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> | |
31f3e0b3 | 20 | #include <linux/fdtable.h> |
1da177e4 | 21 | #include <linux/fs.h> |
7a91bf7f | 22 | #include <linux/fsnotify.h> |
1da177e4 LT |
23 | #include <linux/slab.h> |
24 | #include <linux/init.h> | |
1da177e4 LT |
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> | |
07f3f05c | 35 | #include "internal.h" |
1da177e4 | 36 | |
1da177e4 | 37 | |
fa3536cc | 38 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
39 | EXPORT_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 | |
44 | EXPORT_SYMBOL(dcache_lock); | |
45 | ||
e18b890b | 46 | static 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 |
61 | static unsigned int d_hash_mask __read_mostly; |
62 | static unsigned int d_hash_shift __read_mostly; | |
63 | static struct hlist_head *dentry_hashtable __read_mostly; | |
1da177e4 LT |
64 | |
65 | /* Statistics gathering. */ | |
66 | struct dentry_stat_t dentry_stat = { | |
67 | .age_limit = 45, | |
68 | }; | |
69 | ||
b3423415 | 70 | static void __d_free(struct dentry *dentry) |
1da177e4 | 71 | { |
1da177e4 LT |
72 | if (dname_external(dentry)) |
73 | kfree(dentry->d_name.name); | |
74 | kmem_cache_free(dentry_cache, dentry); | |
75 | } | |
76 | ||
b3423415 ED |
77 | static void d_callback(struct rcu_head *head) |
78 | { | |
79 | struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu); | |
80 | __d_free(dentry); | |
81 | } | |
82 | ||
1da177e4 LT |
83 | /* |
84 | * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry | |
85 | * inside dcache_lock. | |
86 | */ | |
87 | static void d_free(struct dentry *dentry) | |
88 | { | |
89 | if (dentry->d_op && dentry->d_op->d_release) | |
90 | dentry->d_op->d_release(dentry); | |
b3423415 | 91 | /* if dentry was never inserted into hash, immediate free is OK */ |
e8462caa | 92 | if (hlist_unhashed(&dentry->d_hash)) |
b3423415 ED |
93 | __d_free(dentry); |
94 | else | |
95 | call_rcu(&dentry->d_u.d_rcu, d_callback); | |
1da177e4 LT |
96 | } |
97 | ||
98 | /* | |
99 | * Release the dentry's inode, using the filesystem | |
100 | * d_iput() operation if defined. | |
1da177e4 | 101 | */ |
858119e1 | 102 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 MS |
103 | __releases(dentry->d_lock) |
104 | __releases(dcache_lock) | |
1da177e4 LT |
105 | { |
106 | struct inode *inode = dentry->d_inode; | |
107 | if (inode) { | |
108 | dentry->d_inode = NULL; | |
109 | list_del_init(&dentry->d_alias); | |
110 | spin_unlock(&dentry->d_lock); | |
111 | spin_unlock(&dcache_lock); | |
f805fbda LT |
112 | if (!inode->i_nlink) |
113 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
114 | if (dentry->d_op && dentry->d_op->d_iput) |
115 | dentry->d_op->d_iput(dentry, inode); | |
116 | else | |
117 | iput(inode); | |
118 | } else { | |
119 | spin_unlock(&dentry->d_lock); | |
120 | spin_unlock(&dcache_lock); | |
121 | } | |
122 | } | |
123 | ||
da3bbdd4 KM |
124 | /* |
125 | * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held. | |
126 | */ | |
127 | static void dentry_lru_add(struct dentry *dentry) | |
128 | { | |
129 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
130 | dentry->d_sb->s_nr_dentry_unused++; | |
131 | dentry_stat.nr_unused++; | |
132 | } | |
133 | ||
134 | static void dentry_lru_add_tail(struct dentry *dentry) | |
135 | { | |
136 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
137 | dentry->d_sb->s_nr_dentry_unused++; | |
138 | dentry_stat.nr_unused++; | |
139 | } | |
140 | ||
141 | static void dentry_lru_del(struct dentry *dentry) | |
142 | { | |
143 | if (!list_empty(&dentry->d_lru)) { | |
144 | list_del(&dentry->d_lru); | |
145 | dentry->d_sb->s_nr_dentry_unused--; | |
146 | dentry_stat.nr_unused--; | |
147 | } | |
148 | } | |
149 | ||
150 | static void dentry_lru_del_init(struct dentry *dentry) | |
151 | { | |
152 | if (likely(!list_empty(&dentry->d_lru))) { | |
153 | list_del_init(&dentry->d_lru); | |
154 | dentry->d_sb->s_nr_dentry_unused--; | |
155 | dentry_stat.nr_unused--; | |
156 | } | |
157 | } | |
158 | ||
d52b9086 MS |
159 | /** |
160 | * d_kill - kill dentry and return parent | |
161 | * @dentry: dentry to kill | |
162 | * | |
31f3e0b3 | 163 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
164 | * |
165 | * If this is the root of the dentry tree, return NULL. | |
166 | */ | |
167 | static struct dentry *d_kill(struct dentry *dentry) | |
31f3e0b3 MS |
168 | __releases(dentry->d_lock) |
169 | __releases(dcache_lock) | |
d52b9086 MS |
170 | { |
171 | struct dentry *parent; | |
172 | ||
173 | list_del(&dentry->d_u.d_child); | |
174 | dentry_stat.nr_dentry--; /* For d_free, below */ | |
175 | /*drops the locks, at that point nobody can reach this dentry */ | |
176 | dentry_iput(dentry); | |
177 | parent = dentry->d_parent; | |
178 | d_free(dentry); | |
179 | return dentry == parent ? NULL : parent; | |
180 | } | |
181 | ||
1da177e4 LT |
182 | /* |
183 | * This is dput | |
184 | * | |
185 | * This is complicated by the fact that we do not want to put | |
186 | * dentries that are no longer on any hash chain on the unused | |
187 | * list: we'd much rather just get rid of them immediately. | |
188 | * | |
189 | * However, that implies that we have to traverse the dentry | |
190 | * tree upwards to the parents which might _also_ now be | |
191 | * scheduled for deletion (it may have been only waiting for | |
192 | * its last child to go away). | |
193 | * | |
194 | * This tail recursion is done by hand as we don't want to depend | |
195 | * on the compiler to always get this right (gcc generally doesn't). | |
196 | * Real recursion would eat up our stack space. | |
197 | */ | |
198 | ||
199 | /* | |
200 | * dput - release a dentry | |
201 | * @dentry: dentry to release | |
202 | * | |
203 | * Release a dentry. This will drop the usage count and if appropriate | |
204 | * call the dentry unlink method as well as removing it from the queues and | |
205 | * releasing its resources. If the parent dentries were scheduled for release | |
206 | * they too may now get deleted. | |
207 | * | |
208 | * no dcache lock, please. | |
209 | */ | |
210 | ||
211 | void dput(struct dentry *dentry) | |
212 | { | |
213 | if (!dentry) | |
214 | return; | |
215 | ||
216 | repeat: | |
217 | if (atomic_read(&dentry->d_count) == 1) | |
218 | might_sleep(); | |
219 | if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock)) | |
220 | return; | |
221 | ||
222 | spin_lock(&dentry->d_lock); | |
223 | if (atomic_read(&dentry->d_count)) { | |
224 | spin_unlock(&dentry->d_lock); | |
225 | spin_unlock(&dcache_lock); | |
226 | return; | |
227 | } | |
228 | ||
229 | /* | |
230 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
231 | */ | |
232 | if (dentry->d_op && dentry->d_op->d_delete) { | |
233 | if (dentry->d_op->d_delete(dentry)) | |
234 | goto unhash_it; | |
235 | } | |
236 | /* Unreachable? Get rid of it */ | |
237 | if (d_unhashed(dentry)) | |
238 | goto kill_it; | |
239 | if (list_empty(&dentry->d_lru)) { | |
240 | dentry->d_flags |= DCACHE_REFERENCED; | |
da3bbdd4 | 241 | dentry_lru_add(dentry); |
1da177e4 LT |
242 | } |
243 | spin_unlock(&dentry->d_lock); | |
244 | spin_unlock(&dcache_lock); | |
245 | return; | |
246 | ||
247 | unhash_it: | |
248 | __d_drop(dentry); | |
d52b9086 | 249 | kill_it: |
da3bbdd4 KM |
250 | /* if dentry was on the d_lru list delete it from there */ |
251 | dentry_lru_del(dentry); | |
d52b9086 MS |
252 | dentry = d_kill(dentry); |
253 | if (dentry) | |
254 | goto repeat; | |
1da177e4 LT |
255 | } |
256 | ||
257 | /** | |
258 | * d_invalidate - invalidate a dentry | |
259 | * @dentry: dentry to invalidate | |
260 | * | |
261 | * Try to invalidate the dentry if it turns out to be | |
262 | * possible. If there are other dentries that can be | |
263 | * reached through this one we can't delete it and we | |
264 | * return -EBUSY. On success we return 0. | |
265 | * | |
266 | * no dcache lock. | |
267 | */ | |
268 | ||
269 | int d_invalidate(struct dentry * dentry) | |
270 | { | |
271 | /* | |
272 | * If it's already been dropped, return OK. | |
273 | */ | |
274 | spin_lock(&dcache_lock); | |
275 | if (d_unhashed(dentry)) { | |
276 | spin_unlock(&dcache_lock); | |
277 | return 0; | |
278 | } | |
279 | /* | |
280 | * Check whether to do a partial shrink_dcache | |
281 | * to get rid of unused child entries. | |
282 | */ | |
283 | if (!list_empty(&dentry->d_subdirs)) { | |
284 | spin_unlock(&dcache_lock); | |
285 | shrink_dcache_parent(dentry); | |
286 | spin_lock(&dcache_lock); | |
287 | } | |
288 | ||
289 | /* | |
290 | * Somebody else still using it? | |
291 | * | |
292 | * If it's a directory, we can't drop it | |
293 | * for fear of somebody re-populating it | |
294 | * with children (even though dropping it | |
295 | * would make it unreachable from the root, | |
296 | * we might still populate it if it was a | |
297 | * working directory or similar). | |
298 | */ | |
299 | spin_lock(&dentry->d_lock); | |
300 | if (atomic_read(&dentry->d_count) > 1) { | |
301 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { | |
302 | spin_unlock(&dentry->d_lock); | |
303 | spin_unlock(&dcache_lock); | |
304 | return -EBUSY; | |
305 | } | |
306 | } | |
307 | ||
308 | __d_drop(dentry); | |
309 | spin_unlock(&dentry->d_lock); | |
310 | spin_unlock(&dcache_lock); | |
311 | return 0; | |
312 | } | |
313 | ||
314 | /* This should be called _only_ with dcache_lock held */ | |
315 | ||
316 | static inline struct dentry * __dget_locked(struct dentry *dentry) | |
317 | { | |
318 | atomic_inc(&dentry->d_count); | |
da3bbdd4 | 319 | dentry_lru_del_init(dentry); |
1da177e4 LT |
320 | return dentry; |
321 | } | |
322 | ||
323 | struct dentry * dget_locked(struct dentry *dentry) | |
324 | { | |
325 | return __dget_locked(dentry); | |
326 | } | |
327 | ||
328 | /** | |
329 | * d_find_alias - grab a hashed alias of inode | |
330 | * @inode: inode in question | |
331 | * @want_discon: flag, used by d_splice_alias, to request | |
332 | * that only a DISCONNECTED alias be returned. | |
333 | * | |
334 | * If inode has a hashed alias, or is a directory and has any alias, | |
335 | * acquire the reference to alias and return it. Otherwise return NULL. | |
336 | * Notice that if inode is a directory there can be only one alias and | |
337 | * it can be unhashed only if it has no children, or if it is the root | |
338 | * of a filesystem. | |
339 | * | |
21c0d8fd | 340 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 341 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 342 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 LT |
343 | */ |
344 | ||
345 | static struct dentry * __d_find_alias(struct inode *inode, int want_discon) | |
346 | { | |
347 | struct list_head *head, *next, *tmp; | |
348 | struct dentry *alias, *discon_alias=NULL; | |
349 | ||
350 | head = &inode->i_dentry; | |
351 | next = inode->i_dentry.next; | |
352 | while (next != head) { | |
353 | tmp = next; | |
354 | next = tmp->next; | |
355 | prefetch(next); | |
356 | alias = list_entry(tmp, struct dentry, d_alias); | |
357 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
21c0d8fd N |
358 | if (IS_ROOT(alias) && |
359 | (alias->d_flags & DCACHE_DISCONNECTED)) | |
1da177e4 LT |
360 | discon_alias = alias; |
361 | else if (!want_discon) { | |
362 | __dget_locked(alias); | |
363 | return alias; | |
364 | } | |
365 | } | |
366 | } | |
367 | if (discon_alias) | |
368 | __dget_locked(discon_alias); | |
369 | return discon_alias; | |
370 | } | |
371 | ||
372 | struct dentry * d_find_alias(struct inode *inode) | |
373 | { | |
214fda1f DH |
374 | struct dentry *de = NULL; |
375 | ||
376 | if (!list_empty(&inode->i_dentry)) { | |
377 | spin_lock(&dcache_lock); | |
378 | de = __d_find_alias(inode, 0); | |
379 | spin_unlock(&dcache_lock); | |
380 | } | |
1da177e4 LT |
381 | return de; |
382 | } | |
383 | ||
384 | /* | |
385 | * Try to kill dentries associated with this inode. | |
386 | * WARNING: you must own a reference to inode. | |
387 | */ | |
388 | void d_prune_aliases(struct inode *inode) | |
389 | { | |
0cdca3f9 | 390 | struct dentry *dentry; |
1da177e4 LT |
391 | restart: |
392 | spin_lock(&dcache_lock); | |
0cdca3f9 | 393 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 LT |
394 | spin_lock(&dentry->d_lock); |
395 | if (!atomic_read(&dentry->d_count)) { | |
396 | __dget_locked(dentry); | |
397 | __d_drop(dentry); | |
398 | spin_unlock(&dentry->d_lock); | |
399 | spin_unlock(&dcache_lock); | |
400 | dput(dentry); | |
401 | goto restart; | |
402 | } | |
403 | spin_unlock(&dentry->d_lock); | |
404 | } | |
405 | spin_unlock(&dcache_lock); | |
406 | } | |
407 | ||
408 | /* | |
d702ccb3 AM |
409 | * Throw away a dentry - free the inode, dput the parent. This requires that |
410 | * the LRU list has already been removed. | |
411 | * | |
85864e10 MS |
412 | * Try to prune ancestors as well. This is necessary to prevent |
413 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
414 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 415 | */ |
85864e10 | 416 | static void prune_one_dentry(struct dentry * dentry) |
31f3e0b3 MS |
417 | __releases(dentry->d_lock) |
418 | __releases(dcache_lock) | |
419 | __acquires(dcache_lock) | |
1da177e4 | 420 | { |
1da177e4 | 421 | __d_drop(dentry); |
d52b9086 | 422 | dentry = d_kill(dentry); |
d52b9086 MS |
423 | |
424 | /* | |
425 | * Prune ancestors. Locking is simpler than in dput(), | |
426 | * because dcache_lock needs to be taken anyway. | |
427 | */ | |
1da177e4 | 428 | spin_lock(&dcache_lock); |
d52b9086 MS |
429 | while (dentry) { |
430 | if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock)) | |
431 | return; | |
432 | ||
433 | if (dentry->d_op && dentry->d_op->d_delete) | |
434 | dentry->d_op->d_delete(dentry); | |
da3bbdd4 | 435 | dentry_lru_del_init(dentry); |
d52b9086 MS |
436 | __d_drop(dentry); |
437 | dentry = d_kill(dentry); | |
438 | spin_lock(&dcache_lock); | |
439 | } | |
1da177e4 LT |
440 | } |
441 | ||
da3bbdd4 KM |
442 | /* |
443 | * Shrink the dentry LRU on a given superblock. | |
444 | * @sb : superblock to shrink dentry LRU. | |
445 | * @count: If count is NULL, we prune all dentries on superblock. | |
446 | * @flags: If flags is non-zero, we need to do special processing based on | |
447 | * which flags are set. This means we don't need to maintain multiple | |
448 | * similar copies of this loop. | |
1da177e4 | 449 | */ |
da3bbdd4 | 450 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) |
1da177e4 | 451 | { |
da3bbdd4 KM |
452 | LIST_HEAD(referenced); |
453 | LIST_HEAD(tmp); | |
454 | struct dentry *dentry; | |
455 | int cnt = 0; | |
1da177e4 | 456 | |
da3bbdd4 KM |
457 | BUG_ON(!sb); |
458 | BUG_ON((flags & DCACHE_REFERENCED) && count == NULL); | |
459 | spin_lock(&dcache_lock); | |
460 | if (count != NULL) | |
461 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
462 | cnt = *count; | |
463 | restart: | |
464 | if (count == NULL) | |
465 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
466 | else { | |
467 | while (!list_empty(&sb->s_dentry_lru)) { | |
468 | dentry = list_entry(sb->s_dentry_lru.prev, | |
469 | struct dentry, d_lru); | |
470 | BUG_ON(dentry->d_sb != sb); | |
471 | ||
472 | spin_lock(&dentry->d_lock); | |
473 | /* | |
474 | * If we are honouring the DCACHE_REFERENCED flag and | |
475 | * the dentry has this flag set, don't free it. Clear | |
476 | * the flag and put it back on the LRU. | |
0feae5c4 | 477 | */ |
da3bbdd4 KM |
478 | if ((flags & DCACHE_REFERENCED) |
479 | && (dentry->d_flags & DCACHE_REFERENCED)) { | |
480 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
481 | list_move_tail(&dentry->d_lru, &referenced); | |
482 | spin_unlock(&dentry->d_lock); | |
483 | } else { | |
484 | list_move_tail(&dentry->d_lru, &tmp); | |
485 | spin_unlock(&dentry->d_lock); | |
486 | cnt--; | |
487 | if (!cnt) | |
488 | break; | |
0feae5c4 N |
489 | } |
490 | } | |
da3bbdd4 KM |
491 | } |
492 | while (!list_empty(&tmp)) { | |
493 | dentry = list_entry(tmp.prev, struct dentry, d_lru); | |
494 | dentry_lru_del_init(dentry); | |
495 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
496 | /* |
497 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
498 | * the LRU because of laziness during lookup. Do not free |
499 | * it - just keep it off the LRU list. | |
1da177e4 | 500 | */ |
da3bbdd4 KM |
501 | if (atomic_read(&dentry->d_count)) { |
502 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
503 | continue; |
504 | } | |
da3bbdd4 KM |
505 | prune_one_dentry(dentry); |
506 | /* dentry->d_lock was dropped in prune_one_dentry() */ | |
507 | cond_resched_lock(&dcache_lock); | |
508 | } | |
509 | if (count == NULL && !list_empty(&sb->s_dentry_lru)) | |
510 | goto restart; | |
511 | if (count != NULL) | |
512 | *count = cnt; | |
513 | if (!list_empty(&referenced)) | |
514 | list_splice(&referenced, &sb->s_dentry_lru); | |
515 | spin_unlock(&dcache_lock); | |
516 | } | |
517 | ||
518 | /** | |
519 | * prune_dcache - shrink the dcache | |
520 | * @count: number of entries to try to free | |
521 | * | |
522 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
523 | * need to unmount something (at which point we need to unuse all dentries). | |
524 | * | |
525 | * This function may fail to free any resources if all the dentries are in use. | |
526 | */ | |
527 | static void prune_dcache(int count) | |
528 | { | |
529 | struct super_block *sb; | |
530 | int w_count; | |
531 | int unused = dentry_stat.nr_unused; | |
532 | int prune_ratio; | |
533 | int pruned; | |
534 | ||
535 | if (unused == 0 || count == 0) | |
536 | return; | |
537 | spin_lock(&dcache_lock); | |
538 | restart: | |
539 | if (count >= unused) | |
540 | prune_ratio = 1; | |
541 | else | |
542 | prune_ratio = unused / count; | |
543 | spin_lock(&sb_lock); | |
544 | list_for_each_entry(sb, &super_blocks, s_list) { | |
545 | if (sb->s_nr_dentry_unused == 0) | |
1da177e4 | 546 | continue; |
da3bbdd4 KM |
547 | sb->s_count++; |
548 | /* Now, we reclaim unused dentrins with fairness. | |
549 | * We reclaim them same percentage from each superblock. | |
550 | * We calculate number of dentries to scan on this sb | |
551 | * as follows, but the implementation is arranged to avoid | |
552 | * overflows: | |
553 | * number of dentries to scan on this sb = | |
554 | * count * (number of dentries on this sb / | |
555 | * number of dentries in the machine) | |
0feae5c4 | 556 | */ |
da3bbdd4 KM |
557 | spin_unlock(&sb_lock); |
558 | if (prune_ratio != 1) | |
559 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
560 | else | |
561 | w_count = sb->s_nr_dentry_unused; | |
562 | pruned = w_count; | |
0feae5c4 | 563 | /* |
da3bbdd4 KM |
564 | * We need to be sure this filesystem isn't being unmounted, |
565 | * otherwise we could race with generic_shutdown_super(), and | |
566 | * end up holding a reference to an inode while the filesystem | |
567 | * is unmounted. So we try to get s_umount, and make sure | |
568 | * s_root isn't NULL. | |
0feae5c4 | 569 | */ |
da3bbdd4 KM |
570 | if (down_read_trylock(&sb->s_umount)) { |
571 | if ((sb->s_root != NULL) && | |
572 | (!list_empty(&sb->s_dentry_lru))) { | |
573 | spin_unlock(&dcache_lock); | |
574 | __shrink_dcache_sb(sb, &w_count, | |
575 | DCACHE_REFERENCED); | |
576 | pruned -= w_count; | |
577 | spin_lock(&dcache_lock); | |
0feae5c4 | 578 | } |
da3bbdd4 | 579 | up_read(&sb->s_umount); |
0feae5c4 | 580 | } |
da3bbdd4 KM |
581 | spin_lock(&sb_lock); |
582 | count -= pruned; | |
6eac3f93 | 583 | /* |
da3bbdd4 KM |
584 | * restart only when sb is no longer on the list and |
585 | * we have more work to do. | |
0feae5c4 | 586 | */ |
da3bbdd4 KM |
587 | if (__put_super_and_need_restart(sb) && count > 0) { |
588 | spin_unlock(&sb_lock); | |
589 | goto restart; | |
590 | } | |
1da177e4 | 591 | } |
da3bbdd4 | 592 | spin_unlock(&sb_lock); |
1da177e4 LT |
593 | spin_unlock(&dcache_lock); |
594 | } | |
595 | ||
1da177e4 LT |
596 | /** |
597 | * shrink_dcache_sb - shrink dcache for a superblock | |
598 | * @sb: superblock | |
599 | * | |
600 | * Shrink the dcache for the specified super block. This | |
601 | * is used to free the dcache before unmounting a file | |
602 | * system | |
603 | */ | |
1da177e4 LT |
604 | void shrink_dcache_sb(struct super_block * sb) |
605 | { | |
da3bbdd4 | 606 | __shrink_dcache_sb(sb, NULL, 0); |
1da177e4 LT |
607 | } |
608 | ||
c636ebdb DH |
609 | /* |
610 | * destroy a single subtree of dentries for unmount | |
611 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
612 | * locking | |
613 | */ | |
614 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
615 | { | |
616 | struct dentry *parent; | |
f8713576 | 617 | unsigned detached = 0; |
c636ebdb DH |
618 | |
619 | BUG_ON(!IS_ROOT(dentry)); | |
620 | ||
621 | /* detach this root from the system */ | |
622 | spin_lock(&dcache_lock); | |
da3bbdd4 | 623 | dentry_lru_del_init(dentry); |
c636ebdb DH |
624 | __d_drop(dentry); |
625 | spin_unlock(&dcache_lock); | |
626 | ||
627 | for (;;) { | |
628 | /* descend to the first leaf in the current subtree */ | |
629 | while (!list_empty(&dentry->d_subdirs)) { | |
630 | struct dentry *loop; | |
631 | ||
632 | /* this is a branch with children - detach all of them | |
633 | * from the system in one go */ | |
634 | spin_lock(&dcache_lock); | |
635 | list_for_each_entry(loop, &dentry->d_subdirs, | |
636 | d_u.d_child) { | |
da3bbdd4 | 637 | dentry_lru_del_init(loop); |
c636ebdb DH |
638 | __d_drop(loop); |
639 | cond_resched_lock(&dcache_lock); | |
640 | } | |
641 | spin_unlock(&dcache_lock); | |
642 | ||
643 | /* move to the first child */ | |
644 | dentry = list_entry(dentry->d_subdirs.next, | |
645 | struct dentry, d_u.d_child); | |
646 | } | |
647 | ||
648 | /* consume the dentries from this leaf up through its parents | |
649 | * until we find one with children or run out altogether */ | |
650 | do { | |
651 | struct inode *inode; | |
652 | ||
653 | if (atomic_read(&dentry->d_count) != 0) { | |
654 | printk(KERN_ERR | |
655 | "BUG: Dentry %p{i=%lx,n=%s}" | |
656 | " still in use (%d)" | |
657 | " [unmount of %s %s]\n", | |
658 | dentry, | |
659 | dentry->d_inode ? | |
660 | dentry->d_inode->i_ino : 0UL, | |
661 | dentry->d_name.name, | |
662 | atomic_read(&dentry->d_count), | |
663 | dentry->d_sb->s_type->name, | |
664 | dentry->d_sb->s_id); | |
665 | BUG(); | |
666 | } | |
667 | ||
668 | parent = dentry->d_parent; | |
669 | if (parent == dentry) | |
670 | parent = NULL; | |
671 | else | |
672 | atomic_dec(&parent->d_count); | |
673 | ||
674 | list_del(&dentry->d_u.d_child); | |
f8713576 | 675 | detached++; |
c636ebdb DH |
676 | |
677 | inode = dentry->d_inode; | |
678 | if (inode) { | |
679 | dentry->d_inode = NULL; | |
680 | list_del_init(&dentry->d_alias); | |
681 | if (dentry->d_op && dentry->d_op->d_iput) | |
682 | dentry->d_op->d_iput(dentry, inode); | |
683 | else | |
684 | iput(inode); | |
685 | } | |
686 | ||
687 | d_free(dentry); | |
688 | ||
689 | /* finished when we fall off the top of the tree, | |
690 | * otherwise we ascend to the parent and move to the | |
691 | * next sibling if there is one */ | |
692 | if (!parent) | |
f8713576 | 693 | goto out; |
c636ebdb DH |
694 | |
695 | dentry = parent; | |
696 | ||
697 | } while (list_empty(&dentry->d_subdirs)); | |
698 | ||
699 | dentry = list_entry(dentry->d_subdirs.next, | |
700 | struct dentry, d_u.d_child); | |
701 | } | |
f8713576 DH |
702 | out: |
703 | /* several dentries were freed, need to correct nr_dentry */ | |
704 | spin_lock(&dcache_lock); | |
705 | dentry_stat.nr_dentry -= detached; | |
706 | spin_unlock(&dcache_lock); | |
c636ebdb DH |
707 | } |
708 | ||
709 | /* | |
710 | * destroy the dentries attached to a superblock on unmounting | |
711 | * - we don't need to use dentry->d_lock, and only need dcache_lock when | |
712 | * removing the dentry from the system lists and hashes because: | |
713 | * - the superblock is detached from all mountings and open files, so the | |
714 | * dentry trees will not be rearranged by the VFS | |
715 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
716 | * any dentries belonging to this superblock that it comes across | |
717 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
718 | * in this superblock | |
719 | */ | |
720 | void shrink_dcache_for_umount(struct super_block *sb) | |
721 | { | |
722 | struct dentry *dentry; | |
723 | ||
724 | if (down_read_trylock(&sb->s_umount)) | |
725 | BUG(); | |
726 | ||
727 | dentry = sb->s_root; | |
728 | sb->s_root = NULL; | |
729 | atomic_dec(&dentry->d_count); | |
730 | shrink_dcache_for_umount_subtree(dentry); | |
731 | ||
732 | while (!hlist_empty(&sb->s_anon)) { | |
733 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
734 | shrink_dcache_for_umount_subtree(dentry); | |
735 | } | |
736 | } | |
737 | ||
1da177e4 LT |
738 | /* |
739 | * Search for at least 1 mount point in the dentry's subdirs. | |
740 | * We descend to the next level whenever the d_subdirs | |
741 | * list is non-empty and continue searching. | |
742 | */ | |
743 | ||
744 | /** | |
745 | * have_submounts - check for mounts over a dentry | |
746 | * @parent: dentry to check. | |
747 | * | |
748 | * Return true if the parent or its subdirectories contain | |
749 | * a mount point | |
750 | */ | |
751 | ||
752 | int have_submounts(struct dentry *parent) | |
753 | { | |
754 | struct dentry *this_parent = parent; | |
755 | struct list_head *next; | |
756 | ||
757 | spin_lock(&dcache_lock); | |
758 | if (d_mountpoint(parent)) | |
759 | goto positive; | |
760 | repeat: | |
761 | next = this_parent->d_subdirs.next; | |
762 | resume: | |
763 | while (next != &this_parent->d_subdirs) { | |
764 | struct list_head *tmp = next; | |
5160ee6f | 765 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
766 | next = tmp->next; |
767 | /* Have we found a mount point ? */ | |
768 | if (d_mountpoint(dentry)) | |
769 | goto positive; | |
770 | if (!list_empty(&dentry->d_subdirs)) { | |
771 | this_parent = dentry; | |
772 | goto repeat; | |
773 | } | |
774 | } | |
775 | /* | |
776 | * All done at this level ... ascend and resume the search. | |
777 | */ | |
778 | if (this_parent != parent) { | |
5160ee6f | 779 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
780 | this_parent = this_parent->d_parent; |
781 | goto resume; | |
782 | } | |
783 | spin_unlock(&dcache_lock); | |
784 | return 0; /* No mount points found in tree */ | |
785 | positive: | |
786 | spin_unlock(&dcache_lock); | |
787 | return 1; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Search the dentry child list for the specified parent, | |
792 | * and move any unused dentries to the end of the unused | |
793 | * list for prune_dcache(). We descend to the next level | |
794 | * whenever the d_subdirs list is non-empty and continue | |
795 | * searching. | |
796 | * | |
797 | * It returns zero iff there are no unused children, | |
798 | * otherwise it returns the number of children moved to | |
799 | * the end of the unused list. This may not be the total | |
800 | * number of unused children, because select_parent can | |
801 | * drop the lock and return early due to latency | |
802 | * constraints. | |
803 | */ | |
804 | static int select_parent(struct dentry * parent) | |
805 | { | |
806 | struct dentry *this_parent = parent; | |
807 | struct list_head *next; | |
808 | int found = 0; | |
809 | ||
810 | spin_lock(&dcache_lock); | |
811 | repeat: | |
812 | next = this_parent->d_subdirs.next; | |
813 | resume: | |
814 | while (next != &this_parent->d_subdirs) { | |
815 | struct list_head *tmp = next; | |
5160ee6f | 816 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
817 | next = tmp->next; |
818 | ||
da3bbdd4 | 819 | dentry_lru_del_init(dentry); |
1da177e4 LT |
820 | /* |
821 | * move only zero ref count dentries to the end | |
822 | * of the unused list for prune_dcache | |
823 | */ | |
824 | if (!atomic_read(&dentry->d_count)) { | |
da3bbdd4 | 825 | dentry_lru_add_tail(dentry); |
1da177e4 LT |
826 | found++; |
827 | } | |
828 | ||
829 | /* | |
830 | * We can return to the caller if we have found some (this | |
831 | * ensures forward progress). We'll be coming back to find | |
832 | * the rest. | |
833 | */ | |
834 | if (found && need_resched()) | |
835 | goto out; | |
836 | ||
837 | /* | |
838 | * Descend a level if the d_subdirs list is non-empty. | |
839 | */ | |
840 | if (!list_empty(&dentry->d_subdirs)) { | |
841 | this_parent = dentry; | |
1da177e4 LT |
842 | goto repeat; |
843 | } | |
844 | } | |
845 | /* | |
846 | * All done at this level ... ascend and resume the search. | |
847 | */ | |
848 | if (this_parent != parent) { | |
5160ee6f | 849 | next = this_parent->d_u.d_child.next; |
1da177e4 | 850 | this_parent = this_parent->d_parent; |
1da177e4 LT |
851 | goto resume; |
852 | } | |
853 | out: | |
854 | spin_unlock(&dcache_lock); | |
855 | return found; | |
856 | } | |
857 | ||
858 | /** | |
859 | * shrink_dcache_parent - prune dcache | |
860 | * @parent: parent of entries to prune | |
861 | * | |
862 | * Prune the dcache to remove unused children of the parent dentry. | |
863 | */ | |
864 | ||
865 | void shrink_dcache_parent(struct dentry * parent) | |
866 | { | |
da3bbdd4 | 867 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
868 | int found; |
869 | ||
870 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 871 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 LT |
872 | } |
873 | ||
1da177e4 LT |
874 | /* |
875 | * Scan `nr' dentries and return the number which remain. | |
876 | * | |
877 | * We need to avoid reentering the filesystem if the caller is performing a | |
878 | * GFP_NOFS allocation attempt. One example deadlock is: | |
879 | * | |
880 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
881 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
882 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
883 | * | |
884 | * In this case we return -1 to tell the caller that we baled. | |
885 | */ | |
27496a8c | 886 | static int shrink_dcache_memory(int nr, gfp_t gfp_mask) |
1da177e4 LT |
887 | { |
888 | if (nr) { | |
889 | if (!(gfp_mask & __GFP_FS)) | |
890 | return -1; | |
da3bbdd4 | 891 | prune_dcache(nr); |
1da177e4 LT |
892 | } |
893 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
894 | } | |
895 | ||
8e1f936b RR |
896 | static struct shrinker dcache_shrinker = { |
897 | .shrink = shrink_dcache_memory, | |
898 | .seeks = DEFAULT_SEEKS, | |
899 | }; | |
900 | ||
1da177e4 LT |
901 | /** |
902 | * d_alloc - allocate a dcache entry | |
903 | * @parent: parent of entry to allocate | |
904 | * @name: qstr of the name | |
905 | * | |
906 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
907 | * available. On a success the dentry is returned. The name passed in is | |
908 | * copied and the copy passed in may be reused after this call. | |
909 | */ | |
910 | ||
911 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
912 | { | |
913 | struct dentry *dentry; | |
914 | char *dname; | |
915 | ||
e12ba74d | 916 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
917 | if (!dentry) |
918 | return NULL; | |
919 | ||
920 | if (name->len > DNAME_INLINE_LEN-1) { | |
921 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
922 | if (!dname) { | |
923 | kmem_cache_free(dentry_cache, dentry); | |
924 | return NULL; | |
925 | } | |
926 | } else { | |
927 | dname = dentry->d_iname; | |
928 | } | |
929 | dentry->d_name.name = dname; | |
930 | ||
931 | dentry->d_name.len = name->len; | |
932 | dentry->d_name.hash = name->hash; | |
933 | memcpy(dname, name->name, name->len); | |
934 | dname[name->len] = 0; | |
935 | ||
936 | atomic_set(&dentry->d_count, 1); | |
937 | dentry->d_flags = DCACHE_UNHASHED; | |
938 | spin_lock_init(&dentry->d_lock); | |
939 | dentry->d_inode = NULL; | |
940 | dentry->d_parent = NULL; | |
941 | dentry->d_sb = NULL; | |
942 | dentry->d_op = NULL; | |
943 | dentry->d_fsdata = NULL; | |
944 | dentry->d_mounted = 0; | |
47ba87e0 | 945 | #ifdef CONFIG_PROFILING |
1da177e4 | 946 | dentry->d_cookie = NULL; |
47ba87e0 | 947 | #endif |
1da177e4 LT |
948 | INIT_HLIST_NODE(&dentry->d_hash); |
949 | INIT_LIST_HEAD(&dentry->d_lru); | |
950 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
951 | INIT_LIST_HEAD(&dentry->d_alias); | |
952 | ||
953 | if (parent) { | |
954 | dentry->d_parent = dget(parent); | |
955 | dentry->d_sb = parent->d_sb; | |
956 | } else { | |
5160ee6f | 957 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
958 | } |
959 | ||
960 | spin_lock(&dcache_lock); | |
961 | if (parent) | |
5160ee6f | 962 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1da177e4 LT |
963 | dentry_stat.nr_dentry++; |
964 | spin_unlock(&dcache_lock); | |
965 | ||
966 | return dentry; | |
967 | } | |
968 | ||
969 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
970 | { | |
971 | struct qstr q; | |
972 | ||
973 | q.name = name; | |
974 | q.len = strlen(name); | |
975 | q.hash = full_name_hash(q.name, q.len); | |
976 | return d_alloc(parent, &q); | |
977 | } | |
978 | ||
979 | /** | |
980 | * d_instantiate - fill in inode information for a dentry | |
981 | * @entry: dentry to complete | |
982 | * @inode: inode to attach to this dentry | |
983 | * | |
984 | * Fill in inode information in the entry. | |
985 | * | |
986 | * This turns negative dentries into productive full members | |
987 | * of society. | |
988 | * | |
989 | * NOTE! This assumes that the inode count has been incremented | |
990 | * (or otherwise set) by the caller to indicate that it is now | |
991 | * in use by the dcache. | |
992 | */ | |
993 | ||
994 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
995 | { | |
28133c7b | 996 | BUG_ON(!list_empty(&entry->d_alias)); |
1da177e4 LT |
997 | spin_lock(&dcache_lock); |
998 | if (inode) | |
999 | list_add(&entry->d_alias, &inode->i_dentry); | |
1000 | entry->d_inode = inode; | |
c32ccd87 | 1001 | fsnotify_d_instantiate(entry, inode); |
1da177e4 LT |
1002 | spin_unlock(&dcache_lock); |
1003 | security_d_instantiate(entry, inode); | |
1004 | } | |
1005 | ||
1006 | /** | |
1007 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1008 | * @entry: dentry to instantiate | |
1009 | * @inode: inode to attach to this dentry | |
1010 | * | |
1011 | * Fill in inode information in the entry. On success, it returns NULL. | |
1012 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1013 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1014 | * |
1015 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1016 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1017 | * |
1018 | * This also assumes that the inode count has been incremented | |
1019 | * (or otherwise set) by the caller to indicate that it is now | |
1020 | * in use by the dcache. | |
1da177e4 | 1021 | */ |
770bfad8 DH |
1022 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1023 | struct inode *inode) | |
1da177e4 LT |
1024 | { |
1025 | struct dentry *alias; | |
1026 | int len = entry->d_name.len; | |
1027 | const char *name = entry->d_name.name; | |
1028 | unsigned int hash = entry->d_name.hash; | |
1029 | ||
770bfad8 DH |
1030 | if (!inode) { |
1031 | entry->d_inode = NULL; | |
1032 | return NULL; | |
1033 | } | |
1034 | ||
1da177e4 LT |
1035 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1036 | struct qstr *qstr = &alias->d_name; | |
1037 | ||
1038 | if (qstr->hash != hash) | |
1039 | continue; | |
1040 | if (alias->d_parent != entry->d_parent) | |
1041 | continue; | |
1042 | if (qstr->len != len) | |
1043 | continue; | |
1044 | if (memcmp(qstr->name, name, len)) | |
1045 | continue; | |
1046 | dget_locked(alias); | |
1da177e4 LT |
1047 | return alias; |
1048 | } | |
770bfad8 | 1049 | |
1da177e4 | 1050 | list_add(&entry->d_alias, &inode->i_dentry); |
1da177e4 | 1051 | entry->d_inode = inode; |
c32ccd87 | 1052 | fsnotify_d_instantiate(entry, inode); |
1da177e4 LT |
1053 | return NULL; |
1054 | } | |
770bfad8 DH |
1055 | |
1056 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1057 | { | |
1058 | struct dentry *result; | |
1059 | ||
1060 | BUG_ON(!list_empty(&entry->d_alias)); | |
1061 | ||
1062 | spin_lock(&dcache_lock); | |
1063 | result = __d_instantiate_unique(entry, inode); | |
1064 | spin_unlock(&dcache_lock); | |
1065 | ||
1066 | if (!result) { | |
1067 | security_d_instantiate(entry, inode); | |
1068 | return NULL; | |
1069 | } | |
1070 | ||
1071 | BUG_ON(!d_unhashed(result)); | |
1072 | iput(inode); | |
1073 | return result; | |
1074 | } | |
1075 | ||
1da177e4 LT |
1076 | EXPORT_SYMBOL(d_instantiate_unique); |
1077 | ||
1078 | /** | |
1079 | * d_alloc_root - allocate root dentry | |
1080 | * @root_inode: inode to allocate the root for | |
1081 | * | |
1082 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1083 | * instantiated and returned. %NULL is returned if there is insufficient | |
1084 | * memory or the inode passed is %NULL. | |
1085 | */ | |
1086 | ||
1087 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1088 | { | |
1089 | struct dentry *res = NULL; | |
1090 | ||
1091 | if (root_inode) { | |
1092 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1093 | ||
1094 | res = d_alloc(NULL, &name); | |
1095 | if (res) { | |
1096 | res->d_sb = root_inode->i_sb; | |
1097 | res->d_parent = res; | |
1098 | d_instantiate(res, root_inode); | |
1099 | } | |
1100 | } | |
1101 | return res; | |
1102 | } | |
1103 | ||
1104 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1105 | unsigned long hash) | |
1106 | { | |
1107 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1108 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1109 | return dentry_hashtable + (hash & D_HASHMASK); | |
1110 | } | |
1111 | ||
1112 | /** | |
1113 | * d_alloc_anon - allocate an anonymous dentry | |
1114 | * @inode: inode to allocate the dentry for | |
1115 | * | |
1116 | * This is similar to d_alloc_root. It is used by filesystems when | |
1117 | * creating a dentry for a given inode, often in the process of | |
1118 | * mapping a filehandle to a dentry. The returned dentry may be | |
1119 | * anonymous, or may have a full name (if the inode was already | |
1120 | * in the cache). The file system may need to make further | |
1121 | * efforts to connect this dentry into the dcache properly. | |
1122 | * | |
1123 | * When called on a directory inode, we must ensure that | |
1124 | * the inode only ever has one dentry. If a dentry is | |
1125 | * found, that is returned instead of allocating a new one. | |
1126 | * | |
1127 | * On successful return, the reference to the inode has been transferred | |
1128 | * to the dentry. If %NULL is returned (indicating kmalloc failure), | |
1129 | * the reference on the inode has not been released. | |
1130 | */ | |
1131 | ||
1132 | struct dentry * d_alloc_anon(struct inode *inode) | |
1133 | { | |
1134 | static const struct qstr anonstring = { .name = "" }; | |
1135 | struct dentry *tmp; | |
1136 | struct dentry *res; | |
1137 | ||
1138 | if ((res = d_find_alias(inode))) { | |
1139 | iput(inode); | |
1140 | return res; | |
1141 | } | |
1142 | ||
1143 | tmp = d_alloc(NULL, &anonstring); | |
1144 | if (!tmp) | |
1145 | return NULL; | |
1146 | ||
1147 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ | |
1148 | ||
1149 | spin_lock(&dcache_lock); | |
1150 | res = __d_find_alias(inode, 0); | |
1151 | if (!res) { | |
1152 | /* attach a disconnected dentry */ | |
1153 | res = tmp; | |
1154 | tmp = NULL; | |
1155 | spin_lock(&res->d_lock); | |
1156 | res->d_sb = inode->i_sb; | |
1157 | res->d_parent = res; | |
1158 | res->d_inode = inode; | |
1159 | res->d_flags |= DCACHE_DISCONNECTED; | |
1160 | res->d_flags &= ~DCACHE_UNHASHED; | |
1161 | list_add(&res->d_alias, &inode->i_dentry); | |
1162 | hlist_add_head(&res->d_hash, &inode->i_sb->s_anon); | |
1163 | spin_unlock(&res->d_lock); | |
1164 | ||
1165 | inode = NULL; /* don't drop reference */ | |
1166 | } | |
1167 | spin_unlock(&dcache_lock); | |
1168 | ||
1169 | if (inode) | |
1170 | iput(inode); | |
1171 | if (tmp) | |
1172 | dput(tmp); | |
1173 | return res; | |
1174 | } | |
1175 | ||
1176 | ||
1177 | /** | |
1178 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1179 | * @inode: the inode which may have a disconnected dentry | |
1180 | * @dentry: a negative dentry which we want to point to the inode. | |
1181 | * | |
1182 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1183 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1184 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1185 | * | |
1186 | * This is needed in the lookup routine of any filesystem that is exportable | |
1187 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1188 | * | |
1189 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1190 | * is returned. This matches the expected return value of ->lookup. | |
1191 | * | |
1192 | */ | |
1193 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1194 | { | |
1195 | struct dentry *new = NULL; | |
1196 | ||
21c0d8fd | 1197 | if (inode && S_ISDIR(inode->i_mode)) { |
1da177e4 LT |
1198 | spin_lock(&dcache_lock); |
1199 | new = __d_find_alias(inode, 1); | |
1200 | if (new) { | |
1201 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
c32ccd87 | 1202 | fsnotify_d_instantiate(new, inode); |
1da177e4 LT |
1203 | spin_unlock(&dcache_lock); |
1204 | security_d_instantiate(new, inode); | |
1205 | d_rehash(dentry); | |
1206 | d_move(new, dentry); | |
1207 | iput(inode); | |
1208 | } else { | |
1209 | /* d_instantiate takes dcache_lock, so we do it by hand */ | |
1210 | list_add(&dentry->d_alias, &inode->i_dentry); | |
1211 | dentry->d_inode = inode; | |
c32ccd87 | 1212 | fsnotify_d_instantiate(dentry, inode); |
1da177e4 LT |
1213 | spin_unlock(&dcache_lock); |
1214 | security_d_instantiate(dentry, inode); | |
1215 | d_rehash(dentry); | |
1216 | } | |
1217 | } else | |
1218 | d_add(dentry, inode); | |
1219 | return new; | |
1220 | } | |
1221 | ||
1222 | ||
1223 | /** | |
1224 | * d_lookup - search for a dentry | |
1225 | * @parent: parent dentry | |
1226 | * @name: qstr of name we wish to find | |
1227 | * | |
1228 | * Searches the children of the parent dentry for the name in question. If | |
1229 | * the dentry is found its reference count is incremented and the dentry | |
1230 | * is returned. The caller must use d_put to free the entry when it has | |
1231 | * finished using it. %NULL is returned on failure. | |
1232 | * | |
1233 | * __d_lookup is dcache_lock free. The hash list is protected using RCU. | |
1234 | * Memory barriers are used while updating and doing lockless traversal. | |
1235 | * To avoid races with d_move while rename is happening, d_lock is used. | |
1236 | * | |
1237 | * Overflows in memcmp(), while d_move, are avoided by keeping the length | |
1238 | * and name pointer in one structure pointed by d_qstr. | |
1239 | * | |
1240 | * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while | |
1241 | * lookup is going on. | |
1242 | * | |
da3bbdd4 | 1243 | * The dentry unused LRU is not updated even if lookup finds the required dentry |
1da177e4 LT |
1244 | * in there. It is updated in places such as prune_dcache, shrink_dcache_sb, |
1245 | * select_parent and __dget_locked. This laziness saves lookup from dcache_lock | |
1246 | * acquisition. | |
1247 | * | |
1248 | * d_lookup() is protected against the concurrent renames in some unrelated | |
1249 | * directory using the seqlockt_t rename_lock. | |
1250 | */ | |
1251 | ||
1252 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) | |
1253 | { | |
1254 | struct dentry * dentry = NULL; | |
1255 | unsigned long seq; | |
1256 | ||
1257 | do { | |
1258 | seq = read_seqbegin(&rename_lock); | |
1259 | dentry = __d_lookup(parent, name); | |
1260 | if (dentry) | |
1261 | break; | |
1262 | } while (read_seqretry(&rename_lock, seq)); | |
1263 | return dentry; | |
1264 | } | |
1265 | ||
1266 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) | |
1267 | { | |
1268 | unsigned int len = name->len; | |
1269 | unsigned int hash = name->hash; | |
1270 | const unsigned char *str = name->name; | |
1271 | struct hlist_head *head = d_hash(parent,hash); | |
1272 | struct dentry *found = NULL; | |
1273 | struct hlist_node *node; | |
665a7583 | 1274 | struct dentry *dentry; |
1da177e4 LT |
1275 | |
1276 | rcu_read_lock(); | |
1277 | ||
665a7583 | 1278 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1279 | struct qstr *qstr; |
1280 | ||
1da177e4 LT |
1281 | if (dentry->d_name.hash != hash) |
1282 | continue; | |
1283 | if (dentry->d_parent != parent) | |
1284 | continue; | |
1285 | ||
1286 | spin_lock(&dentry->d_lock); | |
1287 | ||
1288 | /* | |
1289 | * Recheck the dentry after taking the lock - d_move may have | |
1290 | * changed things. Don't bother checking the hash because we're | |
1291 | * about to compare the whole name anyway. | |
1292 | */ | |
1293 | if (dentry->d_parent != parent) | |
1294 | goto next; | |
1295 | ||
1296 | /* | |
1297 | * It is safe to compare names since d_move() cannot | |
1298 | * change the qstr (protected by d_lock). | |
1299 | */ | |
1300 | qstr = &dentry->d_name; | |
1301 | if (parent->d_op && parent->d_op->d_compare) { | |
1302 | if (parent->d_op->d_compare(parent, qstr, name)) | |
1303 | goto next; | |
1304 | } else { | |
1305 | if (qstr->len != len) | |
1306 | goto next; | |
1307 | if (memcmp(qstr->name, str, len)) | |
1308 | goto next; | |
1309 | } | |
1310 | ||
1311 | if (!d_unhashed(dentry)) { | |
1312 | atomic_inc(&dentry->d_count); | |
1313 | found = dentry; | |
1314 | } | |
1315 | spin_unlock(&dentry->d_lock); | |
1316 | break; | |
1317 | next: | |
1318 | spin_unlock(&dentry->d_lock); | |
1319 | } | |
1320 | rcu_read_unlock(); | |
1321 | ||
1322 | return found; | |
1323 | } | |
1324 | ||
3e7e241f EB |
1325 | /** |
1326 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1327 | * @dir: Directory to search in | |
1328 | * @name: qstr of name we wish to find | |
1329 | * | |
1330 | * On hash failure or on lookup failure NULL is returned. | |
1331 | */ | |
1332 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1333 | { | |
1334 | struct dentry *dentry = NULL; | |
1335 | ||
1336 | /* | |
1337 | * Check for a fs-specific hash function. Note that we must | |
1338 | * calculate the standard hash first, as the d_op->d_hash() | |
1339 | * routine may choose to leave the hash value unchanged. | |
1340 | */ | |
1341 | name->hash = full_name_hash(name->name, name->len); | |
1342 | if (dir->d_op && dir->d_op->d_hash) { | |
1343 | if (dir->d_op->d_hash(dir, name) < 0) | |
1344 | goto out; | |
1345 | } | |
1346 | dentry = d_lookup(dir, name); | |
1347 | out: | |
1348 | return dentry; | |
1349 | } | |
1350 | ||
1da177e4 LT |
1351 | /** |
1352 | * d_validate - verify dentry provided from insecure source | |
1353 | * @dentry: The dentry alleged to be valid child of @dparent | |
1354 | * @dparent: The parent dentry (known to be valid) | |
1355 | * @hash: Hash of the dentry | |
1356 | * @len: Length of the name | |
1357 | * | |
1358 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1359 | * This is used by ncpfs in its readdir implementation. | |
1360 | * Zero is returned in the dentry is invalid. | |
1361 | */ | |
1362 | ||
1363 | int d_validate(struct dentry *dentry, struct dentry *dparent) | |
1364 | { | |
1365 | struct hlist_head *base; | |
1366 | struct hlist_node *lhp; | |
1367 | ||
1368 | /* Check whether the ptr might be valid at all.. */ | |
1369 | if (!kmem_ptr_validate(dentry_cache, dentry)) | |
1370 | goto out; | |
1371 | ||
1372 | if (dentry->d_parent != dparent) | |
1373 | goto out; | |
1374 | ||
1375 | spin_lock(&dcache_lock); | |
1376 | base = d_hash(dparent, dentry->d_name.hash); | |
1377 | hlist_for_each(lhp,base) { | |
665a7583 | 1378 | /* hlist_for_each_entry_rcu() not required for d_hash list |
1da177e4 LT |
1379 | * as it is parsed under dcache_lock |
1380 | */ | |
1381 | if (dentry == hlist_entry(lhp, struct dentry, d_hash)) { | |
1382 | __dget_locked(dentry); | |
1383 | spin_unlock(&dcache_lock); | |
1384 | return 1; | |
1385 | } | |
1386 | } | |
1387 | spin_unlock(&dcache_lock); | |
1388 | out: | |
1389 | return 0; | |
1390 | } | |
1391 | ||
1392 | /* | |
1393 | * When a file is deleted, we have two options: | |
1394 | * - turn this dentry into a negative dentry | |
1395 | * - unhash this dentry and free it. | |
1396 | * | |
1397 | * Usually, we want to just turn this into | |
1398 | * a negative dentry, but if anybody else is | |
1399 | * currently using the dentry or the inode | |
1400 | * we can't do that and we fall back on removing | |
1401 | * it from the hash queues and waiting for | |
1402 | * it to be deleted later when it has no users | |
1403 | */ | |
1404 | ||
1405 | /** | |
1406 | * d_delete - delete a dentry | |
1407 | * @dentry: The dentry to delete | |
1408 | * | |
1409 | * Turn the dentry into a negative dentry if possible, otherwise | |
1410 | * remove it from the hash queues so it can be deleted later | |
1411 | */ | |
1412 | ||
1413 | void d_delete(struct dentry * dentry) | |
1414 | { | |
7a91bf7f | 1415 | int isdir = 0; |
1da177e4 LT |
1416 | /* |
1417 | * Are we the only user? | |
1418 | */ | |
1419 | spin_lock(&dcache_lock); | |
1420 | spin_lock(&dentry->d_lock); | |
7a91bf7f | 1421 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
1da177e4 LT |
1422 | if (atomic_read(&dentry->d_count) == 1) { |
1423 | dentry_iput(dentry); | |
7a91bf7f | 1424 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1425 | return; |
1426 | } | |
1427 | ||
1428 | if (!d_unhashed(dentry)) | |
1429 | __d_drop(dentry); | |
1430 | ||
1431 | spin_unlock(&dentry->d_lock); | |
1432 | spin_unlock(&dcache_lock); | |
7a91bf7f JM |
1433 | |
1434 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 LT |
1435 | } |
1436 | ||
1437 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1438 | { | |
1439 | ||
1440 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1441 | hlist_add_head_rcu(&entry->d_hash, list); | |
1442 | } | |
1443 | ||
770bfad8 DH |
1444 | static void _d_rehash(struct dentry * entry) |
1445 | { | |
1446 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1447 | } | |
1448 | ||
1da177e4 LT |
1449 | /** |
1450 | * d_rehash - add an entry back to the hash | |
1451 | * @entry: dentry to add to the hash | |
1452 | * | |
1453 | * Adds a dentry to the hash according to its name. | |
1454 | */ | |
1455 | ||
1456 | void d_rehash(struct dentry * entry) | |
1457 | { | |
1da177e4 LT |
1458 | spin_lock(&dcache_lock); |
1459 | spin_lock(&entry->d_lock); | |
770bfad8 | 1460 | _d_rehash(entry); |
1da177e4 LT |
1461 | spin_unlock(&entry->d_lock); |
1462 | spin_unlock(&dcache_lock); | |
1463 | } | |
1464 | ||
1465 | #define do_switch(x,y) do { \ | |
1466 | __typeof__ (x) __tmp = x; \ | |
1467 | x = y; y = __tmp; } while (0) | |
1468 | ||
1469 | /* | |
1470 | * When switching names, the actual string doesn't strictly have to | |
1471 | * be preserved in the target - because we're dropping the target | |
1472 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1473 | * the new name before we switch. | |
1474 | * | |
1475 | * Note that we have to be a lot more careful about getting the hash | |
1476 | * switched - we have to switch the hash value properly even if it | |
1477 | * then no longer matches the actual (corrupted) string of the target. | |
1478 | * The hash value has to match the hash queue that the dentry is on.. | |
1479 | */ | |
1480 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
1481 | { | |
1482 | if (dname_external(target)) { | |
1483 | if (dname_external(dentry)) { | |
1484 | /* | |
1485 | * Both external: swap the pointers | |
1486 | */ | |
1487 | do_switch(target->d_name.name, dentry->d_name.name); | |
1488 | } else { | |
1489 | /* | |
1490 | * dentry:internal, target:external. Steal target's | |
1491 | * storage and make target internal. | |
1492 | */ | |
321bcf92 BF |
1493 | memcpy(target->d_iname, dentry->d_name.name, |
1494 | dentry->d_name.len + 1); | |
1da177e4 LT |
1495 | dentry->d_name.name = target->d_name.name; |
1496 | target->d_name.name = target->d_iname; | |
1497 | } | |
1498 | } else { | |
1499 | if (dname_external(dentry)) { | |
1500 | /* | |
1501 | * dentry:external, target:internal. Give dentry's | |
1502 | * storage to target and make dentry internal | |
1503 | */ | |
1504 | memcpy(dentry->d_iname, target->d_name.name, | |
1505 | target->d_name.len + 1); | |
1506 | target->d_name.name = dentry->d_name.name; | |
1507 | dentry->d_name.name = dentry->d_iname; | |
1508 | } else { | |
1509 | /* | |
1510 | * Both are internal. Just copy target to dentry | |
1511 | */ | |
1512 | memcpy(dentry->d_iname, target->d_name.name, | |
1513 | target->d_name.len + 1); | |
1514 | } | |
1515 | } | |
1516 | } | |
1517 | ||
1518 | /* | |
1519 | * We cannibalize "target" when moving dentry on top of it, | |
1520 | * because it's going to be thrown away anyway. We could be more | |
1521 | * polite about it, though. | |
1522 | * | |
1523 | * This forceful removal will result in ugly /proc output if | |
1524 | * somebody holds a file open that got deleted due to a rename. | |
1525 | * We could be nicer about the deleted file, and let it show | |
bc154b1e BF |
1526 | * up under the name it had before it was deleted rather than |
1527 | * under the original name of the file that was moved on top of it. | |
1da177e4 LT |
1528 | */ |
1529 | ||
9eaef27b TM |
1530 | /* |
1531 | * d_move_locked - move a dentry | |
1da177e4 LT |
1532 | * @dentry: entry to move |
1533 | * @target: new dentry | |
1534 | * | |
1535 | * Update the dcache to reflect the move of a file name. Negative | |
1536 | * dcache entries should not be moved in this way. | |
1537 | */ | |
9eaef27b | 1538 | static void d_move_locked(struct dentry * dentry, struct dentry * target) |
1da177e4 LT |
1539 | { |
1540 | struct hlist_head *list; | |
1541 | ||
1542 | if (!dentry->d_inode) | |
1543 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1544 | ||
1da177e4 LT |
1545 | write_seqlock(&rename_lock); |
1546 | /* | |
1547 | * XXXX: do we really need to take target->d_lock? | |
1548 | */ | |
1549 | if (target < dentry) { | |
1550 | spin_lock(&target->d_lock); | |
a90b9c05 | 1551 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1552 | } else { |
1553 | spin_lock(&dentry->d_lock); | |
a90b9c05 | 1554 | spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1555 | } |
1556 | ||
1557 | /* Move the dentry to the target hash queue, if on different bucket */ | |
f77e3498 | 1558 | if (d_unhashed(dentry)) |
1da177e4 LT |
1559 | goto already_unhashed; |
1560 | ||
1561 | hlist_del_rcu(&dentry->d_hash); | |
1562 | ||
1563 | already_unhashed: | |
1564 | list = d_hash(target->d_parent, target->d_name.hash); | |
1565 | __d_rehash(dentry, list); | |
1566 | ||
1567 | /* Unhash the target: dput() will then get rid of it */ | |
1568 | __d_drop(target); | |
1569 | ||
5160ee6f ED |
1570 | list_del(&dentry->d_u.d_child); |
1571 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
1572 | |
1573 | /* Switch the names.. */ | |
1574 | switch_names(dentry, target); | |
1575 | do_switch(dentry->d_name.len, target->d_name.len); | |
1576 | do_switch(dentry->d_name.hash, target->d_name.hash); | |
1577 | ||
1578 | /* ... and switch the parents */ | |
1579 | if (IS_ROOT(dentry)) { | |
1580 | dentry->d_parent = target->d_parent; | |
1581 | target->d_parent = target; | |
5160ee6f | 1582 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 LT |
1583 | } else { |
1584 | do_switch(dentry->d_parent, target->d_parent); | |
1585 | ||
1586 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 1587 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
1588 | } |
1589 | ||
5160ee6f | 1590 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 1591 | spin_unlock(&target->d_lock); |
c32ccd87 | 1592 | fsnotify_d_move(dentry); |
1da177e4 LT |
1593 | spin_unlock(&dentry->d_lock); |
1594 | write_sequnlock(&rename_lock); | |
9eaef27b TM |
1595 | } |
1596 | ||
1597 | /** | |
1598 | * d_move - move a dentry | |
1599 | * @dentry: entry to move | |
1600 | * @target: new dentry | |
1601 | * | |
1602 | * Update the dcache to reflect the move of a file name. Negative | |
1603 | * dcache entries should not be moved in this way. | |
1604 | */ | |
1605 | ||
1606 | void d_move(struct dentry * dentry, struct dentry * target) | |
1607 | { | |
1608 | spin_lock(&dcache_lock); | |
1609 | d_move_locked(dentry, target); | |
1da177e4 LT |
1610 | spin_unlock(&dcache_lock); |
1611 | } | |
1612 | ||
9eaef27b TM |
1613 | /* |
1614 | * Helper that returns 1 if p1 is a parent of p2, else 0 | |
1615 | */ | |
1616 | static int d_isparent(struct dentry *p1, struct dentry *p2) | |
1617 | { | |
1618 | struct dentry *p; | |
1619 | ||
1620 | for (p = p2; p->d_parent != p; p = p->d_parent) { | |
1621 | if (p->d_parent == p1) | |
1622 | return 1; | |
1623 | } | |
1624 | return 0; | |
1625 | } | |
1626 | ||
1627 | /* | |
1628 | * This helper attempts to cope with remotely renamed directories | |
1629 | * | |
1630 | * It assumes that the caller is already holding | |
1631 | * dentry->d_parent->d_inode->i_mutex and the dcache_lock | |
1632 | * | |
1633 | * Note: If ever the locking in lock_rename() changes, then please | |
1634 | * remember to update this too... | |
9eaef27b TM |
1635 | */ |
1636 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
31f3e0b3 | 1637 | __releases(dcache_lock) |
9eaef27b TM |
1638 | { |
1639 | struct mutex *m1 = NULL, *m2 = NULL; | |
1640 | struct dentry *ret; | |
1641 | ||
1642 | /* If alias and dentry share a parent, then no extra locks required */ | |
1643 | if (alias->d_parent == dentry->d_parent) | |
1644 | goto out_unalias; | |
1645 | ||
1646 | /* Check for loops */ | |
1647 | ret = ERR_PTR(-ELOOP); | |
1648 | if (d_isparent(alias, dentry)) | |
1649 | goto out_err; | |
1650 | ||
1651 | /* See lock_rename() */ | |
1652 | ret = ERR_PTR(-EBUSY); | |
1653 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
1654 | goto out_err; | |
1655 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
1656 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
1657 | goto out_err; | |
1658 | m2 = &alias->d_parent->d_inode->i_mutex; | |
1659 | out_unalias: | |
1660 | d_move_locked(alias, dentry); | |
1661 | ret = alias; | |
1662 | out_err: | |
1663 | spin_unlock(&dcache_lock); | |
1664 | if (m2) | |
1665 | mutex_unlock(m2); | |
1666 | if (m1) | |
1667 | mutex_unlock(m1); | |
1668 | return ret; | |
1669 | } | |
1670 | ||
770bfad8 DH |
1671 | /* |
1672 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
1673 | * named dentry in place of the dentry to be replaced. | |
1674 | */ | |
1675 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
1676 | { | |
1677 | struct dentry *dparent, *aparent; | |
1678 | ||
1679 | switch_names(dentry, anon); | |
1680 | do_switch(dentry->d_name.len, anon->d_name.len); | |
1681 | do_switch(dentry->d_name.hash, anon->d_name.hash); | |
1682 | ||
1683 | dparent = dentry->d_parent; | |
1684 | aparent = anon->d_parent; | |
1685 | ||
1686 | dentry->d_parent = (aparent == anon) ? dentry : aparent; | |
1687 | list_del(&dentry->d_u.d_child); | |
1688 | if (!IS_ROOT(dentry)) | |
1689 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
1690 | else | |
1691 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
1692 | ||
1693 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
1694 | list_del(&anon->d_u.d_child); | |
1695 | if (!IS_ROOT(anon)) | |
1696 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
1697 | else | |
1698 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
1699 | ||
1700 | anon->d_flags &= ~DCACHE_DISCONNECTED; | |
1701 | } | |
1702 | ||
1703 | /** | |
1704 | * d_materialise_unique - introduce an inode into the tree | |
1705 | * @dentry: candidate dentry | |
1706 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
1707 | * | |
1708 | * Introduces an dentry into the tree, substituting an extant disconnected | |
1709 | * root directory alias in its place if there is one | |
1710 | */ | |
1711 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
1712 | { | |
9eaef27b | 1713 | struct dentry *actual; |
770bfad8 DH |
1714 | |
1715 | BUG_ON(!d_unhashed(dentry)); | |
1716 | ||
1717 | spin_lock(&dcache_lock); | |
1718 | ||
1719 | if (!inode) { | |
1720 | actual = dentry; | |
1721 | dentry->d_inode = NULL; | |
1722 | goto found_lock; | |
1723 | } | |
1724 | ||
9eaef27b TM |
1725 | if (S_ISDIR(inode->i_mode)) { |
1726 | struct dentry *alias; | |
1727 | ||
1728 | /* Does an aliased dentry already exist? */ | |
1729 | alias = __d_find_alias(inode, 0); | |
1730 | if (alias) { | |
1731 | actual = alias; | |
1732 | /* Is this an anonymous mountpoint that we could splice | |
1733 | * into our tree? */ | |
1734 | if (IS_ROOT(alias)) { | |
1735 | spin_lock(&alias->d_lock); | |
1736 | __d_materialise_dentry(dentry, alias); | |
1737 | __d_drop(alias); | |
1738 | goto found; | |
1739 | } | |
1740 | /* Nope, but we must(!) avoid directory aliasing */ | |
1741 | actual = __d_unalias(dentry, alias); | |
1742 | if (IS_ERR(actual)) | |
1743 | dput(alias); | |
1744 | goto out_nolock; | |
1745 | } | |
770bfad8 DH |
1746 | } |
1747 | ||
1748 | /* Add a unique reference */ | |
1749 | actual = __d_instantiate_unique(dentry, inode); | |
1750 | if (!actual) | |
1751 | actual = dentry; | |
1752 | else if (unlikely(!d_unhashed(actual))) | |
1753 | goto shouldnt_be_hashed; | |
1754 | ||
1755 | found_lock: | |
1756 | spin_lock(&actual->d_lock); | |
1757 | found: | |
1758 | _d_rehash(actual); | |
1759 | spin_unlock(&actual->d_lock); | |
1760 | spin_unlock(&dcache_lock); | |
9eaef27b | 1761 | out_nolock: |
770bfad8 DH |
1762 | if (actual == dentry) { |
1763 | security_d_instantiate(dentry, inode); | |
1764 | return NULL; | |
1765 | } | |
1766 | ||
1767 | iput(inode); | |
1768 | return actual; | |
1769 | ||
770bfad8 DH |
1770 | shouldnt_be_hashed: |
1771 | spin_unlock(&dcache_lock); | |
1772 | BUG(); | |
770bfad8 DH |
1773 | } |
1774 | ||
cdd16d02 | 1775 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
1776 | { |
1777 | *buflen -= namelen; | |
1778 | if (*buflen < 0) | |
1779 | return -ENAMETOOLONG; | |
1780 | *buffer -= namelen; | |
1781 | memcpy(*buffer, str, namelen); | |
1782 | return 0; | |
1783 | } | |
1784 | ||
cdd16d02 MS |
1785 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
1786 | { | |
1787 | return prepend(buffer, buflen, name->name, name->len); | |
1788 | } | |
1789 | ||
1da177e4 | 1790 | /** |
31f3e0b3 | 1791 | * __d_path - return the path of a dentry |
9d1bc601 MS |
1792 | * @path: the dentry/vfsmount to report |
1793 | * @root: root vfsmnt/dentry (may be modified by this function) | |
1da177e4 LT |
1794 | * @buffer: buffer to return value in |
1795 | * @buflen: buffer length | |
1796 | * | |
552ce544 LT |
1797 | * Convert a dentry into an ASCII path name. If the entry has been deleted |
1798 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1da177e4 | 1799 | * |
552ce544 LT |
1800 | * Returns the buffer or an error code if the path was too long. |
1801 | * | |
1802 | * "buflen" should be positive. Caller holds the dcache_lock. | |
9d1bc601 MS |
1803 | * |
1804 | * If path is not reachable from the supplied root, then the value of | |
1805 | * root is changed (without modifying refcounts). | |
1da177e4 | 1806 | */ |
9d1bc601 MS |
1807 | char *__d_path(const struct path *path, struct path *root, |
1808 | char *buffer, int buflen) | |
1da177e4 | 1809 | { |
9d1bc601 MS |
1810 | struct dentry *dentry = path->dentry; |
1811 | struct vfsmount *vfsmnt = path->mnt; | |
cdd16d02 MS |
1812 | char *end = buffer + buflen; |
1813 | char *retval; | |
6092d048 | 1814 | |
be285c71 | 1815 | spin_lock(&vfsmount_lock); |
6092d048 RP |
1816 | prepend(&end, &buflen, "\0", 1); |
1817 | if (!IS_ROOT(dentry) && d_unhashed(dentry) && | |
1818 | (prepend(&end, &buflen, " (deleted)", 10) != 0)) | |
552ce544 | 1819 | goto Elong; |
552ce544 LT |
1820 | |
1821 | if (buflen < 1) | |
1822 | goto Elong; | |
1823 | /* Get '/' right */ | |
1824 | retval = end-1; | |
1825 | *retval = '/'; | |
1826 | ||
1827 | for (;;) { | |
1da177e4 LT |
1828 | struct dentry * parent; |
1829 | ||
329c97f0 | 1830 | if (dentry == root->dentry && vfsmnt == root->mnt) |
552ce544 | 1831 | break; |
1da177e4 | 1832 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 1833 | /* Global root? */ |
1da177e4 | 1834 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
1835 | goto global_root; |
1836 | } | |
1837 | dentry = vfsmnt->mnt_mountpoint; | |
1838 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
1839 | continue; |
1840 | } | |
1841 | parent = dentry->d_parent; | |
1842 | prefetch(parent); | |
cdd16d02 | 1843 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 | 1844 | (prepend(&end, &buflen, "/", 1) != 0)) |
552ce544 | 1845 | goto Elong; |
552ce544 | 1846 | retval = end; |
1da177e4 LT |
1847 | dentry = parent; |
1848 | } | |
1849 | ||
be285c71 AG |
1850 | out: |
1851 | spin_unlock(&vfsmount_lock); | |
552ce544 | 1852 | return retval; |
1da177e4 LT |
1853 | |
1854 | global_root: | |
6092d048 | 1855 | retval += 1; /* hit the slash */ |
cdd16d02 | 1856 | if (prepend_name(&retval, &buflen, &dentry->d_name) != 0) |
1da177e4 | 1857 | goto Elong; |
9d1bc601 MS |
1858 | root->mnt = vfsmnt; |
1859 | root->dentry = dentry; | |
be285c71 AG |
1860 | goto out; |
1861 | ||
1da177e4 | 1862 | Elong: |
be285c71 AG |
1863 | retval = ERR_PTR(-ENAMETOOLONG); |
1864 | goto out; | |
1da177e4 LT |
1865 | } |
1866 | ||
a03a8a70 JB |
1867 | /** |
1868 | * d_path - return the path of a dentry | |
cf28b486 | 1869 | * @path: path to report |
a03a8a70 JB |
1870 | * @buf: buffer to return value in |
1871 | * @buflen: buffer length | |
1872 | * | |
1873 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
1874 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1875 | * | |
1876 | * Returns the buffer or an error code if the path was too long. | |
1877 | * | |
31f3e0b3 | 1878 | * "buflen" should be positive. |
a03a8a70 | 1879 | */ |
20d4fdc1 | 1880 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 LT |
1881 | { |
1882 | char *res; | |
6ac08c39 | 1883 | struct path root; |
9d1bc601 | 1884 | struct path tmp; |
1da177e4 | 1885 | |
c23fbb6b ED |
1886 | /* |
1887 | * We have various synthetic filesystems that never get mounted. On | |
1888 | * these filesystems dentries are never used for lookup purposes, and | |
1889 | * thus don't need to be hashed. They also don't need a name until a | |
1890 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
1891 | * below allows us to generate a name for these objects on demand: | |
1892 | */ | |
cf28b486 JB |
1893 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
1894 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 1895 | |
1da177e4 | 1896 | read_lock(¤t->fs->lock); |
6ac08c39 | 1897 | root = current->fs->root; |
6092d048 | 1898 | path_get(&root); |
1da177e4 | 1899 | read_unlock(¤t->fs->lock); |
552ce544 | 1900 | spin_lock(&dcache_lock); |
9d1bc601 MS |
1901 | tmp = root; |
1902 | res = __d_path(path, &tmp, buf, buflen); | |
552ce544 | 1903 | spin_unlock(&dcache_lock); |
6ac08c39 | 1904 | path_put(&root); |
1da177e4 LT |
1905 | return res; |
1906 | } | |
1907 | ||
c23fbb6b ED |
1908 | /* |
1909 | * Helper function for dentry_operations.d_dname() members | |
1910 | */ | |
1911 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
1912 | const char *fmt, ...) | |
1913 | { | |
1914 | va_list args; | |
1915 | char temp[64]; | |
1916 | int sz; | |
1917 | ||
1918 | va_start(args, fmt); | |
1919 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
1920 | va_end(args); | |
1921 | ||
1922 | if (sz > sizeof(temp) || sz > buflen) | |
1923 | return ERR_PTR(-ENAMETOOLONG); | |
1924 | ||
1925 | buffer += buflen - sz; | |
1926 | return memcpy(buffer, temp, sz); | |
1927 | } | |
1928 | ||
6092d048 RP |
1929 | /* |
1930 | * Write full pathname from the root of the filesystem into the buffer. | |
1931 | */ | |
1932 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
1933 | { | |
1934 | char *end = buf + buflen; | |
1935 | char *retval; | |
1936 | ||
1937 | spin_lock(&dcache_lock); | |
1938 | prepend(&end, &buflen, "\0", 1); | |
1939 | if (!IS_ROOT(dentry) && d_unhashed(dentry) && | |
1940 | (prepend(&end, &buflen, "//deleted", 9) != 0)) | |
1941 | goto Elong; | |
1942 | if (buflen < 1) | |
1943 | goto Elong; | |
1944 | /* Get '/' right */ | |
1945 | retval = end-1; | |
1946 | *retval = '/'; | |
1947 | ||
cdd16d02 MS |
1948 | while (!IS_ROOT(dentry)) { |
1949 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 1950 | |
6092d048 | 1951 | prefetch(parent); |
cdd16d02 | 1952 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 RP |
1953 | (prepend(&end, &buflen, "/", 1) != 0)) |
1954 | goto Elong; | |
1955 | ||
1956 | retval = end; | |
1957 | dentry = parent; | |
1958 | } | |
1959 | spin_unlock(&dcache_lock); | |
1960 | return retval; | |
1961 | Elong: | |
1962 | spin_unlock(&dcache_lock); | |
1963 | return ERR_PTR(-ENAMETOOLONG); | |
1964 | } | |
1965 | ||
1da177e4 LT |
1966 | /* |
1967 | * NOTE! The user-level library version returns a | |
1968 | * character pointer. The kernel system call just | |
1969 | * returns the length of the buffer filled (which | |
1970 | * includes the ending '\0' character), or a negative | |
1971 | * error value. So libc would do something like | |
1972 | * | |
1973 | * char *getcwd(char * buf, size_t size) | |
1974 | * { | |
1975 | * int retval; | |
1976 | * | |
1977 | * retval = sys_getcwd(buf, size); | |
1978 | * if (retval >= 0) | |
1979 | * return buf; | |
1980 | * errno = -retval; | |
1981 | * return NULL; | |
1982 | * } | |
1983 | */ | |
1984 | asmlinkage long sys_getcwd(char __user *buf, unsigned long size) | |
1985 | { | |
552ce544 | 1986 | int error; |
6ac08c39 | 1987 | struct path pwd, root; |
552ce544 | 1988 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
1989 | |
1990 | if (!page) | |
1991 | return -ENOMEM; | |
1992 | ||
1993 | read_lock(¤t->fs->lock); | |
6ac08c39 | 1994 | pwd = current->fs->pwd; |
6092d048 | 1995 | path_get(&pwd); |
6ac08c39 | 1996 | root = current->fs->root; |
6092d048 | 1997 | path_get(&root); |
1da177e4 LT |
1998 | read_unlock(¤t->fs->lock); |
1999 | ||
552ce544 LT |
2000 | error = -ENOENT; |
2001 | /* Has the current directory has been unlinked? */ | |
2002 | spin_lock(&dcache_lock); | |
cdd16d02 | 2003 | if (IS_ROOT(pwd.dentry) || !d_unhashed(pwd.dentry)) { |
552ce544 | 2004 | unsigned long len; |
9d1bc601 | 2005 | struct path tmp = root; |
552ce544 | 2006 | char * cwd; |
1da177e4 | 2007 | |
9d1bc601 | 2008 | cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE); |
552ce544 LT |
2009 | spin_unlock(&dcache_lock); |
2010 | ||
2011 | error = PTR_ERR(cwd); | |
2012 | if (IS_ERR(cwd)) | |
2013 | goto out; | |
2014 | ||
2015 | error = -ERANGE; | |
2016 | len = PAGE_SIZE + page - cwd; | |
2017 | if (len <= size) { | |
2018 | error = len; | |
2019 | if (copy_to_user(buf, cwd, len)) | |
2020 | error = -EFAULT; | |
2021 | } | |
2022 | } else | |
2023 | spin_unlock(&dcache_lock); | |
1da177e4 LT |
2024 | |
2025 | out: | |
6ac08c39 JB |
2026 | path_put(&pwd); |
2027 | path_put(&root); | |
1da177e4 LT |
2028 | free_page((unsigned long) page); |
2029 | return error; | |
2030 | } | |
2031 | ||
2032 | /* | |
2033 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2034 | * | |
2035 | * Trivially implemented using the dcache structure | |
2036 | */ | |
2037 | ||
2038 | /** | |
2039 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2040 | * @new_dentry: new dentry | |
2041 | * @old_dentry: old dentry | |
2042 | * | |
2043 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2044 | * Returns 0 otherwise. | |
2045 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2046 | */ | |
2047 | ||
2048 | int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry) | |
2049 | { | |
2050 | int result; | |
2051 | struct dentry * saved = new_dentry; | |
2052 | unsigned long seq; | |
2053 | ||
2054 | /* need rcu_readlock to protect against the d_parent trashing due to | |
2055 | * d_move | |
2056 | */ | |
2057 | rcu_read_lock(); | |
2058 | do { | |
2059 | /* for restarting inner loop in case of seq retry */ | |
2060 | new_dentry = saved; | |
2061 | result = 0; | |
2062 | seq = read_seqbegin(&rename_lock); | |
2063 | for (;;) { | |
2064 | if (new_dentry != old_dentry) { | |
2065 | struct dentry * parent = new_dentry->d_parent; | |
2066 | if (parent == new_dentry) | |
2067 | break; | |
2068 | new_dentry = parent; | |
2069 | continue; | |
2070 | } | |
2071 | result = 1; | |
2072 | break; | |
2073 | } | |
2074 | } while (read_seqretry(&rename_lock, seq)); | |
2075 | rcu_read_unlock(); | |
2076 | ||
2077 | return result; | |
2078 | } | |
2079 | ||
2080 | void d_genocide(struct dentry *root) | |
2081 | { | |
2082 | struct dentry *this_parent = root; | |
2083 | struct list_head *next; | |
2084 | ||
2085 | spin_lock(&dcache_lock); | |
2086 | repeat: | |
2087 | next = this_parent->d_subdirs.next; | |
2088 | resume: | |
2089 | while (next != &this_parent->d_subdirs) { | |
2090 | struct list_head *tmp = next; | |
5160ee6f | 2091 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
2092 | next = tmp->next; |
2093 | if (d_unhashed(dentry)||!dentry->d_inode) | |
2094 | continue; | |
2095 | if (!list_empty(&dentry->d_subdirs)) { | |
2096 | this_parent = dentry; | |
2097 | goto repeat; | |
2098 | } | |
2099 | atomic_dec(&dentry->d_count); | |
2100 | } | |
2101 | if (this_parent != root) { | |
5160ee6f | 2102 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
2103 | atomic_dec(&this_parent->d_count); |
2104 | this_parent = this_parent->d_parent; | |
2105 | goto resume; | |
2106 | } | |
2107 | spin_unlock(&dcache_lock); | |
2108 | } | |
2109 | ||
2110 | /** | |
2111 | * find_inode_number - check for dentry with name | |
2112 | * @dir: directory to check | |
2113 | * @name: Name to find. | |
2114 | * | |
2115 | * Check whether a dentry already exists for the given name, | |
2116 | * and return the inode number if it has an inode. Otherwise | |
2117 | * 0 is returned. | |
2118 | * | |
2119 | * This routine is used to post-process directory listings for | |
2120 | * filesystems using synthetic inode numbers, and is necessary | |
2121 | * to keep getcwd() working. | |
2122 | */ | |
2123 | ||
2124 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2125 | { | |
2126 | struct dentry * dentry; | |
2127 | ino_t ino = 0; | |
2128 | ||
3e7e241f EB |
2129 | dentry = d_hash_and_lookup(dir, name); |
2130 | if (dentry) { | |
1da177e4 LT |
2131 | if (dentry->d_inode) |
2132 | ino = dentry->d_inode->i_ino; | |
2133 | dput(dentry); | |
2134 | } | |
1da177e4 LT |
2135 | return ino; |
2136 | } | |
2137 | ||
2138 | static __initdata unsigned long dhash_entries; | |
2139 | static int __init set_dhash_entries(char *str) | |
2140 | { | |
2141 | if (!str) | |
2142 | return 0; | |
2143 | dhash_entries = simple_strtoul(str, &str, 0); | |
2144 | return 1; | |
2145 | } | |
2146 | __setup("dhash_entries=", set_dhash_entries); | |
2147 | ||
2148 | static void __init dcache_init_early(void) | |
2149 | { | |
2150 | int loop; | |
2151 | ||
2152 | /* If hashes are distributed across NUMA nodes, defer | |
2153 | * hash allocation until vmalloc space is available. | |
2154 | */ | |
2155 | if (hashdist) | |
2156 | return; | |
2157 | ||
2158 | dentry_hashtable = | |
2159 | alloc_large_system_hash("Dentry cache", | |
2160 | sizeof(struct hlist_head), | |
2161 | dhash_entries, | |
2162 | 13, | |
2163 | HASH_EARLY, | |
2164 | &d_hash_shift, | |
2165 | &d_hash_mask, | |
2166 | 0); | |
2167 | ||
2168 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2169 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2170 | } | |
2171 | ||
74bf17cf | 2172 | static void __init dcache_init(void) |
1da177e4 LT |
2173 | { |
2174 | int loop; | |
2175 | ||
2176 | /* | |
2177 | * A constructor could be added for stable state like the lists, | |
2178 | * but it is probably not worth it because of the cache nature | |
2179 | * of the dcache. | |
2180 | */ | |
0a31bd5f CL |
2181 | dentry_cache = KMEM_CACHE(dentry, |
2182 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2183 | |
8e1f936b | 2184 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2185 | |
2186 | /* Hash may have been set up in dcache_init_early */ | |
2187 | if (!hashdist) | |
2188 | return; | |
2189 | ||
2190 | dentry_hashtable = | |
2191 | alloc_large_system_hash("Dentry cache", | |
2192 | sizeof(struct hlist_head), | |
2193 | dhash_entries, | |
2194 | 13, | |
2195 | 0, | |
2196 | &d_hash_shift, | |
2197 | &d_hash_mask, | |
2198 | 0); | |
2199 | ||
2200 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2201 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2202 | } | |
2203 | ||
2204 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2205 | struct kmem_cache *names_cachep __read_mostly; |
1da177e4 LT |
2206 | |
2207 | /* SLAB cache for file structures */ | |
e18b890b | 2208 | struct kmem_cache *filp_cachep __read_mostly; |
1da177e4 LT |
2209 | |
2210 | EXPORT_SYMBOL(d_genocide); | |
2211 | ||
1da177e4 LT |
2212 | void __init vfs_caches_init_early(void) |
2213 | { | |
2214 | dcache_init_early(); | |
2215 | inode_init_early(); | |
2216 | } | |
2217 | ||
2218 | void __init vfs_caches_init(unsigned long mempages) | |
2219 | { | |
2220 | unsigned long reserve; | |
2221 | ||
2222 | /* Base hash sizes on available memory, with a reserve equal to | |
2223 | 150% of current kernel size */ | |
2224 | ||
2225 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2226 | mempages -= reserve; | |
2227 | ||
2228 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2229 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 LT |
2230 | |
2231 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
20c2df83 | 2232 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2233 | |
74bf17cf DC |
2234 | dcache_init(); |
2235 | inode_init(); | |
1da177e4 | 2236 | files_init(mempages); |
74bf17cf | 2237 | mnt_init(); |
1da177e4 LT |
2238 | bdev_cache_init(); |
2239 | chrdev_init(); | |
2240 | } | |
2241 | ||
2242 | EXPORT_SYMBOL(d_alloc); | |
2243 | EXPORT_SYMBOL(d_alloc_anon); | |
2244 | EXPORT_SYMBOL(d_alloc_root); | |
2245 | EXPORT_SYMBOL(d_delete); | |
2246 | EXPORT_SYMBOL(d_find_alias); | |
2247 | EXPORT_SYMBOL(d_instantiate); | |
2248 | EXPORT_SYMBOL(d_invalidate); | |
2249 | EXPORT_SYMBOL(d_lookup); | |
2250 | EXPORT_SYMBOL(d_move); | |
770bfad8 | 2251 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
1da177e4 LT |
2252 | EXPORT_SYMBOL(d_path); |
2253 | EXPORT_SYMBOL(d_prune_aliases); | |
2254 | EXPORT_SYMBOL(d_rehash); | |
2255 | EXPORT_SYMBOL(d_splice_alias); | |
2256 | EXPORT_SYMBOL(d_validate); | |
2257 | EXPORT_SYMBOL(dget_locked); | |
2258 | EXPORT_SYMBOL(dput); | |
2259 | EXPORT_SYMBOL(find_inode_number); | |
2260 | EXPORT_SYMBOL(have_submounts); | |
2261 | EXPORT_SYMBOL(names_cachep); | |
2262 | EXPORT_SYMBOL(shrink_dcache_parent); | |
2263 | EXPORT_SYMBOL(shrink_dcache_sb); |