]>
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: | |
b23fb0a6 NP |
40 | * dcache_inode_lock protects: |
41 | * - i_dentry, d_alias, d_inode | |
23044507 NP |
42 | * dcache_hash_lock protects: |
43 | * - the dcache hash table, s_anon lists | |
44 | * dcache_lru_lock protects: | |
45 | * - the dcache lru lists and counters | |
46 | * d_lock protects: | |
47 | * - d_flags | |
48 | * - d_name | |
49 | * - d_lru | |
b7ab39f6 | 50 | * - d_count |
da502956 | 51 | * - d_unhashed() |
2fd6b7f5 NP |
52 | * - d_parent and d_subdirs |
53 | * - childrens' d_child and d_parent | |
b23fb0a6 | 54 | * - d_alias, d_inode |
789680d1 NP |
55 | * |
56 | * Ordering: | |
b5c84bf6 NP |
57 | * dcache_inode_lock |
58 | * dentry->d_lock | |
59 | * dcache_lru_lock | |
60 | * dcache_hash_lock | |
789680d1 | 61 | * |
da502956 NP |
62 | * If there is an ancestor relationship: |
63 | * dentry->d_parent->...->d_parent->d_lock | |
64 | * ... | |
65 | * dentry->d_parent->d_lock | |
66 | * dentry->d_lock | |
67 | * | |
68 | * If no ancestor relationship: | |
789680d1 NP |
69 | * if (dentry1 < dentry2) |
70 | * dentry1->d_lock | |
71 | * dentry2->d_lock | |
72 | */ | |
fa3536cc | 73 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
74 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
75 | ||
b23fb0a6 | 76 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_inode_lock); |
789680d1 | 77 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_hash_lock); |
23044507 | 78 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
74c3cbe3 | 79 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 80 | |
949854d0 | 81 | EXPORT_SYMBOL(rename_lock); |
b23fb0a6 | 82 | EXPORT_SYMBOL(dcache_inode_lock); |
1da177e4 | 83 | |
e18b890b | 84 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 LT |
85 | |
86 | #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname)) | |
87 | ||
88 | /* | |
89 | * This is the single most critical data structure when it comes | |
90 | * to the dcache: the hashtable for lookups. Somebody should try | |
91 | * to make this good - I've just made it work. | |
92 | * | |
93 | * This hash-function tries to avoid losing too many bits of hash | |
94 | * information, yet avoid using a prime hash-size or similar. | |
95 | */ | |
96 | #define D_HASHBITS d_hash_shift | |
97 | #define D_HASHMASK d_hash_mask | |
98 | ||
fa3536cc ED |
99 | static unsigned int d_hash_mask __read_mostly; |
100 | static unsigned int d_hash_shift __read_mostly; | |
101 | static struct hlist_head *dentry_hashtable __read_mostly; | |
1da177e4 LT |
102 | |
103 | /* Statistics gathering. */ | |
104 | struct dentry_stat_t dentry_stat = { | |
105 | .age_limit = 45, | |
106 | }; | |
107 | ||
3e880fb5 | 108 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
109 | |
110 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
111 | static int get_nr_dentry(void) |
112 | { | |
113 | int i; | |
114 | int sum = 0; | |
115 | for_each_possible_cpu(i) | |
116 | sum += per_cpu(nr_dentry, i); | |
117 | return sum < 0 ? 0 : sum; | |
118 | } | |
119 | ||
312d3ca8 CH |
120 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
121 | size_t *lenp, loff_t *ppos) | |
122 | { | |
3e880fb5 | 123 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
124 | return proc_dointvec(table, write, buffer, lenp, ppos); |
125 | } | |
126 | #endif | |
127 | ||
9c82ab9c | 128 | static void __d_free(struct rcu_head *head) |
1da177e4 | 129 | { |
9c82ab9c CH |
130 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
131 | ||
fd217f4d | 132 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
133 | if (dname_external(dentry)) |
134 | kfree(dentry->d_name.name); | |
135 | kmem_cache_free(dentry_cache, dentry); | |
136 | } | |
137 | ||
138 | /* | |
b5c84bf6 | 139 | * no locks, please. |
1da177e4 LT |
140 | */ |
141 | static void d_free(struct dentry *dentry) | |
142 | { | |
b7ab39f6 | 143 | BUG_ON(dentry->d_count); |
3e880fb5 | 144 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
145 | if (dentry->d_op && dentry->d_op->d_release) |
146 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 147 | |
b3423415 | 148 | /* if dentry was never inserted into hash, immediate free is OK */ |
e8462caa | 149 | if (hlist_unhashed(&dentry->d_hash)) |
9c82ab9c | 150 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 151 | else |
9c82ab9c | 152 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
153 | } |
154 | ||
155 | /* | |
156 | * Release the dentry's inode, using the filesystem | |
157 | * d_iput() operation if defined. | |
1da177e4 | 158 | */ |
858119e1 | 159 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 160 | __releases(dentry->d_lock) |
b23fb0a6 | 161 | __releases(dcache_inode_lock) |
1da177e4 LT |
162 | { |
163 | struct inode *inode = dentry->d_inode; | |
164 | if (inode) { | |
165 | dentry->d_inode = NULL; | |
166 | list_del_init(&dentry->d_alias); | |
167 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 168 | spin_unlock(&dcache_inode_lock); |
f805fbda LT |
169 | if (!inode->i_nlink) |
170 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
171 | if (dentry->d_op && dentry->d_op->d_iput) |
172 | dentry->d_op->d_iput(dentry, inode); | |
173 | else | |
174 | iput(inode); | |
175 | } else { | |
176 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 177 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
178 | } |
179 | } | |
180 | ||
da3bbdd4 | 181 | /* |
23044507 | 182 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
183 | */ |
184 | static void dentry_lru_add(struct dentry *dentry) | |
185 | { | |
a4633357 | 186 | if (list_empty(&dentry->d_lru)) { |
23044507 | 187 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
188 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
189 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 190 | dentry_stat.nr_unused++; |
23044507 | 191 | spin_unlock(&dcache_lru_lock); |
a4633357 | 192 | } |
da3bbdd4 KM |
193 | } |
194 | ||
23044507 NP |
195 | static void __dentry_lru_del(struct dentry *dentry) |
196 | { | |
197 | list_del_init(&dentry->d_lru); | |
198 | dentry->d_sb->s_nr_dentry_unused--; | |
199 | dentry_stat.nr_unused--; | |
200 | } | |
201 | ||
da3bbdd4 KM |
202 | static void dentry_lru_del(struct dentry *dentry) |
203 | { | |
204 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
205 | spin_lock(&dcache_lru_lock); |
206 | __dentry_lru_del(dentry); | |
207 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
208 | } |
209 | } | |
210 | ||
a4633357 | 211 | static void dentry_lru_move_tail(struct dentry *dentry) |
da3bbdd4 | 212 | { |
23044507 | 213 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
214 | if (list_empty(&dentry->d_lru)) { |
215 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
216 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 217 | dentry_stat.nr_unused++; |
a4633357 CH |
218 | } else { |
219 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
da3bbdd4 | 220 | } |
23044507 | 221 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
222 | } |
223 | ||
d52b9086 MS |
224 | /** |
225 | * d_kill - kill dentry and return parent | |
226 | * @dentry: dentry to kill | |
227 | * | |
31f3e0b3 | 228 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
229 | * |
230 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 231 | * |
b5c84bf6 NP |
232 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
233 | * d_kill. | |
d52b9086 | 234 | */ |
2fd6b7f5 | 235 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 236 | __releases(dentry->d_lock) |
2fd6b7f5 | 237 | __releases(parent->d_lock) |
b23fb0a6 | 238 | __releases(dcache_inode_lock) |
d52b9086 | 239 | { |
949854d0 | 240 | dentry->d_parent = NULL; |
d52b9086 | 241 | list_del(&dentry->d_u.d_child); |
2fd6b7f5 NP |
242 | if (parent) |
243 | spin_unlock(&parent->d_lock); | |
d52b9086 | 244 | dentry_iput(dentry); |
b7ab39f6 NP |
245 | /* |
246 | * dentry_iput drops the locks, at which point nobody (except | |
247 | * transient RCU lookups) can reach this dentry. | |
248 | */ | |
d52b9086 | 249 | d_free(dentry); |
871c0067 | 250 | return parent; |
d52b9086 MS |
251 | } |
252 | ||
789680d1 NP |
253 | /** |
254 | * d_drop - drop a dentry | |
255 | * @dentry: dentry to drop | |
256 | * | |
257 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
258 | * be found through a VFS lookup any more. Note that this is different from | |
259 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
260 | * possible, giving a successful _negative_ lookup, while d_drop will | |
261 | * just make the cache lookup fail. | |
262 | * | |
263 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
264 | * reason (NFS timeouts or autofs deletes). | |
265 | * | |
266 | * __d_drop requires dentry->d_lock. | |
267 | */ | |
268 | void __d_drop(struct dentry *dentry) | |
269 | { | |
270 | if (!(dentry->d_flags & DCACHE_UNHASHED)) { | |
271 | dentry->d_flags |= DCACHE_UNHASHED; | |
272 | spin_lock(&dcache_hash_lock); | |
273 | hlist_del_rcu(&dentry->d_hash); | |
274 | spin_unlock(&dcache_hash_lock); | |
275 | } | |
276 | } | |
277 | EXPORT_SYMBOL(__d_drop); | |
278 | ||
279 | void d_drop(struct dentry *dentry) | |
280 | { | |
789680d1 NP |
281 | spin_lock(&dentry->d_lock); |
282 | __d_drop(dentry); | |
283 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
284 | } |
285 | EXPORT_SYMBOL(d_drop); | |
286 | ||
1da177e4 LT |
287 | /* |
288 | * This is dput | |
289 | * | |
290 | * This is complicated by the fact that we do not want to put | |
291 | * dentries that are no longer on any hash chain on the unused | |
292 | * list: we'd much rather just get rid of them immediately. | |
293 | * | |
294 | * However, that implies that we have to traverse the dentry | |
295 | * tree upwards to the parents which might _also_ now be | |
296 | * scheduled for deletion (it may have been only waiting for | |
297 | * its last child to go away). | |
298 | * | |
299 | * This tail recursion is done by hand as we don't want to depend | |
300 | * on the compiler to always get this right (gcc generally doesn't). | |
301 | * Real recursion would eat up our stack space. | |
302 | */ | |
303 | ||
304 | /* | |
305 | * dput - release a dentry | |
306 | * @dentry: dentry to release | |
307 | * | |
308 | * Release a dentry. This will drop the usage count and if appropriate | |
309 | * call the dentry unlink method as well as removing it from the queues and | |
310 | * releasing its resources. If the parent dentries were scheduled for release | |
311 | * they too may now get deleted. | |
312 | * | |
313 | * no dcache lock, please. | |
314 | */ | |
315 | ||
316 | void dput(struct dentry *dentry) | |
317 | { | |
2fd6b7f5 | 318 | struct dentry *parent; |
1da177e4 LT |
319 | if (!dentry) |
320 | return; | |
321 | ||
322 | repeat: | |
b7ab39f6 | 323 | if (dentry->d_count == 1) |
1da177e4 | 324 | might_sleep(); |
1da177e4 | 325 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
326 | BUG_ON(!dentry->d_count); |
327 | if (dentry->d_count > 1) { | |
328 | dentry->d_count--; | |
1da177e4 | 329 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
330 | return; |
331 | } | |
332 | ||
1da177e4 LT |
333 | if (dentry->d_op && dentry->d_op->d_delete) { |
334 | if (dentry->d_op->d_delete(dentry)) | |
61f3dee4 | 335 | goto kill_it; |
1da177e4 | 336 | } |
265ac902 | 337 | |
1da177e4 LT |
338 | /* Unreachable? Get rid of it */ |
339 | if (d_unhashed(dentry)) | |
340 | goto kill_it; | |
265ac902 NP |
341 | |
342 | /* Otherwise leave it cached and ensure it's on the LRU */ | |
343 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 344 | dentry_lru_add(dentry); |
265ac902 | 345 | |
61f3dee4 NP |
346 | dentry->d_count--; |
347 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
348 | return; |
349 | ||
d52b9086 | 350 | kill_it: |
61f3dee4 NP |
351 | if (!spin_trylock(&dcache_inode_lock)) { |
352 | relock: | |
353 | spin_unlock(&dentry->d_lock); | |
354 | cpu_relax(); | |
355 | goto repeat; | |
356 | } | |
357 | if (IS_ROOT(dentry)) | |
358 | parent = NULL; | |
359 | else | |
360 | parent = dentry->d_parent; | |
361 | if (parent && !spin_trylock(&parent->d_lock)) { | |
362 | spin_unlock(&dcache_inode_lock); | |
363 | goto relock; | |
364 | } | |
365 | dentry->d_count--; | |
da3bbdd4 KM |
366 | /* if dentry was on the d_lru list delete it from there */ |
367 | dentry_lru_del(dentry); | |
61f3dee4 NP |
368 | /* if it was on the hash (d_delete case), then remove it */ |
369 | __d_drop(dentry); | |
2fd6b7f5 | 370 | dentry = d_kill(dentry, parent); |
d52b9086 MS |
371 | if (dentry) |
372 | goto repeat; | |
1da177e4 | 373 | } |
ec4f8605 | 374 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
375 | |
376 | /** | |
377 | * d_invalidate - invalidate a dentry | |
378 | * @dentry: dentry to invalidate | |
379 | * | |
380 | * Try to invalidate the dentry if it turns out to be | |
381 | * possible. If there are other dentries that can be | |
382 | * reached through this one we can't delete it and we | |
383 | * return -EBUSY. On success we return 0. | |
384 | * | |
385 | * no dcache lock. | |
386 | */ | |
387 | ||
388 | int d_invalidate(struct dentry * dentry) | |
389 | { | |
390 | /* | |
391 | * If it's already been dropped, return OK. | |
392 | */ | |
da502956 | 393 | spin_lock(&dentry->d_lock); |
1da177e4 | 394 | if (d_unhashed(dentry)) { |
da502956 | 395 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
396 | return 0; |
397 | } | |
398 | /* | |
399 | * Check whether to do a partial shrink_dcache | |
400 | * to get rid of unused child entries. | |
401 | */ | |
402 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 403 | spin_unlock(&dentry->d_lock); |
1da177e4 | 404 | shrink_dcache_parent(dentry); |
da502956 | 405 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
406 | } |
407 | ||
408 | /* | |
409 | * Somebody else still using it? | |
410 | * | |
411 | * If it's a directory, we can't drop it | |
412 | * for fear of somebody re-populating it | |
413 | * with children (even though dropping it | |
414 | * would make it unreachable from the root, | |
415 | * we might still populate it if it was a | |
416 | * working directory or similar). | |
417 | */ | |
b7ab39f6 | 418 | if (dentry->d_count > 1) { |
1da177e4 LT |
419 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
420 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
421 | return -EBUSY; |
422 | } | |
423 | } | |
424 | ||
425 | __d_drop(dentry); | |
426 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
427 | return 0; |
428 | } | |
ec4f8605 | 429 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 430 | |
b5c84bf6 | 431 | /* This must be called with d_lock held */ |
dc0474be | 432 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 433 | { |
b7ab39f6 | 434 | dentry->d_count++; |
23044507 NP |
435 | } |
436 | ||
dc0474be | 437 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 438 | { |
23044507 | 439 | spin_lock(&dentry->d_lock); |
dc0474be | 440 | __dget_dlock(dentry); |
23044507 | 441 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
442 | } |
443 | ||
b7ab39f6 NP |
444 | struct dentry *dget_parent(struct dentry *dentry) |
445 | { | |
446 | struct dentry *ret; | |
447 | ||
448 | repeat: | |
449 | spin_lock(&dentry->d_lock); | |
450 | ret = dentry->d_parent; | |
451 | if (!ret) | |
452 | goto out; | |
453 | if (dentry == ret) { | |
454 | ret->d_count++; | |
455 | goto out; | |
456 | } | |
457 | if (!spin_trylock(&ret->d_lock)) { | |
458 | spin_unlock(&dentry->d_lock); | |
459 | cpu_relax(); | |
460 | goto repeat; | |
461 | } | |
462 | BUG_ON(!ret->d_count); | |
463 | ret->d_count++; | |
464 | spin_unlock(&ret->d_lock); | |
465 | out: | |
466 | spin_unlock(&dentry->d_lock); | |
467 | return ret; | |
468 | } | |
469 | EXPORT_SYMBOL(dget_parent); | |
470 | ||
1da177e4 LT |
471 | /** |
472 | * d_find_alias - grab a hashed alias of inode | |
473 | * @inode: inode in question | |
474 | * @want_discon: flag, used by d_splice_alias, to request | |
475 | * that only a DISCONNECTED alias be returned. | |
476 | * | |
477 | * If inode has a hashed alias, or is a directory and has any alias, | |
478 | * acquire the reference to alias and return it. Otherwise return NULL. | |
479 | * Notice that if inode is a directory there can be only one alias and | |
480 | * it can be unhashed only if it has no children, or if it is the root | |
481 | * of a filesystem. | |
482 | * | |
21c0d8fd | 483 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 484 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 485 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 486 | */ |
da502956 | 487 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 488 | { |
da502956 | 489 | struct dentry *alias, *discon_alias; |
1da177e4 | 490 | |
da502956 NP |
491 | again: |
492 | discon_alias = NULL; | |
493 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
494 | spin_lock(&alias->d_lock); | |
1da177e4 | 495 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 496 | if (IS_ROOT(alias) && |
da502956 | 497 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 498 | discon_alias = alias; |
da502956 | 499 | } else if (!want_discon) { |
dc0474be | 500 | __dget_dlock(alias); |
da502956 NP |
501 | spin_unlock(&alias->d_lock); |
502 | return alias; | |
503 | } | |
504 | } | |
505 | spin_unlock(&alias->d_lock); | |
506 | } | |
507 | if (discon_alias) { | |
508 | alias = discon_alias; | |
509 | spin_lock(&alias->d_lock); | |
510 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
511 | if (IS_ROOT(alias) && | |
512 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 513 | __dget_dlock(alias); |
da502956 | 514 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
515 | return alias; |
516 | } | |
517 | } | |
da502956 NP |
518 | spin_unlock(&alias->d_lock); |
519 | goto again; | |
1da177e4 | 520 | } |
da502956 | 521 | return NULL; |
1da177e4 LT |
522 | } |
523 | ||
da502956 | 524 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 525 | { |
214fda1f DH |
526 | struct dentry *de = NULL; |
527 | ||
528 | if (!list_empty(&inode->i_dentry)) { | |
b23fb0a6 | 529 | spin_lock(&dcache_inode_lock); |
214fda1f | 530 | de = __d_find_alias(inode, 0); |
b23fb0a6 | 531 | spin_unlock(&dcache_inode_lock); |
214fda1f | 532 | } |
1da177e4 LT |
533 | return de; |
534 | } | |
ec4f8605 | 535 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
536 | |
537 | /* | |
538 | * Try to kill dentries associated with this inode. | |
539 | * WARNING: you must own a reference to inode. | |
540 | */ | |
541 | void d_prune_aliases(struct inode *inode) | |
542 | { | |
0cdca3f9 | 543 | struct dentry *dentry; |
1da177e4 | 544 | restart: |
b23fb0a6 | 545 | spin_lock(&dcache_inode_lock); |
0cdca3f9 | 546 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 547 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 548 | if (!dentry->d_count) { |
dc0474be | 549 | __dget_dlock(dentry); |
1da177e4 LT |
550 | __d_drop(dentry); |
551 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 552 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
553 | dput(dentry); |
554 | goto restart; | |
555 | } | |
556 | spin_unlock(&dentry->d_lock); | |
557 | } | |
b23fb0a6 | 558 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 559 | } |
ec4f8605 | 560 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
561 | |
562 | /* | |
d702ccb3 AM |
563 | * Throw away a dentry - free the inode, dput the parent. This requires that |
564 | * the LRU list has already been removed. | |
565 | * | |
85864e10 MS |
566 | * Try to prune ancestors as well. This is necessary to prevent |
567 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
568 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 569 | */ |
2fd6b7f5 | 570 | static void prune_one_dentry(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 571 | __releases(dentry->d_lock) |
2fd6b7f5 | 572 | __releases(parent->d_lock) |
b23fb0a6 | 573 | __releases(dcache_inode_lock) |
1da177e4 | 574 | { |
1da177e4 | 575 | __d_drop(dentry); |
2fd6b7f5 | 576 | dentry = d_kill(dentry, parent); |
d52b9086 MS |
577 | |
578 | /* | |
b5c84bf6 | 579 | * Prune ancestors. |
d52b9086 | 580 | */ |
d52b9086 | 581 | while (dentry) { |
b23fb0a6 | 582 | spin_lock(&dcache_inode_lock); |
2fd6b7f5 | 583 | again: |
b7ab39f6 | 584 | spin_lock(&dentry->d_lock); |
2fd6b7f5 NP |
585 | if (IS_ROOT(dentry)) |
586 | parent = NULL; | |
587 | else | |
588 | parent = dentry->d_parent; | |
589 | if (parent && !spin_trylock(&parent->d_lock)) { | |
590 | spin_unlock(&dentry->d_lock); | |
591 | goto again; | |
592 | } | |
b7ab39f6 NP |
593 | dentry->d_count--; |
594 | if (dentry->d_count) { | |
2fd6b7f5 NP |
595 | if (parent) |
596 | spin_unlock(&parent->d_lock); | |
b7ab39f6 | 597 | spin_unlock(&dentry->d_lock); |
b23fb0a6 | 598 | spin_unlock(&dcache_inode_lock); |
d52b9086 | 599 | return; |
23044507 | 600 | } |
d52b9086 | 601 | |
a4633357 | 602 | dentry_lru_del(dentry); |
d52b9086 | 603 | __d_drop(dentry); |
2fd6b7f5 | 604 | dentry = d_kill(dentry, parent); |
d52b9086 | 605 | } |
1da177e4 LT |
606 | } |
607 | ||
3049cfe2 | 608 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 609 | { |
da3bbdd4 | 610 | struct dentry *dentry; |
da3bbdd4 | 611 | |
3049cfe2 | 612 | while (!list_empty(list)) { |
2fd6b7f5 NP |
613 | struct dentry *parent; |
614 | ||
3049cfe2 | 615 | dentry = list_entry(list->prev, struct dentry, d_lru); |
23044507 NP |
616 | |
617 | if (!spin_trylock(&dentry->d_lock)) { | |
2fd6b7f5 | 618 | relock: |
23044507 NP |
619 | spin_unlock(&dcache_lru_lock); |
620 | cpu_relax(); | |
621 | spin_lock(&dcache_lru_lock); | |
622 | continue; | |
623 | } | |
624 | ||
1da177e4 LT |
625 | /* |
626 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
627 | * the LRU because of laziness during lookup. Do not free |
628 | * it - just keep it off the LRU list. | |
1da177e4 | 629 | */ |
b7ab39f6 | 630 | if (dentry->d_count) { |
2fd6b7f5 | 631 | __dentry_lru_del(dentry); |
da3bbdd4 | 632 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
633 | continue; |
634 | } | |
2fd6b7f5 NP |
635 | if (IS_ROOT(dentry)) |
636 | parent = NULL; | |
637 | else | |
638 | parent = dentry->d_parent; | |
639 | if (parent && !spin_trylock(&parent->d_lock)) { | |
640 | spin_unlock(&dentry->d_lock); | |
641 | goto relock; | |
642 | } | |
643 | __dentry_lru_del(dentry); | |
23044507 NP |
644 | spin_unlock(&dcache_lru_lock); |
645 | ||
2fd6b7f5 | 646 | prune_one_dentry(dentry, parent); |
b5c84bf6 | 647 | /* dcache_inode_lock and dentry->d_lock dropped */ |
b23fb0a6 | 648 | spin_lock(&dcache_inode_lock); |
23044507 | 649 | spin_lock(&dcache_lru_lock); |
da3bbdd4 | 650 | } |
3049cfe2 CH |
651 | } |
652 | ||
653 | /** | |
654 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
655 | * @sb: superblock to shrink dentry LRU. | |
656 | * @count: number of entries to prune | |
657 | * @flags: flags to control the dentry processing | |
658 | * | |
659 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
660 | */ | |
661 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
662 | { | |
663 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
664 | struct dentry *dentry; | |
665 | LIST_HEAD(referenced); | |
666 | LIST_HEAD(tmp); | |
667 | int cnt = *count; | |
668 | ||
b23fb0a6 | 669 | spin_lock(&dcache_inode_lock); |
23044507 NP |
670 | relock: |
671 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
672 | while (!list_empty(&sb->s_dentry_lru)) { |
673 | dentry = list_entry(sb->s_dentry_lru.prev, | |
674 | struct dentry, d_lru); | |
675 | BUG_ON(dentry->d_sb != sb); | |
676 | ||
23044507 NP |
677 | if (!spin_trylock(&dentry->d_lock)) { |
678 | spin_unlock(&dcache_lru_lock); | |
679 | cpu_relax(); | |
680 | goto relock; | |
681 | } | |
682 | ||
3049cfe2 CH |
683 | /* |
684 | * If we are honouring the DCACHE_REFERENCED flag and the | |
685 | * dentry has this flag set, don't free it. Clear the flag | |
686 | * and put it back on the LRU. | |
687 | */ | |
23044507 NP |
688 | if (flags & DCACHE_REFERENCED && |
689 | dentry->d_flags & DCACHE_REFERENCED) { | |
690 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
691 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 692 | spin_unlock(&dentry->d_lock); |
23044507 NP |
693 | } else { |
694 | list_move_tail(&dentry->d_lru, &tmp); | |
695 | spin_unlock(&dentry->d_lock); | |
696 | if (!--cnt) | |
697 | break; | |
3049cfe2 | 698 | } |
23044507 | 699 | /* XXX: re-add cond_resched_lock when dcache_lock goes away */ |
3049cfe2 CH |
700 | } |
701 | ||
702 | *count = cnt; | |
703 | shrink_dentry_list(&tmp); | |
704 | ||
da3bbdd4 KM |
705 | if (!list_empty(&referenced)) |
706 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 707 | spin_unlock(&dcache_lru_lock); |
b23fb0a6 | 708 | spin_unlock(&dcache_inode_lock); |
da3bbdd4 KM |
709 | } |
710 | ||
711 | /** | |
712 | * prune_dcache - shrink the dcache | |
713 | * @count: number of entries to try to free | |
714 | * | |
715 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
716 | * need to unmount something (at which point we need to unuse all dentries). | |
717 | * | |
718 | * This function may fail to free any resources if all the dentries are in use. | |
719 | */ | |
720 | static void prune_dcache(int count) | |
721 | { | |
dca33252 | 722 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 723 | int w_count; |
86c8749e | 724 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
725 | int prune_ratio; |
726 | int pruned; | |
727 | ||
728 | if (unused == 0 || count == 0) | |
729 | return; | |
da3bbdd4 KM |
730 | if (count >= unused) |
731 | prune_ratio = 1; | |
732 | else | |
733 | prune_ratio = unused / count; | |
734 | spin_lock(&sb_lock); | |
dca33252 | 735 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
736 | if (list_empty(&sb->s_instances)) |
737 | continue; | |
da3bbdd4 | 738 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 739 | continue; |
da3bbdd4 KM |
740 | sb->s_count++; |
741 | /* Now, we reclaim unused dentrins with fairness. | |
742 | * We reclaim them same percentage from each superblock. | |
743 | * We calculate number of dentries to scan on this sb | |
744 | * as follows, but the implementation is arranged to avoid | |
745 | * overflows: | |
746 | * number of dentries to scan on this sb = | |
747 | * count * (number of dentries on this sb / | |
748 | * number of dentries in the machine) | |
0feae5c4 | 749 | */ |
da3bbdd4 KM |
750 | spin_unlock(&sb_lock); |
751 | if (prune_ratio != 1) | |
752 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
753 | else | |
754 | w_count = sb->s_nr_dentry_unused; | |
755 | pruned = w_count; | |
0feae5c4 | 756 | /* |
da3bbdd4 KM |
757 | * We need to be sure this filesystem isn't being unmounted, |
758 | * otherwise we could race with generic_shutdown_super(), and | |
759 | * end up holding a reference to an inode while the filesystem | |
760 | * is unmounted. So we try to get s_umount, and make sure | |
761 | * s_root isn't NULL. | |
0feae5c4 | 762 | */ |
da3bbdd4 KM |
763 | if (down_read_trylock(&sb->s_umount)) { |
764 | if ((sb->s_root != NULL) && | |
765 | (!list_empty(&sb->s_dentry_lru))) { | |
da3bbdd4 KM |
766 | __shrink_dcache_sb(sb, &w_count, |
767 | DCACHE_REFERENCED); | |
768 | pruned -= w_count; | |
0feae5c4 | 769 | } |
da3bbdd4 | 770 | up_read(&sb->s_umount); |
0feae5c4 | 771 | } |
da3bbdd4 | 772 | spin_lock(&sb_lock); |
dca33252 AV |
773 | if (p) |
774 | __put_super(p); | |
da3bbdd4 | 775 | count -= pruned; |
dca33252 | 776 | p = sb; |
79893c17 AV |
777 | /* more work left to do? */ |
778 | if (count <= 0) | |
779 | break; | |
1da177e4 | 780 | } |
dca33252 AV |
781 | if (p) |
782 | __put_super(p); | |
da3bbdd4 | 783 | spin_unlock(&sb_lock); |
1da177e4 LT |
784 | } |
785 | ||
1da177e4 LT |
786 | /** |
787 | * shrink_dcache_sb - shrink dcache for a superblock | |
788 | * @sb: superblock | |
789 | * | |
3049cfe2 CH |
790 | * Shrink the dcache for the specified super block. This is used to free |
791 | * the dcache before unmounting a file system. | |
1da177e4 | 792 | */ |
3049cfe2 | 793 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 794 | { |
3049cfe2 CH |
795 | LIST_HEAD(tmp); |
796 | ||
b23fb0a6 | 797 | spin_lock(&dcache_inode_lock); |
23044507 | 798 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
799 | while (!list_empty(&sb->s_dentry_lru)) { |
800 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
801 | shrink_dentry_list(&tmp); | |
802 | } | |
23044507 | 803 | spin_unlock(&dcache_lru_lock); |
b23fb0a6 | 804 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 805 | } |
ec4f8605 | 806 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 807 | |
c636ebdb DH |
808 | /* |
809 | * destroy a single subtree of dentries for unmount | |
810 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
811 | * locking | |
812 | */ | |
813 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
814 | { | |
815 | struct dentry *parent; | |
f8713576 | 816 | unsigned detached = 0; |
c636ebdb DH |
817 | |
818 | BUG_ON(!IS_ROOT(dentry)); | |
819 | ||
820 | /* detach this root from the system */ | |
23044507 | 821 | spin_lock(&dentry->d_lock); |
a4633357 | 822 | dentry_lru_del(dentry); |
c636ebdb | 823 | __d_drop(dentry); |
da502956 | 824 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
825 | |
826 | for (;;) { | |
827 | /* descend to the first leaf in the current subtree */ | |
828 | while (!list_empty(&dentry->d_subdirs)) { | |
829 | struct dentry *loop; | |
830 | ||
831 | /* this is a branch with children - detach all of them | |
832 | * from the system in one go */ | |
2fd6b7f5 | 833 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
834 | list_for_each_entry(loop, &dentry->d_subdirs, |
835 | d_u.d_child) { | |
2fd6b7f5 NP |
836 | spin_lock_nested(&loop->d_lock, |
837 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 838 | dentry_lru_del(loop); |
c636ebdb | 839 | __d_drop(loop); |
da502956 | 840 | spin_unlock(&loop->d_lock); |
c636ebdb | 841 | } |
2fd6b7f5 | 842 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
843 | |
844 | /* move to the first child */ | |
845 | dentry = list_entry(dentry->d_subdirs.next, | |
846 | struct dentry, d_u.d_child); | |
847 | } | |
848 | ||
849 | /* consume the dentries from this leaf up through its parents | |
850 | * until we find one with children or run out altogether */ | |
851 | do { | |
852 | struct inode *inode; | |
853 | ||
b7ab39f6 | 854 | if (dentry->d_count != 0) { |
c636ebdb DH |
855 | printk(KERN_ERR |
856 | "BUG: Dentry %p{i=%lx,n=%s}" | |
857 | " still in use (%d)" | |
858 | " [unmount of %s %s]\n", | |
859 | dentry, | |
860 | dentry->d_inode ? | |
861 | dentry->d_inode->i_ino : 0UL, | |
862 | dentry->d_name.name, | |
b7ab39f6 | 863 | dentry->d_count, |
c636ebdb DH |
864 | dentry->d_sb->s_type->name, |
865 | dentry->d_sb->s_id); | |
866 | BUG(); | |
867 | } | |
868 | ||
2fd6b7f5 | 869 | if (IS_ROOT(dentry)) { |
c636ebdb | 870 | parent = NULL; |
2fd6b7f5 NP |
871 | list_del(&dentry->d_u.d_child); |
872 | } else { | |
871c0067 | 873 | parent = dentry->d_parent; |
b7ab39f6 NP |
874 | spin_lock(&parent->d_lock); |
875 | parent->d_count--; | |
2fd6b7f5 | 876 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 877 | spin_unlock(&parent->d_lock); |
871c0067 | 878 | } |
c636ebdb | 879 | |
f8713576 | 880 | detached++; |
c636ebdb DH |
881 | |
882 | inode = dentry->d_inode; | |
883 | if (inode) { | |
884 | dentry->d_inode = NULL; | |
885 | list_del_init(&dentry->d_alias); | |
886 | if (dentry->d_op && dentry->d_op->d_iput) | |
887 | dentry->d_op->d_iput(dentry, inode); | |
888 | else | |
889 | iput(inode); | |
890 | } | |
891 | ||
892 | d_free(dentry); | |
893 | ||
894 | /* finished when we fall off the top of the tree, | |
895 | * otherwise we ascend to the parent and move to the | |
896 | * next sibling if there is one */ | |
897 | if (!parent) | |
312d3ca8 | 898 | return; |
c636ebdb | 899 | dentry = parent; |
c636ebdb DH |
900 | } while (list_empty(&dentry->d_subdirs)); |
901 | ||
902 | dentry = list_entry(dentry->d_subdirs.next, | |
903 | struct dentry, d_u.d_child); | |
904 | } | |
905 | } | |
906 | ||
907 | /* | |
908 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 909 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
910 | * - the superblock is detached from all mountings and open files, so the |
911 | * dentry trees will not be rearranged by the VFS | |
912 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
913 | * any dentries belonging to this superblock that it comes across | |
914 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
915 | * in this superblock | |
916 | */ | |
917 | void shrink_dcache_for_umount(struct super_block *sb) | |
918 | { | |
919 | struct dentry *dentry; | |
920 | ||
921 | if (down_read_trylock(&sb->s_umount)) | |
922 | BUG(); | |
923 | ||
924 | dentry = sb->s_root; | |
925 | sb->s_root = NULL; | |
b7ab39f6 NP |
926 | spin_lock(&dentry->d_lock); |
927 | dentry->d_count--; | |
928 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
929 | shrink_dcache_for_umount_subtree(dentry); |
930 | ||
931 | while (!hlist_empty(&sb->s_anon)) { | |
932 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
933 | shrink_dcache_for_umount_subtree(dentry); | |
934 | } | |
935 | } | |
936 | ||
1da177e4 LT |
937 | /* |
938 | * Search for at least 1 mount point in the dentry's subdirs. | |
939 | * We descend to the next level whenever the d_subdirs | |
940 | * list is non-empty and continue searching. | |
941 | */ | |
942 | ||
943 | /** | |
944 | * have_submounts - check for mounts over a dentry | |
945 | * @parent: dentry to check. | |
946 | * | |
947 | * Return true if the parent or its subdirectories contain | |
948 | * a mount point | |
949 | */ | |
1da177e4 LT |
950 | int have_submounts(struct dentry *parent) |
951 | { | |
949854d0 | 952 | struct dentry *this_parent; |
1da177e4 | 953 | struct list_head *next; |
949854d0 | 954 | unsigned seq; |
58db63d0 | 955 | int locked = 0; |
949854d0 | 956 | |
949854d0 | 957 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
958 | again: |
959 | this_parent = parent; | |
1da177e4 | 960 | |
1da177e4 LT |
961 | if (d_mountpoint(parent)) |
962 | goto positive; | |
2fd6b7f5 | 963 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
964 | repeat: |
965 | next = this_parent->d_subdirs.next; | |
966 | resume: | |
967 | while (next != &this_parent->d_subdirs) { | |
968 | struct list_head *tmp = next; | |
5160ee6f | 969 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 970 | next = tmp->next; |
2fd6b7f5 NP |
971 | |
972 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 973 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
974 | if (d_mountpoint(dentry)) { |
975 | spin_unlock(&dentry->d_lock); | |
976 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 977 | goto positive; |
2fd6b7f5 | 978 | } |
1da177e4 | 979 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
980 | spin_unlock(&this_parent->d_lock); |
981 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 982 | this_parent = dentry; |
2fd6b7f5 | 983 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
984 | goto repeat; |
985 | } | |
2fd6b7f5 | 986 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
987 | } |
988 | /* | |
989 | * All done at this level ... ascend and resume the search. | |
990 | */ | |
991 | if (this_parent != parent) { | |
949854d0 NP |
992 | struct dentry *tmp; |
993 | struct dentry *child; | |
994 | ||
995 | tmp = this_parent->d_parent; | |
996 | rcu_read_lock(); | |
2fd6b7f5 | 997 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
998 | child = this_parent; |
999 | this_parent = tmp; | |
2fd6b7f5 | 1000 | spin_lock(&this_parent->d_lock); |
949854d0 NP |
1001 | /* might go back up the wrong parent if we have had a rename |
1002 | * or deletion */ | |
1003 | if (this_parent != child->d_parent || | |
58db63d0 | 1004 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 1005 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1006 | rcu_read_unlock(); |
1007 | goto rename_retry; | |
1008 | } | |
1009 | rcu_read_unlock(); | |
1010 | next = child->d_u.d_child.next; | |
1da177e4 LT |
1011 | goto resume; |
1012 | } | |
2fd6b7f5 | 1013 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1014 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1015 | goto rename_retry; |
58db63d0 NP |
1016 | if (locked) |
1017 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1018 | return 0; /* No mount points found in tree */ |
1019 | positive: | |
58db63d0 | 1020 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1021 | goto rename_retry; |
58db63d0 NP |
1022 | if (locked) |
1023 | write_sequnlock(&rename_lock); | |
1da177e4 | 1024 | return 1; |
58db63d0 NP |
1025 | |
1026 | rename_retry: | |
1027 | locked = 1; | |
1028 | write_seqlock(&rename_lock); | |
1029 | goto again; | |
1da177e4 | 1030 | } |
ec4f8605 | 1031 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1032 | |
1033 | /* | |
1034 | * Search the dentry child list for the specified parent, | |
1035 | * and move any unused dentries to the end of the unused | |
1036 | * list for prune_dcache(). We descend to the next level | |
1037 | * whenever the d_subdirs list is non-empty and continue | |
1038 | * searching. | |
1039 | * | |
1040 | * It returns zero iff there are no unused children, | |
1041 | * otherwise it returns the number of children moved to | |
1042 | * the end of the unused list. This may not be the total | |
1043 | * number of unused children, because select_parent can | |
1044 | * drop the lock and return early due to latency | |
1045 | * constraints. | |
1046 | */ | |
1047 | static int select_parent(struct dentry * parent) | |
1048 | { | |
949854d0 | 1049 | struct dentry *this_parent; |
1da177e4 | 1050 | struct list_head *next; |
949854d0 | 1051 | unsigned seq; |
1da177e4 | 1052 | int found = 0; |
58db63d0 | 1053 | int locked = 0; |
1da177e4 | 1054 | |
949854d0 | 1055 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1056 | again: |
1057 | this_parent = parent; | |
2fd6b7f5 | 1058 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1059 | repeat: |
1060 | next = this_parent->d_subdirs.next; | |
1061 | resume: | |
1062 | while (next != &this_parent->d_subdirs) { | |
1063 | struct list_head *tmp = next; | |
5160ee6f | 1064 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1065 | next = tmp->next; |
1066 | ||
2fd6b7f5 | 1067 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1068 | |
1da177e4 LT |
1069 | /* |
1070 | * move only zero ref count dentries to the end | |
1071 | * of the unused list for prune_dcache | |
1072 | */ | |
b7ab39f6 | 1073 | if (!dentry->d_count) { |
a4633357 | 1074 | dentry_lru_move_tail(dentry); |
1da177e4 | 1075 | found++; |
a4633357 CH |
1076 | } else { |
1077 | dentry_lru_del(dentry); | |
1da177e4 LT |
1078 | } |
1079 | ||
1080 | /* | |
1081 | * We can return to the caller if we have found some (this | |
1082 | * ensures forward progress). We'll be coming back to find | |
1083 | * the rest. | |
1084 | */ | |
2fd6b7f5 NP |
1085 | if (found && need_resched()) { |
1086 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1087 | goto out; |
2fd6b7f5 | 1088 | } |
1da177e4 LT |
1089 | |
1090 | /* | |
1091 | * Descend a level if the d_subdirs list is non-empty. | |
1092 | */ | |
1093 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1094 | spin_unlock(&this_parent->d_lock); |
1095 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1096 | this_parent = dentry; |
2fd6b7f5 | 1097 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1098 | goto repeat; |
1099 | } | |
2fd6b7f5 NP |
1100 | |
1101 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1102 | } |
1103 | /* | |
1104 | * All done at this level ... ascend and resume the search. | |
1105 | */ | |
1106 | if (this_parent != parent) { | |
2fd6b7f5 | 1107 | struct dentry *tmp; |
949854d0 NP |
1108 | struct dentry *child; |
1109 | ||
2fd6b7f5 | 1110 | tmp = this_parent->d_parent; |
949854d0 | 1111 | rcu_read_lock(); |
2fd6b7f5 | 1112 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1113 | child = this_parent; |
2fd6b7f5 NP |
1114 | this_parent = tmp; |
1115 | spin_lock(&this_parent->d_lock); | |
949854d0 NP |
1116 | /* might go back up the wrong parent if we have had a rename |
1117 | * or deletion */ | |
1118 | if (this_parent != child->d_parent || | |
58db63d0 | 1119 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 1120 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1121 | rcu_read_unlock(); |
1122 | goto rename_retry; | |
1123 | } | |
1124 | rcu_read_unlock(); | |
1125 | next = child->d_u.d_child.next; | |
1da177e4 LT |
1126 | goto resume; |
1127 | } | |
1128 | out: | |
2fd6b7f5 | 1129 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1130 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1131 | goto rename_retry; |
58db63d0 NP |
1132 | if (locked) |
1133 | write_sequnlock(&rename_lock); | |
1da177e4 | 1134 | return found; |
58db63d0 NP |
1135 | |
1136 | rename_retry: | |
1137 | if (found) | |
1138 | return found; | |
1139 | locked = 1; | |
1140 | write_seqlock(&rename_lock); | |
1141 | goto again; | |
1da177e4 LT |
1142 | } |
1143 | ||
1144 | /** | |
1145 | * shrink_dcache_parent - prune dcache | |
1146 | * @parent: parent of entries to prune | |
1147 | * | |
1148 | * Prune the dcache to remove unused children of the parent dentry. | |
1149 | */ | |
1150 | ||
1151 | void shrink_dcache_parent(struct dentry * parent) | |
1152 | { | |
da3bbdd4 | 1153 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1154 | int found; |
1155 | ||
1156 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 1157 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 1158 | } |
ec4f8605 | 1159 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1160 | |
1da177e4 LT |
1161 | /* |
1162 | * Scan `nr' dentries and return the number which remain. | |
1163 | * | |
1164 | * We need to avoid reentering the filesystem if the caller is performing a | |
1165 | * GFP_NOFS allocation attempt. One example deadlock is: | |
1166 | * | |
1167 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
1168 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
1169 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
1170 | * | |
1171 | * In this case we return -1 to tell the caller that we baled. | |
1172 | */ | |
7f8275d0 | 1173 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
1174 | { |
1175 | if (nr) { | |
1176 | if (!(gfp_mask & __GFP_FS)) | |
1177 | return -1; | |
da3bbdd4 | 1178 | prune_dcache(nr); |
1da177e4 | 1179 | } |
312d3ca8 | 1180 | |
86c8749e | 1181 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
1182 | } |
1183 | ||
8e1f936b RR |
1184 | static struct shrinker dcache_shrinker = { |
1185 | .shrink = shrink_dcache_memory, | |
1186 | .seeks = DEFAULT_SEEKS, | |
1187 | }; | |
1188 | ||
1da177e4 LT |
1189 | /** |
1190 | * d_alloc - allocate a dcache entry | |
1191 | * @parent: parent of entry to allocate | |
1192 | * @name: qstr of the name | |
1193 | * | |
1194 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1195 | * available. On a success the dentry is returned. The name passed in is | |
1196 | * copied and the copy passed in may be reused after this call. | |
1197 | */ | |
1198 | ||
1199 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1200 | { | |
1201 | struct dentry *dentry; | |
1202 | char *dname; | |
1203 | ||
e12ba74d | 1204 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1205 | if (!dentry) |
1206 | return NULL; | |
1207 | ||
1208 | if (name->len > DNAME_INLINE_LEN-1) { | |
1209 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1210 | if (!dname) { | |
1211 | kmem_cache_free(dentry_cache, dentry); | |
1212 | return NULL; | |
1213 | } | |
1214 | } else { | |
1215 | dname = dentry->d_iname; | |
1216 | } | |
1217 | dentry->d_name.name = dname; | |
1218 | ||
1219 | dentry->d_name.len = name->len; | |
1220 | dentry->d_name.hash = name->hash; | |
1221 | memcpy(dname, name->name, name->len); | |
1222 | dname[name->len] = 0; | |
1223 | ||
b7ab39f6 | 1224 | dentry->d_count = 1; |
1da177e4 LT |
1225 | dentry->d_flags = DCACHE_UNHASHED; |
1226 | spin_lock_init(&dentry->d_lock); | |
1227 | dentry->d_inode = NULL; | |
1228 | dentry->d_parent = NULL; | |
1229 | dentry->d_sb = NULL; | |
1230 | dentry->d_op = NULL; | |
1231 | dentry->d_fsdata = NULL; | |
1232 | dentry->d_mounted = 0; | |
1da177e4 LT |
1233 | INIT_HLIST_NODE(&dentry->d_hash); |
1234 | INIT_LIST_HEAD(&dentry->d_lru); | |
1235 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1236 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1237 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
1238 | |
1239 | if (parent) { | |
2fd6b7f5 | 1240 | spin_lock(&parent->d_lock); |
89ad485f NP |
1241 | /* |
1242 | * don't need child lock because it is not subject | |
1243 | * to concurrency here | |
1244 | */ | |
dc0474be NP |
1245 | __dget_dlock(parent); |
1246 | dentry->d_parent = parent; | |
1da177e4 | 1247 | dentry->d_sb = parent->d_sb; |
5160ee6f | 1248 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
2fd6b7f5 | 1249 | spin_unlock(&parent->d_lock); |
2fd6b7f5 | 1250 | } |
1da177e4 | 1251 | |
3e880fb5 | 1252 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1253 | |
1da177e4 LT |
1254 | return dentry; |
1255 | } | |
ec4f8605 | 1256 | EXPORT_SYMBOL(d_alloc); |
1da177e4 LT |
1257 | |
1258 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
1259 | { | |
1260 | struct qstr q; | |
1261 | ||
1262 | q.name = name; | |
1263 | q.len = strlen(name); | |
1264 | q.hash = full_name_hash(q.name, q.len); | |
1265 | return d_alloc(parent, &q); | |
1266 | } | |
ef26ca97 | 1267 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1268 | |
360da900 OH |
1269 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1270 | { | |
b23fb0a6 | 1271 | spin_lock(&dentry->d_lock); |
360da900 OH |
1272 | if (inode) |
1273 | list_add(&dentry->d_alias, &inode->i_dentry); | |
1274 | dentry->d_inode = inode; | |
b23fb0a6 | 1275 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1276 | fsnotify_d_instantiate(dentry, inode); |
1277 | } | |
1278 | ||
1da177e4 LT |
1279 | /** |
1280 | * d_instantiate - fill in inode information for a dentry | |
1281 | * @entry: dentry to complete | |
1282 | * @inode: inode to attach to this dentry | |
1283 | * | |
1284 | * Fill in inode information in the entry. | |
1285 | * | |
1286 | * This turns negative dentries into productive full members | |
1287 | * of society. | |
1288 | * | |
1289 | * NOTE! This assumes that the inode count has been incremented | |
1290 | * (or otherwise set) by the caller to indicate that it is now | |
1291 | * in use by the dcache. | |
1292 | */ | |
1293 | ||
1294 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1295 | { | |
28133c7b | 1296 | BUG_ON(!list_empty(&entry->d_alias)); |
b23fb0a6 | 1297 | spin_lock(&dcache_inode_lock); |
360da900 | 1298 | __d_instantiate(entry, inode); |
b23fb0a6 | 1299 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1300 | security_d_instantiate(entry, inode); |
1301 | } | |
ec4f8605 | 1302 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1303 | |
1304 | /** | |
1305 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1306 | * @entry: dentry to instantiate | |
1307 | * @inode: inode to attach to this dentry | |
1308 | * | |
1309 | * Fill in inode information in the entry. On success, it returns NULL. | |
1310 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1311 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1312 | * |
1313 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1314 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1315 | * |
1316 | * This also assumes that the inode count has been incremented | |
1317 | * (or otherwise set) by the caller to indicate that it is now | |
1318 | * in use by the dcache. | |
1da177e4 | 1319 | */ |
770bfad8 DH |
1320 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1321 | struct inode *inode) | |
1da177e4 LT |
1322 | { |
1323 | struct dentry *alias; | |
1324 | int len = entry->d_name.len; | |
1325 | const char *name = entry->d_name.name; | |
1326 | unsigned int hash = entry->d_name.hash; | |
1327 | ||
770bfad8 | 1328 | if (!inode) { |
360da900 | 1329 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1330 | return NULL; |
1331 | } | |
1332 | ||
1da177e4 LT |
1333 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1334 | struct qstr *qstr = &alias->d_name; | |
1335 | ||
9abca360 NP |
1336 | /* |
1337 | * Don't need alias->d_lock here, because aliases with | |
1338 | * d_parent == entry->d_parent are not subject to name or | |
1339 | * parent changes, because the parent inode i_mutex is held. | |
1340 | */ | |
1da177e4 LT |
1341 | if (qstr->hash != hash) |
1342 | continue; | |
1343 | if (alias->d_parent != entry->d_parent) | |
1344 | continue; | |
1345 | if (qstr->len != len) | |
1346 | continue; | |
1347 | if (memcmp(qstr->name, name, len)) | |
1348 | continue; | |
dc0474be | 1349 | __dget(alias); |
1da177e4 LT |
1350 | return alias; |
1351 | } | |
770bfad8 | 1352 | |
360da900 | 1353 | __d_instantiate(entry, inode); |
1da177e4 LT |
1354 | return NULL; |
1355 | } | |
770bfad8 DH |
1356 | |
1357 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1358 | { | |
1359 | struct dentry *result; | |
1360 | ||
1361 | BUG_ON(!list_empty(&entry->d_alias)); | |
1362 | ||
b23fb0a6 | 1363 | spin_lock(&dcache_inode_lock); |
770bfad8 | 1364 | result = __d_instantiate_unique(entry, inode); |
b23fb0a6 | 1365 | spin_unlock(&dcache_inode_lock); |
770bfad8 DH |
1366 | |
1367 | if (!result) { | |
1368 | security_d_instantiate(entry, inode); | |
1369 | return NULL; | |
1370 | } | |
1371 | ||
1372 | BUG_ON(!d_unhashed(result)); | |
1373 | iput(inode); | |
1374 | return result; | |
1375 | } | |
1376 | ||
1da177e4 LT |
1377 | EXPORT_SYMBOL(d_instantiate_unique); |
1378 | ||
1379 | /** | |
1380 | * d_alloc_root - allocate root dentry | |
1381 | * @root_inode: inode to allocate the root for | |
1382 | * | |
1383 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1384 | * instantiated and returned. %NULL is returned if there is insufficient | |
1385 | * memory or the inode passed is %NULL. | |
1386 | */ | |
1387 | ||
1388 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1389 | { | |
1390 | struct dentry *res = NULL; | |
1391 | ||
1392 | if (root_inode) { | |
1393 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1394 | ||
1395 | res = d_alloc(NULL, &name); | |
1396 | if (res) { | |
1397 | res->d_sb = root_inode->i_sb; | |
1398 | res->d_parent = res; | |
1399 | d_instantiate(res, root_inode); | |
1400 | } | |
1401 | } | |
1402 | return res; | |
1403 | } | |
ec4f8605 | 1404 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 LT |
1405 | |
1406 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1407 | unsigned long hash) | |
1408 | { | |
1409 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1410 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1411 | return dentry_hashtable + (hash & D_HASHMASK); | |
1412 | } | |
1413 | ||
4ea3ada2 CH |
1414 | /** |
1415 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1416 | * @inode: inode to allocate the dentry for | |
1417 | * | |
1418 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1419 | * similar open by handle operations. The returned dentry may be anonymous, | |
1420 | * or may have a full name (if the inode was already in the cache). | |
1421 | * | |
1422 | * When called on a directory inode, we must ensure that the inode only ever | |
1423 | * has one dentry. If a dentry is found, that is returned instead of | |
1424 | * allocating a new one. | |
1425 | * | |
1426 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1427 | * to the dentry. In case of an error the reference on the inode is released. |
1428 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1429 | * be passed in and will be the error will be propagate to the return value, | |
1430 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1431 | */ |
1432 | struct dentry *d_obtain_alias(struct inode *inode) | |
1433 | { | |
9308a612 CH |
1434 | static const struct qstr anonstring = { .name = "" }; |
1435 | struct dentry *tmp; | |
1436 | struct dentry *res; | |
4ea3ada2 CH |
1437 | |
1438 | if (!inode) | |
44003728 | 1439 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1440 | if (IS_ERR(inode)) |
1441 | return ERR_CAST(inode); | |
1442 | ||
9308a612 CH |
1443 | res = d_find_alias(inode); |
1444 | if (res) | |
1445 | goto out_iput; | |
1446 | ||
1447 | tmp = d_alloc(NULL, &anonstring); | |
1448 | if (!tmp) { | |
1449 | res = ERR_PTR(-ENOMEM); | |
1450 | goto out_iput; | |
4ea3ada2 | 1451 | } |
9308a612 CH |
1452 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1453 | ||
b5c84bf6 | 1454 | |
b23fb0a6 | 1455 | spin_lock(&dcache_inode_lock); |
9308a612 CH |
1456 | res = __d_find_alias(inode, 0); |
1457 | if (res) { | |
b23fb0a6 | 1458 | spin_unlock(&dcache_inode_lock); |
9308a612 CH |
1459 | dput(tmp); |
1460 | goto out_iput; | |
1461 | } | |
1462 | ||
1463 | /* attach a disconnected dentry */ | |
1464 | spin_lock(&tmp->d_lock); | |
1465 | tmp->d_sb = inode->i_sb; | |
1466 | tmp->d_inode = inode; | |
1467 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1468 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1469 | list_add(&tmp->d_alias, &inode->i_dentry); | |
789680d1 | 1470 | spin_lock(&dcache_hash_lock); |
9308a612 | 1471 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); |
789680d1 | 1472 | spin_unlock(&dcache_hash_lock); |
9308a612 | 1473 | spin_unlock(&tmp->d_lock); |
b23fb0a6 | 1474 | spin_unlock(&dcache_inode_lock); |
9308a612 | 1475 | |
9308a612 CH |
1476 | return tmp; |
1477 | ||
1478 | out_iput: | |
1479 | iput(inode); | |
1480 | return res; | |
4ea3ada2 | 1481 | } |
adc48720 | 1482 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1483 | |
1484 | /** | |
1485 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1486 | * @inode: the inode which may have a disconnected dentry | |
1487 | * @dentry: a negative dentry which we want to point to the inode. | |
1488 | * | |
1489 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1490 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1491 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1492 | * | |
1493 | * This is needed in the lookup routine of any filesystem that is exportable | |
1494 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1495 | * | |
1496 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1497 | * is returned. This matches the expected return value of ->lookup. | |
1498 | * | |
1499 | */ | |
1500 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1501 | { | |
1502 | struct dentry *new = NULL; | |
1503 | ||
21c0d8fd | 1504 | if (inode && S_ISDIR(inode->i_mode)) { |
b23fb0a6 | 1505 | spin_lock(&dcache_inode_lock); |
1da177e4 LT |
1506 | new = __d_find_alias(inode, 1); |
1507 | if (new) { | |
1508 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
b23fb0a6 | 1509 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 1510 | security_d_instantiate(new, inode); |
1da177e4 LT |
1511 | d_move(new, dentry); |
1512 | iput(inode); | |
1513 | } else { | |
b5c84bf6 | 1514 | /* already taking dcache_inode_lock, so d_add() by hand */ |
360da900 | 1515 | __d_instantiate(dentry, inode); |
b23fb0a6 | 1516 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1517 | security_d_instantiate(dentry, inode); |
1518 | d_rehash(dentry); | |
1519 | } | |
1520 | } else | |
1521 | d_add(dentry, inode); | |
1522 | return new; | |
1523 | } | |
ec4f8605 | 1524 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1525 | |
9403540c BN |
1526 | /** |
1527 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1528 | * @inode: the inode case-insensitive lookup has found | |
1529 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1530 | * @name: the case-exact name to be associated with the returned dentry | |
1531 | * | |
1532 | * This is to avoid filling the dcache with case-insensitive names to the | |
1533 | * same inode, only the actual correct case is stored in the dcache for | |
1534 | * case-insensitive filesystems. | |
1535 | * | |
1536 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1537 | * already exists in in the dcache, use it and return it. | |
1538 | * | |
1539 | * If no entry exists with the exact case name, allocate new dentry with | |
1540 | * the exact case, and return the spliced entry. | |
1541 | */ | |
e45b590b | 1542 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1543 | struct qstr *name) |
1544 | { | |
1545 | int error; | |
1546 | struct dentry *found; | |
1547 | struct dentry *new; | |
1548 | ||
b6520c81 CH |
1549 | /* |
1550 | * First check if a dentry matching the name already exists, | |
1551 | * if not go ahead and create it now. | |
1552 | */ | |
9403540c | 1553 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1554 | if (!found) { |
1555 | new = d_alloc(dentry->d_parent, name); | |
1556 | if (!new) { | |
1557 | error = -ENOMEM; | |
1558 | goto err_out; | |
1559 | } | |
b6520c81 | 1560 | |
9403540c BN |
1561 | found = d_splice_alias(inode, new); |
1562 | if (found) { | |
1563 | dput(new); | |
1564 | return found; | |
1565 | } | |
1566 | return new; | |
1567 | } | |
b6520c81 CH |
1568 | |
1569 | /* | |
1570 | * If a matching dentry exists, and it's not negative use it. | |
1571 | * | |
1572 | * Decrement the reference count to balance the iget() done | |
1573 | * earlier on. | |
1574 | */ | |
9403540c BN |
1575 | if (found->d_inode) { |
1576 | if (unlikely(found->d_inode != inode)) { | |
1577 | /* This can't happen because bad inodes are unhashed. */ | |
1578 | BUG_ON(!is_bad_inode(inode)); | |
1579 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1580 | } | |
9403540c BN |
1581 | iput(inode); |
1582 | return found; | |
1583 | } | |
b6520c81 | 1584 | |
9403540c BN |
1585 | /* |
1586 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1587 | * already has a dentry. |
9403540c | 1588 | */ |
b23fb0a6 | 1589 | spin_lock(&dcache_inode_lock); |
b6520c81 | 1590 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1591 | __d_instantiate(found, inode); |
b23fb0a6 | 1592 | spin_unlock(&dcache_inode_lock); |
9403540c BN |
1593 | security_d_instantiate(found, inode); |
1594 | return found; | |
1595 | } | |
b6520c81 | 1596 | |
9403540c | 1597 | /* |
b6520c81 CH |
1598 | * In case a directory already has a (disconnected) entry grab a |
1599 | * reference to it, move it in place and use it. | |
9403540c BN |
1600 | */ |
1601 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
dc0474be | 1602 | __dget(new); |
b23fb0a6 | 1603 | spin_unlock(&dcache_inode_lock); |
9403540c | 1604 | security_d_instantiate(found, inode); |
9403540c | 1605 | d_move(new, found); |
9403540c | 1606 | iput(inode); |
9403540c | 1607 | dput(found); |
9403540c BN |
1608 | return new; |
1609 | ||
1610 | err_out: | |
1611 | iput(inode); | |
1612 | return ERR_PTR(error); | |
1613 | } | |
ec4f8605 | 1614 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 LT |
1615 | |
1616 | /** | |
1617 | * d_lookup - search for a dentry | |
1618 | * @parent: parent dentry | |
1619 | * @name: qstr of name we wish to find | |
b04f784e | 1620 | * Returns: dentry, or NULL |
1da177e4 | 1621 | * |
b04f784e NP |
1622 | * d_lookup searches the children of the parent dentry for the name in |
1623 | * question. If the dentry is found its reference count is incremented and the | |
1624 | * dentry is returned. The caller must use dput to free the entry when it has | |
1625 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1626 | */ |
1da177e4 LT |
1627 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) |
1628 | { | |
1629 | struct dentry * dentry = NULL; | |
949854d0 | 1630 | unsigned seq; |
1da177e4 LT |
1631 | |
1632 | do { | |
1633 | seq = read_seqbegin(&rename_lock); | |
1634 | dentry = __d_lookup(parent, name); | |
1635 | if (dentry) | |
1636 | break; | |
1637 | } while (read_seqretry(&rename_lock, seq)); | |
1638 | return dentry; | |
1639 | } | |
ec4f8605 | 1640 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1641 | |
b04f784e NP |
1642 | /* |
1643 | * __d_lookup - search for a dentry (racy) | |
1644 | * @parent: parent dentry | |
1645 | * @name: qstr of name we wish to find | |
1646 | * Returns: dentry, or NULL | |
1647 | * | |
1648 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1649 | * false-negative result due to unrelated rename activity. | |
1650 | * | |
1651 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1652 | * however it must be used carefully, eg. with a following d_lookup in | |
1653 | * the case of failure. | |
1654 | * | |
1655 | * __d_lookup callers must be commented. | |
1656 | */ | |
1da177e4 LT |
1657 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) |
1658 | { | |
1659 | unsigned int len = name->len; | |
1660 | unsigned int hash = name->hash; | |
1661 | const unsigned char *str = name->name; | |
1662 | struct hlist_head *head = d_hash(parent,hash); | |
1663 | struct dentry *found = NULL; | |
1664 | struct hlist_node *node; | |
665a7583 | 1665 | struct dentry *dentry; |
1da177e4 | 1666 | |
b04f784e NP |
1667 | /* |
1668 | * The hash list is protected using RCU. | |
1669 | * | |
1670 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1671 | * with d_move(). | |
1672 | * | |
1673 | * It is possible that concurrent renames can mess up our list | |
1674 | * walk here and result in missing our dentry, resulting in the | |
1675 | * false-negative result. d_lookup() protects against concurrent | |
1676 | * renames using rename_lock seqlock. | |
1677 | * | |
1678 | * See Documentation/vfs/dcache-locking.txt for more details. | |
1679 | */ | |
1da177e4 LT |
1680 | rcu_read_lock(); |
1681 | ||
665a7583 | 1682 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1683 | struct qstr *qstr; |
1684 | ||
1da177e4 LT |
1685 | if (dentry->d_name.hash != hash) |
1686 | continue; | |
1687 | if (dentry->d_parent != parent) | |
1688 | continue; | |
1689 | ||
1690 | spin_lock(&dentry->d_lock); | |
1691 | ||
1692 | /* | |
1693 | * Recheck the dentry after taking the lock - d_move may have | |
b04f784e NP |
1694 | * changed things. Don't bother checking the hash because |
1695 | * we're about to compare the whole name anyway. | |
1da177e4 LT |
1696 | */ |
1697 | if (dentry->d_parent != parent) | |
1698 | goto next; | |
1699 | ||
d0185c08 LT |
1700 | /* non-existing due to RCU? */ |
1701 | if (d_unhashed(dentry)) | |
1702 | goto next; | |
1703 | ||
1da177e4 LT |
1704 | /* |
1705 | * It is safe to compare names since d_move() cannot | |
1706 | * change the qstr (protected by d_lock). | |
1707 | */ | |
1708 | qstr = &dentry->d_name; | |
1709 | if (parent->d_op && parent->d_op->d_compare) { | |
621e155a NP |
1710 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1711 | dentry, dentry->d_inode, | |
1712 | qstr->len, qstr->name, name)) | |
1da177e4 LT |
1713 | goto next; |
1714 | } else { | |
1715 | if (qstr->len != len) | |
1716 | goto next; | |
1717 | if (memcmp(qstr->name, str, len)) | |
1718 | goto next; | |
1719 | } | |
1720 | ||
b7ab39f6 | 1721 | dentry->d_count++; |
d0185c08 | 1722 | found = dentry; |
1da177e4 LT |
1723 | spin_unlock(&dentry->d_lock); |
1724 | break; | |
1725 | next: | |
1726 | spin_unlock(&dentry->d_lock); | |
1727 | } | |
1728 | rcu_read_unlock(); | |
1729 | ||
1730 | return found; | |
1731 | } | |
1732 | ||
3e7e241f EB |
1733 | /** |
1734 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1735 | * @dir: Directory to search in | |
1736 | * @name: qstr of name we wish to find | |
1737 | * | |
1738 | * On hash failure or on lookup failure NULL is returned. | |
1739 | */ | |
1740 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1741 | { | |
1742 | struct dentry *dentry = NULL; | |
1743 | ||
1744 | /* | |
1745 | * Check for a fs-specific hash function. Note that we must | |
1746 | * calculate the standard hash first, as the d_op->d_hash() | |
1747 | * routine may choose to leave the hash value unchanged. | |
1748 | */ | |
1749 | name->hash = full_name_hash(name->name, name->len); | |
1750 | if (dir->d_op && dir->d_op->d_hash) { | |
b1e6a015 | 1751 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1752 | goto out; |
1753 | } | |
1754 | dentry = d_lookup(dir, name); | |
1755 | out: | |
1756 | return dentry; | |
1757 | } | |
1758 | ||
1da177e4 | 1759 | /** |
786a5e15 | 1760 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 LT |
1761 | * @dentry: The dentry alleged to be valid child of @dparent |
1762 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1763 | * |
1764 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1765 | * This is used by ncpfs in its readdir implementation. | |
1766 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1767 | * |
1768 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1769 | */ |
d3a23e16 | 1770 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1771 | { |
786a5e15 | 1772 | struct dentry *child; |
d3a23e16 | 1773 | |
2fd6b7f5 | 1774 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
1775 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
1776 | if (dentry == child) { | |
2fd6b7f5 | 1777 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 1778 | __dget_dlock(dentry); |
2fd6b7f5 NP |
1779 | spin_unlock(&dentry->d_lock); |
1780 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
1781 | return 1; |
1782 | } | |
1783 | } | |
2fd6b7f5 | 1784 | spin_unlock(&dparent->d_lock); |
786a5e15 | 1785 | |
1da177e4 LT |
1786 | return 0; |
1787 | } | |
ec4f8605 | 1788 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1789 | |
1790 | /* | |
1791 | * When a file is deleted, we have two options: | |
1792 | * - turn this dentry into a negative dentry | |
1793 | * - unhash this dentry and free it. | |
1794 | * | |
1795 | * Usually, we want to just turn this into | |
1796 | * a negative dentry, but if anybody else is | |
1797 | * currently using the dentry or the inode | |
1798 | * we can't do that and we fall back on removing | |
1799 | * it from the hash queues and waiting for | |
1800 | * it to be deleted later when it has no users | |
1801 | */ | |
1802 | ||
1803 | /** | |
1804 | * d_delete - delete a dentry | |
1805 | * @dentry: The dentry to delete | |
1806 | * | |
1807 | * Turn the dentry into a negative dentry if possible, otherwise | |
1808 | * remove it from the hash queues so it can be deleted later | |
1809 | */ | |
1810 | ||
1811 | void d_delete(struct dentry * dentry) | |
1812 | { | |
7a91bf7f | 1813 | int isdir = 0; |
1da177e4 LT |
1814 | /* |
1815 | * Are we the only user? | |
1816 | */ | |
357f8e65 | 1817 | again: |
1da177e4 | 1818 | spin_lock(&dentry->d_lock); |
7a91bf7f | 1819 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
b7ab39f6 | 1820 | if (dentry->d_count == 1) { |
357f8e65 NP |
1821 | if (!spin_trylock(&dcache_inode_lock)) { |
1822 | spin_unlock(&dentry->d_lock); | |
1823 | cpu_relax(); | |
1824 | goto again; | |
1825 | } | |
13e3c5e5 | 1826 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
1da177e4 | 1827 | dentry_iput(dentry); |
7a91bf7f | 1828 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1829 | return; |
1830 | } | |
1831 | ||
1832 | if (!d_unhashed(dentry)) | |
1833 | __d_drop(dentry); | |
1834 | ||
1835 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
1836 | |
1837 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1838 | } |
ec4f8605 | 1839 | EXPORT_SYMBOL(d_delete); |
1da177e4 LT |
1840 | |
1841 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1842 | { | |
1843 | ||
1844 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1845 | hlist_add_head_rcu(&entry->d_hash, list); | |
1846 | } | |
1847 | ||
770bfad8 DH |
1848 | static void _d_rehash(struct dentry * entry) |
1849 | { | |
1850 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1851 | } | |
1852 | ||
1da177e4 LT |
1853 | /** |
1854 | * d_rehash - add an entry back to the hash | |
1855 | * @entry: dentry to add to the hash | |
1856 | * | |
1857 | * Adds a dentry to the hash according to its name. | |
1858 | */ | |
1859 | ||
1860 | void d_rehash(struct dentry * entry) | |
1861 | { | |
1da177e4 | 1862 | spin_lock(&entry->d_lock); |
789680d1 | 1863 | spin_lock(&dcache_hash_lock); |
770bfad8 | 1864 | _d_rehash(entry); |
789680d1 | 1865 | spin_unlock(&dcache_hash_lock); |
1da177e4 | 1866 | spin_unlock(&entry->d_lock); |
1da177e4 | 1867 | } |
ec4f8605 | 1868 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 1869 | |
fb2d5b86 NP |
1870 | /** |
1871 | * dentry_update_name_case - update case insensitive dentry with a new name | |
1872 | * @dentry: dentry to be updated | |
1873 | * @name: new name | |
1874 | * | |
1875 | * Update a case insensitive dentry with new case of name. | |
1876 | * | |
1877 | * dentry must have been returned by d_lookup with name @name. Old and new | |
1878 | * name lengths must match (ie. no d_compare which allows mismatched name | |
1879 | * lengths). | |
1880 | * | |
1881 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
1882 | * keep renames and concurrent inserts, and readdir(2) away). | |
1883 | */ | |
1884 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
1885 | { | |
1886 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
1887 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | |
1888 | ||
fb2d5b86 NP |
1889 | spin_lock(&dentry->d_lock); |
1890 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); | |
1891 | spin_unlock(&dentry->d_lock); | |
fb2d5b86 NP |
1892 | } |
1893 | EXPORT_SYMBOL(dentry_update_name_case); | |
1894 | ||
1da177e4 LT |
1895 | static void switch_names(struct dentry *dentry, struct dentry *target) |
1896 | { | |
1897 | if (dname_external(target)) { | |
1898 | if (dname_external(dentry)) { | |
1899 | /* | |
1900 | * Both external: swap the pointers | |
1901 | */ | |
9a8d5bb4 | 1902 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
1903 | } else { |
1904 | /* | |
1905 | * dentry:internal, target:external. Steal target's | |
1906 | * storage and make target internal. | |
1907 | */ | |
321bcf92 BF |
1908 | memcpy(target->d_iname, dentry->d_name.name, |
1909 | dentry->d_name.len + 1); | |
1da177e4 LT |
1910 | dentry->d_name.name = target->d_name.name; |
1911 | target->d_name.name = target->d_iname; | |
1912 | } | |
1913 | } else { | |
1914 | if (dname_external(dentry)) { | |
1915 | /* | |
1916 | * dentry:external, target:internal. Give dentry's | |
1917 | * storage to target and make dentry internal | |
1918 | */ | |
1919 | memcpy(dentry->d_iname, target->d_name.name, | |
1920 | target->d_name.len + 1); | |
1921 | target->d_name.name = dentry->d_name.name; | |
1922 | dentry->d_name.name = dentry->d_iname; | |
1923 | } else { | |
1924 | /* | |
1925 | * Both are internal. Just copy target to dentry | |
1926 | */ | |
1927 | memcpy(dentry->d_iname, target->d_name.name, | |
1928 | target->d_name.len + 1); | |
dc711ca3 AV |
1929 | dentry->d_name.len = target->d_name.len; |
1930 | return; | |
1da177e4 LT |
1931 | } |
1932 | } | |
9a8d5bb4 | 1933 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
1934 | } |
1935 | ||
2fd6b7f5 NP |
1936 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
1937 | { | |
1938 | /* | |
1939 | * XXXX: do we really need to take target->d_lock? | |
1940 | */ | |
1941 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
1942 | spin_lock(&target->d_parent->d_lock); | |
1943 | else { | |
1944 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
1945 | spin_lock(&dentry->d_parent->d_lock); | |
1946 | spin_lock_nested(&target->d_parent->d_lock, | |
1947 | DENTRY_D_LOCK_NESTED); | |
1948 | } else { | |
1949 | spin_lock(&target->d_parent->d_lock); | |
1950 | spin_lock_nested(&dentry->d_parent->d_lock, | |
1951 | DENTRY_D_LOCK_NESTED); | |
1952 | } | |
1953 | } | |
1954 | if (target < dentry) { | |
1955 | spin_lock_nested(&target->d_lock, 2); | |
1956 | spin_lock_nested(&dentry->d_lock, 3); | |
1957 | } else { | |
1958 | spin_lock_nested(&dentry->d_lock, 2); | |
1959 | spin_lock_nested(&target->d_lock, 3); | |
1960 | } | |
1961 | } | |
1962 | ||
1963 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
1964 | struct dentry *target) | |
1965 | { | |
1966 | if (target->d_parent != dentry->d_parent) | |
1967 | spin_unlock(&dentry->d_parent->d_lock); | |
1968 | if (target->d_parent != target) | |
1969 | spin_unlock(&target->d_parent->d_lock); | |
1970 | } | |
1971 | ||
1da177e4 | 1972 | /* |
2fd6b7f5 NP |
1973 | * When switching names, the actual string doesn't strictly have to |
1974 | * be preserved in the target - because we're dropping the target | |
1975 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1976 | * the new name before we switch. | |
1977 | * | |
1978 | * Note that we have to be a lot more careful about getting the hash | |
1979 | * switched - we have to switch the hash value properly even if it | |
1980 | * then no longer matches the actual (corrupted) string of the target. | |
1981 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 1982 | */ |
9eaef27b | 1983 | /* |
b5c84bf6 | 1984 | * d_move - move a dentry |
1da177e4 LT |
1985 | * @dentry: entry to move |
1986 | * @target: new dentry | |
1987 | * | |
1988 | * Update the dcache to reflect the move of a file name. Negative | |
1989 | * dcache entries should not be moved in this way. | |
1990 | */ | |
b5c84bf6 | 1991 | void d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 1992 | { |
1da177e4 LT |
1993 | if (!dentry->d_inode) |
1994 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
1995 | ||
2fd6b7f5 NP |
1996 | BUG_ON(d_ancestor(dentry, target)); |
1997 | BUG_ON(d_ancestor(target, dentry)); | |
1998 | ||
1da177e4 | 1999 | write_seqlock(&rename_lock); |
2fd6b7f5 NP |
2000 | |
2001 | dentry_lock_for_move(dentry, target); | |
1da177e4 LT |
2002 | |
2003 | /* Move the dentry to the target hash queue, if on different bucket */ | |
789680d1 NP |
2004 | spin_lock(&dcache_hash_lock); |
2005 | if (!d_unhashed(dentry)) | |
2006 | hlist_del_rcu(&dentry->d_hash); | |
2007 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); | |
2008 | spin_unlock(&dcache_hash_lock); | |
1da177e4 LT |
2009 | |
2010 | /* Unhash the target: dput() will then get rid of it */ | |
2011 | __d_drop(target); | |
2012 | ||
5160ee6f ED |
2013 | list_del(&dentry->d_u.d_child); |
2014 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2015 | |
2016 | /* Switch the names.. */ | |
2017 | switch_names(dentry, target); | |
9a8d5bb4 | 2018 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2019 | |
2020 | /* ... and switch the parents */ | |
2021 | if (IS_ROOT(dentry)) { | |
2022 | dentry->d_parent = target->d_parent; | |
2023 | target->d_parent = target; | |
5160ee6f | 2024 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2025 | } else { |
9a8d5bb4 | 2026 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2027 | |
2028 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2029 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2030 | } |
2031 | ||
5160ee6f | 2032 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 NP |
2033 | |
2034 | dentry_unlock_parents_for_move(dentry, target); | |
1da177e4 | 2035 | spin_unlock(&target->d_lock); |
c32ccd87 | 2036 | fsnotify_d_move(dentry); |
1da177e4 LT |
2037 | spin_unlock(&dentry->d_lock); |
2038 | write_sequnlock(&rename_lock); | |
9eaef27b | 2039 | } |
ec4f8605 | 2040 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2041 | |
e2761a11 OH |
2042 | /** |
2043 | * d_ancestor - search for an ancestor | |
2044 | * @p1: ancestor dentry | |
2045 | * @p2: child dentry | |
2046 | * | |
2047 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2048 | * an ancestor of p2, else NULL. | |
9eaef27b | 2049 | */ |
e2761a11 | 2050 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2051 | { |
2052 | struct dentry *p; | |
2053 | ||
871c0067 | 2054 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2055 | if (p->d_parent == p1) |
e2761a11 | 2056 | return p; |
9eaef27b | 2057 | } |
e2761a11 | 2058 | return NULL; |
9eaef27b TM |
2059 | } |
2060 | ||
2061 | /* | |
2062 | * This helper attempts to cope with remotely renamed directories | |
2063 | * | |
2064 | * It assumes that the caller is already holding | |
b5c84bf6 | 2065 | * dentry->d_parent->d_inode->i_mutex and the dcache_inode_lock |
9eaef27b TM |
2066 | * |
2067 | * Note: If ever the locking in lock_rename() changes, then please | |
2068 | * remember to update this too... | |
9eaef27b TM |
2069 | */ |
2070 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
b23fb0a6 | 2071 | __releases(dcache_inode_lock) |
9eaef27b TM |
2072 | { |
2073 | struct mutex *m1 = NULL, *m2 = NULL; | |
2074 | struct dentry *ret; | |
2075 | ||
2076 | /* If alias and dentry share a parent, then no extra locks required */ | |
2077 | if (alias->d_parent == dentry->d_parent) | |
2078 | goto out_unalias; | |
2079 | ||
2080 | /* Check for loops */ | |
2081 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 2082 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
2083 | goto out_err; |
2084 | ||
2085 | /* See lock_rename() */ | |
2086 | ret = ERR_PTR(-EBUSY); | |
2087 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2088 | goto out_err; | |
2089 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2090 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2091 | goto out_err; | |
2092 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2093 | out_unalias: | |
b5c84bf6 | 2094 | d_move(alias, dentry); |
9eaef27b TM |
2095 | ret = alias; |
2096 | out_err: | |
b23fb0a6 | 2097 | spin_unlock(&dcache_inode_lock); |
9eaef27b TM |
2098 | if (m2) |
2099 | mutex_unlock(m2); | |
2100 | if (m1) | |
2101 | mutex_unlock(m1); | |
2102 | return ret; | |
2103 | } | |
2104 | ||
770bfad8 DH |
2105 | /* |
2106 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2107 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2108 | * returns with anon->d_lock held! |
770bfad8 DH |
2109 | */ |
2110 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2111 | { | |
2112 | struct dentry *dparent, *aparent; | |
2113 | ||
2fd6b7f5 | 2114 | dentry_lock_for_move(anon, dentry); |
770bfad8 DH |
2115 | |
2116 | dparent = dentry->d_parent; | |
2117 | aparent = anon->d_parent; | |
2118 | ||
2fd6b7f5 NP |
2119 | switch_names(dentry, anon); |
2120 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2121 | ||
770bfad8 DH |
2122 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2123 | list_del(&dentry->d_u.d_child); | |
2124 | if (!IS_ROOT(dentry)) | |
2125 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2126 | else | |
2127 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2128 | ||
2129 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2130 | list_del(&anon->d_u.d_child); | |
2131 | if (!IS_ROOT(anon)) | |
2132 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2133 | else | |
2134 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2135 | ||
2fd6b7f5 NP |
2136 | dentry_unlock_parents_for_move(anon, dentry); |
2137 | spin_unlock(&dentry->d_lock); | |
2138 | ||
2139 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2140 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2141 | } | |
2142 | ||
2143 | /** | |
2144 | * d_materialise_unique - introduce an inode into the tree | |
2145 | * @dentry: candidate dentry | |
2146 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2147 | * | |
2148 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2149 | * root directory alias in its place if there is one | |
2150 | */ | |
2151 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2152 | { | |
9eaef27b | 2153 | struct dentry *actual; |
770bfad8 DH |
2154 | |
2155 | BUG_ON(!d_unhashed(dentry)); | |
2156 | ||
770bfad8 DH |
2157 | if (!inode) { |
2158 | actual = dentry; | |
360da900 | 2159 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2160 | d_rehash(actual); |
2161 | goto out_nolock; | |
770bfad8 DH |
2162 | } |
2163 | ||
357f8e65 NP |
2164 | spin_lock(&dcache_inode_lock); |
2165 | ||
9eaef27b TM |
2166 | if (S_ISDIR(inode->i_mode)) { |
2167 | struct dentry *alias; | |
2168 | ||
2169 | /* Does an aliased dentry already exist? */ | |
2170 | alias = __d_find_alias(inode, 0); | |
2171 | if (alias) { | |
2172 | actual = alias; | |
2173 | /* Is this an anonymous mountpoint that we could splice | |
2174 | * into our tree? */ | |
2175 | if (IS_ROOT(alias)) { | |
9eaef27b TM |
2176 | __d_materialise_dentry(dentry, alias); |
2177 | __d_drop(alias); | |
2178 | goto found; | |
2179 | } | |
2180 | /* Nope, but we must(!) avoid directory aliasing */ | |
2181 | actual = __d_unalias(dentry, alias); | |
2182 | if (IS_ERR(actual)) | |
2183 | dput(alias); | |
2184 | goto out_nolock; | |
2185 | } | |
770bfad8 DH |
2186 | } |
2187 | ||
2188 | /* Add a unique reference */ | |
2189 | actual = __d_instantiate_unique(dentry, inode); | |
2190 | if (!actual) | |
2191 | actual = dentry; | |
357f8e65 NP |
2192 | else |
2193 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2194 | |
770bfad8 DH |
2195 | spin_lock(&actual->d_lock); |
2196 | found: | |
789680d1 | 2197 | spin_lock(&dcache_hash_lock); |
770bfad8 | 2198 | _d_rehash(actual); |
789680d1 | 2199 | spin_unlock(&dcache_hash_lock); |
770bfad8 | 2200 | spin_unlock(&actual->d_lock); |
b23fb0a6 | 2201 | spin_unlock(&dcache_inode_lock); |
9eaef27b | 2202 | out_nolock: |
770bfad8 DH |
2203 | if (actual == dentry) { |
2204 | security_d_instantiate(dentry, inode); | |
2205 | return NULL; | |
2206 | } | |
2207 | ||
2208 | iput(inode); | |
2209 | return actual; | |
770bfad8 | 2210 | } |
ec4f8605 | 2211 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2212 | |
cdd16d02 | 2213 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2214 | { |
2215 | *buflen -= namelen; | |
2216 | if (*buflen < 0) | |
2217 | return -ENAMETOOLONG; | |
2218 | *buffer -= namelen; | |
2219 | memcpy(*buffer, str, namelen); | |
2220 | return 0; | |
2221 | } | |
2222 | ||
cdd16d02 MS |
2223 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2224 | { | |
2225 | return prepend(buffer, buflen, name->name, name->len); | |
2226 | } | |
2227 | ||
1da177e4 | 2228 | /** |
f2eb6575 MS |
2229 | * Prepend path string to a buffer |
2230 | * | |
9d1bc601 MS |
2231 | * @path: the dentry/vfsmount to report |
2232 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2233 | * @buffer: pointer to the end of the buffer |
2234 | * @buflen: pointer to buffer length | |
552ce544 | 2235 | * |
949854d0 | 2236 | * Caller holds the rename_lock. |
9d1bc601 MS |
2237 | * |
2238 | * If path is not reachable from the supplied root, then the value of | |
2239 | * root is changed (without modifying refcounts). | |
1da177e4 | 2240 | */ |
f2eb6575 MS |
2241 | static int prepend_path(const struct path *path, struct path *root, |
2242 | char **buffer, int *buflen) | |
1da177e4 | 2243 | { |
9d1bc601 MS |
2244 | struct dentry *dentry = path->dentry; |
2245 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2246 | bool slash = false; |
2247 | int error = 0; | |
6092d048 | 2248 | |
99b7db7b | 2249 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2250 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2251 | struct dentry * parent; |
2252 | ||
1da177e4 | 2253 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2254 | /* Global root? */ |
1da177e4 | 2255 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2256 | goto global_root; |
2257 | } | |
2258 | dentry = vfsmnt->mnt_mountpoint; | |
2259 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2260 | continue; |
2261 | } | |
2262 | parent = dentry->d_parent; | |
2263 | prefetch(parent); | |
9abca360 | 2264 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2265 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2266 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2267 | if (!error) |
2268 | error = prepend(buffer, buflen, "/", 1); | |
2269 | if (error) | |
2270 | break; | |
2271 | ||
2272 | slash = true; | |
1da177e4 LT |
2273 | dentry = parent; |
2274 | } | |
2275 | ||
be285c71 | 2276 | out: |
f2eb6575 MS |
2277 | if (!error && !slash) |
2278 | error = prepend(buffer, buflen, "/", 1); | |
2279 | ||
99b7db7b | 2280 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2281 | return error; |
1da177e4 LT |
2282 | |
2283 | global_root: | |
98dc568b MS |
2284 | /* |
2285 | * Filesystems needing to implement special "root names" | |
2286 | * should do so with ->d_dname() | |
2287 | */ | |
2288 | if (IS_ROOT(dentry) && | |
2289 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2290 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2291 | (int) dentry->d_name.len, dentry->d_name.name); | |
2292 | } | |
9d1bc601 MS |
2293 | root->mnt = vfsmnt; |
2294 | root->dentry = dentry; | |
be285c71 | 2295 | goto out; |
f2eb6575 | 2296 | } |
be285c71 | 2297 | |
f2eb6575 MS |
2298 | /** |
2299 | * __d_path - return the path of a dentry | |
2300 | * @path: the dentry/vfsmount to report | |
2301 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2302 | * @buf: buffer to return value in |
f2eb6575 MS |
2303 | * @buflen: buffer length |
2304 | * | |
ffd1f4ed | 2305 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2306 | * |
2307 | * Returns a pointer into the buffer or an error code if the | |
2308 | * path was too long. | |
2309 | * | |
be148247 | 2310 | * "buflen" should be positive. |
f2eb6575 MS |
2311 | * |
2312 | * If path is not reachable from the supplied root, then the value of | |
2313 | * root is changed (without modifying refcounts). | |
2314 | */ | |
2315 | char *__d_path(const struct path *path, struct path *root, | |
2316 | char *buf, int buflen) | |
2317 | { | |
2318 | char *res = buf + buflen; | |
2319 | int error; | |
2320 | ||
2321 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2322 | write_seqlock(&rename_lock); |
f2eb6575 | 2323 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2324 | write_sequnlock(&rename_lock); |
be148247 | 2325 | |
f2eb6575 MS |
2326 | if (error) |
2327 | return ERR_PTR(error); | |
f2eb6575 | 2328 | return res; |
1da177e4 LT |
2329 | } |
2330 | ||
ffd1f4ed MS |
2331 | /* |
2332 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2333 | */ | |
2334 | static int path_with_deleted(const struct path *path, struct path *root, | |
2335 | char **buf, int *buflen) | |
2336 | { | |
2337 | prepend(buf, buflen, "\0", 1); | |
2338 | if (d_unlinked(path->dentry)) { | |
2339 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2340 | if (error) | |
2341 | return error; | |
2342 | } | |
2343 | ||
2344 | return prepend_path(path, root, buf, buflen); | |
2345 | } | |
2346 | ||
8df9d1a4 MS |
2347 | static int prepend_unreachable(char **buffer, int *buflen) |
2348 | { | |
2349 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2350 | } | |
2351 | ||
a03a8a70 JB |
2352 | /** |
2353 | * d_path - return the path of a dentry | |
cf28b486 | 2354 | * @path: path to report |
a03a8a70 JB |
2355 | * @buf: buffer to return value in |
2356 | * @buflen: buffer length | |
2357 | * | |
2358 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2359 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2360 | * | |
52afeefb AV |
2361 | * Returns a pointer into the buffer or an error code if the path was |
2362 | * too long. Note: Callers should use the returned pointer, not the passed | |
2363 | * in buffer, to use the name! The implementation often starts at an offset | |
2364 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2365 | * |
31f3e0b3 | 2366 | * "buflen" should be positive. |
a03a8a70 | 2367 | */ |
20d4fdc1 | 2368 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2369 | { |
ffd1f4ed | 2370 | char *res = buf + buflen; |
6ac08c39 | 2371 | struct path root; |
9d1bc601 | 2372 | struct path tmp; |
ffd1f4ed | 2373 | int error; |
1da177e4 | 2374 | |
c23fbb6b ED |
2375 | /* |
2376 | * We have various synthetic filesystems that never get mounted. On | |
2377 | * these filesystems dentries are never used for lookup purposes, and | |
2378 | * thus don't need to be hashed. They also don't need a name until a | |
2379 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2380 | * below allows us to generate a name for these objects on demand: | |
2381 | */ | |
cf28b486 JB |
2382 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2383 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2384 | |
f7ad3c6b | 2385 | get_fs_root(current->fs, &root); |
949854d0 | 2386 | write_seqlock(&rename_lock); |
9d1bc601 | 2387 | tmp = root; |
ffd1f4ed MS |
2388 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2389 | if (error) | |
2390 | res = ERR_PTR(error); | |
949854d0 | 2391 | write_sequnlock(&rename_lock); |
6ac08c39 | 2392 | path_put(&root); |
1da177e4 LT |
2393 | return res; |
2394 | } | |
ec4f8605 | 2395 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2396 | |
8df9d1a4 MS |
2397 | /** |
2398 | * d_path_with_unreachable - return the path of a dentry | |
2399 | * @path: path to report | |
2400 | * @buf: buffer to return value in | |
2401 | * @buflen: buffer length | |
2402 | * | |
2403 | * The difference from d_path() is that this prepends "(unreachable)" | |
2404 | * to paths which are unreachable from the current process' root. | |
2405 | */ | |
2406 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2407 | { | |
2408 | char *res = buf + buflen; | |
2409 | struct path root; | |
2410 | struct path tmp; | |
2411 | int error; | |
2412 | ||
2413 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2414 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2415 | ||
2416 | get_fs_root(current->fs, &root); | |
949854d0 | 2417 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2418 | tmp = root; |
2419 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2420 | if (!error && !path_equal(&tmp, &root)) | |
2421 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2422 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2423 | path_put(&root); |
2424 | if (error) | |
2425 | res = ERR_PTR(error); | |
2426 | ||
2427 | return res; | |
2428 | } | |
2429 | ||
c23fbb6b ED |
2430 | /* |
2431 | * Helper function for dentry_operations.d_dname() members | |
2432 | */ | |
2433 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2434 | const char *fmt, ...) | |
2435 | { | |
2436 | va_list args; | |
2437 | char temp[64]; | |
2438 | int sz; | |
2439 | ||
2440 | va_start(args, fmt); | |
2441 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2442 | va_end(args); | |
2443 | ||
2444 | if (sz > sizeof(temp) || sz > buflen) | |
2445 | return ERR_PTR(-ENAMETOOLONG); | |
2446 | ||
2447 | buffer += buflen - sz; | |
2448 | return memcpy(buffer, temp, sz); | |
2449 | } | |
2450 | ||
6092d048 RP |
2451 | /* |
2452 | * Write full pathname from the root of the filesystem into the buffer. | |
2453 | */ | |
ec2447c2 | 2454 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2455 | { |
2456 | char *end = buf + buflen; | |
2457 | char *retval; | |
2458 | ||
6092d048 | 2459 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2460 | if (buflen < 1) |
2461 | goto Elong; | |
2462 | /* Get '/' right */ | |
2463 | retval = end-1; | |
2464 | *retval = '/'; | |
2465 | ||
cdd16d02 MS |
2466 | while (!IS_ROOT(dentry)) { |
2467 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2468 | int error; |
6092d048 | 2469 | |
6092d048 | 2470 | prefetch(parent); |
9abca360 NP |
2471 | spin_lock(&dentry->d_lock); |
2472 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2473 | spin_unlock(&dentry->d_lock); | |
2474 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2475 | goto Elong; |
2476 | ||
2477 | retval = end; | |
2478 | dentry = parent; | |
2479 | } | |
c103135c AV |
2480 | return retval; |
2481 | Elong: | |
2482 | return ERR_PTR(-ENAMETOOLONG); | |
2483 | } | |
ec2447c2 NP |
2484 | |
2485 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2486 | { | |
2487 | char *retval; | |
2488 | ||
949854d0 | 2489 | write_seqlock(&rename_lock); |
ec2447c2 | 2490 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2491 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2492 | |
2493 | return retval; | |
2494 | } | |
2495 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2496 | |
2497 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2498 | { | |
2499 | char *p = NULL; | |
2500 | char *retval; | |
2501 | ||
949854d0 | 2502 | write_seqlock(&rename_lock); |
c103135c AV |
2503 | if (d_unlinked(dentry)) { |
2504 | p = buf + buflen; | |
2505 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2506 | goto Elong; | |
2507 | buflen++; | |
2508 | } | |
2509 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2510 | write_sequnlock(&rename_lock); |
c103135c AV |
2511 | if (!IS_ERR(retval) && p) |
2512 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2513 | return retval; |
2514 | Elong: | |
6092d048 RP |
2515 | return ERR_PTR(-ENAMETOOLONG); |
2516 | } | |
2517 | ||
1da177e4 LT |
2518 | /* |
2519 | * NOTE! The user-level library version returns a | |
2520 | * character pointer. The kernel system call just | |
2521 | * returns the length of the buffer filled (which | |
2522 | * includes the ending '\0' character), or a negative | |
2523 | * error value. So libc would do something like | |
2524 | * | |
2525 | * char *getcwd(char * buf, size_t size) | |
2526 | * { | |
2527 | * int retval; | |
2528 | * | |
2529 | * retval = sys_getcwd(buf, size); | |
2530 | * if (retval >= 0) | |
2531 | * return buf; | |
2532 | * errno = -retval; | |
2533 | * return NULL; | |
2534 | * } | |
2535 | */ | |
3cdad428 | 2536 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2537 | { |
552ce544 | 2538 | int error; |
6ac08c39 | 2539 | struct path pwd, root; |
552ce544 | 2540 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2541 | |
2542 | if (!page) | |
2543 | return -ENOMEM; | |
2544 | ||
f7ad3c6b | 2545 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2546 | |
552ce544 | 2547 | error = -ENOENT; |
949854d0 | 2548 | write_seqlock(&rename_lock); |
f3da392e | 2549 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2550 | unsigned long len; |
9d1bc601 | 2551 | struct path tmp = root; |
8df9d1a4 MS |
2552 | char *cwd = page + PAGE_SIZE; |
2553 | int buflen = PAGE_SIZE; | |
1da177e4 | 2554 | |
8df9d1a4 MS |
2555 | prepend(&cwd, &buflen, "\0", 1); |
2556 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2557 | write_sequnlock(&rename_lock); |
552ce544 | 2558 | |
8df9d1a4 | 2559 | if (error) |
552ce544 LT |
2560 | goto out; |
2561 | ||
8df9d1a4 MS |
2562 | /* Unreachable from current root */ |
2563 | if (!path_equal(&tmp, &root)) { | |
2564 | error = prepend_unreachable(&cwd, &buflen); | |
2565 | if (error) | |
2566 | goto out; | |
2567 | } | |
2568 | ||
552ce544 LT |
2569 | error = -ERANGE; |
2570 | len = PAGE_SIZE + page - cwd; | |
2571 | if (len <= size) { | |
2572 | error = len; | |
2573 | if (copy_to_user(buf, cwd, len)) | |
2574 | error = -EFAULT; | |
2575 | } | |
949854d0 NP |
2576 | } else { |
2577 | write_sequnlock(&rename_lock); | |
949854d0 | 2578 | } |
1da177e4 LT |
2579 | |
2580 | out: | |
6ac08c39 JB |
2581 | path_put(&pwd); |
2582 | path_put(&root); | |
1da177e4 LT |
2583 | free_page((unsigned long) page); |
2584 | return error; | |
2585 | } | |
2586 | ||
2587 | /* | |
2588 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2589 | * | |
2590 | * Trivially implemented using the dcache structure | |
2591 | */ | |
2592 | ||
2593 | /** | |
2594 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2595 | * @new_dentry: new dentry | |
2596 | * @old_dentry: old dentry | |
2597 | * | |
2598 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2599 | * Returns 0 otherwise. | |
2600 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2601 | */ | |
2602 | ||
e2761a11 | 2603 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2604 | { |
2605 | int result; | |
949854d0 | 2606 | unsigned seq; |
1da177e4 | 2607 | |
e2761a11 OH |
2608 | if (new_dentry == old_dentry) |
2609 | return 1; | |
2610 | ||
e2761a11 | 2611 | do { |
1da177e4 | 2612 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2613 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2614 | /* |
2615 | * Need rcu_readlock to protect against the d_parent trashing | |
2616 | * due to d_move | |
2617 | */ | |
2618 | rcu_read_lock(); | |
e2761a11 | 2619 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2620 | result = 1; |
e2761a11 OH |
2621 | else |
2622 | result = 0; | |
949854d0 | 2623 | rcu_read_unlock(); |
1da177e4 | 2624 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2625 | |
2626 | return result; | |
2627 | } | |
2628 | ||
2096f759 AV |
2629 | int path_is_under(struct path *path1, struct path *path2) |
2630 | { | |
2631 | struct vfsmount *mnt = path1->mnt; | |
2632 | struct dentry *dentry = path1->dentry; | |
2633 | int res; | |
99b7db7b NP |
2634 | |
2635 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2636 | if (mnt != path2->mnt) { |
2637 | for (;;) { | |
2638 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2639 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2640 | return 0; |
2641 | } | |
2642 | if (mnt->mnt_parent == path2->mnt) | |
2643 | break; | |
2644 | mnt = mnt->mnt_parent; | |
2645 | } | |
2646 | dentry = mnt->mnt_mountpoint; | |
2647 | } | |
2648 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2649 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2650 | return res; |
2651 | } | |
2652 | EXPORT_SYMBOL(path_is_under); | |
2653 | ||
1da177e4 LT |
2654 | void d_genocide(struct dentry *root) |
2655 | { | |
949854d0 | 2656 | struct dentry *this_parent; |
1da177e4 | 2657 | struct list_head *next; |
949854d0 | 2658 | unsigned seq; |
58db63d0 | 2659 | int locked = 0; |
1da177e4 | 2660 | |
949854d0 | 2661 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2662 | again: |
2663 | this_parent = root; | |
2fd6b7f5 | 2664 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2665 | repeat: |
2666 | next = this_parent->d_subdirs.next; | |
2667 | resume: | |
2668 | while (next != &this_parent->d_subdirs) { | |
2669 | struct list_head *tmp = next; | |
5160ee6f | 2670 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2671 | next = tmp->next; |
949854d0 | 2672 | |
da502956 NP |
2673 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2674 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2675 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2676 | continue; |
da502956 | 2677 | } |
1da177e4 | 2678 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2679 | spin_unlock(&this_parent->d_lock); |
2680 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2681 | this_parent = dentry; |
2fd6b7f5 | 2682 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2683 | goto repeat; |
2684 | } | |
949854d0 NP |
2685 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2686 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2687 | dentry->d_count--; | |
2688 | } | |
b7ab39f6 | 2689 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2690 | } |
2691 | if (this_parent != root) { | |
949854d0 NP |
2692 | struct dentry *tmp; |
2693 | struct dentry *child; | |
2694 | ||
2695 | tmp = this_parent->d_parent; | |
2696 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { | |
2697 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2698 | this_parent->d_count--; | |
2699 | } | |
2700 | rcu_read_lock(); | |
b7ab39f6 | 2701 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
2702 | child = this_parent; |
2703 | this_parent = tmp; | |
2fd6b7f5 | 2704 | spin_lock(&this_parent->d_lock); |
949854d0 NP |
2705 | /* might go back up the wrong parent if we have had a rename |
2706 | * or deletion */ | |
2707 | if (this_parent != child->d_parent || | |
58db63d0 | 2708 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 2709 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
2710 | rcu_read_unlock(); |
2711 | goto rename_retry; | |
2712 | } | |
2713 | rcu_read_unlock(); | |
2714 | next = child->d_u.d_child.next; | |
1da177e4 LT |
2715 | goto resume; |
2716 | } | |
2fd6b7f5 | 2717 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2718 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2719 | goto rename_retry; |
58db63d0 NP |
2720 | if (locked) |
2721 | write_sequnlock(&rename_lock); | |
2722 | return; | |
2723 | ||
2724 | rename_retry: | |
2725 | locked = 1; | |
2726 | write_seqlock(&rename_lock); | |
2727 | goto again; | |
1da177e4 LT |
2728 | } |
2729 | ||
2730 | /** | |
2731 | * find_inode_number - check for dentry with name | |
2732 | * @dir: directory to check | |
2733 | * @name: Name to find. | |
2734 | * | |
2735 | * Check whether a dentry already exists for the given name, | |
2736 | * and return the inode number if it has an inode. Otherwise | |
2737 | * 0 is returned. | |
2738 | * | |
2739 | * This routine is used to post-process directory listings for | |
2740 | * filesystems using synthetic inode numbers, and is necessary | |
2741 | * to keep getcwd() working. | |
2742 | */ | |
2743 | ||
2744 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2745 | { | |
2746 | struct dentry * dentry; | |
2747 | ino_t ino = 0; | |
2748 | ||
3e7e241f EB |
2749 | dentry = d_hash_and_lookup(dir, name); |
2750 | if (dentry) { | |
1da177e4 LT |
2751 | if (dentry->d_inode) |
2752 | ino = dentry->d_inode->i_ino; | |
2753 | dput(dentry); | |
2754 | } | |
1da177e4 LT |
2755 | return ino; |
2756 | } | |
ec4f8605 | 2757 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2758 | |
2759 | static __initdata unsigned long dhash_entries; | |
2760 | static int __init set_dhash_entries(char *str) | |
2761 | { | |
2762 | if (!str) | |
2763 | return 0; | |
2764 | dhash_entries = simple_strtoul(str, &str, 0); | |
2765 | return 1; | |
2766 | } | |
2767 | __setup("dhash_entries=", set_dhash_entries); | |
2768 | ||
2769 | static void __init dcache_init_early(void) | |
2770 | { | |
2771 | int loop; | |
2772 | ||
2773 | /* If hashes are distributed across NUMA nodes, defer | |
2774 | * hash allocation until vmalloc space is available. | |
2775 | */ | |
2776 | if (hashdist) | |
2777 | return; | |
2778 | ||
2779 | dentry_hashtable = | |
2780 | alloc_large_system_hash("Dentry cache", | |
2781 | sizeof(struct hlist_head), | |
2782 | dhash_entries, | |
2783 | 13, | |
2784 | HASH_EARLY, | |
2785 | &d_hash_shift, | |
2786 | &d_hash_mask, | |
2787 | 0); | |
2788 | ||
2789 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2790 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2791 | } | |
2792 | ||
74bf17cf | 2793 | static void __init dcache_init(void) |
1da177e4 LT |
2794 | { |
2795 | int loop; | |
2796 | ||
2797 | /* | |
2798 | * A constructor could be added for stable state like the lists, | |
2799 | * but it is probably not worth it because of the cache nature | |
2800 | * of the dcache. | |
2801 | */ | |
0a31bd5f CL |
2802 | dentry_cache = KMEM_CACHE(dentry, |
2803 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2804 | |
8e1f936b | 2805 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2806 | |
2807 | /* Hash may have been set up in dcache_init_early */ | |
2808 | if (!hashdist) | |
2809 | return; | |
2810 | ||
2811 | dentry_hashtable = | |
2812 | alloc_large_system_hash("Dentry cache", | |
2813 | sizeof(struct hlist_head), | |
2814 | dhash_entries, | |
2815 | 13, | |
2816 | 0, | |
2817 | &d_hash_shift, | |
2818 | &d_hash_mask, | |
2819 | 0); | |
2820 | ||
2821 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2822 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2823 | } | |
2824 | ||
2825 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2826 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2827 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2828 | |
1da177e4 LT |
2829 | EXPORT_SYMBOL(d_genocide); |
2830 | ||
1da177e4 LT |
2831 | void __init vfs_caches_init_early(void) |
2832 | { | |
2833 | dcache_init_early(); | |
2834 | inode_init_early(); | |
2835 | } | |
2836 | ||
2837 | void __init vfs_caches_init(unsigned long mempages) | |
2838 | { | |
2839 | unsigned long reserve; | |
2840 | ||
2841 | /* Base hash sizes on available memory, with a reserve equal to | |
2842 | 150% of current kernel size */ | |
2843 | ||
2844 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2845 | mempages -= reserve; | |
2846 | ||
2847 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2848 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2849 | |
74bf17cf DC |
2850 | dcache_init(); |
2851 | inode_init(); | |
1da177e4 | 2852 | files_init(mempages); |
74bf17cf | 2853 | mnt_init(); |
1da177e4 LT |
2854 | bdev_cache_init(); |
2855 | chrdev_init(); | |
2856 | } |