	* the lock.
871	*/
872	rcu_read_lock();
873	ret = dentry->d_parent;
874	spin_lock(&ret->d_lock);
875	if (unlikely(ret != dentry->d_parent)) {
876	spin_unlock(&ret->d_lock);
877	rcu_read_unlock();
878	goto repeat;
879	}
880	rcu_read_unlock();
881	BUG_ON(!ret->d_lockref.count);
882	ret->d_lockref.count++;
883	spin_unlock(&ret->d_lock);
884	return ret;
885	}
886	EXPORT_SYMBOL(dget_parent);
887
888	/**
889	* d_find_alias - grab a hashed alias of inode
890	* @inode: inode in question
891	*
892	* If inode has a hashed alias, or is a directory and has any alias,
893	* acquire the reference to alias and return it. Otherwise return NULL.
894	* Notice that if inode is a directory there can be only one alias and
895	* it can be unhashed only if it has no children, or if it is the root
896	* of a filesystem, or if the directory was renamed and d_revalidate
897	* was the first vfs operation to notice.
898	*
899	* If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
900	* any other hashed alias over that one.
901	*/
902	static struct dentry *__d_find_alias(struct inode *inode)
903	{
904	struct dentry *alias, *discon_alias;
905
906	again:
907	discon_alias = NULL;
908	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
909	spin_lock(&alias->d_lock);
910	if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
911	if (IS_ROOT(alias) &&
912	(alias->d_flags & DCACHE_DISCONNECTED)) {
913	discon_alias = alias;
914	} else {
915	__dget_dlock(alias);
916	spin_unlock(&alias->d_lock);
917	return alias;
918	}
919	}
920	spin_unlock(&alias->d_lock);
921	}
922	if (discon_alias) {
923	alias = discon_alias;
924	spin_lock(&alias->d_lock);
925	if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
926	__dget_dlock(alias);
927	spin_unlock(&alias->d_lock);
928	return alias;
929	}
930	spin_unlock(&alias->d_lock);
931	goto again;
932	}
933	return NULL;
934	}
935
936	struct dentry *d_find_alias(struct inode *inode)
937	{
938	struct dentry *de = NULL;
939
940	if (!hlist_empty(&inode->i_dentry)) {
941	spin_lock(&inode->i_lock);
942	de = __d_find_alias(inode);
943	spin_unlock(&inode->i_lock);
944	}
945	return de;
946	}
947	EXPORT_SYMBOL(d_find_alias);
948
949	/*
950	* Try to kill dentries associated with this inode.
951	* WARNING: you must own a reference to inode.
952	*/
953	void d_prune_aliases(struct inode *inode)
954	{
955	struct dentry *dentry;
956	restart:
957	spin_lock(&inode->i_lock);
958	hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) {
959	spin_lock(&dentry->d_lock);
960	if (!dentry->d_lockref.count) {
961	struct dentry *parent = lock_parent(dentry);
962	if (likely(!dentry->d_lockref.count)) {
963	__dentry_kill(dentry);
964	dput(parent);
965	goto restart;
966	}
967	if (parent)
968	spin_unlock(&parent->d_lock);
969	}
970	spin_unlock(&dentry->d_lock);
971	}
972	spin_unlock(&inode->i_lock);
973	}
974	EXPORT_SYMBOL(d_prune_aliases);
975
976	static void shrink_dentry_list(struct list_head *list)
977	{
978	struct dentry *dentry, *parent;
979
980	while (!list_empty(list)) {
981	struct inode *inode;
982	dentry = list_entry(list->prev, struct dentry, d_lru);
983	spin_lock(&dentry->d_lock);
984	parent = lock_parent(dentry);
985
986	/*
987	* The dispose list is isolated and dentries are not accounted
988	* to the LRU here, so we can simply remove it from the list
989	* here regardless of whether it is referenced or not.
990	*/
991	d_shrink_del(dentry);
992
993	/*
994	* We found an inuse dentry which was not removed from
995	* the LRU because of laziness during lookup. Do not free it.
996	*/
997	if (dentry->d_lockref.count > 0) {
998	spin_unlock(&dentry->d_lock);
999	if (parent)
1000	spin_unlock(&parent->d_lock);
1001	continue;
1002	}
1003
1004
1005	if (unlikely(dentry->d_flags & DCACHE_DENTRY_KILLED)) {
1006	bool can_free = dentry->d_flags & DCACHE_MAY_FREE;
1007	spin_unlock(&dentry->d_lock);
1008	if (parent)
1009	spin_unlock(&parent->d_lock);
1010	if (can_free)
1011	dentry_free(dentry);
1012	continue;
1013	}
1014
1015	inode = dentry->d_inode;
1016	if (inode && unlikely(!spin_trylock(&inode->i_lock))) {
1017	d_shrink_add(dentry, list);
1018	spin_unlock(&dentry->d_lock);
1019	if (parent)
1020	spin_unlock(&parent->d_lock);
1021	continue;
1022	}
1023
1024	__dentry_kill(dentry);
1025
1026	/*
1027	* We need to prune ancestors too. This is necessary to prevent
1028	* quadratic behavior of shrink_dcache_parent(), but is also
1029	* expected to be beneficial in reducing dentry cache
1030	* fragmentation.
1031	*/
1032	dentry = parent;
1033	while (dentry && !lockref_put_or_lock(&dentry->d_lockref)) {
1034	parent = lock_parent(dentry);
1035	if (dentry->d_lockref.count != 1) {
1036	dentry->d_lockref.count--;
1037	spin_unlock(&dentry->d_lock);
1038	if (parent)
1039	spin_unlock(&parent->d_lock);
1040	break;
1041	}
1042	inode = dentry->d_inode; /* can't be NULL */
1043	if (unlikely(!spin_trylock(&inode->i_lock))) {
1044	spin_unlock(&dentry->d_lock);
1045	if (parent)
1046	spin_unlock(&parent->d_lock);
1047	cpu_relax();
1048	continue;
1049	}
1050	__dentry_kill(dentry);
1051	dentry = parent;
1052	}
1053	}
1054	}
1055
1056	static enum lru_status dentry_lru_isolate(struct list_head *item,
1057	struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1058	{
1059	struct list_head *freeable = arg;
1060	struct dentry *dentry = container_of(item, struct dentry, d_lru);
1061
1062
1063	/*
1064	* we are inverting the lru lock/dentry->d_lock here,
1065	* so use a trylock. If we fail to get the lock, just skip
1066	* it
1067	*/
1068	if (!spin_trylock(&dentry->d_lock))
1069	return LRU_SKIP;
1070
1071	/*
1072	* Referenced dentries are still in use. If they have active
1073	* counts, just remove them from the LRU. Otherwise give them
1074	* another pass through the LRU.
1075	*/
1076	if (dentry->d_lockref.count) {
1077	d_lru_isolate(lru, dentry);
1078	spin_unlock(&dentry->d_lock);
1079	return LRU_REMOVED;
1080	}
1081
1082	if (dentry->d_flags & DCACHE_REFERENCED) {
1083	dentry->d_flags &= ~DCACHE_REFERENCED;
1084	spin_unlock(&dentry->d_lock);
1085
1086	/*
1087	* The list move itself will be made by the common LRU code. At
1088	* this point, we've dropped the dentry->d_lock but keep the
1089	* lru lock. This is safe to do, since every list movement is
1090	* protected by the lru lock even if both locks are held.
1091	*
1092	* This is guaranteed by the fact that all LRU management
1093	* functions are intermediated by the LRU API calls like
1094	* list_lru_add and list_lru_del. List movement in this file
1095	* only ever occur through this functions or through callbacks
1096	* like this one, that are called from the LRU API.
1097	*
1098	* The only exceptions to this are functions like
1099	* shrink_dentry_list, and code that first checks for the
1100	* DCACHE_SHRINK_LIST flag. Those are guaranteed to be
1101	* operating only with stack provided lists after they are
1102	* properly isolated from the main list. It is thus, always a
1103	* local access.
1104	*/
1105	return LRU_ROTATE;
1106	}
1107
1108	d_lru_shrink_move(lru, dentry, freeable);
1109	spin_unlock(&dentry->d_lock);
1110
1111	return LRU_REMOVED;
1112	}
1113
1114	/**
1115	* prune_dcache_sb - shrink the dcache
1116	* @sb: superblock
1117	* @sc: shrink control, passed to list_lru_shrink_walk()
1118	*
1119	* Attempt to shrink the superblock dcache LRU by @sc->nr_to_scan entries. This
1120	* is done when we need more memory and called from the superblock shrinker
1121	* function.
1122	*
1123	* This function may fail to free any resources if all the dentries are in
1124	* use.
1125	*/
1126	long prune_dcache_sb(struct super_block *sb, struct shrink_control *sc)
1127	{
1128	LIST_HEAD(dispose);
1129	long freed;
1130
1131	freed = list_lru_shrink_walk(&sb->s_dentry_lru, sc,
1132	dentry_lru_isolate, &dispose);
1133	shrink_dentry_list(&dispose);
1134	return freed;
1135	}
1136
1137	static enum lru_status dentry_lru_isolate_shrink(struct list_head *item,
1138	struct list_lru_one *lru, spinlock_t *lru_lock, void *arg)
1139	{
1140	struct list_head *freeable = arg;
1141	struct dentry *dentry = container_of(item, struct dentry, d_lru);
1142
1143	/*
1144	* we are inverting the lru lock/dentry->d_lock here,
1145	* so use a trylock. If we fail to get the lock, just skip
1146	* it
1147	*/
1148	if (!spin_trylock(&dentry->d_lock))
1149	return LRU_SKIP;
1150
1151	d_lru_shrink_move(lru, dentry, freeable);
1152	spin_unlock(&dentry->d_lock);
1153
1154	return LRU_REMOVED;
1155	}
1156
1157
1158	/**
1159	* shrink_dcache_sb - shrink dcache for a superblock
1160	* @sb: superblock
1161	*
1162	* Shrink the dcache for the specified super block. This is used to free
1163	* the dcache before unmounting a file system.
1164	*/
1165	void shrink_dcache_sb(struct super_block *sb)
1166	{
1167	do {
1168	LIST_HEAD(dispose);
1169
1170	list_lru_walk(&sb->s_dentry_lru,
1171	dentry_lru_isolate_shrink, &dispose, 1024);
1172	shrink_dentry_list(&dispose);
1173	cond_resched();
1174	} while (list_lru_count(&sb->s_dentry_lru) > 0);
1175	}
1176	EXPORT_SYMBOL(shrink_dcache_sb);
1177
1178	/**
1179	* enum d_walk_ret - action to talke during tree walk
1180	* @D_WALK_CONTINUE: contrinue walk
1181	* @D_WALK_QUIT: quit walk
1182	* @D_WALK_NORETRY: quit when retry is needed
1183	* @D_WALK_SKIP: skip this dentry and its children
1184	*/
1185	enum d_walk_ret {
1186	D_WALK_CONTINUE,
1187	D_WALK_QUIT,
1188	D_WALK_NORETRY,
1189	D_WALK_SKIP,
1190	};
1191
1192	/**
1193	* d_walk - walk the dentry tree
1194	* @parent: start of walk
1195	* @data: data passed to @enter() and @finish()
1196	* @enter: callback when first entering the dentry
1197	* @finish: callback when successfully finished the walk
1198	*
1199	* The @enter() and @finish() callbacks are called with d_lock held.
1200	*/
1201	static void d_walk(struct dentry *parent, void *data,
1202	enum d_walk_ret (*enter)(void *, struct dentry *),
1203	void (*finish)(void *))
1204	{
1205	struct dentry *this_parent;
1206	struct list_head *next;
1207	unsigned seq = 0;
1208	enum d_walk_ret ret;
1209	bool retry = true;
1210
1211	again:
1212	read_seqbegin_or_lock(&rename_lock, &seq);
1213	this_parent = parent;
1214	spin_lock(&this_parent->d_lock);
1215
1216	ret = enter(data, this_parent);
1217	switch (ret) {
1218	case D_WALK_CONTINUE:
1219	break;
1220	case D_WALK_QUIT:
1221	case D_WALK_SKIP:
1222	goto out_unlock;
1223	case D_WALK_NORETRY:
1224	retry = false;
1225	break;
1226	}
1227	repeat:
1228	next = this_parent->d_subdirs.next;
1229	resume:
1230	while (next != &this_parent->d_subdirs) {
1231	struct list_head *tmp = next;
1232	struct dentry *dentry = list_entry(tmp, struct dentry, d_child);
1233	next = tmp->next;
1234
1235	if (unlikely(dentry->d_flags & DCACHE_DENTRY_CURSOR))
1236	continue;
1237
1238	spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1239
1240	ret = enter(data, dentry);
1241	switch (ret) {
1242	case D_WALK_CONTINUE:
1243	break;
1244	case D_WALK_QUIT:
1245	spin_unlock(&dentry->d_lock);
1246	goto out_unlock;
1247	case D_WALK_NORETRY:
1248	retry = false;
1249	break;
1250	case D_WALK_SKIP:
1251	spin_unlock(&dentry->d_lock);
1252	continue;
1253	}
1254
1255	if (!list_empty(&dentry->d_subdirs)) {
1256	spin_unlock(&this_parent->d_lock);
1257	spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1258	this_parent = dentry;
1259	spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1260	goto repeat;
1261	}
1262	spin_unlock(&dentry->d_lock);
1263	}
1264	/*
1265	* All done at this level ... ascend and resume the search.
1266	*/
1267	rcu_read_lock();
1268	ascend:
1269	if (this_parent != parent) {
1270	struct dentry *child = this_parent;
1271	this_parent = child->d_parent;
1272
1273	spin_unlock(&child->d_lock);
1274	spin_lock(&this_parent->d_lock);
1275
1276	/* might go back up the wrong parent if we have had a rename. */
1277	if (need_seqretry(&rename_lock, seq))
1278	goto rename_retry;
1279	/* go into the first sibling still alive */
1280	do {
1281	next = child->d_child.next;
1282	if (next == &this_parent->d_subdirs)
1283	goto ascend;
1284	child = list_entry(next, struct dentry, d_child);
1285	} while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED));
1286	rcu_read_unlock();
1287	goto resume;
1288	}
1289	if (need_seqretry(&rename_lock, seq))
1290	goto rename_retry;
1291	rcu_read_unlock();
1292	if (finish)
1293	finish(data);
1294
1295	out_unlock:
1296	spin_unlock(&this_parent->d_lock);
1297	done_seqretry(&rename_lock, seq);
1298	return;
1299
1300	rename_retry:
1301	spin_unlock(&this_parent->d_lock);
1302	rcu_read_unlock();
1303	BUG_ON(seq & 1);
1304	if (!retry)
1305	return;
1306	seq = 1;
1307	goto again;
1308	}
1309
1310	/*
1311	* Search for at least 1 mount point in the dentry's subdirs.
1312	* We descend to the next level whenever the d_subdirs
1313	* list is non-empty and continue searching.
1314	*/
1315
1316	static enum d_walk_ret check_mount(void *data, struct dentry *dentry)
1317	{
1318	int *ret = data;
1319	if (d_mountpoint(dentry)) {
1320	*ret = 1;
1321	return D_WALK_QUIT;
1322	}
1323	return D_WALK_CONTINUE;
1324	}
1325
1326	/**
1327	* have_submounts - check for mounts over a dentry
1328	* @parent: dentry to check.
1329	*
1330	* Return true if the parent or its subdirectories contain
1331	* a mount point
1332	*/
1333	int have_submounts(struct dentry *parent)
1334	{
1335	int ret = 0;
1336
1337	d_walk(parent, &ret, check_mount, NULL);
1338
1339	return ret;
1340	}
1341	EXPORT_SYMBOL(have_submounts);
1342
1343	/*
1344	* Called by mount code to set a mountpoint and check if the mountpoint is
1345	* reachable (e.g. NFS can unhash a directory dentry and then the complete
1346	* subtree can become unreachable).
1347	*
1348	* Only one of d_invalidate() and d_set_mounted() must succeed. For
1349	* this reason take rename_lock and d_lock on dentry and ancestors.
1350	*/
1351	int d_set_mounted(struct dentry *dentry)
1352	{
1353	struct dentry *p;
1354	int ret = -ENOENT;
1355	write_seqlock(&rename_lock);
1356	for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) {
1357	/* Need exclusion wrt. d_invalidate() */
1358	spin_lock(&p->d_lock);
1359	if (unlikely(d_unhashed(p))) {
1360	spin_unlock(&p->d_lock);
1361	goto out;
1362	}
1363	spin_unlock(&p->d_lock);
1364	}
1365	spin_lock(&dentry->d_lock);
1366	if (!d_unlinked(dentry)) {
1367	ret = -EBUSY;
1368	if (!d_mountpoint(dentry)) {
1369	dentry->d_flags |= DCACHE_MOUNTED;
1370	ret = 0;
1371	}
1372	}
1373	spin_unlock(&dentry->d_lock);
1374	out:
1375	write_sequnlock(&rename_lock);
1376	return ret;
1377	}
1378
1379	/*
1380	* Search the dentry child list of the specified parent,
1381	* and move any unused dentries to the end of the unused
1382	* list for prune_dcache(). We descend to the next level
1383	* whenever the d_subdirs list is non-empty and continue
1384	* searching.
1385	*
1386	* It returns zero iff there are no unused children,
1387	* otherwise it returns the number of children moved to
1388	* the end of the unused list. This may not be the total
1389	* number of unused children, because select_parent can
1390	* drop the lock and return early due to latency
1391	* constraints.
1392	*/
1393
1394	struct select_data {
1395	struct dentry *start;
1396	struct list_head dispose;
1397	int found;
1398	};
1399
1400	static enum d_walk_ret select_collect(void *_data, struct dentry *dentry)
1401	{
1402	struct select_data *data = _data;
1403	enum d_walk_ret ret = D_WALK_CONTINUE;
1404
1405	if (data->start == dentry)
1406	goto out;
1407
1408	if (dentry->d_flags & DCACHE_SHRINK_LIST) {
1409	data->found++;
1410	} else {
1411	if (dentry->d_flags & DCACHE_LRU_LIST)
1412	d_lru_del(dentry);
1413	if (!dentry->d_lockref.count) {
1414	d_shrink_add(dentry, &data->dispose);
1415	data->found++;
1416	}
1417	}
1418	/*
1419	* We can return to the caller if we have found some (this
1420	* ensures forward progress). We'll be coming back to find
1421	* the rest.
1422	*/
1423	if (!list_empty(&data->dispose))
1424	ret = need_resched() ? D_WALK_QUIT : D_WALK_NORETRY;
1425	out:
1426	return ret;
1427	}
1428
1429	/**
1430	* shrink_dcache_parent - prune dcache
1431	* @parent: parent of entries to prune
1432	*
1433	* Prune the dcache to remove unused children of the parent dentry.
1434	*/
1435	void shrink_dcache_parent(struct dentry *parent)
1436	{
1437	for (;;) {
1438	struct select_data data;
1439
1440	INIT_LIST_HEAD(&data.dispose);
1441	data.start = parent;
1442	data.found = 0;
1443
1444	d_walk(parent, &data, select_collect, NULL);
1445	if (!data.found)
1446	break;
1447
1448	shrink_dentry_list(&data.dispose);
1449	cond_resched();
1450	}
1451	}
1452	EXPORT_SYMBOL(shrink_dcache_parent);
1453
1454	static enum d_walk_ret umount_check(void *_data, struct dentry *dentry)
1455	{
1456	/* it has busy descendents; complain about those instead */
1457	if (!list_empty(&dentry->d_subdirs))
1458	return D_WALK_CONTINUE;
1459
1460	/* root with refcount 1 is fine */
1461	if (dentry == _data && dentry->d_lockref.count == 1)
1462	return D_WALK_CONTINUE;
1463
1464	printk(KERN_ERR "BUG: Dentry %p{i=%lx,n=%pd} "
1465	" still in use (%d) [unmount of %s %s]\n",
1466	dentry,
1467	dentry->d_inode ?
1468	dentry->d_inode->i_ino : 0UL,
1469	dentry,
1470	dentry->d_lockref.count,
1471	dentry->d_sb->s_type->name,
1472	dentry->d_sb->s_id);
1473	WARN_ON(1);
1474	return D_WALK_CONTINUE;
1475	}
1476
1477	static void do_one_tree(struct dentry *dentry)
1478	{
1479	shrink_dcache_parent(dentry);
1480	d_walk(dentry, dentry, umount_check, NULL);
1481	d_drop(dentry);
1482	dput(dentry);
1483	}
1484
1485	/*
1486	* destroy the dentries attached to a superblock on unmounting
1487	*/
1488	void shrink_dcache_for_umount(struct super_block *sb)
1489	{
1490	struct dentry *dentry;
1491
1492	WARN(down_read_trylock(&sb->s_umount), "s_umount should've been locked");
1493
1494	dentry = sb->s_root;
1495	sb->s_root = NULL;
1496	do_one_tree(dentry);
1497
1498	while (!hlist_bl_empty(&sb->s_anon)) {
1499	dentry = dget(hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash));
1500	do_one_tree(dentry);
1501	}
1502	}
1503
1504	struct detach_data {
1505	struct select_data select;
1506	struct dentry *mountpoint;
1507	};
1508	static enum d_walk_ret detach_and_collect(void *_data, struct dentry *dentry)
1509	{
1510	struct detach_data *data = _data;
1511
1512	if (d_mountpoint(dentry)) {
1513	__dget_dlock(dentry);
1514	data->mountpoint = dentry;
1515	return D_WALK_QUIT;
1516	}
1517
1518	return select_collect(&data->select, dentry);
1519	}
1520
1521	static void check_and_drop(void *_data)
1522	{
1523	struct detach_data *data = _data;
1524
1525	if (!data->mountpoint && list_empty(&data->select.dispose))
1526	__d_drop(data->select.start);
1527	}
1528
1529	/**
1530	* d_invalidate - detach submounts, prune dcache, and drop
1531	* @dentry: dentry to invalidate (aka detach, prune and drop)
1532	*
1533	* no dcache lock.
1534	*
1535	* The final d_drop is done as an atomic operation relative to
1536	* rename_lock ensuring there are no races with d_set_mounted. This
1537	* ensures there are no unhashed dentries on the path to a mountpoint.
1538	*/
1539	void d_invalidate(struct dentry *dentry)
1540	{
1541	/*
1542	* If it's already been dropped, return OK.
1543	*/
1544	spin_lock(&dentry->d_lock);
1545	if (d_unhashed(dentry)) {
1546	spin_unlock(&dentry->d_lock);
1547	return;
1548	}
1549	spin_unlock(&dentry->d_lock);
1550
1551	/* Negative dentries can be dropped without further checks */
1552	if (!dentry->d_inode) {
1553	d_drop(dentry);
1554	return;
1555	}
1556
1557	for (;;) {
1558	struct detach_data data;
1559
1560	data.mountpoint = NULL;
1561	INIT_LIST_HEAD(&data.select.dispose);
1562	data.select.start = dentry;
1563	data.select.found = 0;
1564
1565	d_walk(dentry, &data, detach_and_collect, check_and_drop);
1566
1567	if (!list_empty(&data.select.dispose))
1568	shrink_dentry_list(&data.select.dispose);
1569	else if (!data.mountpoint)
1570	return;
1571
1572	if (data.mountpoint) {
1573	detach_mounts(data.mountpoint);
1574	dput(data.mountpoint);
1575	}
1576	cond_resched();
1577	}
1578	}
1579	EXPORT_SYMBOL(d_invalidate);
1580
1581	/**
1582	* __d_alloc - allocate a dcache entry
1583	* @sb: filesystem it will belong to
1584	* @name: qstr of the name
1585	*
1586	* Allocates a dentry. It returns %NULL if there is insufficient memory
1587	* available. On a success the dentry is returned. The name passed in is
1588	* copied and the copy passed in may be reused after this call.
1589	*/
1590
1591	struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1592	{
1593	struct dentry *dentry;
1594	char *dname;
1595	int err;
1596
1597	dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1598	if (!dentry)
1599	return NULL;
1600
1601	/*
1602	* We guarantee that the inline name is always NUL-terminated.
1603	* This way the memcpy() done by the name switching in rename
1604	* will still always have a NUL at the end, even if we might
1605	* be overwriting an internal NUL character
1606	*/
1607	dentry->d_iname[DNAME_INLINE_LEN-1] = 0;
1608	if (unlikely(!name)) {
1609	static const struct qstr anon = QSTR_INIT("/", 1);
1610	name = &anon;
1611	dname = dentry->d_iname;
1612	} else if (name->len > DNAME_INLINE_LEN-1) {
1613	size_t size = offsetof(struct external_name, name[1]);
1614	struct external_name *p = kmalloc(size + name->len,
1615	GFP_KERNEL_ACCOUNT);
1616	if (!p) {
1617	kmem_cache_free(dentry_cache, dentry);
1618	return NULL;
1619	}
1620	atomic_set(&p->u.count, 1);
1621	dname = p->name;
1622	if (IS_ENABLED(CONFIG_DCACHE_WORD_ACCESS))
1623	kasan_unpoison_shadow(dname,
1624	round_up(name->len + 1, sizeof(unsigned long)));
1625	} else {
1626	dname = dentry->d_iname;
1627	}
1628
1629	dentry->d_name.len = name->len;
1630	dentry->d_name.hash = name->hash;
1631	memcpy(dname, name->name, name->len);
1632	dname[name->len] = 0;
1633
1634	/* Make sure we always see the terminating NUL character */
1635	smp_wmb();
1636	dentry->d_name.name = dname;
1637
1638	dentry->d_lockref.count = 1;
1639	dentry->d_flags = 0;
1640	spin_lock_init(&dentry->d_lock);
1641	seqcount_init(&dentry->d_seq);
1642	dentry->d_inode = NULL;
1643	dentry->d_parent = dentry;
1644	dentry->d_sb = sb;
1645	dentry->d_op = NULL;
1646	dentry->d_fsdata = NULL;
1647	INIT_HLIST_BL_NODE(&dentry->d_hash);
1648	INIT_LIST_HEAD(&dentry->d_lru);
1649	INIT_LIST_HEAD(&dentry->d_subdirs);
1650	INIT_HLIST_NODE(&dentry->d_u.d_alias);
1651	INIT_LIST_HEAD(&dentry->d_child);
1652	d_set_d_op(dentry, dentry->d_sb->s_d_op);
1653
1654	if (dentry->d_op && dentry->d_op->d_init) {
1655	err = dentry->d_op->d_init(dentry);
1656	if (err) {
1657	if (dname_external(dentry))
1658	kfree(external_name(dentry));
1659	kmem_cache_free(dentry_cache, dentry);
1660	return NULL;
1661	}
1662	}
1663
1664	this_cpu_inc(nr_dentry);
1665
1666	return dentry;
1667	}
1668
1669	/**
1670	* d_alloc - allocate a dcache entry
1671	* @parent: parent of entry to allocate
1672	* @name: qstr of the name
1673	*
1674	* Allocates a dentry. It returns %NULL if there is insufficient memory
1675	* available. On a success the dentry is returned. The name passed in is
1676	* copied and the copy passed in may be reused after this call.
1677	*/
1678	struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1679	{
1680	struct dentry *dentry = __d_alloc(parent->d_sb, name);
1681	if (!dentry)
1682	return NULL;
1683	dentry->d_flags |= DCACHE_RCUACCESS;
1684	spin_lock(&parent->d_lock);
1685	/*
1686	* don't need child lock because it is not subject
1687	* to concurrency here
1688	*/
1689	__dget_dlock(parent);
1690	dentry->d_parent = parent;
1691	list_add(&dentry->d_child, &parent->d_subdirs);
1692	spin_unlock(&parent->d_lock);
1693
1694	return dentry;
1695	}
1696	EXPORT_SYMBOL(d_alloc);
1697
1698	struct dentry *d_alloc_cursor(struct dentry * parent)
1699	{
1700	struct dentry *dentry = __d_alloc(parent->d_sb, NULL);
1701	if (dentry) {
1702	dentry->d_flags |= DCACHE_RCUACCESS | DCACHE_DENTRY_CURSOR;
1703	dentry->d_parent = dget(parent);
1704	}
1705	return dentry;
1706	}
1707
1708	/**
1709	* d_alloc_pseudo - allocate a dentry (for lookup-less filesystems)
1710	* @sb: the superblock
1711	* @name: qstr of the name
1712	*
1713	* For a filesystem that just pins its dentries in memory and never
1714	* performs lookups at all, return an unhashed IS_ROOT dentry.
1715	*/
1716	struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1717	{
1718	return __d_alloc(sb, name);
1719	}
1720	EXPORT_SYMBOL(d_alloc_pseudo);
1721
1722	struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1723	{
1724	struct qstr q;
1725
1726	q.name = name;
1727	q.hash_len = hashlen_string(parent, name);
1728	return d_alloc(parent, &q);
1729	}
1730	EXPORT_SYMBOL(d_alloc_name);
1731
1732	void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1733	{
1734	WARN_ON_ONCE(dentry->d_op);
1735	WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1736	DCACHE_OP_COMPARE |
1737	DCACHE_OP_REVALIDATE |
1738	DCACHE_OP_WEAK_REVALIDATE |
1739	DCACHE_OP_DELETE |
1740	DCACHE_OP_REAL));
1741	dentry->d_op = op;
1742	if (!op)
1743	return;
1744	if (op->d_hash)
1745	dentry->d_flags |= DCACHE_OP_HASH;
1746	if (op->d_compare)
1747	dentry->d_flags |= DCACHE_OP_COMPARE;
1748	if (op->d_revalidate)
1749	dentry->d_flags |= DCACHE_OP_REVALIDATE;
1750	if (op->d_weak_revalidate)
1751	dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE;
1752	if (op->d_delete)
1753	dentry->d_flags |= DCACHE_OP_DELETE;
1754	if (op->d_prune)
1755	dentry->d_flags |= DCACHE_OP_PRUNE;
1756	if (op->d_real)
1757	dentry->d_flags |= DCACHE_OP_REAL;
1758
1759	}
1760	EXPORT_SYMBOL(d_set_d_op);
1761
1762
1763	/*
1764	* d_set_fallthru - Mark a dentry as falling through to a lower layer
1765	* @dentry - The dentry to mark
1766	*
1767	* Mark a dentry as falling through to the lower layer (as set with
1768	* d_pin_lower()). This flag may be recorded on the medium.
1769	*/
1770	void d_set_fallthru(struct dentry *dentry)
1771	{
1772	spin_lock(&dentry->d_lock);
1773	dentry->d_flags |= DCACHE_FALLTHRU;
1774	spin_unlock(&dentry->d_lock);
1775	}
1776	EXPORT_SYMBOL(d_set_fallthru);
1777
1778	static unsigned d_flags_for_inode(struct inode *inode)
1779	{
1780	unsigned add_flags = DCACHE_REGULAR_TYPE;
1781
1782	if (!inode)
1783	return DCACHE_MISS_TYPE;
1784
1785	if (S_ISDIR(inode->i_mode)) {
1786	add_flags = DCACHE_DIRECTORY_TYPE;
1787	if (unlikely(!(inode->i_opflags & IOP_LOOKUP))) {
1788	if (unlikely(!inode->i_op->lookup))
1789	add_flags = DCACHE_AUTODIR_TYPE;
1790	else
1791	inode->i_opflags |= IOP_LOOKUP;
1792	}
1793	goto type_determined;
1794	}
1795
1796	if (unlikely(!(inode->i_opflags & IOP_NOFOLLOW))) {
1797	if (unlikely(inode->i_op->get_link)) {
1798	add_flags = DCACHE_SYMLINK_TYPE;
1799	goto type_determined;
1800	}
1801	inode->i_opflags |= IOP_NOFOLLOW;
1802	}
1803
1804	if (unlikely(!S_ISREG(inode->i_mode)))
1805	add_flags = DCACHE_SPECIAL_TYPE;
1806
1807	type_determined:
1808	if (unlikely(IS_AUTOMOUNT(inode)))
1809	add_flags |= DCACHE_NEED_AUTOMOUNT;
1810	return add_flags;
1811	}
1812
1813	static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1814	{
1815	unsigned add_flags = d_flags_for_inode(inode);
1816	WARN_ON(d_in_lookup(dentry));
1817
1818	spin_lock(&dentry->d_lock);
1819	hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
1820	raw_write_seqcount_begin(&dentry->d_seq);
1821	__d_set_inode_and_type(dentry, inode, add_flags);
1822	raw_write_seqcount_end(&dentry->d_seq);
1823	fsnotify_update_flags(dentry);
1824	spin_unlock(&dentry->d_lock);
1825	}
1826
1827	/**
1828	* d_instantiate - fill in inode information for a dentry
1829	* @entry: dentry to complete
1830	* @inode: inode to attach to this dentry
1831	*
1832	* Fill in inode information in the entry.
1833	*
1834	* This turns negative dentries into productive full members
1835	* of society.
1836	*
1837	* NOTE! This assumes that the inode count has been incremented
1838	* (or otherwise set) by the caller to indicate that it is now
1839	* in use by the dcache.
1840	*/
1841
1842	void d_instantiate(struct dentry *entry, struct inode * inode)
1843	{
1844	BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1845	if (inode) {
1846	security_d_instantiate(entry, inode);
1847	spin_lock(&inode->i_lock);
1848	__d_instantiate(entry, inode);
1849	spin_unlock(&inode->i_lock);
1850	}
1851	}
1852	EXPORT_SYMBOL(d_instantiate);
1853
1854	/*
1855	* This should be equivalent to d_instantiate() + unlock_new_inode(),
1856	* with lockdep-related part of unlock_new_inode() done before
1857	* anything else. Use that instead of open-coding d_instantiate()/
1858	* unlock_new_inode() combinations.
1859	*/
1860	void d_instantiate_new(struct dentry *entry, struct inode *inode)
1861	{
1862	BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1863	BUG_ON(!inode);
1864	lockdep_annotate_inode_mutex_key(inode);
1865	security_d_instantiate(entry, inode);
1866	spin_lock(&inode->i_lock);
1867	__d_instantiate(entry, inode);
1868	WARN_ON(!(inode->i_state & I_NEW));
1869	inode->i_state &= ~I_NEW;
1870	smp_mb();
1871	wake_up_bit(&inode->i_state, __I_NEW);
1872	spin_unlock(&inode->i_lock);
1873	}
1874	EXPORT_SYMBOL(d_instantiate_new);
1875
1876	/**
1877	* d_instantiate_no_diralias - instantiate a non-aliased dentry
1878	* @entry: dentry to complete
1879	* @inode: inode to attach to this dentry
1880	*
1881	* Fill in inode information in the entry. If a directory alias is found, then
1882	* return an error (and drop inode). Together with d_materialise_unique() this
1883	* guarantees that a directory inode may never have more than one alias.
1884	*/
1885	int d_instantiate_no_diralias(struct dentry *entry, struct inode *inode)
1886	{
1887	BUG_ON(!hlist_unhashed(&entry->d_u.d_alias));
1888
1889	security_d_instantiate(entry, inode);
1890	spin_lock(&inode->i_lock);
1891	if (S_ISDIR(inode->i_mode) && !hlist_empty(&inode->i_dentry)) {
1892	spin_unlock(&inode->i_lock);
1893	iput(inode);
1894	return -EBUSY;
1895	}
1896	__d_instantiate(entry, inode);
1897	spin_unlock(&inode->i_lock);
1898
1899	return 0;
1900	}
1901	EXPORT_SYMBOL(d_instantiate_no_diralias);
1902
1903	struct dentry *d_make_root(struct inode *root_inode)
1904	{
1905	struct dentry *res = NULL;
1906
1907	if (root_inode) {
1908	res = __d_alloc(root_inode->i_sb, NULL);
1909	if (res) {
1910	res->d_flags |= DCACHE_RCUACCESS;
1911	d_instantiate(res, root_inode);
1912	} else {
1913	iput(root_inode);
1914	}
1915	}
1916	return res;
1917	}
1918	EXPORT_SYMBOL(d_make_root);
1919
1920	static struct dentry * __d_find_any_alias(struct inode *inode)
1921	{
1922	struct dentry *alias;
1923
1924	if (hlist_empty(&inode->i_dentry))
1925	return NULL;
1926	alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias);
1927	__dget(alias);
1928	return alias;
1929	}
1930
1931	/**
1932	* d_find_any_alias - find any alias for a given inode
1933	* @inode: inode to find an alias for
1934	*
1935	* If any aliases exist for the given inode, take and return a
1936	* reference for one of them. If no aliases exist, return %NULL.
1937	*/
1938	struct dentry *d_find_any_alias(struct inode *inode)
1939	{
1940	struct dentry *de;
1941
1942	spin_lock(&inode->i_lock);
1943	de = __d_find_any_alias(inode);
1944	spin_unlock(&inode->i_lock);
1945	return de;
1946	}
1947	EXPORT_SYMBOL(d_find_any_alias);
1948
1949	static struct dentry *__d_obtain_alias(struct inode *inode, int disconnected)
1950	{
1951	struct dentry *tmp;
1952	struct dentry *res;
1953	unsigned add_flags;
1954
1955	if (!inode)
1956	return ERR_PTR(-ESTALE);
1957	if (IS_ERR(inode))
1958	return ERR_CAST(inode);
1959
1960	res = d_find_any_alias(inode);
1961	if (res)
1962	goto out_iput;
1963
1964	tmp = __d_alloc(inode->i_sb, NULL);
1965	if (!tmp) {
1966	res = ERR_PTR(-ENOMEM);
1967	goto out_iput;
1968	}
1969
1970	security_d_instantiate(tmp, inode);
1971	spin_lock(&inode->i_lock);
1972	res = __d_find_any_alias(inode);
1973	if (res) {
1974	spin_unlock(&inode->i_lock);
1975	dput(tmp);
1976	goto out_iput;
1977	}
1978
1979	/* attach a disconnected dentry */
1980	add_flags = d_flags_for_inode(inode);
1981
1982	if (disconnected)
1983	add_flags |= DCACHE_DISCONNECTED;
1984
1985	spin_lock(&tmp->d_lock);
1986	__d_set_inode_and_type(tmp, inode, add_flags);
1987	hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry);
1988	hlist_bl_lock(&tmp->d_sb->s_anon);
1989	hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1990	hlist_bl_unlock(&tmp->d_sb->s_anon);
1991	spin_unlock(&tmp->d_lock);
1992	spin_unlock(&inode->i_lock);
1993
1994	return tmp;
1995
1996	out_iput:
1997	iput(inode);
1998	return res;
1999	}
2000
2001	/**
2002	* d_obtain_alias - find or allocate a DISCONNECTED dentry for a given inode
2003	* @inode: inode to allocate the dentry for
2004	*
2005	* Obtain a dentry for an inode resulting from NFS filehandle conversion or
2006	* similar open by handle operations. The returned dentry may be anonymous,
2007	* or may have a full name (if the inode was already in the cache).
2008	*
2009	* When called on a directory inode, we must ensure that the inode only ever
2010	* has one dentry. If a dentry is found, that is returned instead of
2011	* allocating a new one.
2012	*
2013	* On successful return, the reference to the inode has been transferred
2014	* to the dentry. In case of an error the reference on the inode is released.
2015	* To make it easier to use in export operations a %NULL or IS_ERR inode may
2016	* be passed in and the error will be propagated to the return value,
2017	* with a %NULL @inode replaced by ERR_PTR(-ESTALE).
2018	*/
2019	struct dentry *d_obtain_alias(struct inode *inode)
2020	{
2021	return __d_obtain_alias(inode, 1);
2022	}
2023	EXPORT_SYMBOL(d_obtain_alias);
2024
2025	/**
2026	* d_obtain_root - find or allocate a dentry for a given inode
2027	* @inode: inode to allocate the dentry for
2028	*
2029	* Obtain an IS_ROOT dentry for the root of a filesystem.
2030	*
2031	* We must ensure that directory inodes only ever have one dentry. If a
2032	* dentry is found, that is returned instead of allocating a new one.
2033	*
2034	* On successful return, the reference to the inode has been transferred
2035	* to the dentry. In case of an error the reference on the inode is
2036	* released. A %NULL or IS_ERR inode may be passed in and will be the
2037	* error will be propagate to the return value, with a %NULL @inode
2038	* replaced by ERR_PTR(-ESTALE).
2039	*/
2040	struct dentry *d_obtain_root(struct inode *inode)
2041	{
2042	return __d_obtain_alias(inode, 0);
2043	}
2044	EXPORT_SYMBOL(d_obtain_root);
2045
2046	/**
2047	* d_add_ci - lookup or allocate new dentry with case-exact name
2048	* @inode: the inode case-insensitive lookup has found
2049	* @dentry: the negative dentry that was passed to the parent's lookup func
2050	* @name: the case-exact name to be associated with the returned dentry
2051	*
2052	* This is to avoid filling the dcache with case-insensitive names to the
2053	* same inode, only the actual correct case is stored in the dcache for
2054	* case-insensitive filesystems.
2055	*
2056	* For a case-insensitive lookup match and if the the case-exact dentry
2057	* already exists in in the dcache, use it and return it.
2058	*
2059	* If no entry exists with the exact case name, allocate new dentry with
2060	* the exact case, and return the spliced entry.
2061	*/
2062	struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
2063	struct qstr *name)
2064	{
2065	struct dentry *found, *res;
2066
2067	/*
2068	* First check if a dentry matching the name already exists,
2069	* if not go ahead and create it now.
2070	*/
2071	found = d_hash_and_lookup(dentry->d_parent, name);
2072	if (found) {
2073	iput(inode);
2074	return found;
2075	}
2076	if (d_in_lookup(dentry)) {
2077	found = d_alloc_parallel(dentry->d_parent, name,
2078	dentry->d_wait);
2079	if (IS_ERR(found) || !d_in_lookup(found)) {
2080	iput(inode);
2081	return found;
2082	}
2083	} else {
2084	found = d_alloc(dentry->d_parent, name);
2085	if (!found) {
2086	iput(inode);
2087	return ERR_PTR(-ENOMEM);
2088	}
2089	}
2090	res = d_splice_alias(inode, found);
2091	if (res) {
2092	dput(found);
2093	return res;
2094	}
2095	return found;
2096	}
2097	EXPORT_SYMBOL(d_add_ci);
2098
2099
2100	static inline bool d_same_name(const struct dentry *dentry,
2101	const struct dentry *parent,
2102	const struct qstr *name)
2103	{
2104	if (likely(!(parent->d_flags & DCACHE_OP_COMPARE))) {
2105	if (dentry->d_name.len != name->len)
2106	return false;
2107	return dentry_cmp(dentry, name->name, name->len) == 0;
2108	}
2109	return parent->d_op->d_compare(dentry,
2110	dentry->d_name.len, dentry->d_name.name,
2111	name) == 0;
2112	}
2113
2114	/**
2115	* __d_lookup_rcu - search for a dentry (racy, store-free)
2116	* @parent: parent dentry
2117	* @name: qstr of name we wish to find
2118	* @seqp: returns d_seq value at the point where the dentry was found
2119	* Returns: dentry, or NULL
2120	*
2121	* __d_lookup_rcu is the dcache lookup function for rcu-walk name
2122	* resolution (store-free path walking) design described in
2123	* Documentation/filesystems/path-lookup.txt.
2124	*
2125	* This is not to be used outside core vfs.
2126	*
2127	* __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
2128	* held, and rcu_read_lock held. The returned dentry must not be stored into
2129	* without taking d_lock and checking d_seq sequence count against @seq
2130	* returned here.
2131	*
2132	* A refcount may be taken on the found dentry with the d_rcu_to_refcount
2133	* function.
2134	*
2135	* Alternatively, __d_lookup_rcu may be called again to look up the child of
2136	* the returned dentry, so long as its parent's seqlock is checked after the
2137	* child is looked up. Thus, an interlocking stepping of sequence lock checks
2138	* is formed, giving integrity down the path walk.
2139	*
2140	* NOTE! The caller *has* to check the resulting dentry against the sequence
2141	* number we've returned before using any of the resulting dentry state!
2142	*/
2143	struct dentry *__d_lookup_rcu(const struct dentry *parent,
2144	const struct qstr *name,
2145	unsigned *seqp)
2146	{
2147	u64 hashlen = name->hash_len;
2148	const unsigned char *str = name->name;
2149	struct hlist_bl_head *b = d_hash(hashlen_hash(hashlen));
2150	struct hlist_bl_node *node;
2151	struct dentry *dentry;
2152
2153	/*
2154	* Note: There is significant duplication with __d_lookup_rcu which is
2155	* required to prevent single threaded performance regressions
2156	* especially on architectures where smp_rmb (in seqcounts) are costly.
2157	* Keep the two functions in sync.
2158	*/
2159
2160	/*
2161	* The hash list is protected using RCU.
2162	*
2163	* Carefully use d_seq when comparing a candidate dentry, to avoid
2164	* races with d_move().
2165	*
2166	* It is possible that concurrent renames can mess up our list
2167	* walk here and result in missing our dentry, resulting in the
2168	* false-negative result. d_lookup() protects against concurrent
2169	* renames using rename_lock seqlock.
2170	*
2171	* See Documentation/filesystems/path-lookup.txt for more details.
2172	*/
2173	hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2174	unsigned seq;
2175
2176	seqretry:
2177	/*
2178	* The dentry sequence count protects us from concurrent
2179	* renames, and thus protects parent and name fields.
2180	*
2181	* The caller must perform a seqcount check in order
2182	* to do anything useful with the returned dentry.
2183	*
2184	* NOTE! We do a "raw" seqcount_begin here. That means that
2185	* we don't wait for the sequence count to stabilize if it
2186	* is in the middle of a sequence change. If we do the slow
2187	* dentry compare, we will do seqretries until it is stable,
2188	* and if we end up with a successful lookup, we actually
2189	* want to exit RCU lookup anyway.
2190	*
2191	* Note that raw_seqcount_begin still *does* smp_rmb(), so
2192	* we are still guaranteed NUL-termination of ->d_name.name.
2193	*/
2194	seq = raw_seqcount_begin(&dentry->d_seq);
2195	if (dentry->d_parent != parent)
2196	continue;
2197	if (d_unhashed(dentry))
2198	continue;
2199
2200	if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
2201	int tlen;
2202	const char *tname;
2203	if (dentry->d_name.hash != hashlen_hash(hashlen))
2204	continue;
2205	tlen = dentry->d_name.len;
2206	tname = dentry->d_name.name;
2207	/* we want a consistent (name,len) pair */
2208	if (read_seqcount_retry(&dentry->d_seq, seq)) {
2209	cpu_relax();
2210	goto seqretry;
2211	}
2212	if (parent->d_op->d_compare(dentry,
2213	tlen, tname, name) != 0)
2214	continue;
2215	} else {
2216	if (dentry->d_name.hash_len != hashlen)
2217	continue;
2218	if (dentry_cmp(dentry, str, hashlen_len(hashlen)) != 0)
2219	continue;
2220	}
2221	*seqp = seq;
2222	return dentry;
2223	}
2224	return NULL;
2225	}
2226
2227	/**
2228	* d_lookup - search for a dentry
2229	* @parent: parent dentry
2230	* @name: qstr of name we wish to find
2231	* Returns: dentry, or NULL
2232	*
2233	* d_lookup searches the children of the parent dentry for the name in
2234	* question. If the dentry is found its reference count is incremented and the
2235	* dentry is returned. The caller must use dput to free the entry when it has
2236	* finished using it. %NULL is returned if the dentry does not exist.
2237	*/
2238	struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name)
2239	{
2240	struct dentry *dentry;
2241	unsigned seq;
2242
2243	do {
2244	seq = read_seqbegin(&rename_lock);
2245	dentry = __d_lookup(parent, name);
2246	if (dentry)
2247	break;
2248	} while (read_seqretry(&rename_lock, seq));
2249	return dentry;
2250	}
2251	EXPORT_SYMBOL(d_lookup);
2252
2253	/**
2254	* __d_lookup - search for a dentry (racy)
2255	* @parent: parent dentry
2256	* @name: qstr of name we wish to find
2257	* Returns: dentry, or NULL
2258	*
2259	* __d_lookup is like d_lookup, however it may (rarely) return a
2260	* false-negative result due to unrelated rename activity.
2261	*
2262	* __d_lookup is slightly faster by avoiding rename_lock read seqlock,
2263	* however it must be used carefully, eg. with a following d_lookup in
2264	* the case of failure.
2265	*
2266	* __d_lookup callers must be commented.
2267	*/
2268	struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name)
2269	{
2270	unsigned int hash = name->hash;
2271	struct hlist_bl_head *b = d_hash(hash);
2272	struct hlist_bl_node *node;
2273	struct dentry *found = NULL;
2274	struct dentry *dentry;
2275
2276	/*
2277	* Note: There is significant duplication with __d_lookup_rcu which is
2278	* required to prevent single threaded performance regressions
2279	* especially on architectures where smp_rmb (in seqcounts) are costly.
2280	* Keep the two functions in sync.
2281	*/
2282
2283	/*
2284	* The hash list is protected using RCU.
2285	*
2286	* Take d_lock when comparing a candidate dentry, to avoid races
2287	* with d_move().
2288	*
2289	* It is possible that concurrent renames can mess up our list
2290	* walk here and result in missing our dentry, resulting in the
2291	* false-negative result. d_lookup() protects against concurrent
2292	* renames using rename_lock seqlock.
2293	*
2294	* See Documentation/filesystems/path-lookup.txt for more details.
2295	*/
2296	rcu_read_lock();
2297
2298	hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
2299
2300	if (dentry->d_name.hash != hash)
2301	continue;
2302
2303	spin_lock(&dentry->d_lock);
2304	if (dentry->d_parent != parent)
2305	goto next;
2306	if (d_unhashed(dentry))
2307	goto next;
2308
2309	if (!d_same_name(dentry, parent, name))
2310	goto next;
2311
2312	dentry->d_lockref.count++;
2313	found = dentry;
2314	spin_unlock(&dentry->d_lock);
2315	break;
2316	next:
2317	spin_unlock(&dentry->d_lock);
2318	}
2319	rcu_read_unlock();
2320
2321	return found;
2322	}
2323
2324	/**
2325	* d_hash_and_lookup - hash the qstr then search for a dentry
2326	* @dir: Directory to search in
2327	* @name: qstr of name we wish to find
2328	*
2329	* On lookup failure NULL is returned; on bad name - ERR_PTR(-error)
2330	*/
2331	struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
2332	{
2333	/*
2334	* Check for a fs-specific hash function. Note that we must
2335	* calculate the standard hash first, as the d_op->d_hash()
2336	* routine may choose to leave the hash value unchanged.
2337	*/
2338	name->hash = full_name_hash(dir, name->name, name->len);
2339	if (dir->d_flags & DCACHE_OP_HASH) {
2340	int err = dir->d_op->d_hash(dir, name);
2341	if (unlikely(err < 0))
2342	return ERR_PTR(err);
2343	}
2344	return d_lookup(dir, name);
2345	}
2346	EXPORT_SYMBOL(d_hash_and_lookup);
2347
2348	/*
2349	* When a file is deleted, we have two options:
2350	* - turn this dentry into a negative dentry
2351	* - unhash this dentry and free it.
2352	*
2353	* Usually, we want to just turn this into
2354	* a negative dentry, but if anybody else is
2355	* currently using the dentry or the inode
2356	* we can't do that and we fall back on removing
2357	* it from the hash queues and waiting for
2358	* it to be deleted later when it has no users
2359	*/
2360
2361	/**
2362	* d_delete - delete a dentry
2363	* @dentry: The dentry to delete
2364	*
2365	* Turn the dentry into a negative dentry if possible, otherwise
2366	* remove it from the hash queues so it can be deleted later
2367	*/
2368
2369	void d_delete(struct dentry * dentry)
2370	{
2371	struct inode *inode;
2372	int isdir = 0;
2373	/*
2374	* Are we the only user?
2375	*/
2376	again:
2377	spin_lock(&dentry->d_lock);
2378	inode = dentry->d_inode;
2379	isdir = S_ISDIR(inode->i_mode);
2380	if (dentry->d_lockref.count == 1) {
2381	if (!spin_trylock(&inode->i_lock)) {
2382	spin_unlock(&dentry->d_lock);
2383	cpu_relax();
2384	goto again;
2385	}
2386	dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2387	dentry_unlink_inode(dentry);
2388	fsnotify_nameremove(dentry, isdir);
2389	return;
2390	}
2391
2392	if (!d_unhashed(dentry))
2393	__d_drop(dentry);
2394
2395	spin_unlock(&dentry->d_lock);
2396
2397	fsnotify_nameremove(dentry, isdir);
2398	}
2399	EXPORT_SYMBOL(d_delete);
2400
2401	static void __d_rehash(struct dentry *entry)
2402	{
2403	struct hlist_bl_head *b = d_hash(entry->d_name.hash);
2404
2405	hlist_bl_lock(b);
2406	hlist_bl_add_head_rcu(&entry->d_hash, b);
2407	hlist_bl_unlock(b);
2408	}
2409
2410	/**
2411	* d_rehash - add an entry back to the hash
2412	* @entry: dentry to add to the hash
2413	*
2414	* Adds a dentry to the hash according to its name.
2415	*/
2416
2417	void d_rehash(struct dentry * entry)
2418	{
2419	spin_lock(&entry->d_lock);
2420	__d_rehash(entry);
2421	spin_unlock(&entry->d_lock);
2422	}
2423	EXPORT_SYMBOL(d_rehash);
2424
2425	static inline unsigned start_dir_add(struct inode *dir)
2426	{
2427
2428	for (;;) {
2429	unsigned n = dir->i_dir_seq;
2430	if (!(n & 1) && cmpxchg(&dir->i_dir_seq, n, n + 1) == n)
2431	return n;
2432	cpu_relax();
2433	}
2434	}
2435
2436	static inline void end_dir_add(struct inode *dir, unsigned n)
2437	{
2438	smp_store_release(&dir->i_dir_seq, n + 2);
2439	}
2440
2441	static void d_wait_lookup(struct dentry *dentry)
2442	{
2443	if (d_in_lookup(dentry)) {
2444	DECLARE_WAITQUEUE(wait, current);
2445	add_wait_queue(dentry->d_wait, &wait);
2446	do {
2447	set_current_state(TASK_UNINTERRUPTIBLE);
2448	spin_unlock(&dentry->d_lock);
2449	schedule();
2450	spin_lock(&dentry->d_lock);
2451	} while (d_in_lookup(dentry));
2452	}
2453	}
2454
2455	struct dentry *d_alloc_parallel(struct dentry *parent,
2456	const struct qstr *name,
2457	wait_queue_head_t *wq)
2458	{
2459	unsigned int hash = name->hash;
2460	struct hlist_bl_head *b = in_lookup_hash(parent, hash);
2461	struct hlist_bl_node *node;
2462	struct dentry *new = d_alloc(parent, name);
2463	struct dentry *dentry;
2464	unsigned seq, r_seq, d_seq;
2465
2466	if (unlikely(!new))
2467	return ERR_PTR(-ENOMEM);
2468
2469	retry:
2470	rcu_read_lock();
2471	seq = smp_load_acquire(&parent->d_inode->i_dir_seq);
2472	r_seq = read_seqbegin(&rename_lock);
2473	dentry = __d_lookup_rcu(parent, name, &d_seq);
2474	if (unlikely(dentry)) {
2475	if (!lockref_get_not_dead(&dentry->d_lockref)) {
2476	rcu_read_unlock();
2477	goto retry;
2478	}
2479	if (read_seqcount_retry(&dentry->d_seq, d_seq)) {
2480	rcu_read_unlock();
2481	dput(dentry);
2482	goto retry;
2483	}
2484	rcu_read_unlock();
2485	dput(new);
2486	return dentry;
2487	}
2488	if (unlikely(read_seqretry(&rename_lock, r_seq))) {
2489	rcu_read_unlock();
2490	goto retry;
2491	}
2492
2493	if (unlikely(seq & 1)) {
2494	rcu_read_unlock();
2495	goto retry;
2496	}
2497
2498	hlist_bl_lock(b);
2499	if (unlikely(READ_ONCE(parent->d_inode->i_dir_seq) != seq)) {
2500	hlist_bl_unlock(b);
2501	rcu_read_unlock();
2502	goto retry;
2503	}
2504	/*
2505	* No changes for the parent since the beginning of d_lookup().
2506	* Since all removals from the chain happen with hlist_bl_lock(),
2507	* any potential in-lookup matches are going to stay here until
2508	* we unlock the chain. All fields are stable in everything
2509	* we encounter.
2510	*/
2511	hlist_bl_for_each_entry(dentry, node, b, d_u.d_in_lookup_hash) {
2512	if (dentry->d_name.hash != hash)
2513	continue;
2514	if (dentry->d_parent != parent)
2515	continue;
2516	if (!d_same_name(dentry, parent, name))
2517	continue;
2518	hlist_bl_unlock(b);
2519	/* now we can try to grab a reference */
2520	if (!lockref_get_not_dead(&dentry->d_lockref)) {
2521	rcu_read_unlock();
2522	goto retry;
2523	}
2524
2525	rcu_read_unlock();
2526	/*
2527	* somebody is likely to be still doing lookup for it;
2528	* wait for them to finish
2529	*/
2530	spin_lock(&dentry->d_lock);
2531	d_wait_lookup(dentry);
2532	/*
2533	* it's not in-lookup anymore; in principle we should repeat
2534	* everything from dcache lookup, but it's likely to be what
2535	* d_lookup() would've found anyway. If it is, just return it;
2536	* otherwise we really have to repeat the whole thing.
2537	*/
2538	if (unlikely(dentry->d_name.hash != hash))
2539	goto mismatch;
2540	if (unlikely(dentry->d_parent != parent))
2541	goto mismatch;
2542	if (unlikely(d_unhashed(dentry)))
2543	goto mismatch;
2544	if (unlikely(!d_same_name(dentry, parent, name)))
2545	goto mismatch;
2546	/* OK, it *is* a hashed match; return it */
2547	spin_unlock(&dentry->d_lock);
2548	dput(new);
2549	return dentry;
2550	}
2551	rcu_read_unlock();
2552	/* we can't take ->d_lock here; it's OK, though. */
2553	new->d_flags |= DCACHE_PAR_LOOKUP;
2554	new->d_wait = wq;
2555	hlist_bl_add_head_rcu(&new->d_u.d_in_lookup_hash, b);
2556	hlist_bl_unlock(b);
2557	return new;
2558	mismatch:
2559	spin_unlock(&dentry->d_lock);
2560	dput(dentry);
2561	goto retry;
2562	}
2563	EXPORT_SYMBOL(d_alloc_parallel);
2564
2565	void __d_lookup_done(struct dentry *dentry)
2566	{
2567	struct hlist_bl_head *b = in_lookup_hash(dentry->d_parent,
2568	dentry->d_name.hash);
2569	hlist_bl_lock(b);
2570	dentry->d_flags &= ~DCACHE_PAR_LOOKUP;
2571	__hlist_bl_del(&dentry->d_u.d_in_lookup_hash);
2572	wake_up_all(dentry->d_wait);
2573	dentry->d_wait = NULL;
2574	hlist_bl_unlock(b);
2575	INIT_HLIST_NODE(&dentry->d_u.d_alias);
2576	INIT_LIST_HEAD(&dentry->d_lru);
2577	}
2578	EXPORT_SYMBOL(__d_lookup_done);
2579
2580	/* inode->i_lock held if inode is non-NULL */
2581
2582	static inline void __d_add(struct dentry *dentry, struct inode *inode)
2583	{
2584	struct inode *dir = NULL;
2585	unsigned n;
2586	spin_lock(&dentry->d_lock);
2587	if (unlikely(d_in_lookup(dentry))) {
2588	dir = dentry->d_parent->d_inode;
2589	n = start_dir_add(dir);
2590	__d_lookup_done(dentry);
2591	}
2592	if (inode) {
2593	unsigned add_flags = d_flags_for_inode(inode);
2594	hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry);
2595	raw_write_seqcount_begin(&dentry->d_seq);
2596	__d_set_inode_and_type(dentry, inode, add_flags);
2597	raw_write_seqcount_end(&dentry->d_seq);
2598	fsnotify_update_flags(dentry);
2599	}
2600	__d_rehash(dentry);
2601	if (dir)
2602	end_dir_add(dir, n);
2603	spin_unlock(&dentry->d_lock);
2604	if (inode)
2605	spin_unlock(&inode->i_lock);
2606	}
2607
2608	/**
2609	* d_add - add dentry to hash queues
2610	* @entry: dentry to add
2611	* @inode: The inode to attach to this dentry
2612	*
2613	* This adds the entry to the hash queues and initializes @inode.
2614	* The entry was actually filled in earlier during d_alloc().
2615	*/
2616
2617	void d_add(struct dentry *entry, struct inode *inode)
2618	{
2619	if (inode) {
2620	security_d_instantiate(entry, inode);
2621	spin_lock(&inode->i_lock);
2622	}
2623	__d_add(entry, inode);
2624	}
2625	EXPORT_SYMBOL(d_add);
2626
2627	/**
2628	* d_exact_alias - find and hash an exact unhashed alias
2629	* @entry: dentry to add
2630	* @inode: The inode to go with this dentry
2631	*
2632	* If an unhashed dentry with the same name/parent and desired
2633	* inode already exists, hash and return it. Otherwise, return
2634	* NULL.
2635	*
2636	* Parent directory should be locked.
2637	*/
2638	struct dentry *d_exact_alias(struct dentry *entry, struct inode *inode)
2639	{
2640	struct dentry *alias;
2641	unsigned int hash = entry->d_name.hash;
2642
2643	spin_lock(&inode->i_lock);
2644	hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) {
2645	/*
2646	* Don't need alias->d_lock here, because aliases with
2647	* d_parent == entry->d_parent are not subject to name or
2648	* parent changes, because the parent inode i_mutex is held.
2649	*/
2650	if (alias->d_name.hash != hash)
2651	continue;
2652	if (alias->d_parent != entry->d_parent)
2653	continue;
2654	if (!d_same_name(alias, entry->d_parent, &entry->d_name))
2655	continue;
2656	spin_lock(&alias->d_lock);
2657	if (!d_unhashed(alias)) {
2658	spin_unlock(&alias->d_lock);
2659	alias = NULL;
2660	} else {
2661	__dget_dlock(alias);
2662	__d_rehash(alias);
2663	spin_unlock(&alias->d_lock);
2664	}
2665	spin_unlock(&inode->i_lock);
2666	return alias;
2667	}
2668	spin_unlock(&inode->i_lock);
2669	return NULL;
2670	}
2671	EXPORT_SYMBOL(d_exact_alias);
2672
2673	/**
2674	* dentry_update_name_case - update case insensitive dentry with a new name
2675	* @dentry: dentry to be updated
2676	* @name: new name
2677	*
2678	* Update a case insensitive dentry with new case of name.
2679	*
2680	* dentry must have been returned by d_lookup with name @name. Old and new
2681	* name lengths must match (ie. no d_compare which allows mismatched name
2682	* lengths).
2683	*
2684	* Parent inode i_mutex must be held over d_lookup and into this call (to
2685	* keep renames and concurrent inserts, and readdir(2) away).
2686	*/
2687	void dentry_update_name_case(struct dentry *dentry, const struct qstr *name)
2688	{
2689	BUG_ON(!inode_is_locked(dentry->d_parent->d_inode));
2690	BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2691
2692	spin_lock(&dentry->d_lock);
2693	write_seqcount_begin(&dentry->d_seq);
2694	memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2695	write_seqcount_end(&dentry->d_seq);
2696	spin_unlock(&dentry->d_lock);
2697	}
2698	EXPORT_SYMBOL(dentry_update_name_case);
2699
2700	static void swap_names(struct dentry *dentry, struct dentry *target)
2701	{
2702	if (unlikely(dname_external(target))) {
2703	if (unlikely(dname_external(dentry))) {
2704	/*
2705	* Both external: swap the pointers
2706	*/
2707	swap(target->d_name.name, dentry->d_name.name);
2708	} else {
2709	/*
2710	* dentry:internal, target:external. Steal target's
2711	* storage and make target internal.
2712	*/
2713	memcpy(target->d_iname, dentry->d_name.name,
2714	dentry->d_name.len + 1);
2715	dentry->d_name.name = target->d_name.name;
2716	target->d_name.name = target->d_iname;
2717	}
2718	} else {
2719	if (unlikely(dname_external(dentry))) {
2720	/*
2721	* dentry:external, target:internal. Give dentry's
2722	* storage to target and make dentry internal
2723	*/
2724	memcpy(dentry->d_iname, target->d_name.name,
2725	target->d_name.len + 1);
2726	target->d_name.name = dentry->d_name.name;
2727	dentry->d_name.name = dentry->d_iname;
2728	} else {
2729	/*
2730	* Both are internal.
2731	*/
2732	unsigned int i;
2733	BUILD_BUG_ON(!IS_ALIGNED(DNAME_INLINE_LEN, sizeof(long)));
2734	kmemcheck_mark_initialized(dentry->d_iname, DNAME_INLINE_LEN);
2735	kmemcheck_mark_initialized(target->d_iname, DNAME_INLINE_LEN);
2736	for (i = 0; i < DNAME_INLINE_LEN / sizeof(long); i++) {
2737	swap(((long *) &dentry->d_iname)[i],
2738	((long *) &target->d_iname)[i]);
2739	}
2740	}
2741	}
2742	swap(dentry->d_name.hash_len, target->d_name.hash_len);
2743	}
2744
2745	static void copy_name(struct dentry *dentry, struct dentry *target)
2746	{
2747	struct external_name *old_name = NULL;
2748	if (unlikely(dname_external(dentry)))
2749	old_name = external_name(dentry);
2750	if (unlikely(dname_external(target))) {
2751	atomic_inc(&external_name(target)->u.count);
2752	dentry->d_name = target->d_name;
2753	} else {
2754	memcpy(dentry->d_iname, target->d_name.name,
2755	target->d_name.len + 1);
2756	dentry->d_name.name = dentry->d_iname;
2757	dentry->d_name.hash_len = target->d_name.hash_len;
2758	}
2759	if (old_name && likely(atomic_dec_and_test(&old_name->u.count)))
2760	kfree_rcu(old_name, u.head);
2761	}
2762
2763	static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2764	{
2765	/*
2766	* XXXX: do we really need to take target->d_lock?
2767	*/
2768	if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2769	spin_lock(&target->d_parent->d_lock);
2770	else {
2771	if (d_ancestor(dentry->d_parent, target->d_parent)) {
2772	spin_lock(&dentry->d_parent->d_lock);
2773	spin_lock_nested(&target->d_parent->d_lock,
2774	DENTRY_D_LOCK_NESTED);
2775	} else {
2776	spin_lock(&target->d_parent->d_lock);
2777	spin_lock_nested(&dentry->d_parent->d_lock,
2778	DENTRY_D_LOCK_NESTED);
2779	}
2780	}
2781	if (target < dentry) {
2782	spin_lock_nested(&target->d_lock, 2);
2783	spin_lock_nested(&dentry->d_lock, 3);
2784	} else {
2785	spin_lock_nested(&dentry->d_lock, 2);
2786	spin_lock_nested(&target->d_lock, 3);
2787	}
2788	}
2789
2790	static void dentry_unlock_for_move(struct dentry *dentry, struct dentry *target)
2791	{
2792	if (target->d_parent != dentry->d_parent)
2793	spin_unlock(&dentry->d_parent->d_lock);
2794	if (target->d_parent != target)
2795	spin_unlock(&target->d_parent->d_lock);
2796	spin_unlock(&target->d_lock);
2797	spin_unlock(&dentry->d_lock);
2798	}
2799
2800	/*
2801	* When switching names, the actual string doesn't strictly have to
2802	* be preserved in the target - because we're dropping the target
2803	* anyway. As such, we can just do a simple memcpy() to copy over
2804	* the new name before we switch, unless we are going to rehash
2805	* it. Note that if we *do* unhash the target, we are not allowed
2806	* to rehash it without giving it a new name/hash key - whether
2807	* we swap or overwrite the names here, resulting name won't match
2808	* the reality in filesystem; it's only there for d_path() purposes.
2809	* Note that all of this is happening under rename_lock, so the
2810	* any hash lookup seeing it in the middle of manipulations will
2811	* be discarded anyway. So we do not care what happens to the hash
2812	* key in that case.
2813	*/
2814	/*
2815	* __d_move - move a dentry
2816	* @dentry: entry to move
2817	* @target: new dentry
2818	* @exchange: exchange the two dentries
2819	*
2820	* Update the dcache to reflect the move of a file name. Negative
2821	* dcache entries should not be moved in this way. Caller must hold
2822	* rename_lock, the i_mutex of the source and target directories,
2823	* and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2824	*/
2825	static void __d_move(struct dentry *dentry, struct dentry *target,
2826	bool exchange)
2827	{
2828	struct inode *dir = NULL;
2829	unsigned n;
2830	if (!dentry->d_inode)
2831	printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2832
2833	BUG_ON(d_ancestor(dentry, target));
2834	BUG_ON(d_ancestor(target, dentry));
2835
2836	dentry_lock_for_move(dentry, target);
2837	if (unlikely(d_in_lookup(target))) {
2838	dir = target->d_parent->d_inode;
2839	n = start_dir_add(dir);
2840	__d_lookup_done(target);
2841	}
2842
2843	write_seqcount_begin(&dentry->d_seq);
2844	write_seqcount_begin_nested(&target->d_seq, DENTRY_D_LOCK_NESTED);
2845
2846	/* unhash both */
2847	/* ___d_drop does write_seqcount_barrier, but they're OK to nest. */
2848	___d_drop(dentry);
2849	___d_drop(target);
2850
2851	/* Switch the names.. */
2852	if (exchange)
2853	swap_names(dentry, target);
2854	else
2855	copy_name(dentry, target);
2856
2857	/* rehash in new place(s) */
2858	__d_rehash(dentry);
2859	if (exchange)
2860	__d_rehash(target);
2861	else
2862	target->d_hash.pprev = NULL;
2863
2864	/* ... and switch them in the tree */
2865	if (IS_ROOT(dentry)) {
2866	/* splicing a tree */
2867	dentry->d_flags |= DCACHE_RCUACCESS;
2868	dentry->d_parent = target->d_parent;
2869	target->d_parent = target;
2870	list_del_init(&target->d_child);
2871	list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2872	} else {
2873	/* swapping two dentries */
2874	swap(dentry->d_parent, target->d_parent);
2875	list_move(&target->d_child, &target->d_parent->d_subdirs);
2876	list_move(&dentry->d_child, &dentry->d_parent->d_subdirs);
2877	if (exchange)
2878	fsnotify_update_flags(target);
2879	fsnotify_update_flags(dentry);
2880	}
2881
2882	write_seqcount_end(&target->d_seq);
2883	write_seqcount_end(&dentry->d_seq);
2884
2885	if (dir)
2886	end_dir_add(dir, n);
2887	dentry_unlock_for_move(dentry, target);
2888	}
2889
2890	/*
2891	* d_move - move a dentry
2892	* @dentry: entry to move
2893	* @target: new dentry
2894	*
2895	* Update the dcache to reflect the move of a file name. Negative
2896	* dcache entries should not be moved in this way. See the locking
2897	* requirements for __d_move.
2898	*/
2899	void d_move(struct dentry *dentry, struct dentry *target)
2900	{
2901	write_seqlock(&rename_lock);
2902	__d_move(dentry, target, false);
2903	write_sequnlock(&rename_lock);
2904	}
2905	EXPORT_SYMBOL(d_move);
2906
2907	/*
2908	* d_exchange - exchange two dentries
2909	* @dentry1: first dentry
2910	* @dentry2: second dentry
2911	*/
2912	void d_exchange(struct dentry *dentry1, struct dentry *dentry2)
2913	{
2914	write_seqlock(&rename_lock);
2915
2916	WARN_ON(!dentry1->d_inode);
2917	WARN_ON(!dentry2->d_inode);
2918	WARN_ON(IS_ROOT(dentry1));
2919	WARN_ON(IS_ROOT(dentry2));
2920
2921	__d_move(dentry1, dentry2, true);
2922
2923	write_sequnlock(&rename_lock);
2924	}
2925
2926	/**
2927	* d_ancestor - search for an ancestor
2928	* @p1: ancestor dentry
2929	* @p2: child dentry
2930	*
2931	* Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2932	* an ancestor of p2, else NULL.
2933	*/
2934	struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2935	{
2936	struct dentry *p;
2937
2938	for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2939	if (p->d_parent == p1)
2940	return p;
2941	}
2942	return NULL;
2943	}
2944
2945	/*
2946	* This helper attempts to cope with remotely renamed directories
2947	*
2948	* It assumes that the caller is already holding
2949	* dentry->d_parent->d_inode->i_mutex, and rename_lock
2950	*
2951	* Note: If ever the locking in lock_rename() changes, then please
2952	* remember to update this too...
2953	*/
2954	static int __d_unalias(struct inode *inode,
2955	struct dentry *dentry, struct dentry *alias)
2956	{
2957	struct mutex *m1 = NULL;
2958	struct rw_semaphore *m2 = NULL;
2959	int ret = -ESTALE;
2960
2961	/* If alias and dentry share a parent, then no extra locks required */
2962	if (alias->d_parent == dentry->d_parent)
2963	goto out_unalias;
2964
2965	/* See lock_rename() */
2966	if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2967	goto out_err;
2968	m1 = &dentry->d_sb->s_vfs_rename_mutex;
2969	if (!inode_trylock_shared(alias->d_parent->d_inode))
2970	goto out_err;
2971	m2 = &alias->d_parent->d_inode->i_rwsem;
2972	out_unalias:
2973	__d_move(alias, dentry, false);
2974	ret = 0;
2975	out_err:
2976	if (m2)
2977	up_read(m2);
2978	if (m1)
2979	mutex_unlock(m1);
2980	return ret;
2981	}
2982
2983	/**
2984	* d_splice_alias - splice a disconnected dentry into the tree if one exists
2985	* @inode: the inode which may have a disconnected dentry
2986	* @dentry: a negative dentry which we want to point to the inode.
2987	*
2988	* If inode is a directory and has an IS_ROOT alias, then d_move that in
2989	* place of the given dentry and return it, else simply d_add the inode
2990	* to the dentry and return NULL.
2991	*
2992	* If a non-IS_ROOT directory is found, the filesystem is corrupt, and
2993	* we should error out: directories can't have multiple aliases.
2994	*
2995	* This is needed in the lookup routine of any filesystem that is exportable
2996	* (via knfsd) so that we can build dcache paths to directories effectively.
2997	*
2998	* If a dentry was found and moved, then it is returned. Otherwise NULL
2999	* is returned. This matches the expected return value of ->lookup.
3000	*
3001	* Cluster filesystems may call this function with a negative, hashed dentry.
3002	* In that case, we know that the inode will be a regular file, and also this
3003	* will only occur during atomic_open. So we need to check for the dentry
3004	* being already hashed only in the final case.
3005	*/
3006	struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
3007	{
3008	if (IS_ERR(inode))
3009	return ERR_CAST(inode);
3010
3011	BUG_ON(!d_unhashed(dentry));
3012
3013	if (!inode)
3014	goto out;
3015
3016	security_d_instantiate(dentry, inode);
3017	spin_lock(&inode->i_lock);
3018	if (S_ISDIR(inode->i_mode)) {
3019	struct dentry *new = __d_find_any_alias(inode);
3020	if (unlikely(new)) {
3021	/* The reference to new ensures it remains an alias */
3022	spin_unlock(&inode->i_lock);
3023	write_seqlock(&rename_lock);
3024	if (unlikely(d_ancestor(new, dentry))) {
3025	write_sequnlock(&rename_lock);
3026	dput(new);
3027	new = ERR_PTR(-ELOOP);
3028	pr_warn_ratelimited(
3029	"VFS: Lookup of '%s' in %s %s"
3030	" would have caused loop\n",
3031	dentry->d_name.name,
3032	inode->i_sb->s_type->name,
3033	inode->i_sb->s_id);
3034	} else if (!IS_ROOT(new)) {
3035	int err = __d_unalias(inode, dentry, new);
3036	write_sequnlock(&rename_lock);
3037	if (err) {
3038	dput(new);
3039	new = ERR_PTR(err);
3040	}
3041	} else {
3042	__d_move(new, dentry, false);
3043	write_sequnlock(&rename_lock);
3044	}
3045	iput(inode);
3046	return new;
3047	}
3048	}
3049	out:
3050	__d_add(dentry, inode);
3051	return NULL;
3052	}
3053	EXPORT_SYMBOL(d_splice_alias);
3054
3055	static int prepend(char **buffer, int *buflen, const char *str, int namelen)
3056	{
3057	*buflen -= namelen;
3058	if (*buflen < 0)
3059	return -ENAMETOOLONG;
3060	*buffer -= namelen;
3061	memcpy(*buffer, str, namelen);
3062	return 0;
3063	}
3064
3065	/**
3066	* prepend_name - prepend a pathname in front of current buffer pointer
3067	* @buffer: buffer pointer
3068	* @buflen: allocated length of the buffer
3069	* @name: name string and length qstr structure
3070	*
3071	* With RCU path tracing, it may race with d_move(). Use ACCESS_ONCE() to
3072	* make sure that either the old or the new name pointer and length are
3073	* fetched. However, there may be mismatch between length and pointer.
3074	* The length cannot be trusted, we need to copy it byte-by-byte until
3075	* the length is reached or a null byte is found. It also prepends "/" at
3076	* the beginning of the name. The sequence number check at the caller will
3077	* retry it again when a d_move() does happen. So any garbage in the buffer
3078	* due to mismatched pointer and length will be discarded.
3079	*
3080	* Data dependency barrier is needed to make sure that we see that terminating
3081	* NUL. Alpha strikes again, film at 11...
3082	*/
3083	static int prepend_name(char **buffer, int *buflen, const struct qstr *name)
3084	{
3085	const char *dname = ACCESS_ONCE(name->name);
3086	u32 dlen = ACCESS_ONCE(name->len);
3087	char *p;
3088
3089	smp_read_barrier_depends();
3090
3091	*buflen -= dlen + 1;
3092	if (*buflen < 0)
3093	return -ENAMETOOLONG;
3094	p = *buffer -= dlen + 1;
3095	*p++ = '/';
3096	while (dlen--) {
3097	char c = *dname++;
3098	if (!c)
3099	break;
3100	*p++ = c;
3101	}
3102	return 0;
3103	}
3104
3105	/**
3106	* prepend_path - Prepend path string to a buffer
3107	* @path: the dentry/vfsmount to report
3108	* @root: root vfsmnt/dentry
3109	* @buffer: pointer to the end of the buffer
3110	* @buflen: pointer to buffer length
3111	*
3112	* The function will first try to write out the pathname without taking any
3113	* lock other than the RCU read lock to make sure that dentries won't go away.
3114	* It only checks the sequence number of the global rename_lock as any change
3115	* in the dentry's d_seq will be preceded by changes in the rename_lock
3116	* sequence number. If the sequence number had been changed, it will restart
3117	* the whole pathname back-tracing sequence again by taking the rename_lock.
3118	* In this case, there is no need to take the RCU read lock as the recursive
3119	* parent pointer references will keep the dentry chain alive as long as no
3120	* rename operation is performed.
3121	*/
3122	static int prepend_path(const struct path *path,
3123	const struct path *root,
3124	char **buffer, int *buflen)
3125	{
3126	struct dentry *dentry;
3127	struct vfsmount *vfsmnt;
3128	struct mount *mnt;
3129	int error = 0;
3130	unsigned seq, m_seq = 0;
3131	char *bptr;
3132	int blen;
3133
3134	rcu_read_lock();
3135	restart_mnt:
3136	read_seqbegin_or_lock(&mount_lock, &m_seq);
3137	seq = 0;
3138	rcu_read_lock();
3139	restart:
3140	bptr = *buffer;
3141	blen = *buflen;
3142	error = 0;
3143	dentry = path->dentry;
3144	vfsmnt = path->mnt;
3145	mnt = real_mount(vfsmnt);
3146	read_seqbegin_or_lock(&rename_lock, &seq);
3147	while (dentry != root->dentry || vfsmnt != root->mnt) {
3148	struct dentry * parent;
3149
3150	if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
3151	struct mount *parent = ACCESS_ONCE(mnt->mnt_parent);
3152	/* Escaped? */
3153	if (dentry != vfsmnt->mnt_root) {
3154	bptr = *buffer;
3155	blen = *buflen;
3156	error = 3;
3157	break;
3158	}
3159	/* Global root? */
3160	if (mnt != parent) {
3161	dentry = ACCESS_ONCE(mnt->mnt_mountpoint);
3162	mnt = parent;
3163	vfsmnt = &mnt->mnt;
3164	continue;
3165	}
3166	if (!error)
3167	error = is_mounted(vfsmnt) ? 1 : 2;
3168	break;
3169	}
3170	parent = dentry->d_parent;
3171	prefetch(parent);
3172	error = prepend_name(&bptr, &blen, &dentry->d_name);
3173	if (error)
3174	break;
3175
3176	dentry = parent;
3177	}
3178	if (!(seq & 1))
3179	rcu_read_unlock();
3180	if (need_seqretry(&rename_lock, seq)) {
3181	seq = 1;
3182	goto restart;
3183	}
3184	done_seqretry(&rename_lock, seq);
3185
3186	if (!(m_seq & 1))
3187	rcu_read_unlock();
3188	if (need_seqretry(&mount_lock, m_seq)) {
3189	m_seq = 1;
3190	goto restart_mnt;
3191	}
3192	done_seqretry(&mount_lock, m_seq);
3193
3194	if (error >= 0 && bptr == *buffer) {
3195	if (--blen < 0)
3196	error = -ENAMETOOLONG;
3197	else
3198	*--bptr = '/';
3199	}
3200	*buffer = bptr;
3201	*buflen = blen;
3202	return error;
3203	}
3204
3205	/**
3206	* __d_path - return the path of a dentry
3207	* @path: the dentry/vfsmount to report
3208	* @root: root vfsmnt/dentry
3209	* @buf: buffer to return value in
3210	* @buflen: buffer length
3211	*
3212	* Convert a dentry into an ASCII path name.
3213	*
3214	* Returns a pointer into the buffer or an error code if the
3215	* path was too long.
3216	*
3217	* "buflen" should be positive.
3218	*
3219	* If the path is not reachable from the supplied root, return %NULL.
3220	*/
3221	char *__d_path(const struct path *path,
3222	const struct path *root,
3223	char *buf, int buflen)
3224	{
3225	char *res = buf + buflen;
3226	int error;
3227
3228	prepend(&res, &buflen, "\0", 1);
3229	error = prepend_path(path, root, &res, &buflen);
3230
3231	if (error < 0)
3232	return ERR_PTR(error);
3233	if (error > 0)
3234	return NULL;
3235	return res;
3236	}
3237
3238	char *d_absolute_path(const struct path *path,
3239	char *buf, int buflen)
3240	{
3241	struct path root = {};
3242	char *res = buf + buflen;
3243	int error;
3244
3245	prepend(&res, &buflen, "\0", 1);
3246	error = prepend_path(path, &root, &res, &buflen);
3247
3248	if (error > 1)
3249	error = -EINVAL;
3250	if (error < 0)
3251	return ERR_PTR(error);
3252	return res;
3253	}
3254	EXPORT_SYMBOL(d_absolute_path);
3255
3256	/*
3257	* same as __d_path but appends "(deleted)" for unlinked files.
3258	*/
3259	static int path_with_deleted(const struct path *path,
3260	const struct path *root,
3261	char **buf, int *buflen)
3262	{
3263	prepend(buf, buflen, "\0", 1);
3264	if (d_unlinked(path->dentry)) {
3265	int error = prepend(buf, buflen, " (deleted)", 10);
3266	if (error)
3267	return error;
3268	}
3269
3270	return prepend_path(path, root, buf, buflen);
3271	}
3272
3273	static int prepend_unreachable(char **buffer, int *buflen)
3274	{
3275	return prepend(buffer, buflen, "(unreachable)", 13);
3276	}
3277
3278	static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
3279	{
3280	unsigned seq;
3281
3282	do {
3283	seq = read_seqcount_begin(&fs->seq);
3284	*root = fs->root;
3285	} while (read_seqcount_retry(&fs->seq, seq));
3286	}
3287
3288	/**
3289	* d_path - return the path of a dentry
3290	* @path: path to report
3291	* @buf: buffer to return value in
3292	* @buflen: buffer length
3293	*
3294	* Convert a dentry into an ASCII path name. If the entry has been deleted
3295	* the string " (deleted)" is appended. Note that this is ambiguous.
3296	*
3297	* Returns a pointer into the buffer or an error code if the path was
3298	* too long. Note: Callers should use the returned pointer, not the passed
3299	* in buffer, to use the name! The implementation often starts at an offset
3300	* into the buffer, and may leave 0 bytes at the start.
3301	*
3302	* "buflen" should be positive.
3303	*/
3304	char *d_path(const struct path *path, char *buf, int buflen)
3305	{
3306	char *res = buf + buflen;
3307	struct path root;
3308	int error;
3309
3310	/*
3311	* We have various synthetic filesystems that never get mounted. On
3312	* these filesystems dentries are never used for lookup purposes, and
3313	* thus don't need to be hashed. They also don't need a name until a
3314	* user wants to identify the object in /proc/pid/fd/. The little hack
3315	* below allows us to generate a name for these objects on demand:
3316	*
3317	* Some pseudo inodes are mountable. When they are mounted
3318	* path->dentry == path->mnt->mnt_root. In that case don't call d_dname
3319	* and instead have d_path return the mounted path.
3320	*/
3321	if (path->dentry->d_op && path->dentry->d_op->d_dname &&
3322	(!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root))
3323	return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
3324
3325	rcu_read_lock();
3326	get_fs_root_rcu(current->fs, &root);
3327	error = path_with_deleted(path, &root, &res, &buflen);
3328	rcu_read_unlock();
3329
3330	if (error < 0)
3331	res = ERR_PTR(error);
3332	return res;
3333	}
3334	EXPORT_SYMBOL(d_path);
3335
3336	/*
3337	* Helper function for dentry_operations.d_dname() members
3338	*/
3339	char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
3340	const char *fmt, ...)
3341	{
3342	va_list args;
3343	char temp[64];
3344	int sz;
3345
3346	va_start(args, fmt);
3347	sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
3348	va_end(args);
3349
3350	if (sz > sizeof(temp) || sz > buflen)
3351	return ERR_PTR(-ENAMETOOLONG);
3352
3353	buffer += buflen - sz;
3354	return memcpy(buffer, temp, sz);
3355	}
3356
3357	char *simple_dname(struct dentry *dentry, char *buffer, int buflen)
3358	{
3359	char *end = buffer + buflen;
3360	/* these dentries are never renamed, so d_lock is not needed */
3361	if (prepend(&end, &buflen, " (deleted)", 11) ||
3362	prepend(&end, &buflen, dentry->d_name.name, dentry->d_name.len) ||
3363	prepend(&end, &buflen, "/", 1))
3364	end = ERR_PTR(-ENAMETOOLONG);
3365	return end;
3366	}
3367	EXPORT_SYMBOL(simple_dname);
3368
3369	/*
3370	* Write full pathname from the root of the filesystem into the buffer.
3371	*/
3372	static char *__dentry_path(struct dentry *d, char *buf, int buflen)
3373	{
3374	struct dentry *dentry;
3375	char *end, *retval;
3376	int len, seq = 0;
3377	int error = 0;
3378
3379	if (buflen < 2)
3380	goto Elong;
3381
3382	rcu_read_lock();
3383	restart:
3384	dentry = d;
3385	end = buf + buflen;
3386	len = buflen;
3387	prepend(&end, &len, "\0", 1);
3388	/* Get '/' right */
3389	retval = end-1;
3390	*retval = '/';
3391	read_seqbegin_or_lock(&rename_lock, &seq);
3392	while (!IS_ROOT(dentry)) {
3393	struct dentry *parent = dentry->d_parent;
3394
3395	prefetch(parent);
3396	error = prepend_name(&end, &len, &dentry->d_name);
3397	if (error)
3398	break;
3399
3400	retval = end;
3401	dentry = parent;
3402	}
3403	if (!(seq & 1))
3404	rcu_read_unlock();
3405	if (need_seqretry(&rename_lock, seq)) {
3406	seq = 1;
3407	goto restart;
3408	}
3409	done_seqretry(&rename_lock, seq);
3410	if (error)
3411	goto Elong;
3412	return retval;
3413	Elong:
3414	return ERR_PTR(-ENAMETOOLONG);
3415	}
3416
3417	char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
3418	{
3419	return __dentry_path(dentry, buf, buflen);
3420	}
3421	EXPORT_SYMBOL(dentry_path_raw);
3422
3423	char *dentry_path(struct dentry *dentry, char *buf, int buflen)
3424	{
3425	char *p = NULL;
3426	char *retval;
3427
3428	if (d_unlinked(dentry)) {
3429	p = buf + buflen;
3430	if (prepend(&p, &buflen, "//deleted", 10) != 0)
3431	goto Elong;
3432	buflen++;
3433	}
3434	retval = __dentry_path(dentry, buf, buflen);
3435	if (!IS_ERR(retval) && p)
3436	*p = '/'; /* restore '/' overriden with '\0' */
3437	return retval;
3438	Elong:
3439	return ERR_PTR(-ENAMETOOLONG);
3440	}
3441
3442	static void get_fs_root_and_pwd_rcu(struct fs_struct *fs, struct path *root,
3443	struct path *pwd)
3444	{
3445	unsigned seq;
3446
3447	do {
3448	seq = read_seqcount_begin(&fs->seq);
3449	*root = fs->root;
3450	*pwd = fs->pwd;
3451	} while (read_seqcount_retry(&fs->seq, seq));
3452	}
3453
3454	/*
3455	* NOTE! The user-level library version returns a
3456	* character pointer. The kernel system call just
3457	* returns the length of the buffer filled (which
3458	* includes the ending '\0' character), or a negative
3459	* error value. So libc would do something like
3460	*
3461	* char *getcwd(char * buf, size_t size)
3462	* {
3463	* int retval;
3464	*
3465	* retval = sys_getcwd(buf, size);
3466	* if (retval >= 0)
3467	* return buf;
3468	* errno = -retval;
3469	* return NULL;
3470	* }
3471	*/
3472	SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
3473	{
3474	int error;
3475	struct path pwd, root;
3476	char *page = __getname();
3477
3478	if (!page)
3479	return -ENOMEM;
3480
3481	rcu_read_lock();
3482	get_fs_root_and_pwd_rcu(current->fs, &root, &pwd);
3483
3484	error = -ENOENT;
3485	if (!d_unlinked(pwd.dentry)) {
3486	unsigned long len;
3487	char *cwd = page + PATH_MAX;
3488	int buflen = PATH_MAX;
3489
3490	prepend(&cwd, &buflen, "\0", 1);
3491	error = prepend_path(&pwd, &root, &cwd, &buflen);
3492	rcu_read_unlock();
3493
3494	if (error < 0)
3495	goto out;
3496
3497	/* Unreachable from current root */
3498	if (error > 0) {
3499	error = prepend_unreachable(&cwd, &buflen);
3500	if (error)
3501	goto out;
3502	}
3503
3504	error = -ERANGE;
3505	len = PATH_MAX + page - cwd;
3506	if (len <= size) {
3507	error = len;
3508	if (copy_to_user(buf, cwd, len))
3509	error = -EFAULT;
3510	}
3511	} else {
3512	rcu_read_unlock();
3513	}
3514
3515	out:
3516	__putname(page);
3517	return error;
3518	}
3519
3520	/*
3521	* Test whether new_dentry is a subdirectory of old_dentry.
3522	*
3523	* Trivially implemented using the dcache structure
3524	*/
3525
3526	/**
3527	* is_subdir - is new dentry a subdirectory of old_dentry
3528	* @new_dentry: new dentry
3529	* @old_dentry: old dentry
3530	*
3531	* Returns true if new_dentry is a subdirectory of the parent (at any depth).
3532	* Returns false otherwise.
3533	* Caller must ensure that "new_dentry" is pinned before calling is_subdir()
3534	*/
3535
3536	bool is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
3537	{
3538	bool result;
3539	unsigned seq;
3540
3541	if (new_dentry == old_dentry)
3542	return true;
3543
3544	do {
3545	/* for restarting inner loop in case of seq retry */
3546	seq = read_seqbegin(&rename_lock);
3547	/*
3548	* Need rcu_readlock to protect against the d_parent trashing
3549	* due to d_move
3550	*/
3551	rcu_read_lock();
3552	if (d_ancestor(old_dentry, new_dentry))
3553	result = true;
3554	else
3555	result = false;
3556	rcu_read_unlock();
3557	} while (read_seqretry(&rename_lock, seq));
3558
3559	return result;
3560	}
3561
3562	static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry)
3563	{
3564	struct dentry *root = data;
3565	if (dentry != root) {
3566	if (d_unhashed(dentry) || !dentry->d_inode)
3567	return D_WALK_SKIP;
3568
3569	if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
3570	dentry->d_flags |= DCACHE_GENOCIDE;
3571	dentry->d_lockref.count--;
3572	}
3573	}
3574	return D_WALK_CONTINUE;
3575	}
3576
3577	void d_genocide(struct dentry *parent)
3578	{
3579	d_walk(parent, parent, d_genocide_kill, NULL);
3580	}
3581
3582	void d_tmpfile(struct dentry *dentry, struct inode *inode)
3583	{
3584	inode_dec_link_count(inode);
3585	BUG_ON(dentry->d_name.name != dentry->d_iname ||
3586	!hlist_unhashed(&dentry->d_u.d_alias) ||
3587	!d_unlinked(dentry));
3588	spin_lock(&dentry->d_parent->d_lock);
3589	spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
3590	dentry->d_name.len = sprintf(dentry->d_iname, "#%llu",
3591	(unsigned long long)inode->i_ino);
3592	spin_unlock(&dentry->d_lock);
3593	spin_unlock(&dentry->d_parent->d_lock);
3594	d_instantiate(dentry, inode);
3595	}
3596	EXPORT_SYMBOL(d_tmpfile);
3597
3598	static __initdata unsigned long dhash_entries;
3599	static int __init set_dhash_entries(char *str)
3600	{
3601	if (!str)
3602	return 0;
3603	dhash_entries = simple_strtoul(str, &str, 0);
3604	return 1;
3605	}
3606	__setup("dhash_entries=", set_dhash_entries);
3607
3608	static void __init dcache_init_early(void)
3609	{
3610	unsigned int loop;
3611
3612	/* If hashes are distributed across NUMA nodes, defer
3613	* hash allocation until vmalloc space is available.
3614	*/
3615	if (hashdist)
3616	return;
3617
3618	dentry_hashtable =
3619	alloc_large_system_hash("Dentry cache",
3620	sizeof(struct hlist_bl_head),
3621	dhash_entries,
3622	13,
3623	HASH_EARLY,
3624	&d_hash_shift,
3625	&d_hash_mask,
3626	0,
3627	0);
3628
3629	for (loop = 0; loop < (1U << d_hash_shift); loop++)
3630	INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3631	}
3632
3633	static void __init dcache_init(void)
3634	{
3635	unsigned int loop;
3636
3637	/*
3638	* A constructor could be added for stable state like the lists,
3639	* but it is probably not worth it because of the cache nature
3640	* of the dcache.
3641	*/
3642	dentry_cache = KMEM_CACHE(dentry,
3643	SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD|SLAB_ACCOUNT);
3644
3645	/* Hash may have been set up in dcache_init_early */
3646	if (!hashdist)
3647	return;
3648
3649	dentry_hashtable =
3650	alloc_large_system_hash("Dentry cache",
3651	sizeof(struct hlist_bl_head),
3652	dhash_entries,
3653	13,
3654	0,
3655	&d_hash_shift,
3656	&d_hash_mask,
3657	0,
3658	0);
3659
3660	for (loop = 0; loop < (1U << d_hash_shift); loop++)
3661	INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3662	}
3663
3664	/* SLAB cache for __getname() consumers */
3665	struct kmem_cache *names_cachep __read_mostly;
3666	EXPORT_SYMBOL(names_cachep);
3667
3668	EXPORT_SYMBOL(d_genocide);
3669
3670	void __init vfs_caches_init_early(void)
3671	{
3672	dcache_init_early();
3673	inode_init_early();
3674	}
3675
3676	void __init vfs_caches_init(void)
3677	{
3678	names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3679	SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3680
3681	dcache_init();
3682	inode_init();
3683	files_init();
3684	files_maxfiles_init();
3685	mnt_init();
3686	bdev_cache_init();
3687	chrdev_init();
3688	}
3689