USB: serial: add quatech2 usb to serial driver
[GitHub/mt8127/android_kernel_alcatel_ttab.git] / fs / dcache.c
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
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)
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
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 */
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
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
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 */
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 */
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 */
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 */
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 */
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
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 */
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 */
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
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 */
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_DISCONNECTED;
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 */
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 */
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
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 */
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 */
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 */
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
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 */
630 static inline void __dget_dlock(struct dentry *dentry)
631 {
632 dentry->d_count++;
633 }
634
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
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 */
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
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 */
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 */
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
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 */
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 */
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 */
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 */
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 */
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_DISCONNECTED) ||
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 */
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 locked = 1;
1120 write_seqlock(&rename_lock);
1121 goto again;
1122 }
1123 EXPORT_SYMBOL(have_submounts);
1124
1125 /*
1126 * Search the dentry child list for the specified parent,
1127 * and move any unused dentries to the end of the unused
1128 * list for prune_dcache(). We descend to the next level
1129 * whenever the d_subdirs list is non-empty and continue
1130 * searching.
1131 *
1132 * It returns zero iff there are no unused children,
1133 * otherwise it returns the number of children moved to
1134 * the end of the unused list. This may not be the total
1135 * number of unused children, because select_parent can
1136 * drop the lock and return early due to latency
1137 * constraints.
1138 */
1139 static int select_parent(struct dentry *parent, struct list_head *dispose)
1140 {
1141 struct dentry *this_parent;
1142 struct list_head *next;
1143 unsigned seq;
1144 int found = 0;
1145 int locked = 0;
1146
1147 seq = read_seqbegin(&rename_lock);
1148 again:
1149 this_parent = parent;
1150 spin_lock(&this_parent->d_lock);
1151 repeat:
1152 next = this_parent->d_subdirs.next;
1153 resume:
1154 while (next != &this_parent->d_subdirs) {
1155 struct list_head *tmp = next;
1156 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1157 next = tmp->next;
1158
1159 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1160
1161 /*
1162 * move only zero ref count dentries to the dispose list.
1163 *
1164 * Those which are presently on the shrink list, being processed
1165 * by shrink_dentry_list(), shouldn't be moved. Otherwise the
1166 * loop in shrink_dcache_parent() might not make any progress
1167 * and loop forever.
1168 */
1169 if (dentry->d_count) {
1170 dentry_lru_del(dentry);
1171 } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) {
1172 dentry_lru_move_list(dentry, dispose);
1173 dentry->d_flags |= DCACHE_SHRINK_LIST;
1174 found++;
1175 }
1176 /*
1177 * We can return to the caller if we have found some (this
1178 * ensures forward progress). We'll be coming back to find
1179 * the rest.
1180 */
1181 if (found && need_resched()) {
1182 spin_unlock(&dentry->d_lock);
1183 goto out;
1184 }
1185
1186 /*
1187 * Descend a level if the d_subdirs list is non-empty.
1188 */
1189 if (!list_empty(&dentry->d_subdirs)) {
1190 spin_unlock(&this_parent->d_lock);
1191 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
1192 this_parent = dentry;
1193 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
1194 goto repeat;
1195 }
1196
1197 spin_unlock(&dentry->d_lock);
1198 }
1199 /*
1200 * All done at this level ... ascend and resume the search.
1201 */
1202 if (this_parent != parent) {
1203 struct dentry *child = this_parent;
1204 this_parent = try_to_ascend(this_parent, locked, seq);
1205 if (!this_parent)
1206 goto rename_retry;
1207 next = child->d_u.d_child.next;
1208 goto resume;
1209 }
1210 out:
1211 spin_unlock(&this_parent->d_lock);
1212 if (!locked && read_seqretry(&rename_lock, seq))
1213 goto rename_retry;
1214 if (locked)
1215 write_sequnlock(&rename_lock);
1216 return found;
1217
1218 rename_retry:
1219 if (found)
1220 return found;
1221 locked = 1;
1222 write_seqlock(&rename_lock);
1223 goto again;
1224 }
1225
1226 /**
1227 * shrink_dcache_parent - prune dcache
1228 * @parent: parent of entries to prune
1229 *
1230 * Prune the dcache to remove unused children of the parent dentry.
1231 */
1232 void shrink_dcache_parent(struct dentry * parent)
1233 {
1234 LIST_HEAD(dispose);
1235 int found;
1236
1237 while ((found = select_parent(parent, &dispose)) != 0)
1238 shrink_dentry_list(&dispose);
1239 }
1240 EXPORT_SYMBOL(shrink_dcache_parent);
1241
1242 /**
1243 * __d_alloc - allocate a dcache entry
1244 * @sb: filesystem it will belong to
1245 * @name: qstr of the name
1246 *
1247 * Allocates a dentry. It returns %NULL if there is insufficient memory
1248 * available. On a success the dentry is returned. The name passed in is
1249 * copied and the copy passed in may be reused after this call.
1250 */
1251
1252 struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name)
1253 {
1254 struct dentry *dentry;
1255 char *dname;
1256
1257 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1258 if (!dentry)
1259 return NULL;
1260
1261 if (name->len > DNAME_INLINE_LEN-1) {
1262 dname = kmalloc(name->len + 1, GFP_KERNEL);
1263 if (!dname) {
1264 kmem_cache_free(dentry_cache, dentry);
1265 return NULL;
1266 }
1267 } else {
1268 dname = dentry->d_iname;
1269 }
1270 dentry->d_name.name = dname;
1271
1272 dentry->d_name.len = name->len;
1273 dentry->d_name.hash = name->hash;
1274 memcpy(dname, name->name, name->len);
1275 dname[name->len] = 0;
1276
1277 dentry->d_count = 1;
1278 dentry->d_flags = 0;
1279 spin_lock_init(&dentry->d_lock);
1280 seqcount_init(&dentry->d_seq);
1281 dentry->d_inode = NULL;
1282 dentry->d_parent = dentry;
1283 dentry->d_sb = sb;
1284 dentry->d_op = NULL;
1285 dentry->d_fsdata = NULL;
1286 INIT_HLIST_BL_NODE(&dentry->d_hash);
1287 INIT_LIST_HEAD(&dentry->d_lru);
1288 INIT_LIST_HEAD(&dentry->d_subdirs);
1289 INIT_LIST_HEAD(&dentry->d_alias);
1290 INIT_LIST_HEAD(&dentry->d_u.d_child);
1291 d_set_d_op(dentry, dentry->d_sb->s_d_op);
1292
1293 this_cpu_inc(nr_dentry);
1294
1295 return dentry;
1296 }
1297
1298 /**
1299 * d_alloc - allocate a dcache entry
1300 * @parent: parent of entry to allocate
1301 * @name: qstr of the name
1302 *
1303 * Allocates a dentry. It returns %NULL if there is insufficient memory
1304 * available. On a success the dentry is returned. The name passed in is
1305 * copied and the copy passed in may be reused after this call.
1306 */
1307 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
1308 {
1309 struct dentry *dentry = __d_alloc(parent->d_sb, name);
1310 if (!dentry)
1311 return NULL;
1312
1313 spin_lock(&parent->d_lock);
1314 /*
1315 * don't need child lock because it is not subject
1316 * to concurrency here
1317 */
1318 __dget_dlock(parent);
1319 dentry->d_parent = parent;
1320 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1321 spin_unlock(&parent->d_lock);
1322
1323 return dentry;
1324 }
1325 EXPORT_SYMBOL(d_alloc);
1326
1327 struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name)
1328 {
1329 struct dentry *dentry = __d_alloc(sb, name);
1330 if (dentry)
1331 dentry->d_flags |= DCACHE_DISCONNECTED;
1332 return dentry;
1333 }
1334 EXPORT_SYMBOL(d_alloc_pseudo);
1335
1336 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
1337 {
1338 struct qstr q;
1339
1340 q.name = name;
1341 q.len = strlen(name);
1342 q.hash = full_name_hash(q.name, q.len);
1343 return d_alloc(parent, &q);
1344 }
1345 EXPORT_SYMBOL(d_alloc_name);
1346
1347 void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op)
1348 {
1349 WARN_ON_ONCE(dentry->d_op);
1350 WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH |
1351 DCACHE_OP_COMPARE |
1352 DCACHE_OP_REVALIDATE |
1353 DCACHE_OP_DELETE ));
1354 dentry->d_op = op;
1355 if (!op)
1356 return;
1357 if (op->d_hash)
1358 dentry->d_flags |= DCACHE_OP_HASH;
1359 if (op->d_compare)
1360 dentry->d_flags |= DCACHE_OP_COMPARE;
1361 if (op->d_revalidate)
1362 dentry->d_flags |= DCACHE_OP_REVALIDATE;
1363 if (op->d_delete)
1364 dentry->d_flags |= DCACHE_OP_DELETE;
1365 if (op->d_prune)
1366 dentry->d_flags |= DCACHE_OP_PRUNE;
1367
1368 }
1369 EXPORT_SYMBOL(d_set_d_op);
1370
1371 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1372 {
1373 spin_lock(&dentry->d_lock);
1374 if (inode) {
1375 if (unlikely(IS_AUTOMOUNT(inode)))
1376 dentry->d_flags |= DCACHE_NEED_AUTOMOUNT;
1377 list_add(&dentry->d_alias, &inode->i_dentry);
1378 }
1379 dentry->d_inode = inode;
1380 dentry_rcuwalk_barrier(dentry);
1381 spin_unlock(&dentry->d_lock);
1382 fsnotify_d_instantiate(dentry, inode);
1383 }
1384
1385 /**
1386 * d_instantiate - fill in inode information for a dentry
1387 * @entry: dentry to complete
1388 * @inode: inode to attach to this dentry
1389 *
1390 * Fill in inode information in the entry.
1391 *
1392 * This turns negative dentries into productive full members
1393 * of society.
1394 *
1395 * NOTE! This assumes that the inode count has been incremented
1396 * (or otherwise set) by the caller to indicate that it is now
1397 * in use by the dcache.
1398 */
1399
1400 void d_instantiate(struct dentry *entry, struct inode * inode)
1401 {
1402 BUG_ON(!list_empty(&entry->d_alias));
1403 if (inode)
1404 spin_lock(&inode->i_lock);
1405 __d_instantiate(entry, inode);
1406 if (inode)
1407 spin_unlock(&inode->i_lock);
1408 security_d_instantiate(entry, inode);
1409 }
1410 EXPORT_SYMBOL(d_instantiate);
1411
1412 /**
1413 * d_instantiate_unique - instantiate a non-aliased dentry
1414 * @entry: dentry to instantiate
1415 * @inode: inode to attach to this dentry
1416 *
1417 * Fill in inode information in the entry. On success, it returns NULL.
1418 * If an unhashed alias of "entry" already exists, then we return the
1419 * aliased dentry instead and drop one reference to inode.
1420 *
1421 * Note that in order to avoid conflicts with rename() etc, the caller
1422 * had better be holding the parent directory semaphore.
1423 *
1424 * This also assumes that the inode count has been incremented
1425 * (or otherwise set) by the caller to indicate that it is now
1426 * in use by the dcache.
1427 */
1428 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1429 struct inode *inode)
1430 {
1431 struct dentry *alias;
1432 int len = entry->d_name.len;
1433 const char *name = entry->d_name.name;
1434 unsigned int hash = entry->d_name.hash;
1435
1436 if (!inode) {
1437 __d_instantiate(entry, NULL);
1438 return NULL;
1439 }
1440
1441 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1442 struct qstr *qstr = &alias->d_name;
1443
1444 /*
1445 * Don't need alias->d_lock here, because aliases with
1446 * d_parent == entry->d_parent are not subject to name or
1447 * parent changes, because the parent inode i_mutex is held.
1448 */
1449 if (qstr->hash != hash)
1450 continue;
1451 if (alias->d_parent != entry->d_parent)
1452 continue;
1453 if (dentry_cmp(qstr->name, qstr->len, name, len))
1454 continue;
1455 __dget(alias);
1456 return alias;
1457 }
1458
1459 __d_instantiate(entry, inode);
1460 return NULL;
1461 }
1462
1463 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1464 {
1465 struct dentry *result;
1466
1467 BUG_ON(!list_empty(&entry->d_alias));
1468
1469 if (inode)
1470 spin_lock(&inode->i_lock);
1471 result = __d_instantiate_unique(entry, inode);
1472 if (inode)
1473 spin_unlock(&inode->i_lock);
1474
1475 if (!result) {
1476 security_d_instantiate(entry, inode);
1477 return NULL;
1478 }
1479
1480 BUG_ON(!d_unhashed(result));
1481 iput(inode);
1482 return result;
1483 }
1484
1485 EXPORT_SYMBOL(d_instantiate_unique);
1486
1487 struct dentry *d_make_root(struct inode *root_inode)
1488 {
1489 struct dentry *res = NULL;
1490
1491 if (root_inode) {
1492 static const struct qstr name = { .name = "/", .len = 1 };
1493
1494 res = __d_alloc(root_inode->i_sb, &name);
1495 if (res)
1496 d_instantiate(res, root_inode);
1497 else
1498 iput(root_inode);
1499 }
1500 return res;
1501 }
1502 EXPORT_SYMBOL(d_make_root);
1503
1504 static struct dentry * __d_find_any_alias(struct inode *inode)
1505 {
1506 struct dentry *alias;
1507
1508 if (list_empty(&inode->i_dentry))
1509 return NULL;
1510 alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias);
1511 __dget(alias);
1512 return alias;
1513 }
1514
1515 /**
1516 * d_find_any_alias - find any alias for a given inode
1517 * @inode: inode to find an alias for
1518 *
1519 * If any aliases exist for the given inode, take and return a
1520 * reference for one of them. If no aliases exist, return %NULL.
1521 */
1522 struct dentry *d_find_any_alias(struct inode *inode)
1523 {
1524 struct dentry *de;
1525
1526 spin_lock(&inode->i_lock);
1527 de = __d_find_any_alias(inode);
1528 spin_unlock(&inode->i_lock);
1529 return de;
1530 }
1531 EXPORT_SYMBOL(d_find_any_alias);
1532
1533 /**
1534 * d_obtain_alias - find or allocate a dentry for a given inode
1535 * @inode: inode to allocate the dentry for
1536 *
1537 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1538 * similar open by handle operations. The returned dentry may be anonymous,
1539 * or may have a full name (if the inode was already in the cache).
1540 *
1541 * When called on a directory inode, we must ensure that the inode only ever
1542 * has one dentry. If a dentry is found, that is returned instead of
1543 * allocating a new one.
1544 *
1545 * On successful return, the reference to the inode has been transferred
1546 * to the dentry. In case of an error the reference on the inode is released.
1547 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1548 * be passed in and will be the error will be propagate to the return value,
1549 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1550 */
1551 struct dentry *d_obtain_alias(struct inode *inode)
1552 {
1553 static const struct qstr anonstring = { .name = "" };
1554 struct dentry *tmp;
1555 struct dentry *res;
1556
1557 if (!inode)
1558 return ERR_PTR(-ESTALE);
1559 if (IS_ERR(inode))
1560 return ERR_CAST(inode);
1561
1562 res = d_find_any_alias(inode);
1563 if (res)
1564 goto out_iput;
1565
1566 tmp = __d_alloc(inode->i_sb, &anonstring);
1567 if (!tmp) {
1568 res = ERR_PTR(-ENOMEM);
1569 goto out_iput;
1570 }
1571
1572 spin_lock(&inode->i_lock);
1573 res = __d_find_any_alias(inode);
1574 if (res) {
1575 spin_unlock(&inode->i_lock);
1576 dput(tmp);
1577 goto out_iput;
1578 }
1579
1580 /* attach a disconnected dentry */
1581 spin_lock(&tmp->d_lock);
1582 tmp->d_inode = inode;
1583 tmp->d_flags |= DCACHE_DISCONNECTED;
1584 list_add(&tmp->d_alias, &inode->i_dentry);
1585 hlist_bl_lock(&tmp->d_sb->s_anon);
1586 hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon);
1587 hlist_bl_unlock(&tmp->d_sb->s_anon);
1588 spin_unlock(&tmp->d_lock);
1589 spin_unlock(&inode->i_lock);
1590 security_d_instantiate(tmp, inode);
1591
1592 return tmp;
1593
1594 out_iput:
1595 if (res && !IS_ERR(res))
1596 security_d_instantiate(res, inode);
1597 iput(inode);
1598 return res;
1599 }
1600 EXPORT_SYMBOL(d_obtain_alias);
1601
1602 /**
1603 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1604 * @inode: the inode which may have a disconnected dentry
1605 * @dentry: a negative dentry which we want to point to the inode.
1606 *
1607 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1608 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1609 * and return it, else simply d_add the inode to the dentry and return NULL.
1610 *
1611 * This is needed in the lookup routine of any filesystem that is exportable
1612 * (via knfsd) so that we can build dcache paths to directories effectively.
1613 *
1614 * If a dentry was found and moved, then it is returned. Otherwise NULL
1615 * is returned. This matches the expected return value of ->lookup.
1616 *
1617 */
1618 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1619 {
1620 struct dentry *new = NULL;
1621
1622 if (IS_ERR(inode))
1623 return ERR_CAST(inode);
1624
1625 if (inode && S_ISDIR(inode->i_mode)) {
1626 spin_lock(&inode->i_lock);
1627 new = __d_find_alias(inode, 1);
1628 if (new) {
1629 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1630 spin_unlock(&inode->i_lock);
1631 security_d_instantiate(new, inode);
1632 d_move(new, dentry);
1633 iput(inode);
1634 } else {
1635 /* already taking inode->i_lock, so d_add() by hand */
1636 __d_instantiate(dentry, inode);
1637 spin_unlock(&inode->i_lock);
1638 security_d_instantiate(dentry, inode);
1639 d_rehash(dentry);
1640 }
1641 } else
1642 d_add(dentry, inode);
1643 return new;
1644 }
1645 EXPORT_SYMBOL(d_splice_alias);
1646
1647 /**
1648 * d_add_ci - lookup or allocate new dentry with case-exact name
1649 * @inode: the inode case-insensitive lookup has found
1650 * @dentry: the negative dentry that was passed to the parent's lookup func
1651 * @name: the case-exact name to be associated with the returned dentry
1652 *
1653 * This is to avoid filling the dcache with case-insensitive names to the
1654 * same inode, only the actual correct case is stored in the dcache for
1655 * case-insensitive filesystems.
1656 *
1657 * For a case-insensitive lookup match and if the the case-exact dentry
1658 * already exists in in the dcache, use it and return it.
1659 *
1660 * If no entry exists with the exact case name, allocate new dentry with
1661 * the exact case, and return the spliced entry.
1662 */
1663 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1664 struct qstr *name)
1665 {
1666 int error;
1667 struct dentry *found;
1668 struct dentry *new;
1669
1670 /*
1671 * First check if a dentry matching the name already exists,
1672 * if not go ahead and create it now.
1673 */
1674 found = d_hash_and_lookup(dentry->d_parent, name);
1675 if (!found) {
1676 new = d_alloc(dentry->d_parent, name);
1677 if (!new) {
1678 error = -ENOMEM;
1679 goto err_out;
1680 }
1681
1682 found = d_splice_alias(inode, new);
1683 if (found) {
1684 dput(new);
1685 return found;
1686 }
1687 return new;
1688 }
1689
1690 /*
1691 * If a matching dentry exists, and it's not negative use it.
1692 *
1693 * Decrement the reference count to balance the iget() done
1694 * earlier on.
1695 */
1696 if (found->d_inode) {
1697 if (unlikely(found->d_inode != inode)) {
1698 /* This can't happen because bad inodes are unhashed. */
1699 BUG_ON(!is_bad_inode(inode));
1700 BUG_ON(!is_bad_inode(found->d_inode));
1701 }
1702 iput(inode);
1703 return found;
1704 }
1705
1706 /*
1707 * We are going to instantiate this dentry, unhash it and clear the
1708 * lookup flag so we can do that.
1709 */
1710 if (unlikely(d_need_lookup(found)))
1711 d_clear_need_lookup(found);
1712
1713 /*
1714 * Negative dentry: instantiate it unless the inode is a directory and
1715 * already has a dentry.
1716 */
1717 new = d_splice_alias(inode, found);
1718 if (new) {
1719 dput(found);
1720 found = new;
1721 }
1722 return found;
1723
1724 err_out:
1725 iput(inode);
1726 return ERR_PTR(error);
1727 }
1728 EXPORT_SYMBOL(d_add_ci);
1729
1730 /**
1731 * __d_lookup_rcu - search for a dentry (racy, store-free)
1732 * @parent: parent dentry
1733 * @name: qstr of name we wish to find
1734 * @seqp: returns d_seq value at the point where the dentry was found
1735 * @inode: returns dentry->d_inode when the inode was found valid.
1736 * Returns: dentry, or NULL
1737 *
1738 * __d_lookup_rcu is the dcache lookup function for rcu-walk name
1739 * resolution (store-free path walking) design described in
1740 * Documentation/filesystems/path-lookup.txt.
1741 *
1742 * This is not to be used outside core vfs.
1743 *
1744 * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock
1745 * held, and rcu_read_lock held. The returned dentry must not be stored into
1746 * without taking d_lock and checking d_seq sequence count against @seq
1747 * returned here.
1748 *
1749 * A refcount may be taken on the found dentry with the __d_rcu_to_refcount
1750 * function.
1751 *
1752 * Alternatively, __d_lookup_rcu may be called again to look up the child of
1753 * the returned dentry, so long as its parent's seqlock is checked after the
1754 * child is looked up. Thus, an interlocking stepping of sequence lock checks
1755 * is formed, giving integrity down the path walk.
1756 */
1757 struct dentry *__d_lookup_rcu(const struct dentry *parent,
1758 const struct qstr *name,
1759 unsigned *seqp, struct inode **inode)
1760 {
1761 unsigned int len = name->len;
1762 unsigned int hash = name->hash;
1763 const unsigned char *str = name->name;
1764 struct hlist_bl_head *b = d_hash(parent, hash);
1765 struct hlist_bl_node *node;
1766 struct dentry *dentry;
1767
1768 /*
1769 * Note: There is significant duplication with __d_lookup_rcu which is
1770 * required to prevent single threaded performance regressions
1771 * especially on architectures where smp_rmb (in seqcounts) are costly.
1772 * Keep the two functions in sync.
1773 */
1774
1775 /*
1776 * The hash list is protected using RCU.
1777 *
1778 * Carefully use d_seq when comparing a candidate dentry, to avoid
1779 * races with d_move().
1780 *
1781 * It is possible that concurrent renames can mess up our list
1782 * walk here and result in missing our dentry, resulting in the
1783 * false-negative result. d_lookup() protects against concurrent
1784 * renames using rename_lock seqlock.
1785 *
1786 * See Documentation/filesystems/path-lookup.txt for more details.
1787 */
1788 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1789 unsigned seq;
1790 struct inode *i;
1791 const char *tname;
1792 int tlen;
1793
1794 if (dentry->d_name.hash != hash)
1795 continue;
1796
1797 seqretry:
1798 seq = read_seqcount_begin(&dentry->d_seq);
1799 if (dentry->d_parent != parent)
1800 continue;
1801 if (d_unhashed(dentry))
1802 continue;
1803 tlen = dentry->d_name.len;
1804 tname = dentry->d_name.name;
1805 i = dentry->d_inode;
1806 prefetch(tname);
1807 /*
1808 * This seqcount check is required to ensure name and
1809 * len are loaded atomically, so as not to walk off the
1810 * edge of memory when walking. If we could load this
1811 * atomically some other way, we could drop this check.
1812 */
1813 if (read_seqcount_retry(&dentry->d_seq, seq))
1814 goto seqretry;
1815 if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) {
1816 if (parent->d_op->d_compare(parent, *inode,
1817 dentry, i,
1818 tlen, tname, name))
1819 continue;
1820 } else {
1821 if (dentry_cmp(tname, tlen, str, len))
1822 continue;
1823 }
1824 /*
1825 * No extra seqcount check is required after the name
1826 * compare. The caller must perform a seqcount check in
1827 * order to do anything useful with the returned dentry
1828 * anyway.
1829 */
1830 *seqp = seq;
1831 *inode = i;
1832 return dentry;
1833 }
1834 return NULL;
1835 }
1836
1837 /**
1838 * d_lookup - search for a dentry
1839 * @parent: parent dentry
1840 * @name: qstr of name we wish to find
1841 * Returns: dentry, or NULL
1842 *
1843 * d_lookup searches the children of the parent dentry for the name in
1844 * question. If the dentry is found its reference count is incremented and the
1845 * dentry is returned. The caller must use dput to free the entry when it has
1846 * finished using it. %NULL is returned if the dentry does not exist.
1847 */
1848 struct dentry *d_lookup(struct dentry *parent, struct qstr *name)
1849 {
1850 struct dentry *dentry;
1851 unsigned seq;
1852
1853 do {
1854 seq = read_seqbegin(&rename_lock);
1855 dentry = __d_lookup(parent, name);
1856 if (dentry)
1857 break;
1858 } while (read_seqretry(&rename_lock, seq));
1859 return dentry;
1860 }
1861 EXPORT_SYMBOL(d_lookup);
1862
1863 /**
1864 * __d_lookup - search for a dentry (racy)
1865 * @parent: parent dentry
1866 * @name: qstr of name we wish to find
1867 * Returns: dentry, or NULL
1868 *
1869 * __d_lookup is like d_lookup, however it may (rarely) return a
1870 * false-negative result due to unrelated rename activity.
1871 *
1872 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1873 * however it must be used carefully, eg. with a following d_lookup in
1874 * the case of failure.
1875 *
1876 * __d_lookup callers must be commented.
1877 */
1878 struct dentry *__d_lookup(struct dentry *parent, struct qstr *name)
1879 {
1880 unsigned int len = name->len;
1881 unsigned int hash = name->hash;
1882 const unsigned char *str = name->name;
1883 struct hlist_bl_head *b = d_hash(parent, hash);
1884 struct hlist_bl_node *node;
1885 struct dentry *found = NULL;
1886 struct dentry *dentry;
1887
1888 /*
1889 * Note: There is significant duplication with __d_lookup_rcu which is
1890 * required to prevent single threaded performance regressions
1891 * especially on architectures where smp_rmb (in seqcounts) are costly.
1892 * Keep the two functions in sync.
1893 */
1894
1895 /*
1896 * The hash list is protected using RCU.
1897 *
1898 * Take d_lock when comparing a candidate dentry, to avoid races
1899 * with d_move().
1900 *
1901 * It is possible that concurrent renames can mess up our list
1902 * walk here and result in missing our dentry, resulting in the
1903 * false-negative result. d_lookup() protects against concurrent
1904 * renames using rename_lock seqlock.
1905 *
1906 * See Documentation/filesystems/path-lookup.txt for more details.
1907 */
1908 rcu_read_lock();
1909
1910 hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) {
1911 const char *tname;
1912 int tlen;
1913
1914 if (dentry->d_name.hash != hash)
1915 continue;
1916
1917 spin_lock(&dentry->d_lock);
1918 if (dentry->d_parent != parent)
1919 goto next;
1920 if (d_unhashed(dentry))
1921 goto next;
1922
1923 /*
1924 * It is safe to compare names since d_move() cannot
1925 * change the qstr (protected by d_lock).
1926 */
1927 tlen = dentry->d_name.len;
1928 tname = dentry->d_name.name;
1929 if (parent->d_flags & DCACHE_OP_COMPARE) {
1930 if (parent->d_op->d_compare(parent, parent->d_inode,
1931 dentry, dentry->d_inode,
1932 tlen, tname, name))
1933 goto next;
1934 } else {
1935 if (dentry_cmp(tname, tlen, str, len))
1936 goto next;
1937 }
1938
1939 dentry->d_count++;
1940 found = dentry;
1941 spin_unlock(&dentry->d_lock);
1942 break;
1943 next:
1944 spin_unlock(&dentry->d_lock);
1945 }
1946 rcu_read_unlock();
1947
1948 return found;
1949 }
1950
1951 /**
1952 * d_hash_and_lookup - hash the qstr then search for a dentry
1953 * @dir: Directory to search in
1954 * @name: qstr of name we wish to find
1955 *
1956 * On hash failure or on lookup failure NULL is returned.
1957 */
1958 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1959 {
1960 struct dentry *dentry = NULL;
1961
1962 /*
1963 * Check for a fs-specific hash function. Note that we must
1964 * calculate the standard hash first, as the d_op->d_hash()
1965 * routine may choose to leave the hash value unchanged.
1966 */
1967 name->hash = full_name_hash(name->name, name->len);
1968 if (dir->d_flags & DCACHE_OP_HASH) {
1969 if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0)
1970 goto out;
1971 }
1972 dentry = d_lookup(dir, name);
1973 out:
1974 return dentry;
1975 }
1976
1977 /**
1978 * d_validate - verify dentry provided from insecure source (deprecated)
1979 * @dentry: The dentry alleged to be valid child of @dparent
1980 * @dparent: The parent dentry (known to be valid)
1981 *
1982 * An insecure source has sent us a dentry, here we verify it and dget() it.
1983 * This is used by ncpfs in its readdir implementation.
1984 * Zero is returned in the dentry is invalid.
1985 *
1986 * This function is slow for big directories, and deprecated, do not use it.
1987 */
1988 int d_validate(struct dentry *dentry, struct dentry *dparent)
1989 {
1990 struct dentry *child;
1991
1992 spin_lock(&dparent->d_lock);
1993 list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) {
1994 if (dentry == child) {
1995 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1996 __dget_dlock(dentry);
1997 spin_unlock(&dentry->d_lock);
1998 spin_unlock(&dparent->d_lock);
1999 return 1;
2000 }
2001 }
2002 spin_unlock(&dparent->d_lock);
2003
2004 return 0;
2005 }
2006 EXPORT_SYMBOL(d_validate);
2007
2008 /*
2009 * When a file is deleted, we have two options:
2010 * - turn this dentry into a negative dentry
2011 * - unhash this dentry and free it.
2012 *
2013 * Usually, we want to just turn this into
2014 * a negative dentry, but if anybody else is
2015 * currently using the dentry or the inode
2016 * we can't do that and we fall back on removing
2017 * it from the hash queues and waiting for
2018 * it to be deleted later when it has no users
2019 */
2020
2021 /**
2022 * d_delete - delete a dentry
2023 * @dentry: The dentry to delete
2024 *
2025 * Turn the dentry into a negative dentry if possible, otherwise
2026 * remove it from the hash queues so it can be deleted later
2027 */
2028
2029 void d_delete(struct dentry * dentry)
2030 {
2031 struct inode *inode;
2032 int isdir = 0;
2033 /*
2034 * Are we the only user?
2035 */
2036 again:
2037 spin_lock(&dentry->d_lock);
2038 inode = dentry->d_inode;
2039 isdir = S_ISDIR(inode->i_mode);
2040 if (dentry->d_count == 1) {
2041 if (inode && !spin_trylock(&inode->i_lock)) {
2042 spin_unlock(&dentry->d_lock);
2043 cpu_relax();
2044 goto again;
2045 }
2046 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
2047 dentry_unlink_inode(dentry);
2048 fsnotify_nameremove(dentry, isdir);
2049 return;
2050 }
2051
2052 if (!d_unhashed(dentry))
2053 __d_drop(dentry);
2054
2055 spin_unlock(&dentry->d_lock);
2056
2057 fsnotify_nameremove(dentry, isdir);
2058 }
2059 EXPORT_SYMBOL(d_delete);
2060
2061 static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b)
2062 {
2063 BUG_ON(!d_unhashed(entry));
2064 hlist_bl_lock(b);
2065 entry->d_flags |= DCACHE_RCUACCESS;
2066 hlist_bl_add_head_rcu(&entry->d_hash, b);
2067 hlist_bl_unlock(b);
2068 }
2069
2070 static void _d_rehash(struct dentry * entry)
2071 {
2072 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
2073 }
2074
2075 /**
2076 * d_rehash - add an entry back to the hash
2077 * @entry: dentry to add to the hash
2078 *
2079 * Adds a dentry to the hash according to its name.
2080 */
2081
2082 void d_rehash(struct dentry * entry)
2083 {
2084 spin_lock(&entry->d_lock);
2085 _d_rehash(entry);
2086 spin_unlock(&entry->d_lock);
2087 }
2088 EXPORT_SYMBOL(d_rehash);
2089
2090 /**
2091 * dentry_update_name_case - update case insensitive dentry with a new name
2092 * @dentry: dentry to be updated
2093 * @name: new name
2094 *
2095 * Update a case insensitive dentry with new case of name.
2096 *
2097 * dentry must have been returned by d_lookup with name @name. Old and new
2098 * name lengths must match (ie. no d_compare which allows mismatched name
2099 * lengths).
2100 *
2101 * Parent inode i_mutex must be held over d_lookup and into this call (to
2102 * keep renames and concurrent inserts, and readdir(2) away).
2103 */
2104 void dentry_update_name_case(struct dentry *dentry, struct qstr *name)
2105 {
2106 BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex));
2107 BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */
2108
2109 spin_lock(&dentry->d_lock);
2110 write_seqcount_begin(&dentry->d_seq);
2111 memcpy((unsigned char *)dentry->d_name.name, name->name, name->len);
2112 write_seqcount_end(&dentry->d_seq);
2113 spin_unlock(&dentry->d_lock);
2114 }
2115 EXPORT_SYMBOL(dentry_update_name_case);
2116
2117 static void switch_names(struct dentry *dentry, struct dentry *target)
2118 {
2119 if (dname_external(target)) {
2120 if (dname_external(dentry)) {
2121 /*
2122 * Both external: swap the pointers
2123 */
2124 swap(target->d_name.name, dentry->d_name.name);
2125 } else {
2126 /*
2127 * dentry:internal, target:external. Steal target's
2128 * storage and make target internal.
2129 */
2130 memcpy(target->d_iname, dentry->d_name.name,
2131 dentry->d_name.len + 1);
2132 dentry->d_name.name = target->d_name.name;
2133 target->d_name.name = target->d_iname;
2134 }
2135 } else {
2136 if (dname_external(dentry)) {
2137 /*
2138 * dentry:external, target:internal. Give dentry's
2139 * storage to target and make dentry internal
2140 */
2141 memcpy(dentry->d_iname, target->d_name.name,
2142 target->d_name.len + 1);
2143 target->d_name.name = dentry->d_name.name;
2144 dentry->d_name.name = dentry->d_iname;
2145 } else {
2146 /*
2147 * Both are internal. Just copy target to dentry
2148 */
2149 memcpy(dentry->d_iname, target->d_name.name,
2150 target->d_name.len + 1);
2151 dentry->d_name.len = target->d_name.len;
2152 return;
2153 }
2154 }
2155 swap(dentry->d_name.len, target->d_name.len);
2156 }
2157
2158 static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target)
2159 {
2160 /*
2161 * XXXX: do we really need to take target->d_lock?
2162 */
2163 if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent)
2164 spin_lock(&target->d_parent->d_lock);
2165 else {
2166 if (d_ancestor(dentry->d_parent, target->d_parent)) {
2167 spin_lock(&dentry->d_parent->d_lock);
2168 spin_lock_nested(&target->d_parent->d_lock,
2169 DENTRY_D_LOCK_NESTED);
2170 } else {
2171 spin_lock(&target->d_parent->d_lock);
2172 spin_lock_nested(&dentry->d_parent->d_lock,
2173 DENTRY_D_LOCK_NESTED);
2174 }
2175 }
2176 if (target < dentry) {
2177 spin_lock_nested(&target->d_lock, 2);
2178 spin_lock_nested(&dentry->d_lock, 3);
2179 } else {
2180 spin_lock_nested(&dentry->d_lock, 2);
2181 spin_lock_nested(&target->d_lock, 3);
2182 }
2183 }
2184
2185 static void dentry_unlock_parents_for_move(struct dentry *dentry,
2186 struct dentry *target)
2187 {
2188 if (target->d_parent != dentry->d_parent)
2189 spin_unlock(&dentry->d_parent->d_lock);
2190 if (target->d_parent != target)
2191 spin_unlock(&target->d_parent->d_lock);
2192 }
2193
2194 /*
2195 * When switching names, the actual string doesn't strictly have to
2196 * be preserved in the target - because we're dropping the target
2197 * anyway. As such, we can just do a simple memcpy() to copy over
2198 * the new name before we switch.
2199 *
2200 * Note that we have to be a lot more careful about getting the hash
2201 * switched - we have to switch the hash value properly even if it
2202 * then no longer matches the actual (corrupted) string of the target.
2203 * The hash value has to match the hash queue that the dentry is on..
2204 */
2205 /*
2206 * __d_move - move a dentry
2207 * @dentry: entry to move
2208 * @target: new dentry
2209 *
2210 * Update the dcache to reflect the move of a file name. Negative
2211 * dcache entries should not be moved in this way. Caller must hold
2212 * rename_lock, the i_mutex of the source and target directories,
2213 * and the sb->s_vfs_rename_mutex if they differ. See lock_rename().
2214 */
2215 static void __d_move(struct dentry * dentry, struct dentry * target)
2216 {
2217 if (!dentry->d_inode)
2218 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
2219
2220 BUG_ON(d_ancestor(dentry, target));
2221 BUG_ON(d_ancestor(target, dentry));
2222
2223 dentry_lock_for_move(dentry, target);
2224
2225 write_seqcount_begin(&dentry->d_seq);
2226 write_seqcount_begin(&target->d_seq);
2227
2228 /* __d_drop does write_seqcount_barrier, but they're OK to nest. */
2229
2230 /*
2231 * Move the dentry to the target hash queue. Don't bother checking
2232 * for the same hash queue because of how unlikely it is.
2233 */
2234 __d_drop(dentry);
2235 __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash));
2236
2237 /* Unhash the target: dput() will then get rid of it */
2238 __d_drop(target);
2239
2240 list_del(&dentry->d_u.d_child);
2241 list_del(&target->d_u.d_child);
2242
2243 /* Switch the names.. */
2244 switch_names(dentry, target);
2245 swap(dentry->d_name.hash, target->d_name.hash);
2246
2247 /* ... and switch the parents */
2248 if (IS_ROOT(dentry)) {
2249 dentry->d_parent = target->d_parent;
2250 target->d_parent = target;
2251 INIT_LIST_HEAD(&target->d_u.d_child);
2252 } else {
2253 swap(dentry->d_parent, target->d_parent);
2254
2255 /* And add them back to the (new) parent lists */
2256 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
2257 }
2258
2259 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2260
2261 write_seqcount_end(&target->d_seq);
2262 write_seqcount_end(&dentry->d_seq);
2263
2264 dentry_unlock_parents_for_move(dentry, target);
2265 spin_unlock(&target->d_lock);
2266 fsnotify_d_move(dentry);
2267 spin_unlock(&dentry->d_lock);
2268 }
2269
2270 /*
2271 * d_move - move a dentry
2272 * @dentry: entry to move
2273 * @target: new dentry
2274 *
2275 * Update the dcache to reflect the move of a file name. Negative
2276 * dcache entries should not be moved in this way. See the locking
2277 * requirements for __d_move.
2278 */
2279 void d_move(struct dentry *dentry, struct dentry *target)
2280 {
2281 write_seqlock(&rename_lock);
2282 __d_move(dentry, target);
2283 write_sequnlock(&rename_lock);
2284 }
2285 EXPORT_SYMBOL(d_move);
2286
2287 /**
2288 * d_ancestor - search for an ancestor
2289 * @p1: ancestor dentry
2290 * @p2: child dentry
2291 *
2292 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
2293 * an ancestor of p2, else NULL.
2294 */
2295 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
2296 {
2297 struct dentry *p;
2298
2299 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
2300 if (p->d_parent == p1)
2301 return p;
2302 }
2303 return NULL;
2304 }
2305
2306 /*
2307 * This helper attempts to cope with remotely renamed directories
2308 *
2309 * It assumes that the caller is already holding
2310 * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock
2311 *
2312 * Note: If ever the locking in lock_rename() changes, then please
2313 * remember to update this too...
2314 */
2315 static struct dentry *__d_unalias(struct inode *inode,
2316 struct dentry *dentry, struct dentry *alias)
2317 {
2318 struct mutex *m1 = NULL, *m2 = NULL;
2319 struct dentry *ret;
2320
2321 /* If alias and dentry share a parent, then no extra locks required */
2322 if (alias->d_parent == dentry->d_parent)
2323 goto out_unalias;
2324
2325 /* See lock_rename() */
2326 ret = ERR_PTR(-EBUSY);
2327 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
2328 goto out_err;
2329 m1 = &dentry->d_sb->s_vfs_rename_mutex;
2330 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
2331 goto out_err;
2332 m2 = &alias->d_parent->d_inode->i_mutex;
2333 out_unalias:
2334 __d_move(alias, dentry);
2335 ret = alias;
2336 out_err:
2337 spin_unlock(&inode->i_lock);
2338 if (m2)
2339 mutex_unlock(m2);
2340 if (m1)
2341 mutex_unlock(m1);
2342 return ret;
2343 }
2344
2345 /*
2346 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
2347 * named dentry in place of the dentry to be replaced.
2348 * returns with anon->d_lock held!
2349 */
2350 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
2351 {
2352 struct dentry *dparent, *aparent;
2353
2354 dentry_lock_for_move(anon, dentry);
2355
2356 write_seqcount_begin(&dentry->d_seq);
2357 write_seqcount_begin(&anon->d_seq);
2358
2359 dparent = dentry->d_parent;
2360 aparent = anon->d_parent;
2361
2362 switch_names(dentry, anon);
2363 swap(dentry->d_name.hash, anon->d_name.hash);
2364
2365 dentry->d_parent = (aparent == anon) ? dentry : aparent;
2366 list_del(&dentry->d_u.d_child);
2367 if (!IS_ROOT(dentry))
2368 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
2369 else
2370 INIT_LIST_HEAD(&dentry->d_u.d_child);
2371
2372 anon->d_parent = (dparent == dentry) ? anon : dparent;
2373 list_del(&anon->d_u.d_child);
2374 if (!IS_ROOT(anon))
2375 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
2376 else
2377 INIT_LIST_HEAD(&anon->d_u.d_child);
2378
2379 write_seqcount_end(&dentry->d_seq);
2380 write_seqcount_end(&anon->d_seq);
2381
2382 dentry_unlock_parents_for_move(anon, dentry);
2383 spin_unlock(&dentry->d_lock);
2384
2385 /* anon->d_lock still locked, returns locked */
2386 anon->d_flags &= ~DCACHE_DISCONNECTED;
2387 }
2388
2389 /**
2390 * d_materialise_unique - introduce an inode into the tree
2391 * @dentry: candidate dentry
2392 * @inode: inode to bind to the dentry, to which aliases may be attached
2393 *
2394 * Introduces an dentry into the tree, substituting an extant disconnected
2395 * root directory alias in its place if there is one. Caller must hold the
2396 * i_mutex of the parent directory.
2397 */
2398 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
2399 {
2400 struct dentry *actual;
2401
2402 BUG_ON(!d_unhashed(dentry));
2403
2404 if (!inode) {
2405 actual = dentry;
2406 __d_instantiate(dentry, NULL);
2407 d_rehash(actual);
2408 goto out_nolock;
2409 }
2410
2411 spin_lock(&inode->i_lock);
2412
2413 if (S_ISDIR(inode->i_mode)) {
2414 struct dentry *alias;
2415
2416 /* Does an aliased dentry already exist? */
2417 alias = __d_find_alias(inode, 0);
2418 if (alias) {
2419 actual = alias;
2420 write_seqlock(&rename_lock);
2421
2422 if (d_ancestor(alias, dentry)) {
2423 /* Check for loops */
2424 actual = ERR_PTR(-ELOOP);
2425 spin_unlock(&inode->i_lock);
2426 } else if (IS_ROOT(alias)) {
2427 /* Is this an anonymous mountpoint that we
2428 * could splice into our tree? */
2429 __d_materialise_dentry(dentry, alias);
2430 write_sequnlock(&rename_lock);
2431 __d_drop(alias);
2432 goto found;
2433 } else {
2434 /* Nope, but we must(!) avoid directory
2435 * aliasing. This drops inode->i_lock */
2436 actual = __d_unalias(inode, dentry, alias);
2437 }
2438 write_sequnlock(&rename_lock);
2439 if (IS_ERR(actual)) {
2440 if (PTR_ERR(actual) == -ELOOP)
2441 pr_warn_ratelimited(
2442 "VFS: Lookup of '%s' in %s %s"
2443 " would have caused loop\n",
2444 dentry->d_name.name,
2445 inode->i_sb->s_type->name,
2446 inode->i_sb->s_id);
2447 dput(alias);
2448 }
2449 goto out_nolock;
2450 }
2451 }
2452
2453 /* Add a unique reference */
2454 actual = __d_instantiate_unique(dentry, inode);
2455 if (!actual)
2456 actual = dentry;
2457 else
2458 BUG_ON(!d_unhashed(actual));
2459
2460 spin_lock(&actual->d_lock);
2461 found:
2462 _d_rehash(actual);
2463 spin_unlock(&actual->d_lock);
2464 spin_unlock(&inode->i_lock);
2465 out_nolock:
2466 if (actual == dentry) {
2467 security_d_instantiate(dentry, inode);
2468 return NULL;
2469 }
2470
2471 iput(inode);
2472 return actual;
2473 }
2474 EXPORT_SYMBOL_GPL(d_materialise_unique);
2475
2476 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
2477 {
2478 *buflen -= namelen;
2479 if (*buflen < 0)
2480 return -ENAMETOOLONG;
2481 *buffer -= namelen;
2482 memcpy(*buffer, str, namelen);
2483 return 0;
2484 }
2485
2486 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
2487 {
2488 return prepend(buffer, buflen, name->name, name->len);
2489 }
2490
2491 /**
2492 * prepend_path - Prepend path string to a buffer
2493 * @path: the dentry/vfsmount to report
2494 * @root: root vfsmnt/dentry
2495 * @buffer: pointer to the end of the buffer
2496 * @buflen: pointer to buffer length
2497 *
2498 * Caller holds the rename_lock.
2499 */
2500 static int prepend_path(const struct path *path,
2501 const struct path *root,
2502 char **buffer, int *buflen)
2503 {
2504 struct dentry *dentry = path->dentry;
2505 struct vfsmount *vfsmnt = path->mnt;
2506 struct mount *mnt = real_mount(vfsmnt);
2507 bool slash = false;
2508 int error = 0;
2509
2510 br_read_lock(vfsmount_lock);
2511 while (dentry != root->dentry || vfsmnt != root->mnt) {
2512 struct dentry * parent;
2513
2514 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
2515 /* Global root? */
2516 if (!mnt_has_parent(mnt))
2517 goto global_root;
2518 dentry = mnt->mnt_mountpoint;
2519 mnt = mnt->mnt_parent;
2520 vfsmnt = &mnt->mnt;
2521 continue;
2522 }
2523 parent = dentry->d_parent;
2524 prefetch(parent);
2525 spin_lock(&dentry->d_lock);
2526 error = prepend_name(buffer, buflen, &dentry->d_name);
2527 spin_unlock(&dentry->d_lock);
2528 if (!error)
2529 error = prepend(buffer, buflen, "/", 1);
2530 if (error)
2531 break;
2532
2533 slash = true;
2534 dentry = parent;
2535 }
2536
2537 if (!error && !slash)
2538 error = prepend(buffer, buflen, "/", 1);
2539
2540 out:
2541 br_read_unlock(vfsmount_lock);
2542 return error;
2543
2544 global_root:
2545 /*
2546 * Filesystems needing to implement special "root names"
2547 * should do so with ->d_dname()
2548 */
2549 if (IS_ROOT(dentry) &&
2550 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
2551 WARN(1, "Root dentry has weird name <%.*s>\n",
2552 (int) dentry->d_name.len, dentry->d_name.name);
2553 }
2554 if (!slash)
2555 error = prepend(buffer, buflen, "/", 1);
2556 if (!error)
2557 error = real_mount(vfsmnt)->mnt_ns ? 1 : 2;
2558 goto out;
2559 }
2560
2561 /**
2562 * __d_path - return the path of a dentry
2563 * @path: the dentry/vfsmount to report
2564 * @root: root vfsmnt/dentry
2565 * @buf: buffer to return value in
2566 * @buflen: buffer length
2567 *
2568 * Convert a dentry into an ASCII path name.
2569 *
2570 * Returns a pointer into the buffer or an error code if the
2571 * path was too long.
2572 *
2573 * "buflen" should be positive.
2574 *
2575 * If the path is not reachable from the supplied root, return %NULL.
2576 */
2577 char *__d_path(const struct path *path,
2578 const struct path *root,
2579 char *buf, int buflen)
2580 {
2581 char *res = buf + buflen;
2582 int error;
2583
2584 prepend(&res, &buflen, "\0", 1);
2585 write_seqlock(&rename_lock);
2586 error = prepend_path(path, root, &res, &buflen);
2587 write_sequnlock(&rename_lock);
2588
2589 if (error < 0)
2590 return ERR_PTR(error);
2591 if (error > 0)
2592 return NULL;
2593 return res;
2594 }
2595
2596 char *d_absolute_path(const struct path *path,
2597 char *buf, int buflen)
2598 {
2599 struct path root = {};
2600 char *res = buf + buflen;
2601 int error;
2602
2603 prepend(&res, &buflen, "\0", 1);
2604 write_seqlock(&rename_lock);
2605 error = prepend_path(path, &root, &res, &buflen);
2606 write_sequnlock(&rename_lock);
2607
2608 if (error > 1)
2609 error = -EINVAL;
2610 if (error < 0)
2611 return ERR_PTR(error);
2612 return res;
2613 }
2614
2615 /*
2616 * same as __d_path but appends "(deleted)" for unlinked files.
2617 */
2618 static int path_with_deleted(const struct path *path,
2619 const struct path *root,
2620 char **buf, int *buflen)
2621 {
2622 prepend(buf, buflen, "\0", 1);
2623 if (d_unlinked(path->dentry)) {
2624 int error = prepend(buf, buflen, " (deleted)", 10);
2625 if (error)
2626 return error;
2627 }
2628
2629 return prepend_path(path, root, buf, buflen);
2630 }
2631
2632 static int prepend_unreachable(char **buffer, int *buflen)
2633 {
2634 return prepend(buffer, buflen, "(unreachable)", 13);
2635 }
2636
2637 /**
2638 * d_path - return the path of a dentry
2639 * @path: path to report
2640 * @buf: buffer to return value in
2641 * @buflen: buffer length
2642 *
2643 * Convert a dentry into an ASCII path name. If the entry has been deleted
2644 * the string " (deleted)" is appended. Note that this is ambiguous.
2645 *
2646 * Returns a pointer into the buffer or an error code if the path was
2647 * too long. Note: Callers should use the returned pointer, not the passed
2648 * in buffer, to use the name! The implementation often starts at an offset
2649 * into the buffer, and may leave 0 bytes at the start.
2650 *
2651 * "buflen" should be positive.
2652 */
2653 char *d_path(const struct path *path, char *buf, int buflen)
2654 {
2655 char *res = buf + buflen;
2656 struct path root;
2657 int error;
2658
2659 /*
2660 * We have various synthetic filesystems that never get mounted. On
2661 * these filesystems dentries are never used for lookup purposes, and
2662 * thus don't need to be hashed. They also don't need a name until a
2663 * user wants to identify the object in /proc/pid/fd/. The little hack
2664 * below allows us to generate a name for these objects on demand:
2665 */
2666 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2667 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2668
2669 get_fs_root(current->fs, &root);
2670 write_seqlock(&rename_lock);
2671 error = path_with_deleted(path, &root, &res, &buflen);
2672 if (error < 0)
2673 res = ERR_PTR(error);
2674 write_sequnlock(&rename_lock);
2675 path_put(&root);
2676 return res;
2677 }
2678 EXPORT_SYMBOL(d_path);
2679
2680 /**
2681 * d_path_with_unreachable - return the path of a dentry
2682 * @path: path to report
2683 * @buf: buffer to return value in
2684 * @buflen: buffer length
2685 *
2686 * The difference from d_path() is that this prepends "(unreachable)"
2687 * to paths which are unreachable from the current process' root.
2688 */
2689 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2690 {
2691 char *res = buf + buflen;
2692 struct path root;
2693 int error;
2694
2695 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2696 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2697
2698 get_fs_root(current->fs, &root);
2699 write_seqlock(&rename_lock);
2700 error = path_with_deleted(path, &root, &res, &buflen);
2701 if (error > 0)
2702 error = prepend_unreachable(&res, &buflen);
2703 write_sequnlock(&rename_lock);
2704 path_put(&root);
2705 if (error)
2706 res = ERR_PTR(error);
2707
2708 return res;
2709 }
2710
2711 /*
2712 * Helper function for dentry_operations.d_dname() members
2713 */
2714 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2715 const char *fmt, ...)
2716 {
2717 va_list args;
2718 char temp[64];
2719 int sz;
2720
2721 va_start(args, fmt);
2722 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2723 va_end(args);
2724
2725 if (sz > sizeof(temp) || sz > buflen)
2726 return ERR_PTR(-ENAMETOOLONG);
2727
2728 buffer += buflen - sz;
2729 return memcpy(buffer, temp, sz);
2730 }
2731
2732 /*
2733 * Write full pathname from the root of the filesystem into the buffer.
2734 */
2735 static char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2736 {
2737 char *end = buf + buflen;
2738 char *retval;
2739
2740 prepend(&end, &buflen, "\0", 1);
2741 if (buflen < 1)
2742 goto Elong;
2743 /* Get '/' right */
2744 retval = end-1;
2745 *retval = '/';
2746
2747 while (!IS_ROOT(dentry)) {
2748 struct dentry *parent = dentry->d_parent;
2749 int error;
2750
2751 prefetch(parent);
2752 spin_lock(&dentry->d_lock);
2753 error = prepend_name(&end, &buflen, &dentry->d_name);
2754 spin_unlock(&dentry->d_lock);
2755 if (error != 0 || prepend(&end, &buflen, "/", 1) != 0)
2756 goto Elong;
2757
2758 retval = end;
2759 dentry = parent;
2760 }
2761 return retval;
2762 Elong:
2763 return ERR_PTR(-ENAMETOOLONG);
2764 }
2765
2766 char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen)
2767 {
2768 char *retval;
2769
2770 write_seqlock(&rename_lock);
2771 retval = __dentry_path(dentry, buf, buflen);
2772 write_sequnlock(&rename_lock);
2773
2774 return retval;
2775 }
2776 EXPORT_SYMBOL(dentry_path_raw);
2777
2778 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2779 {
2780 char *p = NULL;
2781 char *retval;
2782
2783 write_seqlock(&rename_lock);
2784 if (d_unlinked(dentry)) {
2785 p = buf + buflen;
2786 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2787 goto Elong;
2788 buflen++;
2789 }
2790 retval = __dentry_path(dentry, buf, buflen);
2791 write_sequnlock(&rename_lock);
2792 if (!IS_ERR(retval) && p)
2793 *p = '/'; /* restore '/' overriden with '\0' */
2794 return retval;
2795 Elong:
2796 return ERR_PTR(-ENAMETOOLONG);
2797 }
2798
2799 /*
2800 * NOTE! The user-level library version returns a
2801 * character pointer. The kernel system call just
2802 * returns the length of the buffer filled (which
2803 * includes the ending '\0' character), or a negative
2804 * error value. So libc would do something like
2805 *
2806 * char *getcwd(char * buf, size_t size)
2807 * {
2808 * int retval;
2809 *
2810 * retval = sys_getcwd(buf, size);
2811 * if (retval >= 0)
2812 * return buf;
2813 * errno = -retval;
2814 * return NULL;
2815 * }
2816 */
2817 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2818 {
2819 int error;
2820 struct path pwd, root;
2821 char *page = (char *) __get_free_page(GFP_USER);
2822
2823 if (!page)
2824 return -ENOMEM;
2825
2826 get_fs_root_and_pwd(current->fs, &root, &pwd);
2827
2828 error = -ENOENT;
2829 write_seqlock(&rename_lock);
2830 if (!d_unlinked(pwd.dentry)) {
2831 unsigned long len;
2832 char *cwd = page + PAGE_SIZE;
2833 int buflen = PAGE_SIZE;
2834
2835 prepend(&cwd, &buflen, "\0", 1);
2836 error = prepend_path(&pwd, &root, &cwd, &buflen);
2837 write_sequnlock(&rename_lock);
2838
2839 if (error < 0)
2840 goto out;
2841
2842 /* Unreachable from current root */
2843 if (error > 0) {
2844 error = prepend_unreachable(&cwd, &buflen);
2845 if (error)
2846 goto out;
2847 }
2848
2849 error = -ERANGE;
2850 len = PAGE_SIZE + page - cwd;
2851 if (len <= size) {
2852 error = len;
2853 if (copy_to_user(buf, cwd, len))
2854 error = -EFAULT;
2855 }
2856 } else {
2857 write_sequnlock(&rename_lock);
2858 }
2859
2860 out:
2861 path_put(&pwd);
2862 path_put(&root);
2863 free_page((unsigned long) page);
2864 return error;
2865 }
2866
2867 /*
2868 * Test whether new_dentry is a subdirectory of old_dentry.
2869 *
2870 * Trivially implemented using the dcache structure
2871 */
2872
2873 /**
2874 * is_subdir - is new dentry a subdirectory of old_dentry
2875 * @new_dentry: new dentry
2876 * @old_dentry: old dentry
2877 *
2878 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2879 * Returns 0 otherwise.
2880 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2881 */
2882
2883 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2884 {
2885 int result;
2886 unsigned seq;
2887
2888 if (new_dentry == old_dentry)
2889 return 1;
2890
2891 do {
2892 /* for restarting inner loop in case of seq retry */
2893 seq = read_seqbegin(&rename_lock);
2894 /*
2895 * Need rcu_readlock to protect against the d_parent trashing
2896 * due to d_move
2897 */
2898 rcu_read_lock();
2899 if (d_ancestor(old_dentry, new_dentry))
2900 result = 1;
2901 else
2902 result = 0;
2903 rcu_read_unlock();
2904 } while (read_seqretry(&rename_lock, seq));
2905
2906 return result;
2907 }
2908
2909 void d_genocide(struct dentry *root)
2910 {
2911 struct dentry *this_parent;
2912 struct list_head *next;
2913 unsigned seq;
2914 int locked = 0;
2915
2916 seq = read_seqbegin(&rename_lock);
2917 again:
2918 this_parent = root;
2919 spin_lock(&this_parent->d_lock);
2920 repeat:
2921 next = this_parent->d_subdirs.next;
2922 resume:
2923 while (next != &this_parent->d_subdirs) {
2924 struct list_head *tmp = next;
2925 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2926 next = tmp->next;
2927
2928 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
2929 if (d_unhashed(dentry) || !dentry->d_inode) {
2930 spin_unlock(&dentry->d_lock);
2931 continue;
2932 }
2933 if (!list_empty(&dentry->d_subdirs)) {
2934 spin_unlock(&this_parent->d_lock);
2935 spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_);
2936 this_parent = dentry;
2937 spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_);
2938 goto repeat;
2939 }
2940 if (!(dentry->d_flags & DCACHE_GENOCIDE)) {
2941 dentry->d_flags |= DCACHE_GENOCIDE;
2942 dentry->d_count--;
2943 }
2944 spin_unlock(&dentry->d_lock);
2945 }
2946 if (this_parent != root) {
2947 struct dentry *child = this_parent;
2948 if (!(this_parent->d_flags & DCACHE_GENOCIDE)) {
2949 this_parent->d_flags |= DCACHE_GENOCIDE;
2950 this_parent->d_count--;
2951 }
2952 this_parent = try_to_ascend(this_parent, locked, seq);
2953 if (!this_parent)
2954 goto rename_retry;
2955 next = child->d_u.d_child.next;
2956 goto resume;
2957 }
2958 spin_unlock(&this_parent->d_lock);
2959 if (!locked && read_seqretry(&rename_lock, seq))
2960 goto rename_retry;
2961 if (locked)
2962 write_sequnlock(&rename_lock);
2963 return;
2964
2965 rename_retry:
2966 locked = 1;
2967 write_seqlock(&rename_lock);
2968 goto again;
2969 }
2970
2971 /**
2972 * find_inode_number - check for dentry with name
2973 * @dir: directory to check
2974 * @name: Name to find.
2975 *
2976 * Check whether a dentry already exists for the given name,
2977 * and return the inode number if it has an inode. Otherwise
2978 * 0 is returned.
2979 *
2980 * This routine is used to post-process directory listings for
2981 * filesystems using synthetic inode numbers, and is necessary
2982 * to keep getcwd() working.
2983 */
2984
2985 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2986 {
2987 struct dentry * dentry;
2988 ino_t ino = 0;
2989
2990 dentry = d_hash_and_lookup(dir, name);
2991 if (dentry) {
2992 if (dentry->d_inode)
2993 ino = dentry->d_inode->i_ino;
2994 dput(dentry);
2995 }
2996 return ino;
2997 }
2998 EXPORT_SYMBOL(find_inode_number);
2999
3000 static __initdata unsigned long dhash_entries;
3001 static int __init set_dhash_entries(char *str)
3002 {
3003 if (!str)
3004 return 0;
3005 dhash_entries = simple_strtoul(str, &str, 0);
3006 return 1;
3007 }
3008 __setup("dhash_entries=", set_dhash_entries);
3009
3010 static void __init dcache_init_early(void)
3011 {
3012 unsigned int loop;
3013
3014 /* If hashes are distributed across NUMA nodes, defer
3015 * hash allocation until vmalloc space is available.
3016 */
3017 if (hashdist)
3018 return;
3019
3020 dentry_hashtable =
3021 alloc_large_system_hash("Dentry cache",
3022 sizeof(struct hlist_bl_head),
3023 dhash_entries,
3024 13,
3025 HASH_EARLY,
3026 &d_hash_shift,
3027 &d_hash_mask,
3028 0);
3029
3030 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3031 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3032 }
3033
3034 static void __init dcache_init(void)
3035 {
3036 unsigned int loop;
3037
3038 /*
3039 * A constructor could be added for stable state like the lists,
3040 * but it is probably not worth it because of the cache nature
3041 * of the dcache.
3042 */
3043 dentry_cache = KMEM_CACHE(dentry,
3044 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
3045
3046 /* Hash may have been set up in dcache_init_early */
3047 if (!hashdist)
3048 return;
3049
3050 dentry_hashtable =
3051 alloc_large_system_hash("Dentry cache",
3052 sizeof(struct hlist_bl_head),
3053 dhash_entries,
3054 13,
3055 0,
3056 &d_hash_shift,
3057 &d_hash_mask,
3058 0);
3059
3060 for (loop = 0; loop < (1U << d_hash_shift); loop++)
3061 INIT_HLIST_BL_HEAD(dentry_hashtable + loop);
3062 }
3063
3064 /* SLAB cache for __getname() consumers */
3065 struct kmem_cache *names_cachep __read_mostly;
3066 EXPORT_SYMBOL(names_cachep);
3067
3068 EXPORT_SYMBOL(d_genocide);
3069
3070 void __init vfs_caches_init_early(void)
3071 {
3072 dcache_init_early();
3073 inode_init_early();
3074 }
3075
3076 void __init vfs_caches_init(unsigned long mempages)
3077 {
3078 unsigned long reserve;
3079
3080 /* Base hash sizes on available memory, with a reserve equal to
3081 150% of current kernel size */
3082
3083 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
3084 mempages -= reserve;
3085
3086 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
3087 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
3088
3089 dcache_init();
3090 inode_init();
3091 files_init(mempages);
3092 mnt_init();
3093 bdev_cache_init();
3094 chrdev_init();
3095 }