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 | ||
17 | #include <linux/config.h> | |
18 | #include <linux/syscalls.h> | |
19 | #include <linux/string.h> | |
20 | #include <linux/mm.h> | |
21 | #include <linux/fs.h> | |
7a91bf7f | 22 | #include <linux/fsnotify.h> |
1da177e4 LT |
23 | #include <linux/slab.h> |
24 | #include <linux/init.h> | |
25 | #include <linux/smp_lock.h> | |
26 | #include <linux/hash.h> | |
27 | #include <linux/cache.h> | |
28 | #include <linux/module.h> | |
29 | #include <linux/mount.h> | |
30 | #include <linux/file.h> | |
31 | #include <asm/uaccess.h> | |
32 | #include <linux/security.h> | |
33 | #include <linux/seqlock.h> | |
34 | #include <linux/swap.h> | |
35 | #include <linux/bootmem.h> | |
36 | ||
37 | /* #define DCACHE_DEBUG 1 */ | |
38 | ||
39 | int sysctl_vfs_cache_pressure = 100; | |
40 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); | |
41 | ||
42 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock); | |
75c96f85 | 43 | static seqlock_t rename_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED; |
1da177e4 LT |
44 | |
45 | EXPORT_SYMBOL(dcache_lock); | |
46 | ||
47 | static kmem_cache_t *dentry_cache; | |
48 | ||
49 | #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname)) | |
50 | ||
51 | /* | |
52 | * This is the single most critical data structure when it comes | |
53 | * to the dcache: the hashtable for lookups. Somebody should try | |
54 | * to make this good - I've just made it work. | |
55 | * | |
56 | * This hash-function tries to avoid losing too many bits of hash | |
57 | * information, yet avoid using a prime hash-size or similar. | |
58 | */ | |
59 | #define D_HASHBITS d_hash_shift | |
60 | #define D_HASHMASK d_hash_mask | |
61 | ||
62 | static unsigned int d_hash_mask; | |
63 | static unsigned int d_hash_shift; | |
64 | static struct hlist_head *dentry_hashtable; | |
65 | static LIST_HEAD(dentry_unused); | |
66 | ||
67 | /* Statistics gathering. */ | |
68 | struct dentry_stat_t dentry_stat = { | |
69 | .age_limit = 45, | |
70 | }; | |
71 | ||
72 | static void d_callback(struct rcu_head *head) | |
73 | { | |
5160ee6f | 74 | struct dentry * dentry = container_of(head, struct dentry, d_u.d_rcu); |
1da177e4 LT |
75 | |
76 | if (dname_external(dentry)) | |
77 | kfree(dentry->d_name.name); | |
78 | kmem_cache_free(dentry_cache, dentry); | |
79 | } | |
80 | ||
81 | /* | |
82 | * no dcache_lock, please. The caller must decrement dentry_stat.nr_dentry | |
83 | * inside dcache_lock. | |
84 | */ | |
85 | static void d_free(struct dentry *dentry) | |
86 | { | |
87 | if (dentry->d_op && dentry->d_op->d_release) | |
88 | dentry->d_op->d_release(dentry); | |
5160ee6f | 89 | call_rcu(&dentry->d_u.d_rcu, d_callback); |
1da177e4 LT |
90 | } |
91 | ||
92 | /* | |
93 | * Release the dentry's inode, using the filesystem | |
94 | * d_iput() operation if defined. | |
95 | * Called with dcache_lock and per dentry lock held, drops both. | |
96 | */ | |
858119e1 | 97 | static void dentry_iput(struct dentry * dentry) |
1da177e4 LT |
98 | { |
99 | struct inode *inode = dentry->d_inode; | |
100 | if (inode) { | |
101 | dentry->d_inode = NULL; | |
102 | list_del_init(&dentry->d_alias); | |
103 | spin_unlock(&dentry->d_lock); | |
104 | spin_unlock(&dcache_lock); | |
f805fbda LT |
105 | if (!inode->i_nlink) |
106 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
107 | if (dentry->d_op && dentry->d_op->d_iput) |
108 | dentry->d_op->d_iput(dentry, inode); | |
109 | else | |
110 | iput(inode); | |
111 | } else { | |
112 | spin_unlock(&dentry->d_lock); | |
113 | spin_unlock(&dcache_lock); | |
114 | } | |
115 | } | |
116 | ||
117 | /* | |
118 | * This is dput | |
119 | * | |
120 | * This is complicated by the fact that we do not want to put | |
121 | * dentries that are no longer on any hash chain on the unused | |
122 | * list: we'd much rather just get rid of them immediately. | |
123 | * | |
124 | * However, that implies that we have to traverse the dentry | |
125 | * tree upwards to the parents which might _also_ now be | |
126 | * scheduled for deletion (it may have been only waiting for | |
127 | * its last child to go away). | |
128 | * | |
129 | * This tail recursion is done by hand as we don't want to depend | |
130 | * on the compiler to always get this right (gcc generally doesn't). | |
131 | * Real recursion would eat up our stack space. | |
132 | */ | |
133 | ||
134 | /* | |
135 | * dput - release a dentry | |
136 | * @dentry: dentry to release | |
137 | * | |
138 | * Release a dentry. This will drop the usage count and if appropriate | |
139 | * call the dentry unlink method as well as removing it from the queues and | |
140 | * releasing its resources. If the parent dentries were scheduled for release | |
141 | * they too may now get deleted. | |
142 | * | |
143 | * no dcache lock, please. | |
144 | */ | |
145 | ||
146 | void dput(struct dentry *dentry) | |
147 | { | |
148 | if (!dentry) | |
149 | return; | |
150 | ||
151 | repeat: | |
152 | if (atomic_read(&dentry->d_count) == 1) | |
153 | might_sleep(); | |
154 | if (!atomic_dec_and_lock(&dentry->d_count, &dcache_lock)) | |
155 | return; | |
156 | ||
157 | spin_lock(&dentry->d_lock); | |
158 | if (atomic_read(&dentry->d_count)) { | |
159 | spin_unlock(&dentry->d_lock); | |
160 | spin_unlock(&dcache_lock); | |
161 | return; | |
162 | } | |
163 | ||
164 | /* | |
165 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
166 | */ | |
167 | if (dentry->d_op && dentry->d_op->d_delete) { | |
168 | if (dentry->d_op->d_delete(dentry)) | |
169 | goto unhash_it; | |
170 | } | |
171 | /* Unreachable? Get rid of it */ | |
172 | if (d_unhashed(dentry)) | |
173 | goto kill_it; | |
174 | if (list_empty(&dentry->d_lru)) { | |
175 | dentry->d_flags |= DCACHE_REFERENCED; | |
176 | list_add(&dentry->d_lru, &dentry_unused); | |
177 | dentry_stat.nr_unused++; | |
178 | } | |
179 | spin_unlock(&dentry->d_lock); | |
180 | spin_unlock(&dcache_lock); | |
181 | return; | |
182 | ||
183 | unhash_it: | |
184 | __d_drop(dentry); | |
185 | ||
186 | kill_it: { | |
187 | struct dentry *parent; | |
188 | ||
189 | /* If dentry was on d_lru list | |
190 | * delete it from there | |
191 | */ | |
192 | if (!list_empty(&dentry->d_lru)) { | |
193 | list_del(&dentry->d_lru); | |
194 | dentry_stat.nr_unused--; | |
195 | } | |
5160ee6f | 196 | list_del(&dentry->d_u.d_child); |
1da177e4 LT |
197 | dentry_stat.nr_dentry--; /* For d_free, below */ |
198 | /*drops the locks, at that point nobody can reach this dentry */ | |
199 | dentry_iput(dentry); | |
200 | parent = dentry->d_parent; | |
201 | d_free(dentry); | |
202 | if (dentry == parent) | |
203 | return; | |
204 | dentry = parent; | |
205 | goto repeat; | |
206 | } | |
207 | } | |
208 | ||
209 | /** | |
210 | * d_invalidate - invalidate a dentry | |
211 | * @dentry: dentry to invalidate | |
212 | * | |
213 | * Try to invalidate the dentry if it turns out to be | |
214 | * possible. If there are other dentries that can be | |
215 | * reached through this one we can't delete it and we | |
216 | * return -EBUSY. On success we return 0. | |
217 | * | |
218 | * no dcache lock. | |
219 | */ | |
220 | ||
221 | int d_invalidate(struct dentry * dentry) | |
222 | { | |
223 | /* | |
224 | * If it's already been dropped, return OK. | |
225 | */ | |
226 | spin_lock(&dcache_lock); | |
227 | if (d_unhashed(dentry)) { | |
228 | spin_unlock(&dcache_lock); | |
229 | return 0; | |
230 | } | |
231 | /* | |
232 | * Check whether to do a partial shrink_dcache | |
233 | * to get rid of unused child entries. | |
234 | */ | |
235 | if (!list_empty(&dentry->d_subdirs)) { | |
236 | spin_unlock(&dcache_lock); | |
237 | shrink_dcache_parent(dentry); | |
238 | spin_lock(&dcache_lock); | |
239 | } | |
240 | ||
241 | /* | |
242 | * Somebody else still using it? | |
243 | * | |
244 | * If it's a directory, we can't drop it | |
245 | * for fear of somebody re-populating it | |
246 | * with children (even though dropping it | |
247 | * would make it unreachable from the root, | |
248 | * we might still populate it if it was a | |
249 | * working directory or similar). | |
250 | */ | |
251 | spin_lock(&dentry->d_lock); | |
252 | if (atomic_read(&dentry->d_count) > 1) { | |
253 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { | |
254 | spin_unlock(&dentry->d_lock); | |
255 | spin_unlock(&dcache_lock); | |
256 | return -EBUSY; | |
257 | } | |
258 | } | |
259 | ||
260 | __d_drop(dentry); | |
261 | spin_unlock(&dentry->d_lock); | |
262 | spin_unlock(&dcache_lock); | |
263 | return 0; | |
264 | } | |
265 | ||
266 | /* This should be called _only_ with dcache_lock held */ | |
267 | ||
268 | static inline struct dentry * __dget_locked(struct dentry *dentry) | |
269 | { | |
270 | atomic_inc(&dentry->d_count); | |
271 | if (!list_empty(&dentry->d_lru)) { | |
272 | dentry_stat.nr_unused--; | |
273 | list_del_init(&dentry->d_lru); | |
274 | } | |
275 | return dentry; | |
276 | } | |
277 | ||
278 | struct dentry * dget_locked(struct dentry *dentry) | |
279 | { | |
280 | return __dget_locked(dentry); | |
281 | } | |
282 | ||
283 | /** | |
284 | * d_find_alias - grab a hashed alias of inode | |
285 | * @inode: inode in question | |
286 | * @want_discon: flag, used by d_splice_alias, to request | |
287 | * that only a DISCONNECTED alias be returned. | |
288 | * | |
289 | * If inode has a hashed alias, or is a directory and has any alias, | |
290 | * acquire the reference to alias and return it. Otherwise return NULL. | |
291 | * Notice that if inode is a directory there can be only one alias and | |
292 | * it can be unhashed only if it has no children, or if it is the root | |
293 | * of a filesystem. | |
294 | * | |
295 | * If the inode has a DCACHE_DISCONNECTED alias, then prefer | |
296 | * any other hashed alias over that one unless @want_discon is set, | |
297 | * in which case only return a DCACHE_DISCONNECTED alias. | |
298 | */ | |
299 | ||
300 | static struct dentry * __d_find_alias(struct inode *inode, int want_discon) | |
301 | { | |
302 | struct list_head *head, *next, *tmp; | |
303 | struct dentry *alias, *discon_alias=NULL; | |
304 | ||
305 | head = &inode->i_dentry; | |
306 | next = inode->i_dentry.next; | |
307 | while (next != head) { | |
308 | tmp = next; | |
309 | next = tmp->next; | |
310 | prefetch(next); | |
311 | alias = list_entry(tmp, struct dentry, d_alias); | |
312 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
313 | if (alias->d_flags & DCACHE_DISCONNECTED) | |
314 | discon_alias = alias; | |
315 | else if (!want_discon) { | |
316 | __dget_locked(alias); | |
317 | return alias; | |
318 | } | |
319 | } | |
320 | } | |
321 | if (discon_alias) | |
322 | __dget_locked(discon_alias); | |
323 | return discon_alias; | |
324 | } | |
325 | ||
326 | struct dentry * d_find_alias(struct inode *inode) | |
327 | { | |
214fda1f DH |
328 | struct dentry *de = NULL; |
329 | ||
330 | if (!list_empty(&inode->i_dentry)) { | |
331 | spin_lock(&dcache_lock); | |
332 | de = __d_find_alias(inode, 0); | |
333 | spin_unlock(&dcache_lock); | |
334 | } | |
1da177e4 LT |
335 | return de; |
336 | } | |
337 | ||
338 | /* | |
339 | * Try to kill dentries associated with this inode. | |
340 | * WARNING: you must own a reference to inode. | |
341 | */ | |
342 | void d_prune_aliases(struct inode *inode) | |
343 | { | |
0cdca3f9 | 344 | struct dentry *dentry; |
1da177e4 LT |
345 | restart: |
346 | spin_lock(&dcache_lock); | |
0cdca3f9 | 347 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 LT |
348 | spin_lock(&dentry->d_lock); |
349 | if (!atomic_read(&dentry->d_count)) { | |
350 | __dget_locked(dentry); | |
351 | __d_drop(dentry); | |
352 | spin_unlock(&dentry->d_lock); | |
353 | spin_unlock(&dcache_lock); | |
354 | dput(dentry); | |
355 | goto restart; | |
356 | } | |
357 | spin_unlock(&dentry->d_lock); | |
358 | } | |
359 | spin_unlock(&dcache_lock); | |
360 | } | |
361 | ||
362 | /* | |
363 | * Throw away a dentry - free the inode, dput the parent. | |
364 | * This requires that the LRU list has already been | |
365 | * removed. | |
366 | * Called with dcache_lock, drops it and then regains. | |
367 | */ | |
368 | static inline void prune_one_dentry(struct dentry * dentry) | |
369 | { | |
370 | struct dentry * parent; | |
371 | ||
372 | __d_drop(dentry); | |
5160ee6f | 373 | list_del(&dentry->d_u.d_child); |
1da177e4 LT |
374 | dentry_stat.nr_dentry--; /* For d_free, below */ |
375 | dentry_iput(dentry); | |
376 | parent = dentry->d_parent; | |
377 | d_free(dentry); | |
378 | if (parent != dentry) | |
379 | dput(parent); | |
380 | spin_lock(&dcache_lock); | |
381 | } | |
382 | ||
383 | /** | |
384 | * prune_dcache - shrink the dcache | |
385 | * @count: number of entries to try and free | |
386 | * | |
387 | * Shrink the dcache. This is done when we need | |
388 | * more memory, or simply when we need to unmount | |
389 | * something (at which point we need to unuse | |
390 | * all dentries). | |
391 | * | |
392 | * This function may fail to free any resources if | |
393 | * all the dentries are in use. | |
394 | */ | |
395 | ||
396 | static void prune_dcache(int count) | |
397 | { | |
398 | spin_lock(&dcache_lock); | |
399 | for (; count ; count--) { | |
400 | struct dentry *dentry; | |
401 | struct list_head *tmp; | |
402 | ||
403 | cond_resched_lock(&dcache_lock); | |
404 | ||
405 | tmp = dentry_unused.prev; | |
406 | if (tmp == &dentry_unused) | |
407 | break; | |
408 | list_del_init(tmp); | |
409 | prefetch(dentry_unused.prev); | |
410 | dentry_stat.nr_unused--; | |
411 | dentry = list_entry(tmp, struct dentry, d_lru); | |
412 | ||
413 | spin_lock(&dentry->d_lock); | |
414 | /* | |
415 | * We found an inuse dentry which was not removed from | |
416 | * dentry_unused because of laziness during lookup. Do not free | |
417 | * it - just keep it off the dentry_unused list. | |
418 | */ | |
419 | if (atomic_read(&dentry->d_count)) { | |
420 | spin_unlock(&dentry->d_lock); | |
421 | continue; | |
422 | } | |
423 | /* If the dentry was recently referenced, don't free it. */ | |
424 | if (dentry->d_flags & DCACHE_REFERENCED) { | |
425 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
426 | list_add(&dentry->d_lru, &dentry_unused); | |
427 | dentry_stat.nr_unused++; | |
428 | spin_unlock(&dentry->d_lock); | |
429 | continue; | |
430 | } | |
431 | prune_one_dentry(dentry); | |
432 | } | |
433 | spin_unlock(&dcache_lock); | |
434 | } | |
435 | ||
436 | /* | |
437 | * Shrink the dcache for the specified super block. | |
438 | * This allows us to unmount a device without disturbing | |
439 | * the dcache for the other devices. | |
440 | * | |
441 | * This implementation makes just two traversals of the | |
442 | * unused list. On the first pass we move the selected | |
443 | * dentries to the most recent end, and on the second | |
444 | * pass we free them. The second pass must restart after | |
445 | * each dput(), but since the target dentries are all at | |
446 | * the end, it's really just a single traversal. | |
447 | */ | |
448 | ||
449 | /** | |
450 | * shrink_dcache_sb - shrink dcache for a superblock | |
451 | * @sb: superblock | |
452 | * | |
453 | * Shrink the dcache for the specified super block. This | |
454 | * is used to free the dcache before unmounting a file | |
455 | * system | |
456 | */ | |
457 | ||
458 | void shrink_dcache_sb(struct super_block * sb) | |
459 | { | |
460 | struct list_head *tmp, *next; | |
461 | struct dentry *dentry; | |
462 | ||
463 | /* | |
464 | * Pass one ... move the dentries for the specified | |
465 | * superblock to the most recent end of the unused list. | |
466 | */ | |
467 | spin_lock(&dcache_lock); | |
0cdca3f9 | 468 | list_for_each_safe(tmp, next, &dentry_unused) { |
1da177e4 LT |
469 | dentry = list_entry(tmp, struct dentry, d_lru); |
470 | if (dentry->d_sb != sb) | |
471 | continue; | |
472 | list_del(tmp); | |
473 | list_add(tmp, &dentry_unused); | |
474 | } | |
475 | ||
476 | /* | |
477 | * Pass two ... free the dentries for this superblock. | |
478 | */ | |
479 | repeat: | |
0cdca3f9 | 480 | list_for_each_safe(tmp, next, &dentry_unused) { |
1da177e4 LT |
481 | dentry = list_entry(tmp, struct dentry, d_lru); |
482 | if (dentry->d_sb != sb) | |
483 | continue; | |
484 | dentry_stat.nr_unused--; | |
485 | list_del_init(tmp); | |
486 | spin_lock(&dentry->d_lock); | |
487 | if (atomic_read(&dentry->d_count)) { | |
488 | spin_unlock(&dentry->d_lock); | |
489 | continue; | |
490 | } | |
491 | prune_one_dentry(dentry); | |
492 | goto repeat; | |
493 | } | |
494 | spin_unlock(&dcache_lock); | |
495 | } | |
496 | ||
497 | /* | |
498 | * Search for at least 1 mount point in the dentry's subdirs. | |
499 | * We descend to the next level whenever the d_subdirs | |
500 | * list is non-empty and continue searching. | |
501 | */ | |
502 | ||
503 | /** | |
504 | * have_submounts - check for mounts over a dentry | |
505 | * @parent: dentry to check. | |
506 | * | |
507 | * Return true if the parent or its subdirectories contain | |
508 | * a mount point | |
509 | */ | |
510 | ||
511 | int have_submounts(struct dentry *parent) | |
512 | { | |
513 | struct dentry *this_parent = parent; | |
514 | struct list_head *next; | |
515 | ||
516 | spin_lock(&dcache_lock); | |
517 | if (d_mountpoint(parent)) | |
518 | goto positive; | |
519 | repeat: | |
520 | next = this_parent->d_subdirs.next; | |
521 | resume: | |
522 | while (next != &this_parent->d_subdirs) { | |
523 | struct list_head *tmp = next; | |
5160ee6f | 524 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
525 | next = tmp->next; |
526 | /* Have we found a mount point ? */ | |
527 | if (d_mountpoint(dentry)) | |
528 | goto positive; | |
529 | if (!list_empty(&dentry->d_subdirs)) { | |
530 | this_parent = dentry; | |
531 | goto repeat; | |
532 | } | |
533 | } | |
534 | /* | |
535 | * All done at this level ... ascend and resume the search. | |
536 | */ | |
537 | if (this_parent != parent) { | |
5160ee6f | 538 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
539 | this_parent = this_parent->d_parent; |
540 | goto resume; | |
541 | } | |
542 | spin_unlock(&dcache_lock); | |
543 | return 0; /* No mount points found in tree */ | |
544 | positive: | |
545 | spin_unlock(&dcache_lock); | |
546 | return 1; | |
547 | } | |
548 | ||
549 | /* | |
550 | * Search the dentry child list for the specified parent, | |
551 | * and move any unused dentries to the end of the unused | |
552 | * list for prune_dcache(). We descend to the next level | |
553 | * whenever the d_subdirs list is non-empty and continue | |
554 | * searching. | |
555 | * | |
556 | * It returns zero iff there are no unused children, | |
557 | * otherwise it returns the number of children moved to | |
558 | * the end of the unused list. This may not be the total | |
559 | * number of unused children, because select_parent can | |
560 | * drop the lock and return early due to latency | |
561 | * constraints. | |
562 | */ | |
563 | static int select_parent(struct dentry * parent) | |
564 | { | |
565 | struct dentry *this_parent = parent; | |
566 | struct list_head *next; | |
567 | int found = 0; | |
568 | ||
569 | spin_lock(&dcache_lock); | |
570 | repeat: | |
571 | next = this_parent->d_subdirs.next; | |
572 | resume: | |
573 | while (next != &this_parent->d_subdirs) { | |
574 | struct list_head *tmp = next; | |
5160ee6f | 575 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
576 | next = tmp->next; |
577 | ||
578 | if (!list_empty(&dentry->d_lru)) { | |
579 | dentry_stat.nr_unused--; | |
580 | list_del_init(&dentry->d_lru); | |
581 | } | |
582 | /* | |
583 | * move only zero ref count dentries to the end | |
584 | * of the unused list for prune_dcache | |
585 | */ | |
586 | if (!atomic_read(&dentry->d_count)) { | |
587 | list_add(&dentry->d_lru, dentry_unused.prev); | |
588 | dentry_stat.nr_unused++; | |
589 | found++; | |
590 | } | |
591 | ||
592 | /* | |
593 | * We can return to the caller if we have found some (this | |
594 | * ensures forward progress). We'll be coming back to find | |
595 | * the rest. | |
596 | */ | |
597 | if (found && need_resched()) | |
598 | goto out; | |
599 | ||
600 | /* | |
601 | * Descend a level if the d_subdirs list is non-empty. | |
602 | */ | |
603 | if (!list_empty(&dentry->d_subdirs)) { | |
604 | this_parent = dentry; | |
605 | #ifdef DCACHE_DEBUG | |
606 | printk(KERN_DEBUG "select_parent: descending to %s/%s, found=%d\n", | |
607 | dentry->d_parent->d_name.name, dentry->d_name.name, found); | |
608 | #endif | |
609 | goto repeat; | |
610 | } | |
611 | } | |
612 | /* | |
613 | * All done at this level ... ascend and resume the search. | |
614 | */ | |
615 | if (this_parent != parent) { | |
5160ee6f | 616 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
617 | this_parent = this_parent->d_parent; |
618 | #ifdef DCACHE_DEBUG | |
619 | printk(KERN_DEBUG "select_parent: ascending to %s/%s, found=%d\n", | |
620 | this_parent->d_parent->d_name.name, this_parent->d_name.name, found); | |
621 | #endif | |
622 | goto resume; | |
623 | } | |
624 | out: | |
625 | spin_unlock(&dcache_lock); | |
626 | return found; | |
627 | } | |
628 | ||
629 | /** | |
630 | * shrink_dcache_parent - prune dcache | |
631 | * @parent: parent of entries to prune | |
632 | * | |
633 | * Prune the dcache to remove unused children of the parent dentry. | |
634 | */ | |
635 | ||
636 | void shrink_dcache_parent(struct dentry * parent) | |
637 | { | |
638 | int found; | |
639 | ||
640 | while ((found = select_parent(parent)) != 0) | |
641 | prune_dcache(found); | |
642 | } | |
643 | ||
644 | /** | |
645 | * shrink_dcache_anon - further prune the cache | |
646 | * @head: head of d_hash list of dentries to prune | |
647 | * | |
648 | * Prune the dentries that are anonymous | |
649 | * | |
665a7583 | 650 | * parsing d_hash list does not hlist_for_each_entry_rcu() as it |
1da177e4 LT |
651 | * done under dcache_lock. |
652 | * | |
653 | */ | |
654 | void shrink_dcache_anon(struct hlist_head *head) | |
655 | { | |
656 | struct hlist_node *lp; | |
657 | int found; | |
658 | do { | |
659 | found = 0; | |
660 | spin_lock(&dcache_lock); | |
661 | hlist_for_each(lp, head) { | |
662 | struct dentry *this = hlist_entry(lp, struct dentry, d_hash); | |
663 | if (!list_empty(&this->d_lru)) { | |
664 | dentry_stat.nr_unused--; | |
665 | list_del_init(&this->d_lru); | |
666 | } | |
667 | ||
668 | /* | |
669 | * move only zero ref count dentries to the end | |
670 | * of the unused list for prune_dcache | |
671 | */ | |
672 | if (!atomic_read(&this->d_count)) { | |
673 | list_add_tail(&this->d_lru, &dentry_unused); | |
674 | dentry_stat.nr_unused++; | |
675 | found++; | |
676 | } | |
677 | } | |
678 | spin_unlock(&dcache_lock); | |
679 | prune_dcache(found); | |
680 | } while(found); | |
681 | } | |
682 | ||
683 | /* | |
684 | * Scan `nr' dentries and return the number which remain. | |
685 | * | |
686 | * We need to avoid reentering the filesystem if the caller is performing a | |
687 | * GFP_NOFS allocation attempt. One example deadlock is: | |
688 | * | |
689 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
690 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
691 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
692 | * | |
693 | * In this case we return -1 to tell the caller that we baled. | |
694 | */ | |
27496a8c | 695 | static int shrink_dcache_memory(int nr, gfp_t gfp_mask) |
1da177e4 LT |
696 | { |
697 | if (nr) { | |
698 | if (!(gfp_mask & __GFP_FS)) | |
699 | return -1; | |
700 | prune_dcache(nr); | |
701 | } | |
702 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; | |
703 | } | |
704 | ||
705 | /** | |
706 | * d_alloc - allocate a dcache entry | |
707 | * @parent: parent of entry to allocate | |
708 | * @name: qstr of the name | |
709 | * | |
710 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
711 | * available. On a success the dentry is returned. The name passed in is | |
712 | * copied and the copy passed in may be reused after this call. | |
713 | */ | |
714 | ||
715 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
716 | { | |
717 | struct dentry *dentry; | |
718 | char *dname; | |
719 | ||
720 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); | |
721 | if (!dentry) | |
722 | return NULL; | |
723 | ||
724 | if (name->len > DNAME_INLINE_LEN-1) { | |
725 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
726 | if (!dname) { | |
727 | kmem_cache_free(dentry_cache, dentry); | |
728 | return NULL; | |
729 | } | |
730 | } else { | |
731 | dname = dentry->d_iname; | |
732 | } | |
733 | dentry->d_name.name = dname; | |
734 | ||
735 | dentry->d_name.len = name->len; | |
736 | dentry->d_name.hash = name->hash; | |
737 | memcpy(dname, name->name, name->len); | |
738 | dname[name->len] = 0; | |
739 | ||
740 | atomic_set(&dentry->d_count, 1); | |
741 | dentry->d_flags = DCACHE_UNHASHED; | |
742 | spin_lock_init(&dentry->d_lock); | |
743 | dentry->d_inode = NULL; | |
744 | dentry->d_parent = NULL; | |
745 | dentry->d_sb = NULL; | |
746 | dentry->d_op = NULL; | |
747 | dentry->d_fsdata = NULL; | |
748 | dentry->d_mounted = 0; | |
47ba87e0 | 749 | #ifdef CONFIG_PROFILING |
1da177e4 | 750 | dentry->d_cookie = NULL; |
47ba87e0 | 751 | #endif |
1da177e4 LT |
752 | INIT_HLIST_NODE(&dentry->d_hash); |
753 | INIT_LIST_HEAD(&dentry->d_lru); | |
754 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
755 | INIT_LIST_HEAD(&dentry->d_alias); | |
756 | ||
757 | if (parent) { | |
758 | dentry->d_parent = dget(parent); | |
759 | dentry->d_sb = parent->d_sb; | |
760 | } else { | |
5160ee6f | 761 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
762 | } |
763 | ||
764 | spin_lock(&dcache_lock); | |
765 | if (parent) | |
5160ee6f | 766 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1da177e4 LT |
767 | dentry_stat.nr_dentry++; |
768 | spin_unlock(&dcache_lock); | |
769 | ||
770 | return dentry; | |
771 | } | |
772 | ||
773 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
774 | { | |
775 | struct qstr q; | |
776 | ||
777 | q.name = name; | |
778 | q.len = strlen(name); | |
779 | q.hash = full_name_hash(q.name, q.len); | |
780 | return d_alloc(parent, &q); | |
781 | } | |
782 | ||
783 | /** | |
784 | * d_instantiate - fill in inode information for a dentry | |
785 | * @entry: dentry to complete | |
786 | * @inode: inode to attach to this dentry | |
787 | * | |
788 | * Fill in inode information in the entry. | |
789 | * | |
790 | * This turns negative dentries into productive full members | |
791 | * of society. | |
792 | * | |
793 | * NOTE! This assumes that the inode count has been incremented | |
794 | * (or otherwise set) by the caller to indicate that it is now | |
795 | * in use by the dcache. | |
796 | */ | |
797 | ||
798 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
799 | { | |
800 | if (!list_empty(&entry->d_alias)) BUG(); | |
801 | spin_lock(&dcache_lock); | |
802 | if (inode) | |
803 | list_add(&entry->d_alias, &inode->i_dentry); | |
804 | entry->d_inode = inode; | |
805 | spin_unlock(&dcache_lock); | |
806 | security_d_instantiate(entry, inode); | |
807 | } | |
808 | ||
809 | /** | |
810 | * d_instantiate_unique - instantiate a non-aliased dentry | |
811 | * @entry: dentry to instantiate | |
812 | * @inode: inode to attach to this dentry | |
813 | * | |
814 | * Fill in inode information in the entry. On success, it returns NULL. | |
815 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 816 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
817 | * |
818 | * Note that in order to avoid conflicts with rename() etc, the caller | |
819 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
820 | * |
821 | * This also assumes that the inode count has been incremented | |
822 | * (or otherwise set) by the caller to indicate that it is now | |
823 | * in use by the dcache. | |
1da177e4 LT |
824 | */ |
825 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
826 | { | |
827 | struct dentry *alias; | |
828 | int len = entry->d_name.len; | |
829 | const char *name = entry->d_name.name; | |
830 | unsigned int hash = entry->d_name.hash; | |
831 | ||
832 | BUG_ON(!list_empty(&entry->d_alias)); | |
833 | spin_lock(&dcache_lock); | |
834 | if (!inode) | |
835 | goto do_negative; | |
836 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
837 | struct qstr *qstr = &alias->d_name; | |
838 | ||
839 | if (qstr->hash != hash) | |
840 | continue; | |
841 | if (alias->d_parent != entry->d_parent) | |
842 | continue; | |
843 | if (qstr->len != len) | |
844 | continue; | |
845 | if (memcmp(qstr->name, name, len)) | |
846 | continue; | |
847 | dget_locked(alias); | |
848 | spin_unlock(&dcache_lock); | |
849 | BUG_ON(!d_unhashed(alias)); | |
e866cfa9 | 850 | iput(inode); |
1da177e4 LT |
851 | return alias; |
852 | } | |
853 | list_add(&entry->d_alias, &inode->i_dentry); | |
854 | do_negative: | |
855 | entry->d_inode = inode; | |
856 | spin_unlock(&dcache_lock); | |
857 | security_d_instantiate(entry, inode); | |
858 | return NULL; | |
859 | } | |
860 | EXPORT_SYMBOL(d_instantiate_unique); | |
861 | ||
862 | /** | |
863 | * d_alloc_root - allocate root dentry | |
864 | * @root_inode: inode to allocate the root for | |
865 | * | |
866 | * Allocate a root ("/") dentry for the inode given. The inode is | |
867 | * instantiated and returned. %NULL is returned if there is insufficient | |
868 | * memory or the inode passed is %NULL. | |
869 | */ | |
870 | ||
871 | struct dentry * d_alloc_root(struct inode * root_inode) | |
872 | { | |
873 | struct dentry *res = NULL; | |
874 | ||
875 | if (root_inode) { | |
876 | static const struct qstr name = { .name = "/", .len = 1 }; | |
877 | ||
878 | res = d_alloc(NULL, &name); | |
879 | if (res) { | |
880 | res->d_sb = root_inode->i_sb; | |
881 | res->d_parent = res; | |
882 | d_instantiate(res, root_inode); | |
883 | } | |
884 | } | |
885 | return res; | |
886 | } | |
887 | ||
888 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
889 | unsigned long hash) | |
890 | { | |
891 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
892 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
893 | return dentry_hashtable + (hash & D_HASHMASK); | |
894 | } | |
895 | ||
896 | /** | |
897 | * d_alloc_anon - allocate an anonymous dentry | |
898 | * @inode: inode to allocate the dentry for | |
899 | * | |
900 | * This is similar to d_alloc_root. It is used by filesystems when | |
901 | * creating a dentry for a given inode, often in the process of | |
902 | * mapping a filehandle to a dentry. The returned dentry may be | |
903 | * anonymous, or may have a full name (if the inode was already | |
904 | * in the cache). The file system may need to make further | |
905 | * efforts to connect this dentry into the dcache properly. | |
906 | * | |
907 | * When called on a directory inode, we must ensure that | |
908 | * the inode only ever has one dentry. If a dentry is | |
909 | * found, that is returned instead of allocating a new one. | |
910 | * | |
911 | * On successful return, the reference to the inode has been transferred | |
912 | * to the dentry. If %NULL is returned (indicating kmalloc failure), | |
913 | * the reference on the inode has not been released. | |
914 | */ | |
915 | ||
916 | struct dentry * d_alloc_anon(struct inode *inode) | |
917 | { | |
918 | static const struct qstr anonstring = { .name = "" }; | |
919 | struct dentry *tmp; | |
920 | struct dentry *res; | |
921 | ||
922 | if ((res = d_find_alias(inode))) { | |
923 | iput(inode); | |
924 | return res; | |
925 | } | |
926 | ||
927 | tmp = d_alloc(NULL, &anonstring); | |
928 | if (!tmp) | |
929 | return NULL; | |
930 | ||
931 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ | |
932 | ||
933 | spin_lock(&dcache_lock); | |
934 | res = __d_find_alias(inode, 0); | |
935 | if (!res) { | |
936 | /* attach a disconnected dentry */ | |
937 | res = tmp; | |
938 | tmp = NULL; | |
939 | spin_lock(&res->d_lock); | |
940 | res->d_sb = inode->i_sb; | |
941 | res->d_parent = res; | |
942 | res->d_inode = inode; | |
943 | res->d_flags |= DCACHE_DISCONNECTED; | |
944 | res->d_flags &= ~DCACHE_UNHASHED; | |
945 | list_add(&res->d_alias, &inode->i_dentry); | |
946 | hlist_add_head(&res->d_hash, &inode->i_sb->s_anon); | |
947 | spin_unlock(&res->d_lock); | |
948 | ||
949 | inode = NULL; /* don't drop reference */ | |
950 | } | |
951 | spin_unlock(&dcache_lock); | |
952 | ||
953 | if (inode) | |
954 | iput(inode); | |
955 | if (tmp) | |
956 | dput(tmp); | |
957 | return res; | |
958 | } | |
959 | ||
960 | ||
961 | /** | |
962 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
963 | * @inode: the inode which may have a disconnected dentry | |
964 | * @dentry: a negative dentry which we want to point to the inode. | |
965 | * | |
966 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
967 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
968 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
969 | * | |
970 | * This is needed in the lookup routine of any filesystem that is exportable | |
971 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
972 | * | |
973 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
974 | * is returned. This matches the expected return value of ->lookup. | |
975 | * | |
976 | */ | |
977 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
978 | { | |
979 | struct dentry *new = NULL; | |
980 | ||
981 | if (inode) { | |
982 | spin_lock(&dcache_lock); | |
983 | new = __d_find_alias(inode, 1); | |
984 | if (new) { | |
985 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
986 | spin_unlock(&dcache_lock); | |
987 | security_d_instantiate(new, inode); | |
988 | d_rehash(dentry); | |
989 | d_move(new, dentry); | |
990 | iput(inode); | |
991 | } else { | |
992 | /* d_instantiate takes dcache_lock, so we do it by hand */ | |
993 | list_add(&dentry->d_alias, &inode->i_dentry); | |
994 | dentry->d_inode = inode; | |
995 | spin_unlock(&dcache_lock); | |
996 | security_d_instantiate(dentry, inode); | |
997 | d_rehash(dentry); | |
998 | } | |
999 | } else | |
1000 | d_add(dentry, inode); | |
1001 | return new; | |
1002 | } | |
1003 | ||
1004 | ||
1005 | /** | |
1006 | * d_lookup - search for a dentry | |
1007 | * @parent: parent dentry | |
1008 | * @name: qstr of name we wish to find | |
1009 | * | |
1010 | * Searches the children of the parent dentry for the name in question. If | |
1011 | * the dentry is found its reference count is incremented and the dentry | |
1012 | * is returned. The caller must use d_put to free the entry when it has | |
1013 | * finished using it. %NULL is returned on failure. | |
1014 | * | |
1015 | * __d_lookup is dcache_lock free. The hash list is protected using RCU. | |
1016 | * Memory barriers are used while updating and doing lockless traversal. | |
1017 | * To avoid races with d_move while rename is happening, d_lock is used. | |
1018 | * | |
1019 | * Overflows in memcmp(), while d_move, are avoided by keeping the length | |
1020 | * and name pointer in one structure pointed by d_qstr. | |
1021 | * | |
1022 | * rcu_read_lock() and rcu_read_unlock() are used to disable preemption while | |
1023 | * lookup is going on. | |
1024 | * | |
1025 | * dentry_unused list is not updated even if lookup finds the required dentry | |
1026 | * in there. It is updated in places such as prune_dcache, shrink_dcache_sb, | |
1027 | * select_parent and __dget_locked. This laziness saves lookup from dcache_lock | |
1028 | * acquisition. | |
1029 | * | |
1030 | * d_lookup() is protected against the concurrent renames in some unrelated | |
1031 | * directory using the seqlockt_t rename_lock. | |
1032 | */ | |
1033 | ||
1034 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) | |
1035 | { | |
1036 | struct dentry * dentry = NULL; | |
1037 | unsigned long seq; | |
1038 | ||
1039 | do { | |
1040 | seq = read_seqbegin(&rename_lock); | |
1041 | dentry = __d_lookup(parent, name); | |
1042 | if (dentry) | |
1043 | break; | |
1044 | } while (read_seqretry(&rename_lock, seq)); | |
1045 | return dentry; | |
1046 | } | |
1047 | ||
1048 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) | |
1049 | { | |
1050 | unsigned int len = name->len; | |
1051 | unsigned int hash = name->hash; | |
1052 | const unsigned char *str = name->name; | |
1053 | struct hlist_head *head = d_hash(parent,hash); | |
1054 | struct dentry *found = NULL; | |
1055 | struct hlist_node *node; | |
665a7583 | 1056 | struct dentry *dentry; |
1da177e4 LT |
1057 | |
1058 | rcu_read_lock(); | |
1059 | ||
665a7583 | 1060 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1061 | struct qstr *qstr; |
1062 | ||
1da177e4 LT |
1063 | if (dentry->d_name.hash != hash) |
1064 | continue; | |
1065 | if (dentry->d_parent != parent) | |
1066 | continue; | |
1067 | ||
1068 | spin_lock(&dentry->d_lock); | |
1069 | ||
1070 | /* | |
1071 | * Recheck the dentry after taking the lock - d_move may have | |
1072 | * changed things. Don't bother checking the hash because we're | |
1073 | * about to compare the whole name anyway. | |
1074 | */ | |
1075 | if (dentry->d_parent != parent) | |
1076 | goto next; | |
1077 | ||
1078 | /* | |
1079 | * It is safe to compare names since d_move() cannot | |
1080 | * change the qstr (protected by d_lock). | |
1081 | */ | |
1082 | qstr = &dentry->d_name; | |
1083 | if (parent->d_op && parent->d_op->d_compare) { | |
1084 | if (parent->d_op->d_compare(parent, qstr, name)) | |
1085 | goto next; | |
1086 | } else { | |
1087 | if (qstr->len != len) | |
1088 | goto next; | |
1089 | if (memcmp(qstr->name, str, len)) | |
1090 | goto next; | |
1091 | } | |
1092 | ||
1093 | if (!d_unhashed(dentry)) { | |
1094 | atomic_inc(&dentry->d_count); | |
1095 | found = dentry; | |
1096 | } | |
1097 | spin_unlock(&dentry->d_lock); | |
1098 | break; | |
1099 | next: | |
1100 | spin_unlock(&dentry->d_lock); | |
1101 | } | |
1102 | rcu_read_unlock(); | |
1103 | ||
1104 | return found; | |
1105 | } | |
1106 | ||
1107 | /** | |
1108 | * d_validate - verify dentry provided from insecure source | |
1109 | * @dentry: The dentry alleged to be valid child of @dparent | |
1110 | * @dparent: The parent dentry (known to be valid) | |
1111 | * @hash: Hash of the dentry | |
1112 | * @len: Length of the name | |
1113 | * | |
1114 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1115 | * This is used by ncpfs in its readdir implementation. | |
1116 | * Zero is returned in the dentry is invalid. | |
1117 | */ | |
1118 | ||
1119 | int d_validate(struct dentry *dentry, struct dentry *dparent) | |
1120 | { | |
1121 | struct hlist_head *base; | |
1122 | struct hlist_node *lhp; | |
1123 | ||
1124 | /* Check whether the ptr might be valid at all.. */ | |
1125 | if (!kmem_ptr_validate(dentry_cache, dentry)) | |
1126 | goto out; | |
1127 | ||
1128 | if (dentry->d_parent != dparent) | |
1129 | goto out; | |
1130 | ||
1131 | spin_lock(&dcache_lock); | |
1132 | base = d_hash(dparent, dentry->d_name.hash); | |
1133 | hlist_for_each(lhp,base) { | |
665a7583 | 1134 | /* hlist_for_each_entry_rcu() not required for d_hash list |
1da177e4 LT |
1135 | * as it is parsed under dcache_lock |
1136 | */ | |
1137 | if (dentry == hlist_entry(lhp, struct dentry, d_hash)) { | |
1138 | __dget_locked(dentry); | |
1139 | spin_unlock(&dcache_lock); | |
1140 | return 1; | |
1141 | } | |
1142 | } | |
1143 | spin_unlock(&dcache_lock); | |
1144 | out: | |
1145 | return 0; | |
1146 | } | |
1147 | ||
1148 | /* | |
1149 | * When a file is deleted, we have two options: | |
1150 | * - turn this dentry into a negative dentry | |
1151 | * - unhash this dentry and free it. | |
1152 | * | |
1153 | * Usually, we want to just turn this into | |
1154 | * a negative dentry, but if anybody else is | |
1155 | * currently using the dentry or the inode | |
1156 | * we can't do that and we fall back on removing | |
1157 | * it from the hash queues and waiting for | |
1158 | * it to be deleted later when it has no users | |
1159 | */ | |
1160 | ||
1161 | /** | |
1162 | * d_delete - delete a dentry | |
1163 | * @dentry: The dentry to delete | |
1164 | * | |
1165 | * Turn the dentry into a negative dentry if possible, otherwise | |
1166 | * remove it from the hash queues so it can be deleted later | |
1167 | */ | |
1168 | ||
1169 | void d_delete(struct dentry * dentry) | |
1170 | { | |
7a91bf7f | 1171 | int isdir = 0; |
1da177e4 LT |
1172 | /* |
1173 | * Are we the only user? | |
1174 | */ | |
1175 | spin_lock(&dcache_lock); | |
1176 | spin_lock(&dentry->d_lock); | |
7a91bf7f | 1177 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
1da177e4 LT |
1178 | if (atomic_read(&dentry->d_count) == 1) { |
1179 | dentry_iput(dentry); | |
7a91bf7f | 1180 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1181 | return; |
1182 | } | |
1183 | ||
1184 | if (!d_unhashed(dentry)) | |
1185 | __d_drop(dentry); | |
1186 | ||
1187 | spin_unlock(&dentry->d_lock); | |
1188 | spin_unlock(&dcache_lock); | |
7a91bf7f JM |
1189 | |
1190 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 LT |
1191 | } |
1192 | ||
1193 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1194 | { | |
1195 | ||
1196 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1197 | hlist_add_head_rcu(&entry->d_hash, list); | |
1198 | } | |
1199 | ||
1200 | /** | |
1201 | * d_rehash - add an entry back to the hash | |
1202 | * @entry: dentry to add to the hash | |
1203 | * | |
1204 | * Adds a dentry to the hash according to its name. | |
1205 | */ | |
1206 | ||
1207 | void d_rehash(struct dentry * entry) | |
1208 | { | |
1209 | struct hlist_head *list = d_hash(entry->d_parent, entry->d_name.hash); | |
1210 | ||
1211 | spin_lock(&dcache_lock); | |
1212 | spin_lock(&entry->d_lock); | |
1213 | __d_rehash(entry, list); | |
1214 | spin_unlock(&entry->d_lock); | |
1215 | spin_unlock(&dcache_lock); | |
1216 | } | |
1217 | ||
1218 | #define do_switch(x,y) do { \ | |
1219 | __typeof__ (x) __tmp = x; \ | |
1220 | x = y; y = __tmp; } while (0) | |
1221 | ||
1222 | /* | |
1223 | * When switching names, the actual string doesn't strictly have to | |
1224 | * be preserved in the target - because we're dropping the target | |
1225 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1226 | * the new name before we switch. | |
1227 | * | |
1228 | * Note that we have to be a lot more careful about getting the hash | |
1229 | * switched - we have to switch the hash value properly even if it | |
1230 | * then no longer matches the actual (corrupted) string of the target. | |
1231 | * The hash value has to match the hash queue that the dentry is on.. | |
1232 | */ | |
1233 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
1234 | { | |
1235 | if (dname_external(target)) { | |
1236 | if (dname_external(dentry)) { | |
1237 | /* | |
1238 | * Both external: swap the pointers | |
1239 | */ | |
1240 | do_switch(target->d_name.name, dentry->d_name.name); | |
1241 | } else { | |
1242 | /* | |
1243 | * dentry:internal, target:external. Steal target's | |
1244 | * storage and make target internal. | |
1245 | */ | |
1246 | dentry->d_name.name = target->d_name.name; | |
1247 | target->d_name.name = target->d_iname; | |
1248 | } | |
1249 | } else { | |
1250 | if (dname_external(dentry)) { | |
1251 | /* | |
1252 | * dentry:external, target:internal. Give dentry's | |
1253 | * storage to target and make dentry internal | |
1254 | */ | |
1255 | memcpy(dentry->d_iname, target->d_name.name, | |
1256 | target->d_name.len + 1); | |
1257 | target->d_name.name = dentry->d_name.name; | |
1258 | dentry->d_name.name = dentry->d_iname; | |
1259 | } else { | |
1260 | /* | |
1261 | * Both are internal. Just copy target to dentry | |
1262 | */ | |
1263 | memcpy(dentry->d_iname, target->d_name.name, | |
1264 | target->d_name.len + 1); | |
1265 | } | |
1266 | } | |
1267 | } | |
1268 | ||
1269 | /* | |
1270 | * We cannibalize "target" when moving dentry on top of it, | |
1271 | * because it's going to be thrown away anyway. We could be more | |
1272 | * polite about it, though. | |
1273 | * | |
1274 | * This forceful removal will result in ugly /proc output if | |
1275 | * somebody holds a file open that got deleted due to a rename. | |
1276 | * We could be nicer about the deleted file, and let it show | |
1277 | * up under the name it got deleted rather than the name that | |
1278 | * deleted it. | |
1279 | */ | |
1280 | ||
1281 | /** | |
1282 | * d_move - move a dentry | |
1283 | * @dentry: entry to move | |
1284 | * @target: new dentry | |
1285 | * | |
1286 | * Update the dcache to reflect the move of a file name. Negative | |
1287 | * dcache entries should not be moved in this way. | |
1288 | */ | |
1289 | ||
1290 | void d_move(struct dentry * dentry, struct dentry * target) | |
1291 | { | |
1292 | struct hlist_head *list; | |
1293 | ||
1294 | if (!dentry->d_inode) | |
1295 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1296 | ||
1297 | spin_lock(&dcache_lock); | |
1298 | write_seqlock(&rename_lock); | |
1299 | /* | |
1300 | * XXXX: do we really need to take target->d_lock? | |
1301 | */ | |
1302 | if (target < dentry) { | |
1303 | spin_lock(&target->d_lock); | |
1304 | spin_lock(&dentry->d_lock); | |
1305 | } else { | |
1306 | spin_lock(&dentry->d_lock); | |
1307 | spin_lock(&target->d_lock); | |
1308 | } | |
1309 | ||
1310 | /* Move the dentry to the target hash queue, if on different bucket */ | |
1311 | if (dentry->d_flags & DCACHE_UNHASHED) | |
1312 | goto already_unhashed; | |
1313 | ||
1314 | hlist_del_rcu(&dentry->d_hash); | |
1315 | ||
1316 | already_unhashed: | |
1317 | list = d_hash(target->d_parent, target->d_name.hash); | |
1318 | __d_rehash(dentry, list); | |
1319 | ||
1320 | /* Unhash the target: dput() will then get rid of it */ | |
1321 | __d_drop(target); | |
1322 | ||
5160ee6f ED |
1323 | list_del(&dentry->d_u.d_child); |
1324 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
1325 | |
1326 | /* Switch the names.. */ | |
1327 | switch_names(dentry, target); | |
1328 | do_switch(dentry->d_name.len, target->d_name.len); | |
1329 | do_switch(dentry->d_name.hash, target->d_name.hash); | |
1330 | ||
1331 | /* ... and switch the parents */ | |
1332 | if (IS_ROOT(dentry)) { | |
1333 | dentry->d_parent = target->d_parent; | |
1334 | target->d_parent = target; | |
5160ee6f | 1335 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 LT |
1336 | } else { |
1337 | do_switch(dentry->d_parent, target->d_parent); | |
1338 | ||
1339 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 1340 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
1341 | } |
1342 | ||
5160ee6f | 1343 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 LT |
1344 | spin_unlock(&target->d_lock); |
1345 | spin_unlock(&dentry->d_lock); | |
1346 | write_sequnlock(&rename_lock); | |
1347 | spin_unlock(&dcache_lock); | |
1348 | } | |
1349 | ||
1350 | /** | |
1351 | * d_path - return the path of a dentry | |
1352 | * @dentry: dentry to report | |
1353 | * @vfsmnt: vfsmnt to which the dentry belongs | |
1354 | * @root: root dentry | |
1355 | * @rootmnt: vfsmnt to which the root dentry belongs | |
1356 | * @buffer: buffer to return value in | |
1357 | * @buflen: buffer length | |
1358 | * | |
1359 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
1360 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
1361 | * | |
1362 | * Returns the buffer or an error code if the path was too long. | |
1363 | * | |
1364 | * "buflen" should be positive. Caller holds the dcache_lock. | |
1365 | */ | |
1366 | static char * __d_path( struct dentry *dentry, struct vfsmount *vfsmnt, | |
1367 | struct dentry *root, struct vfsmount *rootmnt, | |
1368 | char *buffer, int buflen) | |
1369 | { | |
1370 | char * end = buffer+buflen; | |
1371 | char * retval; | |
1372 | int namelen; | |
1373 | ||
1374 | *--end = '\0'; | |
1375 | buflen--; | |
1376 | if (!IS_ROOT(dentry) && d_unhashed(dentry)) { | |
1377 | buflen -= 10; | |
1378 | end -= 10; | |
1379 | if (buflen < 0) | |
1380 | goto Elong; | |
1381 | memcpy(end, " (deleted)", 10); | |
1382 | } | |
1383 | ||
1384 | if (buflen < 1) | |
1385 | goto Elong; | |
1386 | /* Get '/' right */ | |
1387 | retval = end-1; | |
1388 | *retval = '/'; | |
1389 | ||
1390 | for (;;) { | |
1391 | struct dentry * parent; | |
1392 | ||
1393 | if (dentry == root && vfsmnt == rootmnt) | |
1394 | break; | |
1395 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { | |
1396 | /* Global root? */ | |
1397 | spin_lock(&vfsmount_lock); | |
1398 | if (vfsmnt->mnt_parent == vfsmnt) { | |
1399 | spin_unlock(&vfsmount_lock); | |
1400 | goto global_root; | |
1401 | } | |
1402 | dentry = vfsmnt->mnt_mountpoint; | |
1403 | vfsmnt = vfsmnt->mnt_parent; | |
1404 | spin_unlock(&vfsmount_lock); | |
1405 | continue; | |
1406 | } | |
1407 | parent = dentry->d_parent; | |
1408 | prefetch(parent); | |
1409 | namelen = dentry->d_name.len; | |
1410 | buflen -= namelen + 1; | |
1411 | if (buflen < 0) | |
1412 | goto Elong; | |
1413 | end -= namelen; | |
1414 | memcpy(end, dentry->d_name.name, namelen); | |
1415 | *--end = '/'; | |
1416 | retval = end; | |
1417 | dentry = parent; | |
1418 | } | |
1419 | ||
1420 | return retval; | |
1421 | ||
1422 | global_root: | |
1423 | namelen = dentry->d_name.len; | |
1424 | buflen -= namelen; | |
1425 | if (buflen < 0) | |
1426 | goto Elong; | |
1427 | retval -= namelen-1; /* hit the slash */ | |
1428 | memcpy(retval, dentry->d_name.name, namelen); | |
1429 | return retval; | |
1430 | Elong: | |
1431 | return ERR_PTR(-ENAMETOOLONG); | |
1432 | } | |
1433 | ||
1434 | /* write full pathname into buffer and return start of pathname */ | |
1435 | char * d_path(struct dentry *dentry, struct vfsmount *vfsmnt, | |
1436 | char *buf, int buflen) | |
1437 | { | |
1438 | char *res; | |
1439 | struct vfsmount *rootmnt; | |
1440 | struct dentry *root; | |
1441 | ||
1442 | read_lock(¤t->fs->lock); | |
1443 | rootmnt = mntget(current->fs->rootmnt); | |
1444 | root = dget(current->fs->root); | |
1445 | read_unlock(¤t->fs->lock); | |
1446 | spin_lock(&dcache_lock); | |
1447 | res = __d_path(dentry, vfsmnt, root, rootmnt, buf, buflen); | |
1448 | spin_unlock(&dcache_lock); | |
1449 | dput(root); | |
1450 | mntput(rootmnt); | |
1451 | return res; | |
1452 | } | |
1453 | ||
1454 | /* | |
1455 | * NOTE! The user-level library version returns a | |
1456 | * character pointer. The kernel system call just | |
1457 | * returns the length of the buffer filled (which | |
1458 | * includes the ending '\0' character), or a negative | |
1459 | * error value. So libc would do something like | |
1460 | * | |
1461 | * char *getcwd(char * buf, size_t size) | |
1462 | * { | |
1463 | * int retval; | |
1464 | * | |
1465 | * retval = sys_getcwd(buf, size); | |
1466 | * if (retval >= 0) | |
1467 | * return buf; | |
1468 | * errno = -retval; | |
1469 | * return NULL; | |
1470 | * } | |
1471 | */ | |
1472 | asmlinkage long sys_getcwd(char __user *buf, unsigned long size) | |
1473 | { | |
1474 | int error; | |
1475 | struct vfsmount *pwdmnt, *rootmnt; | |
1476 | struct dentry *pwd, *root; | |
1477 | char *page = (char *) __get_free_page(GFP_USER); | |
1478 | ||
1479 | if (!page) | |
1480 | return -ENOMEM; | |
1481 | ||
1482 | read_lock(¤t->fs->lock); | |
1483 | pwdmnt = mntget(current->fs->pwdmnt); | |
1484 | pwd = dget(current->fs->pwd); | |
1485 | rootmnt = mntget(current->fs->rootmnt); | |
1486 | root = dget(current->fs->root); | |
1487 | read_unlock(¤t->fs->lock); | |
1488 | ||
1489 | error = -ENOENT; | |
1490 | /* Has the current directory has been unlinked? */ | |
1491 | spin_lock(&dcache_lock); | |
1492 | if (pwd->d_parent == pwd || !d_unhashed(pwd)) { | |
1493 | unsigned long len; | |
1494 | char * cwd; | |
1495 | ||
1496 | cwd = __d_path(pwd, pwdmnt, root, rootmnt, page, PAGE_SIZE); | |
1497 | spin_unlock(&dcache_lock); | |
1498 | ||
1499 | error = PTR_ERR(cwd); | |
1500 | if (IS_ERR(cwd)) | |
1501 | goto out; | |
1502 | ||
1503 | error = -ERANGE; | |
1504 | len = PAGE_SIZE + page - cwd; | |
1505 | if (len <= size) { | |
1506 | error = len; | |
1507 | if (copy_to_user(buf, cwd, len)) | |
1508 | error = -EFAULT; | |
1509 | } | |
1510 | } else | |
1511 | spin_unlock(&dcache_lock); | |
1512 | ||
1513 | out: | |
1514 | dput(pwd); | |
1515 | mntput(pwdmnt); | |
1516 | dput(root); | |
1517 | mntput(rootmnt); | |
1518 | free_page((unsigned long) page); | |
1519 | return error; | |
1520 | } | |
1521 | ||
1522 | /* | |
1523 | * Test whether new_dentry is a subdirectory of old_dentry. | |
1524 | * | |
1525 | * Trivially implemented using the dcache structure | |
1526 | */ | |
1527 | ||
1528 | /** | |
1529 | * is_subdir - is new dentry a subdirectory of old_dentry | |
1530 | * @new_dentry: new dentry | |
1531 | * @old_dentry: old dentry | |
1532 | * | |
1533 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
1534 | * Returns 0 otherwise. | |
1535 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
1536 | */ | |
1537 | ||
1538 | int is_subdir(struct dentry * new_dentry, struct dentry * old_dentry) | |
1539 | { | |
1540 | int result; | |
1541 | struct dentry * saved = new_dentry; | |
1542 | unsigned long seq; | |
1543 | ||
1544 | /* need rcu_readlock to protect against the d_parent trashing due to | |
1545 | * d_move | |
1546 | */ | |
1547 | rcu_read_lock(); | |
1548 | do { | |
1549 | /* for restarting inner loop in case of seq retry */ | |
1550 | new_dentry = saved; | |
1551 | result = 0; | |
1552 | seq = read_seqbegin(&rename_lock); | |
1553 | for (;;) { | |
1554 | if (new_dentry != old_dentry) { | |
1555 | struct dentry * parent = new_dentry->d_parent; | |
1556 | if (parent == new_dentry) | |
1557 | break; | |
1558 | new_dentry = parent; | |
1559 | continue; | |
1560 | } | |
1561 | result = 1; | |
1562 | break; | |
1563 | } | |
1564 | } while (read_seqretry(&rename_lock, seq)); | |
1565 | rcu_read_unlock(); | |
1566 | ||
1567 | return result; | |
1568 | } | |
1569 | ||
1570 | void d_genocide(struct dentry *root) | |
1571 | { | |
1572 | struct dentry *this_parent = root; | |
1573 | struct list_head *next; | |
1574 | ||
1575 | spin_lock(&dcache_lock); | |
1576 | repeat: | |
1577 | next = this_parent->d_subdirs.next; | |
1578 | resume: | |
1579 | while (next != &this_parent->d_subdirs) { | |
1580 | struct list_head *tmp = next; | |
5160ee6f | 1581 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1582 | next = tmp->next; |
1583 | if (d_unhashed(dentry)||!dentry->d_inode) | |
1584 | continue; | |
1585 | if (!list_empty(&dentry->d_subdirs)) { | |
1586 | this_parent = dentry; | |
1587 | goto repeat; | |
1588 | } | |
1589 | atomic_dec(&dentry->d_count); | |
1590 | } | |
1591 | if (this_parent != root) { | |
5160ee6f | 1592 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
1593 | atomic_dec(&this_parent->d_count); |
1594 | this_parent = this_parent->d_parent; | |
1595 | goto resume; | |
1596 | } | |
1597 | spin_unlock(&dcache_lock); | |
1598 | } | |
1599 | ||
1600 | /** | |
1601 | * find_inode_number - check for dentry with name | |
1602 | * @dir: directory to check | |
1603 | * @name: Name to find. | |
1604 | * | |
1605 | * Check whether a dentry already exists for the given name, | |
1606 | * and return the inode number if it has an inode. Otherwise | |
1607 | * 0 is returned. | |
1608 | * | |
1609 | * This routine is used to post-process directory listings for | |
1610 | * filesystems using synthetic inode numbers, and is necessary | |
1611 | * to keep getcwd() working. | |
1612 | */ | |
1613 | ||
1614 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
1615 | { | |
1616 | struct dentry * dentry; | |
1617 | ino_t ino = 0; | |
1618 | ||
1619 | /* | |
1620 | * Check for a fs-specific hash function. Note that we must | |
1621 | * calculate the standard hash first, as the d_op->d_hash() | |
1622 | * routine may choose to leave the hash value unchanged. | |
1623 | */ | |
1624 | name->hash = full_name_hash(name->name, name->len); | |
1625 | if (dir->d_op && dir->d_op->d_hash) | |
1626 | { | |
1627 | if (dir->d_op->d_hash(dir, name) != 0) | |
1628 | goto out; | |
1629 | } | |
1630 | ||
1631 | dentry = d_lookup(dir, name); | |
1632 | if (dentry) | |
1633 | { | |
1634 | if (dentry->d_inode) | |
1635 | ino = dentry->d_inode->i_ino; | |
1636 | dput(dentry); | |
1637 | } | |
1638 | out: | |
1639 | return ino; | |
1640 | } | |
1641 | ||
1642 | static __initdata unsigned long dhash_entries; | |
1643 | static int __init set_dhash_entries(char *str) | |
1644 | { | |
1645 | if (!str) | |
1646 | return 0; | |
1647 | dhash_entries = simple_strtoul(str, &str, 0); | |
1648 | return 1; | |
1649 | } | |
1650 | __setup("dhash_entries=", set_dhash_entries); | |
1651 | ||
1652 | static void __init dcache_init_early(void) | |
1653 | { | |
1654 | int loop; | |
1655 | ||
1656 | /* If hashes are distributed across NUMA nodes, defer | |
1657 | * hash allocation until vmalloc space is available. | |
1658 | */ | |
1659 | if (hashdist) | |
1660 | return; | |
1661 | ||
1662 | dentry_hashtable = | |
1663 | alloc_large_system_hash("Dentry cache", | |
1664 | sizeof(struct hlist_head), | |
1665 | dhash_entries, | |
1666 | 13, | |
1667 | HASH_EARLY, | |
1668 | &d_hash_shift, | |
1669 | &d_hash_mask, | |
1670 | 0); | |
1671 | ||
1672 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
1673 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
1674 | } | |
1675 | ||
1676 | static void __init dcache_init(unsigned long mempages) | |
1677 | { | |
1678 | int loop; | |
1679 | ||
1680 | /* | |
1681 | * A constructor could be added for stable state like the lists, | |
1682 | * but it is probably not worth it because of the cache nature | |
1683 | * of the dcache. | |
1684 | */ | |
1685 | dentry_cache = kmem_cache_create("dentry_cache", | |
1686 | sizeof(struct dentry), | |
1687 | 0, | |
b0196009 PJ |
1688 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| |
1689 | SLAB_MEM_SPREAD), | |
1da177e4 LT |
1690 | NULL, NULL); |
1691 | ||
1692 | set_shrinker(DEFAULT_SEEKS, shrink_dcache_memory); | |
1693 | ||
1694 | /* Hash may have been set up in dcache_init_early */ | |
1695 | if (!hashdist) | |
1696 | return; | |
1697 | ||
1698 | dentry_hashtable = | |
1699 | alloc_large_system_hash("Dentry cache", | |
1700 | sizeof(struct hlist_head), | |
1701 | dhash_entries, | |
1702 | 13, | |
1703 | 0, | |
1704 | &d_hash_shift, | |
1705 | &d_hash_mask, | |
1706 | 0); | |
1707 | ||
1708 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
1709 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
1710 | } | |
1711 | ||
1712 | /* SLAB cache for __getname() consumers */ | |
1713 | kmem_cache_t *names_cachep; | |
1714 | ||
1715 | /* SLAB cache for file structures */ | |
1716 | kmem_cache_t *filp_cachep; | |
1717 | ||
1718 | EXPORT_SYMBOL(d_genocide); | |
1719 | ||
1720 | extern void bdev_cache_init(void); | |
1721 | extern void chrdev_init(void); | |
1722 | ||
1723 | void __init vfs_caches_init_early(void) | |
1724 | { | |
1725 | dcache_init_early(); | |
1726 | inode_init_early(); | |
1727 | } | |
1728 | ||
1729 | void __init vfs_caches_init(unsigned long mempages) | |
1730 | { | |
1731 | unsigned long reserve; | |
1732 | ||
1733 | /* Base hash sizes on available memory, with a reserve equal to | |
1734 | 150% of current kernel size */ | |
1735 | ||
1736 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
1737 | mempages -= reserve; | |
1738 | ||
1739 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
1740 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); | |
1741 | ||
1742 | filp_cachep = kmem_cache_create("filp", sizeof(struct file), 0, | |
529bf6be | 1743 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL, NULL); |
1da177e4 LT |
1744 | |
1745 | dcache_init(mempages); | |
1746 | inode_init(mempages); | |
1747 | files_init(mempages); | |
1748 | mnt_init(mempages); | |
1749 | bdev_cache_init(); | |
1750 | chrdev_init(); | |
1751 | } | |
1752 | ||
1753 | EXPORT_SYMBOL(d_alloc); | |
1754 | EXPORT_SYMBOL(d_alloc_anon); | |
1755 | EXPORT_SYMBOL(d_alloc_root); | |
1756 | EXPORT_SYMBOL(d_delete); | |
1757 | EXPORT_SYMBOL(d_find_alias); | |
1758 | EXPORT_SYMBOL(d_instantiate); | |
1759 | EXPORT_SYMBOL(d_invalidate); | |
1760 | EXPORT_SYMBOL(d_lookup); | |
1761 | EXPORT_SYMBOL(d_move); | |
1762 | EXPORT_SYMBOL(d_path); | |
1763 | EXPORT_SYMBOL(d_prune_aliases); | |
1764 | EXPORT_SYMBOL(d_rehash); | |
1765 | EXPORT_SYMBOL(d_splice_alias); | |
1766 | EXPORT_SYMBOL(d_validate); | |
1767 | EXPORT_SYMBOL(dget_locked); | |
1768 | EXPORT_SYMBOL(dput); | |
1769 | EXPORT_SYMBOL(find_inode_number); | |
1770 | EXPORT_SYMBOL(have_submounts); | |
1771 | EXPORT_SYMBOL(names_cachep); | |
1772 | EXPORT_SYMBOL(shrink_dcache_parent); | |
1773 | EXPORT_SYMBOL(shrink_dcache_sb); |