]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/dcache.c
BKL: Remove BKL from HFS
[mirror_ubuntu-artful-kernel.git] / fs / dcache.c
CommitLineData
1da177e4
LT
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
1da177e4
LT
17#include <linux/syscalls.h>
18#include <linux/string.h>
19#include <linux/mm.h>
20#include <linux/fs.h>
7a91bf7f 21#include <linux/fsnotify.h>
1da177e4
LT
22#include <linux/slab.h>
23#include <linux/init.h>
1da177e4
LT
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>
5ad4e53b 34#include <linux/fs_struct.h>
613afbf8 35#include <linux/hardirq.h>
07f3f05c 36#include "internal.h"
1da177e4 37
fa3536cc 38int sysctl_vfs_cache_pressure __read_mostly = 100;
1da177e4
LT
39EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure);
40
41 __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock);
74c3cbe3 42__cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock);
1da177e4
LT
43
44EXPORT_SYMBOL(dcache_lock);
45
e18b890b 46static struct kmem_cache *dentry_cache __read_mostly;
1da177e4
LT
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
fa3536cc
ED
61static unsigned int d_hash_mask __read_mostly;
62static unsigned int d_hash_shift __read_mostly;
63static struct hlist_head *dentry_hashtable __read_mostly;
1da177e4
LT
64
65/* Statistics gathering. */
66struct dentry_stat_t dentry_stat = {
67 .age_limit = 45,
68};
69
b3423415 70static void __d_free(struct dentry *dentry)
1da177e4 71{
fd217f4d 72 WARN_ON(!list_empty(&dentry->d_alias));
1da177e4
LT
73 if (dname_external(dentry))
74 kfree(dentry->d_name.name);
75 kmem_cache_free(dentry_cache, dentry);
76}
77
b3423415
ED
78static void d_callback(struct rcu_head *head)
79{
80 struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu);
81 __d_free(dentry);
82}
83
1da177e4
LT
84/*
85 * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry
86 * inside dcache_lock.
87 */
88static void d_free(struct dentry *dentry)
89{
90 if (dentry->d_op && dentry->d_op->d_release)
91 dentry->d_op->d_release(dentry);
b3423415 92 /* if dentry was never inserted into hash, immediate free is OK */
e8462caa 93 if (hlist_unhashed(&dentry->d_hash))
b3423415
ED
94 __d_free(dentry);
95 else
96 call_rcu(&dentry->d_u.d_rcu, d_callback);
1da177e4
LT
97}
98
99/*
100 * Release the dentry's inode, using the filesystem
101 * d_iput() operation if defined.
1da177e4 102 */
858119e1 103static void dentry_iput(struct dentry * dentry)
31f3e0b3
MS
104 __releases(dentry->d_lock)
105 __releases(dcache_lock)
1da177e4
LT
106{
107 struct inode *inode = dentry->d_inode;
108 if (inode) {
109 dentry->d_inode = NULL;
110 list_del_init(&dentry->d_alias);
111 spin_unlock(&dentry->d_lock);
112 spin_unlock(&dcache_lock);
f805fbda
LT
113 if (!inode->i_nlink)
114 fsnotify_inoderemove(inode);
1da177e4
LT
115 if (dentry->d_op && dentry->d_op->d_iput)
116 dentry->d_op->d_iput(dentry, inode);
117 else
118 iput(inode);
119 } else {
120 spin_unlock(&dentry->d_lock);
121 spin_unlock(&dcache_lock);
122 }
123}
124
da3bbdd4
KM
125/*
126 * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held.
127 */
128static void dentry_lru_add(struct dentry *dentry)
129{
130 list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
131 dentry->d_sb->s_nr_dentry_unused++;
132 dentry_stat.nr_unused++;
133}
134
135static void dentry_lru_add_tail(struct dentry *dentry)
136{
137 list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru);
138 dentry->d_sb->s_nr_dentry_unused++;
139 dentry_stat.nr_unused++;
140}
141
142static void dentry_lru_del(struct dentry *dentry)
143{
144 if (!list_empty(&dentry->d_lru)) {
145 list_del(&dentry->d_lru);
146 dentry->d_sb->s_nr_dentry_unused--;
147 dentry_stat.nr_unused--;
148 }
149}
150
151static void dentry_lru_del_init(struct dentry *dentry)
152{
153 if (likely(!list_empty(&dentry->d_lru))) {
154 list_del_init(&dentry->d_lru);
155 dentry->d_sb->s_nr_dentry_unused--;
156 dentry_stat.nr_unused--;
157 }
158}
159
d52b9086
MS
160/**
161 * d_kill - kill dentry and return parent
162 * @dentry: dentry to kill
163 *
31f3e0b3 164 * The dentry must already be unhashed and removed from the LRU.
d52b9086
MS
165 *
166 * If this is the root of the dentry tree, return NULL.
167 */
168static struct dentry *d_kill(struct dentry *dentry)
31f3e0b3
MS
169 __releases(dentry->d_lock)
170 __releases(dcache_lock)
d52b9086
MS
171{
172 struct dentry *parent;
173
174 list_del(&dentry->d_u.d_child);
175 dentry_stat.nr_dentry--; /* For d_free, below */
176 /*drops the locks, at that point nobody can reach this dentry */
177 dentry_iput(dentry);
871c0067
OH
178 if (IS_ROOT(dentry))
179 parent = NULL;
180 else
181 parent = dentry->d_parent;
d52b9086 182 d_free(dentry);
871c0067 183 return parent;
d52b9086
MS
184}
185
1da177e4
LT
186/*
187 * This is dput
188 *
189 * This is complicated by the fact that we do not want to put
190 * dentries that are no longer on any hash chain on the unused
191 * list: we'd much rather just get rid of them immediately.
192 *
193 * However, that implies that we have to traverse the dentry
194 * tree upwards to the parents which might _also_ now be
195 * scheduled for deletion (it may have been only waiting for
196 * its last child to go away).
197 *
198 * This tail recursion is done by hand as we don't want to depend
199 * on the compiler to always get this right (gcc generally doesn't).
200 * Real recursion would eat up our stack space.
201 */
202
203/*
204 * dput - release a dentry
205 * @dentry: dentry to release
206 *
207 * Release a dentry. This will drop the usage count and if appropriate
208 * call the dentry unlink method as well as removing it from the queues and
209 * releasing its resources. If the parent dentries were scheduled for release
210 * they too may now get deleted.
211 *
212 * no dcache lock, please.
213 */
214
215void dput(struct dentry *dentry)
216{
217 if (!dentry)
218 return;
219
220repeat:
221 if (atomic_read(&dentry->d_count) == 1)
222 might_sleep();
223 if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock))
224 return;
225
226 spin_lock(&dentry->d_lock);
227 if (atomic_read(&dentry->d_count)) {
228 spin_unlock(&dentry->d_lock);
229 spin_unlock(&dcache_lock);
230 return;
231 }
232
233 /*
234 * AV: ->d_delete() is _NOT_ allowed to block now.
235 */
236 if (dentry->d_op && dentry->d_op->d_delete) {
237 if (dentry->d_op->d_delete(dentry))
238 goto unhash_it;
239 }
240 /* Unreachable? Get rid of it */
241 if (d_unhashed(dentry))
242 goto kill_it;
243 if (list_empty(&dentry->d_lru)) {
244 dentry->d_flags |= DCACHE_REFERENCED;
da3bbdd4 245 dentry_lru_add(dentry);
1da177e4
LT
246 }
247 spin_unlock(&dentry->d_lock);
248 spin_unlock(&dcache_lock);
249 return;
250
251unhash_it:
252 __d_drop(dentry);
d52b9086 253kill_it:
da3bbdd4
KM
254 /* if dentry was on the d_lru list delete it from there */
255 dentry_lru_del(dentry);
d52b9086
MS
256 dentry = d_kill(dentry);
257 if (dentry)
258 goto repeat;
1da177e4 259}
ec4f8605 260EXPORT_SYMBOL(dput);
1da177e4
LT
261
262/**
263 * d_invalidate - invalidate a dentry
264 * @dentry: dentry to invalidate
265 *
266 * Try to invalidate the dentry if it turns out to be
267 * possible. If there are other dentries that can be
268 * reached through this one we can't delete it and we
269 * return -EBUSY. On success we return 0.
270 *
271 * no dcache lock.
272 */
273
274int d_invalidate(struct dentry * dentry)
275{
276 /*
277 * If it's already been dropped, return OK.
278 */
279 spin_lock(&dcache_lock);
280 if (d_unhashed(dentry)) {
281 spin_unlock(&dcache_lock);
282 return 0;
283 }
284 /*
285 * Check whether to do a partial shrink_dcache
286 * to get rid of unused child entries.
287 */
288 if (!list_empty(&dentry->d_subdirs)) {
289 spin_unlock(&dcache_lock);
290 shrink_dcache_parent(dentry);
291 spin_lock(&dcache_lock);
292 }
293
294 /*
295 * Somebody else still using it?
296 *
297 * If it's a directory, we can't drop it
298 * for fear of somebody re-populating it
299 * with children (even though dropping it
300 * would make it unreachable from the root,
301 * we might still populate it if it was a
302 * working directory or similar).
303 */
304 spin_lock(&dentry->d_lock);
305 if (atomic_read(&dentry->d_count) > 1) {
306 if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) {
307 spin_unlock(&dentry->d_lock);
308 spin_unlock(&dcache_lock);
309 return -EBUSY;
310 }
311 }
312
313 __d_drop(dentry);
314 spin_unlock(&dentry->d_lock);
315 spin_unlock(&dcache_lock);
316 return 0;
317}
ec4f8605 318EXPORT_SYMBOL(d_invalidate);
1da177e4
LT
319
320/* This should be called _only_ with dcache_lock held */
321
322static inline struct dentry * __dget_locked(struct dentry *dentry)
323{
324 atomic_inc(&dentry->d_count);
da3bbdd4 325 dentry_lru_del_init(dentry);
1da177e4
LT
326 return dentry;
327}
328
329struct dentry * dget_locked(struct dentry *dentry)
330{
331 return __dget_locked(dentry);
332}
ec4f8605 333EXPORT_SYMBOL(dget_locked);
1da177e4
LT
334
335/**
336 * d_find_alias - grab a hashed alias of inode
337 * @inode: inode in question
338 * @want_discon: flag, used by d_splice_alias, to request
339 * that only a DISCONNECTED alias be returned.
340 *
341 * If inode has a hashed alias, or is a directory and has any alias,
342 * acquire the reference to alias and return it. Otherwise return NULL.
343 * Notice that if inode is a directory there can be only one alias and
344 * it can be unhashed only if it has no children, or if it is the root
345 * of a filesystem.
346 *
21c0d8fd 347 * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer
1da177e4 348 * any other hashed alias over that one unless @want_discon is set,
21c0d8fd 349 * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias.
1da177e4
LT
350 */
351
352static struct dentry * __d_find_alias(struct inode *inode, int want_discon)
353{
354 struct list_head *head, *next, *tmp;
355 struct dentry *alias, *discon_alias=NULL;
356
357 head = &inode->i_dentry;
358 next = inode->i_dentry.next;
359 while (next != head) {
360 tmp = next;
361 next = tmp->next;
362 prefetch(next);
363 alias = list_entry(tmp, struct dentry, d_alias);
364 if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) {
21c0d8fd
N
365 if (IS_ROOT(alias) &&
366 (alias->d_flags & DCACHE_DISCONNECTED))
1da177e4
LT
367 discon_alias = alias;
368 else if (!want_discon) {
369 __dget_locked(alias);
370 return alias;
371 }
372 }
373 }
374 if (discon_alias)
375 __dget_locked(discon_alias);
376 return discon_alias;
377}
378
379struct dentry * d_find_alias(struct inode *inode)
380{
214fda1f
DH
381 struct dentry *de = NULL;
382
383 if (!list_empty(&inode->i_dentry)) {
384 spin_lock(&dcache_lock);
385 de = __d_find_alias(inode, 0);
386 spin_unlock(&dcache_lock);
387 }
1da177e4
LT
388 return de;
389}
ec4f8605 390EXPORT_SYMBOL(d_find_alias);
1da177e4
LT
391
392/*
393 * Try to kill dentries associated with this inode.
394 * WARNING: you must own a reference to inode.
395 */
396void d_prune_aliases(struct inode *inode)
397{
0cdca3f9 398 struct dentry *dentry;
1da177e4
LT
399restart:
400 spin_lock(&dcache_lock);
0cdca3f9 401 list_for_each_entry(dentry, &inode->i_dentry, d_alias) {
1da177e4
LT
402 spin_lock(&dentry->d_lock);
403 if (!atomic_read(&dentry->d_count)) {
404 __dget_locked(dentry);
405 __d_drop(dentry);
406 spin_unlock(&dentry->d_lock);
407 spin_unlock(&dcache_lock);
408 dput(dentry);
409 goto restart;
410 }
411 spin_unlock(&dentry->d_lock);
412 }
413 spin_unlock(&dcache_lock);
414}
ec4f8605 415EXPORT_SYMBOL(d_prune_aliases);
1da177e4
LT
416
417/*
d702ccb3
AM
418 * Throw away a dentry - free the inode, dput the parent. This requires that
419 * the LRU list has already been removed.
420 *
85864e10
MS
421 * Try to prune ancestors as well. This is necessary to prevent
422 * quadratic behavior of shrink_dcache_parent(), but is also expected
423 * to be beneficial in reducing dentry cache fragmentation.
1da177e4 424 */
85864e10 425static void prune_one_dentry(struct dentry * dentry)
31f3e0b3
MS
426 __releases(dentry->d_lock)
427 __releases(dcache_lock)
428 __acquires(dcache_lock)
1da177e4 429{
1da177e4 430 __d_drop(dentry);
d52b9086 431 dentry = d_kill(dentry);
d52b9086
MS
432
433 /*
434 * Prune ancestors. Locking is simpler than in dput(),
435 * because dcache_lock needs to be taken anyway.
436 */
1da177e4 437 spin_lock(&dcache_lock);
d52b9086
MS
438 while (dentry) {
439 if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock))
440 return;
441
442 if (dentry->d_op && dentry->d_op->d_delete)
443 dentry->d_op->d_delete(dentry);
da3bbdd4 444 dentry_lru_del_init(dentry);
d52b9086
MS
445 __d_drop(dentry);
446 dentry = d_kill(dentry);
447 spin_lock(&dcache_lock);
448 }
1da177e4
LT
449}
450
da3bbdd4
KM
451/*
452 * Shrink the dentry LRU on a given superblock.
453 * @sb : superblock to shrink dentry LRU.
454 * @count: If count is NULL, we prune all dentries on superblock.
455 * @flags: If flags is non-zero, we need to do special processing based on
456 * which flags are set. This means we don't need to maintain multiple
457 * similar copies of this loop.
1da177e4 458 */
da3bbdd4 459static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags)
1da177e4 460{
da3bbdd4
KM
461 LIST_HEAD(referenced);
462 LIST_HEAD(tmp);
463 struct dentry *dentry;
464 int cnt = 0;
1da177e4 465
da3bbdd4
KM
466 BUG_ON(!sb);
467 BUG_ON((flags & DCACHE_REFERENCED) && count == NULL);
468 spin_lock(&dcache_lock);
469 if (count != NULL)
470 /* called from prune_dcache() and shrink_dcache_parent() */
471 cnt = *count;
472restart:
473 if (count == NULL)
474 list_splice_init(&sb->s_dentry_lru, &tmp);
475 else {
476 while (!list_empty(&sb->s_dentry_lru)) {
477 dentry = list_entry(sb->s_dentry_lru.prev,
478 struct dentry, d_lru);
479 BUG_ON(dentry->d_sb != sb);
480
481 spin_lock(&dentry->d_lock);
482 /*
483 * If we are honouring the DCACHE_REFERENCED flag and
484 * the dentry has this flag set, don't free it. Clear
485 * the flag and put it back on the LRU.
0feae5c4 486 */
da3bbdd4
KM
487 if ((flags & DCACHE_REFERENCED)
488 && (dentry->d_flags & DCACHE_REFERENCED)) {
489 dentry->d_flags &= ~DCACHE_REFERENCED;
c490d79b 490 list_move(&dentry->d_lru, &referenced);
da3bbdd4
KM
491 spin_unlock(&dentry->d_lock);
492 } else {
493 list_move_tail(&dentry->d_lru, &tmp);
494 spin_unlock(&dentry->d_lock);
495 cnt--;
496 if (!cnt)
497 break;
0feae5c4 498 }
f3c6ba98 499 cond_resched_lock(&dcache_lock);
0feae5c4 500 }
da3bbdd4
KM
501 }
502 while (!list_empty(&tmp)) {
503 dentry = list_entry(tmp.prev, struct dentry, d_lru);
504 dentry_lru_del_init(dentry);
505 spin_lock(&dentry->d_lock);
1da177e4
LT
506 /*
507 * We found an inuse dentry which was not removed from
da3bbdd4
KM
508 * the LRU because of laziness during lookup. Do not free
509 * it - just keep it off the LRU list.
1da177e4 510 */
da3bbdd4
KM
511 if (atomic_read(&dentry->d_count)) {
512 spin_unlock(&dentry->d_lock);
1da177e4
LT
513 continue;
514 }
da3bbdd4
KM
515 prune_one_dentry(dentry);
516 /* dentry->d_lock was dropped in prune_one_dentry() */
517 cond_resched_lock(&dcache_lock);
518 }
519 if (count == NULL && !list_empty(&sb->s_dentry_lru))
520 goto restart;
521 if (count != NULL)
522 *count = cnt;
523 if (!list_empty(&referenced))
524 list_splice(&referenced, &sb->s_dentry_lru);
525 spin_unlock(&dcache_lock);
526}
527
528/**
529 * prune_dcache - shrink the dcache
530 * @count: number of entries to try to free
531 *
532 * Shrink the dcache. This is done when we need more memory, or simply when we
533 * need to unmount something (at which point we need to unuse all dentries).
534 *
535 * This function may fail to free any resources if all the dentries are in use.
536 */
537static void prune_dcache(int count)
538{
dca33252 539 struct super_block *sb, *p = NULL;
da3bbdd4
KM
540 int w_count;
541 int unused = dentry_stat.nr_unused;
542 int prune_ratio;
543 int pruned;
544
545 if (unused == 0 || count == 0)
546 return;
547 spin_lock(&dcache_lock);
da3bbdd4
KM
548 if (count >= unused)
549 prune_ratio = 1;
550 else
551 prune_ratio = unused / count;
552 spin_lock(&sb_lock);
dca33252 553 list_for_each_entry(sb, &super_blocks, s_list) {
551de6f3
AV
554 if (list_empty(&sb->s_instances))
555 continue;
da3bbdd4 556 if (sb->s_nr_dentry_unused == 0)
1da177e4 557 continue;
da3bbdd4
KM
558 sb->s_count++;
559 /* Now, we reclaim unused dentrins with fairness.
560 * We reclaim them same percentage from each superblock.
561 * We calculate number of dentries to scan on this sb
562 * as follows, but the implementation is arranged to avoid
563 * overflows:
564 * number of dentries to scan on this sb =
565 * count * (number of dentries on this sb /
566 * number of dentries in the machine)
0feae5c4 567 */
da3bbdd4
KM
568 spin_unlock(&sb_lock);
569 if (prune_ratio != 1)
570 w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1;
571 else
572 w_count = sb->s_nr_dentry_unused;
573 pruned = w_count;
0feae5c4 574 /*
da3bbdd4
KM
575 * We need to be sure this filesystem isn't being unmounted,
576 * otherwise we could race with generic_shutdown_super(), and
577 * end up holding a reference to an inode while the filesystem
578 * is unmounted. So we try to get s_umount, and make sure
579 * s_root isn't NULL.
0feae5c4 580 */
da3bbdd4
KM
581 if (down_read_trylock(&sb->s_umount)) {
582 if ((sb->s_root != NULL) &&
583 (!list_empty(&sb->s_dentry_lru))) {
584 spin_unlock(&dcache_lock);
585 __shrink_dcache_sb(sb, &w_count,
586 DCACHE_REFERENCED);
587 pruned -= w_count;
588 spin_lock(&dcache_lock);
0feae5c4 589 }
da3bbdd4 590 up_read(&sb->s_umount);
0feae5c4 591 }
da3bbdd4 592 spin_lock(&sb_lock);
dca33252
AV
593 if (p)
594 __put_super(p);
da3bbdd4 595 count -= pruned;
dca33252 596 p = sb;
79893c17
AV
597 /* more work left to do? */
598 if (count <= 0)
599 break;
1da177e4 600 }
dca33252
AV
601 if (p)
602 __put_super(p);
da3bbdd4 603 spin_unlock(&sb_lock);
1da177e4
LT
604 spin_unlock(&dcache_lock);
605}
606
1da177e4
LT
607/**
608 * shrink_dcache_sb - shrink dcache for a superblock
609 * @sb: superblock
610 *
611 * Shrink the dcache for the specified super block. This
612 * is used to free the dcache before unmounting a file
613 * system
614 */
1da177e4
LT
615void shrink_dcache_sb(struct super_block * sb)
616{
da3bbdd4 617 __shrink_dcache_sb(sb, NULL, 0);
1da177e4 618}
ec4f8605 619EXPORT_SYMBOL(shrink_dcache_sb);
1da177e4 620
c636ebdb
DH
621/*
622 * destroy a single subtree of dentries for unmount
623 * - see the comments on shrink_dcache_for_umount() for a description of the
624 * locking
625 */
626static void shrink_dcache_for_umount_subtree(struct dentry *dentry)
627{
628 struct dentry *parent;
f8713576 629 unsigned detached = 0;
c636ebdb
DH
630
631 BUG_ON(!IS_ROOT(dentry));
632
633 /* detach this root from the system */
634 spin_lock(&dcache_lock);
da3bbdd4 635 dentry_lru_del_init(dentry);
c636ebdb
DH
636 __d_drop(dentry);
637 spin_unlock(&dcache_lock);
638
639 for (;;) {
640 /* descend to the first leaf in the current subtree */
641 while (!list_empty(&dentry->d_subdirs)) {
642 struct dentry *loop;
643
644 /* this is a branch with children - detach all of them
645 * from the system in one go */
646 spin_lock(&dcache_lock);
647 list_for_each_entry(loop, &dentry->d_subdirs,
648 d_u.d_child) {
da3bbdd4 649 dentry_lru_del_init(loop);
c636ebdb
DH
650 __d_drop(loop);
651 cond_resched_lock(&dcache_lock);
652 }
653 spin_unlock(&dcache_lock);
654
655 /* move to the first child */
656 dentry = list_entry(dentry->d_subdirs.next,
657 struct dentry, d_u.d_child);
658 }
659
660 /* consume the dentries from this leaf up through its parents
661 * until we find one with children or run out altogether */
662 do {
663 struct inode *inode;
664
665 if (atomic_read(&dentry->d_count) != 0) {
666 printk(KERN_ERR
667 "BUG: Dentry %p{i=%lx,n=%s}"
668 " still in use (%d)"
669 " [unmount of %s %s]\n",
670 dentry,
671 dentry->d_inode ?
672 dentry->d_inode->i_ino : 0UL,
673 dentry->d_name.name,
674 atomic_read(&dentry->d_count),
675 dentry->d_sb->s_type->name,
676 dentry->d_sb->s_id);
677 BUG();
678 }
679
871c0067 680 if (IS_ROOT(dentry))
c636ebdb 681 parent = NULL;
871c0067
OH
682 else {
683 parent = dentry->d_parent;
c636ebdb 684 atomic_dec(&parent->d_count);
871c0067 685 }
c636ebdb
DH
686
687 list_del(&dentry->d_u.d_child);
f8713576 688 detached++;
c636ebdb
DH
689
690 inode = dentry->d_inode;
691 if (inode) {
692 dentry->d_inode = NULL;
693 list_del_init(&dentry->d_alias);
694 if (dentry->d_op && dentry->d_op->d_iput)
695 dentry->d_op->d_iput(dentry, inode);
696 else
697 iput(inode);
698 }
699
700 d_free(dentry);
701
702 /* finished when we fall off the top of the tree,
703 * otherwise we ascend to the parent and move to the
704 * next sibling if there is one */
705 if (!parent)
f8713576 706 goto out;
c636ebdb
DH
707
708 dentry = parent;
709
710 } while (list_empty(&dentry->d_subdirs));
711
712 dentry = list_entry(dentry->d_subdirs.next,
713 struct dentry, d_u.d_child);
714 }
f8713576
DH
715out:
716 /* several dentries were freed, need to correct nr_dentry */
717 spin_lock(&dcache_lock);
718 dentry_stat.nr_dentry -= detached;
719 spin_unlock(&dcache_lock);
c636ebdb
DH
720}
721
722/*
723 * destroy the dentries attached to a superblock on unmounting
724 * - we don't need to use dentry->d_lock, and only need dcache_lock when
725 * removing the dentry from the system lists and hashes because:
726 * - the superblock is detached from all mountings and open files, so the
727 * dentry trees will not be rearranged by the VFS
728 * - s_umount is write-locked, so the memory pressure shrinker will ignore
729 * any dentries belonging to this superblock that it comes across
730 * - the filesystem itself is no longer permitted to rearrange the dentries
731 * in this superblock
732 */
733void shrink_dcache_for_umount(struct super_block *sb)
734{
735 struct dentry *dentry;
736
737 if (down_read_trylock(&sb->s_umount))
738 BUG();
739
740 dentry = sb->s_root;
741 sb->s_root = NULL;
742 atomic_dec(&dentry->d_count);
743 shrink_dcache_for_umount_subtree(dentry);
744
745 while (!hlist_empty(&sb->s_anon)) {
746 dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash);
747 shrink_dcache_for_umount_subtree(dentry);
748 }
749}
750
1da177e4
LT
751/*
752 * Search for at least 1 mount point in the dentry's subdirs.
753 * We descend to the next level whenever the d_subdirs
754 * list is non-empty and continue searching.
755 */
756
757/**
758 * have_submounts - check for mounts over a dentry
759 * @parent: dentry to check.
760 *
761 * Return true if the parent or its subdirectories contain
762 * a mount point
763 */
764
765int have_submounts(struct dentry *parent)
766{
767 struct dentry *this_parent = parent;
768 struct list_head *next;
769
770 spin_lock(&dcache_lock);
771 if (d_mountpoint(parent))
772 goto positive;
773repeat:
774 next = this_parent->d_subdirs.next;
775resume:
776 while (next != &this_parent->d_subdirs) {
777 struct list_head *tmp = next;
5160ee6f 778 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1da177e4
LT
779 next = tmp->next;
780 /* Have we found a mount point ? */
781 if (d_mountpoint(dentry))
782 goto positive;
783 if (!list_empty(&dentry->d_subdirs)) {
784 this_parent = dentry;
785 goto repeat;
786 }
787 }
788 /*
789 * All done at this level ... ascend and resume the search.
790 */
791 if (this_parent != parent) {
5160ee6f 792 next = this_parent->d_u.d_child.next;
1da177e4
LT
793 this_parent = this_parent->d_parent;
794 goto resume;
795 }
796 spin_unlock(&dcache_lock);
797 return 0; /* No mount points found in tree */
798positive:
799 spin_unlock(&dcache_lock);
800 return 1;
801}
ec4f8605 802EXPORT_SYMBOL(have_submounts);
1da177e4
LT
803
804/*
805 * Search the dentry child list for the specified parent,
806 * and move any unused dentries to the end of the unused
807 * list for prune_dcache(). We descend to the next level
808 * whenever the d_subdirs list is non-empty and continue
809 * searching.
810 *
811 * It returns zero iff there are no unused children,
812 * otherwise it returns the number of children moved to
813 * the end of the unused list. This may not be the total
814 * number of unused children, because select_parent can
815 * drop the lock and return early due to latency
816 * constraints.
817 */
818static int select_parent(struct dentry * parent)
819{
820 struct dentry *this_parent = parent;
821 struct list_head *next;
822 int found = 0;
823
824 spin_lock(&dcache_lock);
825repeat:
826 next = this_parent->d_subdirs.next;
827resume:
828 while (next != &this_parent->d_subdirs) {
829 struct list_head *tmp = next;
5160ee6f 830 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1da177e4
LT
831 next = tmp->next;
832
da3bbdd4 833 dentry_lru_del_init(dentry);
1da177e4
LT
834 /*
835 * move only zero ref count dentries to the end
836 * of the unused list for prune_dcache
837 */
838 if (!atomic_read(&dentry->d_count)) {
da3bbdd4 839 dentry_lru_add_tail(dentry);
1da177e4
LT
840 found++;
841 }
842
843 /*
844 * We can return to the caller if we have found some (this
845 * ensures forward progress). We'll be coming back to find
846 * the rest.
847 */
848 if (found && need_resched())
849 goto out;
850
851 /*
852 * Descend a level if the d_subdirs list is non-empty.
853 */
854 if (!list_empty(&dentry->d_subdirs)) {
855 this_parent = dentry;
1da177e4
LT
856 goto repeat;
857 }
858 }
859 /*
860 * All done at this level ... ascend and resume the search.
861 */
862 if (this_parent != parent) {
5160ee6f 863 next = this_parent->d_u.d_child.next;
1da177e4 864 this_parent = this_parent->d_parent;
1da177e4
LT
865 goto resume;
866 }
867out:
868 spin_unlock(&dcache_lock);
869 return found;
870}
871
872/**
873 * shrink_dcache_parent - prune dcache
874 * @parent: parent of entries to prune
875 *
876 * Prune the dcache to remove unused children of the parent dentry.
877 */
878
879void shrink_dcache_parent(struct dentry * parent)
880{
da3bbdd4 881 struct super_block *sb = parent->d_sb;
1da177e4
LT
882 int found;
883
884 while ((found = select_parent(parent)) != 0)
da3bbdd4 885 __shrink_dcache_sb(sb, &found, 0);
1da177e4 886}
ec4f8605 887EXPORT_SYMBOL(shrink_dcache_parent);
1da177e4 888
1da177e4
LT
889/*
890 * Scan `nr' dentries and return the number which remain.
891 *
892 * We need to avoid reentering the filesystem if the caller is performing a
893 * GFP_NOFS allocation attempt. One example deadlock is:
894 *
895 * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache->
896 * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode->
897 * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK.
898 *
899 * In this case we return -1 to tell the caller that we baled.
900 */
7f8275d0 901static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask)
1da177e4
LT
902{
903 if (nr) {
904 if (!(gfp_mask & __GFP_FS))
905 return -1;
da3bbdd4 906 prune_dcache(nr);
1da177e4
LT
907 }
908 return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure;
909}
910
8e1f936b
RR
911static struct shrinker dcache_shrinker = {
912 .shrink = shrink_dcache_memory,
913 .seeks = DEFAULT_SEEKS,
914};
915
1da177e4
LT
916/**
917 * d_alloc - allocate a dcache entry
918 * @parent: parent of entry to allocate
919 * @name: qstr of the name
920 *
921 * Allocates a dentry. It returns %NULL if there is insufficient memory
922 * available. On a success the dentry is returned. The name passed in is
923 * copied and the copy passed in may be reused after this call.
924 */
925
926struct dentry *d_alloc(struct dentry * parent, const struct qstr *name)
927{
928 struct dentry *dentry;
929 char *dname;
930
e12ba74d 931 dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL);
1da177e4
LT
932 if (!dentry)
933 return NULL;
934
935 if (name->len > DNAME_INLINE_LEN-1) {
936 dname = kmalloc(name->len + 1, GFP_KERNEL);
937 if (!dname) {
938 kmem_cache_free(dentry_cache, dentry);
939 return NULL;
940 }
941 } else {
942 dname = dentry->d_iname;
943 }
944 dentry->d_name.name = dname;
945
946 dentry->d_name.len = name->len;
947 dentry->d_name.hash = name->hash;
948 memcpy(dname, name->name, name->len);
949 dname[name->len] = 0;
950
951 atomic_set(&dentry->d_count, 1);
952 dentry->d_flags = DCACHE_UNHASHED;
953 spin_lock_init(&dentry->d_lock);
954 dentry->d_inode = NULL;
955 dentry->d_parent = NULL;
956 dentry->d_sb = NULL;
957 dentry->d_op = NULL;
958 dentry->d_fsdata = NULL;
959 dentry->d_mounted = 0;
1da177e4
LT
960 INIT_HLIST_NODE(&dentry->d_hash);
961 INIT_LIST_HEAD(&dentry->d_lru);
962 INIT_LIST_HEAD(&dentry->d_subdirs);
963 INIT_LIST_HEAD(&dentry->d_alias);
964
965 if (parent) {
966 dentry->d_parent = dget(parent);
967 dentry->d_sb = parent->d_sb;
968 } else {
5160ee6f 969 INIT_LIST_HEAD(&dentry->d_u.d_child);
1da177e4
LT
970 }
971
972 spin_lock(&dcache_lock);
973 if (parent)
5160ee6f 974 list_add(&dentry->d_u.d_child, &parent->d_subdirs);
1da177e4
LT
975 dentry_stat.nr_dentry++;
976 spin_unlock(&dcache_lock);
977
978 return dentry;
979}
ec4f8605 980EXPORT_SYMBOL(d_alloc);
1da177e4
LT
981
982struct dentry *d_alloc_name(struct dentry *parent, const char *name)
983{
984 struct qstr q;
985
986 q.name = name;
987 q.len = strlen(name);
988 q.hash = full_name_hash(q.name, q.len);
989 return d_alloc(parent, &q);
990}
ef26ca97 991EXPORT_SYMBOL(d_alloc_name);
1da177e4 992
360da900
OH
993/* the caller must hold dcache_lock */
994static void __d_instantiate(struct dentry *dentry, struct inode *inode)
995{
996 if (inode)
997 list_add(&dentry->d_alias, &inode->i_dentry);
998 dentry->d_inode = inode;
999 fsnotify_d_instantiate(dentry, inode);
1000}
1001
1da177e4
LT
1002/**
1003 * d_instantiate - fill in inode information for a dentry
1004 * @entry: dentry to complete
1005 * @inode: inode to attach to this dentry
1006 *
1007 * Fill in inode information in the entry.
1008 *
1009 * This turns negative dentries into productive full members
1010 * of society.
1011 *
1012 * NOTE! This assumes that the inode count has been incremented
1013 * (or otherwise set) by the caller to indicate that it is now
1014 * in use by the dcache.
1015 */
1016
1017void d_instantiate(struct dentry *entry, struct inode * inode)
1018{
28133c7b 1019 BUG_ON(!list_empty(&entry->d_alias));
1da177e4 1020 spin_lock(&dcache_lock);
360da900 1021 __d_instantiate(entry, inode);
1da177e4
LT
1022 spin_unlock(&dcache_lock);
1023 security_d_instantiate(entry, inode);
1024}
ec4f8605 1025EXPORT_SYMBOL(d_instantiate);
1da177e4
LT
1026
1027/**
1028 * d_instantiate_unique - instantiate a non-aliased dentry
1029 * @entry: dentry to instantiate
1030 * @inode: inode to attach to this dentry
1031 *
1032 * Fill in inode information in the entry. On success, it returns NULL.
1033 * If an unhashed alias of "entry" already exists, then we return the
e866cfa9 1034 * aliased dentry instead and drop one reference to inode.
1da177e4
LT
1035 *
1036 * Note that in order to avoid conflicts with rename() etc, the caller
1037 * had better be holding the parent directory semaphore.
e866cfa9
OD
1038 *
1039 * This also assumes that the inode count has been incremented
1040 * (or otherwise set) by the caller to indicate that it is now
1041 * in use by the dcache.
1da177e4 1042 */
770bfad8
DH
1043static struct dentry *__d_instantiate_unique(struct dentry *entry,
1044 struct inode *inode)
1da177e4
LT
1045{
1046 struct dentry *alias;
1047 int len = entry->d_name.len;
1048 const char *name = entry->d_name.name;
1049 unsigned int hash = entry->d_name.hash;
1050
770bfad8 1051 if (!inode) {
360da900 1052 __d_instantiate(entry, NULL);
770bfad8
DH
1053 return NULL;
1054 }
1055
1da177e4
LT
1056 list_for_each_entry(alias, &inode->i_dentry, d_alias) {
1057 struct qstr *qstr = &alias->d_name;
1058
1059 if (qstr->hash != hash)
1060 continue;
1061 if (alias->d_parent != entry->d_parent)
1062 continue;
1063 if (qstr->len != len)
1064 continue;
1065 if (memcmp(qstr->name, name, len))
1066 continue;
1067 dget_locked(alias);
1da177e4
LT
1068 return alias;
1069 }
770bfad8 1070
360da900 1071 __d_instantiate(entry, inode);
1da177e4
LT
1072 return NULL;
1073}
770bfad8
DH
1074
1075struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode)
1076{
1077 struct dentry *result;
1078
1079 BUG_ON(!list_empty(&entry->d_alias));
1080
1081 spin_lock(&dcache_lock);
1082 result = __d_instantiate_unique(entry, inode);
1083 spin_unlock(&dcache_lock);
1084
1085 if (!result) {
1086 security_d_instantiate(entry, inode);
1087 return NULL;
1088 }
1089
1090 BUG_ON(!d_unhashed(result));
1091 iput(inode);
1092 return result;
1093}
1094
1da177e4
LT
1095EXPORT_SYMBOL(d_instantiate_unique);
1096
1097/**
1098 * d_alloc_root - allocate root dentry
1099 * @root_inode: inode to allocate the root for
1100 *
1101 * Allocate a root ("/") dentry for the inode given. The inode is
1102 * instantiated and returned. %NULL is returned if there is insufficient
1103 * memory or the inode passed is %NULL.
1104 */
1105
1106struct dentry * d_alloc_root(struct inode * root_inode)
1107{
1108 struct dentry *res = NULL;
1109
1110 if (root_inode) {
1111 static const struct qstr name = { .name = "/", .len = 1 };
1112
1113 res = d_alloc(NULL, &name);
1114 if (res) {
1115 res->d_sb = root_inode->i_sb;
1116 res->d_parent = res;
1117 d_instantiate(res, root_inode);
1118 }
1119 }
1120 return res;
1121}
ec4f8605 1122EXPORT_SYMBOL(d_alloc_root);
1da177e4
LT
1123
1124static inline struct hlist_head *d_hash(struct dentry *parent,
1125 unsigned long hash)
1126{
1127 hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES;
1128 hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS);
1129 return dentry_hashtable + (hash & D_HASHMASK);
1130}
1131
4ea3ada2
CH
1132/**
1133 * d_obtain_alias - find or allocate a dentry for a given inode
1134 * @inode: inode to allocate the dentry for
1135 *
1136 * Obtain a dentry for an inode resulting from NFS filehandle conversion or
1137 * similar open by handle operations. The returned dentry may be anonymous,
1138 * or may have a full name (if the inode was already in the cache).
1139 *
1140 * When called on a directory inode, we must ensure that the inode only ever
1141 * has one dentry. If a dentry is found, that is returned instead of
1142 * allocating a new one.
1143 *
1144 * On successful return, the reference to the inode has been transferred
44003728
CH
1145 * to the dentry. In case of an error the reference on the inode is released.
1146 * To make it easier to use in export operations a %NULL or IS_ERR inode may
1147 * be passed in and will be the error will be propagate to the return value,
1148 * with a %NULL @inode replaced by ERR_PTR(-ESTALE).
4ea3ada2
CH
1149 */
1150struct dentry *d_obtain_alias(struct inode *inode)
1151{
9308a612
CH
1152 static const struct qstr anonstring = { .name = "" };
1153 struct dentry *tmp;
1154 struct dentry *res;
4ea3ada2
CH
1155
1156 if (!inode)
44003728 1157 return ERR_PTR(-ESTALE);
4ea3ada2
CH
1158 if (IS_ERR(inode))
1159 return ERR_CAST(inode);
1160
9308a612
CH
1161 res = d_find_alias(inode);
1162 if (res)
1163 goto out_iput;
1164
1165 tmp = d_alloc(NULL, &anonstring);
1166 if (!tmp) {
1167 res = ERR_PTR(-ENOMEM);
1168 goto out_iput;
4ea3ada2 1169 }
9308a612
CH
1170 tmp->d_parent = tmp; /* make sure dput doesn't croak */
1171
1172 spin_lock(&dcache_lock);
1173 res = __d_find_alias(inode, 0);
1174 if (res) {
1175 spin_unlock(&dcache_lock);
1176 dput(tmp);
1177 goto out_iput;
1178 }
1179
1180 /* attach a disconnected dentry */
1181 spin_lock(&tmp->d_lock);
1182 tmp->d_sb = inode->i_sb;
1183 tmp->d_inode = inode;
1184 tmp->d_flags |= DCACHE_DISCONNECTED;
1185 tmp->d_flags &= ~DCACHE_UNHASHED;
1186 list_add(&tmp->d_alias, &inode->i_dentry);
1187 hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon);
1188 spin_unlock(&tmp->d_lock);
1189
1190 spin_unlock(&dcache_lock);
1191 return tmp;
1192
1193 out_iput:
1194 iput(inode);
1195 return res;
4ea3ada2 1196}
adc48720 1197EXPORT_SYMBOL(d_obtain_alias);
1da177e4
LT
1198
1199/**
1200 * d_splice_alias - splice a disconnected dentry into the tree if one exists
1201 * @inode: the inode which may have a disconnected dentry
1202 * @dentry: a negative dentry which we want to point to the inode.
1203 *
1204 * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and
1205 * DCACHE_DISCONNECTED), then d_move that in place of the given dentry
1206 * and return it, else simply d_add the inode to the dentry and return NULL.
1207 *
1208 * This is needed in the lookup routine of any filesystem that is exportable
1209 * (via knfsd) so that we can build dcache paths to directories effectively.
1210 *
1211 * If a dentry was found and moved, then it is returned. Otherwise NULL
1212 * is returned. This matches the expected return value of ->lookup.
1213 *
1214 */
1215struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry)
1216{
1217 struct dentry *new = NULL;
1218
21c0d8fd 1219 if (inode && S_ISDIR(inode->i_mode)) {
1da177e4
LT
1220 spin_lock(&dcache_lock);
1221 new = __d_find_alias(inode, 1);
1222 if (new) {
1223 BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED));
1224 spin_unlock(&dcache_lock);
1225 security_d_instantiate(new, inode);
1da177e4
LT
1226 d_move(new, dentry);
1227 iput(inode);
1228 } else {
360da900
OH
1229 /* already taking dcache_lock, so d_add() by hand */
1230 __d_instantiate(dentry, inode);
1da177e4
LT
1231 spin_unlock(&dcache_lock);
1232 security_d_instantiate(dentry, inode);
1233 d_rehash(dentry);
1234 }
1235 } else
1236 d_add(dentry, inode);
1237 return new;
1238}
ec4f8605 1239EXPORT_SYMBOL(d_splice_alias);
1da177e4 1240
9403540c
BN
1241/**
1242 * d_add_ci - lookup or allocate new dentry with case-exact name
1243 * @inode: the inode case-insensitive lookup has found
1244 * @dentry: the negative dentry that was passed to the parent's lookup func
1245 * @name: the case-exact name to be associated with the returned dentry
1246 *
1247 * This is to avoid filling the dcache with case-insensitive names to the
1248 * same inode, only the actual correct case is stored in the dcache for
1249 * case-insensitive filesystems.
1250 *
1251 * For a case-insensitive lookup match and if the the case-exact dentry
1252 * already exists in in the dcache, use it and return it.
1253 *
1254 * If no entry exists with the exact case name, allocate new dentry with
1255 * the exact case, and return the spliced entry.
1256 */
e45b590b 1257struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode,
9403540c
BN
1258 struct qstr *name)
1259{
1260 int error;
1261 struct dentry *found;
1262 struct dentry *new;
1263
b6520c81
CH
1264 /*
1265 * First check if a dentry matching the name already exists,
1266 * if not go ahead and create it now.
1267 */
9403540c 1268 found = d_hash_and_lookup(dentry->d_parent, name);
9403540c
BN
1269 if (!found) {
1270 new = d_alloc(dentry->d_parent, name);
1271 if (!new) {
1272 error = -ENOMEM;
1273 goto err_out;
1274 }
b6520c81 1275
9403540c
BN
1276 found = d_splice_alias(inode, new);
1277 if (found) {
1278 dput(new);
1279 return found;
1280 }
1281 return new;
1282 }
b6520c81
CH
1283
1284 /*
1285 * If a matching dentry exists, and it's not negative use it.
1286 *
1287 * Decrement the reference count to balance the iget() done
1288 * earlier on.
1289 */
9403540c
BN
1290 if (found->d_inode) {
1291 if (unlikely(found->d_inode != inode)) {
1292 /* This can't happen because bad inodes are unhashed. */
1293 BUG_ON(!is_bad_inode(inode));
1294 BUG_ON(!is_bad_inode(found->d_inode));
1295 }
9403540c
BN
1296 iput(inode);
1297 return found;
1298 }
b6520c81 1299
9403540c
BN
1300 /*
1301 * Negative dentry: instantiate it unless the inode is a directory and
b6520c81 1302 * already has a dentry.
9403540c 1303 */
9403540c 1304 spin_lock(&dcache_lock);
b6520c81 1305 if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) {
360da900 1306 __d_instantiate(found, inode);
9403540c
BN
1307 spin_unlock(&dcache_lock);
1308 security_d_instantiate(found, inode);
1309 return found;
1310 }
b6520c81 1311
9403540c 1312 /*
b6520c81
CH
1313 * In case a directory already has a (disconnected) entry grab a
1314 * reference to it, move it in place and use it.
9403540c
BN
1315 */
1316 new = list_entry(inode->i_dentry.next, struct dentry, d_alias);
1317 dget_locked(new);
1318 spin_unlock(&dcache_lock);
9403540c 1319 security_d_instantiate(found, inode);
9403540c 1320 d_move(new, found);
9403540c 1321 iput(inode);
9403540c 1322 dput(found);
9403540c
BN
1323 return new;
1324
1325err_out:
1326 iput(inode);
1327 return ERR_PTR(error);
1328}
ec4f8605 1329EXPORT_SYMBOL(d_add_ci);
1da177e4
LT
1330
1331/**
1332 * d_lookup - search for a dentry
1333 * @parent: parent dentry
1334 * @name: qstr of name we wish to find
b04f784e 1335 * Returns: dentry, or NULL
1da177e4 1336 *
b04f784e
NP
1337 * d_lookup searches the children of the parent dentry for the name in
1338 * question. If the dentry is found its reference count is incremented and the
1339 * dentry is returned. The caller must use dput to free the entry when it has
1340 * finished using it. %NULL is returned if the dentry does not exist.
1da177e4 1341 */
1da177e4
LT
1342struct dentry * d_lookup(struct dentry * parent, struct qstr * name)
1343{
1344 struct dentry * dentry = NULL;
1345 unsigned long seq;
1346
1347 do {
1348 seq = read_seqbegin(&rename_lock);
1349 dentry = __d_lookup(parent, name);
1350 if (dentry)
1351 break;
1352 } while (read_seqretry(&rename_lock, seq));
1353 return dentry;
1354}
ec4f8605 1355EXPORT_SYMBOL(d_lookup);
1da177e4 1356
b04f784e
NP
1357/*
1358 * __d_lookup - search for a dentry (racy)
1359 * @parent: parent dentry
1360 * @name: qstr of name we wish to find
1361 * Returns: dentry, or NULL
1362 *
1363 * __d_lookup is like d_lookup, however it may (rarely) return a
1364 * false-negative result due to unrelated rename activity.
1365 *
1366 * __d_lookup is slightly faster by avoiding rename_lock read seqlock,
1367 * however it must be used carefully, eg. with a following d_lookup in
1368 * the case of failure.
1369 *
1370 * __d_lookup callers must be commented.
1371 */
1da177e4
LT
1372struct dentry * __d_lookup(struct dentry * parent, struct qstr * name)
1373{
1374 unsigned int len = name->len;
1375 unsigned int hash = name->hash;
1376 const unsigned char *str = name->name;
1377 struct hlist_head *head = d_hash(parent,hash);
1378 struct dentry *found = NULL;
1379 struct hlist_node *node;
665a7583 1380 struct dentry *dentry;
1da177e4 1381
b04f784e
NP
1382 /*
1383 * The hash list is protected using RCU.
1384 *
1385 * Take d_lock when comparing a candidate dentry, to avoid races
1386 * with d_move().
1387 *
1388 * It is possible that concurrent renames can mess up our list
1389 * walk here and result in missing our dentry, resulting in the
1390 * false-negative result. d_lookup() protects against concurrent
1391 * renames using rename_lock seqlock.
1392 *
1393 * See Documentation/vfs/dcache-locking.txt for more details.
1394 */
1da177e4
LT
1395 rcu_read_lock();
1396
665a7583 1397 hlist_for_each_entry_rcu(dentry, node, head, d_hash) {
1da177e4
LT
1398 struct qstr *qstr;
1399
1da177e4
LT
1400 if (dentry->d_name.hash != hash)
1401 continue;
1402 if (dentry->d_parent != parent)
1403 continue;
1404
1405 spin_lock(&dentry->d_lock);
1406
1407 /*
1408 * Recheck the dentry after taking the lock - d_move may have
b04f784e
NP
1409 * changed things. Don't bother checking the hash because
1410 * we're about to compare the whole name anyway.
1da177e4
LT
1411 */
1412 if (dentry->d_parent != parent)
1413 goto next;
1414
d0185c08
LT
1415 /* non-existing due to RCU? */
1416 if (d_unhashed(dentry))
1417 goto next;
1418
1da177e4
LT
1419 /*
1420 * It is safe to compare names since d_move() cannot
1421 * change the qstr (protected by d_lock).
1422 */
1423 qstr = &dentry->d_name;
1424 if (parent->d_op && parent->d_op->d_compare) {
1425 if (parent->d_op->d_compare(parent, qstr, name))
1426 goto next;
1427 } else {
1428 if (qstr->len != len)
1429 goto next;
1430 if (memcmp(qstr->name, str, len))
1431 goto next;
1432 }
1433
d0185c08
LT
1434 atomic_inc(&dentry->d_count);
1435 found = dentry;
1da177e4
LT
1436 spin_unlock(&dentry->d_lock);
1437 break;
1438next:
1439 spin_unlock(&dentry->d_lock);
1440 }
1441 rcu_read_unlock();
1442
1443 return found;
1444}
1445
3e7e241f
EB
1446/**
1447 * d_hash_and_lookup - hash the qstr then search for a dentry
1448 * @dir: Directory to search in
1449 * @name: qstr of name we wish to find
1450 *
1451 * On hash failure or on lookup failure NULL is returned.
1452 */
1453struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name)
1454{
1455 struct dentry *dentry = NULL;
1456
1457 /*
1458 * Check for a fs-specific hash function. Note that we must
1459 * calculate the standard hash first, as the d_op->d_hash()
1460 * routine may choose to leave the hash value unchanged.
1461 */
1462 name->hash = full_name_hash(name->name, name->len);
1463 if (dir->d_op && dir->d_op->d_hash) {
1464 if (dir->d_op->d_hash(dir, name) < 0)
1465 goto out;
1466 }
1467 dentry = d_lookup(dir, name);
1468out:
1469 return dentry;
1470}
1471
1da177e4
LT
1472/**
1473 * d_validate - verify dentry provided from insecure source
1474 * @dentry: The dentry alleged to be valid child of @dparent
1475 * @dparent: The parent dentry (known to be valid)
1da177e4
LT
1476 *
1477 * An insecure source has sent us a dentry, here we verify it and dget() it.
1478 * This is used by ncpfs in its readdir implementation.
1479 * Zero is returned in the dentry is invalid.
1480 */
1481
1482int d_validate(struct dentry *dentry, struct dentry *dparent)
1483{
1484 struct hlist_head *base;
1485 struct hlist_node *lhp;
1486
1487 /* Check whether the ptr might be valid at all.. */
1488 if (!kmem_ptr_validate(dentry_cache, dentry))
1489 goto out;
1490
1491 if (dentry->d_parent != dparent)
1492 goto out;
1493
1494 spin_lock(&dcache_lock);
1495 base = d_hash(dparent, dentry->d_name.hash);
1496 hlist_for_each(lhp,base) {
665a7583 1497 /* hlist_for_each_entry_rcu() not required for d_hash list
1da177e4
LT
1498 * as it is parsed under dcache_lock
1499 */
1500 if (dentry == hlist_entry(lhp, struct dentry, d_hash)) {
1501 __dget_locked(dentry);
1502 spin_unlock(&dcache_lock);
1503 return 1;
1504 }
1505 }
1506 spin_unlock(&dcache_lock);
1507out:
1508 return 0;
1509}
ec4f8605 1510EXPORT_SYMBOL(d_validate);
1da177e4
LT
1511
1512/*
1513 * When a file is deleted, we have two options:
1514 * - turn this dentry into a negative dentry
1515 * - unhash this dentry and free it.
1516 *
1517 * Usually, we want to just turn this into
1518 * a negative dentry, but if anybody else is
1519 * currently using the dentry or the inode
1520 * we can't do that and we fall back on removing
1521 * it from the hash queues and waiting for
1522 * it to be deleted later when it has no users
1523 */
1524
1525/**
1526 * d_delete - delete a dentry
1527 * @dentry: The dentry to delete
1528 *
1529 * Turn the dentry into a negative dentry if possible, otherwise
1530 * remove it from the hash queues so it can be deleted later
1531 */
1532
1533void d_delete(struct dentry * dentry)
1534{
7a91bf7f 1535 int isdir = 0;
1da177e4
LT
1536 /*
1537 * Are we the only user?
1538 */
1539 spin_lock(&dcache_lock);
1540 spin_lock(&dentry->d_lock);
7a91bf7f 1541 isdir = S_ISDIR(dentry->d_inode->i_mode);
1da177e4 1542 if (atomic_read(&dentry->d_count) == 1) {
13e3c5e5 1543 dentry->d_flags &= ~DCACHE_CANT_MOUNT;
1da177e4 1544 dentry_iput(dentry);
7a91bf7f 1545 fsnotify_nameremove(dentry, isdir);
1da177e4
LT
1546 return;
1547 }
1548
1549 if (!d_unhashed(dentry))
1550 __d_drop(dentry);
1551
1552 spin_unlock(&dentry->d_lock);
1553 spin_unlock(&dcache_lock);
7a91bf7f
JM
1554
1555 fsnotify_nameremove(dentry, isdir);
1da177e4 1556}
ec4f8605 1557EXPORT_SYMBOL(d_delete);
1da177e4
LT
1558
1559static void __d_rehash(struct dentry * entry, struct hlist_head *list)
1560{
1561
1562 entry->d_flags &= ~DCACHE_UNHASHED;
1563 hlist_add_head_rcu(&entry->d_hash, list);
1564}
1565
770bfad8
DH
1566static void _d_rehash(struct dentry * entry)
1567{
1568 __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash));
1569}
1570
1da177e4
LT
1571/**
1572 * d_rehash - add an entry back to the hash
1573 * @entry: dentry to add to the hash
1574 *
1575 * Adds a dentry to the hash according to its name.
1576 */
1577
1578void d_rehash(struct dentry * entry)
1579{
1da177e4
LT
1580 spin_lock(&dcache_lock);
1581 spin_lock(&entry->d_lock);
770bfad8 1582 _d_rehash(entry);
1da177e4
LT
1583 spin_unlock(&entry->d_lock);
1584 spin_unlock(&dcache_lock);
1585}
ec4f8605 1586EXPORT_SYMBOL(d_rehash);
1da177e4 1587
1da177e4
LT
1588/*
1589 * When switching names, the actual string doesn't strictly have to
1590 * be preserved in the target - because we're dropping the target
1591 * anyway. As such, we can just do a simple memcpy() to copy over
1592 * the new name before we switch.
1593 *
1594 * Note that we have to be a lot more careful about getting the hash
1595 * switched - we have to switch the hash value properly even if it
1596 * then no longer matches the actual (corrupted) string of the target.
1597 * The hash value has to match the hash queue that the dentry is on..
1598 */
1599static void switch_names(struct dentry *dentry, struct dentry *target)
1600{
1601 if (dname_external(target)) {
1602 if (dname_external(dentry)) {
1603 /*
1604 * Both external: swap the pointers
1605 */
9a8d5bb4 1606 swap(target->d_name.name, dentry->d_name.name);
1da177e4
LT
1607 } else {
1608 /*
1609 * dentry:internal, target:external. Steal target's
1610 * storage and make target internal.
1611 */
321bcf92
BF
1612 memcpy(target->d_iname, dentry->d_name.name,
1613 dentry->d_name.len + 1);
1da177e4
LT
1614 dentry->d_name.name = target->d_name.name;
1615 target->d_name.name = target->d_iname;
1616 }
1617 } else {
1618 if (dname_external(dentry)) {
1619 /*
1620 * dentry:external, target:internal. Give dentry's
1621 * storage to target and make dentry internal
1622 */
1623 memcpy(dentry->d_iname, target->d_name.name,
1624 target->d_name.len + 1);
1625 target->d_name.name = dentry->d_name.name;
1626 dentry->d_name.name = dentry->d_iname;
1627 } else {
1628 /*
1629 * Both are internal. Just copy target to dentry
1630 */
1631 memcpy(dentry->d_iname, target->d_name.name,
1632 target->d_name.len + 1);
dc711ca3
AV
1633 dentry->d_name.len = target->d_name.len;
1634 return;
1da177e4
LT
1635 }
1636 }
9a8d5bb4 1637 swap(dentry->d_name.len, target->d_name.len);
1da177e4
LT
1638}
1639
1640/*
1641 * We cannibalize "target" when moving dentry on top of it,
1642 * because it's going to be thrown away anyway. We could be more
1643 * polite about it, though.
1644 *
1645 * This forceful removal will result in ugly /proc output if
1646 * somebody holds a file open that got deleted due to a rename.
1647 * We could be nicer about the deleted file, and let it show
bc154b1e
BF
1648 * up under the name it had before it was deleted rather than
1649 * under the original name of the file that was moved on top of it.
1da177e4
LT
1650 */
1651
9eaef27b
TM
1652/*
1653 * d_move_locked - move a dentry
1da177e4
LT
1654 * @dentry: entry to move
1655 * @target: new dentry
1656 *
1657 * Update the dcache to reflect the move of a file name. Negative
1658 * dcache entries should not be moved in this way.
1659 */
9eaef27b 1660static void d_move_locked(struct dentry * dentry, struct dentry * target)
1da177e4
LT
1661{
1662 struct hlist_head *list;
1663
1664 if (!dentry->d_inode)
1665 printk(KERN_WARNING "VFS: moving negative dcache entry\n");
1666
1da177e4
LT
1667 write_seqlock(&rename_lock);
1668 /*
1669 * XXXX: do we really need to take target->d_lock?
1670 */
1671 if (target < dentry) {
1672 spin_lock(&target->d_lock);
a90b9c05 1673 spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED);
1da177e4
LT
1674 } else {
1675 spin_lock(&dentry->d_lock);
a90b9c05 1676 spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED);
1da177e4
LT
1677 }
1678
1679 /* Move the dentry to the target hash queue, if on different bucket */
f77e3498 1680 if (d_unhashed(dentry))
1da177e4
LT
1681 goto already_unhashed;
1682
1683 hlist_del_rcu(&dentry->d_hash);
1684
1685already_unhashed:
1686 list = d_hash(target->d_parent, target->d_name.hash);
1687 __d_rehash(dentry, list);
1688
1689 /* Unhash the target: dput() will then get rid of it */
1690 __d_drop(target);
1691
5160ee6f
ED
1692 list_del(&dentry->d_u.d_child);
1693 list_del(&target->d_u.d_child);
1da177e4
LT
1694
1695 /* Switch the names.. */
1696 switch_names(dentry, target);
9a8d5bb4 1697 swap(dentry->d_name.hash, target->d_name.hash);
1da177e4
LT
1698
1699 /* ... and switch the parents */
1700 if (IS_ROOT(dentry)) {
1701 dentry->d_parent = target->d_parent;
1702 target->d_parent = target;
5160ee6f 1703 INIT_LIST_HEAD(&target->d_u.d_child);
1da177e4 1704 } else {
9a8d5bb4 1705 swap(dentry->d_parent, target->d_parent);
1da177e4
LT
1706
1707 /* And add them back to the (new) parent lists */
5160ee6f 1708 list_add(&target->d_u.d_child, &target->d_parent->d_subdirs);
1da177e4
LT
1709 }
1710
5160ee6f 1711 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1da177e4 1712 spin_unlock(&target->d_lock);
c32ccd87 1713 fsnotify_d_move(dentry);
1da177e4
LT
1714 spin_unlock(&dentry->d_lock);
1715 write_sequnlock(&rename_lock);
9eaef27b
TM
1716}
1717
1718/**
1719 * d_move - move a dentry
1720 * @dentry: entry to move
1721 * @target: new dentry
1722 *
1723 * Update the dcache to reflect the move of a file name. Negative
1724 * dcache entries should not be moved in this way.
1725 */
1726
1727void d_move(struct dentry * dentry, struct dentry * target)
1728{
1729 spin_lock(&dcache_lock);
1730 d_move_locked(dentry, target);
1da177e4
LT
1731 spin_unlock(&dcache_lock);
1732}
ec4f8605 1733EXPORT_SYMBOL(d_move);
1da177e4 1734
e2761a11
OH
1735/**
1736 * d_ancestor - search for an ancestor
1737 * @p1: ancestor dentry
1738 * @p2: child dentry
1739 *
1740 * Returns the ancestor dentry of p2 which is a child of p1, if p1 is
1741 * an ancestor of p2, else NULL.
9eaef27b 1742 */
e2761a11 1743struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2)
9eaef27b
TM
1744{
1745 struct dentry *p;
1746
871c0067 1747 for (p = p2; !IS_ROOT(p); p = p->d_parent) {
9eaef27b 1748 if (p->d_parent == p1)
e2761a11 1749 return p;
9eaef27b 1750 }
e2761a11 1751 return NULL;
9eaef27b
TM
1752}
1753
1754/*
1755 * This helper attempts to cope with remotely renamed directories
1756 *
1757 * It assumes that the caller is already holding
1758 * dentry->d_parent->d_inode->i_mutex and the dcache_lock
1759 *
1760 * Note: If ever the locking in lock_rename() changes, then please
1761 * remember to update this too...
9eaef27b
TM
1762 */
1763static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias)
31f3e0b3 1764 __releases(dcache_lock)
9eaef27b
TM
1765{
1766 struct mutex *m1 = NULL, *m2 = NULL;
1767 struct dentry *ret;
1768
1769 /* If alias and dentry share a parent, then no extra locks required */
1770 if (alias->d_parent == dentry->d_parent)
1771 goto out_unalias;
1772
1773 /* Check for loops */
1774 ret = ERR_PTR(-ELOOP);
e2761a11 1775 if (d_ancestor(alias, dentry))
9eaef27b
TM
1776 goto out_err;
1777
1778 /* See lock_rename() */
1779 ret = ERR_PTR(-EBUSY);
1780 if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex))
1781 goto out_err;
1782 m1 = &dentry->d_sb->s_vfs_rename_mutex;
1783 if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex))
1784 goto out_err;
1785 m2 = &alias->d_parent->d_inode->i_mutex;
1786out_unalias:
1787 d_move_locked(alias, dentry);
1788 ret = alias;
1789out_err:
1790 spin_unlock(&dcache_lock);
1791 if (m2)
1792 mutex_unlock(m2);
1793 if (m1)
1794 mutex_unlock(m1);
1795 return ret;
1796}
1797
770bfad8
DH
1798/*
1799 * Prepare an anonymous dentry for life in the superblock's dentry tree as a
1800 * named dentry in place of the dentry to be replaced.
1801 */
1802static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon)
1803{
1804 struct dentry *dparent, *aparent;
1805
1806 switch_names(dentry, anon);
9a8d5bb4 1807 swap(dentry->d_name.hash, anon->d_name.hash);
770bfad8
DH
1808
1809 dparent = dentry->d_parent;
1810 aparent = anon->d_parent;
1811
1812 dentry->d_parent = (aparent == anon) ? dentry : aparent;
1813 list_del(&dentry->d_u.d_child);
1814 if (!IS_ROOT(dentry))
1815 list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs);
1816 else
1817 INIT_LIST_HEAD(&dentry->d_u.d_child);
1818
1819 anon->d_parent = (dparent == dentry) ? anon : dparent;
1820 list_del(&anon->d_u.d_child);
1821 if (!IS_ROOT(anon))
1822 list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs);
1823 else
1824 INIT_LIST_HEAD(&anon->d_u.d_child);
1825
1826 anon->d_flags &= ~DCACHE_DISCONNECTED;
1827}
1828
1829/**
1830 * d_materialise_unique - introduce an inode into the tree
1831 * @dentry: candidate dentry
1832 * @inode: inode to bind to the dentry, to which aliases may be attached
1833 *
1834 * Introduces an dentry into the tree, substituting an extant disconnected
1835 * root directory alias in its place if there is one
1836 */
1837struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode)
1838{
9eaef27b 1839 struct dentry *actual;
770bfad8
DH
1840
1841 BUG_ON(!d_unhashed(dentry));
1842
1843 spin_lock(&dcache_lock);
1844
1845 if (!inode) {
1846 actual = dentry;
360da900 1847 __d_instantiate(dentry, NULL);
770bfad8
DH
1848 goto found_lock;
1849 }
1850
9eaef27b
TM
1851 if (S_ISDIR(inode->i_mode)) {
1852 struct dentry *alias;
1853
1854 /* Does an aliased dentry already exist? */
1855 alias = __d_find_alias(inode, 0);
1856 if (alias) {
1857 actual = alias;
1858 /* Is this an anonymous mountpoint that we could splice
1859 * into our tree? */
1860 if (IS_ROOT(alias)) {
1861 spin_lock(&alias->d_lock);
1862 __d_materialise_dentry(dentry, alias);
1863 __d_drop(alias);
1864 goto found;
1865 }
1866 /* Nope, but we must(!) avoid directory aliasing */
1867 actual = __d_unalias(dentry, alias);
1868 if (IS_ERR(actual))
1869 dput(alias);
1870 goto out_nolock;
1871 }
770bfad8
DH
1872 }
1873
1874 /* Add a unique reference */
1875 actual = __d_instantiate_unique(dentry, inode);
1876 if (!actual)
1877 actual = dentry;
1878 else if (unlikely(!d_unhashed(actual)))
1879 goto shouldnt_be_hashed;
1880
1881found_lock:
1882 spin_lock(&actual->d_lock);
1883found:
1884 _d_rehash(actual);
1885 spin_unlock(&actual->d_lock);
1886 spin_unlock(&dcache_lock);
9eaef27b 1887out_nolock:
770bfad8
DH
1888 if (actual == dentry) {
1889 security_d_instantiate(dentry, inode);
1890 return NULL;
1891 }
1892
1893 iput(inode);
1894 return actual;
1895
770bfad8
DH
1896shouldnt_be_hashed:
1897 spin_unlock(&dcache_lock);
1898 BUG();
770bfad8 1899}
ec4f8605 1900EXPORT_SYMBOL_GPL(d_materialise_unique);
770bfad8 1901
cdd16d02 1902static int prepend(char **buffer, int *buflen, const char *str, int namelen)
6092d048
RP
1903{
1904 *buflen -= namelen;
1905 if (*buflen < 0)
1906 return -ENAMETOOLONG;
1907 *buffer -= namelen;
1908 memcpy(*buffer, str, namelen);
1909 return 0;
1910}
1911
cdd16d02
MS
1912static int prepend_name(char **buffer, int *buflen, struct qstr *name)
1913{
1914 return prepend(buffer, buflen, name->name, name->len);
1915}
1916
1da177e4 1917/**
f2eb6575
MS
1918 * Prepend path string to a buffer
1919 *
9d1bc601
MS
1920 * @path: the dentry/vfsmount to report
1921 * @root: root vfsmnt/dentry (may be modified by this function)
f2eb6575
MS
1922 * @buffer: pointer to the end of the buffer
1923 * @buflen: pointer to buffer length
552ce544 1924 *
f2eb6575 1925 * Caller holds the dcache_lock.
9d1bc601
MS
1926 *
1927 * If path is not reachable from the supplied root, then the value of
1928 * root is changed (without modifying refcounts).
1da177e4 1929 */
f2eb6575
MS
1930static int prepend_path(const struct path *path, struct path *root,
1931 char **buffer, int *buflen)
1da177e4 1932{
9d1bc601
MS
1933 struct dentry *dentry = path->dentry;
1934 struct vfsmount *vfsmnt = path->mnt;
f2eb6575
MS
1935 bool slash = false;
1936 int error = 0;
6092d048 1937
99b7db7b 1938 br_read_lock(vfsmount_lock);
f2eb6575 1939 while (dentry != root->dentry || vfsmnt != root->mnt) {
1da177e4
LT
1940 struct dentry * parent;
1941
1da177e4 1942 if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) {
552ce544 1943 /* Global root? */
1da177e4 1944 if (vfsmnt->mnt_parent == vfsmnt) {
1da177e4
LT
1945 goto global_root;
1946 }
1947 dentry = vfsmnt->mnt_mountpoint;
1948 vfsmnt = vfsmnt->mnt_parent;
1da177e4
LT
1949 continue;
1950 }
1951 parent = dentry->d_parent;
1952 prefetch(parent);
f2eb6575
MS
1953 error = prepend_name(buffer, buflen, &dentry->d_name);
1954 if (!error)
1955 error = prepend(buffer, buflen, "/", 1);
1956 if (error)
1957 break;
1958
1959 slash = true;
1da177e4
LT
1960 dentry = parent;
1961 }
1962
be285c71 1963out:
f2eb6575
MS
1964 if (!error && !slash)
1965 error = prepend(buffer, buflen, "/", 1);
1966
99b7db7b 1967 br_read_unlock(vfsmount_lock);
f2eb6575 1968 return error;
1da177e4
LT
1969
1970global_root:
98dc568b
MS
1971 /*
1972 * Filesystems needing to implement special "root names"
1973 * should do so with ->d_dname()
1974 */
1975 if (IS_ROOT(dentry) &&
1976 (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) {
1977 WARN(1, "Root dentry has weird name <%.*s>\n",
1978 (int) dentry->d_name.len, dentry->d_name.name);
1979 }
9d1bc601
MS
1980 root->mnt = vfsmnt;
1981 root->dentry = dentry;
be285c71 1982 goto out;
f2eb6575 1983}
be285c71 1984
f2eb6575
MS
1985/**
1986 * __d_path - return the path of a dentry
1987 * @path: the dentry/vfsmount to report
1988 * @root: root vfsmnt/dentry (may be modified by this function)
cd956a1c 1989 * @buf: buffer to return value in
f2eb6575
MS
1990 * @buflen: buffer length
1991 *
ffd1f4ed 1992 * Convert a dentry into an ASCII path name.
f2eb6575
MS
1993 *
1994 * Returns a pointer into the buffer or an error code if the
1995 * path was too long.
1996 *
1997 * "buflen" should be positive. Caller holds the dcache_lock.
1998 *
1999 * If path is not reachable from the supplied root, then the value of
2000 * root is changed (without modifying refcounts).
2001 */
2002char *__d_path(const struct path *path, struct path *root,
2003 char *buf, int buflen)
2004{
2005 char *res = buf + buflen;
2006 int error;
2007
2008 prepend(&res, &buflen, "\0", 1);
f2eb6575
MS
2009 error = prepend_path(path, root, &res, &buflen);
2010 if (error)
2011 return ERR_PTR(error);
2012
2013 return res;
1da177e4
LT
2014}
2015
ffd1f4ed
MS
2016/*
2017 * same as __d_path but appends "(deleted)" for unlinked files.
2018 */
2019static int path_with_deleted(const struct path *path, struct path *root,
2020 char **buf, int *buflen)
2021{
2022 prepend(buf, buflen, "\0", 1);
2023 if (d_unlinked(path->dentry)) {
2024 int error = prepend(buf, buflen, " (deleted)", 10);
2025 if (error)
2026 return error;
2027 }
2028
2029 return prepend_path(path, root, buf, buflen);
2030}
2031
8df9d1a4
MS
2032static int prepend_unreachable(char **buffer, int *buflen)
2033{
2034 return prepend(buffer, buflen, "(unreachable)", 13);
2035}
2036
a03a8a70
JB
2037/**
2038 * d_path - return the path of a dentry
cf28b486 2039 * @path: path to report
a03a8a70
JB
2040 * @buf: buffer to return value in
2041 * @buflen: buffer length
2042 *
2043 * Convert a dentry into an ASCII path name. If the entry has been deleted
2044 * the string " (deleted)" is appended. Note that this is ambiguous.
2045 *
52afeefb
AV
2046 * Returns a pointer into the buffer or an error code if the path was
2047 * too long. Note: Callers should use the returned pointer, not the passed
2048 * in buffer, to use the name! The implementation often starts at an offset
2049 * into the buffer, and may leave 0 bytes at the start.
a03a8a70 2050 *
31f3e0b3 2051 * "buflen" should be positive.
a03a8a70 2052 */
20d4fdc1 2053char *d_path(const struct path *path, char *buf, int buflen)
1da177e4 2054{
ffd1f4ed 2055 char *res = buf + buflen;
6ac08c39 2056 struct path root;
9d1bc601 2057 struct path tmp;
ffd1f4ed 2058 int error;
1da177e4 2059
c23fbb6b
ED
2060 /*
2061 * We have various synthetic filesystems that never get mounted. On
2062 * these filesystems dentries are never used for lookup purposes, and
2063 * thus don't need to be hashed. They also don't need a name until a
2064 * user wants to identify the object in /proc/pid/fd/. The little hack
2065 * below allows us to generate a name for these objects on demand:
2066 */
cf28b486
JB
2067 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2068 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
c23fbb6b 2069
f7ad3c6b 2070 get_fs_root(current->fs, &root);
552ce544 2071 spin_lock(&dcache_lock);
9d1bc601 2072 tmp = root;
ffd1f4ed
MS
2073 error = path_with_deleted(path, &tmp, &res, &buflen);
2074 if (error)
2075 res = ERR_PTR(error);
552ce544 2076 spin_unlock(&dcache_lock);
6ac08c39 2077 path_put(&root);
1da177e4
LT
2078 return res;
2079}
ec4f8605 2080EXPORT_SYMBOL(d_path);
1da177e4 2081
8df9d1a4
MS
2082/**
2083 * d_path_with_unreachable - return the path of a dentry
2084 * @path: path to report
2085 * @buf: buffer to return value in
2086 * @buflen: buffer length
2087 *
2088 * The difference from d_path() is that this prepends "(unreachable)"
2089 * to paths which are unreachable from the current process' root.
2090 */
2091char *d_path_with_unreachable(const struct path *path, char *buf, int buflen)
2092{
2093 char *res = buf + buflen;
2094 struct path root;
2095 struct path tmp;
2096 int error;
2097
2098 if (path->dentry->d_op && path->dentry->d_op->d_dname)
2099 return path->dentry->d_op->d_dname(path->dentry, buf, buflen);
2100
2101 get_fs_root(current->fs, &root);
2102 spin_lock(&dcache_lock);
2103 tmp = root;
2104 error = path_with_deleted(path, &tmp, &res, &buflen);
2105 if (!error && !path_equal(&tmp, &root))
2106 error = prepend_unreachable(&res, &buflen);
2107 spin_unlock(&dcache_lock);
2108 path_put(&root);
2109 if (error)
2110 res = ERR_PTR(error);
2111
2112 return res;
2113}
2114
c23fbb6b
ED
2115/*
2116 * Helper function for dentry_operations.d_dname() members
2117 */
2118char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen,
2119 const char *fmt, ...)
2120{
2121 va_list args;
2122 char temp[64];
2123 int sz;
2124
2125 va_start(args, fmt);
2126 sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1;
2127 va_end(args);
2128
2129 if (sz > sizeof(temp) || sz > buflen)
2130 return ERR_PTR(-ENAMETOOLONG);
2131
2132 buffer += buflen - sz;
2133 return memcpy(buffer, temp, sz);
2134}
2135
6092d048
RP
2136/*
2137 * Write full pathname from the root of the filesystem into the buffer.
2138 */
c103135c 2139char *__dentry_path(struct dentry *dentry, char *buf, int buflen)
6092d048
RP
2140{
2141 char *end = buf + buflen;
2142 char *retval;
2143
6092d048 2144 prepend(&end, &buflen, "\0", 1);
6092d048
RP
2145 if (buflen < 1)
2146 goto Elong;
2147 /* Get '/' right */
2148 retval = end-1;
2149 *retval = '/';
2150
cdd16d02
MS
2151 while (!IS_ROOT(dentry)) {
2152 struct dentry *parent = dentry->d_parent;
6092d048 2153
6092d048 2154 prefetch(parent);
cdd16d02 2155 if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) ||
6092d048
RP
2156 (prepend(&end, &buflen, "/", 1) != 0))
2157 goto Elong;
2158
2159 retval = end;
2160 dentry = parent;
2161 }
c103135c
AV
2162 return retval;
2163Elong:
2164 return ERR_PTR(-ENAMETOOLONG);
2165}
2166EXPORT_SYMBOL(__dentry_path);
2167
2168char *dentry_path(struct dentry *dentry, char *buf, int buflen)
2169{
2170 char *p = NULL;
2171 char *retval;
2172
2173 spin_lock(&dcache_lock);
2174 if (d_unlinked(dentry)) {
2175 p = buf + buflen;
2176 if (prepend(&p, &buflen, "//deleted", 10) != 0)
2177 goto Elong;
2178 buflen++;
2179 }
2180 retval = __dentry_path(dentry, buf, buflen);
6092d048 2181 spin_unlock(&dcache_lock);
c103135c
AV
2182 if (!IS_ERR(retval) && p)
2183 *p = '/'; /* restore '/' overriden with '\0' */
6092d048
RP
2184 return retval;
2185Elong:
2186 spin_unlock(&dcache_lock);
2187 return ERR_PTR(-ENAMETOOLONG);
2188}
2189
1da177e4
LT
2190/*
2191 * NOTE! The user-level library version returns a
2192 * character pointer. The kernel system call just
2193 * returns the length of the buffer filled (which
2194 * includes the ending '\0' character), or a negative
2195 * error value. So libc would do something like
2196 *
2197 * char *getcwd(char * buf, size_t size)
2198 * {
2199 * int retval;
2200 *
2201 * retval = sys_getcwd(buf, size);
2202 * if (retval >= 0)
2203 * return buf;
2204 * errno = -retval;
2205 * return NULL;
2206 * }
2207 */
3cdad428 2208SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size)
1da177e4 2209{
552ce544 2210 int error;
6ac08c39 2211 struct path pwd, root;
552ce544 2212 char *page = (char *) __get_free_page(GFP_USER);
1da177e4
LT
2213
2214 if (!page)
2215 return -ENOMEM;
2216
f7ad3c6b 2217 get_fs_root_and_pwd(current->fs, &root, &pwd);
1da177e4 2218
552ce544 2219 error = -ENOENT;
552ce544 2220 spin_lock(&dcache_lock);
f3da392e 2221 if (!d_unlinked(pwd.dentry)) {
552ce544 2222 unsigned long len;
9d1bc601 2223 struct path tmp = root;
8df9d1a4
MS
2224 char *cwd = page + PAGE_SIZE;
2225 int buflen = PAGE_SIZE;
1da177e4 2226
8df9d1a4
MS
2227 prepend(&cwd, &buflen, "\0", 1);
2228 error = prepend_path(&pwd, &tmp, &cwd, &buflen);
552ce544
LT
2229 spin_unlock(&dcache_lock);
2230
8df9d1a4 2231 if (error)
552ce544
LT
2232 goto out;
2233
8df9d1a4
MS
2234 /* Unreachable from current root */
2235 if (!path_equal(&tmp, &root)) {
2236 error = prepend_unreachable(&cwd, &buflen);
2237 if (error)
2238 goto out;
2239 }
2240
552ce544
LT
2241 error = -ERANGE;
2242 len = PAGE_SIZE + page - cwd;
2243 if (len <= size) {
2244 error = len;
2245 if (copy_to_user(buf, cwd, len))
2246 error = -EFAULT;
2247 }
2248 } else
2249 spin_unlock(&dcache_lock);
1da177e4
LT
2250
2251out:
6ac08c39
JB
2252 path_put(&pwd);
2253 path_put(&root);
1da177e4
LT
2254 free_page((unsigned long) page);
2255 return error;
2256}
2257
2258/*
2259 * Test whether new_dentry is a subdirectory of old_dentry.
2260 *
2261 * Trivially implemented using the dcache structure
2262 */
2263
2264/**
2265 * is_subdir - is new dentry a subdirectory of old_dentry
2266 * @new_dentry: new dentry
2267 * @old_dentry: old dentry
2268 *
2269 * Returns 1 if new_dentry is a subdirectory of the parent (at any depth).
2270 * Returns 0 otherwise.
2271 * Caller must ensure that "new_dentry" is pinned before calling is_subdir()
2272 */
2273
e2761a11 2274int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry)
1da177e4
LT
2275{
2276 int result;
1da177e4
LT
2277 unsigned long seq;
2278
e2761a11
OH
2279 if (new_dentry == old_dentry)
2280 return 1;
2281
2282 /*
2283 * Need rcu_readlock to protect against the d_parent trashing
2284 * due to d_move
1da177e4
LT
2285 */
2286 rcu_read_lock();
e2761a11 2287 do {
1da177e4 2288 /* for restarting inner loop in case of seq retry */
1da177e4 2289 seq = read_seqbegin(&rename_lock);
e2761a11 2290 if (d_ancestor(old_dentry, new_dentry))
1da177e4 2291 result = 1;
e2761a11
OH
2292 else
2293 result = 0;
1da177e4
LT
2294 } while (read_seqretry(&rename_lock, seq));
2295 rcu_read_unlock();
2296
2297 return result;
2298}
2299
2096f759
AV
2300int path_is_under(struct path *path1, struct path *path2)
2301{
2302 struct vfsmount *mnt = path1->mnt;
2303 struct dentry *dentry = path1->dentry;
2304 int res;
99b7db7b
NP
2305
2306 br_read_lock(vfsmount_lock);
2096f759
AV
2307 if (mnt != path2->mnt) {
2308 for (;;) {
2309 if (mnt->mnt_parent == mnt) {
99b7db7b 2310 br_read_unlock(vfsmount_lock);
2096f759
AV
2311 return 0;
2312 }
2313 if (mnt->mnt_parent == path2->mnt)
2314 break;
2315 mnt = mnt->mnt_parent;
2316 }
2317 dentry = mnt->mnt_mountpoint;
2318 }
2319 res = is_subdir(dentry, path2->dentry);
99b7db7b 2320 br_read_unlock(vfsmount_lock);
2096f759
AV
2321 return res;
2322}
2323EXPORT_SYMBOL(path_is_under);
2324
1da177e4
LT
2325void d_genocide(struct dentry *root)
2326{
2327 struct dentry *this_parent = root;
2328 struct list_head *next;
2329
2330 spin_lock(&dcache_lock);
2331repeat:
2332 next = this_parent->d_subdirs.next;
2333resume:
2334 while (next != &this_parent->d_subdirs) {
2335 struct list_head *tmp = next;
5160ee6f 2336 struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child);
1da177e4
LT
2337 next = tmp->next;
2338 if (d_unhashed(dentry)||!dentry->d_inode)
2339 continue;
2340 if (!list_empty(&dentry->d_subdirs)) {
2341 this_parent = dentry;
2342 goto repeat;
2343 }
2344 atomic_dec(&dentry->d_count);
2345 }
2346 if (this_parent != root) {
5160ee6f 2347 next = this_parent->d_u.d_child.next;
1da177e4
LT
2348 atomic_dec(&this_parent->d_count);
2349 this_parent = this_parent->d_parent;
2350 goto resume;
2351 }
2352 spin_unlock(&dcache_lock);
2353}
2354
2355/**
2356 * find_inode_number - check for dentry with name
2357 * @dir: directory to check
2358 * @name: Name to find.
2359 *
2360 * Check whether a dentry already exists for the given name,
2361 * and return the inode number if it has an inode. Otherwise
2362 * 0 is returned.
2363 *
2364 * This routine is used to post-process directory listings for
2365 * filesystems using synthetic inode numbers, and is necessary
2366 * to keep getcwd() working.
2367 */
2368
2369ino_t find_inode_number(struct dentry *dir, struct qstr *name)
2370{
2371 struct dentry * dentry;
2372 ino_t ino = 0;
2373
3e7e241f
EB
2374 dentry = d_hash_and_lookup(dir, name);
2375 if (dentry) {
1da177e4
LT
2376 if (dentry->d_inode)
2377 ino = dentry->d_inode->i_ino;
2378 dput(dentry);
2379 }
1da177e4
LT
2380 return ino;
2381}
ec4f8605 2382EXPORT_SYMBOL(find_inode_number);
1da177e4
LT
2383
2384static __initdata unsigned long dhash_entries;
2385static int __init set_dhash_entries(char *str)
2386{
2387 if (!str)
2388 return 0;
2389 dhash_entries = simple_strtoul(str, &str, 0);
2390 return 1;
2391}
2392__setup("dhash_entries=", set_dhash_entries);
2393
2394static void __init dcache_init_early(void)
2395{
2396 int loop;
2397
2398 /* If hashes are distributed across NUMA nodes, defer
2399 * hash allocation until vmalloc space is available.
2400 */
2401 if (hashdist)
2402 return;
2403
2404 dentry_hashtable =
2405 alloc_large_system_hash("Dentry cache",
2406 sizeof(struct hlist_head),
2407 dhash_entries,
2408 13,
2409 HASH_EARLY,
2410 &d_hash_shift,
2411 &d_hash_mask,
2412 0);
2413
2414 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2415 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2416}
2417
74bf17cf 2418static void __init dcache_init(void)
1da177e4
LT
2419{
2420 int loop;
2421
2422 /*
2423 * A constructor could be added for stable state like the lists,
2424 * but it is probably not worth it because of the cache nature
2425 * of the dcache.
2426 */
0a31bd5f
CL
2427 dentry_cache = KMEM_CACHE(dentry,
2428 SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD);
1da177e4 2429
8e1f936b 2430 register_shrinker(&dcache_shrinker);
1da177e4
LT
2431
2432 /* Hash may have been set up in dcache_init_early */
2433 if (!hashdist)
2434 return;
2435
2436 dentry_hashtable =
2437 alloc_large_system_hash("Dentry cache",
2438 sizeof(struct hlist_head),
2439 dhash_entries,
2440 13,
2441 0,
2442 &d_hash_shift,
2443 &d_hash_mask,
2444 0);
2445
2446 for (loop = 0; loop < (1 << d_hash_shift); loop++)
2447 INIT_HLIST_HEAD(&dentry_hashtable[loop]);
2448}
2449
2450/* SLAB cache for __getname() consumers */
e18b890b 2451struct kmem_cache *names_cachep __read_mostly;
ec4f8605 2452EXPORT_SYMBOL(names_cachep);
1da177e4 2453
1da177e4
LT
2454EXPORT_SYMBOL(d_genocide);
2455
1da177e4
LT
2456void __init vfs_caches_init_early(void)
2457{
2458 dcache_init_early();
2459 inode_init_early();
2460}
2461
2462void __init vfs_caches_init(unsigned long mempages)
2463{
2464 unsigned long reserve;
2465
2466 /* Base hash sizes on available memory, with a reserve equal to
2467 150% of current kernel size */
2468
2469 reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1);
2470 mempages -= reserve;
2471
2472 names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0,
20c2df83 2473 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1da177e4 2474
74bf17cf
DC
2475 dcache_init();
2476 inode_init();
1da177e4 2477 files_init(mempages);
74bf17cf 2478 mnt_init();
1da177e4
LT
2479 bdev_cache_init();
2480 chrdev_init();
2481}