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: | |
57 | * dcache_lock | |
b23fb0a6 NP |
58 | * dcache_inode_lock |
59 | * dentry->d_lock | |
60 | * dcache_lru_lock | |
61 | * dcache_hash_lock | |
789680d1 | 62 | * |
da502956 NP |
63 | * If there is an ancestor relationship: |
64 | * dentry->d_parent->...->d_parent->d_lock | |
65 | * ... | |
66 | * dentry->d_parent->d_lock | |
67 | * dentry->d_lock | |
68 | * | |
69 | * If no ancestor relationship: | |
789680d1 NP |
70 | * if (dentry1 < dentry2) |
71 | * dentry1->d_lock | |
72 | * dentry2->d_lock | |
73 | */ | |
fa3536cc | 74 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
75 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
76 | ||
b23fb0a6 | 77 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_inode_lock); |
789680d1 | 78 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_hash_lock); |
23044507 | 79 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
789680d1 | 80 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lock); |
74c3cbe3 | 81 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 82 | |
b23fb0a6 | 83 | EXPORT_SYMBOL(dcache_inode_lock); |
1da177e4 LT |
84 | EXPORT_SYMBOL(dcache_lock); |
85 | ||
e18b890b | 86 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 LT |
87 | |
88 | #define DNAME_INLINE_LEN (sizeof(struct dentry)-offsetof(struct dentry,d_iname)) | |
89 | ||
90 | /* | |
91 | * This is the single most critical data structure when it comes | |
92 | * to the dcache: the hashtable for lookups. Somebody should try | |
93 | * to make this good - I've just made it work. | |
94 | * | |
95 | * This hash-function tries to avoid losing too many bits of hash | |
96 | * information, yet avoid using a prime hash-size or similar. | |
97 | */ | |
98 | #define D_HASHBITS d_hash_shift | |
99 | #define D_HASHMASK d_hash_mask | |
100 | ||
fa3536cc ED |
101 | static unsigned int d_hash_mask __read_mostly; |
102 | static unsigned int d_hash_shift __read_mostly; | |
103 | static struct hlist_head *dentry_hashtable __read_mostly; | |
1da177e4 LT |
104 | |
105 | /* Statistics gathering. */ | |
106 | struct dentry_stat_t dentry_stat = { | |
107 | .age_limit = 45, | |
108 | }; | |
109 | ||
3e880fb5 | 110 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
111 | |
112 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
113 | static int get_nr_dentry(void) |
114 | { | |
115 | int i; | |
116 | int sum = 0; | |
117 | for_each_possible_cpu(i) | |
118 | sum += per_cpu(nr_dentry, i); | |
119 | return sum < 0 ? 0 : sum; | |
120 | } | |
121 | ||
312d3ca8 CH |
122 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
123 | size_t *lenp, loff_t *ppos) | |
124 | { | |
3e880fb5 | 125 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
126 | return proc_dointvec(table, write, buffer, lenp, ppos); |
127 | } | |
128 | #endif | |
129 | ||
9c82ab9c | 130 | static void __d_free(struct rcu_head *head) |
1da177e4 | 131 | { |
9c82ab9c CH |
132 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
133 | ||
fd217f4d | 134 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
135 | if (dname_external(dentry)) |
136 | kfree(dentry->d_name.name); | |
137 | kmem_cache_free(dentry_cache, dentry); | |
138 | } | |
139 | ||
140 | /* | |
312d3ca8 | 141 | * no dcache_lock, please. |
1da177e4 LT |
142 | */ |
143 | static void d_free(struct dentry *dentry) | |
144 | { | |
b7ab39f6 | 145 | BUG_ON(dentry->d_count); |
3e880fb5 | 146 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
147 | if (dentry->d_op && dentry->d_op->d_release) |
148 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 149 | |
b3423415 | 150 | /* if dentry was never inserted into hash, immediate free is OK */ |
e8462caa | 151 | if (hlist_unhashed(&dentry->d_hash)) |
9c82ab9c | 152 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 153 | else |
9c82ab9c | 154 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
155 | } |
156 | ||
157 | /* | |
158 | * Release the dentry's inode, using the filesystem | |
159 | * d_iput() operation if defined. | |
1da177e4 | 160 | */ |
858119e1 | 161 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 162 | __releases(dentry->d_lock) |
b23fb0a6 | 163 | __releases(dcache_inode_lock) |
31f3e0b3 | 164 | __releases(dcache_lock) |
1da177e4 LT |
165 | { |
166 | struct inode *inode = dentry->d_inode; | |
167 | if (inode) { | |
168 | dentry->d_inode = NULL; | |
169 | list_del_init(&dentry->d_alias); | |
170 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 171 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 172 | spin_unlock(&dcache_lock); |
f805fbda LT |
173 | if (!inode->i_nlink) |
174 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
175 | if (dentry->d_op && dentry->d_op->d_iput) |
176 | dentry->d_op->d_iput(dentry, inode); | |
177 | else | |
178 | iput(inode); | |
179 | } else { | |
180 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 181 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
182 | spin_unlock(&dcache_lock); |
183 | } | |
184 | } | |
185 | ||
da3bbdd4 | 186 | /* |
23044507 | 187 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
188 | */ |
189 | static void dentry_lru_add(struct dentry *dentry) | |
190 | { | |
a4633357 | 191 | if (list_empty(&dentry->d_lru)) { |
23044507 | 192 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
193 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
194 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 195 | dentry_stat.nr_unused++; |
23044507 | 196 | spin_unlock(&dcache_lru_lock); |
a4633357 | 197 | } |
da3bbdd4 KM |
198 | } |
199 | ||
23044507 NP |
200 | static void __dentry_lru_del(struct dentry *dentry) |
201 | { | |
202 | list_del_init(&dentry->d_lru); | |
203 | dentry->d_sb->s_nr_dentry_unused--; | |
204 | dentry_stat.nr_unused--; | |
205 | } | |
206 | ||
da3bbdd4 KM |
207 | static void dentry_lru_del(struct dentry *dentry) |
208 | { | |
209 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
210 | spin_lock(&dcache_lru_lock); |
211 | __dentry_lru_del(dentry); | |
212 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
213 | } |
214 | } | |
215 | ||
a4633357 | 216 | static void dentry_lru_move_tail(struct dentry *dentry) |
da3bbdd4 | 217 | { |
23044507 | 218 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
219 | if (list_empty(&dentry->d_lru)) { |
220 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
221 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 222 | dentry_stat.nr_unused++; |
a4633357 CH |
223 | } else { |
224 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
da3bbdd4 | 225 | } |
23044507 | 226 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
227 | } |
228 | ||
d52b9086 MS |
229 | /** |
230 | * d_kill - kill dentry and return parent | |
231 | * @dentry: dentry to kill | |
232 | * | |
31f3e0b3 | 233 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
234 | * |
235 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 236 | * |
2fd6b7f5 NP |
237 | * dcache_lock and d_lock and d_parent->d_lock must be held by caller, and |
238 | * are dropped by d_kill. | |
d52b9086 | 239 | */ |
2fd6b7f5 | 240 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 241 | __releases(dentry->d_lock) |
2fd6b7f5 | 242 | __releases(parent->d_lock) |
b23fb0a6 | 243 | __releases(dcache_inode_lock) |
31f3e0b3 | 244 | __releases(dcache_lock) |
d52b9086 | 245 | { |
d52b9086 | 246 | list_del(&dentry->d_u.d_child); |
2fd6b7f5 NP |
247 | if (parent) |
248 | spin_unlock(&parent->d_lock); | |
d52b9086 | 249 | dentry_iput(dentry); |
b7ab39f6 NP |
250 | /* |
251 | * dentry_iput drops the locks, at which point nobody (except | |
252 | * transient RCU lookups) can reach this dentry. | |
253 | */ | |
d52b9086 | 254 | d_free(dentry); |
871c0067 | 255 | return parent; |
d52b9086 MS |
256 | } |
257 | ||
789680d1 NP |
258 | /** |
259 | * d_drop - drop a dentry | |
260 | * @dentry: dentry to drop | |
261 | * | |
262 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
263 | * be found through a VFS lookup any more. Note that this is different from | |
264 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
265 | * possible, giving a successful _negative_ lookup, while d_drop will | |
266 | * just make the cache lookup fail. | |
267 | * | |
268 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
269 | * reason (NFS timeouts or autofs deletes). | |
270 | * | |
271 | * __d_drop requires dentry->d_lock. | |
272 | */ | |
273 | void __d_drop(struct dentry *dentry) | |
274 | { | |
275 | if (!(dentry->d_flags & DCACHE_UNHASHED)) { | |
276 | dentry->d_flags |= DCACHE_UNHASHED; | |
277 | spin_lock(&dcache_hash_lock); | |
278 | hlist_del_rcu(&dentry->d_hash); | |
279 | spin_unlock(&dcache_hash_lock); | |
280 | } | |
281 | } | |
282 | EXPORT_SYMBOL(__d_drop); | |
283 | ||
284 | void d_drop(struct dentry *dentry) | |
285 | { | |
286 | spin_lock(&dcache_lock); | |
287 | spin_lock(&dentry->d_lock); | |
288 | __d_drop(dentry); | |
289 | spin_unlock(&dentry->d_lock); | |
290 | spin_unlock(&dcache_lock); | |
291 | } | |
292 | EXPORT_SYMBOL(d_drop); | |
293 | ||
1da177e4 LT |
294 | /* |
295 | * This is dput | |
296 | * | |
297 | * This is complicated by the fact that we do not want to put | |
298 | * dentries that are no longer on any hash chain on the unused | |
299 | * list: we'd much rather just get rid of them immediately. | |
300 | * | |
301 | * However, that implies that we have to traverse the dentry | |
302 | * tree upwards to the parents which might _also_ now be | |
303 | * scheduled for deletion (it may have been only waiting for | |
304 | * its last child to go away). | |
305 | * | |
306 | * This tail recursion is done by hand as we don't want to depend | |
307 | * on the compiler to always get this right (gcc generally doesn't). | |
308 | * Real recursion would eat up our stack space. | |
309 | */ | |
310 | ||
311 | /* | |
312 | * dput - release a dentry | |
313 | * @dentry: dentry to release | |
314 | * | |
315 | * Release a dentry. This will drop the usage count and if appropriate | |
316 | * call the dentry unlink method as well as removing it from the queues and | |
317 | * releasing its resources. If the parent dentries were scheduled for release | |
318 | * they too may now get deleted. | |
319 | * | |
320 | * no dcache lock, please. | |
321 | */ | |
322 | ||
323 | void dput(struct dentry *dentry) | |
324 | { | |
2fd6b7f5 | 325 | struct dentry *parent; |
1da177e4 LT |
326 | if (!dentry) |
327 | return; | |
328 | ||
329 | repeat: | |
b7ab39f6 | 330 | if (dentry->d_count == 1) |
1da177e4 | 331 | might_sleep(); |
1da177e4 | 332 | spin_lock(&dentry->d_lock); |
2fd6b7f5 NP |
333 | if (IS_ROOT(dentry)) |
334 | parent = NULL; | |
335 | else | |
336 | parent = dentry->d_parent; | |
b7ab39f6 NP |
337 | if (dentry->d_count == 1) { |
338 | if (!spin_trylock(&dcache_lock)) { | |
339 | /* | |
340 | * Something of a livelock possibility we could avoid | |
341 | * by taking dcache_lock and trying again, but we | |
342 | * want to reduce dcache_lock anyway so this will | |
343 | * get improved. | |
344 | */ | |
b23fb0a6 | 345 | drop1: |
b7ab39f6 NP |
346 | spin_unlock(&dentry->d_lock); |
347 | goto repeat; | |
348 | } | |
b23fb0a6 NP |
349 | if (!spin_trylock(&dcache_inode_lock)) { |
350 | drop2: | |
2fd6b7f5 | 351 | spin_unlock(&dcache_lock); |
b23fb0a6 NP |
352 | goto drop1; |
353 | } | |
354 | if (parent && !spin_trylock(&parent->d_lock)) { | |
355 | spin_unlock(&dcache_inode_lock); | |
356 | goto drop2; | |
2fd6b7f5 | 357 | } |
b7ab39f6 NP |
358 | } |
359 | dentry->d_count--; | |
360 | if (dentry->d_count) { | |
1da177e4 | 361 | spin_unlock(&dentry->d_lock); |
2fd6b7f5 NP |
362 | if (parent) |
363 | spin_unlock(&parent->d_lock); | |
1da177e4 LT |
364 | spin_unlock(&dcache_lock); |
365 | return; | |
366 | } | |
367 | ||
368 | /* | |
369 | * AV: ->d_delete() is _NOT_ allowed to block now. | |
370 | */ | |
371 | if (dentry->d_op && dentry->d_op->d_delete) { | |
372 | if (dentry->d_op->d_delete(dentry)) | |
373 | goto unhash_it; | |
374 | } | |
265ac902 | 375 | |
1da177e4 LT |
376 | /* Unreachable? Get rid of it */ |
377 | if (d_unhashed(dentry)) | |
378 | goto kill_it; | |
265ac902 NP |
379 | |
380 | /* Otherwise leave it cached and ensure it's on the LRU */ | |
381 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 382 | dentry_lru_add(dentry); |
265ac902 | 383 | |
1da177e4 | 384 | spin_unlock(&dentry->d_lock); |
2fd6b7f5 NP |
385 | if (parent) |
386 | spin_unlock(&parent->d_lock); | |
b23fb0a6 | 387 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
388 | spin_unlock(&dcache_lock); |
389 | return; | |
390 | ||
391 | unhash_it: | |
392 | __d_drop(dentry); | |
d52b9086 | 393 | kill_it: |
da3bbdd4 KM |
394 | /* if dentry was on the d_lru list delete it from there */ |
395 | dentry_lru_del(dentry); | |
2fd6b7f5 | 396 | dentry = d_kill(dentry, parent); |
d52b9086 MS |
397 | if (dentry) |
398 | goto repeat; | |
1da177e4 | 399 | } |
ec4f8605 | 400 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
401 | |
402 | /** | |
403 | * d_invalidate - invalidate a dentry | |
404 | * @dentry: dentry to invalidate | |
405 | * | |
406 | * Try to invalidate the dentry if it turns out to be | |
407 | * possible. If there are other dentries that can be | |
408 | * reached through this one we can't delete it and we | |
409 | * return -EBUSY. On success we return 0. | |
410 | * | |
411 | * no dcache lock. | |
412 | */ | |
413 | ||
414 | int d_invalidate(struct dentry * dentry) | |
415 | { | |
416 | /* | |
417 | * If it's already been dropped, return OK. | |
418 | */ | |
419 | spin_lock(&dcache_lock); | |
da502956 | 420 | spin_lock(&dentry->d_lock); |
1da177e4 | 421 | if (d_unhashed(dentry)) { |
da502956 | 422 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
423 | spin_unlock(&dcache_lock); |
424 | return 0; | |
425 | } | |
426 | /* | |
427 | * Check whether to do a partial shrink_dcache | |
428 | * to get rid of unused child entries. | |
429 | */ | |
430 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 431 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
432 | spin_unlock(&dcache_lock); |
433 | shrink_dcache_parent(dentry); | |
434 | spin_lock(&dcache_lock); | |
da502956 | 435 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
436 | } |
437 | ||
438 | /* | |
439 | * Somebody else still using it? | |
440 | * | |
441 | * If it's a directory, we can't drop it | |
442 | * for fear of somebody re-populating it | |
443 | * with children (even though dropping it | |
444 | * would make it unreachable from the root, | |
445 | * we might still populate it if it was a | |
446 | * working directory or similar). | |
447 | */ | |
b7ab39f6 | 448 | if (dentry->d_count > 1) { |
1da177e4 LT |
449 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
450 | spin_unlock(&dentry->d_lock); | |
451 | spin_unlock(&dcache_lock); | |
452 | return -EBUSY; | |
453 | } | |
454 | } | |
455 | ||
456 | __d_drop(dentry); | |
457 | spin_unlock(&dentry->d_lock); | |
458 | spin_unlock(&dcache_lock); | |
459 | return 0; | |
460 | } | |
ec4f8605 | 461 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 462 | |
b7ab39f6 | 463 | /* This must be called with dcache_lock and d_lock held */ |
23044507 NP |
464 | static inline struct dentry * __dget_locked_dlock(struct dentry *dentry) |
465 | { | |
b7ab39f6 | 466 | dentry->d_count++; |
23044507 NP |
467 | dentry_lru_del(dentry); |
468 | return dentry; | |
469 | } | |
470 | ||
b7ab39f6 | 471 | /* This should be called _only_ with dcache_lock held */ |
1da177e4 LT |
472 | static inline struct dentry * __dget_locked(struct dentry *dentry) |
473 | { | |
23044507 | 474 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 475 | __dget_locked_dlock(dentry); |
23044507 | 476 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
477 | return dentry; |
478 | } | |
479 | ||
b7ab39f6 NP |
480 | struct dentry * dget_locked_dlock(struct dentry *dentry) |
481 | { | |
482 | return __dget_locked_dlock(dentry); | |
483 | } | |
484 | ||
1da177e4 LT |
485 | struct dentry * dget_locked(struct dentry *dentry) |
486 | { | |
487 | return __dget_locked(dentry); | |
488 | } | |
ec4f8605 | 489 | EXPORT_SYMBOL(dget_locked); |
1da177e4 | 490 | |
b7ab39f6 NP |
491 | struct dentry *dget_parent(struct dentry *dentry) |
492 | { | |
493 | struct dentry *ret; | |
494 | ||
495 | repeat: | |
496 | spin_lock(&dentry->d_lock); | |
497 | ret = dentry->d_parent; | |
498 | if (!ret) | |
499 | goto out; | |
500 | if (dentry == ret) { | |
501 | ret->d_count++; | |
502 | goto out; | |
503 | } | |
504 | if (!spin_trylock(&ret->d_lock)) { | |
505 | spin_unlock(&dentry->d_lock); | |
506 | cpu_relax(); | |
507 | goto repeat; | |
508 | } | |
509 | BUG_ON(!ret->d_count); | |
510 | ret->d_count++; | |
511 | spin_unlock(&ret->d_lock); | |
512 | out: | |
513 | spin_unlock(&dentry->d_lock); | |
514 | return ret; | |
515 | } | |
516 | EXPORT_SYMBOL(dget_parent); | |
517 | ||
1da177e4 LT |
518 | /** |
519 | * d_find_alias - grab a hashed alias of inode | |
520 | * @inode: inode in question | |
521 | * @want_discon: flag, used by d_splice_alias, to request | |
522 | * that only a DISCONNECTED alias be returned. | |
523 | * | |
524 | * If inode has a hashed alias, or is a directory and has any alias, | |
525 | * acquire the reference to alias and return it. Otherwise return NULL. | |
526 | * Notice that if inode is a directory there can be only one alias and | |
527 | * it can be unhashed only if it has no children, or if it is the root | |
528 | * of a filesystem. | |
529 | * | |
21c0d8fd | 530 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 531 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 532 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 533 | */ |
da502956 | 534 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 535 | { |
da502956 | 536 | struct dentry *alias, *discon_alias; |
1da177e4 | 537 | |
da502956 NP |
538 | again: |
539 | discon_alias = NULL; | |
540 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
541 | spin_lock(&alias->d_lock); | |
1da177e4 | 542 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 543 | if (IS_ROOT(alias) && |
da502956 | 544 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 545 | discon_alias = alias; |
da502956 NP |
546 | } else if (!want_discon) { |
547 | __dget_locked_dlock(alias); | |
548 | spin_unlock(&alias->d_lock); | |
549 | return alias; | |
550 | } | |
551 | } | |
552 | spin_unlock(&alias->d_lock); | |
553 | } | |
554 | if (discon_alias) { | |
555 | alias = discon_alias; | |
556 | spin_lock(&alias->d_lock); | |
557 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
558 | if (IS_ROOT(alias) && | |
559 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
560 | __dget_locked_dlock(alias); | |
561 | spin_unlock(&alias->d_lock); | |
1da177e4 LT |
562 | return alias; |
563 | } | |
564 | } | |
da502956 NP |
565 | spin_unlock(&alias->d_lock); |
566 | goto again; | |
1da177e4 | 567 | } |
da502956 | 568 | return NULL; |
1da177e4 LT |
569 | } |
570 | ||
da502956 | 571 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 572 | { |
214fda1f DH |
573 | struct dentry *de = NULL; |
574 | ||
575 | if (!list_empty(&inode->i_dentry)) { | |
576 | spin_lock(&dcache_lock); | |
b23fb0a6 | 577 | spin_lock(&dcache_inode_lock); |
214fda1f | 578 | de = __d_find_alias(inode, 0); |
b23fb0a6 | 579 | spin_unlock(&dcache_inode_lock); |
214fda1f DH |
580 | spin_unlock(&dcache_lock); |
581 | } | |
1da177e4 LT |
582 | return de; |
583 | } | |
ec4f8605 | 584 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
585 | |
586 | /* | |
587 | * Try to kill dentries associated with this inode. | |
588 | * WARNING: you must own a reference to inode. | |
589 | */ | |
590 | void d_prune_aliases(struct inode *inode) | |
591 | { | |
0cdca3f9 | 592 | struct dentry *dentry; |
1da177e4 LT |
593 | restart: |
594 | spin_lock(&dcache_lock); | |
b23fb0a6 | 595 | spin_lock(&dcache_inode_lock); |
0cdca3f9 | 596 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 597 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 598 | if (!dentry->d_count) { |
23044507 | 599 | __dget_locked_dlock(dentry); |
1da177e4 LT |
600 | __d_drop(dentry); |
601 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 602 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
603 | spin_unlock(&dcache_lock); |
604 | dput(dentry); | |
605 | goto restart; | |
606 | } | |
607 | spin_unlock(&dentry->d_lock); | |
608 | } | |
b23fb0a6 | 609 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
610 | spin_unlock(&dcache_lock); |
611 | } | |
ec4f8605 | 612 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
613 | |
614 | /* | |
d702ccb3 AM |
615 | * Throw away a dentry - free the inode, dput the parent. This requires that |
616 | * the LRU list has already been removed. | |
617 | * | |
85864e10 MS |
618 | * Try to prune ancestors as well. This is necessary to prevent |
619 | * quadratic behavior of shrink_dcache_parent(), but is also expected | |
620 | * to be beneficial in reducing dentry cache fragmentation. | |
1da177e4 | 621 | */ |
2fd6b7f5 | 622 | static void prune_one_dentry(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 623 | __releases(dentry->d_lock) |
2fd6b7f5 | 624 | __releases(parent->d_lock) |
b23fb0a6 | 625 | __releases(dcache_inode_lock) |
31f3e0b3 | 626 | __releases(dcache_lock) |
1da177e4 | 627 | { |
1da177e4 | 628 | __d_drop(dentry); |
2fd6b7f5 | 629 | dentry = d_kill(dentry, parent); |
d52b9086 MS |
630 | |
631 | /* | |
632 | * Prune ancestors. Locking is simpler than in dput(), | |
633 | * because dcache_lock needs to be taken anyway. | |
634 | */ | |
d52b9086 | 635 | while (dentry) { |
23044507 | 636 | spin_lock(&dcache_lock); |
b23fb0a6 | 637 | spin_lock(&dcache_inode_lock); |
2fd6b7f5 | 638 | again: |
b7ab39f6 | 639 | spin_lock(&dentry->d_lock); |
2fd6b7f5 NP |
640 | if (IS_ROOT(dentry)) |
641 | parent = NULL; | |
642 | else | |
643 | parent = dentry->d_parent; | |
644 | if (parent && !spin_trylock(&parent->d_lock)) { | |
645 | spin_unlock(&dentry->d_lock); | |
646 | goto again; | |
647 | } | |
b7ab39f6 NP |
648 | dentry->d_count--; |
649 | if (dentry->d_count) { | |
2fd6b7f5 NP |
650 | if (parent) |
651 | spin_unlock(&parent->d_lock); | |
b7ab39f6 | 652 | spin_unlock(&dentry->d_lock); |
b23fb0a6 | 653 | spin_unlock(&dcache_inode_lock); |
23044507 | 654 | spin_unlock(&dcache_lock); |
d52b9086 | 655 | return; |
23044507 | 656 | } |
d52b9086 | 657 | |
a4633357 | 658 | dentry_lru_del(dentry); |
d52b9086 | 659 | __d_drop(dentry); |
2fd6b7f5 | 660 | dentry = d_kill(dentry, parent); |
d52b9086 | 661 | } |
1da177e4 LT |
662 | } |
663 | ||
3049cfe2 | 664 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 665 | { |
da3bbdd4 | 666 | struct dentry *dentry; |
da3bbdd4 | 667 | |
3049cfe2 | 668 | while (!list_empty(list)) { |
2fd6b7f5 NP |
669 | struct dentry *parent; |
670 | ||
3049cfe2 | 671 | dentry = list_entry(list->prev, struct dentry, d_lru); |
23044507 NP |
672 | |
673 | if (!spin_trylock(&dentry->d_lock)) { | |
2fd6b7f5 | 674 | relock: |
23044507 NP |
675 | spin_unlock(&dcache_lru_lock); |
676 | cpu_relax(); | |
677 | spin_lock(&dcache_lru_lock); | |
678 | continue; | |
679 | } | |
680 | ||
1da177e4 LT |
681 | /* |
682 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
683 | * the LRU because of laziness during lookup. Do not free |
684 | * it - just keep it off the LRU list. | |
1da177e4 | 685 | */ |
b7ab39f6 | 686 | if (dentry->d_count) { |
2fd6b7f5 | 687 | __dentry_lru_del(dentry); |
da3bbdd4 | 688 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
689 | continue; |
690 | } | |
2fd6b7f5 NP |
691 | if (IS_ROOT(dentry)) |
692 | parent = NULL; | |
693 | else | |
694 | parent = dentry->d_parent; | |
695 | if (parent && !spin_trylock(&parent->d_lock)) { | |
696 | spin_unlock(&dentry->d_lock); | |
697 | goto relock; | |
698 | } | |
699 | __dentry_lru_del(dentry); | |
23044507 NP |
700 | spin_unlock(&dcache_lru_lock); |
701 | ||
2fd6b7f5 | 702 | prune_one_dentry(dentry, parent); |
b23fb0a6 | 703 | /* dcache_lock, dcache_inode_lock and dentry->d_lock dropped */ |
23044507 | 704 | spin_lock(&dcache_lock); |
b23fb0a6 | 705 | spin_lock(&dcache_inode_lock); |
23044507 | 706 | spin_lock(&dcache_lru_lock); |
da3bbdd4 | 707 | } |
3049cfe2 CH |
708 | } |
709 | ||
710 | /** | |
711 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
712 | * @sb: superblock to shrink dentry LRU. | |
713 | * @count: number of entries to prune | |
714 | * @flags: flags to control the dentry processing | |
715 | * | |
716 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
717 | */ | |
718 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
719 | { | |
720 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
721 | struct dentry *dentry; | |
722 | LIST_HEAD(referenced); | |
723 | LIST_HEAD(tmp); | |
724 | int cnt = *count; | |
725 | ||
726 | spin_lock(&dcache_lock); | |
b23fb0a6 | 727 | spin_lock(&dcache_inode_lock); |
23044507 NP |
728 | relock: |
729 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
730 | while (!list_empty(&sb->s_dentry_lru)) { |
731 | dentry = list_entry(sb->s_dentry_lru.prev, | |
732 | struct dentry, d_lru); | |
733 | BUG_ON(dentry->d_sb != sb); | |
734 | ||
23044507 NP |
735 | if (!spin_trylock(&dentry->d_lock)) { |
736 | spin_unlock(&dcache_lru_lock); | |
737 | cpu_relax(); | |
738 | goto relock; | |
739 | } | |
740 | ||
3049cfe2 CH |
741 | /* |
742 | * If we are honouring the DCACHE_REFERENCED flag and the | |
743 | * dentry has this flag set, don't free it. Clear the flag | |
744 | * and put it back on the LRU. | |
745 | */ | |
23044507 NP |
746 | if (flags & DCACHE_REFERENCED && |
747 | dentry->d_flags & DCACHE_REFERENCED) { | |
748 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
749 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 750 | spin_unlock(&dentry->d_lock); |
23044507 NP |
751 | } else { |
752 | list_move_tail(&dentry->d_lru, &tmp); | |
753 | spin_unlock(&dentry->d_lock); | |
754 | if (!--cnt) | |
755 | break; | |
3049cfe2 | 756 | } |
23044507 | 757 | /* XXX: re-add cond_resched_lock when dcache_lock goes away */ |
3049cfe2 CH |
758 | } |
759 | ||
760 | *count = cnt; | |
761 | shrink_dentry_list(&tmp); | |
762 | ||
da3bbdd4 KM |
763 | if (!list_empty(&referenced)) |
764 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 765 | spin_unlock(&dcache_lru_lock); |
b23fb0a6 | 766 | spin_unlock(&dcache_inode_lock); |
da3bbdd4 KM |
767 | spin_unlock(&dcache_lock); |
768 | } | |
769 | ||
770 | /** | |
771 | * prune_dcache - shrink the dcache | |
772 | * @count: number of entries to try to free | |
773 | * | |
774 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
775 | * need to unmount something (at which point we need to unuse all dentries). | |
776 | * | |
777 | * This function may fail to free any resources if all the dentries are in use. | |
778 | */ | |
779 | static void prune_dcache(int count) | |
780 | { | |
dca33252 | 781 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 782 | int w_count; |
86c8749e | 783 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
784 | int prune_ratio; |
785 | int pruned; | |
786 | ||
787 | if (unused == 0 || count == 0) | |
788 | return; | |
789 | spin_lock(&dcache_lock); | |
da3bbdd4 KM |
790 | if (count >= unused) |
791 | prune_ratio = 1; | |
792 | else | |
793 | prune_ratio = unused / count; | |
794 | spin_lock(&sb_lock); | |
dca33252 | 795 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
796 | if (list_empty(&sb->s_instances)) |
797 | continue; | |
da3bbdd4 | 798 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 799 | continue; |
da3bbdd4 KM |
800 | sb->s_count++; |
801 | /* Now, we reclaim unused dentrins with fairness. | |
802 | * We reclaim them same percentage from each superblock. | |
803 | * We calculate number of dentries to scan on this sb | |
804 | * as follows, but the implementation is arranged to avoid | |
805 | * overflows: | |
806 | * number of dentries to scan on this sb = | |
807 | * count * (number of dentries on this sb / | |
808 | * number of dentries in the machine) | |
0feae5c4 | 809 | */ |
da3bbdd4 KM |
810 | spin_unlock(&sb_lock); |
811 | if (prune_ratio != 1) | |
812 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
813 | else | |
814 | w_count = sb->s_nr_dentry_unused; | |
815 | pruned = w_count; | |
0feae5c4 | 816 | /* |
da3bbdd4 KM |
817 | * We need to be sure this filesystem isn't being unmounted, |
818 | * otherwise we could race with generic_shutdown_super(), and | |
819 | * end up holding a reference to an inode while the filesystem | |
820 | * is unmounted. So we try to get s_umount, and make sure | |
821 | * s_root isn't NULL. | |
0feae5c4 | 822 | */ |
da3bbdd4 KM |
823 | if (down_read_trylock(&sb->s_umount)) { |
824 | if ((sb->s_root != NULL) && | |
825 | (!list_empty(&sb->s_dentry_lru))) { | |
826 | spin_unlock(&dcache_lock); | |
827 | __shrink_dcache_sb(sb, &w_count, | |
828 | DCACHE_REFERENCED); | |
829 | pruned -= w_count; | |
830 | spin_lock(&dcache_lock); | |
0feae5c4 | 831 | } |
da3bbdd4 | 832 | up_read(&sb->s_umount); |
0feae5c4 | 833 | } |
da3bbdd4 | 834 | spin_lock(&sb_lock); |
dca33252 AV |
835 | if (p) |
836 | __put_super(p); | |
da3bbdd4 | 837 | count -= pruned; |
dca33252 | 838 | p = sb; |
79893c17 AV |
839 | /* more work left to do? */ |
840 | if (count <= 0) | |
841 | break; | |
1da177e4 | 842 | } |
dca33252 AV |
843 | if (p) |
844 | __put_super(p); | |
da3bbdd4 | 845 | spin_unlock(&sb_lock); |
1da177e4 LT |
846 | spin_unlock(&dcache_lock); |
847 | } | |
848 | ||
1da177e4 LT |
849 | /** |
850 | * shrink_dcache_sb - shrink dcache for a superblock | |
851 | * @sb: superblock | |
852 | * | |
3049cfe2 CH |
853 | * Shrink the dcache for the specified super block. This is used to free |
854 | * the dcache before unmounting a file system. | |
1da177e4 | 855 | */ |
3049cfe2 | 856 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 857 | { |
3049cfe2 CH |
858 | LIST_HEAD(tmp); |
859 | ||
860 | spin_lock(&dcache_lock); | |
b23fb0a6 | 861 | spin_lock(&dcache_inode_lock); |
23044507 | 862 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
863 | while (!list_empty(&sb->s_dentry_lru)) { |
864 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
865 | shrink_dentry_list(&tmp); | |
866 | } | |
23044507 | 867 | spin_unlock(&dcache_lru_lock); |
b23fb0a6 | 868 | spin_unlock(&dcache_inode_lock); |
3049cfe2 | 869 | spin_unlock(&dcache_lock); |
1da177e4 | 870 | } |
ec4f8605 | 871 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 872 | |
c636ebdb DH |
873 | /* |
874 | * destroy a single subtree of dentries for unmount | |
875 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
876 | * locking | |
877 | */ | |
878 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
879 | { | |
880 | struct dentry *parent; | |
f8713576 | 881 | unsigned detached = 0; |
c636ebdb DH |
882 | |
883 | BUG_ON(!IS_ROOT(dentry)); | |
884 | ||
885 | /* detach this root from the system */ | |
886 | spin_lock(&dcache_lock); | |
23044507 | 887 | spin_lock(&dentry->d_lock); |
a4633357 | 888 | dentry_lru_del(dentry); |
c636ebdb | 889 | __d_drop(dentry); |
da502956 | 890 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
891 | spin_unlock(&dcache_lock); |
892 | ||
893 | for (;;) { | |
894 | /* descend to the first leaf in the current subtree */ | |
895 | while (!list_empty(&dentry->d_subdirs)) { | |
896 | struct dentry *loop; | |
897 | ||
898 | /* this is a branch with children - detach all of them | |
899 | * from the system in one go */ | |
900 | spin_lock(&dcache_lock); | |
2fd6b7f5 | 901 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
902 | list_for_each_entry(loop, &dentry->d_subdirs, |
903 | d_u.d_child) { | |
2fd6b7f5 NP |
904 | spin_lock_nested(&loop->d_lock, |
905 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 906 | dentry_lru_del(loop); |
c636ebdb | 907 | __d_drop(loop); |
da502956 | 908 | spin_unlock(&loop->d_lock); |
c636ebdb | 909 | } |
2fd6b7f5 | 910 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
911 | spin_unlock(&dcache_lock); |
912 | ||
913 | /* move to the first child */ | |
914 | dentry = list_entry(dentry->d_subdirs.next, | |
915 | struct dentry, d_u.d_child); | |
916 | } | |
917 | ||
918 | /* consume the dentries from this leaf up through its parents | |
919 | * until we find one with children or run out altogether */ | |
920 | do { | |
921 | struct inode *inode; | |
922 | ||
b7ab39f6 | 923 | if (dentry->d_count != 0) { |
c636ebdb DH |
924 | printk(KERN_ERR |
925 | "BUG: Dentry %p{i=%lx,n=%s}" | |
926 | " still in use (%d)" | |
927 | " [unmount of %s %s]\n", | |
928 | dentry, | |
929 | dentry->d_inode ? | |
930 | dentry->d_inode->i_ino : 0UL, | |
931 | dentry->d_name.name, | |
b7ab39f6 | 932 | dentry->d_count, |
c636ebdb DH |
933 | dentry->d_sb->s_type->name, |
934 | dentry->d_sb->s_id); | |
935 | BUG(); | |
936 | } | |
937 | ||
2fd6b7f5 | 938 | if (IS_ROOT(dentry)) { |
c636ebdb | 939 | parent = NULL; |
2fd6b7f5 NP |
940 | list_del(&dentry->d_u.d_child); |
941 | } else { | |
871c0067 | 942 | parent = dentry->d_parent; |
b7ab39f6 NP |
943 | spin_lock(&parent->d_lock); |
944 | parent->d_count--; | |
2fd6b7f5 | 945 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 946 | spin_unlock(&parent->d_lock); |
871c0067 | 947 | } |
c636ebdb | 948 | |
f8713576 | 949 | detached++; |
c636ebdb DH |
950 | |
951 | inode = dentry->d_inode; | |
952 | if (inode) { | |
953 | dentry->d_inode = NULL; | |
954 | list_del_init(&dentry->d_alias); | |
955 | if (dentry->d_op && dentry->d_op->d_iput) | |
956 | dentry->d_op->d_iput(dentry, inode); | |
957 | else | |
958 | iput(inode); | |
959 | } | |
960 | ||
961 | d_free(dentry); | |
962 | ||
963 | /* finished when we fall off the top of the tree, | |
964 | * otherwise we ascend to the parent and move to the | |
965 | * next sibling if there is one */ | |
966 | if (!parent) | |
312d3ca8 | 967 | return; |
c636ebdb | 968 | dentry = parent; |
c636ebdb DH |
969 | } while (list_empty(&dentry->d_subdirs)); |
970 | ||
971 | dentry = list_entry(dentry->d_subdirs.next, | |
972 | struct dentry, d_u.d_child); | |
973 | } | |
974 | } | |
975 | ||
976 | /* | |
977 | * destroy the dentries attached to a superblock on unmounting | |
978 | * - we don't need to use dentry->d_lock, and only need dcache_lock when | |
979 | * removing the dentry from the system lists and hashes because: | |
980 | * - the superblock is detached from all mountings and open files, so the | |
981 | * dentry trees will not be rearranged by the VFS | |
982 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
983 | * any dentries belonging to this superblock that it comes across | |
984 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
985 | * in this superblock | |
986 | */ | |
987 | void shrink_dcache_for_umount(struct super_block *sb) | |
988 | { | |
989 | struct dentry *dentry; | |
990 | ||
991 | if (down_read_trylock(&sb->s_umount)) | |
992 | BUG(); | |
993 | ||
994 | dentry = sb->s_root; | |
995 | sb->s_root = NULL; | |
b7ab39f6 NP |
996 | spin_lock(&dentry->d_lock); |
997 | dentry->d_count--; | |
998 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
999 | shrink_dcache_for_umount_subtree(dentry); |
1000 | ||
1001 | while (!hlist_empty(&sb->s_anon)) { | |
1002 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
1003 | shrink_dcache_for_umount_subtree(dentry); | |
1004 | } | |
1005 | } | |
1006 | ||
1da177e4 LT |
1007 | /* |
1008 | * Search for at least 1 mount point in the dentry's subdirs. | |
1009 | * We descend to the next level whenever the d_subdirs | |
1010 | * list is non-empty and continue searching. | |
1011 | */ | |
1012 | ||
1013 | /** | |
1014 | * have_submounts - check for mounts over a dentry | |
1015 | * @parent: dentry to check. | |
1016 | * | |
1017 | * Return true if the parent or its subdirectories contain | |
1018 | * a mount point | |
1019 | */ | |
1020 | ||
1021 | int have_submounts(struct dentry *parent) | |
1022 | { | |
1023 | struct dentry *this_parent = parent; | |
1024 | struct list_head *next; | |
1025 | ||
1026 | spin_lock(&dcache_lock); | |
1027 | if (d_mountpoint(parent)) | |
1028 | goto positive; | |
2fd6b7f5 | 1029 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1030 | repeat: |
1031 | next = this_parent->d_subdirs.next; | |
1032 | resume: | |
1033 | while (next != &this_parent->d_subdirs) { | |
1034 | struct list_head *tmp = next; | |
5160ee6f | 1035 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1036 | next = tmp->next; |
2fd6b7f5 NP |
1037 | |
1038 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1039 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1040 | if (d_mountpoint(dentry)) { |
1041 | spin_unlock(&dentry->d_lock); | |
1042 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1043 | goto positive; |
2fd6b7f5 | 1044 | } |
1da177e4 | 1045 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1046 | spin_unlock(&this_parent->d_lock); |
1047 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1048 | this_parent = dentry; |
2fd6b7f5 | 1049 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1050 | goto repeat; |
1051 | } | |
2fd6b7f5 | 1052 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1053 | } |
1054 | /* | |
1055 | * All done at this level ... ascend and resume the search. | |
1056 | */ | |
1057 | if (this_parent != parent) { | |
5160ee6f | 1058 | next = this_parent->d_u.d_child.next; |
2fd6b7f5 | 1059 | spin_unlock(&this_parent->d_lock); |
1da177e4 | 1060 | this_parent = this_parent->d_parent; |
2fd6b7f5 | 1061 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1062 | goto resume; |
1063 | } | |
2fd6b7f5 | 1064 | spin_unlock(&this_parent->d_lock); |
1da177e4 LT |
1065 | spin_unlock(&dcache_lock); |
1066 | return 0; /* No mount points found in tree */ | |
1067 | positive: | |
1068 | spin_unlock(&dcache_lock); | |
1069 | return 1; | |
1070 | } | |
ec4f8605 | 1071 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1072 | |
1073 | /* | |
1074 | * Search the dentry child list for the specified parent, | |
1075 | * and move any unused dentries to the end of the unused | |
1076 | * list for prune_dcache(). We descend to the next level | |
1077 | * whenever the d_subdirs list is non-empty and continue | |
1078 | * searching. | |
1079 | * | |
1080 | * It returns zero iff there are no unused children, | |
1081 | * otherwise it returns the number of children moved to | |
1082 | * the end of the unused list. This may not be the total | |
1083 | * number of unused children, because select_parent can | |
1084 | * drop the lock and return early due to latency | |
1085 | * constraints. | |
1086 | */ | |
1087 | static int select_parent(struct dentry * parent) | |
1088 | { | |
1089 | struct dentry *this_parent = parent; | |
1090 | struct list_head *next; | |
1091 | int found = 0; | |
1092 | ||
1093 | spin_lock(&dcache_lock); | |
2fd6b7f5 | 1094 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1095 | repeat: |
1096 | next = this_parent->d_subdirs.next; | |
1097 | resume: | |
1098 | while (next != &this_parent->d_subdirs) { | |
1099 | struct list_head *tmp = next; | |
5160ee6f | 1100 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1101 | next = tmp->next; |
2fd6b7f5 | 1102 | BUG_ON(this_parent == dentry); |
1da177e4 | 1103 | |
2fd6b7f5 | 1104 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1105 | |
1da177e4 LT |
1106 | /* |
1107 | * move only zero ref count dentries to the end | |
1108 | * of the unused list for prune_dcache | |
1109 | */ | |
b7ab39f6 | 1110 | if (!dentry->d_count) { |
a4633357 | 1111 | dentry_lru_move_tail(dentry); |
1da177e4 | 1112 | found++; |
a4633357 CH |
1113 | } else { |
1114 | dentry_lru_del(dentry); | |
1da177e4 LT |
1115 | } |
1116 | ||
1117 | /* | |
1118 | * We can return to the caller if we have found some (this | |
1119 | * ensures forward progress). We'll be coming back to find | |
1120 | * the rest. | |
1121 | */ | |
2fd6b7f5 NP |
1122 | if (found && need_resched()) { |
1123 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1124 | goto out; |
2fd6b7f5 | 1125 | } |
1da177e4 LT |
1126 | |
1127 | /* | |
1128 | * Descend a level if the d_subdirs list is non-empty. | |
1129 | */ | |
1130 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1131 | spin_unlock(&this_parent->d_lock); |
1132 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1133 | this_parent = dentry; |
2fd6b7f5 | 1134 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1135 | goto repeat; |
1136 | } | |
2fd6b7f5 NP |
1137 | |
1138 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1139 | } |
1140 | /* | |
1141 | * All done at this level ... ascend and resume the search. | |
1142 | */ | |
1143 | if (this_parent != parent) { | |
2fd6b7f5 | 1144 | struct dentry *tmp; |
5160ee6f | 1145 | next = this_parent->d_u.d_child.next; |
2fd6b7f5 NP |
1146 | tmp = this_parent->d_parent; |
1147 | spin_unlock(&this_parent->d_lock); | |
1148 | BUG_ON(tmp == this_parent); | |
1149 | this_parent = tmp; | |
1150 | spin_lock(&this_parent->d_lock); | |
1da177e4 LT |
1151 | goto resume; |
1152 | } | |
1153 | out: | |
2fd6b7f5 | 1154 | spin_unlock(&this_parent->d_lock); |
1da177e4 LT |
1155 | spin_unlock(&dcache_lock); |
1156 | return found; | |
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 NP |
1255 | spin_lock(&dcache_lock); |
1256 | spin_lock(&parent->d_lock); | |
1257 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1258 | dentry->d_parent = dget_dlock(parent); | |
1da177e4 | 1259 | dentry->d_sb = parent->d_sb; |
5160ee6f | 1260 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
2fd6b7f5 NP |
1261 | spin_unlock(&dentry->d_lock); |
1262 | spin_unlock(&parent->d_lock); | |
1263 | spin_unlock(&dcache_lock); | |
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 | /* the caller must hold dcache_lock */ |
1284 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) | |
1285 | { | |
b23fb0a6 | 1286 | spin_lock(&dentry->d_lock); |
360da900 OH |
1287 | if (inode) |
1288 | list_add(&dentry->d_alias, &inode->i_dentry); | |
1289 | dentry->d_inode = inode; | |
b23fb0a6 | 1290 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1291 | fsnotify_d_instantiate(dentry, inode); |
1292 | } | |
1293 | ||
1da177e4 LT |
1294 | /** |
1295 | * d_instantiate - fill in inode information for a dentry | |
1296 | * @entry: dentry to complete | |
1297 | * @inode: inode to attach to this dentry | |
1298 | * | |
1299 | * Fill in inode information in the entry. | |
1300 | * | |
1301 | * This turns negative dentries into productive full members | |
1302 | * of society. | |
1303 | * | |
1304 | * NOTE! This assumes that the inode count has been incremented | |
1305 | * (or otherwise set) by the caller to indicate that it is now | |
1306 | * in use by the dcache. | |
1307 | */ | |
1308 | ||
1309 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1310 | { | |
28133c7b | 1311 | BUG_ON(!list_empty(&entry->d_alias)); |
1da177e4 | 1312 | spin_lock(&dcache_lock); |
b23fb0a6 | 1313 | spin_lock(&dcache_inode_lock); |
360da900 | 1314 | __d_instantiate(entry, inode); |
b23fb0a6 | 1315 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1316 | spin_unlock(&dcache_lock); |
1317 | security_d_instantiate(entry, inode); | |
1318 | } | |
ec4f8605 | 1319 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1320 | |
1321 | /** | |
1322 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1323 | * @entry: dentry to instantiate | |
1324 | * @inode: inode to attach to this dentry | |
1325 | * | |
1326 | * Fill in inode information in the entry. On success, it returns NULL. | |
1327 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1328 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1329 | * |
1330 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1331 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1332 | * |
1333 | * This also assumes that the inode count has been incremented | |
1334 | * (or otherwise set) by the caller to indicate that it is now | |
1335 | * in use by the dcache. | |
1da177e4 | 1336 | */ |
770bfad8 DH |
1337 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1338 | struct inode *inode) | |
1da177e4 LT |
1339 | { |
1340 | struct dentry *alias; | |
1341 | int len = entry->d_name.len; | |
1342 | const char *name = entry->d_name.name; | |
1343 | unsigned int hash = entry->d_name.hash; | |
1344 | ||
770bfad8 | 1345 | if (!inode) { |
360da900 | 1346 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1347 | return NULL; |
1348 | } | |
1349 | ||
1da177e4 LT |
1350 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1351 | struct qstr *qstr = &alias->d_name; | |
1352 | ||
9abca360 NP |
1353 | /* |
1354 | * Don't need alias->d_lock here, because aliases with | |
1355 | * d_parent == entry->d_parent are not subject to name or | |
1356 | * parent changes, because the parent inode i_mutex is held. | |
1357 | */ | |
1da177e4 LT |
1358 | if (qstr->hash != hash) |
1359 | continue; | |
1360 | if (alias->d_parent != entry->d_parent) | |
1361 | continue; | |
1362 | if (qstr->len != len) | |
1363 | continue; | |
1364 | if (memcmp(qstr->name, name, len)) | |
1365 | continue; | |
1366 | dget_locked(alias); | |
1da177e4 LT |
1367 | return alias; |
1368 | } | |
770bfad8 | 1369 | |
360da900 | 1370 | __d_instantiate(entry, inode); |
1da177e4 LT |
1371 | return NULL; |
1372 | } | |
770bfad8 DH |
1373 | |
1374 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1375 | { | |
1376 | struct dentry *result; | |
1377 | ||
1378 | BUG_ON(!list_empty(&entry->d_alias)); | |
1379 | ||
1380 | spin_lock(&dcache_lock); | |
b23fb0a6 | 1381 | spin_lock(&dcache_inode_lock); |
770bfad8 | 1382 | result = __d_instantiate_unique(entry, inode); |
b23fb0a6 | 1383 | spin_unlock(&dcache_inode_lock); |
770bfad8 DH |
1384 | spin_unlock(&dcache_lock); |
1385 | ||
1386 | if (!result) { | |
1387 | security_d_instantiate(entry, inode); | |
1388 | return NULL; | |
1389 | } | |
1390 | ||
1391 | BUG_ON(!d_unhashed(result)); | |
1392 | iput(inode); | |
1393 | return result; | |
1394 | } | |
1395 | ||
1da177e4 LT |
1396 | EXPORT_SYMBOL(d_instantiate_unique); |
1397 | ||
1398 | /** | |
1399 | * d_alloc_root - allocate root dentry | |
1400 | * @root_inode: inode to allocate the root for | |
1401 | * | |
1402 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1403 | * instantiated and returned. %NULL is returned if there is insufficient | |
1404 | * memory or the inode passed is %NULL. | |
1405 | */ | |
1406 | ||
1407 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1408 | { | |
1409 | struct dentry *res = NULL; | |
1410 | ||
1411 | if (root_inode) { | |
1412 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1413 | ||
1414 | res = d_alloc(NULL, &name); | |
1415 | if (res) { | |
1416 | res->d_sb = root_inode->i_sb; | |
1417 | res->d_parent = res; | |
1418 | d_instantiate(res, root_inode); | |
1419 | } | |
1420 | } | |
1421 | return res; | |
1422 | } | |
ec4f8605 | 1423 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 LT |
1424 | |
1425 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1426 | unsigned long hash) | |
1427 | { | |
1428 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1429 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1430 | return dentry_hashtable + (hash & D_HASHMASK); | |
1431 | } | |
1432 | ||
4ea3ada2 CH |
1433 | /** |
1434 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1435 | * @inode: inode to allocate the dentry for | |
1436 | * | |
1437 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1438 | * similar open by handle operations. The returned dentry may be anonymous, | |
1439 | * or may have a full name (if the inode was already in the cache). | |
1440 | * | |
1441 | * When called on a directory inode, we must ensure that the inode only ever | |
1442 | * has one dentry. If a dentry is found, that is returned instead of | |
1443 | * allocating a new one. | |
1444 | * | |
1445 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1446 | * to the dentry. In case of an error the reference on the inode is released. |
1447 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1448 | * be passed in and will be the error will be propagate to the return value, | |
1449 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1450 | */ |
1451 | struct dentry *d_obtain_alias(struct inode *inode) | |
1452 | { | |
9308a612 CH |
1453 | static const struct qstr anonstring = { .name = "" }; |
1454 | struct dentry *tmp; | |
1455 | struct dentry *res; | |
4ea3ada2 CH |
1456 | |
1457 | if (!inode) | |
44003728 | 1458 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1459 | if (IS_ERR(inode)) |
1460 | return ERR_CAST(inode); | |
1461 | ||
9308a612 CH |
1462 | res = d_find_alias(inode); |
1463 | if (res) | |
1464 | goto out_iput; | |
1465 | ||
1466 | tmp = d_alloc(NULL, &anonstring); | |
1467 | if (!tmp) { | |
1468 | res = ERR_PTR(-ENOMEM); | |
1469 | goto out_iput; | |
4ea3ada2 | 1470 | } |
9308a612 CH |
1471 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1472 | ||
1473 | spin_lock(&dcache_lock); | |
b23fb0a6 | 1474 | spin_lock(&dcache_inode_lock); |
9308a612 CH |
1475 | res = __d_find_alias(inode, 0); |
1476 | if (res) { | |
b23fb0a6 | 1477 | spin_unlock(&dcache_inode_lock); |
9308a612 CH |
1478 | spin_unlock(&dcache_lock); |
1479 | dput(tmp); | |
1480 | goto out_iput; | |
1481 | } | |
1482 | ||
1483 | /* attach a disconnected dentry */ | |
1484 | spin_lock(&tmp->d_lock); | |
1485 | tmp->d_sb = inode->i_sb; | |
1486 | tmp->d_inode = inode; | |
1487 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1488 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1489 | list_add(&tmp->d_alias, &inode->i_dentry); | |
789680d1 | 1490 | spin_lock(&dcache_hash_lock); |
9308a612 | 1491 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); |
789680d1 | 1492 | spin_unlock(&dcache_hash_lock); |
9308a612 | 1493 | spin_unlock(&tmp->d_lock); |
b23fb0a6 | 1494 | spin_unlock(&dcache_inode_lock); |
9308a612 CH |
1495 | |
1496 | spin_unlock(&dcache_lock); | |
1497 | return tmp; | |
1498 | ||
1499 | out_iput: | |
1500 | iput(inode); | |
1501 | return res; | |
4ea3ada2 | 1502 | } |
adc48720 | 1503 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1504 | |
1505 | /** | |
1506 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1507 | * @inode: the inode which may have a disconnected dentry | |
1508 | * @dentry: a negative dentry which we want to point to the inode. | |
1509 | * | |
1510 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1511 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1512 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1513 | * | |
1514 | * This is needed in the lookup routine of any filesystem that is exportable | |
1515 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1516 | * | |
1517 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1518 | * is returned. This matches the expected return value of ->lookup. | |
1519 | * | |
1520 | */ | |
1521 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1522 | { | |
1523 | struct dentry *new = NULL; | |
1524 | ||
21c0d8fd | 1525 | if (inode && S_ISDIR(inode->i_mode)) { |
1da177e4 | 1526 | spin_lock(&dcache_lock); |
b23fb0a6 | 1527 | spin_lock(&dcache_inode_lock); |
1da177e4 LT |
1528 | new = __d_find_alias(inode, 1); |
1529 | if (new) { | |
1530 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
b23fb0a6 | 1531 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1532 | spin_unlock(&dcache_lock); |
1533 | security_d_instantiate(new, inode); | |
1da177e4 LT |
1534 | d_move(new, dentry); |
1535 | iput(inode); | |
1536 | } else { | |
360da900 OH |
1537 | /* already taking dcache_lock, so d_add() by hand */ |
1538 | __d_instantiate(dentry, inode); | |
b23fb0a6 | 1539 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1540 | spin_unlock(&dcache_lock); |
1541 | security_d_instantiate(dentry, inode); | |
1542 | d_rehash(dentry); | |
1543 | } | |
1544 | } else | |
1545 | d_add(dentry, inode); | |
1546 | return new; | |
1547 | } | |
ec4f8605 | 1548 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1549 | |
9403540c BN |
1550 | /** |
1551 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1552 | * @inode: the inode case-insensitive lookup has found | |
1553 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1554 | * @name: the case-exact name to be associated with the returned dentry | |
1555 | * | |
1556 | * This is to avoid filling the dcache with case-insensitive names to the | |
1557 | * same inode, only the actual correct case is stored in the dcache for | |
1558 | * case-insensitive filesystems. | |
1559 | * | |
1560 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1561 | * already exists in in the dcache, use it and return it. | |
1562 | * | |
1563 | * If no entry exists with the exact case name, allocate new dentry with | |
1564 | * the exact case, and return the spliced entry. | |
1565 | */ | |
e45b590b | 1566 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1567 | struct qstr *name) |
1568 | { | |
1569 | int error; | |
1570 | struct dentry *found; | |
1571 | struct dentry *new; | |
1572 | ||
b6520c81 CH |
1573 | /* |
1574 | * First check if a dentry matching the name already exists, | |
1575 | * if not go ahead and create it now. | |
1576 | */ | |
9403540c | 1577 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1578 | if (!found) { |
1579 | new = d_alloc(dentry->d_parent, name); | |
1580 | if (!new) { | |
1581 | error = -ENOMEM; | |
1582 | goto err_out; | |
1583 | } | |
b6520c81 | 1584 | |
9403540c BN |
1585 | found = d_splice_alias(inode, new); |
1586 | if (found) { | |
1587 | dput(new); | |
1588 | return found; | |
1589 | } | |
1590 | return new; | |
1591 | } | |
b6520c81 CH |
1592 | |
1593 | /* | |
1594 | * If a matching dentry exists, and it's not negative use it. | |
1595 | * | |
1596 | * Decrement the reference count to balance the iget() done | |
1597 | * earlier on. | |
1598 | */ | |
9403540c BN |
1599 | if (found->d_inode) { |
1600 | if (unlikely(found->d_inode != inode)) { | |
1601 | /* This can't happen because bad inodes are unhashed. */ | |
1602 | BUG_ON(!is_bad_inode(inode)); | |
1603 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1604 | } | |
9403540c BN |
1605 | iput(inode); |
1606 | return found; | |
1607 | } | |
b6520c81 | 1608 | |
9403540c BN |
1609 | /* |
1610 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1611 | * already has a dentry. |
9403540c | 1612 | */ |
9403540c | 1613 | spin_lock(&dcache_lock); |
b23fb0a6 | 1614 | spin_lock(&dcache_inode_lock); |
b6520c81 | 1615 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1616 | __d_instantiate(found, inode); |
b23fb0a6 | 1617 | spin_unlock(&dcache_inode_lock); |
9403540c BN |
1618 | spin_unlock(&dcache_lock); |
1619 | security_d_instantiate(found, inode); | |
1620 | return found; | |
1621 | } | |
b6520c81 | 1622 | |
9403540c | 1623 | /* |
b6520c81 CH |
1624 | * In case a directory already has a (disconnected) entry grab a |
1625 | * reference to it, move it in place and use it. | |
9403540c BN |
1626 | */ |
1627 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
1628 | dget_locked(new); | |
b23fb0a6 | 1629 | spin_unlock(&dcache_inode_lock); |
9403540c | 1630 | spin_unlock(&dcache_lock); |
9403540c | 1631 | security_d_instantiate(found, inode); |
9403540c | 1632 | d_move(new, found); |
9403540c | 1633 | iput(inode); |
9403540c | 1634 | dput(found); |
9403540c BN |
1635 | return new; |
1636 | ||
1637 | err_out: | |
1638 | iput(inode); | |
1639 | return ERR_PTR(error); | |
1640 | } | |
ec4f8605 | 1641 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 LT |
1642 | |
1643 | /** | |
1644 | * d_lookup - search for a dentry | |
1645 | * @parent: parent dentry | |
1646 | * @name: qstr of name we wish to find | |
b04f784e | 1647 | * Returns: dentry, or NULL |
1da177e4 | 1648 | * |
b04f784e NP |
1649 | * d_lookup searches the children of the parent dentry for the name in |
1650 | * question. If the dentry is found its reference count is incremented and the | |
1651 | * dentry is returned. The caller must use dput to free the entry when it has | |
1652 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1653 | */ |
1da177e4 LT |
1654 | struct dentry * d_lookup(struct dentry * parent, struct qstr * name) |
1655 | { | |
1656 | struct dentry * dentry = NULL; | |
1657 | unsigned long seq; | |
1658 | ||
1659 | do { | |
1660 | seq = read_seqbegin(&rename_lock); | |
1661 | dentry = __d_lookup(parent, name); | |
1662 | if (dentry) | |
1663 | break; | |
1664 | } while (read_seqretry(&rename_lock, seq)); | |
1665 | return dentry; | |
1666 | } | |
ec4f8605 | 1667 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1668 | |
b04f784e NP |
1669 | /* |
1670 | * __d_lookup - search for a dentry (racy) | |
1671 | * @parent: parent dentry | |
1672 | * @name: qstr of name we wish to find | |
1673 | * Returns: dentry, or NULL | |
1674 | * | |
1675 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1676 | * false-negative result due to unrelated rename activity. | |
1677 | * | |
1678 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1679 | * however it must be used carefully, eg. with a following d_lookup in | |
1680 | * the case of failure. | |
1681 | * | |
1682 | * __d_lookup callers must be commented. | |
1683 | */ | |
1da177e4 LT |
1684 | struct dentry * __d_lookup(struct dentry * parent, struct qstr * name) |
1685 | { | |
1686 | unsigned int len = name->len; | |
1687 | unsigned int hash = name->hash; | |
1688 | const unsigned char *str = name->name; | |
1689 | struct hlist_head *head = d_hash(parent,hash); | |
1690 | struct dentry *found = NULL; | |
1691 | struct hlist_node *node; | |
665a7583 | 1692 | struct dentry *dentry; |
1da177e4 | 1693 | |
b04f784e NP |
1694 | /* |
1695 | * The hash list is protected using RCU. | |
1696 | * | |
1697 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1698 | * with d_move(). | |
1699 | * | |
1700 | * It is possible that concurrent renames can mess up our list | |
1701 | * walk here and result in missing our dentry, resulting in the | |
1702 | * false-negative result. d_lookup() protects against concurrent | |
1703 | * renames using rename_lock seqlock. | |
1704 | * | |
1705 | * See Documentation/vfs/dcache-locking.txt for more details. | |
1706 | */ | |
1da177e4 LT |
1707 | rcu_read_lock(); |
1708 | ||
665a7583 | 1709 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
1da177e4 LT |
1710 | struct qstr *qstr; |
1711 | ||
1da177e4 LT |
1712 | if (dentry->d_name.hash != hash) |
1713 | continue; | |
1714 | if (dentry->d_parent != parent) | |
1715 | continue; | |
1716 | ||
1717 | spin_lock(&dentry->d_lock); | |
1718 | ||
1719 | /* | |
1720 | * Recheck the dentry after taking the lock - d_move may have | |
b04f784e NP |
1721 | * changed things. Don't bother checking the hash because |
1722 | * we're about to compare the whole name anyway. | |
1da177e4 LT |
1723 | */ |
1724 | if (dentry->d_parent != parent) | |
1725 | goto next; | |
1726 | ||
d0185c08 LT |
1727 | /* non-existing due to RCU? */ |
1728 | if (d_unhashed(dentry)) | |
1729 | goto next; | |
1730 | ||
1da177e4 LT |
1731 | /* |
1732 | * It is safe to compare names since d_move() cannot | |
1733 | * change the qstr (protected by d_lock). | |
1734 | */ | |
1735 | qstr = &dentry->d_name; | |
1736 | if (parent->d_op && parent->d_op->d_compare) { | |
621e155a NP |
1737 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1738 | dentry, dentry->d_inode, | |
1739 | qstr->len, qstr->name, name)) | |
1da177e4 LT |
1740 | goto next; |
1741 | } else { | |
1742 | if (qstr->len != len) | |
1743 | goto next; | |
1744 | if (memcmp(qstr->name, str, len)) | |
1745 | goto next; | |
1746 | } | |
1747 | ||
b7ab39f6 | 1748 | dentry->d_count++; |
d0185c08 | 1749 | found = dentry; |
1da177e4 LT |
1750 | spin_unlock(&dentry->d_lock); |
1751 | break; | |
1752 | next: | |
1753 | spin_unlock(&dentry->d_lock); | |
1754 | } | |
1755 | rcu_read_unlock(); | |
1756 | ||
1757 | return found; | |
1758 | } | |
1759 | ||
3e7e241f EB |
1760 | /** |
1761 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1762 | * @dir: Directory to search in | |
1763 | * @name: qstr of name we wish to find | |
1764 | * | |
1765 | * On hash failure or on lookup failure NULL is returned. | |
1766 | */ | |
1767 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1768 | { | |
1769 | struct dentry *dentry = NULL; | |
1770 | ||
1771 | /* | |
1772 | * Check for a fs-specific hash function. Note that we must | |
1773 | * calculate the standard hash first, as the d_op->d_hash() | |
1774 | * routine may choose to leave the hash value unchanged. | |
1775 | */ | |
1776 | name->hash = full_name_hash(name->name, name->len); | |
1777 | if (dir->d_op && dir->d_op->d_hash) { | |
b1e6a015 | 1778 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1779 | goto out; |
1780 | } | |
1781 | dentry = d_lookup(dir, name); | |
1782 | out: | |
1783 | return dentry; | |
1784 | } | |
1785 | ||
1da177e4 | 1786 | /** |
786a5e15 | 1787 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 LT |
1788 | * @dentry: The dentry alleged to be valid child of @dparent |
1789 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1790 | * |
1791 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1792 | * This is used by ncpfs in its readdir implementation. | |
1793 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1794 | * |
1795 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1796 | */ |
d3a23e16 | 1797 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1798 | { |
786a5e15 | 1799 | struct dentry *child; |
d3a23e16 NP |
1800 | |
1801 | spin_lock(&dcache_lock); | |
2fd6b7f5 | 1802 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
1803 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
1804 | if (dentry == child) { | |
2fd6b7f5 NP |
1805 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
1806 | __dget_locked_dlock(dentry); | |
1807 | spin_unlock(&dentry->d_lock); | |
1808 | spin_unlock(&dparent->d_lock); | |
d3a23e16 | 1809 | spin_unlock(&dcache_lock); |
1da177e4 LT |
1810 | return 1; |
1811 | } | |
1812 | } | |
2fd6b7f5 | 1813 | spin_unlock(&dparent->d_lock); |
d3a23e16 | 1814 | spin_unlock(&dcache_lock); |
786a5e15 | 1815 | |
1da177e4 LT |
1816 | return 0; |
1817 | } | |
ec4f8605 | 1818 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1819 | |
1820 | /* | |
1821 | * When a file is deleted, we have two options: | |
1822 | * - turn this dentry into a negative dentry | |
1823 | * - unhash this dentry and free it. | |
1824 | * | |
1825 | * Usually, we want to just turn this into | |
1826 | * a negative dentry, but if anybody else is | |
1827 | * currently using the dentry or the inode | |
1828 | * we can't do that and we fall back on removing | |
1829 | * it from the hash queues and waiting for | |
1830 | * it to be deleted later when it has no users | |
1831 | */ | |
1832 | ||
1833 | /** | |
1834 | * d_delete - delete a dentry | |
1835 | * @dentry: The dentry to delete | |
1836 | * | |
1837 | * Turn the dentry into a negative dentry if possible, otherwise | |
1838 | * remove it from the hash queues so it can be deleted later | |
1839 | */ | |
1840 | ||
1841 | void d_delete(struct dentry * dentry) | |
1842 | { | |
7a91bf7f | 1843 | int isdir = 0; |
1da177e4 LT |
1844 | /* |
1845 | * Are we the only user? | |
1846 | */ | |
1847 | spin_lock(&dcache_lock); | |
b23fb0a6 | 1848 | spin_lock(&dcache_inode_lock); |
1da177e4 | 1849 | spin_lock(&dentry->d_lock); |
7a91bf7f | 1850 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
b7ab39f6 | 1851 | if (dentry->d_count == 1) { |
13e3c5e5 | 1852 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
1da177e4 | 1853 | dentry_iput(dentry); |
7a91bf7f | 1854 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1855 | return; |
1856 | } | |
1857 | ||
1858 | if (!d_unhashed(dentry)) | |
1859 | __d_drop(dentry); | |
1860 | ||
1861 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 1862 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 1863 | spin_unlock(&dcache_lock); |
7a91bf7f JM |
1864 | |
1865 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1866 | } |
ec4f8605 | 1867 | EXPORT_SYMBOL(d_delete); |
1da177e4 LT |
1868 | |
1869 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
1870 | { | |
1871 | ||
1872 | entry->d_flags &= ~DCACHE_UNHASHED; | |
1873 | hlist_add_head_rcu(&entry->d_hash, list); | |
1874 | } | |
1875 | ||
770bfad8 DH |
1876 | static void _d_rehash(struct dentry * entry) |
1877 | { | |
1878 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1879 | } | |
1880 | ||
1da177e4 LT |
1881 | /** |
1882 | * d_rehash - add an entry back to the hash | |
1883 | * @entry: dentry to add to the hash | |
1884 | * | |
1885 | * Adds a dentry to the hash according to its name. | |
1886 | */ | |
1887 | ||
1888 | void d_rehash(struct dentry * entry) | |
1889 | { | |
1da177e4 LT |
1890 | spin_lock(&dcache_lock); |
1891 | spin_lock(&entry->d_lock); | |
789680d1 | 1892 | spin_lock(&dcache_hash_lock); |
770bfad8 | 1893 | _d_rehash(entry); |
789680d1 | 1894 | spin_unlock(&dcache_hash_lock); |
1da177e4 LT |
1895 | spin_unlock(&entry->d_lock); |
1896 | spin_unlock(&dcache_lock); | |
1897 | } | |
ec4f8605 | 1898 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 1899 | |
fb2d5b86 NP |
1900 | /** |
1901 | * dentry_update_name_case - update case insensitive dentry with a new name | |
1902 | * @dentry: dentry to be updated | |
1903 | * @name: new name | |
1904 | * | |
1905 | * Update a case insensitive dentry with new case of name. | |
1906 | * | |
1907 | * dentry must have been returned by d_lookup with name @name. Old and new | |
1908 | * name lengths must match (ie. no d_compare which allows mismatched name | |
1909 | * lengths). | |
1910 | * | |
1911 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
1912 | * keep renames and concurrent inserts, and readdir(2) away). | |
1913 | */ | |
1914 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
1915 | { | |
1916 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
1917 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | |
1918 | ||
1919 | spin_lock(&dcache_lock); | |
1920 | spin_lock(&dentry->d_lock); | |
1921 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); | |
1922 | spin_unlock(&dentry->d_lock); | |
1923 | spin_unlock(&dcache_lock); | |
1924 | } | |
1925 | EXPORT_SYMBOL(dentry_update_name_case); | |
1926 | ||
1da177e4 LT |
1927 | static void switch_names(struct dentry *dentry, struct dentry *target) |
1928 | { | |
1929 | if (dname_external(target)) { | |
1930 | if (dname_external(dentry)) { | |
1931 | /* | |
1932 | * Both external: swap the pointers | |
1933 | */ | |
9a8d5bb4 | 1934 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
1935 | } else { |
1936 | /* | |
1937 | * dentry:internal, target:external. Steal target's | |
1938 | * storage and make target internal. | |
1939 | */ | |
321bcf92 BF |
1940 | memcpy(target->d_iname, dentry->d_name.name, |
1941 | dentry->d_name.len + 1); | |
1da177e4 LT |
1942 | dentry->d_name.name = target->d_name.name; |
1943 | target->d_name.name = target->d_iname; | |
1944 | } | |
1945 | } else { | |
1946 | if (dname_external(dentry)) { | |
1947 | /* | |
1948 | * dentry:external, target:internal. Give dentry's | |
1949 | * storage to target and make dentry internal | |
1950 | */ | |
1951 | memcpy(dentry->d_iname, target->d_name.name, | |
1952 | target->d_name.len + 1); | |
1953 | target->d_name.name = dentry->d_name.name; | |
1954 | dentry->d_name.name = dentry->d_iname; | |
1955 | } else { | |
1956 | /* | |
1957 | * Both are internal. Just copy target to dentry | |
1958 | */ | |
1959 | memcpy(dentry->d_iname, target->d_name.name, | |
1960 | target->d_name.len + 1); | |
dc711ca3 AV |
1961 | dentry->d_name.len = target->d_name.len; |
1962 | return; | |
1da177e4 LT |
1963 | } |
1964 | } | |
9a8d5bb4 | 1965 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
1966 | } |
1967 | ||
2fd6b7f5 NP |
1968 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
1969 | { | |
1970 | /* | |
1971 | * XXXX: do we really need to take target->d_lock? | |
1972 | */ | |
1973 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
1974 | spin_lock(&target->d_parent->d_lock); | |
1975 | else { | |
1976 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
1977 | spin_lock(&dentry->d_parent->d_lock); | |
1978 | spin_lock_nested(&target->d_parent->d_lock, | |
1979 | DENTRY_D_LOCK_NESTED); | |
1980 | } else { | |
1981 | spin_lock(&target->d_parent->d_lock); | |
1982 | spin_lock_nested(&dentry->d_parent->d_lock, | |
1983 | DENTRY_D_LOCK_NESTED); | |
1984 | } | |
1985 | } | |
1986 | if (target < dentry) { | |
1987 | spin_lock_nested(&target->d_lock, 2); | |
1988 | spin_lock_nested(&dentry->d_lock, 3); | |
1989 | } else { | |
1990 | spin_lock_nested(&dentry->d_lock, 2); | |
1991 | spin_lock_nested(&target->d_lock, 3); | |
1992 | } | |
1993 | } | |
1994 | ||
1995 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
1996 | struct dentry *target) | |
1997 | { | |
1998 | if (target->d_parent != dentry->d_parent) | |
1999 | spin_unlock(&dentry->d_parent->d_lock); | |
2000 | if (target->d_parent != target) | |
2001 | spin_unlock(&target->d_parent->d_lock); | |
2002 | } | |
2003 | ||
1da177e4 | 2004 | /* |
2fd6b7f5 NP |
2005 | * When switching names, the actual string doesn't strictly have to |
2006 | * be preserved in the target - because we're dropping the target | |
2007 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2008 | * the new name before we switch. | |
2009 | * | |
2010 | * Note that we have to be a lot more careful about getting the hash | |
2011 | * switched - we have to switch the hash value properly even if it | |
2012 | * then no longer matches the actual (corrupted) string of the target. | |
2013 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2014 | */ |
9eaef27b TM |
2015 | /* |
2016 | * d_move_locked - move a dentry | |
1da177e4 LT |
2017 | * @dentry: entry to move |
2018 | * @target: new dentry | |
2019 | * | |
2020 | * Update the dcache to reflect the move of a file name. Negative | |
2021 | * dcache entries should not be moved in this way. | |
2022 | */ | |
9eaef27b | 2023 | static void d_move_locked(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2024 | { |
1da177e4 LT |
2025 | if (!dentry->d_inode) |
2026 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2027 | ||
2fd6b7f5 NP |
2028 | BUG_ON(d_ancestor(dentry, target)); |
2029 | BUG_ON(d_ancestor(target, dentry)); | |
2030 | ||
1da177e4 | 2031 | write_seqlock(&rename_lock); |
2fd6b7f5 NP |
2032 | |
2033 | dentry_lock_for_move(dentry, target); | |
1da177e4 LT |
2034 | |
2035 | /* Move the dentry to the target hash queue, if on different bucket */ | |
789680d1 NP |
2036 | spin_lock(&dcache_hash_lock); |
2037 | if (!d_unhashed(dentry)) | |
2038 | hlist_del_rcu(&dentry->d_hash); | |
2039 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); | |
2040 | spin_unlock(&dcache_hash_lock); | |
1da177e4 LT |
2041 | |
2042 | /* Unhash the target: dput() will then get rid of it */ | |
2043 | __d_drop(target); | |
2044 | ||
5160ee6f ED |
2045 | list_del(&dentry->d_u.d_child); |
2046 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2047 | |
2048 | /* Switch the names.. */ | |
2049 | switch_names(dentry, target); | |
9a8d5bb4 | 2050 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2051 | |
2052 | /* ... and switch the parents */ | |
2053 | if (IS_ROOT(dentry)) { | |
2054 | dentry->d_parent = target->d_parent; | |
2055 | target->d_parent = target; | |
5160ee6f | 2056 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2057 | } else { |
9a8d5bb4 | 2058 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2059 | |
2060 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2061 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2062 | } |
2063 | ||
5160ee6f | 2064 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 NP |
2065 | |
2066 | dentry_unlock_parents_for_move(dentry, target); | |
1da177e4 | 2067 | spin_unlock(&target->d_lock); |
c32ccd87 | 2068 | fsnotify_d_move(dentry); |
1da177e4 LT |
2069 | spin_unlock(&dentry->d_lock); |
2070 | write_sequnlock(&rename_lock); | |
9eaef27b TM |
2071 | } |
2072 | ||
2073 | /** | |
2074 | * d_move - move a dentry | |
2075 | * @dentry: entry to move | |
2076 | * @target: new dentry | |
2077 | * | |
2078 | * Update the dcache to reflect the move of a file name. Negative | |
2079 | * dcache entries should not be moved in this way. | |
2080 | */ | |
2081 | ||
2082 | void d_move(struct dentry * dentry, struct dentry * target) | |
2083 | { | |
2084 | spin_lock(&dcache_lock); | |
2085 | d_move_locked(dentry, target); | |
1da177e4 LT |
2086 | spin_unlock(&dcache_lock); |
2087 | } | |
ec4f8605 | 2088 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2089 | |
e2761a11 OH |
2090 | /** |
2091 | * d_ancestor - search for an ancestor | |
2092 | * @p1: ancestor dentry | |
2093 | * @p2: child dentry | |
2094 | * | |
2095 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2096 | * an ancestor of p2, else NULL. | |
9eaef27b | 2097 | */ |
e2761a11 | 2098 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2099 | { |
2100 | struct dentry *p; | |
2101 | ||
871c0067 | 2102 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2103 | if (p->d_parent == p1) |
e2761a11 | 2104 | return p; |
9eaef27b | 2105 | } |
e2761a11 | 2106 | return NULL; |
9eaef27b TM |
2107 | } |
2108 | ||
2109 | /* | |
2110 | * This helper attempts to cope with remotely renamed directories | |
2111 | * | |
2112 | * It assumes that the caller is already holding | |
2113 | * dentry->d_parent->d_inode->i_mutex and the dcache_lock | |
2114 | * | |
2115 | * Note: If ever the locking in lock_rename() changes, then please | |
2116 | * remember to update this too... | |
9eaef27b TM |
2117 | */ |
2118 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
31f3e0b3 | 2119 | __releases(dcache_lock) |
b23fb0a6 | 2120 | __releases(dcache_inode_lock) |
9eaef27b TM |
2121 | { |
2122 | struct mutex *m1 = NULL, *m2 = NULL; | |
2123 | struct dentry *ret; | |
2124 | ||
2125 | /* If alias and dentry share a parent, then no extra locks required */ | |
2126 | if (alias->d_parent == dentry->d_parent) | |
2127 | goto out_unalias; | |
2128 | ||
2129 | /* Check for loops */ | |
2130 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 2131 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
2132 | goto out_err; |
2133 | ||
2134 | /* See lock_rename() */ | |
2135 | ret = ERR_PTR(-EBUSY); | |
2136 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2137 | goto out_err; | |
2138 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2139 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2140 | goto out_err; | |
2141 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2142 | out_unalias: | |
2143 | d_move_locked(alias, dentry); | |
2144 | ret = alias; | |
2145 | out_err: | |
b23fb0a6 | 2146 | spin_unlock(&dcache_inode_lock); |
9eaef27b TM |
2147 | spin_unlock(&dcache_lock); |
2148 | if (m2) | |
2149 | mutex_unlock(m2); | |
2150 | if (m1) | |
2151 | mutex_unlock(m1); | |
2152 | return ret; | |
2153 | } | |
2154 | ||
770bfad8 DH |
2155 | /* |
2156 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2157 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2158 | * returns with anon->d_lock held! |
770bfad8 DH |
2159 | */ |
2160 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2161 | { | |
2162 | struct dentry *dparent, *aparent; | |
2163 | ||
2fd6b7f5 | 2164 | dentry_lock_for_move(anon, dentry); |
770bfad8 DH |
2165 | |
2166 | dparent = dentry->d_parent; | |
2167 | aparent = anon->d_parent; | |
2168 | ||
2fd6b7f5 NP |
2169 | switch_names(dentry, anon); |
2170 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2171 | ||
770bfad8 DH |
2172 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2173 | list_del(&dentry->d_u.d_child); | |
2174 | if (!IS_ROOT(dentry)) | |
2175 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2176 | else | |
2177 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2178 | ||
2179 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2180 | list_del(&anon->d_u.d_child); | |
2181 | if (!IS_ROOT(anon)) | |
2182 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2183 | else | |
2184 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2185 | ||
2fd6b7f5 NP |
2186 | dentry_unlock_parents_for_move(anon, dentry); |
2187 | spin_unlock(&dentry->d_lock); | |
2188 | ||
2189 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2190 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2191 | } | |
2192 | ||
2193 | /** | |
2194 | * d_materialise_unique - introduce an inode into the tree | |
2195 | * @dentry: candidate dentry | |
2196 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2197 | * | |
2198 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2199 | * root directory alias in its place if there is one | |
2200 | */ | |
2201 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2202 | { | |
9eaef27b | 2203 | struct dentry *actual; |
770bfad8 DH |
2204 | |
2205 | BUG_ON(!d_unhashed(dentry)); | |
2206 | ||
2207 | spin_lock(&dcache_lock); | |
b23fb0a6 | 2208 | spin_lock(&dcache_inode_lock); |
770bfad8 DH |
2209 | |
2210 | if (!inode) { | |
2211 | actual = dentry; | |
360da900 | 2212 | __d_instantiate(dentry, NULL); |
770bfad8 DH |
2213 | goto found_lock; |
2214 | } | |
2215 | ||
9eaef27b TM |
2216 | if (S_ISDIR(inode->i_mode)) { |
2217 | struct dentry *alias; | |
2218 | ||
2219 | /* Does an aliased dentry already exist? */ | |
2220 | alias = __d_find_alias(inode, 0); | |
2221 | if (alias) { | |
2222 | actual = alias; | |
2223 | /* Is this an anonymous mountpoint that we could splice | |
2224 | * into our tree? */ | |
2225 | if (IS_ROOT(alias)) { | |
9eaef27b TM |
2226 | __d_materialise_dentry(dentry, alias); |
2227 | __d_drop(alias); | |
2228 | goto found; | |
2229 | } | |
2230 | /* Nope, but we must(!) avoid directory aliasing */ | |
2231 | actual = __d_unalias(dentry, alias); | |
2232 | if (IS_ERR(actual)) | |
2233 | dput(alias); | |
2234 | goto out_nolock; | |
2235 | } | |
770bfad8 DH |
2236 | } |
2237 | ||
2238 | /* Add a unique reference */ | |
2239 | actual = __d_instantiate_unique(dentry, inode); | |
2240 | if (!actual) | |
2241 | actual = dentry; | |
2242 | else if (unlikely(!d_unhashed(actual))) | |
2243 | goto shouldnt_be_hashed; | |
2244 | ||
2245 | found_lock: | |
2246 | spin_lock(&actual->d_lock); | |
2247 | found: | |
789680d1 | 2248 | spin_lock(&dcache_hash_lock); |
770bfad8 | 2249 | _d_rehash(actual); |
789680d1 | 2250 | spin_unlock(&dcache_hash_lock); |
770bfad8 | 2251 | spin_unlock(&actual->d_lock); |
b23fb0a6 | 2252 | spin_unlock(&dcache_inode_lock); |
770bfad8 | 2253 | spin_unlock(&dcache_lock); |
9eaef27b | 2254 | out_nolock: |
770bfad8 DH |
2255 | if (actual == dentry) { |
2256 | security_d_instantiate(dentry, inode); | |
2257 | return NULL; | |
2258 | } | |
2259 | ||
2260 | iput(inode); | |
2261 | return actual; | |
2262 | ||
770bfad8 | 2263 | shouldnt_be_hashed: |
b23fb0a6 | 2264 | spin_unlock(&dcache_inode_lock); |
770bfad8 DH |
2265 | spin_unlock(&dcache_lock); |
2266 | BUG(); | |
770bfad8 | 2267 | } |
ec4f8605 | 2268 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2269 | |
cdd16d02 | 2270 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2271 | { |
2272 | *buflen -= namelen; | |
2273 | if (*buflen < 0) | |
2274 | return -ENAMETOOLONG; | |
2275 | *buffer -= namelen; | |
2276 | memcpy(*buffer, str, namelen); | |
2277 | return 0; | |
2278 | } | |
2279 | ||
cdd16d02 MS |
2280 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2281 | { | |
2282 | return prepend(buffer, buflen, name->name, name->len); | |
2283 | } | |
2284 | ||
1da177e4 | 2285 | /** |
f2eb6575 MS |
2286 | * Prepend path string to a buffer |
2287 | * | |
9d1bc601 MS |
2288 | * @path: the dentry/vfsmount to report |
2289 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2290 | * @buffer: pointer to the end of the buffer |
2291 | * @buflen: pointer to buffer length | |
552ce544 | 2292 | * |
f2eb6575 | 2293 | * Caller holds the dcache_lock. |
9d1bc601 MS |
2294 | * |
2295 | * If path is not reachable from the supplied root, then the value of | |
2296 | * root is changed (without modifying refcounts). | |
1da177e4 | 2297 | */ |
f2eb6575 MS |
2298 | static int prepend_path(const struct path *path, struct path *root, |
2299 | char **buffer, int *buflen) | |
1da177e4 | 2300 | { |
9d1bc601 MS |
2301 | struct dentry *dentry = path->dentry; |
2302 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2303 | bool slash = false; |
2304 | int error = 0; | |
6092d048 | 2305 | |
99b7db7b | 2306 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2307 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2308 | struct dentry * parent; |
2309 | ||
1da177e4 | 2310 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2311 | /* Global root? */ |
1da177e4 | 2312 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2313 | goto global_root; |
2314 | } | |
2315 | dentry = vfsmnt->mnt_mountpoint; | |
2316 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2317 | continue; |
2318 | } | |
2319 | parent = dentry->d_parent; | |
2320 | prefetch(parent); | |
9abca360 | 2321 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2322 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2323 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2324 | if (!error) |
2325 | error = prepend(buffer, buflen, "/", 1); | |
2326 | if (error) | |
2327 | break; | |
2328 | ||
2329 | slash = true; | |
1da177e4 LT |
2330 | dentry = parent; |
2331 | } | |
2332 | ||
be285c71 | 2333 | out: |
f2eb6575 MS |
2334 | if (!error && !slash) |
2335 | error = prepend(buffer, buflen, "/", 1); | |
2336 | ||
99b7db7b | 2337 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2338 | return error; |
1da177e4 LT |
2339 | |
2340 | global_root: | |
98dc568b MS |
2341 | /* |
2342 | * Filesystems needing to implement special "root names" | |
2343 | * should do so with ->d_dname() | |
2344 | */ | |
2345 | if (IS_ROOT(dentry) && | |
2346 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2347 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2348 | (int) dentry->d_name.len, dentry->d_name.name); | |
2349 | } | |
9d1bc601 MS |
2350 | root->mnt = vfsmnt; |
2351 | root->dentry = dentry; | |
be285c71 | 2352 | goto out; |
f2eb6575 | 2353 | } |
be285c71 | 2354 | |
f2eb6575 MS |
2355 | /** |
2356 | * __d_path - return the path of a dentry | |
2357 | * @path: the dentry/vfsmount to report | |
2358 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2359 | * @buf: buffer to return value in |
f2eb6575 MS |
2360 | * @buflen: buffer length |
2361 | * | |
ffd1f4ed | 2362 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2363 | * |
2364 | * Returns a pointer into the buffer or an error code if the | |
2365 | * path was too long. | |
2366 | * | |
be148247 | 2367 | * "buflen" should be positive. |
f2eb6575 MS |
2368 | * |
2369 | * If path is not reachable from the supplied root, then the value of | |
2370 | * root is changed (without modifying refcounts). | |
2371 | */ | |
2372 | char *__d_path(const struct path *path, struct path *root, | |
2373 | char *buf, int buflen) | |
2374 | { | |
2375 | char *res = buf + buflen; | |
2376 | int error; | |
2377 | ||
2378 | prepend(&res, &buflen, "\0", 1); | |
be148247 | 2379 | spin_lock(&dcache_lock); |
f2eb6575 | 2380 | error = prepend_path(path, root, &res, &buflen); |
be148247 CH |
2381 | spin_unlock(&dcache_lock); |
2382 | ||
f2eb6575 MS |
2383 | if (error) |
2384 | return ERR_PTR(error); | |
f2eb6575 | 2385 | return res; |
1da177e4 LT |
2386 | } |
2387 | ||
ffd1f4ed MS |
2388 | /* |
2389 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2390 | */ | |
2391 | static int path_with_deleted(const struct path *path, struct path *root, | |
2392 | char **buf, int *buflen) | |
2393 | { | |
2394 | prepend(buf, buflen, "\0", 1); | |
2395 | if (d_unlinked(path->dentry)) { | |
2396 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2397 | if (error) | |
2398 | return error; | |
2399 | } | |
2400 | ||
2401 | return prepend_path(path, root, buf, buflen); | |
2402 | } | |
2403 | ||
8df9d1a4 MS |
2404 | static int prepend_unreachable(char **buffer, int *buflen) |
2405 | { | |
2406 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2407 | } | |
2408 | ||
a03a8a70 JB |
2409 | /** |
2410 | * d_path - return the path of a dentry | |
cf28b486 | 2411 | * @path: path to report |
a03a8a70 JB |
2412 | * @buf: buffer to return value in |
2413 | * @buflen: buffer length | |
2414 | * | |
2415 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2416 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2417 | * | |
52afeefb AV |
2418 | * Returns a pointer into the buffer or an error code if the path was |
2419 | * too long. Note: Callers should use the returned pointer, not the passed | |
2420 | * in buffer, to use the name! The implementation often starts at an offset | |
2421 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2422 | * |
31f3e0b3 | 2423 | * "buflen" should be positive. |
a03a8a70 | 2424 | */ |
20d4fdc1 | 2425 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2426 | { |
ffd1f4ed | 2427 | char *res = buf + buflen; |
6ac08c39 | 2428 | struct path root; |
9d1bc601 | 2429 | struct path tmp; |
ffd1f4ed | 2430 | int error; |
1da177e4 | 2431 | |
c23fbb6b ED |
2432 | /* |
2433 | * We have various synthetic filesystems that never get mounted. On | |
2434 | * these filesystems dentries are never used for lookup purposes, and | |
2435 | * thus don't need to be hashed. They also don't need a name until a | |
2436 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2437 | * below allows us to generate a name for these objects on demand: | |
2438 | */ | |
cf28b486 JB |
2439 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2440 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2441 | |
f7ad3c6b | 2442 | get_fs_root(current->fs, &root); |
552ce544 | 2443 | spin_lock(&dcache_lock); |
9d1bc601 | 2444 | tmp = root; |
ffd1f4ed MS |
2445 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2446 | if (error) | |
2447 | res = ERR_PTR(error); | |
552ce544 | 2448 | spin_unlock(&dcache_lock); |
6ac08c39 | 2449 | path_put(&root); |
1da177e4 LT |
2450 | return res; |
2451 | } | |
ec4f8605 | 2452 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2453 | |
8df9d1a4 MS |
2454 | /** |
2455 | * d_path_with_unreachable - return the path of a dentry | |
2456 | * @path: path to report | |
2457 | * @buf: buffer to return value in | |
2458 | * @buflen: buffer length | |
2459 | * | |
2460 | * The difference from d_path() is that this prepends "(unreachable)" | |
2461 | * to paths which are unreachable from the current process' root. | |
2462 | */ | |
2463 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2464 | { | |
2465 | char *res = buf + buflen; | |
2466 | struct path root; | |
2467 | struct path tmp; | |
2468 | int error; | |
2469 | ||
2470 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2471 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2472 | ||
2473 | get_fs_root(current->fs, &root); | |
2474 | spin_lock(&dcache_lock); | |
2475 | tmp = root; | |
2476 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2477 | if (!error && !path_equal(&tmp, &root)) | |
2478 | error = prepend_unreachable(&res, &buflen); | |
2479 | spin_unlock(&dcache_lock); | |
2480 | path_put(&root); | |
2481 | if (error) | |
2482 | res = ERR_PTR(error); | |
2483 | ||
2484 | return res; | |
2485 | } | |
2486 | ||
c23fbb6b ED |
2487 | /* |
2488 | * Helper function for dentry_operations.d_dname() members | |
2489 | */ | |
2490 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2491 | const char *fmt, ...) | |
2492 | { | |
2493 | va_list args; | |
2494 | char temp[64]; | |
2495 | int sz; | |
2496 | ||
2497 | va_start(args, fmt); | |
2498 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2499 | va_end(args); | |
2500 | ||
2501 | if (sz > sizeof(temp) || sz > buflen) | |
2502 | return ERR_PTR(-ENAMETOOLONG); | |
2503 | ||
2504 | buffer += buflen - sz; | |
2505 | return memcpy(buffer, temp, sz); | |
2506 | } | |
2507 | ||
6092d048 RP |
2508 | /* |
2509 | * Write full pathname from the root of the filesystem into the buffer. | |
2510 | */ | |
ec2447c2 | 2511 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2512 | { |
2513 | char *end = buf + buflen; | |
2514 | char *retval; | |
2515 | ||
6092d048 | 2516 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2517 | if (buflen < 1) |
2518 | goto Elong; | |
2519 | /* Get '/' right */ | |
2520 | retval = end-1; | |
2521 | *retval = '/'; | |
2522 | ||
cdd16d02 MS |
2523 | while (!IS_ROOT(dentry)) { |
2524 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2525 | int error; |
6092d048 | 2526 | |
6092d048 | 2527 | prefetch(parent); |
9abca360 NP |
2528 | spin_lock(&dentry->d_lock); |
2529 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2530 | spin_unlock(&dentry->d_lock); | |
2531 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2532 | goto Elong; |
2533 | ||
2534 | retval = end; | |
2535 | dentry = parent; | |
2536 | } | |
c103135c AV |
2537 | return retval; |
2538 | Elong: | |
2539 | return ERR_PTR(-ENAMETOOLONG); | |
2540 | } | |
ec2447c2 NP |
2541 | |
2542 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2543 | { | |
2544 | char *retval; | |
2545 | ||
2546 | spin_lock(&dcache_lock); | |
2547 | retval = __dentry_path(dentry, buf, buflen); | |
2548 | spin_unlock(&dcache_lock); | |
2549 | ||
2550 | return retval; | |
2551 | } | |
2552 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2553 | |
2554 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2555 | { | |
2556 | char *p = NULL; | |
2557 | char *retval; | |
2558 | ||
2559 | spin_lock(&dcache_lock); | |
2560 | if (d_unlinked(dentry)) { | |
2561 | p = buf + buflen; | |
2562 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2563 | goto Elong; | |
2564 | buflen++; | |
2565 | } | |
2566 | retval = __dentry_path(dentry, buf, buflen); | |
6092d048 | 2567 | spin_unlock(&dcache_lock); |
c103135c AV |
2568 | if (!IS_ERR(retval) && p) |
2569 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2570 | return retval; |
2571 | Elong: | |
2572 | spin_unlock(&dcache_lock); | |
2573 | return ERR_PTR(-ENAMETOOLONG); | |
2574 | } | |
2575 | ||
1da177e4 LT |
2576 | /* |
2577 | * NOTE! The user-level library version returns a | |
2578 | * character pointer. The kernel system call just | |
2579 | * returns the length of the buffer filled (which | |
2580 | * includes the ending '\0' character), or a negative | |
2581 | * error value. So libc would do something like | |
2582 | * | |
2583 | * char *getcwd(char * buf, size_t size) | |
2584 | * { | |
2585 | * int retval; | |
2586 | * | |
2587 | * retval = sys_getcwd(buf, size); | |
2588 | * if (retval >= 0) | |
2589 | * return buf; | |
2590 | * errno = -retval; | |
2591 | * return NULL; | |
2592 | * } | |
2593 | */ | |
3cdad428 | 2594 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2595 | { |
552ce544 | 2596 | int error; |
6ac08c39 | 2597 | struct path pwd, root; |
552ce544 | 2598 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2599 | |
2600 | if (!page) | |
2601 | return -ENOMEM; | |
2602 | ||
f7ad3c6b | 2603 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2604 | |
552ce544 | 2605 | error = -ENOENT; |
552ce544 | 2606 | spin_lock(&dcache_lock); |
f3da392e | 2607 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2608 | unsigned long len; |
9d1bc601 | 2609 | struct path tmp = root; |
8df9d1a4 MS |
2610 | char *cwd = page + PAGE_SIZE; |
2611 | int buflen = PAGE_SIZE; | |
1da177e4 | 2612 | |
8df9d1a4 MS |
2613 | prepend(&cwd, &buflen, "\0", 1); |
2614 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
552ce544 LT |
2615 | spin_unlock(&dcache_lock); |
2616 | ||
8df9d1a4 | 2617 | if (error) |
552ce544 LT |
2618 | goto out; |
2619 | ||
8df9d1a4 MS |
2620 | /* Unreachable from current root */ |
2621 | if (!path_equal(&tmp, &root)) { | |
2622 | error = prepend_unreachable(&cwd, &buflen); | |
2623 | if (error) | |
2624 | goto out; | |
2625 | } | |
2626 | ||
552ce544 LT |
2627 | error = -ERANGE; |
2628 | len = PAGE_SIZE + page - cwd; | |
2629 | if (len <= size) { | |
2630 | error = len; | |
2631 | if (copy_to_user(buf, cwd, len)) | |
2632 | error = -EFAULT; | |
2633 | } | |
2634 | } else | |
2635 | spin_unlock(&dcache_lock); | |
1da177e4 LT |
2636 | |
2637 | out: | |
6ac08c39 JB |
2638 | path_put(&pwd); |
2639 | path_put(&root); | |
1da177e4 LT |
2640 | free_page((unsigned long) page); |
2641 | return error; | |
2642 | } | |
2643 | ||
2644 | /* | |
2645 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2646 | * | |
2647 | * Trivially implemented using the dcache structure | |
2648 | */ | |
2649 | ||
2650 | /** | |
2651 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2652 | * @new_dentry: new dentry | |
2653 | * @old_dentry: old dentry | |
2654 | * | |
2655 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2656 | * Returns 0 otherwise. | |
2657 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2658 | */ | |
2659 | ||
e2761a11 | 2660 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2661 | { |
2662 | int result; | |
1da177e4 LT |
2663 | unsigned long seq; |
2664 | ||
e2761a11 OH |
2665 | if (new_dentry == old_dentry) |
2666 | return 1; | |
2667 | ||
2668 | /* | |
2669 | * Need rcu_readlock to protect against the d_parent trashing | |
2670 | * due to d_move | |
1da177e4 LT |
2671 | */ |
2672 | rcu_read_lock(); | |
e2761a11 | 2673 | do { |
1da177e4 | 2674 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2675 | seq = read_seqbegin(&rename_lock); |
e2761a11 | 2676 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2677 | result = 1; |
e2761a11 OH |
2678 | else |
2679 | result = 0; | |
1da177e4 LT |
2680 | } while (read_seqretry(&rename_lock, seq)); |
2681 | rcu_read_unlock(); | |
2682 | ||
2683 | return result; | |
2684 | } | |
2685 | ||
2096f759 AV |
2686 | int path_is_under(struct path *path1, struct path *path2) |
2687 | { | |
2688 | struct vfsmount *mnt = path1->mnt; | |
2689 | struct dentry *dentry = path1->dentry; | |
2690 | int res; | |
99b7db7b NP |
2691 | |
2692 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2693 | if (mnt != path2->mnt) { |
2694 | for (;;) { | |
2695 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2696 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2697 | return 0; |
2698 | } | |
2699 | if (mnt->mnt_parent == path2->mnt) | |
2700 | break; | |
2701 | mnt = mnt->mnt_parent; | |
2702 | } | |
2703 | dentry = mnt->mnt_mountpoint; | |
2704 | } | |
2705 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2706 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2707 | return res; |
2708 | } | |
2709 | EXPORT_SYMBOL(path_is_under); | |
2710 | ||
1da177e4 LT |
2711 | void d_genocide(struct dentry *root) |
2712 | { | |
2713 | struct dentry *this_parent = root; | |
2714 | struct list_head *next; | |
2715 | ||
2716 | spin_lock(&dcache_lock); | |
2fd6b7f5 | 2717 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2718 | repeat: |
2719 | next = this_parent->d_subdirs.next; | |
2720 | resume: | |
2721 | while (next != &this_parent->d_subdirs) { | |
2722 | struct list_head *tmp = next; | |
5160ee6f | 2723 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2724 | next = tmp->next; |
da502956 NP |
2725 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2726 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2727 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2728 | continue; |
da502956 | 2729 | } |
1da177e4 | 2730 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2731 | spin_unlock(&this_parent->d_lock); |
2732 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2733 | this_parent = dentry; |
2fd6b7f5 | 2734 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2735 | goto repeat; |
2736 | } | |
b7ab39f6 NP |
2737 | dentry->d_count--; |
2738 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
2739 | } |
2740 | if (this_parent != root) { | |
5160ee6f | 2741 | next = this_parent->d_u.d_child.next; |
b7ab39f6 NP |
2742 | this_parent->d_count--; |
2743 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 2744 | this_parent = this_parent->d_parent; |
2fd6b7f5 | 2745 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2746 | goto resume; |
2747 | } | |
2fd6b7f5 | 2748 | spin_unlock(&this_parent->d_lock); |
1da177e4 LT |
2749 | spin_unlock(&dcache_lock); |
2750 | } | |
2751 | ||
2752 | /** | |
2753 | * find_inode_number - check for dentry with name | |
2754 | * @dir: directory to check | |
2755 | * @name: Name to find. | |
2756 | * | |
2757 | * Check whether a dentry already exists for the given name, | |
2758 | * and return the inode number if it has an inode. Otherwise | |
2759 | * 0 is returned. | |
2760 | * | |
2761 | * This routine is used to post-process directory listings for | |
2762 | * filesystems using synthetic inode numbers, and is necessary | |
2763 | * to keep getcwd() working. | |
2764 | */ | |
2765 | ||
2766 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2767 | { | |
2768 | struct dentry * dentry; | |
2769 | ino_t ino = 0; | |
2770 | ||
3e7e241f EB |
2771 | dentry = d_hash_and_lookup(dir, name); |
2772 | if (dentry) { | |
1da177e4 LT |
2773 | if (dentry->d_inode) |
2774 | ino = dentry->d_inode->i_ino; | |
2775 | dput(dentry); | |
2776 | } | |
1da177e4 LT |
2777 | return ino; |
2778 | } | |
ec4f8605 | 2779 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2780 | |
2781 | static __initdata unsigned long dhash_entries; | |
2782 | static int __init set_dhash_entries(char *str) | |
2783 | { | |
2784 | if (!str) | |
2785 | return 0; | |
2786 | dhash_entries = simple_strtoul(str, &str, 0); | |
2787 | return 1; | |
2788 | } | |
2789 | __setup("dhash_entries=", set_dhash_entries); | |
2790 | ||
2791 | static void __init dcache_init_early(void) | |
2792 | { | |
2793 | int loop; | |
2794 | ||
2795 | /* If hashes are distributed across NUMA nodes, defer | |
2796 | * hash allocation until vmalloc space is available. | |
2797 | */ | |
2798 | if (hashdist) | |
2799 | return; | |
2800 | ||
2801 | dentry_hashtable = | |
2802 | alloc_large_system_hash("Dentry cache", | |
2803 | sizeof(struct hlist_head), | |
2804 | dhash_entries, | |
2805 | 13, | |
2806 | HASH_EARLY, | |
2807 | &d_hash_shift, | |
2808 | &d_hash_mask, | |
2809 | 0); | |
2810 | ||
2811 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2812 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2813 | } | |
2814 | ||
74bf17cf | 2815 | static void __init dcache_init(void) |
1da177e4 LT |
2816 | { |
2817 | int loop; | |
2818 | ||
2819 | /* | |
2820 | * A constructor could be added for stable state like the lists, | |
2821 | * but it is probably not worth it because of the cache nature | |
2822 | * of the dcache. | |
2823 | */ | |
0a31bd5f CL |
2824 | dentry_cache = KMEM_CACHE(dentry, |
2825 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2826 | |
8e1f936b | 2827 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2828 | |
2829 | /* Hash may have been set up in dcache_init_early */ | |
2830 | if (!hashdist) | |
2831 | return; | |
2832 | ||
2833 | dentry_hashtable = | |
2834 | alloc_large_system_hash("Dentry cache", | |
2835 | sizeof(struct hlist_head), | |
2836 | dhash_entries, | |
2837 | 13, | |
2838 | 0, | |
2839 | &d_hash_shift, | |
2840 | &d_hash_mask, | |
2841 | 0); | |
2842 | ||
2843 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2844 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2845 | } | |
2846 | ||
2847 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2848 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2849 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2850 | |
1da177e4 LT |
2851 | EXPORT_SYMBOL(d_genocide); |
2852 | ||
1da177e4 LT |
2853 | void __init vfs_caches_init_early(void) |
2854 | { | |
2855 | dcache_init_early(); | |
2856 | inode_init_early(); | |
2857 | } | |
2858 | ||
2859 | void __init vfs_caches_init(unsigned long mempages) | |
2860 | { | |
2861 | unsigned long reserve; | |
2862 | ||
2863 | /* Base hash sizes on available memory, with a reserve equal to | |
2864 | 150% of current kernel size */ | |
2865 | ||
2866 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2867 | mempages -= reserve; | |
2868 | ||
2869 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 2870 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 2871 | |
74bf17cf DC |
2872 | dcache_init(); |
2873 | inode_init(); | |
1da177e4 | 2874 | files_init(mempages); |
74bf17cf | 2875 | mnt_init(); |
1da177e4 LT |
2876 | bdev_cache_init(); |
2877 | chrdev_init(); | |
2878 | } |