Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/dcache.c | |
3 | * | |
4 | * Complete reimplementation | |
5 | * (C) 1997 Thomas Schoebel-Theuer, | |
6 | * with heavy changes by Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Notes on the allocation strategy: | |
11 | * | |
12 | * The dcache is a master of the icache - whenever a dcache entry | |
13 | * exists, the inode will always exist. "iput()" is done either when | |
14 | * the dcache entry is deleted or garbage collected. | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
26 | #include <linux/module.h> | |
27 | #include <linux/mount.h> | |
28 | #include <linux/file.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/seqlock.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/bootmem.h> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
07f3f05c | 36 | #include "internal.h" |
1da177e4 | 37 | |
789680d1 NP |
38 | /* |
39 | * Usage: | |
b23fb0a6 NP |
40 | * dcache_inode_lock protects: |
41 | * - i_dentry, d_alias, d_inode | |
23044507 NP |
42 | * dcache_hash_lock protects: |
43 | * - the dcache hash table, s_anon lists | |
44 | * dcache_lru_lock protects: | |
45 | * - the dcache lru lists and counters | |
46 | * d_lock protects: | |
47 | * - d_flags | |
48 | * - d_name | |
49 | * - d_lru | |
b7ab39f6 | 50 | * - d_count |
da502956 | 51 | * - d_unhashed() |
2fd6b7f5 NP |
52 | * - d_parent and d_subdirs |
53 | * - childrens' d_child and d_parent | |
b23fb0a6 | 54 | * - d_alias, d_inode |
789680d1 NP |
55 | * |
56 | * Ordering: | |
b5c84bf6 NP |
57 | * dcache_inode_lock |
58 | * dentry->d_lock | |
59 | * dcache_lru_lock | |
60 | * dcache_hash_lock | |
789680d1 | 61 | * |
da502956 NP |
62 | * If there is an ancestor relationship: |
63 | * dentry->d_parent->...->d_parent->d_lock | |
64 | * ... | |
65 | * dentry->d_parent->d_lock | |
66 | * dentry->d_lock | |
67 | * | |
68 | * If no ancestor relationship: | |
789680d1 NP |
69 | * if (dentry1 < dentry2) |
70 | * dentry1->d_lock | |
71 | * dentry2->d_lock | |
72 | */ | |
fa3536cc | 73 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
74 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
75 | ||
b23fb0a6 | 76 | __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_inode_lock); |
789680d1 | 77 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_hash_lock); |
23044507 | 78 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
74c3cbe3 | 79 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 80 | |
949854d0 | 81 | EXPORT_SYMBOL(rename_lock); |
b23fb0a6 | 82 | EXPORT_SYMBOL(dcache_inode_lock); |
1da177e4 | 83 | |
e18b890b | 84 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 85 | |
1da177e4 LT |
86 | /* |
87 | * This is the single most critical data structure when it comes | |
88 | * to the dcache: the hashtable for lookups. Somebody should try | |
89 | * to make this good - I've just made it work. | |
90 | * | |
91 | * This hash-function tries to avoid losing too many bits of hash | |
92 | * information, yet avoid using a prime hash-size or similar. | |
93 | */ | |
94 | #define D_HASHBITS d_hash_shift | |
95 | #define D_HASHMASK d_hash_mask | |
96 | ||
fa3536cc ED |
97 | static unsigned int d_hash_mask __read_mostly; |
98 | static unsigned int d_hash_shift __read_mostly; | |
99 | static struct hlist_head *dentry_hashtable __read_mostly; | |
1da177e4 LT |
100 | |
101 | /* Statistics gathering. */ | |
102 | struct dentry_stat_t dentry_stat = { | |
103 | .age_limit = 45, | |
104 | }; | |
105 | ||
3e880fb5 | 106 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
107 | |
108 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
109 | static int get_nr_dentry(void) |
110 | { | |
111 | int i; | |
112 | int sum = 0; | |
113 | for_each_possible_cpu(i) | |
114 | sum += per_cpu(nr_dentry, i); | |
115 | return sum < 0 ? 0 : sum; | |
116 | } | |
117 | ||
312d3ca8 CH |
118 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
119 | size_t *lenp, loff_t *ppos) | |
120 | { | |
3e880fb5 | 121 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
122 | return proc_dointvec(table, write, buffer, lenp, ppos); |
123 | } | |
124 | #endif | |
125 | ||
9c82ab9c | 126 | static void __d_free(struct rcu_head *head) |
1da177e4 | 127 | { |
9c82ab9c CH |
128 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
129 | ||
fd217f4d | 130 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
131 | if (dname_external(dentry)) |
132 | kfree(dentry->d_name.name); | |
133 | kmem_cache_free(dentry_cache, dentry); | |
134 | } | |
135 | ||
136 | /* | |
b5c84bf6 | 137 | * no locks, please. |
1da177e4 LT |
138 | */ |
139 | static void d_free(struct dentry *dentry) | |
140 | { | |
b7ab39f6 | 141 | BUG_ON(dentry->d_count); |
3e880fb5 | 142 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
143 | if (dentry->d_op && dentry->d_op->d_release) |
144 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 145 | |
b3423415 | 146 | /* if dentry was never inserted into hash, immediate free is OK */ |
e8462caa | 147 | if (hlist_unhashed(&dentry->d_hash)) |
9c82ab9c | 148 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 149 | else |
9c82ab9c | 150 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
151 | } |
152 | ||
31e6b01f NP |
153 | /** |
154 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
155 | * After this call, in-progress rcu-walk path lookup will fail. This | |
156 | * should be called after unhashing, and after changing d_inode (if | |
157 | * the dentry has not already been unhashed). | |
158 | */ | |
159 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
160 | { | |
161 | assert_spin_locked(&dentry->d_lock); | |
162 | /* Go through a barrier */ | |
163 | write_seqcount_barrier(&dentry->d_seq); | |
164 | } | |
165 | ||
1da177e4 LT |
166 | /* |
167 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
168 | * d_iput() operation if defined. Dentry has no refcount |
169 | * and is unhashed. | |
1da177e4 | 170 | */ |
858119e1 | 171 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 172 | __releases(dentry->d_lock) |
b23fb0a6 | 173 | __releases(dcache_inode_lock) |
1da177e4 LT |
174 | { |
175 | struct inode *inode = dentry->d_inode; | |
176 | if (inode) { | |
177 | dentry->d_inode = NULL; | |
178 | list_del_init(&dentry->d_alias); | |
179 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 180 | spin_unlock(&dcache_inode_lock); |
f805fbda LT |
181 | if (!inode->i_nlink) |
182 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
183 | if (dentry->d_op && dentry->d_op->d_iput) |
184 | dentry->d_op->d_iput(dentry, inode); | |
185 | else | |
186 | iput(inode); | |
187 | } else { | |
188 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 189 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
190 | } |
191 | } | |
192 | ||
31e6b01f NP |
193 | /* |
194 | * Release the dentry's inode, using the filesystem | |
195 | * d_iput() operation if defined. dentry remains in-use. | |
196 | */ | |
197 | static void dentry_unlink_inode(struct dentry * dentry) | |
198 | __releases(dentry->d_lock) | |
199 | __releases(dcache_inode_lock) | |
200 | { | |
201 | struct inode *inode = dentry->d_inode; | |
202 | dentry->d_inode = NULL; | |
203 | list_del_init(&dentry->d_alias); | |
204 | dentry_rcuwalk_barrier(dentry); | |
205 | spin_unlock(&dentry->d_lock); | |
206 | spin_unlock(&dcache_inode_lock); | |
207 | if (!inode->i_nlink) | |
208 | fsnotify_inoderemove(inode); | |
209 | if (dentry->d_op && dentry->d_op->d_iput) | |
210 | dentry->d_op->d_iput(dentry, inode); | |
211 | else | |
212 | iput(inode); | |
213 | } | |
214 | ||
da3bbdd4 | 215 | /* |
23044507 | 216 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
217 | */ |
218 | static void dentry_lru_add(struct dentry *dentry) | |
219 | { | |
a4633357 | 220 | if (list_empty(&dentry->d_lru)) { |
23044507 | 221 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
222 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
223 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 224 | dentry_stat.nr_unused++; |
23044507 | 225 | spin_unlock(&dcache_lru_lock); |
a4633357 | 226 | } |
da3bbdd4 KM |
227 | } |
228 | ||
23044507 NP |
229 | static void __dentry_lru_del(struct dentry *dentry) |
230 | { | |
231 | list_del_init(&dentry->d_lru); | |
232 | dentry->d_sb->s_nr_dentry_unused--; | |
233 | dentry_stat.nr_unused--; | |
234 | } | |
235 | ||
da3bbdd4 KM |
236 | static void dentry_lru_del(struct dentry *dentry) |
237 | { | |
238 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
239 | spin_lock(&dcache_lru_lock); |
240 | __dentry_lru_del(dentry); | |
241 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
242 | } |
243 | } | |
244 | ||
a4633357 | 245 | static void dentry_lru_move_tail(struct dentry *dentry) |
da3bbdd4 | 246 | { |
23044507 | 247 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
248 | if (list_empty(&dentry->d_lru)) { |
249 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
250 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 251 | dentry_stat.nr_unused++; |
a4633357 CH |
252 | } else { |
253 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
da3bbdd4 | 254 | } |
23044507 | 255 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
256 | } |
257 | ||
d52b9086 MS |
258 | /** |
259 | * d_kill - kill dentry and return parent | |
260 | * @dentry: dentry to kill | |
261 | * | |
31f3e0b3 | 262 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
263 | * |
264 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 265 | * |
b5c84bf6 NP |
266 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
267 | * d_kill. | |
d52b9086 | 268 | */ |
2fd6b7f5 | 269 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 270 | __releases(dentry->d_lock) |
2fd6b7f5 | 271 | __releases(parent->d_lock) |
b23fb0a6 | 272 | __releases(dcache_inode_lock) |
d52b9086 | 273 | { |
949854d0 | 274 | dentry->d_parent = NULL; |
d52b9086 | 275 | list_del(&dentry->d_u.d_child); |
2fd6b7f5 NP |
276 | if (parent) |
277 | spin_unlock(&parent->d_lock); | |
d52b9086 | 278 | dentry_iput(dentry); |
b7ab39f6 NP |
279 | /* |
280 | * dentry_iput drops the locks, at which point nobody (except | |
281 | * transient RCU lookups) can reach this dentry. | |
282 | */ | |
d52b9086 | 283 | d_free(dentry); |
871c0067 | 284 | return parent; |
d52b9086 MS |
285 | } |
286 | ||
789680d1 NP |
287 | /** |
288 | * d_drop - drop a dentry | |
289 | * @dentry: dentry to drop | |
290 | * | |
291 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
292 | * be found through a VFS lookup any more. Note that this is different from | |
293 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
294 | * possible, giving a successful _negative_ lookup, while d_drop will | |
295 | * just make the cache lookup fail. | |
296 | * | |
297 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
298 | * reason (NFS timeouts or autofs deletes). | |
299 | * | |
300 | * __d_drop requires dentry->d_lock. | |
301 | */ | |
302 | void __d_drop(struct dentry *dentry) | |
303 | { | |
304 | if (!(dentry->d_flags & DCACHE_UNHASHED)) { | |
305 | dentry->d_flags |= DCACHE_UNHASHED; | |
306 | spin_lock(&dcache_hash_lock); | |
307 | hlist_del_rcu(&dentry->d_hash); | |
308 | spin_unlock(&dcache_hash_lock); | |
31e6b01f | 309 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
310 | } |
311 | } | |
312 | EXPORT_SYMBOL(__d_drop); | |
313 | ||
314 | void d_drop(struct dentry *dentry) | |
315 | { | |
789680d1 NP |
316 | spin_lock(&dentry->d_lock); |
317 | __d_drop(dentry); | |
318 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
319 | } |
320 | EXPORT_SYMBOL(d_drop); | |
321 | ||
77812a1e NP |
322 | /* |
323 | * Finish off a dentry we've decided to kill. | |
324 | * dentry->d_lock must be held, returns with it unlocked. | |
325 | * If ref is non-zero, then decrement the refcount too. | |
326 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
327 | */ | |
328 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
329 | __releases(dentry->d_lock) | |
330 | { | |
331 | struct dentry *parent; | |
332 | ||
333 | if (!spin_trylock(&dcache_inode_lock)) { | |
334 | relock: | |
335 | spin_unlock(&dentry->d_lock); | |
336 | cpu_relax(); | |
337 | return dentry; /* try again with same dentry */ | |
338 | } | |
339 | if (IS_ROOT(dentry)) | |
340 | parent = NULL; | |
341 | else | |
342 | parent = dentry->d_parent; | |
343 | if (parent && !spin_trylock(&parent->d_lock)) { | |
344 | spin_unlock(&dcache_inode_lock); | |
345 | goto relock; | |
346 | } | |
31e6b01f | 347 | |
77812a1e NP |
348 | if (ref) |
349 | dentry->d_count--; | |
350 | /* if dentry was on the d_lru list delete it from there */ | |
351 | dentry_lru_del(dentry); | |
352 | /* if it was on the hash then remove it */ | |
353 | __d_drop(dentry); | |
354 | return d_kill(dentry, parent); | |
355 | } | |
356 | ||
1da177e4 LT |
357 | /* |
358 | * This is dput | |
359 | * | |
360 | * This is complicated by the fact that we do not want to put | |
361 | * dentries that are no longer on any hash chain on the unused | |
362 | * list: we'd much rather just get rid of them immediately. | |
363 | * | |
364 | * However, that implies that we have to traverse the dentry | |
365 | * tree upwards to the parents which might _also_ now be | |
366 | * scheduled for deletion (it may have been only waiting for | |
367 | * its last child to go away). | |
368 | * | |
369 | * This tail recursion is done by hand as we don't want to depend | |
370 | * on the compiler to always get this right (gcc generally doesn't). | |
371 | * Real recursion would eat up our stack space. | |
372 | */ | |
373 | ||
374 | /* | |
375 | * dput - release a dentry | |
376 | * @dentry: dentry to release | |
377 | * | |
378 | * Release a dentry. This will drop the usage count and if appropriate | |
379 | * call the dentry unlink method as well as removing it from the queues and | |
380 | * releasing its resources. If the parent dentries were scheduled for release | |
381 | * they too may now get deleted. | |
1da177e4 | 382 | */ |
1da177e4 LT |
383 | void dput(struct dentry *dentry) |
384 | { | |
385 | if (!dentry) | |
386 | return; | |
387 | ||
388 | repeat: | |
b7ab39f6 | 389 | if (dentry->d_count == 1) |
1da177e4 | 390 | might_sleep(); |
1da177e4 | 391 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
392 | BUG_ON(!dentry->d_count); |
393 | if (dentry->d_count > 1) { | |
394 | dentry->d_count--; | |
1da177e4 | 395 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
396 | return; |
397 | } | |
398 | ||
fb045adb | 399 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 400 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 401 | goto kill_it; |
1da177e4 | 402 | } |
265ac902 | 403 | |
1da177e4 LT |
404 | /* Unreachable? Get rid of it */ |
405 | if (d_unhashed(dentry)) | |
406 | goto kill_it; | |
265ac902 NP |
407 | |
408 | /* Otherwise leave it cached and ensure it's on the LRU */ | |
409 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 410 | dentry_lru_add(dentry); |
265ac902 | 411 | |
61f3dee4 NP |
412 | dentry->d_count--; |
413 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
414 | return; |
415 | ||
d52b9086 | 416 | kill_it: |
77812a1e | 417 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
418 | if (dentry) |
419 | goto repeat; | |
1da177e4 | 420 | } |
ec4f8605 | 421 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
422 | |
423 | /** | |
424 | * d_invalidate - invalidate a dentry | |
425 | * @dentry: dentry to invalidate | |
426 | * | |
427 | * Try to invalidate the dentry if it turns out to be | |
428 | * possible. If there are other dentries that can be | |
429 | * reached through this one we can't delete it and we | |
430 | * return -EBUSY. On success we return 0. | |
431 | * | |
432 | * no dcache lock. | |
433 | */ | |
434 | ||
435 | int d_invalidate(struct dentry * dentry) | |
436 | { | |
437 | /* | |
438 | * If it's already been dropped, return OK. | |
439 | */ | |
da502956 | 440 | spin_lock(&dentry->d_lock); |
1da177e4 | 441 | if (d_unhashed(dentry)) { |
da502956 | 442 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
443 | return 0; |
444 | } | |
445 | /* | |
446 | * Check whether to do a partial shrink_dcache | |
447 | * to get rid of unused child entries. | |
448 | */ | |
449 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 450 | spin_unlock(&dentry->d_lock); |
1da177e4 | 451 | shrink_dcache_parent(dentry); |
da502956 | 452 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
453 | } |
454 | ||
455 | /* | |
456 | * Somebody else still using it? | |
457 | * | |
458 | * If it's a directory, we can't drop it | |
459 | * for fear of somebody re-populating it | |
460 | * with children (even though dropping it | |
461 | * would make it unreachable from the root, | |
462 | * we might still populate it if it was a | |
463 | * working directory or similar). | |
464 | */ | |
b7ab39f6 | 465 | if (dentry->d_count > 1) { |
1da177e4 LT |
466 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
467 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
468 | return -EBUSY; |
469 | } | |
470 | } | |
471 | ||
472 | __d_drop(dentry); | |
473 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
474 | return 0; |
475 | } | |
ec4f8605 | 476 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 477 | |
b5c84bf6 | 478 | /* This must be called with d_lock held */ |
dc0474be | 479 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 480 | { |
b7ab39f6 | 481 | dentry->d_count++; |
23044507 NP |
482 | } |
483 | ||
dc0474be | 484 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 485 | { |
23044507 | 486 | spin_lock(&dentry->d_lock); |
dc0474be | 487 | __dget_dlock(dentry); |
23044507 | 488 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
489 | } |
490 | ||
b7ab39f6 NP |
491 | struct dentry *dget_parent(struct dentry *dentry) |
492 | { | |
493 | struct dentry *ret; | |
494 | ||
495 | repeat: | |
a734eb45 NP |
496 | /* |
497 | * Don't need rcu_dereference because we re-check it was correct under | |
498 | * the lock. | |
499 | */ | |
500 | rcu_read_lock(); | |
b7ab39f6 | 501 | ret = dentry->d_parent; |
a734eb45 NP |
502 | if (!ret) { |
503 | rcu_read_unlock(); | |
b7ab39f6 NP |
504 | goto out; |
505 | } | |
a734eb45 NP |
506 | spin_lock(&ret->d_lock); |
507 | if (unlikely(ret != dentry->d_parent)) { | |
508 | spin_unlock(&ret->d_lock); | |
509 | rcu_read_unlock(); | |
b7ab39f6 NP |
510 | goto repeat; |
511 | } | |
a734eb45 | 512 | rcu_read_unlock(); |
b7ab39f6 NP |
513 | BUG_ON(!ret->d_count); |
514 | ret->d_count++; | |
515 | spin_unlock(&ret->d_lock); | |
516 | out: | |
b7ab39f6 NP |
517 | return ret; |
518 | } | |
519 | EXPORT_SYMBOL(dget_parent); | |
520 | ||
1da177e4 LT |
521 | /** |
522 | * d_find_alias - grab a hashed alias of inode | |
523 | * @inode: inode in question | |
524 | * @want_discon: flag, used by d_splice_alias, to request | |
525 | * that only a DISCONNECTED alias be returned. | |
526 | * | |
527 | * If inode has a hashed alias, or is a directory and has any alias, | |
528 | * acquire the reference to alias and return it. Otherwise return NULL. | |
529 | * Notice that if inode is a directory there can be only one alias and | |
530 | * it can be unhashed only if it has no children, or if it is the root | |
531 | * of a filesystem. | |
532 | * | |
21c0d8fd | 533 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 534 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 535 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 536 | */ |
da502956 | 537 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 538 | { |
da502956 | 539 | struct dentry *alias, *discon_alias; |
1da177e4 | 540 | |
da502956 NP |
541 | again: |
542 | discon_alias = NULL; | |
543 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
544 | spin_lock(&alias->d_lock); | |
1da177e4 | 545 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 546 | if (IS_ROOT(alias) && |
da502956 | 547 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 548 | discon_alias = alias; |
da502956 | 549 | } else if (!want_discon) { |
dc0474be | 550 | __dget_dlock(alias); |
da502956 NP |
551 | spin_unlock(&alias->d_lock); |
552 | return alias; | |
553 | } | |
554 | } | |
555 | spin_unlock(&alias->d_lock); | |
556 | } | |
557 | if (discon_alias) { | |
558 | alias = discon_alias; | |
559 | spin_lock(&alias->d_lock); | |
560 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
561 | if (IS_ROOT(alias) && | |
562 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 563 | __dget_dlock(alias); |
da502956 | 564 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
565 | return alias; |
566 | } | |
567 | } | |
da502956 NP |
568 | spin_unlock(&alias->d_lock); |
569 | goto again; | |
1da177e4 | 570 | } |
da502956 | 571 | return NULL; |
1da177e4 LT |
572 | } |
573 | ||
da502956 | 574 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 575 | { |
214fda1f DH |
576 | struct dentry *de = NULL; |
577 | ||
578 | if (!list_empty(&inode->i_dentry)) { | |
b23fb0a6 | 579 | spin_lock(&dcache_inode_lock); |
214fda1f | 580 | de = __d_find_alias(inode, 0); |
b23fb0a6 | 581 | spin_unlock(&dcache_inode_lock); |
214fda1f | 582 | } |
1da177e4 LT |
583 | return de; |
584 | } | |
ec4f8605 | 585 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
586 | |
587 | /* | |
588 | * Try to kill dentries associated with this inode. | |
589 | * WARNING: you must own a reference to inode. | |
590 | */ | |
591 | void d_prune_aliases(struct inode *inode) | |
592 | { | |
0cdca3f9 | 593 | struct dentry *dentry; |
1da177e4 | 594 | restart: |
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) { |
dc0474be | 599 | __dget_dlock(dentry); |
1da177e4 LT |
600 | __d_drop(dentry); |
601 | spin_unlock(&dentry->d_lock); | |
b23fb0a6 | 602 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
603 | dput(dentry); |
604 | goto restart; | |
605 | } | |
606 | spin_unlock(&dentry->d_lock); | |
607 | } | |
b23fb0a6 | 608 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 609 | } |
ec4f8605 | 610 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
611 | |
612 | /* | |
77812a1e NP |
613 | * Try to throw away a dentry - free the inode, dput the parent. |
614 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
615 | * Releases dentry->d_lock. | |
d702ccb3 | 616 | * |
77812a1e | 617 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 618 | */ |
77812a1e | 619 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 620 | __releases(dentry->d_lock) |
1da177e4 | 621 | { |
77812a1e | 622 | struct dentry *parent; |
d52b9086 | 623 | |
77812a1e | 624 | parent = dentry_kill(dentry, 0); |
d52b9086 | 625 | /* |
77812a1e NP |
626 | * If dentry_kill returns NULL, we have nothing more to do. |
627 | * if it returns the same dentry, trylocks failed. In either | |
628 | * case, just loop again. | |
629 | * | |
630 | * Otherwise, we need to prune ancestors too. This is necessary | |
631 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
632 | * is also expected to be beneficial in reducing dentry cache | |
633 | * fragmentation. | |
d52b9086 | 634 | */ |
77812a1e NP |
635 | if (!parent) |
636 | return; | |
637 | if (parent == dentry) | |
638 | return; | |
639 | ||
640 | /* Prune ancestors. */ | |
641 | dentry = parent; | |
d52b9086 | 642 | while (dentry) { |
b7ab39f6 | 643 | spin_lock(&dentry->d_lock); |
89e60548 NP |
644 | if (dentry->d_count > 1) { |
645 | dentry->d_count--; | |
646 | spin_unlock(&dentry->d_lock); | |
647 | return; | |
648 | } | |
77812a1e | 649 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 650 | } |
1da177e4 LT |
651 | } |
652 | ||
3049cfe2 | 653 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 654 | { |
da3bbdd4 | 655 | struct dentry *dentry; |
da3bbdd4 | 656 | |
ec33679d NP |
657 | rcu_read_lock(); |
658 | for (;;) { | |
ec33679d NP |
659 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
660 | if (&dentry->d_lru == list) | |
661 | break; /* empty */ | |
662 | spin_lock(&dentry->d_lock); | |
663 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
664 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
665 | continue; |
666 | } | |
667 | ||
1da177e4 LT |
668 | /* |
669 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
670 | * the LRU because of laziness during lookup. Do not free |
671 | * it - just keep it off the LRU list. | |
1da177e4 | 672 | */ |
b7ab39f6 | 673 | if (dentry->d_count) { |
ec33679d | 674 | dentry_lru_del(dentry); |
da3bbdd4 | 675 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
676 | continue; |
677 | } | |
ec33679d | 678 | |
ec33679d | 679 | rcu_read_unlock(); |
77812a1e NP |
680 | |
681 | try_prune_one_dentry(dentry); | |
682 | ||
ec33679d | 683 | rcu_read_lock(); |
da3bbdd4 | 684 | } |
ec33679d | 685 | rcu_read_unlock(); |
3049cfe2 CH |
686 | } |
687 | ||
688 | /** | |
689 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
690 | * @sb: superblock to shrink dentry LRU. | |
691 | * @count: number of entries to prune | |
692 | * @flags: flags to control the dentry processing | |
693 | * | |
694 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
695 | */ | |
696 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
697 | { | |
698 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
699 | struct dentry *dentry; | |
700 | LIST_HEAD(referenced); | |
701 | LIST_HEAD(tmp); | |
702 | int cnt = *count; | |
703 | ||
23044507 NP |
704 | relock: |
705 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
706 | while (!list_empty(&sb->s_dentry_lru)) { |
707 | dentry = list_entry(sb->s_dentry_lru.prev, | |
708 | struct dentry, d_lru); | |
709 | BUG_ON(dentry->d_sb != sb); | |
710 | ||
23044507 NP |
711 | if (!spin_trylock(&dentry->d_lock)) { |
712 | spin_unlock(&dcache_lru_lock); | |
713 | cpu_relax(); | |
714 | goto relock; | |
715 | } | |
716 | ||
3049cfe2 CH |
717 | /* |
718 | * If we are honouring the DCACHE_REFERENCED flag and the | |
719 | * dentry has this flag set, don't free it. Clear the flag | |
720 | * and put it back on the LRU. | |
721 | */ | |
23044507 NP |
722 | if (flags & DCACHE_REFERENCED && |
723 | dentry->d_flags & DCACHE_REFERENCED) { | |
724 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
725 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 726 | spin_unlock(&dentry->d_lock); |
23044507 NP |
727 | } else { |
728 | list_move_tail(&dentry->d_lru, &tmp); | |
729 | spin_unlock(&dentry->d_lock); | |
730 | if (!--cnt) | |
731 | break; | |
3049cfe2 | 732 | } |
ec33679d | 733 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 734 | } |
da3bbdd4 KM |
735 | if (!list_empty(&referenced)) |
736 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 737 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
738 | |
739 | shrink_dentry_list(&tmp); | |
740 | ||
741 | *count = cnt; | |
da3bbdd4 KM |
742 | } |
743 | ||
744 | /** | |
745 | * prune_dcache - shrink the dcache | |
746 | * @count: number of entries to try to free | |
747 | * | |
748 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
749 | * need to unmount something (at which point we need to unuse all dentries). | |
750 | * | |
751 | * This function may fail to free any resources if all the dentries are in use. | |
752 | */ | |
753 | static void prune_dcache(int count) | |
754 | { | |
dca33252 | 755 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 756 | int w_count; |
86c8749e | 757 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
758 | int prune_ratio; |
759 | int pruned; | |
760 | ||
761 | if (unused == 0 || count == 0) | |
762 | return; | |
da3bbdd4 KM |
763 | if (count >= unused) |
764 | prune_ratio = 1; | |
765 | else | |
766 | prune_ratio = unused / count; | |
767 | spin_lock(&sb_lock); | |
dca33252 | 768 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
769 | if (list_empty(&sb->s_instances)) |
770 | continue; | |
da3bbdd4 | 771 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 772 | continue; |
da3bbdd4 KM |
773 | sb->s_count++; |
774 | /* Now, we reclaim unused dentrins with fairness. | |
775 | * We reclaim them same percentage from each superblock. | |
776 | * We calculate number of dentries to scan on this sb | |
777 | * as follows, but the implementation is arranged to avoid | |
778 | * overflows: | |
779 | * number of dentries to scan on this sb = | |
780 | * count * (number of dentries on this sb / | |
781 | * number of dentries in the machine) | |
0feae5c4 | 782 | */ |
da3bbdd4 KM |
783 | spin_unlock(&sb_lock); |
784 | if (prune_ratio != 1) | |
785 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
786 | else | |
787 | w_count = sb->s_nr_dentry_unused; | |
788 | pruned = w_count; | |
0feae5c4 | 789 | /* |
da3bbdd4 KM |
790 | * We need to be sure this filesystem isn't being unmounted, |
791 | * otherwise we could race with generic_shutdown_super(), and | |
792 | * end up holding a reference to an inode while the filesystem | |
793 | * is unmounted. So we try to get s_umount, and make sure | |
794 | * s_root isn't NULL. | |
0feae5c4 | 795 | */ |
da3bbdd4 KM |
796 | if (down_read_trylock(&sb->s_umount)) { |
797 | if ((sb->s_root != NULL) && | |
798 | (!list_empty(&sb->s_dentry_lru))) { | |
da3bbdd4 KM |
799 | __shrink_dcache_sb(sb, &w_count, |
800 | DCACHE_REFERENCED); | |
801 | pruned -= w_count; | |
0feae5c4 | 802 | } |
da3bbdd4 | 803 | up_read(&sb->s_umount); |
0feae5c4 | 804 | } |
da3bbdd4 | 805 | spin_lock(&sb_lock); |
dca33252 AV |
806 | if (p) |
807 | __put_super(p); | |
da3bbdd4 | 808 | count -= pruned; |
dca33252 | 809 | p = sb; |
79893c17 AV |
810 | /* more work left to do? */ |
811 | if (count <= 0) | |
812 | break; | |
1da177e4 | 813 | } |
dca33252 AV |
814 | if (p) |
815 | __put_super(p); | |
da3bbdd4 | 816 | spin_unlock(&sb_lock); |
1da177e4 LT |
817 | } |
818 | ||
1da177e4 LT |
819 | /** |
820 | * shrink_dcache_sb - shrink dcache for a superblock | |
821 | * @sb: superblock | |
822 | * | |
3049cfe2 CH |
823 | * Shrink the dcache for the specified super block. This is used to free |
824 | * the dcache before unmounting a file system. | |
1da177e4 | 825 | */ |
3049cfe2 | 826 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 827 | { |
3049cfe2 CH |
828 | LIST_HEAD(tmp); |
829 | ||
23044507 | 830 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
831 | while (!list_empty(&sb->s_dentry_lru)) { |
832 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 833 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 834 | shrink_dentry_list(&tmp); |
ec33679d | 835 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 836 | } |
23044507 | 837 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 838 | } |
ec4f8605 | 839 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 840 | |
c636ebdb DH |
841 | /* |
842 | * destroy a single subtree of dentries for unmount | |
843 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
844 | * locking | |
845 | */ | |
846 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
847 | { | |
848 | struct dentry *parent; | |
f8713576 | 849 | unsigned detached = 0; |
c636ebdb DH |
850 | |
851 | BUG_ON(!IS_ROOT(dentry)); | |
852 | ||
853 | /* detach this root from the system */ | |
23044507 | 854 | spin_lock(&dentry->d_lock); |
a4633357 | 855 | dentry_lru_del(dentry); |
c636ebdb | 856 | __d_drop(dentry); |
da502956 | 857 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
858 | |
859 | for (;;) { | |
860 | /* descend to the first leaf in the current subtree */ | |
861 | while (!list_empty(&dentry->d_subdirs)) { | |
862 | struct dentry *loop; | |
863 | ||
864 | /* this is a branch with children - detach all of them | |
865 | * from the system in one go */ | |
2fd6b7f5 | 866 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
867 | list_for_each_entry(loop, &dentry->d_subdirs, |
868 | d_u.d_child) { | |
2fd6b7f5 NP |
869 | spin_lock_nested(&loop->d_lock, |
870 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 871 | dentry_lru_del(loop); |
c636ebdb | 872 | __d_drop(loop); |
da502956 | 873 | spin_unlock(&loop->d_lock); |
c636ebdb | 874 | } |
2fd6b7f5 | 875 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
876 | |
877 | /* move to the first child */ | |
878 | dentry = list_entry(dentry->d_subdirs.next, | |
879 | struct dentry, d_u.d_child); | |
880 | } | |
881 | ||
882 | /* consume the dentries from this leaf up through its parents | |
883 | * until we find one with children or run out altogether */ | |
884 | do { | |
885 | struct inode *inode; | |
886 | ||
b7ab39f6 | 887 | if (dentry->d_count != 0) { |
c636ebdb DH |
888 | printk(KERN_ERR |
889 | "BUG: Dentry %p{i=%lx,n=%s}" | |
890 | " still in use (%d)" | |
891 | " [unmount of %s %s]\n", | |
892 | dentry, | |
893 | dentry->d_inode ? | |
894 | dentry->d_inode->i_ino : 0UL, | |
895 | dentry->d_name.name, | |
b7ab39f6 | 896 | dentry->d_count, |
c636ebdb DH |
897 | dentry->d_sb->s_type->name, |
898 | dentry->d_sb->s_id); | |
899 | BUG(); | |
900 | } | |
901 | ||
2fd6b7f5 | 902 | if (IS_ROOT(dentry)) { |
c636ebdb | 903 | parent = NULL; |
2fd6b7f5 NP |
904 | list_del(&dentry->d_u.d_child); |
905 | } else { | |
871c0067 | 906 | parent = dentry->d_parent; |
b7ab39f6 NP |
907 | spin_lock(&parent->d_lock); |
908 | parent->d_count--; | |
2fd6b7f5 | 909 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 910 | spin_unlock(&parent->d_lock); |
871c0067 | 911 | } |
c636ebdb | 912 | |
f8713576 | 913 | detached++; |
c636ebdb DH |
914 | |
915 | inode = dentry->d_inode; | |
916 | if (inode) { | |
917 | dentry->d_inode = NULL; | |
918 | list_del_init(&dentry->d_alias); | |
919 | if (dentry->d_op && dentry->d_op->d_iput) | |
920 | dentry->d_op->d_iput(dentry, inode); | |
921 | else | |
922 | iput(inode); | |
923 | } | |
924 | ||
925 | d_free(dentry); | |
926 | ||
927 | /* finished when we fall off the top of the tree, | |
928 | * otherwise we ascend to the parent and move to the | |
929 | * next sibling if there is one */ | |
930 | if (!parent) | |
312d3ca8 | 931 | return; |
c636ebdb | 932 | dentry = parent; |
c636ebdb DH |
933 | } while (list_empty(&dentry->d_subdirs)); |
934 | ||
935 | dentry = list_entry(dentry->d_subdirs.next, | |
936 | struct dentry, d_u.d_child); | |
937 | } | |
938 | } | |
939 | ||
940 | /* | |
941 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 942 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
943 | * - the superblock is detached from all mountings and open files, so the |
944 | * dentry trees will not be rearranged by the VFS | |
945 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
946 | * any dentries belonging to this superblock that it comes across | |
947 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
948 | * in this superblock | |
949 | */ | |
950 | void shrink_dcache_for_umount(struct super_block *sb) | |
951 | { | |
952 | struct dentry *dentry; | |
953 | ||
954 | if (down_read_trylock(&sb->s_umount)) | |
955 | BUG(); | |
956 | ||
957 | dentry = sb->s_root; | |
958 | sb->s_root = NULL; | |
b7ab39f6 NP |
959 | spin_lock(&dentry->d_lock); |
960 | dentry->d_count--; | |
961 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
962 | shrink_dcache_for_umount_subtree(dentry); |
963 | ||
964 | while (!hlist_empty(&sb->s_anon)) { | |
965 | dentry = hlist_entry(sb->s_anon.first, struct dentry, d_hash); | |
966 | shrink_dcache_for_umount_subtree(dentry); | |
967 | } | |
968 | } | |
969 | ||
1da177e4 LT |
970 | /* |
971 | * Search for at least 1 mount point in the dentry's subdirs. | |
972 | * We descend to the next level whenever the d_subdirs | |
973 | * list is non-empty and continue searching. | |
974 | */ | |
975 | ||
976 | /** | |
977 | * have_submounts - check for mounts over a dentry | |
978 | * @parent: dentry to check. | |
979 | * | |
980 | * Return true if the parent or its subdirectories contain | |
981 | * a mount point | |
982 | */ | |
1da177e4 LT |
983 | int have_submounts(struct dentry *parent) |
984 | { | |
949854d0 | 985 | struct dentry *this_parent; |
1da177e4 | 986 | struct list_head *next; |
949854d0 | 987 | unsigned seq; |
58db63d0 | 988 | int locked = 0; |
949854d0 | 989 | |
949854d0 | 990 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
991 | again: |
992 | this_parent = parent; | |
1da177e4 | 993 | |
1da177e4 LT |
994 | if (d_mountpoint(parent)) |
995 | goto positive; | |
2fd6b7f5 | 996 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
997 | repeat: |
998 | next = this_parent->d_subdirs.next; | |
999 | resume: | |
1000 | while (next != &this_parent->d_subdirs) { | |
1001 | struct list_head *tmp = next; | |
5160ee6f | 1002 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1003 | next = tmp->next; |
2fd6b7f5 NP |
1004 | |
1005 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1006 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1007 | if (d_mountpoint(dentry)) { |
1008 | spin_unlock(&dentry->d_lock); | |
1009 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1010 | goto positive; |
2fd6b7f5 | 1011 | } |
1da177e4 | 1012 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1013 | spin_unlock(&this_parent->d_lock); |
1014 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1015 | this_parent = dentry; |
2fd6b7f5 | 1016 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1017 | goto repeat; |
1018 | } | |
2fd6b7f5 | 1019 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1020 | } |
1021 | /* | |
1022 | * All done at this level ... ascend and resume the search. | |
1023 | */ | |
1024 | if (this_parent != parent) { | |
949854d0 NP |
1025 | struct dentry *tmp; |
1026 | struct dentry *child; | |
1027 | ||
1028 | tmp = this_parent->d_parent; | |
1029 | rcu_read_lock(); | |
2fd6b7f5 | 1030 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1031 | child = this_parent; |
1032 | this_parent = tmp; | |
2fd6b7f5 | 1033 | spin_lock(&this_parent->d_lock); |
949854d0 NP |
1034 | /* might go back up the wrong parent if we have had a rename |
1035 | * or deletion */ | |
1036 | if (this_parent != child->d_parent || | |
58db63d0 | 1037 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 1038 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1039 | rcu_read_unlock(); |
1040 | goto rename_retry; | |
1041 | } | |
1042 | rcu_read_unlock(); | |
1043 | next = child->d_u.d_child.next; | |
1da177e4 LT |
1044 | goto resume; |
1045 | } | |
2fd6b7f5 | 1046 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1047 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1048 | goto rename_retry; |
58db63d0 NP |
1049 | if (locked) |
1050 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1051 | return 0; /* No mount points found in tree */ |
1052 | positive: | |
58db63d0 | 1053 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1054 | goto rename_retry; |
58db63d0 NP |
1055 | if (locked) |
1056 | write_sequnlock(&rename_lock); | |
1da177e4 | 1057 | return 1; |
58db63d0 NP |
1058 | |
1059 | rename_retry: | |
1060 | locked = 1; | |
1061 | write_seqlock(&rename_lock); | |
1062 | goto again; | |
1da177e4 | 1063 | } |
ec4f8605 | 1064 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1065 | |
1066 | /* | |
1067 | * Search the dentry child list for the specified parent, | |
1068 | * and move any unused dentries to the end of the unused | |
1069 | * list for prune_dcache(). We descend to the next level | |
1070 | * whenever the d_subdirs list is non-empty and continue | |
1071 | * searching. | |
1072 | * | |
1073 | * It returns zero iff there are no unused children, | |
1074 | * otherwise it returns the number of children moved to | |
1075 | * the end of the unused list. This may not be the total | |
1076 | * number of unused children, because select_parent can | |
1077 | * drop the lock and return early due to latency | |
1078 | * constraints. | |
1079 | */ | |
1080 | static int select_parent(struct dentry * parent) | |
1081 | { | |
949854d0 | 1082 | struct dentry *this_parent; |
1da177e4 | 1083 | struct list_head *next; |
949854d0 | 1084 | unsigned seq; |
1da177e4 | 1085 | int found = 0; |
58db63d0 | 1086 | int locked = 0; |
1da177e4 | 1087 | |
949854d0 | 1088 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1089 | again: |
1090 | this_parent = parent; | |
2fd6b7f5 | 1091 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1092 | repeat: |
1093 | next = this_parent->d_subdirs.next; | |
1094 | resume: | |
1095 | while (next != &this_parent->d_subdirs) { | |
1096 | struct list_head *tmp = next; | |
5160ee6f | 1097 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1098 | next = tmp->next; |
1099 | ||
2fd6b7f5 | 1100 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1101 | |
1da177e4 LT |
1102 | /* |
1103 | * move only zero ref count dentries to the end | |
1104 | * of the unused list for prune_dcache | |
1105 | */ | |
b7ab39f6 | 1106 | if (!dentry->d_count) { |
a4633357 | 1107 | dentry_lru_move_tail(dentry); |
1da177e4 | 1108 | found++; |
a4633357 CH |
1109 | } else { |
1110 | dentry_lru_del(dentry); | |
1da177e4 LT |
1111 | } |
1112 | ||
1113 | /* | |
1114 | * We can return to the caller if we have found some (this | |
1115 | * ensures forward progress). We'll be coming back to find | |
1116 | * the rest. | |
1117 | */ | |
2fd6b7f5 NP |
1118 | if (found && need_resched()) { |
1119 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1120 | goto out; |
2fd6b7f5 | 1121 | } |
1da177e4 LT |
1122 | |
1123 | /* | |
1124 | * Descend a level if the d_subdirs list is non-empty. | |
1125 | */ | |
1126 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1127 | spin_unlock(&this_parent->d_lock); |
1128 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1129 | this_parent = dentry; |
2fd6b7f5 | 1130 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1131 | goto repeat; |
1132 | } | |
2fd6b7f5 NP |
1133 | |
1134 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1135 | } |
1136 | /* | |
1137 | * All done at this level ... ascend and resume the search. | |
1138 | */ | |
1139 | if (this_parent != parent) { | |
2fd6b7f5 | 1140 | struct dentry *tmp; |
949854d0 NP |
1141 | struct dentry *child; |
1142 | ||
2fd6b7f5 | 1143 | tmp = this_parent->d_parent; |
949854d0 | 1144 | rcu_read_lock(); |
2fd6b7f5 | 1145 | spin_unlock(&this_parent->d_lock); |
949854d0 | 1146 | child = this_parent; |
2fd6b7f5 NP |
1147 | this_parent = tmp; |
1148 | spin_lock(&this_parent->d_lock); | |
949854d0 NP |
1149 | /* might go back up the wrong parent if we have had a rename |
1150 | * or deletion */ | |
1151 | if (this_parent != child->d_parent || | |
58db63d0 | 1152 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 1153 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
1154 | rcu_read_unlock(); |
1155 | goto rename_retry; | |
1156 | } | |
1157 | rcu_read_unlock(); | |
1158 | next = child->d_u.d_child.next; | |
1da177e4 LT |
1159 | goto resume; |
1160 | } | |
1161 | out: | |
2fd6b7f5 | 1162 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1163 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1164 | goto rename_retry; |
58db63d0 NP |
1165 | if (locked) |
1166 | write_sequnlock(&rename_lock); | |
1da177e4 | 1167 | return found; |
58db63d0 NP |
1168 | |
1169 | rename_retry: | |
1170 | if (found) | |
1171 | return found; | |
1172 | locked = 1; | |
1173 | write_seqlock(&rename_lock); | |
1174 | goto again; | |
1da177e4 LT |
1175 | } |
1176 | ||
1177 | /** | |
1178 | * shrink_dcache_parent - prune dcache | |
1179 | * @parent: parent of entries to prune | |
1180 | * | |
1181 | * Prune the dcache to remove unused children of the parent dentry. | |
1182 | */ | |
1183 | ||
1184 | void shrink_dcache_parent(struct dentry * parent) | |
1185 | { | |
da3bbdd4 | 1186 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1187 | int found; |
1188 | ||
1189 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 1190 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 1191 | } |
ec4f8605 | 1192 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1193 | |
1da177e4 LT |
1194 | /* |
1195 | * Scan `nr' dentries and return the number which remain. | |
1196 | * | |
1197 | * We need to avoid reentering the filesystem if the caller is performing a | |
1198 | * GFP_NOFS allocation attempt. One example deadlock is: | |
1199 | * | |
1200 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
1201 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
1202 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
1203 | * | |
1204 | * In this case we return -1 to tell the caller that we baled. | |
1205 | */ | |
7f8275d0 | 1206 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
1207 | { |
1208 | if (nr) { | |
1209 | if (!(gfp_mask & __GFP_FS)) | |
1210 | return -1; | |
da3bbdd4 | 1211 | prune_dcache(nr); |
1da177e4 | 1212 | } |
312d3ca8 | 1213 | |
86c8749e | 1214 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
1215 | } |
1216 | ||
8e1f936b RR |
1217 | static struct shrinker dcache_shrinker = { |
1218 | .shrink = shrink_dcache_memory, | |
1219 | .seeks = DEFAULT_SEEKS, | |
1220 | }; | |
1221 | ||
1da177e4 LT |
1222 | /** |
1223 | * d_alloc - allocate a dcache entry | |
1224 | * @parent: parent of entry to allocate | |
1225 | * @name: qstr of the name | |
1226 | * | |
1227 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1228 | * available. On a success the dentry is returned. The name passed in is | |
1229 | * copied and the copy passed in may be reused after this call. | |
1230 | */ | |
1231 | ||
1232 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1233 | { | |
1234 | struct dentry *dentry; | |
1235 | char *dname; | |
1236 | ||
e12ba74d | 1237 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1238 | if (!dentry) |
1239 | return NULL; | |
1240 | ||
1241 | if (name->len > DNAME_INLINE_LEN-1) { | |
1242 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1243 | if (!dname) { | |
1244 | kmem_cache_free(dentry_cache, dentry); | |
1245 | return NULL; | |
1246 | } | |
1247 | } else { | |
1248 | dname = dentry->d_iname; | |
1249 | } | |
1250 | dentry->d_name.name = dname; | |
1251 | ||
1252 | dentry->d_name.len = name->len; | |
1253 | dentry->d_name.hash = name->hash; | |
1254 | memcpy(dname, name->name, name->len); | |
1255 | dname[name->len] = 0; | |
1256 | ||
b7ab39f6 | 1257 | dentry->d_count = 1; |
1da177e4 LT |
1258 | dentry->d_flags = DCACHE_UNHASHED; |
1259 | spin_lock_init(&dentry->d_lock); | |
31e6b01f | 1260 | seqcount_init(&dentry->d_seq); |
1da177e4 LT |
1261 | dentry->d_inode = NULL; |
1262 | dentry->d_parent = NULL; | |
1263 | dentry->d_sb = NULL; | |
1264 | dentry->d_op = NULL; | |
1265 | dentry->d_fsdata = NULL; | |
1da177e4 LT |
1266 | INIT_HLIST_NODE(&dentry->d_hash); |
1267 | INIT_LIST_HEAD(&dentry->d_lru); | |
1268 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1269 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1270 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
1271 | |
1272 | if (parent) { | |
2fd6b7f5 | 1273 | spin_lock(&parent->d_lock); |
89ad485f NP |
1274 | /* |
1275 | * don't need child lock because it is not subject | |
1276 | * to concurrency here | |
1277 | */ | |
dc0474be NP |
1278 | __dget_dlock(parent); |
1279 | dentry->d_parent = parent; | |
1da177e4 | 1280 | dentry->d_sb = parent->d_sb; |
5160ee6f | 1281 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
2fd6b7f5 | 1282 | spin_unlock(&parent->d_lock); |
2fd6b7f5 | 1283 | } |
1da177e4 | 1284 | |
3e880fb5 | 1285 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1286 | |
1da177e4 LT |
1287 | return dentry; |
1288 | } | |
ec4f8605 | 1289 | EXPORT_SYMBOL(d_alloc); |
1da177e4 LT |
1290 | |
1291 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) | |
1292 | { | |
1293 | struct qstr q; | |
1294 | ||
1295 | q.name = name; | |
1296 | q.len = strlen(name); | |
1297 | q.hash = full_name_hash(q.name, q.len); | |
1298 | return d_alloc(parent, &q); | |
1299 | } | |
ef26ca97 | 1300 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1301 | |
fb045adb NP |
1302 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1303 | { | |
1304 | BUG_ON(dentry->d_op); | |
1305 | BUG_ON(dentry->d_flags & (DCACHE_OP_HASH | | |
1306 | DCACHE_OP_COMPARE | | |
1307 | DCACHE_OP_REVALIDATE | | |
1308 | DCACHE_OP_DELETE )); | |
1309 | dentry->d_op = op; | |
1310 | if (!op) | |
1311 | return; | |
1312 | if (op->d_hash) | |
1313 | dentry->d_flags |= DCACHE_OP_HASH; | |
1314 | if (op->d_compare) | |
1315 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1316 | if (op->d_revalidate) | |
1317 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1318 | if (op->d_delete) | |
1319 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1320 | ||
1321 | } | |
1322 | EXPORT_SYMBOL(d_set_d_op); | |
1323 | ||
360da900 OH |
1324 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1325 | { | |
b23fb0a6 | 1326 | spin_lock(&dentry->d_lock); |
360da900 OH |
1327 | if (inode) |
1328 | list_add(&dentry->d_alias, &inode->i_dentry); | |
1329 | dentry->d_inode = inode; | |
31e6b01f | 1330 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1331 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1332 | fsnotify_d_instantiate(dentry, inode); |
1333 | } | |
1334 | ||
1da177e4 LT |
1335 | /** |
1336 | * d_instantiate - fill in inode information for a dentry | |
1337 | * @entry: dentry to complete | |
1338 | * @inode: inode to attach to this dentry | |
1339 | * | |
1340 | * Fill in inode information in the entry. | |
1341 | * | |
1342 | * This turns negative dentries into productive full members | |
1343 | * of society. | |
1344 | * | |
1345 | * NOTE! This assumes that the inode count has been incremented | |
1346 | * (or otherwise set) by the caller to indicate that it is now | |
1347 | * in use by the dcache. | |
1348 | */ | |
1349 | ||
1350 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1351 | { | |
28133c7b | 1352 | BUG_ON(!list_empty(&entry->d_alias)); |
b23fb0a6 | 1353 | spin_lock(&dcache_inode_lock); |
360da900 | 1354 | __d_instantiate(entry, inode); |
b23fb0a6 | 1355 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1356 | security_d_instantiate(entry, inode); |
1357 | } | |
ec4f8605 | 1358 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1359 | |
1360 | /** | |
1361 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1362 | * @entry: dentry to instantiate | |
1363 | * @inode: inode to attach to this dentry | |
1364 | * | |
1365 | * Fill in inode information in the entry. On success, it returns NULL. | |
1366 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1367 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1368 | * |
1369 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1370 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1371 | * |
1372 | * This also assumes that the inode count has been incremented | |
1373 | * (or otherwise set) by the caller to indicate that it is now | |
1374 | * in use by the dcache. | |
1da177e4 | 1375 | */ |
770bfad8 DH |
1376 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1377 | struct inode *inode) | |
1da177e4 LT |
1378 | { |
1379 | struct dentry *alias; | |
1380 | int len = entry->d_name.len; | |
1381 | const char *name = entry->d_name.name; | |
1382 | unsigned int hash = entry->d_name.hash; | |
1383 | ||
770bfad8 | 1384 | if (!inode) { |
360da900 | 1385 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1386 | return NULL; |
1387 | } | |
1388 | ||
1da177e4 LT |
1389 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1390 | struct qstr *qstr = &alias->d_name; | |
1391 | ||
9abca360 NP |
1392 | /* |
1393 | * Don't need alias->d_lock here, because aliases with | |
1394 | * d_parent == entry->d_parent are not subject to name or | |
1395 | * parent changes, because the parent inode i_mutex is held. | |
1396 | */ | |
1da177e4 LT |
1397 | if (qstr->hash != hash) |
1398 | continue; | |
1399 | if (alias->d_parent != entry->d_parent) | |
1400 | continue; | |
1401 | if (qstr->len != len) | |
1402 | continue; | |
1403 | if (memcmp(qstr->name, name, len)) | |
1404 | continue; | |
dc0474be | 1405 | __dget(alias); |
1da177e4 LT |
1406 | return alias; |
1407 | } | |
770bfad8 | 1408 | |
360da900 | 1409 | __d_instantiate(entry, inode); |
1da177e4 LT |
1410 | return NULL; |
1411 | } | |
770bfad8 DH |
1412 | |
1413 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1414 | { | |
1415 | struct dentry *result; | |
1416 | ||
1417 | BUG_ON(!list_empty(&entry->d_alias)); | |
1418 | ||
b23fb0a6 | 1419 | spin_lock(&dcache_inode_lock); |
770bfad8 | 1420 | result = __d_instantiate_unique(entry, inode); |
b23fb0a6 | 1421 | spin_unlock(&dcache_inode_lock); |
770bfad8 DH |
1422 | |
1423 | if (!result) { | |
1424 | security_d_instantiate(entry, inode); | |
1425 | return NULL; | |
1426 | } | |
1427 | ||
1428 | BUG_ON(!d_unhashed(result)); | |
1429 | iput(inode); | |
1430 | return result; | |
1431 | } | |
1432 | ||
1da177e4 LT |
1433 | EXPORT_SYMBOL(d_instantiate_unique); |
1434 | ||
1435 | /** | |
1436 | * d_alloc_root - allocate root dentry | |
1437 | * @root_inode: inode to allocate the root for | |
1438 | * | |
1439 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1440 | * instantiated and returned. %NULL is returned if there is insufficient | |
1441 | * memory or the inode passed is %NULL. | |
1442 | */ | |
1443 | ||
1444 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1445 | { | |
1446 | struct dentry *res = NULL; | |
1447 | ||
1448 | if (root_inode) { | |
1449 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1450 | ||
1451 | res = d_alloc(NULL, &name); | |
1452 | if (res) { | |
1453 | res->d_sb = root_inode->i_sb; | |
1454 | res->d_parent = res; | |
1455 | d_instantiate(res, root_inode); | |
1456 | } | |
1457 | } | |
1458 | return res; | |
1459 | } | |
ec4f8605 | 1460 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 LT |
1461 | |
1462 | static inline struct hlist_head *d_hash(struct dentry *parent, | |
1463 | unsigned long hash) | |
1464 | { | |
1465 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
1466 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
1467 | return dentry_hashtable + (hash & D_HASHMASK); | |
1468 | } | |
1469 | ||
4ea3ada2 CH |
1470 | /** |
1471 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1472 | * @inode: inode to allocate the dentry for | |
1473 | * | |
1474 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1475 | * similar open by handle operations. The returned dentry may be anonymous, | |
1476 | * or may have a full name (if the inode was already in the cache). | |
1477 | * | |
1478 | * When called on a directory inode, we must ensure that the inode only ever | |
1479 | * has one dentry. If a dentry is found, that is returned instead of | |
1480 | * allocating a new one. | |
1481 | * | |
1482 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1483 | * to the dentry. In case of an error the reference on the inode is released. |
1484 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1485 | * be passed in and will be the error will be propagate to the return value, | |
1486 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1487 | */ |
1488 | struct dentry *d_obtain_alias(struct inode *inode) | |
1489 | { | |
9308a612 CH |
1490 | static const struct qstr anonstring = { .name = "" }; |
1491 | struct dentry *tmp; | |
1492 | struct dentry *res; | |
4ea3ada2 CH |
1493 | |
1494 | if (!inode) | |
44003728 | 1495 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1496 | if (IS_ERR(inode)) |
1497 | return ERR_CAST(inode); | |
1498 | ||
9308a612 CH |
1499 | res = d_find_alias(inode); |
1500 | if (res) | |
1501 | goto out_iput; | |
1502 | ||
1503 | tmp = d_alloc(NULL, &anonstring); | |
1504 | if (!tmp) { | |
1505 | res = ERR_PTR(-ENOMEM); | |
1506 | goto out_iput; | |
4ea3ada2 | 1507 | } |
9308a612 CH |
1508 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1509 | ||
b5c84bf6 | 1510 | |
b23fb0a6 | 1511 | spin_lock(&dcache_inode_lock); |
9308a612 CH |
1512 | res = __d_find_alias(inode, 0); |
1513 | if (res) { | |
b23fb0a6 | 1514 | spin_unlock(&dcache_inode_lock); |
9308a612 CH |
1515 | dput(tmp); |
1516 | goto out_iput; | |
1517 | } | |
1518 | ||
1519 | /* attach a disconnected dentry */ | |
1520 | spin_lock(&tmp->d_lock); | |
1521 | tmp->d_sb = inode->i_sb; | |
1522 | tmp->d_inode = inode; | |
1523 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
1524 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1525 | list_add(&tmp->d_alias, &inode->i_dentry); | |
789680d1 | 1526 | spin_lock(&dcache_hash_lock); |
9308a612 | 1527 | hlist_add_head(&tmp->d_hash, &inode->i_sb->s_anon); |
789680d1 | 1528 | spin_unlock(&dcache_hash_lock); |
9308a612 | 1529 | spin_unlock(&tmp->d_lock); |
b23fb0a6 | 1530 | spin_unlock(&dcache_inode_lock); |
9308a612 | 1531 | |
9308a612 CH |
1532 | return tmp; |
1533 | ||
1534 | out_iput: | |
1535 | iput(inode); | |
1536 | return res; | |
4ea3ada2 | 1537 | } |
adc48720 | 1538 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1539 | |
1540 | /** | |
1541 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1542 | * @inode: the inode which may have a disconnected dentry | |
1543 | * @dentry: a negative dentry which we want to point to the inode. | |
1544 | * | |
1545 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1546 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1547 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1548 | * | |
1549 | * This is needed in the lookup routine of any filesystem that is exportable | |
1550 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1551 | * | |
1552 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1553 | * is returned. This matches the expected return value of ->lookup. | |
1554 | * | |
1555 | */ | |
1556 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1557 | { | |
1558 | struct dentry *new = NULL; | |
1559 | ||
21c0d8fd | 1560 | if (inode && S_ISDIR(inode->i_mode)) { |
b23fb0a6 | 1561 | spin_lock(&dcache_inode_lock); |
1da177e4 LT |
1562 | new = __d_find_alias(inode, 1); |
1563 | if (new) { | |
1564 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
b23fb0a6 | 1565 | spin_unlock(&dcache_inode_lock); |
1da177e4 | 1566 | security_d_instantiate(new, inode); |
1da177e4 LT |
1567 | d_move(new, dentry); |
1568 | iput(inode); | |
1569 | } else { | |
b5c84bf6 | 1570 | /* already taking dcache_inode_lock, so d_add() by hand */ |
360da900 | 1571 | __d_instantiate(dentry, inode); |
b23fb0a6 | 1572 | spin_unlock(&dcache_inode_lock); |
1da177e4 LT |
1573 | security_d_instantiate(dentry, inode); |
1574 | d_rehash(dentry); | |
1575 | } | |
1576 | } else | |
1577 | d_add(dentry, inode); | |
1578 | return new; | |
1579 | } | |
ec4f8605 | 1580 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1581 | |
9403540c BN |
1582 | /** |
1583 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1584 | * @inode: the inode case-insensitive lookup has found | |
1585 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1586 | * @name: the case-exact name to be associated with the returned dentry | |
1587 | * | |
1588 | * This is to avoid filling the dcache with case-insensitive names to the | |
1589 | * same inode, only the actual correct case is stored in the dcache for | |
1590 | * case-insensitive filesystems. | |
1591 | * | |
1592 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1593 | * already exists in in the dcache, use it and return it. | |
1594 | * | |
1595 | * If no entry exists with the exact case name, allocate new dentry with | |
1596 | * the exact case, and return the spliced entry. | |
1597 | */ | |
e45b590b | 1598 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1599 | struct qstr *name) |
1600 | { | |
1601 | int error; | |
1602 | struct dentry *found; | |
1603 | struct dentry *new; | |
1604 | ||
b6520c81 CH |
1605 | /* |
1606 | * First check if a dentry matching the name already exists, | |
1607 | * if not go ahead and create it now. | |
1608 | */ | |
9403540c | 1609 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1610 | if (!found) { |
1611 | new = d_alloc(dentry->d_parent, name); | |
1612 | if (!new) { | |
1613 | error = -ENOMEM; | |
1614 | goto err_out; | |
1615 | } | |
b6520c81 | 1616 | |
9403540c BN |
1617 | found = d_splice_alias(inode, new); |
1618 | if (found) { | |
1619 | dput(new); | |
1620 | return found; | |
1621 | } | |
1622 | return new; | |
1623 | } | |
b6520c81 CH |
1624 | |
1625 | /* | |
1626 | * If a matching dentry exists, and it's not negative use it. | |
1627 | * | |
1628 | * Decrement the reference count to balance the iget() done | |
1629 | * earlier on. | |
1630 | */ | |
9403540c BN |
1631 | if (found->d_inode) { |
1632 | if (unlikely(found->d_inode != inode)) { | |
1633 | /* This can't happen because bad inodes are unhashed. */ | |
1634 | BUG_ON(!is_bad_inode(inode)); | |
1635 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1636 | } | |
9403540c BN |
1637 | iput(inode); |
1638 | return found; | |
1639 | } | |
b6520c81 | 1640 | |
9403540c BN |
1641 | /* |
1642 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1643 | * already has a dentry. |
9403540c | 1644 | */ |
b23fb0a6 | 1645 | spin_lock(&dcache_inode_lock); |
b6520c81 | 1646 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1647 | __d_instantiate(found, inode); |
b23fb0a6 | 1648 | spin_unlock(&dcache_inode_lock); |
9403540c BN |
1649 | security_d_instantiate(found, inode); |
1650 | return found; | |
1651 | } | |
b6520c81 | 1652 | |
9403540c | 1653 | /* |
b6520c81 CH |
1654 | * In case a directory already has a (disconnected) entry grab a |
1655 | * reference to it, move it in place and use it. | |
9403540c BN |
1656 | */ |
1657 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
dc0474be | 1658 | __dget(new); |
b23fb0a6 | 1659 | spin_unlock(&dcache_inode_lock); |
9403540c | 1660 | security_d_instantiate(found, inode); |
9403540c | 1661 | d_move(new, found); |
9403540c | 1662 | iput(inode); |
9403540c | 1663 | dput(found); |
9403540c BN |
1664 | return new; |
1665 | ||
1666 | err_out: | |
1667 | iput(inode); | |
1668 | return ERR_PTR(error); | |
1669 | } | |
ec4f8605 | 1670 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1671 | |
31e6b01f NP |
1672 | /** |
1673 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1674 | * @parent: parent dentry | |
1675 | * @name: qstr of name we wish to find | |
1676 | * @seq: returns d_seq value at the point where the dentry was found | |
1677 | * @inode: returns dentry->d_inode when the inode was found valid. | |
1678 | * Returns: dentry, or NULL | |
1679 | * | |
1680 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1681 | * resolution (store-free path walking) design described in | |
1682 | * Documentation/filesystems/path-lookup.txt. | |
1683 | * | |
1684 | * This is not to be used outside core vfs. | |
1685 | * | |
1686 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1687 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1688 | * without taking d_lock and checking d_seq sequence count against @seq | |
1689 | * returned here. | |
1690 | * | |
1691 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1692 | * function. | |
1693 | * | |
1694 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1695 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1696 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1697 | * is formed, giving integrity down the path walk. | |
1698 | */ | |
1699 | struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name, | |
1700 | unsigned *seq, struct inode **inode) | |
1701 | { | |
1702 | unsigned int len = name->len; | |
1703 | unsigned int hash = name->hash; | |
1704 | const unsigned char *str = name->name; | |
1705 | struct hlist_head *head = d_hash(parent, hash); | |
1706 | struct hlist_node *node; | |
1707 | struct dentry *dentry; | |
1708 | ||
1709 | /* | |
1710 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1711 | * required to prevent single threaded performance regressions | |
1712 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1713 | * Keep the two functions in sync. | |
1714 | */ | |
1715 | ||
1716 | /* | |
1717 | * The hash list is protected using RCU. | |
1718 | * | |
1719 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1720 | * races with d_move(). | |
1721 | * | |
1722 | * It is possible that concurrent renames can mess up our list | |
1723 | * walk here and result in missing our dentry, resulting in the | |
1724 | * false-negative result. d_lookup() protects against concurrent | |
1725 | * renames using rename_lock seqlock. | |
1726 | * | |
1727 | * See Documentation/vfs/dcache-locking.txt for more details. | |
1728 | */ | |
1729 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { | |
1730 | struct inode *i; | |
1731 | const char *tname; | |
1732 | int tlen; | |
1733 | ||
1734 | if (dentry->d_name.hash != hash) | |
1735 | continue; | |
1736 | ||
1737 | seqretry: | |
1738 | *seq = read_seqcount_begin(&dentry->d_seq); | |
1739 | if (dentry->d_parent != parent) | |
1740 | continue; | |
1741 | if (d_unhashed(dentry)) | |
1742 | continue; | |
1743 | tlen = dentry->d_name.len; | |
1744 | tname = dentry->d_name.name; | |
1745 | i = dentry->d_inode; | |
1746 | /* | |
1747 | * This seqcount check is required to ensure name and | |
1748 | * len are loaded atomically, so as not to walk off the | |
1749 | * edge of memory when walking. If we could load this | |
1750 | * atomically some other way, we could drop this check. | |
1751 | */ | |
1752 | if (read_seqcount_retry(&dentry->d_seq, *seq)) | |
1753 | goto seqretry; | |
fb045adb | 1754 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
31e6b01f NP |
1755 | if (parent->d_op->d_compare(parent, *inode, |
1756 | dentry, i, | |
1757 | tlen, tname, name)) | |
1758 | continue; | |
1759 | } else { | |
1760 | if (tlen != len) | |
1761 | continue; | |
1762 | if (memcmp(tname, str, tlen)) | |
1763 | continue; | |
1764 | } | |
1765 | /* | |
1766 | * No extra seqcount check is required after the name | |
1767 | * compare. The caller must perform a seqcount check in | |
1768 | * order to do anything useful with the returned dentry | |
1769 | * anyway. | |
1770 | */ | |
1771 | *inode = i; | |
1772 | return dentry; | |
1773 | } | |
1774 | return NULL; | |
1775 | } | |
1776 | ||
1da177e4 LT |
1777 | /** |
1778 | * d_lookup - search for a dentry | |
1779 | * @parent: parent dentry | |
1780 | * @name: qstr of name we wish to find | |
b04f784e | 1781 | * Returns: dentry, or NULL |
1da177e4 | 1782 | * |
b04f784e NP |
1783 | * d_lookup searches the children of the parent dentry for the name in |
1784 | * question. If the dentry is found its reference count is incremented and the | |
1785 | * dentry is returned. The caller must use dput to free the entry when it has | |
1786 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1787 | */ |
31e6b01f | 1788 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 | 1789 | { |
31e6b01f | 1790 | struct dentry *dentry; |
949854d0 | 1791 | unsigned seq; |
1da177e4 LT |
1792 | |
1793 | do { | |
1794 | seq = read_seqbegin(&rename_lock); | |
1795 | dentry = __d_lookup(parent, name); | |
1796 | if (dentry) | |
1797 | break; | |
1798 | } while (read_seqretry(&rename_lock, seq)); | |
1799 | return dentry; | |
1800 | } | |
ec4f8605 | 1801 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1802 | |
31e6b01f | 1803 | /** |
b04f784e NP |
1804 | * __d_lookup - search for a dentry (racy) |
1805 | * @parent: parent dentry | |
1806 | * @name: qstr of name we wish to find | |
1807 | * Returns: dentry, or NULL | |
1808 | * | |
1809 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1810 | * false-negative result due to unrelated rename activity. | |
1811 | * | |
1812 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1813 | * however it must be used carefully, eg. with a following d_lookup in | |
1814 | * the case of failure. | |
1815 | * | |
1816 | * __d_lookup callers must be commented. | |
1817 | */ | |
31e6b01f | 1818 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 LT |
1819 | { |
1820 | unsigned int len = name->len; | |
1821 | unsigned int hash = name->hash; | |
1822 | const unsigned char *str = name->name; | |
1823 | struct hlist_head *head = d_hash(parent,hash); | |
1da177e4 | 1824 | struct hlist_node *node; |
31e6b01f | 1825 | struct dentry *found = NULL; |
665a7583 | 1826 | struct dentry *dentry; |
1da177e4 | 1827 | |
31e6b01f NP |
1828 | /* |
1829 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1830 | * required to prevent single threaded performance regressions | |
1831 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1832 | * Keep the two functions in sync. | |
1833 | */ | |
1834 | ||
b04f784e NP |
1835 | /* |
1836 | * The hash list is protected using RCU. | |
1837 | * | |
1838 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1839 | * with d_move(). | |
1840 | * | |
1841 | * It is possible that concurrent renames can mess up our list | |
1842 | * walk here and result in missing our dentry, resulting in the | |
1843 | * false-negative result. d_lookup() protects against concurrent | |
1844 | * renames using rename_lock seqlock. | |
1845 | * | |
1846 | * See Documentation/vfs/dcache-locking.txt for more details. | |
1847 | */ | |
1da177e4 LT |
1848 | rcu_read_lock(); |
1849 | ||
665a7583 | 1850 | hlist_for_each_entry_rcu(dentry, node, head, d_hash) { |
31e6b01f NP |
1851 | const char *tname; |
1852 | int tlen; | |
1da177e4 | 1853 | |
1da177e4 LT |
1854 | if (dentry->d_name.hash != hash) |
1855 | continue; | |
1da177e4 LT |
1856 | |
1857 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1858 | if (dentry->d_parent != parent) |
1859 | goto next; | |
d0185c08 LT |
1860 | if (d_unhashed(dentry)) |
1861 | goto next; | |
1862 | ||
1da177e4 LT |
1863 | /* |
1864 | * It is safe to compare names since d_move() cannot | |
1865 | * change the qstr (protected by d_lock). | |
1866 | */ | |
31e6b01f NP |
1867 | tlen = dentry->d_name.len; |
1868 | tname = dentry->d_name.name; | |
fb045adb | 1869 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
621e155a NP |
1870 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1871 | dentry, dentry->d_inode, | |
31e6b01f | 1872 | tlen, tname, name)) |
1da177e4 LT |
1873 | goto next; |
1874 | } else { | |
31e6b01f | 1875 | if (tlen != len) |
1da177e4 | 1876 | goto next; |
31e6b01f | 1877 | if (memcmp(tname, str, tlen)) |
1da177e4 LT |
1878 | goto next; |
1879 | } | |
1880 | ||
b7ab39f6 | 1881 | dentry->d_count++; |
d0185c08 | 1882 | found = dentry; |
1da177e4 LT |
1883 | spin_unlock(&dentry->d_lock); |
1884 | break; | |
1885 | next: | |
1886 | spin_unlock(&dentry->d_lock); | |
1887 | } | |
1888 | rcu_read_unlock(); | |
1889 | ||
1890 | return found; | |
1891 | } | |
1892 | ||
3e7e241f EB |
1893 | /** |
1894 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1895 | * @dir: Directory to search in | |
1896 | * @name: qstr of name we wish to find | |
1897 | * | |
1898 | * On hash failure or on lookup failure NULL is returned. | |
1899 | */ | |
1900 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1901 | { | |
1902 | struct dentry *dentry = NULL; | |
1903 | ||
1904 | /* | |
1905 | * Check for a fs-specific hash function. Note that we must | |
1906 | * calculate the standard hash first, as the d_op->d_hash() | |
1907 | * routine may choose to leave the hash value unchanged. | |
1908 | */ | |
1909 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1910 | if (dir->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 1911 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1912 | goto out; |
1913 | } | |
1914 | dentry = d_lookup(dir, name); | |
1915 | out: | |
1916 | return dentry; | |
1917 | } | |
1918 | ||
1da177e4 | 1919 | /** |
786a5e15 | 1920 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 LT |
1921 | * @dentry: The dentry alleged to be valid child of @dparent |
1922 | * @dparent: The parent dentry (known to be valid) | |
1da177e4 LT |
1923 | * |
1924 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1925 | * This is used by ncpfs in its readdir implementation. | |
1926 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1927 | * |
1928 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1929 | */ |
d3a23e16 | 1930 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1931 | { |
786a5e15 | 1932 | struct dentry *child; |
d3a23e16 | 1933 | |
2fd6b7f5 | 1934 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
1935 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
1936 | if (dentry == child) { | |
2fd6b7f5 | 1937 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 1938 | __dget_dlock(dentry); |
2fd6b7f5 NP |
1939 | spin_unlock(&dentry->d_lock); |
1940 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
1941 | return 1; |
1942 | } | |
1943 | } | |
2fd6b7f5 | 1944 | spin_unlock(&dparent->d_lock); |
786a5e15 | 1945 | |
1da177e4 LT |
1946 | return 0; |
1947 | } | |
ec4f8605 | 1948 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1949 | |
1950 | /* | |
1951 | * When a file is deleted, we have two options: | |
1952 | * - turn this dentry into a negative dentry | |
1953 | * - unhash this dentry and free it. | |
1954 | * | |
1955 | * Usually, we want to just turn this into | |
1956 | * a negative dentry, but if anybody else is | |
1957 | * currently using the dentry or the inode | |
1958 | * we can't do that and we fall back on removing | |
1959 | * it from the hash queues and waiting for | |
1960 | * it to be deleted later when it has no users | |
1961 | */ | |
1962 | ||
1963 | /** | |
1964 | * d_delete - delete a dentry | |
1965 | * @dentry: The dentry to delete | |
1966 | * | |
1967 | * Turn the dentry into a negative dentry if possible, otherwise | |
1968 | * remove it from the hash queues so it can be deleted later | |
1969 | */ | |
1970 | ||
1971 | void d_delete(struct dentry * dentry) | |
1972 | { | |
7a91bf7f | 1973 | int isdir = 0; |
1da177e4 LT |
1974 | /* |
1975 | * Are we the only user? | |
1976 | */ | |
357f8e65 | 1977 | again: |
1da177e4 | 1978 | spin_lock(&dentry->d_lock); |
7a91bf7f | 1979 | isdir = S_ISDIR(dentry->d_inode->i_mode); |
b7ab39f6 | 1980 | if (dentry->d_count == 1) { |
357f8e65 NP |
1981 | if (!spin_trylock(&dcache_inode_lock)) { |
1982 | spin_unlock(&dentry->d_lock); | |
1983 | cpu_relax(); | |
1984 | goto again; | |
1985 | } | |
13e3c5e5 | 1986 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 1987 | dentry_unlink_inode(dentry); |
7a91bf7f | 1988 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1989 | return; |
1990 | } | |
1991 | ||
1992 | if (!d_unhashed(dentry)) | |
1993 | __d_drop(dentry); | |
1994 | ||
1995 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
1996 | |
1997 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1998 | } |
ec4f8605 | 1999 | EXPORT_SYMBOL(d_delete); |
1da177e4 LT |
2000 | |
2001 | static void __d_rehash(struct dentry * entry, struct hlist_head *list) | |
2002 | { | |
2003 | ||
2004 | entry->d_flags &= ~DCACHE_UNHASHED; | |
2005 | hlist_add_head_rcu(&entry->d_hash, list); | |
2006 | } | |
2007 | ||
770bfad8 DH |
2008 | static void _d_rehash(struct dentry * entry) |
2009 | { | |
2010 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2011 | } | |
2012 | ||
1da177e4 LT |
2013 | /** |
2014 | * d_rehash - add an entry back to the hash | |
2015 | * @entry: dentry to add to the hash | |
2016 | * | |
2017 | * Adds a dentry to the hash according to its name. | |
2018 | */ | |
2019 | ||
2020 | void d_rehash(struct dentry * entry) | |
2021 | { | |
1da177e4 | 2022 | spin_lock(&entry->d_lock); |
789680d1 | 2023 | spin_lock(&dcache_hash_lock); |
770bfad8 | 2024 | _d_rehash(entry); |
789680d1 | 2025 | spin_unlock(&dcache_hash_lock); |
1da177e4 | 2026 | spin_unlock(&entry->d_lock); |
1da177e4 | 2027 | } |
ec4f8605 | 2028 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2029 | |
fb2d5b86 NP |
2030 | /** |
2031 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2032 | * @dentry: dentry to be updated | |
2033 | * @name: new name | |
2034 | * | |
2035 | * Update a case insensitive dentry with new case of name. | |
2036 | * | |
2037 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2038 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2039 | * lengths). | |
2040 | * | |
2041 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2042 | * keep renames and concurrent inserts, and readdir(2) away). | |
2043 | */ | |
2044 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2045 | { | |
2046 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
2047 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | |
2048 | ||
fb2d5b86 | 2049 | spin_lock(&dentry->d_lock); |
31e6b01f | 2050 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2051 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2052 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2053 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2054 | } |
2055 | EXPORT_SYMBOL(dentry_update_name_case); | |
2056 | ||
1da177e4 LT |
2057 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2058 | { | |
2059 | if (dname_external(target)) { | |
2060 | if (dname_external(dentry)) { | |
2061 | /* | |
2062 | * Both external: swap the pointers | |
2063 | */ | |
9a8d5bb4 | 2064 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2065 | } else { |
2066 | /* | |
2067 | * dentry:internal, target:external. Steal target's | |
2068 | * storage and make target internal. | |
2069 | */ | |
321bcf92 BF |
2070 | memcpy(target->d_iname, dentry->d_name.name, |
2071 | dentry->d_name.len + 1); | |
1da177e4 LT |
2072 | dentry->d_name.name = target->d_name.name; |
2073 | target->d_name.name = target->d_iname; | |
2074 | } | |
2075 | } else { | |
2076 | if (dname_external(dentry)) { | |
2077 | /* | |
2078 | * dentry:external, target:internal. Give dentry's | |
2079 | * storage to target and make dentry internal | |
2080 | */ | |
2081 | memcpy(dentry->d_iname, target->d_name.name, | |
2082 | target->d_name.len + 1); | |
2083 | target->d_name.name = dentry->d_name.name; | |
2084 | dentry->d_name.name = dentry->d_iname; | |
2085 | } else { | |
2086 | /* | |
2087 | * Both are internal. Just copy target to dentry | |
2088 | */ | |
2089 | memcpy(dentry->d_iname, target->d_name.name, | |
2090 | target->d_name.len + 1); | |
dc711ca3 AV |
2091 | dentry->d_name.len = target->d_name.len; |
2092 | return; | |
1da177e4 LT |
2093 | } |
2094 | } | |
9a8d5bb4 | 2095 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2096 | } |
2097 | ||
2fd6b7f5 NP |
2098 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2099 | { | |
2100 | /* | |
2101 | * XXXX: do we really need to take target->d_lock? | |
2102 | */ | |
2103 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2104 | spin_lock(&target->d_parent->d_lock); | |
2105 | else { | |
2106 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2107 | spin_lock(&dentry->d_parent->d_lock); | |
2108 | spin_lock_nested(&target->d_parent->d_lock, | |
2109 | DENTRY_D_LOCK_NESTED); | |
2110 | } else { | |
2111 | spin_lock(&target->d_parent->d_lock); | |
2112 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2113 | DENTRY_D_LOCK_NESTED); | |
2114 | } | |
2115 | } | |
2116 | if (target < dentry) { | |
2117 | spin_lock_nested(&target->d_lock, 2); | |
2118 | spin_lock_nested(&dentry->d_lock, 3); | |
2119 | } else { | |
2120 | spin_lock_nested(&dentry->d_lock, 2); | |
2121 | spin_lock_nested(&target->d_lock, 3); | |
2122 | } | |
2123 | } | |
2124 | ||
2125 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2126 | struct dentry *target) | |
2127 | { | |
2128 | if (target->d_parent != dentry->d_parent) | |
2129 | spin_unlock(&dentry->d_parent->d_lock); | |
2130 | if (target->d_parent != target) | |
2131 | spin_unlock(&target->d_parent->d_lock); | |
2132 | } | |
2133 | ||
1da177e4 | 2134 | /* |
2fd6b7f5 NP |
2135 | * When switching names, the actual string doesn't strictly have to |
2136 | * be preserved in the target - because we're dropping the target | |
2137 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2138 | * the new name before we switch. | |
2139 | * | |
2140 | * Note that we have to be a lot more careful about getting the hash | |
2141 | * switched - we have to switch the hash value properly even if it | |
2142 | * then no longer matches the actual (corrupted) string of the target. | |
2143 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2144 | */ |
9eaef27b | 2145 | /* |
b5c84bf6 | 2146 | * d_move - move a dentry |
1da177e4 LT |
2147 | * @dentry: entry to move |
2148 | * @target: new dentry | |
2149 | * | |
2150 | * Update the dcache to reflect the move of a file name. Negative | |
2151 | * dcache entries should not be moved in this way. | |
2152 | */ | |
b5c84bf6 | 2153 | void d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2154 | { |
1da177e4 LT |
2155 | if (!dentry->d_inode) |
2156 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2157 | ||
2fd6b7f5 NP |
2158 | BUG_ON(d_ancestor(dentry, target)); |
2159 | BUG_ON(d_ancestor(target, dentry)); | |
2160 | ||
1da177e4 | 2161 | write_seqlock(&rename_lock); |
2fd6b7f5 NP |
2162 | |
2163 | dentry_lock_for_move(dentry, target); | |
1da177e4 | 2164 | |
31e6b01f NP |
2165 | write_seqcount_begin(&dentry->d_seq); |
2166 | write_seqcount_begin(&target->d_seq); | |
2167 | ||
1da177e4 | 2168 | /* Move the dentry to the target hash queue, if on different bucket */ |
789680d1 NP |
2169 | spin_lock(&dcache_hash_lock); |
2170 | if (!d_unhashed(dentry)) | |
2171 | hlist_del_rcu(&dentry->d_hash); | |
2172 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); | |
2173 | spin_unlock(&dcache_hash_lock); | |
1da177e4 LT |
2174 | |
2175 | /* Unhash the target: dput() will then get rid of it */ | |
31e6b01f | 2176 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
1da177e4 LT |
2177 | __d_drop(target); |
2178 | ||
5160ee6f ED |
2179 | list_del(&dentry->d_u.d_child); |
2180 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2181 | |
2182 | /* Switch the names.. */ | |
2183 | switch_names(dentry, target); | |
9a8d5bb4 | 2184 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2185 | |
2186 | /* ... and switch the parents */ | |
2187 | if (IS_ROOT(dentry)) { | |
2188 | dentry->d_parent = target->d_parent; | |
2189 | target->d_parent = target; | |
5160ee6f | 2190 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2191 | } else { |
9a8d5bb4 | 2192 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2193 | |
2194 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2195 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2196 | } |
2197 | ||
5160ee6f | 2198 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2199 | |
31e6b01f NP |
2200 | write_seqcount_end(&target->d_seq); |
2201 | write_seqcount_end(&dentry->d_seq); | |
2202 | ||
2fd6b7f5 | 2203 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2204 | spin_unlock(&target->d_lock); |
c32ccd87 | 2205 | fsnotify_d_move(dentry); |
1da177e4 LT |
2206 | spin_unlock(&dentry->d_lock); |
2207 | write_sequnlock(&rename_lock); | |
9eaef27b | 2208 | } |
ec4f8605 | 2209 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2210 | |
e2761a11 OH |
2211 | /** |
2212 | * d_ancestor - search for an ancestor | |
2213 | * @p1: ancestor dentry | |
2214 | * @p2: child dentry | |
2215 | * | |
2216 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2217 | * an ancestor of p2, else NULL. | |
9eaef27b | 2218 | */ |
e2761a11 | 2219 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2220 | { |
2221 | struct dentry *p; | |
2222 | ||
871c0067 | 2223 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2224 | if (p->d_parent == p1) |
e2761a11 | 2225 | return p; |
9eaef27b | 2226 | } |
e2761a11 | 2227 | return NULL; |
9eaef27b TM |
2228 | } |
2229 | ||
2230 | /* | |
2231 | * This helper attempts to cope with remotely renamed directories | |
2232 | * | |
2233 | * It assumes that the caller is already holding | |
b5c84bf6 | 2234 | * dentry->d_parent->d_inode->i_mutex and the dcache_inode_lock |
9eaef27b TM |
2235 | * |
2236 | * Note: If ever the locking in lock_rename() changes, then please | |
2237 | * remember to update this too... | |
9eaef27b TM |
2238 | */ |
2239 | static struct dentry *__d_unalias(struct dentry *dentry, struct dentry *alias) | |
b23fb0a6 | 2240 | __releases(dcache_inode_lock) |
9eaef27b TM |
2241 | { |
2242 | struct mutex *m1 = NULL, *m2 = NULL; | |
2243 | struct dentry *ret; | |
2244 | ||
2245 | /* If alias and dentry share a parent, then no extra locks required */ | |
2246 | if (alias->d_parent == dentry->d_parent) | |
2247 | goto out_unalias; | |
2248 | ||
2249 | /* Check for loops */ | |
2250 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 2251 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
2252 | goto out_err; |
2253 | ||
2254 | /* See lock_rename() */ | |
2255 | ret = ERR_PTR(-EBUSY); | |
2256 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2257 | goto out_err; | |
2258 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2259 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2260 | goto out_err; | |
2261 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2262 | out_unalias: | |
b5c84bf6 | 2263 | d_move(alias, dentry); |
9eaef27b TM |
2264 | ret = alias; |
2265 | out_err: | |
b23fb0a6 | 2266 | spin_unlock(&dcache_inode_lock); |
9eaef27b TM |
2267 | if (m2) |
2268 | mutex_unlock(m2); | |
2269 | if (m1) | |
2270 | mutex_unlock(m1); | |
2271 | return ret; | |
2272 | } | |
2273 | ||
770bfad8 DH |
2274 | /* |
2275 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2276 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2277 | * returns with anon->d_lock held! |
770bfad8 DH |
2278 | */ |
2279 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2280 | { | |
2281 | struct dentry *dparent, *aparent; | |
2282 | ||
2fd6b7f5 | 2283 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2284 | |
31e6b01f NP |
2285 | write_seqcount_begin(&dentry->d_seq); |
2286 | write_seqcount_begin(&anon->d_seq); | |
2287 | ||
770bfad8 DH |
2288 | dparent = dentry->d_parent; |
2289 | aparent = anon->d_parent; | |
2290 | ||
2fd6b7f5 NP |
2291 | switch_names(dentry, anon); |
2292 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2293 | ||
770bfad8 DH |
2294 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2295 | list_del(&dentry->d_u.d_child); | |
2296 | if (!IS_ROOT(dentry)) | |
2297 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2298 | else | |
2299 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2300 | ||
2301 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2302 | list_del(&anon->d_u.d_child); | |
2303 | if (!IS_ROOT(anon)) | |
2304 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2305 | else | |
2306 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2307 | ||
31e6b01f NP |
2308 | write_seqcount_end(&dentry->d_seq); |
2309 | write_seqcount_end(&anon->d_seq); | |
2310 | ||
2fd6b7f5 NP |
2311 | dentry_unlock_parents_for_move(anon, dentry); |
2312 | spin_unlock(&dentry->d_lock); | |
2313 | ||
2314 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2315 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2316 | } | |
2317 | ||
2318 | /** | |
2319 | * d_materialise_unique - introduce an inode into the tree | |
2320 | * @dentry: candidate dentry | |
2321 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2322 | * | |
2323 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2324 | * root directory alias in its place if there is one | |
2325 | */ | |
2326 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2327 | { | |
9eaef27b | 2328 | struct dentry *actual; |
770bfad8 DH |
2329 | |
2330 | BUG_ON(!d_unhashed(dentry)); | |
2331 | ||
770bfad8 DH |
2332 | if (!inode) { |
2333 | actual = dentry; | |
360da900 | 2334 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2335 | d_rehash(actual); |
2336 | goto out_nolock; | |
770bfad8 DH |
2337 | } |
2338 | ||
357f8e65 NP |
2339 | spin_lock(&dcache_inode_lock); |
2340 | ||
9eaef27b TM |
2341 | if (S_ISDIR(inode->i_mode)) { |
2342 | struct dentry *alias; | |
2343 | ||
2344 | /* Does an aliased dentry already exist? */ | |
2345 | alias = __d_find_alias(inode, 0); | |
2346 | if (alias) { | |
2347 | actual = alias; | |
2348 | /* Is this an anonymous mountpoint that we could splice | |
2349 | * into our tree? */ | |
2350 | if (IS_ROOT(alias)) { | |
9eaef27b TM |
2351 | __d_materialise_dentry(dentry, alias); |
2352 | __d_drop(alias); | |
2353 | goto found; | |
2354 | } | |
2355 | /* Nope, but we must(!) avoid directory aliasing */ | |
2356 | actual = __d_unalias(dentry, alias); | |
2357 | if (IS_ERR(actual)) | |
2358 | dput(alias); | |
2359 | goto out_nolock; | |
2360 | } | |
770bfad8 DH |
2361 | } |
2362 | ||
2363 | /* Add a unique reference */ | |
2364 | actual = __d_instantiate_unique(dentry, inode); | |
2365 | if (!actual) | |
2366 | actual = dentry; | |
357f8e65 NP |
2367 | else |
2368 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2369 | |
770bfad8 DH |
2370 | spin_lock(&actual->d_lock); |
2371 | found: | |
789680d1 | 2372 | spin_lock(&dcache_hash_lock); |
770bfad8 | 2373 | _d_rehash(actual); |
789680d1 | 2374 | spin_unlock(&dcache_hash_lock); |
770bfad8 | 2375 | spin_unlock(&actual->d_lock); |
b23fb0a6 | 2376 | spin_unlock(&dcache_inode_lock); |
9eaef27b | 2377 | out_nolock: |
770bfad8 DH |
2378 | if (actual == dentry) { |
2379 | security_d_instantiate(dentry, inode); | |
2380 | return NULL; | |
2381 | } | |
2382 | ||
2383 | iput(inode); | |
2384 | return actual; | |
770bfad8 | 2385 | } |
ec4f8605 | 2386 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2387 | |
cdd16d02 | 2388 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2389 | { |
2390 | *buflen -= namelen; | |
2391 | if (*buflen < 0) | |
2392 | return -ENAMETOOLONG; | |
2393 | *buffer -= namelen; | |
2394 | memcpy(*buffer, str, namelen); | |
2395 | return 0; | |
2396 | } | |
2397 | ||
cdd16d02 MS |
2398 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2399 | { | |
2400 | return prepend(buffer, buflen, name->name, name->len); | |
2401 | } | |
2402 | ||
1da177e4 | 2403 | /** |
f2eb6575 MS |
2404 | * Prepend path string to a buffer |
2405 | * | |
9d1bc601 MS |
2406 | * @path: the dentry/vfsmount to report |
2407 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2408 | * @buffer: pointer to the end of the buffer |
2409 | * @buflen: pointer to buffer length | |
552ce544 | 2410 | * |
949854d0 | 2411 | * Caller holds the rename_lock. |
9d1bc601 MS |
2412 | * |
2413 | * If path is not reachable from the supplied root, then the value of | |
2414 | * root is changed (without modifying refcounts). | |
1da177e4 | 2415 | */ |
f2eb6575 MS |
2416 | static int prepend_path(const struct path *path, struct path *root, |
2417 | char **buffer, int *buflen) | |
1da177e4 | 2418 | { |
9d1bc601 MS |
2419 | struct dentry *dentry = path->dentry; |
2420 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2421 | bool slash = false; |
2422 | int error = 0; | |
6092d048 | 2423 | |
99b7db7b | 2424 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2425 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2426 | struct dentry * parent; |
2427 | ||
1da177e4 | 2428 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2429 | /* Global root? */ |
1da177e4 | 2430 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2431 | goto global_root; |
2432 | } | |
2433 | dentry = vfsmnt->mnt_mountpoint; | |
2434 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2435 | continue; |
2436 | } | |
2437 | parent = dentry->d_parent; | |
2438 | prefetch(parent); | |
9abca360 | 2439 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2440 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2441 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2442 | if (!error) |
2443 | error = prepend(buffer, buflen, "/", 1); | |
2444 | if (error) | |
2445 | break; | |
2446 | ||
2447 | slash = true; | |
1da177e4 LT |
2448 | dentry = parent; |
2449 | } | |
2450 | ||
be285c71 | 2451 | out: |
f2eb6575 MS |
2452 | if (!error && !slash) |
2453 | error = prepend(buffer, buflen, "/", 1); | |
2454 | ||
99b7db7b | 2455 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2456 | return error; |
1da177e4 LT |
2457 | |
2458 | global_root: | |
98dc568b MS |
2459 | /* |
2460 | * Filesystems needing to implement special "root names" | |
2461 | * should do so with ->d_dname() | |
2462 | */ | |
2463 | if (IS_ROOT(dentry) && | |
2464 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2465 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2466 | (int) dentry->d_name.len, dentry->d_name.name); | |
2467 | } | |
9d1bc601 MS |
2468 | root->mnt = vfsmnt; |
2469 | root->dentry = dentry; | |
be285c71 | 2470 | goto out; |
f2eb6575 | 2471 | } |
be285c71 | 2472 | |
f2eb6575 MS |
2473 | /** |
2474 | * __d_path - return the path of a dentry | |
2475 | * @path: the dentry/vfsmount to report | |
2476 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2477 | * @buf: buffer to return value in |
f2eb6575 MS |
2478 | * @buflen: buffer length |
2479 | * | |
ffd1f4ed | 2480 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2481 | * |
2482 | * Returns a pointer into the buffer or an error code if the | |
2483 | * path was too long. | |
2484 | * | |
be148247 | 2485 | * "buflen" should be positive. |
f2eb6575 MS |
2486 | * |
2487 | * If path is not reachable from the supplied root, then the value of | |
2488 | * root is changed (without modifying refcounts). | |
2489 | */ | |
2490 | char *__d_path(const struct path *path, struct path *root, | |
2491 | char *buf, int buflen) | |
2492 | { | |
2493 | char *res = buf + buflen; | |
2494 | int error; | |
2495 | ||
2496 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2497 | write_seqlock(&rename_lock); |
f2eb6575 | 2498 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2499 | write_sequnlock(&rename_lock); |
be148247 | 2500 | |
f2eb6575 MS |
2501 | if (error) |
2502 | return ERR_PTR(error); | |
f2eb6575 | 2503 | return res; |
1da177e4 LT |
2504 | } |
2505 | ||
ffd1f4ed MS |
2506 | /* |
2507 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2508 | */ | |
2509 | static int path_with_deleted(const struct path *path, struct path *root, | |
2510 | char **buf, int *buflen) | |
2511 | { | |
2512 | prepend(buf, buflen, "\0", 1); | |
2513 | if (d_unlinked(path->dentry)) { | |
2514 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2515 | if (error) | |
2516 | return error; | |
2517 | } | |
2518 | ||
2519 | return prepend_path(path, root, buf, buflen); | |
2520 | } | |
2521 | ||
8df9d1a4 MS |
2522 | static int prepend_unreachable(char **buffer, int *buflen) |
2523 | { | |
2524 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2525 | } | |
2526 | ||
a03a8a70 JB |
2527 | /** |
2528 | * d_path - return the path of a dentry | |
cf28b486 | 2529 | * @path: path to report |
a03a8a70 JB |
2530 | * @buf: buffer to return value in |
2531 | * @buflen: buffer length | |
2532 | * | |
2533 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2534 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2535 | * | |
52afeefb AV |
2536 | * Returns a pointer into the buffer or an error code if the path was |
2537 | * too long. Note: Callers should use the returned pointer, not the passed | |
2538 | * in buffer, to use the name! The implementation often starts at an offset | |
2539 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2540 | * |
31f3e0b3 | 2541 | * "buflen" should be positive. |
a03a8a70 | 2542 | */ |
20d4fdc1 | 2543 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2544 | { |
ffd1f4ed | 2545 | char *res = buf + buflen; |
6ac08c39 | 2546 | struct path root; |
9d1bc601 | 2547 | struct path tmp; |
ffd1f4ed | 2548 | int error; |
1da177e4 | 2549 | |
c23fbb6b ED |
2550 | /* |
2551 | * We have various synthetic filesystems that never get mounted. On | |
2552 | * these filesystems dentries are never used for lookup purposes, and | |
2553 | * thus don't need to be hashed. They also don't need a name until a | |
2554 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2555 | * below allows us to generate a name for these objects on demand: | |
2556 | */ | |
cf28b486 JB |
2557 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2558 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2559 | |
f7ad3c6b | 2560 | get_fs_root(current->fs, &root); |
949854d0 | 2561 | write_seqlock(&rename_lock); |
9d1bc601 | 2562 | tmp = root; |
ffd1f4ed MS |
2563 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2564 | if (error) | |
2565 | res = ERR_PTR(error); | |
949854d0 | 2566 | write_sequnlock(&rename_lock); |
6ac08c39 | 2567 | path_put(&root); |
1da177e4 LT |
2568 | return res; |
2569 | } | |
ec4f8605 | 2570 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2571 | |
8df9d1a4 MS |
2572 | /** |
2573 | * d_path_with_unreachable - return the path of a dentry | |
2574 | * @path: path to report | |
2575 | * @buf: buffer to return value in | |
2576 | * @buflen: buffer length | |
2577 | * | |
2578 | * The difference from d_path() is that this prepends "(unreachable)" | |
2579 | * to paths which are unreachable from the current process' root. | |
2580 | */ | |
2581 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2582 | { | |
2583 | char *res = buf + buflen; | |
2584 | struct path root; | |
2585 | struct path tmp; | |
2586 | int error; | |
2587 | ||
2588 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2589 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2590 | ||
2591 | get_fs_root(current->fs, &root); | |
949854d0 | 2592 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2593 | tmp = root; |
2594 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2595 | if (!error && !path_equal(&tmp, &root)) | |
2596 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2597 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2598 | path_put(&root); |
2599 | if (error) | |
2600 | res = ERR_PTR(error); | |
2601 | ||
2602 | return res; | |
2603 | } | |
2604 | ||
c23fbb6b ED |
2605 | /* |
2606 | * Helper function for dentry_operations.d_dname() members | |
2607 | */ | |
2608 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2609 | const char *fmt, ...) | |
2610 | { | |
2611 | va_list args; | |
2612 | char temp[64]; | |
2613 | int sz; | |
2614 | ||
2615 | va_start(args, fmt); | |
2616 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2617 | va_end(args); | |
2618 | ||
2619 | if (sz > sizeof(temp) || sz > buflen) | |
2620 | return ERR_PTR(-ENAMETOOLONG); | |
2621 | ||
2622 | buffer += buflen - sz; | |
2623 | return memcpy(buffer, temp, sz); | |
2624 | } | |
2625 | ||
6092d048 RP |
2626 | /* |
2627 | * Write full pathname from the root of the filesystem into the buffer. | |
2628 | */ | |
ec2447c2 | 2629 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2630 | { |
2631 | char *end = buf + buflen; | |
2632 | char *retval; | |
2633 | ||
6092d048 | 2634 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2635 | if (buflen < 1) |
2636 | goto Elong; | |
2637 | /* Get '/' right */ | |
2638 | retval = end-1; | |
2639 | *retval = '/'; | |
2640 | ||
cdd16d02 MS |
2641 | while (!IS_ROOT(dentry)) { |
2642 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2643 | int error; |
6092d048 | 2644 | |
6092d048 | 2645 | prefetch(parent); |
9abca360 NP |
2646 | spin_lock(&dentry->d_lock); |
2647 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2648 | spin_unlock(&dentry->d_lock); | |
2649 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2650 | goto Elong; |
2651 | ||
2652 | retval = end; | |
2653 | dentry = parent; | |
2654 | } | |
c103135c AV |
2655 | return retval; |
2656 | Elong: | |
2657 | return ERR_PTR(-ENAMETOOLONG); | |
2658 | } | |
ec2447c2 NP |
2659 | |
2660 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2661 | { | |
2662 | char *retval; | |
2663 | ||
949854d0 | 2664 | write_seqlock(&rename_lock); |
ec2447c2 | 2665 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2666 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2667 | |
2668 | return retval; | |
2669 | } | |
2670 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2671 | |
2672 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2673 | { | |
2674 | char *p = NULL; | |
2675 | char *retval; | |
2676 | ||
949854d0 | 2677 | write_seqlock(&rename_lock); |
c103135c AV |
2678 | if (d_unlinked(dentry)) { |
2679 | p = buf + buflen; | |
2680 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2681 | goto Elong; | |
2682 | buflen++; | |
2683 | } | |
2684 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2685 | write_sequnlock(&rename_lock); |
c103135c AV |
2686 | if (!IS_ERR(retval) && p) |
2687 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2688 | return retval; |
2689 | Elong: | |
6092d048 RP |
2690 | return ERR_PTR(-ENAMETOOLONG); |
2691 | } | |
2692 | ||
1da177e4 LT |
2693 | /* |
2694 | * NOTE! The user-level library version returns a | |
2695 | * character pointer. The kernel system call just | |
2696 | * returns the length of the buffer filled (which | |
2697 | * includes the ending '\0' character), or a negative | |
2698 | * error value. So libc would do something like | |
2699 | * | |
2700 | * char *getcwd(char * buf, size_t size) | |
2701 | * { | |
2702 | * int retval; | |
2703 | * | |
2704 | * retval = sys_getcwd(buf, size); | |
2705 | * if (retval >= 0) | |
2706 | * return buf; | |
2707 | * errno = -retval; | |
2708 | * return NULL; | |
2709 | * } | |
2710 | */ | |
3cdad428 | 2711 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2712 | { |
552ce544 | 2713 | int error; |
6ac08c39 | 2714 | struct path pwd, root; |
552ce544 | 2715 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2716 | |
2717 | if (!page) | |
2718 | return -ENOMEM; | |
2719 | ||
f7ad3c6b | 2720 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2721 | |
552ce544 | 2722 | error = -ENOENT; |
949854d0 | 2723 | write_seqlock(&rename_lock); |
f3da392e | 2724 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2725 | unsigned long len; |
9d1bc601 | 2726 | struct path tmp = root; |
8df9d1a4 MS |
2727 | char *cwd = page + PAGE_SIZE; |
2728 | int buflen = PAGE_SIZE; | |
1da177e4 | 2729 | |
8df9d1a4 MS |
2730 | prepend(&cwd, &buflen, "\0", 1); |
2731 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2732 | write_sequnlock(&rename_lock); |
552ce544 | 2733 | |
8df9d1a4 | 2734 | if (error) |
552ce544 LT |
2735 | goto out; |
2736 | ||
8df9d1a4 MS |
2737 | /* Unreachable from current root */ |
2738 | if (!path_equal(&tmp, &root)) { | |
2739 | error = prepend_unreachable(&cwd, &buflen); | |
2740 | if (error) | |
2741 | goto out; | |
2742 | } | |
2743 | ||
552ce544 LT |
2744 | error = -ERANGE; |
2745 | len = PAGE_SIZE + page - cwd; | |
2746 | if (len <= size) { | |
2747 | error = len; | |
2748 | if (copy_to_user(buf, cwd, len)) | |
2749 | error = -EFAULT; | |
2750 | } | |
949854d0 NP |
2751 | } else { |
2752 | write_sequnlock(&rename_lock); | |
949854d0 | 2753 | } |
1da177e4 LT |
2754 | |
2755 | out: | |
6ac08c39 JB |
2756 | path_put(&pwd); |
2757 | path_put(&root); | |
1da177e4 LT |
2758 | free_page((unsigned long) page); |
2759 | return error; | |
2760 | } | |
2761 | ||
2762 | /* | |
2763 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2764 | * | |
2765 | * Trivially implemented using the dcache structure | |
2766 | */ | |
2767 | ||
2768 | /** | |
2769 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2770 | * @new_dentry: new dentry | |
2771 | * @old_dentry: old dentry | |
2772 | * | |
2773 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2774 | * Returns 0 otherwise. | |
2775 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2776 | */ | |
2777 | ||
e2761a11 | 2778 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2779 | { |
2780 | int result; | |
949854d0 | 2781 | unsigned seq; |
1da177e4 | 2782 | |
e2761a11 OH |
2783 | if (new_dentry == old_dentry) |
2784 | return 1; | |
2785 | ||
e2761a11 | 2786 | do { |
1da177e4 | 2787 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2788 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2789 | /* |
2790 | * Need rcu_readlock to protect against the d_parent trashing | |
2791 | * due to d_move | |
2792 | */ | |
2793 | rcu_read_lock(); | |
e2761a11 | 2794 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2795 | result = 1; |
e2761a11 OH |
2796 | else |
2797 | result = 0; | |
949854d0 | 2798 | rcu_read_unlock(); |
1da177e4 | 2799 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2800 | |
2801 | return result; | |
2802 | } | |
2803 | ||
2096f759 AV |
2804 | int path_is_under(struct path *path1, struct path *path2) |
2805 | { | |
2806 | struct vfsmount *mnt = path1->mnt; | |
2807 | struct dentry *dentry = path1->dentry; | |
2808 | int res; | |
99b7db7b NP |
2809 | |
2810 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2811 | if (mnt != path2->mnt) { |
2812 | for (;;) { | |
2813 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2814 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2815 | return 0; |
2816 | } | |
2817 | if (mnt->mnt_parent == path2->mnt) | |
2818 | break; | |
2819 | mnt = mnt->mnt_parent; | |
2820 | } | |
2821 | dentry = mnt->mnt_mountpoint; | |
2822 | } | |
2823 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2824 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2825 | return res; |
2826 | } | |
2827 | EXPORT_SYMBOL(path_is_under); | |
2828 | ||
1da177e4 LT |
2829 | void d_genocide(struct dentry *root) |
2830 | { | |
949854d0 | 2831 | struct dentry *this_parent; |
1da177e4 | 2832 | struct list_head *next; |
949854d0 | 2833 | unsigned seq; |
58db63d0 | 2834 | int locked = 0; |
1da177e4 | 2835 | |
949854d0 | 2836 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2837 | again: |
2838 | this_parent = root; | |
2fd6b7f5 | 2839 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2840 | repeat: |
2841 | next = this_parent->d_subdirs.next; | |
2842 | resume: | |
2843 | while (next != &this_parent->d_subdirs) { | |
2844 | struct list_head *tmp = next; | |
5160ee6f | 2845 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2846 | next = tmp->next; |
949854d0 | 2847 | |
da502956 NP |
2848 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2849 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2850 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2851 | continue; |
da502956 | 2852 | } |
1da177e4 | 2853 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2854 | spin_unlock(&this_parent->d_lock); |
2855 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2856 | this_parent = dentry; |
2fd6b7f5 | 2857 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2858 | goto repeat; |
2859 | } | |
949854d0 NP |
2860 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2861 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2862 | dentry->d_count--; | |
2863 | } | |
b7ab39f6 | 2864 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2865 | } |
2866 | if (this_parent != root) { | |
949854d0 NP |
2867 | struct dentry *tmp; |
2868 | struct dentry *child; | |
2869 | ||
2870 | tmp = this_parent->d_parent; | |
2871 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { | |
2872 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2873 | this_parent->d_count--; | |
2874 | } | |
2875 | rcu_read_lock(); | |
b7ab39f6 | 2876 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
2877 | child = this_parent; |
2878 | this_parent = tmp; | |
2fd6b7f5 | 2879 | spin_lock(&this_parent->d_lock); |
949854d0 NP |
2880 | /* might go back up the wrong parent if we have had a rename |
2881 | * or deletion */ | |
2882 | if (this_parent != child->d_parent || | |
58db63d0 | 2883 | (!locked && read_seqretry(&rename_lock, seq))) { |
949854d0 | 2884 | spin_unlock(&this_parent->d_lock); |
949854d0 NP |
2885 | rcu_read_unlock(); |
2886 | goto rename_retry; | |
2887 | } | |
2888 | rcu_read_unlock(); | |
2889 | next = child->d_u.d_child.next; | |
1da177e4 LT |
2890 | goto resume; |
2891 | } | |
2fd6b7f5 | 2892 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2893 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2894 | goto rename_retry; |
58db63d0 NP |
2895 | if (locked) |
2896 | write_sequnlock(&rename_lock); | |
2897 | return; | |
2898 | ||
2899 | rename_retry: | |
2900 | locked = 1; | |
2901 | write_seqlock(&rename_lock); | |
2902 | goto again; | |
1da177e4 LT |
2903 | } |
2904 | ||
2905 | /** | |
2906 | * find_inode_number - check for dentry with name | |
2907 | * @dir: directory to check | |
2908 | * @name: Name to find. | |
2909 | * | |
2910 | * Check whether a dentry already exists for the given name, | |
2911 | * and return the inode number if it has an inode. Otherwise | |
2912 | * 0 is returned. | |
2913 | * | |
2914 | * This routine is used to post-process directory listings for | |
2915 | * filesystems using synthetic inode numbers, and is necessary | |
2916 | * to keep getcwd() working. | |
2917 | */ | |
2918 | ||
2919 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2920 | { | |
2921 | struct dentry * dentry; | |
2922 | ino_t ino = 0; | |
2923 | ||
3e7e241f EB |
2924 | dentry = d_hash_and_lookup(dir, name); |
2925 | if (dentry) { | |
1da177e4 LT |
2926 | if (dentry->d_inode) |
2927 | ino = dentry->d_inode->i_ino; | |
2928 | dput(dentry); | |
2929 | } | |
1da177e4 LT |
2930 | return ino; |
2931 | } | |
ec4f8605 | 2932 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2933 | |
2934 | static __initdata unsigned long dhash_entries; | |
2935 | static int __init set_dhash_entries(char *str) | |
2936 | { | |
2937 | if (!str) | |
2938 | return 0; | |
2939 | dhash_entries = simple_strtoul(str, &str, 0); | |
2940 | return 1; | |
2941 | } | |
2942 | __setup("dhash_entries=", set_dhash_entries); | |
2943 | ||
2944 | static void __init dcache_init_early(void) | |
2945 | { | |
2946 | int loop; | |
2947 | ||
2948 | /* If hashes are distributed across NUMA nodes, defer | |
2949 | * hash allocation until vmalloc space is available. | |
2950 | */ | |
2951 | if (hashdist) | |
2952 | return; | |
2953 | ||
2954 | dentry_hashtable = | |
2955 | alloc_large_system_hash("Dentry cache", | |
2956 | sizeof(struct hlist_head), | |
2957 | dhash_entries, | |
2958 | 13, | |
2959 | HASH_EARLY, | |
2960 | &d_hash_shift, | |
2961 | &d_hash_mask, | |
2962 | 0); | |
2963 | ||
2964 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2965 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2966 | } | |
2967 | ||
74bf17cf | 2968 | static void __init dcache_init(void) |
1da177e4 LT |
2969 | { |
2970 | int loop; | |
2971 | ||
2972 | /* | |
2973 | * A constructor could be added for stable state like the lists, | |
2974 | * but it is probably not worth it because of the cache nature | |
2975 | * of the dcache. | |
2976 | */ | |
0a31bd5f CL |
2977 | dentry_cache = KMEM_CACHE(dentry, |
2978 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 2979 | |
8e1f936b | 2980 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
2981 | |
2982 | /* Hash may have been set up in dcache_init_early */ | |
2983 | if (!hashdist) | |
2984 | return; | |
2985 | ||
2986 | dentry_hashtable = | |
2987 | alloc_large_system_hash("Dentry cache", | |
2988 | sizeof(struct hlist_head), | |
2989 | dhash_entries, | |
2990 | 13, | |
2991 | 0, | |
2992 | &d_hash_shift, | |
2993 | &d_hash_mask, | |
2994 | 0); | |
2995 | ||
2996 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
2997 | INIT_HLIST_HEAD(&dentry_hashtable[loop]); | |
2998 | } | |
2999 | ||
3000 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3001 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3002 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3003 | |
1da177e4 LT |
3004 | EXPORT_SYMBOL(d_genocide); |
3005 | ||
1da177e4 LT |
3006 | void __init vfs_caches_init_early(void) |
3007 | { | |
3008 | dcache_init_early(); | |
3009 | inode_init_early(); | |
3010 | } | |
3011 | ||
3012 | void __init vfs_caches_init(unsigned long mempages) | |
3013 | { | |
3014 | unsigned long reserve; | |
3015 | ||
3016 | /* Base hash sizes on available memory, with a reserve equal to | |
3017 | 150% of current kernel size */ | |
3018 | ||
3019 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3020 | mempages -= reserve; | |
3021 | ||
3022 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3023 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3024 | |
74bf17cf DC |
3025 | dcache_init(); |
3026 | inode_init(); | |
1da177e4 | 3027 | files_init(mempages); |
74bf17cf | 3028 | mnt_init(); |
1da177e4 LT |
3029 | bdev_cache_init(); |
3030 | chrdev_init(); | |
3031 | } |