]>
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 | |
789680d1 NP |
38 | /* |
39 | * Usage: | |
23044507 NP |
40 | * dcache_hash_lock protects: |
41 | * - the dcache hash table, s_anon lists | |
42 | * dcache_lru_lock protects: | |
43 | * - the dcache lru lists and counters | |
44 | * d_lock protects: | |
45 | * - d_flags | |
46 | * - d_name | |
47 | * - d_lru | |
b7ab39f6 | 48 | * - d_count |
da502956 | 49 | * - d_unhashed() |
789680d1 NP |
50 | * |
51 | * Ordering: | |
52 | * dcache_lock | |
53 | * dentry->d_lock | |
23044507 | 54 | * dcache_lru_lock |
789680d1 NP |
55 | * dcache_hash_lock |
56 | * | |
da502956 NP |
57 | * If there is an ancestor relationship: |
58 | * dentry->d_parent->...->d_parent->d_lock | |
59 | * ... | |
60 | * dentry->d_parent->d_lock | |
61 | * dentry->d_lock | |
62 | * | |
63 | * If no ancestor relationship: | |
789680d1 NP |
64 | * if (dentry1 < dentry2) |
65 | * dentry1->d_lock | |
66 | * dentry2->d_lock | |
67 | */ | |
fa3536cc | 68 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
69 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
70 | ||
789680d1 | 71 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_hash_lock); |
23044507 | 72 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
789680d1 | 73 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock); |
74c3cbe3 | 74 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 LT |
75 | |
76 | EXPORT_SYMBOL(dcache_lock); | |
77 | ||
e18b890b | 78 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 LT |
79 | |
80 | #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname)) | |
81 | ||
82 | /* | |
83 | * This is the single most critical data structure when it comes | |
84 | * to the dcache: the hashtable for lookups. Somebody should try | |
85 | * to make this good - I've just made it work. | |
86 | * | |
87 | * This hash-function tries to avoid losing too many bits of hash | |
88 | * information, yet avoid using a prime hash-size or similar. | |
89 | */ | |
90 | #define D_HASHBITS d_hash_shift | |
91 | #define D_HASHMASK d_hash_mask | |
92 | ||
fa3536cc ED |
93 | static unsigned int d_hash_mask __read_mostly; |
94 | static unsigned int d_hash_shift __read_mostly; | |
95 | static struct hlist_head *dentry_hashtable __read_mostly; | |
1da177e4 LT |
96 | |
97 | /* Statistics gathering. */ | |
98 | struct dentry_stat_t dentry_stat = { | |
99 | .age_limit = 45, | |
100 | }; | |
101 | ||
3e880fb5 | 102 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
103 | |
104 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
105 | static int get_nr_dentry(void) |
106 | { | |
107 | int i; | |
108 | int sum = 0; | |
109 | for_each_possible_cpu(i) | |
110 | sum += per_cpu(nr_dentry, i); | |
111 | return sum < 0 ? 0 : sum; | |
112 | } | |
113 | ||
312d3ca8 CH |
114 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
115 | size_t *lenp, loff_t *ppos) | |
116 | { | |
3e880fb5 | 117 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
118 | return proc_dointvec(table, write, buffer, lenp, ppos); |
119 | } | |
120 | #endif | |
121 | ||
9c82ab9c | 122 | static void __d_free(struct rcu_head *head) |
1da177e4 | 123 | { |
9c82ab9c CH |
124 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
125 | ||
fd217f4d | 126 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
127 | if (dname_external(dentry)) |
128 | kfree(dentry->d_name.name); | |
129 | kmem_cache_free(dentry_cache, dentry); | |
130 | } | |
131 | ||
132 | /* | |
312d3ca8 | 133 | * no dcache_lock, please. |
1da177e4 LT |
134 | */ |
135 | static void d_free(struct dentry *dentry) | |
136 | { | |
b7ab39f6 | 137 | BUG_ON(dentry->d_count); |
3e880fb5 | 138 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
139 | if (dentry->d_op && dentry->d_op->d_release) |
140 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 141 | |
b3423415 | 142 | /* if dentry was never inserted into hash, immediate free is OK */ |
e8462caa | 143 | if (hlist_unhashed(&dentry->d_hash)) |
9c82ab9c | 144 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 145 | else |
9c82ab9c | 146 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
147 | } |
148 | ||
149 | /* | |
150 | * Release the dentry's inode, using the filesystem | |
151 | * d_iput() operation if defined. | |
1da177e4 | 152 | */ |
858119e1 | 153 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 MS |
154 | __releases(dentry->d_lock) |
155 | __releases(dcache_lock) | |
1da177e4 LT |
156 | { |
157 | struct inode *inode = dentry->d_inode; | |
158 | if (inode) { | |
159 | dentry->d_inode = NULL; | |
160 | list_del_init(&dentry->d_alias); | |
161 | spin_unlock(&dentry->d_lock); | |
162 | spin_unlock(&dcache_lock); | |
f805fbda LT |
163 | if (!inode->i_nlink) |
164 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
165 | if (dentry->d_op && dentry->d_op->d_iput) |
166 | dentry->d_op->d_iput(dentry, inode); | |
167 | else | |
168 | iput(inode); | |
169 | } else { | |
170 | spin_unlock(&dentry->d_lock); | |
171 | spin_unlock(&dcache_lock); | |
172 | } | |
173 | } | |
174 | ||
da3bbdd4 | 175 | /* |
23044507 | 176 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
177 | */ |
178 | static void dentry_lru_add(struct dentry *dentry) | |
179 | { | |
a4633357 | 180 | if (list_empty(&dentry->d_lru)) { |
23044507 | 181 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
182 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
183 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 184 | dentry_stat.nr_unused++; |
23044507 | 185 | spin_unlock(&dcache_lru_lock); |
a4633357 | 186 | } |
da3bbdd4 KM |
187 | } |
188 | ||
23044507 NP |
189 | static void __dentry_lru_del(struct dentry *dentry) |
190 | { | |
191 | list_del_init(&dentry->d_lru); | |
192 | dentry->d_sb->s_nr_dentry_unused--; | |
193 | dentry_stat.nr_unused--; | |
194 | } | |
195 | ||
da3bbdd4 KM |
196 | static void dentry_lru_del(struct dentry *dentry) |
197 | { | |
198 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
199 | spin_lock(&dcache_lru_lock); |
200 | __dentry_lru_del(dentry); | |
201 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
202 | } |
203 | } | |
204 | ||
a4633357 | 205 | static void dentry_lru_move_tail(struct dentry *dentry) |
da3bbdd4 | 206 | { |
23044507 | 207 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
208 | if (list_empty(&dentry->d_lru)) { |
209 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
210 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 211 | dentry_stat.nr_unused++; |
a4633357 CH |
212 | } else { |
213 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
da3bbdd4 | 214 | } |
23044507 | 215 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
216 | } |
217 | ||
d52b9086 MS |
218 | /** |
219 | * d_kill - kill dentry and return parent | |
220 | * @dentry: dentry to kill | |
221 | * | |
31f3e0b3 | 222 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
223 | * |
224 | * If this is the root of the dentry tree, return NULL. | |
23044507 NP |
225 | * |
226 | * dcache_lock and d_lock must be held by caller, are dropped by d_kill. | |
d52b9086 MS |
227 | */ |
228 | static struct dentry *d_kill(struct dentry *dentry) | |
31f3e0b3 MS |
229 | __releases(dentry->d_lock) |
230 | __releases(dcache_lock) | |
d52b9086 MS |
231 | { |
232 | struct dentry *parent; | |
233 | ||
234 | list_del(&dentry->d_u.d_child); | |
d52b9086 | 235 | dentry_iput(dentry); |
b7ab39f6 NP |
236 | /* |
237 | * dentry_iput drops the locks, at which point nobody (except | |
238 | * transient RCU lookups) can reach this dentry. | |
239 | */ | |
871c0067 OH |
240 | if (IS_ROOT(dentry)) |
241 | parent = NULL; | |
242 | else | |
243 | parent = dentry->d_parent; | |
d52b9086 | 244 | d_free(dentry); |
871c0067 | 245 | return parent; |
d52b9086 MS |
246 | } |
247 | ||
789680d1 NP |
248 | /** |
249 | * d_drop - drop a dentry | |
250 | * @dentry: dentry to drop | |
251 | * | |
252 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
253 | * be found through a VFS lookup any more. Note that this is different from | |
254 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
255 | * possible, giving a successful _negative_ lookup, while d_drop will | |
256 | * just make the cache lookup fail. | |
257 | * | |
258 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
259 | * reason (NFS timeouts or autofs deletes). | |
260 | * | |
261 | * __d_drop requires dentry->d_lock. | |
262 | */ | |
263 | void __d_drop(struct dentry *dentry) | |
264 | { | |
265 | if (!(dentry->d_flags & DCACHE_UNHASHED)) { | |
266 | dentry->d_flags |= DCACHE_UNHASHED; | |
267 | spin_lock(&dcache_hash_lock); | |
268 | hlist_del_rcu(&dentry->d_hash); | |
269 | spin_unlock(&dcache_hash_lock); | |
270 | } | |
271 | } | |
272 | EXPORT_SYMBOL(__d_drop); | |
273 | ||
274 | void d_drop(struct dentry *dentry) | |
275 | { | |
276 | spin_lock(&dcache_lock); | |
277 | spin_lock(&dentry->d_lock); | |
278 | __d_drop(dentry); | |
279 | spin_unlock(&dentry->d_lock); | |
280 | spin_unlock(&dcache_lock); | |
281 | } | |
282 | EXPORT_SYMBOL(d_drop); | |
283 | ||
1da177e4 LT |
284 | /* |
285 | * This is dput | |
286 | * | |
287 | * This is complicated by the fact that we do not want to put | |
288 | * dentries that are no longer on any hash chain on the unused | |
289 | * list: we'd much rather just get rid of them immediately. | |
290 | * | |
291 | * However, that implies that we have to traverse the dentry | |
292 | * tree upwards to the parents which might _also_ now be | |
293 | * scheduled for deletion (it may have been only waiting for | |
294 | * its last child to go away). | |
295 | * | |
296 | * This tail recursion is done by hand as we don't want to depend | |
297 | * on the compiler to always get this right (gcc generally doesn't). | |
298 | * Real recursion would eat up our stack space. | |
299 | */ | |
300 | ||
301 | /* | |
302 | * dput - release a dentry | |
303 | * @dentry: dentry to release | |
304 | * | |
305 | * Release a dentry. This will drop the usage count and if appropriate | |
306 | * call the dentry unlink method as well as removing it from the queues and | |
307 | * releasing its resources. If the parent dentries were scheduled for release | |
308 | * they too may now get deleted. | |
309 | * | |
310 | * no dcache lock, please. | |
311 | */ | |
312 | ||
313 | void dput(struct dentry *dentry) | |
314 | { | |
315 | if (!dentry) | |
316 | return; | |
317 | ||
318 | repeat: | |
b7ab39f6 | 319 | if (dentry->d_count == 1) |
1da177e4 | 320 | might_sleep(); |
1da177e4 | 321 | spin_lock(&dentry->d_lock); |
b7ab39f6 NP |
322 | if (dentry->d_count == 1) { |
323 | if (!spin_trylock(&dcache_lock)) { | |
324 | /* | |
325 | * Something of a livelock possibility we could avoid | |
326 | * by taking dcache_lock and trying again, but we | |
327 | * want to reduce dcache_lock anyway so this will | |
328 | * get improved. | |
329 | */ | |
330 | spin_unlock(&dentry->d_lock); | |
331 | goto repeat; | |
332 | } | |
333 | } | |
334 | dentry->d_count--; | |
335 | if (dentry->d_count) { | |
1da177e4 LT |
336 | spin_unlock(&dentry->d_lock); |
337 | spin_unlock(&dcache_lock); | |
338 | return; | |
339 | } | |
340 | ||
341 | /* | |
342 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
343 | */ | |
344 | if (dentry->d_op && dentry->d_op->d_delete) { | |
345 | if (dentry->d_op->d_delete(dentry)) | |
346 | goto unhash_it; | |
347 | } | |
265ac902 | 348 | |
1da177e4 LT |
349 | /* Unreachable? Get rid of it */ |
350 | if (d_unhashed(dentry)) | |
351 | goto kill_it; | |
265ac902 NP |
352 | |
353 | /* Otherwise leave it cached and ensure it's on the LRU */ | |
354 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 355 | dentry_lru_add(dentry); |
265ac902 | 356 | |
1da177e4 LT |
357 | spin_unlock(&dentry->d_lock); |
358 | spin_unlock(&dcache_lock); | |
359 | return; | |
360 | ||
361 | unhash_it: | |
362 | __d_drop(dentry); | |
d52b9086 | 363 | kill_it: |
da3bbdd4 KM |
364 | /* if dentry was on the d_lru list delete it from there */ |
365 | dentry_lru_del(dentry); | |
d52b9086 MS |
366 | dentry = d_kill(dentry); |
367 | if (dentry) | |
368 | goto repeat; | |
1da177e4 | 369 | } |
ec4f8605 | 370 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
371 | |
372 | /** | |
373 | * d_invalidate - invalidate a dentry | |
374 | * @dentry: dentry to invalidate | |
375 | * | |
376 | * Try to invalidate the dentry if it turns out to be | |
377 | * possible. If there are other dentries that can be | |
378 | * reached through this one we can't delete it and we | |
379 | * return -EBUSY. On success we return 0. | |
380 | * | |
381 | * no dcache lock. | |
382 | */ | |
383 | ||
384 | int d_invalidate(struct dentry * dentry) | |
385 | { | |
386 | /* | |
387 | * If it's already been dropped, return OK. | |
388 | */ | |
389 | spin_lock(&dcache_lock); | |
da502956 | 390 | spin_lock(&dentry->d_lock); |
1da177e4 | 391 | if (d_unhashed(dentry)) { |
da502956 | 392 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
393 | spin_unlock(&dcache_lock); |
394 | return 0; | |
395 | } | |
396 | /* | |
397 | * Check whether to do a partial shrink_dcache | |
398 | * to get rid of unused child entries. | |
399 | */ | |
400 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 401 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
402 | spin_unlock(&dcache_lock); |
403 | shrink_dcache_parent(dentry); | |
404 | spin_lock(&dcache_lock); | |
da502956 | 405 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
406 | } |
407 | ||
408 | /* | |
409 | * Somebody else still using it? | |
410 | * | |
411 | * If it's a directory, we can't drop it | |
412 | * for fear of somebody re-populating it | |
413 | * with children (even though dropping it | |
414 | * would make it unreachable from the root, | |
415 | * we might still populate it if it was a | |
416 | * working directory or similar). | |
417 | */ | |
b7ab39f6 | 418 | if (dentry->d_count > 1) { |
1da177e4 LT |
419 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
420 | spin_unlock(&dentry->d_lock); | |
421 | spin_unlock(&dcache_lock); | |
422 | return -EBUSY; | |
423 | } | |
424 | } | |
425 | ||
426 | __d_drop(dentry); | |
427 | spin_unlock(&dentry->d_lock); | |
428 | spin_unlock(&dcache_lock); | |
429 | return 0; | |
430 | } | |
ec4f8605 | 431 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 432 | |
b7ab39f6 | 433 | /* This must be called with dcache_lock and d_lock held */ |
23044507 NP |
434 | static inline struct dentry * __dget_locked_dlock(struct dentry *dentry) |
435 | { | |
b7ab39f6 | 436 | dentry->d_count++; |
23044507 NP |
437 | dentry_lru_del(dentry); |
438 | return dentry; | |
439 | } | |
440 | ||
b7ab39f6 | 441 | /* This should be called _only_ with dcache_lock held */ |
1da177e4 LT |
442 | static inline struct dentry * __dget_locked(struct dentry *dentry) |
443 | { | |
23044507 | 444 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 445 | __dget_locked_dlock(dentry); |
23044507 | 446 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
447 | return dentry; |
448 | } | |
449 | ||
b7ab39f6 NP |
450 | struct dentry * dget_locked_dlock(struct dentry *dentry) |
451 | { | |
452 | return __dget_locked_dlock(dentry); | |
453 | } | |
454 | ||
1da177e4 LT |
455 | struct dentry * dget_locked(struct dentry *dentry) |
456 | { | |
457 | return __dget_locked(dentry); | |
458 | } | |
ec4f8605 | 459 | EXPORT_SYMBOL(dget_locked); |
1da177e4 | 460 | |
b7ab39f6 NP |
461 | struct dentry *dget_parent(struct dentry *dentry) |
462 | { | |
463 | struct dentry *ret; | |
464 | ||
465 | repeat: | |
466 | spin_lock(&dentry->d_lock); | |
467 | ret = dentry->d_parent; | |
468 | if (!ret) | |
469 | goto out; | |
470 | if (dentry == ret) { | |
471 | ret->d_count++; | |
472 | goto out; | |
473 | } | |
474 | if (!spin_trylock(&ret->d_lock)) { | |
475 | spin_unlock(&dentry->d_lock); | |
476 | cpu_relax(); | |
477 | goto repeat; | |
478 | } | |
479 | BUG_ON(!ret->d_count); | |
480 | ret->d_count++; | |
481 | spin_unlock(&ret->d_lock); | |
482 | out: | |
483 | spin_unlock(&dentry->d_lock); | |
484 | return ret; | |
485 | } | |
486 | EXPORT_SYMBOL(dget_parent); | |
487 | ||
1da177e4 LT |
488 | /** |
489 | * d_find_alias - grab a hashed alias of inode | |
490 | * @inode: inode in question | |
491 | * @want_discon: flag, used by d_splice_alias, to request | |
492 | * that only a DISCONNECTED alias be returned. | |
493 | * | |
494 | * If inode has a hashed alias, or is a directory and has any alias, | |
495 | * acquire the reference to alias and return it. Otherwise return NULL. | |
496 | * Notice that if inode is a directory there can be only one alias and | |
497 | * it can be unhashed only if it has no children, or if it is the root | |
498 | * of a filesystem. | |
499 | * | |
21c0d8fd | 500 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 501 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 502 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 503 | */ |
da502956 | 504 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 505 | { |
da502956 | 506 | struct dentry *alias, *discon_alias; |
1da177e4 | 507 | |
da502956 NP |
508 | again: |
509 | discon_alias = NULL; | |
510 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
511 | spin_lock(&alias->d_lock); | |
1da177e4 | 512 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 513 | if (IS_ROOT(alias) && |
da502956 | 514 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 515 | discon_alias = alias; |
da502956 NP |
516 | } else if (!want_discon) { |
517 | __dget_locked_dlock(alias); | |
518 | spin_unlock(&alias->d_lock); | |
519 | return alias; | |
520 | } | |
521 | } | |
522 | spin_unlock(&alias->d_lock); | |
523 | } | |
524 | if (discon_alias) { | |
525 | alias = discon_alias; | |
526 | spin_lock(&alias->d_lock); | |
527 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
528 | if (IS_ROOT(alias) && | |
529 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
530 | __dget_locked_dlock(alias); | |
531 | spin_unlock(&alias->d_lock); | |
1da177e4 LT |
532 | return alias; |
533 | } | |
534 | } | |
da502956 NP |
535 | spin_unlock(&alias->d_lock); |
536 | goto again; | |
1da177e4 | 537 | } |
da502956 | 538 | return NULL; |
1da177e4 LT |
539 | } |
540 | ||
da502956 | 541 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 542 | { |
214fda1f DH |
543 | struct dentry *de = NULL; |
544 | ||
545 | if (!list_empty(&inode->i_dentry)) { | |
546 | spin_lock(&dcache_lock); | |
547 | de = __d_find_alias(inode, 0); | |
548 | spin_unlock(&dcache_lock); | |
549 | } | |
1da177e4 LT |
550 | return de; |
551 | } | |
ec4f8605 | 552 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
553 | |
554 | /* | |
555 | * Try to kill dentries associated with this inode. | |
556 | * WARNING: you must own a reference to inode. | |
557 | */ | |
558 | void d_prune_aliases(struct inode *inode) | |
559 | { | |
0cdca3f9 | 560 | struct dentry *dentry; |
1da177e4 LT |
561 | restart: |
562 | spin_lock(&dcache_lock); | |
0cdca3f9 | 563 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 564 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 565 | if (!dentry->d_count) { |
23044507 | 566 | __dget_locked_dlock(dentry); |
1da177e4 LT |
567 | __d_drop(dentry); |
568 | spin_unlock(&dentry->d_lock); | |
569 | spin_unlock(&dcache_lock); | |
570 | dput(dentry); | |
571 | goto restart; | |
572 | } | |
573 | spin_unlock(&dentry->d_lock); | |
574 | } | |
575 | spin_unlock(&dcache_lock); | |
576 | } | |
ec4f8605 | 577 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
578 | |
579 | /* | |
d702ccb3 AM |
580 | * Throw away a dentry - free the inode, dput the parent. This requires that |
581 | * the LRU list has already been removed. | |
582 | * | |
85864e10 MS |
583 | * Try to prune ancestors as well. This is necessary to prevent |
584 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
585 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 586 | */ |
85864e10 | 587 | static void prune_one_dentry(struct dentry * dentry) |
31f3e0b3 MS |
588 | __releases(dentry->d_lock) |
589 | __releases(dcache_lock) | |
1da177e4 | 590 | { |
1da177e4 | 591 | __d_drop(dentry); |
d52b9086 | 592 | dentry = d_kill(dentry); |
d52b9086 MS |
593 | |
594 | /* | |
595 | * Prune ancestors. Locking is simpler than in dput(), | |
596 | * because dcache_lock needs to be taken anyway. | |
597 | */ | |
d52b9086 | 598 | while (dentry) { |
23044507 | 599 | spin_lock(&dcache_lock); |
b7ab39f6 NP |
600 | spin_lock(&dentry->d_lock); |
601 | dentry->d_count--; | |
602 | if (dentry->d_count) { | |
603 | spin_unlock(&dentry->d_lock); | |
23044507 | 604 | spin_unlock(&dcache_lock); |
d52b9086 | 605 | return; |
23044507 | 606 | } |
d52b9086 | 607 | |
a4633357 | 608 | dentry_lru_del(dentry); |
d52b9086 MS |
609 | __d_drop(dentry); |
610 | dentry = d_kill(dentry); | |
d52b9086 | 611 | } |
1da177e4 LT |
612 | } |
613 | ||
3049cfe2 | 614 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 615 | { |
da3bbdd4 | 616 | struct dentry *dentry; |
da3bbdd4 | 617 | |
3049cfe2 CH |
618 | while (!list_empty(list)) { |
619 | dentry = list_entry(list->prev, struct dentry, d_lru); | |
23044507 NP |
620 | |
621 | if (!spin_trylock(&dentry->d_lock)) { | |
622 | spin_unlock(&dcache_lru_lock); | |
623 | cpu_relax(); | |
624 | spin_lock(&dcache_lru_lock); | |
625 | continue; | |
626 | } | |
627 | ||
628 | __dentry_lru_del(dentry); | |
3049cfe2 | 629 | |
1da177e4 LT |
630 | /* |
631 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
632 | * the LRU because of laziness during lookup. Do not free |
633 | * it - just keep it off the LRU list. | |
1da177e4 | 634 | */ |
b7ab39f6 | 635 | if (dentry->d_count) { |
da3bbdd4 | 636 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
637 | continue; |
638 | } | |
23044507 NP |
639 | spin_unlock(&dcache_lru_lock); |
640 | ||
da3bbdd4 | 641 | prune_one_dentry(dentry); |
23044507 NP |
642 | /* dcache_lock and dentry->d_lock dropped */ |
643 | spin_lock(&dcache_lock); | |
644 | spin_lock(&dcache_lru_lock); | |
da3bbdd4 | 645 | } |
3049cfe2 CH |
646 | } |
647 | ||
648 | /** | |
649 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
650 | * @sb: superblock to shrink dentry LRU. | |
651 | * @count: number of entries to prune | |
652 | * @flags: flags to control the dentry processing | |
653 | * | |
654 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
655 | */ | |
656 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
657 | { | |
658 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
659 | struct dentry *dentry; | |
660 | LIST_HEAD(referenced); | |
661 | LIST_HEAD(tmp); | |
662 | int cnt = *count; | |
663 | ||
664 | spin_lock(&dcache_lock); | |
23044507 NP |
665 | relock: |
666 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
667 | while (!list_empty(&sb->s_dentry_lru)) { |
668 | dentry = list_entry(sb->s_dentry_lru.prev, | |
669 | struct dentry, d_lru); | |
670 | BUG_ON(dentry->d_sb != sb); | |
671 | ||
23044507 NP |
672 | if (!spin_trylock(&dentry->d_lock)) { |
673 | spin_unlock(&dcache_lru_lock); | |
674 | cpu_relax(); | |
675 | goto relock; | |
676 | } | |
677 | ||
3049cfe2 CH |
678 | /* |
679 | * If we are honouring the DCACHE_REFERENCED flag and the | |
680 | * dentry has this flag set, don't free it. Clear the flag | |
681 | * and put it back on the LRU. | |
682 | */ | |
23044507 NP |
683 | if (flags & DCACHE_REFERENCED && |
684 | dentry->d_flags & DCACHE_REFERENCED) { | |
685 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
686 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 687 | spin_unlock(&dentry->d_lock); |
23044507 NP |
688 | } else { |
689 | list_move_tail(&dentry->d_lru, &tmp); | |
690 | spin_unlock(&dentry->d_lock); | |
691 | if (!--cnt) | |
692 | break; | |
3049cfe2 | 693 | } |
23044507 | 694 | /* XXX: re-add cond_resched_lock when dcache_lock goes away */ |
3049cfe2 CH |
695 | } |
696 | ||
697 | *count = cnt; | |
698 | shrink_dentry_list(&tmp); | |
699 | ||
da3bbdd4 KM |
700 | if (!list_empty(&referenced)) |
701 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 702 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 | 703 | spin_unlock(&dcache_lock); |
3049cfe2 | 704 | |
da3bbdd4 KM |
705 | } |
706 | ||
707 | /** | |
708 | * prune_dcache - shrink the dcache | |
709 | * @count: number of entries to try to free | |
710 | * | |
711 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
712 | * need to unmount something (at which point we need to unuse all dentries). | |
713 | * | |
714 | * This function may fail to free any resources if all the dentries are in use. | |
715 | */ | |
716 | static void prune_dcache(int count) | |
717 | { | |
dca33252 | 718 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 719 | int w_count; |
86c8749e | 720 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
721 | int prune_ratio; |
722 | int pruned; | |
723 | ||
724 | if (unused == 0 || count == 0) | |
725 | return; | |
726 | spin_lock(&dcache_lock); | |
da3bbdd4 KM |
727 | if (count >= unused) |
728 | prune_ratio = 1; | |
729 | else | |
730 | prune_ratio = unused / count; | |
731 | spin_lock(&sb_lock); | |
dca33252 | 732 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
733 | if (list_empty(&sb->s_instances)) |
734 | continue; | |
da3bbdd4 | 735 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 736 | continue; |
da3bbdd4 KM |
737 | sb->s_count++; |
738 | /* Now, we reclaim unused dentrins with fairness. | |
739 | * We reclaim them same percentage from each superblock. | |
740 | * We calculate number of dentries to scan on this sb | |
741 | * as follows, but the implementation is arranged to avoid | |
742 | * overflows: | |
743 | * number of dentries to scan on this sb = | |
744 | * count * (number of dentries on this sb / | |
745 | * number of dentries in the machine) | |
0feae5c4 | 746 | */ |
da3bbdd4 KM |
747 | spin_unlock(&sb_lock); |
748 | if (prune_ratio != 1) | |
749 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
750 | else | |
751 | w_count = sb->s_nr_dentry_unused; | |
752 | pruned = w_count; | |
0feae5c4 | 753 | /* |
da3bbdd4 KM |
754 | * We need to be sure this filesystem isn't being unmounted, |
755 | * otherwise we could race with generic_shutdown_super(), and | |
756 | * end up holding a reference to an inode while the filesystem | |
757 | * is unmounted. So we try to get s_umount, and make sure | |
758 | * s_root isn't NULL. | |
0feae5c4 | 759 | */ |
da3bbdd4 KM |
760 | if (down_read_trylock(&sb->s_umount)) { |
761 | if ((sb->s_root != NULL) && | |
762 | (!list_empty(&sb->s_dentry_lru))) { | |
763 | spin_unlock(&dcache_lock); | |
764 | __shrink_dcache_sb(sb, &w_count, | |
765 | DCACHE_REFERENCED); | |
766 | pruned -= w_count; | |
767 | spin_lock(&dcache_lock); | |
0feae5c4 | 768 | } |
da3bbdd4 | 769 | up_read(&sb->s_umount); |
0feae5c4 | 770 | } |
da3bbdd4 | 771 | spin_lock(&sb_lock); |
dca33252 AV |
772 | if (p) |
773 | __put_super(p); | |
da3bbdd4 | 774 | count -= pruned; |
dca33252 | 775 | p = sb; |
79893c17 AV |
776 | /* more work left to do? */ |
777 | if (count <= 0) | |
778 | break; | |
1da177e4 | 779 | } |
dca33252 AV |
780 | if (p) |
781 | __put_super(p); | |
da3bbdd4 | 782 | spin_unlock(&sb_lock); |
1da177e4 LT |
783 | spin_unlock(&dcache_lock); |
784 | } | |
785 | ||
1da177e4 LT |
786 | /** |
787 | * shrink_dcache_sb - shrink dcache for a superblock | |
788 | * @sb: superblock | |
789 | * | |
3049cfe2 CH |
790 | * Shrink the dcache for the specified super block. This is used to free |
791 | * the dcache before unmounting a file system. | |
1da177e4 | 792 | */ |
3049cfe2 | 793 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 794 | { |
3049cfe2 CH |
795 | LIST_HEAD(tmp); |
796 | ||
797 | spin_lock(&dcache_lock); | |
23044507 | 798 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
799 | while (!list_empty(&sb->s_dentry_lru)) { |
800 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
801 | shrink_dentry_list(&tmp); | |
802 | } | |
23044507 | 803 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 804 | spin_unlock(&dcache_lock); |
1da177e4 | 805 | } |
ec4f8605 | 806 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 807 | |
c636ebdb DH |
808 | /* |
809 | * destroy a single subtree of dentries for unmount | |
810 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
811 | * locking | |
812 | */ | |
813 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
814 | { | |
815 | struct dentry *parent; | |
f8713576 | 816 | unsigned detached = 0; |
c636ebdb DH |
817 | |
818 | BUG_ON(!IS_ROOT(dentry)); | |
819 | ||
820 | /* detach this root from the system */ | |
821 | spin_lock(&dcache_lock); | |
23044507 | 822 | spin_lock(&dentry->d_lock); |
a4633357 | 823 | dentry_lru_del(dentry); |
c636ebdb | 824 | __d_drop(dentry); |
da502956 | 825 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
826 | spin_unlock(&dcache_lock); |
827 | ||
828 | for (;;) { | |
829 | /* descend to the first leaf in the current subtree */ | |
830 | while (!list_empty(&dentry->d_subdirs)) { | |
831 | struct dentry *loop; | |
832 | ||
833 | /* this is a branch with children - detach all of them | |
834 | * from the system in one go */ | |
835 | spin_lock(&dcache_lock); | |
836 | list_for_each_entry(loop, &dentry->d_subdirs, | |
837 | d_u.d_child) { | |
23044507 | 838 | spin_lock(&loop->d_lock); |
a4633357 | 839 | dentry_lru_del(loop); |
c636ebdb | 840 | __d_drop(loop); |
da502956 | 841 | spin_unlock(&loop->d_lock); |
c636ebdb DH |
842 | cond_resched_lock(&dcache_lock); |
843 | } | |
844 | spin_unlock(&dcache_lock); | |
845 | ||
846 | /* move to the first child */ | |
847 | dentry = list_entry(dentry->d_subdirs.next, | |
848 | struct dentry, d_u.d_child); | |
849 | } | |
850 | ||
851 | /* consume the dentries from this leaf up through its parents | |
852 | * until we find one with children or run out altogether */ | |
853 | do { | |
854 | struct inode *inode; | |
855 | ||
b7ab39f6 | 856 | if (dentry->d_count != 0) { |
c636ebdb DH |
857 | printk(KERN_ERR |
858 | "BUG: Dentry %p{i=%lx,n=%s}" | |
859 | " still in use (%d)" | |
860 | " [unmount of %s %s]\n", | |
861 | dentry, | |
862 | dentry->d_inode ? | |
863 | dentry->d_inode->i_ino : 0UL, | |
864 | dentry->d_name.name, | |
b7ab39f6 | 865 | dentry->d_count, |
c636ebdb DH |
866 | dentry->d_sb->s_type->name, |
867 | dentry->d_sb->s_id); | |
868 | BUG(); | |
869 | } | |
870 | ||
871c0067 | 871 | if (IS_ROOT(dentry)) |
c636ebdb | 872 | parent = NULL; |
871c0067 OH |
873 | else { |
874 | parent = dentry->d_parent; | |
b7ab39f6 NP |
875 | spin_lock(&parent->d_lock); |
876 | parent->d_count--; | |
877 | spin_unlock(&parent->d_lock); | |
871c0067 | 878 | } |
c636ebdb DH |
879 | |
880 | list_del(&dentry->d_u.d_child); | |
f8713576 | 881 | detached++; |
c636ebdb DH |
882 | |
883 | inode = dentry->d_inode; | |
884 | if (inode) { | |
885 | dentry->d_inode = NULL; | |
886 | list_del_init(&dentry->d_alias); | |
887 | if (dentry->d_op && dentry->d_op->d_iput) | |
888 | dentry->d_op->d_iput(dentry, inode); | |
889 | else | |
890 | iput(inode); | |
891 | } | |
892 | ||
893 | d_free(dentry); | |
894 | ||
895 | /* finished when we fall off the top of the tree, | |
896 | * otherwise we ascend to the parent and move to the | |
897 | * next sibling if there is one */ | |
898 | if (!parent) | |
312d3ca8 | 899 | return; |
c636ebdb | 900 | dentry = parent; |
c636ebdb DH |
901 | } while (list_empty(&dentry->d_subdirs)); |
902 | ||
903 | dentry = list_entry(dentry->d_subdirs.next, | |
904 | struct dentry, d_u.d_child); | |
905 | } | |
906 | } | |
907 | ||
908 | /* | |
909 | * destroy the dentries attached to a superblock on unmounting | |
910 | * - we don't need to use dentry->d_lock, and only need dcache_lock when | |
911 | * removing the dentry from the system lists and hashes because: | |
912 | * - the superblock is detached from all mountings and open files, so the | |
913 | * dentry trees will not be rearranged by the VFS | |
914 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
915 | * any dentries belonging to this superblock that it comes across | |
916 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
917 | * in this superblock | |
918 | */ | |
919 | void shrink_dcache_for_umount(struct super_block *sb) | |
920 | { | |
921 | struct dentry *dentry; | |
922 | ||
923 | if (down_read_trylock(&sb->s_umount)) | |
924 | BUG(); | |
925 | ||
926 | dentry = sb->s_root; | |
927 | sb->s_root = NULL; | |
b7ab39f6 NP |
928 | spin_lock(&dentry->d_lock); |
929 | dentry->d_count--; | |
930 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
931 | shrink_dcache_for_umount_subtree(dentry); |
932 | ||
933 | while (!hlist_empty(&sb->s_anon)) { | |
934 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
935 | shrink_dcache_for_umount_subtree(dentry); | |
936 | } | |
937 | } | |
938 | ||
1da177e4 LT |
939 | /* |
940 | * Search for at least 1 mount point in the dentry's subdirs. | |
941 | * We descend to the next level whenever the d_subdirs | |
942 | * list is non-empty and continue searching. | |
943 | */ | |
944 | ||
945 | /** | |
946 | * have_submounts - check for mounts over a dentry | |
947 | * @parent: dentry to check. | |
948 | * | |
949 | * Return true if the parent or its subdirectories contain | |
950 | * a mount point | |
951 | */ | |
952 | ||
953 | int have_submounts(struct dentry *parent) | |
954 | { | |
955 | struct dentry *this_parent = parent; | |
956 | struct list_head *next; | |
957 | ||
958 | spin_lock(&dcache_lock); | |
959 | if (d_mountpoint(parent)) | |
960 | goto positive; | |
961 | repeat: | |
962 | next = this_parent->d_subdirs.next; | |
963 | resume: | |
964 | while (next != &this_parent->d_subdirs) { | |
965 | struct list_head *tmp = next; | |
5160ee6f | 966 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
967 | next = tmp->next; |
968 | /* Have we found a mount point ? */ | |
969 | if (d_mountpoint(dentry)) | |
970 | goto positive; | |
971 | if (!list_empty(&dentry->d_subdirs)) { | |
972 | this_parent = dentry; | |
973 | goto repeat; | |
974 | } | |
975 | } | |
976 | /* | |
977 | * All done at this level ... ascend and resume the search. | |
978 | */ | |
979 | if (this_parent != parent) { | |
5160ee6f | 980 | next = this_parent->d_u.d_child.next; |
1da177e4 LT |
981 | this_parent = this_parent->d_parent; |
982 | goto resume; | |
983 | } | |
984 | spin_unlock(&dcache_lock); | |
985 | return 0; /* No mount points found in tree */ | |
986 | positive: | |
987 | spin_unlock(&dcache_lock); | |
988 | return 1; | |
989 | } | |
ec4f8605 | 990 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
991 | |
992 | /* | |
993 | * Search the dentry child list for the specified parent, | |
994 | * and move any unused dentries to the end of the unused | |
995 | * list for prune_dcache(). We descend to the next level | |
996 | * whenever the d_subdirs list is non-empty and continue | |
997 | * searching. | |
998 | * | |
999 | * It returns zero iff there are no unused children, | |
1000 | * otherwise it returns the number of children moved to | |
1001 | * the end of the unused list. This may not be the total | |
1002 | * number of unused children, because select_parent can | |
1003 | * drop the lock and return early due to latency | |
1004 | * constraints. | |
1005 | */ | |
1006 | static int select_parent(struct dentry * parent) | |
1007 | { | |
1008 | struct dentry *this_parent = parent; | |
1009 | struct list_head *next; | |
1010 | int found = 0; | |
1011 | ||
1012 | spin_lock(&dcache_lock); | |
1013 | repeat: | |
1014 | next = this_parent->d_subdirs.next; | |
1015 | resume: | |
1016 | while (next != &this_parent->d_subdirs) { | |
1017 | struct list_head *tmp = next; | |
5160ee6f | 1018 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1019 | next = tmp->next; |
1020 | ||
23044507 NP |
1021 | spin_lock(&dentry->d_lock); |
1022 | ||
1da177e4 LT |
1023 | /* |
1024 | * move only zero ref count dentries to the end | |
1025 | * of the unused list for prune_dcache | |
1026 | */ | |
b7ab39f6 | 1027 | if (!dentry->d_count) { |
a4633357 | 1028 | dentry_lru_move_tail(dentry); |
1da177e4 | 1029 | found++; |
a4633357 CH |
1030 | } else { |
1031 | dentry_lru_del(dentry); | |
1da177e4 LT |
1032 | } |
1033 | ||
23044507 NP |
1034 | spin_unlock(&dentry->d_lock); |
1035 | ||
1da177e4 LT |
1036 | /* |
1037 | * We can return to the caller if we have found some (this | |
1038 | * ensures forward progress). We'll be coming back to find | |
1039 | * the rest. | |
1040 | */ | |
1041 | if (found && need_resched()) | |
1042 | goto out; | |
1043 | ||
1044 | /* | |
1045 | * Descend a level if the d_subdirs list is non-empty. | |
1046 | */ | |
1047 | if (!list_empty(&dentry->d_subdirs)) { | |
1048 | this_parent = dentry; | |
1da177e4 LT |
1049 | goto repeat; |
1050 | } | |
1051 | } | |
1052 | /* | |
1053 | * All done at this level ... ascend and resume the search. | |
1054 | */ | |
1055 | if (this_parent != parent) { | |
5160ee6f | 1056 | next = this_parent->d_u.d_child.next; |
1da177e4 | 1057 | this_parent = this_parent->d_parent; |
1da177e4 LT |
1058 | goto resume; |
1059 | } | |
1060 | out: | |
1061 | spin_unlock(&dcache_lock); | |
1062 | return found; | |
1063 | } | |
1064 | ||
1065 | /** | |
1066 | * shrink_dcache_parent - prune dcache | |
1067 | * @parent: parent of entries to prune | |
1068 | * | |
1069 | * Prune the dcache to remove unused children of the parent dentry. | |
1070 | */ | |
1071 | ||
1072 | void shrink_dcache_parent(struct dentry * parent) | |
1073 | { | |
da3bbdd4 | 1074 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1075 | int found; |
1076 | ||
1077 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 1078 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 1079 | } |
ec4f8605 | 1080 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1081 | |
1da177e4 LT |
1082 | /* |
1083 | * Scan `nr' dentries and return the number which remain. | |
1084 | * | |
1085 | * We need to avoid reentering the filesystem if the caller is performing a | |
1086 | * GFP_NOFS allocation attempt. One example deadlock is: | |
1087 | * | |
1088 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
1089 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
1090 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
1091 | * | |
1092 | * In this case we return -1 to tell the caller that we baled. | |
1093 | */ | |
7f8275d0 | 1094 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
1095 | { |
1096 | if (nr) { | |
1097 | if (!(gfp_mask & __GFP_FS)) | |
1098 | return -1; | |
da3bbdd4 | 1099 | prune_dcache(nr); |
1da177e4 | 1100 | } |
312d3ca8 | 1101 | |
86c8749e | 1102 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
1103 | } |
1104 | ||
8e1f936b RR |
1105 | static struct shrinker dcache_shrinker = { |
1106 | .shrink = shrink_dcache_memory, | |
1107 | .seeks = DEFAULT_SEEKS, | |
1108 | }; | |
1109 | ||
1da177e4 LT |
1110 | /** |
1111 | * d_alloc - allocate a dcache entry | |
1112 | * @parent: parent of entry to allocate | |
1113 | * @name: qstr of the name | |
1114 | * | |
1115 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1116 | * available. On a success the dentry is returned. The name passed in is | |
1117 | * copied and the copy passed in may be reused after this call. | |
1118 | */ | |
1119 | ||
1120 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1121 | { | |
1122 | struct dentry *dentry; | |
1123 | char *dname; | |
1124 | ||
e12ba74d | 1125 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1126 | if (!dentry) |
1127 | return NULL; | |
1128 | ||
1129 | if (name->len > DNAME_INLINE_LEN-1) { | |
1130 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1131 | if (!dname) { | |
1132 | kmem_cache_free(dentry_cache, dentry); | |
1133 | return NULL; | |
1134 | } | |
1135 | } else { | |
1136 | dname = dentry->d_iname; | |
1137 | } | |
1138 | dentry->d_name.name = dname; | |
1139 | ||
1140 | dentry->d_name.len = name->len; | |
1141 | dentry->d_name.hash = name->hash; | |
1142 | memcpy(dname, name->name, name->len); | |
1143 | dname[name->len] = 0; | |
1144 | ||
b7ab39f6 | 1145 | dentry->d_count = 1; |
1da177e4 LT |
1146 | dentry->d_flags = DCACHE_UNHASHED; |
1147 | spin_lock_init(&dentry->d_lock); | |
1148 | dentry->d_inode = NULL; | |
1149 | dentry->d_parent = NULL; | |
1150 | dentry->d_sb = NULL; | |
1151 | dentry->d_op = NULL; | |
1152 | dentry->d_fsdata = NULL; | |
1153 | dentry->d_mounted = 0; | |
1da177e4 LT |
1154 | INIT_HLIST_NODE(&dentry->d_hash); |
1155 | INIT_LIST_HEAD(&dentry->d_lru); | |
1156 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1157 | INIT_LIST_HEAD(&dentry->d_alias); | |
1158 | ||
1159 | if (parent) { | |
1160 | dentry->d_parent = dget(parent); | |
1161 | dentry->d_sb = parent->d_sb; | |
1162 | } else { | |
5160ee6f | 1163 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
1164 | } |
1165 | ||
1166 | spin_lock(&dcache_lock); | |
1167 | if (parent) | |
5160ee6f | 1168 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
1da177e4 LT |
1169 | spin_unlock(&dcache_lock); |
1170 | ||
3e880fb5 | 1171 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1172 | |
1da177e4 LT |
1173 | return dentry; |
1174 | } | |
ec4f8605 | 1175 | EXPORT_SYMBOL(d_alloc); |
1da177e4 LT |
1176 | |
1177 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
1178 | { | |
1179 | struct qstr q; | |
1180 | ||
1181 | q.name = name; | |
1182 | q.len = strlen(name); | |
1183 | q.hash = full_name_hash(q.name, q.len); | |
1184 | return d_alloc(parent, &q); | |
1185 | } | |
ef26ca97 | 1186 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1187 | |
360da900 OH |
1188 | /* the caller must hold dcache_lock */ |
1189 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | |
1190 | { | |
1191 | if (inode) | |
1192 | list_add(&dentry->d_alias, &inode->i_dentry); | |
1193 | dentry->d_inode = inode; | |
1194 | fsnotify_d_instantiate(dentry, inode); | |
1195 | } | |
1196 | ||
1da177e4 LT |
1197 | /** |
1198 | * d_instantiate - fill in inode information for a dentry | |
1199 | * @entry: dentry to complete | |
1200 | * @inode: inode to attach to this dentry | |
1201 | * | |
1202 | * Fill in inode information in the entry. | |
1203 | * | |
1204 | * This turns negative dentries into productive full members | |
1205 | * of society. | |
1206 | * | |
1207 | * NOTE! This assumes that the inode count has been incremented | |
1208 | * (or otherwise set) by the caller to indicate that it is now | |
1209 | * in use by the dcache. | |
1210 | */ | |
1211 | ||
1212 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1213 | { | |
28133c7b | 1214 | BUG_ON(!list_empty(&entry->d_alias)); |
1da177e4 | 1215 | spin_lock(&dcache_lock); |
360da900 | 1216 | __d_instantiate(entry, inode); |
1da177e4 LT |
1217 | spin_unlock(&dcache_lock); |
1218 | security_d_instantiate(entry, inode); | |
1219 | } | |
ec4f8605 | 1220 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1221 | |
1222 | /** | |
1223 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1224 | * @entry: dentry to instantiate | |
1225 | * @inode: inode to attach to this dentry | |
1226 | * | |
1227 | * Fill in inode information in the entry. On success, it returns NULL. | |
1228 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1229 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1230 | * |
1231 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1232 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1233 | * |
1234 | * This also assumes that the inode count has been incremented | |
1235 | * (or otherwise set) by the caller to indicate that it is now | |
1236 | * in use by the dcache. | |
1da177e4 | 1237 | */ |
770bfad8 DH |
1238 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1239 | struct inode *inode) | |
1da177e4 LT |
1240 | { |
1241 | struct dentry *alias; | |
1242 | int len = entry->d_name.len; | |
1243 | const char *name = entry->d_name.name; | |
1244 | unsigned int hash = entry->d_name.hash; | |
1245 | ||
770bfad8 | 1246 | if (!inode) { |
360da900 | 1247 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1248 | return NULL; |
1249 | } | |
1250 | ||
1da177e4 LT |
1251 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1252 | struct qstr *qstr = &alias->d_name; | |
1253 | ||
1254 | if (qstr->hash != hash) | |
1255 | continue; | |
1256 | if (alias->d_parent != entry->d_parent) | |
1257 | continue; | |
1258 | if (qstr->len != len) | |
1259 | continue; | |
1260 | if (memcmp(qstr->name, name, len)) | |
1261 | continue; | |
1262 | dget_locked(alias); | |
1da177e4 LT |
1263 | return alias; |
1264 | } | |
770bfad8 | 1265 | |
360da900 | 1266 | __d_instantiate(entry, inode); |
1da177e4 LT |
1267 | return NULL; |
1268 | } | |
770bfad8 DH |
1269 | |
1270 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1271 | { | |
1272 | struct dentry *result; | |
1273 | ||
1274 | BUG_ON(!list_empty(&entry->d_alias)); | |
1275 | ||
1276 | spin_lock(&dcache_lock); | |
1277 | result = __d_instantiate_unique(entry, inode); | |
1278 | spin_unlock(&dcache_lock); | |
1279 | ||
1280 | if (!result) { | |
1281 | security_d_instantiate(entry, inode); | |
1282 | return NULL; | |
1283 | } | |
1284 | ||
1285 | BUG_ON(!d_unhashed(result)); | |
1286 | iput(inode); | |
1287 | return result; | |
1288 | } | |
1289 | ||
1da177e4 LT |
1290 | EXPORT_SYMBOL(d_instantiate_unique); |
1291 | ||
1292 | /** | |
1293 | * d_alloc_root - allocate root dentry | |
1294 | * @root_inode: inode to allocate the root for | |
1295 | * | |
1296 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1297 | * instantiated and returned. %NULL is returned if there is insufficient | |
1298 | * memory or the inode passed is %NULL. | |
1299 | */ | |
1300 | ||
1301 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1302 | { | |
1303 | struct dentry *res = NULL; | |
1304 | ||
1305 | if (root_inode) { | |
1306 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1307 | ||
1308 | res = d_alloc(NULL, &name); | |
1309 | if (res) { | |
1310 | res->d_sb = root_inode->i_sb; | |
1311 | res->d_parent = res; | |
1312 | d_instantiate(res, root_inode); | |
1313 | } | |
1314 | } | |
1315 | return res; | |
1316 | } | |
ec4f8605 | 1317 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 LT |
1318 | |
1319 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1320 | unsigned long hash) | |
1321 | { | |
1322 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1323 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1324 | return dentry_hashtable + (hash & D_HASHMASK); | |
1325 | } | |
1326 | ||
4ea3ada2 CH |
1327 | /** |
1328 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1329 | * @inode: inode to allocate the dentry for | |
1330 | * | |
1331 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1332 | * similar open by handle operations. The returned dentry may be anonymous, | |
1333 | * or may have a full name (if the inode was already in the cache). | |
1334 | * | |
1335 | * When called on a directory inode, we must ensure that the inode only ever | |
1336 | * has one dentry. If a dentry is found, that is returned instead of | |
1337 | * allocating a new one. | |
1338 | * | |
1339 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1340 | * to the dentry. In case of an error the reference on the inode is released. |
1341 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1342 | * be passed in and will be the error will be propagate to the return value, | |
1343 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1344 | */ |
1345 | struct dentry *d_obtain_alias(struct inode *inode) | |
1346 | { | |
9308a612 CH |
1347 | static const struct qstr anonstring = { .name = "" }; |
1348 | struct dentry *tmp; | |
1349 | struct dentry *res; | |
4ea3ada2 CH |
1350 | |
1351 | if (!inode) | |
44003728 | 1352 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1353 | if (IS_ERR(inode)) |
1354 | return ERR_CAST(inode); | |
1355 | ||
9308a612 CH |
1356 | res = d_find_alias(inode); |
1357 | if (res) | |
1358 | goto out_iput; | |
1359 | ||
1360 | tmp = d_alloc(NULL, &anonstring); | |
1361 | if (!tmp) { | |
1362 | res = ERR_PTR(-ENOMEM); | |
1363 | goto out_iput; | |
4ea3ada2 | 1364 | } |
9308a612 CH |
1365 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1366 | ||
1367 | spin_lock(&dcache_lock); | |
1368 | res = __d_find_alias(inode, 0); | |
1369 | if (res) { | |
1370 | spin_unlock(&dcache_lock); | |
1371 | dput(tmp); | |
1372 | goto out_iput; | |
1373 | } | |
1374 | ||
1375 | /* attach a disconnected dentry */ | |
1376 | spin_lock(&tmp->d_lock); | |
1377 | tmp->d_sb = inode->i_sb; | |
1378 | tmp->d_inode = inode; | |
1379 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1380 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1381 | list_add(&tmp->d_alias, &inode->i_dentry); | |
789680d1 | 1382 | spin_lock(&dcache_hash_lock); |
9308a612 | 1383 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); |
789680d1 | 1384 | spin_unlock(&dcache_hash_lock); |
9308a612 CH |
1385 | spin_unlock(&tmp->d_lock); |
1386 | ||
1387 | spin_unlock(&dcache_lock); | |
1388 | return tmp; | |
1389 | ||
1390 | out_iput: | |
1391 | iput(inode); | |
1392 | return res; | |
4ea3ada2 | 1393 | } |
adc48720 | 1394 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1395 | |
1396 | /** | |
1397 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1398 | * @inode: the inode which may have a disconnected dentry | |
1399 | * @dentry: a negative dentry which we want to point to the inode. | |
1400 | * | |
1401 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1402 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1403 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1404 | * | |
1405 | * This is needed in the lookup routine of any filesystem that is exportable | |
1406 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1407 | * | |
1408 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1409 | * is returned. This matches the expected return value of ->lookup. | |
1410 | * | |
1411 | */ | |
1412 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1413 | { | |
1414 | struct dentry *new = NULL; | |
1415 | ||
21c0d8fd | 1416 | if (inode && S_ISDIR(inode->i_mode)) { |
1da177e4 LT |
1417 | spin_lock(&dcache_lock); |
1418 | new = __d_find_alias(inode, 1); | |
1419 | if (new) { | |
1420 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
1421 | spin_unlock(&dcache_lock); | |
1422 | security_d_instantiate(new, inode); | |
1da177e4 LT |
1423 | d_move(new, dentry); |
1424 | iput(inode); | |
1425 | } else { | |
360da900 OH |
1426 | /* already taking dcache_lock, so d_add() by hand */ |
1427 | __d_instantiate(dentry, inode); | |
1da177e4 LT |
1428 | spin_unlock(&dcache_lock); |
1429 | security_d_instantiate(dentry, inode); | |
1430 | d_rehash(dentry); | |
1431 | } | |
1432 | } else | |
1433 | d_add(dentry, inode); | |
1434 | return new; | |
1435 | } | |
ec4f8605 | 1436 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1437 | |
9403540c BN |
1438 | /** |
1439 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1440 | * @inode: the inode case-insensitive lookup has found | |
1441 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1442 | * @name: the case-exact name to be associated with the returned dentry | |
1443 | * | |
1444 | * This is to avoid filling the dcache with case-insensitive names to the | |
1445 | * same inode, only the actual correct case is stored in the dcache for | |
1446 | * case-insensitive filesystems. | |
1447 | * | |
1448 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1449 | * already exists in in the dcache, use it and return it. | |
1450 | * | |
1451 | * If no entry exists with the exact case name, allocate new dentry with | |
1452 | * the exact case, and return the spliced entry. | |
1453 | */ | |
e45b590b | 1454 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1455 | struct qstr *name) |
1456 | { | |
1457 | int error; | |
1458 | struct dentry *found; | |
1459 | struct dentry *new; | |
1460 | ||
b6520c81 CH |
1461 | /* |
1462 | * First check if a dentry matching the name already exists, | |
1463 | * if not go ahead and create it now. | |
1464 | */ | |
9403540c | 1465 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1466 | if (!found) { |
1467 | new = d_alloc(dentry->d_parent, name); | |
1468 | if (!new) { | |
1469 | error = -ENOMEM; | |
1470 | goto err_out; | |
1471 | } | |
b6520c81 | 1472 | |
9403540c BN |
1473 | found = d_splice_alias(inode, new); |
1474 | if (found) { | |
1475 | dput(new); | |
1476 | return found; | |
1477 | } | |
1478 | return new; | |
1479 | } | |
b6520c81 CH |
1480 | |
1481 | /* | |
1482 | * If a matching dentry exists, and it's not negative use it. | |
1483 | * | |
1484 | * Decrement the reference count to balance the iget() done | |
1485 | * earlier on. | |
1486 | */ | |
9403540c BN |
1487 | if (found->d_inode) { |
1488 | if (unlikely(found->d_inode != inode)) { | |
1489 | /* This can't happen because bad inodes are unhashed. */ | |
1490 | BUG_ON(!is_bad_inode(inode)); | |
1491 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1492 | } | |
9403540c BN |
1493 | iput(inode); |
1494 | return found; | |
1495 | } | |
b6520c81 | 1496 | |
9403540c BN |
1497 | /* |
1498 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1499 | * already has a dentry. |
9403540c | 1500 | */ |
9403540c | 1501 | spin_lock(&dcache_lock); |
b6520c81 | 1502 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1503 | __d_instantiate(found, inode); |
9403540c BN |
1504 | spin_unlock(&dcache_lock); |
1505 | security_d_instantiate(found, inode); | |
1506 | return found; | |
1507 | } | |
b6520c81 | 1508 | |
9403540c | 1509 | /* |
b6520c81 CH |
1510 | * In case a directory already has a (disconnected) entry grab a |
1511 | * reference to it, move it in place and use it. | |
9403540c BN |
1512 | */ |
1513 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
1514 | dget_locked(new); | |
1515 | spin_unlock(&dcache_lock); | |
9403540c | 1516 | security_d_instantiate(found, inode); |
9403540c | 1517 | d_move(new, found); |
9403540c | 1518 | iput(inode); |
9403540c | 1519 | dput(found); |
9403540c BN |
1520 | return new; |
1521 | ||
1522 | err_out: | |
1523 | iput(inode); | |
1524 | return ERR_PTR(error); | |
1525 | } | |
ec4f8605 | 1526 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 LT |
1527 | |
1528 | /** | |
1529 | * d_lookup - search for a dentry | |
1530 | * @parent: parent dentry | |
1531 | * @name: qstr of name we wish to find | |
b04f784e | 1532 | * Returns: dentry, or NULL |
1da177e4 | 1533 | * |
b04f784e NP |
1534 | * d_lookup searches the children of the parent dentry for the name in |
1535 | * question. If the dentry is found its reference count is incremented and the | |
1536 | * dentry is returned. The caller must use dput to free the entry when it has | |
1537 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1538 | */ |
1da177e4 LT |
1539 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) |
1540 | { | |
1541 | struct dentry * dentry = NULL; | |
1542 | unsigned long seq; | |
1543 | ||
1544 | do { | |
1545 | seq = read_seqbegin(&rename_lock); | |
1546 | dentry = __d_lookup(parent, name); | |
1547 | if (dentry) | |
1548 | break; | |
1549 | } while (read_seqretry(&rename_lock, seq)); | |
1550 | return dentry; | |
1551 | } | |
ec4f8605 | 1552 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1553 | |
b04f784e NP |
1554 | /* |
1555 | * __d_lookup - search for a dentry (racy) | |
1556 | * @parent: parent dentry | |
1557 | * @name: qstr of name we wish to find | |
1558 | * Returns: dentry, or NULL | |
1559 | * | |
1560 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1561 | * false-negative result due to unrelated rename activity. | |
1562 | * | |
1563 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1564 | * however it must be used carefully, eg. with a following d_lookup in | |
1565 | * the case of failure. | |
1566 | * | |
1567 | * __d_lookup callers must be commented. | |
1568 | */ | |
1da177e4 LT |
1569 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) |
1570 | { | |
1571 | unsigned int len = name->len; | |
1572 | unsigned int hash = name->hash; | |
1573 | const unsigned char *str = name->name; | |
1574 | struct hlist_head *head = d_hash(parent,hash); | |
1575 | struct dentry *found = NULL; | |
1576 | struct hlist_node *node; | |
665a7583 | 1577 | struct dentry *dentry; |
1da177e4 | 1578 | |
b04f784e NP |
1579 | /* |
1580 | * The hash list is protected using RCU. | |
1581 | * | |
1582 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1583 | * with d_move(). | |
1584 | * | |
1585 | * It is possible that concurrent renames can mess up our list | |
1586 | * walk here and result in missing our dentry, resulting in the | |
1587 | * false-negative result. d_lookup() protects against concurrent | |
1588 | * renames using rename_lock seqlock. | |
1589 | * | |
1590 | * See Documentation/vfs/dcache-locking.txt for more details. | |
1591 | */ | |
1da177e4 LT |
1592 | rcu_read_lock(); |
1593 | ||
665a7583 | 1594 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1595 | struct qstr *qstr; |
1596 | ||
1da177e4 LT |
1597 | if (dentry->d_name.hash != hash) |
1598 | continue; | |
1599 | if (dentry->d_parent != parent) | |
1600 | continue; | |
1601 | ||
1602 | spin_lock(&dentry->d_lock); | |
1603 | ||
1604 | /* | |
1605 | * Recheck the dentry after taking the lock - d_move may have | |
b04f784e NP |
1606 | * changed things. Don't bother checking the hash because |
1607 | * we're about to compare the whole name anyway. | |
1da177e4 LT |
1608 | */ |
1609 | if (dentry->d_parent != parent) | |
1610 | goto next; | |
1611 | ||
d0185c08 LT |
1612 | /* non-existing due to RCU? */ |
1613 | if (d_unhashed(dentry)) | |
1614 | goto next; | |
1615 | ||
1da177e4 LT |
1616 | /* |
1617 | * It is safe to compare names since d_move() cannot | |
1618 | * change the qstr (protected by d_lock). | |
1619 | */ | |
1620 | qstr = &dentry->d_name; | |
1621 | if (parent->d_op && parent->d_op->d_compare) { | |
621e155a NP |
1622 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1623 | dentry, dentry->d_inode, | |
1624 | qstr->len, qstr->name, name)) | |
1da177e4 LT |
1625 | goto next; |
1626 | } else { | |
1627 | if (qstr->len != len) | |
1628 | goto next; | |
1629 | if (memcmp(qstr->name, str, len)) | |
1630 | goto next; | |
1631 | } | |
1632 | ||
b7ab39f6 | 1633 | dentry->d_count++; |
d0185c08 | 1634 | found = dentry; |
1da177e4 LT |
1635 | spin_unlock(&dentry->d_lock); |
1636 | break; | |
1637 | next: | |
1638 | spin_unlock(&dentry->d_lock); | |
1639 | } | |
1640 | rcu_read_unlock(); | |
1641 | ||
1642 | return found; | |
1643 | } | |
1644 | ||
3e7e241f EB |
1645 | /** |
1646 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1647 | * @dir: Directory to search in | |
1648 | * @name: qstr of name we wish to find | |
1649 | * | |
1650 | * On hash failure or on lookup failure NULL is returned. | |
1651 | */ | |
1652 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1653 | { | |
1654 | struct dentry *dentry = NULL; | |
1655 | ||
1656 | /* | |
1657 | * Check for a fs-specific hash function. Note that we must | |
1658 | * calculate the standard hash first, as the d_op->d_hash() | |
1659 | * routine may choose to leave the hash value unchanged. | |
1660 | */ | |
1661 | name->hash = full_name_hash(name->name, name->len); | |
1662 | if (dir->d_op && dir->d_op->d_hash) { | |
b1e6a015 | 1663 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1664 | goto out; |
1665 | } | |
1666 | dentry = d_lookup(dir, name); | |
1667 | out: | |
1668 | return dentry; | |
1669 | } | |
1670 | ||
1da177e4 | 1671 | /** |
786a5e15 | 1672 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 LT |
1673 | * @dentry: The dentry alleged to be valid child of @dparent |
1674 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1675 | * |
1676 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1677 | * This is used by ncpfs in its readdir implementation. | |
1678 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1679 | * |
1680 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1681 | */ |
d3a23e16 | 1682 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1683 | { |
786a5e15 | 1684 | struct dentry *child; |
d3a23e16 NP |
1685 | |
1686 | spin_lock(&dcache_lock); | |
786a5e15 NP |
1687 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
1688 | if (dentry == child) { | |
d3a23e16 NP |
1689 | __dget_locked(dentry); |
1690 | spin_unlock(&dcache_lock); | |
1da177e4 LT |
1691 | return 1; |
1692 | } | |
1693 | } | |
d3a23e16 | 1694 | spin_unlock(&dcache_lock); |
786a5e15 | 1695 | |
1da177e4 LT |
1696 | return 0; |
1697 | } | |
ec4f8605 | 1698 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1699 | |
1700 | /* | |
1701 | * When a file is deleted, we have two options: | |
1702 | * - turn this dentry into a negative dentry | |
1703 | * - unhash this dentry and free it. | |
1704 | * | |
1705 | * Usually, we want to just turn this into | |
1706 | * a negative dentry, but if anybody else is | |
1707 | * currently using the dentry or the inode | |
1708 | * we can't do that and we fall back on removing | |
1709 | * it from the hash queues and waiting for | |
1710 | * it to be deleted later when it has no users | |
1711 | */ | |
1712 | ||
1713 | /** | |
1714 | * d_delete - delete a dentry | |
1715 | * @dentry: The dentry to delete | |
1716 | * | |
1717 | * Turn the dentry into a negative dentry if possible, otherwise | |
1718 | * remove it from the hash queues so it can be deleted later | |
1719 | */ | |
1720 | ||
1721 | void d_delete(struct dentry * dentry) | |
1722 | { | |
7a91bf7f | 1723 | int isdir = 0; |
1da177e4 LT |
1724 | /* |
1725 | * Are we the only user? | |
1726 | */ | |
1727 | spin_lock(&dcache_lock); | |
1728 | spin_lock(&dentry->d_lock); | |
7a91bf7f | 1729 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
b7ab39f6 | 1730 | if (dentry->d_count == 1) { |
13e3c5e5 | 1731 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
1da177e4 | 1732 | dentry_iput(dentry); |
7a91bf7f | 1733 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1734 | return; |
1735 | } | |
1736 | ||
1737 | if (!d_unhashed(dentry)) | |
1738 | __d_drop(dentry); | |
1739 | ||
1740 | spin_unlock(&dentry->d_lock); | |
1741 | spin_unlock(&dcache_lock); | |
7a91bf7f JM |
1742 | |
1743 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1744 | } |
ec4f8605 | 1745 | EXPORT_SYMBOL(d_delete); |
1da177e4 LT |
1746 | |
1747 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1748 | { | |
1749 | ||
1750 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1751 | hlist_add_head_rcu(&entry->d_hash, list); | |
1752 | } | |
1753 | ||
770bfad8 DH |
1754 | static void _d_rehash(struct dentry * entry) |
1755 | { | |
1756 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1757 | } | |
1758 | ||
1da177e4 LT |
1759 | /** |
1760 | * d_rehash - add an entry back to the hash | |
1761 | * @entry: dentry to add to the hash | |
1762 | * | |
1763 | * Adds a dentry to the hash according to its name. | |
1764 | */ | |
1765 | ||
1766 | void d_rehash(struct dentry * entry) | |
1767 | { | |
1da177e4 LT |
1768 | spin_lock(&dcache_lock); |
1769 | spin_lock(&entry->d_lock); | |
789680d1 | 1770 | spin_lock(&dcache_hash_lock); |
770bfad8 | 1771 | _d_rehash(entry); |
789680d1 | 1772 | spin_unlock(&dcache_hash_lock); |
1da177e4 LT |
1773 | spin_unlock(&entry->d_lock); |
1774 | spin_unlock(&dcache_lock); | |
1775 | } | |
ec4f8605 | 1776 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 1777 | |
fb2d5b86 NP |
1778 | /** |
1779 | * dentry_update_name_case - update case insensitive dentry with a new name | |
1780 | * @dentry: dentry to be updated | |
1781 | * @name: new name | |
1782 | * | |
1783 | * Update a case insensitive dentry with new case of name. | |
1784 | * | |
1785 | * dentry must have been returned by d_lookup with name @name. Old and new | |
1786 | * name lengths must match (ie. no d_compare which allows mismatched name | |
1787 | * lengths). | |
1788 | * | |
1789 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
1790 | * keep renames and concurrent inserts, and readdir(2) away). | |
1791 | */ | |
1792 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
1793 | { | |
1794 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
1795 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | |
1796 | ||
1797 | spin_lock(&dcache_lock); | |
1798 | spin_lock(&dentry->d_lock); | |
1799 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); | |
1800 | spin_unlock(&dentry->d_lock); | |
1801 | spin_unlock(&dcache_lock); | |
1802 | } | |
1803 | EXPORT_SYMBOL(dentry_update_name_case); | |
1804 | ||
1da177e4 LT |
1805 | /* |
1806 | * When switching names, the actual string doesn't strictly have to | |
1807 | * be preserved in the target - because we're dropping the target | |
1808 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1809 | * the new name before we switch. | |
1810 | * | |
1811 | * Note that we have to be a lot more careful about getting the hash | |
1812 | * switched - we have to switch the hash value properly even if it | |
1813 | * then no longer matches the actual (corrupted) string of the target. | |
1814 | * The hash value has to match the hash queue that the dentry is on.. | |
1815 | */ | |
1816 | static void switch_names(struct dentry *dentry, struct dentry *target) | |
1817 | { | |
1818 | if (dname_external(target)) { | |
1819 | if (dname_external(dentry)) { | |
1820 | /* | |
1821 | * Both external: swap the pointers | |
1822 | */ | |
9a8d5bb4 | 1823 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
1824 | } else { |
1825 | /* | |
1826 | * dentry:internal, target:external. Steal target's | |
1827 | * storage and make target internal. | |
1828 | */ | |
321bcf92 BF |
1829 | memcpy(target->d_iname, dentry->d_name.name, |
1830 | dentry->d_name.len + 1); | |
1da177e4 LT |
1831 | dentry->d_name.name = target->d_name.name; |
1832 | target->d_name.name = target->d_iname; | |
1833 | } | |
1834 | } else { | |
1835 | if (dname_external(dentry)) { | |
1836 | /* | |
1837 | * dentry:external, target:internal. Give dentry's | |
1838 | * storage to target and make dentry internal | |
1839 | */ | |
1840 | memcpy(dentry->d_iname, target->d_name.name, | |
1841 | target->d_name.len + 1); | |
1842 | target->d_name.name = dentry->d_name.name; | |
1843 | dentry->d_name.name = dentry->d_iname; | |
1844 | } else { | |
1845 | /* | |
1846 | * Both are internal. Just copy target to dentry | |
1847 | */ | |
1848 | memcpy(dentry->d_iname, target->d_name.name, | |
1849 | target->d_name.len + 1); | |
dc711ca3 AV |
1850 | dentry->d_name.len = target->d_name.len; |
1851 | return; | |
1da177e4 LT |
1852 | } |
1853 | } | |
9a8d5bb4 | 1854 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
1855 | } |
1856 | ||
1857 | /* | |
1858 | * We cannibalize "target" when moving dentry on top of it, | |
1859 | * because it's going to be thrown away anyway. We could be more | |
1860 | * polite about it, though. | |
1861 | * | |
1862 | * This forceful removal will result in ugly /proc output if | |
1863 | * somebody holds a file open that got deleted due to a rename. | |
1864 | * We could be nicer about the deleted file, and let it show | |
bc154b1e BF |
1865 | * up under the name it had before it was deleted rather than |
1866 | * under the original name of the file that was moved on top of it. | |
1da177e4 LT |
1867 | */ |
1868 | ||
9eaef27b TM |
1869 | /* |
1870 | * d_move_locked - move a dentry | |
1da177e4 LT |
1871 | * @dentry: entry to move |
1872 | * @target: new dentry | |
1873 | * | |
1874 | * Update the dcache to reflect the move of a file name. Negative | |
1875 | * dcache entries should not be moved in this way. | |
1876 | */ | |
9eaef27b | 1877 | static void d_move_locked(struct dentry * dentry, struct dentry * target) |
1da177e4 | 1878 | { |
1da177e4 LT |
1879 | if (!dentry->d_inode) |
1880 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1881 | ||
1da177e4 LT |
1882 | write_seqlock(&rename_lock); |
1883 | /* | |
1884 | * XXXX: do we really need to take target->d_lock? | |
1885 | */ | |
da502956 NP |
1886 | if (d_ancestor(dentry, target)) { |
1887 | spin_lock(&dentry->d_lock); | |
1888 | spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED); | |
1889 | } else if (d_ancestor(target, dentry) || target < dentry) { | |
1da177e4 | 1890 | spin_lock(&target->d_lock); |
a90b9c05 | 1891 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1892 | } else { |
1893 | spin_lock(&dentry->d_lock); | |
a90b9c05 | 1894 | spin_lock_nested(&target->d_lock, DENTRY_D_LOCK_NESTED); |
1da177e4 LT |
1895 | } |
1896 | ||
1897 | /* Move the dentry to the target hash queue, if on different bucket */ | |
789680d1 NP |
1898 | spin_lock(&dcache_hash_lock); |
1899 | if (!d_unhashed(dentry)) | |
1900 | hlist_del_rcu(&dentry->d_hash); | |
1901 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); | |
1902 | spin_unlock(&dcache_hash_lock); | |
1da177e4 LT |
1903 | |
1904 | /* Unhash the target: dput() will then get rid of it */ | |
1905 | __d_drop(target); | |
1906 | ||
5160ee6f ED |
1907 | list_del(&dentry->d_u.d_child); |
1908 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
1909 | |
1910 | /* Switch the names.. */ | |
1911 | switch_names(dentry, target); | |
9a8d5bb4 | 1912 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
1913 | |
1914 | /* ... and switch the parents */ | |
1915 | if (IS_ROOT(dentry)) { | |
1916 | dentry->d_parent = target->d_parent; | |
1917 | target->d_parent = target; | |
5160ee6f | 1918 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 1919 | } else { |
9a8d5bb4 | 1920 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
1921 | |
1922 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 1923 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
1924 | } |
1925 | ||
5160ee6f | 1926 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
1da177e4 | 1927 | spin_unlock(&target->d_lock); |
c32ccd87 | 1928 | fsnotify_d_move(dentry); |
1da177e4 LT |
1929 | spin_unlock(&dentry->d_lock); |
1930 | write_sequnlock(&rename_lock); | |
9eaef27b TM |
1931 | } |
1932 | ||
1933 | /** | |
1934 | * d_move - move a dentry | |
1935 | * @dentry: entry to move | |
1936 | * @target: new dentry | |
1937 | * | |
1938 | * Update the dcache to reflect the move of a file name. Negative | |
1939 | * dcache entries should not be moved in this way. | |
1940 | */ | |
1941 | ||
1942 | void d_move(struct dentry * dentry, struct dentry * target) | |
1943 | { | |
1944 | spin_lock(&dcache_lock); | |
1945 | d_move_locked(dentry, target); | |
1da177e4 LT |
1946 | spin_unlock(&dcache_lock); |
1947 | } | |
ec4f8605 | 1948 | EXPORT_SYMBOL(d_move); |
1da177e4 | 1949 | |
e2761a11 OH |
1950 | /** |
1951 | * d_ancestor - search for an ancestor | |
1952 | * @p1: ancestor dentry | |
1953 | * @p2: child dentry | |
1954 | * | |
1955 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
1956 | * an ancestor of p2, else NULL. | |
9eaef27b | 1957 | */ |
e2761a11 | 1958 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
1959 | { |
1960 | struct dentry *p; | |
1961 | ||
871c0067 | 1962 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 1963 | if (p->d_parent == p1) |
e2761a11 | 1964 | return p; |
9eaef27b | 1965 | } |
e2761a11 | 1966 | return NULL; |
9eaef27b TM |
1967 | } |
1968 | ||
1969 | /* | |
1970 | * This helper attempts to cope with remotely renamed directories | |
1971 | * | |
1972 | * It assumes that the caller is already holding | |
1973 | * dentry->d_parent->d_inode->i_mutex and the dcache_lock | |
1974 | * | |
1975 | * Note: If ever the locking in lock_rename() changes, then please | |
1976 | * remember to update this too... | |
9eaef27b TM |
1977 | */ |
1978 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
31f3e0b3 | 1979 | __releases(dcache_lock) |
9eaef27b TM |
1980 | { |
1981 | struct mutex *m1 = NULL, *m2 = NULL; | |
1982 | struct dentry *ret; | |
1983 | ||
1984 | /* If alias and dentry share a parent, then no extra locks required */ | |
1985 | if (alias->d_parent == dentry->d_parent) | |
1986 | goto out_unalias; | |
1987 | ||
1988 | /* Check for loops */ | |
1989 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 1990 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
1991 | goto out_err; |
1992 | ||
1993 | /* See lock_rename() */ | |
1994 | ret = ERR_PTR(-EBUSY); | |
1995 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
1996 | goto out_err; | |
1997 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
1998 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
1999 | goto out_err; | |
2000 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2001 | out_unalias: | |
2002 | d_move_locked(alias, dentry); | |
2003 | ret = alias; | |
2004 | out_err: | |
2005 | spin_unlock(&dcache_lock); | |
2006 | if (m2) | |
2007 | mutex_unlock(m2); | |
2008 | if (m1) | |
2009 | mutex_unlock(m1); | |
2010 | return ret; | |
2011 | } | |
2012 | ||
770bfad8 DH |
2013 | /* |
2014 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2015 | * named dentry in place of the dentry to be replaced. | |
2016 | */ | |
2017 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2018 | { | |
2019 | struct dentry *dparent, *aparent; | |
2020 | ||
2021 | switch_names(dentry, anon); | |
9a8d5bb4 | 2022 | swap(dentry->d_name.hash, anon->d_name.hash); |
770bfad8 DH |
2023 | |
2024 | dparent = dentry->d_parent; | |
2025 | aparent = anon->d_parent; | |
2026 | ||
2027 | dentry->d_parent = (aparent == anon) ? dentry : aparent; | |
2028 | list_del(&dentry->d_u.d_child); | |
2029 | if (!IS_ROOT(dentry)) | |
2030 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2031 | else | |
2032 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2033 | ||
2034 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2035 | list_del(&anon->d_u.d_child); | |
2036 | if (!IS_ROOT(anon)) | |
2037 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2038 | else | |
2039 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2040 | ||
2041 | anon->d_flags &= ~DCACHE_DISCONNECTED; | |
2042 | } | |
2043 | ||
2044 | /** | |
2045 | * d_materialise_unique - introduce an inode into the tree | |
2046 | * @dentry: candidate dentry | |
2047 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2048 | * | |
2049 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2050 | * root directory alias in its place if there is one | |
2051 | */ | |
2052 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2053 | { | |
9eaef27b | 2054 | struct dentry *actual; |
770bfad8 DH |
2055 | |
2056 | BUG_ON(!d_unhashed(dentry)); | |
2057 | ||
2058 | spin_lock(&dcache_lock); | |
2059 | ||
2060 | if (!inode) { | |
2061 | actual = dentry; | |
360da900 | 2062 | __d_instantiate(dentry, NULL); |
770bfad8 DH |
2063 | goto found_lock; |
2064 | } | |
2065 | ||
9eaef27b TM |
2066 | if (S_ISDIR(inode->i_mode)) { |
2067 | struct dentry *alias; | |
2068 | ||
2069 | /* Does an aliased dentry already exist? */ | |
2070 | alias = __d_find_alias(inode, 0); | |
2071 | if (alias) { | |
2072 | actual = alias; | |
2073 | /* Is this an anonymous mountpoint that we could splice | |
2074 | * into our tree? */ | |
2075 | if (IS_ROOT(alias)) { | |
2076 | spin_lock(&alias->d_lock); | |
2077 | __d_materialise_dentry(dentry, alias); | |
2078 | __d_drop(alias); | |
2079 | goto found; | |
2080 | } | |
2081 | /* Nope, but we must(!) avoid directory aliasing */ | |
2082 | actual = __d_unalias(dentry, alias); | |
2083 | if (IS_ERR(actual)) | |
2084 | dput(alias); | |
2085 | goto out_nolock; | |
2086 | } | |
770bfad8 DH |
2087 | } |
2088 | ||
2089 | /* Add a unique reference */ | |
2090 | actual = __d_instantiate_unique(dentry, inode); | |
2091 | if (!actual) | |
2092 | actual = dentry; | |
2093 | else if (unlikely(!d_unhashed(actual))) | |
2094 | goto shouldnt_be_hashed; | |
2095 | ||
2096 | found_lock: | |
2097 | spin_lock(&actual->d_lock); | |
2098 | found: | |
789680d1 | 2099 | spin_lock(&dcache_hash_lock); |
770bfad8 | 2100 | _d_rehash(actual); |
789680d1 | 2101 | spin_unlock(&dcache_hash_lock); |
770bfad8 DH |
2102 | spin_unlock(&actual->d_lock); |
2103 | spin_unlock(&dcache_lock); | |
9eaef27b | 2104 | out_nolock: |
770bfad8 DH |
2105 | if (actual == dentry) { |
2106 | security_d_instantiate(dentry, inode); | |
2107 | return NULL; | |
2108 | } | |
2109 | ||
2110 | iput(inode); | |
2111 | return actual; | |
2112 | ||
770bfad8 DH |
2113 | shouldnt_be_hashed: |
2114 | spin_unlock(&dcache_lock); | |
2115 | BUG(); | |
770bfad8 | 2116 | } |
ec4f8605 | 2117 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2118 | |
cdd16d02 | 2119 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2120 | { |
2121 | *buflen -= namelen; | |
2122 | if (*buflen < 0) | |
2123 | return -ENAMETOOLONG; | |
2124 | *buffer -= namelen; | |
2125 | memcpy(*buffer, str, namelen); | |
2126 | return 0; | |
2127 | } | |
2128 | ||
cdd16d02 MS |
2129 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2130 | { | |
2131 | return prepend(buffer, buflen, name->name, name->len); | |
2132 | } | |
2133 | ||
1da177e4 | 2134 | /** |
f2eb6575 MS |
2135 | * Prepend path string to a buffer |
2136 | * | |
9d1bc601 MS |
2137 | * @path: the dentry/vfsmount to report |
2138 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2139 | * @buffer: pointer to the end of the buffer |
2140 | * @buflen: pointer to buffer length | |
552ce544 | 2141 | * |
f2eb6575 | 2142 | * Caller holds the dcache_lock. |
9d1bc601 MS |
2143 | * |
2144 | * If path is not reachable from the supplied root, then the value of | |
2145 | * root is changed (without modifying refcounts). | |
1da177e4 | 2146 | */ |
f2eb6575 MS |
2147 | static int prepend_path(const struct path *path, struct path *root, |
2148 | char **buffer, int *buflen) | |
1da177e4 | 2149 | { |
9d1bc601 MS |
2150 | struct dentry *dentry = path->dentry; |
2151 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2152 | bool slash = false; |
2153 | int error = 0; | |
6092d048 | 2154 | |
99b7db7b | 2155 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2156 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2157 | struct dentry * parent; |
2158 | ||
1da177e4 | 2159 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2160 | /* Global root? */ |
1da177e4 | 2161 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2162 | goto global_root; |
2163 | } | |
2164 | dentry = vfsmnt->mnt_mountpoint; | |
2165 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2166 | continue; |
2167 | } | |
2168 | parent = dentry->d_parent; | |
2169 | prefetch(parent); | |
f2eb6575 MS |
2170 | error = prepend_name(buffer, buflen, &dentry->d_name); |
2171 | if (!error) | |
2172 | error = prepend(buffer, buflen, "/", 1); | |
2173 | if (error) | |
2174 | break; | |
2175 | ||
2176 | slash = true; | |
1da177e4 LT |
2177 | dentry = parent; |
2178 | } | |
2179 | ||
be285c71 | 2180 | out: |
f2eb6575 MS |
2181 | if (!error && !slash) |
2182 | error = prepend(buffer, buflen, "/", 1); | |
2183 | ||
99b7db7b | 2184 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2185 | return error; |
1da177e4 LT |
2186 | |
2187 | global_root: | |
98dc568b MS |
2188 | /* |
2189 | * Filesystems needing to implement special "root names" | |
2190 | * should do so with ->d_dname() | |
2191 | */ | |
2192 | if (IS_ROOT(dentry) && | |
2193 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2194 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2195 | (int) dentry->d_name.len, dentry->d_name.name); | |
2196 | } | |
9d1bc601 MS |
2197 | root->mnt = vfsmnt; |
2198 | root->dentry = dentry; | |
be285c71 | 2199 | goto out; |
f2eb6575 | 2200 | } |
be285c71 | 2201 | |
f2eb6575 MS |
2202 | /** |
2203 | * __d_path - return the path of a dentry | |
2204 | * @path: the dentry/vfsmount to report | |
2205 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2206 | * @buf: buffer to return value in |
f2eb6575 MS |
2207 | * @buflen: buffer length |
2208 | * | |
ffd1f4ed | 2209 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2210 | * |
2211 | * Returns a pointer into the buffer or an error code if the | |
2212 | * path was too long. | |
2213 | * | |
be148247 | 2214 | * "buflen" should be positive. |
f2eb6575 MS |
2215 | * |
2216 | * If path is not reachable from the supplied root, then the value of | |
2217 | * root is changed (without modifying refcounts). | |
2218 | */ | |
2219 | char *__d_path(const struct path *path, struct path *root, | |
2220 | char *buf, int buflen) | |
2221 | { | |
2222 | char *res = buf + buflen; | |
2223 | int error; | |
2224 | ||
2225 | prepend(&res, &buflen, "\0", 1); | |
be148247 | 2226 | spin_lock(&dcache_lock); |
f2eb6575 | 2227 | error = prepend_path(path, root, &res, &buflen); |
be148247 CH |
2228 | spin_unlock(&dcache_lock); |
2229 | ||
f2eb6575 MS |
2230 | if (error) |
2231 | return ERR_PTR(error); | |
f2eb6575 | 2232 | return res; |
1da177e4 LT |
2233 | } |
2234 | ||
ffd1f4ed MS |
2235 | /* |
2236 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2237 | */ | |
2238 | static int path_with_deleted(const struct path *path, struct path *root, | |
2239 | char **buf, int *buflen) | |
2240 | { | |
2241 | prepend(buf, buflen, "\0", 1); | |
2242 | if (d_unlinked(path->dentry)) { | |
2243 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2244 | if (error) | |
2245 | return error; | |
2246 | } | |
2247 | ||
2248 | return prepend_path(path, root, buf, buflen); | |
2249 | } | |
2250 | ||
8df9d1a4 MS |
2251 | static int prepend_unreachable(char **buffer, int *buflen) |
2252 | { | |
2253 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2254 | } | |
2255 | ||
a03a8a70 JB |
2256 | /** |
2257 | * d_path - return the path of a dentry | |
cf28b486 | 2258 | * @path: path to report |
a03a8a70 JB |
2259 | * @buf: buffer to return value in |
2260 | * @buflen: buffer length | |
2261 | * | |
2262 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2263 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2264 | * | |
52afeefb AV |
2265 | * Returns a pointer into the buffer or an error code if the path was |
2266 | * too long. Note: Callers should use the returned pointer, not the passed | |
2267 | * in buffer, to use the name! The implementation often starts at an offset | |
2268 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2269 | * |
31f3e0b3 | 2270 | * "buflen" should be positive. |
a03a8a70 | 2271 | */ |
20d4fdc1 | 2272 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2273 | { |
ffd1f4ed | 2274 | char *res = buf + buflen; |
6ac08c39 | 2275 | struct path root; |
9d1bc601 | 2276 | struct path tmp; |
ffd1f4ed | 2277 | int error; |
1da177e4 | 2278 | |
c23fbb6b ED |
2279 | /* |
2280 | * We have various synthetic filesystems that never get mounted. On | |
2281 | * these filesystems dentries are never used for lookup purposes, and | |
2282 | * thus don't need to be hashed. They also don't need a name until a | |
2283 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2284 | * below allows us to generate a name for these objects on demand: | |
2285 | */ | |
cf28b486 JB |
2286 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2287 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2288 | |
f7ad3c6b | 2289 | get_fs_root(current->fs, &root); |
552ce544 | 2290 | spin_lock(&dcache_lock); |
9d1bc601 | 2291 | tmp = root; |
ffd1f4ed MS |
2292 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2293 | if (error) | |
2294 | res = ERR_PTR(error); | |
552ce544 | 2295 | spin_unlock(&dcache_lock); |
6ac08c39 | 2296 | path_put(&root); |
1da177e4 LT |
2297 | return res; |
2298 | } | |
ec4f8605 | 2299 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2300 | |
8df9d1a4 MS |
2301 | /** |
2302 | * d_path_with_unreachable - return the path of a dentry | |
2303 | * @path: path to report | |
2304 | * @buf: buffer to return value in | |
2305 | * @buflen: buffer length | |
2306 | * | |
2307 | * The difference from d_path() is that this prepends "(unreachable)" | |
2308 | * to paths which are unreachable from the current process' root. | |
2309 | */ | |
2310 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2311 | { | |
2312 | char *res = buf + buflen; | |
2313 | struct path root; | |
2314 | struct path tmp; | |
2315 | int error; | |
2316 | ||
2317 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2318 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2319 | ||
2320 | get_fs_root(current->fs, &root); | |
2321 | spin_lock(&dcache_lock); | |
2322 | tmp = root; | |
2323 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2324 | if (!error && !path_equal(&tmp, &root)) | |
2325 | error = prepend_unreachable(&res, &buflen); | |
2326 | spin_unlock(&dcache_lock); | |
2327 | path_put(&root); | |
2328 | if (error) | |
2329 | res = ERR_PTR(error); | |
2330 | ||
2331 | return res; | |
2332 | } | |
2333 | ||
c23fbb6b ED |
2334 | /* |
2335 | * Helper function for dentry_operations.d_dname() members | |
2336 | */ | |
2337 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2338 | const char *fmt, ...) | |
2339 | { | |
2340 | va_list args; | |
2341 | char temp[64]; | |
2342 | int sz; | |
2343 | ||
2344 | va_start(args, fmt); | |
2345 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2346 | va_end(args); | |
2347 | ||
2348 | if (sz > sizeof(temp) || sz > buflen) | |
2349 | return ERR_PTR(-ENAMETOOLONG); | |
2350 | ||
2351 | buffer += buflen - sz; | |
2352 | return memcpy(buffer, temp, sz); | |
2353 | } | |
2354 | ||
6092d048 RP |
2355 | /* |
2356 | * Write full pathname from the root of the filesystem into the buffer. | |
2357 | */ | |
ec2447c2 | 2358 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2359 | { |
2360 | char *end = buf + buflen; | |
2361 | char *retval; | |
2362 | ||
6092d048 | 2363 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2364 | if (buflen < 1) |
2365 | goto Elong; | |
2366 | /* Get '/' right */ | |
2367 | retval = end-1; | |
2368 | *retval = '/'; | |
2369 | ||
cdd16d02 MS |
2370 | while (!IS_ROOT(dentry)) { |
2371 | struct dentry *parent = dentry->d_parent; | |
6092d048 | 2372 | |
6092d048 | 2373 | prefetch(parent); |
cdd16d02 | 2374 | if ((prepend_name(&end, &buflen, &dentry->d_name) != 0) || |
6092d048 RP |
2375 | (prepend(&end, &buflen, "/", 1) != 0)) |
2376 | goto Elong; | |
2377 | ||
2378 | retval = end; | |
2379 | dentry = parent; | |
2380 | } | |
c103135c AV |
2381 | return retval; |
2382 | Elong: | |
2383 | return ERR_PTR(-ENAMETOOLONG); | |
2384 | } | |
ec2447c2 NP |
2385 | |
2386 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2387 | { | |
2388 | char *retval; | |
2389 | ||
2390 | spin_lock(&dcache_lock); | |
2391 | retval = __dentry_path(dentry, buf, buflen); | |
2392 | spin_unlock(&dcache_lock); | |
2393 | ||
2394 | return retval; | |
2395 | } | |
2396 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2397 | |
2398 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2399 | { | |
2400 | char *p = NULL; | |
2401 | char *retval; | |
2402 | ||
2403 | spin_lock(&dcache_lock); | |
2404 | if (d_unlinked(dentry)) { | |
2405 | p = buf + buflen; | |
2406 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2407 | goto Elong; | |
2408 | buflen++; | |
2409 | } | |
2410 | retval = __dentry_path(dentry, buf, buflen); | |
6092d048 | 2411 | spin_unlock(&dcache_lock); |
c103135c AV |
2412 | if (!IS_ERR(retval) && p) |
2413 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2414 | return retval; |
2415 | Elong: | |
2416 | spin_unlock(&dcache_lock); | |
2417 | return ERR_PTR(-ENAMETOOLONG); | |
2418 | } | |
2419 | ||
1da177e4 LT |
2420 | /* |
2421 | * NOTE! The user-level library version returns a | |
2422 | * character pointer. The kernel system call just | |
2423 | * returns the length of the buffer filled (which | |
2424 | * includes the ending '\0' character), or a negative | |
2425 | * error value. So libc would do something like | |
2426 | * | |
2427 | * char *getcwd(char * buf, size_t size) | |
2428 | * { | |
2429 | * int retval; | |
2430 | * | |
2431 | * retval = sys_getcwd(buf, size); | |
2432 | * if (retval >= 0) | |
2433 | * return buf; | |
2434 | * errno = -retval; | |
2435 | * return NULL; | |
2436 | * } | |
2437 | */ | |
3cdad428 | 2438 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2439 | { |
552ce544 | 2440 | int error; |
6ac08c39 | 2441 | struct path pwd, root; |
552ce544 | 2442 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2443 | |
2444 | if (!page) | |
2445 | return -ENOMEM; | |
2446 | ||
f7ad3c6b | 2447 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2448 | |
552ce544 | 2449 | error = -ENOENT; |
552ce544 | 2450 | spin_lock(&dcache_lock); |
f3da392e | 2451 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2452 | unsigned long len; |
9d1bc601 | 2453 | struct path tmp = root; |
8df9d1a4 MS |
2454 | char *cwd = page + PAGE_SIZE; |
2455 | int buflen = PAGE_SIZE; | |
1da177e4 | 2456 | |
8df9d1a4 MS |
2457 | prepend(&cwd, &buflen, "\0", 1); |
2458 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
552ce544 LT |
2459 | spin_unlock(&dcache_lock); |
2460 | ||
8df9d1a4 | 2461 | if (error) |
552ce544 LT |
2462 | goto out; |
2463 | ||
8df9d1a4 MS |
2464 | /* Unreachable from current root */ |
2465 | if (!path_equal(&tmp, &root)) { | |
2466 | error = prepend_unreachable(&cwd, &buflen); | |
2467 | if (error) | |
2468 | goto out; | |
2469 | } | |
2470 | ||
552ce544 LT |
2471 | error = -ERANGE; |
2472 | len = PAGE_SIZE + page - cwd; | |
2473 | if (len <= size) { | |
2474 | error = len; | |
2475 | if (copy_to_user(buf, cwd, len)) | |
2476 | error = -EFAULT; | |
2477 | } | |
2478 | } else | |
2479 | spin_unlock(&dcache_lock); | |
1da177e4 LT |
2480 | |
2481 | out: | |
6ac08c39 JB |
2482 | path_put(&pwd); |
2483 | path_put(&root); | |
1da177e4 LT |
2484 | free_page((unsigned long) page); |
2485 | return error; | |
2486 | } | |
2487 | ||
2488 | /* | |
2489 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2490 | * | |
2491 | * Trivially implemented using the dcache structure | |
2492 | */ | |
2493 | ||
2494 | /** | |
2495 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2496 | * @new_dentry: new dentry | |
2497 | * @old_dentry: old dentry | |
2498 | * | |
2499 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2500 | * Returns 0 otherwise. | |
2501 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2502 | */ | |
2503 | ||
e2761a11 | 2504 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2505 | { |
2506 | int result; | |
1da177e4 LT |
2507 | unsigned long seq; |
2508 | ||
e2761a11 OH |
2509 | if (new_dentry == old_dentry) |
2510 | return 1; | |
2511 | ||
2512 | /* | |
2513 | * Need rcu_readlock to protect against the d_parent trashing | |
2514 | * due to d_move | |
1da177e4 LT |
2515 | */ |
2516 | rcu_read_lock(); | |
e2761a11 | 2517 | do { |
1da177e4 | 2518 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2519 | seq = read_seqbegin(&rename_lock); |
e2761a11 | 2520 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2521 | result = 1; |
e2761a11 OH |
2522 | else |
2523 | result = 0; | |
1da177e4 LT |
2524 | } while (read_seqretry(&rename_lock, seq)); |
2525 | rcu_read_unlock(); | |
2526 | ||
2527 | return result; | |
2528 | } | |
2529 | ||
2096f759 AV |
2530 | int path_is_under(struct path *path1, struct path *path2) |
2531 | { | |
2532 | struct vfsmount *mnt = path1->mnt; | |
2533 | struct dentry *dentry = path1->dentry; | |
2534 | int res; | |
99b7db7b NP |
2535 | |
2536 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2537 | if (mnt != path2->mnt) { |
2538 | for (;;) { | |
2539 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2540 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2541 | return 0; |
2542 | } | |
2543 | if (mnt->mnt_parent == path2->mnt) | |
2544 | break; | |
2545 | mnt = mnt->mnt_parent; | |
2546 | } | |
2547 | dentry = mnt->mnt_mountpoint; | |
2548 | } | |
2549 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2550 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2551 | return res; |
2552 | } | |
2553 | EXPORT_SYMBOL(path_is_under); | |
2554 | ||
1da177e4 LT |
2555 | void d_genocide(struct dentry *root) |
2556 | { | |
2557 | struct dentry *this_parent = root; | |
2558 | struct list_head *next; | |
2559 | ||
2560 | spin_lock(&dcache_lock); | |
2561 | repeat: | |
2562 | next = this_parent->d_subdirs.next; | |
2563 | resume: | |
2564 | while (next != &this_parent->d_subdirs) { | |
2565 | struct list_head *tmp = next; | |
5160ee6f | 2566 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2567 | next = tmp->next; |
da502956 NP |
2568 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2569 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2570 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2571 | continue; |
da502956 | 2572 | } |
1da177e4 | 2573 | if (!list_empty(&dentry->d_subdirs)) { |
da502956 | 2574 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2575 | this_parent = dentry; |
2576 | goto repeat; | |
2577 | } | |
b7ab39f6 NP |
2578 | dentry->d_count--; |
2579 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2580 | } |
2581 | if (this_parent != root) { | |
5160ee6f | 2582 | next = this_parent->d_u.d_child.next; |
b7ab39f6 NP |
2583 | spin_lock(&this_parent->d_lock); |
2584 | this_parent->d_count--; | |
2585 | spin_unlock(&this_parent->d_lock); | |
1da177e4 LT |
2586 | this_parent = this_parent->d_parent; |
2587 | goto resume; | |
2588 | } | |
2589 | spin_unlock(&dcache_lock); | |
2590 | } | |
2591 | ||
2592 | /** | |
2593 | * find_inode_number - check for dentry with name | |
2594 | * @dir: directory to check | |
2595 | * @name: Name to find. | |
2596 | * | |
2597 | * Check whether a dentry already exists for the given name, | |
2598 | * and return the inode number if it has an inode. Otherwise | |
2599 | * 0 is returned. | |
2600 | * | |
2601 | * This routine is used to post-process directory listings for | |
2602 | * filesystems using synthetic inode numbers, and is necessary | |
2603 | * to keep getcwd() working. | |
2604 | */ | |
2605 | ||
2606 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2607 | { | |
2608 | struct dentry * dentry; | |
2609 | ino_t ino = 0; | |
2610 | ||
3e7e241f EB |
2611 | dentry = d_hash_and_lookup(dir, name); |
2612 | if (dentry) { | |
1da177e4 LT |
2613 | if (dentry->d_inode) |
2614 | ino = dentry->d_inode->i_ino; | |
2615 | dput(dentry); | |
2616 | } | |
1da177e4 LT |
2617 | return ino; |
2618 | } | |
ec4f8605 | 2619 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2620 | |
2621 | static __initdata unsigned long dhash_entries; | |
2622 | static int __init set_dhash_entries(char *str) | |
2623 | { | |
2624 | if (!str) | |
2625 | return 0; | |
2626 | dhash_entries = simple_strtoul(str, &str, 0); | |
2627 | return 1; | |
2628 | } | |
2629 | __setup("dhash_entries=", set_dhash_entries); | |
2630 | ||
2631 | static void __init dcache_init_early(void) | |
2632 | { | |
2633 | int loop; | |
2634 | ||
2635 | /* If hashes are distributed across NUMA nodes, defer | |
2636 | * hash allocation until vmalloc space is available. | |
2637 | */ | |
2638 | if (hashdist) | |
2639 | return; | |
2640 | ||
2641 | dentry_hashtable = | |
2642 | alloc_large_system_hash("Dentry cache", | |
2643 | sizeof(struct hlist_head), | |
2644 | dhash_entries, | |
2645 | 13, | |
2646 | HASH_EARLY, | |
2647 | &d_hash_shift, | |
2648 | &d_hash_mask, | |
2649 | 0); | |
2650 | ||
2651 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2652 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2653 | } | |
2654 | ||
74bf17cf | 2655 | static void __init dcache_init(void) |
1da177e4 LT |
2656 | { |
2657 | int loop; | |
2658 | ||
2659 | /* | |
2660 | * A constructor could be added for stable state like the lists, | |
2661 | * but it is probably not worth it because of the cache nature | |
2662 | * of the dcache. | |
2663 | */ | |
0a31bd5f CL |
2664 | dentry_cache = KMEM_CACHE(dentry, |
2665 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2666 | |
8e1f936b | 2667 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2668 | |
2669 | /* Hash may have been set up in dcache_init_early */ | |
2670 | if (!hashdist) | |
2671 | return; | |
2672 | ||
2673 | dentry_hashtable = | |
2674 | alloc_large_system_hash("Dentry cache", | |
2675 | sizeof(struct hlist_head), | |
2676 | dhash_entries, | |
2677 | 13, | |
2678 | 0, | |
2679 | &d_hash_shift, | |
2680 | &d_hash_mask, | |
2681 | 0); | |
2682 | ||
2683 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2684 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2685 | } | |
2686 | ||
2687 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2688 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2689 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2690 | |
1da177e4 LT |
2691 | EXPORT_SYMBOL(d_genocide); |
2692 | ||
1da177e4 LT |
2693 | void __init vfs_caches_init_early(void) |
2694 | { | |
2695 | dcache_init_early(); | |
2696 | inode_init_early(); | |
2697 | } | |
2698 | ||
2699 | void __init vfs_caches_init(unsigned long mempages) | |
2700 | { | |
2701 | unsigned long reserve; | |
2702 | ||
2703 | /* Base hash sizes on available memory, with a reserve equal to | |
2704 | 150% of current kernel size */ | |
2705 | ||
2706 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2707 | mempages -= reserve; | |
2708 | ||
2709 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2710 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2711 | |
74bf17cf DC |
2712 | dcache_init(); |
2713 | inode_init(); | |
1da177e4 | 2714 | files_init(mempages); |
74bf17cf | 2715 | mnt_init(); |
1da177e4 LT |
2716 | bdev_cache_init(); |
2717 | chrdev_init(); | |
2718 | } |