• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 "internal.h"
41 #include "mount.h"
42 
43 /*
44  * Usage:
45  * dcache->d_inode->i_lock protects:
46  *   - i_dentry, d_alias, d_inode of aliases
47  * dcache_hash_bucket lock protects:
48  *   - the dcache hash table
49  * s_anon bl list spinlock protects:
50  *   - the s_anon list (see __d_drop)
51  * dcache_lru_lock protects:
52  *   - the dcache lru lists and counters
53  * d_lock protects:
54  *   - d_flags
55  *   - d_name
56  *   - d_lru
57  *   - d_count
58  *   - d_unhashed()
59  *   - d_parent and d_subdirs
60  *   - childrens' d_child and d_parent
61  *   - d_alias, d_inode
62  *
63  * Ordering:
64  * dentry->d_inode->i_lock
65  *   dentry->d_lock
66  *     dcache_lru_lock
67  *     dcache_hash_bucket lock
68  *     s_anon lock
69  *
70  * If there is an ancestor relationship:
71  * dentry->d_parent->...->d_parent->d_lock
72  *   ...
73  *     dentry->d_parent->d_lock
74  *       dentry->d_lock
75  *
76  * If no ancestor relationship:
77  * if (dentry1 < dentry2)
78  *   dentry1->d_lock
79  *     dentry2->d_lock
80  */
81 int sysctl_vfs_cache_pressure __read_mostly = 100;
82 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
83 
84 static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock);
85 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
86 
87 EXPORT_SYMBOL(rename_lock);
88 
89 static struct kmem_cache *dentry_cache __read_mostly;
90 
91 /*
92  * This is the single most critical data structure when it comes
93  * to the dcache: the hashtable for lookups. Somebody should try
94  * to make this good - I've just made it work.
95  *
96  * This hash-function tries to avoid losing too many bits of hash
97  * information, yet avoid using a prime hash-size or similar.
98  */
99 #define D_HASHBITS     d_hash_shift
100 #define D_HASHMASK     d_hash_mask
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 
d_hash(const struct dentry * parent,unsigned int hash)107 static inline struct hlist_bl_head *d_hash(const struct dentry *parent,
108 					unsigned int hash)
109 {
110 	hash += (unsigned long) parent / L1_CACHE_BYTES;
111 	hash = hash + (hash >> D_HASHBITS);
112 	return dentry_hashtable + (hash & D_HASHMASK);
113 }
114 
115 /* Statistics gathering. */
116 struct dentry_stat_t dentry_stat = {
117 	.age_limit = 45,
118 };
119 
120 static DEFINE_PER_CPU(unsigned int, nr_dentry);
121 
122 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
get_nr_dentry(void)123 static int get_nr_dentry(void)
124 {
125 	int i;
126 	int sum = 0;
127 	for_each_possible_cpu(i)
128 		sum += per_cpu(nr_dentry, i);
129 	return sum < 0 ? 0 : sum;
130 }
131 
proc_nr_dentry(ctl_table * table,int write,void __user * buffer,size_t * lenp,loff_t * ppos)132 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
133 		   size_t *lenp, loff_t *ppos)
134 {
135 	dentry_stat.nr_dentry = get_nr_dentry();
136 	return proc_dointvec(table, write, buffer, lenp, ppos);
137 }
138 #endif
139 
140 /*
141  * Compare 2 name strings, return 0 if they match, otherwise non-zero.
142  * The strings are both count bytes long, and count is non-zero.
143  */
144 #ifdef CONFIG_DCACHE_WORD_ACCESS
145 
146 #include <asm/word-at-a-time.h>
147 /*
148  * NOTE! 'cs' and 'scount' come from a dentry, so it has a
149  * aligned allocation for this particular component. We don't
150  * strictly need the load_unaligned_zeropad() safety, but it
151  * doesn't hurt either.
152  *
153  * In contrast, 'ct' and 'tcount' can be from a pathname, and do
154  * need the careful unaligned handling.
155  */
dentry_cmp(const unsigned char * cs,size_t scount,const unsigned char * ct,size_t tcount)156 static inline int dentry_cmp(const unsigned char *cs, size_t scount,
157 				const unsigned char *ct, size_t tcount)
158 {
159 	unsigned long a,b,mask;
160 
161 	if (unlikely(scount != tcount))
162 		return 1;
163 
164 	for (;;) {
165 		a = load_unaligned_zeropad(cs);
166 		b = load_unaligned_zeropad(ct);
167 		if (tcount < sizeof(unsigned long))
168 			break;
169 		if (unlikely(a != b))
170 			return 1;
171 		cs += sizeof(unsigned long);
172 		ct += sizeof(unsigned long);
173 		tcount -= sizeof(unsigned long);
174 		if (!tcount)
175 			return 0;
176 	}
177 	mask = ~(~0ul << tcount*8);
178 	return unlikely(!!((a ^ b) & mask));
179 }
180 
181 #else
182 
dentry_cmp(const unsigned char * cs,size_t scount,const unsigned char * ct,size_t tcount)183 static inline int dentry_cmp(const unsigned char *cs, size_t scount,
184 				const unsigned char *ct, size_t tcount)
185 {
186 	if (scount != tcount)
187 		return 1;
188 
189 	do {
190 		if (*cs != *ct)
191 			return 1;
192 		cs++;
193 		ct++;
194 		tcount--;
195 	} while (tcount);
196 	return 0;
197 }
198 
199 #endif
200 
__d_free(struct rcu_head * head)201 static void __d_free(struct rcu_head *head)
202 {
203 	struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
204 
205 	WARN_ON(!list_empty(&dentry->d_alias));
206 	if (dname_external(dentry))
207 		kfree(dentry->d_name.name);
208 	kmem_cache_free(dentry_cache, dentry);
209 }
210 
211 /*
212  * no locks, please.
213  */
d_free(struct dentry * dentry)214 static void d_free(struct dentry *dentry)
215 {
216 	BUG_ON(dentry->d_count);
217 	this_cpu_dec(nr_dentry);
218 	if (dentry->d_op && dentry->d_op->d_release)
219 		dentry->d_op->d_release(dentry);
220 
221 	/* if dentry was never visible to RCU, immediate free is OK */
222 	if (!(dentry->d_flags & DCACHE_RCUACCESS))
223 		__d_free(&dentry->d_u.d_rcu);
224 	else
225 		call_rcu(&dentry->d_u.d_rcu, __d_free);
226 }
227 
228 /**
229  * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups
230  * @dentry: the target dentry
231  * After this call, in-progress rcu-walk path lookup will fail. This
232  * should be called after unhashing, and after changing d_inode (if
233  * the dentry has not already been unhashed).
234  */
dentry_rcuwalk_barrier(struct dentry * dentry)235 static inline void dentry_rcuwalk_barrier(struct dentry *dentry)
236 {
237 	assert_spin_locked(&dentry->d_lock);
238 	/* Go through a barrier */
239 	write_seqcount_barrier(&dentry->d_seq);
240 }
241 
242 /*
243  * Release the dentry's inode, using the filesystem
244  * d_iput() operation if defined. Dentry has no refcount
245  * and is unhashed.
246  */
dentry_iput(struct dentry * dentry)247 static void dentry_iput(struct dentry * dentry)
248 	__releases(dentry->d_lock)
249 	__releases(dentry->d_inode->i_lock)
250 {
251 	struct inode *inode = dentry->d_inode;
252 	if (inode) {
253 		dentry->d_inode = NULL;
254 		list_del_init(&dentry->d_alias);
255 		spin_unlock(&dentry->d_lock);
256 		spin_unlock(&inode->i_lock);
257 		if (!inode->i_nlink)
258 			fsnotify_inoderemove(inode);
259 		if (dentry->d_op && dentry->d_op->d_iput)
260 			dentry->d_op->d_iput(dentry, inode);
261 		else
262 			iput(inode);
263 	} else {
264 		spin_unlock(&dentry->d_lock);
265 	}
266 }
267 
268 /*
269  * Release the dentry's inode, using the filesystem
270  * d_iput() operation if defined. dentry remains in-use.
271  */
dentry_unlink_inode(struct dentry * dentry)272 static void dentry_unlink_inode(struct dentry * dentry)
273 	__releases(dentry->d_lock)
274 	__releases(dentry->d_inode->i_lock)
275 {
276 	struct inode *inode = dentry->d_inode;
277 	dentry->d_inode = NULL;
278 	list_del_init(&dentry->d_alias);
279 	dentry_rcuwalk_barrier(dentry);
280 	spin_unlock(&dentry->d_lock);
281 	spin_unlock(&inode->i_lock);
282 	if (!inode->i_nlink)
283 		fsnotify_inoderemove(inode);
284 	if (dentry->d_op && dentry->d_op->d_iput)
285 		dentry->d_op->d_iput(dentry, inode);
286 	else
287 		iput(inode);
288 }
289 
290 /*
291  * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held.
292  */
dentry_lru_add(struct dentry * dentry)293 static void dentry_lru_add(struct dentry *dentry)
294 {
295 	if (list_empty(&dentry->d_lru)) {
296 		spin_lock(&dcache_lru_lock);
297 		list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
298 		dentry->d_sb->s_nr_dentry_unused++;
299 		dentry_stat.nr_unused++;
300 		spin_unlock(&dcache_lru_lock);
301 	}
302 }
303 
__dentry_lru_del(struct dentry * dentry)304 static void __dentry_lru_del(struct dentry *dentry)
305 {
306 	list_del_init(&dentry->d_lru);
307 	dentry->d_flags &= ~DCACHE_SHRINK_LIST;
308 	dentry->d_sb->s_nr_dentry_unused--;
309 	dentry_stat.nr_unused--;
310 }
311 
312 /*
313  * Remove a dentry with references from the LRU.
314  */
dentry_lru_del(struct dentry * dentry)315 static void dentry_lru_del(struct dentry *dentry)
316 {
317 	if (!list_empty(&dentry->d_lru)) {
318 		spin_lock(&dcache_lru_lock);
319 		__dentry_lru_del(dentry);
320 		spin_unlock(&dcache_lru_lock);
321 	}
322 }
323 
324 /*
325  * Remove a dentry that is unreferenced and about to be pruned
326  * (unhashed and destroyed) from the LRU, and inform the file system.
327  * This wrapper should be called _prior_ to unhashing a victim dentry.
328  */
dentry_lru_prune(struct dentry * dentry)329 static void dentry_lru_prune(struct dentry *dentry)
330 {
331 	if (!list_empty(&dentry->d_lru)) {
332 		if (dentry->d_flags & DCACHE_OP_PRUNE)
333 			dentry->d_op->d_prune(dentry);
334 
335 		spin_lock(&dcache_lru_lock);
336 		__dentry_lru_del(dentry);
337 		spin_unlock(&dcache_lru_lock);
338 	}
339 }
340 
dentry_lru_move_list(struct dentry * dentry,struct list_head * list)341 static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list)
342 {
343 	spin_lock(&dcache_lru_lock);
344 	if (list_empty(&dentry->d_lru)) {
345 		list_add_tail(&dentry->d_lru, list);
346 		dentry->d_sb->s_nr_dentry_unused++;
347 		dentry_stat.nr_unused++;
348 	} else {
349 		list_move_tail(&dentry->d_lru, list);
350 	}
351 	spin_unlock(&dcache_lru_lock);
352 }
353 
354 /**
355  * d_kill - kill dentry and return parent
356  * @dentry: dentry to kill
357  * @parent: parent dentry
358  *
359  * The dentry must already be unhashed and removed from the LRU.
360  *
361  * If this is the root of the dentry tree, return NULL.
362  *
363  * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by
364  * d_kill.
365  */
d_kill(struct dentry * dentry,struct dentry * parent)366 static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent)
367 	__releases(dentry->d_lock)
368 	__releases(parent->d_lock)
369 	__releases(dentry->d_inode->i_lock)
370 {
371 	list_del(&dentry->d_u.d_child);
372 	/*
373 	 * Inform try_to_ascend() that we are no longer attached to the
374 	 * dentry tree
375 	 */
376 	dentry->d_flags |= DCACHE_DENTRY_KILLED;
377 	if (parent)
378 		spin_unlock(&parent->d_lock);
379 	dentry_iput(dentry);
380 	/*
381 	 * dentry_iput drops the locks, at which point nobody (except
382 	 * transient RCU lookups) can reach this dentry.
383 	 */
384 	d_free(dentry);
385 	return parent;
386 }
387 
388 /*
389  * Unhash a dentry without inserting an RCU walk barrier or checking that
390  * dentry->d_lock is locked.  The caller must take care of that, if
391  * appropriate.
392  */
__d_shrink(struct dentry * dentry)393 static void __d_shrink(struct dentry *dentry)
394 {
395 	if (!d_unhashed(dentry)) {
396 		struct hlist_bl_head *b;
397 		if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED))
398 			b = &dentry->d_sb->s_anon;
399 		else
400 			b = d_hash(dentry->d_parent, dentry->d_name.hash);
401 
402 		hlist_bl_lock(b);
403 		__hlist_bl_del(&dentry->d_hash);
404 		dentry->d_hash.pprev = NULL;
405 		hlist_bl_unlock(b);
406 	}
407 }
408 
409 /**
410  * d_drop - drop a dentry
411  * @dentry: dentry to drop
412  *
413  * d_drop() unhashes the entry from the parent dentry hashes, so that it won't
414  * be found through a VFS lookup any more. Note that this is different from
415  * deleting the dentry - d_delete will try to mark the dentry negative if
416  * possible, giving a successful _negative_ lookup, while d_drop will
417  * just make the cache lookup fail.
418  *
419  * d_drop() is used mainly for stuff that wants to invalidate a dentry for some
420  * reason (NFS timeouts or autofs deletes).
421  *
422  * __d_drop requires dentry->d_lock.
423  */
__d_drop(struct dentry * dentry)424 void __d_drop(struct dentry *dentry)
425 {
426 	if (!d_unhashed(dentry)) {
427 		__d_shrink(dentry);
428 		dentry_rcuwalk_barrier(dentry);
429 	}
430 }
431 EXPORT_SYMBOL(__d_drop);
432 
d_drop(struct dentry * dentry)433 void d_drop(struct dentry *dentry)
434 {
435 	spin_lock(&dentry->d_lock);
436 	__d_drop(dentry);
437 	spin_unlock(&dentry->d_lock);
438 }
439 EXPORT_SYMBOL(d_drop);
440 
441 /*
442  * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag
443  * @dentry: dentry to drop
444  *
445  * This is called when we do a lookup on a placeholder dentry that needed to be
446  * looked up.  The dentry should have been hashed in order for it to be found by
447  * the lookup code, but now needs to be unhashed while we do the actual lookup
448  * and clear the DCACHE_NEED_LOOKUP flag.
449  */
d_clear_need_lookup(struct dentry * dentry)450 void d_clear_need_lookup(struct dentry *dentry)
451 {
452 	spin_lock(&dentry->d_lock);
453 	__d_drop(dentry);
454 	dentry->d_flags &= ~DCACHE_NEED_LOOKUP;
455 	spin_unlock(&dentry->d_lock);
456 }
457 EXPORT_SYMBOL(d_clear_need_lookup);
458 
459 /*
460  * Finish off a dentry we've decided to kill.
461  * dentry->d_lock must be held, returns with it unlocked.
462  * If ref is non-zero, then decrement the refcount too.
463  * Returns dentry requiring refcount drop, or NULL if we're done.
464  */
dentry_kill(struct dentry * dentry,int ref)465 static inline struct dentry *dentry_kill(struct dentry *dentry, int ref)
466 	__releases(dentry->d_lock)
467 {
468 	struct inode *inode;
469 	struct dentry *parent;
470 
471 	inode = dentry->d_inode;
472 	if (inode && !spin_trylock(&inode->i_lock)) {
473 relock:
474 		spin_unlock(&dentry->d_lock);
475 		cpu_relax();
476 		return dentry; /* try again with same dentry */
477 	}
478 	if (IS_ROOT(dentry))
479 		parent = NULL;
480 	else
481 		parent = dentry->d_parent;
482 	if (parent && !spin_trylock(&parent->d_lock)) {
483 		if (inode)
484 			spin_unlock(&inode->i_lock);
485 		goto relock;
486 	}
487 
488 	if (ref)
489 		dentry->d_count--;
490 	/*
491 	 * if dentry was on the d_lru list delete it from there.
492 	 * inform the fs via d_prune that this dentry is about to be
493 	 * unhashed and destroyed.
494 	 */
495 	dentry_lru_prune(dentry);
496 	/* if it was on the hash then remove it */
497 	__d_drop(dentry);
498 	return d_kill(dentry, parent);
499 }
500 
501 /*
502  * This is dput
503  *
504  * This is complicated by the fact that we do not want to put
505  * dentries that are no longer on any hash chain on the unused
506  * list: we'd much rather just get rid of them immediately.
507  *
508  * However, that implies that we have to traverse the dentry
509  * tree upwards to the parents which might _also_ now be
510  * scheduled for deletion (it may have been only waiting for
511  * its last child to go away).
512  *
513  * This tail recursion is done by hand as we don't want to depend
514  * on the compiler to always get this right (gcc generally doesn't).
515  * Real recursion would eat up our stack space.
516  */
517 
518 /*
519  * dput - release a dentry
520  * @dentry: dentry to release
521  *
522  * Release a dentry. This will drop the usage count and if appropriate
523  * call the dentry unlink method as well as removing it from the queues and
524  * releasing its resources. If the parent dentries were scheduled for release
525  * they too may now get deleted.
526  */
dput(struct dentry * dentry)527 void dput(struct dentry *dentry)
528 {
529 	if (!dentry)
530 		return;
531 
532 repeat:
533 	if (dentry->d_count == 1)
534 		might_sleep();
535 	spin_lock(&dentry->d_lock);
536 	BUG_ON(!dentry->d_count);
537 	if (dentry->d_count > 1) {
538 		dentry->d_count--;
539 		spin_unlock(&dentry->d_lock);
540 		return;
541 	}
542 
543 	if (dentry->d_flags & DCACHE_OP_DELETE) {
544 		if (dentry->d_op->d_delete(dentry))
545 			goto kill_it;
546 	}
547 
548 	/* Unreachable? Get rid of it */
549  	if (d_unhashed(dentry))
550 		goto kill_it;
551 
552 	/*
553 	 * If this dentry needs lookup, don't set the referenced flag so that it
554 	 * is more likely to be cleaned up by the dcache shrinker in case of
555 	 * memory pressure.
556 	 */
557 	if (!d_need_lookup(dentry))
558 		dentry->d_flags |= DCACHE_REFERENCED;
559 	dentry_lru_add(dentry);
560 
561 	dentry->d_count--;
562 	spin_unlock(&dentry->d_lock);
563 	return;
564 
565 kill_it:
566 	dentry = dentry_kill(dentry, 1);
567 	if (dentry)
568 		goto repeat;
569 }
570 EXPORT_SYMBOL(dput);
571 
572 /**
573  * d_invalidate - invalidate a dentry
574  * @dentry: dentry to invalidate
575  *
576  * Try to invalidate the dentry if it turns out to be
577  * possible. If there are other dentries that can be
578  * reached through this one we can't delete it and we
579  * return -EBUSY. On success we return 0.
580  *
581  * no dcache lock.
582  */
583 
d_invalidate(struct dentry * dentry)584 int d_invalidate(struct dentry * dentry)
585 {
586 	/*
587 	 * If it's already been dropped, return OK.
588 	 */
589 	spin_lock(&dentry->d_lock);
590 	if (d_unhashed(dentry)) {
591 		spin_unlock(&dentry->d_lock);
592 		return 0;
593 	}
594 	/*
595 	 * Check whether to do a partial shrink_dcache
596 	 * to get rid of unused child entries.
597 	 */
598 	if (!list_empty(&dentry->d_subdirs)) {
599 		spin_unlock(&dentry->d_lock);
600 		shrink_dcache_parent(dentry);
601 		spin_lock(&dentry->d_lock);
602 	}
603 
604 	/*
605 	 * Somebody else still using it?
606 	 *
607 	 * If it's a directory, we can't drop it
608 	 * for fear of somebody re-populating it
609 	 * with children (even though dropping it
610 	 * would make it unreachable from the root,
611 	 * we might still populate it if it was a
612 	 * working directory or similar).
613 	 * We also need to leave mountpoints alone,
614 	 * directory or not.
615 	 */
616 	if (dentry->d_count > 1 && dentry->d_inode) {
617 		if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) {
618 			spin_unlock(&dentry->d_lock);
619 			return -EBUSY;
620 		}
621 	}
622 
623 	__d_drop(dentry);
624 	spin_unlock(&dentry->d_lock);
625 	return 0;
626 }
627 EXPORT_SYMBOL(d_invalidate);
628 
629 /* This must be called with d_lock held */
__dget_dlock(struct dentry * dentry)630 static inline void __dget_dlock(struct dentry *dentry)
631 {
632 	dentry->d_count++;
633 }
634 
__dget(struct dentry * dentry)635 static inline void __dget(struct dentry *dentry)
636 {
637 	spin_lock(&dentry->d_lock);
638 	__dget_dlock(dentry);
639 	spin_unlock(&dentry->d_lock);
640 }
641 
dget_parent(struct dentry * dentry)642 struct dentry *dget_parent(struct dentry *dentry)
643 {
644 	struct dentry *ret;
645 
646 repeat:
647 	/*
648 	 * Don't need rcu_dereference because we re-check it was correct under
649 	 * the lock.
650 	 */
651 	rcu_read_lock();
652 	ret = dentry->d_parent;
653 	spin_lock(&ret->d_lock);
654 	if (unlikely(ret != dentry->d_parent)) {
655 		spin_unlock(&ret->d_lock);
656 		rcu_read_unlock();
657 		goto repeat;
658 	}
659 	rcu_read_unlock();
660 	BUG_ON(!ret->d_count);
661 	ret->d_count++;
662 	spin_unlock(&ret->d_lock);
663 	return ret;
664 }
665 EXPORT_SYMBOL(dget_parent);
666 
667 /**
668  * d_find_alias - grab a hashed alias of inode
669  * @inode: inode in question
670  * @want_discon:  flag, used by d_splice_alias, to request
671  *          that only a DISCONNECTED alias be returned.
672  *
673  * If inode has a hashed alias, or is a directory and has any alias,
674  * acquire the reference to alias and return it. Otherwise return NULL.
675  * Notice that if inode is a directory there can be only one alias and
676  * it can be unhashed only if it has no children, or if it is the root
677  * of a filesystem.
678  *
679  * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
680  * any other hashed alias over that one unless @want_discon is set,
681  * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
682  */
__d_find_alias(struct inode * inode,int want_discon)683 static struct dentry *__d_find_alias(struct inode *inode, int want_discon)
684 {
685 	struct dentry *alias, *discon_alias;
686 
687 again:
688 	discon_alias = NULL;
689 	list_for_each_entry(alias, &inode->i_dentry, d_alias) {
690 		spin_lock(&alias->d_lock);
691  		if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
692 			if (IS_ROOT(alias) &&
693 			    (alias->d_flags & DCACHE_DISCONNECTED)) {
694 				discon_alias = alias;
695 			} else if (!want_discon) {
696 				__dget_dlock(alias);
697 				spin_unlock(&alias->d_lock);
698 				return alias;
699 			}
700 		}
701 		spin_unlock(&alias->d_lock);
702 	}
703 	if (discon_alias) {
704 		alias = discon_alias;
705 		spin_lock(&alias->d_lock);
706 		if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
707 			if (IS_ROOT(alias) &&
708 			    (alias->d_flags & DCACHE_DISCONNECTED)) {
709 				__dget_dlock(alias);
710 				spin_unlock(&alias->d_lock);
711 				return alias;
712 			}
713 		}
714 		spin_unlock(&alias->d_lock);
715 		goto again;
716 	}
717 	return NULL;
718 }
719 
d_find_alias(struct inode * inode)720 struct dentry *d_find_alias(struct inode *inode)
721 {
722 	struct dentry *de = NULL;
723 
724 	if (!list_empty(&inode->i_dentry)) {
725 		spin_lock(&inode->i_lock);
726 		de = __d_find_alias(inode, 0);
727 		spin_unlock(&inode->i_lock);
728 	}
729 	return de;
730 }
731 EXPORT_SYMBOL(d_find_alias);
732 
733 /*
734  *	Try to kill dentries associated with this inode.
735  * WARNING: you must own a reference to inode.
736  */
d_prune_aliases(struct inode * inode)737 void d_prune_aliases(struct inode *inode)
738 {
739 	struct dentry *dentry;
740 restart:
741 	spin_lock(&inode->i_lock);
742 	list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
743 		spin_lock(&dentry->d_lock);
744 		if (!dentry->d_count) {
745 			__dget_dlock(dentry);
746 			__d_drop(dentry);
747 			spin_unlock(&dentry->d_lock);
748 			spin_unlock(&inode->i_lock);
749 			dput(dentry);
750 			goto restart;
751 		}
752 		spin_unlock(&dentry->d_lock);
753 	}
754 	spin_unlock(&inode->i_lock);
755 }
756 EXPORT_SYMBOL(d_prune_aliases);
757 
758 /*
759  * Try to throw away a dentry - free the inode, dput the parent.
760  * Requires dentry->d_lock is held, and dentry->d_count == 0.
761  * Releases dentry->d_lock.
762  *
763  * This may fail if locks cannot be acquired no problem, just try again.
764  */
try_prune_one_dentry(struct dentry * dentry)765 static void try_prune_one_dentry(struct dentry *dentry)
766 	__releases(dentry->d_lock)
767 {
768 	struct dentry *parent;
769 
770 	parent = dentry_kill(dentry, 0);
771 	/*
772 	 * If dentry_kill returns NULL, we have nothing more to do.
773 	 * if it returns the same dentry, trylocks failed. In either
774 	 * case, just loop again.
775 	 *
776 	 * Otherwise, we need to prune ancestors too. This is necessary
777 	 * to prevent quadratic behavior of shrink_dcache_parent(), but
778 	 * is also expected to be beneficial in reducing dentry cache
779 	 * fragmentation.
780 	 */
781 	if (!parent)
782 		return;
783 	if (parent == dentry)
784 		return;
785 
786 	/* Prune ancestors. */
787 	dentry = parent;
788 	while (dentry) {
789 		spin_lock(&dentry->d_lock);
790 		if (dentry->d_count > 1) {
791 			dentry->d_count--;
792 			spin_unlock(&dentry->d_lock);
793 			return;
794 		}
795 		dentry = dentry_kill(dentry, 1);
796 	}
797 }
798 
shrink_dentry_list(struct list_head * list)799 static void shrink_dentry_list(struct list_head *list)
800 {
801 	struct dentry *dentry;
802 
803 	rcu_read_lock();
804 	for (;;) {
805 		dentry = list_entry_rcu(list->prev, struct dentry, d_lru);
806 		if (&dentry->d_lru == list)
807 			break; /* empty */
808 		spin_lock(&dentry->d_lock);
809 		if (dentry != list_entry(list->prev, struct dentry, d_lru)) {
810 			spin_unlock(&dentry->d_lock);
811 			continue;
812 		}
813 
814 		/*
815 		 * We found an inuse dentry which was not removed from
816 		 * the LRU because of laziness during lookup.  Do not free
817 		 * it - just keep it off the LRU list.
818 		 */
819 		if (dentry->d_count) {
820 			dentry_lru_del(dentry);
821 			spin_unlock(&dentry->d_lock);
822 			continue;
823 		}
824 
825 		rcu_read_unlock();
826 
827 		try_prune_one_dentry(dentry);
828 
829 		rcu_read_lock();
830 	}
831 	rcu_read_unlock();
832 }
833 
834 /**
835  * prune_dcache_sb - shrink the dcache
836  * @sb: superblock
837  * @count: number of entries to try to free
838  *
839  * Attempt to shrink the superblock dcache LRU by @count entries. This is
840  * done when we need more memory an called from the superblock shrinker
841  * function.
842  *
843  * This function may fail to free any resources if all the dentries are in
844  * use.
845  */
prune_dcache_sb(struct super_block * sb,int count)846 void prune_dcache_sb(struct super_block *sb, int count)
847 {
848 	struct dentry *dentry;
849 	LIST_HEAD(referenced);
850 	LIST_HEAD(tmp);
851 
852 relock:
853 	spin_lock(&dcache_lru_lock);
854 	while (!list_empty(&sb->s_dentry_lru)) {
855 		dentry = list_entry(sb->s_dentry_lru.prev,
856 				struct dentry, d_lru);
857 		BUG_ON(dentry->d_sb != sb);
858 
859 		if (!spin_trylock(&dentry->d_lock)) {
860 			spin_unlock(&dcache_lru_lock);
861 			cpu_relax();
862 			goto relock;
863 		}
864 
865 		if (dentry->d_flags & DCACHE_REFERENCED) {
866 			dentry->d_flags &= ~DCACHE_REFERENCED;
867 			list_move(&dentry->d_lru, &referenced);
868 			spin_unlock(&dentry->d_lock);
869 		} else {
870 			list_move_tail(&dentry->d_lru, &tmp);
871 			dentry->d_flags |= DCACHE_SHRINK_LIST;
872 			spin_unlock(&dentry->d_lock);
873 			if (!--count)
874 				break;
875 		}
876 		cond_resched_lock(&dcache_lru_lock);
877 	}
878 	if (!list_empty(&referenced))
879 		list_splice(&referenced, &sb->s_dentry_lru);
880 	spin_unlock(&dcache_lru_lock);
881 
882 	shrink_dentry_list(&tmp);
883 }
884 
885 /**
886  * shrink_dcache_sb - shrink dcache for a superblock
887  * @sb: superblock
888  *
889  * Shrink the dcache for the specified super block. This is used to free
890  * the dcache before unmounting a file system.
891  */
shrink_dcache_sb(struct super_block * sb)892 void shrink_dcache_sb(struct super_block *sb)
893 {
894 	LIST_HEAD(tmp);
895 
896 	spin_lock(&dcache_lru_lock);
897 	while (!list_empty(&sb->s_dentry_lru)) {
898 		list_splice_init(&sb->s_dentry_lru, &tmp);
899 		spin_unlock(&dcache_lru_lock);
900 		shrink_dentry_list(&tmp);
901 		spin_lock(&dcache_lru_lock);
902 	}
903 	spin_unlock(&dcache_lru_lock);
904 }
905 EXPORT_SYMBOL(shrink_dcache_sb);
906 
907 /*
908  * destroy a single subtree of dentries for unmount
909  * - see the comments on shrink_dcache_for_umount() for a description of the
910  *   locking
911  */
shrink_dcache_for_umount_subtree(struct dentry * dentry)912 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
913 {
914 	struct dentry *parent;
915 
916 	BUG_ON(!IS_ROOT(dentry));
917 
918 	for (;;) {
919 		/* descend to the first leaf in the current subtree */
920 		while (!list_empty(&dentry->d_subdirs))
921 			dentry = list_entry(dentry->d_subdirs.next,
922 					    struct dentry, d_u.d_child);
923 
924 		/* consume the dentries from this leaf up through its parents
925 		 * until we find one with children or run out altogether */
926 		do {
927 			struct inode *inode;
928 
929 			/*
930 			 * remove the dentry from the lru, and inform
931 			 * the fs that this dentry is about to be
932 			 * unhashed and destroyed.
933 			 */
934 			dentry_lru_prune(dentry);
935 			__d_shrink(dentry);
936 
937 			if (dentry->d_count != 0) {
938 				printk(KERN_ERR
939 				       "BUG: Dentry %p{i=%lx,n=%s}"
940 				       " still in use (%d)"
941 				       " [unmount of %s %s]\n",
942 				       dentry,
943 				       dentry->d_inode ?
944 				       dentry->d_inode->i_ino : 0UL,
945 				       dentry->d_name.name,
946 				       dentry->d_count,
947 				       dentry->d_sb->s_type->name,
948 				       dentry->d_sb->s_id);
949 				BUG();
950 			}
951 
952 			if (IS_ROOT(dentry)) {
953 				parent = NULL;
954 				list_del(&dentry->d_u.d_child);
955 			} else {
956 				parent = dentry->d_parent;
957 				parent->d_count--;
958 				list_del(&dentry->d_u.d_child);
959 			}
960 
961 			inode = dentry->d_inode;
962 			if (inode) {
963 				dentry->d_inode = NULL;
964 				list_del_init(&dentry->d_alias);
965 				if (dentry->d_op && dentry->d_op->d_iput)
966 					dentry->d_op->d_iput(dentry, inode);
967 				else
968 					iput(inode);
969 			}
970 
971 			d_free(dentry);
972 
973 			/* finished when we fall off the top of the tree,
974 			 * otherwise we ascend to the parent and move to the
975 			 * next sibling if there is one */
976 			if (!parent)
977 				return;
978 			dentry = parent;
979 		} while (list_empty(&dentry->d_subdirs));
980 
981 		dentry = list_entry(dentry->d_subdirs.next,
982 				    struct dentry, d_u.d_child);
983 	}
984 }
985 
986 /*
987  * destroy the dentries attached to a superblock on unmounting
988  * - we don't need to use dentry->d_lock because:
989  *   - the superblock is detached from all mountings and open files, so the
990  *     dentry trees will not be rearranged by the VFS
991  *   - s_umount is write-locked, so the memory pressure shrinker will ignore
992  *     any dentries belonging to this superblock that it comes across
993  *   - the filesystem itself is no longer permitted to rearrange the dentries
994  *     in this superblock
995  */
shrink_dcache_for_umount(struct super_block * sb)996 void shrink_dcache_for_umount(struct super_block *sb)
997 {
998 	struct dentry *dentry;
999 
1000 	if (down_read_trylock(&sb->s_umount))
1001 		BUG();
1002 
1003 	dentry = sb->s_root;
1004 	sb->s_root = NULL;
1005 	dentry->d_count--;
1006 	shrink_dcache_for_umount_subtree(dentry);
1007 
1008 	while (!hlist_bl_empty(&sb->s_anon)) {
1009 		dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash);
1010 		shrink_dcache_for_umount_subtree(dentry);
1011 	}
1012 }
1013 
1014 /*
1015  * This tries to ascend one level of parenthood, but
1016  * we can race with renaming, so we need to re-check
1017  * the parenthood after dropping the lock and check
1018  * that the sequence number still matches.
1019  */
try_to_ascend(struct dentry * old,int locked,unsigned seq)1020 static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq)
1021 {
1022 	struct dentry *new = old->d_parent;
1023 
1024 	rcu_read_lock();
1025 	spin_unlock(&old->d_lock);
1026 	spin_lock(&new->d_lock);
1027 
1028 	/*
1029 	 * might go back up the wrong parent if we have had a rename
1030 	 * or deletion
1031 	 */
1032 	if (new != old->d_parent ||
1033 		 (old->d_flags & DCACHE_DENTRY_KILLED) ||
1034 		 (!locked && read_seqretry(&rename_lock, seq))) {
1035 		spin_unlock(&new->d_lock);
1036 		new = NULL;
1037 	}
1038 	rcu_read_unlock();
1039 	return new;
1040 }
1041 
1042 
1043 /*
1044  * Search for at least 1 mount point in the dentry's subdirs.
1045  * We descend to the next level whenever the d_subdirs
1046  * list is non-empty and continue searching.
1047  */
1048 
1049 /**
1050  * have_submounts - check for mounts over a dentry
1051  * @parent: dentry to check.
1052  *
1053  * Return true if the parent or its subdirectories contain
1054  * a mount point
1055  */
have_submounts(struct dentry * parent)1056 int have_submounts(struct dentry *parent)
1057 {
1058 	struct dentry *this_parent;
1059 	struct list_head *next;
1060 	unsigned seq;
1061 	int locked = 0;
1062 
1063 	seq = read_seqbegin(&rename_lock);
1064 again:
1065 	this_parent = parent;
1066 
1067 	if (d_mountpoint(parent))
1068 		goto positive;
1069 	spin_lock(&this_parent->d_lock);
1070 repeat:
1071 	next = this_parent->d_subdirs.next;
1072 resume:
1073 	while (next != &this_parent->d_subdirs) {
1074 		struct list_head *tmp = next;
1075 		struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1076 		next = tmp->next;
1077 
1078 		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1079 		/* Have we found a mount point ? */
1080 		if (d_mountpoint(dentry)) {
1081 			spin_unlock(&dentry->d_lock);
1082 			spin_unlock(&this_parent->d_lock);
1083 			goto positive;
1084 		}
1085 		if (!list_empty(&dentry->d_subdirs)) {
1086 			spin_unlock(&this_parent->d_lock);
1087 			spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1088 			this_parent = dentry;
1089 			spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1090 			goto repeat;
1091 		}
1092 		spin_unlock(&dentry->d_lock);
1093 	}
1094 	/*
1095 	 * All done at this level ... ascend and resume the search.
1096 	 */
1097 	if (this_parent != parent) {
1098 		struct dentry *child = this_parent;
1099 		this_parent = try_to_ascend(this_parent, locked, seq);
1100 		if (!this_parent)
1101 			goto rename_retry;
1102 		next = child->d_u.d_child.next;
1103 		goto resume;
1104 	}
1105 	spin_unlock(&this_parent->d_lock);
1106 	if (!locked && read_seqretry(&rename_lock, seq))
1107 		goto rename_retry;
1108 	if (locked)
1109 		write_sequnlock(&rename_lock);
1110 	return 0; /* No mount points found in tree */
1111 positive:
1112 	if (!locked && read_seqretry(&rename_lock, seq))
1113 		goto rename_retry;
1114 	if (locked)
1115 		write_sequnlock(&rename_lock);
1116 	return 1;
1117 
1118 rename_retry:
1119 	if (locked)
1120 		goto again;
1121 	locked = 1;
1122 	write_seqlock(&rename_lock);
1123 	goto again;
1124 }
1125 EXPORT_SYMBOL(have_submounts);
1126 
1127 /*
1128  * Search the dentry child list for the specified parent,
1129  * and move any unused dentries to the end of the unused
1130  * list for prune_dcache(). We descend to the next level
1131  * whenever the d_subdirs list is non-empty and continue
1132  * searching.
1133  *
1134  * It returns zero iff there are no unused children,
1135  * otherwise  it returns the number of children moved to
1136  * the end of the unused list. This may not be the total
1137  * number of unused children, because select_parent can
1138  * drop the lock and return early due to latency
1139  * constraints.
1140  */
select_parent(struct dentry * parent,struct list_head * dispose)1141 static int select_parent(struct dentry *parent, struct list_head *dispose)
1142 {
1143 	struct dentry *this_parent;
1144 	struct list_head *next;
1145 	unsigned seq;
1146 	int found = 0;
1147 	int locked = 0;
1148 
1149 	seq = read_seqbegin(&rename_lock);
1150 again:
1151 	this_parent = parent;
1152 	spin_lock(&this_parent->d_lock);
1153 repeat:
1154 	next = this_parent->d_subdirs.next;
1155 resume:
1156 	while (next != &this_parent->d_subdirs) {
1157 		struct list_head *tmp = next;
1158 		struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1159 		next = tmp->next;
1160 
1161 		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1162 
1163 		/*
1164 		 * move only zero ref count dentries to the dispose list.
1165 		 *
1166 		 * Those which are presently on the shrink list, being processed
1167 		 * by shrink_dentry_list(), shouldn't be moved.  Otherwise the
1168 		 * loop in shrink_dcache_parent() might not make any progress
1169 		 * and loop forever.
1170 		 */
1171 		if (dentry->d_count) {
1172 			dentry_lru_del(dentry);
1173 		} else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1174 			dentry_lru_move_list(dentry, dispose);
1175 			dentry->d_flags |= DCACHE_SHRINK_LIST;
1176 			found++;
1177 		}
1178 		/*
1179 		 * We can return to the caller if we have found some (this
1180 		 * ensures forward progress). We'll be coming back to find
1181 		 * the rest.
1182 		 */
1183 		if (found && need_resched()) {
1184 			spin_unlock(&dentry->d_lock);
1185 			goto out;
1186 		}
1187 
1188 		/*
1189 		 * Descend a level if the d_subdirs list is non-empty.
1190 		 */
1191 		if (!list_empty(&dentry->d_subdirs)) {
1192 			spin_unlock(&this_parent->d_lock);
1193 			spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1194 			this_parent = dentry;
1195 			spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1196 			goto repeat;
1197 		}
1198 
1199 		spin_unlock(&dentry->d_lock);
1200 	}
1201 	/*
1202 	 * All done at this level ... ascend and resume the search.
1203 	 */
1204 	if (this_parent != parent) {
1205 		struct dentry *child = this_parent;
1206 		this_parent = try_to_ascend(this_parent, locked, seq);
1207 		if (!this_parent)
1208 			goto rename_retry;
1209 		next = child->d_u.d_child.next;
1210 		goto resume;
1211 	}
1212 out:
1213 	spin_unlock(&this_parent->d_lock);
1214 	if (!locked && read_seqretry(&rename_lock, seq))
1215 		goto rename_retry;
1216 	if (locked)
1217 		write_sequnlock(&rename_lock);
1218 	return found;
1219 
1220 rename_retry:
1221 	if (found)
1222 		return found;
1223 	if (locked)
1224 		goto again;
1225 	locked = 1;
1226 	write_seqlock(&rename_lock);
1227 	goto again;
1228 }
1229 
1230 /**
1231  * shrink_dcache_parent - prune dcache
1232  * @parent: parent of entries to prune
1233  *
1234  * Prune the dcache to remove unused children of the parent dentry.
1235  */
shrink_dcache_parent(struct dentry * parent)1236 void shrink_dcache_parent(struct dentry * parent)
1237 {
1238 	LIST_HEAD(dispose);
1239 	int found;
1240 
1241 	while ((found = select_parent(parent, &dispose)) != 0) {
1242 		shrink_dentry_list(&dispose);
1243 		cond_resched();
1244 	}
1245 }
1246 EXPORT_SYMBOL(shrink_dcache_parent);
1247 
1248 /**
1249  * __d_alloc	-	allocate a dcache entry
1250  * @sb: filesystem it will belong to
1251  * @name: qstr of the name
1252  *
1253  * Allocates a dentry. It returns %NULL if there is insufficient memory
1254  * available. On a success the dentry is returned. The name passed in is
1255  * copied and the copy passed in may be reused after this call.
1256  */
1257 
__d_alloc(struct super_block * sb,const struct qstr * name)1258 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1259 {
1260 	struct dentry *dentry;
1261 	char *dname;
1262 
1263 	dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1264 	if (!dentry)
1265 		return NULL;
1266 
1267 	if (name->len > DNAME_INLINE_LEN-1) {
1268 		dname = kmalloc(name->len + 1, GFP_KERNEL);
1269 		if (!dname) {
1270 			kmem_cache_free(dentry_cache, dentry);
1271 			return NULL;
1272 		}
1273 	} else  {
1274 		dname = dentry->d_iname;
1275 	}
1276 	dentry->d_name.name = dname;
1277 
1278 	dentry->d_name.len = name->len;
1279 	dentry->d_name.hash = name->hash;
1280 	memcpy(dname, name->name, name->len);
1281 	dname[name->len] = 0;
1282 
1283 	dentry->d_count = 1;
1284 	dentry->d_flags = 0;
1285 	spin_lock_init(&dentry->d_lock);
1286 	seqcount_init(&dentry->d_seq);
1287 	dentry->d_inode = NULL;
1288 	dentry->d_parent = dentry;
1289 	dentry->d_sb = sb;
1290 	dentry->d_op = NULL;
1291 	dentry->d_fsdata = NULL;
1292 	INIT_HLIST_BL_NODE(&dentry->d_hash);
1293 	INIT_LIST_HEAD(&dentry->d_lru);
1294 	INIT_LIST_HEAD(&dentry->d_subdirs);
1295 	INIT_LIST_HEAD(&dentry->d_alias);
1296 	INIT_LIST_HEAD(&dentry->d_u.d_child);
1297 	d_set_d_op(dentry, dentry->d_sb->s_d_op);
1298 
1299 	this_cpu_inc(nr_dentry);
1300 
1301 	return dentry;
1302 }
1303 
1304 /**
1305  * d_alloc	-	allocate a dcache entry
1306  * @parent: parent of entry to allocate
1307  * @name: qstr of the name
1308  *
1309  * Allocates a dentry. It returns %NULL if there is insufficient memory
1310  * available. On a success the dentry is returned. The name passed in is
1311  * copied and the copy passed in may be reused after this call.
1312  */
d_alloc(struct dentry * parent,const struct qstr * name)1313 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1314 {
1315 	struct dentry *dentry = __d_alloc(parent->d_sb, name);
1316 	if (!dentry)
1317 		return NULL;
1318 
1319 	spin_lock(&parent->d_lock);
1320 	/*
1321 	 * don't need child lock because it is not subject
1322 	 * to concurrency here
1323 	 */
1324 	__dget_dlock(parent);
1325 	dentry->d_parent = parent;
1326 	list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1327 	spin_unlock(&parent->d_lock);
1328 
1329 	return dentry;
1330 }
1331 EXPORT_SYMBOL(d_alloc);
1332 
d_alloc_pseudo(struct super_block * sb,const struct qstr * name)1333 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1334 {
1335 	struct dentry *dentry = __d_alloc(sb, name);
1336 	if (dentry)
1337 		dentry->d_flags |= DCACHE_DISCONNECTED;
1338 	return dentry;
1339 }
1340 EXPORT_SYMBOL(d_alloc_pseudo);
1341 
d_alloc_name(struct dentry * parent,const char * name)1342 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1343 {
1344 	struct qstr q;
1345 
1346 	q.name = name;
1347 	q.len = strlen(name);
1348 	q.hash = full_name_hash(q.name, q.len);
1349 	return d_alloc(parent, &q);
1350 }
1351 EXPORT_SYMBOL(d_alloc_name);
1352 
d_set_d_op(struct dentry * dentry,const struct dentry_operations * op)1353 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1354 {
1355 	WARN_ON_ONCE(dentry->d_op);
1356 	WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH	|
1357 				DCACHE_OP_COMPARE	|
1358 				DCACHE_OP_REVALIDATE	|
1359 				DCACHE_OP_DELETE ));
1360 	dentry->d_op = op;
1361 	if (!op)
1362 		return;
1363 	if (op->d_hash)
1364 		dentry->d_flags |= DCACHE_OP_HASH;
1365 	if (op->d_compare)
1366 		dentry->d_flags |= DCACHE_OP_COMPARE;
1367 	if (op->d_revalidate)
1368 		dentry->d_flags |= DCACHE_OP_REVALIDATE;
1369 	if (op->d_delete)
1370 		dentry->d_flags |= DCACHE_OP_DELETE;
1371 	if (op->d_prune)
1372 		dentry->d_flags |= DCACHE_OP_PRUNE;
1373 
1374 }
1375 EXPORT_SYMBOL(d_set_d_op);
1376 
__d_instantiate(struct dentry * dentry,struct inode * inode)1377 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1378 {
1379 	spin_lock(&dentry->d_lock);
1380 	if (inode) {
1381 		if (unlikely(IS_AUTOMOUNT(inode)))
1382 			dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1383 		list_add(&dentry->d_alias, &inode->i_dentry);
1384 	}
1385 	dentry->d_inode = inode;
1386 	dentry_rcuwalk_barrier(dentry);
1387 	spin_unlock(&dentry->d_lock);
1388 	fsnotify_d_instantiate(dentry, inode);
1389 }
1390 
1391 /**
1392  * d_instantiate - fill in inode information for a dentry
1393  * @entry: dentry to complete
1394  * @inode: inode to attach to this dentry
1395  *
1396  * Fill in inode information in the entry.
1397  *
1398  * This turns negative dentries into productive full members
1399  * of society.
1400  *
1401  * NOTE! This assumes that the inode count has been incremented
1402  * (or otherwise set) by the caller to indicate that it is now
1403  * in use by the dcache.
1404  */
1405 
d_instantiate(struct dentry * entry,struct inode * inode)1406 void d_instantiate(struct dentry *entry, struct inode * inode)
1407 {
1408 	BUG_ON(!list_empty(&entry->d_alias));
1409 	if (inode)
1410 		spin_lock(&inode->i_lock);
1411 	__d_instantiate(entry, inode);
1412 	if (inode)
1413 		spin_unlock(&inode->i_lock);
1414 	security_d_instantiate(entry, inode);
1415 }
1416 EXPORT_SYMBOL(d_instantiate);
1417 
1418 /**
1419  * d_instantiate_unique - instantiate a non-aliased dentry
1420  * @entry: dentry to instantiate
1421  * @inode: inode to attach to this dentry
1422  *
1423  * Fill in inode information in the entry. On success, it returns NULL.
1424  * If an unhashed alias of "entry" already exists, then we return the
1425  * aliased dentry instead and drop one reference to inode.
1426  *
1427  * Note that in order to avoid conflicts with rename() etc, the caller
1428  * had better be holding the parent directory semaphore.
1429  *
1430  * This also assumes that the inode count has been incremented
1431  * (or otherwise set) by the caller to indicate that it is now
1432  * in use by the dcache.
1433  */
__d_instantiate_unique(struct dentry * entry,struct inode * inode)1434 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1435 					     struct inode *inode)
1436 {
1437 	struct dentry *alias;
1438 	int len = entry->d_name.len;
1439 	const char *name = entry->d_name.name;
1440 	unsigned int hash = entry->d_name.hash;
1441 
1442 	if (!inode) {
1443 		__d_instantiate(entry, NULL);
1444 		return NULL;
1445 	}
1446 
1447 	list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1448 		struct qstr *qstr = &alias->d_name;
1449 
1450 		/*
1451 		 * Don't need alias->d_lock here, because aliases with
1452 		 * d_parent == entry->d_parent are not subject to name or
1453 		 * parent changes, because the parent inode i_mutex is held.
1454 		 */
1455 		if (qstr->hash != hash)
1456 			continue;
1457 		if (alias->d_parent != entry->d_parent)
1458 			continue;
1459 		if (dentry_cmp(qstr->name, qstr->len, name, len))
1460 			continue;
1461 		__dget(alias);
1462 		return alias;
1463 	}
1464 
1465 	__d_instantiate(entry, inode);
1466 	return NULL;
1467 }
1468 
d_instantiate_unique(struct dentry * entry,struct inode * inode)1469 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1470 {
1471 	struct dentry *result;
1472 
1473 	BUG_ON(!list_empty(&entry->d_alias));
1474 
1475 	if (inode)
1476 		spin_lock(&inode->i_lock);
1477 	result = __d_instantiate_unique(entry, inode);
1478 	if (inode)
1479 		spin_unlock(&inode->i_lock);
1480 
1481 	if (!result) {
1482 		security_d_instantiate(entry, inode);
1483 		return NULL;
1484 	}
1485 
1486 	BUG_ON(!d_unhashed(result));
1487 	iput(inode);
1488 	return result;
1489 }
1490 
1491 EXPORT_SYMBOL(d_instantiate_unique);
1492 
d_make_root(struct inode * root_inode)1493 struct dentry *d_make_root(struct inode *root_inode)
1494 {
1495 	struct dentry *res = NULL;
1496 
1497 	if (root_inode) {
1498 		static const struct qstr name = { .name = "/", .len = 1 };
1499 
1500 		res = __d_alloc(root_inode->i_sb, &name);
1501 		if (res)
1502 			d_instantiate(res, root_inode);
1503 		else
1504 			iput(root_inode);
1505 	}
1506 	return res;
1507 }
1508 EXPORT_SYMBOL(d_make_root);
1509 
__d_find_any_alias(struct inode * inode)1510 static struct dentry * __d_find_any_alias(struct inode *inode)
1511 {
1512 	struct dentry *alias;
1513 
1514 	if (list_empty(&inode->i_dentry))
1515 		return NULL;
1516 	alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1517 	__dget(alias);
1518 	return alias;
1519 }
1520 
1521 /**
1522  * d_find_any_alias - find any alias for a given inode
1523  * @inode: inode to find an alias for
1524  *
1525  * If any aliases exist for the given inode, take and return a
1526  * reference for one of them.  If no aliases exist, return %NULL.
1527  */
d_find_any_alias(struct inode * inode)1528 struct dentry *d_find_any_alias(struct inode *inode)
1529 {
1530 	struct dentry *de;
1531 
1532 	spin_lock(&inode->i_lock);
1533 	de = __d_find_any_alias(inode);
1534 	spin_unlock(&inode->i_lock);
1535 	return de;
1536 }
1537 EXPORT_SYMBOL(d_find_any_alias);
1538 
1539 /**
1540  * d_obtain_alias - find or allocate a dentry for a given inode
1541  * @inode: inode to allocate the dentry for
1542  *
1543  * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1544  * similar open by handle operations.  The returned dentry may be anonymous,
1545  * or may have a full name (if the inode was already in the cache).
1546  *
1547  * When called on a directory inode, we must ensure that the inode only ever
1548  * has one dentry.  If a dentry is found, that is returned instead of
1549  * allocating a new one.
1550  *
1551  * On successful return, the reference to the inode has been transferred
1552  * to the dentry.  In case of an error the reference on the inode is released.
1553  * To make it easier to use in export operations a %NULL or IS_ERR inode may
1554  * be passed in and will be the error will be propagate to the return value,
1555  * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1556  */
d_obtain_alias(struct inode * inode)1557 struct dentry *d_obtain_alias(struct inode *inode)
1558 {
1559 	static const struct qstr anonstring = { .name = "/", .len = 1 };
1560 	struct dentry *tmp;
1561 	struct dentry *res;
1562 
1563 	if (!inode)
1564 		return ERR_PTR(-ESTALE);
1565 	if (IS_ERR(inode))
1566 		return ERR_CAST(inode);
1567 
1568 	res = d_find_any_alias(inode);
1569 	if (res)
1570 		goto out_iput;
1571 
1572 	tmp = __d_alloc(inode->i_sb, &anonstring);
1573 	if (!tmp) {
1574 		res = ERR_PTR(-ENOMEM);
1575 		goto out_iput;
1576 	}
1577 
1578 	spin_lock(&inode->i_lock);
1579 	res = __d_find_any_alias(inode);
1580 	if (res) {
1581 		spin_unlock(&inode->i_lock);
1582 		dput(tmp);
1583 		goto out_iput;
1584 	}
1585 
1586 	/* attach a disconnected dentry */
1587 	spin_lock(&tmp->d_lock);
1588 	tmp->d_inode = inode;
1589 	tmp->d_flags |= DCACHE_DISCONNECTED;
1590 	list_add(&tmp->d_alias, &inode->i_dentry);
1591 	hlist_bl_lock(&tmp->d_sb->s_anon);
1592 	hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1593 	hlist_bl_unlock(&tmp->d_sb->s_anon);
1594 	spin_unlock(&tmp->d_lock);
1595 	spin_unlock(&inode->i_lock);
1596 	security_d_instantiate(tmp, inode);
1597 
1598 	return tmp;
1599 
1600  out_iput:
1601 	if (res && !IS_ERR(res))
1602 		security_d_instantiate(res, inode);
1603 	iput(inode);
1604 	return res;
1605 }
1606 EXPORT_SYMBOL(d_obtain_alias);
1607 
1608 /**
1609  * d_splice_alias - splice a disconnected dentry into the tree if one exists
1610  * @inode:  the inode which may have a disconnected dentry
1611  * @dentry: a negative dentry which we want to point to the inode.
1612  *
1613  * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1614  * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1615  * and return it, else simply d_add the inode to the dentry and return NULL.
1616  *
1617  * This is needed in the lookup routine of any filesystem that is exportable
1618  * (via knfsd) so that we can build dcache paths to directories effectively.
1619  *
1620  * If a dentry was found and moved, then it is returned.  Otherwise NULL
1621  * is returned.  This matches the expected return value of ->lookup.
1622  *
1623  */
d_splice_alias(struct inode * inode,struct dentry * dentry)1624 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1625 {
1626 	struct dentry *new = NULL;
1627 
1628 	if (IS_ERR(inode))
1629 		return ERR_CAST(inode);
1630 
1631 	if (inode && S_ISDIR(inode->i_mode)) {
1632 		spin_lock(&inode->i_lock);
1633 		new = __d_find_alias(inode, 1);
1634 		if (new) {
1635 			BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1636 			spin_unlock(&inode->i_lock);
1637 			security_d_instantiate(new, inode);
1638 			d_move(new, dentry);
1639 			iput(inode);
1640 		} else {
1641 			/* already taking inode->i_lock, so d_add() by hand */
1642 			__d_instantiate(dentry, inode);
1643 			spin_unlock(&inode->i_lock);
1644 			security_d_instantiate(dentry, inode);
1645 			d_rehash(dentry);
1646 		}
1647 	} else
1648 		d_add(dentry, inode);
1649 	return new;
1650 }
1651 EXPORT_SYMBOL(d_splice_alias);
1652 
1653 /**
1654  * d_add_ci - lookup or allocate new dentry with case-exact name
1655  * @inode:  the inode case-insensitive lookup has found
1656  * @dentry: the negative dentry that was passed to the parent's lookup func
1657  * @name:   the case-exact name to be associated with the returned dentry
1658  *
1659  * This is to avoid filling the dcache with case-insensitive names to the
1660  * same inode, only the actual correct case is stored in the dcache for
1661  * case-insensitive filesystems.
1662  *
1663  * For a case-insensitive lookup match and if the the case-exact dentry
1664  * already exists in in the dcache, use it and return it.
1665  *
1666  * If no entry exists with the exact case name, allocate new dentry with
1667  * the exact case, and return the spliced entry.
1668  */
d_add_ci(struct dentry * dentry,struct inode * inode,struct qstr * name)1669 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1670 			struct qstr *name)
1671 {
1672 	int error;
1673 	struct dentry *found;
1674 	struct dentry *new;
1675 
1676 	/*
1677 	 * First check if a dentry matching the name already exists,
1678 	 * if not go ahead and create it now.
1679 	 */
1680 	found = d_hash_and_lookup(dentry->d_parent, name);
1681 	if (!found) {
1682 		new = d_alloc(dentry->d_parent, name);
1683 		if (!new) {
1684 			error = -ENOMEM;
1685 			goto err_out;
1686 		}
1687 
1688 		found = d_splice_alias(inode, new);
1689 		if (found) {
1690 			dput(new);
1691 			return found;
1692 		}
1693 		return new;
1694 	}
1695 
1696 	/*
1697 	 * If a matching dentry exists, and it's not negative use it.
1698 	 *
1699 	 * Decrement the reference count to balance the iget() done
1700 	 * earlier on.
1701 	 */
1702 	if (found->d_inode) {
1703 		if (unlikely(found->d_inode != inode)) {
1704 			/* This can't happen because bad inodes are unhashed. */
1705 			BUG_ON(!is_bad_inode(inode));
1706 			BUG_ON(!is_bad_inode(found->d_inode));
1707 		}
1708 		iput(inode);
1709 		return found;
1710 	}
1711 
1712 	/*
1713 	 * We are going to instantiate this dentry, unhash it and clear the
1714 	 * lookup flag so we can do that.
1715 	 */
1716 	if (unlikely(d_need_lookup(found)))
1717 		d_clear_need_lookup(found);
1718 
1719 	/*
1720 	 * Negative dentry: instantiate it unless the inode is a directory and
1721 	 * already has a dentry.
1722 	 */
1723 	new = d_splice_alias(inode, found);
1724 	if (new) {
1725 		dput(found);
1726 		found = new;
1727 	}
1728 	return found;
1729 
1730 err_out:
1731 	iput(inode);
1732 	return ERR_PTR(error);
1733 }
1734 EXPORT_SYMBOL(d_add_ci);
1735 
1736 /**
1737  * __d_lookup_rcu - search for a dentry (racy, store-free)
1738  * @parent: parent dentry
1739  * @name: qstr of name we wish to find
1740  * @seqp: returns d_seq value at the point where the dentry was found
1741  * @inode: returns dentry->d_inode when the inode was found valid.
1742  * Returns: dentry, or NULL
1743  *
1744  * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1745  * resolution (store-free path walking) design described in
1746  * Documentation/filesystems/path-lookup.txt.
1747  *
1748  * This is not to be used outside core vfs.
1749  *
1750  * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1751  * held, and rcu_read_lock held. The returned dentry must not be stored into
1752  * without taking d_lock and checking d_seq sequence count against @seq
1753  * returned here.
1754  *
1755  * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1756  * function.
1757  *
1758  * Alternatively, __d_lookup_rcu may be called again to look up the child of
1759  * the returned dentry, so long as its parent's seqlock is checked after the
1760  * child is looked up. Thus, an interlocking stepping of sequence lock checks
1761  * is formed, giving integrity down the path walk.
1762  */
__d_lookup_rcu(const struct dentry * parent,const struct qstr * name,unsigned * seqp,struct inode ** inode)1763 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1764 				const struct qstr *name,
1765 				unsigned *seqp, struct inode **inode)
1766 {
1767 	unsigned int len = name->len;
1768 	unsigned int hash = name->hash;
1769 	const unsigned char *str = name->name;
1770 	struct hlist_bl_head *b = d_hash(parent, hash);
1771 	struct hlist_bl_node *node;
1772 	struct dentry *dentry;
1773 
1774 	/*
1775 	 * Note: There is significant duplication with __d_lookup_rcu which is
1776 	 * required to prevent single threaded performance regressions
1777 	 * especially on architectures where smp_rmb (in seqcounts) are costly.
1778 	 * Keep the two functions in sync.
1779 	 */
1780 
1781 	/*
1782 	 * The hash list is protected using RCU.
1783 	 *
1784 	 * Carefully use d_seq when comparing a candidate dentry, to avoid
1785 	 * races with d_move().
1786 	 *
1787 	 * It is possible that concurrent renames can mess up our list
1788 	 * walk here and result in missing our dentry, resulting in the
1789 	 * false-negative result. d_lookup() protects against concurrent
1790 	 * renames using rename_lock seqlock.
1791 	 *
1792 	 * See Documentation/filesystems/path-lookup.txt for more details.
1793 	 */
1794 	hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1795 		unsigned seq;
1796 		struct inode *i;
1797 		const char *tname;
1798 		int tlen;
1799 
1800 		if (dentry->d_name.hash != hash)
1801 			continue;
1802 
1803 seqretry:
1804 		seq = read_seqcount_begin(&dentry->d_seq);
1805 		if (dentry->d_parent != parent)
1806 			continue;
1807 		if (d_unhashed(dentry))
1808 			continue;
1809 		tlen = dentry->d_name.len;
1810 		tname = dentry->d_name.name;
1811 		i = dentry->d_inode;
1812 		prefetch(tname);
1813 		/*
1814 		 * This seqcount check is required to ensure name and
1815 		 * len are loaded atomically, so as not to walk off the
1816 		 * edge of memory when walking. If we could load this
1817 		 * atomically some other way, we could drop this check.
1818 		 */
1819 		if (read_seqcount_retry(&dentry->d_seq, seq))
1820 			goto seqretry;
1821 		if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1822 			if (parent->d_op->d_compare(parent, *inode,
1823 						dentry, i,
1824 						tlen, tname, name))
1825 				continue;
1826 		} else {
1827 			if (dentry_cmp(tname, tlen, str, len))
1828 				continue;
1829 		}
1830 		/*
1831 		 * No extra seqcount check is required after the name
1832 		 * compare. The caller must perform a seqcount check in
1833 		 * order to do anything useful with the returned dentry
1834 		 * anyway.
1835 		 */
1836 		*seqp = seq;
1837 		*inode = i;
1838 		return dentry;
1839 	}
1840 	return NULL;
1841 }
1842 
1843 /**
1844  * d_lookup - search for a dentry
1845  * @parent: parent dentry
1846  * @name: qstr of name we wish to find
1847  * Returns: dentry, or NULL
1848  *
1849  * d_lookup searches the children of the parent dentry for the name in
1850  * question. If the dentry is found its reference count is incremented and the
1851  * dentry is returned. The caller must use dput to free the entry when it has
1852  * finished using it. %NULL is returned if the dentry does not exist.
1853  */
d_lookup(struct dentry * parent,struct qstr * name)1854 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1855 {
1856 	struct dentry *dentry;
1857 	unsigned seq;
1858 
1859         do {
1860                 seq = read_seqbegin(&rename_lock);
1861                 dentry = __d_lookup(parent, name);
1862                 if (dentry)
1863 			break;
1864 	} while (read_seqretry(&rename_lock, seq));
1865 	return dentry;
1866 }
1867 EXPORT_SYMBOL(d_lookup);
1868 
1869 /**
1870  * __d_lookup - search for a dentry (racy)
1871  * @parent: parent dentry
1872  * @name: qstr of name we wish to find
1873  * Returns: dentry, or NULL
1874  *
1875  * __d_lookup is like d_lookup, however it may (rarely) return a
1876  * false-negative result due to unrelated rename activity.
1877  *
1878  * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1879  * however it must be used carefully, eg. with a following d_lookup in
1880  * the case of failure.
1881  *
1882  * __d_lookup callers must be commented.
1883  */
__d_lookup(struct dentry * parent,struct qstr * name)1884 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1885 {
1886 	unsigned int len = name->len;
1887 	unsigned int hash = name->hash;
1888 	const unsigned char *str = name->name;
1889 	struct hlist_bl_head *b = d_hash(parent, hash);
1890 	struct hlist_bl_node *node;
1891 	struct dentry *found = NULL;
1892 	struct dentry *dentry;
1893 
1894 	/*
1895 	 * Note: There is significant duplication with __d_lookup_rcu which is
1896 	 * required to prevent single threaded performance regressions
1897 	 * especially on architectures where smp_rmb (in seqcounts) are costly.
1898 	 * Keep the two functions in sync.
1899 	 */
1900 
1901 	/*
1902 	 * The hash list is protected using RCU.
1903 	 *
1904 	 * Take d_lock when comparing a candidate dentry, to avoid races
1905 	 * with d_move().
1906 	 *
1907 	 * It is possible that concurrent renames can mess up our list
1908 	 * walk here and result in missing our dentry, resulting in the
1909 	 * false-negative result. d_lookup() protects against concurrent
1910 	 * renames using rename_lock seqlock.
1911 	 *
1912 	 * See Documentation/filesystems/path-lookup.txt for more details.
1913 	 */
1914 	rcu_read_lock();
1915 
1916 	hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1917 		const char *tname;
1918 		int tlen;
1919 
1920 		if (dentry->d_name.hash != hash)
1921 			continue;
1922 
1923 		spin_lock(&dentry->d_lock);
1924 		if (dentry->d_parent != parent)
1925 			goto next;
1926 		if (d_unhashed(dentry))
1927 			goto next;
1928 
1929 		/*
1930 		 * It is safe to compare names since d_move() cannot
1931 		 * change the qstr (protected by d_lock).
1932 		 */
1933 		tlen = dentry->d_name.len;
1934 		tname = dentry->d_name.name;
1935 		if (parent->d_flags & DCACHE_OP_COMPARE) {
1936 			if (parent->d_op->d_compare(parent, parent->d_inode,
1937 						dentry, dentry->d_inode,
1938 						tlen, tname, name))
1939 				goto next;
1940 		} else {
1941 			if (dentry_cmp(tname, tlen, str, len))
1942 				goto next;
1943 		}
1944 
1945 		dentry->d_count++;
1946 		found = dentry;
1947 		spin_unlock(&dentry->d_lock);
1948 		break;
1949 next:
1950 		spin_unlock(&dentry->d_lock);
1951  	}
1952  	rcu_read_unlock();
1953 
1954  	return found;
1955 }
1956 
1957 /**
1958  * d_hash_and_lookup - hash the qstr then search for a dentry
1959  * @dir: Directory to search in
1960  * @name: qstr of name we wish to find
1961  *
1962  * On hash failure or on lookup failure NULL is returned.
1963  */
d_hash_and_lookup(struct dentry * dir,struct qstr * name)1964 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1965 {
1966 	struct dentry *dentry = NULL;
1967 
1968 	/*
1969 	 * Check for a fs-specific hash function. Note that we must
1970 	 * calculate the standard hash first, as the d_op->d_hash()
1971 	 * routine may choose to leave the hash value unchanged.
1972 	 */
1973 	name->hash = full_name_hash(name->name, name->len);
1974 	if (dir->d_flags & DCACHE_OP_HASH) {
1975 		if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1976 			goto out;
1977 	}
1978 	dentry = d_lookup(dir, name);
1979 out:
1980 	return dentry;
1981 }
1982 
1983 /**
1984  * d_validate - verify dentry provided from insecure source (deprecated)
1985  * @dentry: The dentry alleged to be valid child of @dparent
1986  * @dparent: The parent dentry (known to be valid)
1987  *
1988  * An insecure source has sent us a dentry, here we verify it and dget() it.
1989  * This is used by ncpfs in its readdir implementation.
1990  * Zero is returned in the dentry is invalid.
1991  *
1992  * This function is slow for big directories, and deprecated, do not use it.
1993  */
d_validate(struct dentry * dentry,struct dentry * dparent)1994 int d_validate(struct dentry *dentry, struct dentry *dparent)
1995 {
1996 	struct dentry *child;
1997 
1998 	spin_lock(&dparent->d_lock);
1999 	list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
2000 		if (dentry == child) {
2001 			spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2002 			__dget_dlock(dentry);
2003 			spin_unlock(&dentry->d_lock);
2004 			spin_unlock(&dparent->d_lock);
2005 			return 1;
2006 		}
2007 	}
2008 	spin_unlock(&dparent->d_lock);
2009 
2010 	return 0;
2011 }
2012 EXPORT_SYMBOL(d_validate);
2013 
2014 /*
2015  * When a file is deleted, we have two options:
2016  * - turn this dentry into a negative dentry
2017  * - unhash this dentry and free it.
2018  *
2019  * Usually, we want to just turn this into
2020  * a negative dentry, but if anybody else is
2021  * currently using the dentry or the inode
2022  * we can't do that and we fall back on removing
2023  * it from the hash queues and waiting for
2024  * it to be deleted later when it has no users
2025  */
2026 
2027 /**
2028  * d_delete - delete a dentry
2029  * @dentry: The dentry to delete
2030  *
2031  * Turn the dentry into a negative dentry if possible, otherwise
2032  * remove it from the hash queues so it can be deleted later
2033  */
2034 
d_delete(struct dentry * dentry)2035 void d_delete(struct dentry * dentry)
2036 {
2037 	struct inode *inode;
2038 	int isdir = 0;
2039 	/*
2040 	 * Are we the only user?
2041 	 */
2042 again:
2043 	spin_lock(&dentry->d_lock);
2044 	inode = dentry->d_inode;
2045 	isdir = S_ISDIR(inode->i_mode);
2046 	if (dentry->d_count == 1) {
2047 		if (inode && !spin_trylock(&inode->i_lock)) {
2048 			spin_unlock(&dentry->d_lock);
2049 			cpu_relax();
2050 			goto again;
2051 		}
2052 		dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2053 		dentry_unlink_inode(dentry);
2054 		fsnotify_nameremove(dentry, isdir);
2055 		return;
2056 	}
2057 
2058 	if (!d_unhashed(dentry))
2059 		__d_drop(dentry);
2060 
2061 	spin_unlock(&dentry->d_lock);
2062 
2063 	fsnotify_nameremove(dentry, isdir);
2064 }
2065 EXPORT_SYMBOL(d_delete);
2066 
__d_rehash(struct dentry * entry,struct hlist_bl_head * b)2067 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2068 {
2069 	BUG_ON(!d_unhashed(entry));
2070 	hlist_bl_lock(b);
2071 	entry->d_flags |= DCACHE_RCUACCESS;
2072 	hlist_bl_add_head_rcu(&entry->d_hash, b);
2073 	hlist_bl_unlock(b);
2074 }
2075 
_d_rehash(struct dentry * entry)2076 static void _d_rehash(struct dentry * entry)
2077 {
2078 	__d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2079 }
2080 
2081 /**
2082  * d_rehash	- add an entry back to the hash
2083  * @entry: dentry to add to the hash
2084  *
2085  * Adds a dentry to the hash according to its name.
2086  */
2087 
d_rehash(struct dentry * entry)2088 void d_rehash(struct dentry * entry)
2089 {
2090 	spin_lock(&entry->d_lock);
2091 	_d_rehash(entry);
2092 	spin_unlock(&entry->d_lock);
2093 }
2094 EXPORT_SYMBOL(d_rehash);
2095 
2096 /**
2097  * dentry_update_name_case - update case insensitive dentry with a new name
2098  * @dentry: dentry to be updated
2099  * @name: new name
2100  *
2101  * Update a case insensitive dentry with new case of name.
2102  *
2103  * dentry must have been returned by d_lookup with name @name. Old and new
2104  * name lengths must match (ie. no d_compare which allows mismatched name
2105  * lengths).
2106  *
2107  * Parent inode i_mutex must be held over d_lookup and into this call (to
2108  * keep renames and concurrent inserts, and readdir(2) away).
2109  */
dentry_update_name_case(struct dentry * dentry,struct qstr * name)2110 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2111 {
2112 	BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2113 	BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2114 
2115 	spin_lock(&dentry->d_lock);
2116 	write_seqcount_begin(&dentry->d_seq);
2117 	memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2118 	write_seqcount_end(&dentry->d_seq);
2119 	spin_unlock(&dentry->d_lock);
2120 }
2121 EXPORT_SYMBOL(dentry_update_name_case);
2122 
switch_names(struct dentry * dentry,struct dentry * target)2123 static void switch_names(struct dentry *dentry, struct dentry *target)
2124 {
2125 	if (dname_external(target)) {
2126 		if (dname_external(dentry)) {
2127 			/*
2128 			 * Both external: swap the pointers
2129 			 */
2130 			swap(target->d_name.name, dentry->d_name.name);
2131 		} else {
2132 			/*
2133 			 * dentry:internal, target:external.  Steal target's
2134 			 * storage and make target internal.
2135 			 */
2136 			memcpy(target->d_iname, dentry->d_name.name,
2137 					dentry->d_name.len + 1);
2138 			dentry->d_name.name = target->d_name.name;
2139 			target->d_name.name = target->d_iname;
2140 		}
2141 	} else {
2142 		if (dname_external(dentry)) {
2143 			/*
2144 			 * dentry:external, target:internal.  Give dentry's
2145 			 * storage to target and make dentry internal
2146 			 */
2147 			memcpy(dentry->d_iname, target->d_name.name,
2148 					target->d_name.len + 1);
2149 			target->d_name.name = dentry->d_name.name;
2150 			dentry->d_name.name = dentry->d_iname;
2151 		} else {
2152 			/*
2153 			 * Both are internal.  Just copy target to dentry
2154 			 */
2155 			memcpy(dentry->d_iname, target->d_name.name,
2156 					target->d_name.len + 1);
2157 			dentry->d_name.len = target->d_name.len;
2158 			return;
2159 		}
2160 	}
2161 	swap(dentry->d_name.len, target->d_name.len);
2162 }
2163 
dentry_lock_for_move(struct dentry * dentry,struct dentry * target)2164 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2165 {
2166 	/*
2167 	 * XXXX: do we really need to take target->d_lock?
2168 	 */
2169 	if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2170 		spin_lock(&target->d_parent->d_lock);
2171 	else {
2172 		if (d_ancestor(dentry->d_parent, target->d_parent)) {
2173 			spin_lock(&dentry->d_parent->d_lock);
2174 			spin_lock_nested(&target->d_parent->d_lock,
2175 						DENTRY_D_LOCK_NESTED);
2176 		} else {
2177 			spin_lock(&target->d_parent->d_lock);
2178 			spin_lock_nested(&dentry->d_parent->d_lock,
2179 						DENTRY_D_LOCK_NESTED);
2180 		}
2181 	}
2182 	if (target < dentry) {
2183 		spin_lock_nested(&target->d_lock, 2);
2184 		spin_lock_nested(&dentry->d_lock, 3);
2185 	} else {
2186 		spin_lock_nested(&dentry->d_lock, 2);
2187 		spin_lock_nested(&target->d_lock, 3);
2188 	}
2189 }
2190 
dentry_unlock_parents_for_move(struct dentry * dentry,struct dentry * target)2191 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2192 					struct dentry *target)
2193 {
2194 	if (target->d_parent != dentry->d_parent)
2195 		spin_unlock(&dentry->d_parent->d_lock);
2196 	if (target->d_parent != target)
2197 		spin_unlock(&target->d_parent->d_lock);
2198 }
2199 
2200 /*
2201  * When switching names, the actual string doesn't strictly have to
2202  * be preserved in the target - because we're dropping the target
2203  * anyway. As such, we can just do a simple memcpy() to copy over
2204  * the new name before we switch.
2205  *
2206  * Note that we have to be a lot more careful about getting the hash
2207  * switched - we have to switch the hash value properly even if it
2208  * then no longer matches the actual (corrupted) string of the target.
2209  * The hash value has to match the hash queue that the dentry is on..
2210  */
2211 /*
2212  * __d_move - move a dentry
2213  * @dentry: entry to move
2214  * @target: new dentry
2215  *
2216  * Update the dcache to reflect the move of a file name. Negative
2217  * dcache entries should not be moved in this way. Caller must hold
2218  * rename_lock, the i_mutex of the source and target directories,
2219  * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2220  */
__d_move(struct dentry * dentry,struct dentry * target)2221 static void __d_move(struct dentry * dentry, struct dentry * target)
2222 {
2223 	if (!dentry->d_inode)
2224 		printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2225 
2226 	BUG_ON(d_ancestor(dentry, target));
2227 	BUG_ON(d_ancestor(target, dentry));
2228 
2229 	dentry_lock_for_move(dentry, target);
2230 
2231 	write_seqcount_begin(&dentry->d_seq);
2232 	write_seqcount_begin(&target->d_seq);
2233 
2234 	/* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2235 
2236 	/*
2237 	 * Move the dentry to the target hash queue. Don't bother checking
2238 	 * for the same hash queue because of how unlikely it is.
2239 	 */
2240 	__d_drop(dentry);
2241 	__d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2242 
2243 	/* Unhash the target: dput() will then get rid of it */
2244 	__d_drop(target);
2245 
2246 	list_del(&dentry->d_u.d_child);
2247 	list_del(&target->d_u.d_child);
2248 
2249 	/* Switch the names.. */
2250 	switch_names(dentry, target);
2251 	swap(dentry->d_name.hash, target->d_name.hash);
2252 
2253 	/* ... and switch the parents */
2254 	if (IS_ROOT(dentry)) {
2255 		dentry->d_parent = target->d_parent;
2256 		target->d_parent = target;
2257 		INIT_LIST_HEAD(&target->d_u.d_child);
2258 	} else {
2259 		swap(dentry->d_parent, target->d_parent);
2260 
2261 		/* And add them back to the (new) parent lists */
2262 		list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2263 	}
2264 
2265 	list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2266 
2267 	write_seqcount_end(&target->d_seq);
2268 	write_seqcount_end(&dentry->d_seq);
2269 
2270 	dentry_unlock_parents_for_move(dentry, target);
2271 	spin_unlock(&target->d_lock);
2272 	fsnotify_d_move(dentry);
2273 	spin_unlock(&dentry->d_lock);
2274 }
2275 
2276 /*
2277  * d_move - move a dentry
2278  * @dentry: entry to move
2279  * @target: new dentry
2280  *
2281  * Update the dcache to reflect the move of a file name. Negative
2282  * dcache entries should not be moved in this way. See the locking
2283  * requirements for __d_move.
2284  */
d_move(struct dentry * dentry,struct dentry * target)2285 void d_move(struct dentry *dentry, struct dentry *target)
2286 {
2287 	write_seqlock(&rename_lock);
2288 	__d_move(dentry, target);
2289 	write_sequnlock(&rename_lock);
2290 }
2291 EXPORT_SYMBOL(d_move);
2292 
2293 /**
2294  * d_ancestor - search for an ancestor
2295  * @p1: ancestor dentry
2296  * @p2: child dentry
2297  *
2298  * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2299  * an ancestor of p2, else NULL.
2300  */
d_ancestor(struct dentry * p1,struct dentry * p2)2301 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2302 {
2303 	struct dentry *p;
2304 
2305 	for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2306 		if (p->d_parent == p1)
2307 			return p;
2308 	}
2309 	return NULL;
2310 }
2311 
2312 /*
2313  * This helper attempts to cope with remotely renamed directories
2314  *
2315  * It assumes that the caller is already holding
2316  * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2317  *
2318  * Note: If ever the locking in lock_rename() changes, then please
2319  * remember to update this too...
2320  */
__d_unalias(struct inode * inode,struct dentry * dentry,struct dentry * alias)2321 static struct dentry *__d_unalias(struct inode *inode,
2322 		struct dentry *dentry, struct dentry *alias)
2323 {
2324 	struct mutex *m1 = NULL, *m2 = NULL;
2325 	struct dentry *ret;
2326 
2327 	/* If alias and dentry share a parent, then no extra locks required */
2328 	if (alias->d_parent == dentry->d_parent)
2329 		goto out_unalias;
2330 
2331 	/* See lock_rename() */
2332 	ret = ERR_PTR(-EBUSY);
2333 	if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2334 		goto out_err;
2335 	m1 = &dentry->d_sb->s_vfs_rename_mutex;
2336 	if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2337 		goto out_err;
2338 	m2 = &alias->d_parent->d_inode->i_mutex;
2339 out_unalias:
2340 	__d_move(alias, dentry);
2341 	ret = alias;
2342 out_err:
2343 	spin_unlock(&inode->i_lock);
2344 	if (m2)
2345 		mutex_unlock(m2);
2346 	if (m1)
2347 		mutex_unlock(m1);
2348 	return ret;
2349 }
2350 
2351 /*
2352  * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2353  * named dentry in place of the dentry to be replaced.
2354  * returns with anon->d_lock held!
2355  */
__d_materialise_dentry(struct dentry * dentry,struct dentry * anon)2356 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2357 {
2358 	struct dentry *dparent, *aparent;
2359 
2360 	dentry_lock_for_move(anon, dentry);
2361 
2362 	write_seqcount_begin(&dentry->d_seq);
2363 	write_seqcount_begin(&anon->d_seq);
2364 
2365 	dparent = dentry->d_parent;
2366 	aparent = anon->d_parent;
2367 
2368 	switch_names(dentry, anon);
2369 	swap(dentry->d_name.hash, anon->d_name.hash);
2370 
2371 	dentry->d_parent = (aparent == anon) ? dentry : aparent;
2372 	list_del(&dentry->d_u.d_child);
2373 	if (!IS_ROOT(dentry))
2374 		list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2375 	else
2376 		INIT_LIST_HEAD(&dentry->d_u.d_child);
2377 
2378 	anon->d_parent = (dparent == dentry) ? anon : dparent;
2379 	list_del(&anon->d_u.d_child);
2380 	if (!IS_ROOT(anon))
2381 		list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2382 	else
2383 		INIT_LIST_HEAD(&anon->d_u.d_child);
2384 
2385 	write_seqcount_end(&dentry->d_seq);
2386 	write_seqcount_end(&anon->d_seq);
2387 
2388 	dentry_unlock_parents_for_move(anon, dentry);
2389 	spin_unlock(&dentry->d_lock);
2390 
2391 	/* anon->d_lock still locked, returns locked */
2392 	anon->d_flags &= ~DCACHE_DISCONNECTED;
2393 }
2394 
2395 /**
2396  * d_materialise_unique - introduce an inode into the tree
2397  * @dentry: candidate dentry
2398  * @inode: inode to bind to the dentry, to which aliases may be attached
2399  *
2400  * Introduces an dentry into the tree, substituting an extant disconnected
2401  * root directory alias in its place if there is one. Caller must hold the
2402  * i_mutex of the parent directory.
2403  */
d_materialise_unique(struct dentry * dentry,struct inode * inode)2404 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2405 {
2406 	struct dentry *actual;
2407 
2408 	BUG_ON(!d_unhashed(dentry));
2409 
2410 	if (!inode) {
2411 		actual = dentry;
2412 		__d_instantiate(dentry, NULL);
2413 		d_rehash(actual);
2414 		goto out_nolock;
2415 	}
2416 
2417 	spin_lock(&inode->i_lock);
2418 
2419 	if (S_ISDIR(inode->i_mode)) {
2420 		struct dentry *alias;
2421 
2422 		/* Does an aliased dentry already exist? */
2423 		alias = __d_find_alias(inode, 0);
2424 		if (alias) {
2425 			actual = alias;
2426 			write_seqlock(&rename_lock);
2427 
2428 			if (d_ancestor(alias, dentry)) {
2429 				/* Check for loops */
2430 				actual = ERR_PTR(-ELOOP);
2431 				spin_unlock(&inode->i_lock);
2432 			} else if (IS_ROOT(alias)) {
2433 				/* Is this an anonymous mountpoint that we
2434 				 * could splice into our tree? */
2435 				__d_materialise_dentry(dentry, alias);
2436 				write_sequnlock(&rename_lock);
2437 				__d_drop(alias);
2438 				goto found;
2439 			} else {
2440 				/* Nope, but we must(!) avoid directory
2441 				 * aliasing. This drops inode->i_lock */
2442 				actual = __d_unalias(inode, dentry, alias);
2443 			}
2444 			write_sequnlock(&rename_lock);
2445 			if (IS_ERR(actual)) {
2446 				if (PTR_ERR(actual) == -ELOOP)
2447 					pr_warn_ratelimited(
2448 						"VFS: Lookup of '%s' in %s %s"
2449 						" would have caused loop\n",
2450 						dentry->d_name.name,
2451 						inode->i_sb->s_type->name,
2452 						inode->i_sb->s_id);
2453 				dput(alias);
2454 			}
2455 			goto out_nolock;
2456 		}
2457 	}
2458 
2459 	/* Add a unique reference */
2460 	actual = __d_instantiate_unique(dentry, inode);
2461 	if (!actual)
2462 		actual = dentry;
2463 	else
2464 		BUG_ON(!d_unhashed(actual));
2465 
2466 	spin_lock(&actual->d_lock);
2467 found:
2468 	_d_rehash(actual);
2469 	spin_unlock(&actual->d_lock);
2470 	spin_unlock(&inode->i_lock);
2471 out_nolock:
2472 	if (actual == dentry) {
2473 		security_d_instantiate(dentry, inode);
2474 		return NULL;
2475 	}
2476 
2477 	iput(inode);
2478 	return actual;
2479 }
2480 EXPORT_SYMBOL_GPL(d_materialise_unique);
2481 
prepend(char ** buffer,int * buflen,const char * str,int namelen)2482 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2483 {
2484 	*buflen -= namelen;
2485 	if (*buflen < 0)
2486 		return -ENAMETOOLONG;
2487 	*buffer -= namelen;
2488 	memcpy(*buffer, str, namelen);
2489 	return 0;
2490 }
2491 
prepend_name(char ** buffer,int * buflen,struct qstr * name)2492 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2493 {
2494 	return prepend(buffer, buflen, name->name, name->len);
2495 }
2496 
2497 /**
2498  * prepend_path - Prepend path string to a buffer
2499  * @path: the dentry/vfsmount to report
2500  * @root: root vfsmnt/dentry
2501  * @buffer: pointer to the end of the buffer
2502  * @buflen: pointer to buffer length
2503  *
2504  * Caller holds the rename_lock.
2505  */
prepend_path(const struct path * path,const struct path * root,char ** buffer,int * buflen)2506 static int prepend_path(const struct path *path,
2507 			const struct path *root,
2508 			char **buffer, int *buflen)
2509 {
2510 	struct dentry *dentry = path->dentry;
2511 	struct vfsmount *vfsmnt = path->mnt;
2512 	struct mount *mnt = real_mount(vfsmnt);
2513 	bool slash = false;
2514 	int error = 0;
2515 
2516 	br_read_lock(&vfsmount_lock);
2517 	while (dentry != root->dentry || vfsmnt != root->mnt) {
2518 		struct dentry * parent;
2519 
2520 		if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2521 			/* Global root? */
2522 			if (!mnt_has_parent(mnt))
2523 				goto global_root;
2524 			dentry = mnt->mnt_mountpoint;
2525 			mnt = mnt->mnt_parent;
2526 			vfsmnt = &mnt->mnt;
2527 			continue;
2528 		}
2529 		parent = dentry->d_parent;
2530 		prefetch(parent);
2531 		spin_lock(&dentry->d_lock);
2532 		error = prepend_name(buffer, buflen, &dentry->d_name);
2533 		spin_unlock(&dentry->d_lock);
2534 		if (!error)
2535 			error = prepend(buffer, buflen, "/", 1);
2536 		if (error)
2537 			break;
2538 
2539 		slash = true;
2540 		dentry = parent;
2541 	}
2542 
2543 	if (!error && !slash)
2544 		error = prepend(buffer, buflen, "/", 1);
2545 
2546 out:
2547 	br_read_unlock(&vfsmount_lock);
2548 	return error;
2549 
2550 global_root:
2551 	/*
2552 	 * Filesystems needing to implement special "root names"
2553 	 * should do so with ->d_dname()
2554 	 */
2555 	if (IS_ROOT(dentry) &&
2556 	    (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2557 		WARN(1, "Root dentry has weird name <%.*s>\n",
2558 		     (int) dentry->d_name.len, dentry->d_name.name);
2559 	}
2560 	if (!slash)
2561 		error = prepend(buffer, buflen, "/", 1);
2562 	if (!error)
2563 		error = is_mounted(vfsmnt) ? 1 : 2;
2564 	goto out;
2565 }
2566 
2567 /**
2568  * __d_path - return the path of a dentry
2569  * @path: the dentry/vfsmount to report
2570  * @root: root vfsmnt/dentry
2571  * @buf: buffer to return value in
2572  * @buflen: buffer length
2573  *
2574  * Convert a dentry into an ASCII path name.
2575  *
2576  * Returns a pointer into the buffer or an error code if the
2577  * path was too long.
2578  *
2579  * "buflen" should be positive.
2580  *
2581  * If the path is not reachable from the supplied root, return %NULL.
2582  */
__d_path(const struct path * path,const struct path * root,char * buf,int buflen)2583 char *__d_path(const struct path *path,
2584 	       const struct path *root,
2585 	       char *buf, int buflen)
2586 {
2587 	char *res = buf + buflen;
2588 	int error;
2589 
2590 	prepend(&res, &buflen, "\0", 1);
2591 	write_seqlock(&rename_lock);
2592 	error = prepend_path(path, root, &res, &buflen);
2593 	write_sequnlock(&rename_lock);
2594 
2595 	if (error < 0)
2596 		return ERR_PTR(error);
2597 	if (error > 0)
2598 		return NULL;
2599 	return res;
2600 }
2601 
d_absolute_path(const struct path * path,char * buf,int buflen)2602 char *d_absolute_path(const struct path *path,
2603 	       char *buf, int buflen)
2604 {
2605 	struct path root = {};
2606 	char *res = buf + buflen;
2607 	int error;
2608 
2609 	prepend(&res, &buflen, "\0", 1);
2610 	write_seqlock(&rename_lock);
2611 	error = prepend_path(path, &root, &res, &buflen);
2612 	write_sequnlock(&rename_lock);
2613 
2614 	if (error > 1)
2615 		error = -EINVAL;
2616 	if (error < 0)
2617 		return ERR_PTR(error);
2618 	return res;
2619 }
2620 
2621 /*
2622  * same as __d_path but appends "(deleted)" for unlinked files.
2623  */
path_with_deleted(const struct path * path,const struct path * root,char ** buf,int * buflen)2624 static int path_with_deleted(const struct path *path,
2625 			     const struct path *root,
2626 			     char **buf, int *buflen)
2627 {
2628 	prepend(buf, buflen, "\0", 1);
2629 	if (d_unlinked(path->dentry)) {
2630 		int error = prepend(buf, buflen, " (deleted)", 10);
2631 		if (error)
2632 			return error;
2633 	}
2634 
2635 	return prepend_path(path, root, buf, buflen);
2636 }
2637 
prepend_unreachable(char ** buffer,int * buflen)2638 static int prepend_unreachable(char **buffer, int *buflen)
2639 {
2640 	return prepend(buffer, buflen, "(unreachable)", 13);
2641 }
2642 
2643 /**
2644  * d_path - return the path of a dentry
2645  * @path: path to report
2646  * @buf: buffer to return value in
2647  * @buflen: buffer length
2648  *
2649  * Convert a dentry into an ASCII path name. If the entry has been deleted
2650  * the string " (deleted)" is appended. Note that this is ambiguous.
2651  *
2652  * Returns a pointer into the buffer or an error code if the path was
2653  * too long. Note: Callers should use the returned pointer, not the passed
2654  * in buffer, to use the name! The implementation often starts at an offset
2655  * into the buffer, and may leave 0 bytes at the start.
2656  *
2657  * "buflen" should be positive.
2658  */
d_path(const struct path * path,char * buf,int buflen)2659 char *d_path(const struct path *path, char *buf, int buflen)
2660 {
2661 	char *res = buf + buflen;
2662 	struct path root;
2663 	int error;
2664 
2665 	/*
2666 	 * We have various synthetic filesystems that never get mounted.  On
2667 	 * these filesystems dentries are never used for lookup purposes, and
2668 	 * thus don't need to be hashed.  They also don't need a name until a
2669 	 * user wants to identify the object in /proc/pid/fd/.  The little hack
2670 	 * below allows us to generate a name for these objects on demand:
2671 	 */
2672 	if (path->dentry->d_op && path->dentry->d_op->d_dname)
2673 		return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2674 
2675 	get_fs_root(current->fs, &root);
2676 	write_seqlock(&rename_lock);
2677 	error = path_with_deleted(path, &root, &res, &buflen);
2678 	if (error < 0)
2679 		res = ERR_PTR(error);
2680 	write_sequnlock(&rename_lock);
2681 	path_put(&root);
2682 	return res;
2683 }
2684 EXPORT_SYMBOL(d_path);
2685 
2686 /**
2687  * d_path_with_unreachable - return the path of a dentry
2688  * @path: path to report
2689  * @buf: buffer to return value in
2690  * @buflen: buffer length
2691  *
2692  * The difference from d_path() is that this prepends "(unreachable)"
2693  * to paths which are unreachable from the current process' root.
2694  */
d_path_with_unreachable(const struct path * path,char * buf,int buflen)2695 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2696 {
2697 	char *res = buf + buflen;
2698 	struct path root;
2699 	int error;
2700 
2701 	if (path->dentry->d_op && path->dentry->d_op->d_dname)
2702 		return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2703 
2704 	get_fs_root(current->fs, &root);
2705 	write_seqlock(&rename_lock);
2706 	error = path_with_deleted(path, &root, &res, &buflen);
2707 	if (error > 0)
2708 		error = prepend_unreachable(&res, &buflen);
2709 	write_sequnlock(&rename_lock);
2710 	path_put(&root);
2711 	if (error)
2712 		res =  ERR_PTR(error);
2713 
2714 	return res;
2715 }
2716 
2717 /*
2718  * Helper function for dentry_operations.d_dname() members
2719  */
dynamic_dname(struct dentry * dentry,char * buffer,int buflen,const char * fmt,...)2720 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2721 			const char *fmt, ...)
2722 {
2723 	va_list args;
2724 	char temp[64];
2725 	int sz;
2726 
2727 	va_start(args, fmt);
2728 	sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2729 	va_end(args);
2730 
2731 	if (sz > sizeof(temp) || sz > buflen)
2732 		return ERR_PTR(-ENAMETOOLONG);
2733 
2734 	buffer += buflen - sz;
2735 	return memcpy(buffer, temp, sz);
2736 }
2737 
2738 /*
2739  * Write full pathname from the root of the filesystem into the buffer.
2740  */
__dentry_path(struct dentry * dentry,char * buf,int buflen)2741 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2742 {
2743 	char *end = buf + buflen;
2744 	char *retval;
2745 
2746 	prepend(&end, &buflen, "\0", 1);
2747 	if (buflen < 1)
2748 		goto Elong;
2749 	/* Get '/' right */
2750 	retval = end-1;
2751 	*retval = '/';
2752 
2753 	while (!IS_ROOT(dentry)) {
2754 		struct dentry *parent = dentry->d_parent;
2755 		int error;
2756 
2757 		prefetch(parent);
2758 		spin_lock(&dentry->d_lock);
2759 		error = prepend_name(&end, &buflen, &dentry->d_name);
2760 		spin_unlock(&dentry->d_lock);
2761 		if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2762 			goto Elong;
2763 
2764 		retval = end;
2765 		dentry = parent;
2766 	}
2767 	return retval;
2768 Elong:
2769 	return ERR_PTR(-ENAMETOOLONG);
2770 }
2771 
dentry_path_raw(struct dentry * dentry,char * buf,int buflen)2772 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2773 {
2774 	char *retval;
2775 
2776 	write_seqlock(&rename_lock);
2777 	retval = __dentry_path(dentry, buf, buflen);
2778 	write_sequnlock(&rename_lock);
2779 
2780 	return retval;
2781 }
2782 EXPORT_SYMBOL(dentry_path_raw);
2783 
dentry_path(struct dentry * dentry,char * buf,int buflen)2784 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2785 {
2786 	char *p = NULL;
2787 	char *retval;
2788 
2789 	write_seqlock(&rename_lock);
2790 	if (d_unlinked(dentry)) {
2791 		p = buf + buflen;
2792 		if (prepend(&p, &buflen, "//deleted", 10) != 0)
2793 			goto Elong;
2794 		buflen++;
2795 	}
2796 	retval = __dentry_path(dentry, buf, buflen);
2797 	write_sequnlock(&rename_lock);
2798 	if (!IS_ERR(retval) && p)
2799 		*p = '/';	/* restore '/' overriden with '\0' */
2800 	return retval;
2801 Elong:
2802 	return ERR_PTR(-ENAMETOOLONG);
2803 }
2804 
2805 /*
2806  * NOTE! The user-level library version returns a
2807  * character pointer. The kernel system call just
2808  * returns the length of the buffer filled (which
2809  * includes the ending '\0' character), or a negative
2810  * error value. So libc would do something like
2811  *
2812  *	char *getcwd(char * buf, size_t size)
2813  *	{
2814  *		int retval;
2815  *
2816  *		retval = sys_getcwd(buf, size);
2817  *		if (retval >= 0)
2818  *			return buf;
2819  *		errno = -retval;
2820  *		return NULL;
2821  *	}
2822  */
SYSCALL_DEFINE2(getcwd,char __user *,buf,unsigned long,size)2823 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2824 {
2825 	int error;
2826 	struct path pwd, root;
2827 	char *page = (char *) __get_free_page(GFP_USER);
2828 
2829 	if (!page)
2830 		return -ENOMEM;
2831 
2832 	get_fs_root_and_pwd(current->fs, &root, &pwd);
2833 
2834 	error = -ENOENT;
2835 	write_seqlock(&rename_lock);
2836 	if (!d_unlinked(pwd.dentry)) {
2837 		unsigned long len;
2838 		char *cwd = page + PAGE_SIZE;
2839 		int buflen = PAGE_SIZE;
2840 
2841 		prepend(&cwd, &buflen, "\0", 1);
2842 		error = prepend_path(&pwd, &root, &cwd, &buflen);
2843 		write_sequnlock(&rename_lock);
2844 
2845 		if (error < 0)
2846 			goto out;
2847 
2848 		/* Unreachable from current root */
2849 		if (error > 0) {
2850 			error = prepend_unreachable(&cwd, &buflen);
2851 			if (error)
2852 				goto out;
2853 		}
2854 
2855 		error = -ERANGE;
2856 		len = PAGE_SIZE + page - cwd;
2857 		if (len <= size) {
2858 			error = len;
2859 			if (copy_to_user(buf, cwd, len))
2860 				error = -EFAULT;
2861 		}
2862 	} else {
2863 		write_sequnlock(&rename_lock);
2864 	}
2865 
2866 out:
2867 	path_put(&pwd);
2868 	path_put(&root);
2869 	free_page((unsigned long) page);
2870 	return error;
2871 }
2872 
2873 /*
2874  * Test whether new_dentry is a subdirectory of old_dentry.
2875  *
2876  * Trivially implemented using the dcache structure
2877  */
2878 
2879 /**
2880  * is_subdir - is new dentry a subdirectory of old_dentry
2881  * @new_dentry: new dentry
2882  * @old_dentry: old dentry
2883  *
2884  * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2885  * Returns 0 otherwise.
2886  * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2887  */
2888 
is_subdir(struct dentry * new_dentry,struct dentry * old_dentry)2889 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2890 {
2891 	int result;
2892 	unsigned seq;
2893 
2894 	if (new_dentry == old_dentry)
2895 		return 1;
2896 
2897 	do {
2898 		/* for restarting inner loop in case of seq retry */
2899 		seq = read_seqbegin(&rename_lock);
2900 		/*
2901 		 * Need rcu_readlock to protect against the d_parent trashing
2902 		 * due to d_move
2903 		 */
2904 		rcu_read_lock();
2905 		if (d_ancestor(old_dentry, new_dentry))
2906 			result = 1;
2907 		else
2908 			result = 0;
2909 		rcu_read_unlock();
2910 	} while (read_seqretry(&rename_lock, seq));
2911 
2912 	return result;
2913 }
2914 
d_genocide(struct dentry * root)2915 void d_genocide(struct dentry *root)
2916 {
2917 	struct dentry *this_parent;
2918 	struct list_head *next;
2919 	unsigned seq;
2920 	int locked = 0;
2921 
2922 	seq = read_seqbegin(&rename_lock);
2923 again:
2924 	this_parent = root;
2925 	spin_lock(&this_parent->d_lock);
2926 repeat:
2927 	next = this_parent->d_subdirs.next;
2928 resume:
2929 	while (next != &this_parent->d_subdirs) {
2930 		struct list_head *tmp = next;
2931 		struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2932 		next = tmp->next;
2933 
2934 		spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2935 		if (d_unhashed(dentry) || !dentry->d_inode) {
2936 			spin_unlock(&dentry->d_lock);
2937 			continue;
2938 		}
2939 		if (!list_empty(&dentry->d_subdirs)) {
2940 			spin_unlock(&this_parent->d_lock);
2941 			spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2942 			this_parent = dentry;
2943 			spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2944 			goto repeat;
2945 		}
2946 		if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2947 			dentry->d_flags |= DCACHE_GENOCIDE;
2948 			dentry->d_count--;
2949 		}
2950 		spin_unlock(&dentry->d_lock);
2951 	}
2952 	if (this_parent != root) {
2953 		struct dentry *child = this_parent;
2954 		if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2955 			this_parent->d_flags |= DCACHE_GENOCIDE;
2956 			this_parent->d_count--;
2957 		}
2958 		this_parent = try_to_ascend(this_parent, locked, seq);
2959 		if (!this_parent)
2960 			goto rename_retry;
2961 		next = child->d_u.d_child.next;
2962 		goto resume;
2963 	}
2964 	spin_unlock(&this_parent->d_lock);
2965 	if (!locked && read_seqretry(&rename_lock, seq))
2966 		goto rename_retry;
2967 	if (locked)
2968 		write_sequnlock(&rename_lock);
2969 	return;
2970 
2971 rename_retry:
2972 	if (locked)
2973 		goto again;
2974 	locked = 1;
2975 	write_seqlock(&rename_lock);
2976 	goto again;
2977 }
2978 
2979 /**
2980  * find_inode_number - check for dentry with name
2981  * @dir: directory to check
2982  * @name: Name to find.
2983  *
2984  * Check whether a dentry already exists for the given name,
2985  * and return the inode number if it has an inode. Otherwise
2986  * 0 is returned.
2987  *
2988  * This routine is used to post-process directory listings for
2989  * filesystems using synthetic inode numbers, and is necessary
2990  * to keep getcwd() working.
2991  */
2992 
find_inode_number(struct dentry * dir,struct qstr * name)2993 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2994 {
2995 	struct dentry * dentry;
2996 	ino_t ino = 0;
2997 
2998 	dentry = d_hash_and_lookup(dir, name);
2999 	if (dentry) {
3000 		if (dentry->d_inode)
3001 			ino = dentry->d_inode->i_ino;
3002 		dput(dentry);
3003 	}
3004 	return ino;
3005 }
3006 EXPORT_SYMBOL(find_inode_number);
3007 
3008 static __initdata unsigned long dhash_entries;
set_dhash_entries(char * str)3009 static int __init set_dhash_entries(char *str)
3010 {
3011 	if (!str)
3012 		return 0;
3013 	dhash_entries = simple_strtoul(str, &str, 0);
3014 	return 1;
3015 }
3016 __setup("dhash_entries=", set_dhash_entries);
3017 
dcache_init_early(void)3018 static void __init dcache_init_early(void)
3019 {
3020 	unsigned int loop;
3021 
3022 	/* If hashes are distributed across NUMA nodes, defer
3023 	 * hash allocation until vmalloc space is available.
3024 	 */
3025 	if (hashdist)
3026 		return;
3027 
3028 	dentry_hashtable =
3029 		alloc_large_system_hash("Dentry cache",
3030 					sizeof(struct hlist_bl_head),
3031 					dhash_entries,
3032 					13,
3033 					HASH_EARLY,
3034 					&d_hash_shift,
3035 					&d_hash_mask,
3036 					0);
3037 
3038 	for (loop = 0; loop < (1U << d_hash_shift); loop++)
3039 		INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3040 }
3041 
dcache_init(void)3042 static void __init dcache_init(void)
3043 {
3044 	unsigned int loop;
3045 
3046 	/*
3047 	 * A constructor could be added for stable state like the lists,
3048 	 * but it is probably not worth it because of the cache nature
3049 	 * of the dcache.
3050 	 */
3051 	dentry_cache = KMEM_CACHE(dentry,
3052 		SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3053 
3054 	/* Hash may have been set up in dcache_init_early */
3055 	if (!hashdist)
3056 		return;
3057 
3058 	dentry_hashtable =
3059 		alloc_large_system_hash("Dentry cache",
3060 					sizeof(struct hlist_bl_head),
3061 					dhash_entries,
3062 					13,
3063 					0,
3064 					&d_hash_shift,
3065 					&d_hash_mask,
3066 					0);
3067 
3068 	for (loop = 0; loop < (1U << d_hash_shift); loop++)
3069 		INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3070 }
3071 
3072 /* SLAB cache for __getname() consumers */
3073 struct kmem_cache *names_cachep __read_mostly;
3074 EXPORT_SYMBOL(names_cachep);
3075 
3076 EXPORT_SYMBOL(d_genocide);
3077 
vfs_caches_init_early(void)3078 void __init vfs_caches_init_early(void)
3079 {
3080 	dcache_init_early();
3081 	inode_init_early();
3082 }
3083 
vfs_caches_init(unsigned long mempages)3084 void __init vfs_caches_init(unsigned long mempages)
3085 {
3086 	unsigned long reserve;
3087 
3088 	/* Base hash sizes on available memory, with a reserve equal to
3089            150% of current kernel size */
3090 
3091 	reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3092 	mempages -= reserve;
3093 
3094 	names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3095 			SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3096 
3097 	dcache_init();
3098 	inode_init();
3099 	files_init(mempages);
3100 	mnt_init();
3101 	bdev_cache_init();
3102 	chrdev_init();
3103 }
3104