]>
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 | ||
9c82ab9c | 70 | static void __d_free(struct rcu_head *head) |
1da177e4 | 71 | { |
9c82ab9c CH |
72 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
73 | ||
fd217f4d | 74 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
75 | if (dname_external(dentry)) |
76 | kfree(dentry->d_name.name); | |
77 | kmem_cache_free(dentry_cache, dentry); | |
78 | } | |
79 | ||
80 | /* | |
81 | * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry | |
82 | * inside dcache_lock. | |
83 | */ | |
84 | static void d_free(struct dentry *dentry) | |
85 | { | |
86 | if (dentry->d_op && dentry->d_op->d_release) | |
87 | dentry->d_op->d_release(dentry); | |
b3423415 | 88 | /* if dentry was never inserted into hash, immediate free is OK */ |
e8462caa | 89 | if (hlist_unhashed(&dentry->d_hash)) |
9c82ab9c | 90 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 91 | else |
9c82ab9c | 92 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
93 | } |
94 | ||
95 | /* | |
96 | * Release the dentry's inode, using the filesystem | |
97 | * d_iput() operation if defined. | |
1da177e4 | 98 | */ |
858119e1 | 99 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 MS |
100 | __releases(dentry->d_lock) |
101 | __releases(dcache_lock) | |
1da177e4 LT |
102 | { |
103 | struct inode *inode = dentry->d_inode; | |
104 | if (inode) { | |
105 | dentry->d_inode = NULL; | |
106 | list_del_init(&dentry->d_alias); | |
107 | spin_unlock(&dentry->d_lock); | |
108 | spin_unlock(&dcache_lock); | |
f805fbda LT |
109 | if (!inode->i_nlink) |
110 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
111 | if (dentry->d_op && dentry->d_op->d_iput) |
112 | dentry->d_op->d_iput(dentry, inode); | |
113 | else | |
114 | iput(inode); | |
115 | } else { | |
116 | spin_unlock(&dentry->d_lock); | |
117 | spin_unlock(&dcache_lock); | |
118 | } | |
119 | } | |
120 | ||
da3bbdd4 KM |
121 | /* |
122 | * dentry_lru_(add|add_tail|del|del_init) must be called with dcache_lock held. | |
123 | */ | |
124 | static void dentry_lru_add(struct dentry *dentry) | |
125 | { | |
126 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
127 | dentry->d_sb->s_nr_dentry_unused++; | |
128 | dentry_stat.nr_unused++; | |
129 | } | |
130 | ||
131 | static void dentry_lru_add_tail(struct dentry *dentry) | |
132 | { | |
133 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
134 | dentry->d_sb->s_nr_dentry_unused++; | |
135 | dentry_stat.nr_unused++; | |
136 | } | |
137 | ||
138 | static void dentry_lru_del(struct dentry *dentry) | |
139 | { | |
140 | if (!list_empty(&dentry->d_lru)) { | |
141 | list_del(&dentry->d_lru); | |
142 | dentry->d_sb->s_nr_dentry_unused--; | |
143 | dentry_stat.nr_unused--; | |
144 | } | |
145 | } | |
146 | ||
147 | static void dentry_lru_del_init(struct dentry *dentry) | |
148 | { | |
149 | if (likely(!list_empty(&dentry->d_lru))) { | |
150 | list_del_init(&dentry->d_lru); | |
151 | dentry->d_sb->s_nr_dentry_unused--; | |
152 | dentry_stat.nr_unused--; | |
153 | } | |
154 | } | |
155 | ||
d52b9086 MS |
156 | /** |
157 | * d_kill - kill dentry and return parent | |
158 | * @dentry: dentry to kill | |
159 | * | |
31f3e0b3 | 160 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
161 | * |
162 | * If this is the root of the dentry tree, return NULL. | |
163 | */ | |
164 | static struct dentry *d_kill(struct dentry *dentry) | |
31f3e0b3 MS |
165 | __releases(dentry->d_lock) |
166 | __releases(dcache_lock) | |
d52b9086 MS |
167 | { |
168 | struct dentry *parent; | |
169 | ||
170 | list_del(&dentry->d_u.d_child); | |
171 | dentry_stat.nr_dentry--; /* For d_free, below */ | |
172 | /*drops the locks, at that point nobody can reach this dentry */ | |
173 | dentry_iput(dentry); | |
871c0067 OH |
174 | if (IS_ROOT(dentry)) |
175 | parent = NULL; | |
176 | else | |
177 | parent = dentry->d_parent; | |
d52b9086 | 178 | d_free(dentry); |
871c0067 | 179 | return parent; |
d52b9086 MS |
180 | } |
181 | ||
1da177e4 LT |
182 | /* |
183 | * This is dput | |
184 | * | |
185 | * This is complicated by the fact that we do not want to put | |
186 | * dentries that are no longer on any hash chain on the unused | |
187 | * list: we'd much rather just get rid of them immediately. | |
188 | * | |
189 | * However, that implies that we have to traverse the dentry | |
190 | * tree upwards to the parents which might _also_ now be | |
191 | * scheduled for deletion (it may have been only waiting for | |
192 | * its last child to go away). | |
193 | * | |
194 | * This tail recursion is done by hand as we don't want to depend | |
195 | * on the compiler to always get this right (gcc generally doesn't). | |
196 | * Real recursion would eat up our stack space. | |
197 | */ | |
198 | ||
199 | /* | |
200 | * dput - release a dentry | |
201 | * @dentry: dentry to release | |
202 | * | |
203 | * Release a dentry. This will drop the usage count and if appropriate | |
204 | * call the dentry unlink method as well as removing it from the queues and | |
205 | * releasing its resources. If the parent dentries were scheduled for release | |
206 | * they too may now get deleted. | |
207 | * | |
208 | * no dcache lock, please. | |
209 | */ | |
210 | ||
211 | void dput(struct dentry *dentry) | |
212 | { | |
213 | if (!dentry) | |
214 | return; | |
215 | ||
216 | repeat: | |
217 | if (atomic_read(&dentry->d_count) == 1) | |
218 | might_sleep(); | |
219 | if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock)) | |
220 | return; | |
221 | ||
222 | spin_lock(&dentry->d_lock); | |
223 | if (atomic_read(&dentry->d_count)) { | |
224 | spin_unlock(&dentry->d_lock); | |
225 | spin_unlock(&dcache_lock); | |
226 | return; | |
227 | } | |
228 | ||
229 | /* | |
230 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
231 | */ | |
232 | if (dentry->d_op && dentry->d_op->d_delete) { | |
233 | if (dentry->d_op->d_delete(dentry)) | |
234 | goto unhash_it; | |
235 | } | |
236 | /* Unreachable? Get rid of it */ | |
237 | if (d_unhashed(dentry)) | |
238 | goto kill_it; | |
239 | if (list_empty(&dentry->d_lru)) { | |
240 | dentry->d_flags |= DCACHE_REFERENCED; | |
da3bbdd4 | 241 | dentry_lru_add(dentry); |
1da177e4 LT |
242 | } |
243 | spin_unlock(&dentry->d_lock); | |
244 | spin_unlock(&dcache_lock); | |
245 | return; | |
246 | ||
247 | unhash_it: | |
248 | __d_drop(dentry); | |
d52b9086 | 249 | kill_it: |
da3bbdd4 KM |
250 | /* if dentry was on the d_lru list delete it from there */ |
251 | dentry_lru_del(dentry); | |
d52b9086 MS |
252 | dentry = d_kill(dentry); |
253 | if (dentry) | |
254 | goto repeat; | |
1da177e4 | 255 | } |
ec4f8605 | 256 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
257 | |
258 | /** | |
259 | * d_invalidate - invalidate a dentry | |
260 | * @dentry: dentry to invalidate | |
261 | * | |
262 | * Try to invalidate the dentry if it turns out to be | |
263 | * possible. If there are other dentries that can be | |
264 | * reached through this one we can't delete it and we | |
265 | * return -EBUSY. On success we return 0. | |
266 | * | |
267 | * no dcache lock. | |
268 | */ | |
269 | ||
270 | int d_invalidate(struct dentry * dentry) | |
271 | { | |
272 | /* | |
273 | * If it's already been dropped, return OK. | |
274 | */ | |
275 | spin_lock(&dcache_lock); | |
276 | if (d_unhashed(dentry)) { | |
277 | spin_unlock(&dcache_lock); | |
278 | return 0; | |
279 | } | |
280 | /* | |
281 | * Check whether to do a partial shrink_dcache | |
282 | * to get rid of unused child entries. | |
283 | */ | |
284 | if (!list_empty(&dentry->d_subdirs)) { | |
285 | spin_unlock(&dcache_lock); | |
286 | shrink_dcache_parent(dentry); | |
287 | spin_lock(&dcache_lock); | |
288 | } | |
289 | ||
290 | /* | |
291 | * Somebody else still using it? | |
292 | * | |
293 | * If it's a directory, we can't drop it | |
294 | * for fear of somebody re-populating it | |
295 | * with children (even though dropping it | |
296 | * would make it unreachable from the root, | |
297 | * we might still populate it if it was a | |
298 | * working directory or similar). | |
299 | */ | |
300 | spin_lock(&dentry->d_lock); | |
301 | if (atomic_read(&dentry->d_count) > 1) { | |
302 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { | |
303 | spin_unlock(&dentry->d_lock); | |
304 | spin_unlock(&dcache_lock); | |
305 | return -EBUSY; | |
306 | } | |
307 | } | |
308 | ||
309 | __d_drop(dentry); | |
310 | spin_unlock(&dentry->d_lock); | |
311 | spin_unlock(&dcache_lock); | |
312 | return 0; | |
313 | } | |
ec4f8605 | 314 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 LT |
315 | |
316 | /* This should be called _only_ with dcache_lock held */ | |
317 | ||
318 | static inline struct dentry * __dget_locked(struct dentry *dentry) | |
319 | { | |
320 | atomic_inc(&dentry->d_count); | |
da3bbdd4 | 321 | dentry_lru_del_init(dentry); |
1da177e4 LT |
322 | return dentry; |
323 | } | |
324 | ||
325 | struct dentry * dget_locked(struct dentry *dentry) | |
326 | { | |
327 | return __dget_locked(dentry); | |
328 | } | |
ec4f8605 | 329 | EXPORT_SYMBOL(dget_locked); |
1da177e4 LT |
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 | } | |
ec4f8605 | 386 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
387 | |
388 | /* | |
389 | * Try to kill dentries associated with this inode. | |
390 | * WARNING: you must own a reference to inode. | |
391 | */ | |
392 | void d_prune_aliases(struct inode *inode) | |
393 | { | |
0cdca3f9 | 394 | struct dentry *dentry; |
1da177e4 LT |
395 | restart: |
396 | spin_lock(&dcache_lock); | |
0cdca3f9 | 397 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 LT |
398 | spin_lock(&dentry->d_lock); |
399 | if (!atomic_read(&dentry->d_count)) { | |
400 | __dget_locked(dentry); | |
401 | __d_drop(dentry); | |
402 | spin_unlock(&dentry->d_lock); | |
403 | spin_unlock(&dcache_lock); | |
404 | dput(dentry); | |
405 | goto restart; | |
406 | } | |
407 | spin_unlock(&dentry->d_lock); | |
408 | } | |
409 | spin_unlock(&dcache_lock); | |
410 | } | |
ec4f8605 | 411 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
412 | |
413 | /* | |
d702ccb3 AM |
414 | * Throw away a dentry - free the inode, dput the parent. This requires that |
415 | * the LRU list has already been removed. | |
416 | * | |
85864e10 MS |
417 | * Try to prune ancestors as well. This is necessary to prevent |
418 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
419 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 420 | */ |
85864e10 | 421 | static void prune_one_dentry(struct dentry * dentry) |
31f3e0b3 MS |
422 | __releases(dentry->d_lock) |
423 | __releases(dcache_lock) | |
424 | __acquires(dcache_lock) | |
1da177e4 | 425 | { |
1da177e4 | 426 | __d_drop(dentry); |
d52b9086 | 427 | dentry = d_kill(dentry); |
d52b9086 MS |
428 | |
429 | /* | |
430 | * Prune ancestors. Locking is simpler than in dput(), | |
431 | * because dcache_lock needs to be taken anyway. | |
432 | */ | |
1da177e4 | 433 | spin_lock(&dcache_lock); |
d52b9086 MS |
434 | while (dentry) { |
435 | if (!atomic_dec_and_lock(&dentry->d_count, &dentry->d_lock)) | |
436 | return; | |
437 | ||
438 | if (dentry->d_op && dentry->d_op->d_delete) | |
439 | dentry->d_op->d_delete(dentry); | |
da3bbdd4 | 440 | dentry_lru_del_init(dentry); |
d52b9086 MS |
441 | __d_drop(dentry); |
442 | dentry = d_kill(dentry); | |
443 | spin_lock(&dcache_lock); | |
444 | } | |
1da177e4 LT |
445 | } |
446 | ||
da3bbdd4 KM |
447 | /* |
448 | * Shrink the dentry LRU on a given superblock. | |
449 | * @sb : superblock to shrink dentry LRU. | |
450 | * @count: If count is NULL, we prune all dentries on superblock. | |
451 | * @flags: If flags is non-zero, we need to do special processing based on | |
452 | * which flags are set. This means we don't need to maintain multiple | |
453 | * similar copies of this loop. | |
1da177e4 | 454 | */ |
da3bbdd4 | 455 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) |
1da177e4 | 456 | { |
da3bbdd4 KM |
457 | LIST_HEAD(referenced); |
458 | LIST_HEAD(tmp); | |
459 | struct dentry *dentry; | |
460 | int cnt = 0; | |
1da177e4 | 461 | |
da3bbdd4 KM |
462 | BUG_ON(!sb); |
463 | BUG_ON((flags & DCACHE_REFERENCED) && count == NULL); | |
464 | spin_lock(&dcache_lock); | |
465 | if (count != NULL) | |
466 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
467 | cnt = *count; | |
468 | restart: | |
469 | if (count == NULL) | |
470 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
471 | else { | |
472 | while (!list_empty(&sb->s_dentry_lru)) { | |
473 | dentry = list_entry(sb->s_dentry_lru.prev, | |
474 | struct dentry, d_lru); | |
475 | BUG_ON(dentry->d_sb != sb); | |
476 | ||
477 | spin_lock(&dentry->d_lock); | |
478 | /* | |
479 | * If we are honouring the DCACHE_REFERENCED flag and | |
480 | * the dentry has this flag set, don't free it. Clear | |
481 | * the flag and put it back on the LRU. | |
0feae5c4 | 482 | */ |
da3bbdd4 KM |
483 | if ((flags & DCACHE_REFERENCED) |
484 | && (dentry->d_flags & DCACHE_REFERENCED)) { | |
485 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
c490d79b | 486 | list_move(&dentry->d_lru, &referenced); |
da3bbdd4 KM |
487 | spin_unlock(&dentry->d_lock); |
488 | } else { | |
489 | list_move_tail(&dentry->d_lru, &tmp); | |
490 | spin_unlock(&dentry->d_lock); | |
491 | cnt--; | |
492 | if (!cnt) | |
493 | break; | |
0feae5c4 | 494 | } |
f3c6ba98 | 495 | cond_resched_lock(&dcache_lock); |
0feae5c4 | 496 | } |
da3bbdd4 KM |
497 | } |
498 | while (!list_empty(&tmp)) { | |
499 | dentry = list_entry(tmp.prev, struct dentry, d_lru); | |
500 | dentry_lru_del_init(dentry); | |
501 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
502 | /* |
503 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
504 | * the LRU because of laziness during lookup. Do not free |
505 | * it - just keep it off the LRU list. | |
1da177e4 | 506 | */ |
da3bbdd4 KM |
507 | if (atomic_read(&dentry->d_count)) { |
508 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
509 | continue; |
510 | } | |
da3bbdd4 KM |
511 | prune_one_dentry(dentry); |
512 | /* dentry->d_lock was dropped in prune_one_dentry() */ | |
513 | cond_resched_lock(&dcache_lock); | |
514 | } | |
515 | if (count == NULL && !list_empty(&sb->s_dentry_lru)) | |
516 | goto restart; | |
517 | if (count != NULL) | |
518 | *count = cnt; | |
519 | if (!list_empty(&referenced)) | |
520 | list_splice(&referenced, &sb->s_dentry_lru); | |
521 | spin_unlock(&dcache_lock); | |
522 | } | |
523 | ||
524 | /** | |
525 | * prune_dcache - shrink the dcache | |
526 | * @count: number of entries to try to free | |
527 | * | |
528 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
529 | * need to unmount something (at which point we need to unuse all dentries). | |
530 | * | |
531 | * This function may fail to free any resources if all the dentries are in use. | |
532 | */ | |
533 | static void prune_dcache(int count) | |
534 | { | |
dca33252 | 535 | struct super_block *sb, *p = NULL; |
da3bbdd4 KM |
536 | int w_count; |
537 | int unused = dentry_stat.nr_unused; | |
538 | int prune_ratio; | |
539 | int pruned; | |
540 | ||
541 | if (unused == 0 || count == 0) | |
542 | return; | |
543 | spin_lock(&dcache_lock); | |
da3bbdd4 KM |
544 | if (count >= unused) |
545 | prune_ratio = 1; | |
546 | else | |
547 | prune_ratio = unused / count; | |
548 | spin_lock(&sb_lock); | |
dca33252 | 549 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
550 | if (list_empty(&sb->s_instances)) |
551 | continue; | |
da3bbdd4 | 552 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 553 | continue; |
da3bbdd4 KM |
554 | sb->s_count++; |
555 | /* Now, we reclaim unused dentrins with fairness. | |
556 | * We reclaim them same percentage from each superblock. | |
557 | * We calculate number of dentries to scan on this sb | |
558 | * as follows, but the implementation is arranged to avoid | |
559 | * overflows: | |
560 | * number of dentries to scan on this sb = | |
561 | * count * (number of dentries on this sb / | |
562 | * number of dentries in the machine) | |
0feae5c4 | 563 | */ |
da3bbdd4 KM |
564 | spin_unlock(&sb_lock); |
565 | if (prune_ratio != 1) | |
566 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
567 | else | |
568 | w_count = sb->s_nr_dentry_unused; | |
569 | pruned = w_count; | |
0feae5c4 | 570 | /* |
da3bbdd4 KM |
571 | * We need to be sure this filesystem isn't being unmounted, |
572 | * otherwise we could race with generic_shutdown_super(), and | |
573 | * end up holding a reference to an inode while the filesystem | |
574 | * is unmounted. So we try to get s_umount, and make sure | |
575 | * s_root isn't NULL. | |
0feae5c4 | 576 | */ |
da3bbdd4 KM |
577 | if (down_read_trylock(&sb->s_umount)) { |
578 | if ((sb->s_root != NULL) && | |
579 | (!list_empty(&sb->s_dentry_lru))) { | |
580 | spin_unlock(&dcache_lock); | |
581 | __shrink_dcache_sb(sb, &w_count, | |
582 | DCACHE_REFERENCED); | |
583 | pruned -= w_count; | |
584 | spin_lock(&dcache_lock); | |
0feae5c4 | 585 | } |
da3bbdd4 | 586 | up_read(&sb->s_umount); |
0feae5c4 | 587 | } |
da3bbdd4 | 588 | spin_lock(&sb_lock); |
dca33252 AV |
589 | if (p) |
590 | __put_super(p); | |
da3bbdd4 | 591 | count -= pruned; |
dca33252 | 592 | p = sb; |
79893c17 AV |
593 | /* more work left to do? */ |
594 | if (count <= 0) | |
595 | break; | |
1da177e4 | 596 | } |
dca33252 AV |
597 | if (p) |
598 | __put_super(p); | |
da3bbdd4 | 599 | spin_unlock(&sb_lock); |
1da177e4 LT |
600 | spin_unlock(&dcache_lock); |
601 | } | |
602 | ||
1da177e4 LT |
603 | /** |
604 | * shrink_dcache_sb - shrink dcache for a superblock | |
605 | * @sb: superblock | |
606 | * | |
607 | * Shrink the dcache for the specified super block. This | |
608 | * is used to free the dcache before unmounting a file | |
609 | * system | |
610 | */ | |
1da177e4 LT |
611 | void shrink_dcache_sb(struct super_block * sb) |
612 | { | |
da3bbdd4 | 613 | __shrink_dcache_sb(sb, NULL, 0); |
1da177e4 | 614 | } |
ec4f8605 | 615 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 616 | |
c636ebdb DH |
617 | /* |
618 | * destroy a single subtree of dentries for unmount | |
619 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
620 | * locking | |
621 | */ | |
622 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
623 | { | |
624 | struct dentry *parent; | |
f8713576 | 625 | unsigned detached = 0; |
c636ebdb DH |
626 | |
627 | BUG_ON(!IS_ROOT(dentry)); | |
628 | ||
629 | /* detach this root from the system */ | |
630 | spin_lock(&dcache_lock); | |
da3bbdd4 | 631 | dentry_lru_del_init(dentry); |
c636ebdb DH |
632 | __d_drop(dentry); |
633 | spin_unlock(&dcache_lock); | |
634 | ||
635 | for (;;) { | |
636 | /* descend to the first leaf in the current subtree */ | |
637 | while (!list_empty(&dentry->d_subdirs)) { | |
638 | struct dentry *loop; | |
639 | ||
640 | /* this is a branch with children - detach all of them | |
641 | * from the system in one go */ | |
642 | spin_lock(&dcache_lock); | |
643 | list_for_each_entry(loop, &dentry->d_subdirs, | |
644 | d_u.d_child) { | |
da3bbdd4 | 645 | dentry_lru_del_init(loop); |
c636ebdb DH |
646 | __d_drop(loop); |
647 | cond_resched_lock(&dcache_lock); | |
648 | } | |
649 | spin_unlock(&dcache_lock); | |
650 | ||
651 | /* move to the first child */ | |
652 | dentry = list_entry(dentry->d_subdirs.next, | |
653 | struct dentry, d_u.d_child); | |
654 | } | |
655 | ||
656 | /* consume the dentries from this leaf up through its parents | |
657 | * until we find one with children or run out altogether */ | |
658 | do { | |
659 | struct inode *inode; | |
660 | ||
661 | if (atomic_read(&dentry->d_count) != 0) { | |
662 | printk(KERN_ERR | |
663 | "BUG: Dentry %p{i=%lx,n=%s}" | |
664 | " still in use (%d)" | |
665 | " [unmount of %s %s]\n", | |
666 | dentry, | |
667 | dentry->d_inode ? | |
668 | dentry->d_inode->i_ino : 0UL, | |
669 | dentry->d_name.name, | |
670 | atomic_read(&dentry->d_count), | |
671 | dentry->d_sb->s_type->name, | |
672 | dentry->d_sb->s_id); | |
673 | BUG(); | |
674 | } | |
675 | ||
871c0067 | 676 | if (IS_ROOT(dentry)) |
c636ebdb | 677 | parent = NULL; |
871c0067 OH |
678 | else { |
679 | parent = dentry->d_parent; | |
c636ebdb | 680 | atomic_dec(&parent->d_count); |
871c0067 | 681 | } |
c636ebdb DH |
682 | |
683 | list_del(&dentry->d_u.d_child); | |
f8713576 | 684 | detached++; |
c636ebdb DH |
685 | |
686 | inode = dentry->d_inode; | |
687 | if (inode) { | |
688 | dentry->d_inode = NULL; | |
689 | list_del_init(&dentry->d_alias); | |
690 | if (dentry->d_op && dentry->d_op->d_iput) | |
691 | dentry->d_op->d_iput(dentry, inode); | |
692 | else | |
693 | iput(inode); | |
694 | } | |
695 | ||
696 | d_free(dentry); | |
697 | ||
698 | /* finished when we fall off the top of the tree, | |
699 | * otherwise we ascend to the parent and move to the | |
700 | * next sibling if there is one */ | |
701 | if (!parent) | |
f8713576 | 702 | goto out; |
c636ebdb DH |
703 | |
704 | dentry = parent; | |
705 | ||
706 | } while (list_empty(&dentry->d_subdirs)); | |
707 | ||
708 | dentry = list_entry(dentry->d_subdirs.next, | |
709 | struct dentry, d_u.d_child); | |
710 | } | |
f8713576 DH |
711 | out: |
712 | /* several dentries were freed, need to correct nr_dentry */ | |
713 | spin_lock(&dcache_lock); | |
714 | dentry_stat.nr_dentry -= detached; | |
715 | spin_unlock(&dcache_lock); | |
c636ebdb DH |
716 | } |
717 | ||
718 | /* | |
719 | * destroy the dentries attached to a superblock on unmounting | |
720 | * - we don't need to use dentry->d_lock, and only need dcache_lock when | |
721 | * removing the dentry from the system lists and hashes because: | |
722 | * - the superblock is detached from all mountings and open files, so the | |
723 | * dentry trees will not be rearranged by the VFS | |
724 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
725 | * any dentries belonging to this superblock that it comes across | |
726 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
727 | * in this superblock | |
728 | */ | |
729 | void shrink_dcache_for_umount(struct super_block *sb) | |
730 | { | |
731 | struct dentry *dentry; | |
732 | ||
733 | if (down_read_trylock(&sb->s_umount)) | |
734 | BUG(); | |
735 | ||
736 | dentry = sb->s_root; | |
737 | sb->s_root = NULL; | |
738 | atomic_dec(&dentry->d_count); | |
739 | shrink_dcache_for_umount_subtree(dentry); | |
740 | ||
741 | while (!hlist_empty(&sb->s_anon)) { | |
742 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
743 | shrink_dcache_for_umount_subtree(dentry); | |
744 | } | |
745 | } | |
746 | ||
1da177e4 LT |
747 | /* |
748 | * Search for at least 1 mount point in the dentry's subdirs. | |
749 | * We descend to the next level whenever the d_subdirs | |
750 | * list is non-empty and continue searching. | |
751 | */ | |
752 | ||
753 | /** | |
754 | * have_submounts - check for mounts over a dentry | |
755 | * @parent: dentry to check. | |
756 | * | |
757 | * Return true if the parent or its subdirectories contain | |
758 | * a mount point | |
759 | */ | |
760 | ||
761 | int have_submounts(struct dentry *parent) | |
762 | { | |
763 | struct dentry *this_parent = parent; | |
764 | struct list_head *next; | |
765 | ||
766 | spin_lock(&dcache_lock); | |
767 | if (d_mountpoint(parent)) | |
768 | goto positive; | |
769 | repeat: | |
770 | next = this_parent->d_subdirs.next; | |
771 | resume: | |
772 | while (next != &this_parent->d_subdirs) { | |
773 | struct list_head *tmp = next; | |
5160ee6f | 774 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
775 | next = tmp->next; |
776 | /* Have we found a mount point ? */ | |
777 | if (d_mountpoint(dentry)) | |
778 | goto positive; | |
779 | if (!list_empty(&dentry->d_subdirs)) { | |
780 | this_parent = dentry; | |
781 | goto repeat; | |
782 | } | |
783 | } | |
784 | /* | |
785 | * All done at this level ... ascend and resume the search. | |
786 | */ | |
787 | if (this_parent != parent) { | |
5160ee6f | 788 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
789 | this_parent = this_parent->d_parent; |
790 | goto resume; | |
791 | } | |
792 | spin_unlock(&dcache_lock); | |
793 | return 0; /* No mount points found in tree */ | |
794 | positive: | |
795 | spin_unlock(&dcache_lock); | |
796 | return 1; | |
797 | } | |
ec4f8605 | 798 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
799 | |
800 | /* | |
801 | * Search the dentry child list for the specified parent, | |
802 | * and move any unused dentries to the end of the unused | |
803 | * list for prune_dcache(). We descend to the next level | |
804 | * whenever the d_subdirs list is non-empty and continue | |
805 | * searching. | |
806 | * | |
807 | * It returns zero iff there are no unused children, | |
808 | * otherwise it returns the number of children moved to | |
809 | * the end of the unused list. This may not be the total | |
810 | * number of unused children, because select_parent can | |
811 | * drop the lock and return early due to latency | |
812 | * constraints. | |
813 | */ | |
814 | static int select_parent(struct dentry * parent) | |
815 | { | |
816 | struct dentry *this_parent = parent; | |
817 | struct list_head *next; | |
818 | int found = 0; | |
819 | ||
820 | spin_lock(&dcache_lock); | |
821 | repeat: | |
822 | next = this_parent->d_subdirs.next; | |
823 | resume: | |
824 | while (next != &this_parent->d_subdirs) { | |
825 | struct list_head *tmp = next; | |
5160ee6f | 826 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
827 | next = tmp->next; |
828 | ||
da3bbdd4 | 829 | dentry_lru_del_init(dentry); |
1da177e4 LT |
830 | /* |
831 | * move only zero ref count dentries to the end | |
832 | * of the unused list for prune_dcache | |
833 | */ | |
834 | if (!atomic_read(&dentry->d_count)) { | |
da3bbdd4 | 835 | dentry_lru_add_tail(dentry); |
1da177e4 LT |
836 | found++; |
837 | } | |
838 | ||
839 | /* | |
840 | * We can return to the caller if we have found some (this | |
841 | * ensures forward progress). We'll be coming back to find | |
842 | * the rest. | |
843 | */ | |
844 | if (found && need_resched()) | |
845 | goto out; | |
846 | ||
847 | /* | |
848 | * Descend a level if the d_subdirs list is non-empty. | |
849 | */ | |
850 | if (!list_empty(&dentry->d_subdirs)) { | |
851 | this_parent = dentry; | |
1da177e4 LT |
852 | goto repeat; |
853 | } | |
854 | } | |
855 | /* | |
856 | * All done at this level ... ascend and resume the search. | |
857 | */ | |
858 | if (this_parent != parent) { | |
5160ee6f | 859 | next = this_parent->d_u.d_child.next; |
1da177e4 | 860 | this_parent = this_parent->d_parent; |
1da177e4 LT |
861 | goto resume; |
862 | } | |
863 | out: | |
864 | spin_unlock(&dcache_lock); | |
865 | return found; | |
866 | } | |
867 | ||
868 | /** | |
869 | * shrink_dcache_parent - prune dcache | |
870 | * @parent: parent of entries to prune | |
871 | * | |
872 | * Prune the dcache to remove unused children of the parent dentry. | |
873 | */ | |
874 | ||
875 | void shrink_dcache_parent(struct dentry * parent) | |
876 | { | |
da3bbdd4 | 877 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
878 | int found; |
879 | ||
880 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 881 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 882 | } |
ec4f8605 | 883 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 884 | |
1da177e4 LT |
885 | /* |
886 | * Scan `nr' dentries and return the number which remain. | |
887 | * | |
888 | * We need to avoid reentering the filesystem if the caller is performing a | |
889 | * GFP_NOFS allocation attempt. One example deadlock is: | |
890 | * | |
891 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
892 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
893 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
894 | * | |
895 | * In this case we return -1 to tell the caller that we baled. | |
896 | */ | |
7f8275d0 | 897 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
898 | { |
899 | if (nr) { | |
900 | if (!(gfp_mask & __GFP_FS)) | |
901 | return -1; | |
da3bbdd4 | 902 | prune_dcache(nr); |
1da177e4 LT |
903 | } |
904 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
905 | } | |
906 | ||
8e1f936b RR |
907 | static struct shrinker dcache_shrinker = { |
908 | .shrink = shrink_dcache_memory, | |
909 | .seeks = DEFAULT_SEEKS, | |
910 | }; | |
911 | ||
1da177e4 LT |
912 | /** |
913 | * d_alloc - allocate a dcache entry | |
914 | * @parent: parent of entry to allocate | |
915 | * @name: qstr of the name | |
916 | * | |
917 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
918 | * available. On a success the dentry is returned. The name passed in is | |
919 | * copied and the copy passed in may be reused after this call. | |
920 | */ | |
921 | ||
922 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
923 | { | |
924 | struct dentry *dentry; | |
925 | char *dname; | |
926 | ||
e12ba74d | 927 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
928 | if (!dentry) |
929 | return NULL; | |
930 | ||
931 | if (name->len > DNAME_INLINE_LEN-1) { | |
932 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
933 | if (!dname) { | |
934 | kmem_cache_free(dentry_cache, dentry); | |
935 | return NULL; | |
936 | } | |
937 | } else { | |
938 | dname = dentry->d_iname; | |
939 | } | |
940 | dentry->d_name.name = dname; | |
941 | ||
942 | dentry->d_name.len = name->len; | |
943 | dentry->d_name.hash = name->hash; | |
944 | memcpy(dname, name->name, name->len); | |
945 | dname[name->len] = 0; | |
946 | ||
947 | atomic_set(&dentry->d_count, 1); | |
948 | dentry->d_flags = DCACHE_UNHASHED; | |
949 | spin_lock_init(&dentry->d_lock); | |
950 | dentry->d_inode = NULL; | |
951 | dentry->d_parent = NULL; | |
952 | dentry->d_sb = NULL; | |
953 | dentry->d_op = NULL; | |
954 | dentry->d_fsdata = NULL; | |
955 | dentry->d_mounted = 0; | |
1da177e4 LT |
956 | INIT_HLIST_NODE(&dentry->d_hash); |
957 | INIT_LIST_HEAD(&dentry->d_lru); | |
958 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
959 | INIT_LIST_HEAD(&dentry->d_alias); | |
960 | ||
961 | if (parent) { | |
962 | dentry->d_parent = dget(parent); | |
963 | dentry->d_sb = parent->d_sb; | |
964 | } else { | |
5160ee6f | 965 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
966 | } |
967 | ||
968 | spin_lock(&dcache_lock); | |
969 | if (parent) | |
5160ee6f | 970 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1da177e4 LT |
971 | dentry_stat.nr_dentry++; |
972 | spin_unlock(&dcache_lock); | |
973 | ||
974 | return dentry; | |
975 | } | |
ec4f8605 | 976 | EXPORT_SYMBOL(d_alloc); |
1da177e4 LT |
977 | |
978 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
979 | { | |
980 | struct qstr q; | |
981 | ||
982 | q.name = name; | |
983 | q.len = strlen(name); | |
984 | q.hash = full_name_hash(q.name, q.len); | |
985 | return d_alloc(parent, &q); | |
986 | } | |
ef26ca97 | 987 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 988 | |
360da900 OH |
989 | /* the caller must hold dcache_lock */ |
990 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | |
991 | { | |
992 | if (inode) | |
993 | list_add(&dentry->d_alias, &inode->i_dentry); | |
994 | dentry->d_inode = inode; | |
995 | fsnotify_d_instantiate(dentry, inode); | |
996 | } | |
997 | ||
1da177e4 LT |
998 | /** |
999 | * d_instantiate - fill in inode information for a dentry | |
1000 | * @entry: dentry to complete | |
1001 | * @inode: inode to attach to this dentry | |
1002 | * | |
1003 | * Fill in inode information in the entry. | |
1004 | * | |
1005 | * This turns negative dentries into productive full members | |
1006 | * of society. | |
1007 | * | |
1008 | * NOTE! This assumes that the inode count has been incremented | |
1009 | * (or otherwise set) by the caller to indicate that it is now | |
1010 | * in use by the dcache. | |
1011 | */ | |
1012 | ||
1013 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1014 | { | |
28133c7b | 1015 | BUG_ON(!list_empty(&entry->d_alias)); |
1da177e4 | 1016 | spin_lock(&dcache_lock); |
360da900 | 1017 | __d_instantiate(entry, inode); |
1da177e4 LT |
1018 | spin_unlock(&dcache_lock); |
1019 | security_d_instantiate(entry, inode); | |
1020 | } | |
ec4f8605 | 1021 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1022 | |
1023 | /** | |
1024 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1025 | * @entry: dentry to instantiate | |
1026 | * @inode: inode to attach to this dentry | |
1027 | * | |
1028 | * Fill in inode information in the entry. On success, it returns NULL. | |
1029 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1030 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1031 | * |
1032 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1033 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1034 | * |
1035 | * This also assumes that the inode count has been incremented | |
1036 | * (or otherwise set) by the caller to indicate that it is now | |
1037 | * in use by the dcache. | |
1da177e4 | 1038 | */ |
770bfad8 DH |
1039 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1040 | struct inode *inode) | |
1da177e4 LT |
1041 | { |
1042 | struct dentry *alias; | |
1043 | int len = entry->d_name.len; | |
1044 | const char *name = entry->d_name.name; | |
1045 | unsigned int hash = entry->d_name.hash; | |
1046 | ||
770bfad8 | 1047 | if (!inode) { |
360da900 | 1048 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1049 | return NULL; |
1050 | } | |
1051 | ||
1da177e4 LT |
1052 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1053 | struct qstr *qstr = &alias->d_name; | |
1054 | ||
1055 | if (qstr->hash != hash) | |
1056 | continue; | |
1057 | if (alias->d_parent != entry->d_parent) | |
1058 | continue; | |
1059 | if (qstr->len != len) | |
1060 | continue; | |
1061 | if (memcmp(qstr->name, name, len)) | |
1062 | continue; | |
1063 | dget_locked(alias); | |
1da177e4 LT |
1064 | return alias; |
1065 | } | |
770bfad8 | 1066 | |
360da900 | 1067 | __d_instantiate(entry, inode); |
1da177e4 LT |
1068 | return NULL; |
1069 | } | |
770bfad8 DH |
1070 | |
1071 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1072 | { | |
1073 | struct dentry *result; | |
1074 | ||
1075 | BUG_ON(!list_empty(&entry->d_alias)); | |
1076 | ||
1077 | spin_lock(&dcache_lock); | |
1078 | result = __d_instantiate_unique(entry, inode); | |
1079 | spin_unlock(&dcache_lock); | |
1080 | ||
1081 | if (!result) { | |
1082 | security_d_instantiate(entry, inode); | |
1083 | return NULL; | |
1084 | } | |
1085 | ||
1086 | BUG_ON(!d_unhashed(result)); | |
1087 | iput(inode); | |
1088 | return result; | |
1089 | } | |
1090 | ||
1da177e4 LT |
1091 | EXPORT_SYMBOL(d_instantiate_unique); |
1092 | ||
1093 | /** | |
1094 | * d_alloc_root - allocate root dentry | |
1095 | * @root_inode: inode to allocate the root for | |
1096 | * | |
1097 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1098 | * instantiated and returned. %NULL is returned if there is insufficient | |
1099 | * memory or the inode passed is %NULL. | |
1100 | */ | |
1101 | ||
1102 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1103 | { | |
1104 | struct dentry *res = NULL; | |
1105 | ||
1106 | if (root_inode) { | |
1107 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1108 | ||
1109 | res = d_alloc(NULL, &name); | |
1110 | if (res) { | |
1111 | res->d_sb = root_inode->i_sb; | |
1112 | res->d_parent = res; | |
1113 | d_instantiate(res, root_inode); | |
1114 | } | |
1115 | } | |
1116 | return res; | |
1117 | } | |
ec4f8605 | 1118 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 LT |
1119 | |
1120 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1121 | unsigned long hash) | |
1122 | { | |
1123 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1124 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1125 | return dentry_hashtable + (hash & D_HASHMASK); | |
1126 | } | |
1127 | ||
4ea3ada2 CH |
1128 | /** |
1129 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1130 | * @inode: inode to allocate the dentry for | |
1131 | * | |
1132 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1133 | * similar open by handle operations. The returned dentry may be anonymous, | |
1134 | * or may have a full name (if the inode was already in the cache). | |
1135 | * | |
1136 | * When called on a directory inode, we must ensure that the inode only ever | |
1137 | * has one dentry. If a dentry is found, that is returned instead of | |
1138 | * allocating a new one. | |
1139 | * | |
1140 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1141 | * to the dentry. In case of an error the reference on the inode is released. |
1142 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1143 | * be passed in and will be the error will be propagate to the return value, | |
1144 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1145 | */ |
1146 | struct dentry *d_obtain_alias(struct inode *inode) | |
1147 | { | |
9308a612 CH |
1148 | static const struct qstr anonstring = { .name = "" }; |
1149 | struct dentry *tmp; | |
1150 | struct dentry *res; | |
4ea3ada2 CH |
1151 | |
1152 | if (!inode) | |
44003728 | 1153 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1154 | if (IS_ERR(inode)) |
1155 | return ERR_CAST(inode); | |
1156 | ||
9308a612 CH |
1157 | res = d_find_alias(inode); |
1158 | if (res) | |
1159 | goto out_iput; | |
1160 | ||
1161 | tmp = d_alloc(NULL, &anonstring); | |
1162 | if (!tmp) { | |
1163 | res = ERR_PTR(-ENOMEM); | |
1164 | goto out_iput; | |
4ea3ada2 | 1165 | } |
9308a612 CH |
1166 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1167 | ||
1168 | spin_lock(&dcache_lock); | |
1169 | res = __d_find_alias(inode, 0); | |
1170 | if (res) { | |
1171 | spin_unlock(&dcache_lock); | |
1172 | dput(tmp); | |
1173 | goto out_iput; | |
1174 | } | |
1175 | ||
1176 | /* attach a disconnected dentry */ | |
1177 | spin_lock(&tmp->d_lock); | |
1178 | tmp->d_sb = inode->i_sb; | |
1179 | tmp->d_inode = inode; | |
1180 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1181 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1182 | list_add(&tmp->d_alias, &inode->i_dentry); | |
1183 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); | |
1184 | spin_unlock(&tmp->d_lock); | |
1185 | ||
1186 | spin_unlock(&dcache_lock); | |
1187 | return tmp; | |
1188 | ||
1189 | out_iput: | |
1190 | iput(inode); | |
1191 | return res; | |
4ea3ada2 | 1192 | } |
adc48720 | 1193 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1194 | |
1195 | /** | |
1196 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1197 | * @inode: the inode which may have a disconnected dentry | |
1198 | * @dentry: a negative dentry which we want to point to the inode. | |
1199 | * | |
1200 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1201 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1202 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1203 | * | |
1204 | * This is needed in the lookup routine of any filesystem that is exportable | |
1205 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1206 | * | |
1207 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1208 | * is returned. This matches the expected return value of ->lookup. | |
1209 | * | |
1210 | */ | |
1211 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1212 | { | |
1213 | struct dentry *new = NULL; | |
1214 | ||
21c0d8fd | 1215 | if (inode && S_ISDIR(inode->i_mode)) { |
1da177e4 LT |
1216 | spin_lock(&dcache_lock); |
1217 | new = __d_find_alias(inode, 1); | |
1218 | if (new) { | |
1219 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
1220 | spin_unlock(&dcache_lock); | |
1221 | security_d_instantiate(new, inode); | |
1da177e4 LT |
1222 | d_move(new, dentry); |
1223 | iput(inode); | |
1224 | } else { | |
360da900 OH |
1225 | /* already taking dcache_lock, so d_add() by hand */ |
1226 | __d_instantiate(dentry, inode); | |
1da177e4 LT |
1227 | spin_unlock(&dcache_lock); |
1228 | security_d_instantiate(dentry, inode); | |
1229 | d_rehash(dentry); | |
1230 | } | |
1231 | } else | |
1232 | d_add(dentry, inode); | |
1233 | return new; | |
1234 | } | |
ec4f8605 | 1235 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1236 | |
9403540c BN |
1237 | /** |
1238 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1239 | * @inode: the inode case-insensitive lookup has found | |
1240 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1241 | * @name: the case-exact name to be associated with the returned dentry | |
1242 | * | |
1243 | * This is to avoid filling the dcache with case-insensitive names to the | |
1244 | * same inode, only the actual correct case is stored in the dcache for | |
1245 | * case-insensitive filesystems. | |
1246 | * | |
1247 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1248 | * already exists in in the dcache, use it and return it. | |
1249 | * | |
1250 | * If no entry exists with the exact case name, allocate new dentry with | |
1251 | * the exact case, and return the spliced entry. | |
1252 | */ | |
e45b590b | 1253 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1254 | struct qstr *name) |
1255 | { | |
1256 | int error; | |
1257 | struct dentry *found; | |
1258 | struct dentry *new; | |
1259 | ||
b6520c81 CH |
1260 | /* |
1261 | * First check if a dentry matching the name already exists, | |
1262 | * if not go ahead and create it now. | |
1263 | */ | |
9403540c | 1264 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1265 | if (!found) { |
1266 | new = d_alloc(dentry->d_parent, name); | |
1267 | if (!new) { | |
1268 | error = -ENOMEM; | |
1269 | goto err_out; | |
1270 | } | |
b6520c81 | 1271 | |
9403540c BN |
1272 | found = d_splice_alias(inode, new); |
1273 | if (found) { | |
1274 | dput(new); | |
1275 | return found; | |
1276 | } | |
1277 | return new; | |
1278 | } | |
b6520c81 CH |
1279 | |
1280 | /* | |
1281 | * If a matching dentry exists, and it's not negative use it. | |
1282 | * | |
1283 | * Decrement the reference count to balance the iget() done | |
1284 | * earlier on. | |
1285 | */ | |
9403540c BN |
1286 | if (found->d_inode) { |
1287 | if (unlikely(found->d_inode != inode)) { | |
1288 | /* This can't happen because bad inodes are unhashed. */ | |
1289 | BUG_ON(!is_bad_inode(inode)); | |
1290 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1291 | } | |
9403540c BN |
1292 | iput(inode); |
1293 | return found; | |
1294 | } | |
b6520c81 | 1295 | |
9403540c BN |
1296 | /* |
1297 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1298 | * already has a dentry. |
9403540c | 1299 | */ |
9403540c | 1300 | spin_lock(&dcache_lock); |
b6520c81 | 1301 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1302 | __d_instantiate(found, inode); |
9403540c BN |
1303 | spin_unlock(&dcache_lock); |
1304 | security_d_instantiate(found, inode); | |
1305 | return found; | |
1306 | } | |
b6520c81 | 1307 | |
9403540c | 1308 | /* |
b6520c81 CH |
1309 | * In case a directory already has a (disconnected) entry grab a |
1310 | * reference to it, move it in place and use it. | |
9403540c BN |
1311 | */ |
1312 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
1313 | dget_locked(new); | |
1314 | spin_unlock(&dcache_lock); | |
9403540c | 1315 | security_d_instantiate(found, inode); |
9403540c | 1316 | d_move(new, found); |
9403540c | 1317 | iput(inode); |
9403540c | 1318 | dput(found); |
9403540c BN |
1319 | return new; |
1320 | ||
1321 | err_out: | |
1322 | iput(inode); | |
1323 | return ERR_PTR(error); | |
1324 | } | |
ec4f8605 | 1325 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 LT |
1326 | |
1327 | /** | |
1328 | * d_lookup - search for a dentry | |
1329 | * @parent: parent dentry | |
1330 | * @name: qstr of name we wish to find | |
b04f784e | 1331 | * Returns: dentry, or NULL |
1da177e4 | 1332 | * |
b04f784e NP |
1333 | * d_lookup searches the children of the parent dentry for the name in |
1334 | * question. If the dentry is found its reference count is incremented and the | |
1335 | * dentry is returned. The caller must use dput to free the entry when it has | |
1336 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1337 | */ |
1da177e4 LT |
1338 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) |
1339 | { | |
1340 | struct dentry * dentry = NULL; | |
1341 | unsigned long seq; | |
1342 | ||
1343 | do { | |
1344 | seq = read_seqbegin(&rename_lock); | |
1345 | dentry = __d_lookup(parent, name); | |
1346 | if (dentry) | |
1347 | break; | |
1348 | } while (read_seqretry(&rename_lock, seq)); | |
1349 | return dentry; | |
1350 | } | |
ec4f8605 | 1351 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1352 | |
b04f784e NP |
1353 | /* |
1354 | * __d_lookup - search for a dentry (racy) | |
1355 | * @parent: parent dentry | |
1356 | * @name: qstr of name we wish to find | |
1357 | * Returns: dentry, or NULL | |
1358 | * | |
1359 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1360 | * false-negative result due to unrelated rename activity. | |
1361 | * | |
1362 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1363 | * however it must be used carefully, eg. with a following d_lookup in | |
1364 | * the case of failure. | |
1365 | * | |
1366 | * __d_lookup callers must be commented. | |
1367 | */ | |
1da177e4 LT |
1368 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) |
1369 | { | |
1370 | unsigned int len = name->len; | |
1371 | unsigned int hash = name->hash; | |
1372 | const unsigned char *str = name->name; | |
1373 | struct hlist_head *head = d_hash(parent,hash); | |
1374 | struct dentry *found = NULL; | |
1375 | struct hlist_node *node; | |
665a7583 | 1376 | struct dentry *dentry; |
1da177e4 | 1377 | |
b04f784e NP |
1378 | /* |
1379 | * The hash list is protected using RCU. | |
1380 | * | |
1381 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1382 | * with d_move(). | |
1383 | * | |
1384 | * It is possible that concurrent renames can mess up our list | |
1385 | * walk here and result in missing our dentry, resulting in the | |
1386 | * false-negative result. d_lookup() protects against concurrent | |
1387 | * renames using rename_lock seqlock. | |
1388 | * | |
1389 | * See Documentation/vfs/dcache-locking.txt for more details. | |
1390 | */ | |
1da177e4 LT |
1391 | rcu_read_lock(); |
1392 | ||
665a7583 | 1393 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1394 | struct qstr *qstr; |
1395 | ||
1da177e4 LT |
1396 | if (dentry->d_name.hash != hash) |
1397 | continue; | |
1398 | if (dentry->d_parent != parent) | |
1399 | continue; | |
1400 | ||
1401 | spin_lock(&dentry->d_lock); | |
1402 | ||
1403 | /* | |
1404 | * Recheck the dentry after taking the lock - d_move may have | |
b04f784e NP |
1405 | * changed things. Don't bother checking the hash because |
1406 | * we're about to compare the whole name anyway. | |
1da177e4 LT |
1407 | */ |
1408 | if (dentry->d_parent != parent) | |
1409 | goto next; | |
1410 | ||
d0185c08 LT |
1411 | /* non-existing due to RCU? */ |
1412 | if (d_unhashed(dentry)) | |
1413 | goto next; | |
1414 | ||
1da177e4 LT |
1415 | /* |
1416 | * It is safe to compare names since d_move() cannot | |
1417 | * change the qstr (protected by d_lock). | |
1418 | */ | |
1419 | qstr = &dentry->d_name; | |
1420 | if (parent->d_op && parent->d_op->d_compare) { | |
1421 | if (parent->d_op->d_compare(parent, qstr, name)) | |
1422 | goto next; | |
1423 | } else { | |
1424 | if (qstr->len != len) | |
1425 | goto next; | |
1426 | if (memcmp(qstr->name, str, len)) | |
1427 | goto next; | |
1428 | } | |
1429 | ||
d0185c08 LT |
1430 | atomic_inc(&dentry->d_count); |
1431 | found = dentry; | |
1da177e4 LT |
1432 | spin_unlock(&dentry->d_lock); |
1433 | break; | |
1434 | next: | |
1435 | spin_unlock(&dentry->d_lock); | |
1436 | } | |
1437 | rcu_read_unlock(); | |
1438 | ||
1439 | return found; | |
1440 | } | |
1441 | ||
3e7e241f EB |
1442 | /** |
1443 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1444 | * @dir: Directory to search in | |
1445 | * @name: qstr of name we wish to find | |
1446 | * | |
1447 | * On hash failure or on lookup failure NULL is returned. | |
1448 | */ | |
1449 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1450 | { | |
1451 | struct dentry *dentry = NULL; | |
1452 | ||
1453 | /* | |
1454 | * Check for a fs-specific hash function. Note that we must | |
1455 | * calculate the standard hash first, as the d_op->d_hash() | |
1456 | * routine may choose to leave the hash value unchanged. | |
1457 | */ | |
1458 | name->hash = full_name_hash(name->name, name->len); | |
1459 | if (dir->d_op && dir->d_op->d_hash) { | |
1460 | if (dir->d_op->d_hash(dir, name) < 0) | |
1461 | goto out; | |
1462 | } | |
1463 | dentry = d_lookup(dir, name); | |
1464 | out: | |
1465 | return dentry; | |
1466 | } | |
1467 | ||
1da177e4 LT |
1468 | /** |
1469 | * d_validate - verify dentry provided from insecure source | |
1470 | * @dentry: The dentry alleged to be valid child of @dparent | |
1471 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1472 | * |
1473 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1474 | * This is used by ncpfs in its readdir implementation. | |
1475 | * Zero is returned in the dentry is invalid. | |
1476 | */ | |
1477 | ||
1478 | int d_validate(struct dentry *dentry, struct dentry *dparent) | |
1479 | { | |
1480 | struct hlist_head *base; | |
1481 | struct hlist_node *lhp; | |
1482 | ||
1483 | /* Check whether the ptr might be valid at all.. */ | |
1484 | if (!kmem_ptr_validate(dentry_cache, dentry)) | |
1485 | goto out; | |
1486 | ||
1487 | if (dentry->d_parent != dparent) | |
1488 | goto out; | |
1489 | ||
1490 | spin_lock(&dcache_lock); | |
1491 | base = d_hash(dparent, dentry->d_name.hash); | |
1492 | hlist_for_each(lhp,base) { | |
665a7583 | 1493 | /* hlist_for_each_entry_rcu() not required for d_hash list |
1da177e4 LT |
1494 | * as it is parsed under dcache_lock |
1495 | */ | |
1496 | if (dentry == hlist_entry(lhp, struct dentry, d_hash)) { | |
1497 | __dget_locked(dentry); | |
1498 | spin_unlock(&dcache_lock); | |
1499 | return 1; | |
1500 | } | |
1501 | } | |
1502 | spin_unlock(&dcache_lock); | |
1503 | out: | |
1504 | return 0; | |
1505 | } | |
ec4f8605 | 1506 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1507 | |
1508 | /* | |
1509 | * When a file is deleted, we have two options: | |
1510 | * - turn this dentry into a negative dentry | |
1511 | * - unhash this dentry and free it. | |
1512 | * | |
1513 | * Usually, we want to just turn this into | |
1514 | * a negative dentry, but if anybody else is | |
1515 | * currently using the dentry or the inode | |
1516 | * we can't do that and we fall back on removing | |
1517 | * it from the hash queues and waiting for | |
1518 | * it to be deleted later when it has no users | |
1519 | */ | |
1520 | ||
1521 | /** | |
1522 | * d_delete - delete a dentry | |
1523 | * @dentry: The dentry to delete | |
1524 | * | |
1525 | * Turn the dentry into a negative dentry if possible, otherwise | |
1526 | * remove it from the hash queues so it can be deleted later | |
1527 | */ | |
1528 | ||
1529 | void d_delete(struct dentry * dentry) | |
1530 | { | |
7a91bf7f | 1531 | int isdir = 0; |
1da177e4 LT |
1532 | /* |
1533 | * Are we the only user? | |
1534 | */ | |
1535 | spin_lock(&dcache_lock); | |
1536 | spin_lock(&dentry->d_lock); | |
7a91bf7f | 1537 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
1da177e4 | 1538 | if (atomic_read(&dentry->d_count) == 1) { |
13e3c5e5 | 1539 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
1da177e4 | 1540 | dentry_iput(dentry); |
7a91bf7f | 1541 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1542 | return; |
1543 | } | |
1544 | ||
1545 | if (!d_unhashed(dentry)) | |
1546 | __d_drop(dentry); | |
1547 | ||
1548 | spin_unlock(&dentry->d_lock); | |
1549 | spin_unlock(&dcache_lock); | |
7a91bf7f JM |
1550 | |
1551 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1552 | } |
ec4f8605 | 1553 | EXPORT_SYMBOL(d_delete); |
1da177e4 LT |
1554 | |
1555 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1556 | { | |
1557 | ||
1558 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1559 | hlist_add_head_rcu(&entry->d_hash, list); | |
1560 | } | |
1561 | ||
770bfad8 DH |
1562 | static void _d_rehash(struct dentry * entry) |
1563 | { | |
1564 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1565 | } | |
1566 | ||
1da177e4 LT |
1567 | /** |
1568 | * d_rehash - add an entry back to the hash | |
1569 | * @entry: dentry to add to the hash | |
1570 | * | |
1571 | * Adds a dentry to the hash according to its name. | |
1572 | */ | |
1573 | ||
1574 | void d_rehash(struct dentry * entry) | |
1575 | { | |
1da177e4 LT |
1576 | spin_lock(&dcache_lock); |
1577 | spin_lock(&entry->d_lock); | |
770bfad8 | 1578 | _d_rehash(entry); |
1da177e4 LT |
1579 | spin_unlock(&entry->d_lock); |
1580 | spin_unlock(&dcache_lock); | |
1581 | } | |
ec4f8605 | 1582 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 1583 | |
1da177e4 LT |
1584 | /* |
1585 | * When switching names, the actual string doesn't strictly have to | |
1586 | * be preserved in the target - because we're dropping the target | |
1587 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1588 | * the new name before we switch. | |
1589 | * | |
1590 | * Note that we have to be a lot more careful about getting the hash | |
1591 | * switched - we have to switch the hash value properly even if it | |
1592 | * then no longer matches the actual (corrupted) string of the target. | |
1593 | * The hash value has to match the hash queue that the dentry is on.. | |
1594 | */ | |
1595 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
1596 | { | |
1597 | if (dname_external(target)) { | |
1598 | if (dname_external(dentry)) { | |
1599 | /* | |
1600 | * Both external: swap the pointers | |
1601 | */ | |
9a8d5bb4 | 1602 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
1603 | } else { |
1604 | /* | |
1605 | * dentry:internal, target:external. Steal target's | |
1606 | * storage and make target internal. | |
1607 | */ | |
321bcf92 BF |
1608 | memcpy(target->d_iname, dentry->d_name.name, |
1609 | dentry->d_name.len + 1); | |
1da177e4 LT |
1610 | dentry->d_name.name = target->d_name.name; |
1611 | target->d_name.name = target->d_iname; | |
1612 | } | |
1613 | } else { | |
1614 | if (dname_external(dentry)) { | |
1615 | /* | |
1616 | * dentry:external, target:internal. Give dentry's | |
1617 | * storage to target and make dentry internal | |
1618 | */ | |
1619 | memcpy(dentry->d_iname, target->d_name.name, | |
1620 | target->d_name.len + 1); | |
1621 | target->d_name.name = dentry->d_name.name; | |
1622 | dentry->d_name.name = dentry->d_iname; | |
1623 | } else { | |
1624 | /* | |
1625 | * Both are internal. Just copy target to dentry | |
1626 | */ | |
1627 | memcpy(dentry->d_iname, target->d_name.name, | |
1628 | target->d_name.len + 1); | |
dc711ca3 AV |
1629 | dentry->d_name.len = target->d_name.len; |
1630 | return; | |
1da177e4 LT |
1631 | } |
1632 | } | |
9a8d5bb4 | 1633 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
1634 | } |
1635 | ||
1636 | /* | |
1637 | * We cannibalize "target" when moving dentry on top of it, | |
1638 | * because it's going to be thrown away anyway. We could be more | |
1639 | * polite about it, though. | |
1640 | * | |
1641 | * This forceful removal will result in ugly /proc output if | |
1642 | * somebody holds a file open that got deleted due to a rename. | |
1643 | * We could be nicer about the deleted file, and let it show | |
bc154b1e BF |
1644 | * up under the name it had before it was deleted rather than |
1645 | * under the original name of the file that was moved on top of it. | |
1da177e4 LT |
1646 | */ |
1647 | ||
9eaef27b TM |
1648 | /* |
1649 | * d_move_locked - move a dentry | |
1da177e4 LT |
1650 | * @dentry: entry to move |
1651 | * @target: new dentry | |
1652 | * | |
1653 | * Update the dcache to reflect the move of a file name. Negative | |
1654 | * dcache entries should not be moved in this way. | |
1655 | */ | |
9eaef27b | 1656 | static void d_move_locked(struct dentry * dentry, struct dentry * target) |
1da177e4 LT |
1657 | { |
1658 | struct hlist_head *list; | |
1659 | ||
1660 | if (!dentry->d_inode) | |
1661 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1662 | ||
1da177e4 LT |
1663 | write_seqlock(&rename_lock); |
1664 | /* | |
1665 | * XXXX: do we really need to take target->d_lock? | |
1666 | */ | |
1667 | if (target < dentry) { | |
1668 | spin_lock(&target->d_lock); | |
a90b9c05 | 1669 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1670 | } else { |
1671 | spin_lock(&dentry->d_lock); | |
a90b9c05 | 1672 | spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1673 | } |
1674 | ||
1675 | /* Move the dentry to the target hash queue, if on different bucket */ | |
f77e3498 | 1676 | if (d_unhashed(dentry)) |
1da177e4 LT |
1677 | goto already_unhashed; |
1678 | ||
1679 | hlist_del_rcu(&dentry->d_hash); | |
1680 | ||
1681 | already_unhashed: | |
1682 | list = d_hash(target->d_parent, target->d_name.hash); | |
1683 | __d_rehash(dentry, list); | |
1684 | ||
1685 | /* Unhash the target: dput() will then get rid of it */ | |
1686 | __d_drop(target); | |
1687 | ||
5160ee6f ED |
1688 | list_del(&dentry->d_u.d_child); |
1689 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
1690 | |
1691 | /* Switch the names.. */ | |
1692 | switch_names(dentry, target); | |
9a8d5bb4 | 1693 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
1694 | |
1695 | /* ... and switch the parents */ | |
1696 | if (IS_ROOT(dentry)) { | |
1697 | dentry->d_parent = target->d_parent; | |
1698 | target->d_parent = target; | |
5160ee6f | 1699 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 1700 | } else { |
9a8d5bb4 | 1701 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
1702 | |
1703 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 1704 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
1705 | } |
1706 | ||
5160ee6f | 1707 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 1708 | spin_unlock(&target->d_lock); |
c32ccd87 | 1709 | fsnotify_d_move(dentry); |
1da177e4 LT |
1710 | spin_unlock(&dentry->d_lock); |
1711 | write_sequnlock(&rename_lock); | |
9eaef27b TM |
1712 | } |
1713 | ||
1714 | /** | |
1715 | * d_move - move a dentry | |
1716 | * @dentry: entry to move | |
1717 | * @target: new dentry | |
1718 | * | |
1719 | * Update the dcache to reflect the move of a file name. Negative | |
1720 | * dcache entries should not be moved in this way. | |
1721 | */ | |
1722 | ||
1723 | void d_move(struct dentry * dentry, struct dentry * target) | |
1724 | { | |
1725 | spin_lock(&dcache_lock); | |
1726 | d_move_locked(dentry, target); | |
1da177e4 LT |
1727 | spin_unlock(&dcache_lock); |
1728 | } | |
ec4f8605 | 1729 | EXPORT_SYMBOL(d_move); |
1da177e4 | 1730 | |
e2761a11 OH |
1731 | /** |
1732 | * d_ancestor - search for an ancestor | |
1733 | * @p1: ancestor dentry | |
1734 | * @p2: child dentry | |
1735 | * | |
1736 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
1737 | * an ancestor of p2, else NULL. | |
9eaef27b | 1738 | */ |
e2761a11 | 1739 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
1740 | { |
1741 | struct dentry *p; | |
1742 | ||
871c0067 | 1743 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 1744 | if (p->d_parent == p1) |
e2761a11 | 1745 | return p; |
9eaef27b | 1746 | } |
e2761a11 | 1747 | return NULL; |
9eaef27b TM |
1748 | } |
1749 | ||
1750 | /* | |
1751 | * This helper attempts to cope with remotely renamed directories | |
1752 | * | |
1753 | * It assumes that the caller is already holding | |
1754 | * dentry->d_parent->d_inode->i_mutex and the dcache_lock | |
1755 | * | |
1756 | * Note: If ever the locking in lock_rename() changes, then please | |
1757 | * remember to update this too... | |
9eaef27b TM |
1758 | */ |
1759 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
31f3e0b3 | 1760 | __releases(dcache_lock) |
9eaef27b TM |
1761 | { |
1762 | struct mutex *m1 = NULL, *m2 = NULL; | |
1763 | struct dentry *ret; | |
1764 | ||
1765 | /* If alias and dentry share a parent, then no extra locks required */ | |
1766 | if (alias->d_parent == dentry->d_parent) | |
1767 | goto out_unalias; | |
1768 | ||
1769 | /* Check for loops */ | |
1770 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 1771 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
1772 | goto out_err; |
1773 | ||
1774 | /* See lock_rename() */ | |
1775 | ret = ERR_PTR(-EBUSY); | |
1776 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
1777 | goto out_err; | |
1778 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
1779 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
1780 | goto out_err; | |
1781 | m2 = &alias->d_parent->d_inode->i_mutex; | |
1782 | out_unalias: | |
1783 | d_move_locked(alias, dentry); | |
1784 | ret = alias; | |
1785 | out_err: | |
1786 | spin_unlock(&dcache_lock); | |
1787 | if (m2) | |
1788 | mutex_unlock(m2); | |
1789 | if (m1) | |
1790 | mutex_unlock(m1); | |
1791 | return ret; | |
1792 | } | |
1793 | ||
770bfad8 DH |
1794 | /* |
1795 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
1796 | * named dentry in place of the dentry to be replaced. | |
1797 | */ | |
1798 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
1799 | { | |
1800 | struct dentry *dparent, *aparent; | |
1801 | ||
1802 | switch_names(dentry, anon); | |
9a8d5bb4 | 1803 | swap(dentry->d_name.hash, anon->d_name.hash); |
770bfad8 DH |
1804 | |
1805 | dparent = dentry->d_parent; | |
1806 | aparent = anon->d_parent; | |
1807 | ||
1808 | dentry->d_parent = (aparent == anon) ? dentry : aparent; | |
1809 | list_del(&dentry->d_u.d_child); | |
1810 | if (!IS_ROOT(dentry)) | |
1811 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
1812 | else | |
1813 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
1814 | ||
1815 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
1816 | list_del(&anon->d_u.d_child); | |
1817 | if (!IS_ROOT(anon)) | |
1818 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
1819 | else | |
1820 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
1821 | ||
1822 | anon->d_flags &= ~DCACHE_DISCONNECTED; | |
1823 | } | |
1824 | ||
1825 | /** | |
1826 | * d_materialise_unique - introduce an inode into the tree | |
1827 | * @dentry: candidate dentry | |
1828 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
1829 | * | |
1830 | * Introduces an dentry into the tree, substituting an extant disconnected | |
1831 | * root directory alias in its place if there is one | |
1832 | */ | |
1833 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
1834 | { | |
9eaef27b | 1835 | struct dentry *actual; |
770bfad8 DH |
1836 | |
1837 | BUG_ON(!d_unhashed(dentry)); | |
1838 | ||
1839 | spin_lock(&dcache_lock); | |
1840 | ||
1841 | if (!inode) { | |
1842 | actual = dentry; | |
360da900 | 1843 | __d_instantiate(dentry, NULL); |
770bfad8 DH |
1844 | goto found_lock; |
1845 | } | |
1846 | ||
9eaef27b TM |
1847 | if (S_ISDIR(inode->i_mode)) { |
1848 | struct dentry *alias; | |
1849 | ||
1850 | /* Does an aliased dentry already exist? */ | |
1851 | alias = __d_find_alias(inode, 0); | |
1852 | if (alias) { | |
1853 | actual = alias; | |
1854 | /* Is this an anonymous mountpoint that we could splice | |
1855 | * into our tree? */ | |
1856 | if (IS_ROOT(alias)) { | |
1857 | spin_lock(&alias->d_lock); | |
1858 | __d_materialise_dentry(dentry, alias); | |
1859 | __d_drop(alias); | |
1860 | goto found; | |
1861 | } | |
1862 | /* Nope, but we must(!) avoid directory aliasing */ | |
1863 | actual = __d_unalias(dentry, alias); | |
1864 | if (IS_ERR(actual)) | |
1865 | dput(alias); | |
1866 | goto out_nolock; | |
1867 | } | |
770bfad8 DH |
1868 | } |
1869 | ||
1870 | /* Add a unique reference */ | |
1871 | actual = __d_instantiate_unique(dentry, inode); | |
1872 | if (!actual) | |
1873 | actual = dentry; | |
1874 | else if (unlikely(!d_unhashed(actual))) | |
1875 | goto shouldnt_be_hashed; | |
1876 | ||
1877 | found_lock: | |
1878 | spin_lock(&actual->d_lock); | |
1879 | found: | |
1880 | _d_rehash(actual); | |
1881 | spin_unlock(&actual->d_lock); | |
1882 | spin_unlock(&dcache_lock); | |
9eaef27b | 1883 | out_nolock: |
770bfad8 DH |
1884 | if (actual == dentry) { |
1885 | security_d_instantiate(dentry, inode); | |
1886 | return NULL; | |
1887 | } | |
1888 | ||
1889 | iput(inode); | |
1890 | return actual; | |
1891 | ||
770bfad8 DH |
1892 | shouldnt_be_hashed: |
1893 | spin_unlock(&dcache_lock); | |
1894 | BUG(); | |
770bfad8 | 1895 | } |
ec4f8605 | 1896 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 1897 | |
cdd16d02 | 1898 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
1899 | { |
1900 | *buflen -= namelen; | |
1901 | if (*buflen < 0) | |
1902 | return -ENAMETOOLONG; | |
1903 | *buffer -= namelen; | |
1904 | memcpy(*buffer, str, namelen); | |
1905 | return 0; | |
1906 | } | |
1907 | ||
cdd16d02 MS |
1908 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
1909 | { | |
1910 | return prepend(buffer, buflen, name->name, name->len); | |
1911 | } | |
1912 | ||
1da177e4 | 1913 | /** |
f2eb6575 MS |
1914 | * Prepend path string to a buffer |
1915 | * | |
9d1bc601 MS |
1916 | * @path: the dentry/vfsmount to report |
1917 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
1918 | * @buffer: pointer to the end of the buffer |
1919 | * @buflen: pointer to buffer length | |
552ce544 | 1920 | * |
f2eb6575 | 1921 | * Caller holds the dcache_lock. |
9d1bc601 MS |
1922 | * |
1923 | * If path is not reachable from the supplied root, then the value of | |
1924 | * root is changed (without modifying refcounts). | |
1da177e4 | 1925 | */ |
f2eb6575 MS |
1926 | static int prepend_path(const struct path *path, struct path *root, |
1927 | char **buffer, int *buflen) | |
1da177e4 | 1928 | { |
9d1bc601 MS |
1929 | struct dentry *dentry = path->dentry; |
1930 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
1931 | bool slash = false; |
1932 | int error = 0; | |
6092d048 | 1933 | |
99b7db7b | 1934 | br_read_lock(vfsmount_lock); |
f2eb6575 | 1935 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
1936 | struct dentry * parent; |
1937 | ||
1da177e4 | 1938 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 1939 | /* Global root? */ |
1da177e4 | 1940 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
1941 | goto global_root; |
1942 | } | |
1943 | dentry = vfsmnt->mnt_mountpoint; | |
1944 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
1945 | continue; |
1946 | } | |
1947 | parent = dentry->d_parent; | |
1948 | prefetch(parent); | |
f2eb6575 MS |
1949 | error = prepend_name(buffer, buflen, &dentry->d_name); |
1950 | if (!error) | |
1951 | error = prepend(buffer, buflen, "/", 1); | |
1952 | if (error) | |
1953 | break; | |
1954 | ||
1955 | slash = true; | |
1da177e4 LT |
1956 | dentry = parent; |
1957 | } | |
1958 | ||
be285c71 | 1959 | out: |
f2eb6575 MS |
1960 | if (!error && !slash) |
1961 | error = prepend(buffer, buflen, "/", 1); | |
1962 | ||
99b7db7b | 1963 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 1964 | return error; |
1da177e4 LT |
1965 | |
1966 | global_root: | |
98dc568b MS |
1967 | /* |
1968 | * Filesystems needing to implement special "root names" | |
1969 | * should do so with ->d_dname() | |
1970 | */ | |
1971 | if (IS_ROOT(dentry) && | |
1972 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
1973 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
1974 | (int) dentry->d_name.len, dentry->d_name.name); | |
1975 | } | |
9d1bc601 MS |
1976 | root->mnt = vfsmnt; |
1977 | root->dentry = dentry; | |
be285c71 | 1978 | goto out; |
f2eb6575 | 1979 | } |
be285c71 | 1980 | |
f2eb6575 MS |
1981 | /** |
1982 | * __d_path - return the path of a dentry | |
1983 | * @path: the dentry/vfsmount to report | |
1984 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 1985 | * @buf: buffer to return value in |
f2eb6575 MS |
1986 | * @buflen: buffer length |
1987 | * | |
ffd1f4ed | 1988 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
1989 | * |
1990 | * Returns a pointer into the buffer or an error code if the | |
1991 | * path was too long. | |
1992 | * | |
be148247 | 1993 | * "buflen" should be positive. |
f2eb6575 MS |
1994 | * |
1995 | * If path is not reachable from the supplied root, then the value of | |
1996 | * root is changed (without modifying refcounts). | |
1997 | */ | |
1998 | char *__d_path(const struct path *path, struct path *root, | |
1999 | char *buf, int buflen) | |
2000 | { | |
2001 | char *res = buf + buflen; | |
2002 | int error; | |
2003 | ||
2004 | prepend(&res, &buflen, "\0", 1); | |
be148247 | 2005 | spin_lock(&dcache_lock); |
f2eb6575 | 2006 | error = prepend_path(path, root, &res, &buflen); |
be148247 CH |
2007 | spin_unlock(&dcache_lock); |
2008 | ||
f2eb6575 MS |
2009 | if (error) |
2010 | return ERR_PTR(error); | |
f2eb6575 | 2011 | return res; |
1da177e4 LT |
2012 | } |
2013 | ||
ffd1f4ed MS |
2014 | /* |
2015 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2016 | */ | |
2017 | static int path_with_deleted(const struct path *path, struct path *root, | |
2018 | char **buf, int *buflen) | |
2019 | { | |
2020 | prepend(buf, buflen, "\0", 1); | |
2021 | if (d_unlinked(path->dentry)) { | |
2022 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2023 | if (error) | |
2024 | return error; | |
2025 | } | |
2026 | ||
2027 | return prepend_path(path, root, buf, buflen); | |
2028 | } | |
2029 | ||
8df9d1a4 MS |
2030 | static int prepend_unreachable(char **buffer, int *buflen) |
2031 | { | |
2032 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2033 | } | |
2034 | ||
a03a8a70 JB |
2035 | /** |
2036 | * d_path - return the path of a dentry | |
cf28b486 | 2037 | * @path: path to report |
a03a8a70 JB |
2038 | * @buf: buffer to return value in |
2039 | * @buflen: buffer length | |
2040 | * | |
2041 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2042 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2043 | * | |
52afeefb AV |
2044 | * Returns a pointer into the buffer or an error code if the path was |
2045 | * too long. Note: Callers should use the returned pointer, not the passed | |
2046 | * in buffer, to use the name! The implementation often starts at an offset | |
2047 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2048 | * |
31f3e0b3 | 2049 | * "buflen" should be positive. |
a03a8a70 | 2050 | */ |
20d4fdc1 | 2051 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2052 | { |
ffd1f4ed | 2053 | char *res = buf + buflen; |
6ac08c39 | 2054 | struct path root; |
9d1bc601 | 2055 | struct path tmp; |
ffd1f4ed | 2056 | int error; |
1da177e4 | 2057 | |
c23fbb6b ED |
2058 | /* |
2059 | * We have various synthetic filesystems that never get mounted. On | |
2060 | * these filesystems dentries are never used for lookup purposes, and | |
2061 | * thus don't need to be hashed. They also don't need a name until a | |
2062 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2063 | * below allows us to generate a name for these objects on demand: | |
2064 | */ | |
cf28b486 JB |
2065 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2066 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2067 | |
f7ad3c6b | 2068 | get_fs_root(current->fs, &root); |
552ce544 | 2069 | spin_lock(&dcache_lock); |
9d1bc601 | 2070 | tmp = root; |
ffd1f4ed MS |
2071 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2072 | if (error) | |
2073 | res = ERR_PTR(error); | |
552ce544 | 2074 | spin_unlock(&dcache_lock); |
6ac08c39 | 2075 | path_put(&root); |
1da177e4 LT |
2076 | return res; |
2077 | } | |
ec4f8605 | 2078 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2079 | |
8df9d1a4 MS |
2080 | /** |
2081 | * d_path_with_unreachable - return the path of a dentry | |
2082 | * @path: path to report | |
2083 | * @buf: buffer to return value in | |
2084 | * @buflen: buffer length | |
2085 | * | |
2086 | * The difference from d_path() is that this prepends "(unreachable)" | |
2087 | * to paths which are unreachable from the current process' root. | |
2088 | */ | |
2089 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2090 | { | |
2091 | char *res = buf + buflen; | |
2092 | struct path root; | |
2093 | struct path tmp; | |
2094 | int error; | |
2095 | ||
2096 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2097 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2098 | ||
2099 | get_fs_root(current->fs, &root); | |
2100 | spin_lock(&dcache_lock); | |
2101 | tmp = root; | |
2102 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2103 | if (!error && !path_equal(&tmp, &root)) | |
2104 | error = prepend_unreachable(&res, &buflen); | |
2105 | spin_unlock(&dcache_lock); | |
2106 | path_put(&root); | |
2107 | if (error) | |
2108 | res = ERR_PTR(error); | |
2109 | ||
2110 | return res; | |
2111 | } | |
2112 | ||
c23fbb6b ED |
2113 | /* |
2114 | * Helper function for dentry_operations.d_dname() members | |
2115 | */ | |
2116 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2117 | const char *fmt, ...) | |
2118 | { | |
2119 | va_list args; | |
2120 | char temp[64]; | |
2121 | int sz; | |
2122 | ||
2123 | va_start(args, fmt); | |
2124 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2125 | va_end(args); | |
2126 | ||
2127 | if (sz > sizeof(temp) || sz > buflen) | |
2128 | return ERR_PTR(-ENAMETOOLONG); | |
2129 | ||
2130 | buffer += buflen - sz; | |
2131 | return memcpy(buffer, temp, sz); | |
2132 | } | |
2133 | ||
6092d048 RP |
2134 | /* |
2135 | * Write full pathname from the root of the filesystem into the buffer. | |
2136 | */ | |
c103135c | 2137 | char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2138 | { |
2139 | char *end = buf + buflen; | |
2140 | char *retval; | |
2141 | ||
6092d048 | 2142 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2143 | if (buflen < 1) |
2144 | goto Elong; | |
2145 | /* Get '/' right */ | |
2146 | retval = end-1; | |
2147 | *retval = '/'; | |
2148 | ||
cdd16d02 MS |
2149 | while (!IS_ROOT(dentry)) { |
2150 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 2151 | |
6092d048 | 2152 | prefetch(parent); |
cdd16d02 | 2153 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 RP |
2154 | (prepend(&end, &buflen, "/", 1) != 0)) |
2155 | goto Elong; | |
2156 | ||
2157 | retval = end; | |
2158 | dentry = parent; | |
2159 | } | |
c103135c AV |
2160 | return retval; |
2161 | Elong: | |
2162 | return ERR_PTR(-ENAMETOOLONG); | |
2163 | } | |
2164 | EXPORT_SYMBOL(__dentry_path); | |
2165 | ||
2166 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2167 | { | |
2168 | char *p = NULL; | |
2169 | char *retval; | |
2170 | ||
2171 | spin_lock(&dcache_lock); | |
2172 | if (d_unlinked(dentry)) { | |
2173 | p = buf + buflen; | |
2174 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2175 | goto Elong; | |
2176 | buflen++; | |
2177 | } | |
2178 | retval = __dentry_path(dentry, buf, buflen); | |
6092d048 | 2179 | spin_unlock(&dcache_lock); |
c103135c AV |
2180 | if (!IS_ERR(retval) && p) |
2181 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2182 | return retval; |
2183 | Elong: | |
2184 | spin_unlock(&dcache_lock); | |
2185 | return ERR_PTR(-ENAMETOOLONG); | |
2186 | } | |
2187 | ||
1da177e4 LT |
2188 | /* |
2189 | * NOTE! The user-level library version returns a | |
2190 | * character pointer. The kernel system call just | |
2191 | * returns the length of the buffer filled (which | |
2192 | * includes the ending '\0' character), or a negative | |
2193 | * error value. So libc would do something like | |
2194 | * | |
2195 | * char *getcwd(char * buf, size_t size) | |
2196 | * { | |
2197 | * int retval; | |
2198 | * | |
2199 | * retval = sys_getcwd(buf, size); | |
2200 | * if (retval >= 0) | |
2201 | * return buf; | |
2202 | * errno = -retval; | |
2203 | * return NULL; | |
2204 | * } | |
2205 | */ | |
3cdad428 | 2206 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2207 | { |
552ce544 | 2208 | int error; |
6ac08c39 | 2209 | struct path pwd, root; |
552ce544 | 2210 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2211 | |
2212 | if (!page) | |
2213 | return -ENOMEM; | |
2214 | ||
f7ad3c6b | 2215 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2216 | |
552ce544 | 2217 | error = -ENOENT; |
552ce544 | 2218 | spin_lock(&dcache_lock); |
f3da392e | 2219 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2220 | unsigned long len; |
9d1bc601 | 2221 | struct path tmp = root; |
8df9d1a4 MS |
2222 | char *cwd = page + PAGE_SIZE; |
2223 | int buflen = PAGE_SIZE; | |
1da177e4 | 2224 | |
8df9d1a4 MS |
2225 | prepend(&cwd, &buflen, "\0", 1); |
2226 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
552ce544 LT |
2227 | spin_unlock(&dcache_lock); |
2228 | ||
8df9d1a4 | 2229 | if (error) |
552ce544 LT |
2230 | goto out; |
2231 | ||
8df9d1a4 MS |
2232 | /* Unreachable from current root */ |
2233 | if (!path_equal(&tmp, &root)) { | |
2234 | error = prepend_unreachable(&cwd, &buflen); | |
2235 | if (error) | |
2236 | goto out; | |
2237 | } | |
2238 | ||
552ce544 LT |
2239 | error = -ERANGE; |
2240 | len = PAGE_SIZE + page - cwd; | |
2241 | if (len <= size) { | |
2242 | error = len; | |
2243 | if (copy_to_user(buf, cwd, len)) | |
2244 | error = -EFAULT; | |
2245 | } | |
2246 | } else | |
2247 | spin_unlock(&dcache_lock); | |
1da177e4 LT |
2248 | |
2249 | out: | |
6ac08c39 JB |
2250 | path_put(&pwd); |
2251 | path_put(&root); | |
1da177e4 LT |
2252 | free_page((unsigned long) page); |
2253 | return error; | |
2254 | } | |
2255 | ||
2256 | /* | |
2257 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2258 | * | |
2259 | * Trivially implemented using the dcache structure | |
2260 | */ | |
2261 | ||
2262 | /** | |
2263 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2264 | * @new_dentry: new dentry | |
2265 | * @old_dentry: old dentry | |
2266 | * | |
2267 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2268 | * Returns 0 otherwise. | |
2269 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2270 | */ | |
2271 | ||
e2761a11 | 2272 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2273 | { |
2274 | int result; | |
1da177e4 LT |
2275 | unsigned long seq; |
2276 | ||
e2761a11 OH |
2277 | if (new_dentry == old_dentry) |
2278 | return 1; | |
2279 | ||
2280 | /* | |
2281 | * Need rcu_readlock to protect against the d_parent trashing | |
2282 | * due to d_move | |
1da177e4 LT |
2283 | */ |
2284 | rcu_read_lock(); | |
e2761a11 | 2285 | do { |
1da177e4 | 2286 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2287 | seq = read_seqbegin(&rename_lock); |
e2761a11 | 2288 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2289 | result = 1; |
e2761a11 OH |
2290 | else |
2291 | result = 0; | |
1da177e4 LT |
2292 | } while (read_seqretry(&rename_lock, seq)); |
2293 | rcu_read_unlock(); | |
2294 | ||
2295 | return result; | |
2296 | } | |
2297 | ||
2096f759 AV |
2298 | int path_is_under(struct path *path1, struct path *path2) |
2299 | { | |
2300 | struct vfsmount *mnt = path1->mnt; | |
2301 | struct dentry *dentry = path1->dentry; | |
2302 | int res; | |
99b7db7b NP |
2303 | |
2304 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2305 | if (mnt != path2->mnt) { |
2306 | for (;;) { | |
2307 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2308 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2309 | return 0; |
2310 | } | |
2311 | if (mnt->mnt_parent == path2->mnt) | |
2312 | break; | |
2313 | mnt = mnt->mnt_parent; | |
2314 | } | |
2315 | dentry = mnt->mnt_mountpoint; | |
2316 | } | |
2317 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2318 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2319 | return res; |
2320 | } | |
2321 | EXPORT_SYMBOL(path_is_under); | |
2322 | ||
1da177e4 LT |
2323 | void d_genocide(struct dentry *root) |
2324 | { | |
2325 | struct dentry *this_parent = root; | |
2326 | struct list_head *next; | |
2327 | ||
2328 | spin_lock(&dcache_lock); | |
2329 | repeat: | |
2330 | next = this_parent->d_subdirs.next; | |
2331 | resume: | |
2332 | while (next != &this_parent->d_subdirs) { | |
2333 | struct list_head *tmp = next; | |
5160ee6f | 2334 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
2335 | next = tmp->next; |
2336 | if (d_unhashed(dentry)||!dentry->d_inode) | |
2337 | continue; | |
2338 | if (!list_empty(&dentry->d_subdirs)) { | |
2339 | this_parent = dentry; | |
2340 | goto repeat; | |
2341 | } | |
2342 | atomic_dec(&dentry->d_count); | |
2343 | } | |
2344 | if (this_parent != root) { | |
5160ee6f | 2345 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
2346 | atomic_dec(&this_parent->d_count); |
2347 | this_parent = this_parent->d_parent; | |
2348 | goto resume; | |
2349 | } | |
2350 | spin_unlock(&dcache_lock); | |
2351 | } | |
2352 | ||
2353 | /** | |
2354 | * find_inode_number - check for dentry with name | |
2355 | * @dir: directory to check | |
2356 | * @name: Name to find. | |
2357 | * | |
2358 | * Check whether a dentry already exists for the given name, | |
2359 | * and return the inode number if it has an inode. Otherwise | |
2360 | * 0 is returned. | |
2361 | * | |
2362 | * This routine is used to post-process directory listings for | |
2363 | * filesystems using synthetic inode numbers, and is necessary | |
2364 | * to keep getcwd() working. | |
2365 | */ | |
2366 | ||
2367 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2368 | { | |
2369 | struct dentry * dentry; | |
2370 | ino_t ino = 0; | |
2371 | ||
3e7e241f EB |
2372 | dentry = d_hash_and_lookup(dir, name); |
2373 | if (dentry) { | |
1da177e4 LT |
2374 | if (dentry->d_inode) |
2375 | ino = dentry->d_inode->i_ino; | |
2376 | dput(dentry); | |
2377 | } | |
1da177e4 LT |
2378 | return ino; |
2379 | } | |
ec4f8605 | 2380 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2381 | |
2382 | static __initdata unsigned long dhash_entries; | |
2383 | static int __init set_dhash_entries(char *str) | |
2384 | { | |
2385 | if (!str) | |
2386 | return 0; | |
2387 | dhash_entries = simple_strtoul(str, &str, 0); | |
2388 | return 1; | |
2389 | } | |
2390 | __setup("dhash_entries=", set_dhash_entries); | |
2391 | ||
2392 | static void __init dcache_init_early(void) | |
2393 | { | |
2394 | int loop; | |
2395 | ||
2396 | /* If hashes are distributed across NUMA nodes, defer | |
2397 | * hash allocation until vmalloc space is available. | |
2398 | */ | |
2399 | if (hashdist) | |
2400 | return; | |
2401 | ||
2402 | dentry_hashtable = | |
2403 | alloc_large_system_hash("Dentry cache", | |
2404 | sizeof(struct hlist_head), | |
2405 | dhash_entries, | |
2406 | 13, | |
2407 | HASH_EARLY, | |
2408 | &d_hash_shift, | |
2409 | &d_hash_mask, | |
2410 | 0); | |
2411 | ||
2412 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2413 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2414 | } | |
2415 | ||
74bf17cf | 2416 | static void __init dcache_init(void) |
1da177e4 LT |
2417 | { |
2418 | int loop; | |
2419 | ||
2420 | /* | |
2421 | * A constructor could be added for stable state like the lists, | |
2422 | * but it is probably not worth it because of the cache nature | |
2423 | * of the dcache. | |
2424 | */ | |
0a31bd5f CL |
2425 | dentry_cache = KMEM_CACHE(dentry, |
2426 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2427 | |
8e1f936b | 2428 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2429 | |
2430 | /* Hash may have been set up in dcache_init_early */ | |
2431 | if (!hashdist) | |
2432 | return; | |
2433 | ||
2434 | dentry_hashtable = | |
2435 | alloc_large_system_hash("Dentry cache", | |
2436 | sizeof(struct hlist_head), | |
2437 | dhash_entries, | |
2438 | 13, | |
2439 | 0, | |
2440 | &d_hash_shift, | |
2441 | &d_hash_mask, | |
2442 | 0); | |
2443 | ||
2444 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2445 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2446 | } | |
2447 | ||
2448 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2449 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2450 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2451 | |
1da177e4 LT |
2452 | EXPORT_SYMBOL(d_genocide); |
2453 | ||
1da177e4 LT |
2454 | void __init vfs_caches_init_early(void) |
2455 | { | |
2456 | dcache_init_early(); | |
2457 | inode_init_early(); | |
2458 | } | |
2459 | ||
2460 | void __init vfs_caches_init(unsigned long mempages) | |
2461 | { | |
2462 | unsigned long reserve; | |
2463 | ||
2464 | /* Base hash sizes on available memory, with a reserve equal to | |
2465 | 150% of current kernel size */ | |
2466 | ||
2467 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2468 | mempages -= reserve; | |
2469 | ||
2470 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2471 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2472 | |
74bf17cf DC |
2473 | dcache_init(); |
2474 | inode_init(); | |
1da177e4 | 2475 | files_init(mempages); |
74bf17cf | 2476 | mnt_init(); |
1da177e4 LT |
2477 | bdev_cache_init(); |
2478 | chrdev_init(); | |
2479 | } |