]> git.proxmox.com Git - mirror_ubuntu-zesty-kernel.git/blob - fs/dcache.c
fs: use percpu counter for nr_dentry and nr_dentry_unused
[mirror_ubuntu-zesty-kernel.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/module.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 "internal.h"
37
38 int sysctl_vfs_cache_pressure __read_mostly = 100;
39 EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
40
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
42 __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
43
44 EXPORT_SYMBOL(dcache_lock);
45
46 static struct kmem_cache *dentry_cache __read_mostly;
47
48 #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname))
49
50 /*
51 * This is the single most critical data structure when it comes
52 * to the dcache: the hashtable for lookups. Somebody should try
53 * to make this good - I've just made it work.
54 *
55 * This hash-function tries to avoid losing too many bits of hash
56 * information, yet avoid using a prime hash-size or similar.
57 */
58 #define D_HASHBITS d_hash_shift
59 #define D_HASHMASK d_hash_mask
60
61 static unsigned int d_hash_mask __read_mostly;
62 static unsigned int d_hash_shift __read_mostly;
63 static struct hlist_head *dentry_hashtable __read_mostly;
64
65 /* Statistics gathering. */
66 struct dentry_stat_t dentry_stat = {
67 .age_limit = 45,
68 };
69
70 static struct percpu_counter nr_dentry __cacheline_aligned_in_smp;
71 static struct percpu_counter nr_dentry_unused __cacheline_aligned_in_smp;
72
73 #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS)
74 int proc_nr_dentry(ctl_table *table, int write, void __user *buffer,
75 size_t *lenp, loff_t *ppos)
76 {
77 dentry_stat.nr_dentry = percpu_counter_sum_positive(&nr_dentry);
78 dentry_stat.nr_unused = percpu_counter_sum_positive(&nr_dentry_unused);
79 return proc_dointvec(table, write, buffer, lenp, ppos);
80 }
81 #endif
82
83 static void __d_free(struct rcu_head *head)
84 {
85 struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu);
86
87 WARN_ON(!list_empty(&dentry->d_alias));
88 if (dname_external(dentry))
89 kfree(dentry->d_name.name);
90 kmem_cache_free(dentry_cache, dentry);
91 }
92
93 /*
94 * no dcache_lock, please.
95 */
96 static void d_free(struct dentry *dentry)
97 {
98 percpu_counter_dec(&nr_dentry);
99 if (dentry->d_op && dentry->d_op->d_release)
100 dentry->d_op->d_release(dentry);
101
102 /* if dentry was never inserted into hash, immediate free is OK */
103 if (hlist_unhashed(&dentry->d_hash))
104 __d_free(&dentry->d_u.d_rcu);
105 else
106 call_rcu(&dentry->d_u.d_rcu, __d_free);
107 }
108
109 /*
110 * Release the dentry's inode, using the filesystem
111 * d_iput() operation if defined.
112 */
113 static void dentry_iput(struct dentry * dentry)
114 __releases(dentry->d_lock)
115 __releases(dcache_lock)
116 {
117 struct inode *inode = dentry->d_inode;
118 if (inode) {
119 dentry->d_inode = NULL;
120 list_del_init(&dentry->d_alias);
121 spin_unlock(&dentry->d_lock);
122 spin_unlock(&dcache_lock);
123 if (!inode->i_nlink)
124 fsnotify_inoderemove(inode);
125 if (dentry->d_op && dentry->d_op->d_iput)
126 dentry->d_op->d_iput(dentry, inode);
127 else
128 iput(inode);
129 } else {
130 spin_unlock(&dentry->d_lock);
131 spin_unlock(&dcache_lock);
132 }
133 }
134
135 /*
136 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
137 */
138 static void dentry_lru_add(struct dentry *dentry)
139 {
140 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
141 dentry->d_sb->s_nr_dentry_unused++;
142 percpu_counter_inc(&nr_dentry_unused);
143 }
144
145 static void dentry_lru_add_tail(struct dentry *dentry)
146 {
147 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
148 dentry->d_sb->s_nr_dentry_unused++;
149 percpu_counter_inc(&nr_dentry_unused);
150 }
151
152 static void dentry_lru_del(struct dentry *dentry)
153 {
154 if (!list_empty(&dentry->d_lru)) {
155 list_del(&dentry->d_lru);
156 dentry->d_sb->s_nr_dentry_unused--;
157 percpu_counter_dec(&nr_dentry_unused);
158 }
159 }
160
161 static void dentry_lru_del_init(struct dentry *dentry)
162 {
163 if (likely(!list_empty(&dentry->d_lru))) {
164 list_del_init(&dentry->d_lru);
165 dentry->d_sb->s_nr_dentry_unused--;
166 percpu_counter_dec(&nr_dentry_unused);
167 }
168 }
169
170 /**
171 * d_kill - kill dentry and return parent
172 * @dentry: dentry to kill
173 *
174 * The dentry must already be unhashed and removed from the LRU.
175 *
176 * If this is the root of the dentry tree, return NULL.
177 */
178 static struct dentry *d_kill(struct dentry *dentry)
179 __releases(dentry->d_lock)
180 __releases(dcache_lock)
181 {
182 struct dentry *parent;
183
184 list_del(&dentry->d_u.d_child);
185 /*drops the locks, at that point nobody can reach this dentry */
186 dentry_iput(dentry);
187 if (IS_ROOT(dentry))
188 parent = NULL;
189 else
190 parent = dentry->d_parent;
191 d_free(dentry);
192 return parent;
193 }
194
195 /*
196 * This is dput
197 *
198 * This is complicated by the fact that we do not want to put
199 * dentries that are no longer on any hash chain on the unused
200 * list: we'd much rather just get rid of them immediately.
201 *
202 * However, that implies that we have to traverse the dentry
203 * tree upwards to the parents which might _also_ now be
204 * scheduled for deletion (it may have been only waiting for
205 * its last child to go away).
206 *
207 * This tail recursion is done by hand as we don't want to depend
208 * on the compiler to always get this right (gcc generally doesn't).
209 * Real recursion would eat up our stack space.
210 */
211
212 /*
213 * dput - release a dentry
214 * @dentry: dentry to release
215 *
216 * Release a dentry. This will drop the usage count and if appropriate
217 * call the dentry unlink method as well as removing it from the queues and
218 * releasing its resources. If the parent dentries were scheduled for release
219 * they too may now get deleted.
220 *
221 * no dcache lock, please.
222 */
223
224 void dput(struct dentry *dentry)
225 {
226 if (!dentry)
227 return;
228
229 repeat:
230 if (atomic_read(&dentry->d_count) == 1)
231 might_sleep();
232 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
233 return;
234
235 spin_lock(&dentry->d_lock);
236 if (atomic_read(&dentry->d_count)) {
237 spin_unlock(&dentry->d_lock);
238 spin_unlock(&dcache_lock);
239 return;
240 }
241
242 /*
243 * AV: ->d_delete() is _NOT_ allowed to block now.
244 */
245 if (dentry->d_op && dentry->d_op->d_delete) {
246 if (dentry->d_op->d_delete(dentry))
247 goto unhash_it;
248 }
249 /* Unreachable? Get rid of it */
250 if (d_unhashed(dentry))
251 goto kill_it;
252 if (list_empty(&dentry->d_lru)) {
253 dentry->d_flags |= DCACHE_REFERENCED;
254 dentry_lru_add(dentry);
255 }
256 spin_unlock(&dentry->d_lock);
257 spin_unlock(&dcache_lock);
258 return;
259
260 unhash_it:
261 __d_drop(dentry);
262 kill_it:
263 /* if dentry was on the d_lru list delete it from there */
264 dentry_lru_del(dentry);
265 dentry = d_kill(dentry);
266 if (dentry)
267 goto repeat;
268 }
269 EXPORT_SYMBOL(dput);
270
271 /**
272 * d_invalidate - invalidate a dentry
273 * @dentry: dentry to invalidate
274 *
275 * Try to invalidate the dentry if it turns out to be
276 * possible. If there are other dentries that can be
277 * reached through this one we can't delete it and we
278 * return -EBUSY. On success we return 0.
279 *
280 * no dcache lock.
281 */
282
283 int d_invalidate(struct dentry * dentry)
284 {
285 /*
286 * If it's already been dropped, return OK.
287 */
288 spin_lock(&dcache_lock);
289 if (d_unhashed(dentry)) {
290 spin_unlock(&dcache_lock);
291 return 0;
292 }
293 /*
294 * Check whether to do a partial shrink_dcache
295 * to get rid of unused child entries.
296 */
297 if (!list_empty(&dentry->d_subdirs)) {
298 spin_unlock(&dcache_lock);
299 shrink_dcache_parent(dentry);
300 spin_lock(&dcache_lock);
301 }
302
303 /*
304 * Somebody else still using it?
305 *
306 * If it's a directory, we can't drop it
307 * for fear of somebody re-populating it
308 * with children (even though dropping it
309 * would make it unreachable from the root,
310 * we might still populate it if it was a
311 * working directory or similar).
312 */
313 spin_lock(&dentry->d_lock);
314 if (atomic_read(&dentry->d_count) > 1) {
315 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
316 spin_unlock(&dentry->d_lock);
317 spin_unlock(&dcache_lock);
318 return -EBUSY;
319 }
320 }
321
322 __d_drop(dentry);
323 spin_unlock(&dentry->d_lock);
324 spin_unlock(&dcache_lock);
325 return 0;
326 }
327 EXPORT_SYMBOL(d_invalidate);
328
329 /* This should be called _only_ with dcache_lock held */
330 static inline struct dentry * __dget_locked(struct dentry *dentry)
331 {
332 atomic_inc(&dentry->d_count);
333 dentry_lru_del_init(dentry);
334 return dentry;
335 }
336
337 struct dentry * dget_locked(struct dentry *dentry)
338 {
339 return __dget_locked(dentry);
340 }
341 EXPORT_SYMBOL(dget_locked);
342
343 /**
344 * d_find_alias - grab a hashed alias of inode
345 * @inode: inode in question
346 * @want_discon: flag, used by d_splice_alias, to request
347 * that only a DISCONNECTED alias be returned.
348 *
349 * If inode has a hashed alias, or is a directory and has any alias,
350 * acquire the reference to alias and return it. Otherwise return NULL.
351 * Notice that if inode is a directory there can be only one alias and
352 * it can be unhashed only if it has no children, or if it is the root
353 * of a filesystem.
354 *
355 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
356 * any other hashed alias over that one unless @want_discon is set,
357 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
358 */
359
360 static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
361 {
362 struct list_head *head, *next, *tmp;
363 struct dentry *alias, *discon_alias=NULL;
364
365 head = &inode->i_dentry;
366 next = inode->i_dentry.next;
367 while (next != head) {
368 tmp = next;
369 next = tmp->next;
370 prefetch(next);
371 alias = list_entry(tmp, struct dentry, d_alias);
372 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
373 if (IS_ROOT(alias) &&
374 (alias->d_flags & DCACHE_DISCONNECTED))
375 discon_alias = alias;
376 else if (!want_discon) {
377 __dget_locked(alias);
378 return alias;
379 }
380 }
381 }
382 if (discon_alias)
383 __dget_locked(discon_alias);
384 return discon_alias;
385 }
386
387 struct dentry * d_find_alias(struct inode *inode)
388 {
389 struct dentry *de = NULL;
390
391 if (!list_empty(&inode->i_dentry)) {
392 spin_lock(&dcache_lock);
393 de = __d_find_alias(inode, 0);
394 spin_unlock(&dcache_lock);
395 }
396 return de;
397 }
398 EXPORT_SYMBOL(d_find_alias);
399
400 /*
401 * Try to kill dentries associated with this inode.
402 * WARNING: you must own a reference to inode.
403 */
404 void d_prune_aliases(struct inode *inode)
405 {
406 struct dentry *dentry;
407 restart:
408 spin_lock(&dcache_lock);
409 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
410 spin_lock(&dentry->d_lock);
411 if (!atomic_read(&dentry->d_count)) {
412 __dget_locked(dentry);
413 __d_drop(dentry);
414 spin_unlock(&dentry->d_lock);
415 spin_unlock(&dcache_lock);
416 dput(dentry);
417 goto restart;
418 }
419 spin_unlock(&dentry->d_lock);
420 }
421 spin_unlock(&dcache_lock);
422 }
423 EXPORT_SYMBOL(d_prune_aliases);
424
425 /*
426 * Throw away a dentry - free the inode, dput the parent. This requires that
427 * the LRU list has already been removed.
428 *
429 * Try to prune ancestors as well. This is necessary to prevent
430 * quadratic behavior of shrink_dcache_parent(), but is also expected
431 * to be beneficial in reducing dentry cache fragmentation.
432 */
433 static void prune_one_dentry(struct dentry * dentry)
434 __releases(dentry->d_lock)
435 __releases(dcache_lock)
436 __acquires(dcache_lock)
437 {
438 __d_drop(dentry);
439 dentry = d_kill(dentry);
440
441 /*
442 * Prune ancestors. Locking is simpler than in dput(),
443 * because dcache_lock needs to be taken anyway.
444 */
445 spin_lock(&dcache_lock);
446 while (dentry) {
447 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
448 return;
449
450 if (dentry->d_op && dentry->d_op->d_delete)
451 dentry->d_op->d_delete(dentry);
452 dentry_lru_del_init(dentry);
453 __d_drop(dentry);
454 dentry = d_kill(dentry);
455 spin_lock(&dcache_lock);
456 }
457 }
458
459 /*
460 * Shrink the dentry LRU on a given superblock.
461 * @sb : superblock to shrink dentry LRU.
462 * @count: If count is NULL, we prune all dentries on superblock.
463 * @flags: If flags is non-zero, we need to do special processing based on
464 * which flags are set. This means we don't need to maintain multiple
465 * similar copies of this loop.
466 */
467 static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
468 {
469 LIST_HEAD(referenced);
470 LIST_HEAD(tmp);
471 struct dentry *dentry;
472 int cnt = 0;
473
474 BUG_ON(!sb);
475 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
476 spin_lock(&dcache_lock);
477 if (count != NULL)
478 /* called from prune_dcache() and shrink_dcache_parent() */
479 cnt = *count;
480 restart:
481 if (count == NULL)
482 list_splice_init(&sb->s_dentry_lru, &tmp);
483 else {
484 while (!list_empty(&sb->s_dentry_lru)) {
485 dentry = list_entry(sb->s_dentry_lru.prev,
486 struct dentry, d_lru);
487 BUG_ON(dentry->d_sb != sb);
488
489 spin_lock(&dentry->d_lock);
490 /*
491 * If we are honouring the DCACHE_REFERENCED flag and
492 * the dentry has this flag set, don't free it. Clear
493 * the flag and put it back on the LRU.
494 */
495 if ((flags & DCACHE_REFERENCED)
496 && (dentry->d_flags & DCACHE_REFERENCED)) {
497 dentry->d_flags &= ~DCACHE_REFERENCED;
498 list_move(&dentry->d_lru, &referenced);
499 spin_unlock(&dentry->d_lock);
500 } else {
501 list_move_tail(&dentry->d_lru, &tmp);
502 spin_unlock(&dentry->d_lock);
503 cnt--;
504 if (!cnt)
505 break;
506 }
507 cond_resched_lock(&dcache_lock);
508 }
509 }
510 while (!list_empty(&tmp)) {
511 dentry = list_entry(tmp.prev, struct dentry, d_lru);
512 dentry_lru_del_init(dentry);
513 spin_lock(&dentry->d_lock);
514 /*
515 * We found an inuse dentry which was not removed from
516 * the LRU because of laziness during lookup. Do not free
517 * it - just keep it off the LRU list.
518 */
519 if (atomic_read(&dentry->d_count)) {
520 spin_unlock(&dentry->d_lock);
521 continue;
522 }
523 prune_one_dentry(dentry);
524 /* dentry->d_lock was dropped in prune_one_dentry() */
525 cond_resched_lock(&dcache_lock);
526 }
527 if (count == NULL && !list_empty(&sb->s_dentry_lru))
528 goto restart;
529 if (count != NULL)
530 *count = cnt;
531 if (!list_empty(&referenced))
532 list_splice(&referenced, &sb->s_dentry_lru);
533 spin_unlock(&dcache_lock);
534 }
535
536 /**
537 * prune_dcache - shrink the dcache
538 * @count: number of entries to try to free
539 *
540 * Shrink the dcache. This is done when we need more memory, or simply when we
541 * need to unmount something (at which point we need to unuse all dentries).
542 *
543 * This function may fail to free any resources if all the dentries are in use.
544 */
545 static void prune_dcache(int count)
546 {
547 struct super_block *sb, *p = NULL;
548 int w_count;
549 int unused = percpu_counter_sum_positive(&nr_dentry_unused);
550 int prune_ratio;
551 int pruned;
552
553 if (unused == 0 || count == 0)
554 return;
555 spin_lock(&dcache_lock);
556 if (count >= unused)
557 prune_ratio = 1;
558 else
559 prune_ratio = unused / count;
560 spin_lock(&sb_lock);
561 list_for_each_entry(sb, &super_blocks, s_list) {
562 if (list_empty(&sb->s_instances))
563 continue;
564 if (sb->s_nr_dentry_unused == 0)
565 continue;
566 sb->s_count++;
567 /* Now, we reclaim unused dentrins with fairness.
568 * We reclaim them same percentage from each superblock.
569 * We calculate number of dentries to scan on this sb
570 * as follows, but the implementation is arranged to avoid
571 * overflows:
572 * number of dentries to scan on this sb =
573 * count * (number of dentries on this sb /
574 * number of dentries in the machine)
575 */
576 spin_unlock(&sb_lock);
577 if (prune_ratio != 1)
578 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
579 else
580 w_count = sb->s_nr_dentry_unused;
581 pruned = w_count;
582 /*
583 * We need to be sure this filesystem isn't being unmounted,
584 * otherwise we could race with generic_shutdown_super(), and
585 * end up holding a reference to an inode while the filesystem
586 * is unmounted. So we try to get s_umount, and make sure
587 * s_root isn't NULL.
588 */
589 if (down_read_trylock(&sb->s_umount)) {
590 if ((sb->s_root != NULL) &&
591 (!list_empty(&sb->s_dentry_lru))) {
592 spin_unlock(&dcache_lock);
593 __shrink_dcache_sb(sb, &w_count,
594 DCACHE_REFERENCED);
595 pruned -= w_count;
596 spin_lock(&dcache_lock);
597 }
598 up_read(&sb->s_umount);
599 }
600 spin_lock(&sb_lock);
601 if (p)
602 __put_super(p);
603 count -= pruned;
604 p = sb;
605 /* more work left to do? */
606 if (count <= 0)
607 break;
608 }
609 if (p)
610 __put_super(p);
611 spin_unlock(&sb_lock);
612 spin_unlock(&dcache_lock);
613 }
614
615 /**
616 * shrink_dcache_sb - shrink dcache for a superblock
617 * @sb: superblock
618 *
619 * Shrink the dcache for the specified super block. This
620 * is used to free the dcache before unmounting a file
621 * system
622 */
623 void shrink_dcache_sb(struct super_block * sb)
624 {
625 __shrink_dcache_sb(sb, NULL, 0);
626 }
627 EXPORT_SYMBOL(shrink_dcache_sb);
628
629 /*
630 * destroy a single subtree of dentries for unmount
631 * - see the comments on shrink_dcache_for_umount() for a description of the
632 * locking
633 */
634 static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
635 {
636 struct dentry *parent;
637 unsigned detached = 0;
638
639 BUG_ON(!IS_ROOT(dentry));
640
641 /* detach this root from the system */
642 spin_lock(&dcache_lock);
643 dentry_lru_del_init(dentry);
644 __d_drop(dentry);
645 spin_unlock(&dcache_lock);
646
647 for (;;) {
648 /* descend to the first leaf in the current subtree */
649 while (!list_empty(&dentry->d_subdirs)) {
650 struct dentry *loop;
651
652 /* this is a branch with children - detach all of them
653 * from the system in one go */
654 spin_lock(&dcache_lock);
655 list_for_each_entry(loop, &dentry->d_subdirs,
656 d_u.d_child) {
657 dentry_lru_del_init(loop);
658 __d_drop(loop);
659 cond_resched_lock(&dcache_lock);
660 }
661 spin_unlock(&dcache_lock);
662
663 /* move to the first child */
664 dentry = list_entry(dentry->d_subdirs.next,
665 struct dentry, d_u.d_child);
666 }
667
668 /* consume the dentries from this leaf up through its parents
669 * until we find one with children or run out altogether */
670 do {
671 struct inode *inode;
672
673 if (atomic_read(&dentry->d_count) != 0) {
674 printk(KERN_ERR
675 "BUG: Dentry %p{i=%lx,n=%s}"
676 " still in use (%d)"
677 " [unmount of %s %s]\n",
678 dentry,
679 dentry->d_inode ?
680 dentry->d_inode->i_ino : 0UL,
681 dentry->d_name.name,
682 atomic_read(&dentry->d_count),
683 dentry->d_sb->s_type->name,
684 dentry->d_sb->s_id);
685 BUG();
686 }
687
688 if (IS_ROOT(dentry))
689 parent = NULL;
690 else {
691 parent = dentry->d_parent;
692 atomic_dec(&parent->d_count);
693 }
694
695 list_del(&dentry->d_u.d_child);
696 detached++;
697
698 inode = dentry->d_inode;
699 if (inode) {
700 dentry->d_inode = NULL;
701 list_del_init(&dentry->d_alias);
702 if (dentry->d_op && dentry->d_op->d_iput)
703 dentry->d_op->d_iput(dentry, inode);
704 else
705 iput(inode);
706 }
707
708 d_free(dentry);
709
710 /* finished when we fall off the top of the tree,
711 * otherwise we ascend to the parent and move to the
712 * next sibling if there is one */
713 if (!parent)
714 return;
715 dentry = parent;
716 } while (list_empty(&dentry->d_subdirs));
717
718 dentry = list_entry(dentry->d_subdirs.next,
719 struct dentry, d_u.d_child);
720 }
721 }
722
723 /*
724 * destroy the dentries attached to a superblock on unmounting
725 * - we don't need to use dentry->d_lock, and only need dcache_lock when
726 * removing the dentry from the system lists and hashes because:
727 * - the superblock is detached from all mountings and open files, so the
728 * dentry trees will not be rearranged by the VFS
729 * - s_umount is write-locked, so the memory pressure shrinker will ignore
730 * any dentries belonging to this superblock that it comes across
731 * - the filesystem itself is no longer permitted to rearrange the dentries
732 * in this superblock
733 */
734 void shrink_dcache_for_umount(struct super_block *sb)
735 {
736 struct dentry *dentry;
737
738 if (down_read_trylock(&sb->s_umount))
739 BUG();
740
741 dentry = sb->s_root;
742 sb->s_root = NULL;
743 atomic_dec(&dentry->d_count);
744 shrink_dcache_for_umount_subtree(dentry);
745
746 while (!hlist_empty(&sb->s_anon)) {
747 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
748 shrink_dcache_for_umount_subtree(dentry);
749 }
750 }
751
752 /*
753 * Search for at least 1 mount point in the dentry's subdirs.
754 * We descend to the next level whenever the d_subdirs
755 * list is non-empty and continue searching.
756 */
757
758 /**
759 * have_submounts - check for mounts over a dentry
760 * @parent: dentry to check.
761 *
762 * Return true if the parent or its subdirectories contain
763 * a mount point
764 */
765
766 int have_submounts(struct dentry *parent)
767 {
768 struct dentry *this_parent = parent;
769 struct list_head *next;
770
771 spin_lock(&dcache_lock);
772 if (d_mountpoint(parent))
773 goto positive;
774 repeat:
775 next = this_parent->d_subdirs.next;
776 resume:
777 while (next != &this_parent->d_subdirs) {
778 struct list_head *tmp = next;
779 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
780 next = tmp->next;
781 /* Have we found a mount point ? */
782 if (d_mountpoint(dentry))
783 goto positive;
784 if (!list_empty(&dentry->d_subdirs)) {
785 this_parent = dentry;
786 goto repeat;
787 }
788 }
789 /*
790 * All done at this level ... ascend and resume the search.
791 */
792 if (this_parent != parent) {
793 next = this_parent->d_u.d_child.next;
794 this_parent = this_parent->d_parent;
795 goto resume;
796 }
797 spin_unlock(&dcache_lock);
798 return 0; /* No mount points found in tree */
799 positive:
800 spin_unlock(&dcache_lock);
801 return 1;
802 }
803 EXPORT_SYMBOL(have_submounts);
804
805 /*
806 * Search the dentry child list for the specified parent,
807 * and move any unused dentries to the end of the unused
808 * list for prune_dcache(). We descend to the next level
809 * whenever the d_subdirs list is non-empty and continue
810 * searching.
811 *
812 * It returns zero iff there are no unused children,
813 * otherwise it returns the number of children moved to
814 * the end of the unused list. This may not be the total
815 * number of unused children, because select_parent can
816 * drop the lock and return early due to latency
817 * constraints.
818 */
819 static int select_parent(struct dentry * parent)
820 {
821 struct dentry *this_parent = parent;
822 struct list_head *next;
823 int found = 0;
824
825 spin_lock(&dcache_lock);
826 repeat:
827 next = this_parent->d_subdirs.next;
828 resume:
829 while (next != &this_parent->d_subdirs) {
830 struct list_head *tmp = next;
831 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
832 next = tmp->next;
833
834 dentry_lru_del_init(dentry);
835 /*
836 * move only zero ref count dentries to the end
837 * of the unused list for prune_dcache
838 */
839 if (!atomic_read(&dentry->d_count)) {
840 dentry_lru_add_tail(dentry);
841 found++;
842 }
843
844 /*
845 * We can return to the caller if we have found some (this
846 * ensures forward progress). We'll be coming back to find
847 * the rest.
848 */
849 if (found && need_resched())
850 goto out;
851
852 /*
853 * Descend a level if the d_subdirs list is non-empty.
854 */
855 if (!list_empty(&dentry->d_subdirs)) {
856 this_parent = dentry;
857 goto repeat;
858 }
859 }
860 /*
861 * All done at this level ... ascend and resume the search.
862 */
863 if (this_parent != parent) {
864 next = this_parent->d_u.d_child.next;
865 this_parent = this_parent->d_parent;
866 goto resume;
867 }
868 out:
869 spin_unlock(&dcache_lock);
870 return found;
871 }
872
873 /**
874 * shrink_dcache_parent - prune dcache
875 * @parent: parent of entries to prune
876 *
877 * Prune the dcache to remove unused children of the parent dentry.
878 */
879
880 void shrink_dcache_parent(struct dentry * parent)
881 {
882 struct super_block *sb = parent->d_sb;
883 int found;
884
885 while ((found = select_parent(parent)) != 0)
886 __shrink_dcache_sb(sb, &found, 0);
887 }
888 EXPORT_SYMBOL(shrink_dcache_parent);
889
890 /*
891 * Scan `nr' dentries and return the number which remain.
892 *
893 * We need to avoid reentering the filesystem if the caller is performing a
894 * GFP_NOFS allocation attempt. One example deadlock is:
895 *
896 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
897 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
898 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
899 *
900 * In this case we return -1 to tell the caller that we baled.
901 */
902 static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
903 {
904 int nr_unused;
905
906 if (nr) {
907 if (!(gfp_mask & __GFP_FS))
908 return -1;
909 prune_dcache(nr);
910 }
911
912 nr_unused = percpu_counter_sum_positive(&nr_dentry_unused);
913 return (nr_unused / 100) * sysctl_vfs_cache_pressure;
914 }
915
916 static struct shrinker dcache_shrinker = {
917 .shrink = shrink_dcache_memory,
918 .seeks = DEFAULT_SEEKS,
919 };
920
921 /**
922 * d_alloc - allocate a dcache entry
923 * @parent: parent of entry to allocate
924 * @name: qstr of the name
925 *
926 * Allocates a dentry. It returns %NULL if there is insufficient memory
927 * available. On a success the dentry is returned. The name passed in is
928 * copied and the copy passed in may be reused after this call.
929 */
930
931 struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
932 {
933 struct dentry *dentry;
934 char *dname;
935
936 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
937 if (!dentry)
938 return NULL;
939
940 if (name->len > DNAME_INLINE_LEN-1) {
941 dname = kmalloc(name->len + 1, GFP_KERNEL);
942 if (!dname) {
943 kmem_cache_free(dentry_cache, dentry);
944 return NULL;
945 }
946 } else {
947 dname = dentry->d_iname;
948 }
949 dentry->d_name.name = dname;
950
951 dentry->d_name.len = name->len;
952 dentry->d_name.hash = name->hash;
953 memcpy(dname, name->name, name->len);
954 dname[name->len] = 0;
955
956 atomic_set(&dentry->d_count, 1);
957 dentry->d_flags = DCACHE_UNHASHED;
958 spin_lock_init(&dentry->d_lock);
959 dentry->d_inode = NULL;
960 dentry->d_parent = NULL;
961 dentry->d_sb = NULL;
962 dentry->d_op = NULL;
963 dentry->d_fsdata = NULL;
964 dentry->d_mounted = 0;
965 INIT_HLIST_NODE(&dentry->d_hash);
966 INIT_LIST_HEAD(&dentry->d_lru);
967 INIT_LIST_HEAD(&dentry->d_subdirs);
968 INIT_LIST_HEAD(&dentry->d_alias);
969
970 if (parent) {
971 dentry->d_parent = dget(parent);
972 dentry->d_sb = parent->d_sb;
973 } else {
974 INIT_LIST_HEAD(&dentry->d_u.d_child);
975 }
976
977 spin_lock(&dcache_lock);
978 if (parent)
979 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
980 spin_unlock(&dcache_lock);
981
982 percpu_counter_inc(&nr_dentry);
983
984 return dentry;
985 }
986 EXPORT_SYMBOL(d_alloc);
987
988 struct dentry *d_alloc_name(struct dentry *parent, const char *name)
989 {
990 struct qstr q;
991
992 q.name = name;
993 q.len = strlen(name);
994 q.hash = full_name_hash(q.name, q.len);
995 return d_alloc(parent, &q);
996 }
997 EXPORT_SYMBOL(d_alloc_name);
998
999 /* the caller must hold dcache_lock */
1000 static void __d_instantiate(struct dentry *dentry, struct inode *inode)
1001 {
1002 if (inode)
1003 list_add(&dentry->d_alias, &inode->i_dentry);
1004 dentry->d_inode = inode;
1005 fsnotify_d_instantiate(dentry, inode);
1006 }
1007
1008 /**
1009 * d_instantiate - fill in inode information for a dentry
1010 * @entry: dentry to complete
1011 * @inode: inode to attach to this dentry
1012 *
1013 * Fill in inode information in the entry.
1014 *
1015 * This turns negative dentries into productive full members
1016 * of society.
1017 *
1018 * NOTE! This assumes that the inode count has been incremented
1019 * (or otherwise set) by the caller to indicate that it is now
1020 * in use by the dcache.
1021 */
1022
1023 void d_instantiate(struct dentry *entry, struct inode * inode)
1024 {
1025 BUG_ON(!list_empty(&entry->d_alias));
1026 spin_lock(&dcache_lock);
1027 __d_instantiate(entry, inode);
1028 spin_unlock(&dcache_lock);
1029 security_d_instantiate(entry, inode);
1030 }
1031 EXPORT_SYMBOL(d_instantiate);
1032
1033 /**
1034 * d_instantiate_unique - instantiate a non-aliased dentry
1035 * @entry: dentry to instantiate
1036 * @inode: inode to attach to this dentry
1037 *
1038 * Fill in inode information in the entry. On success, it returns NULL.
1039 * If an unhashed alias of "entry" already exists, then we return the
1040 * aliased dentry instead and drop one reference to inode.
1041 *
1042 * Note that in order to avoid conflicts with rename() etc, the caller
1043 * had better be holding the parent directory semaphore.
1044 *
1045 * This also assumes that the inode count has been incremented
1046 * (or otherwise set) by the caller to indicate that it is now
1047 * in use by the dcache.
1048 */
1049 static struct dentry *__d_instantiate_unique(struct dentry *entry,
1050 struct inode *inode)
1051 {
1052 struct dentry *alias;
1053 int len = entry->d_name.len;
1054 const char *name = entry->d_name.name;
1055 unsigned int hash = entry->d_name.hash;
1056
1057 if (!inode) {
1058 __d_instantiate(entry, NULL);
1059 return NULL;
1060 }
1061
1062 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1063 struct qstr *qstr = &alias->d_name;
1064
1065 if (qstr->hash != hash)
1066 continue;
1067 if (alias->d_parent != entry->d_parent)
1068 continue;
1069 if (qstr->len != len)
1070 continue;
1071 if (memcmp(qstr->name, name, len))
1072 continue;
1073 dget_locked(alias);
1074 return alias;
1075 }
1076
1077 __d_instantiate(entry, inode);
1078 return NULL;
1079 }
1080
1081 struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1082 {
1083 struct dentry *result;
1084
1085 BUG_ON(!list_empty(&entry->d_alias));
1086
1087 spin_lock(&dcache_lock);
1088 result = __d_instantiate_unique(entry, inode);
1089 spin_unlock(&dcache_lock);
1090
1091 if (!result) {
1092 security_d_instantiate(entry, inode);
1093 return NULL;
1094 }
1095
1096 BUG_ON(!d_unhashed(result));
1097 iput(inode);
1098 return result;
1099 }
1100
1101 EXPORT_SYMBOL(d_instantiate_unique);
1102
1103 /**
1104 * d_alloc_root - allocate root dentry
1105 * @root_inode: inode to allocate the root for
1106 *
1107 * Allocate a root ("/") dentry for the inode given. The inode is
1108 * instantiated and returned. %NULL is returned if there is insufficient
1109 * memory or the inode passed is %NULL.
1110 */
1111
1112 struct dentry * d_alloc_root(struct inode * root_inode)
1113 {
1114 struct dentry *res = NULL;
1115
1116 if (root_inode) {
1117 static const struct qstr name = { .name = "/", .len = 1 };
1118
1119 res = d_alloc(NULL, &name);
1120 if (res) {
1121 res->d_sb = root_inode->i_sb;
1122 res->d_parent = res;
1123 d_instantiate(res, root_inode);
1124 }
1125 }
1126 return res;
1127 }
1128 EXPORT_SYMBOL(d_alloc_root);
1129
1130 static inline struct hlist_head *d_hash(struct dentry *parent,
1131 unsigned long hash)
1132 {
1133 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1134 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1135 return dentry_hashtable + (hash & D_HASHMASK);
1136 }
1137
1138 /**
1139 * d_obtain_alias - find or allocate a dentry for a given inode
1140 * @inode: inode to allocate the dentry for
1141 *
1142 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1143 * similar open by handle operations. The returned dentry may be anonymous,
1144 * or may have a full name (if the inode was already in the cache).
1145 *
1146 * When called on a directory inode, we must ensure that the inode only ever
1147 * has one dentry. If a dentry is found, that is returned instead of
1148 * allocating a new one.
1149 *
1150 * On successful return, the reference to the inode has been transferred
1151 * to the dentry. In case of an error the reference on the inode is released.
1152 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1153 * be passed in and will be the error will be propagate to the return value,
1154 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
1155 */
1156 struct dentry *d_obtain_alias(struct inode *inode)
1157 {
1158 static const struct qstr anonstring = { .name = "" };
1159 struct dentry *tmp;
1160 struct dentry *res;
1161
1162 if (!inode)
1163 return ERR_PTR(-ESTALE);
1164 if (IS_ERR(inode))
1165 return ERR_CAST(inode);
1166
1167 res = d_find_alias(inode);
1168 if (res)
1169 goto out_iput;
1170
1171 tmp = d_alloc(NULL, &anonstring);
1172 if (!tmp) {
1173 res = ERR_PTR(-ENOMEM);
1174 goto out_iput;
1175 }
1176 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1177
1178 spin_lock(&dcache_lock);
1179 res = __d_find_alias(inode, 0);
1180 if (res) {
1181 spin_unlock(&dcache_lock);
1182 dput(tmp);
1183 goto out_iput;
1184 }
1185
1186 /* attach a disconnected dentry */
1187 spin_lock(&tmp->d_lock);
1188 tmp->d_sb = inode->i_sb;
1189 tmp->d_inode = inode;
1190 tmp->d_flags |= DCACHE_DISCONNECTED;
1191 tmp->d_flags &= ~DCACHE_UNHASHED;
1192 list_add(&tmp->d_alias, &inode->i_dentry);
1193 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1194 spin_unlock(&tmp->d_lock);
1195
1196 spin_unlock(&dcache_lock);
1197 return tmp;
1198
1199 out_iput:
1200 iput(inode);
1201 return res;
1202 }
1203 EXPORT_SYMBOL(d_obtain_alias);
1204
1205 /**
1206 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1207 * @inode: the inode which may have a disconnected dentry
1208 * @dentry: a negative dentry which we want to point to the inode.
1209 *
1210 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1211 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1212 * and return it, else simply d_add the inode to the dentry and return NULL.
1213 *
1214 * This is needed in the lookup routine of any filesystem that is exportable
1215 * (via knfsd) so that we can build dcache paths to directories effectively.
1216 *
1217 * If a dentry was found and moved, then it is returned. Otherwise NULL
1218 * is returned. This matches the expected return value of ->lookup.
1219 *
1220 */
1221 struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1222 {
1223 struct dentry *new = NULL;
1224
1225 if (inode && S_ISDIR(inode->i_mode)) {
1226 spin_lock(&dcache_lock);
1227 new = __d_find_alias(inode, 1);
1228 if (new) {
1229 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1230 spin_unlock(&dcache_lock);
1231 security_d_instantiate(new, inode);
1232 d_move(new, dentry);
1233 iput(inode);
1234 } else {
1235 /* already taking dcache_lock, so d_add() by hand */
1236 __d_instantiate(dentry, inode);
1237 spin_unlock(&dcache_lock);
1238 security_d_instantiate(dentry, inode);
1239 d_rehash(dentry);
1240 }
1241 } else
1242 d_add(dentry, inode);
1243 return new;
1244 }
1245 EXPORT_SYMBOL(d_splice_alias);
1246
1247 /**
1248 * d_add_ci - lookup or allocate new dentry with case-exact name
1249 * @inode: the inode case-insensitive lookup has found
1250 * @dentry: the negative dentry that was passed to the parent's lookup func
1251 * @name: the case-exact name to be associated with the returned dentry
1252 *
1253 * This is to avoid filling the dcache with case-insensitive names to the
1254 * same inode, only the actual correct case is stored in the dcache for
1255 * case-insensitive filesystems.
1256 *
1257 * For a case-insensitive lookup match and if the the case-exact dentry
1258 * already exists in in the dcache, use it and return it.
1259 *
1260 * If no entry exists with the exact case name, allocate new dentry with
1261 * the exact case, and return the spliced entry.
1262 */
1263 struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
1264 struct qstr *name)
1265 {
1266 int error;
1267 struct dentry *found;
1268 struct dentry *new;
1269
1270 /*
1271 * First check if a dentry matching the name already exists,
1272 * if not go ahead and create it now.
1273 */
1274 found = d_hash_and_lookup(dentry->d_parent, name);
1275 if (!found) {
1276 new = d_alloc(dentry->d_parent, name);
1277 if (!new) {
1278 error = -ENOMEM;
1279 goto err_out;
1280 }
1281
1282 found = d_splice_alias(inode, new);
1283 if (found) {
1284 dput(new);
1285 return found;
1286 }
1287 return new;
1288 }
1289
1290 /*
1291 * If a matching dentry exists, and it's not negative use it.
1292 *
1293 * Decrement the reference count to balance the iget() done
1294 * earlier on.
1295 */
1296 if (found->d_inode) {
1297 if (unlikely(found->d_inode != inode)) {
1298 /* This can't happen because bad inodes are unhashed. */
1299 BUG_ON(!is_bad_inode(inode));
1300 BUG_ON(!is_bad_inode(found->d_inode));
1301 }
1302 iput(inode);
1303 return found;
1304 }
1305
1306 /*
1307 * Negative dentry: instantiate it unless the inode is a directory and
1308 * already has a dentry.
1309 */
1310 spin_lock(&dcache_lock);
1311 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
1312 __d_instantiate(found, inode);
1313 spin_unlock(&dcache_lock);
1314 security_d_instantiate(found, inode);
1315 return found;
1316 }
1317
1318 /*
1319 * In case a directory already has a (disconnected) entry grab a
1320 * reference to it, move it in place and use it.
1321 */
1322 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1323 dget_locked(new);
1324 spin_unlock(&dcache_lock);
1325 security_d_instantiate(found, inode);
1326 d_move(new, found);
1327 iput(inode);
1328 dput(found);
1329 return new;
1330
1331 err_out:
1332 iput(inode);
1333 return ERR_PTR(error);
1334 }
1335 EXPORT_SYMBOL(d_add_ci);
1336
1337 /**
1338 * d_lookup - search for a dentry
1339 * @parent: parent dentry
1340 * @name: qstr of name we wish to find
1341 * Returns: dentry, or NULL
1342 *
1343 * d_lookup searches the children of the parent dentry for the name in
1344 * question. If the dentry is found its reference count is incremented and the
1345 * dentry is returned. The caller must use dput to free the entry when it has
1346 * finished using it. %NULL is returned if the dentry does not exist.
1347 */
1348 struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1349 {
1350 struct dentry * dentry = NULL;
1351 unsigned long seq;
1352
1353 do {
1354 seq = read_seqbegin(&rename_lock);
1355 dentry = __d_lookup(parent, name);
1356 if (dentry)
1357 break;
1358 } while (read_seqretry(&rename_lock, seq));
1359 return dentry;
1360 }
1361 EXPORT_SYMBOL(d_lookup);
1362
1363 /*
1364 * __d_lookup - search for a dentry (racy)
1365 * @parent: parent dentry
1366 * @name: qstr of name we wish to find
1367 * Returns: dentry, or NULL
1368 *
1369 * __d_lookup is like d_lookup, however it may (rarely) return a
1370 * false-negative result due to unrelated rename activity.
1371 *
1372 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1373 * however it must be used carefully, eg. with a following d_lookup in
1374 * the case of failure.
1375 *
1376 * __d_lookup callers must be commented.
1377 */
1378 struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1379 {
1380 unsigned int len = name->len;
1381 unsigned int hash = name->hash;
1382 const unsigned char *str = name->name;
1383 struct hlist_head *head = d_hash(parent,hash);
1384 struct dentry *found = NULL;
1385 struct hlist_node *node;
1386 struct dentry *dentry;
1387
1388 /*
1389 * The hash list is protected using RCU.
1390 *
1391 * Take d_lock when comparing a candidate dentry, to avoid races
1392 * with d_move().
1393 *
1394 * It is possible that concurrent renames can mess up our list
1395 * walk here and result in missing our dentry, resulting in the
1396 * false-negative result. d_lookup() protects against concurrent
1397 * renames using rename_lock seqlock.
1398 *
1399 * See Documentation/vfs/dcache-locking.txt for more details.
1400 */
1401 rcu_read_lock();
1402
1403 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1404 struct qstr *qstr;
1405
1406 if (dentry->d_name.hash != hash)
1407 continue;
1408 if (dentry->d_parent != parent)
1409 continue;
1410
1411 spin_lock(&dentry->d_lock);
1412
1413 /*
1414 * Recheck the dentry after taking the lock - d_move may have
1415 * changed things. Don't bother checking the hash because
1416 * we're about to compare the whole name anyway.
1417 */
1418 if (dentry->d_parent != parent)
1419 goto next;
1420
1421 /* non-existing due to RCU? */
1422 if (d_unhashed(dentry))
1423 goto next;
1424
1425 /*
1426 * It is safe to compare names since d_move() cannot
1427 * change the qstr (protected by d_lock).
1428 */
1429 qstr = &dentry->d_name;
1430 if (parent->d_op && parent->d_op->d_compare) {
1431 if (parent->d_op->d_compare(parent, qstr, name))
1432 goto next;
1433 } else {
1434 if (qstr->len != len)
1435 goto next;
1436 if (memcmp(qstr->name, str, len))
1437 goto next;
1438 }
1439
1440 atomic_inc(&dentry->d_count);
1441 found = dentry;
1442 spin_unlock(&dentry->d_lock);
1443 break;
1444 next:
1445 spin_unlock(&dentry->d_lock);
1446 }
1447 rcu_read_unlock();
1448
1449 return found;
1450 }
1451
1452 /**
1453 * d_hash_and_lookup - hash the qstr then search for a dentry
1454 * @dir: Directory to search in
1455 * @name: qstr of name we wish to find
1456 *
1457 * On hash failure or on lookup failure NULL is returned.
1458 */
1459 struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1460 {
1461 struct dentry *dentry = NULL;
1462
1463 /*
1464 * Check for a fs-specific hash function. Note that we must
1465 * calculate the standard hash first, as the d_op->d_hash()
1466 * routine may choose to leave the hash value unchanged.
1467 */
1468 name->hash = full_name_hash(name->name, name->len);
1469 if (dir->d_op && dir->d_op->d_hash) {
1470 if (dir->d_op->d_hash(dir, name) < 0)
1471 goto out;
1472 }
1473 dentry = d_lookup(dir, name);
1474 out:
1475 return dentry;
1476 }
1477
1478 /**
1479 * d_validate - verify dentry provided from insecure source
1480 * @dentry: The dentry alleged to be valid child of @dparent
1481 * @dparent: The parent dentry (known to be valid)
1482 *
1483 * An insecure source has sent us a dentry, here we verify it and dget() it.
1484 * This is used by ncpfs in its readdir implementation.
1485 * Zero is returned in the dentry is invalid.
1486 */
1487
1488 int d_validate(struct dentry *dentry, struct dentry *dparent)
1489 {
1490 struct hlist_head *base;
1491 struct hlist_node *lhp;
1492
1493 /* Check whether the ptr might be valid at all.. */
1494 if (!kmem_ptr_validate(dentry_cache, dentry))
1495 goto out;
1496
1497 if (dentry->d_parent != dparent)
1498 goto out;
1499
1500 spin_lock(&dcache_lock);
1501 base = d_hash(dparent, dentry->d_name.hash);
1502 hlist_for_each(lhp,base) {
1503 /* hlist_for_each_entry_rcu() not required for d_hash list
1504 * as it is parsed under dcache_lock
1505 */
1506 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1507 __dget_locked(dentry);
1508 spin_unlock(&dcache_lock);
1509 return 1;
1510 }
1511 }
1512 spin_unlock(&dcache_lock);
1513 out:
1514 return 0;
1515 }
1516 EXPORT_SYMBOL(d_validate);
1517
1518 /*
1519 * When a file is deleted, we have two options:
1520 * - turn this dentry into a negative dentry
1521 * - unhash this dentry and free it.
1522 *
1523 * Usually, we want to just turn this into
1524 * a negative dentry, but if anybody else is
1525 * currently using the dentry or the inode
1526 * we can't do that and we fall back on removing
1527 * it from the hash queues and waiting for
1528 * it to be deleted later when it has no users
1529 */
1530
1531 /**
1532 * d_delete - delete a dentry
1533 * @dentry: The dentry to delete
1534 *
1535 * Turn the dentry into a negative dentry if possible, otherwise
1536 * remove it from the hash queues so it can be deleted later
1537 */
1538
1539 void d_delete(struct dentry * dentry)
1540 {
1541 int isdir = 0;
1542 /*
1543 * Are we the only user?
1544 */
1545 spin_lock(&dcache_lock);
1546 spin_lock(&dentry->d_lock);
1547 isdir = S_ISDIR(dentry->d_inode->i_mode);
1548 if (atomic_read(&dentry->d_count) == 1) {
1549 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1550 dentry_iput(dentry);
1551 fsnotify_nameremove(dentry, isdir);
1552 return;
1553 }
1554
1555 if (!d_unhashed(dentry))
1556 __d_drop(dentry);
1557
1558 spin_unlock(&dentry->d_lock);
1559 spin_unlock(&dcache_lock);
1560
1561 fsnotify_nameremove(dentry, isdir);
1562 }
1563 EXPORT_SYMBOL(d_delete);
1564
1565 static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1566 {
1567
1568 entry->d_flags &= ~DCACHE_UNHASHED;
1569 hlist_add_head_rcu(&entry->d_hash, list);
1570 }
1571
1572 static void _d_rehash(struct dentry * entry)
1573 {
1574 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1575 }
1576
1577 /**
1578 * d_rehash - add an entry back to the hash
1579 * @entry: dentry to add to the hash
1580 *
1581 * Adds a dentry to the hash according to its name.
1582 */
1583
1584 void d_rehash(struct dentry * entry)
1585 {
1586 spin_lock(&dcache_lock);
1587 spin_lock(&entry->d_lock);
1588 _d_rehash(entry);
1589 spin_unlock(&entry->d_lock);
1590 spin_unlock(&dcache_lock);
1591 }
1592 EXPORT_SYMBOL(d_rehash);
1593
1594 /*
1595 * When switching names, the actual string doesn't strictly have to
1596 * be preserved in the target - because we're dropping the target
1597 * anyway. As such, we can just do a simple memcpy() to copy over
1598 * the new name before we switch.
1599 *
1600 * Note that we have to be a lot more careful about getting the hash
1601 * switched - we have to switch the hash value properly even if it
1602 * then no longer matches the actual (corrupted) string of the target.
1603 * The hash value has to match the hash queue that the dentry is on..
1604 */
1605 static void switch_names(struct dentry *dentry, struct dentry *target)
1606 {
1607 if (dname_external(target)) {
1608 if (dname_external(dentry)) {
1609 /*
1610 * Both external: swap the pointers
1611 */
1612 swap(target->d_name.name, dentry->d_name.name);
1613 } else {
1614 /*
1615 * dentry:internal, target:external. Steal target's
1616 * storage and make target internal.
1617 */
1618 memcpy(target->d_iname, dentry->d_name.name,
1619 dentry->d_name.len + 1);
1620 dentry->d_name.name = target->d_name.name;
1621 target->d_name.name = target->d_iname;
1622 }
1623 } else {
1624 if (dname_external(dentry)) {
1625 /*
1626 * dentry:external, target:internal. Give dentry's
1627 * storage to target and make dentry internal
1628 */
1629 memcpy(dentry->d_iname, target->d_name.name,
1630 target->d_name.len + 1);
1631 target->d_name.name = dentry->d_name.name;
1632 dentry->d_name.name = dentry->d_iname;
1633 } else {
1634 /*
1635 * Both are internal. Just copy target to dentry
1636 */
1637 memcpy(dentry->d_iname, target->d_name.name,
1638 target->d_name.len + 1);
1639 dentry->d_name.len = target->d_name.len;
1640 return;
1641 }
1642 }
1643 swap(dentry->d_name.len, target->d_name.len);
1644 }
1645
1646 /*
1647 * We cannibalize "target" when moving dentry on top of it,
1648 * because it's going to be thrown away anyway. We could be more
1649 * polite about it, though.
1650 *
1651 * This forceful removal will result in ugly /proc output if
1652 * somebody holds a file open that got deleted due to a rename.
1653 * We could be nicer about the deleted file, and let it show
1654 * up under the name it had before it was deleted rather than
1655 * under the original name of the file that was moved on top of it.
1656 */
1657
1658 /*
1659 * d_move_locked - move a dentry
1660 * @dentry: entry to move
1661 * @target: new dentry
1662 *
1663 * Update the dcache to reflect the move of a file name. Negative
1664 * dcache entries should not be moved in this way.
1665 */
1666 static void d_move_locked(struct dentry * dentry, struct dentry * target)
1667 {
1668 struct hlist_head *list;
1669
1670 if (!dentry->d_inode)
1671 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1672
1673 write_seqlock(&rename_lock);
1674 /*
1675 * XXXX: do we really need to take target->d_lock?
1676 */
1677 if (target < dentry) {
1678 spin_lock(&target->d_lock);
1679 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1680 } else {
1681 spin_lock(&dentry->d_lock);
1682 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
1683 }
1684
1685 /* Move the dentry to the target hash queue, if on different bucket */
1686 if (d_unhashed(dentry))
1687 goto already_unhashed;
1688
1689 hlist_del_rcu(&dentry->d_hash);
1690
1691 already_unhashed:
1692 list = d_hash(target->d_parent, target->d_name.hash);
1693 __d_rehash(dentry, list);
1694
1695 /* Unhash the target: dput() will then get rid of it */
1696 __d_drop(target);
1697
1698 list_del(&dentry->d_u.d_child);
1699 list_del(&target->d_u.d_child);
1700
1701 /* Switch the names.. */
1702 switch_names(dentry, target);
1703 swap(dentry->d_name.hash, target->d_name.hash);
1704
1705 /* ... and switch the parents */
1706 if (IS_ROOT(dentry)) {
1707 dentry->d_parent = target->d_parent;
1708 target->d_parent = target;
1709 INIT_LIST_HEAD(&target->d_u.d_child);
1710 } else {
1711 swap(dentry->d_parent, target->d_parent);
1712
1713 /* And add them back to the (new) parent lists */
1714 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1715 }
1716
1717 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1718 spin_unlock(&target->d_lock);
1719 fsnotify_d_move(dentry);
1720 spin_unlock(&dentry->d_lock);
1721 write_sequnlock(&rename_lock);
1722 }
1723
1724 /**
1725 * d_move - move a dentry
1726 * @dentry: entry to move
1727 * @target: new dentry
1728 *
1729 * Update the dcache to reflect the move of a file name. Negative
1730 * dcache entries should not be moved in this way.
1731 */
1732
1733 void d_move(struct dentry * dentry, struct dentry * target)
1734 {
1735 spin_lock(&dcache_lock);
1736 d_move_locked(dentry, target);
1737 spin_unlock(&dcache_lock);
1738 }
1739 EXPORT_SYMBOL(d_move);
1740
1741 /**
1742 * d_ancestor - search for an ancestor
1743 * @p1: ancestor dentry
1744 * @p2: child dentry
1745 *
1746 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1747 * an ancestor of p2, else NULL.
1748 */
1749 struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
1750 {
1751 struct dentry *p;
1752
1753 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
1754 if (p->d_parent == p1)
1755 return p;
1756 }
1757 return NULL;
1758 }
1759
1760 /*
1761 * This helper attempts to cope with remotely renamed directories
1762 *
1763 * It assumes that the caller is already holding
1764 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1765 *
1766 * Note: If ever the locking in lock_rename() changes, then please
1767 * remember to update this too...
1768 */
1769 static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
1770 __releases(dcache_lock)
1771 {
1772 struct mutex *m1 = NULL, *m2 = NULL;
1773 struct dentry *ret;
1774
1775 /* If alias and dentry share a parent, then no extra locks required */
1776 if (alias->d_parent == dentry->d_parent)
1777 goto out_unalias;
1778
1779 /* Check for loops */
1780 ret = ERR_PTR(-ELOOP);
1781 if (d_ancestor(alias, dentry))
1782 goto out_err;
1783
1784 /* See lock_rename() */
1785 ret = ERR_PTR(-EBUSY);
1786 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1787 goto out_err;
1788 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1789 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1790 goto out_err;
1791 m2 = &alias->d_parent->d_inode->i_mutex;
1792 out_unalias:
1793 d_move_locked(alias, dentry);
1794 ret = alias;
1795 out_err:
1796 spin_unlock(&dcache_lock);
1797 if (m2)
1798 mutex_unlock(m2);
1799 if (m1)
1800 mutex_unlock(m1);
1801 return ret;
1802 }
1803
1804 /*
1805 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1806 * named dentry in place of the dentry to be replaced.
1807 */
1808 static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1809 {
1810 struct dentry *dparent, *aparent;
1811
1812 switch_names(dentry, anon);
1813 swap(dentry->d_name.hash, anon->d_name.hash);
1814
1815 dparent = dentry->d_parent;
1816 aparent = anon->d_parent;
1817
1818 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1819 list_del(&dentry->d_u.d_child);
1820 if (!IS_ROOT(dentry))
1821 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1822 else
1823 INIT_LIST_HEAD(&dentry->d_u.d_child);
1824
1825 anon->d_parent = (dparent == dentry) ? anon : dparent;
1826 list_del(&anon->d_u.d_child);
1827 if (!IS_ROOT(anon))
1828 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1829 else
1830 INIT_LIST_HEAD(&anon->d_u.d_child);
1831
1832 anon->d_flags &= ~DCACHE_DISCONNECTED;
1833 }
1834
1835 /**
1836 * d_materialise_unique - introduce an inode into the tree
1837 * @dentry: candidate dentry
1838 * @inode: inode to bind to the dentry, to which aliases may be attached
1839 *
1840 * Introduces an dentry into the tree, substituting an extant disconnected
1841 * root directory alias in its place if there is one
1842 */
1843 struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1844 {
1845 struct dentry *actual;
1846
1847 BUG_ON(!d_unhashed(dentry));
1848
1849 spin_lock(&dcache_lock);
1850
1851 if (!inode) {
1852 actual = dentry;
1853 __d_instantiate(dentry, NULL);
1854 goto found_lock;
1855 }
1856
1857 if (S_ISDIR(inode->i_mode)) {
1858 struct dentry *alias;
1859
1860 /* Does an aliased dentry already exist? */
1861 alias = __d_find_alias(inode, 0);
1862 if (alias) {
1863 actual = alias;
1864 /* Is this an anonymous mountpoint that we could splice
1865 * into our tree? */
1866 if (IS_ROOT(alias)) {
1867 spin_lock(&alias->d_lock);
1868 __d_materialise_dentry(dentry, alias);
1869 __d_drop(alias);
1870 goto found;
1871 }
1872 /* Nope, but we must(!) avoid directory aliasing */
1873 actual = __d_unalias(dentry, alias);
1874 if (IS_ERR(actual))
1875 dput(alias);
1876 goto out_nolock;
1877 }
1878 }
1879
1880 /* Add a unique reference */
1881 actual = __d_instantiate_unique(dentry, inode);
1882 if (!actual)
1883 actual = dentry;
1884 else if (unlikely(!d_unhashed(actual)))
1885 goto shouldnt_be_hashed;
1886
1887 found_lock:
1888 spin_lock(&actual->d_lock);
1889 found:
1890 _d_rehash(actual);
1891 spin_unlock(&actual->d_lock);
1892 spin_unlock(&dcache_lock);
1893 out_nolock:
1894 if (actual == dentry) {
1895 security_d_instantiate(dentry, inode);
1896 return NULL;
1897 }
1898
1899 iput(inode);
1900 return actual;
1901
1902 shouldnt_be_hashed:
1903 spin_unlock(&dcache_lock);
1904 BUG();
1905 }
1906 EXPORT_SYMBOL_GPL(d_materialise_unique);
1907
1908 static int prepend(char **buffer, int *buflen, const char *str, int namelen)
1909 {
1910 *buflen -= namelen;
1911 if (*buflen < 0)
1912 return -ENAMETOOLONG;
1913 *buffer -= namelen;
1914 memcpy(*buffer, str, namelen);
1915 return 0;
1916 }
1917
1918 static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1919 {
1920 return prepend(buffer, buflen, name->name, name->len);
1921 }
1922
1923 /**
1924 * Prepend path string to a buffer
1925 *
1926 * @path: the dentry/vfsmount to report
1927 * @root: root vfsmnt/dentry (may be modified by this function)
1928 * @buffer: pointer to the end of the buffer
1929 * @buflen: pointer to buffer length
1930 *
1931 * Caller holds the dcache_lock.
1932 *
1933 * If path is not reachable from the supplied root, then the value of
1934 * root is changed (without modifying refcounts).
1935 */
1936 static int prepend_path(const struct path *path, struct path *root,
1937 char **buffer, int *buflen)
1938 {
1939 struct dentry *dentry = path->dentry;
1940 struct vfsmount *vfsmnt = path->mnt;
1941 bool slash = false;
1942 int error = 0;
1943
1944 br_read_lock(vfsmount_lock);
1945 while (dentry != root->dentry || vfsmnt != root->mnt) {
1946 struct dentry * parent;
1947
1948 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
1949 /* Global root? */
1950 if (vfsmnt->mnt_parent == vfsmnt) {
1951 goto global_root;
1952 }
1953 dentry = vfsmnt->mnt_mountpoint;
1954 vfsmnt = vfsmnt->mnt_parent;
1955 continue;
1956 }
1957 parent = dentry->d_parent;
1958 prefetch(parent);
1959 error = prepend_name(buffer, buflen, &dentry->d_name);
1960 if (!error)
1961 error = prepend(buffer, buflen, "/", 1);
1962 if (error)
1963 break;
1964
1965 slash = true;
1966 dentry = parent;
1967 }
1968
1969 out:
1970 if (!error && !slash)
1971 error = prepend(buffer, buflen, "/", 1);
1972
1973 br_read_unlock(vfsmount_lock);
1974 return error;
1975
1976 global_root:
1977 /*
1978 * Filesystems needing to implement special "root names"
1979 * should do so with ->d_dname()
1980 */
1981 if (IS_ROOT(dentry) &&
1982 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
1983 WARN(1, "Root dentry has weird name <%.*s>\n",
1984 (int) dentry->d_name.len, dentry->d_name.name);
1985 }
1986 root->mnt = vfsmnt;
1987 root->dentry = dentry;
1988 goto out;
1989 }
1990
1991 /**
1992 * __d_path - return the path of a dentry
1993 * @path: the dentry/vfsmount to report
1994 * @root: root vfsmnt/dentry (may be modified by this function)
1995 * @buf: buffer to return value in
1996 * @buflen: buffer length
1997 *
1998 * Convert a dentry into an ASCII path name.
1999 *
2000 * Returns a pointer into the buffer or an error code if the
2001 * path was too long.
2002 *
2003 * "buflen" should be positive.
2004 *
2005 * If path is not reachable from the supplied root, then the value of
2006 * root is changed (without modifying refcounts).
2007 */
2008 char *__d_path(const struct path *path, struct path *root,
2009 char *buf, int buflen)
2010 {
2011 char *res = buf + buflen;
2012 int error;
2013
2014 prepend(&res, &buflen, "\0", 1);
2015 spin_lock(&dcache_lock);
2016 error = prepend_path(path, root, &res, &buflen);
2017 spin_unlock(&dcache_lock);
2018
2019 if (error)
2020 return ERR_PTR(error);
2021 return res;
2022 }
2023
2024 /*
2025 * same as __d_path but appends "(deleted)" for unlinked files.
2026 */
2027 static int path_with_deleted(const struct path *path, struct path *root,
2028 char **buf, int *buflen)
2029 {
2030 prepend(buf, buflen, "\0", 1);
2031 if (d_unlinked(path->dentry)) {
2032 int error = prepend(buf, buflen, " (deleted)", 10);
2033 if (error)
2034 return error;
2035 }
2036
2037 return prepend_path(path, root, buf, buflen);
2038 }
2039
2040 static int prepend_unreachable(char **buffer, int *buflen)
2041 {
2042 return prepend(buffer, buflen, "(unreachable)", 13);
2043 }
2044
2045 /**
2046 * d_path - return the path of a dentry
2047 * @path: path to report
2048 * @buf: buffer to return value in
2049 * @buflen: buffer length
2050 *
2051 * Convert a dentry into an ASCII path name. If the entry has been deleted
2052 * the string " (deleted)" is appended. Note that this is ambiguous.
2053 *
2054 * Returns a pointer into the buffer or an error code if the path was
2055 * too long. Note: Callers should use the returned pointer, not the passed
2056 * in buffer, to use the name! The implementation often starts at an offset
2057 * into the buffer, and may leave 0 bytes at the start.
2058 *
2059 * "buflen" should be positive.
2060 */
2061 char *d_path(const struct path *path, char *buf, int buflen)
2062 {
2063 char *res = buf + buflen;
2064 struct path root;
2065 struct path tmp;
2066 int error;
2067
2068 /*
2069 * We have various synthetic filesystems that never get mounted. On
2070 * these filesystems dentries are never used for lookup purposes, and
2071 * thus don't need to be hashed. They also don't need a name until a
2072 * user wants to identify the object in /proc/pid/fd/. The little hack
2073 * below allows us to generate a name for these objects on demand:
2074 */
2075 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2076 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2077
2078 get_fs_root(current->fs, &root);
2079 spin_lock(&dcache_lock);
2080 tmp = root;
2081 error = path_with_deleted(path, &tmp, &res, &buflen);
2082 if (error)
2083 res = ERR_PTR(error);
2084 spin_unlock(&dcache_lock);
2085 path_put(&root);
2086 return res;
2087 }
2088 EXPORT_SYMBOL(d_path);
2089
2090 /**
2091 * d_path_with_unreachable - return the path of a dentry
2092 * @path: path to report
2093 * @buf: buffer to return value in
2094 * @buflen: buffer length
2095 *
2096 * The difference from d_path() is that this prepends "(unreachable)"
2097 * to paths which are unreachable from the current process' root.
2098 */
2099 char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2100 {
2101 char *res = buf + buflen;
2102 struct path root;
2103 struct path tmp;
2104 int error;
2105
2106 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2107 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2108
2109 get_fs_root(current->fs, &root);
2110 spin_lock(&dcache_lock);
2111 tmp = root;
2112 error = path_with_deleted(path, &tmp, &res, &buflen);
2113 if (!error && !path_equal(&tmp, &root))
2114 error = prepend_unreachable(&res, &buflen);
2115 spin_unlock(&dcache_lock);
2116 path_put(&root);
2117 if (error)
2118 res = ERR_PTR(error);
2119
2120 return res;
2121 }
2122
2123 /*
2124 * Helper function for dentry_operations.d_dname() members
2125 */
2126 char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2127 const char *fmt, ...)
2128 {
2129 va_list args;
2130 char temp[64];
2131 int sz;
2132
2133 va_start(args, fmt);
2134 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2135 va_end(args);
2136
2137 if (sz > sizeof(temp) || sz > buflen)
2138 return ERR_PTR(-ENAMETOOLONG);
2139
2140 buffer += buflen - sz;
2141 return memcpy(buffer, temp, sz);
2142 }
2143
2144 /*
2145 * Write full pathname from the root of the filesystem into the buffer.
2146 */
2147 char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
2148 {
2149 char *end = buf + buflen;
2150 char *retval;
2151
2152 prepend(&end, &buflen, "\0", 1);
2153 if (buflen < 1)
2154 goto Elong;
2155 /* Get '/' right */
2156 retval = end-1;
2157 *retval = '/';
2158
2159 while (!IS_ROOT(dentry)) {
2160 struct dentry *parent = dentry->d_parent;
2161
2162 prefetch(parent);
2163 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
2164 (prepend(&end, &buflen, "/", 1) != 0))
2165 goto Elong;
2166
2167 retval = end;
2168 dentry = parent;
2169 }
2170 return retval;
2171 Elong:
2172 return ERR_PTR(-ENAMETOOLONG);
2173 }
2174 EXPORT_SYMBOL(__dentry_path);
2175
2176 char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2177 {
2178 char *p = NULL;
2179 char *retval;
2180
2181 spin_lock(&dcache_lock);
2182 if (d_unlinked(dentry)) {
2183 p = buf + buflen;
2184 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2185 goto Elong;
2186 buflen++;
2187 }
2188 retval = __dentry_path(dentry, buf, buflen);
2189 spin_unlock(&dcache_lock);
2190 if (!IS_ERR(retval) && p)
2191 *p = '/'; /* restore '/' overriden with '\0' */
2192 return retval;
2193 Elong:
2194 spin_unlock(&dcache_lock);
2195 return ERR_PTR(-ENAMETOOLONG);
2196 }
2197
2198 /*
2199 * NOTE! The user-level library version returns a
2200 * character pointer. The kernel system call just
2201 * returns the length of the buffer filled (which
2202 * includes the ending '\0' character), or a negative
2203 * error value. So libc would do something like
2204 *
2205 * char *getcwd(char * buf, size_t size)
2206 * {
2207 * int retval;
2208 *
2209 * retval = sys_getcwd(buf, size);
2210 * if (retval >= 0)
2211 * return buf;
2212 * errno = -retval;
2213 * return NULL;
2214 * }
2215 */
2216 SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
2217 {
2218 int error;
2219 struct path pwd, root;
2220 char *page = (char *) __get_free_page(GFP_USER);
2221
2222 if (!page)
2223 return -ENOMEM;
2224
2225 get_fs_root_and_pwd(current->fs, &root, &pwd);
2226
2227 error = -ENOENT;
2228 spin_lock(&dcache_lock);
2229 if (!d_unlinked(pwd.dentry)) {
2230 unsigned long len;
2231 struct path tmp = root;
2232 char *cwd = page + PAGE_SIZE;
2233 int buflen = PAGE_SIZE;
2234
2235 prepend(&cwd, &buflen, "\0", 1);
2236 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
2237 spin_unlock(&dcache_lock);
2238
2239 if (error)
2240 goto out;
2241
2242 /* Unreachable from current root */
2243 if (!path_equal(&tmp, &root)) {
2244 error = prepend_unreachable(&cwd, &buflen);
2245 if (error)
2246 goto out;
2247 }
2248
2249 error = -ERANGE;
2250 len = PAGE_SIZE + page - cwd;
2251 if (len <= size) {
2252 error = len;
2253 if (copy_to_user(buf, cwd, len))
2254 error = -EFAULT;
2255 }
2256 } else
2257 spin_unlock(&dcache_lock);
2258
2259 out:
2260 path_put(&pwd);
2261 path_put(&root);
2262 free_page((unsigned long) page);
2263 return error;
2264 }
2265
2266 /*
2267 * Test whether new_dentry is a subdirectory of old_dentry.
2268 *
2269 * Trivially implemented using the dcache structure
2270 */
2271
2272 /**
2273 * is_subdir - is new dentry a subdirectory of old_dentry
2274 * @new_dentry: new dentry
2275 * @old_dentry: old dentry
2276 *
2277 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2278 * Returns 0 otherwise.
2279 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2280 */
2281
2282 int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
2283 {
2284 int result;
2285 unsigned long seq;
2286
2287 if (new_dentry == old_dentry)
2288 return 1;
2289
2290 /*
2291 * Need rcu_readlock to protect against the d_parent trashing
2292 * due to d_move
2293 */
2294 rcu_read_lock();
2295 do {
2296 /* for restarting inner loop in case of seq retry */
2297 seq = read_seqbegin(&rename_lock);
2298 if (d_ancestor(old_dentry, new_dentry))
2299 result = 1;
2300 else
2301 result = 0;
2302 } while (read_seqretry(&rename_lock, seq));
2303 rcu_read_unlock();
2304
2305 return result;
2306 }
2307
2308 int path_is_under(struct path *path1, struct path *path2)
2309 {
2310 struct vfsmount *mnt = path1->mnt;
2311 struct dentry *dentry = path1->dentry;
2312 int res;
2313
2314 br_read_lock(vfsmount_lock);
2315 if (mnt != path2->mnt) {
2316 for (;;) {
2317 if (mnt->mnt_parent == mnt) {
2318 br_read_unlock(vfsmount_lock);
2319 return 0;
2320 }
2321 if (mnt->mnt_parent == path2->mnt)
2322 break;
2323 mnt = mnt->mnt_parent;
2324 }
2325 dentry = mnt->mnt_mountpoint;
2326 }
2327 res = is_subdir(dentry, path2->dentry);
2328 br_read_unlock(vfsmount_lock);
2329 return res;
2330 }
2331 EXPORT_SYMBOL(path_is_under);
2332
2333 void d_genocide(struct dentry *root)
2334 {
2335 struct dentry *this_parent = root;
2336 struct list_head *next;
2337
2338 spin_lock(&dcache_lock);
2339 repeat:
2340 next = this_parent->d_subdirs.next;
2341 resume:
2342 while (next != &this_parent->d_subdirs) {
2343 struct list_head *tmp = next;
2344 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
2345 next = tmp->next;
2346 if (d_unhashed(dentry)||!dentry->d_inode)
2347 continue;
2348 if (!list_empty(&dentry->d_subdirs)) {
2349 this_parent = dentry;
2350 goto repeat;
2351 }
2352 atomic_dec(&dentry->d_count);
2353 }
2354 if (this_parent != root) {
2355 next = this_parent->d_u.d_child.next;
2356 atomic_dec(&this_parent->d_count);
2357 this_parent = this_parent->d_parent;
2358 goto resume;
2359 }
2360 spin_unlock(&dcache_lock);
2361 }
2362
2363 /**
2364 * find_inode_number - check for dentry with name
2365 * @dir: directory to check
2366 * @name: Name to find.
2367 *
2368 * Check whether a dentry already exists for the given name,
2369 * and return the inode number if it has an inode. Otherwise
2370 * 0 is returned.
2371 *
2372 * This routine is used to post-process directory listings for
2373 * filesystems using synthetic inode numbers, and is necessary
2374 * to keep getcwd() working.
2375 */
2376
2377 ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2378 {
2379 struct dentry * dentry;
2380 ino_t ino = 0;
2381
2382 dentry = d_hash_and_lookup(dir, name);
2383 if (dentry) {
2384 if (dentry->d_inode)
2385 ino = dentry->d_inode->i_ino;
2386 dput(dentry);
2387 }
2388 return ino;
2389 }
2390 EXPORT_SYMBOL(find_inode_number);
2391
2392 static __initdata unsigned long dhash_entries;
2393 static int __init set_dhash_entries(char *str)
2394 {
2395 if (!str)
2396 return 0;
2397 dhash_entries = simple_strtoul(str, &str, 0);
2398 return 1;
2399 }
2400 __setup("dhash_entries=", set_dhash_entries);
2401
2402 static void __init dcache_init_early(void)
2403 {
2404 int loop;
2405
2406 /* If hashes are distributed across NUMA nodes, defer
2407 * hash allocation until vmalloc space is available.
2408 */
2409 if (hashdist)
2410 return;
2411
2412 dentry_hashtable =
2413 alloc_large_system_hash("Dentry cache",
2414 sizeof(struct hlist_head),
2415 dhash_entries,
2416 13,
2417 HASH_EARLY,
2418 &d_hash_shift,
2419 &d_hash_mask,
2420 0);
2421
2422 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2423 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2424 }
2425
2426 static void __init dcache_init(void)
2427 {
2428 int loop;
2429
2430 percpu_counter_init(&nr_dentry, 0);
2431 percpu_counter_init(&nr_dentry_unused, 0);
2432
2433 /*
2434 * A constructor could be added for stable state like the lists,
2435 * but it is probably not worth it because of the cache nature
2436 * of the dcache.
2437 */
2438 dentry_cache = KMEM_CACHE(dentry,
2439 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
2440
2441 register_shrinker(&dcache_shrinker);
2442
2443 /* Hash may have been set up in dcache_init_early */
2444 if (!hashdist)
2445 return;
2446
2447 dentry_hashtable =
2448 alloc_large_system_hash("Dentry cache",
2449 sizeof(struct hlist_head),
2450 dhash_entries,
2451 13,
2452 0,
2453 &d_hash_shift,
2454 &d_hash_mask,
2455 0);
2456
2457 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2458 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2459 }
2460
2461 /* SLAB cache for __getname() consumers */
2462 struct kmem_cache *names_cachep __read_mostly;
2463 EXPORT_SYMBOL(names_cachep);
2464
2465 EXPORT_SYMBOL(d_genocide);
2466
2467 void __init vfs_caches_init_early(void)
2468 {
2469 dcache_init_early();
2470 inode_init_early();
2471 }
2472
2473 void __init vfs_caches_init(unsigned long mempages)
2474 {
2475 unsigned long reserve;
2476
2477 /* Base hash sizes on available memory, with a reserve equal to
2478 150% of current kernel size */
2479
2480 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2481 mempages -= reserve;
2482
2483 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
2484 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
2485
2486 dcache_init();
2487 inode_init();
2488 files_init(mempages);
2489 mnt_init();
2490 bdev_cache_init();
2491 chrdev_init();
2492 }