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