Commit | Line | Data |
---|---|---|
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> | |
24 | #include <linux/smp_lock.h> | |
25 | #include <linux/hash.h> | |
26 | #include <linux/cache.h> | |
27 | #include <linux/module.h> | |
28 | #include <linux/mount.h> | |
29 | #include <linux/file.h> | |
30 | #include <asm/uaccess.h> | |
31 | #include <linux/security.h> | |
32 | #include <linux/seqlock.h> | |
33 | #include <linux/swap.h> | |
34 | #include <linux/bootmem.h> | |
35 | ||
1da177e4 | 36 | |
fa3536cc | 37 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
38 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
39 | ||
40 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock); | |
e4d91918 | 41 | static __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 LT |
42 | |
43 | EXPORT_SYMBOL(dcache_lock); | |
44 | ||
fa3536cc | 45 | static kmem_cache_t *dentry_cache __read_mostly; |
1da177e4 LT |
46 | |
47 | #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname)) | |
48 | ||
49 | /* | |
50 | * This is the single most critical data structure when it comes | |
51 | * to the dcache: the hashtable for lookups. Somebody should try | |
52 | * to make this good - I've just made it work. | |
53 | * | |
54 | * This hash-function tries to avoid losing too many bits of hash | |
55 | * information, yet avoid using a prime hash-size or similar. | |
56 | */ | |
57 | #define D_HASHBITS d_hash_shift | |
58 | #define D_HASHMASK d_hash_mask | |
59 | ||
fa3536cc ED |
60 | static unsigned int d_hash_mask __read_mostly; |
61 | static unsigned int d_hash_shift __read_mostly; | |
62 | static struct hlist_head *dentry_hashtable __read_mostly; | |
1da177e4 LT |
63 | static LIST_HEAD(dentry_unused); |
64 | ||
65 | /* Statistics gathering. */ | |
66 | struct dentry_stat_t dentry_stat = { | |
67 | .age_limit = 45, | |
68 | }; | |
69 | ||
70 | static void d_callback(struct rcu_head *head) | |
71 | { | |
5160ee6f | 72 | struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu); |
1da177e4 LT |
73 | |
74 | if (dname_external(dentry)) | |
75 | kfree(dentry->d_name.name); | |
76 | kmem_cache_free(dentry_cache, dentry); | |
77 | } | |
78 | ||
79 | /* | |
80 | * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry | |
81 | * inside dcache_lock. | |
82 | */ | |
83 | static void d_free(struct dentry *dentry) | |
84 | { | |
85 | if (dentry->d_op && dentry->d_op->d_release) | |
86 | dentry->d_op->d_release(dentry); | |
5160ee6f | 87 | call_rcu(&dentry->d_u.d_rcu, d_callback); |
1da177e4 LT |
88 | } |
89 | ||
90 | /* | |
91 | * Release the dentry's inode, using the filesystem | |
92 | * d_iput() operation if defined. | |
93 | * Called with dcache_lock and per dentry lock held, drops both. | |
94 | */ | |
858119e1 | 95 | static void dentry_iput(struct dentry * dentry) |
1da177e4 LT |
96 | { |
97 | struct inode *inode = dentry->d_inode; | |
98 | if (inode) { | |
99 | dentry->d_inode = NULL; | |
100 | list_del_init(&dentry->d_alias); | |
101 | spin_unlock(&dentry->d_lock); | |
102 | spin_unlock(&dcache_lock); | |
f805fbda LT |
103 | if (!inode->i_nlink) |
104 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
105 | if (dentry->d_op && dentry->d_op->d_iput) |
106 | dentry->d_op->d_iput(dentry, inode); | |
107 | else | |
108 | iput(inode); | |
109 | } else { | |
110 | spin_unlock(&dentry->d_lock); | |
111 | spin_unlock(&dcache_lock); | |
112 | } | |
113 | } | |
114 | ||
115 | /* | |
116 | * This is dput | |
117 | * | |
118 | * This is complicated by the fact that we do not want to put | |
119 | * dentries that are no longer on any hash chain on the unused | |
120 | * list: we'd much rather just get rid of them immediately. | |
121 | * | |
122 | * However, that implies that we have to traverse the dentry | |
123 | * tree upwards to the parents which might _also_ now be | |
124 | * scheduled for deletion (it may have been only waiting for | |
125 | * its last child to go away). | |
126 | * | |
127 | * This tail recursion is done by hand as we don't want to depend | |
128 | * on the compiler to always get this right (gcc generally doesn't). | |
129 | * Real recursion would eat up our stack space. | |
130 | */ | |
131 | ||
132 | /* | |
133 | * dput - release a dentry | |
134 | * @dentry: dentry to release | |
135 | * | |
136 | * Release a dentry. This will drop the usage count and if appropriate | |
137 | * call the dentry unlink method as well as removing it from the queues and | |
138 | * releasing its resources. If the parent dentries were scheduled for release | |
139 | * they too may now get deleted. | |
140 | * | |
141 | * no dcache lock, please. | |
142 | */ | |
143 | ||
144 | void dput(struct dentry *dentry) | |
145 | { | |
146 | if (!dentry) | |
147 | return; | |
148 | ||
149 | repeat: | |
150 | if (atomic_read(&dentry->d_count) == 1) | |
151 | might_sleep(); | |
152 | if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock)) | |
153 | return; | |
154 | ||
155 | spin_lock(&dentry->d_lock); | |
156 | if (atomic_read(&dentry->d_count)) { | |
157 | spin_unlock(&dentry->d_lock); | |
158 | spin_unlock(&dcache_lock); | |
159 | return; | |
160 | } | |
161 | ||
162 | /* | |
163 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
164 | */ | |
165 | if (dentry->d_op && dentry->d_op->d_delete) { | |
166 | if (dentry->d_op->d_delete(dentry)) | |
167 | goto unhash_it; | |
168 | } | |
169 | /* Unreachable? Get rid of it */ | |
170 | if (d_unhashed(dentry)) | |
171 | goto kill_it; | |
172 | if (list_empty(&dentry->d_lru)) { | |
173 | dentry->d_flags |= DCACHE_REFERENCED; | |
174 | list_add(&dentry->d_lru, &dentry_unused); | |
175 | dentry_stat.nr_unused++; | |
176 | } | |
177 | spin_unlock(&dentry->d_lock); | |
178 | spin_unlock(&dcache_lock); | |
179 | return; | |
180 | ||
181 | unhash_it: | |
182 | __d_drop(dentry); | |
183 | ||
184 | kill_it: { | |
185 | struct dentry *parent; | |
186 | ||
187 | /* If dentry was on d_lru list | |
188 | * delete it from there | |
189 | */ | |
190 | if (!list_empty(&dentry->d_lru)) { | |
191 | list_del(&dentry->d_lru); | |
192 | dentry_stat.nr_unused--; | |
193 | } | |
5160ee6f | 194 | list_del(&dentry->d_u.d_child); |
1da177e4 LT |
195 | dentry_stat.nr_dentry--; /* For d_free, below */ |
196 | /*drops the locks, at that point nobody can reach this dentry */ | |
197 | dentry_iput(dentry); | |
198 | parent = dentry->d_parent; | |
199 | d_free(dentry); | |
200 | if (dentry == parent) | |
201 | return; | |
202 | dentry = parent; | |
203 | goto repeat; | |
204 | } | |
205 | } | |
206 | ||
207 | /** | |
208 | * d_invalidate - invalidate a dentry | |
209 | * @dentry: dentry to invalidate | |
210 | * | |
211 | * Try to invalidate the dentry if it turns out to be | |
212 | * possible. If there are other dentries that can be | |
213 | * reached through this one we can't delete it and we | |
214 | * return -EBUSY. On success we return 0. | |
215 | * | |
216 | * no dcache lock. | |
217 | */ | |
218 | ||
219 | int d_invalidate(struct dentry * dentry) | |
220 | { | |
221 | /* | |
222 | * If it's already been dropped, return OK. | |
223 | */ | |
224 | spin_lock(&dcache_lock); | |
225 | if (d_unhashed(dentry)) { | |
226 | spin_unlock(&dcache_lock); | |
227 | return 0; | |
228 | } | |
229 | /* | |
230 | * Check whether to do a partial shrink_dcache | |
231 | * to get rid of unused child entries. | |
232 | */ | |
233 | if (!list_empty(&dentry->d_subdirs)) { | |
234 | spin_unlock(&dcache_lock); | |
235 | shrink_dcache_parent(dentry); | |
236 | spin_lock(&dcache_lock); | |
237 | } | |
238 | ||
239 | /* | |
240 | * Somebody else still using it? | |
241 | * | |
242 | * If it's a directory, we can't drop it | |
243 | * for fear of somebody re-populating it | |
244 | * with children (even though dropping it | |
245 | * would make it unreachable from the root, | |
246 | * we might still populate it if it was a | |
247 | * working directory or similar). | |
248 | */ | |
249 | spin_lock(&dentry->d_lock); | |
250 | if (atomic_read(&dentry->d_count) > 1) { | |
251 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { | |
252 | spin_unlock(&dentry->d_lock); | |
253 | spin_unlock(&dcache_lock); | |
254 | return -EBUSY; | |
255 | } | |
256 | } | |
257 | ||
258 | __d_drop(dentry); | |
259 | spin_unlock(&dentry->d_lock); | |
260 | spin_unlock(&dcache_lock); | |
261 | return 0; | |
262 | } | |
263 | ||
264 | /* This should be called _only_ with dcache_lock held */ | |
265 | ||
266 | static inline struct dentry * __dget_locked(struct dentry *dentry) | |
267 | { | |
268 | atomic_inc(&dentry->d_count); | |
269 | if (!list_empty(&dentry->d_lru)) { | |
270 | dentry_stat.nr_unused--; | |
271 | list_del_init(&dentry->d_lru); | |
272 | } | |
273 | return dentry; | |
274 | } | |
275 | ||
276 | struct dentry * dget_locked(struct dentry *dentry) | |
277 | { | |
278 | return __dget_locked(dentry); | |
279 | } | |
280 | ||
281 | /** | |
282 | * d_find_alias - grab a hashed alias of inode | |
283 | * @inode: inode in question | |
284 | * @want_discon: flag, used by d_splice_alias, to request | |
285 | * that only a DISCONNECTED alias be returned. | |
286 | * | |
287 | * If inode has a hashed alias, or is a directory and has any alias, | |
288 | * acquire the reference to alias and return it. Otherwise return NULL. | |
289 | * Notice that if inode is a directory there can be only one alias and | |
290 | * it can be unhashed only if it has no children, or if it is the root | |
291 | * of a filesystem. | |
292 | * | |
293 | * If the inode has a DCACHE_DISCONNECTED alias, then prefer | |
294 | * any other hashed alias over that one unless @want_discon is set, | |
295 | * in which case only return a DCACHE_DISCONNECTED alias. | |
296 | */ | |
297 | ||
298 | static struct dentry * __d_find_alias(struct inode *inode, int want_discon) | |
299 | { | |
300 | struct list_head *head, *next, *tmp; | |
301 | struct dentry *alias, *discon_alias=NULL; | |
302 | ||
303 | head = &inode->i_dentry; | |
304 | next = inode->i_dentry.next; | |
305 | while (next != head) { | |
306 | tmp = next; | |
307 | next = tmp->next; | |
308 | prefetch(next); | |
309 | alias = list_entry(tmp, struct dentry, d_alias); | |
310 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
311 | if (alias->d_flags & DCACHE_DISCONNECTED) | |
312 | discon_alias = alias; | |
313 | else if (!want_discon) { | |
314 | __dget_locked(alias); | |
315 | return alias; | |
316 | } | |
317 | } | |
318 | } | |
319 | if (discon_alias) | |
320 | __dget_locked(discon_alias); | |
321 | return discon_alias; | |
322 | } | |
323 | ||
324 | struct dentry * d_find_alias(struct inode *inode) | |
325 | { | |
214fda1f DH |
326 | struct dentry *de = NULL; |
327 | ||
328 | if (!list_empty(&inode->i_dentry)) { | |
329 | spin_lock(&dcache_lock); | |
330 | de = __d_find_alias(inode, 0); | |
331 | spin_unlock(&dcache_lock); | |
332 | } | |
1da177e4 LT |
333 | return de; |
334 | } | |
335 | ||
336 | /* | |
337 | * Try to kill dentries associated with this inode. | |
338 | * WARNING: you must own a reference to inode. | |
339 | */ | |
340 | void d_prune_aliases(struct inode *inode) | |
341 | { | |
0cdca3f9 | 342 | struct dentry *dentry; |
1da177e4 LT |
343 | restart: |
344 | spin_lock(&dcache_lock); | |
0cdca3f9 | 345 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 LT |
346 | spin_lock(&dentry->d_lock); |
347 | if (!atomic_read(&dentry->d_count)) { | |
348 | __dget_locked(dentry); | |
349 | __d_drop(dentry); | |
350 | spin_unlock(&dentry->d_lock); | |
351 | spin_unlock(&dcache_lock); | |
352 | dput(dentry); | |
353 | goto restart; | |
354 | } | |
355 | spin_unlock(&dentry->d_lock); | |
356 | } | |
357 | spin_unlock(&dcache_lock); | |
358 | } | |
359 | ||
360 | /* | |
d702ccb3 AM |
361 | * Throw away a dentry - free the inode, dput the parent. This requires that |
362 | * the LRU list has already been removed. | |
363 | * | |
1da177e4 | 364 | * Called with dcache_lock, drops it and then regains. |
d702ccb3 | 365 | * Called with dentry->d_lock held, drops it. |
1da177e4 | 366 | */ |
d702ccb3 | 367 | static void prune_one_dentry(struct dentry * dentry) |
1da177e4 LT |
368 | { |
369 | struct dentry * parent; | |
370 | ||
371 | __d_drop(dentry); | |
5160ee6f | 372 | list_del(&dentry->d_u.d_child); |
1da177e4 LT |
373 | dentry_stat.nr_dentry--; /* For d_free, below */ |
374 | dentry_iput(dentry); | |
375 | parent = dentry->d_parent; | |
376 | d_free(dentry); | |
377 | if (parent != dentry) | |
378 | dput(parent); | |
379 | spin_lock(&dcache_lock); | |
380 | } | |
381 | ||
382 | /** | |
383 | * prune_dcache - shrink the dcache | |
384 | * @count: number of entries to try and free | |
0feae5c4 N |
385 | * @sb: if given, ignore dentries for other superblocks |
386 | * which are being unmounted. | |
1da177e4 LT |
387 | * |
388 | * Shrink the dcache. This is done when we need | |
389 | * more memory, or simply when we need to unmount | |
390 | * something (at which point we need to unuse | |
391 | * all dentries). | |
392 | * | |
393 | * This function may fail to free any resources if | |
394 | * all the dentries are in use. | |
395 | */ | |
396 | ||
0feae5c4 | 397 | static void prune_dcache(int count, struct super_block *sb) |
1da177e4 LT |
398 | { |
399 | spin_lock(&dcache_lock); | |
400 | for (; count ; count--) { | |
401 | struct dentry *dentry; | |
402 | struct list_head *tmp; | |
0feae5c4 | 403 | struct rw_semaphore *s_umount; |
1da177e4 LT |
404 | |
405 | cond_resched_lock(&dcache_lock); | |
406 | ||
407 | tmp = dentry_unused.prev; | |
f58a1ebb | 408 | if (sb) { |
0feae5c4 N |
409 | /* Try to find a dentry for this sb, but don't try |
410 | * too hard, if they aren't near the tail they will | |
411 | * be moved down again soon | |
412 | */ | |
413 | int skip = count; | |
414 | while (skip && tmp != &dentry_unused && | |
415 | list_entry(tmp, struct dentry, d_lru)->d_sb != sb) { | |
416 | skip--; | |
417 | tmp = tmp->prev; | |
418 | } | |
419 | } | |
1da177e4 LT |
420 | if (tmp == &dentry_unused) |
421 | break; | |
422 | list_del_init(tmp); | |
423 | prefetch(dentry_unused.prev); | |
424 | dentry_stat.nr_unused--; | |
425 | dentry = list_entry(tmp, struct dentry, d_lru); | |
426 | ||
427 | spin_lock(&dentry->d_lock); | |
428 | /* | |
429 | * We found an inuse dentry which was not removed from | |
430 | * dentry_unused because of laziness during lookup. Do not free | |
431 | * it - just keep it off the dentry_unused list. | |
432 | */ | |
433 | if (atomic_read(&dentry->d_count)) { | |
434 | spin_unlock(&dentry->d_lock); | |
435 | continue; | |
436 | } | |
437 | /* If the dentry was recently referenced, don't free it. */ | |
438 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
439 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
440 | list_add(&dentry->d_lru, &dentry_unused); | |
441 | dentry_stat.nr_unused++; | |
442 | spin_unlock(&dentry->d_lock); | |
443 | continue; | |
444 | } | |
0feae5c4 N |
445 | /* |
446 | * If the dentry is not DCACHED_REFERENCED, it is time | |
447 | * to remove it from the dcache, provided the super block is | |
448 | * NULL (which means we are trying to reclaim memory) | |
449 | * or this dentry belongs to the same super block that | |
450 | * we want to shrink. | |
451 | */ | |
452 | /* | |
453 | * If this dentry is for "my" filesystem, then I can prune it | |
454 | * without taking the s_umount lock (I already hold it). | |
455 | */ | |
456 | if (sb && dentry->d_sb == sb) { | |
457 | prune_one_dentry(dentry); | |
458 | continue; | |
459 | } | |
460 | /* | |
461 | * ...otherwise we need to be sure this filesystem isn't being | |
462 | * unmounted, otherwise we could race with | |
463 | * generic_shutdown_super(), and end up holding a reference to | |
464 | * an inode while the filesystem is unmounted. | |
465 | * So we try to get s_umount, and make sure s_root isn't NULL. | |
466 | * (Take a local copy of s_umount to avoid a use-after-free of | |
467 | * `dentry'). | |
468 | */ | |
469 | s_umount = &dentry->d_sb->s_umount; | |
470 | if (down_read_trylock(s_umount)) { | |
471 | if (dentry->d_sb->s_root != NULL) { | |
472 | prune_one_dentry(dentry); | |
473 | up_read(s_umount); | |
474 | continue; | |
475 | } | |
476 | up_read(s_umount); | |
477 | } | |
478 | spin_unlock(&dentry->d_lock); | |
479 | /* Cannot remove the first dentry, and it isn't appropriate | |
480 | * to move it to the head of the list, so give up, and try | |
481 | * later | |
482 | */ | |
483 | break; | |
1da177e4 LT |
484 | } |
485 | spin_unlock(&dcache_lock); | |
486 | } | |
487 | ||
488 | /* | |
489 | * Shrink the dcache for the specified super block. | |
490 | * This allows us to unmount a device without disturbing | |
491 | * the dcache for the other devices. | |
492 | * | |
493 | * This implementation makes just two traversals of the | |
494 | * unused list. On the first pass we move the selected | |
495 | * dentries to the most recent end, and on the second | |
496 | * pass we free them. The second pass must restart after | |
497 | * each dput(), but since the target dentries are all at | |
498 | * the end, it's really just a single traversal. | |
499 | */ | |
500 | ||
501 | /** | |
502 | * shrink_dcache_sb - shrink dcache for a superblock | |
503 | * @sb: superblock | |
504 | * | |
505 | * Shrink the dcache for the specified super block. This | |
506 | * is used to free the dcache before unmounting a file | |
507 | * system | |
508 | */ | |
509 | ||
510 | void shrink_dcache_sb(struct super_block * sb) | |
511 | { | |
512 | struct list_head *tmp, *next; | |
513 | struct dentry *dentry; | |
514 | ||
515 | /* | |
516 | * Pass one ... move the dentries for the specified | |
517 | * superblock to the most recent end of the unused list. | |
518 | */ | |
519 | spin_lock(&dcache_lock); | |
0cdca3f9 | 520 | list_for_each_safe(tmp, next, &dentry_unused) { |
1da177e4 LT |
521 | dentry = list_entry(tmp, struct dentry, d_lru); |
522 | if (dentry->d_sb != sb) | |
523 | continue; | |
1bfba4e8 | 524 | list_move(tmp, &dentry_unused); |
1da177e4 LT |
525 | } |
526 | ||
527 | /* | |
528 | * Pass two ... free the dentries for this superblock. | |
529 | */ | |
530 | repeat: | |
0cdca3f9 | 531 | list_for_each_safe(tmp, next, &dentry_unused) { |
1da177e4 LT |
532 | dentry = list_entry(tmp, struct dentry, d_lru); |
533 | if (dentry->d_sb != sb) | |
534 | continue; | |
535 | dentry_stat.nr_unused--; | |
536 | list_del_init(tmp); | |
537 | spin_lock(&dentry->d_lock); | |
538 | if (atomic_read(&dentry->d_count)) { | |
539 | spin_unlock(&dentry->d_lock); | |
540 | continue; | |
541 | } | |
542 | prune_one_dentry(dentry); | |
2ab13460 | 543 | cond_resched_lock(&dcache_lock); |
1da177e4 LT |
544 | goto repeat; |
545 | } | |
546 | spin_unlock(&dcache_lock); | |
547 | } | |
548 | ||
549 | /* | |
550 | * Search for at least 1 mount point in the dentry's subdirs. | |
551 | * We descend to the next level whenever the d_subdirs | |
552 | * list is non-empty and continue searching. | |
553 | */ | |
554 | ||
555 | /** | |
556 | * have_submounts - check for mounts over a dentry | |
557 | * @parent: dentry to check. | |
558 | * | |
559 | * Return true if the parent or its subdirectories contain | |
560 | * a mount point | |
561 | */ | |
562 | ||
563 | int have_submounts(struct dentry *parent) | |
564 | { | |
565 | struct dentry *this_parent = parent; | |
566 | struct list_head *next; | |
567 | ||
568 | spin_lock(&dcache_lock); | |
569 | if (d_mountpoint(parent)) | |
570 | goto positive; | |
571 | repeat: | |
572 | next = this_parent->d_subdirs.next; | |
573 | resume: | |
574 | while (next != &this_parent->d_subdirs) { | |
575 | struct list_head *tmp = next; | |
5160ee6f | 576 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
577 | next = tmp->next; |
578 | /* Have we found a mount point ? */ | |
579 | if (d_mountpoint(dentry)) | |
580 | goto positive; | |
581 | if (!list_empty(&dentry->d_subdirs)) { | |
582 | this_parent = dentry; | |
583 | goto repeat; | |
584 | } | |
585 | } | |
586 | /* | |
587 | * All done at this level ... ascend and resume the search. | |
588 | */ | |
589 | if (this_parent != parent) { | |
5160ee6f | 590 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
591 | this_parent = this_parent->d_parent; |
592 | goto resume; | |
593 | } | |
594 | spin_unlock(&dcache_lock); | |
595 | return 0; /* No mount points found in tree */ | |
596 | positive: | |
597 | spin_unlock(&dcache_lock); | |
598 | return 1; | |
599 | } | |
600 | ||
601 | /* | |
602 | * Search the dentry child list for the specified parent, | |
603 | * and move any unused dentries to the end of the unused | |
604 | * list for prune_dcache(). We descend to the next level | |
605 | * whenever the d_subdirs list is non-empty and continue | |
606 | * searching. | |
607 | * | |
608 | * It returns zero iff there are no unused children, | |
609 | * otherwise it returns the number of children moved to | |
610 | * the end of the unused list. This may not be the total | |
611 | * number of unused children, because select_parent can | |
612 | * drop the lock and return early due to latency | |
613 | * constraints. | |
614 | */ | |
615 | static int select_parent(struct dentry * parent) | |
616 | { | |
617 | struct dentry *this_parent = parent; | |
618 | struct list_head *next; | |
619 | int found = 0; | |
620 | ||
621 | spin_lock(&dcache_lock); | |
622 | repeat: | |
623 | next = this_parent->d_subdirs.next; | |
624 | resume: | |
625 | while (next != &this_parent->d_subdirs) { | |
626 | struct list_head *tmp = next; | |
5160ee6f | 627 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
628 | next = tmp->next; |
629 | ||
630 | if (!list_empty(&dentry->d_lru)) { | |
631 | dentry_stat.nr_unused--; | |
632 | list_del_init(&dentry->d_lru); | |
633 | } | |
634 | /* | |
635 | * move only zero ref count dentries to the end | |
636 | * of the unused list for prune_dcache | |
637 | */ | |
638 | if (!atomic_read(&dentry->d_count)) { | |
8e13059a | 639 | list_add_tail(&dentry->d_lru, &dentry_unused); |
1da177e4 LT |
640 | dentry_stat.nr_unused++; |
641 | found++; | |
642 | } | |
643 | ||
644 | /* | |
645 | * We can return to the caller if we have found some (this | |
646 | * ensures forward progress). We'll be coming back to find | |
647 | * the rest. | |
648 | */ | |
649 | if (found && need_resched()) | |
650 | goto out; | |
651 | ||
652 | /* | |
653 | * Descend a level if the d_subdirs list is non-empty. | |
654 | */ | |
655 | if (!list_empty(&dentry->d_subdirs)) { | |
656 | this_parent = dentry; | |
1da177e4 LT |
657 | goto repeat; |
658 | } | |
659 | } | |
660 | /* | |
661 | * All done at this level ... ascend and resume the search. | |
662 | */ | |
663 | if (this_parent != parent) { | |
5160ee6f | 664 | next = this_parent->d_u.d_child.next; |
1da177e4 | 665 | this_parent = this_parent->d_parent; |
1da177e4 LT |
666 | goto resume; |
667 | } | |
668 | out: | |
669 | spin_unlock(&dcache_lock); | |
670 | return found; | |
671 | } | |
672 | ||
673 | /** | |
674 | * shrink_dcache_parent - prune dcache | |
675 | * @parent: parent of entries to prune | |
676 | * | |
677 | * Prune the dcache to remove unused children of the parent dentry. | |
678 | */ | |
679 | ||
680 | void shrink_dcache_parent(struct dentry * parent) | |
681 | { | |
682 | int found; | |
683 | ||
684 | while ((found = select_parent(parent)) != 0) | |
0feae5c4 | 685 | prune_dcache(found, parent->d_sb); |
1da177e4 LT |
686 | } |
687 | ||
1da177e4 LT |
688 | /* |
689 | * Scan `nr' dentries and return the number which remain. | |
690 | * | |
691 | * We need to avoid reentering the filesystem if the caller is performing a | |
692 | * GFP_NOFS allocation attempt. One example deadlock is: | |
693 | * | |
694 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
695 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
696 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
697 | * | |
698 | * In this case we return -1 to tell the caller that we baled. | |
699 | */ | |
27496a8c | 700 | static int shrink_dcache_memory(int nr, gfp_t gfp_mask) |
1da177e4 LT |
701 | { |
702 | if (nr) { | |
703 | if (!(gfp_mask & __GFP_FS)) | |
704 | return -1; | |
0feae5c4 | 705 | prune_dcache(nr, NULL); |
1da177e4 LT |
706 | } |
707 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
708 | } | |
709 | ||
710 | /** | |
711 | * d_alloc - allocate a dcache entry | |
712 | * @parent: parent of entry to allocate | |
713 | * @name: qstr of the name | |
714 | * | |
715 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
716 | * available. On a success the dentry is returned. The name passed in is | |
717 | * copied and the copy passed in may be reused after this call. | |
718 | */ | |
719 | ||
720 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
721 | { | |
722 | struct dentry *dentry; | |
723 | char *dname; | |
724 | ||
725 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); | |
726 | if (!dentry) | |
727 | return NULL; | |
728 | ||
729 | if (name->len > DNAME_INLINE_LEN-1) { | |
730 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
731 | if (!dname) { | |
732 | kmem_cache_free(dentry_cache, dentry); | |
733 | return NULL; | |
734 | } | |
735 | } else { | |
736 | dname = dentry->d_iname; | |
737 | } | |
738 | dentry->d_name.name = dname; | |
739 | ||
740 | dentry->d_name.len = name->len; | |
741 | dentry->d_name.hash = name->hash; | |
742 | memcpy(dname, name->name, name->len); | |
743 | dname[name->len] = 0; | |
744 | ||
745 | atomic_set(&dentry->d_count, 1); | |
746 | dentry->d_flags = DCACHE_UNHASHED; | |
747 | spin_lock_init(&dentry->d_lock); | |
748 | dentry->d_inode = NULL; | |
749 | dentry->d_parent = NULL; | |
750 | dentry->d_sb = NULL; | |
751 | dentry->d_op = NULL; | |
752 | dentry->d_fsdata = NULL; | |
753 | dentry->d_mounted = 0; | |
47ba87e0 | 754 | #ifdef CONFIG_PROFILING |
1da177e4 | 755 | dentry->d_cookie = NULL; |
47ba87e0 | 756 | #endif |
1da177e4 LT |
757 | INIT_HLIST_NODE(&dentry->d_hash); |
758 | INIT_LIST_HEAD(&dentry->d_lru); | |
759 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
760 | INIT_LIST_HEAD(&dentry->d_alias); | |
761 | ||
762 | if (parent) { | |
763 | dentry->d_parent = dget(parent); | |
764 | dentry->d_sb = parent->d_sb; | |
765 | } else { | |
5160ee6f | 766 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
767 | } |
768 | ||
769 | spin_lock(&dcache_lock); | |
770 | if (parent) | |
5160ee6f | 771 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1da177e4 LT |
772 | dentry_stat.nr_dentry++; |
773 | spin_unlock(&dcache_lock); | |
774 | ||
775 | return dentry; | |
776 | } | |
777 | ||
778 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
779 | { | |
780 | struct qstr q; | |
781 | ||
782 | q.name = name; | |
783 | q.len = strlen(name); | |
784 | q.hash = full_name_hash(q.name, q.len); | |
785 | return d_alloc(parent, &q); | |
786 | } | |
787 | ||
788 | /** | |
789 | * d_instantiate - fill in inode information for a dentry | |
790 | * @entry: dentry to complete | |
791 | * @inode: inode to attach to this dentry | |
792 | * | |
793 | * Fill in inode information in the entry. | |
794 | * | |
795 | * This turns negative dentries into productive full members | |
796 | * of society. | |
797 | * | |
798 | * NOTE! This assumes that the inode count has been incremented | |
799 | * (or otherwise set) by the caller to indicate that it is now | |
800 | * in use by the dcache. | |
801 | */ | |
802 | ||
803 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
804 | { | |
28133c7b | 805 | BUG_ON(!list_empty(&entry->d_alias)); |
1da177e4 LT |
806 | spin_lock(&dcache_lock); |
807 | if (inode) | |
808 | list_add(&entry->d_alias, &inode->i_dentry); | |
809 | entry->d_inode = inode; | |
c32ccd87 | 810 | fsnotify_d_instantiate(entry, inode); |
1da177e4 LT |
811 | spin_unlock(&dcache_lock); |
812 | security_d_instantiate(entry, inode); | |
813 | } | |
814 | ||
815 | /** | |
816 | * d_instantiate_unique - instantiate a non-aliased dentry | |
817 | * @entry: dentry to instantiate | |
818 | * @inode: inode to attach to this dentry | |
819 | * | |
820 | * Fill in inode information in the entry. On success, it returns NULL. | |
821 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 822 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
823 | * |
824 | * Note that in order to avoid conflicts with rename() etc, the caller | |
825 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
826 | * |
827 | * This also assumes that the inode count has been incremented | |
828 | * (or otherwise set) by the caller to indicate that it is now | |
829 | * in use by the dcache. | |
1da177e4 | 830 | */ |
770bfad8 DH |
831 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
832 | struct inode *inode) | |
1da177e4 LT |
833 | { |
834 | struct dentry *alias; | |
835 | int len = entry->d_name.len; | |
836 | const char *name = entry->d_name.name; | |
837 | unsigned int hash = entry->d_name.hash; | |
838 | ||
770bfad8 DH |
839 | if (!inode) { |
840 | entry->d_inode = NULL; | |
841 | return NULL; | |
842 | } | |
843 | ||
1da177e4 LT |
844 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
845 | struct qstr *qstr = &alias->d_name; | |
846 | ||
847 | if (qstr->hash != hash) | |
848 | continue; | |
849 | if (alias->d_parent != entry->d_parent) | |
850 | continue; | |
851 | if (qstr->len != len) | |
852 | continue; | |
853 | if (memcmp(qstr->name, name, len)) | |
854 | continue; | |
855 | dget_locked(alias); | |
1da177e4 LT |
856 | return alias; |
857 | } | |
770bfad8 | 858 | |
1da177e4 | 859 | list_add(&entry->d_alias, &inode->i_dentry); |
1da177e4 | 860 | entry->d_inode = inode; |
c32ccd87 | 861 | fsnotify_d_instantiate(entry, inode); |
1da177e4 LT |
862 | return NULL; |
863 | } | |
770bfad8 DH |
864 | |
865 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
866 | { | |
867 | struct dentry *result; | |
868 | ||
869 | BUG_ON(!list_empty(&entry->d_alias)); | |
870 | ||
871 | spin_lock(&dcache_lock); | |
872 | result = __d_instantiate_unique(entry, inode); | |
873 | spin_unlock(&dcache_lock); | |
874 | ||
875 | if (!result) { | |
876 | security_d_instantiate(entry, inode); | |
877 | return NULL; | |
878 | } | |
879 | ||
880 | BUG_ON(!d_unhashed(result)); | |
881 | iput(inode); | |
882 | return result; | |
883 | } | |
884 | ||
1da177e4 LT |
885 | EXPORT_SYMBOL(d_instantiate_unique); |
886 | ||
887 | /** | |
888 | * d_alloc_root - allocate root dentry | |
889 | * @root_inode: inode to allocate the root for | |
890 | * | |
891 | * Allocate a root ("/") dentry for the inode given. The inode is | |
892 | * instantiated and returned. %NULL is returned if there is insufficient | |
893 | * memory or the inode passed is %NULL. | |
894 | */ | |
895 | ||
896 | struct dentry * d_alloc_root(struct inode * root_inode) | |
897 | { | |
898 | struct dentry *res = NULL; | |
899 | ||
900 | if (root_inode) { | |
901 | static const struct qstr name = { .name = "/", .len = 1 }; | |
902 | ||
903 | res = d_alloc(NULL, &name); | |
904 | if (res) { | |
905 | res->d_sb = root_inode->i_sb; | |
906 | res->d_parent = res; | |
907 | d_instantiate(res, root_inode); | |
908 | } | |
909 | } | |
910 | return res; | |
911 | } | |
912 | ||
913 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
914 | unsigned long hash) | |
915 | { | |
916 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
917 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
918 | return dentry_hashtable + (hash & D_HASHMASK); | |
919 | } | |
920 | ||
921 | /** | |
922 | * d_alloc_anon - allocate an anonymous dentry | |
923 | * @inode: inode to allocate the dentry for | |
924 | * | |
925 | * This is similar to d_alloc_root. It is used by filesystems when | |
926 | * creating a dentry for a given inode, often in the process of | |
927 | * mapping a filehandle to a dentry. The returned dentry may be | |
928 | * anonymous, or may have a full name (if the inode was already | |
929 | * in the cache). The file system may need to make further | |
930 | * efforts to connect this dentry into the dcache properly. | |
931 | * | |
932 | * When called on a directory inode, we must ensure that | |
933 | * the inode only ever has one dentry. If a dentry is | |
934 | * found, that is returned instead of allocating a new one. | |
935 | * | |
936 | * On successful return, the reference to the inode has been transferred | |
937 | * to the dentry. If %NULL is returned (indicating kmalloc failure), | |
938 | * the reference on the inode has not been released. | |
939 | */ | |
940 | ||
941 | struct dentry * d_alloc_anon(struct inode *inode) | |
942 | { | |
943 | static const struct qstr anonstring = { .name = "" }; | |
944 | struct dentry *tmp; | |
945 | struct dentry *res; | |
946 | ||
947 | if ((res = d_find_alias(inode))) { | |
948 | iput(inode); | |
949 | return res; | |
950 | } | |
951 | ||
952 | tmp = d_alloc(NULL, &anonstring); | |
953 | if (!tmp) | |
954 | return NULL; | |
955 | ||
956 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ | |
957 | ||
958 | spin_lock(&dcache_lock); | |
959 | res = __d_find_alias(inode, 0); | |
960 | if (!res) { | |
961 | /* attach a disconnected dentry */ | |
962 | res = tmp; | |
963 | tmp = NULL; | |
964 | spin_lock(&res->d_lock); | |
965 | res->d_sb = inode->i_sb; | |
966 | res->d_parent = res; | |
967 | res->d_inode = inode; | |
968 | res->d_flags |= DCACHE_DISCONNECTED; | |
969 | res->d_flags &= ~DCACHE_UNHASHED; | |
970 | list_add(&res->d_alias, &inode->i_dentry); | |
971 | hlist_add_head(&res->d_hash, &inode->i_sb->s_anon); | |
972 | spin_unlock(&res->d_lock); | |
973 | ||
974 | inode = NULL; /* don't drop reference */ | |
975 | } | |
976 | spin_unlock(&dcache_lock); | |
977 | ||
978 | if (inode) | |
979 | iput(inode); | |
980 | if (tmp) | |
981 | dput(tmp); | |
982 | return res; | |
983 | } | |
984 | ||
985 | ||
986 | /** | |
987 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
988 | * @inode: the inode which may have a disconnected dentry | |
989 | * @dentry: a negative dentry which we want to point to the inode. | |
990 | * | |
991 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
992 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
993 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
994 | * | |
995 | * This is needed in the lookup routine of any filesystem that is exportable | |
996 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
997 | * | |
998 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
999 | * is returned. This matches the expected return value of ->lookup. | |
1000 | * | |
1001 | */ | |
1002 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1003 | { | |
1004 | struct dentry *new = NULL; | |
1005 | ||
1006 | if (inode) { | |
1007 | spin_lock(&dcache_lock); | |
1008 | new = __d_find_alias(inode, 1); | |
1009 | if (new) { | |
1010 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
c32ccd87 | 1011 | fsnotify_d_instantiate(new, inode); |
1da177e4 LT |
1012 | spin_unlock(&dcache_lock); |
1013 | security_d_instantiate(new, inode); | |
1014 | d_rehash(dentry); | |
1015 | d_move(new, dentry); | |
1016 | iput(inode); | |
1017 | } else { | |
1018 | /* d_instantiate takes dcache_lock, so we do it by hand */ | |
1019 | list_add(&dentry->d_alias, &inode->i_dentry); | |
1020 | dentry->d_inode = inode; | |
c32ccd87 | 1021 | fsnotify_d_instantiate(dentry, inode); |
1da177e4 LT |
1022 | spin_unlock(&dcache_lock); |
1023 | security_d_instantiate(dentry, inode); | |
1024 | d_rehash(dentry); | |
1025 | } | |
1026 | } else | |
1027 | d_add(dentry, inode); | |
1028 | return new; | |
1029 | } | |
1030 | ||
1031 | ||
1032 | /** | |
1033 | * d_lookup - search for a dentry | |
1034 | * @parent: parent dentry | |
1035 | * @name: qstr of name we wish to find | |
1036 | * | |
1037 | * Searches the children of the parent dentry for the name in question. If | |
1038 | * the dentry is found its reference count is incremented and the dentry | |
1039 | * is returned. The caller must use d_put to free the entry when it has | |
1040 | * finished using it. %NULL is returned on failure. | |
1041 | * | |
1042 | * __d_lookup is dcache_lock free. The hash list is protected using RCU. | |
1043 | * Memory barriers are used while updating and doing lockless traversal. | |
1044 | * To avoid races with d_move while rename is happening, d_lock is used. | |
1045 | * | |
1046 | * Overflows in memcmp(), while d_move, are avoided by keeping the length | |
1047 | * and name pointer in one structure pointed by d_qstr. | |
1048 | * | |
1049 | * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while | |
1050 | * lookup is going on. | |
1051 | * | |
1052 | * dentry_unused list is not updated even if lookup finds the required dentry | |
1053 | * in there. It is updated in places such as prune_dcache, shrink_dcache_sb, | |
1054 | * select_parent and __dget_locked. This laziness saves lookup from dcache_lock | |
1055 | * acquisition. | |
1056 | * | |
1057 | * d_lookup() is protected against the concurrent renames in some unrelated | |
1058 | * directory using the seqlockt_t rename_lock. | |
1059 | */ | |
1060 | ||
1061 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) | |
1062 | { | |
1063 | struct dentry * dentry = NULL; | |
1064 | unsigned long seq; | |
1065 | ||
1066 | do { | |
1067 | seq = read_seqbegin(&rename_lock); | |
1068 | dentry = __d_lookup(parent, name); | |
1069 | if (dentry) | |
1070 | break; | |
1071 | } while (read_seqretry(&rename_lock, seq)); | |
1072 | return dentry; | |
1073 | } | |
1074 | ||
1075 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) | |
1076 | { | |
1077 | unsigned int len = name->len; | |
1078 | unsigned int hash = name->hash; | |
1079 | const unsigned char *str = name->name; | |
1080 | struct hlist_head *head = d_hash(parent,hash); | |
1081 | struct dentry *found = NULL; | |
1082 | struct hlist_node *node; | |
665a7583 | 1083 | struct dentry *dentry; |
1da177e4 LT |
1084 | |
1085 | rcu_read_lock(); | |
1086 | ||
665a7583 | 1087 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1088 | struct qstr *qstr; |
1089 | ||
1da177e4 LT |
1090 | if (dentry->d_name.hash != hash) |
1091 | continue; | |
1092 | if (dentry->d_parent != parent) | |
1093 | continue; | |
1094 | ||
1095 | spin_lock(&dentry->d_lock); | |
1096 | ||
1097 | /* | |
1098 | * Recheck the dentry after taking the lock - d_move may have | |
1099 | * changed things. Don't bother checking the hash because we're | |
1100 | * about to compare the whole name anyway. | |
1101 | */ | |
1102 | if (dentry->d_parent != parent) | |
1103 | goto next; | |
1104 | ||
1105 | /* | |
1106 | * It is safe to compare names since d_move() cannot | |
1107 | * change the qstr (protected by d_lock). | |
1108 | */ | |
1109 | qstr = &dentry->d_name; | |
1110 | if (parent->d_op && parent->d_op->d_compare) { | |
1111 | if (parent->d_op->d_compare(parent, qstr, name)) | |
1112 | goto next; | |
1113 | } else { | |
1114 | if (qstr->len != len) | |
1115 | goto next; | |
1116 | if (memcmp(qstr->name, str, len)) | |
1117 | goto next; | |
1118 | } | |
1119 | ||
1120 | if (!d_unhashed(dentry)) { | |
1121 | atomic_inc(&dentry->d_count); | |
1122 | found = dentry; | |
1123 | } | |
1124 | spin_unlock(&dentry->d_lock); | |
1125 | break; | |
1126 | next: | |
1127 | spin_unlock(&dentry->d_lock); | |
1128 | } | |
1129 | rcu_read_unlock(); | |
1130 | ||
1131 | return found; | |
1132 | } | |
1133 | ||
3e7e241f EB |
1134 | /** |
1135 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1136 | * @dir: Directory to search in | |
1137 | * @name: qstr of name we wish to find | |
1138 | * | |
1139 | * On hash failure or on lookup failure NULL is returned. | |
1140 | */ | |
1141 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1142 | { | |
1143 | struct dentry *dentry = NULL; | |
1144 | ||
1145 | /* | |
1146 | * Check for a fs-specific hash function. Note that we must | |
1147 | * calculate the standard hash first, as the d_op->d_hash() | |
1148 | * routine may choose to leave the hash value unchanged. | |
1149 | */ | |
1150 | name->hash = full_name_hash(name->name, name->len); | |
1151 | if (dir->d_op && dir->d_op->d_hash) { | |
1152 | if (dir->d_op->d_hash(dir, name) < 0) | |
1153 | goto out; | |
1154 | } | |
1155 | dentry = d_lookup(dir, name); | |
1156 | out: | |
1157 | return dentry; | |
1158 | } | |
1159 | ||
1da177e4 LT |
1160 | /** |
1161 | * d_validate - verify dentry provided from insecure source | |
1162 | * @dentry: The dentry alleged to be valid child of @dparent | |
1163 | * @dparent: The parent dentry (known to be valid) | |
1164 | * @hash: Hash of the dentry | |
1165 | * @len: Length of the name | |
1166 | * | |
1167 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1168 | * This is used by ncpfs in its readdir implementation. | |
1169 | * Zero is returned in the dentry is invalid. | |
1170 | */ | |
1171 | ||
1172 | int d_validate(struct dentry *dentry, struct dentry *dparent) | |
1173 | { | |
1174 | struct hlist_head *base; | |
1175 | struct hlist_node *lhp; | |
1176 | ||
1177 | /* Check whether the ptr might be valid at all.. */ | |
1178 | if (!kmem_ptr_validate(dentry_cache, dentry)) | |
1179 | goto out; | |
1180 | ||
1181 | if (dentry->d_parent != dparent) | |
1182 | goto out; | |
1183 | ||
1184 | spin_lock(&dcache_lock); | |
1185 | base = d_hash(dparent, dentry->d_name.hash); | |
1186 | hlist_for_each(lhp,base) { | |
665a7583 | 1187 | /* hlist_for_each_entry_rcu() not required for d_hash list |
1da177e4 LT |
1188 | * as it is parsed under dcache_lock |
1189 | */ | |
1190 | if (dentry == hlist_entry(lhp, struct dentry, d_hash)) { | |
1191 | __dget_locked(dentry); | |
1192 | spin_unlock(&dcache_lock); | |
1193 | return 1; | |
1194 | } | |
1195 | } | |
1196 | spin_unlock(&dcache_lock); | |
1197 | out: | |
1198 | return 0; | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * When a file is deleted, we have two options: | |
1203 | * - turn this dentry into a negative dentry | |
1204 | * - unhash this dentry and free it. | |
1205 | * | |
1206 | * Usually, we want to just turn this into | |
1207 | * a negative dentry, but if anybody else is | |
1208 | * currently using the dentry or the inode | |
1209 | * we can't do that and we fall back on removing | |
1210 | * it from the hash queues and waiting for | |
1211 | * it to be deleted later when it has no users | |
1212 | */ | |
1213 | ||
1214 | /** | |
1215 | * d_delete - delete a dentry | |
1216 | * @dentry: The dentry to delete | |
1217 | * | |
1218 | * Turn the dentry into a negative dentry if possible, otherwise | |
1219 | * remove it from the hash queues so it can be deleted later | |
1220 | */ | |
1221 | ||
1222 | void d_delete(struct dentry * dentry) | |
1223 | { | |
7a91bf7f | 1224 | int isdir = 0; |
1da177e4 LT |
1225 | /* |
1226 | * Are we the only user? | |
1227 | */ | |
1228 | spin_lock(&dcache_lock); | |
1229 | spin_lock(&dentry->d_lock); | |
7a91bf7f | 1230 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
1da177e4 LT |
1231 | if (atomic_read(&dentry->d_count) == 1) { |
1232 | dentry_iput(dentry); | |
7a91bf7f | 1233 | fsnotify_nameremove(dentry, isdir); |
7a2bd3f7 AG |
1234 | |
1235 | /* remove this and other inotify debug checks after 2.6.18 */ | |
1236 | dentry->d_flags &= ~DCACHE_INOTIFY_PARENT_WATCHED; | |
1da177e4 LT |
1237 | return; |
1238 | } | |
1239 | ||
1240 | if (!d_unhashed(dentry)) | |
1241 | __d_drop(dentry); | |
1242 | ||
1243 | spin_unlock(&dentry->d_lock); | |
1244 | spin_unlock(&dcache_lock); | |
7a91bf7f JM |
1245 | |
1246 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 LT |
1247 | } |
1248 | ||
1249 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1250 | { | |
1251 | ||
1252 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1253 | hlist_add_head_rcu(&entry->d_hash, list); | |
1254 | } | |
1255 | ||
770bfad8 DH |
1256 | static void _d_rehash(struct dentry * entry) |
1257 | { | |
1258 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1259 | } | |
1260 | ||
1da177e4 LT |
1261 | /** |
1262 | * d_rehash - add an entry back to the hash | |
1263 | * @entry: dentry to add to the hash | |
1264 | * | |
1265 | * Adds a dentry to the hash according to its name. | |
1266 | */ | |
1267 | ||
1268 | void d_rehash(struct dentry * entry) | |
1269 | { | |
1da177e4 LT |
1270 | spin_lock(&dcache_lock); |
1271 | spin_lock(&entry->d_lock); | |
770bfad8 | 1272 | _d_rehash(entry); |
1da177e4 LT |
1273 | spin_unlock(&entry->d_lock); |
1274 | spin_unlock(&dcache_lock); | |
1275 | } | |
1276 | ||
1277 | #define do_switch(x,y) do { \ | |
1278 | __typeof__ (x) __tmp = x; \ | |
1279 | x = y; y = __tmp; } while (0) | |
1280 | ||
1281 | /* | |
1282 | * When switching names, the actual string doesn't strictly have to | |
1283 | * be preserved in the target - because we're dropping the target | |
1284 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1285 | * the new name before we switch. | |
1286 | * | |
1287 | * Note that we have to be a lot more careful about getting the hash | |
1288 | * switched - we have to switch the hash value properly even if it | |
1289 | * then no longer matches the actual (corrupted) string of the target. | |
1290 | * The hash value has to match the hash queue that the dentry is on.. | |
1291 | */ | |
1292 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
1293 | { | |
1294 | if (dname_external(target)) { | |
1295 | if (dname_external(dentry)) { | |
1296 | /* | |
1297 | * Both external: swap the pointers | |
1298 | */ | |
1299 | do_switch(target->d_name.name, dentry->d_name.name); | |
1300 | } else { | |
1301 | /* | |
1302 | * dentry:internal, target:external. Steal target's | |
1303 | * storage and make target internal. | |
1304 | */ | |
1305 | dentry->d_name.name = target->d_name.name; | |
1306 | target->d_name.name = target->d_iname; | |
1307 | } | |
1308 | } else { | |
1309 | if (dname_external(dentry)) { | |
1310 | /* | |
1311 | * dentry:external, target:internal. Give dentry's | |
1312 | * storage to target and make dentry internal | |
1313 | */ | |
1314 | memcpy(dentry->d_iname, target->d_name.name, | |
1315 | target->d_name.len + 1); | |
1316 | target->d_name.name = dentry->d_name.name; | |
1317 | dentry->d_name.name = dentry->d_iname; | |
1318 | } else { | |
1319 | /* | |
1320 | * Both are internal. Just copy target to dentry | |
1321 | */ | |
1322 | memcpy(dentry->d_iname, target->d_name.name, | |
1323 | target->d_name.len + 1); | |
1324 | } | |
1325 | } | |
1326 | } | |
1327 | ||
1328 | /* | |
1329 | * We cannibalize "target" when moving dentry on top of it, | |
1330 | * because it's going to be thrown away anyway. We could be more | |
1331 | * polite about it, though. | |
1332 | * | |
1333 | * This forceful removal will result in ugly /proc output if | |
1334 | * somebody holds a file open that got deleted due to a rename. | |
1335 | * We could be nicer about the deleted file, and let it show | |
1336 | * up under the name it got deleted rather than the name that | |
1337 | * deleted it. | |
1338 | */ | |
1339 | ||
1340 | /** | |
1341 | * d_move - move a dentry | |
1342 | * @dentry: entry to move | |
1343 | * @target: new dentry | |
1344 | * | |
1345 | * Update the dcache to reflect the move of a file name. Negative | |
1346 | * dcache entries should not be moved in this way. | |
1347 | */ | |
1348 | ||
1349 | void d_move(struct dentry * dentry, struct dentry * target) | |
1350 | { | |
1351 | struct hlist_head *list; | |
1352 | ||
1353 | if (!dentry->d_inode) | |
1354 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1355 | ||
1356 | spin_lock(&dcache_lock); | |
1357 | write_seqlock(&rename_lock); | |
1358 | /* | |
1359 | * XXXX: do we really need to take target->d_lock? | |
1360 | */ | |
1361 | if (target < dentry) { | |
1362 | spin_lock(&target->d_lock); | |
a90b9c05 | 1363 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1364 | } else { |
1365 | spin_lock(&dentry->d_lock); | |
a90b9c05 | 1366 | spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1367 | } |
1368 | ||
1369 | /* Move the dentry to the target hash queue, if on different bucket */ | |
1370 | if (dentry->d_flags & DCACHE_UNHASHED) | |
1371 | goto already_unhashed; | |
1372 | ||
1373 | hlist_del_rcu(&dentry->d_hash); | |
1374 | ||
1375 | already_unhashed: | |
1376 | list = d_hash(target->d_parent, target->d_name.hash); | |
1377 | __d_rehash(dentry, list); | |
1378 | ||
1379 | /* Unhash the target: dput() will then get rid of it */ | |
1380 | __d_drop(target); | |
1381 | ||
5160ee6f ED |
1382 | list_del(&dentry->d_u.d_child); |
1383 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
1384 | |
1385 | /* Switch the names.. */ | |
1386 | switch_names(dentry, target); | |
1387 | do_switch(dentry->d_name.len, target->d_name.len); | |
1388 | do_switch(dentry->d_name.hash, target->d_name.hash); | |
1389 | ||
1390 | /* ... and switch the parents */ | |
1391 | if (IS_ROOT(dentry)) { | |
1392 | dentry->d_parent = target->d_parent; | |
1393 | target->d_parent = target; | |
5160ee6f | 1394 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 LT |
1395 | } else { |
1396 | do_switch(dentry->d_parent, target->d_parent); | |
1397 | ||
1398 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 1399 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
1400 | } |
1401 | ||
5160ee6f | 1402 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 1403 | spin_unlock(&target->d_lock); |
c32ccd87 | 1404 | fsnotify_d_move(dentry); |
1da177e4 LT |
1405 | spin_unlock(&dentry->d_lock); |
1406 | write_sequnlock(&rename_lock); | |
1407 | spin_unlock(&dcache_lock); | |
1408 | } | |
1409 | ||
770bfad8 DH |
1410 | /* |
1411 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
1412 | * named dentry in place of the dentry to be replaced. | |
1413 | */ | |
1414 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
1415 | { | |
1416 | struct dentry *dparent, *aparent; | |
1417 | ||
1418 | switch_names(dentry, anon); | |
1419 | do_switch(dentry->d_name.len, anon->d_name.len); | |
1420 | do_switch(dentry->d_name.hash, anon->d_name.hash); | |
1421 | ||
1422 | dparent = dentry->d_parent; | |
1423 | aparent = anon->d_parent; | |
1424 | ||
1425 | dentry->d_parent = (aparent == anon) ? dentry : aparent; | |
1426 | list_del(&dentry->d_u.d_child); | |
1427 | if (!IS_ROOT(dentry)) | |
1428 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
1429 | else | |
1430 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
1431 | ||
1432 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
1433 | list_del(&anon->d_u.d_child); | |
1434 | if (!IS_ROOT(anon)) | |
1435 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
1436 | else | |
1437 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
1438 | ||
1439 | anon->d_flags &= ~DCACHE_DISCONNECTED; | |
1440 | } | |
1441 | ||
1442 | /** | |
1443 | * d_materialise_unique - introduce an inode into the tree | |
1444 | * @dentry: candidate dentry | |
1445 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
1446 | * | |
1447 | * Introduces an dentry into the tree, substituting an extant disconnected | |
1448 | * root directory alias in its place if there is one | |
1449 | */ | |
1450 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
1451 | { | |
1452 | struct dentry *alias, *actual; | |
1453 | ||
1454 | BUG_ON(!d_unhashed(dentry)); | |
1455 | ||
1456 | spin_lock(&dcache_lock); | |
1457 | ||
1458 | if (!inode) { | |
1459 | actual = dentry; | |
1460 | dentry->d_inode = NULL; | |
1461 | goto found_lock; | |
1462 | } | |
1463 | ||
1464 | /* See if a disconnected directory already exists as an anonymous root | |
1465 | * that we should splice into the tree instead */ | |
1466 | if (S_ISDIR(inode->i_mode) && (alias = __d_find_alias(inode, 1))) { | |
1467 | spin_lock(&alias->d_lock); | |
1468 | ||
1469 | /* Is this a mountpoint that we could splice into our tree? */ | |
1470 | if (IS_ROOT(alias)) | |
1471 | goto connect_mountpoint; | |
1472 | ||
1473 | if (alias->d_name.len == dentry->d_name.len && | |
1474 | alias->d_parent == dentry->d_parent && | |
1475 | memcmp(alias->d_name.name, | |
1476 | dentry->d_name.name, | |
1477 | dentry->d_name.len) == 0) | |
1478 | goto replace_with_alias; | |
1479 | ||
1480 | spin_unlock(&alias->d_lock); | |
1481 | ||
1482 | /* Doh! Seem to be aliasing directories for some reason... */ | |
1483 | dput(alias); | |
1484 | } | |
1485 | ||
1486 | /* Add a unique reference */ | |
1487 | actual = __d_instantiate_unique(dentry, inode); | |
1488 | if (!actual) | |
1489 | actual = dentry; | |
1490 | else if (unlikely(!d_unhashed(actual))) | |
1491 | goto shouldnt_be_hashed; | |
1492 | ||
1493 | found_lock: | |
1494 | spin_lock(&actual->d_lock); | |
1495 | found: | |
1496 | _d_rehash(actual); | |
1497 | spin_unlock(&actual->d_lock); | |
1498 | spin_unlock(&dcache_lock); | |
1499 | ||
1500 | if (actual == dentry) { | |
1501 | security_d_instantiate(dentry, inode); | |
1502 | return NULL; | |
1503 | } | |
1504 | ||
1505 | iput(inode); | |
1506 | return actual; | |
1507 | ||
1508 | /* Convert the anonymous/root alias into an ordinary dentry */ | |
1509 | connect_mountpoint: | |
1510 | __d_materialise_dentry(dentry, alias); | |
1511 | ||
1512 | /* Replace the candidate dentry with the alias in the tree */ | |
1513 | replace_with_alias: | |
1514 | __d_drop(alias); | |
1515 | actual = alias; | |
1516 | goto found; | |
1517 | ||
1518 | shouldnt_be_hashed: | |
1519 | spin_unlock(&dcache_lock); | |
1520 | BUG(); | |
1521 | goto shouldnt_be_hashed; | |
1522 | } | |
1523 | ||
1da177e4 LT |
1524 | /** |
1525 | * d_path - return the path of a dentry | |
1526 | * @dentry: dentry to report | |
1527 | * @vfsmnt: vfsmnt to which the dentry belongs | |
1528 | * @root: root dentry | |
1529 | * @rootmnt: vfsmnt to which the root dentry belongs | |
1530 | * @buffer: buffer to return value in | |
1531 | * @buflen: buffer length | |
1532 | * | |
1533 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
1534 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1535 | * | |
1536 | * Returns the buffer or an error code if the path was too long. | |
1537 | * | |
1538 | * "buflen" should be positive. Caller holds the dcache_lock. | |
1539 | */ | |
1540 | static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt, | |
1541 | struct dentry *root, struct vfsmount *rootmnt, | |
1542 | char *buffer, int buflen) | |
1543 | { | |
1544 | char * end = buffer+buflen; | |
1545 | char * retval; | |
1546 | int namelen; | |
1547 | ||
1548 | *--end = '\0'; | |
1549 | buflen--; | |
1550 | if (!IS_ROOT(dentry) && d_unhashed(dentry)) { | |
1551 | buflen -= 10; | |
1552 | end -= 10; | |
1553 | if (buflen < 0) | |
1554 | goto Elong; | |
1555 | memcpy(end, " (deleted)", 10); | |
1556 | } | |
1557 | ||
1558 | if (buflen < 1) | |
1559 | goto Elong; | |
1560 | /* Get '/' right */ | |
1561 | retval = end-1; | |
1562 | *retval = '/'; | |
1563 | ||
1564 | for (;;) { | |
1565 | struct dentry * parent; | |
1566 | ||
1567 | if (dentry == root && vfsmnt == rootmnt) | |
1568 | break; | |
1569 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { | |
1570 | /* Global root? */ | |
1571 | spin_lock(&vfsmount_lock); | |
1572 | if (vfsmnt->mnt_parent == vfsmnt) { | |
1573 | spin_unlock(&vfsmount_lock); | |
1574 | goto global_root; | |
1575 | } | |
1576 | dentry = vfsmnt->mnt_mountpoint; | |
1577 | vfsmnt = vfsmnt->mnt_parent; | |
1578 | spin_unlock(&vfsmount_lock); | |
1579 | continue; | |
1580 | } | |
1581 | parent = dentry->d_parent; | |
1582 | prefetch(parent); | |
1583 | namelen = dentry->d_name.len; | |
1584 | buflen -= namelen + 1; | |
1585 | if (buflen < 0) | |
1586 | goto Elong; | |
1587 | end -= namelen; | |
1588 | memcpy(end, dentry->d_name.name, namelen); | |
1589 | *--end = '/'; | |
1590 | retval = end; | |
1591 | dentry = parent; | |
1592 | } | |
1593 | ||
1594 | return retval; | |
1595 | ||
1596 | global_root: | |
1597 | namelen = dentry->d_name.len; | |
1598 | buflen -= namelen; | |
1599 | if (buflen < 0) | |
1600 | goto Elong; | |
1601 | retval -= namelen-1; /* hit the slash */ | |
1602 | memcpy(retval, dentry->d_name.name, namelen); | |
1603 | return retval; | |
1604 | Elong: | |
1605 | return ERR_PTR(-ENAMETOOLONG); | |
1606 | } | |
1607 | ||
1608 | /* write full pathname into buffer and return start of pathname */ | |
1609 | char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt, | |
1610 | char *buf, int buflen) | |
1611 | { | |
1612 | char *res; | |
1613 | struct vfsmount *rootmnt; | |
1614 | struct dentry *root; | |
1615 | ||
1616 | read_lock(¤t->fs->lock); | |
1617 | rootmnt = mntget(current->fs->rootmnt); | |
1618 | root = dget(current->fs->root); | |
1619 | read_unlock(¤t->fs->lock); | |
1620 | spin_lock(&dcache_lock); | |
1621 | res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen); | |
1622 | spin_unlock(&dcache_lock); | |
1623 | dput(root); | |
1624 | mntput(rootmnt); | |
1625 | return res; | |
1626 | } | |
1627 | ||
1628 | /* | |
1629 | * NOTE! The user-level library version returns a | |
1630 | * character pointer. The kernel system call just | |
1631 | * returns the length of the buffer filled (which | |
1632 | * includes the ending '\0' character), or a negative | |
1633 | * error value. So libc would do something like | |
1634 | * | |
1635 | * char *getcwd(char * buf, size_t size) | |
1636 | * { | |
1637 | * int retval; | |
1638 | * | |
1639 | * retval = sys_getcwd(buf, size); | |
1640 | * if (retval >= 0) | |
1641 | * return buf; | |
1642 | * errno = -retval; | |
1643 | * return NULL; | |
1644 | * } | |
1645 | */ | |
1646 | asmlinkage long sys_getcwd(char __user *buf, unsigned long size) | |
1647 | { | |
1648 | int error; | |
1649 | struct vfsmount *pwdmnt, *rootmnt; | |
1650 | struct dentry *pwd, *root; | |
1651 | char *page = (char *) __get_free_page(GFP_USER); | |
1652 | ||
1653 | if (!page) | |
1654 | return -ENOMEM; | |
1655 | ||
1656 | read_lock(¤t->fs->lock); | |
1657 | pwdmnt = mntget(current->fs->pwdmnt); | |
1658 | pwd = dget(current->fs->pwd); | |
1659 | rootmnt = mntget(current->fs->rootmnt); | |
1660 | root = dget(current->fs->root); | |
1661 | read_unlock(¤t->fs->lock); | |
1662 | ||
1663 | error = -ENOENT; | |
1664 | /* Has the current directory has been unlinked? */ | |
1665 | spin_lock(&dcache_lock); | |
1666 | if (pwd->d_parent == pwd || !d_unhashed(pwd)) { | |
1667 | unsigned long len; | |
1668 | char * cwd; | |
1669 | ||
1670 | cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE); | |
1671 | spin_unlock(&dcache_lock); | |
1672 | ||
1673 | error = PTR_ERR(cwd); | |
1674 | if (IS_ERR(cwd)) | |
1675 | goto out; | |
1676 | ||
1677 | error = -ERANGE; | |
1678 | len = PAGE_SIZE + page - cwd; | |
1679 | if (len <= size) { | |
1680 | error = len; | |
1681 | if (copy_to_user(buf, cwd, len)) | |
1682 | error = -EFAULT; | |
1683 | } | |
1684 | } else | |
1685 | spin_unlock(&dcache_lock); | |
1686 | ||
1687 | out: | |
1688 | dput(pwd); | |
1689 | mntput(pwdmnt); | |
1690 | dput(root); | |
1691 | mntput(rootmnt); | |
1692 | free_page((unsigned long) page); | |
1693 | return error; | |
1694 | } | |
1695 | ||
1696 | /* | |
1697 | * Test whether new_dentry is a subdirectory of old_dentry. | |
1698 | * | |
1699 | * Trivially implemented using the dcache structure | |
1700 | */ | |
1701 | ||
1702 | /** | |
1703 | * is_subdir - is new dentry a subdirectory of old_dentry | |
1704 | * @new_dentry: new dentry | |
1705 | * @old_dentry: old dentry | |
1706 | * | |
1707 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
1708 | * Returns 0 otherwise. | |
1709 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
1710 | */ | |
1711 | ||
1712 | int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry) | |
1713 | { | |
1714 | int result; | |
1715 | struct dentry * saved = new_dentry; | |
1716 | unsigned long seq; | |
1717 | ||
1718 | /* need rcu_readlock to protect against the d_parent trashing due to | |
1719 | * d_move | |
1720 | */ | |
1721 | rcu_read_lock(); | |
1722 | do { | |
1723 | /* for restarting inner loop in case of seq retry */ | |
1724 | new_dentry = saved; | |
1725 | result = 0; | |
1726 | seq = read_seqbegin(&rename_lock); | |
1727 | for (;;) { | |
1728 | if (new_dentry != old_dentry) { | |
1729 | struct dentry * parent = new_dentry->d_parent; | |
1730 | if (parent == new_dentry) | |
1731 | break; | |
1732 | new_dentry = parent; | |
1733 | continue; | |
1734 | } | |
1735 | result = 1; | |
1736 | break; | |
1737 | } | |
1738 | } while (read_seqretry(&rename_lock, seq)); | |
1739 | rcu_read_unlock(); | |
1740 | ||
1741 | return result; | |
1742 | } | |
1743 | ||
1744 | void d_genocide(struct dentry *root) | |
1745 | { | |
1746 | struct dentry *this_parent = root; | |
1747 | struct list_head *next; | |
1748 | ||
1749 | spin_lock(&dcache_lock); | |
1750 | repeat: | |
1751 | next = this_parent->d_subdirs.next; | |
1752 | resume: | |
1753 | while (next != &this_parent->d_subdirs) { | |
1754 | struct list_head *tmp = next; | |
5160ee6f | 1755 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1756 | next = tmp->next; |
1757 | if (d_unhashed(dentry)||!dentry->d_inode) | |
1758 | continue; | |
1759 | if (!list_empty(&dentry->d_subdirs)) { | |
1760 | this_parent = dentry; | |
1761 | goto repeat; | |
1762 | } | |
1763 | atomic_dec(&dentry->d_count); | |
1764 | } | |
1765 | if (this_parent != root) { | |
5160ee6f | 1766 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
1767 | atomic_dec(&this_parent->d_count); |
1768 | this_parent = this_parent->d_parent; | |
1769 | goto resume; | |
1770 | } | |
1771 | spin_unlock(&dcache_lock); | |
1772 | } | |
1773 | ||
1774 | /** | |
1775 | * find_inode_number - check for dentry with name | |
1776 | * @dir: directory to check | |
1777 | * @name: Name to find. | |
1778 | * | |
1779 | * Check whether a dentry already exists for the given name, | |
1780 | * and return the inode number if it has an inode. Otherwise | |
1781 | * 0 is returned. | |
1782 | * | |
1783 | * This routine is used to post-process directory listings for | |
1784 | * filesystems using synthetic inode numbers, and is necessary | |
1785 | * to keep getcwd() working. | |
1786 | */ | |
1787 | ||
1788 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
1789 | { | |
1790 | struct dentry * dentry; | |
1791 | ino_t ino = 0; | |
1792 | ||
3e7e241f EB |
1793 | dentry = d_hash_and_lookup(dir, name); |
1794 | if (dentry) { | |
1da177e4 LT |
1795 | if (dentry->d_inode) |
1796 | ino = dentry->d_inode->i_ino; | |
1797 | dput(dentry); | |
1798 | } | |
1da177e4 LT |
1799 | return ino; |
1800 | } | |
1801 | ||
1802 | static __initdata unsigned long dhash_entries; | |
1803 | static int __init set_dhash_entries(char *str) | |
1804 | { | |
1805 | if (!str) | |
1806 | return 0; | |
1807 | dhash_entries = simple_strtoul(str, &str, 0); | |
1808 | return 1; | |
1809 | } | |
1810 | __setup("dhash_entries=", set_dhash_entries); | |
1811 | ||
1812 | static void __init dcache_init_early(void) | |
1813 | { | |
1814 | int loop; | |
1815 | ||
1816 | /* If hashes are distributed across NUMA nodes, defer | |
1817 | * hash allocation until vmalloc space is available. | |
1818 | */ | |
1819 | if (hashdist) | |
1820 | return; | |
1821 | ||
1822 | dentry_hashtable = | |
1823 | alloc_large_system_hash("Dentry cache", | |
1824 | sizeof(struct hlist_head), | |
1825 | dhash_entries, | |
1826 | 13, | |
1827 | HASH_EARLY, | |
1828 | &d_hash_shift, | |
1829 | &d_hash_mask, | |
1830 | 0); | |
1831 | ||
1832 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
1833 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
1834 | } | |
1835 | ||
1836 | static void __init dcache_init(unsigned long mempages) | |
1837 | { | |
1838 | int loop; | |
1839 | ||
1840 | /* | |
1841 | * A constructor could be added for stable state like the lists, | |
1842 | * but it is probably not worth it because of the cache nature | |
1843 | * of the dcache. | |
1844 | */ | |
1845 | dentry_cache = kmem_cache_create("dentry_cache", | |
1846 | sizeof(struct dentry), | |
1847 | 0, | |
b0196009 PJ |
1848 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| |
1849 | SLAB_MEM_SPREAD), | |
1da177e4 LT |
1850 | NULL, NULL); |
1851 | ||
1852 | set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory); | |
1853 | ||
1854 | /* Hash may have been set up in dcache_init_early */ | |
1855 | if (!hashdist) | |
1856 | return; | |
1857 | ||
1858 | dentry_hashtable = | |
1859 | alloc_large_system_hash("Dentry cache", | |
1860 | sizeof(struct hlist_head), | |
1861 | dhash_entries, | |
1862 | 13, | |
1863 | 0, | |
1864 | &d_hash_shift, | |
1865 | &d_hash_mask, | |
1866 | 0); | |
1867 | ||
1868 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
1869 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
1870 | } | |
1871 | ||
1872 | /* SLAB cache for __getname() consumers */ | |
fa3536cc | 1873 | kmem_cache_t *names_cachep __read_mostly; |
1da177e4 LT |
1874 | |
1875 | /* SLAB cache for file structures */ | |
fa3536cc | 1876 | kmem_cache_t *filp_cachep __read_mostly; |
1da177e4 LT |
1877 | |
1878 | EXPORT_SYMBOL(d_genocide); | |
1879 | ||
1880 | extern void bdev_cache_init(void); | |
1881 | extern void chrdev_init(void); | |
1882 | ||
1883 | void __init vfs_caches_init_early(void) | |
1884 | { | |
1885 | dcache_init_early(); | |
1886 | inode_init_early(); | |
1887 | } | |
1888 | ||
1889 | void __init vfs_caches_init(unsigned long mempages) | |
1890 | { | |
1891 | unsigned long reserve; | |
1892 | ||
1893 | /* Base hash sizes on available memory, with a reserve equal to | |
1894 | 150% of current kernel size */ | |
1895 | ||
1896 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
1897 | mempages -= reserve; | |
1898 | ||
1899 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
1900 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1901 | ||
1902 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
529bf6be | 1903 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
1da177e4 LT |
1904 | |
1905 | dcache_init(mempages); | |
1906 | inode_init(mempages); | |
1907 | files_init(mempages); | |
1908 | mnt_init(mempages); | |
1909 | bdev_cache_init(); | |
1910 | chrdev_init(); | |
1911 | } | |
1912 | ||
1913 | EXPORT_SYMBOL(d_alloc); | |
1914 | EXPORT_SYMBOL(d_alloc_anon); | |
1915 | EXPORT_SYMBOL(d_alloc_root); | |
1916 | EXPORT_SYMBOL(d_delete); | |
1917 | EXPORT_SYMBOL(d_find_alias); | |
1918 | EXPORT_SYMBOL(d_instantiate); | |
1919 | EXPORT_SYMBOL(d_invalidate); | |
1920 | EXPORT_SYMBOL(d_lookup); | |
1921 | EXPORT_SYMBOL(d_move); | |
770bfad8 | 1922 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
1da177e4 LT |
1923 | EXPORT_SYMBOL(d_path); |
1924 | EXPORT_SYMBOL(d_prune_aliases); | |
1925 | EXPORT_SYMBOL(d_rehash); | |
1926 | EXPORT_SYMBOL(d_splice_alias); | |
1927 | EXPORT_SYMBOL(d_validate); | |
1928 | EXPORT_SYMBOL(dget_locked); | |
1929 | EXPORT_SYMBOL(dput); | |
1930 | EXPORT_SYMBOL(find_inode_number); | |
1931 | EXPORT_SYMBOL(have_submounts); | |
1932 | EXPORT_SYMBOL(names_cachep); | |
1933 | EXPORT_SYMBOL(shrink_dcache_parent); | |
1934 | EXPORT_SYMBOL(shrink_dcache_sb); |