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 */ |
23044507 NP |
432 | static inline struct dentry * __dget_locked_dlock(struct dentry *dentry) |
433 | { | |
b7ab39f6 | 434 | dentry->d_count++; |
23044507 NP |
435 | dentry_lru_del(dentry); |
436 | return dentry; | |
437 | } | |
438 | ||
b5c84bf6 | 439 | /* This must be called with d_lock held */ |
1da177e4 LT |
440 | static inline struct dentry * __dget_locked(struct dentry *dentry) |
441 | { | |
23044507 | 442 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 443 | __dget_locked_dlock(dentry); |
23044507 | 444 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
445 | return dentry; |
446 | } | |
447 | ||
b7ab39f6 NP |
448 | struct dentry * dget_locked_dlock(struct dentry *dentry) |
449 | { | |
450 | return __dget_locked_dlock(dentry); | |
451 | } | |
452 | ||
1da177e4 LT |
453 | struct dentry * dget_locked(struct dentry *dentry) |
454 | { | |
455 | return __dget_locked(dentry); | |
456 | } | |
ec4f8605 | 457 | EXPORT_SYMBOL(dget_locked); |
1da177e4 | 458 | |
b7ab39f6 NP |
459 | struct dentry *dget_parent(struct dentry *dentry) |
460 | { | |
461 | struct dentry *ret; | |
462 | ||
463 | repeat: | |
464 | spin_lock(&dentry->d_lock); | |
465 | ret = dentry->d_parent; | |
466 | if (!ret) | |
467 | goto out; | |
468 | if (dentry == ret) { | |
469 | ret->d_count++; | |
470 | goto out; | |
471 | } | |
472 | if (!spin_trylock(&ret->d_lock)) { | |
473 | spin_unlock(&dentry->d_lock); | |
474 | cpu_relax(); | |
475 | goto repeat; | |
476 | } | |
477 | BUG_ON(!ret->d_count); | |
478 | ret->d_count++; | |
479 | spin_unlock(&ret->d_lock); | |
480 | out: | |
481 | spin_unlock(&dentry->d_lock); | |
482 | return ret; | |
483 | } | |
484 | EXPORT_SYMBOL(dget_parent); | |
485 | ||
1da177e4 LT |
486 | /** |
487 | * d_find_alias - grab a hashed alias of inode | |
488 | * @inode: inode in question | |
489 | * @want_discon: flag, used by d_splice_alias, to request | |
490 | * that only a DISCONNECTED alias be returned. | |
491 | * | |
492 | * If inode has a hashed alias, or is a directory and has any alias, | |
493 | * acquire the reference to alias and return it. Otherwise return NULL. | |
494 | * Notice that if inode is a directory there can be only one alias and | |
495 | * it can be unhashed only if it has no children, or if it is the root | |
496 | * of a filesystem. | |
497 | * | |
21c0d8fd | 498 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 499 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 500 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 501 | */ |
da502956 | 502 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 503 | { |
da502956 | 504 | struct dentry *alias, *discon_alias; |
1da177e4 | 505 | |
da502956 NP |
506 | again: |
507 | discon_alias = NULL; | |
508 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
509 | spin_lock(&alias->d_lock); | |
1da177e4 | 510 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 511 | if (IS_ROOT(alias) && |
da502956 | 512 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 513 | discon_alias = alias; |
da502956 NP |
514 | } else if (!want_discon) { |
515 | __dget_locked_dlock(alias); | |
516 | spin_unlock(&alias->d_lock); | |
517 | return alias; | |
518 | } | |
519 | } | |
520 | spin_unlock(&alias->d_lock); | |
521 | } | |
522 | if (discon_alias) { | |
523 | alias = discon_alias; | |
524 | spin_lock(&alias->d_lock); | |
525 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
526 | if (IS_ROOT(alias) && | |
527 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
528 | __dget_locked_dlock(alias); | |
529 | spin_unlock(&alias->d_lock); | |
1da177e4 LT |
530 | return alias; |
531 | } | |
532 | } | |
da502956 NP |
533 | spin_unlock(&alias->d_lock); |
534 | goto again; | |
1da177e4 | 535 | } |
da502956 | 536 | return NULL; |
1da177e4 LT |
537 | } |
538 | ||
da502956 | 539 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 540 | { |
214fda1f DH |
541 | struct dentry *de = NULL; |
542 | ||
543 | if (!list_empty(&inode->i_dentry)) { | |
b23fb0a6 | 544 | spin_lock(&dcache_inode_lock); |
214fda1f | 545 | de = __d_find_alias(inode, 0); |
b23fb0a6 | 546 | spin_unlock(&dcache_inode_lock); |
214fda1f | 547 | } |
1da177e4 LT |
548 | return de; |
549 | } | |
ec4f8605 | 550 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
551 | |
552 | /* | |
553 | * Try to kill dentries associated with this inode. | |
554 | * WARNING: you must own a reference to inode. | |
555 | */ | |
556 | void d_prune_aliases(struct inode *inode) | |
557 | { | |
0cdca3f9 | 558 | struct dentry *dentry; |
1da177e4 | 559 | restart: |
b23fb0a6 | 560 | spin_lock(&dcache_inode_lock); |
0cdca3f9 | 561 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 562 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 563 | if (!dentry->d_count) { |
23044507 | 564 | __dget_locked_dlock(dentry); |
1da177e4 LT |
565 | __d_drop(dentry); |
566 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 567 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
568 | dput(dentry); |
569 | goto restart; | |
570 | } | |
571 | spin_unlock(&dentry->d_lock); | |
572 | } | |
b23fb0a6 | 573 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 574 | } |
ec4f8605 | 575 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
576 | |
577 | /* | |
d702ccb3 AM |
578 | * Throw away a dentry - free the inode, dput the parent. This requires that |
579 | * the LRU list has already been removed. | |
580 | * | |
85864e10 MS |
581 | * Try to prune ancestors as well. This is necessary to prevent |
582 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
583 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 584 | */ |
2fd6b7f5 | 585 | static void prune_one_dentry(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 586 | __releases(dentry->d_lock) |
2fd6b7f5 | 587 | __releases(parent->d_lock) |
b23fb0a6 | 588 | __releases(dcache_inode_lock) |
1da177e4 | 589 | { |
1da177e4 | 590 | __d_drop(dentry); |
2fd6b7f5 | 591 | dentry = d_kill(dentry, parent); |
d52b9086 MS |
592 | |
593 | /* | |
b5c84bf6 | 594 | * Prune ancestors. |
d52b9086 | 595 | */ |
d52b9086 | 596 | while (dentry) { |
b23fb0a6 | 597 | spin_lock(&dcache_inode_lock); |
2fd6b7f5 | 598 | again: |
b7ab39f6 | 599 | spin_lock(&dentry->d_lock); |
2fd6b7f5 NP |
600 | if (IS_ROOT(dentry)) |
601 | parent = NULL; | |
602 | else | |
603 | parent = dentry->d_parent; | |
604 | if (parent && !spin_trylock(&parent->d_lock)) { | |
605 | spin_unlock(&dentry->d_lock); | |
606 | goto again; | |
607 | } | |
b7ab39f6 NP |
608 | dentry->d_count--; |
609 | if (dentry->d_count) { | |
2fd6b7f5 NP |
610 | if (parent) |
611 | spin_unlock(&parent->d_lock); | |
b7ab39f6 | 612 | spin_unlock(&dentry->d_lock); |
b23fb0a6 | 613 | spin_unlock(&dcache_inode_lock); |
d52b9086 | 614 | return; |
23044507 | 615 | } |
d52b9086 | 616 | |
a4633357 | 617 | dentry_lru_del(dentry); |
d52b9086 | 618 | __d_drop(dentry); |
2fd6b7f5 | 619 | dentry = d_kill(dentry, parent); |
d52b9086 | 620 | } |
1da177e4 LT |
621 | } |
622 | ||
3049cfe2 | 623 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 624 | { |
da3bbdd4 | 625 | struct dentry *dentry; |
da3bbdd4 | 626 | |
3049cfe2 | 627 | while (!list_empty(list)) { |
2fd6b7f5 NP |
628 | struct dentry *parent; |
629 | ||
3049cfe2 | 630 | dentry = list_entry(list->prev, struct dentry, d_lru); |
23044507 NP |
631 | |
632 | if (!spin_trylock(&dentry->d_lock)) { | |
2fd6b7f5 | 633 | relock: |
23044507 NP |
634 | spin_unlock(&dcache_lru_lock); |
635 | cpu_relax(); | |
636 | spin_lock(&dcache_lru_lock); | |
637 | continue; | |
638 | } | |
639 | ||
1da177e4 LT |
640 | /* |
641 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
642 | * the LRU because of laziness during lookup. Do not free |
643 | * it - just keep it off the LRU list. | |
1da177e4 | 644 | */ |
b7ab39f6 | 645 | if (dentry->d_count) { |
2fd6b7f5 | 646 | __dentry_lru_del(dentry); |
da3bbdd4 | 647 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
648 | continue; |
649 | } | |
2fd6b7f5 NP |
650 | if (IS_ROOT(dentry)) |
651 | parent = NULL; | |
652 | else | |
653 | parent = dentry->d_parent; | |
654 | if (parent && !spin_trylock(&parent->d_lock)) { | |
655 | spin_unlock(&dentry->d_lock); | |
656 | goto relock; | |
657 | } | |
658 | __dentry_lru_del(dentry); | |
23044507 NP |
659 | spin_unlock(&dcache_lru_lock); |
660 | ||
2fd6b7f5 | 661 | prune_one_dentry(dentry, parent); |
b5c84bf6 | 662 | /* dcache_inode_lock and dentry->d_lock dropped */ |
b23fb0a6 | 663 | spin_lock(&dcache_inode_lock); |
23044507 | 664 | spin_lock(&dcache_lru_lock); |
da3bbdd4 | 665 | } |
3049cfe2 CH |
666 | } |
667 | ||
668 | /** | |
669 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
670 | * @sb: superblock to shrink dentry LRU. | |
671 | * @count: number of entries to prune | |
672 | * @flags: flags to control the dentry processing | |
673 | * | |
674 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
675 | */ | |
676 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
677 | { | |
678 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
679 | struct dentry *dentry; | |
680 | LIST_HEAD(referenced); | |
681 | LIST_HEAD(tmp); | |
682 | int cnt = *count; | |
683 | ||
b23fb0a6 | 684 | spin_lock(&dcache_inode_lock); |
23044507 NP |
685 | relock: |
686 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
687 | while (!list_empty(&sb->s_dentry_lru)) { |
688 | dentry = list_entry(sb->s_dentry_lru.prev, | |
689 | struct dentry, d_lru); | |
690 | BUG_ON(dentry->d_sb != sb); | |
691 | ||
23044507 NP |
692 | if (!spin_trylock(&dentry->d_lock)) { |
693 | spin_unlock(&dcache_lru_lock); | |
694 | cpu_relax(); | |
695 | goto relock; | |
696 | } | |
697 | ||
3049cfe2 CH |
698 | /* |
699 | * If we are honouring the DCACHE_REFERENCED flag and the | |
700 | * dentry has this flag set, don't free it. Clear the flag | |
701 | * and put it back on the LRU. | |
702 | */ | |
23044507 NP |
703 | if (flags & DCACHE_REFERENCED && |
704 | dentry->d_flags & DCACHE_REFERENCED) { | |
705 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
706 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 707 | spin_unlock(&dentry->d_lock); |
23044507 NP |
708 | } else { |
709 | list_move_tail(&dentry->d_lru, &tmp); | |
710 | spin_unlock(&dentry->d_lock); | |
711 | if (!--cnt) | |
712 | break; | |
3049cfe2 | 713 | } |
23044507 | 714 | /* XXX: re-add cond_resched_lock when dcache_lock goes away */ |
3049cfe2 CH |
715 | } |
716 | ||
717 | *count = cnt; | |
718 | shrink_dentry_list(&tmp); | |
719 | ||
da3bbdd4 KM |
720 | if (!list_empty(&referenced)) |
721 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 722 | spin_unlock(&dcache_lru_lock); |
b23fb0a6 | 723 | spin_unlock(&dcache_inode_lock); |
da3bbdd4 KM |
724 | } |
725 | ||
726 | /** | |
727 | * prune_dcache - shrink the dcache | |
728 | * @count: number of entries to try to free | |
729 | * | |
730 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
731 | * need to unmount something (at which point we need to unuse all dentries). | |
732 | * | |
733 | * This function may fail to free any resources if all the dentries are in use. | |
734 | */ | |
735 | static void prune_dcache(int count) | |
736 | { | |
dca33252 | 737 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 738 | int w_count; |
86c8749e | 739 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
740 | int prune_ratio; |
741 | int pruned; | |
742 | ||
743 | if (unused == 0 || count == 0) | |
744 | return; | |
da3bbdd4 KM |
745 | if (count >= unused) |
746 | prune_ratio = 1; | |
747 | else | |
748 | prune_ratio = unused / count; | |
749 | spin_lock(&sb_lock); | |
dca33252 | 750 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
751 | if (list_empty(&sb->s_instances)) |
752 | continue; | |
da3bbdd4 | 753 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 754 | continue; |
da3bbdd4 KM |
755 | sb->s_count++; |
756 | /* Now, we reclaim unused dentrins with fairness. | |
757 | * We reclaim them same percentage from each superblock. | |
758 | * We calculate number of dentries to scan on this sb | |
759 | * as follows, but the implementation is arranged to avoid | |
760 | * overflows: | |
761 | * number of dentries to scan on this sb = | |
762 | * count * (number of dentries on this sb / | |
763 | * number of dentries in the machine) | |
0feae5c4 | 764 | */ |
da3bbdd4 KM |
765 | spin_unlock(&sb_lock); |
766 | if (prune_ratio != 1) | |
767 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
768 | else | |
769 | w_count = sb->s_nr_dentry_unused; | |
770 | pruned = w_count; | |
0feae5c4 | 771 | /* |
da3bbdd4 KM |
772 | * We need to be sure this filesystem isn't being unmounted, |
773 | * otherwise we could race with generic_shutdown_super(), and | |
774 | * end up holding a reference to an inode while the filesystem | |
775 | * is unmounted. So we try to get s_umount, and make sure | |
776 | * s_root isn't NULL. | |
0feae5c4 | 777 | */ |
da3bbdd4 KM |
778 | if (down_read_trylock(&sb->s_umount)) { |
779 | if ((sb->s_root != NULL) && | |
780 | (!list_empty(&sb->s_dentry_lru))) { | |
da3bbdd4 KM |
781 | __shrink_dcache_sb(sb, &w_count, |
782 | DCACHE_REFERENCED); | |
783 | pruned -= w_count; | |
0feae5c4 | 784 | } |
da3bbdd4 | 785 | up_read(&sb->s_umount); |
0feae5c4 | 786 | } |
da3bbdd4 | 787 | spin_lock(&sb_lock); |
dca33252 AV |
788 | if (p) |
789 | __put_super(p); | |
da3bbdd4 | 790 | count -= pruned; |
dca33252 | 791 | p = sb; |
79893c17 AV |
792 | /* more work left to do? */ |
793 | if (count <= 0) | |
794 | break; | |
1da177e4 | 795 | } |
dca33252 AV |
796 | if (p) |
797 | __put_super(p); | |
da3bbdd4 | 798 | spin_unlock(&sb_lock); |
1da177e4 LT |
799 | } |
800 | ||
1da177e4 LT |
801 | /** |
802 | * shrink_dcache_sb - shrink dcache for a superblock | |
803 | * @sb: superblock | |
804 | * | |
3049cfe2 CH |
805 | * Shrink the dcache for the specified super block. This is used to free |
806 | * the dcache before unmounting a file system. | |
1da177e4 | 807 | */ |
3049cfe2 | 808 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 809 | { |
3049cfe2 CH |
810 | LIST_HEAD(tmp); |
811 | ||
b23fb0a6 | 812 | spin_lock(&dcache_inode_lock); |
23044507 | 813 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
814 | while (!list_empty(&sb->s_dentry_lru)) { |
815 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
816 | shrink_dentry_list(&tmp); | |
817 | } | |
23044507 | 818 | spin_unlock(&dcache_lru_lock); |
b23fb0a6 | 819 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 820 | } |
ec4f8605 | 821 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 822 | |
c636ebdb DH |
823 | /* |
824 | * destroy a single subtree of dentries for unmount | |
825 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
826 | * locking | |
827 | */ | |
828 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
829 | { | |
830 | struct dentry *parent; | |
f8713576 | 831 | unsigned detached = 0; |
c636ebdb DH |
832 | |
833 | BUG_ON(!IS_ROOT(dentry)); | |
834 | ||
835 | /* detach this root from the system */ | |
23044507 | 836 | spin_lock(&dentry->d_lock); |
a4633357 | 837 | dentry_lru_del(dentry); |
c636ebdb | 838 | __d_drop(dentry); |
da502956 | 839 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
840 | |
841 | for (;;) { | |
842 | /* descend to the first leaf in the current subtree */ | |
843 | while (!list_empty(&dentry->d_subdirs)) { | |
844 | struct dentry *loop; | |
845 | ||
846 | /* this is a branch with children - detach all of them | |
847 | * from the system in one go */ | |
2fd6b7f5 | 848 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
849 | list_for_each_entry(loop, &dentry->d_subdirs, |
850 | d_u.d_child) { | |
2fd6b7f5 NP |
851 | spin_lock_nested(&loop->d_lock, |
852 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 853 | dentry_lru_del(loop); |
c636ebdb | 854 | __d_drop(loop); |
da502956 | 855 | spin_unlock(&loop->d_lock); |
c636ebdb | 856 | } |
2fd6b7f5 | 857 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
858 | |
859 | /* move to the first child */ | |
860 | dentry = list_entry(dentry->d_subdirs.next, | |
861 | struct dentry, d_u.d_child); | |
862 | } | |
863 | ||
864 | /* consume the dentries from this leaf up through its parents | |
865 | * until we find one with children or run out altogether */ | |
866 | do { | |
867 | struct inode *inode; | |
868 | ||
b7ab39f6 | 869 | if (dentry->d_count != 0) { |
c636ebdb DH |
870 | printk(KERN_ERR |
871 | "BUG: Dentry %p{i=%lx,n=%s}" | |
872 | " still in use (%d)" | |
873 | " [unmount of %s %s]\n", | |
874 | dentry, | |
875 | dentry->d_inode ? | |
876 | dentry->d_inode->i_ino : 0UL, | |
877 | dentry->d_name.name, | |
b7ab39f6 | 878 | dentry->d_count, |
c636ebdb DH |
879 | dentry->d_sb->s_type->name, |
880 | dentry->d_sb->s_id); | |
881 | BUG(); | |
882 | } | |
883 | ||
2fd6b7f5 | 884 | if (IS_ROOT(dentry)) { |
c636ebdb | 885 | parent = NULL; |
2fd6b7f5 NP |
886 | list_del(&dentry->d_u.d_child); |
887 | } else { | |
871c0067 | 888 | parent = dentry->d_parent; |
b7ab39f6 NP |
889 | spin_lock(&parent->d_lock); |
890 | parent->d_count--; | |
2fd6b7f5 | 891 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 892 | spin_unlock(&parent->d_lock); |
871c0067 | 893 | } |
c636ebdb | 894 | |
f8713576 | 895 | detached++; |
c636ebdb DH |
896 | |
897 | inode = dentry->d_inode; | |
898 | if (inode) { | |
899 | dentry->d_inode = NULL; | |
900 | list_del_init(&dentry->d_alias); | |
901 | if (dentry->d_op && dentry->d_op->d_iput) | |
902 | dentry->d_op->d_iput(dentry, inode); | |
903 | else | |
904 | iput(inode); | |
905 | } | |
906 | ||
907 | d_free(dentry); | |
908 | ||
909 | /* finished when we fall off the top of the tree, | |
910 | * otherwise we ascend to the parent and move to the | |
911 | * next sibling if there is one */ | |
912 | if (!parent) | |
312d3ca8 | 913 | return; |
c636ebdb | 914 | dentry = parent; |
c636ebdb DH |
915 | } while (list_empty(&dentry->d_subdirs)); |
916 | ||
917 | dentry = list_entry(dentry->d_subdirs.next, | |
918 | struct dentry, d_u.d_child); | |
919 | } | |
920 | } | |
921 | ||
922 | /* | |
923 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 924 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
925 | * - the superblock is detached from all mountings and open files, so the |
926 | * dentry trees will not be rearranged by the VFS | |
927 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
928 | * any dentries belonging to this superblock that it comes across | |
929 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
930 | * in this superblock | |
931 | */ | |
932 | void shrink_dcache_for_umount(struct super_block *sb) | |
933 | { | |
934 | struct dentry *dentry; | |
935 | ||
936 | if (down_read_trylock(&sb->s_umount)) | |
937 | BUG(); | |
938 | ||
939 | dentry = sb->s_root; | |
940 | sb->s_root = NULL; | |
b7ab39f6 NP |
941 | spin_lock(&dentry->d_lock); |
942 | dentry->d_count--; | |
943 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
944 | shrink_dcache_for_umount_subtree(dentry); |
945 | ||
946 | while (!hlist_empty(&sb->s_anon)) { | |
947 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
948 | shrink_dcache_for_umount_subtree(dentry); | |
949 | } | |
950 | } | |
951 | ||
1da177e4 LT |
952 | /* |
953 | * Search for at least 1 mount point in the dentry's subdirs. | |
954 | * We descend to the next level whenever the d_subdirs | |
955 | * list is non-empty and continue searching. | |
956 | */ | |
957 | ||
958 | /** | |
959 | * have_submounts - check for mounts over a dentry | |
960 | * @parent: dentry to check. | |
961 | * | |
962 | * Return true if the parent or its subdirectories contain | |
963 | * a mount point | |
964 | */ | |
1da177e4 LT |
965 | int have_submounts(struct dentry *parent) |
966 | { | |
949854d0 | 967 | struct dentry *this_parent; |
1da177e4 | 968 | struct list_head *next; |
949854d0 | 969 | unsigned seq; |
58db63d0 | 970 | int locked = 0; |
949854d0 | 971 | |
949854d0 | 972 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
973 | again: |
974 | this_parent = parent; | |
1da177e4 | 975 | |
1da177e4 LT |
976 | if (d_mountpoint(parent)) |
977 | goto positive; | |
2fd6b7f5 | 978 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
979 | repeat: |
980 | next = this_parent->d_subdirs.next; | |
981 | resume: | |
982 | while (next != &this_parent->d_subdirs) { | |
983 | struct list_head *tmp = next; | |
5160ee6f | 984 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 985 | next = tmp->next; |
2fd6b7f5 NP |
986 | |
987 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 988 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
989 | if (d_mountpoint(dentry)) { |
990 | spin_unlock(&dentry->d_lock); | |
991 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 992 | goto positive; |
2fd6b7f5 | 993 | } |
1da177e4 | 994 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
995 | spin_unlock(&this_parent->d_lock); |
996 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 997 | this_parent = dentry; |
2fd6b7f5 | 998 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
999 | goto repeat; |
1000 | } | |
2fd6b7f5 | 1001 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1002 | } |
1003 | /* | |
1004 | * All done at this level ... ascend and resume the search. | |
1005 | */ | |
1006 | if (this_parent != parent) { | |
949854d0 NP |
1007 | struct dentry *tmp; |
1008 | struct dentry *child; | |
1009 | ||
1010 | tmp = this_parent->d_parent; | |
1011 | rcu_read_lock(); | |
2fd6b7f5 | 1012 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1013 | child = this_parent; |
1014 | this_parent = tmp; | |
2fd6b7f5 | 1015 | spin_lock(&this_parent->d_lock); |
949854d0 NP |
1016 | /* might go back up the wrong parent if we have had a rename |
1017 | * or deletion */ | |
1018 | if (this_parent != child->d_parent || | |
58db63d0 | 1019 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 1020 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1021 | rcu_read_unlock(); |
1022 | goto rename_retry; | |
1023 | } | |
1024 | rcu_read_unlock(); | |
1025 | next = child->d_u.d_child.next; | |
1da177e4 LT |
1026 | goto resume; |
1027 | } | |
2fd6b7f5 | 1028 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1029 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1030 | goto rename_retry; |
58db63d0 NP |
1031 | if (locked) |
1032 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1033 | return 0; /* No mount points found in tree */ |
1034 | positive: | |
58db63d0 | 1035 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1036 | goto rename_retry; |
58db63d0 NP |
1037 | if (locked) |
1038 | write_sequnlock(&rename_lock); | |
1da177e4 | 1039 | return 1; |
58db63d0 NP |
1040 | |
1041 | rename_retry: | |
1042 | locked = 1; | |
1043 | write_seqlock(&rename_lock); | |
1044 | goto again; | |
1da177e4 | 1045 | } |
ec4f8605 | 1046 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1047 | |
1048 | /* | |
1049 | * Search the dentry child list for the specified parent, | |
1050 | * and move any unused dentries to the end of the unused | |
1051 | * list for prune_dcache(). We descend to the next level | |
1052 | * whenever the d_subdirs list is non-empty and continue | |
1053 | * searching. | |
1054 | * | |
1055 | * It returns zero iff there are no unused children, | |
1056 | * otherwise it returns the number of children moved to | |
1057 | * the end of the unused list. This may not be the total | |
1058 | * number of unused children, because select_parent can | |
1059 | * drop the lock and return early due to latency | |
1060 | * constraints. | |
1061 | */ | |
1062 | static int select_parent(struct dentry * parent) | |
1063 | { | |
949854d0 | 1064 | struct dentry *this_parent; |
1da177e4 | 1065 | struct list_head *next; |
949854d0 | 1066 | unsigned seq; |
1da177e4 | 1067 | int found = 0; |
58db63d0 | 1068 | int locked = 0; |
1da177e4 | 1069 | |
949854d0 | 1070 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1071 | again: |
1072 | this_parent = parent; | |
2fd6b7f5 | 1073 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1074 | repeat: |
1075 | next = this_parent->d_subdirs.next; | |
1076 | resume: | |
1077 | while (next != &this_parent->d_subdirs) { | |
1078 | struct list_head *tmp = next; | |
5160ee6f | 1079 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1080 | next = tmp->next; |
1081 | ||
2fd6b7f5 | 1082 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1083 | |
1da177e4 LT |
1084 | /* |
1085 | * move only zero ref count dentries to the end | |
1086 | * of the unused list for prune_dcache | |
1087 | */ | |
b7ab39f6 | 1088 | if (!dentry->d_count) { |
a4633357 | 1089 | dentry_lru_move_tail(dentry); |
1da177e4 | 1090 | found++; |
a4633357 CH |
1091 | } else { |
1092 | dentry_lru_del(dentry); | |
1da177e4 LT |
1093 | } |
1094 | ||
1095 | /* | |
1096 | * We can return to the caller if we have found some (this | |
1097 | * ensures forward progress). We'll be coming back to find | |
1098 | * the rest. | |
1099 | */ | |
2fd6b7f5 NP |
1100 | if (found && need_resched()) { |
1101 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1102 | goto out; |
2fd6b7f5 | 1103 | } |
1da177e4 LT |
1104 | |
1105 | /* | |
1106 | * Descend a level if the d_subdirs list is non-empty. | |
1107 | */ | |
1108 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1109 | spin_unlock(&this_parent->d_lock); |
1110 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1111 | this_parent = dentry; |
2fd6b7f5 | 1112 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1113 | goto repeat; |
1114 | } | |
2fd6b7f5 NP |
1115 | |
1116 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1117 | } |
1118 | /* | |
1119 | * All done at this level ... ascend and resume the search. | |
1120 | */ | |
1121 | if (this_parent != parent) { | |
2fd6b7f5 | 1122 | struct dentry *tmp; |
949854d0 NP |
1123 | struct dentry *child; |
1124 | ||
2fd6b7f5 | 1125 | tmp = this_parent->d_parent; |
949854d0 | 1126 | rcu_read_lock(); |
2fd6b7f5 | 1127 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1128 | child = this_parent; |
2fd6b7f5 NP |
1129 | this_parent = tmp; |
1130 | spin_lock(&this_parent->d_lock); | |
949854d0 NP |
1131 | /* might go back up the wrong parent if we have had a rename |
1132 | * or deletion */ | |
1133 | if (this_parent != child->d_parent || | |
58db63d0 | 1134 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 1135 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1136 | rcu_read_unlock(); |
1137 | goto rename_retry; | |
1138 | } | |
1139 | rcu_read_unlock(); | |
1140 | next = child->d_u.d_child.next; | |
1da177e4 LT |
1141 | goto resume; |
1142 | } | |
1143 | out: | |
2fd6b7f5 | 1144 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1145 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1146 | goto rename_retry; |
58db63d0 NP |
1147 | if (locked) |
1148 | write_sequnlock(&rename_lock); | |
1da177e4 | 1149 | return found; |
58db63d0 NP |
1150 | |
1151 | rename_retry: | |
1152 | if (found) | |
1153 | return found; | |
1154 | locked = 1; | |
1155 | write_seqlock(&rename_lock); | |
1156 | goto again; | |
1da177e4 LT |
1157 | } |
1158 | ||
1159 | /** | |
1160 | * shrink_dcache_parent - prune dcache | |
1161 | * @parent: parent of entries to prune | |
1162 | * | |
1163 | * Prune the dcache to remove unused children of the parent dentry. | |
1164 | */ | |
1165 | ||
1166 | void shrink_dcache_parent(struct dentry * parent) | |
1167 | { | |
da3bbdd4 | 1168 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1169 | int found; |
1170 | ||
1171 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 1172 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 1173 | } |
ec4f8605 | 1174 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1175 | |
1da177e4 LT |
1176 | /* |
1177 | * Scan `nr' dentries and return the number which remain. | |
1178 | * | |
1179 | * We need to avoid reentering the filesystem if the caller is performing a | |
1180 | * GFP_NOFS allocation attempt. One example deadlock is: | |
1181 | * | |
1182 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
1183 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
1184 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
1185 | * | |
1186 | * In this case we return -1 to tell the caller that we baled. | |
1187 | */ | |
7f8275d0 | 1188 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
1189 | { |
1190 | if (nr) { | |
1191 | if (!(gfp_mask & __GFP_FS)) | |
1192 | return -1; | |
da3bbdd4 | 1193 | prune_dcache(nr); |
1da177e4 | 1194 | } |
312d3ca8 | 1195 | |
86c8749e | 1196 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
1197 | } |
1198 | ||
8e1f936b RR |
1199 | static struct shrinker dcache_shrinker = { |
1200 | .shrink = shrink_dcache_memory, | |
1201 | .seeks = DEFAULT_SEEKS, | |
1202 | }; | |
1203 | ||
1da177e4 LT |
1204 | /** |
1205 | * d_alloc - allocate a dcache entry | |
1206 | * @parent: parent of entry to allocate | |
1207 | * @name: qstr of the name | |
1208 | * | |
1209 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1210 | * available. On a success the dentry is returned. The name passed in is | |
1211 | * copied and the copy passed in may be reused after this call. | |
1212 | */ | |
1213 | ||
1214 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1215 | { | |
1216 | struct dentry *dentry; | |
1217 | char *dname; | |
1218 | ||
e12ba74d | 1219 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1220 | if (!dentry) |
1221 | return NULL; | |
1222 | ||
1223 | if (name->len > DNAME_INLINE_LEN-1) { | |
1224 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1225 | if (!dname) { | |
1226 | kmem_cache_free(dentry_cache, dentry); | |
1227 | return NULL; | |
1228 | } | |
1229 | } else { | |
1230 | dname = dentry->d_iname; | |
1231 | } | |
1232 | dentry->d_name.name = dname; | |
1233 | ||
1234 | dentry->d_name.len = name->len; | |
1235 | dentry->d_name.hash = name->hash; | |
1236 | memcpy(dname, name->name, name->len); | |
1237 | dname[name->len] = 0; | |
1238 | ||
b7ab39f6 | 1239 | dentry->d_count = 1; |
1da177e4 LT |
1240 | dentry->d_flags = DCACHE_UNHASHED; |
1241 | spin_lock_init(&dentry->d_lock); | |
1242 | dentry->d_inode = NULL; | |
1243 | dentry->d_parent = NULL; | |
1244 | dentry->d_sb = NULL; | |
1245 | dentry->d_op = NULL; | |
1246 | dentry->d_fsdata = NULL; | |
1247 | dentry->d_mounted = 0; | |
1da177e4 LT |
1248 | INIT_HLIST_NODE(&dentry->d_hash); |
1249 | INIT_LIST_HEAD(&dentry->d_lru); | |
1250 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1251 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1252 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
1253 | |
1254 | if (parent) { | |
2fd6b7f5 | 1255 | spin_lock(&parent->d_lock); |
89ad485f NP |
1256 | /* |
1257 | * don't need child lock because it is not subject | |
1258 | * to concurrency here | |
1259 | */ | |
2fd6b7f5 | 1260 | dentry->d_parent = dget_dlock(parent); |
1da177e4 | 1261 | dentry->d_sb = parent->d_sb; |
5160ee6f | 1262 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
2fd6b7f5 | 1263 | spin_unlock(&parent->d_lock); |
2fd6b7f5 | 1264 | } |
1da177e4 | 1265 | |
3e880fb5 | 1266 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1267 | |
1da177e4 LT |
1268 | return dentry; |
1269 | } | |
ec4f8605 | 1270 | EXPORT_SYMBOL(d_alloc); |
1da177e4 LT |
1271 | |
1272 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
1273 | { | |
1274 | struct qstr q; | |
1275 | ||
1276 | q.name = name; | |
1277 | q.len = strlen(name); | |
1278 | q.hash = full_name_hash(q.name, q.len); | |
1279 | return d_alloc(parent, &q); | |
1280 | } | |
ef26ca97 | 1281 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1282 | |
360da900 OH |
1283 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1284 | { | |
b23fb0a6 | 1285 | spin_lock(&dentry->d_lock); |
360da900 OH |
1286 | if (inode) |
1287 | list_add(&dentry->d_alias, &inode->i_dentry); | |
1288 | dentry->d_inode = inode; | |
b23fb0a6 | 1289 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1290 | fsnotify_d_instantiate(dentry, inode); |
1291 | } | |
1292 | ||
1da177e4 LT |
1293 | /** |
1294 | * d_instantiate - fill in inode information for a dentry | |
1295 | * @entry: dentry to complete | |
1296 | * @inode: inode to attach to this dentry | |
1297 | * | |
1298 | * Fill in inode information in the entry. | |
1299 | * | |
1300 | * This turns negative dentries into productive full members | |
1301 | * of society. | |
1302 | * | |
1303 | * NOTE! This assumes that the inode count has been incremented | |
1304 | * (or otherwise set) by the caller to indicate that it is now | |
1305 | * in use by the dcache. | |
1306 | */ | |
1307 | ||
1308 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1309 | { | |
28133c7b | 1310 | BUG_ON(!list_empty(&entry->d_alias)); |
b23fb0a6 | 1311 | spin_lock(&dcache_inode_lock); |
360da900 | 1312 | __d_instantiate(entry, inode); |
b23fb0a6 | 1313 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1314 | security_d_instantiate(entry, inode); |
1315 | } | |
ec4f8605 | 1316 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1317 | |
1318 | /** | |
1319 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1320 | * @entry: dentry to instantiate | |
1321 | * @inode: inode to attach to this dentry | |
1322 | * | |
1323 | * Fill in inode information in the entry. On success, it returns NULL. | |
1324 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1325 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1326 | * |
1327 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1328 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1329 | * |
1330 | * This also assumes that the inode count has been incremented | |
1331 | * (or otherwise set) by the caller to indicate that it is now | |
1332 | * in use by the dcache. | |
1da177e4 | 1333 | */ |
770bfad8 DH |
1334 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1335 | struct inode *inode) | |
1da177e4 LT |
1336 | { |
1337 | struct dentry *alias; | |
1338 | int len = entry->d_name.len; | |
1339 | const char *name = entry->d_name.name; | |
1340 | unsigned int hash = entry->d_name.hash; | |
1341 | ||
770bfad8 | 1342 | if (!inode) { |
360da900 | 1343 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1344 | return NULL; |
1345 | } | |
1346 | ||
1da177e4 LT |
1347 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1348 | struct qstr *qstr = &alias->d_name; | |
1349 | ||
9abca360 NP |
1350 | /* |
1351 | * Don't need alias->d_lock here, because aliases with | |
1352 | * d_parent == entry->d_parent are not subject to name or | |
1353 | * parent changes, because the parent inode i_mutex is held. | |
1354 | */ | |
1da177e4 LT |
1355 | if (qstr->hash != hash) |
1356 | continue; | |
1357 | if (alias->d_parent != entry->d_parent) | |
1358 | continue; | |
1359 | if (qstr->len != len) | |
1360 | continue; | |
1361 | if (memcmp(qstr->name, name, len)) | |
1362 | continue; | |
1363 | dget_locked(alias); | |
1da177e4 LT |
1364 | return alias; |
1365 | } | |
770bfad8 | 1366 | |
360da900 | 1367 | __d_instantiate(entry, inode); |
1da177e4 LT |
1368 | return NULL; |
1369 | } | |
770bfad8 DH |
1370 | |
1371 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1372 | { | |
1373 | struct dentry *result; | |
1374 | ||
1375 | BUG_ON(!list_empty(&entry->d_alias)); | |
1376 | ||
b23fb0a6 | 1377 | spin_lock(&dcache_inode_lock); |
770bfad8 | 1378 | result = __d_instantiate_unique(entry, inode); |
b23fb0a6 | 1379 | spin_unlock(&dcache_inode_lock); |
770bfad8 DH |
1380 | |
1381 | if (!result) { | |
1382 | security_d_instantiate(entry, inode); | |
1383 | return NULL; | |
1384 | } | |
1385 | ||
1386 | BUG_ON(!d_unhashed(result)); | |
1387 | iput(inode); | |
1388 | return result; | |
1389 | } | |
1390 | ||
1da177e4 LT |
1391 | EXPORT_SYMBOL(d_instantiate_unique); |
1392 | ||
1393 | /** | |
1394 | * d_alloc_root - allocate root dentry | |
1395 | * @root_inode: inode to allocate the root for | |
1396 | * | |
1397 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1398 | * instantiated and returned. %NULL is returned if there is insufficient | |
1399 | * memory or the inode passed is %NULL. | |
1400 | */ | |
1401 | ||
1402 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1403 | { | |
1404 | struct dentry *res = NULL; | |
1405 | ||
1406 | if (root_inode) { | |
1407 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1408 | ||
1409 | res = d_alloc(NULL, &name); | |
1410 | if (res) { | |
1411 | res->d_sb = root_inode->i_sb; | |
1412 | res->d_parent = res; | |
1413 | d_instantiate(res, root_inode); | |
1414 | } | |
1415 | } | |
1416 | return res; | |
1417 | } | |
ec4f8605 | 1418 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 LT |
1419 | |
1420 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1421 | unsigned long hash) | |
1422 | { | |
1423 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1424 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1425 | return dentry_hashtable + (hash & D_HASHMASK); | |
1426 | } | |
1427 | ||
4ea3ada2 CH |
1428 | /** |
1429 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1430 | * @inode: inode to allocate the dentry for | |
1431 | * | |
1432 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1433 | * similar open by handle operations. The returned dentry may be anonymous, | |
1434 | * or may have a full name (if the inode was already in the cache). | |
1435 | * | |
1436 | * When called on a directory inode, we must ensure that the inode only ever | |
1437 | * has one dentry. If a dentry is found, that is returned instead of | |
1438 | * allocating a new one. | |
1439 | * | |
1440 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1441 | * to the dentry. In case of an error the reference on the inode is released. |
1442 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1443 | * be passed in and will be the error will be propagate to the return value, | |
1444 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1445 | */ |
1446 | struct dentry *d_obtain_alias(struct inode *inode) | |
1447 | { | |
9308a612 CH |
1448 | static const struct qstr anonstring = { .name = "" }; |
1449 | struct dentry *tmp; | |
1450 | struct dentry *res; | |
4ea3ada2 CH |
1451 | |
1452 | if (!inode) | |
44003728 | 1453 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1454 | if (IS_ERR(inode)) |
1455 | return ERR_CAST(inode); | |
1456 | ||
9308a612 CH |
1457 | res = d_find_alias(inode); |
1458 | if (res) | |
1459 | goto out_iput; | |
1460 | ||
1461 | tmp = d_alloc(NULL, &anonstring); | |
1462 | if (!tmp) { | |
1463 | res = ERR_PTR(-ENOMEM); | |
1464 | goto out_iput; | |
4ea3ada2 | 1465 | } |
9308a612 CH |
1466 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1467 | ||
b5c84bf6 | 1468 | |
b23fb0a6 | 1469 | spin_lock(&dcache_inode_lock); |
9308a612 CH |
1470 | res = __d_find_alias(inode, 0); |
1471 | if (res) { | |
b23fb0a6 | 1472 | spin_unlock(&dcache_inode_lock); |
9308a612 CH |
1473 | dput(tmp); |
1474 | goto out_iput; | |
1475 | } | |
1476 | ||
1477 | /* attach a disconnected dentry */ | |
1478 | spin_lock(&tmp->d_lock); | |
1479 | tmp->d_sb = inode->i_sb; | |
1480 | tmp->d_inode = inode; | |
1481 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1482 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1483 | list_add(&tmp->d_alias, &inode->i_dentry); | |
789680d1 | 1484 | spin_lock(&dcache_hash_lock); |
9308a612 | 1485 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); |
789680d1 | 1486 | spin_unlock(&dcache_hash_lock); |
9308a612 | 1487 | spin_unlock(&tmp->d_lock); |
b23fb0a6 | 1488 | spin_unlock(&dcache_inode_lock); |
9308a612 | 1489 | |
9308a612 CH |
1490 | return tmp; |
1491 | ||
1492 | out_iput: | |
1493 | iput(inode); | |
1494 | return res; | |
4ea3ada2 | 1495 | } |
adc48720 | 1496 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1497 | |
1498 | /** | |
1499 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1500 | * @inode: the inode which may have a disconnected dentry | |
1501 | * @dentry: a negative dentry which we want to point to the inode. | |
1502 | * | |
1503 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1504 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1505 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1506 | * | |
1507 | * This is needed in the lookup routine of any filesystem that is exportable | |
1508 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1509 | * | |
1510 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1511 | * is returned. This matches the expected return value of ->lookup. | |
1512 | * | |
1513 | */ | |
1514 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1515 | { | |
1516 | struct dentry *new = NULL; | |
1517 | ||
21c0d8fd | 1518 | if (inode && S_ISDIR(inode->i_mode)) { |
b23fb0a6 | 1519 | spin_lock(&dcache_inode_lock); |
1da177e4 LT |
1520 | new = __d_find_alias(inode, 1); |
1521 | if (new) { | |
1522 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
b23fb0a6 | 1523 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 1524 | security_d_instantiate(new, inode); |
1da177e4 LT |
1525 | d_move(new, dentry); |
1526 | iput(inode); | |
1527 | } else { | |
b5c84bf6 | 1528 | /* already taking dcache_inode_lock, so d_add() by hand */ |
360da900 | 1529 | __d_instantiate(dentry, inode); |
b23fb0a6 | 1530 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1531 | security_d_instantiate(dentry, inode); |
1532 | d_rehash(dentry); | |
1533 | } | |
1534 | } else | |
1535 | d_add(dentry, inode); | |
1536 | return new; | |
1537 | } | |
ec4f8605 | 1538 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1539 | |
9403540c BN |
1540 | /** |
1541 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1542 | * @inode: the inode case-insensitive lookup has found | |
1543 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1544 | * @name: the case-exact name to be associated with the returned dentry | |
1545 | * | |
1546 | * This is to avoid filling the dcache with case-insensitive names to the | |
1547 | * same inode, only the actual correct case is stored in the dcache for | |
1548 | * case-insensitive filesystems. | |
1549 | * | |
1550 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1551 | * already exists in in the dcache, use it and return it. | |
1552 | * | |
1553 | * If no entry exists with the exact case name, allocate new dentry with | |
1554 | * the exact case, and return the spliced entry. | |
1555 | */ | |
e45b590b | 1556 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1557 | struct qstr *name) |
1558 | { | |
1559 | int error; | |
1560 | struct dentry *found; | |
1561 | struct dentry *new; | |
1562 | ||
b6520c81 CH |
1563 | /* |
1564 | * First check if a dentry matching the name already exists, | |
1565 | * if not go ahead and create it now. | |
1566 | */ | |
9403540c | 1567 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1568 | if (!found) { |
1569 | new = d_alloc(dentry->d_parent, name); | |
1570 | if (!new) { | |
1571 | error = -ENOMEM; | |
1572 | goto err_out; | |
1573 | } | |
b6520c81 | 1574 | |
9403540c BN |
1575 | found = d_splice_alias(inode, new); |
1576 | if (found) { | |
1577 | dput(new); | |
1578 | return found; | |
1579 | } | |
1580 | return new; | |
1581 | } | |
b6520c81 CH |
1582 | |
1583 | /* | |
1584 | * If a matching dentry exists, and it's not negative use it. | |
1585 | * | |
1586 | * Decrement the reference count to balance the iget() done | |
1587 | * earlier on. | |
1588 | */ | |
9403540c BN |
1589 | if (found->d_inode) { |
1590 | if (unlikely(found->d_inode != inode)) { | |
1591 | /* This can't happen because bad inodes are unhashed. */ | |
1592 | BUG_ON(!is_bad_inode(inode)); | |
1593 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1594 | } | |
9403540c BN |
1595 | iput(inode); |
1596 | return found; | |
1597 | } | |
b6520c81 | 1598 | |
9403540c BN |
1599 | /* |
1600 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1601 | * already has a dentry. |
9403540c | 1602 | */ |
b23fb0a6 | 1603 | spin_lock(&dcache_inode_lock); |
b6520c81 | 1604 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1605 | __d_instantiate(found, inode); |
b23fb0a6 | 1606 | spin_unlock(&dcache_inode_lock); |
9403540c BN |
1607 | security_d_instantiate(found, inode); |
1608 | return found; | |
1609 | } | |
b6520c81 | 1610 | |
9403540c | 1611 | /* |
b6520c81 CH |
1612 | * In case a directory already has a (disconnected) entry grab a |
1613 | * reference to it, move it in place and use it. | |
9403540c BN |
1614 | */ |
1615 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
1616 | dget_locked(new); | |
b23fb0a6 | 1617 | spin_unlock(&dcache_inode_lock); |
9403540c | 1618 | security_d_instantiate(found, inode); |
9403540c | 1619 | d_move(new, found); |
9403540c | 1620 | iput(inode); |
9403540c | 1621 | dput(found); |
9403540c BN |
1622 | return new; |
1623 | ||
1624 | err_out: | |
1625 | iput(inode); | |
1626 | return ERR_PTR(error); | |
1627 | } | |
ec4f8605 | 1628 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 LT |
1629 | |
1630 | /** | |
1631 | * d_lookup - search for a dentry | |
1632 | * @parent: parent dentry | |
1633 | * @name: qstr of name we wish to find | |
b04f784e | 1634 | * Returns: dentry, or NULL |
1da177e4 | 1635 | * |
b04f784e NP |
1636 | * d_lookup searches the children of the parent dentry for the name in |
1637 | * question. If the dentry is found its reference count is incremented and the | |
1638 | * dentry is returned. The caller must use dput to free the entry when it has | |
1639 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1640 | */ |
1da177e4 LT |
1641 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) |
1642 | { | |
1643 | struct dentry * dentry = NULL; | |
949854d0 | 1644 | unsigned seq; |
1da177e4 LT |
1645 | |
1646 | do { | |
1647 | seq = read_seqbegin(&rename_lock); | |
1648 | dentry = __d_lookup(parent, name); | |
1649 | if (dentry) | |
1650 | break; | |
1651 | } while (read_seqretry(&rename_lock, seq)); | |
1652 | return dentry; | |
1653 | } | |
ec4f8605 | 1654 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1655 | |
b04f784e NP |
1656 | /* |
1657 | * __d_lookup - search for a dentry (racy) | |
1658 | * @parent: parent dentry | |
1659 | * @name: qstr of name we wish to find | |
1660 | * Returns: dentry, or NULL | |
1661 | * | |
1662 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1663 | * false-negative result due to unrelated rename activity. | |
1664 | * | |
1665 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1666 | * however it must be used carefully, eg. with a following d_lookup in | |
1667 | * the case of failure. | |
1668 | * | |
1669 | * __d_lookup callers must be commented. | |
1670 | */ | |
1da177e4 LT |
1671 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) |
1672 | { | |
1673 | unsigned int len = name->len; | |
1674 | unsigned int hash = name->hash; | |
1675 | const unsigned char *str = name->name; | |
1676 | struct hlist_head *head = d_hash(parent,hash); | |
1677 | struct dentry *found = NULL; | |
1678 | struct hlist_node *node; | |
665a7583 | 1679 | struct dentry *dentry; |
1da177e4 | 1680 | |
b04f784e NP |
1681 | /* |
1682 | * The hash list is protected using RCU. | |
1683 | * | |
1684 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1685 | * with d_move(). | |
1686 | * | |
1687 | * It is possible that concurrent renames can mess up our list | |
1688 | * walk here and result in missing our dentry, resulting in the | |
1689 | * false-negative result. d_lookup() protects against concurrent | |
1690 | * renames using rename_lock seqlock. | |
1691 | * | |
1692 | * See Documentation/vfs/dcache-locking.txt for more details. | |
1693 | */ | |
1da177e4 LT |
1694 | rcu_read_lock(); |
1695 | ||
665a7583 | 1696 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1697 | struct qstr *qstr; |
1698 | ||
1da177e4 LT |
1699 | if (dentry->d_name.hash != hash) |
1700 | continue; | |
1701 | if (dentry->d_parent != parent) | |
1702 | continue; | |
1703 | ||
1704 | spin_lock(&dentry->d_lock); | |
1705 | ||
1706 | /* | |
1707 | * Recheck the dentry after taking the lock - d_move may have | |
b04f784e NP |
1708 | * changed things. Don't bother checking the hash because |
1709 | * we're about to compare the whole name anyway. | |
1da177e4 LT |
1710 | */ |
1711 | if (dentry->d_parent != parent) | |
1712 | goto next; | |
1713 | ||
d0185c08 LT |
1714 | /* non-existing due to RCU? */ |
1715 | if (d_unhashed(dentry)) | |
1716 | goto next; | |
1717 | ||
1da177e4 LT |
1718 | /* |
1719 | * It is safe to compare names since d_move() cannot | |
1720 | * change the qstr (protected by d_lock). | |
1721 | */ | |
1722 | qstr = &dentry->d_name; | |
1723 | if (parent->d_op && parent->d_op->d_compare) { | |
621e155a NP |
1724 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1725 | dentry, dentry->d_inode, | |
1726 | qstr->len, qstr->name, name)) | |
1da177e4 LT |
1727 | goto next; |
1728 | } else { | |
1729 | if (qstr->len != len) | |
1730 | goto next; | |
1731 | if (memcmp(qstr->name, str, len)) | |
1732 | goto next; | |
1733 | } | |
1734 | ||
b7ab39f6 | 1735 | dentry->d_count++; |
d0185c08 | 1736 | found = dentry; |
1da177e4 LT |
1737 | spin_unlock(&dentry->d_lock); |
1738 | break; | |
1739 | next: | |
1740 | spin_unlock(&dentry->d_lock); | |
1741 | } | |
1742 | rcu_read_unlock(); | |
1743 | ||
1744 | return found; | |
1745 | } | |
1746 | ||
3e7e241f EB |
1747 | /** |
1748 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1749 | * @dir: Directory to search in | |
1750 | * @name: qstr of name we wish to find | |
1751 | * | |
1752 | * On hash failure or on lookup failure NULL is returned. | |
1753 | */ | |
1754 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1755 | { | |
1756 | struct dentry *dentry = NULL; | |
1757 | ||
1758 | /* | |
1759 | * Check for a fs-specific hash function. Note that we must | |
1760 | * calculate the standard hash first, as the d_op->d_hash() | |
1761 | * routine may choose to leave the hash value unchanged. | |
1762 | */ | |
1763 | name->hash = full_name_hash(name->name, name->len); | |
1764 | if (dir->d_op && dir->d_op->d_hash) { | |
b1e6a015 | 1765 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1766 | goto out; |
1767 | } | |
1768 | dentry = d_lookup(dir, name); | |
1769 | out: | |
1770 | return dentry; | |
1771 | } | |
1772 | ||
1da177e4 | 1773 | /** |
786a5e15 | 1774 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 LT |
1775 | * @dentry: The dentry alleged to be valid child of @dparent |
1776 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1777 | * |
1778 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1779 | * This is used by ncpfs in its readdir implementation. | |
1780 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1781 | * |
1782 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1783 | */ |
d3a23e16 | 1784 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1785 | { |
786a5e15 | 1786 | struct dentry *child; |
d3a23e16 | 1787 | |
2fd6b7f5 | 1788 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
1789 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
1790 | if (dentry == child) { | |
2fd6b7f5 NP |
1791 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1792 | __dget_locked_dlock(dentry); | |
1793 | spin_unlock(&dentry->d_lock); | |
1794 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
1795 | return 1; |
1796 | } | |
1797 | } | |
2fd6b7f5 | 1798 | spin_unlock(&dparent->d_lock); |
786a5e15 | 1799 | |
1da177e4 LT |
1800 | return 0; |
1801 | } | |
ec4f8605 | 1802 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1803 | |
1804 | /* | |
1805 | * When a file is deleted, we have two options: | |
1806 | * - turn this dentry into a negative dentry | |
1807 | * - unhash this dentry and free it. | |
1808 | * | |
1809 | * Usually, we want to just turn this into | |
1810 | * a negative dentry, but if anybody else is | |
1811 | * currently using the dentry or the inode | |
1812 | * we can't do that and we fall back on removing | |
1813 | * it from the hash queues and waiting for | |
1814 | * it to be deleted later when it has no users | |
1815 | */ | |
1816 | ||
1817 | /** | |
1818 | * d_delete - delete a dentry | |
1819 | * @dentry: The dentry to delete | |
1820 | * | |
1821 | * Turn the dentry into a negative dentry if possible, otherwise | |
1822 | * remove it from the hash queues so it can be deleted later | |
1823 | */ | |
1824 | ||
1825 | void d_delete(struct dentry * dentry) | |
1826 | { | |
7a91bf7f | 1827 | int isdir = 0; |
1da177e4 LT |
1828 | /* |
1829 | * Are we the only user? | |
1830 | */ | |
357f8e65 | 1831 | again: |
1da177e4 | 1832 | spin_lock(&dentry->d_lock); |
7a91bf7f | 1833 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
b7ab39f6 | 1834 | if (dentry->d_count == 1) { |
357f8e65 NP |
1835 | if (!spin_trylock(&dcache_inode_lock)) { |
1836 | spin_unlock(&dentry->d_lock); | |
1837 | cpu_relax(); | |
1838 | goto again; | |
1839 | } | |
13e3c5e5 | 1840 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
1da177e4 | 1841 | dentry_iput(dentry); |
7a91bf7f | 1842 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1843 | return; |
1844 | } | |
1845 | ||
1846 | if (!d_unhashed(dentry)) | |
1847 | __d_drop(dentry); | |
1848 | ||
1849 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
1850 | |
1851 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1852 | } |
ec4f8605 | 1853 | EXPORT_SYMBOL(d_delete); |
1da177e4 LT |
1854 | |
1855 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1856 | { | |
1857 | ||
1858 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1859 | hlist_add_head_rcu(&entry->d_hash, list); | |
1860 | } | |
1861 | ||
770bfad8 DH |
1862 | static void _d_rehash(struct dentry * entry) |
1863 | { | |
1864 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1865 | } | |
1866 | ||
1da177e4 LT |
1867 | /** |
1868 | * d_rehash - add an entry back to the hash | |
1869 | * @entry: dentry to add to the hash | |
1870 | * | |
1871 | * Adds a dentry to the hash according to its name. | |
1872 | */ | |
1873 | ||
1874 | void d_rehash(struct dentry * entry) | |
1875 | { | |
1da177e4 | 1876 | spin_lock(&entry->d_lock); |
789680d1 | 1877 | spin_lock(&dcache_hash_lock); |
770bfad8 | 1878 | _d_rehash(entry); |
789680d1 | 1879 | spin_unlock(&dcache_hash_lock); |
1da177e4 | 1880 | spin_unlock(&entry->d_lock); |
1da177e4 | 1881 | } |
ec4f8605 | 1882 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 1883 | |
fb2d5b86 NP |
1884 | /** |
1885 | * dentry_update_name_case - update case insensitive dentry with a new name | |
1886 | * @dentry: dentry to be updated | |
1887 | * @name: new name | |
1888 | * | |
1889 | * Update a case insensitive dentry with new case of name. | |
1890 | * | |
1891 | * dentry must have been returned by d_lookup with name @name. Old and new | |
1892 | * name lengths must match (ie. no d_compare which allows mismatched name | |
1893 | * lengths). | |
1894 | * | |
1895 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
1896 | * keep renames and concurrent inserts, and readdir(2) away). | |
1897 | */ | |
1898 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
1899 | { | |
1900 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
1901 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | |
1902 | ||
fb2d5b86 NP |
1903 | spin_lock(&dentry->d_lock); |
1904 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); | |
1905 | spin_unlock(&dentry->d_lock); | |
fb2d5b86 NP |
1906 | } |
1907 | EXPORT_SYMBOL(dentry_update_name_case); | |
1908 | ||
1da177e4 LT |
1909 | static void switch_names(struct dentry *dentry, struct dentry *target) |
1910 | { | |
1911 | if (dname_external(target)) { | |
1912 | if (dname_external(dentry)) { | |
1913 | /* | |
1914 | * Both external: swap the pointers | |
1915 | */ | |
9a8d5bb4 | 1916 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
1917 | } else { |
1918 | /* | |
1919 | * dentry:internal, target:external. Steal target's | |
1920 | * storage and make target internal. | |
1921 | */ | |
321bcf92 BF |
1922 | memcpy(target->d_iname, dentry->d_name.name, |
1923 | dentry->d_name.len + 1); | |
1da177e4 LT |
1924 | dentry->d_name.name = target->d_name.name; |
1925 | target->d_name.name = target->d_iname; | |
1926 | } | |
1927 | } else { | |
1928 | if (dname_external(dentry)) { | |
1929 | /* | |
1930 | * dentry:external, target:internal. Give dentry's | |
1931 | * storage to target and make dentry internal | |
1932 | */ | |
1933 | memcpy(dentry->d_iname, target->d_name.name, | |
1934 | target->d_name.len + 1); | |
1935 | target->d_name.name = dentry->d_name.name; | |
1936 | dentry->d_name.name = dentry->d_iname; | |
1937 | } else { | |
1938 | /* | |
1939 | * Both are internal. Just copy target to dentry | |
1940 | */ | |
1941 | memcpy(dentry->d_iname, target->d_name.name, | |
1942 | target->d_name.len + 1); | |
dc711ca3 AV |
1943 | dentry->d_name.len = target->d_name.len; |
1944 | return; | |
1da177e4 LT |
1945 | } |
1946 | } | |
9a8d5bb4 | 1947 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
1948 | } |
1949 | ||
2fd6b7f5 NP |
1950 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
1951 | { | |
1952 | /* | |
1953 | * XXXX: do we really need to take target->d_lock? | |
1954 | */ | |
1955 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
1956 | spin_lock(&target->d_parent->d_lock); | |
1957 | else { | |
1958 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
1959 | spin_lock(&dentry->d_parent->d_lock); | |
1960 | spin_lock_nested(&target->d_parent->d_lock, | |
1961 | DENTRY_D_LOCK_NESTED); | |
1962 | } else { | |
1963 | spin_lock(&target->d_parent->d_lock); | |
1964 | spin_lock_nested(&dentry->d_parent->d_lock, | |
1965 | DENTRY_D_LOCK_NESTED); | |
1966 | } | |
1967 | } | |
1968 | if (target < dentry) { | |
1969 | spin_lock_nested(&target->d_lock, 2); | |
1970 | spin_lock_nested(&dentry->d_lock, 3); | |
1971 | } else { | |
1972 | spin_lock_nested(&dentry->d_lock, 2); | |
1973 | spin_lock_nested(&target->d_lock, 3); | |
1974 | } | |
1975 | } | |
1976 | ||
1977 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
1978 | struct dentry *target) | |
1979 | { | |
1980 | if (target->d_parent != dentry->d_parent) | |
1981 | spin_unlock(&dentry->d_parent->d_lock); | |
1982 | if (target->d_parent != target) | |
1983 | spin_unlock(&target->d_parent->d_lock); | |
1984 | } | |
1985 | ||
1da177e4 | 1986 | /* |
2fd6b7f5 NP |
1987 | * When switching names, the actual string doesn't strictly have to |
1988 | * be preserved in the target - because we're dropping the target | |
1989 | * anyway. As such, we can just do a simple memcpy() to copy over | |
1990 | * the new name before we switch. | |
1991 | * | |
1992 | * Note that we have to be a lot more careful about getting the hash | |
1993 | * switched - we have to switch the hash value properly even if it | |
1994 | * then no longer matches the actual (corrupted) string of the target. | |
1995 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 1996 | */ |
9eaef27b | 1997 | /* |
b5c84bf6 | 1998 | * d_move - move a dentry |
1da177e4 LT |
1999 | * @dentry: entry to move |
2000 | * @target: new dentry | |
2001 | * | |
2002 | * Update the dcache to reflect the move of a file name. Negative | |
2003 | * dcache entries should not be moved in this way. | |
2004 | */ | |
b5c84bf6 | 2005 | void d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2006 | { |
1da177e4 LT |
2007 | if (!dentry->d_inode) |
2008 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2009 | ||
2fd6b7f5 NP |
2010 | BUG_ON(d_ancestor(dentry, target)); |
2011 | BUG_ON(d_ancestor(target, dentry)); | |
2012 | ||
1da177e4 | 2013 | write_seqlock(&rename_lock); |
2fd6b7f5 NP |
2014 | |
2015 | dentry_lock_for_move(dentry, target); | |
1da177e4 LT |
2016 | |
2017 | /* Move the dentry to the target hash queue, if on different bucket */ | |
789680d1 NP |
2018 | spin_lock(&dcache_hash_lock); |
2019 | if (!d_unhashed(dentry)) | |
2020 | hlist_del_rcu(&dentry->d_hash); | |
2021 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); | |
2022 | spin_unlock(&dcache_hash_lock); | |
1da177e4 LT |
2023 | |
2024 | /* Unhash the target: dput() will then get rid of it */ | |
2025 | __d_drop(target); | |
2026 | ||
5160ee6f ED |
2027 | list_del(&dentry->d_u.d_child); |
2028 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2029 | |
2030 | /* Switch the names.. */ | |
2031 | switch_names(dentry, target); | |
9a8d5bb4 | 2032 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2033 | |
2034 | /* ... and switch the parents */ | |
2035 | if (IS_ROOT(dentry)) { | |
2036 | dentry->d_parent = target->d_parent; | |
2037 | target->d_parent = target; | |
5160ee6f | 2038 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2039 | } else { |
9a8d5bb4 | 2040 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2041 | |
2042 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2043 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2044 | } |
2045 | ||
5160ee6f | 2046 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 NP |
2047 | |
2048 | dentry_unlock_parents_for_move(dentry, target); | |
1da177e4 | 2049 | spin_unlock(&target->d_lock); |
c32ccd87 | 2050 | fsnotify_d_move(dentry); |
1da177e4 LT |
2051 | spin_unlock(&dentry->d_lock); |
2052 | write_sequnlock(&rename_lock); | |
9eaef27b | 2053 | } |
ec4f8605 | 2054 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2055 | |
e2761a11 OH |
2056 | /** |
2057 | * d_ancestor - search for an ancestor | |
2058 | * @p1: ancestor dentry | |
2059 | * @p2: child dentry | |
2060 | * | |
2061 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2062 | * an ancestor of p2, else NULL. | |
9eaef27b | 2063 | */ |
e2761a11 | 2064 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2065 | { |
2066 | struct dentry *p; | |
2067 | ||
871c0067 | 2068 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2069 | if (p->d_parent == p1) |
e2761a11 | 2070 | return p; |
9eaef27b | 2071 | } |
e2761a11 | 2072 | return NULL; |
9eaef27b TM |
2073 | } |
2074 | ||
2075 | /* | |
2076 | * This helper attempts to cope with remotely renamed directories | |
2077 | * | |
2078 | * It assumes that the caller is already holding | |
b5c84bf6 | 2079 | * dentry->d_parent->d_inode->i_mutex and the dcache_inode_lock |
9eaef27b TM |
2080 | * |
2081 | * Note: If ever the locking in lock_rename() changes, then please | |
2082 | * remember to update this too... | |
9eaef27b TM |
2083 | */ |
2084 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
b23fb0a6 | 2085 | __releases(dcache_inode_lock) |
9eaef27b TM |
2086 | { |
2087 | struct mutex *m1 = NULL, *m2 = NULL; | |
2088 | struct dentry *ret; | |
2089 | ||
2090 | /* If alias and dentry share a parent, then no extra locks required */ | |
2091 | if (alias->d_parent == dentry->d_parent) | |
2092 | goto out_unalias; | |
2093 | ||
2094 | /* Check for loops */ | |
2095 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 2096 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
2097 | goto out_err; |
2098 | ||
2099 | /* See lock_rename() */ | |
2100 | ret = ERR_PTR(-EBUSY); | |
2101 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2102 | goto out_err; | |
2103 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2104 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2105 | goto out_err; | |
2106 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2107 | out_unalias: | |
b5c84bf6 | 2108 | d_move(alias, dentry); |
9eaef27b TM |
2109 | ret = alias; |
2110 | out_err: | |
b23fb0a6 | 2111 | spin_unlock(&dcache_inode_lock); |
9eaef27b TM |
2112 | if (m2) |
2113 | mutex_unlock(m2); | |
2114 | if (m1) | |
2115 | mutex_unlock(m1); | |
2116 | return ret; | |
2117 | } | |
2118 | ||
770bfad8 DH |
2119 | /* |
2120 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2121 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2122 | * returns with anon->d_lock held! |
770bfad8 DH |
2123 | */ |
2124 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2125 | { | |
2126 | struct dentry *dparent, *aparent; | |
2127 | ||
2fd6b7f5 | 2128 | dentry_lock_for_move(anon, dentry); |
770bfad8 DH |
2129 | |
2130 | dparent = dentry->d_parent; | |
2131 | aparent = anon->d_parent; | |
2132 | ||
2fd6b7f5 NP |
2133 | switch_names(dentry, anon); |
2134 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2135 | ||
770bfad8 DH |
2136 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2137 | list_del(&dentry->d_u.d_child); | |
2138 | if (!IS_ROOT(dentry)) | |
2139 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2140 | else | |
2141 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2142 | ||
2143 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2144 | list_del(&anon->d_u.d_child); | |
2145 | if (!IS_ROOT(anon)) | |
2146 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2147 | else | |
2148 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2149 | ||
2fd6b7f5 NP |
2150 | dentry_unlock_parents_for_move(anon, dentry); |
2151 | spin_unlock(&dentry->d_lock); | |
2152 | ||
2153 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2154 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2155 | } | |
2156 | ||
2157 | /** | |
2158 | * d_materialise_unique - introduce an inode into the tree | |
2159 | * @dentry: candidate dentry | |
2160 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2161 | * | |
2162 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2163 | * root directory alias in its place if there is one | |
2164 | */ | |
2165 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2166 | { | |
9eaef27b | 2167 | struct dentry *actual; |
770bfad8 DH |
2168 | |
2169 | BUG_ON(!d_unhashed(dentry)); | |
2170 | ||
770bfad8 DH |
2171 | if (!inode) { |
2172 | actual = dentry; | |
360da900 | 2173 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2174 | d_rehash(actual); |
2175 | goto out_nolock; | |
770bfad8 DH |
2176 | } |
2177 | ||
357f8e65 NP |
2178 | spin_lock(&dcache_inode_lock); |
2179 | ||
9eaef27b TM |
2180 | if (S_ISDIR(inode->i_mode)) { |
2181 | struct dentry *alias; | |
2182 | ||
2183 | /* Does an aliased dentry already exist? */ | |
2184 | alias = __d_find_alias(inode, 0); | |
2185 | if (alias) { | |
2186 | actual = alias; | |
2187 | /* Is this an anonymous mountpoint that we could splice | |
2188 | * into our tree? */ | |
2189 | if (IS_ROOT(alias)) { | |
9eaef27b TM |
2190 | __d_materialise_dentry(dentry, alias); |
2191 | __d_drop(alias); | |
2192 | goto found; | |
2193 | } | |
2194 | /* Nope, but we must(!) avoid directory aliasing */ | |
2195 | actual = __d_unalias(dentry, alias); | |
2196 | if (IS_ERR(actual)) | |
2197 | dput(alias); | |
2198 | goto out_nolock; | |
2199 | } | |
770bfad8 DH |
2200 | } |
2201 | ||
2202 | /* Add a unique reference */ | |
2203 | actual = __d_instantiate_unique(dentry, inode); | |
2204 | if (!actual) | |
2205 | actual = dentry; | |
357f8e65 NP |
2206 | else |
2207 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2208 | |
770bfad8 DH |
2209 | spin_lock(&actual->d_lock); |
2210 | found: | |
789680d1 | 2211 | spin_lock(&dcache_hash_lock); |
770bfad8 | 2212 | _d_rehash(actual); |
789680d1 | 2213 | spin_unlock(&dcache_hash_lock); |
770bfad8 | 2214 | spin_unlock(&actual->d_lock); |
b23fb0a6 | 2215 | spin_unlock(&dcache_inode_lock); |
9eaef27b | 2216 | out_nolock: |
770bfad8 DH |
2217 | if (actual == dentry) { |
2218 | security_d_instantiate(dentry, inode); | |
2219 | return NULL; | |
2220 | } | |
2221 | ||
2222 | iput(inode); | |
2223 | return actual; | |
770bfad8 | 2224 | } |
ec4f8605 | 2225 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2226 | |
cdd16d02 | 2227 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2228 | { |
2229 | *buflen -= namelen; | |
2230 | if (*buflen < 0) | |
2231 | return -ENAMETOOLONG; | |
2232 | *buffer -= namelen; | |
2233 | memcpy(*buffer, str, namelen); | |
2234 | return 0; | |
2235 | } | |
2236 | ||
cdd16d02 MS |
2237 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2238 | { | |
2239 | return prepend(buffer, buflen, name->name, name->len); | |
2240 | } | |
2241 | ||
1da177e4 | 2242 | /** |
f2eb6575 MS |
2243 | * Prepend path string to a buffer |
2244 | * | |
9d1bc601 MS |
2245 | * @path: the dentry/vfsmount to report |
2246 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2247 | * @buffer: pointer to the end of the buffer |
2248 | * @buflen: pointer to buffer length | |
552ce544 | 2249 | * |
949854d0 | 2250 | * Caller holds the rename_lock. |
9d1bc601 MS |
2251 | * |
2252 | * If path is not reachable from the supplied root, then the value of | |
2253 | * root is changed (without modifying refcounts). | |
1da177e4 | 2254 | */ |
f2eb6575 MS |
2255 | static int prepend_path(const struct path *path, struct path *root, |
2256 | char **buffer, int *buflen) | |
1da177e4 | 2257 | { |
9d1bc601 MS |
2258 | struct dentry *dentry = path->dentry; |
2259 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2260 | bool slash = false; |
2261 | int error = 0; | |
6092d048 | 2262 | |
99b7db7b | 2263 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2264 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2265 | struct dentry * parent; |
2266 | ||
1da177e4 | 2267 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2268 | /* Global root? */ |
1da177e4 | 2269 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2270 | goto global_root; |
2271 | } | |
2272 | dentry = vfsmnt->mnt_mountpoint; | |
2273 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2274 | continue; |
2275 | } | |
2276 | parent = dentry->d_parent; | |
2277 | prefetch(parent); | |
9abca360 | 2278 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2279 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2280 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2281 | if (!error) |
2282 | error = prepend(buffer, buflen, "/", 1); | |
2283 | if (error) | |
2284 | break; | |
2285 | ||
2286 | slash = true; | |
1da177e4 LT |
2287 | dentry = parent; |
2288 | } | |
2289 | ||
be285c71 | 2290 | out: |
f2eb6575 MS |
2291 | if (!error && !slash) |
2292 | error = prepend(buffer, buflen, "/", 1); | |
2293 | ||
99b7db7b | 2294 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2295 | return error; |
1da177e4 LT |
2296 | |
2297 | global_root: | |
98dc568b MS |
2298 | /* |
2299 | * Filesystems needing to implement special "root names" | |
2300 | * should do so with ->d_dname() | |
2301 | */ | |
2302 | if (IS_ROOT(dentry) && | |
2303 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2304 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2305 | (int) dentry->d_name.len, dentry->d_name.name); | |
2306 | } | |
9d1bc601 MS |
2307 | root->mnt = vfsmnt; |
2308 | root->dentry = dentry; | |
be285c71 | 2309 | goto out; |
f2eb6575 | 2310 | } |
be285c71 | 2311 | |
f2eb6575 MS |
2312 | /** |
2313 | * __d_path - return the path of a dentry | |
2314 | * @path: the dentry/vfsmount to report | |
2315 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2316 | * @buf: buffer to return value in |
f2eb6575 MS |
2317 | * @buflen: buffer length |
2318 | * | |
ffd1f4ed | 2319 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2320 | * |
2321 | * Returns a pointer into the buffer or an error code if the | |
2322 | * path was too long. | |
2323 | * | |
be148247 | 2324 | * "buflen" should be positive. |
f2eb6575 MS |
2325 | * |
2326 | * If path is not reachable from the supplied root, then the value of | |
2327 | * root is changed (without modifying refcounts). | |
2328 | */ | |
2329 | char *__d_path(const struct path *path, struct path *root, | |
2330 | char *buf, int buflen) | |
2331 | { | |
2332 | char *res = buf + buflen; | |
2333 | int error; | |
2334 | ||
2335 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2336 | write_seqlock(&rename_lock); |
f2eb6575 | 2337 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2338 | write_sequnlock(&rename_lock); |
be148247 | 2339 | |
f2eb6575 MS |
2340 | if (error) |
2341 | return ERR_PTR(error); | |
f2eb6575 | 2342 | return res; |
1da177e4 LT |
2343 | } |
2344 | ||
ffd1f4ed MS |
2345 | /* |
2346 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2347 | */ | |
2348 | static int path_with_deleted(const struct path *path, struct path *root, | |
2349 | char **buf, int *buflen) | |
2350 | { | |
2351 | prepend(buf, buflen, "\0", 1); | |
2352 | if (d_unlinked(path->dentry)) { | |
2353 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2354 | if (error) | |
2355 | return error; | |
2356 | } | |
2357 | ||
2358 | return prepend_path(path, root, buf, buflen); | |
2359 | } | |
2360 | ||
8df9d1a4 MS |
2361 | static int prepend_unreachable(char **buffer, int *buflen) |
2362 | { | |
2363 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2364 | } | |
2365 | ||
a03a8a70 JB |
2366 | /** |
2367 | * d_path - return the path of a dentry | |
cf28b486 | 2368 | * @path: path to report |
a03a8a70 JB |
2369 | * @buf: buffer to return value in |
2370 | * @buflen: buffer length | |
2371 | * | |
2372 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2373 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2374 | * | |
52afeefb AV |
2375 | * Returns a pointer into the buffer or an error code if the path was |
2376 | * too long. Note: Callers should use the returned pointer, not the passed | |
2377 | * in buffer, to use the name! The implementation often starts at an offset | |
2378 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2379 | * |
31f3e0b3 | 2380 | * "buflen" should be positive. |
a03a8a70 | 2381 | */ |
20d4fdc1 | 2382 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2383 | { |
ffd1f4ed | 2384 | char *res = buf + buflen; |
6ac08c39 | 2385 | struct path root; |
9d1bc601 | 2386 | struct path tmp; |
ffd1f4ed | 2387 | int error; |
1da177e4 | 2388 | |
c23fbb6b ED |
2389 | /* |
2390 | * We have various synthetic filesystems that never get mounted. On | |
2391 | * these filesystems dentries are never used for lookup purposes, and | |
2392 | * thus don't need to be hashed. They also don't need a name until a | |
2393 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2394 | * below allows us to generate a name for these objects on demand: | |
2395 | */ | |
cf28b486 JB |
2396 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2397 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2398 | |
f7ad3c6b | 2399 | get_fs_root(current->fs, &root); |
949854d0 | 2400 | write_seqlock(&rename_lock); |
9d1bc601 | 2401 | tmp = root; |
ffd1f4ed MS |
2402 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2403 | if (error) | |
2404 | res = ERR_PTR(error); | |
949854d0 | 2405 | write_sequnlock(&rename_lock); |
6ac08c39 | 2406 | path_put(&root); |
1da177e4 LT |
2407 | return res; |
2408 | } | |
ec4f8605 | 2409 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2410 | |
8df9d1a4 MS |
2411 | /** |
2412 | * d_path_with_unreachable - return the path of a dentry | |
2413 | * @path: path to report | |
2414 | * @buf: buffer to return value in | |
2415 | * @buflen: buffer length | |
2416 | * | |
2417 | * The difference from d_path() is that this prepends "(unreachable)" | |
2418 | * to paths which are unreachable from the current process' root. | |
2419 | */ | |
2420 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2421 | { | |
2422 | char *res = buf + buflen; | |
2423 | struct path root; | |
2424 | struct path tmp; | |
2425 | int error; | |
2426 | ||
2427 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2428 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2429 | ||
2430 | get_fs_root(current->fs, &root); | |
949854d0 | 2431 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2432 | tmp = root; |
2433 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2434 | if (!error && !path_equal(&tmp, &root)) | |
2435 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2436 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2437 | path_put(&root); |
2438 | if (error) | |
2439 | res = ERR_PTR(error); | |
2440 | ||
2441 | return res; | |
2442 | } | |
2443 | ||
c23fbb6b ED |
2444 | /* |
2445 | * Helper function for dentry_operations.d_dname() members | |
2446 | */ | |
2447 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2448 | const char *fmt, ...) | |
2449 | { | |
2450 | va_list args; | |
2451 | char temp[64]; | |
2452 | int sz; | |
2453 | ||
2454 | va_start(args, fmt); | |
2455 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2456 | va_end(args); | |
2457 | ||
2458 | if (sz > sizeof(temp) || sz > buflen) | |
2459 | return ERR_PTR(-ENAMETOOLONG); | |
2460 | ||
2461 | buffer += buflen - sz; | |
2462 | return memcpy(buffer, temp, sz); | |
2463 | } | |
2464 | ||
6092d048 RP |
2465 | /* |
2466 | * Write full pathname from the root of the filesystem into the buffer. | |
2467 | */ | |
ec2447c2 | 2468 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2469 | { |
2470 | char *end = buf + buflen; | |
2471 | char *retval; | |
2472 | ||
6092d048 | 2473 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2474 | if (buflen < 1) |
2475 | goto Elong; | |
2476 | /* Get '/' right */ | |
2477 | retval = end-1; | |
2478 | *retval = '/'; | |
2479 | ||
cdd16d02 MS |
2480 | while (!IS_ROOT(dentry)) { |
2481 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2482 | int error; |
6092d048 | 2483 | |
6092d048 | 2484 | prefetch(parent); |
9abca360 NP |
2485 | spin_lock(&dentry->d_lock); |
2486 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2487 | spin_unlock(&dentry->d_lock); | |
2488 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2489 | goto Elong; |
2490 | ||
2491 | retval = end; | |
2492 | dentry = parent; | |
2493 | } | |
c103135c AV |
2494 | return retval; |
2495 | Elong: | |
2496 | return ERR_PTR(-ENAMETOOLONG); | |
2497 | } | |
ec2447c2 NP |
2498 | |
2499 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2500 | { | |
2501 | char *retval; | |
2502 | ||
949854d0 | 2503 | write_seqlock(&rename_lock); |
ec2447c2 | 2504 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2505 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2506 | |
2507 | return retval; | |
2508 | } | |
2509 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2510 | |
2511 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2512 | { | |
2513 | char *p = NULL; | |
2514 | char *retval; | |
2515 | ||
949854d0 | 2516 | write_seqlock(&rename_lock); |
c103135c AV |
2517 | if (d_unlinked(dentry)) { |
2518 | p = buf + buflen; | |
2519 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2520 | goto Elong; | |
2521 | buflen++; | |
2522 | } | |
2523 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2524 | write_sequnlock(&rename_lock); |
c103135c AV |
2525 | if (!IS_ERR(retval) && p) |
2526 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2527 | return retval; |
2528 | Elong: | |
6092d048 RP |
2529 | return ERR_PTR(-ENAMETOOLONG); |
2530 | } | |
2531 | ||
1da177e4 LT |
2532 | /* |
2533 | * NOTE! The user-level library version returns a | |
2534 | * character pointer. The kernel system call just | |
2535 | * returns the length of the buffer filled (which | |
2536 | * includes the ending '\0' character), or a negative | |
2537 | * error value. So libc would do something like | |
2538 | * | |
2539 | * char *getcwd(char * buf, size_t size) | |
2540 | * { | |
2541 | * int retval; | |
2542 | * | |
2543 | * retval = sys_getcwd(buf, size); | |
2544 | * if (retval >= 0) | |
2545 | * return buf; | |
2546 | * errno = -retval; | |
2547 | * return NULL; | |
2548 | * } | |
2549 | */ | |
3cdad428 | 2550 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2551 | { |
552ce544 | 2552 | int error; |
6ac08c39 | 2553 | struct path pwd, root; |
552ce544 | 2554 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2555 | |
2556 | if (!page) | |
2557 | return -ENOMEM; | |
2558 | ||
f7ad3c6b | 2559 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2560 | |
552ce544 | 2561 | error = -ENOENT; |
949854d0 | 2562 | write_seqlock(&rename_lock); |
f3da392e | 2563 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2564 | unsigned long len; |
9d1bc601 | 2565 | struct path tmp = root; |
8df9d1a4 MS |
2566 | char *cwd = page + PAGE_SIZE; |
2567 | int buflen = PAGE_SIZE; | |
1da177e4 | 2568 | |
8df9d1a4 MS |
2569 | prepend(&cwd, &buflen, "\0", 1); |
2570 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2571 | write_sequnlock(&rename_lock); |
552ce544 | 2572 | |
8df9d1a4 | 2573 | if (error) |
552ce544 LT |
2574 | goto out; |
2575 | ||
8df9d1a4 MS |
2576 | /* Unreachable from current root */ |
2577 | if (!path_equal(&tmp, &root)) { | |
2578 | error = prepend_unreachable(&cwd, &buflen); | |
2579 | if (error) | |
2580 | goto out; | |
2581 | } | |
2582 | ||
552ce544 LT |
2583 | error = -ERANGE; |
2584 | len = PAGE_SIZE + page - cwd; | |
2585 | if (len <= size) { | |
2586 | error = len; | |
2587 | if (copy_to_user(buf, cwd, len)) | |
2588 | error = -EFAULT; | |
2589 | } | |
949854d0 NP |
2590 | } else { |
2591 | write_sequnlock(&rename_lock); | |
949854d0 | 2592 | } |
1da177e4 LT |
2593 | |
2594 | out: | |
6ac08c39 JB |
2595 | path_put(&pwd); |
2596 | path_put(&root); | |
1da177e4 LT |
2597 | free_page((unsigned long) page); |
2598 | return error; | |
2599 | } | |
2600 | ||
2601 | /* | |
2602 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2603 | * | |
2604 | * Trivially implemented using the dcache structure | |
2605 | */ | |
2606 | ||
2607 | /** | |
2608 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2609 | * @new_dentry: new dentry | |
2610 | * @old_dentry: old dentry | |
2611 | * | |
2612 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2613 | * Returns 0 otherwise. | |
2614 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2615 | */ | |
2616 | ||
e2761a11 | 2617 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2618 | { |
2619 | int result; | |
949854d0 | 2620 | unsigned seq; |
1da177e4 | 2621 | |
e2761a11 OH |
2622 | if (new_dentry == old_dentry) |
2623 | return 1; | |
2624 | ||
e2761a11 | 2625 | do { |
1da177e4 | 2626 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2627 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2628 | /* |
2629 | * Need rcu_readlock to protect against the d_parent trashing | |
2630 | * due to d_move | |
2631 | */ | |
2632 | rcu_read_lock(); | |
e2761a11 | 2633 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2634 | result = 1; |
e2761a11 OH |
2635 | else |
2636 | result = 0; | |
949854d0 | 2637 | rcu_read_unlock(); |
1da177e4 | 2638 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2639 | |
2640 | return result; | |
2641 | } | |
2642 | ||
2096f759 AV |
2643 | int path_is_under(struct path *path1, struct path *path2) |
2644 | { | |
2645 | struct vfsmount *mnt = path1->mnt; | |
2646 | struct dentry *dentry = path1->dentry; | |
2647 | int res; | |
99b7db7b NP |
2648 | |
2649 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2650 | if (mnt != path2->mnt) { |
2651 | for (;;) { | |
2652 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2653 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2654 | return 0; |
2655 | } | |
2656 | if (mnt->mnt_parent == path2->mnt) | |
2657 | break; | |
2658 | mnt = mnt->mnt_parent; | |
2659 | } | |
2660 | dentry = mnt->mnt_mountpoint; | |
2661 | } | |
2662 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2663 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2664 | return res; |
2665 | } | |
2666 | EXPORT_SYMBOL(path_is_under); | |
2667 | ||
1da177e4 LT |
2668 | void d_genocide(struct dentry *root) |
2669 | { | |
949854d0 | 2670 | struct dentry *this_parent; |
1da177e4 | 2671 | struct list_head *next; |
949854d0 | 2672 | unsigned seq; |
58db63d0 | 2673 | int locked = 0; |
1da177e4 | 2674 | |
949854d0 | 2675 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2676 | again: |
2677 | this_parent = root; | |
2fd6b7f5 | 2678 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2679 | repeat: |
2680 | next = this_parent->d_subdirs.next; | |
2681 | resume: | |
2682 | while (next != &this_parent->d_subdirs) { | |
2683 | struct list_head *tmp = next; | |
5160ee6f | 2684 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2685 | next = tmp->next; |
949854d0 | 2686 | |
da502956 NP |
2687 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2688 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2689 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2690 | continue; |
da502956 | 2691 | } |
1da177e4 | 2692 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2693 | spin_unlock(&this_parent->d_lock); |
2694 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2695 | this_parent = dentry; |
2fd6b7f5 | 2696 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2697 | goto repeat; |
2698 | } | |
949854d0 NP |
2699 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2700 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2701 | dentry->d_count--; | |
2702 | } | |
b7ab39f6 | 2703 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2704 | } |
2705 | if (this_parent != root) { | |
949854d0 NP |
2706 | struct dentry *tmp; |
2707 | struct dentry *child; | |
2708 | ||
2709 | tmp = this_parent->d_parent; | |
2710 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { | |
2711 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2712 | this_parent->d_count--; | |
2713 | } | |
2714 | rcu_read_lock(); | |
b7ab39f6 | 2715 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
2716 | child = this_parent; |
2717 | this_parent = tmp; | |
2fd6b7f5 | 2718 | spin_lock(&this_parent->d_lock); |
949854d0 NP |
2719 | /* might go back up the wrong parent if we have had a rename |
2720 | * or deletion */ | |
2721 | if (this_parent != child->d_parent || | |
58db63d0 | 2722 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 2723 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
2724 | rcu_read_unlock(); |
2725 | goto rename_retry; | |
2726 | } | |
2727 | rcu_read_unlock(); | |
2728 | next = child->d_u.d_child.next; | |
1da177e4 LT |
2729 | goto resume; |
2730 | } | |
2fd6b7f5 | 2731 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2732 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2733 | goto rename_retry; |
58db63d0 NP |
2734 | if (locked) |
2735 | write_sequnlock(&rename_lock); | |
2736 | return; | |
2737 | ||
2738 | rename_retry: | |
2739 | locked = 1; | |
2740 | write_seqlock(&rename_lock); | |
2741 | goto again; | |
1da177e4 LT |
2742 | } |
2743 | ||
2744 | /** | |
2745 | * find_inode_number - check for dentry with name | |
2746 | * @dir: directory to check | |
2747 | * @name: Name to find. | |
2748 | * | |
2749 | * Check whether a dentry already exists for the given name, | |
2750 | * and return the inode number if it has an inode. Otherwise | |
2751 | * 0 is returned. | |
2752 | * | |
2753 | * This routine is used to post-process directory listings for | |
2754 | * filesystems using synthetic inode numbers, and is necessary | |
2755 | * to keep getcwd() working. | |
2756 | */ | |
2757 | ||
2758 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2759 | { | |
2760 | struct dentry * dentry; | |
2761 | ino_t ino = 0; | |
2762 | ||
3e7e241f EB |
2763 | dentry = d_hash_and_lookup(dir, name); |
2764 | if (dentry) { | |
1da177e4 LT |
2765 | if (dentry->d_inode) |
2766 | ino = dentry->d_inode->i_ino; | |
2767 | dput(dentry); | |
2768 | } | |
1da177e4 LT |
2769 | return ino; |
2770 | } | |
ec4f8605 | 2771 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2772 | |
2773 | static __initdata unsigned long dhash_entries; | |
2774 | static int __init set_dhash_entries(char *str) | |
2775 | { | |
2776 | if (!str) | |
2777 | return 0; | |
2778 | dhash_entries = simple_strtoul(str, &str, 0); | |
2779 | return 1; | |
2780 | } | |
2781 | __setup("dhash_entries=", set_dhash_entries); | |
2782 | ||
2783 | static void __init dcache_init_early(void) | |
2784 | { | |
2785 | int loop; | |
2786 | ||
2787 | /* If hashes are distributed across NUMA nodes, defer | |
2788 | * hash allocation until vmalloc space is available. | |
2789 | */ | |
2790 | if (hashdist) | |
2791 | return; | |
2792 | ||
2793 | dentry_hashtable = | |
2794 | alloc_large_system_hash("Dentry cache", | |
2795 | sizeof(struct hlist_head), | |
2796 | dhash_entries, | |
2797 | 13, | |
2798 | HASH_EARLY, | |
2799 | &d_hash_shift, | |
2800 | &d_hash_mask, | |
2801 | 0); | |
2802 | ||
2803 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2804 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2805 | } | |
2806 | ||
74bf17cf | 2807 | static void __init dcache_init(void) |
1da177e4 LT |
2808 | { |
2809 | int loop; | |
2810 | ||
2811 | /* | |
2812 | * A constructor could be added for stable state like the lists, | |
2813 | * but it is probably not worth it because of the cache nature | |
2814 | * of the dcache. | |
2815 | */ | |
0a31bd5f CL |
2816 | dentry_cache = KMEM_CACHE(dentry, |
2817 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2818 | |
8e1f936b | 2819 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2820 | |
2821 | /* Hash may have been set up in dcache_init_early */ | |
2822 | if (!hashdist) | |
2823 | return; | |
2824 | ||
2825 | dentry_hashtable = | |
2826 | alloc_large_system_hash("Dentry cache", | |
2827 | sizeof(struct hlist_head), | |
2828 | dhash_entries, | |
2829 | 13, | |
2830 | 0, | |
2831 | &d_hash_shift, | |
2832 | &d_hash_mask, | |
2833 | 0); | |
2834 | ||
2835 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2836 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2837 | } | |
2838 | ||
2839 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2840 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2841 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2842 | |
1da177e4 LT |
2843 | EXPORT_SYMBOL(d_genocide); |
2844 | ||
1da177e4 LT |
2845 | void __init vfs_caches_init_early(void) |
2846 | { | |
2847 | dcache_init_early(); | |
2848 | inode_init_early(); | |
2849 | } | |
2850 | ||
2851 | void __init vfs_caches_init(unsigned long mempages) | |
2852 | { | |
2853 | unsigned long reserve; | |
2854 | ||
2855 | /* Base hash sizes on available memory, with a reserve equal to | |
2856 | 150% of current kernel size */ | |
2857 | ||
2858 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2859 | mempages -= reserve; | |
2860 | ||
2861 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2862 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2863 | |
74bf17cf DC |
2864 | dcache_init(); |
2865 | inode_init(); | |
1da177e4 | 2866 | files_init(mempages); |
74bf17cf | 2867 | mnt_init(); |
1da177e4 LT |
2868 | bdev_cache_init(); |
2869 | chrdev_init(); | |
2870 | } |