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 | |
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); | |
74c3cbe3 | 41 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 LT |
42 | |
43 | EXPORT_SYMBOL(dcache_lock); | |
44 | ||
e18b890b | 45 | static struct kmem_cache *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 | |
64 | /* Statistics gathering. */ | |
65 | struct dentry_stat_t dentry_stat = { | |
66 | .age_limit = 45, | |
67 | }; | |
68 | ||
b3423415 | 69 | static void __d_free(struct dentry *dentry) |
1da177e4 | 70 | { |
fd217f4d | 71 | WARN_ON(!list_empty(&dentry->d_alias)); |
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); | |
871c0067 OH |
177 | if (IS_ROOT(dentry)) |
178 | parent = NULL; | |
179 | else | |
180 | parent = dentry->d_parent; | |
d52b9086 | 181 | d_free(dentry); |
871c0067 | 182 | return parent; |
d52b9086 MS |
183 | } |
184 | ||
1da177e4 LT |
185 | /* |
186 | * This is dput | |
187 | * | |
188 | * This is complicated by the fact that we do not want to put | |
189 | * dentries that are no longer on any hash chain on the unused | |
190 | * list: we'd much rather just get rid of them immediately. | |
191 | * | |
192 | * However, that implies that we have to traverse the dentry | |
193 | * tree upwards to the parents which might _also_ now be | |
194 | * scheduled for deletion (it may have been only waiting for | |
195 | * its last child to go away). | |
196 | * | |
197 | * This tail recursion is done by hand as we don't want to depend | |
198 | * on the compiler to always get this right (gcc generally doesn't). | |
199 | * Real recursion would eat up our stack space. | |
200 | */ | |
201 | ||
202 | /* | |
203 | * dput - release a dentry | |
204 | * @dentry: dentry to release | |
205 | * | |
206 | * Release a dentry. This will drop the usage count and if appropriate | |
207 | * call the dentry unlink method as well as removing it from the queues and | |
208 | * releasing its resources. If the parent dentries were scheduled for release | |
209 | * they too may now get deleted. | |
210 | * | |
211 | * no dcache lock, please. | |
212 | */ | |
213 | ||
214 | void dput(struct dentry *dentry) | |
215 | { | |
216 | if (!dentry) | |
217 | return; | |
218 | ||
219 | repeat: | |
220 | if (atomic_read(&dentry->d_count) == 1) | |
221 | might_sleep(); | |
222 | if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock)) | |
223 | return; | |
224 | ||
225 | spin_lock(&dentry->d_lock); | |
226 | if (atomic_read(&dentry->d_count)) { | |
227 | spin_unlock(&dentry->d_lock); | |
228 | spin_unlock(&dcache_lock); | |
229 | return; | |
230 | } | |
231 | ||
232 | /* | |
233 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
234 | */ | |
235 | if (dentry->d_op && dentry->d_op->d_delete) { | |
236 | if (dentry->d_op->d_delete(dentry)) | |
237 | goto unhash_it; | |
238 | } | |
239 | /* Unreachable? Get rid of it */ | |
240 | if (d_unhashed(dentry)) | |
241 | goto kill_it; | |
242 | if (list_empty(&dentry->d_lru)) { | |
243 | dentry->d_flags |= DCACHE_REFERENCED; | |
da3bbdd4 | 244 | dentry_lru_add(dentry); |
1da177e4 LT |
245 | } |
246 | spin_unlock(&dentry->d_lock); | |
247 | spin_unlock(&dcache_lock); | |
248 | return; | |
249 | ||
250 | unhash_it: | |
251 | __d_drop(dentry); | |
d52b9086 | 252 | kill_it: |
da3bbdd4 KM |
253 | /* if dentry was on the d_lru list delete it from there */ |
254 | dentry_lru_del(dentry); | |
d52b9086 MS |
255 | dentry = d_kill(dentry); |
256 | if (dentry) | |
257 | goto repeat; | |
1da177e4 LT |
258 | } |
259 | ||
260 | /** | |
261 | * d_invalidate - invalidate a dentry | |
262 | * @dentry: dentry to invalidate | |
263 | * | |
264 | * Try to invalidate the dentry if it turns out to be | |
265 | * possible. If there are other dentries that can be | |
266 | * reached through this one we can't delete it and we | |
267 | * return -EBUSY. On success we return 0. | |
268 | * | |
269 | * no dcache lock. | |
270 | */ | |
271 | ||
272 | int d_invalidate(struct dentry * dentry) | |
273 | { | |
274 | /* | |
275 | * If it's already been dropped, return OK. | |
276 | */ | |
277 | spin_lock(&dcache_lock); | |
278 | if (d_unhashed(dentry)) { | |
279 | spin_unlock(&dcache_lock); | |
280 | return 0; | |
281 | } | |
282 | /* | |
283 | * Check whether to do a partial shrink_dcache | |
284 | * to get rid of unused child entries. | |
285 | */ | |
286 | if (!list_empty(&dentry->d_subdirs)) { | |
287 | spin_unlock(&dcache_lock); | |
288 | shrink_dcache_parent(dentry); | |
289 | spin_lock(&dcache_lock); | |
290 | } | |
291 | ||
292 | /* | |
293 | * Somebody else still using it? | |
294 | * | |
295 | * If it's a directory, we can't drop it | |
296 | * for fear of somebody re-populating it | |
297 | * with children (even though dropping it | |
298 | * would make it unreachable from the root, | |
299 | * we might still populate it if it was a | |
300 | * working directory or similar). | |
301 | */ | |
302 | spin_lock(&dentry->d_lock); | |
303 | if (atomic_read(&dentry->d_count) > 1) { | |
304 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { | |
305 | spin_unlock(&dentry->d_lock); | |
306 | spin_unlock(&dcache_lock); | |
307 | return -EBUSY; | |
308 | } | |
309 | } | |
310 | ||
311 | __d_drop(dentry); | |
312 | spin_unlock(&dentry->d_lock); | |
313 | spin_unlock(&dcache_lock); | |
314 | return 0; | |
315 | } | |
316 | ||
317 | /* This should be called _only_ with dcache_lock held */ | |
318 | ||
319 | static inline struct dentry * __dget_locked(struct dentry *dentry) | |
320 | { | |
321 | atomic_inc(&dentry->d_count); | |
da3bbdd4 | 322 | dentry_lru_del_init(dentry); |
1da177e4 LT |
323 | return dentry; |
324 | } | |
325 | ||
326 | struct dentry * dget_locked(struct dentry *dentry) | |
327 | { | |
328 | return __dget_locked(dentry); | |
329 | } | |
330 | ||
331 | /** | |
332 | * d_find_alias - grab a hashed alias of inode | |
333 | * @inode: inode in question | |
334 | * @want_discon: flag, used by d_splice_alias, to request | |
335 | * that only a DISCONNECTED alias be returned. | |
336 | * | |
337 | * If inode has a hashed alias, or is a directory and has any alias, | |
338 | * acquire the reference to alias and return it. Otherwise return NULL. | |
339 | * Notice that if inode is a directory there can be only one alias and | |
340 | * it can be unhashed only if it has no children, or if it is the root | |
341 | * of a filesystem. | |
342 | * | |
21c0d8fd | 343 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 344 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 345 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 LT |
346 | */ |
347 | ||
348 | static struct dentry * __d_find_alias(struct inode *inode, int want_discon) | |
349 | { | |
350 | struct list_head *head, *next, *tmp; | |
351 | struct dentry *alias, *discon_alias=NULL; | |
352 | ||
353 | head = &inode->i_dentry; | |
354 | next = inode->i_dentry.next; | |
355 | while (next != head) { | |
356 | tmp = next; | |
357 | next = tmp->next; | |
358 | prefetch(next); | |
359 | alias = list_entry(tmp, struct dentry, d_alias); | |
360 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
21c0d8fd N |
361 | if (IS_ROOT(alias) && |
362 | (alias->d_flags & DCACHE_DISCONNECTED)) | |
1da177e4 LT |
363 | discon_alias = alias; |
364 | else if (!want_discon) { | |
365 | __dget_locked(alias); | |
366 | return alias; | |
367 | } | |
368 | } | |
369 | } | |
370 | if (discon_alias) | |
371 | __dget_locked(discon_alias); | |
372 | return discon_alias; | |
373 | } | |
374 | ||
375 | struct dentry * d_find_alias(struct inode *inode) | |
376 | { | |
214fda1f DH |
377 | struct dentry *de = NULL; |
378 | ||
379 | if (!list_empty(&inode->i_dentry)) { | |
380 | spin_lock(&dcache_lock); | |
381 | de = __d_find_alias(inode, 0); | |
382 | spin_unlock(&dcache_lock); | |
383 | } | |
1da177e4 LT |
384 | return de; |
385 | } | |
386 | ||
387 | /* | |
388 | * Try to kill dentries associated with this inode. | |
389 | * WARNING: you must own a reference to inode. | |
390 | */ | |
391 | void d_prune_aliases(struct inode *inode) | |
392 | { | |
0cdca3f9 | 393 | struct dentry *dentry; |
1da177e4 LT |
394 | restart: |
395 | spin_lock(&dcache_lock); | |
0cdca3f9 | 396 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 LT |
397 | spin_lock(&dentry->d_lock); |
398 | if (!atomic_read(&dentry->d_count)) { | |
399 | __dget_locked(dentry); | |
400 | __d_drop(dentry); | |
401 | spin_unlock(&dentry->d_lock); | |
402 | spin_unlock(&dcache_lock); | |
403 | dput(dentry); | |
404 | goto restart; | |
405 | } | |
406 | spin_unlock(&dentry->d_lock); | |
407 | } | |
408 | spin_unlock(&dcache_lock); | |
409 | } | |
410 | ||
411 | /* | |
d702ccb3 AM |
412 | * Throw away a dentry - free the inode, dput the parent. This requires that |
413 | * the LRU list has already been removed. | |
414 | * | |
85864e10 MS |
415 | * Try to prune ancestors as well. This is necessary to prevent |
416 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
417 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 418 | */ |
85864e10 | 419 | static void prune_one_dentry(struct dentry * dentry) |
31f3e0b3 MS |
420 | __releases(dentry->d_lock) |
421 | __releases(dcache_lock) | |
422 | __acquires(dcache_lock) | |
1da177e4 | 423 | { |
1da177e4 | 424 | __d_drop(dentry); |
d52b9086 | 425 | dentry = d_kill(dentry); |
d52b9086 MS |
426 | |
427 | /* | |
428 | * Prune ancestors. Locking is simpler than in dput(), | |
429 | * because dcache_lock needs to be taken anyway. | |
430 | */ | |
1da177e4 | 431 | spin_lock(&dcache_lock); |
d52b9086 MS |
432 | while (dentry) { |
433 | if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock)) | |
434 | return; | |
435 | ||
436 | if (dentry->d_op && dentry->d_op->d_delete) | |
437 | dentry->d_op->d_delete(dentry); | |
da3bbdd4 | 438 | dentry_lru_del_init(dentry); |
d52b9086 MS |
439 | __d_drop(dentry); |
440 | dentry = d_kill(dentry); | |
441 | spin_lock(&dcache_lock); | |
442 | } | |
1da177e4 LT |
443 | } |
444 | ||
da3bbdd4 KM |
445 | /* |
446 | * Shrink the dentry LRU on a given superblock. | |
447 | * @sb : superblock to shrink dentry LRU. | |
448 | * @count: If count is NULL, we prune all dentries on superblock. | |
449 | * @flags: If flags is non-zero, we need to do special processing based on | |
450 | * which flags are set. This means we don't need to maintain multiple | |
451 | * similar copies of this loop. | |
1da177e4 | 452 | */ |
da3bbdd4 | 453 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) |
1da177e4 | 454 | { |
da3bbdd4 KM |
455 | LIST_HEAD(referenced); |
456 | LIST_HEAD(tmp); | |
457 | struct dentry *dentry; | |
458 | int cnt = 0; | |
1da177e4 | 459 | |
da3bbdd4 KM |
460 | BUG_ON(!sb); |
461 | BUG_ON((flags & DCACHE_REFERENCED) && count == NULL); | |
462 | spin_lock(&dcache_lock); | |
463 | if (count != NULL) | |
464 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
465 | cnt = *count; | |
466 | restart: | |
467 | if (count == NULL) | |
468 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
469 | else { | |
470 | while (!list_empty(&sb->s_dentry_lru)) { | |
471 | dentry = list_entry(sb->s_dentry_lru.prev, | |
472 | struct dentry, d_lru); | |
473 | BUG_ON(dentry->d_sb != sb); | |
474 | ||
475 | spin_lock(&dentry->d_lock); | |
476 | /* | |
477 | * If we are honouring the DCACHE_REFERENCED flag and | |
478 | * the dentry has this flag set, don't free it. Clear | |
479 | * the flag and put it back on the LRU. | |
0feae5c4 | 480 | */ |
da3bbdd4 KM |
481 | if ((flags & DCACHE_REFERENCED) |
482 | && (dentry->d_flags & DCACHE_REFERENCED)) { | |
483 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
484 | list_move_tail(&dentry->d_lru, &referenced); | |
485 | spin_unlock(&dentry->d_lock); | |
486 | } else { | |
487 | list_move_tail(&dentry->d_lru, &tmp); | |
488 | spin_unlock(&dentry->d_lock); | |
489 | cnt--; | |
490 | if (!cnt) | |
491 | break; | |
0feae5c4 | 492 | } |
f3c6ba98 | 493 | cond_resched_lock(&dcache_lock); |
0feae5c4 | 494 | } |
da3bbdd4 KM |
495 | } |
496 | while (!list_empty(&tmp)) { | |
497 | dentry = list_entry(tmp.prev, struct dentry, d_lru); | |
498 | dentry_lru_del_init(dentry); | |
499 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
500 | /* |
501 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
502 | * the LRU because of laziness during lookup. Do not free |
503 | * it - just keep it off the LRU list. | |
1da177e4 | 504 | */ |
da3bbdd4 KM |
505 | if (atomic_read(&dentry->d_count)) { |
506 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
507 | continue; |
508 | } | |
da3bbdd4 KM |
509 | prune_one_dentry(dentry); |
510 | /* dentry->d_lock was dropped in prune_one_dentry() */ | |
511 | cond_resched_lock(&dcache_lock); | |
512 | } | |
513 | if (count == NULL && !list_empty(&sb->s_dentry_lru)) | |
514 | goto restart; | |
515 | if (count != NULL) | |
516 | *count = cnt; | |
517 | if (!list_empty(&referenced)) | |
518 | list_splice(&referenced, &sb->s_dentry_lru); | |
519 | spin_unlock(&dcache_lock); | |
520 | } | |
521 | ||
522 | /** | |
523 | * prune_dcache - shrink the dcache | |
524 | * @count: number of entries to try to free | |
525 | * | |
526 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
527 | * need to unmount something (at which point we need to unuse all dentries). | |
528 | * | |
529 | * This function may fail to free any resources if all the dentries are in use. | |
530 | */ | |
531 | static void prune_dcache(int count) | |
532 | { | |
533 | struct super_block *sb; | |
534 | int w_count; | |
535 | int unused = dentry_stat.nr_unused; | |
536 | int prune_ratio; | |
537 | int pruned; | |
538 | ||
539 | if (unused == 0 || count == 0) | |
540 | return; | |
541 | spin_lock(&dcache_lock); | |
542 | restart: | |
543 | if (count >= unused) | |
544 | prune_ratio = 1; | |
545 | else | |
546 | prune_ratio = unused / count; | |
547 | spin_lock(&sb_lock); | |
548 | list_for_each_entry(sb, &super_blocks, s_list) { | |
549 | if (sb->s_nr_dentry_unused == 0) | |
1da177e4 | 550 | continue; |
da3bbdd4 KM |
551 | sb->s_count++; |
552 | /* Now, we reclaim unused dentrins with fairness. | |
553 | * We reclaim them same percentage from each superblock. | |
554 | * We calculate number of dentries to scan on this sb | |
555 | * as follows, but the implementation is arranged to avoid | |
556 | * overflows: | |
557 | * number of dentries to scan on this sb = | |
558 | * count * (number of dentries on this sb / | |
559 | * number of dentries in the machine) | |
0feae5c4 | 560 | */ |
da3bbdd4 KM |
561 | spin_unlock(&sb_lock); |
562 | if (prune_ratio != 1) | |
563 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
564 | else | |
565 | w_count = sb->s_nr_dentry_unused; | |
566 | pruned = w_count; | |
0feae5c4 | 567 | /* |
da3bbdd4 KM |
568 | * We need to be sure this filesystem isn't being unmounted, |
569 | * otherwise we could race with generic_shutdown_super(), and | |
570 | * end up holding a reference to an inode while the filesystem | |
571 | * is unmounted. So we try to get s_umount, and make sure | |
572 | * s_root isn't NULL. | |
0feae5c4 | 573 | */ |
da3bbdd4 KM |
574 | if (down_read_trylock(&sb->s_umount)) { |
575 | if ((sb->s_root != NULL) && | |
576 | (!list_empty(&sb->s_dentry_lru))) { | |
577 | spin_unlock(&dcache_lock); | |
578 | __shrink_dcache_sb(sb, &w_count, | |
579 | DCACHE_REFERENCED); | |
580 | pruned -= w_count; | |
581 | spin_lock(&dcache_lock); | |
0feae5c4 | 582 | } |
da3bbdd4 | 583 | up_read(&sb->s_umount); |
0feae5c4 | 584 | } |
da3bbdd4 KM |
585 | spin_lock(&sb_lock); |
586 | count -= pruned; | |
6eac3f93 | 587 | /* |
da3bbdd4 KM |
588 | * restart only when sb is no longer on the list and |
589 | * we have more work to do. | |
0feae5c4 | 590 | */ |
da3bbdd4 KM |
591 | if (__put_super_and_need_restart(sb) && count > 0) { |
592 | spin_unlock(&sb_lock); | |
593 | goto restart; | |
594 | } | |
1da177e4 | 595 | } |
da3bbdd4 | 596 | spin_unlock(&sb_lock); |
1da177e4 LT |
597 | spin_unlock(&dcache_lock); |
598 | } | |
599 | ||
1da177e4 LT |
600 | /** |
601 | * shrink_dcache_sb - shrink dcache for a superblock | |
602 | * @sb: superblock | |
603 | * | |
604 | * Shrink the dcache for the specified super block. This | |
605 | * is used to free the dcache before unmounting a file | |
606 | * system | |
607 | */ | |
1da177e4 LT |
608 | void shrink_dcache_sb(struct super_block * sb) |
609 | { | |
da3bbdd4 | 610 | __shrink_dcache_sb(sb, NULL, 0); |
1da177e4 LT |
611 | } |
612 | ||
c636ebdb DH |
613 | /* |
614 | * destroy a single subtree of dentries for unmount | |
615 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
616 | * locking | |
617 | */ | |
618 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
619 | { | |
620 | struct dentry *parent; | |
f8713576 | 621 | unsigned detached = 0; |
c636ebdb DH |
622 | |
623 | BUG_ON(!IS_ROOT(dentry)); | |
624 | ||
625 | /* detach this root from the system */ | |
626 | spin_lock(&dcache_lock); | |
da3bbdd4 | 627 | dentry_lru_del_init(dentry); |
c636ebdb DH |
628 | __d_drop(dentry); |
629 | spin_unlock(&dcache_lock); | |
630 | ||
631 | for (;;) { | |
632 | /* descend to the first leaf in the current subtree */ | |
633 | while (!list_empty(&dentry->d_subdirs)) { | |
634 | struct dentry *loop; | |
635 | ||
636 | /* this is a branch with children - detach all of them | |
637 | * from the system in one go */ | |
638 | spin_lock(&dcache_lock); | |
639 | list_for_each_entry(loop, &dentry->d_subdirs, | |
640 | d_u.d_child) { | |
da3bbdd4 | 641 | dentry_lru_del_init(loop); |
c636ebdb DH |
642 | __d_drop(loop); |
643 | cond_resched_lock(&dcache_lock); | |
644 | } | |
645 | spin_unlock(&dcache_lock); | |
646 | ||
647 | /* move to the first child */ | |
648 | dentry = list_entry(dentry->d_subdirs.next, | |
649 | struct dentry, d_u.d_child); | |
650 | } | |
651 | ||
652 | /* consume the dentries from this leaf up through its parents | |
653 | * until we find one with children or run out altogether */ | |
654 | do { | |
655 | struct inode *inode; | |
656 | ||
657 | if (atomic_read(&dentry->d_count) != 0) { | |
658 | printk(KERN_ERR | |
659 | "BUG: Dentry %p{i=%lx,n=%s}" | |
660 | " still in use (%d)" | |
661 | " [unmount of %s %s]\n", | |
662 | dentry, | |
663 | dentry->d_inode ? | |
664 | dentry->d_inode->i_ino : 0UL, | |
665 | dentry->d_name.name, | |
666 | atomic_read(&dentry->d_count), | |
667 | dentry->d_sb->s_type->name, | |
668 | dentry->d_sb->s_id); | |
669 | BUG(); | |
670 | } | |
671 | ||
871c0067 | 672 | if (IS_ROOT(dentry)) |
c636ebdb | 673 | parent = NULL; |
871c0067 OH |
674 | else { |
675 | parent = dentry->d_parent; | |
c636ebdb | 676 | atomic_dec(&parent->d_count); |
871c0067 | 677 | } |
c636ebdb DH |
678 | |
679 | list_del(&dentry->d_u.d_child); | |
f8713576 | 680 | detached++; |
c636ebdb DH |
681 | |
682 | inode = dentry->d_inode; | |
683 | if (inode) { | |
684 | dentry->d_inode = NULL; | |
685 | list_del_init(&dentry->d_alias); | |
686 | if (dentry->d_op && dentry->d_op->d_iput) | |
687 | dentry->d_op->d_iput(dentry, inode); | |
688 | else | |
689 | iput(inode); | |
690 | } | |
691 | ||
692 | d_free(dentry); | |
693 | ||
694 | /* finished when we fall off the top of the tree, | |
695 | * otherwise we ascend to the parent and move to the | |
696 | * next sibling if there is one */ | |
697 | if (!parent) | |
f8713576 | 698 | goto out; |
c636ebdb DH |
699 | |
700 | dentry = parent; | |
701 | ||
702 | } while (list_empty(&dentry->d_subdirs)); | |
703 | ||
704 | dentry = list_entry(dentry->d_subdirs.next, | |
705 | struct dentry, d_u.d_child); | |
706 | } | |
f8713576 DH |
707 | out: |
708 | /* several dentries were freed, need to correct nr_dentry */ | |
709 | spin_lock(&dcache_lock); | |
710 | dentry_stat.nr_dentry -= detached; | |
711 | spin_unlock(&dcache_lock); | |
c636ebdb DH |
712 | } |
713 | ||
714 | /* | |
715 | * destroy the dentries attached to a superblock on unmounting | |
716 | * - we don't need to use dentry->d_lock, and only need dcache_lock when | |
717 | * removing the dentry from the system lists and hashes because: | |
718 | * - the superblock is detached from all mountings and open files, so the | |
719 | * dentry trees will not be rearranged by the VFS | |
720 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
721 | * any dentries belonging to this superblock that it comes across | |
722 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
723 | * in this superblock | |
724 | */ | |
725 | void shrink_dcache_for_umount(struct super_block *sb) | |
726 | { | |
727 | struct dentry *dentry; | |
728 | ||
729 | if (down_read_trylock(&sb->s_umount)) | |
730 | BUG(); | |
731 | ||
732 | dentry = sb->s_root; | |
733 | sb->s_root = NULL; | |
734 | atomic_dec(&dentry->d_count); | |
735 | shrink_dcache_for_umount_subtree(dentry); | |
736 | ||
737 | while (!hlist_empty(&sb->s_anon)) { | |
738 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
739 | shrink_dcache_for_umount_subtree(dentry); | |
740 | } | |
741 | } | |
742 | ||
1da177e4 LT |
743 | /* |
744 | * Search for at least 1 mount point in the dentry's subdirs. | |
745 | * We descend to the next level whenever the d_subdirs | |
746 | * list is non-empty and continue searching. | |
747 | */ | |
748 | ||
749 | /** | |
750 | * have_submounts - check for mounts over a dentry | |
751 | * @parent: dentry to check. | |
752 | * | |
753 | * Return true if the parent or its subdirectories contain | |
754 | * a mount point | |
755 | */ | |
756 | ||
757 | int have_submounts(struct dentry *parent) | |
758 | { | |
759 | struct dentry *this_parent = parent; | |
760 | struct list_head *next; | |
761 | ||
762 | spin_lock(&dcache_lock); | |
763 | if (d_mountpoint(parent)) | |
764 | goto positive; | |
765 | repeat: | |
766 | next = this_parent->d_subdirs.next; | |
767 | resume: | |
768 | while (next != &this_parent->d_subdirs) { | |
769 | struct list_head *tmp = next; | |
5160ee6f | 770 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
771 | next = tmp->next; |
772 | /* Have we found a mount point ? */ | |
773 | if (d_mountpoint(dentry)) | |
774 | goto positive; | |
775 | if (!list_empty(&dentry->d_subdirs)) { | |
776 | this_parent = dentry; | |
777 | goto repeat; | |
778 | } | |
779 | } | |
780 | /* | |
781 | * All done at this level ... ascend and resume the search. | |
782 | */ | |
783 | if (this_parent != parent) { | |
5160ee6f | 784 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
785 | this_parent = this_parent->d_parent; |
786 | goto resume; | |
787 | } | |
788 | spin_unlock(&dcache_lock); | |
789 | return 0; /* No mount points found in tree */ | |
790 | positive: | |
791 | spin_unlock(&dcache_lock); | |
792 | return 1; | |
793 | } | |
794 | ||
795 | /* | |
796 | * Search the dentry child list for the specified parent, | |
797 | * and move any unused dentries to the end of the unused | |
798 | * list for prune_dcache(). We descend to the next level | |
799 | * whenever the d_subdirs list is non-empty and continue | |
800 | * searching. | |
801 | * | |
802 | * It returns zero iff there are no unused children, | |
803 | * otherwise it returns the number of children moved to | |
804 | * the end of the unused list. This may not be the total | |
805 | * number of unused children, because select_parent can | |
806 | * drop the lock and return early due to latency | |
807 | * constraints. | |
808 | */ | |
809 | static int select_parent(struct dentry * parent) | |
810 | { | |
811 | struct dentry *this_parent = parent; | |
812 | struct list_head *next; | |
813 | int found = 0; | |
814 | ||
815 | spin_lock(&dcache_lock); | |
816 | repeat: | |
817 | next = this_parent->d_subdirs.next; | |
818 | resume: | |
819 | while (next != &this_parent->d_subdirs) { | |
820 | struct list_head *tmp = next; | |
5160ee6f | 821 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
822 | next = tmp->next; |
823 | ||
da3bbdd4 | 824 | dentry_lru_del_init(dentry); |
1da177e4 LT |
825 | /* |
826 | * move only zero ref count dentries to the end | |
827 | * of the unused list for prune_dcache | |
828 | */ | |
829 | if (!atomic_read(&dentry->d_count)) { | |
da3bbdd4 | 830 | dentry_lru_add_tail(dentry); |
1da177e4 LT |
831 | found++; |
832 | } | |
833 | ||
834 | /* | |
835 | * We can return to the caller if we have found some (this | |
836 | * ensures forward progress). We'll be coming back to find | |
837 | * the rest. | |
838 | */ | |
839 | if (found && need_resched()) | |
840 | goto out; | |
841 | ||
842 | /* | |
843 | * Descend a level if the d_subdirs list is non-empty. | |
844 | */ | |
845 | if (!list_empty(&dentry->d_subdirs)) { | |
846 | this_parent = dentry; | |
1da177e4 LT |
847 | goto repeat; |
848 | } | |
849 | } | |
850 | /* | |
851 | * All done at this level ... ascend and resume the search. | |
852 | */ | |
853 | if (this_parent != parent) { | |
5160ee6f | 854 | next = this_parent->d_u.d_child.next; |
1da177e4 | 855 | this_parent = this_parent->d_parent; |
1da177e4 LT |
856 | goto resume; |
857 | } | |
858 | out: | |
859 | spin_unlock(&dcache_lock); | |
860 | return found; | |
861 | } | |
862 | ||
863 | /** | |
864 | * shrink_dcache_parent - prune dcache | |
865 | * @parent: parent of entries to prune | |
866 | * | |
867 | * Prune the dcache to remove unused children of the parent dentry. | |
868 | */ | |
869 | ||
870 | void shrink_dcache_parent(struct dentry * parent) | |
871 | { | |
da3bbdd4 | 872 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
873 | int found; |
874 | ||
875 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 876 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 LT |
877 | } |
878 | ||
1da177e4 LT |
879 | /* |
880 | * Scan `nr' dentries and return the number which remain. | |
881 | * | |
882 | * We need to avoid reentering the filesystem if the caller is performing a | |
883 | * GFP_NOFS allocation attempt. One example deadlock is: | |
884 | * | |
885 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
886 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
887 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
888 | * | |
889 | * In this case we return -1 to tell the caller that we baled. | |
890 | */ | |
27496a8c | 891 | static int shrink_dcache_memory(int nr, gfp_t gfp_mask) |
1da177e4 LT |
892 | { |
893 | if (nr) { | |
894 | if (!(gfp_mask & __GFP_FS)) | |
895 | return -1; | |
da3bbdd4 | 896 | prune_dcache(nr); |
1da177e4 LT |
897 | } |
898 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
899 | } | |
900 | ||
8e1f936b RR |
901 | static struct shrinker dcache_shrinker = { |
902 | .shrink = shrink_dcache_memory, | |
903 | .seeks = DEFAULT_SEEKS, | |
904 | }; | |
905 | ||
1da177e4 LT |
906 | /** |
907 | * d_alloc - allocate a dcache entry | |
908 | * @parent: parent of entry to allocate | |
909 | * @name: qstr of the name | |
910 | * | |
911 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
912 | * available. On a success the dentry is returned. The name passed in is | |
913 | * copied and the copy passed in may be reused after this call. | |
914 | */ | |
915 | ||
916 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
917 | { | |
918 | struct dentry *dentry; | |
919 | char *dname; | |
920 | ||
e12ba74d | 921 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
922 | if (!dentry) |
923 | return NULL; | |
924 | ||
925 | if (name->len > DNAME_INLINE_LEN-1) { | |
926 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
927 | if (!dname) { | |
928 | kmem_cache_free(dentry_cache, dentry); | |
929 | return NULL; | |
930 | } | |
931 | } else { | |
932 | dname = dentry->d_iname; | |
933 | } | |
934 | dentry->d_name.name = dname; | |
935 | ||
936 | dentry->d_name.len = name->len; | |
937 | dentry->d_name.hash = name->hash; | |
938 | memcpy(dname, name->name, name->len); | |
939 | dname[name->len] = 0; | |
940 | ||
941 | atomic_set(&dentry->d_count, 1); | |
942 | dentry->d_flags = DCACHE_UNHASHED; | |
943 | spin_lock_init(&dentry->d_lock); | |
944 | dentry->d_inode = NULL; | |
945 | dentry->d_parent = NULL; | |
946 | dentry->d_sb = NULL; | |
947 | dentry->d_op = NULL; | |
948 | dentry->d_fsdata = NULL; | |
949 | dentry->d_mounted = 0; | |
1da177e4 LT |
950 | INIT_HLIST_NODE(&dentry->d_hash); |
951 | INIT_LIST_HEAD(&dentry->d_lru); | |
952 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
953 | INIT_LIST_HEAD(&dentry->d_alias); | |
954 | ||
955 | if (parent) { | |
956 | dentry->d_parent = dget(parent); | |
957 | dentry->d_sb = parent->d_sb; | |
958 | } else { | |
5160ee6f | 959 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
960 | } |
961 | ||
962 | spin_lock(&dcache_lock); | |
963 | if (parent) | |
5160ee6f | 964 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1da177e4 LT |
965 | dentry_stat.nr_dentry++; |
966 | spin_unlock(&dcache_lock); | |
967 | ||
968 | return dentry; | |
969 | } | |
970 | ||
971 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
972 | { | |
973 | struct qstr q; | |
974 | ||
975 | q.name = name; | |
976 | q.len = strlen(name); | |
977 | q.hash = full_name_hash(q.name, q.len); | |
978 | return d_alloc(parent, &q); | |
979 | } | |
980 | ||
360da900 OH |
981 | /* the caller must hold dcache_lock */ |
982 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | |
983 | { | |
984 | if (inode) | |
985 | list_add(&dentry->d_alias, &inode->i_dentry); | |
986 | dentry->d_inode = inode; | |
987 | fsnotify_d_instantiate(dentry, inode); | |
988 | } | |
989 | ||
1da177e4 LT |
990 | /** |
991 | * d_instantiate - fill in inode information for a dentry | |
992 | * @entry: dentry to complete | |
993 | * @inode: inode to attach to this dentry | |
994 | * | |
995 | * Fill in inode information in the entry. | |
996 | * | |
997 | * This turns negative dentries into productive full members | |
998 | * of society. | |
999 | * | |
1000 | * NOTE! This assumes that the inode count has been incremented | |
1001 | * (or otherwise set) by the caller to indicate that it is now | |
1002 | * in use by the dcache. | |
1003 | */ | |
1004 | ||
1005 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1006 | { | |
28133c7b | 1007 | BUG_ON(!list_empty(&entry->d_alias)); |
1da177e4 | 1008 | spin_lock(&dcache_lock); |
360da900 | 1009 | __d_instantiate(entry, inode); |
1da177e4 LT |
1010 | spin_unlock(&dcache_lock); |
1011 | security_d_instantiate(entry, inode); | |
1012 | } | |
1013 | ||
1014 | /** | |
1015 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1016 | * @entry: dentry to instantiate | |
1017 | * @inode: inode to attach to this dentry | |
1018 | * | |
1019 | * Fill in inode information in the entry. On success, it returns NULL. | |
1020 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1021 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1022 | * |
1023 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1024 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1025 | * |
1026 | * This also assumes that the inode count has been incremented | |
1027 | * (or otherwise set) by the caller to indicate that it is now | |
1028 | * in use by the dcache. | |
1da177e4 | 1029 | */ |
770bfad8 DH |
1030 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1031 | struct inode *inode) | |
1da177e4 LT |
1032 | { |
1033 | struct dentry *alias; | |
1034 | int len = entry->d_name.len; | |
1035 | const char *name = entry->d_name.name; | |
1036 | unsigned int hash = entry->d_name.hash; | |
1037 | ||
770bfad8 | 1038 | if (!inode) { |
360da900 | 1039 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1040 | return NULL; |
1041 | } | |
1042 | ||
1da177e4 LT |
1043 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1044 | struct qstr *qstr = &alias->d_name; | |
1045 | ||
1046 | if (qstr->hash != hash) | |
1047 | continue; | |
1048 | if (alias->d_parent != entry->d_parent) | |
1049 | continue; | |
1050 | if (qstr->len != len) | |
1051 | continue; | |
1052 | if (memcmp(qstr->name, name, len)) | |
1053 | continue; | |
1054 | dget_locked(alias); | |
1da177e4 LT |
1055 | return alias; |
1056 | } | |
770bfad8 | 1057 | |
360da900 | 1058 | __d_instantiate(entry, inode); |
1da177e4 LT |
1059 | return NULL; |
1060 | } | |
770bfad8 DH |
1061 | |
1062 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1063 | { | |
1064 | struct dentry *result; | |
1065 | ||
1066 | BUG_ON(!list_empty(&entry->d_alias)); | |
1067 | ||
1068 | spin_lock(&dcache_lock); | |
1069 | result = __d_instantiate_unique(entry, inode); | |
1070 | spin_unlock(&dcache_lock); | |
1071 | ||
1072 | if (!result) { | |
1073 | security_d_instantiate(entry, inode); | |
1074 | return NULL; | |
1075 | } | |
1076 | ||
1077 | BUG_ON(!d_unhashed(result)); | |
1078 | iput(inode); | |
1079 | return result; | |
1080 | } | |
1081 | ||
1da177e4 LT |
1082 | EXPORT_SYMBOL(d_instantiate_unique); |
1083 | ||
1084 | /** | |
1085 | * d_alloc_root - allocate root dentry | |
1086 | * @root_inode: inode to allocate the root for | |
1087 | * | |
1088 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1089 | * instantiated and returned. %NULL is returned if there is insufficient | |
1090 | * memory or the inode passed is %NULL. | |
1091 | */ | |
1092 | ||
1093 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1094 | { | |
1095 | struct dentry *res = NULL; | |
1096 | ||
1097 | if (root_inode) { | |
1098 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1099 | ||
1100 | res = d_alloc(NULL, &name); | |
1101 | if (res) { | |
1102 | res->d_sb = root_inode->i_sb; | |
1103 | res->d_parent = res; | |
1104 | d_instantiate(res, root_inode); | |
1105 | } | |
1106 | } | |
1107 | return res; | |
1108 | } | |
1109 | ||
1110 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1111 | unsigned long hash) | |
1112 | { | |
1113 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1114 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1115 | return dentry_hashtable + (hash & D_HASHMASK); | |
1116 | } | |
1117 | ||
4ea3ada2 CH |
1118 | /** |
1119 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1120 | * @inode: inode to allocate the dentry for | |
1121 | * | |
1122 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1123 | * similar open by handle operations. The returned dentry may be anonymous, | |
1124 | * or may have a full name (if the inode was already in the cache). | |
1125 | * | |
1126 | * When called on a directory inode, we must ensure that the inode only ever | |
1127 | * has one dentry. If a dentry is found, that is returned instead of | |
1128 | * allocating a new one. | |
1129 | * | |
1130 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1131 | * to the dentry. In case of an error the reference on the inode is released. |
1132 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1133 | * be passed in and will be the error will be propagate to the return value, | |
1134 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1135 | */ |
1136 | struct dentry *d_obtain_alias(struct inode *inode) | |
1137 | { | |
9308a612 CH |
1138 | static const struct qstr anonstring = { .name = "" }; |
1139 | struct dentry *tmp; | |
1140 | struct dentry *res; | |
4ea3ada2 CH |
1141 | |
1142 | if (!inode) | |
44003728 | 1143 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1144 | if (IS_ERR(inode)) |
1145 | return ERR_CAST(inode); | |
1146 | ||
9308a612 CH |
1147 | res = d_find_alias(inode); |
1148 | if (res) | |
1149 | goto out_iput; | |
1150 | ||
1151 | tmp = d_alloc(NULL, &anonstring); | |
1152 | if (!tmp) { | |
1153 | res = ERR_PTR(-ENOMEM); | |
1154 | goto out_iput; | |
4ea3ada2 | 1155 | } |
9308a612 CH |
1156 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1157 | ||
1158 | spin_lock(&dcache_lock); | |
1159 | res = __d_find_alias(inode, 0); | |
1160 | if (res) { | |
1161 | spin_unlock(&dcache_lock); | |
1162 | dput(tmp); | |
1163 | goto out_iput; | |
1164 | } | |
1165 | ||
1166 | /* attach a disconnected dentry */ | |
1167 | spin_lock(&tmp->d_lock); | |
1168 | tmp->d_sb = inode->i_sb; | |
1169 | tmp->d_inode = inode; | |
1170 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1171 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1172 | list_add(&tmp->d_alias, &inode->i_dentry); | |
1173 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); | |
1174 | spin_unlock(&tmp->d_lock); | |
1175 | ||
1176 | spin_unlock(&dcache_lock); | |
1177 | return tmp; | |
1178 | ||
1179 | out_iput: | |
1180 | iput(inode); | |
1181 | return res; | |
4ea3ada2 | 1182 | } |
adc48720 | 1183 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1184 | |
1185 | /** | |
1186 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1187 | * @inode: the inode which may have a disconnected dentry | |
1188 | * @dentry: a negative dentry which we want to point to the inode. | |
1189 | * | |
1190 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1191 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1192 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1193 | * | |
1194 | * This is needed in the lookup routine of any filesystem that is exportable | |
1195 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1196 | * | |
1197 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1198 | * is returned. This matches the expected return value of ->lookup. | |
1199 | * | |
1200 | */ | |
1201 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1202 | { | |
1203 | struct dentry *new = NULL; | |
1204 | ||
21c0d8fd | 1205 | if (inode && S_ISDIR(inode->i_mode)) { |
1da177e4 LT |
1206 | spin_lock(&dcache_lock); |
1207 | new = __d_find_alias(inode, 1); | |
1208 | if (new) { | |
1209 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
1210 | spin_unlock(&dcache_lock); | |
1211 | security_d_instantiate(new, inode); | |
1212 | d_rehash(dentry); | |
1213 | d_move(new, dentry); | |
1214 | iput(inode); | |
1215 | } else { | |
360da900 OH |
1216 | /* already taking dcache_lock, so d_add() by hand */ |
1217 | __d_instantiate(dentry, inode); | |
1da177e4 LT |
1218 | spin_unlock(&dcache_lock); |
1219 | security_d_instantiate(dentry, inode); | |
1220 | d_rehash(dentry); | |
1221 | } | |
1222 | } else | |
1223 | d_add(dentry, inode); | |
1224 | return new; | |
1225 | } | |
1226 | ||
9403540c BN |
1227 | /** |
1228 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1229 | * @inode: the inode case-insensitive lookup has found | |
1230 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1231 | * @name: the case-exact name to be associated with the returned dentry | |
1232 | * | |
1233 | * This is to avoid filling the dcache with case-insensitive names to the | |
1234 | * same inode, only the actual correct case is stored in the dcache for | |
1235 | * case-insensitive filesystems. | |
1236 | * | |
1237 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1238 | * already exists in in the dcache, use it and return it. | |
1239 | * | |
1240 | * If no entry exists with the exact case name, allocate new dentry with | |
1241 | * the exact case, and return the spliced entry. | |
1242 | */ | |
e45b590b | 1243 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1244 | struct qstr *name) |
1245 | { | |
1246 | int error; | |
1247 | struct dentry *found; | |
1248 | struct dentry *new; | |
1249 | ||
1250 | /* Does a dentry matching the name exist already? */ | |
1251 | found = d_hash_and_lookup(dentry->d_parent, name); | |
1252 | /* If not, create it now and return */ | |
1253 | if (!found) { | |
1254 | new = d_alloc(dentry->d_parent, name); | |
1255 | if (!new) { | |
1256 | error = -ENOMEM; | |
1257 | goto err_out; | |
1258 | } | |
1259 | found = d_splice_alias(inode, new); | |
1260 | if (found) { | |
1261 | dput(new); | |
1262 | return found; | |
1263 | } | |
1264 | return new; | |
1265 | } | |
1266 | /* Matching dentry exists, check if it is negative. */ | |
1267 | if (found->d_inode) { | |
1268 | if (unlikely(found->d_inode != inode)) { | |
1269 | /* This can't happen because bad inodes are unhashed. */ | |
1270 | BUG_ON(!is_bad_inode(inode)); | |
1271 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1272 | } | |
1273 | /* | |
1274 | * Already have the inode and the dentry attached, decrement | |
1275 | * the reference count to balance the iget() done | |
1276 | * earlier on. We found the dentry using d_lookup() so it | |
1277 | * cannot be disconnected and thus we do not need to worry | |
1278 | * about any NFS/disconnectedness issues here. | |
1279 | */ | |
1280 | iput(inode); | |
1281 | return found; | |
1282 | } | |
1283 | /* | |
1284 | * Negative dentry: instantiate it unless the inode is a directory and | |
1285 | * has a 'disconnected' dentry (i.e. IS_ROOT and DCACHE_DISCONNECTED), | |
1286 | * in which case d_move() that in place of the found dentry. | |
1287 | */ | |
1288 | if (!S_ISDIR(inode->i_mode)) { | |
1289 | /* Not a directory; everything is easy. */ | |
1290 | d_instantiate(found, inode); | |
1291 | return found; | |
1292 | } | |
1293 | spin_lock(&dcache_lock); | |
1294 | if (list_empty(&inode->i_dentry)) { | |
1295 | /* | |
1296 | * Directory without a 'disconnected' dentry; we need to do | |
1297 | * d_instantiate() by hand because it takes dcache_lock which | |
1298 | * we already hold. | |
1299 | */ | |
360da900 | 1300 | __d_instantiate(found, inode); |
9403540c BN |
1301 | spin_unlock(&dcache_lock); |
1302 | security_d_instantiate(found, inode); | |
1303 | return found; | |
1304 | } | |
1305 | /* | |
1306 | * Directory with a 'disconnected' dentry; get a reference to the | |
1307 | * 'disconnected' dentry. | |
1308 | */ | |
1309 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
1310 | dget_locked(new); | |
1311 | spin_unlock(&dcache_lock); | |
1312 | /* Do security vodoo. */ | |
1313 | security_d_instantiate(found, inode); | |
1314 | /* Move new in place of found. */ | |
1315 | d_move(new, found); | |
1316 | /* Balance the iget() we did above. */ | |
1317 | iput(inode); | |
1318 | /* Throw away found. */ | |
1319 | dput(found); | |
1320 | /* Use new as the actual dentry. */ | |
1321 | return new; | |
1322 | ||
1323 | err_out: | |
1324 | iput(inode); | |
1325 | return ERR_PTR(error); | |
1326 | } | |
1da177e4 LT |
1327 | |
1328 | /** | |
1329 | * d_lookup - search for a dentry | |
1330 | * @parent: parent dentry | |
1331 | * @name: qstr of name we wish to find | |
1332 | * | |
1333 | * Searches the children of the parent dentry for the name in question. If | |
1334 | * the dentry is found its reference count is incremented and the dentry | |
be42c4c4 | 1335 | * is returned. The caller must use dput to free the entry when it has |
1da177e4 LT |
1336 | * finished using it. %NULL is returned on failure. |
1337 | * | |
1338 | * __d_lookup is dcache_lock free. The hash list is protected using RCU. | |
1339 | * Memory barriers are used while updating and doing lockless traversal. | |
1340 | * To avoid races with d_move while rename is happening, d_lock is used. | |
1341 | * | |
1342 | * Overflows in memcmp(), while d_move, are avoided by keeping the length | |
1343 | * and name pointer in one structure pointed by d_qstr. | |
1344 | * | |
1345 | * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while | |
1346 | * lookup is going on. | |
1347 | * | |
da3bbdd4 | 1348 | * The dentry unused LRU is not updated even if lookup finds the required dentry |
1da177e4 LT |
1349 | * in there. It is updated in places such as prune_dcache, shrink_dcache_sb, |
1350 | * select_parent and __dget_locked. This laziness saves lookup from dcache_lock | |
1351 | * acquisition. | |
1352 | * | |
1353 | * d_lookup() is protected against the concurrent renames in some unrelated | |
1354 | * directory using the seqlockt_t rename_lock. | |
1355 | */ | |
1356 | ||
1357 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) | |
1358 | { | |
1359 | struct dentry * dentry = NULL; | |
1360 | unsigned long seq; | |
1361 | ||
1362 | do { | |
1363 | seq = read_seqbegin(&rename_lock); | |
1364 | dentry = __d_lookup(parent, name); | |
1365 | if (dentry) | |
1366 | break; | |
1367 | } while (read_seqretry(&rename_lock, seq)); | |
1368 | return dentry; | |
1369 | } | |
1370 | ||
1371 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) | |
1372 | { | |
1373 | unsigned int len = name->len; | |
1374 | unsigned int hash = name->hash; | |
1375 | const unsigned char *str = name->name; | |
1376 | struct hlist_head *head = d_hash(parent,hash); | |
1377 | struct dentry *found = NULL; | |
1378 | struct hlist_node *node; | |
665a7583 | 1379 | struct dentry *dentry; |
1da177e4 LT |
1380 | |
1381 | rcu_read_lock(); | |
1382 | ||
665a7583 | 1383 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1384 | struct qstr *qstr; |
1385 | ||
1da177e4 LT |
1386 | if (dentry->d_name.hash != hash) |
1387 | continue; | |
1388 | if (dentry->d_parent != parent) | |
1389 | continue; | |
1390 | ||
1391 | spin_lock(&dentry->d_lock); | |
1392 | ||
1393 | /* | |
1394 | * Recheck the dentry after taking the lock - d_move may have | |
1395 | * changed things. Don't bother checking the hash because we're | |
1396 | * about to compare the whole name anyway. | |
1397 | */ | |
1398 | if (dentry->d_parent != parent) | |
1399 | goto next; | |
1400 | ||
d0185c08 LT |
1401 | /* non-existing due to RCU? */ |
1402 | if (d_unhashed(dentry)) | |
1403 | goto next; | |
1404 | ||
1da177e4 LT |
1405 | /* |
1406 | * It is safe to compare names since d_move() cannot | |
1407 | * change the qstr (protected by d_lock). | |
1408 | */ | |
1409 | qstr = &dentry->d_name; | |
1410 | if (parent->d_op && parent->d_op->d_compare) { | |
1411 | if (parent->d_op->d_compare(parent, qstr, name)) | |
1412 | goto next; | |
1413 | } else { | |
1414 | if (qstr->len != len) | |
1415 | goto next; | |
1416 | if (memcmp(qstr->name, str, len)) | |
1417 | goto next; | |
1418 | } | |
1419 | ||
d0185c08 LT |
1420 | atomic_inc(&dentry->d_count); |
1421 | found = dentry; | |
1da177e4 LT |
1422 | spin_unlock(&dentry->d_lock); |
1423 | break; | |
1424 | next: | |
1425 | spin_unlock(&dentry->d_lock); | |
1426 | } | |
1427 | rcu_read_unlock(); | |
1428 | ||
1429 | return found; | |
1430 | } | |
1431 | ||
3e7e241f EB |
1432 | /** |
1433 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1434 | * @dir: Directory to search in | |
1435 | * @name: qstr of name we wish to find | |
1436 | * | |
1437 | * On hash failure or on lookup failure NULL is returned. | |
1438 | */ | |
1439 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1440 | { | |
1441 | struct dentry *dentry = NULL; | |
1442 | ||
1443 | /* | |
1444 | * Check for a fs-specific hash function. Note that we must | |
1445 | * calculate the standard hash first, as the d_op->d_hash() | |
1446 | * routine may choose to leave the hash value unchanged. | |
1447 | */ | |
1448 | name->hash = full_name_hash(name->name, name->len); | |
1449 | if (dir->d_op && dir->d_op->d_hash) { | |
1450 | if (dir->d_op->d_hash(dir, name) < 0) | |
1451 | goto out; | |
1452 | } | |
1453 | dentry = d_lookup(dir, name); | |
1454 | out: | |
1455 | return dentry; | |
1456 | } | |
1457 | ||
1da177e4 LT |
1458 | /** |
1459 | * d_validate - verify dentry provided from insecure source | |
1460 | * @dentry: The dentry alleged to be valid child of @dparent | |
1461 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1462 | * |
1463 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1464 | * This is used by ncpfs in its readdir implementation. | |
1465 | * Zero is returned in the dentry is invalid. | |
1466 | */ | |
1467 | ||
1468 | int d_validate(struct dentry *dentry, struct dentry *dparent) | |
1469 | { | |
1470 | struct hlist_head *base; | |
1471 | struct hlist_node *lhp; | |
1472 | ||
1473 | /* Check whether the ptr might be valid at all.. */ | |
1474 | if (!kmem_ptr_validate(dentry_cache, dentry)) | |
1475 | goto out; | |
1476 | ||
1477 | if (dentry->d_parent != dparent) | |
1478 | goto out; | |
1479 | ||
1480 | spin_lock(&dcache_lock); | |
1481 | base = d_hash(dparent, dentry->d_name.hash); | |
1482 | hlist_for_each(lhp,base) { | |
665a7583 | 1483 | /* hlist_for_each_entry_rcu() not required for d_hash list |
1da177e4 LT |
1484 | * as it is parsed under dcache_lock |
1485 | */ | |
1486 | if (dentry == hlist_entry(lhp, struct dentry, d_hash)) { | |
1487 | __dget_locked(dentry); | |
1488 | spin_unlock(&dcache_lock); | |
1489 | return 1; | |
1490 | } | |
1491 | } | |
1492 | spin_unlock(&dcache_lock); | |
1493 | out: | |
1494 | return 0; | |
1495 | } | |
1496 | ||
1497 | /* | |
1498 | * When a file is deleted, we have two options: | |
1499 | * - turn this dentry into a negative dentry | |
1500 | * - unhash this dentry and free it. | |
1501 | * | |
1502 | * Usually, we want to just turn this into | |
1503 | * a negative dentry, but if anybody else is | |
1504 | * currently using the dentry or the inode | |
1505 | * we can't do that and we fall back on removing | |
1506 | * it from the hash queues and waiting for | |
1507 | * it to be deleted later when it has no users | |
1508 | */ | |
1509 | ||
1510 | /** | |
1511 | * d_delete - delete a dentry | |
1512 | * @dentry: The dentry to delete | |
1513 | * | |
1514 | * Turn the dentry into a negative dentry if possible, otherwise | |
1515 | * remove it from the hash queues so it can be deleted later | |
1516 | */ | |
1517 | ||
1518 | void d_delete(struct dentry * dentry) | |
1519 | { | |
7a91bf7f | 1520 | int isdir = 0; |
1da177e4 LT |
1521 | /* |
1522 | * Are we the only user? | |
1523 | */ | |
1524 | spin_lock(&dcache_lock); | |
1525 | spin_lock(&dentry->d_lock); | |
7a91bf7f | 1526 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
1da177e4 LT |
1527 | if (atomic_read(&dentry->d_count) == 1) { |
1528 | dentry_iput(dentry); | |
7a91bf7f | 1529 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1530 | return; |
1531 | } | |
1532 | ||
1533 | if (!d_unhashed(dentry)) | |
1534 | __d_drop(dentry); | |
1535 | ||
1536 | spin_unlock(&dentry->d_lock); | |
1537 | spin_unlock(&dcache_lock); | |
7a91bf7f JM |
1538 | |
1539 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 LT |
1540 | } |
1541 | ||
1542 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1543 | { | |
1544 | ||
1545 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1546 | hlist_add_head_rcu(&entry->d_hash, list); | |
1547 | } | |
1548 | ||
770bfad8 DH |
1549 | static void _d_rehash(struct dentry * entry) |
1550 | { | |
1551 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1552 | } | |
1553 | ||
1da177e4 LT |
1554 | /** |
1555 | * d_rehash - add an entry back to the hash | |
1556 | * @entry: dentry to add to the hash | |
1557 | * | |
1558 | * Adds a dentry to the hash according to its name. | |
1559 | */ | |
1560 | ||
1561 | void d_rehash(struct dentry * entry) | |
1562 | { | |
1da177e4 LT |
1563 | spin_lock(&dcache_lock); |
1564 | spin_lock(&entry->d_lock); | |
770bfad8 | 1565 | _d_rehash(entry); |
1da177e4 LT |
1566 | spin_unlock(&entry->d_lock); |
1567 | spin_unlock(&dcache_lock); | |
1568 | } | |
1569 | ||
1da177e4 LT |
1570 | /* |
1571 | * When switching names, the actual string doesn't strictly have to | |
1572 | * be preserved in the target - because we're dropping the target | |
1573 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1574 | * the new name before we switch. | |
1575 | * | |
1576 | * Note that we have to be a lot more careful about getting the hash | |
1577 | * switched - we have to switch the hash value properly even if it | |
1578 | * then no longer matches the actual (corrupted) string of the target. | |
1579 | * The hash value has to match the hash queue that the dentry is on.. | |
1580 | */ | |
1581 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
1582 | { | |
1583 | if (dname_external(target)) { | |
1584 | if (dname_external(dentry)) { | |
1585 | /* | |
1586 | * Both external: swap the pointers | |
1587 | */ | |
9a8d5bb4 | 1588 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
1589 | } else { |
1590 | /* | |
1591 | * dentry:internal, target:external. Steal target's | |
1592 | * storage and make target internal. | |
1593 | */ | |
321bcf92 BF |
1594 | memcpy(target->d_iname, dentry->d_name.name, |
1595 | dentry->d_name.len + 1); | |
1da177e4 LT |
1596 | dentry->d_name.name = target->d_name.name; |
1597 | target->d_name.name = target->d_iname; | |
1598 | } | |
1599 | } else { | |
1600 | if (dname_external(dentry)) { | |
1601 | /* | |
1602 | * dentry:external, target:internal. Give dentry's | |
1603 | * storage to target and make dentry internal | |
1604 | */ | |
1605 | memcpy(dentry->d_iname, target->d_name.name, | |
1606 | target->d_name.len + 1); | |
1607 | target->d_name.name = dentry->d_name.name; | |
1608 | dentry->d_name.name = dentry->d_iname; | |
1609 | } else { | |
1610 | /* | |
1611 | * Both are internal. Just copy target to dentry | |
1612 | */ | |
1613 | memcpy(dentry->d_iname, target->d_name.name, | |
1614 | target->d_name.len + 1); | |
dc711ca3 AV |
1615 | dentry->d_name.len = target->d_name.len; |
1616 | return; | |
1da177e4 LT |
1617 | } |
1618 | } | |
9a8d5bb4 | 1619 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
1620 | } |
1621 | ||
1622 | /* | |
1623 | * We cannibalize "target" when moving dentry on top of it, | |
1624 | * because it's going to be thrown away anyway. We could be more | |
1625 | * polite about it, though. | |
1626 | * | |
1627 | * This forceful removal will result in ugly /proc output if | |
1628 | * somebody holds a file open that got deleted due to a rename. | |
1629 | * We could be nicer about the deleted file, and let it show | |
bc154b1e BF |
1630 | * up under the name it had before it was deleted rather than |
1631 | * under the original name of the file that was moved on top of it. | |
1da177e4 LT |
1632 | */ |
1633 | ||
9eaef27b TM |
1634 | /* |
1635 | * d_move_locked - move a dentry | |
1da177e4 LT |
1636 | * @dentry: entry to move |
1637 | * @target: new dentry | |
1638 | * | |
1639 | * Update the dcache to reflect the move of a file name. Negative | |
1640 | * dcache entries should not be moved in this way. | |
1641 | */ | |
9eaef27b | 1642 | static void d_move_locked(struct dentry * dentry, struct dentry * target) |
1da177e4 LT |
1643 | { |
1644 | struct hlist_head *list; | |
1645 | ||
1646 | if (!dentry->d_inode) | |
1647 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1648 | ||
1da177e4 LT |
1649 | write_seqlock(&rename_lock); |
1650 | /* | |
1651 | * XXXX: do we really need to take target->d_lock? | |
1652 | */ | |
1653 | if (target < dentry) { | |
1654 | spin_lock(&target->d_lock); | |
a90b9c05 | 1655 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1656 | } else { |
1657 | spin_lock(&dentry->d_lock); | |
a90b9c05 | 1658 | spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1659 | } |
1660 | ||
1661 | /* Move the dentry to the target hash queue, if on different bucket */ | |
f77e3498 | 1662 | if (d_unhashed(dentry)) |
1da177e4 LT |
1663 | goto already_unhashed; |
1664 | ||
1665 | hlist_del_rcu(&dentry->d_hash); | |
1666 | ||
1667 | already_unhashed: | |
1668 | list = d_hash(target->d_parent, target->d_name.hash); | |
1669 | __d_rehash(dentry, list); | |
1670 | ||
1671 | /* Unhash the target: dput() will then get rid of it */ | |
1672 | __d_drop(target); | |
1673 | ||
5160ee6f ED |
1674 | list_del(&dentry->d_u.d_child); |
1675 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
1676 | |
1677 | /* Switch the names.. */ | |
1678 | switch_names(dentry, target); | |
9a8d5bb4 | 1679 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
1680 | |
1681 | /* ... and switch the parents */ | |
1682 | if (IS_ROOT(dentry)) { | |
1683 | dentry->d_parent = target->d_parent; | |
1684 | target->d_parent = target; | |
5160ee6f | 1685 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 1686 | } else { |
9a8d5bb4 | 1687 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
1688 | |
1689 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 1690 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
1691 | } |
1692 | ||
5160ee6f | 1693 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 1694 | spin_unlock(&target->d_lock); |
c32ccd87 | 1695 | fsnotify_d_move(dentry); |
1da177e4 LT |
1696 | spin_unlock(&dentry->d_lock); |
1697 | write_sequnlock(&rename_lock); | |
9eaef27b TM |
1698 | } |
1699 | ||
1700 | /** | |
1701 | * d_move - move a dentry | |
1702 | * @dentry: entry to move | |
1703 | * @target: new dentry | |
1704 | * | |
1705 | * Update the dcache to reflect the move of a file name. Negative | |
1706 | * dcache entries should not be moved in this way. | |
1707 | */ | |
1708 | ||
1709 | void d_move(struct dentry * dentry, struct dentry * target) | |
1710 | { | |
1711 | spin_lock(&dcache_lock); | |
1712 | d_move_locked(dentry, target); | |
1da177e4 LT |
1713 | spin_unlock(&dcache_lock); |
1714 | } | |
1715 | ||
e2761a11 OH |
1716 | /** |
1717 | * d_ancestor - search for an ancestor | |
1718 | * @p1: ancestor dentry | |
1719 | * @p2: child dentry | |
1720 | * | |
1721 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
1722 | * an ancestor of p2, else NULL. | |
9eaef27b | 1723 | */ |
e2761a11 | 1724 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
1725 | { |
1726 | struct dentry *p; | |
1727 | ||
871c0067 | 1728 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 1729 | if (p->d_parent == p1) |
e2761a11 | 1730 | return p; |
9eaef27b | 1731 | } |
e2761a11 | 1732 | return NULL; |
9eaef27b TM |
1733 | } |
1734 | ||
1735 | /* | |
1736 | * This helper attempts to cope with remotely renamed directories | |
1737 | * | |
1738 | * It assumes that the caller is already holding | |
1739 | * dentry->d_parent->d_inode->i_mutex and the dcache_lock | |
1740 | * | |
1741 | * Note: If ever the locking in lock_rename() changes, then please | |
1742 | * remember to update this too... | |
9eaef27b TM |
1743 | */ |
1744 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
31f3e0b3 | 1745 | __releases(dcache_lock) |
9eaef27b TM |
1746 | { |
1747 | struct mutex *m1 = NULL, *m2 = NULL; | |
1748 | struct dentry *ret; | |
1749 | ||
1750 | /* If alias and dentry share a parent, then no extra locks required */ | |
1751 | if (alias->d_parent == dentry->d_parent) | |
1752 | goto out_unalias; | |
1753 | ||
1754 | /* Check for loops */ | |
1755 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 1756 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
1757 | goto out_err; |
1758 | ||
1759 | /* See lock_rename() */ | |
1760 | ret = ERR_PTR(-EBUSY); | |
1761 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
1762 | goto out_err; | |
1763 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
1764 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
1765 | goto out_err; | |
1766 | m2 = &alias->d_parent->d_inode->i_mutex; | |
1767 | out_unalias: | |
1768 | d_move_locked(alias, dentry); | |
1769 | ret = alias; | |
1770 | out_err: | |
1771 | spin_unlock(&dcache_lock); | |
1772 | if (m2) | |
1773 | mutex_unlock(m2); | |
1774 | if (m1) | |
1775 | mutex_unlock(m1); | |
1776 | return ret; | |
1777 | } | |
1778 | ||
770bfad8 DH |
1779 | /* |
1780 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
1781 | * named dentry in place of the dentry to be replaced. | |
1782 | */ | |
1783 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
1784 | { | |
1785 | struct dentry *dparent, *aparent; | |
1786 | ||
1787 | switch_names(dentry, anon); | |
9a8d5bb4 | 1788 | swap(dentry->d_name.hash, anon->d_name.hash); |
770bfad8 DH |
1789 | |
1790 | dparent = dentry->d_parent; | |
1791 | aparent = anon->d_parent; | |
1792 | ||
1793 | dentry->d_parent = (aparent == anon) ? dentry : aparent; | |
1794 | list_del(&dentry->d_u.d_child); | |
1795 | if (!IS_ROOT(dentry)) | |
1796 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
1797 | else | |
1798 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
1799 | ||
1800 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
1801 | list_del(&anon->d_u.d_child); | |
1802 | if (!IS_ROOT(anon)) | |
1803 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
1804 | else | |
1805 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
1806 | ||
1807 | anon->d_flags &= ~DCACHE_DISCONNECTED; | |
1808 | } | |
1809 | ||
1810 | /** | |
1811 | * d_materialise_unique - introduce an inode into the tree | |
1812 | * @dentry: candidate dentry | |
1813 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
1814 | * | |
1815 | * Introduces an dentry into the tree, substituting an extant disconnected | |
1816 | * root directory alias in its place if there is one | |
1817 | */ | |
1818 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
1819 | { | |
9eaef27b | 1820 | struct dentry *actual; |
770bfad8 DH |
1821 | |
1822 | BUG_ON(!d_unhashed(dentry)); | |
1823 | ||
1824 | spin_lock(&dcache_lock); | |
1825 | ||
1826 | if (!inode) { | |
1827 | actual = dentry; | |
360da900 | 1828 | __d_instantiate(dentry, NULL); |
770bfad8 DH |
1829 | goto found_lock; |
1830 | } | |
1831 | ||
9eaef27b TM |
1832 | if (S_ISDIR(inode->i_mode)) { |
1833 | struct dentry *alias; | |
1834 | ||
1835 | /* Does an aliased dentry already exist? */ | |
1836 | alias = __d_find_alias(inode, 0); | |
1837 | if (alias) { | |
1838 | actual = alias; | |
1839 | /* Is this an anonymous mountpoint that we could splice | |
1840 | * into our tree? */ | |
1841 | if (IS_ROOT(alias)) { | |
1842 | spin_lock(&alias->d_lock); | |
1843 | __d_materialise_dentry(dentry, alias); | |
1844 | __d_drop(alias); | |
1845 | goto found; | |
1846 | } | |
1847 | /* Nope, but we must(!) avoid directory aliasing */ | |
1848 | actual = __d_unalias(dentry, alias); | |
1849 | if (IS_ERR(actual)) | |
1850 | dput(alias); | |
1851 | goto out_nolock; | |
1852 | } | |
770bfad8 DH |
1853 | } |
1854 | ||
1855 | /* Add a unique reference */ | |
1856 | actual = __d_instantiate_unique(dentry, inode); | |
1857 | if (!actual) | |
1858 | actual = dentry; | |
1859 | else if (unlikely(!d_unhashed(actual))) | |
1860 | goto shouldnt_be_hashed; | |
1861 | ||
1862 | found_lock: | |
1863 | spin_lock(&actual->d_lock); | |
1864 | found: | |
1865 | _d_rehash(actual); | |
1866 | spin_unlock(&actual->d_lock); | |
1867 | spin_unlock(&dcache_lock); | |
9eaef27b | 1868 | out_nolock: |
770bfad8 DH |
1869 | if (actual == dentry) { |
1870 | security_d_instantiate(dentry, inode); | |
1871 | return NULL; | |
1872 | } | |
1873 | ||
1874 | iput(inode); | |
1875 | return actual; | |
1876 | ||
770bfad8 DH |
1877 | shouldnt_be_hashed: |
1878 | spin_unlock(&dcache_lock); | |
1879 | BUG(); | |
770bfad8 DH |
1880 | } |
1881 | ||
cdd16d02 | 1882 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
1883 | { |
1884 | *buflen -= namelen; | |
1885 | if (*buflen < 0) | |
1886 | return -ENAMETOOLONG; | |
1887 | *buffer -= namelen; | |
1888 | memcpy(*buffer, str, namelen); | |
1889 | return 0; | |
1890 | } | |
1891 | ||
cdd16d02 MS |
1892 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
1893 | { | |
1894 | return prepend(buffer, buflen, name->name, name->len); | |
1895 | } | |
1896 | ||
1da177e4 | 1897 | /** |
31f3e0b3 | 1898 | * __d_path - return the path of a dentry |
9d1bc601 MS |
1899 | * @path: the dentry/vfsmount to report |
1900 | * @root: root vfsmnt/dentry (may be modified by this function) | |
1da177e4 LT |
1901 | * @buffer: buffer to return value in |
1902 | * @buflen: buffer length | |
1903 | * | |
552ce544 LT |
1904 | * Convert a dentry into an ASCII path name. If the entry has been deleted |
1905 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1da177e4 | 1906 | * |
52afeefb AV |
1907 | * Returns a pointer into the buffer or an error code if the |
1908 | * path was too long. | |
552ce544 LT |
1909 | * |
1910 | * "buflen" should be positive. Caller holds the dcache_lock. | |
9d1bc601 MS |
1911 | * |
1912 | * If path is not reachable from the supplied root, then the value of | |
1913 | * root is changed (without modifying refcounts). | |
1da177e4 | 1914 | */ |
9d1bc601 MS |
1915 | char *__d_path(const struct path *path, struct path *root, |
1916 | char *buffer, int buflen) | |
1da177e4 | 1917 | { |
9d1bc601 MS |
1918 | struct dentry *dentry = path->dentry; |
1919 | struct vfsmount *vfsmnt = path->mnt; | |
cdd16d02 MS |
1920 | char *end = buffer + buflen; |
1921 | char *retval; | |
6092d048 | 1922 | |
be285c71 | 1923 | spin_lock(&vfsmount_lock); |
6092d048 RP |
1924 | prepend(&end, &buflen, "\0", 1); |
1925 | if (!IS_ROOT(dentry) && d_unhashed(dentry) && | |
1926 | (prepend(&end, &buflen, " (deleted)", 10) != 0)) | |
552ce544 | 1927 | goto Elong; |
552ce544 LT |
1928 | |
1929 | if (buflen < 1) | |
1930 | goto Elong; | |
1931 | /* Get '/' right */ | |
1932 | retval = end-1; | |
1933 | *retval = '/'; | |
1934 | ||
1935 | for (;;) { | |
1da177e4 LT |
1936 | struct dentry * parent; |
1937 | ||
329c97f0 | 1938 | if (dentry == root->dentry && vfsmnt == root->mnt) |
552ce544 | 1939 | break; |
1da177e4 | 1940 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 1941 | /* Global root? */ |
1da177e4 | 1942 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
1943 | goto global_root; |
1944 | } | |
1945 | dentry = vfsmnt->mnt_mountpoint; | |
1946 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
1947 | continue; |
1948 | } | |
1949 | parent = dentry->d_parent; | |
1950 | prefetch(parent); | |
cdd16d02 | 1951 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 | 1952 | (prepend(&end, &buflen, "/", 1) != 0)) |
552ce544 | 1953 | goto Elong; |
552ce544 | 1954 | retval = end; |
1da177e4 LT |
1955 | dentry = parent; |
1956 | } | |
1957 | ||
be285c71 AG |
1958 | out: |
1959 | spin_unlock(&vfsmount_lock); | |
552ce544 | 1960 | return retval; |
1da177e4 LT |
1961 | |
1962 | global_root: | |
6092d048 | 1963 | retval += 1; /* hit the slash */ |
cdd16d02 | 1964 | if (prepend_name(&retval, &buflen, &dentry->d_name) != 0) |
1da177e4 | 1965 | goto Elong; |
9d1bc601 MS |
1966 | root->mnt = vfsmnt; |
1967 | root->dentry = dentry; | |
be285c71 AG |
1968 | goto out; |
1969 | ||
1da177e4 | 1970 | Elong: |
be285c71 AG |
1971 | retval = ERR_PTR(-ENAMETOOLONG); |
1972 | goto out; | |
1da177e4 LT |
1973 | } |
1974 | ||
a03a8a70 JB |
1975 | /** |
1976 | * d_path - return the path of a dentry | |
cf28b486 | 1977 | * @path: path to report |
a03a8a70 JB |
1978 | * @buf: buffer to return value in |
1979 | * @buflen: buffer length | |
1980 | * | |
1981 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
1982 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1983 | * | |
52afeefb AV |
1984 | * Returns a pointer into the buffer or an error code if the path was |
1985 | * too long. Note: Callers should use the returned pointer, not the passed | |
1986 | * in buffer, to use the name! The implementation often starts at an offset | |
1987 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 1988 | * |
31f3e0b3 | 1989 | * "buflen" should be positive. |
a03a8a70 | 1990 | */ |
20d4fdc1 | 1991 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 LT |
1992 | { |
1993 | char *res; | |
6ac08c39 | 1994 | struct path root; |
9d1bc601 | 1995 | struct path tmp; |
1da177e4 | 1996 | |
c23fbb6b ED |
1997 | /* |
1998 | * We have various synthetic filesystems that never get mounted. On | |
1999 | * these filesystems dentries are never used for lookup purposes, and | |
2000 | * thus don't need to be hashed. They also don't need a name until a | |
2001 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2002 | * below allows us to generate a name for these objects on demand: | |
2003 | */ | |
cf28b486 JB |
2004 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2005 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2006 | |
1da177e4 | 2007 | read_lock(¤t->fs->lock); |
6ac08c39 | 2008 | root = current->fs->root; |
6092d048 | 2009 | path_get(&root); |
1da177e4 | 2010 | read_unlock(¤t->fs->lock); |
552ce544 | 2011 | spin_lock(&dcache_lock); |
9d1bc601 MS |
2012 | tmp = root; |
2013 | res = __d_path(path, &tmp, buf, buflen); | |
552ce544 | 2014 | spin_unlock(&dcache_lock); |
6ac08c39 | 2015 | path_put(&root); |
1da177e4 LT |
2016 | return res; |
2017 | } | |
2018 | ||
c23fbb6b ED |
2019 | /* |
2020 | * Helper function for dentry_operations.d_dname() members | |
2021 | */ | |
2022 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2023 | const char *fmt, ...) | |
2024 | { | |
2025 | va_list args; | |
2026 | char temp[64]; | |
2027 | int sz; | |
2028 | ||
2029 | va_start(args, fmt); | |
2030 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2031 | va_end(args); | |
2032 | ||
2033 | if (sz > sizeof(temp) || sz > buflen) | |
2034 | return ERR_PTR(-ENAMETOOLONG); | |
2035 | ||
2036 | buffer += buflen - sz; | |
2037 | return memcpy(buffer, temp, sz); | |
2038 | } | |
2039 | ||
6092d048 RP |
2040 | /* |
2041 | * Write full pathname from the root of the filesystem into the buffer. | |
2042 | */ | |
2043 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2044 | { | |
2045 | char *end = buf + buflen; | |
2046 | char *retval; | |
2047 | ||
2048 | spin_lock(&dcache_lock); | |
2049 | prepend(&end, &buflen, "\0", 1); | |
2050 | if (!IS_ROOT(dentry) && d_unhashed(dentry) && | |
2051 | (prepend(&end, &buflen, "//deleted", 9) != 0)) | |
2052 | goto Elong; | |
2053 | if (buflen < 1) | |
2054 | goto Elong; | |
2055 | /* Get '/' right */ | |
2056 | retval = end-1; | |
2057 | *retval = '/'; | |
2058 | ||
cdd16d02 MS |
2059 | while (!IS_ROOT(dentry)) { |
2060 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 2061 | |
6092d048 | 2062 | prefetch(parent); |
cdd16d02 | 2063 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 RP |
2064 | (prepend(&end, &buflen, "/", 1) != 0)) |
2065 | goto Elong; | |
2066 | ||
2067 | retval = end; | |
2068 | dentry = parent; | |
2069 | } | |
2070 | spin_unlock(&dcache_lock); | |
2071 | return retval; | |
2072 | Elong: | |
2073 | spin_unlock(&dcache_lock); | |
2074 | return ERR_PTR(-ENAMETOOLONG); | |
2075 | } | |
2076 | ||
1da177e4 LT |
2077 | /* |
2078 | * NOTE! The user-level library version returns a | |
2079 | * character pointer. The kernel system call just | |
2080 | * returns the length of the buffer filled (which | |
2081 | * includes the ending '\0' character), or a negative | |
2082 | * error value. So libc would do something like | |
2083 | * | |
2084 | * char *getcwd(char * buf, size_t size) | |
2085 | * { | |
2086 | * int retval; | |
2087 | * | |
2088 | * retval = sys_getcwd(buf, size); | |
2089 | * if (retval >= 0) | |
2090 | * return buf; | |
2091 | * errno = -retval; | |
2092 | * return NULL; | |
2093 | * } | |
2094 | */ | |
3cdad428 | 2095 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2096 | { |
552ce544 | 2097 | int error; |
6ac08c39 | 2098 | struct path pwd, root; |
552ce544 | 2099 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2100 | |
2101 | if (!page) | |
2102 | return -ENOMEM; | |
2103 | ||
2104 | read_lock(¤t->fs->lock); | |
6ac08c39 | 2105 | pwd = current->fs->pwd; |
6092d048 | 2106 | path_get(&pwd); |
6ac08c39 | 2107 | root = current->fs->root; |
6092d048 | 2108 | path_get(&root); |
1da177e4 LT |
2109 | read_unlock(¤t->fs->lock); |
2110 | ||
552ce544 LT |
2111 | error = -ENOENT; |
2112 | /* Has the current directory has been unlinked? */ | |
2113 | spin_lock(&dcache_lock); | |
cdd16d02 | 2114 | if (IS_ROOT(pwd.dentry) || !d_unhashed(pwd.dentry)) { |
552ce544 | 2115 | unsigned long len; |
9d1bc601 | 2116 | struct path tmp = root; |
552ce544 | 2117 | char * cwd; |
1da177e4 | 2118 | |
9d1bc601 | 2119 | cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE); |
552ce544 LT |
2120 | spin_unlock(&dcache_lock); |
2121 | ||
2122 | error = PTR_ERR(cwd); | |
2123 | if (IS_ERR(cwd)) | |
2124 | goto out; | |
2125 | ||
2126 | error = -ERANGE; | |
2127 | len = PAGE_SIZE + page - cwd; | |
2128 | if (len <= size) { | |
2129 | error = len; | |
2130 | if (copy_to_user(buf, cwd, len)) | |
2131 | error = -EFAULT; | |
2132 | } | |
2133 | } else | |
2134 | spin_unlock(&dcache_lock); | |
1da177e4 LT |
2135 | |
2136 | out: | |
6ac08c39 JB |
2137 | path_put(&pwd); |
2138 | path_put(&root); | |
1da177e4 LT |
2139 | free_page((unsigned long) page); |
2140 | return error; | |
2141 | } | |
2142 | ||
2143 | /* | |
2144 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2145 | * | |
2146 | * Trivially implemented using the dcache structure | |
2147 | */ | |
2148 | ||
2149 | /** | |
2150 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2151 | * @new_dentry: new dentry | |
2152 | * @old_dentry: old dentry | |
2153 | * | |
2154 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2155 | * Returns 0 otherwise. | |
2156 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2157 | */ | |
2158 | ||
e2761a11 | 2159 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2160 | { |
2161 | int result; | |
1da177e4 LT |
2162 | unsigned long seq; |
2163 | ||
e2761a11 OH |
2164 | /* FIXME: This is old behavior, needed? Please check callers. */ |
2165 | if (new_dentry == old_dentry) | |
2166 | return 1; | |
2167 | ||
2168 | /* | |
2169 | * Need rcu_readlock to protect against the d_parent trashing | |
2170 | * due to d_move | |
1da177e4 LT |
2171 | */ |
2172 | rcu_read_lock(); | |
e2761a11 | 2173 | do { |
1da177e4 | 2174 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2175 | seq = read_seqbegin(&rename_lock); |
e2761a11 | 2176 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2177 | result = 1; |
e2761a11 OH |
2178 | else |
2179 | result = 0; | |
1da177e4 LT |
2180 | } while (read_seqretry(&rename_lock, seq)); |
2181 | rcu_read_unlock(); | |
2182 | ||
2183 | return result; | |
2184 | } | |
2185 | ||
2186 | void d_genocide(struct dentry *root) | |
2187 | { | |
2188 | struct dentry *this_parent = root; | |
2189 | struct list_head *next; | |
2190 | ||
2191 | spin_lock(&dcache_lock); | |
2192 | repeat: | |
2193 | next = this_parent->d_subdirs.next; | |
2194 | resume: | |
2195 | while (next != &this_parent->d_subdirs) { | |
2196 | struct list_head *tmp = next; | |
5160ee6f | 2197 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
2198 | next = tmp->next; |
2199 | if (d_unhashed(dentry)||!dentry->d_inode) | |
2200 | continue; | |
2201 | if (!list_empty(&dentry->d_subdirs)) { | |
2202 | this_parent = dentry; | |
2203 | goto repeat; | |
2204 | } | |
2205 | atomic_dec(&dentry->d_count); | |
2206 | } | |
2207 | if (this_parent != root) { | |
5160ee6f | 2208 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
2209 | atomic_dec(&this_parent->d_count); |
2210 | this_parent = this_parent->d_parent; | |
2211 | goto resume; | |
2212 | } | |
2213 | spin_unlock(&dcache_lock); | |
2214 | } | |
2215 | ||
2216 | /** | |
2217 | * find_inode_number - check for dentry with name | |
2218 | * @dir: directory to check | |
2219 | * @name: Name to find. | |
2220 | * | |
2221 | * Check whether a dentry already exists for the given name, | |
2222 | * and return the inode number if it has an inode. Otherwise | |
2223 | * 0 is returned. | |
2224 | * | |
2225 | * This routine is used to post-process directory listings for | |
2226 | * filesystems using synthetic inode numbers, and is necessary | |
2227 | * to keep getcwd() working. | |
2228 | */ | |
2229 | ||
2230 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2231 | { | |
2232 | struct dentry * dentry; | |
2233 | ino_t ino = 0; | |
2234 | ||
3e7e241f EB |
2235 | dentry = d_hash_and_lookup(dir, name); |
2236 | if (dentry) { | |
1da177e4 LT |
2237 | if (dentry->d_inode) |
2238 | ino = dentry->d_inode->i_ino; | |
2239 | dput(dentry); | |
2240 | } | |
1da177e4 LT |
2241 | return ino; |
2242 | } | |
2243 | ||
2244 | static __initdata unsigned long dhash_entries; | |
2245 | static int __init set_dhash_entries(char *str) | |
2246 | { | |
2247 | if (!str) | |
2248 | return 0; | |
2249 | dhash_entries = simple_strtoul(str, &str, 0); | |
2250 | return 1; | |
2251 | } | |
2252 | __setup("dhash_entries=", set_dhash_entries); | |
2253 | ||
2254 | static void __init dcache_init_early(void) | |
2255 | { | |
2256 | int loop; | |
2257 | ||
2258 | /* If hashes are distributed across NUMA nodes, defer | |
2259 | * hash allocation until vmalloc space is available. | |
2260 | */ | |
2261 | if (hashdist) | |
2262 | return; | |
2263 | ||
2264 | dentry_hashtable = | |
2265 | alloc_large_system_hash("Dentry cache", | |
2266 | sizeof(struct hlist_head), | |
2267 | dhash_entries, | |
2268 | 13, | |
2269 | HASH_EARLY, | |
2270 | &d_hash_shift, | |
2271 | &d_hash_mask, | |
2272 | 0); | |
2273 | ||
2274 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2275 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2276 | } | |
2277 | ||
74bf17cf | 2278 | static void __init dcache_init(void) |
1da177e4 LT |
2279 | { |
2280 | int loop; | |
2281 | ||
2282 | /* | |
2283 | * A constructor could be added for stable state like the lists, | |
2284 | * but it is probably not worth it because of the cache nature | |
2285 | * of the dcache. | |
2286 | */ | |
0a31bd5f CL |
2287 | dentry_cache = KMEM_CACHE(dentry, |
2288 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2289 | |
8e1f936b | 2290 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2291 | |
2292 | /* Hash may have been set up in dcache_init_early */ | |
2293 | if (!hashdist) | |
2294 | return; | |
2295 | ||
2296 | dentry_hashtable = | |
2297 | alloc_large_system_hash("Dentry cache", | |
2298 | sizeof(struct hlist_head), | |
2299 | dhash_entries, | |
2300 | 13, | |
2301 | 0, | |
2302 | &d_hash_shift, | |
2303 | &d_hash_mask, | |
2304 | 0); | |
2305 | ||
2306 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2307 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2308 | } | |
2309 | ||
2310 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2311 | struct kmem_cache *names_cachep __read_mostly; |
1da177e4 | 2312 | |
1da177e4 LT |
2313 | EXPORT_SYMBOL(d_genocide); |
2314 | ||
1da177e4 LT |
2315 | void __init vfs_caches_init_early(void) |
2316 | { | |
2317 | dcache_init_early(); | |
2318 | inode_init_early(); | |
2319 | } | |
2320 | ||
2321 | void __init vfs_caches_init(unsigned long mempages) | |
2322 | { | |
2323 | unsigned long reserve; | |
2324 | ||
2325 | /* Base hash sizes on available memory, with a reserve equal to | |
2326 | 150% of current kernel size */ | |
2327 | ||
2328 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2329 | mempages -= reserve; | |
2330 | ||
2331 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2332 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2333 | |
74bf17cf DC |
2334 | dcache_init(); |
2335 | inode_init(); | |
1da177e4 | 2336 | files_init(mempages); |
74bf17cf | 2337 | mnt_init(); |
1da177e4 LT |
2338 | bdev_cache_init(); |
2339 | chrdev_init(); | |
2340 | } | |
2341 | ||
2342 | EXPORT_SYMBOL(d_alloc); | |
1da177e4 LT |
2343 | EXPORT_SYMBOL(d_alloc_root); |
2344 | EXPORT_SYMBOL(d_delete); | |
2345 | EXPORT_SYMBOL(d_find_alias); | |
2346 | EXPORT_SYMBOL(d_instantiate); | |
2347 | EXPORT_SYMBOL(d_invalidate); | |
2348 | EXPORT_SYMBOL(d_lookup); | |
2349 | EXPORT_SYMBOL(d_move); | |
770bfad8 | 2350 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
1da177e4 LT |
2351 | EXPORT_SYMBOL(d_path); |
2352 | EXPORT_SYMBOL(d_prune_aliases); | |
2353 | EXPORT_SYMBOL(d_rehash); | |
2354 | EXPORT_SYMBOL(d_splice_alias); | |
9403540c | 2355 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 LT |
2356 | EXPORT_SYMBOL(d_validate); |
2357 | EXPORT_SYMBOL(dget_locked); | |
2358 | EXPORT_SYMBOL(dput); | |
2359 | EXPORT_SYMBOL(find_inode_number); | |
2360 | EXPORT_SYMBOL(have_submounts); | |
2361 | EXPORT_SYMBOL(names_cachep); | |
2362 | EXPORT_SYMBOL(shrink_dcache_parent); | |
2363 | EXPORT_SYMBOL(shrink_dcache_sb); |