]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/dcache.c | |
3 | * | |
4 | * Complete reimplementation | |
5 | * (C) 1997 Thomas Schoebel-Theuer, | |
6 | * with heavy changes by Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Notes on the allocation strategy: | |
11 | * | |
12 | * The dcache is a master of the icache - whenever a dcache entry | |
13 | * exists, the inode will always exist. "iput()" is done either when | |
14 | * the dcache entry is deleted or garbage collected. | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
26 | #include <linux/module.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/file.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/seqlock.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/bootmem.h> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
07f3f05c | 36 | #include "internal.h" |
1da177e4 | 37 | |
fa3536cc | 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 | { |
fd217f4d | 72 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
73 | if (dname_external(dentry)) |
74 | kfree(dentry->d_name.name); | |
75 | kmem_cache_free(dentry_cache, dentry); | |
76 | } | |
77 | ||
b3423415 ED |
78 | static void d_callback(struct rcu_head *head) |
79 | { | |
80 | struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu); | |
81 | __d_free(dentry); | |
82 | } | |
83 | ||
1da177e4 LT |
84 | /* |
85 | * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry | |
86 | * inside dcache_lock. | |
87 | */ | |
88 | static void d_free(struct dentry *dentry) | |
89 | { | |
90 | if (dentry->d_op && dentry->d_op->d_release) | |
91 | dentry->d_op->d_release(dentry); | |
b3423415 | 92 | /* if dentry was never inserted into hash, immediate free is OK */ |
e8462caa | 93 | if (hlist_unhashed(&dentry->d_hash)) |
b3423415 ED |
94 | __d_free(dentry); |
95 | else | |
96 | call_rcu(&dentry->d_u.d_rcu, d_callback); | |
1da177e4 LT |
97 | } |
98 | ||
99 | /* | |
100 | * Release the dentry's inode, using the filesystem | |
101 | * d_iput() operation if defined. | |
1da177e4 | 102 | */ |
858119e1 | 103 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 MS |
104 | __releases(dentry->d_lock) |
105 | __releases(dcache_lock) | |
1da177e4 LT |
106 | { |
107 | struct inode *inode = dentry->d_inode; | |
108 | if (inode) { | |
109 | dentry->d_inode = NULL; | |
110 | list_del_init(&dentry->d_alias); | |
111 | spin_unlock(&dentry->d_lock); | |
112 | spin_unlock(&dcache_lock); | |
f805fbda LT |
113 | if (!inode->i_nlink) |
114 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
115 | if (dentry->d_op && dentry->d_op->d_iput) |
116 | dentry->d_op->d_iput(dentry, inode); | |
117 | else | |
118 | iput(inode); | |
119 | } else { | |
120 | spin_unlock(&dentry->d_lock); | |
121 | spin_unlock(&dcache_lock); | |
122 | } | |
123 | } | |
124 | ||
da3bbdd4 KM |
125 | /* |
126 | * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held. | |
127 | */ | |
128 | static void dentry_lru_add(struct dentry *dentry) | |
129 | { | |
130 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
131 | dentry->d_sb->s_nr_dentry_unused++; | |
132 | dentry_stat.nr_unused++; | |
133 | } | |
134 | ||
135 | static void dentry_lru_add_tail(struct dentry *dentry) | |
136 | { | |
137 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
138 | dentry->d_sb->s_nr_dentry_unused++; | |
139 | dentry_stat.nr_unused++; | |
140 | } | |
141 | ||
142 | static void dentry_lru_del(struct dentry *dentry) | |
143 | { | |
144 | if (!list_empty(&dentry->d_lru)) { | |
145 | list_del(&dentry->d_lru); | |
146 | dentry->d_sb->s_nr_dentry_unused--; | |
147 | dentry_stat.nr_unused--; | |
148 | } | |
149 | } | |
150 | ||
151 | static void dentry_lru_del_init(struct dentry *dentry) | |
152 | { | |
153 | if (likely(!list_empty(&dentry->d_lru))) { | |
154 | list_del_init(&dentry->d_lru); | |
155 | dentry->d_sb->s_nr_dentry_unused--; | |
156 | dentry_stat.nr_unused--; | |
157 | } | |
158 | } | |
159 | ||
d52b9086 MS |
160 | /** |
161 | * d_kill - kill dentry and return parent | |
162 | * @dentry: dentry to kill | |
163 | * | |
31f3e0b3 | 164 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
165 | * |
166 | * If this is the root of the dentry tree, return NULL. | |
167 | */ | |
168 | static struct dentry *d_kill(struct dentry *dentry) | |
31f3e0b3 MS |
169 | __releases(dentry->d_lock) |
170 | __releases(dcache_lock) | |
d52b9086 MS |
171 | { |
172 | struct dentry *parent; | |
173 | ||
174 | list_del(&dentry->d_u.d_child); | |
175 | dentry_stat.nr_dentry--; /* For d_free, below */ | |
176 | /*drops the locks, at that point nobody can reach this dentry */ | |
177 | dentry_iput(dentry); | |
871c0067 OH |
178 | if (IS_ROOT(dentry)) |
179 | parent = NULL; | |
180 | else | |
181 | parent = dentry->d_parent; | |
d52b9086 | 182 | d_free(dentry); |
871c0067 | 183 | return parent; |
d52b9086 MS |
184 | } |
185 | ||
1da177e4 LT |
186 | /* |
187 | * This is dput | |
188 | * | |
189 | * This is complicated by the fact that we do not want to put | |
190 | * dentries that are no longer on any hash chain on the unused | |
191 | * list: we'd much rather just get rid of them immediately. | |
192 | * | |
193 | * However, that implies that we have to traverse the dentry | |
194 | * tree upwards to the parents which might _also_ now be | |
195 | * scheduled for deletion (it may have been only waiting for | |
196 | * its last child to go away). | |
197 | * | |
198 | * This tail recursion is done by hand as we don't want to depend | |
199 | * on the compiler to always get this right (gcc generally doesn't). | |
200 | * Real recursion would eat up our stack space. | |
201 | */ | |
202 | ||
203 | /* | |
204 | * dput - release a dentry | |
205 | * @dentry: dentry to release | |
206 | * | |
207 | * Release a dentry. This will drop the usage count and if appropriate | |
208 | * call the dentry unlink method as well as removing it from the queues and | |
209 | * releasing its resources. If the parent dentries were scheduled for release | |
210 | * they too may now get deleted. | |
211 | * | |
212 | * no dcache lock, please. | |
213 | */ | |
214 | ||
215 | void dput(struct dentry *dentry) | |
216 | { | |
217 | if (!dentry) | |
218 | return; | |
219 | ||
220 | repeat: | |
221 | if (atomic_read(&dentry->d_count) == 1) | |
222 | might_sleep(); | |
223 | if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock)) | |
224 | return; | |
225 | ||
226 | spin_lock(&dentry->d_lock); | |
227 | if (atomic_read(&dentry->d_count)) { | |
228 | spin_unlock(&dentry->d_lock); | |
229 | spin_unlock(&dcache_lock); | |
230 | return; | |
231 | } | |
232 | ||
233 | /* | |
234 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
235 | */ | |
236 | if (dentry->d_op && dentry->d_op->d_delete) { | |
237 | if (dentry->d_op->d_delete(dentry)) | |
238 | goto unhash_it; | |
239 | } | |
240 | /* Unreachable? Get rid of it */ | |
241 | if (d_unhashed(dentry)) | |
242 | goto kill_it; | |
243 | if (list_empty(&dentry->d_lru)) { | |
244 | dentry->d_flags |= DCACHE_REFERENCED; | |
da3bbdd4 | 245 | dentry_lru_add(dentry); |
1da177e4 LT |
246 | } |
247 | spin_unlock(&dentry->d_lock); | |
248 | spin_unlock(&dcache_lock); | |
249 | return; | |
250 | ||
251 | unhash_it: | |
252 | __d_drop(dentry); | |
d52b9086 | 253 | kill_it: |
da3bbdd4 KM |
254 | /* if dentry was on the d_lru list delete it from there */ |
255 | dentry_lru_del(dentry); | |
d52b9086 MS |
256 | dentry = d_kill(dentry); |
257 | if (dentry) | |
258 | goto repeat; | |
1da177e4 | 259 | } |
ec4f8605 | 260 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
261 | |
262 | /** | |
263 | * d_invalidate - invalidate a dentry | |
264 | * @dentry: dentry to invalidate | |
265 | * | |
266 | * Try to invalidate the dentry if it turns out to be | |
267 | * possible. If there are other dentries that can be | |
268 | * reached through this one we can't delete it and we | |
269 | * return -EBUSY. On success we return 0. | |
270 | * | |
271 | * no dcache lock. | |
272 | */ | |
273 | ||
274 | int d_invalidate(struct dentry * dentry) | |
275 | { | |
276 | /* | |
277 | * If it's already been dropped, return OK. | |
278 | */ | |
279 | spin_lock(&dcache_lock); | |
280 | if (d_unhashed(dentry)) { | |
281 | spin_unlock(&dcache_lock); | |
282 | return 0; | |
283 | } | |
284 | /* | |
285 | * Check whether to do a partial shrink_dcache | |
286 | * to get rid of unused child entries. | |
287 | */ | |
288 | if (!list_empty(&dentry->d_subdirs)) { | |
289 | spin_unlock(&dcache_lock); | |
290 | shrink_dcache_parent(dentry); | |
291 | spin_lock(&dcache_lock); | |
292 | } | |
293 | ||
294 | /* | |
295 | * Somebody else still using it? | |
296 | * | |
297 | * If it's a directory, we can't drop it | |
298 | * for fear of somebody re-populating it | |
299 | * with children (even though dropping it | |
300 | * would make it unreachable from the root, | |
301 | * we might still populate it if it was a | |
302 | * working directory or similar). | |
303 | */ | |
304 | spin_lock(&dentry->d_lock); | |
305 | if (atomic_read(&dentry->d_count) > 1) { | |
306 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { | |
307 | spin_unlock(&dentry->d_lock); | |
308 | spin_unlock(&dcache_lock); | |
309 | return -EBUSY; | |
310 | } | |
311 | } | |
312 | ||
313 | __d_drop(dentry); | |
314 | spin_unlock(&dentry->d_lock); | |
315 | spin_unlock(&dcache_lock); | |
316 | return 0; | |
317 | } | |
ec4f8605 | 318 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 LT |
319 | |
320 | /* This should be called _only_ with dcache_lock held */ | |
321 | ||
322 | static inline struct dentry * __dget_locked(struct dentry *dentry) | |
323 | { | |
324 | atomic_inc(&dentry->d_count); | |
da3bbdd4 | 325 | dentry_lru_del_init(dentry); |
1da177e4 LT |
326 | return dentry; |
327 | } | |
328 | ||
329 | struct dentry * dget_locked(struct dentry *dentry) | |
330 | { | |
331 | return __dget_locked(dentry); | |
332 | } | |
ec4f8605 | 333 | EXPORT_SYMBOL(dget_locked); |
1da177e4 LT |
334 | |
335 | /** | |
336 | * d_find_alias - grab a hashed alias of inode | |
337 | * @inode: inode in question | |
338 | * @want_discon: flag, used by d_splice_alias, to request | |
339 | * that only a DISCONNECTED alias be returned. | |
340 | * | |
341 | * If inode has a hashed alias, or is a directory and has any alias, | |
342 | * acquire the reference to alias and return it. Otherwise return NULL. | |
343 | * Notice that if inode is a directory there can be only one alias and | |
344 | * it can be unhashed only if it has no children, or if it is the root | |
345 | * of a filesystem. | |
346 | * | |
21c0d8fd | 347 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 348 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 349 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 LT |
350 | */ |
351 | ||
352 | static struct dentry * __d_find_alias(struct inode *inode, int want_discon) | |
353 | { | |
354 | struct list_head *head, *next, *tmp; | |
355 | struct dentry *alias, *discon_alias=NULL; | |
356 | ||
357 | head = &inode->i_dentry; | |
358 | next = inode->i_dentry.next; | |
359 | while (next != head) { | |
360 | tmp = next; | |
361 | next = tmp->next; | |
362 | prefetch(next); | |
363 | alias = list_entry(tmp, struct dentry, d_alias); | |
364 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
21c0d8fd N |
365 | if (IS_ROOT(alias) && |
366 | (alias->d_flags & DCACHE_DISCONNECTED)) | |
1da177e4 LT |
367 | discon_alias = alias; |
368 | else if (!want_discon) { | |
369 | __dget_locked(alias); | |
370 | return alias; | |
371 | } | |
372 | } | |
373 | } | |
374 | if (discon_alias) | |
375 | __dget_locked(discon_alias); | |
376 | return discon_alias; | |
377 | } | |
378 | ||
379 | struct dentry * d_find_alias(struct inode *inode) | |
380 | { | |
214fda1f DH |
381 | struct dentry *de = NULL; |
382 | ||
383 | if (!list_empty(&inode->i_dentry)) { | |
384 | spin_lock(&dcache_lock); | |
385 | de = __d_find_alias(inode, 0); | |
386 | spin_unlock(&dcache_lock); | |
387 | } | |
1da177e4 LT |
388 | return de; |
389 | } | |
ec4f8605 | 390 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
391 | |
392 | /* | |
393 | * Try to kill dentries associated with this inode. | |
394 | * WARNING: you must own a reference to inode. | |
395 | */ | |
396 | void d_prune_aliases(struct inode *inode) | |
397 | { | |
0cdca3f9 | 398 | struct dentry *dentry; |
1da177e4 LT |
399 | restart: |
400 | spin_lock(&dcache_lock); | |
0cdca3f9 | 401 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 LT |
402 | spin_lock(&dentry->d_lock); |
403 | if (!atomic_read(&dentry->d_count)) { | |
404 | __dget_locked(dentry); | |
405 | __d_drop(dentry); | |
406 | spin_unlock(&dentry->d_lock); | |
407 | spin_unlock(&dcache_lock); | |
408 | dput(dentry); | |
409 | goto restart; | |
410 | } | |
411 | spin_unlock(&dentry->d_lock); | |
412 | } | |
413 | spin_unlock(&dcache_lock); | |
414 | } | |
ec4f8605 | 415 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
416 | |
417 | /* | |
d702ccb3 AM |
418 | * Throw away a dentry - free the inode, dput the parent. This requires that |
419 | * the LRU list has already been removed. | |
420 | * | |
85864e10 MS |
421 | * Try to prune ancestors as well. This is necessary to prevent |
422 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
423 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 424 | */ |
85864e10 | 425 | static void prune_one_dentry(struct dentry * dentry) |
31f3e0b3 MS |
426 | __releases(dentry->d_lock) |
427 | __releases(dcache_lock) | |
428 | __acquires(dcache_lock) | |
1da177e4 | 429 | { |
1da177e4 | 430 | __d_drop(dentry); |
d52b9086 | 431 | dentry = d_kill(dentry); |
d52b9086 MS |
432 | |
433 | /* | |
434 | * Prune ancestors. Locking is simpler than in dput(), | |
435 | * because dcache_lock needs to be taken anyway. | |
436 | */ | |
1da177e4 | 437 | spin_lock(&dcache_lock); |
d52b9086 MS |
438 | while (dentry) { |
439 | if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock)) | |
440 | return; | |
441 | ||
442 | if (dentry->d_op && dentry->d_op->d_delete) | |
443 | dentry->d_op->d_delete(dentry); | |
da3bbdd4 | 444 | dentry_lru_del_init(dentry); |
d52b9086 MS |
445 | __d_drop(dentry); |
446 | dentry = d_kill(dentry); | |
447 | spin_lock(&dcache_lock); | |
448 | } | |
1da177e4 LT |
449 | } |
450 | ||
da3bbdd4 KM |
451 | /* |
452 | * Shrink the dentry LRU on a given superblock. | |
453 | * @sb : superblock to shrink dentry LRU. | |
454 | * @count: If count is NULL, we prune all dentries on superblock. | |
455 | * @flags: If flags is non-zero, we need to do special processing based on | |
456 | * which flags are set. This means we don't need to maintain multiple | |
457 | * similar copies of this loop. | |
1da177e4 | 458 | */ |
da3bbdd4 | 459 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) |
1da177e4 | 460 | { |
da3bbdd4 KM |
461 | LIST_HEAD(referenced); |
462 | LIST_HEAD(tmp); | |
463 | struct dentry *dentry; | |
464 | int cnt = 0; | |
1da177e4 | 465 | |
da3bbdd4 KM |
466 | BUG_ON(!sb); |
467 | BUG_ON((flags & DCACHE_REFERENCED) && count == NULL); | |
468 | spin_lock(&dcache_lock); | |
469 | if (count != NULL) | |
470 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
471 | cnt = *count; | |
472 | restart: | |
473 | if (count == NULL) | |
474 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
475 | else { | |
476 | while (!list_empty(&sb->s_dentry_lru)) { | |
477 | dentry = list_entry(sb->s_dentry_lru.prev, | |
478 | struct dentry, d_lru); | |
479 | BUG_ON(dentry->d_sb != sb); | |
480 | ||
481 | spin_lock(&dentry->d_lock); | |
482 | /* | |
483 | * If we are honouring the DCACHE_REFERENCED flag and | |
484 | * the dentry has this flag set, don't free it. Clear | |
485 | * the flag and put it back on the LRU. | |
0feae5c4 | 486 | */ |
da3bbdd4 KM |
487 | if ((flags & DCACHE_REFERENCED) |
488 | && (dentry->d_flags & DCACHE_REFERENCED)) { | |
489 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
c490d79b | 490 | list_move(&dentry->d_lru, &referenced); |
da3bbdd4 KM |
491 | spin_unlock(&dentry->d_lock); |
492 | } else { | |
493 | list_move_tail(&dentry->d_lru, &tmp); | |
494 | spin_unlock(&dentry->d_lock); | |
495 | cnt--; | |
496 | if (!cnt) | |
497 | break; | |
0feae5c4 | 498 | } |
f3c6ba98 | 499 | cond_resched_lock(&dcache_lock); |
0feae5c4 | 500 | } |
da3bbdd4 KM |
501 | } |
502 | while (!list_empty(&tmp)) { | |
503 | dentry = list_entry(tmp.prev, struct dentry, d_lru); | |
504 | dentry_lru_del_init(dentry); | |
505 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
506 | /* |
507 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
508 | * the LRU because of laziness during lookup. Do not free |
509 | * it - just keep it off the LRU list. | |
1da177e4 | 510 | */ |
da3bbdd4 KM |
511 | if (atomic_read(&dentry->d_count)) { |
512 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
513 | continue; |
514 | } | |
da3bbdd4 KM |
515 | prune_one_dentry(dentry); |
516 | /* dentry->d_lock was dropped in prune_one_dentry() */ | |
517 | cond_resched_lock(&dcache_lock); | |
518 | } | |
519 | if (count == NULL && !list_empty(&sb->s_dentry_lru)) | |
520 | goto restart; | |
521 | if (count != NULL) | |
522 | *count = cnt; | |
523 | if (!list_empty(&referenced)) | |
524 | list_splice(&referenced, &sb->s_dentry_lru); | |
525 | spin_unlock(&dcache_lock); | |
526 | } | |
527 | ||
528 | /** | |
529 | * prune_dcache - shrink the dcache | |
530 | * @count: number of entries to try to free | |
531 | * | |
532 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
533 | * need to unmount something (at which point we need to unuse all dentries). | |
534 | * | |
535 | * This function may fail to free any resources if all the dentries are in use. | |
536 | */ | |
537 | static void prune_dcache(int count) | |
538 | { | |
dca33252 | 539 | struct super_block *sb, *p = NULL; |
da3bbdd4 KM |
540 | int w_count; |
541 | int unused = dentry_stat.nr_unused; | |
542 | int prune_ratio; | |
543 | int pruned; | |
544 | ||
545 | if (unused == 0 || count == 0) | |
546 | return; | |
547 | spin_lock(&dcache_lock); | |
da3bbdd4 KM |
548 | if (count >= unused) |
549 | prune_ratio = 1; | |
550 | else | |
551 | prune_ratio = unused / count; | |
552 | spin_lock(&sb_lock); | |
dca33252 | 553 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
554 | if (list_empty(&sb->s_instances)) |
555 | continue; | |
da3bbdd4 | 556 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 557 | continue; |
da3bbdd4 KM |
558 | sb->s_count++; |
559 | /* Now, we reclaim unused dentrins with fairness. | |
560 | * We reclaim them same percentage from each superblock. | |
561 | * We calculate number of dentries to scan on this sb | |
562 | * as follows, but the implementation is arranged to avoid | |
563 | * overflows: | |
564 | * number of dentries to scan on this sb = | |
565 | * count * (number of dentries on this sb / | |
566 | * number of dentries in the machine) | |
0feae5c4 | 567 | */ |
da3bbdd4 KM |
568 | spin_unlock(&sb_lock); |
569 | if (prune_ratio != 1) | |
570 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
571 | else | |
572 | w_count = sb->s_nr_dentry_unused; | |
573 | pruned = w_count; | |
0feae5c4 | 574 | /* |
da3bbdd4 KM |
575 | * We need to be sure this filesystem isn't being unmounted, |
576 | * otherwise we could race with generic_shutdown_super(), and | |
577 | * end up holding a reference to an inode while the filesystem | |
578 | * is unmounted. So we try to get s_umount, and make sure | |
579 | * s_root isn't NULL. | |
0feae5c4 | 580 | */ |
da3bbdd4 KM |
581 | if (down_read_trylock(&sb->s_umount)) { |
582 | if ((sb->s_root != NULL) && | |
583 | (!list_empty(&sb->s_dentry_lru))) { | |
584 | spin_unlock(&dcache_lock); | |
585 | __shrink_dcache_sb(sb, &w_count, | |
586 | DCACHE_REFERENCED); | |
587 | pruned -= w_count; | |
588 | spin_lock(&dcache_lock); | |
0feae5c4 | 589 | } |
da3bbdd4 | 590 | up_read(&sb->s_umount); |
0feae5c4 | 591 | } |
da3bbdd4 | 592 | spin_lock(&sb_lock); |
dca33252 AV |
593 | if (p) |
594 | __put_super(p); | |
da3bbdd4 | 595 | count -= pruned; |
dca33252 | 596 | p = sb; |
79893c17 AV |
597 | /* more work left to do? */ |
598 | if (count <= 0) | |
599 | break; | |
1da177e4 | 600 | } |
dca33252 AV |
601 | if (p) |
602 | __put_super(p); | |
da3bbdd4 | 603 | spin_unlock(&sb_lock); |
1da177e4 LT |
604 | spin_unlock(&dcache_lock); |
605 | } | |
606 | ||
1da177e4 LT |
607 | /** |
608 | * shrink_dcache_sb - shrink dcache for a superblock | |
609 | * @sb: superblock | |
610 | * | |
611 | * Shrink the dcache for the specified super block. This | |
612 | * is used to free the dcache before unmounting a file | |
613 | * system | |
614 | */ | |
1da177e4 LT |
615 | void shrink_dcache_sb(struct super_block * sb) |
616 | { | |
da3bbdd4 | 617 | __shrink_dcache_sb(sb, NULL, 0); |
1da177e4 | 618 | } |
ec4f8605 | 619 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 620 | |
c636ebdb DH |
621 | /* |
622 | * destroy a single subtree of dentries for unmount | |
623 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
624 | * locking | |
625 | */ | |
626 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
627 | { | |
628 | struct dentry *parent; | |
f8713576 | 629 | unsigned detached = 0; |
c636ebdb DH |
630 | |
631 | BUG_ON(!IS_ROOT(dentry)); | |
632 | ||
633 | /* detach this root from the system */ | |
634 | spin_lock(&dcache_lock); | |
da3bbdd4 | 635 | dentry_lru_del_init(dentry); |
c636ebdb DH |
636 | __d_drop(dentry); |
637 | spin_unlock(&dcache_lock); | |
638 | ||
639 | for (;;) { | |
640 | /* descend to the first leaf in the current subtree */ | |
641 | while (!list_empty(&dentry->d_subdirs)) { | |
642 | struct dentry *loop; | |
643 | ||
644 | /* this is a branch with children - detach all of them | |
645 | * from the system in one go */ | |
646 | spin_lock(&dcache_lock); | |
647 | list_for_each_entry(loop, &dentry->d_subdirs, | |
648 | d_u.d_child) { | |
da3bbdd4 | 649 | dentry_lru_del_init(loop); |
c636ebdb DH |
650 | __d_drop(loop); |
651 | cond_resched_lock(&dcache_lock); | |
652 | } | |
653 | spin_unlock(&dcache_lock); | |
654 | ||
655 | /* move to the first child */ | |
656 | dentry = list_entry(dentry->d_subdirs.next, | |
657 | struct dentry, d_u.d_child); | |
658 | } | |
659 | ||
660 | /* consume the dentries from this leaf up through its parents | |
661 | * until we find one with children or run out altogether */ | |
662 | do { | |
663 | struct inode *inode; | |
664 | ||
665 | if (atomic_read(&dentry->d_count) != 0) { | |
666 | printk(KERN_ERR | |
667 | "BUG: Dentry %p{i=%lx,n=%s}" | |
668 | " still in use (%d)" | |
669 | " [unmount of %s %s]\n", | |
670 | dentry, | |
671 | dentry->d_inode ? | |
672 | dentry->d_inode->i_ino : 0UL, | |
673 | dentry->d_name.name, | |
674 | atomic_read(&dentry->d_count), | |
675 | dentry->d_sb->s_type->name, | |
676 | dentry->d_sb->s_id); | |
677 | BUG(); | |
678 | } | |
679 | ||
871c0067 | 680 | if (IS_ROOT(dentry)) |
c636ebdb | 681 | parent = NULL; |
871c0067 OH |
682 | else { |
683 | parent = dentry->d_parent; | |
c636ebdb | 684 | atomic_dec(&parent->d_count); |
871c0067 | 685 | } |
c636ebdb DH |
686 | |
687 | list_del(&dentry->d_u.d_child); | |
f8713576 | 688 | detached++; |
c636ebdb DH |
689 | |
690 | inode = dentry->d_inode; | |
691 | if (inode) { | |
692 | dentry->d_inode = NULL; | |
693 | list_del_init(&dentry->d_alias); | |
694 | if (dentry->d_op && dentry->d_op->d_iput) | |
695 | dentry->d_op->d_iput(dentry, inode); | |
696 | else | |
697 | iput(inode); | |
698 | } | |
699 | ||
700 | d_free(dentry); | |
701 | ||
702 | /* finished when we fall off the top of the tree, | |
703 | * otherwise we ascend to the parent and move to the | |
704 | * next sibling if there is one */ | |
705 | if (!parent) | |
f8713576 | 706 | goto out; |
c636ebdb DH |
707 | |
708 | dentry = parent; | |
709 | ||
710 | } while (list_empty(&dentry->d_subdirs)); | |
711 | ||
712 | dentry = list_entry(dentry->d_subdirs.next, | |
713 | struct dentry, d_u.d_child); | |
714 | } | |
f8713576 DH |
715 | out: |
716 | /* several dentries were freed, need to correct nr_dentry */ | |
717 | spin_lock(&dcache_lock); | |
718 | dentry_stat.nr_dentry -= detached; | |
719 | spin_unlock(&dcache_lock); | |
c636ebdb DH |
720 | } |
721 | ||
722 | /* | |
723 | * destroy the dentries attached to a superblock on unmounting | |
724 | * - we don't need to use dentry->d_lock, and only need dcache_lock when | |
725 | * removing the dentry from the system lists and hashes because: | |
726 | * - the superblock is detached from all mountings and open files, so the | |
727 | * dentry trees will not be rearranged by the VFS | |
728 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
729 | * any dentries belonging to this superblock that it comes across | |
730 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
731 | * in this superblock | |
732 | */ | |
733 | void shrink_dcache_for_umount(struct super_block *sb) | |
734 | { | |
735 | struct dentry *dentry; | |
736 | ||
737 | if (down_read_trylock(&sb->s_umount)) | |
738 | BUG(); | |
739 | ||
740 | dentry = sb->s_root; | |
741 | sb->s_root = NULL; | |
742 | atomic_dec(&dentry->d_count); | |
743 | shrink_dcache_for_umount_subtree(dentry); | |
744 | ||
745 | while (!hlist_empty(&sb->s_anon)) { | |
746 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
747 | shrink_dcache_for_umount_subtree(dentry); | |
748 | } | |
749 | } | |
750 | ||
1da177e4 LT |
751 | /* |
752 | * Search for at least 1 mount point in the dentry's subdirs. | |
753 | * We descend to the next level whenever the d_subdirs | |
754 | * list is non-empty and continue searching. | |
755 | */ | |
756 | ||
757 | /** | |
758 | * have_submounts - check for mounts over a dentry | |
759 | * @parent: dentry to check. | |
760 | * | |
761 | * Return true if the parent or its subdirectories contain | |
762 | * a mount point | |
763 | */ | |
764 | ||
765 | int have_submounts(struct dentry *parent) | |
766 | { | |
767 | struct dentry *this_parent = parent; | |
768 | struct list_head *next; | |
769 | ||
770 | spin_lock(&dcache_lock); | |
771 | if (d_mountpoint(parent)) | |
772 | goto positive; | |
773 | repeat: | |
774 | next = this_parent->d_subdirs.next; | |
775 | resume: | |
776 | while (next != &this_parent->d_subdirs) { | |
777 | struct list_head *tmp = next; | |
5160ee6f | 778 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
779 | next = tmp->next; |
780 | /* Have we found a mount point ? */ | |
781 | if (d_mountpoint(dentry)) | |
782 | goto positive; | |
783 | if (!list_empty(&dentry->d_subdirs)) { | |
784 | this_parent = dentry; | |
785 | goto repeat; | |
786 | } | |
787 | } | |
788 | /* | |
789 | * All done at this level ... ascend and resume the search. | |
790 | */ | |
791 | if (this_parent != parent) { | |
5160ee6f | 792 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
793 | this_parent = this_parent->d_parent; |
794 | goto resume; | |
795 | } | |
796 | spin_unlock(&dcache_lock); | |
797 | return 0; /* No mount points found in tree */ | |
798 | positive: | |
799 | spin_unlock(&dcache_lock); | |
800 | return 1; | |
801 | } | |
ec4f8605 | 802 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
803 | |
804 | /* | |
805 | * Search the dentry child list for the specified parent, | |
806 | * and move any unused dentries to the end of the unused | |
807 | * list for prune_dcache(). We descend to the next level | |
808 | * whenever the d_subdirs list is non-empty and continue | |
809 | * searching. | |
810 | * | |
811 | * It returns zero iff there are no unused children, | |
812 | * otherwise it returns the number of children moved to | |
813 | * the end of the unused list. This may not be the total | |
814 | * number of unused children, because select_parent can | |
815 | * drop the lock and return early due to latency | |
816 | * constraints. | |
817 | */ | |
818 | static int select_parent(struct dentry * parent) | |
819 | { | |
820 | struct dentry *this_parent = parent; | |
821 | struct list_head *next; | |
822 | int found = 0; | |
823 | ||
824 | spin_lock(&dcache_lock); | |
825 | repeat: | |
826 | next = this_parent->d_subdirs.next; | |
827 | resume: | |
828 | while (next != &this_parent->d_subdirs) { | |
829 | struct list_head *tmp = next; | |
5160ee6f | 830 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
831 | next = tmp->next; |
832 | ||
da3bbdd4 | 833 | dentry_lru_del_init(dentry); |
1da177e4 LT |
834 | /* |
835 | * move only zero ref count dentries to the end | |
836 | * of the unused list for prune_dcache | |
837 | */ | |
838 | if (!atomic_read(&dentry->d_count)) { | |
da3bbdd4 | 839 | dentry_lru_add_tail(dentry); |
1da177e4 LT |
840 | found++; |
841 | } | |
842 | ||
843 | /* | |
844 | * We can return to the caller if we have found some (this | |
845 | * ensures forward progress). We'll be coming back to find | |
846 | * the rest. | |
847 | */ | |
848 | if (found && need_resched()) | |
849 | goto out; | |
850 | ||
851 | /* | |
852 | * Descend a level if the d_subdirs list is non-empty. | |
853 | */ | |
854 | if (!list_empty(&dentry->d_subdirs)) { | |
855 | this_parent = dentry; | |
1da177e4 LT |
856 | goto repeat; |
857 | } | |
858 | } | |
859 | /* | |
860 | * All done at this level ... ascend and resume the search. | |
861 | */ | |
862 | if (this_parent != parent) { | |
5160ee6f | 863 | next = this_parent->d_u.d_child.next; |
1da177e4 | 864 | this_parent = this_parent->d_parent; |
1da177e4 LT |
865 | goto resume; |
866 | } | |
867 | out: | |
868 | spin_unlock(&dcache_lock); | |
869 | return found; | |
870 | } | |
871 | ||
872 | /** | |
873 | * shrink_dcache_parent - prune dcache | |
874 | * @parent: parent of entries to prune | |
875 | * | |
876 | * Prune the dcache to remove unused children of the parent dentry. | |
877 | */ | |
878 | ||
879 | void shrink_dcache_parent(struct dentry * parent) | |
880 | { | |
da3bbdd4 | 881 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
882 | int found; |
883 | ||
884 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 885 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 886 | } |
ec4f8605 | 887 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 888 | |
1da177e4 LT |
889 | /* |
890 | * Scan `nr' dentries and return the number which remain. | |
891 | * | |
892 | * We need to avoid reentering the filesystem if the caller is performing a | |
893 | * GFP_NOFS allocation attempt. One example deadlock is: | |
894 | * | |
895 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
896 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
897 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
898 | * | |
899 | * In this case we return -1 to tell the caller that we baled. | |
900 | */ | |
7f8275d0 | 901 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
902 | { |
903 | if (nr) { | |
904 | if (!(gfp_mask & __GFP_FS)) | |
905 | return -1; | |
da3bbdd4 | 906 | prune_dcache(nr); |
1da177e4 LT |
907 | } |
908 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
909 | } | |
910 | ||
8e1f936b RR |
911 | static struct shrinker dcache_shrinker = { |
912 | .shrink = shrink_dcache_memory, | |
913 | .seeks = DEFAULT_SEEKS, | |
914 | }; | |
915 | ||
1da177e4 LT |
916 | /** |
917 | * d_alloc - allocate a dcache entry | |
918 | * @parent: parent of entry to allocate | |
919 | * @name: qstr of the name | |
920 | * | |
921 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
922 | * available. On a success the dentry is returned. The name passed in is | |
923 | * copied and the copy passed in may be reused after this call. | |
924 | */ | |
925 | ||
926 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
927 | { | |
928 | struct dentry *dentry; | |
929 | char *dname; | |
930 | ||
e12ba74d | 931 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
932 | if (!dentry) |
933 | return NULL; | |
934 | ||
935 | if (name->len > DNAME_INLINE_LEN-1) { | |
936 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
937 | if (!dname) { | |
938 | kmem_cache_free(dentry_cache, dentry); | |
939 | return NULL; | |
940 | } | |
941 | } else { | |
942 | dname = dentry->d_iname; | |
943 | } | |
944 | dentry->d_name.name = dname; | |
945 | ||
946 | dentry->d_name.len = name->len; | |
947 | dentry->d_name.hash = name->hash; | |
948 | memcpy(dname, name->name, name->len); | |
949 | dname[name->len] = 0; | |
950 | ||
951 | atomic_set(&dentry->d_count, 1); | |
952 | dentry->d_flags = DCACHE_UNHASHED; | |
953 | spin_lock_init(&dentry->d_lock); | |
954 | dentry->d_inode = NULL; | |
955 | dentry->d_parent = NULL; | |
956 | dentry->d_sb = NULL; | |
957 | dentry->d_op = NULL; | |
958 | dentry->d_fsdata = NULL; | |
959 | dentry->d_mounted = 0; | |
1da177e4 LT |
960 | INIT_HLIST_NODE(&dentry->d_hash); |
961 | INIT_LIST_HEAD(&dentry->d_lru); | |
962 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
963 | INIT_LIST_HEAD(&dentry->d_alias); | |
964 | ||
965 | if (parent) { | |
966 | dentry->d_parent = dget(parent); | |
967 | dentry->d_sb = parent->d_sb; | |
968 | } else { | |
5160ee6f | 969 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
970 | } |
971 | ||
972 | spin_lock(&dcache_lock); | |
973 | if (parent) | |
5160ee6f | 974 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1da177e4 LT |
975 | dentry_stat.nr_dentry++; |
976 | spin_unlock(&dcache_lock); | |
977 | ||
978 | return dentry; | |
979 | } | |
ec4f8605 | 980 | EXPORT_SYMBOL(d_alloc); |
1da177e4 LT |
981 | |
982 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
983 | { | |
984 | struct qstr q; | |
985 | ||
986 | q.name = name; | |
987 | q.len = strlen(name); | |
988 | q.hash = full_name_hash(q.name, q.len); | |
989 | return d_alloc(parent, &q); | |
990 | } | |
ef26ca97 | 991 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 992 | |
360da900 OH |
993 | /* the caller must hold dcache_lock */ |
994 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | |
995 | { | |
996 | if (inode) | |
997 | list_add(&dentry->d_alias, &inode->i_dentry); | |
998 | dentry->d_inode = inode; | |
999 | fsnotify_d_instantiate(dentry, inode); | |
1000 | } | |
1001 | ||
1da177e4 LT |
1002 | /** |
1003 | * d_instantiate - fill in inode information for a dentry | |
1004 | * @entry: dentry to complete | |
1005 | * @inode: inode to attach to this dentry | |
1006 | * | |
1007 | * Fill in inode information in the entry. | |
1008 | * | |
1009 | * This turns negative dentries into productive full members | |
1010 | * of society. | |
1011 | * | |
1012 | * NOTE! This assumes that the inode count has been incremented | |
1013 | * (or otherwise set) by the caller to indicate that it is now | |
1014 | * in use by the dcache. | |
1015 | */ | |
1016 | ||
1017 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1018 | { | |
28133c7b | 1019 | BUG_ON(!list_empty(&entry->d_alias)); |
1da177e4 | 1020 | spin_lock(&dcache_lock); |
360da900 | 1021 | __d_instantiate(entry, inode); |
1da177e4 LT |
1022 | spin_unlock(&dcache_lock); |
1023 | security_d_instantiate(entry, inode); | |
1024 | } | |
ec4f8605 | 1025 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1026 | |
1027 | /** | |
1028 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1029 | * @entry: dentry to instantiate | |
1030 | * @inode: inode to attach to this dentry | |
1031 | * | |
1032 | * Fill in inode information in the entry. On success, it returns NULL. | |
1033 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1034 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1035 | * |
1036 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1037 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1038 | * |
1039 | * This also assumes that the inode count has been incremented | |
1040 | * (or otherwise set) by the caller to indicate that it is now | |
1041 | * in use by the dcache. | |
1da177e4 | 1042 | */ |
770bfad8 DH |
1043 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1044 | struct inode *inode) | |
1da177e4 LT |
1045 | { |
1046 | struct dentry *alias; | |
1047 | int len = entry->d_name.len; | |
1048 | const char *name = entry->d_name.name; | |
1049 | unsigned int hash = entry->d_name.hash; | |
1050 | ||
770bfad8 | 1051 | if (!inode) { |
360da900 | 1052 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1053 | return NULL; |
1054 | } | |
1055 | ||
1da177e4 LT |
1056 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1057 | struct qstr *qstr = &alias->d_name; | |
1058 | ||
1059 | if (qstr->hash != hash) | |
1060 | continue; | |
1061 | if (alias->d_parent != entry->d_parent) | |
1062 | continue; | |
1063 | if (qstr->len != len) | |
1064 | continue; | |
1065 | if (memcmp(qstr->name, name, len)) | |
1066 | continue; | |
1067 | dget_locked(alias); | |
1da177e4 LT |
1068 | return alias; |
1069 | } | |
770bfad8 | 1070 | |
360da900 | 1071 | __d_instantiate(entry, inode); |
1da177e4 LT |
1072 | return NULL; |
1073 | } | |
770bfad8 DH |
1074 | |
1075 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1076 | { | |
1077 | struct dentry *result; | |
1078 | ||
1079 | BUG_ON(!list_empty(&entry->d_alias)); | |
1080 | ||
1081 | spin_lock(&dcache_lock); | |
1082 | result = __d_instantiate_unique(entry, inode); | |
1083 | spin_unlock(&dcache_lock); | |
1084 | ||
1085 | if (!result) { | |
1086 | security_d_instantiate(entry, inode); | |
1087 | return NULL; | |
1088 | } | |
1089 | ||
1090 | BUG_ON(!d_unhashed(result)); | |
1091 | iput(inode); | |
1092 | return result; | |
1093 | } | |
1094 | ||
1da177e4 LT |
1095 | EXPORT_SYMBOL(d_instantiate_unique); |
1096 | ||
1097 | /** | |
1098 | * d_alloc_root - allocate root dentry | |
1099 | * @root_inode: inode to allocate the root for | |
1100 | * | |
1101 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1102 | * instantiated and returned. %NULL is returned if there is insufficient | |
1103 | * memory or the inode passed is %NULL. | |
1104 | */ | |
1105 | ||
1106 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1107 | { | |
1108 | struct dentry *res = NULL; | |
1109 | ||
1110 | if (root_inode) { | |
1111 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1112 | ||
1113 | res = d_alloc(NULL, &name); | |
1114 | if (res) { | |
1115 | res->d_sb = root_inode->i_sb; | |
1116 | res->d_parent = res; | |
1117 | d_instantiate(res, root_inode); | |
1118 | } | |
1119 | } | |
1120 | return res; | |
1121 | } | |
ec4f8605 | 1122 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 LT |
1123 | |
1124 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1125 | unsigned long hash) | |
1126 | { | |
1127 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1128 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1129 | return dentry_hashtable + (hash & D_HASHMASK); | |
1130 | } | |
1131 | ||
4ea3ada2 CH |
1132 | /** |
1133 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1134 | * @inode: inode to allocate the dentry for | |
1135 | * | |
1136 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1137 | * similar open by handle operations. The returned dentry may be anonymous, | |
1138 | * or may have a full name (if the inode was already in the cache). | |
1139 | * | |
1140 | * When called on a directory inode, we must ensure that the inode only ever | |
1141 | * has one dentry. If a dentry is found, that is returned instead of | |
1142 | * allocating a new one. | |
1143 | * | |
1144 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1145 | * to the dentry. In case of an error the reference on the inode is released. |
1146 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1147 | * be passed in and will be the error will be propagate to the return value, | |
1148 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1149 | */ |
1150 | struct dentry *d_obtain_alias(struct inode *inode) | |
1151 | { | |
9308a612 CH |
1152 | static const struct qstr anonstring = { .name = "" }; |
1153 | struct dentry *tmp; | |
1154 | struct dentry *res; | |
4ea3ada2 CH |
1155 | |
1156 | if (!inode) | |
44003728 | 1157 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1158 | if (IS_ERR(inode)) |
1159 | return ERR_CAST(inode); | |
1160 | ||
9308a612 CH |
1161 | res = d_find_alias(inode); |
1162 | if (res) | |
1163 | goto out_iput; | |
1164 | ||
1165 | tmp = d_alloc(NULL, &anonstring); | |
1166 | if (!tmp) { | |
1167 | res = ERR_PTR(-ENOMEM); | |
1168 | goto out_iput; | |
4ea3ada2 | 1169 | } |
9308a612 CH |
1170 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1171 | ||
1172 | spin_lock(&dcache_lock); | |
1173 | res = __d_find_alias(inode, 0); | |
1174 | if (res) { | |
1175 | spin_unlock(&dcache_lock); | |
1176 | dput(tmp); | |
1177 | goto out_iput; | |
1178 | } | |
1179 | ||
1180 | /* attach a disconnected dentry */ | |
1181 | spin_lock(&tmp->d_lock); | |
1182 | tmp->d_sb = inode->i_sb; | |
1183 | tmp->d_inode = inode; | |
1184 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1185 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1186 | list_add(&tmp->d_alias, &inode->i_dentry); | |
1187 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); | |
1188 | spin_unlock(&tmp->d_lock); | |
1189 | ||
1190 | spin_unlock(&dcache_lock); | |
1191 | return tmp; | |
1192 | ||
1193 | out_iput: | |
1194 | iput(inode); | |
1195 | return res; | |
4ea3ada2 | 1196 | } |
adc48720 | 1197 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1198 | |
1199 | /** | |
1200 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1201 | * @inode: the inode which may have a disconnected dentry | |
1202 | * @dentry: a negative dentry which we want to point to the inode. | |
1203 | * | |
1204 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1205 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1206 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1207 | * | |
1208 | * This is needed in the lookup routine of any filesystem that is exportable | |
1209 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1210 | * | |
1211 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1212 | * is returned. This matches the expected return value of ->lookup. | |
1213 | * | |
1214 | */ | |
1215 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1216 | { | |
1217 | struct dentry *new = NULL; | |
1218 | ||
21c0d8fd | 1219 | if (inode && S_ISDIR(inode->i_mode)) { |
1da177e4 LT |
1220 | spin_lock(&dcache_lock); |
1221 | new = __d_find_alias(inode, 1); | |
1222 | if (new) { | |
1223 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
1224 | spin_unlock(&dcache_lock); | |
1225 | security_d_instantiate(new, inode); | |
1da177e4 LT |
1226 | d_move(new, dentry); |
1227 | iput(inode); | |
1228 | } else { | |
360da900 OH |
1229 | /* already taking dcache_lock, so d_add() by hand */ |
1230 | __d_instantiate(dentry, inode); | |
1da177e4 LT |
1231 | spin_unlock(&dcache_lock); |
1232 | security_d_instantiate(dentry, inode); | |
1233 | d_rehash(dentry); | |
1234 | } | |
1235 | } else | |
1236 | d_add(dentry, inode); | |
1237 | return new; | |
1238 | } | |
ec4f8605 | 1239 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1240 | |
9403540c BN |
1241 | /** |
1242 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1243 | * @inode: the inode case-insensitive lookup has found | |
1244 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1245 | * @name: the case-exact name to be associated with the returned dentry | |
1246 | * | |
1247 | * This is to avoid filling the dcache with case-insensitive names to the | |
1248 | * same inode, only the actual correct case is stored in the dcache for | |
1249 | * case-insensitive filesystems. | |
1250 | * | |
1251 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1252 | * already exists in in the dcache, use it and return it. | |
1253 | * | |
1254 | * If no entry exists with the exact case name, allocate new dentry with | |
1255 | * the exact case, and return the spliced entry. | |
1256 | */ | |
e45b590b | 1257 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1258 | struct qstr *name) |
1259 | { | |
1260 | int error; | |
1261 | struct dentry *found; | |
1262 | struct dentry *new; | |
1263 | ||
b6520c81 CH |
1264 | /* |
1265 | * First check if a dentry matching the name already exists, | |
1266 | * if not go ahead and create it now. | |
1267 | */ | |
9403540c | 1268 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1269 | if (!found) { |
1270 | new = d_alloc(dentry->d_parent, name); | |
1271 | if (!new) { | |
1272 | error = -ENOMEM; | |
1273 | goto err_out; | |
1274 | } | |
b6520c81 | 1275 | |
9403540c BN |
1276 | found = d_splice_alias(inode, new); |
1277 | if (found) { | |
1278 | dput(new); | |
1279 | return found; | |
1280 | } | |
1281 | return new; | |
1282 | } | |
b6520c81 CH |
1283 | |
1284 | /* | |
1285 | * If a matching dentry exists, and it's not negative use it. | |
1286 | * | |
1287 | * Decrement the reference count to balance the iget() done | |
1288 | * earlier on. | |
1289 | */ | |
9403540c BN |
1290 | if (found->d_inode) { |
1291 | if (unlikely(found->d_inode != inode)) { | |
1292 | /* This can't happen because bad inodes are unhashed. */ | |
1293 | BUG_ON(!is_bad_inode(inode)); | |
1294 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1295 | } | |
9403540c BN |
1296 | iput(inode); |
1297 | return found; | |
1298 | } | |
b6520c81 | 1299 | |
9403540c BN |
1300 | /* |
1301 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1302 | * already has a dentry. |
9403540c | 1303 | */ |
9403540c | 1304 | spin_lock(&dcache_lock); |
b6520c81 | 1305 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1306 | __d_instantiate(found, inode); |
9403540c BN |
1307 | spin_unlock(&dcache_lock); |
1308 | security_d_instantiate(found, inode); | |
1309 | return found; | |
1310 | } | |
b6520c81 | 1311 | |
9403540c | 1312 | /* |
b6520c81 CH |
1313 | * In case a directory already has a (disconnected) entry grab a |
1314 | * reference to it, move it in place and use it. | |
9403540c BN |
1315 | */ |
1316 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
1317 | dget_locked(new); | |
1318 | spin_unlock(&dcache_lock); | |
9403540c | 1319 | security_d_instantiate(found, inode); |
9403540c | 1320 | d_move(new, found); |
9403540c | 1321 | iput(inode); |
9403540c | 1322 | dput(found); |
9403540c BN |
1323 | return new; |
1324 | ||
1325 | err_out: | |
1326 | iput(inode); | |
1327 | return ERR_PTR(error); | |
1328 | } | |
ec4f8605 | 1329 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 LT |
1330 | |
1331 | /** | |
1332 | * d_lookup - search for a dentry | |
1333 | * @parent: parent dentry | |
1334 | * @name: qstr of name we wish to find | |
1335 | * | |
1336 | * Searches the children of the parent dentry for the name in question. If | |
1337 | * the dentry is found its reference count is incremented and the dentry | |
be42c4c4 | 1338 | * is returned. The caller must use dput to free the entry when it has |
1da177e4 LT |
1339 | * finished using it. %NULL is returned on failure. |
1340 | * | |
1341 | * __d_lookup is dcache_lock free. The hash list is protected using RCU. | |
1342 | * Memory barriers are used while updating and doing lockless traversal. | |
1343 | * To avoid races with d_move while rename is happening, d_lock is used. | |
1344 | * | |
1345 | * Overflows in memcmp(), while d_move, are avoided by keeping the length | |
1346 | * and name pointer in one structure pointed by d_qstr. | |
1347 | * | |
1348 | * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while | |
1349 | * lookup is going on. | |
1350 | * | |
da3bbdd4 | 1351 | * The dentry unused LRU is not updated even if lookup finds the required dentry |
1da177e4 LT |
1352 | * in there. It is updated in places such as prune_dcache, shrink_dcache_sb, |
1353 | * select_parent and __dget_locked. This laziness saves lookup from dcache_lock | |
1354 | * acquisition. | |
1355 | * | |
1356 | * d_lookup() is protected against the concurrent renames in some unrelated | |
1357 | * directory using the seqlockt_t rename_lock. | |
1358 | */ | |
1359 | ||
1360 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) | |
1361 | { | |
1362 | struct dentry * dentry = NULL; | |
1363 | unsigned long seq; | |
1364 | ||
1365 | do { | |
1366 | seq = read_seqbegin(&rename_lock); | |
1367 | dentry = __d_lookup(parent, name); | |
1368 | if (dentry) | |
1369 | break; | |
1370 | } while (read_seqretry(&rename_lock, seq)); | |
1371 | return dentry; | |
1372 | } | |
ec4f8605 | 1373 | EXPORT_SYMBOL(d_lookup); |
1da177e4 LT |
1374 | |
1375 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) | |
1376 | { | |
1377 | unsigned int len = name->len; | |
1378 | unsigned int hash = name->hash; | |
1379 | const unsigned char *str = name->name; | |
1380 | struct hlist_head *head = d_hash(parent,hash); | |
1381 | struct dentry *found = NULL; | |
1382 | struct hlist_node *node; | |
665a7583 | 1383 | struct dentry *dentry; |
1da177e4 LT |
1384 | |
1385 | rcu_read_lock(); | |
1386 | ||
665a7583 | 1387 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1388 | struct qstr *qstr; |
1389 | ||
1da177e4 LT |
1390 | if (dentry->d_name.hash != hash) |
1391 | continue; | |
1392 | if (dentry->d_parent != parent) | |
1393 | continue; | |
1394 | ||
1395 | spin_lock(&dentry->d_lock); | |
1396 | ||
1397 | /* | |
1398 | * Recheck the dentry after taking the lock - d_move may have | |
1399 | * changed things. Don't bother checking the hash because we're | |
1400 | * about to compare the whole name anyway. | |
1401 | */ | |
1402 | if (dentry->d_parent != parent) | |
1403 | goto next; | |
1404 | ||
d0185c08 LT |
1405 | /* non-existing due to RCU? */ |
1406 | if (d_unhashed(dentry)) | |
1407 | goto next; | |
1408 | ||
1da177e4 LT |
1409 | /* |
1410 | * It is safe to compare names since d_move() cannot | |
1411 | * change the qstr (protected by d_lock). | |
1412 | */ | |
1413 | qstr = &dentry->d_name; | |
1414 | if (parent->d_op && parent->d_op->d_compare) { | |
1415 | if (parent->d_op->d_compare(parent, qstr, name)) | |
1416 | goto next; | |
1417 | } else { | |
1418 | if (qstr->len != len) | |
1419 | goto next; | |
1420 | if (memcmp(qstr->name, str, len)) | |
1421 | goto next; | |
1422 | } | |
1423 | ||
d0185c08 LT |
1424 | atomic_inc(&dentry->d_count); |
1425 | found = dentry; | |
1da177e4 LT |
1426 | spin_unlock(&dentry->d_lock); |
1427 | break; | |
1428 | next: | |
1429 | spin_unlock(&dentry->d_lock); | |
1430 | } | |
1431 | rcu_read_unlock(); | |
1432 | ||
1433 | return found; | |
1434 | } | |
1435 | ||
3e7e241f EB |
1436 | /** |
1437 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1438 | * @dir: Directory to search in | |
1439 | * @name: qstr of name we wish to find | |
1440 | * | |
1441 | * On hash failure or on lookup failure NULL is returned. | |
1442 | */ | |
1443 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1444 | { | |
1445 | struct dentry *dentry = NULL; | |
1446 | ||
1447 | /* | |
1448 | * Check for a fs-specific hash function. Note that we must | |
1449 | * calculate the standard hash first, as the d_op->d_hash() | |
1450 | * routine may choose to leave the hash value unchanged. | |
1451 | */ | |
1452 | name->hash = full_name_hash(name->name, name->len); | |
1453 | if (dir->d_op && dir->d_op->d_hash) { | |
1454 | if (dir->d_op->d_hash(dir, name) < 0) | |
1455 | goto out; | |
1456 | } | |
1457 | dentry = d_lookup(dir, name); | |
1458 | out: | |
1459 | return dentry; | |
1460 | } | |
1461 | ||
1da177e4 LT |
1462 | /** |
1463 | * d_validate - verify dentry provided from insecure source | |
1464 | * @dentry: The dentry alleged to be valid child of @dparent | |
1465 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1466 | * |
1467 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1468 | * This is used by ncpfs in its readdir implementation. | |
1469 | * Zero is returned in the dentry is invalid. | |
1470 | */ | |
1471 | ||
1472 | int d_validate(struct dentry *dentry, struct dentry *dparent) | |
1473 | { | |
1474 | struct hlist_head *base; | |
1475 | struct hlist_node *lhp; | |
1476 | ||
1477 | /* Check whether the ptr might be valid at all.. */ | |
1478 | if (!kmem_ptr_validate(dentry_cache, dentry)) | |
1479 | goto out; | |
1480 | ||
1481 | if (dentry->d_parent != dparent) | |
1482 | goto out; | |
1483 | ||
1484 | spin_lock(&dcache_lock); | |
1485 | base = d_hash(dparent, dentry->d_name.hash); | |
1486 | hlist_for_each(lhp,base) { | |
665a7583 | 1487 | /* hlist_for_each_entry_rcu() not required for d_hash list |
1da177e4 LT |
1488 | * as it is parsed under dcache_lock |
1489 | */ | |
1490 | if (dentry == hlist_entry(lhp, struct dentry, d_hash)) { | |
1491 | __dget_locked(dentry); | |
1492 | spin_unlock(&dcache_lock); | |
1493 | return 1; | |
1494 | } | |
1495 | } | |
1496 | spin_unlock(&dcache_lock); | |
1497 | out: | |
1498 | return 0; | |
1499 | } | |
ec4f8605 | 1500 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1501 | |
1502 | /* | |
1503 | * When a file is deleted, we have two options: | |
1504 | * - turn this dentry into a negative dentry | |
1505 | * - unhash this dentry and free it. | |
1506 | * | |
1507 | * Usually, we want to just turn this into | |
1508 | * a negative dentry, but if anybody else is | |
1509 | * currently using the dentry or the inode | |
1510 | * we can't do that and we fall back on removing | |
1511 | * it from the hash queues and waiting for | |
1512 | * it to be deleted later when it has no users | |
1513 | */ | |
1514 | ||
1515 | /** | |
1516 | * d_delete - delete a dentry | |
1517 | * @dentry: The dentry to delete | |
1518 | * | |
1519 | * Turn the dentry into a negative dentry if possible, otherwise | |
1520 | * remove it from the hash queues so it can be deleted later | |
1521 | */ | |
1522 | ||
1523 | void d_delete(struct dentry * dentry) | |
1524 | { | |
7a91bf7f | 1525 | int isdir = 0; |
1da177e4 LT |
1526 | /* |
1527 | * Are we the only user? | |
1528 | */ | |
1529 | spin_lock(&dcache_lock); | |
1530 | spin_lock(&dentry->d_lock); | |
7a91bf7f | 1531 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
1da177e4 | 1532 | if (atomic_read(&dentry->d_count) == 1) { |
13e3c5e5 | 1533 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
1da177e4 | 1534 | dentry_iput(dentry); |
7a91bf7f | 1535 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1536 | return; |
1537 | } | |
1538 | ||
1539 | if (!d_unhashed(dentry)) | |
1540 | __d_drop(dentry); | |
1541 | ||
1542 | spin_unlock(&dentry->d_lock); | |
1543 | spin_unlock(&dcache_lock); | |
7a91bf7f JM |
1544 | |
1545 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1546 | } |
ec4f8605 | 1547 | EXPORT_SYMBOL(d_delete); |
1da177e4 LT |
1548 | |
1549 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1550 | { | |
1551 | ||
1552 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1553 | hlist_add_head_rcu(&entry->d_hash, list); | |
1554 | } | |
1555 | ||
770bfad8 DH |
1556 | static void _d_rehash(struct dentry * entry) |
1557 | { | |
1558 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1559 | } | |
1560 | ||
1da177e4 LT |
1561 | /** |
1562 | * d_rehash - add an entry back to the hash | |
1563 | * @entry: dentry to add to the hash | |
1564 | * | |
1565 | * Adds a dentry to the hash according to its name. | |
1566 | */ | |
1567 | ||
1568 | void d_rehash(struct dentry * entry) | |
1569 | { | |
1da177e4 LT |
1570 | spin_lock(&dcache_lock); |
1571 | spin_lock(&entry->d_lock); | |
770bfad8 | 1572 | _d_rehash(entry); |
1da177e4 LT |
1573 | spin_unlock(&entry->d_lock); |
1574 | spin_unlock(&dcache_lock); | |
1575 | } | |
ec4f8605 | 1576 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 1577 | |
1da177e4 LT |
1578 | /* |
1579 | * When switching names, the actual string doesn't strictly have to | |
1580 | * be preserved in the target - because we're dropping the target | |
1581 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1582 | * the new name before we switch. | |
1583 | * | |
1584 | * Note that we have to be a lot more careful about getting the hash | |
1585 | * switched - we have to switch the hash value properly even if it | |
1586 | * then no longer matches the actual (corrupted) string of the target. | |
1587 | * The hash value has to match the hash queue that the dentry is on.. | |
1588 | */ | |
1589 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
1590 | { | |
1591 | if (dname_external(target)) { | |
1592 | if (dname_external(dentry)) { | |
1593 | /* | |
1594 | * Both external: swap the pointers | |
1595 | */ | |
9a8d5bb4 | 1596 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
1597 | } else { |
1598 | /* | |
1599 | * dentry:internal, target:external. Steal target's | |
1600 | * storage and make target internal. | |
1601 | */ | |
321bcf92 BF |
1602 | memcpy(target->d_iname, dentry->d_name.name, |
1603 | dentry->d_name.len + 1); | |
1da177e4 LT |
1604 | dentry->d_name.name = target->d_name.name; |
1605 | target->d_name.name = target->d_iname; | |
1606 | } | |
1607 | } else { | |
1608 | if (dname_external(dentry)) { | |
1609 | /* | |
1610 | * dentry:external, target:internal. Give dentry's | |
1611 | * storage to target and make dentry internal | |
1612 | */ | |
1613 | memcpy(dentry->d_iname, target->d_name.name, | |
1614 | target->d_name.len + 1); | |
1615 | target->d_name.name = dentry->d_name.name; | |
1616 | dentry->d_name.name = dentry->d_iname; | |
1617 | } else { | |
1618 | /* | |
1619 | * Both are internal. Just copy target to dentry | |
1620 | */ | |
1621 | memcpy(dentry->d_iname, target->d_name.name, | |
1622 | target->d_name.len + 1); | |
dc711ca3 AV |
1623 | dentry->d_name.len = target->d_name.len; |
1624 | return; | |
1da177e4 LT |
1625 | } |
1626 | } | |
9a8d5bb4 | 1627 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
1628 | } |
1629 | ||
1630 | /* | |
1631 | * We cannibalize "target" when moving dentry on top of it, | |
1632 | * because it's going to be thrown away anyway. We could be more | |
1633 | * polite about it, though. | |
1634 | * | |
1635 | * This forceful removal will result in ugly /proc output if | |
1636 | * somebody holds a file open that got deleted due to a rename. | |
1637 | * We could be nicer about the deleted file, and let it show | |
bc154b1e BF |
1638 | * up under the name it had before it was deleted rather than |
1639 | * under the original name of the file that was moved on top of it. | |
1da177e4 LT |
1640 | */ |
1641 | ||
9eaef27b TM |
1642 | /* |
1643 | * d_move_locked - move a dentry | |
1da177e4 LT |
1644 | * @dentry: entry to move |
1645 | * @target: new dentry | |
1646 | * | |
1647 | * Update the dcache to reflect the move of a file name. Negative | |
1648 | * dcache entries should not be moved in this way. | |
1649 | */ | |
9eaef27b | 1650 | static void d_move_locked(struct dentry * dentry, struct dentry * target) |
1da177e4 LT |
1651 | { |
1652 | struct hlist_head *list; | |
1653 | ||
1654 | if (!dentry->d_inode) | |
1655 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1656 | ||
1da177e4 LT |
1657 | write_seqlock(&rename_lock); |
1658 | /* | |
1659 | * XXXX: do we really need to take target->d_lock? | |
1660 | */ | |
1661 | if (target < dentry) { | |
1662 | spin_lock(&target->d_lock); | |
a90b9c05 | 1663 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1664 | } else { |
1665 | spin_lock(&dentry->d_lock); | |
a90b9c05 | 1666 | spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1667 | } |
1668 | ||
1669 | /* Move the dentry to the target hash queue, if on different bucket */ | |
f77e3498 | 1670 | if (d_unhashed(dentry)) |
1da177e4 LT |
1671 | goto already_unhashed; |
1672 | ||
1673 | hlist_del_rcu(&dentry->d_hash); | |
1674 | ||
1675 | already_unhashed: | |
1676 | list = d_hash(target->d_parent, target->d_name.hash); | |
1677 | __d_rehash(dentry, list); | |
1678 | ||
1679 | /* Unhash the target: dput() will then get rid of it */ | |
1680 | __d_drop(target); | |
1681 | ||
5160ee6f ED |
1682 | list_del(&dentry->d_u.d_child); |
1683 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
1684 | |
1685 | /* Switch the names.. */ | |
1686 | switch_names(dentry, target); | |
9a8d5bb4 | 1687 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
1688 | |
1689 | /* ... and switch the parents */ | |
1690 | if (IS_ROOT(dentry)) { | |
1691 | dentry->d_parent = target->d_parent; | |
1692 | target->d_parent = target; | |
5160ee6f | 1693 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 1694 | } else { |
9a8d5bb4 | 1695 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
1696 | |
1697 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 1698 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
1699 | } |
1700 | ||
5160ee6f | 1701 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 1702 | spin_unlock(&target->d_lock); |
c32ccd87 | 1703 | fsnotify_d_move(dentry); |
1da177e4 LT |
1704 | spin_unlock(&dentry->d_lock); |
1705 | write_sequnlock(&rename_lock); | |
9eaef27b TM |
1706 | } |
1707 | ||
1708 | /** | |
1709 | * d_move - move a dentry | |
1710 | * @dentry: entry to move | |
1711 | * @target: new dentry | |
1712 | * | |
1713 | * Update the dcache to reflect the move of a file name. Negative | |
1714 | * dcache entries should not be moved in this way. | |
1715 | */ | |
1716 | ||
1717 | void d_move(struct dentry * dentry, struct dentry * target) | |
1718 | { | |
1719 | spin_lock(&dcache_lock); | |
1720 | d_move_locked(dentry, target); | |
1da177e4 LT |
1721 | spin_unlock(&dcache_lock); |
1722 | } | |
ec4f8605 | 1723 | EXPORT_SYMBOL(d_move); |
1da177e4 | 1724 | |
e2761a11 OH |
1725 | /** |
1726 | * d_ancestor - search for an ancestor | |
1727 | * @p1: ancestor dentry | |
1728 | * @p2: child dentry | |
1729 | * | |
1730 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
1731 | * an ancestor of p2, else NULL. | |
9eaef27b | 1732 | */ |
e2761a11 | 1733 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
1734 | { |
1735 | struct dentry *p; | |
1736 | ||
871c0067 | 1737 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 1738 | if (p->d_parent == p1) |
e2761a11 | 1739 | return p; |
9eaef27b | 1740 | } |
e2761a11 | 1741 | return NULL; |
9eaef27b TM |
1742 | } |
1743 | ||
1744 | /* | |
1745 | * This helper attempts to cope with remotely renamed directories | |
1746 | * | |
1747 | * It assumes that the caller is already holding | |
1748 | * dentry->d_parent->d_inode->i_mutex and the dcache_lock | |
1749 | * | |
1750 | * Note: If ever the locking in lock_rename() changes, then please | |
1751 | * remember to update this too... | |
9eaef27b TM |
1752 | */ |
1753 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
31f3e0b3 | 1754 | __releases(dcache_lock) |
9eaef27b TM |
1755 | { |
1756 | struct mutex *m1 = NULL, *m2 = NULL; | |
1757 | struct dentry *ret; | |
1758 | ||
1759 | /* If alias and dentry share a parent, then no extra locks required */ | |
1760 | if (alias->d_parent == dentry->d_parent) | |
1761 | goto out_unalias; | |
1762 | ||
1763 | /* Check for loops */ | |
1764 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 1765 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
1766 | goto out_err; |
1767 | ||
1768 | /* See lock_rename() */ | |
1769 | ret = ERR_PTR(-EBUSY); | |
1770 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
1771 | goto out_err; | |
1772 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
1773 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
1774 | goto out_err; | |
1775 | m2 = &alias->d_parent->d_inode->i_mutex; | |
1776 | out_unalias: | |
1777 | d_move_locked(alias, dentry); | |
1778 | ret = alias; | |
1779 | out_err: | |
1780 | spin_unlock(&dcache_lock); | |
1781 | if (m2) | |
1782 | mutex_unlock(m2); | |
1783 | if (m1) | |
1784 | mutex_unlock(m1); | |
1785 | return ret; | |
1786 | } | |
1787 | ||
770bfad8 DH |
1788 | /* |
1789 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
1790 | * named dentry in place of the dentry to be replaced. | |
1791 | */ | |
1792 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
1793 | { | |
1794 | struct dentry *dparent, *aparent; | |
1795 | ||
1796 | switch_names(dentry, anon); | |
9a8d5bb4 | 1797 | swap(dentry->d_name.hash, anon->d_name.hash); |
770bfad8 DH |
1798 | |
1799 | dparent = dentry->d_parent; | |
1800 | aparent = anon->d_parent; | |
1801 | ||
1802 | dentry->d_parent = (aparent == anon) ? dentry : aparent; | |
1803 | list_del(&dentry->d_u.d_child); | |
1804 | if (!IS_ROOT(dentry)) | |
1805 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
1806 | else | |
1807 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
1808 | ||
1809 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
1810 | list_del(&anon->d_u.d_child); | |
1811 | if (!IS_ROOT(anon)) | |
1812 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
1813 | else | |
1814 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
1815 | ||
1816 | anon->d_flags &= ~DCACHE_DISCONNECTED; | |
1817 | } | |
1818 | ||
1819 | /** | |
1820 | * d_materialise_unique - introduce an inode into the tree | |
1821 | * @dentry: candidate dentry | |
1822 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
1823 | * | |
1824 | * Introduces an dentry into the tree, substituting an extant disconnected | |
1825 | * root directory alias in its place if there is one | |
1826 | */ | |
1827 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
1828 | { | |
9eaef27b | 1829 | struct dentry *actual; |
770bfad8 DH |
1830 | |
1831 | BUG_ON(!d_unhashed(dentry)); | |
1832 | ||
1833 | spin_lock(&dcache_lock); | |
1834 | ||
1835 | if (!inode) { | |
1836 | actual = dentry; | |
360da900 | 1837 | __d_instantiate(dentry, NULL); |
770bfad8 DH |
1838 | goto found_lock; |
1839 | } | |
1840 | ||
9eaef27b TM |
1841 | if (S_ISDIR(inode->i_mode)) { |
1842 | struct dentry *alias; | |
1843 | ||
1844 | /* Does an aliased dentry already exist? */ | |
1845 | alias = __d_find_alias(inode, 0); | |
1846 | if (alias) { | |
1847 | actual = alias; | |
1848 | /* Is this an anonymous mountpoint that we could splice | |
1849 | * into our tree? */ | |
1850 | if (IS_ROOT(alias)) { | |
1851 | spin_lock(&alias->d_lock); | |
1852 | __d_materialise_dentry(dentry, alias); | |
1853 | __d_drop(alias); | |
1854 | goto found; | |
1855 | } | |
1856 | /* Nope, but we must(!) avoid directory aliasing */ | |
1857 | actual = __d_unalias(dentry, alias); | |
1858 | if (IS_ERR(actual)) | |
1859 | dput(alias); | |
1860 | goto out_nolock; | |
1861 | } | |
770bfad8 DH |
1862 | } |
1863 | ||
1864 | /* Add a unique reference */ | |
1865 | actual = __d_instantiate_unique(dentry, inode); | |
1866 | if (!actual) | |
1867 | actual = dentry; | |
1868 | else if (unlikely(!d_unhashed(actual))) | |
1869 | goto shouldnt_be_hashed; | |
1870 | ||
1871 | found_lock: | |
1872 | spin_lock(&actual->d_lock); | |
1873 | found: | |
1874 | _d_rehash(actual); | |
1875 | spin_unlock(&actual->d_lock); | |
1876 | spin_unlock(&dcache_lock); | |
9eaef27b | 1877 | out_nolock: |
770bfad8 DH |
1878 | if (actual == dentry) { |
1879 | security_d_instantiate(dentry, inode); | |
1880 | return NULL; | |
1881 | } | |
1882 | ||
1883 | iput(inode); | |
1884 | return actual; | |
1885 | ||
770bfad8 DH |
1886 | shouldnt_be_hashed: |
1887 | spin_unlock(&dcache_lock); | |
1888 | BUG(); | |
770bfad8 | 1889 | } |
ec4f8605 | 1890 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 1891 | |
cdd16d02 | 1892 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
1893 | { |
1894 | *buflen -= namelen; | |
1895 | if (*buflen < 0) | |
1896 | return -ENAMETOOLONG; | |
1897 | *buffer -= namelen; | |
1898 | memcpy(*buffer, str, namelen); | |
1899 | return 0; | |
1900 | } | |
1901 | ||
cdd16d02 MS |
1902 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
1903 | { | |
1904 | return prepend(buffer, buflen, name->name, name->len); | |
1905 | } | |
1906 | ||
1da177e4 | 1907 | /** |
31f3e0b3 | 1908 | * __d_path - return the path of a dentry |
9d1bc601 MS |
1909 | * @path: the dentry/vfsmount to report |
1910 | * @root: root vfsmnt/dentry (may be modified by this function) | |
1da177e4 LT |
1911 | * @buffer: buffer to return value in |
1912 | * @buflen: buffer length | |
1913 | * | |
552ce544 LT |
1914 | * Convert a dentry into an ASCII path name. If the entry has been deleted |
1915 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1da177e4 | 1916 | * |
52afeefb AV |
1917 | * Returns a pointer into the buffer or an error code if the |
1918 | * path was too long. | |
552ce544 LT |
1919 | * |
1920 | * "buflen" should be positive. Caller holds the dcache_lock. | |
9d1bc601 MS |
1921 | * |
1922 | * If path is not reachable from the supplied root, then the value of | |
1923 | * root is changed (without modifying refcounts). | |
1da177e4 | 1924 | */ |
9d1bc601 MS |
1925 | char *__d_path(const struct path *path, struct path *root, |
1926 | char *buffer, int buflen) | |
1da177e4 | 1927 | { |
9d1bc601 MS |
1928 | struct dentry *dentry = path->dentry; |
1929 | struct vfsmount *vfsmnt = path->mnt; | |
cdd16d02 MS |
1930 | char *end = buffer + buflen; |
1931 | char *retval; | |
6092d048 | 1932 | |
be285c71 | 1933 | spin_lock(&vfsmount_lock); |
6092d048 | 1934 | prepend(&end, &buflen, "\0", 1); |
f3da392e | 1935 | if (d_unlinked(dentry) && |
6092d048 | 1936 | (prepend(&end, &buflen, " (deleted)", 10) != 0)) |
552ce544 | 1937 | goto Elong; |
552ce544 LT |
1938 | |
1939 | if (buflen < 1) | |
1940 | goto Elong; | |
1941 | /* Get '/' right */ | |
1942 | retval = end-1; | |
1943 | *retval = '/'; | |
1944 | ||
1945 | for (;;) { | |
1da177e4 LT |
1946 | struct dentry * parent; |
1947 | ||
329c97f0 | 1948 | if (dentry == root->dentry && vfsmnt == root->mnt) |
552ce544 | 1949 | break; |
1da177e4 | 1950 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 1951 | /* Global root? */ |
1da177e4 | 1952 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
1953 | goto global_root; |
1954 | } | |
1955 | dentry = vfsmnt->mnt_mountpoint; | |
1956 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
1957 | continue; |
1958 | } | |
1959 | parent = dentry->d_parent; | |
1960 | prefetch(parent); | |
cdd16d02 | 1961 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 | 1962 | (prepend(&end, &buflen, "/", 1) != 0)) |
552ce544 | 1963 | goto Elong; |
552ce544 | 1964 | retval = end; |
1da177e4 LT |
1965 | dentry = parent; |
1966 | } | |
1967 | ||
be285c71 AG |
1968 | out: |
1969 | spin_unlock(&vfsmount_lock); | |
552ce544 | 1970 | return retval; |
1da177e4 LT |
1971 | |
1972 | global_root: | |
6092d048 | 1973 | retval += 1; /* hit the slash */ |
cdd16d02 | 1974 | if (prepend_name(&retval, &buflen, &dentry->d_name) != 0) |
1da177e4 | 1975 | goto Elong; |
9d1bc601 MS |
1976 | root->mnt = vfsmnt; |
1977 | root->dentry = dentry; | |
be285c71 AG |
1978 | goto out; |
1979 | ||
1da177e4 | 1980 | Elong: |
be285c71 AG |
1981 | retval = ERR_PTR(-ENAMETOOLONG); |
1982 | goto out; | |
1da177e4 LT |
1983 | } |
1984 | ||
a03a8a70 JB |
1985 | /** |
1986 | * d_path - return the path of a dentry | |
cf28b486 | 1987 | * @path: path to report |
a03a8a70 JB |
1988 | * @buf: buffer to return value in |
1989 | * @buflen: buffer length | |
1990 | * | |
1991 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
1992 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1993 | * | |
52afeefb AV |
1994 | * Returns a pointer into the buffer or an error code if the path was |
1995 | * too long. Note: Callers should use the returned pointer, not the passed | |
1996 | * in buffer, to use the name! The implementation often starts at an offset | |
1997 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 1998 | * |
31f3e0b3 | 1999 | * "buflen" should be positive. |
a03a8a70 | 2000 | */ |
20d4fdc1 | 2001 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 LT |
2002 | { |
2003 | char *res; | |
6ac08c39 | 2004 | struct path root; |
9d1bc601 | 2005 | struct path tmp; |
1da177e4 | 2006 | |
c23fbb6b ED |
2007 | /* |
2008 | * We have various synthetic filesystems that never get mounted. On | |
2009 | * these filesystems dentries are never used for lookup purposes, and | |
2010 | * thus don't need to be hashed. They also don't need a name until a | |
2011 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2012 | * below allows us to generate a name for these objects on demand: | |
2013 | */ | |
cf28b486 JB |
2014 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2015 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2016 | |
1da177e4 | 2017 | read_lock(¤t->fs->lock); |
6ac08c39 | 2018 | root = current->fs->root; |
6092d048 | 2019 | path_get(&root); |
1da177e4 | 2020 | read_unlock(¤t->fs->lock); |
552ce544 | 2021 | spin_lock(&dcache_lock); |
9d1bc601 MS |
2022 | tmp = root; |
2023 | res = __d_path(path, &tmp, buf, buflen); | |
552ce544 | 2024 | spin_unlock(&dcache_lock); |
6ac08c39 | 2025 | path_put(&root); |
1da177e4 LT |
2026 | return res; |
2027 | } | |
ec4f8605 | 2028 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2029 | |
c23fbb6b ED |
2030 | /* |
2031 | * Helper function for dentry_operations.d_dname() members | |
2032 | */ | |
2033 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2034 | const char *fmt, ...) | |
2035 | { | |
2036 | va_list args; | |
2037 | char temp[64]; | |
2038 | int sz; | |
2039 | ||
2040 | va_start(args, fmt); | |
2041 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2042 | va_end(args); | |
2043 | ||
2044 | if (sz > sizeof(temp) || sz > buflen) | |
2045 | return ERR_PTR(-ENAMETOOLONG); | |
2046 | ||
2047 | buffer += buflen - sz; | |
2048 | return memcpy(buffer, temp, sz); | |
2049 | } | |
2050 | ||
6092d048 RP |
2051 | /* |
2052 | * Write full pathname from the root of the filesystem into the buffer. | |
2053 | */ | |
c103135c | 2054 | char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2055 | { |
2056 | char *end = buf + buflen; | |
2057 | char *retval; | |
2058 | ||
6092d048 | 2059 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2060 | if (buflen < 1) |
2061 | goto Elong; | |
2062 | /* Get '/' right */ | |
2063 | retval = end-1; | |
2064 | *retval = '/'; | |
2065 | ||
cdd16d02 MS |
2066 | while (!IS_ROOT(dentry)) { |
2067 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 2068 | |
6092d048 | 2069 | prefetch(parent); |
cdd16d02 | 2070 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 RP |
2071 | (prepend(&end, &buflen, "/", 1) != 0)) |
2072 | goto Elong; | |
2073 | ||
2074 | retval = end; | |
2075 | dentry = parent; | |
2076 | } | |
c103135c AV |
2077 | return retval; |
2078 | Elong: | |
2079 | return ERR_PTR(-ENAMETOOLONG); | |
2080 | } | |
2081 | EXPORT_SYMBOL(__dentry_path); | |
2082 | ||
2083 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2084 | { | |
2085 | char *p = NULL; | |
2086 | char *retval; | |
2087 | ||
2088 | spin_lock(&dcache_lock); | |
2089 | if (d_unlinked(dentry)) { | |
2090 | p = buf + buflen; | |
2091 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2092 | goto Elong; | |
2093 | buflen++; | |
2094 | } | |
2095 | retval = __dentry_path(dentry, buf, buflen); | |
6092d048 | 2096 | spin_unlock(&dcache_lock); |
c103135c AV |
2097 | if (!IS_ERR(retval) && p) |
2098 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2099 | return retval; |
2100 | Elong: | |
2101 | spin_unlock(&dcache_lock); | |
2102 | return ERR_PTR(-ENAMETOOLONG); | |
2103 | } | |
2104 | ||
1da177e4 LT |
2105 | /* |
2106 | * NOTE! The user-level library version returns a | |
2107 | * character pointer. The kernel system call just | |
2108 | * returns the length of the buffer filled (which | |
2109 | * includes the ending '\0' character), or a negative | |
2110 | * error value. So libc would do something like | |
2111 | * | |
2112 | * char *getcwd(char * buf, size_t size) | |
2113 | * { | |
2114 | * int retval; | |
2115 | * | |
2116 | * retval = sys_getcwd(buf, size); | |
2117 | * if (retval >= 0) | |
2118 | * return buf; | |
2119 | * errno = -retval; | |
2120 | * return NULL; | |
2121 | * } | |
2122 | */ | |
3cdad428 | 2123 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2124 | { |
552ce544 | 2125 | int error; |
6ac08c39 | 2126 | struct path pwd, root; |
552ce544 | 2127 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2128 | |
2129 | if (!page) | |
2130 | return -ENOMEM; | |
2131 | ||
2132 | read_lock(¤t->fs->lock); | |
6ac08c39 | 2133 | pwd = current->fs->pwd; |
6092d048 | 2134 | path_get(&pwd); |
6ac08c39 | 2135 | root = current->fs->root; |
6092d048 | 2136 | path_get(&root); |
1da177e4 LT |
2137 | read_unlock(¤t->fs->lock); |
2138 | ||
552ce544 | 2139 | error = -ENOENT; |
552ce544 | 2140 | spin_lock(&dcache_lock); |
f3da392e | 2141 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2142 | unsigned long len; |
9d1bc601 | 2143 | struct path tmp = root; |
552ce544 | 2144 | char * cwd; |
1da177e4 | 2145 | |
9d1bc601 | 2146 | cwd = __d_path(&pwd, &tmp, page, PAGE_SIZE); |
552ce544 LT |
2147 | spin_unlock(&dcache_lock); |
2148 | ||
2149 | error = PTR_ERR(cwd); | |
2150 | if (IS_ERR(cwd)) | |
2151 | goto out; | |
2152 | ||
2153 | error = -ERANGE; | |
2154 | len = PAGE_SIZE + page - cwd; | |
2155 | if (len <= size) { | |
2156 | error = len; | |
2157 | if (copy_to_user(buf, cwd, len)) | |
2158 | error = -EFAULT; | |
2159 | } | |
2160 | } else | |
2161 | spin_unlock(&dcache_lock); | |
1da177e4 LT |
2162 | |
2163 | out: | |
6ac08c39 JB |
2164 | path_put(&pwd); |
2165 | path_put(&root); | |
1da177e4 LT |
2166 | free_page((unsigned long) page); |
2167 | return error; | |
2168 | } | |
2169 | ||
2170 | /* | |
2171 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2172 | * | |
2173 | * Trivially implemented using the dcache structure | |
2174 | */ | |
2175 | ||
2176 | /** | |
2177 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2178 | * @new_dentry: new dentry | |
2179 | * @old_dentry: old dentry | |
2180 | * | |
2181 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2182 | * Returns 0 otherwise. | |
2183 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2184 | */ | |
2185 | ||
e2761a11 | 2186 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2187 | { |
2188 | int result; | |
1da177e4 LT |
2189 | unsigned long seq; |
2190 | ||
e2761a11 OH |
2191 | if (new_dentry == old_dentry) |
2192 | return 1; | |
2193 | ||
2194 | /* | |
2195 | * Need rcu_readlock to protect against the d_parent trashing | |
2196 | * due to d_move | |
1da177e4 LT |
2197 | */ |
2198 | rcu_read_lock(); | |
e2761a11 | 2199 | do { |
1da177e4 | 2200 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2201 | seq = read_seqbegin(&rename_lock); |
e2761a11 | 2202 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2203 | result = 1; |
e2761a11 OH |
2204 | else |
2205 | result = 0; | |
1da177e4 LT |
2206 | } while (read_seqretry(&rename_lock, seq)); |
2207 | rcu_read_unlock(); | |
2208 | ||
2209 | return result; | |
2210 | } | |
2211 | ||
2096f759 AV |
2212 | int path_is_under(struct path *path1, struct path *path2) |
2213 | { | |
2214 | struct vfsmount *mnt = path1->mnt; | |
2215 | struct dentry *dentry = path1->dentry; | |
2216 | int res; | |
2217 | spin_lock(&vfsmount_lock); | |
2218 | if (mnt != path2->mnt) { | |
2219 | for (;;) { | |
2220 | if (mnt->mnt_parent == mnt) { | |
2221 | spin_unlock(&vfsmount_lock); | |
2222 | return 0; | |
2223 | } | |
2224 | if (mnt->mnt_parent == path2->mnt) | |
2225 | break; | |
2226 | mnt = mnt->mnt_parent; | |
2227 | } | |
2228 | dentry = mnt->mnt_mountpoint; | |
2229 | } | |
2230 | res = is_subdir(dentry, path2->dentry); | |
2231 | spin_unlock(&vfsmount_lock); | |
2232 | return res; | |
2233 | } | |
2234 | EXPORT_SYMBOL(path_is_under); | |
2235 | ||
1da177e4 LT |
2236 | void d_genocide(struct dentry *root) |
2237 | { | |
2238 | struct dentry *this_parent = root; | |
2239 | struct list_head *next; | |
2240 | ||
2241 | spin_lock(&dcache_lock); | |
2242 | repeat: | |
2243 | next = this_parent->d_subdirs.next; | |
2244 | resume: | |
2245 | while (next != &this_parent->d_subdirs) { | |
2246 | struct list_head *tmp = next; | |
5160ee6f | 2247 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
2248 | next = tmp->next; |
2249 | if (d_unhashed(dentry)||!dentry->d_inode) | |
2250 | continue; | |
2251 | if (!list_empty(&dentry->d_subdirs)) { | |
2252 | this_parent = dentry; | |
2253 | goto repeat; | |
2254 | } | |
2255 | atomic_dec(&dentry->d_count); | |
2256 | } | |
2257 | if (this_parent != root) { | |
5160ee6f | 2258 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
2259 | atomic_dec(&this_parent->d_count); |
2260 | this_parent = this_parent->d_parent; | |
2261 | goto resume; | |
2262 | } | |
2263 | spin_unlock(&dcache_lock); | |
2264 | } | |
2265 | ||
2266 | /** | |
2267 | * find_inode_number - check for dentry with name | |
2268 | * @dir: directory to check | |
2269 | * @name: Name to find. | |
2270 | * | |
2271 | * Check whether a dentry already exists for the given name, | |
2272 | * and return the inode number if it has an inode. Otherwise | |
2273 | * 0 is returned. | |
2274 | * | |
2275 | * This routine is used to post-process directory listings for | |
2276 | * filesystems using synthetic inode numbers, and is necessary | |
2277 | * to keep getcwd() working. | |
2278 | */ | |
2279 | ||
2280 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2281 | { | |
2282 | struct dentry * dentry; | |
2283 | ino_t ino = 0; | |
2284 | ||
3e7e241f EB |
2285 | dentry = d_hash_and_lookup(dir, name); |
2286 | if (dentry) { | |
1da177e4 LT |
2287 | if (dentry->d_inode) |
2288 | ino = dentry->d_inode->i_ino; | |
2289 | dput(dentry); | |
2290 | } | |
1da177e4 LT |
2291 | return ino; |
2292 | } | |
ec4f8605 | 2293 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2294 | |
2295 | static __initdata unsigned long dhash_entries; | |
2296 | static int __init set_dhash_entries(char *str) | |
2297 | { | |
2298 | if (!str) | |
2299 | return 0; | |
2300 | dhash_entries = simple_strtoul(str, &str, 0); | |
2301 | return 1; | |
2302 | } | |
2303 | __setup("dhash_entries=", set_dhash_entries); | |
2304 | ||
2305 | static void __init dcache_init_early(void) | |
2306 | { | |
2307 | int loop; | |
2308 | ||
2309 | /* If hashes are distributed across NUMA nodes, defer | |
2310 | * hash allocation until vmalloc space is available. | |
2311 | */ | |
2312 | if (hashdist) | |
2313 | return; | |
2314 | ||
2315 | dentry_hashtable = | |
2316 | alloc_large_system_hash("Dentry cache", | |
2317 | sizeof(struct hlist_head), | |
2318 | dhash_entries, | |
2319 | 13, | |
2320 | HASH_EARLY, | |
2321 | &d_hash_shift, | |
2322 | &d_hash_mask, | |
2323 | 0); | |
2324 | ||
2325 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2326 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2327 | } | |
2328 | ||
74bf17cf | 2329 | static void __init dcache_init(void) |
1da177e4 LT |
2330 | { |
2331 | int loop; | |
2332 | ||
2333 | /* | |
2334 | * A constructor could be added for stable state like the lists, | |
2335 | * but it is probably not worth it because of the cache nature | |
2336 | * of the dcache. | |
2337 | */ | |
0a31bd5f CL |
2338 | dentry_cache = KMEM_CACHE(dentry, |
2339 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2340 | |
8e1f936b | 2341 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2342 | |
2343 | /* Hash may have been set up in dcache_init_early */ | |
2344 | if (!hashdist) | |
2345 | return; | |
2346 | ||
2347 | dentry_hashtable = | |
2348 | alloc_large_system_hash("Dentry cache", | |
2349 | sizeof(struct hlist_head), | |
2350 | dhash_entries, | |
2351 | 13, | |
2352 | 0, | |
2353 | &d_hash_shift, | |
2354 | &d_hash_mask, | |
2355 | 0); | |
2356 | ||
2357 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2358 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2359 | } | |
2360 | ||
2361 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2362 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2363 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2364 | |
1da177e4 LT |
2365 | EXPORT_SYMBOL(d_genocide); |
2366 | ||
1da177e4 LT |
2367 | void __init vfs_caches_init_early(void) |
2368 | { | |
2369 | dcache_init_early(); | |
2370 | inode_init_early(); | |
2371 | } | |
2372 | ||
2373 | void __init vfs_caches_init(unsigned long mempages) | |
2374 | { | |
2375 | unsigned long reserve; | |
2376 | ||
2377 | /* Base hash sizes on available memory, with a reserve equal to | |
2378 | 150% of current kernel size */ | |
2379 | ||
2380 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2381 | mempages -= reserve; | |
2382 | ||
2383 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2384 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2385 | |
74bf17cf DC |
2386 | dcache_init(); |
2387 | inode_init(); | |
1da177e4 | 2388 | files_init(mempages); |
74bf17cf | 2389 | mnt_init(); |
1da177e4 LT |
2390 | bdev_cache_init(); |
2391 | chrdev_init(); | |
2392 | } |