]>
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> |
ceb5bdc2 NP |
36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | |
07f3f05c | 38 | #include "internal.h" |
1da177e4 | 39 | |
789680d1 NP |
40 | /* |
41 | * Usage: | |
873feea0 NP |
42 | * dcache->d_inode->i_lock protects: |
43 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
44 | * dcache_hash_bucket lock protects: |
45 | * - the dcache hash table | |
46 | * s_anon bl list spinlock protects: | |
47 | * - the s_anon list (see __d_drop) | |
23044507 NP |
48 | * dcache_lru_lock protects: |
49 | * - the dcache lru lists and counters | |
50 | * d_lock protects: | |
51 | * - d_flags | |
52 | * - d_name | |
53 | * - d_lru | |
b7ab39f6 | 54 | * - d_count |
da502956 | 55 | * - d_unhashed() |
2fd6b7f5 NP |
56 | * - d_parent and d_subdirs |
57 | * - childrens' d_child and d_parent | |
b23fb0a6 | 58 | * - d_alias, d_inode |
789680d1 NP |
59 | * |
60 | * Ordering: | |
873feea0 | 61 | * dentry->d_inode->i_lock |
b5c84bf6 NP |
62 | * dentry->d_lock |
63 | * dcache_lru_lock | |
ceb5bdc2 NP |
64 | * dcache_hash_bucket lock |
65 | * s_anon lock | |
789680d1 | 66 | * |
da502956 NP |
67 | * If there is an ancestor relationship: |
68 | * dentry->d_parent->...->d_parent->d_lock | |
69 | * ... | |
70 | * dentry->d_parent->d_lock | |
71 | * dentry->d_lock | |
72 | * | |
73 | * If no ancestor relationship: | |
789680d1 NP |
74 | * if (dentry1 < dentry2) |
75 | * dentry1->d_lock | |
76 | * dentry2->d_lock | |
77 | */ | |
fa3536cc | 78 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
79 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
80 | ||
23044507 | 81 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
74c3cbe3 | 82 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 83 | |
949854d0 | 84 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 85 | |
e18b890b | 86 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 87 | |
1da177e4 LT |
88 | /* |
89 | * This is the single most critical data structure when it comes | |
90 | * to the dcache: the hashtable for lookups. Somebody should try | |
91 | * to make this good - I've just made it work. | |
92 | * | |
93 | * This hash-function tries to avoid losing too many bits of hash | |
94 | * information, yet avoid using a prime hash-size or similar. | |
95 | */ | |
96 | #define D_HASHBITS d_hash_shift | |
97 | #define D_HASHMASK d_hash_mask | |
98 | ||
fa3536cc ED |
99 | static unsigned int d_hash_mask __read_mostly; |
100 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 NP |
101 | |
102 | struct dcache_hash_bucket { | |
103 | struct hlist_bl_head head; | |
104 | }; | |
105 | static struct dcache_hash_bucket *dentry_hashtable __read_mostly; | |
106 | ||
107 | static inline struct dcache_hash_bucket *d_hash(struct dentry *parent, | |
108 | unsigned long hash) | |
109 | { | |
110 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
111 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
112 | return dentry_hashtable + (hash & D_HASHMASK); | |
113 | } | |
114 | ||
115 | static inline void spin_lock_bucket(struct dcache_hash_bucket *b) | |
116 | { | |
117 | bit_spin_lock(0, (unsigned long *)&b->head.first); | |
118 | } | |
119 | ||
120 | static inline void spin_unlock_bucket(struct dcache_hash_bucket *b) | |
121 | { | |
122 | __bit_spin_unlock(0, (unsigned long *)&b->head.first); | |
123 | } | |
1da177e4 LT |
124 | |
125 | /* Statistics gathering. */ | |
126 | struct dentry_stat_t dentry_stat = { | |
127 | .age_limit = 45, | |
128 | }; | |
129 | ||
3e880fb5 | 130 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
131 | |
132 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
133 | static int get_nr_dentry(void) |
134 | { | |
135 | int i; | |
136 | int sum = 0; | |
137 | for_each_possible_cpu(i) | |
138 | sum += per_cpu(nr_dentry, i); | |
139 | return sum < 0 ? 0 : sum; | |
140 | } | |
141 | ||
312d3ca8 CH |
142 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
143 | size_t *lenp, loff_t *ppos) | |
144 | { | |
3e880fb5 | 145 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
146 | return proc_dointvec(table, write, buffer, lenp, ppos); |
147 | } | |
148 | #endif | |
149 | ||
9c82ab9c | 150 | static void __d_free(struct rcu_head *head) |
1da177e4 | 151 | { |
9c82ab9c CH |
152 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
153 | ||
fd217f4d | 154 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
155 | if (dname_external(dentry)) |
156 | kfree(dentry->d_name.name); | |
157 | kmem_cache_free(dentry_cache, dentry); | |
158 | } | |
159 | ||
160 | /* | |
b5c84bf6 | 161 | * no locks, please. |
1da177e4 LT |
162 | */ |
163 | static void d_free(struct dentry *dentry) | |
164 | { | |
b7ab39f6 | 165 | BUG_ON(dentry->d_count); |
3e880fb5 | 166 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
167 | if (dentry->d_op && dentry->d_op->d_release) |
168 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 169 | |
b3423415 | 170 | /* if dentry was never inserted into hash, immediate free is OK */ |
ceb5bdc2 | 171 | if (hlist_bl_unhashed(&dentry->d_hash)) |
9c82ab9c | 172 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 173 | else |
9c82ab9c | 174 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
175 | } |
176 | ||
31e6b01f NP |
177 | /** |
178 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 179 | * @dentry: the target dentry |
31e6b01f NP |
180 | * After this call, in-progress rcu-walk path lookup will fail. This |
181 | * should be called after unhashing, and after changing d_inode (if | |
182 | * the dentry has not already been unhashed). | |
183 | */ | |
184 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
185 | { | |
186 | assert_spin_locked(&dentry->d_lock); | |
187 | /* Go through a barrier */ | |
188 | write_seqcount_barrier(&dentry->d_seq); | |
189 | } | |
190 | ||
1da177e4 LT |
191 | /* |
192 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
193 | * d_iput() operation if defined. Dentry has no refcount |
194 | * and is unhashed. | |
1da177e4 | 195 | */ |
858119e1 | 196 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 197 | __releases(dentry->d_lock) |
873feea0 | 198 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
199 | { |
200 | struct inode *inode = dentry->d_inode; | |
201 | if (inode) { | |
202 | dentry->d_inode = NULL; | |
203 | list_del_init(&dentry->d_alias); | |
204 | spin_unlock(&dentry->d_lock); | |
873feea0 | 205 | spin_unlock(&inode->i_lock); |
f805fbda LT |
206 | if (!inode->i_nlink) |
207 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
208 | if (dentry->d_op && dentry->d_op->d_iput) |
209 | dentry->d_op->d_iput(dentry, inode); | |
210 | else | |
211 | iput(inode); | |
212 | } else { | |
213 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
214 | } |
215 | } | |
216 | ||
31e6b01f NP |
217 | /* |
218 | * Release the dentry's inode, using the filesystem | |
219 | * d_iput() operation if defined. dentry remains in-use. | |
220 | */ | |
221 | static void dentry_unlink_inode(struct dentry * dentry) | |
222 | __releases(dentry->d_lock) | |
873feea0 | 223 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
224 | { |
225 | struct inode *inode = dentry->d_inode; | |
226 | dentry->d_inode = NULL; | |
227 | list_del_init(&dentry->d_alias); | |
228 | dentry_rcuwalk_barrier(dentry); | |
229 | spin_unlock(&dentry->d_lock); | |
873feea0 | 230 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
231 | if (!inode->i_nlink) |
232 | fsnotify_inoderemove(inode); | |
233 | if (dentry->d_op && dentry->d_op->d_iput) | |
234 | dentry->d_op->d_iput(dentry, inode); | |
235 | else | |
236 | iput(inode); | |
237 | } | |
238 | ||
da3bbdd4 | 239 | /* |
23044507 | 240 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
241 | */ |
242 | static void dentry_lru_add(struct dentry *dentry) | |
243 | { | |
a4633357 | 244 | if (list_empty(&dentry->d_lru)) { |
23044507 | 245 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
246 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
247 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 248 | dentry_stat.nr_unused++; |
23044507 | 249 | spin_unlock(&dcache_lru_lock); |
a4633357 | 250 | } |
da3bbdd4 KM |
251 | } |
252 | ||
23044507 NP |
253 | static void __dentry_lru_del(struct dentry *dentry) |
254 | { | |
255 | list_del_init(&dentry->d_lru); | |
256 | dentry->d_sb->s_nr_dentry_unused--; | |
257 | dentry_stat.nr_unused--; | |
258 | } | |
259 | ||
da3bbdd4 KM |
260 | static void dentry_lru_del(struct dentry *dentry) |
261 | { | |
262 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
263 | spin_lock(&dcache_lru_lock); |
264 | __dentry_lru_del(dentry); | |
265 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
266 | } |
267 | } | |
268 | ||
a4633357 | 269 | static void dentry_lru_move_tail(struct dentry *dentry) |
da3bbdd4 | 270 | { |
23044507 | 271 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
272 | if (list_empty(&dentry->d_lru)) { |
273 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
274 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 275 | dentry_stat.nr_unused++; |
a4633357 CH |
276 | } else { |
277 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
da3bbdd4 | 278 | } |
23044507 | 279 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
280 | } |
281 | ||
d52b9086 MS |
282 | /** |
283 | * d_kill - kill dentry and return parent | |
284 | * @dentry: dentry to kill | |
ff5fdb61 | 285 | * @parent: parent dentry |
d52b9086 | 286 | * |
31f3e0b3 | 287 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
288 | * |
289 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 290 | * |
b5c84bf6 NP |
291 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
292 | * d_kill. | |
d52b9086 | 293 | */ |
2fd6b7f5 | 294 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 295 | __releases(dentry->d_lock) |
2fd6b7f5 | 296 | __releases(parent->d_lock) |
873feea0 | 297 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 298 | { |
d52b9086 | 299 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
300 | /* |
301 | * Inform try_to_ascend() that we are no longer attached to the | |
302 | * dentry tree | |
303 | */ | |
304 | dentry->d_flags |= DCACHE_DISCONNECTED; | |
2fd6b7f5 NP |
305 | if (parent) |
306 | spin_unlock(&parent->d_lock); | |
d52b9086 | 307 | dentry_iput(dentry); |
b7ab39f6 NP |
308 | /* |
309 | * dentry_iput drops the locks, at which point nobody (except | |
310 | * transient RCU lookups) can reach this dentry. | |
311 | */ | |
d52b9086 | 312 | d_free(dentry); |
871c0067 | 313 | return parent; |
d52b9086 MS |
314 | } |
315 | ||
789680d1 NP |
316 | /** |
317 | * d_drop - drop a dentry | |
318 | * @dentry: dentry to drop | |
319 | * | |
320 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
321 | * be found through a VFS lookup any more. Note that this is different from | |
322 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
323 | * possible, giving a successful _negative_ lookup, while d_drop will | |
324 | * just make the cache lookup fail. | |
325 | * | |
326 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
327 | * reason (NFS timeouts or autofs deletes). | |
328 | * | |
329 | * __d_drop requires dentry->d_lock. | |
330 | */ | |
331 | void __d_drop(struct dentry *dentry) | |
332 | { | |
333 | if (!(dentry->d_flags & DCACHE_UNHASHED)) { | |
ceb5bdc2 NP |
334 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) { |
335 | bit_spin_lock(0, | |
336 | (unsigned long *)&dentry->d_sb->s_anon.first); | |
337 | dentry->d_flags |= DCACHE_UNHASHED; | |
338 | hlist_bl_del_init(&dentry->d_hash); | |
339 | __bit_spin_unlock(0, | |
340 | (unsigned long *)&dentry->d_sb->s_anon.first); | |
341 | } else { | |
342 | struct dcache_hash_bucket *b; | |
343 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
344 | spin_lock_bucket(b); | |
345 | /* | |
346 | * We may not actually need to put DCACHE_UNHASHED | |
347 | * manipulations under the hash lock, but follow | |
348 | * the principle of least surprise. | |
349 | */ | |
350 | dentry->d_flags |= DCACHE_UNHASHED; | |
351 | hlist_bl_del_rcu(&dentry->d_hash); | |
352 | spin_unlock_bucket(b); | |
353 | dentry_rcuwalk_barrier(dentry); | |
354 | } | |
789680d1 NP |
355 | } |
356 | } | |
357 | EXPORT_SYMBOL(__d_drop); | |
358 | ||
359 | void d_drop(struct dentry *dentry) | |
360 | { | |
789680d1 NP |
361 | spin_lock(&dentry->d_lock); |
362 | __d_drop(dentry); | |
363 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
364 | } |
365 | EXPORT_SYMBOL(d_drop); | |
366 | ||
77812a1e NP |
367 | /* |
368 | * Finish off a dentry we've decided to kill. | |
369 | * dentry->d_lock must be held, returns with it unlocked. | |
370 | * If ref is non-zero, then decrement the refcount too. | |
371 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
372 | */ | |
373 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
374 | __releases(dentry->d_lock) | |
375 | { | |
873feea0 | 376 | struct inode *inode; |
77812a1e NP |
377 | struct dentry *parent; |
378 | ||
873feea0 NP |
379 | inode = dentry->d_inode; |
380 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
381 | relock: |
382 | spin_unlock(&dentry->d_lock); | |
383 | cpu_relax(); | |
384 | return dentry; /* try again with same dentry */ | |
385 | } | |
386 | if (IS_ROOT(dentry)) | |
387 | parent = NULL; | |
388 | else | |
389 | parent = dentry->d_parent; | |
390 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
391 | if (inode) |
392 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
393 | goto relock; |
394 | } | |
31e6b01f | 395 | |
77812a1e NP |
396 | if (ref) |
397 | dentry->d_count--; | |
398 | /* if dentry was on the d_lru list delete it from there */ | |
399 | dentry_lru_del(dentry); | |
400 | /* if it was on the hash then remove it */ | |
401 | __d_drop(dentry); | |
402 | return d_kill(dentry, parent); | |
403 | } | |
404 | ||
1da177e4 LT |
405 | /* |
406 | * This is dput | |
407 | * | |
408 | * This is complicated by the fact that we do not want to put | |
409 | * dentries that are no longer on any hash chain on the unused | |
410 | * list: we'd much rather just get rid of them immediately. | |
411 | * | |
412 | * However, that implies that we have to traverse the dentry | |
413 | * tree upwards to the parents which might _also_ now be | |
414 | * scheduled for deletion (it may have been only waiting for | |
415 | * its last child to go away). | |
416 | * | |
417 | * This tail recursion is done by hand as we don't want to depend | |
418 | * on the compiler to always get this right (gcc generally doesn't). | |
419 | * Real recursion would eat up our stack space. | |
420 | */ | |
421 | ||
422 | /* | |
423 | * dput - release a dentry | |
424 | * @dentry: dentry to release | |
425 | * | |
426 | * Release a dentry. This will drop the usage count and if appropriate | |
427 | * call the dentry unlink method as well as removing it from the queues and | |
428 | * releasing its resources. If the parent dentries were scheduled for release | |
429 | * they too may now get deleted. | |
1da177e4 | 430 | */ |
1da177e4 LT |
431 | void dput(struct dentry *dentry) |
432 | { | |
433 | if (!dentry) | |
434 | return; | |
435 | ||
436 | repeat: | |
b7ab39f6 | 437 | if (dentry->d_count == 1) |
1da177e4 | 438 | might_sleep(); |
1da177e4 | 439 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
440 | BUG_ON(!dentry->d_count); |
441 | if (dentry->d_count > 1) { | |
442 | dentry->d_count--; | |
1da177e4 | 443 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
444 | return; |
445 | } | |
446 | ||
fb045adb | 447 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 448 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 449 | goto kill_it; |
1da177e4 | 450 | } |
265ac902 | 451 | |
1da177e4 LT |
452 | /* Unreachable? Get rid of it */ |
453 | if (d_unhashed(dentry)) | |
454 | goto kill_it; | |
265ac902 NP |
455 | |
456 | /* Otherwise leave it cached and ensure it's on the LRU */ | |
457 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 458 | dentry_lru_add(dentry); |
265ac902 | 459 | |
61f3dee4 NP |
460 | dentry->d_count--; |
461 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
462 | return; |
463 | ||
d52b9086 | 464 | kill_it: |
77812a1e | 465 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
466 | if (dentry) |
467 | goto repeat; | |
1da177e4 | 468 | } |
ec4f8605 | 469 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
470 | |
471 | /** | |
472 | * d_invalidate - invalidate a dentry | |
473 | * @dentry: dentry to invalidate | |
474 | * | |
475 | * Try to invalidate the dentry if it turns out to be | |
476 | * possible. If there are other dentries that can be | |
477 | * reached through this one we can't delete it and we | |
478 | * return -EBUSY. On success we return 0. | |
479 | * | |
480 | * no dcache lock. | |
481 | */ | |
482 | ||
483 | int d_invalidate(struct dentry * dentry) | |
484 | { | |
485 | /* | |
486 | * If it's already been dropped, return OK. | |
487 | */ | |
da502956 | 488 | spin_lock(&dentry->d_lock); |
1da177e4 | 489 | if (d_unhashed(dentry)) { |
da502956 | 490 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
491 | return 0; |
492 | } | |
493 | /* | |
494 | * Check whether to do a partial shrink_dcache | |
495 | * to get rid of unused child entries. | |
496 | */ | |
497 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 498 | spin_unlock(&dentry->d_lock); |
1da177e4 | 499 | shrink_dcache_parent(dentry); |
da502956 | 500 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
501 | } |
502 | ||
503 | /* | |
504 | * Somebody else still using it? | |
505 | * | |
506 | * If it's a directory, we can't drop it | |
507 | * for fear of somebody re-populating it | |
508 | * with children (even though dropping it | |
509 | * would make it unreachable from the root, | |
510 | * we might still populate it if it was a | |
511 | * working directory or similar). | |
512 | */ | |
b7ab39f6 | 513 | if (dentry->d_count > 1) { |
1da177e4 LT |
514 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
515 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
516 | return -EBUSY; |
517 | } | |
518 | } | |
519 | ||
520 | __d_drop(dentry); | |
521 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
522 | return 0; |
523 | } | |
ec4f8605 | 524 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 525 | |
b5c84bf6 | 526 | /* This must be called with d_lock held */ |
dc0474be | 527 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 528 | { |
b7ab39f6 | 529 | dentry->d_count++; |
23044507 NP |
530 | } |
531 | ||
dc0474be | 532 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 533 | { |
23044507 | 534 | spin_lock(&dentry->d_lock); |
dc0474be | 535 | __dget_dlock(dentry); |
23044507 | 536 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
537 | } |
538 | ||
b7ab39f6 NP |
539 | struct dentry *dget_parent(struct dentry *dentry) |
540 | { | |
541 | struct dentry *ret; | |
542 | ||
543 | repeat: | |
a734eb45 NP |
544 | /* |
545 | * Don't need rcu_dereference because we re-check it was correct under | |
546 | * the lock. | |
547 | */ | |
548 | rcu_read_lock(); | |
b7ab39f6 | 549 | ret = dentry->d_parent; |
a734eb45 NP |
550 | if (!ret) { |
551 | rcu_read_unlock(); | |
b7ab39f6 NP |
552 | goto out; |
553 | } | |
a734eb45 NP |
554 | spin_lock(&ret->d_lock); |
555 | if (unlikely(ret != dentry->d_parent)) { | |
556 | spin_unlock(&ret->d_lock); | |
557 | rcu_read_unlock(); | |
b7ab39f6 NP |
558 | goto repeat; |
559 | } | |
a734eb45 | 560 | rcu_read_unlock(); |
b7ab39f6 NP |
561 | BUG_ON(!ret->d_count); |
562 | ret->d_count++; | |
563 | spin_unlock(&ret->d_lock); | |
564 | out: | |
b7ab39f6 NP |
565 | return ret; |
566 | } | |
567 | EXPORT_SYMBOL(dget_parent); | |
568 | ||
1da177e4 LT |
569 | /** |
570 | * d_find_alias - grab a hashed alias of inode | |
571 | * @inode: inode in question | |
572 | * @want_discon: flag, used by d_splice_alias, to request | |
573 | * that only a DISCONNECTED alias be returned. | |
574 | * | |
575 | * If inode has a hashed alias, or is a directory and has any alias, | |
576 | * acquire the reference to alias and return it. Otherwise return NULL. | |
577 | * Notice that if inode is a directory there can be only one alias and | |
578 | * it can be unhashed only if it has no children, or if it is the root | |
579 | * of a filesystem. | |
580 | * | |
21c0d8fd | 581 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 582 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 583 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 584 | */ |
da502956 | 585 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 586 | { |
da502956 | 587 | struct dentry *alias, *discon_alias; |
1da177e4 | 588 | |
da502956 NP |
589 | again: |
590 | discon_alias = NULL; | |
591 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
592 | spin_lock(&alias->d_lock); | |
1da177e4 | 593 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 594 | if (IS_ROOT(alias) && |
da502956 | 595 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 596 | discon_alias = alias; |
da502956 | 597 | } else if (!want_discon) { |
dc0474be | 598 | __dget_dlock(alias); |
da502956 NP |
599 | spin_unlock(&alias->d_lock); |
600 | return alias; | |
601 | } | |
602 | } | |
603 | spin_unlock(&alias->d_lock); | |
604 | } | |
605 | if (discon_alias) { | |
606 | alias = discon_alias; | |
607 | spin_lock(&alias->d_lock); | |
608 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
609 | if (IS_ROOT(alias) && | |
610 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 611 | __dget_dlock(alias); |
da502956 | 612 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
613 | return alias; |
614 | } | |
615 | } | |
da502956 NP |
616 | spin_unlock(&alias->d_lock); |
617 | goto again; | |
1da177e4 | 618 | } |
da502956 | 619 | return NULL; |
1da177e4 LT |
620 | } |
621 | ||
da502956 | 622 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 623 | { |
214fda1f DH |
624 | struct dentry *de = NULL; |
625 | ||
626 | if (!list_empty(&inode->i_dentry)) { | |
873feea0 | 627 | spin_lock(&inode->i_lock); |
214fda1f | 628 | de = __d_find_alias(inode, 0); |
873feea0 | 629 | spin_unlock(&inode->i_lock); |
214fda1f | 630 | } |
1da177e4 LT |
631 | return de; |
632 | } | |
ec4f8605 | 633 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
634 | |
635 | /* | |
636 | * Try to kill dentries associated with this inode. | |
637 | * WARNING: you must own a reference to inode. | |
638 | */ | |
639 | void d_prune_aliases(struct inode *inode) | |
640 | { | |
0cdca3f9 | 641 | struct dentry *dentry; |
1da177e4 | 642 | restart: |
873feea0 | 643 | spin_lock(&inode->i_lock); |
0cdca3f9 | 644 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 645 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 646 | if (!dentry->d_count) { |
dc0474be | 647 | __dget_dlock(dentry); |
1da177e4 LT |
648 | __d_drop(dentry); |
649 | spin_unlock(&dentry->d_lock); | |
873feea0 | 650 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
651 | dput(dentry); |
652 | goto restart; | |
653 | } | |
654 | spin_unlock(&dentry->d_lock); | |
655 | } | |
873feea0 | 656 | spin_unlock(&inode->i_lock); |
1da177e4 | 657 | } |
ec4f8605 | 658 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
659 | |
660 | /* | |
77812a1e NP |
661 | * Try to throw away a dentry - free the inode, dput the parent. |
662 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
663 | * Releases dentry->d_lock. | |
d702ccb3 | 664 | * |
77812a1e | 665 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 666 | */ |
77812a1e | 667 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 668 | __releases(dentry->d_lock) |
1da177e4 | 669 | { |
77812a1e | 670 | struct dentry *parent; |
d52b9086 | 671 | |
77812a1e | 672 | parent = dentry_kill(dentry, 0); |
d52b9086 | 673 | /* |
77812a1e NP |
674 | * If dentry_kill returns NULL, we have nothing more to do. |
675 | * if it returns the same dentry, trylocks failed. In either | |
676 | * case, just loop again. | |
677 | * | |
678 | * Otherwise, we need to prune ancestors too. This is necessary | |
679 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
680 | * is also expected to be beneficial in reducing dentry cache | |
681 | * fragmentation. | |
d52b9086 | 682 | */ |
77812a1e NP |
683 | if (!parent) |
684 | return; | |
685 | if (parent == dentry) | |
686 | return; | |
687 | ||
688 | /* Prune ancestors. */ | |
689 | dentry = parent; | |
d52b9086 | 690 | while (dentry) { |
b7ab39f6 | 691 | spin_lock(&dentry->d_lock); |
89e60548 NP |
692 | if (dentry->d_count > 1) { |
693 | dentry->d_count--; | |
694 | spin_unlock(&dentry->d_lock); | |
695 | return; | |
696 | } | |
77812a1e | 697 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 698 | } |
1da177e4 LT |
699 | } |
700 | ||
3049cfe2 | 701 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 702 | { |
da3bbdd4 | 703 | struct dentry *dentry; |
da3bbdd4 | 704 | |
ec33679d NP |
705 | rcu_read_lock(); |
706 | for (;;) { | |
ec33679d NP |
707 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
708 | if (&dentry->d_lru == list) | |
709 | break; /* empty */ | |
710 | spin_lock(&dentry->d_lock); | |
711 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
712 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
713 | continue; |
714 | } | |
715 | ||
1da177e4 LT |
716 | /* |
717 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
718 | * the LRU because of laziness during lookup. Do not free |
719 | * it - just keep it off the LRU list. | |
1da177e4 | 720 | */ |
b7ab39f6 | 721 | if (dentry->d_count) { |
ec33679d | 722 | dentry_lru_del(dentry); |
da3bbdd4 | 723 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
724 | continue; |
725 | } | |
ec33679d | 726 | |
ec33679d | 727 | rcu_read_unlock(); |
77812a1e NP |
728 | |
729 | try_prune_one_dentry(dentry); | |
730 | ||
ec33679d | 731 | rcu_read_lock(); |
da3bbdd4 | 732 | } |
ec33679d | 733 | rcu_read_unlock(); |
3049cfe2 CH |
734 | } |
735 | ||
736 | /** | |
737 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
738 | * @sb: superblock to shrink dentry LRU. | |
739 | * @count: number of entries to prune | |
740 | * @flags: flags to control the dentry processing | |
741 | * | |
742 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
743 | */ | |
744 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
745 | { | |
746 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
747 | struct dentry *dentry; | |
748 | LIST_HEAD(referenced); | |
749 | LIST_HEAD(tmp); | |
750 | int cnt = *count; | |
751 | ||
23044507 NP |
752 | relock: |
753 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
754 | while (!list_empty(&sb->s_dentry_lru)) { |
755 | dentry = list_entry(sb->s_dentry_lru.prev, | |
756 | struct dentry, d_lru); | |
757 | BUG_ON(dentry->d_sb != sb); | |
758 | ||
23044507 NP |
759 | if (!spin_trylock(&dentry->d_lock)) { |
760 | spin_unlock(&dcache_lru_lock); | |
761 | cpu_relax(); | |
762 | goto relock; | |
763 | } | |
764 | ||
3049cfe2 CH |
765 | /* |
766 | * If we are honouring the DCACHE_REFERENCED flag and the | |
767 | * dentry has this flag set, don't free it. Clear the flag | |
768 | * and put it back on the LRU. | |
769 | */ | |
23044507 NP |
770 | if (flags & DCACHE_REFERENCED && |
771 | dentry->d_flags & DCACHE_REFERENCED) { | |
772 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
773 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 774 | spin_unlock(&dentry->d_lock); |
23044507 NP |
775 | } else { |
776 | list_move_tail(&dentry->d_lru, &tmp); | |
777 | spin_unlock(&dentry->d_lock); | |
778 | if (!--cnt) | |
779 | break; | |
3049cfe2 | 780 | } |
ec33679d | 781 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 782 | } |
da3bbdd4 KM |
783 | if (!list_empty(&referenced)) |
784 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 785 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
786 | |
787 | shrink_dentry_list(&tmp); | |
788 | ||
789 | *count = cnt; | |
da3bbdd4 KM |
790 | } |
791 | ||
792 | /** | |
793 | * prune_dcache - shrink the dcache | |
794 | * @count: number of entries to try to free | |
795 | * | |
796 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
797 | * need to unmount something (at which point we need to unuse all dentries). | |
798 | * | |
799 | * This function may fail to free any resources if all the dentries are in use. | |
800 | */ | |
801 | static void prune_dcache(int count) | |
802 | { | |
dca33252 | 803 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 804 | int w_count; |
86c8749e | 805 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
806 | int prune_ratio; |
807 | int pruned; | |
808 | ||
809 | if (unused == 0 || count == 0) | |
810 | return; | |
da3bbdd4 KM |
811 | if (count >= unused) |
812 | prune_ratio = 1; | |
813 | else | |
814 | prune_ratio = unused / count; | |
815 | spin_lock(&sb_lock); | |
dca33252 | 816 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
817 | if (list_empty(&sb->s_instances)) |
818 | continue; | |
da3bbdd4 | 819 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 820 | continue; |
da3bbdd4 KM |
821 | sb->s_count++; |
822 | /* Now, we reclaim unused dentrins with fairness. | |
823 | * We reclaim them same percentage from each superblock. | |
824 | * We calculate number of dentries to scan on this sb | |
825 | * as follows, but the implementation is arranged to avoid | |
826 | * overflows: | |
827 | * number of dentries to scan on this sb = | |
828 | * count * (number of dentries on this sb / | |
829 | * number of dentries in the machine) | |
0feae5c4 | 830 | */ |
da3bbdd4 KM |
831 | spin_unlock(&sb_lock); |
832 | if (prune_ratio != 1) | |
833 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
834 | else | |
835 | w_count = sb->s_nr_dentry_unused; | |
836 | pruned = w_count; | |
0feae5c4 | 837 | /* |
da3bbdd4 KM |
838 | * We need to be sure this filesystem isn't being unmounted, |
839 | * otherwise we could race with generic_shutdown_super(), and | |
840 | * end up holding a reference to an inode while the filesystem | |
841 | * is unmounted. So we try to get s_umount, and make sure | |
842 | * s_root isn't NULL. | |
0feae5c4 | 843 | */ |
da3bbdd4 KM |
844 | if (down_read_trylock(&sb->s_umount)) { |
845 | if ((sb->s_root != NULL) && | |
846 | (!list_empty(&sb->s_dentry_lru))) { | |
da3bbdd4 KM |
847 | __shrink_dcache_sb(sb, &w_count, |
848 | DCACHE_REFERENCED); | |
849 | pruned -= w_count; | |
0feae5c4 | 850 | } |
da3bbdd4 | 851 | up_read(&sb->s_umount); |
0feae5c4 | 852 | } |
da3bbdd4 | 853 | spin_lock(&sb_lock); |
dca33252 AV |
854 | if (p) |
855 | __put_super(p); | |
da3bbdd4 | 856 | count -= pruned; |
dca33252 | 857 | p = sb; |
79893c17 AV |
858 | /* more work left to do? */ |
859 | if (count <= 0) | |
860 | break; | |
1da177e4 | 861 | } |
dca33252 AV |
862 | if (p) |
863 | __put_super(p); | |
da3bbdd4 | 864 | spin_unlock(&sb_lock); |
1da177e4 LT |
865 | } |
866 | ||
1da177e4 LT |
867 | /** |
868 | * shrink_dcache_sb - shrink dcache for a superblock | |
869 | * @sb: superblock | |
870 | * | |
3049cfe2 CH |
871 | * Shrink the dcache for the specified super block. This is used to free |
872 | * the dcache before unmounting a file system. | |
1da177e4 | 873 | */ |
3049cfe2 | 874 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 875 | { |
3049cfe2 CH |
876 | LIST_HEAD(tmp); |
877 | ||
23044507 | 878 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
879 | while (!list_empty(&sb->s_dentry_lru)) { |
880 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 881 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 882 | shrink_dentry_list(&tmp); |
ec33679d | 883 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 884 | } |
23044507 | 885 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 886 | } |
ec4f8605 | 887 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 888 | |
c636ebdb DH |
889 | /* |
890 | * destroy a single subtree of dentries for unmount | |
891 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
892 | * locking | |
893 | */ | |
894 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
895 | { | |
896 | struct dentry *parent; | |
f8713576 | 897 | unsigned detached = 0; |
c636ebdb DH |
898 | |
899 | BUG_ON(!IS_ROOT(dentry)); | |
900 | ||
901 | /* detach this root from the system */ | |
23044507 | 902 | spin_lock(&dentry->d_lock); |
a4633357 | 903 | dentry_lru_del(dentry); |
c636ebdb | 904 | __d_drop(dentry); |
da502956 | 905 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
906 | |
907 | for (;;) { | |
908 | /* descend to the first leaf in the current subtree */ | |
909 | while (!list_empty(&dentry->d_subdirs)) { | |
910 | struct dentry *loop; | |
911 | ||
912 | /* this is a branch with children - detach all of them | |
913 | * from the system in one go */ | |
2fd6b7f5 | 914 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
915 | list_for_each_entry(loop, &dentry->d_subdirs, |
916 | d_u.d_child) { | |
2fd6b7f5 NP |
917 | spin_lock_nested(&loop->d_lock, |
918 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 919 | dentry_lru_del(loop); |
c636ebdb | 920 | __d_drop(loop); |
da502956 | 921 | spin_unlock(&loop->d_lock); |
c636ebdb | 922 | } |
2fd6b7f5 | 923 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
924 | |
925 | /* move to the first child */ | |
926 | dentry = list_entry(dentry->d_subdirs.next, | |
927 | struct dentry, d_u.d_child); | |
928 | } | |
929 | ||
930 | /* consume the dentries from this leaf up through its parents | |
931 | * until we find one with children or run out altogether */ | |
932 | do { | |
933 | struct inode *inode; | |
934 | ||
b7ab39f6 | 935 | if (dentry->d_count != 0) { |
c636ebdb DH |
936 | printk(KERN_ERR |
937 | "BUG: Dentry %p{i=%lx,n=%s}" | |
938 | " still in use (%d)" | |
939 | " [unmount of %s %s]\n", | |
940 | dentry, | |
941 | dentry->d_inode ? | |
942 | dentry->d_inode->i_ino : 0UL, | |
943 | dentry->d_name.name, | |
b7ab39f6 | 944 | dentry->d_count, |
c636ebdb DH |
945 | dentry->d_sb->s_type->name, |
946 | dentry->d_sb->s_id); | |
947 | BUG(); | |
948 | } | |
949 | ||
2fd6b7f5 | 950 | if (IS_ROOT(dentry)) { |
c636ebdb | 951 | parent = NULL; |
2fd6b7f5 NP |
952 | list_del(&dentry->d_u.d_child); |
953 | } else { | |
871c0067 | 954 | parent = dentry->d_parent; |
b7ab39f6 NP |
955 | spin_lock(&parent->d_lock); |
956 | parent->d_count--; | |
2fd6b7f5 | 957 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 958 | spin_unlock(&parent->d_lock); |
871c0067 | 959 | } |
c636ebdb | 960 | |
f8713576 | 961 | detached++; |
c636ebdb DH |
962 | |
963 | inode = dentry->d_inode; | |
964 | if (inode) { | |
965 | dentry->d_inode = NULL; | |
966 | list_del_init(&dentry->d_alias); | |
967 | if (dentry->d_op && dentry->d_op->d_iput) | |
968 | dentry->d_op->d_iput(dentry, inode); | |
969 | else | |
970 | iput(inode); | |
971 | } | |
972 | ||
973 | d_free(dentry); | |
974 | ||
975 | /* finished when we fall off the top of the tree, | |
976 | * otherwise we ascend to the parent and move to the | |
977 | * next sibling if there is one */ | |
978 | if (!parent) | |
312d3ca8 | 979 | return; |
c636ebdb | 980 | dentry = parent; |
c636ebdb DH |
981 | } while (list_empty(&dentry->d_subdirs)); |
982 | ||
983 | dentry = list_entry(dentry->d_subdirs.next, | |
984 | struct dentry, d_u.d_child); | |
985 | } | |
986 | } | |
987 | ||
988 | /* | |
989 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 990 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
991 | * - the superblock is detached from all mountings and open files, so the |
992 | * dentry trees will not be rearranged by the VFS | |
993 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
994 | * any dentries belonging to this superblock that it comes across | |
995 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
996 | * in this superblock | |
997 | */ | |
998 | void shrink_dcache_for_umount(struct super_block *sb) | |
999 | { | |
1000 | struct dentry *dentry; | |
1001 | ||
1002 | if (down_read_trylock(&sb->s_umount)) | |
1003 | BUG(); | |
1004 | ||
1005 | dentry = sb->s_root; | |
1006 | sb->s_root = NULL; | |
b7ab39f6 NP |
1007 | spin_lock(&dentry->d_lock); |
1008 | dentry->d_count--; | |
1009 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
1010 | shrink_dcache_for_umount_subtree(dentry); |
1011 | ||
ceb5bdc2 NP |
1012 | while (!hlist_bl_empty(&sb->s_anon)) { |
1013 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
1014 | shrink_dcache_for_umount_subtree(dentry); |
1015 | } | |
1016 | } | |
1017 | ||
c826cb7d LT |
1018 | /* |
1019 | * This tries to ascend one level of parenthood, but | |
1020 | * we can race with renaming, so we need to re-check | |
1021 | * the parenthood after dropping the lock and check | |
1022 | * that the sequence number still matches. | |
1023 | */ | |
1024 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
1025 | { | |
1026 | struct dentry *new = old->d_parent; | |
1027 | ||
1028 | rcu_read_lock(); | |
1029 | spin_unlock(&old->d_lock); | |
1030 | spin_lock(&new->d_lock); | |
1031 | ||
1032 | /* | |
1033 | * might go back up the wrong parent if we have had a rename | |
1034 | * or deletion | |
1035 | */ | |
1036 | if (new != old->d_parent || | |
c83ce989 | 1037 | (old->d_flags & DCACHE_DISCONNECTED) || |
c826cb7d LT |
1038 | (!locked && read_seqretry(&rename_lock, seq))) { |
1039 | spin_unlock(&new->d_lock); | |
1040 | new = NULL; | |
1041 | } | |
1042 | rcu_read_unlock(); | |
1043 | return new; | |
1044 | } | |
1045 | ||
1046 | ||
1da177e4 LT |
1047 | /* |
1048 | * Search for at least 1 mount point in the dentry's subdirs. | |
1049 | * We descend to the next level whenever the d_subdirs | |
1050 | * list is non-empty and continue searching. | |
1051 | */ | |
1052 | ||
1053 | /** | |
1054 | * have_submounts - check for mounts over a dentry | |
1055 | * @parent: dentry to check. | |
1056 | * | |
1057 | * Return true if the parent or its subdirectories contain | |
1058 | * a mount point | |
1059 | */ | |
1da177e4 LT |
1060 | int have_submounts(struct dentry *parent) |
1061 | { | |
949854d0 | 1062 | struct dentry *this_parent; |
1da177e4 | 1063 | struct list_head *next; |
949854d0 | 1064 | unsigned seq; |
58db63d0 | 1065 | int locked = 0; |
949854d0 | 1066 | |
949854d0 | 1067 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1068 | again: |
1069 | this_parent = parent; | |
1da177e4 | 1070 | |
1da177e4 LT |
1071 | if (d_mountpoint(parent)) |
1072 | goto positive; | |
2fd6b7f5 | 1073 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1074 | repeat: |
1075 | next = this_parent->d_subdirs.next; | |
1076 | resume: | |
1077 | while (next != &this_parent->d_subdirs) { | |
1078 | struct list_head *tmp = next; | |
5160ee6f | 1079 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1080 | next = tmp->next; |
2fd6b7f5 NP |
1081 | |
1082 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1083 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1084 | if (d_mountpoint(dentry)) { |
1085 | spin_unlock(&dentry->d_lock); | |
1086 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1087 | goto positive; |
2fd6b7f5 | 1088 | } |
1da177e4 | 1089 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1090 | spin_unlock(&this_parent->d_lock); |
1091 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1092 | this_parent = dentry; |
2fd6b7f5 | 1093 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1094 | goto repeat; |
1095 | } | |
2fd6b7f5 | 1096 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1097 | } |
1098 | /* | |
1099 | * All done at this level ... ascend and resume the search. | |
1100 | */ | |
1101 | if (this_parent != parent) { | |
c826cb7d LT |
1102 | struct dentry *child = this_parent; |
1103 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1104 | if (!this_parent) | |
949854d0 | 1105 | goto rename_retry; |
949854d0 | 1106 | next = child->d_u.d_child.next; |
1da177e4 LT |
1107 | goto resume; |
1108 | } | |
2fd6b7f5 | 1109 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1110 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1111 | goto rename_retry; |
58db63d0 NP |
1112 | if (locked) |
1113 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1114 | return 0; /* No mount points found in tree */ |
1115 | positive: | |
58db63d0 | 1116 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1117 | goto rename_retry; |
58db63d0 NP |
1118 | if (locked) |
1119 | write_sequnlock(&rename_lock); | |
1da177e4 | 1120 | return 1; |
58db63d0 NP |
1121 | |
1122 | rename_retry: | |
1123 | locked = 1; | |
1124 | write_seqlock(&rename_lock); | |
1125 | goto again; | |
1da177e4 | 1126 | } |
ec4f8605 | 1127 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1128 | |
1129 | /* | |
1130 | * Search the dentry child list for the specified parent, | |
1131 | * and move any unused dentries to the end of the unused | |
1132 | * list for prune_dcache(). We descend to the next level | |
1133 | * whenever the d_subdirs list is non-empty and continue | |
1134 | * searching. | |
1135 | * | |
1136 | * It returns zero iff there are no unused children, | |
1137 | * otherwise it returns the number of children moved to | |
1138 | * the end of the unused list. This may not be the total | |
1139 | * number of unused children, because select_parent can | |
1140 | * drop the lock and return early due to latency | |
1141 | * constraints. | |
1142 | */ | |
1143 | static int select_parent(struct dentry * parent) | |
1144 | { | |
949854d0 | 1145 | struct dentry *this_parent; |
1da177e4 | 1146 | struct list_head *next; |
949854d0 | 1147 | unsigned seq; |
1da177e4 | 1148 | int found = 0; |
58db63d0 | 1149 | int locked = 0; |
1da177e4 | 1150 | |
949854d0 | 1151 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1152 | again: |
1153 | this_parent = parent; | |
2fd6b7f5 | 1154 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1155 | repeat: |
1156 | next = this_parent->d_subdirs.next; | |
1157 | resume: | |
1158 | while (next != &this_parent->d_subdirs) { | |
1159 | struct list_head *tmp = next; | |
5160ee6f | 1160 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1161 | next = tmp->next; |
1162 | ||
2fd6b7f5 | 1163 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1164 | |
1da177e4 LT |
1165 | /* |
1166 | * move only zero ref count dentries to the end | |
1167 | * of the unused list for prune_dcache | |
1168 | */ | |
b7ab39f6 | 1169 | if (!dentry->d_count) { |
a4633357 | 1170 | dentry_lru_move_tail(dentry); |
1da177e4 | 1171 | found++; |
a4633357 CH |
1172 | } else { |
1173 | dentry_lru_del(dentry); | |
1da177e4 LT |
1174 | } |
1175 | ||
1176 | /* | |
1177 | * We can return to the caller if we have found some (this | |
1178 | * ensures forward progress). We'll be coming back to find | |
1179 | * the rest. | |
1180 | */ | |
2fd6b7f5 NP |
1181 | if (found && need_resched()) { |
1182 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1183 | goto out; |
2fd6b7f5 | 1184 | } |
1da177e4 LT |
1185 | |
1186 | /* | |
1187 | * Descend a level if the d_subdirs list is non-empty. | |
1188 | */ | |
1189 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1190 | spin_unlock(&this_parent->d_lock); |
1191 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1192 | this_parent = dentry; |
2fd6b7f5 | 1193 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1194 | goto repeat; |
1195 | } | |
2fd6b7f5 NP |
1196 | |
1197 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1198 | } |
1199 | /* | |
1200 | * All done at this level ... ascend and resume the search. | |
1201 | */ | |
1202 | if (this_parent != parent) { | |
c826cb7d LT |
1203 | struct dentry *child = this_parent; |
1204 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1205 | if (!this_parent) | |
949854d0 | 1206 | goto rename_retry; |
949854d0 | 1207 | next = child->d_u.d_child.next; |
1da177e4 LT |
1208 | goto resume; |
1209 | } | |
1210 | out: | |
2fd6b7f5 | 1211 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1212 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1213 | goto rename_retry; |
58db63d0 NP |
1214 | if (locked) |
1215 | write_sequnlock(&rename_lock); | |
1da177e4 | 1216 | return found; |
58db63d0 NP |
1217 | |
1218 | rename_retry: | |
1219 | if (found) | |
1220 | return found; | |
1221 | locked = 1; | |
1222 | write_seqlock(&rename_lock); | |
1223 | goto again; | |
1da177e4 LT |
1224 | } |
1225 | ||
1226 | /** | |
1227 | * shrink_dcache_parent - prune dcache | |
1228 | * @parent: parent of entries to prune | |
1229 | * | |
1230 | * Prune the dcache to remove unused children of the parent dentry. | |
1231 | */ | |
1232 | ||
1233 | void shrink_dcache_parent(struct dentry * parent) | |
1234 | { | |
da3bbdd4 | 1235 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1236 | int found; |
1237 | ||
1238 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 1239 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 1240 | } |
ec4f8605 | 1241 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1242 | |
1da177e4 LT |
1243 | /* |
1244 | * Scan `nr' dentries and return the number which remain. | |
1245 | * | |
1246 | * We need to avoid reentering the filesystem if the caller is performing a | |
1247 | * GFP_NOFS allocation attempt. One example deadlock is: | |
1248 | * | |
1249 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
1250 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
1251 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
1252 | * | |
1253 | * In this case we return -1 to tell the caller that we baled. | |
1254 | */ | |
7f8275d0 | 1255 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
1256 | { |
1257 | if (nr) { | |
1258 | if (!(gfp_mask & __GFP_FS)) | |
1259 | return -1; | |
da3bbdd4 | 1260 | prune_dcache(nr); |
1da177e4 | 1261 | } |
312d3ca8 | 1262 | |
86c8749e | 1263 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
1264 | } |
1265 | ||
8e1f936b RR |
1266 | static struct shrinker dcache_shrinker = { |
1267 | .shrink = shrink_dcache_memory, | |
1268 | .seeks = DEFAULT_SEEKS, | |
1269 | }; | |
1270 | ||
1da177e4 LT |
1271 | /** |
1272 | * d_alloc - allocate a dcache entry | |
1273 | * @parent: parent of entry to allocate | |
1274 | * @name: qstr of the name | |
1275 | * | |
1276 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1277 | * available. On a success the dentry is returned. The name passed in is | |
1278 | * copied and the copy passed in may be reused after this call. | |
1279 | */ | |
1280 | ||
1281 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1282 | { | |
1283 | struct dentry *dentry; | |
1284 | char *dname; | |
1285 | ||
e12ba74d | 1286 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1287 | if (!dentry) |
1288 | return NULL; | |
1289 | ||
1290 | if (name->len > DNAME_INLINE_LEN-1) { | |
1291 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1292 | if (!dname) { | |
1293 | kmem_cache_free(dentry_cache, dentry); | |
1294 | return NULL; | |
1295 | } | |
1296 | } else { | |
1297 | dname = dentry->d_iname; | |
1298 | } | |
1299 | dentry->d_name.name = dname; | |
1300 | ||
1301 | dentry->d_name.len = name->len; | |
1302 | dentry->d_name.hash = name->hash; | |
1303 | memcpy(dname, name->name, name->len); | |
1304 | dname[name->len] = 0; | |
1305 | ||
b7ab39f6 | 1306 | dentry->d_count = 1; |
1da177e4 LT |
1307 | dentry->d_flags = DCACHE_UNHASHED; |
1308 | spin_lock_init(&dentry->d_lock); | |
31e6b01f | 1309 | seqcount_init(&dentry->d_seq); |
1da177e4 LT |
1310 | dentry->d_inode = NULL; |
1311 | dentry->d_parent = NULL; | |
1312 | dentry->d_sb = NULL; | |
1313 | dentry->d_op = NULL; | |
1314 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1315 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1316 | INIT_LIST_HEAD(&dentry->d_lru); |
1317 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1318 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1319 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
1320 | |
1321 | if (parent) { | |
2fd6b7f5 | 1322 | spin_lock(&parent->d_lock); |
89ad485f NP |
1323 | /* |
1324 | * don't need child lock because it is not subject | |
1325 | * to concurrency here | |
1326 | */ | |
dc0474be NP |
1327 | __dget_dlock(parent); |
1328 | dentry->d_parent = parent; | |
1da177e4 | 1329 | dentry->d_sb = parent->d_sb; |
c8aebb0c | 1330 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
5160ee6f | 1331 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
2fd6b7f5 | 1332 | spin_unlock(&parent->d_lock); |
2fd6b7f5 | 1333 | } |
1da177e4 | 1334 | |
3e880fb5 | 1335 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1336 | |
1da177e4 LT |
1337 | return dentry; |
1338 | } | |
ec4f8605 | 1339 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1340 | |
4b936885 NP |
1341 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1342 | { | |
1343 | struct dentry *dentry = d_alloc(NULL, name); | |
1344 | if (dentry) { | |
1345 | dentry->d_sb = sb; | |
c8aebb0c | 1346 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
4b936885 NP |
1347 | dentry->d_parent = dentry; |
1348 | dentry->d_flags |= DCACHE_DISCONNECTED; | |
1349 | } | |
1350 | return dentry; | |
1351 | } | |
1352 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1353 | ||
1da177e4 LT |
1354 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1355 | { | |
1356 | struct qstr q; | |
1357 | ||
1358 | q.name = name; | |
1359 | q.len = strlen(name); | |
1360 | q.hash = full_name_hash(q.name, q.len); | |
1361 | return d_alloc(parent, &q); | |
1362 | } | |
ef26ca97 | 1363 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1364 | |
fb045adb NP |
1365 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1366 | { | |
6f7f7caa LT |
1367 | WARN_ON_ONCE(dentry->d_op); |
1368 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1369 | DCACHE_OP_COMPARE | |
1370 | DCACHE_OP_REVALIDATE | | |
1371 | DCACHE_OP_DELETE )); | |
1372 | dentry->d_op = op; | |
1373 | if (!op) | |
1374 | return; | |
1375 | if (op->d_hash) | |
1376 | dentry->d_flags |= DCACHE_OP_HASH; | |
1377 | if (op->d_compare) | |
1378 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1379 | if (op->d_revalidate) | |
1380 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1381 | if (op->d_delete) | |
1382 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1383 | ||
1384 | } | |
1385 | EXPORT_SYMBOL(d_set_d_op); | |
1386 | ||
360da900 OH |
1387 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1388 | { | |
b23fb0a6 | 1389 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1390 | if (inode) { |
1391 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1392 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
360da900 | 1393 | list_add(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1394 | } |
360da900 | 1395 | dentry->d_inode = inode; |
31e6b01f | 1396 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1397 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1398 | fsnotify_d_instantiate(dentry, inode); |
1399 | } | |
1400 | ||
1da177e4 LT |
1401 | /** |
1402 | * d_instantiate - fill in inode information for a dentry | |
1403 | * @entry: dentry to complete | |
1404 | * @inode: inode to attach to this dentry | |
1405 | * | |
1406 | * Fill in inode information in the entry. | |
1407 | * | |
1408 | * This turns negative dentries into productive full members | |
1409 | * of society. | |
1410 | * | |
1411 | * NOTE! This assumes that the inode count has been incremented | |
1412 | * (or otherwise set) by the caller to indicate that it is now | |
1413 | * in use by the dcache. | |
1414 | */ | |
1415 | ||
1416 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1417 | { | |
28133c7b | 1418 | BUG_ON(!list_empty(&entry->d_alias)); |
873feea0 NP |
1419 | if (inode) |
1420 | spin_lock(&inode->i_lock); | |
360da900 | 1421 | __d_instantiate(entry, inode); |
873feea0 NP |
1422 | if (inode) |
1423 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1424 | security_d_instantiate(entry, inode); |
1425 | } | |
ec4f8605 | 1426 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1427 | |
1428 | /** | |
1429 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1430 | * @entry: dentry to instantiate | |
1431 | * @inode: inode to attach to this dentry | |
1432 | * | |
1433 | * Fill in inode information in the entry. On success, it returns NULL. | |
1434 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1435 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1436 | * |
1437 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1438 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1439 | * |
1440 | * This also assumes that the inode count has been incremented | |
1441 | * (or otherwise set) by the caller to indicate that it is now | |
1442 | * in use by the dcache. | |
1da177e4 | 1443 | */ |
770bfad8 DH |
1444 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1445 | struct inode *inode) | |
1da177e4 LT |
1446 | { |
1447 | struct dentry *alias; | |
1448 | int len = entry->d_name.len; | |
1449 | const char *name = entry->d_name.name; | |
1450 | unsigned int hash = entry->d_name.hash; | |
1451 | ||
770bfad8 | 1452 | if (!inode) { |
360da900 | 1453 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1454 | return NULL; |
1455 | } | |
1456 | ||
1da177e4 LT |
1457 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1458 | struct qstr *qstr = &alias->d_name; | |
1459 | ||
9abca360 NP |
1460 | /* |
1461 | * Don't need alias->d_lock here, because aliases with | |
1462 | * d_parent == entry->d_parent are not subject to name or | |
1463 | * parent changes, because the parent inode i_mutex is held. | |
1464 | */ | |
1da177e4 LT |
1465 | if (qstr->hash != hash) |
1466 | continue; | |
1467 | if (alias->d_parent != entry->d_parent) | |
1468 | continue; | |
9d55c369 | 1469 | if (dentry_cmp(qstr->name, qstr->len, name, len)) |
1da177e4 | 1470 | continue; |
dc0474be | 1471 | __dget(alias); |
1da177e4 LT |
1472 | return alias; |
1473 | } | |
770bfad8 | 1474 | |
360da900 | 1475 | __d_instantiate(entry, inode); |
1da177e4 LT |
1476 | return NULL; |
1477 | } | |
770bfad8 DH |
1478 | |
1479 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1480 | { | |
1481 | struct dentry *result; | |
1482 | ||
1483 | BUG_ON(!list_empty(&entry->d_alias)); | |
1484 | ||
873feea0 NP |
1485 | if (inode) |
1486 | spin_lock(&inode->i_lock); | |
770bfad8 | 1487 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1488 | if (inode) |
1489 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1490 | |
1491 | if (!result) { | |
1492 | security_d_instantiate(entry, inode); | |
1493 | return NULL; | |
1494 | } | |
1495 | ||
1496 | BUG_ON(!d_unhashed(result)); | |
1497 | iput(inode); | |
1498 | return result; | |
1499 | } | |
1500 | ||
1da177e4 LT |
1501 | EXPORT_SYMBOL(d_instantiate_unique); |
1502 | ||
1503 | /** | |
1504 | * d_alloc_root - allocate root dentry | |
1505 | * @root_inode: inode to allocate the root for | |
1506 | * | |
1507 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1508 | * instantiated and returned. %NULL is returned if there is insufficient | |
1509 | * memory or the inode passed is %NULL. | |
1510 | */ | |
1511 | ||
1512 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1513 | { | |
1514 | struct dentry *res = NULL; | |
1515 | ||
1516 | if (root_inode) { | |
1517 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1518 | ||
1519 | res = d_alloc(NULL, &name); | |
1520 | if (res) { | |
1521 | res->d_sb = root_inode->i_sb; | |
c8aebb0c | 1522 | d_set_d_op(res, res->d_sb->s_d_op); |
1da177e4 LT |
1523 | res->d_parent = res; |
1524 | d_instantiate(res, root_inode); | |
1525 | } | |
1526 | } | |
1527 | return res; | |
1528 | } | |
ec4f8605 | 1529 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 | 1530 | |
d891eedb BF |
1531 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1532 | { | |
1533 | struct dentry *alias; | |
1534 | ||
1535 | if (list_empty(&inode->i_dentry)) | |
1536 | return NULL; | |
1537 | alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias); | |
1538 | __dget(alias); | |
1539 | return alias; | |
1540 | } | |
1541 | ||
1542 | static struct dentry * d_find_any_alias(struct inode *inode) | |
1543 | { | |
1544 | struct dentry *de; | |
1545 | ||
1546 | spin_lock(&inode->i_lock); | |
1547 | de = __d_find_any_alias(inode); | |
1548 | spin_unlock(&inode->i_lock); | |
1549 | return de; | |
1550 | } | |
1551 | ||
1552 | ||
4ea3ada2 CH |
1553 | /** |
1554 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1555 | * @inode: inode to allocate the dentry for | |
1556 | * | |
1557 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1558 | * similar open by handle operations. The returned dentry may be anonymous, | |
1559 | * or may have a full name (if the inode was already in the cache). | |
1560 | * | |
1561 | * When called on a directory inode, we must ensure that the inode only ever | |
1562 | * has one dentry. If a dentry is found, that is returned instead of | |
1563 | * allocating a new one. | |
1564 | * | |
1565 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1566 | * to the dentry. In case of an error the reference on the inode is released. |
1567 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1568 | * be passed in and will be the error will be propagate to the return value, | |
1569 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1570 | */ |
1571 | struct dentry *d_obtain_alias(struct inode *inode) | |
1572 | { | |
9308a612 CH |
1573 | static const struct qstr anonstring = { .name = "" }; |
1574 | struct dentry *tmp; | |
1575 | struct dentry *res; | |
4ea3ada2 CH |
1576 | |
1577 | if (!inode) | |
44003728 | 1578 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1579 | if (IS_ERR(inode)) |
1580 | return ERR_CAST(inode); | |
1581 | ||
d891eedb | 1582 | res = d_find_any_alias(inode); |
9308a612 CH |
1583 | if (res) |
1584 | goto out_iput; | |
1585 | ||
1586 | tmp = d_alloc(NULL, &anonstring); | |
1587 | if (!tmp) { | |
1588 | res = ERR_PTR(-ENOMEM); | |
1589 | goto out_iput; | |
4ea3ada2 | 1590 | } |
9308a612 CH |
1591 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1592 | ||
b5c84bf6 | 1593 | |
873feea0 | 1594 | spin_lock(&inode->i_lock); |
d891eedb | 1595 | res = __d_find_any_alias(inode); |
9308a612 | 1596 | if (res) { |
873feea0 | 1597 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1598 | dput(tmp); |
1599 | goto out_iput; | |
1600 | } | |
1601 | ||
1602 | /* attach a disconnected dentry */ | |
1603 | spin_lock(&tmp->d_lock); | |
1604 | tmp->d_sb = inode->i_sb; | |
c8aebb0c | 1605 | d_set_d_op(tmp, tmp->d_sb->s_d_op); |
9308a612 CH |
1606 | tmp->d_inode = inode; |
1607 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
9308a612 | 1608 | list_add(&tmp->d_alias, &inode->i_dentry); |
ceb5bdc2 NP |
1609 | bit_spin_lock(0, (unsigned long *)&tmp->d_sb->s_anon.first); |
1610 | tmp->d_flags &= ~DCACHE_UNHASHED; | |
1611 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); | |
1612 | __bit_spin_unlock(0, (unsigned long *)&tmp->d_sb->s_anon.first); | |
9308a612 | 1613 | spin_unlock(&tmp->d_lock); |
873feea0 | 1614 | spin_unlock(&inode->i_lock); |
24ff6663 | 1615 | security_d_instantiate(tmp, inode); |
9308a612 | 1616 | |
9308a612 CH |
1617 | return tmp; |
1618 | ||
1619 | out_iput: | |
24ff6663 JB |
1620 | if (res && !IS_ERR(res)) |
1621 | security_d_instantiate(res, inode); | |
9308a612 CH |
1622 | iput(inode); |
1623 | return res; | |
4ea3ada2 | 1624 | } |
adc48720 | 1625 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1626 | |
1627 | /** | |
1628 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1629 | * @inode: the inode which may have a disconnected dentry | |
1630 | * @dentry: a negative dentry which we want to point to the inode. | |
1631 | * | |
1632 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1633 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1634 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1635 | * | |
1636 | * This is needed in the lookup routine of any filesystem that is exportable | |
1637 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1638 | * | |
1639 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1640 | * is returned. This matches the expected return value of ->lookup. | |
1641 | * | |
1642 | */ | |
1643 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1644 | { | |
1645 | struct dentry *new = NULL; | |
1646 | ||
21c0d8fd | 1647 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1648 | spin_lock(&inode->i_lock); |
1da177e4 LT |
1649 | new = __d_find_alias(inode, 1); |
1650 | if (new) { | |
1651 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
873feea0 | 1652 | spin_unlock(&inode->i_lock); |
1da177e4 | 1653 | security_d_instantiate(new, inode); |
1da177e4 LT |
1654 | d_move(new, dentry); |
1655 | iput(inode); | |
1656 | } else { | |
873feea0 | 1657 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1658 | __d_instantiate(dentry, inode); |
873feea0 | 1659 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1660 | security_d_instantiate(dentry, inode); |
1661 | d_rehash(dentry); | |
1662 | } | |
1663 | } else | |
1664 | d_add(dentry, inode); | |
1665 | return new; | |
1666 | } | |
ec4f8605 | 1667 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1668 | |
9403540c BN |
1669 | /** |
1670 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1671 | * @inode: the inode case-insensitive lookup has found | |
1672 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1673 | * @name: the case-exact name to be associated with the returned dentry | |
1674 | * | |
1675 | * This is to avoid filling the dcache with case-insensitive names to the | |
1676 | * same inode, only the actual correct case is stored in the dcache for | |
1677 | * case-insensitive filesystems. | |
1678 | * | |
1679 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1680 | * already exists in in the dcache, use it and return it. | |
1681 | * | |
1682 | * If no entry exists with the exact case name, allocate new dentry with | |
1683 | * the exact case, and return the spliced entry. | |
1684 | */ | |
e45b590b | 1685 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1686 | struct qstr *name) |
1687 | { | |
1688 | int error; | |
1689 | struct dentry *found; | |
1690 | struct dentry *new; | |
1691 | ||
b6520c81 CH |
1692 | /* |
1693 | * First check if a dentry matching the name already exists, | |
1694 | * if not go ahead and create it now. | |
1695 | */ | |
9403540c | 1696 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1697 | if (!found) { |
1698 | new = d_alloc(dentry->d_parent, name); | |
1699 | if (!new) { | |
1700 | error = -ENOMEM; | |
1701 | goto err_out; | |
1702 | } | |
b6520c81 | 1703 | |
9403540c BN |
1704 | found = d_splice_alias(inode, new); |
1705 | if (found) { | |
1706 | dput(new); | |
1707 | return found; | |
1708 | } | |
1709 | return new; | |
1710 | } | |
b6520c81 CH |
1711 | |
1712 | /* | |
1713 | * If a matching dentry exists, and it's not negative use it. | |
1714 | * | |
1715 | * Decrement the reference count to balance the iget() done | |
1716 | * earlier on. | |
1717 | */ | |
9403540c BN |
1718 | if (found->d_inode) { |
1719 | if (unlikely(found->d_inode != inode)) { | |
1720 | /* This can't happen because bad inodes are unhashed. */ | |
1721 | BUG_ON(!is_bad_inode(inode)); | |
1722 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1723 | } | |
9403540c BN |
1724 | iput(inode); |
1725 | return found; | |
1726 | } | |
b6520c81 | 1727 | |
9403540c BN |
1728 | /* |
1729 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1730 | * already has a dentry. |
9403540c | 1731 | */ |
873feea0 | 1732 | spin_lock(&inode->i_lock); |
b6520c81 | 1733 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1734 | __d_instantiate(found, inode); |
873feea0 | 1735 | spin_unlock(&inode->i_lock); |
9403540c BN |
1736 | security_d_instantiate(found, inode); |
1737 | return found; | |
1738 | } | |
b6520c81 | 1739 | |
9403540c | 1740 | /* |
b6520c81 CH |
1741 | * In case a directory already has a (disconnected) entry grab a |
1742 | * reference to it, move it in place and use it. | |
9403540c BN |
1743 | */ |
1744 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
dc0474be | 1745 | __dget(new); |
873feea0 | 1746 | spin_unlock(&inode->i_lock); |
9403540c | 1747 | security_d_instantiate(found, inode); |
9403540c | 1748 | d_move(new, found); |
9403540c | 1749 | iput(inode); |
9403540c | 1750 | dput(found); |
9403540c BN |
1751 | return new; |
1752 | ||
1753 | err_out: | |
1754 | iput(inode); | |
1755 | return ERR_PTR(error); | |
1756 | } | |
ec4f8605 | 1757 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1758 | |
31e6b01f NP |
1759 | /** |
1760 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1761 | * @parent: parent dentry | |
1762 | * @name: qstr of name we wish to find | |
1763 | * @seq: returns d_seq value at the point where the dentry was found | |
1764 | * @inode: returns dentry->d_inode when the inode was found valid. | |
1765 | * Returns: dentry, or NULL | |
1766 | * | |
1767 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1768 | * resolution (store-free path walking) design described in | |
1769 | * Documentation/filesystems/path-lookup.txt. | |
1770 | * | |
1771 | * This is not to be used outside core vfs. | |
1772 | * | |
1773 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1774 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1775 | * without taking d_lock and checking d_seq sequence count against @seq | |
1776 | * returned here. | |
1777 | * | |
1778 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1779 | * function. | |
1780 | * | |
1781 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1782 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1783 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1784 | * is formed, giving integrity down the path walk. | |
1785 | */ | |
1786 | struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name, | |
1787 | unsigned *seq, struct inode **inode) | |
1788 | { | |
1789 | unsigned int len = name->len; | |
1790 | unsigned int hash = name->hash; | |
1791 | const unsigned char *str = name->name; | |
ceb5bdc2 NP |
1792 | struct dcache_hash_bucket *b = d_hash(parent, hash); |
1793 | struct hlist_bl_node *node; | |
31e6b01f NP |
1794 | struct dentry *dentry; |
1795 | ||
1796 | /* | |
1797 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1798 | * required to prevent single threaded performance regressions | |
1799 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1800 | * Keep the two functions in sync. | |
1801 | */ | |
1802 | ||
1803 | /* | |
1804 | * The hash list is protected using RCU. | |
1805 | * | |
1806 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1807 | * races with d_move(). | |
1808 | * | |
1809 | * It is possible that concurrent renames can mess up our list | |
1810 | * walk here and result in missing our dentry, resulting in the | |
1811 | * false-negative result. d_lookup() protects against concurrent | |
1812 | * renames using rename_lock seqlock. | |
1813 | * | |
b0a4bb83 | 1814 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1815 | */ |
ceb5bdc2 | 1816 | hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) { |
31e6b01f NP |
1817 | struct inode *i; |
1818 | const char *tname; | |
1819 | int tlen; | |
1820 | ||
1821 | if (dentry->d_name.hash != hash) | |
1822 | continue; | |
1823 | ||
1824 | seqretry: | |
1825 | *seq = read_seqcount_begin(&dentry->d_seq); | |
1826 | if (dentry->d_parent != parent) | |
1827 | continue; | |
1828 | if (d_unhashed(dentry)) | |
1829 | continue; | |
1830 | tlen = dentry->d_name.len; | |
1831 | tname = dentry->d_name.name; | |
1832 | i = dentry->d_inode; | |
e1bb5782 NP |
1833 | prefetch(tname); |
1834 | if (i) | |
1835 | prefetch(i); | |
31e6b01f NP |
1836 | /* |
1837 | * This seqcount check is required to ensure name and | |
1838 | * len are loaded atomically, so as not to walk off the | |
1839 | * edge of memory when walking. If we could load this | |
1840 | * atomically some other way, we could drop this check. | |
1841 | */ | |
1842 | if (read_seqcount_retry(&dentry->d_seq, *seq)) | |
1843 | goto seqretry; | |
fb045adb | 1844 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
31e6b01f NP |
1845 | if (parent->d_op->d_compare(parent, *inode, |
1846 | dentry, i, | |
1847 | tlen, tname, name)) | |
1848 | continue; | |
1849 | } else { | |
9d55c369 | 1850 | if (dentry_cmp(tname, tlen, str, len)) |
31e6b01f NP |
1851 | continue; |
1852 | } | |
1853 | /* | |
1854 | * No extra seqcount check is required after the name | |
1855 | * compare. The caller must perform a seqcount check in | |
1856 | * order to do anything useful with the returned dentry | |
1857 | * anyway. | |
1858 | */ | |
1859 | *inode = i; | |
1860 | return dentry; | |
1861 | } | |
1862 | return NULL; | |
1863 | } | |
1864 | ||
1da177e4 LT |
1865 | /** |
1866 | * d_lookup - search for a dentry | |
1867 | * @parent: parent dentry | |
1868 | * @name: qstr of name we wish to find | |
b04f784e | 1869 | * Returns: dentry, or NULL |
1da177e4 | 1870 | * |
b04f784e NP |
1871 | * d_lookup searches the children of the parent dentry for the name in |
1872 | * question. If the dentry is found its reference count is incremented and the | |
1873 | * dentry is returned. The caller must use dput to free the entry when it has | |
1874 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1875 | */ |
31e6b01f | 1876 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 | 1877 | { |
31e6b01f | 1878 | struct dentry *dentry; |
949854d0 | 1879 | unsigned seq; |
1da177e4 LT |
1880 | |
1881 | do { | |
1882 | seq = read_seqbegin(&rename_lock); | |
1883 | dentry = __d_lookup(parent, name); | |
1884 | if (dentry) | |
1885 | break; | |
1886 | } while (read_seqretry(&rename_lock, seq)); | |
1887 | return dentry; | |
1888 | } | |
ec4f8605 | 1889 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1890 | |
31e6b01f | 1891 | /** |
b04f784e NP |
1892 | * __d_lookup - search for a dentry (racy) |
1893 | * @parent: parent dentry | |
1894 | * @name: qstr of name we wish to find | |
1895 | * Returns: dentry, or NULL | |
1896 | * | |
1897 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1898 | * false-negative result due to unrelated rename activity. | |
1899 | * | |
1900 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1901 | * however it must be used carefully, eg. with a following d_lookup in | |
1902 | * the case of failure. | |
1903 | * | |
1904 | * __d_lookup callers must be commented. | |
1905 | */ | |
31e6b01f | 1906 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 LT |
1907 | { |
1908 | unsigned int len = name->len; | |
1909 | unsigned int hash = name->hash; | |
1910 | const unsigned char *str = name->name; | |
ceb5bdc2 NP |
1911 | struct dcache_hash_bucket *b = d_hash(parent, hash); |
1912 | struct hlist_bl_node *node; | |
31e6b01f | 1913 | struct dentry *found = NULL; |
665a7583 | 1914 | struct dentry *dentry; |
1da177e4 | 1915 | |
31e6b01f NP |
1916 | /* |
1917 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1918 | * required to prevent single threaded performance regressions | |
1919 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1920 | * Keep the two functions in sync. | |
1921 | */ | |
1922 | ||
b04f784e NP |
1923 | /* |
1924 | * The hash list is protected using RCU. | |
1925 | * | |
1926 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1927 | * with d_move(). | |
1928 | * | |
1929 | * It is possible that concurrent renames can mess up our list | |
1930 | * walk here and result in missing our dentry, resulting in the | |
1931 | * false-negative result. d_lookup() protects against concurrent | |
1932 | * renames using rename_lock seqlock. | |
1933 | * | |
b0a4bb83 | 1934 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1935 | */ |
1da177e4 LT |
1936 | rcu_read_lock(); |
1937 | ||
ceb5bdc2 | 1938 | hlist_bl_for_each_entry_rcu(dentry, node, &b->head, d_hash) { |
31e6b01f NP |
1939 | const char *tname; |
1940 | int tlen; | |
1da177e4 | 1941 | |
1da177e4 LT |
1942 | if (dentry->d_name.hash != hash) |
1943 | continue; | |
1da177e4 LT |
1944 | |
1945 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1946 | if (dentry->d_parent != parent) |
1947 | goto next; | |
d0185c08 LT |
1948 | if (d_unhashed(dentry)) |
1949 | goto next; | |
1950 | ||
1da177e4 LT |
1951 | /* |
1952 | * It is safe to compare names since d_move() cannot | |
1953 | * change the qstr (protected by d_lock). | |
1954 | */ | |
31e6b01f NP |
1955 | tlen = dentry->d_name.len; |
1956 | tname = dentry->d_name.name; | |
fb045adb | 1957 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
621e155a NP |
1958 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1959 | dentry, dentry->d_inode, | |
31e6b01f | 1960 | tlen, tname, name)) |
1da177e4 LT |
1961 | goto next; |
1962 | } else { | |
9d55c369 | 1963 | if (dentry_cmp(tname, tlen, str, len)) |
1da177e4 LT |
1964 | goto next; |
1965 | } | |
1966 | ||
b7ab39f6 | 1967 | dentry->d_count++; |
d0185c08 | 1968 | found = dentry; |
1da177e4 LT |
1969 | spin_unlock(&dentry->d_lock); |
1970 | break; | |
1971 | next: | |
1972 | spin_unlock(&dentry->d_lock); | |
1973 | } | |
1974 | rcu_read_unlock(); | |
1975 | ||
1976 | return found; | |
1977 | } | |
1978 | ||
3e7e241f EB |
1979 | /** |
1980 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1981 | * @dir: Directory to search in | |
1982 | * @name: qstr of name we wish to find | |
1983 | * | |
1984 | * On hash failure or on lookup failure NULL is returned. | |
1985 | */ | |
1986 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1987 | { | |
1988 | struct dentry *dentry = NULL; | |
1989 | ||
1990 | /* | |
1991 | * Check for a fs-specific hash function. Note that we must | |
1992 | * calculate the standard hash first, as the d_op->d_hash() | |
1993 | * routine may choose to leave the hash value unchanged. | |
1994 | */ | |
1995 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1996 | if (dir->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 1997 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1998 | goto out; |
1999 | } | |
2000 | dentry = d_lookup(dir, name); | |
2001 | out: | |
2002 | return dentry; | |
2003 | } | |
2004 | ||
1da177e4 | 2005 | /** |
786a5e15 | 2006 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2007 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2008 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2009 | * |
2010 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2011 | * This is used by ncpfs in its readdir implementation. | |
2012 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2013 | * |
2014 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2015 | */ |
d3a23e16 | 2016 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2017 | { |
786a5e15 | 2018 | struct dentry *child; |
d3a23e16 | 2019 | |
2fd6b7f5 | 2020 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2021 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2022 | if (dentry == child) { | |
2fd6b7f5 | 2023 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2024 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2025 | spin_unlock(&dentry->d_lock); |
2026 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2027 | return 1; |
2028 | } | |
2029 | } | |
2fd6b7f5 | 2030 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2031 | |
1da177e4 LT |
2032 | return 0; |
2033 | } | |
ec4f8605 | 2034 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2035 | |
2036 | /* | |
2037 | * When a file is deleted, we have two options: | |
2038 | * - turn this dentry into a negative dentry | |
2039 | * - unhash this dentry and free it. | |
2040 | * | |
2041 | * Usually, we want to just turn this into | |
2042 | * a negative dentry, but if anybody else is | |
2043 | * currently using the dentry or the inode | |
2044 | * we can't do that and we fall back on removing | |
2045 | * it from the hash queues and waiting for | |
2046 | * it to be deleted later when it has no users | |
2047 | */ | |
2048 | ||
2049 | /** | |
2050 | * d_delete - delete a dentry | |
2051 | * @dentry: The dentry to delete | |
2052 | * | |
2053 | * Turn the dentry into a negative dentry if possible, otherwise | |
2054 | * remove it from the hash queues so it can be deleted later | |
2055 | */ | |
2056 | ||
2057 | void d_delete(struct dentry * dentry) | |
2058 | { | |
873feea0 | 2059 | struct inode *inode; |
7a91bf7f | 2060 | int isdir = 0; |
1da177e4 LT |
2061 | /* |
2062 | * Are we the only user? | |
2063 | */ | |
357f8e65 | 2064 | again: |
1da177e4 | 2065 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2066 | inode = dentry->d_inode; |
2067 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 2068 | if (dentry->d_count == 1) { |
873feea0 | 2069 | if (inode && !spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2070 | spin_unlock(&dentry->d_lock); |
2071 | cpu_relax(); | |
2072 | goto again; | |
2073 | } | |
13e3c5e5 | 2074 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2075 | dentry_unlink_inode(dentry); |
7a91bf7f | 2076 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2077 | return; |
2078 | } | |
2079 | ||
2080 | if (!d_unhashed(dentry)) | |
2081 | __d_drop(dentry); | |
2082 | ||
2083 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2084 | |
2085 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2086 | } |
ec4f8605 | 2087 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2088 | |
ceb5bdc2 | 2089 | static void __d_rehash(struct dentry * entry, struct dcache_hash_bucket *b) |
1da177e4 | 2090 | { |
ceb5bdc2 NP |
2091 | BUG_ON(!d_unhashed(entry)); |
2092 | spin_lock_bucket(b); | |
1da177e4 | 2093 | entry->d_flags &= ~DCACHE_UNHASHED; |
ceb5bdc2 NP |
2094 | hlist_bl_add_head_rcu(&entry->d_hash, &b->head); |
2095 | spin_unlock_bucket(b); | |
1da177e4 LT |
2096 | } |
2097 | ||
770bfad8 DH |
2098 | static void _d_rehash(struct dentry * entry) |
2099 | { | |
2100 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2101 | } | |
2102 | ||
1da177e4 LT |
2103 | /** |
2104 | * d_rehash - add an entry back to the hash | |
2105 | * @entry: dentry to add to the hash | |
2106 | * | |
2107 | * Adds a dentry to the hash according to its name. | |
2108 | */ | |
2109 | ||
2110 | void d_rehash(struct dentry * entry) | |
2111 | { | |
1da177e4 | 2112 | spin_lock(&entry->d_lock); |
770bfad8 | 2113 | _d_rehash(entry); |
1da177e4 | 2114 | spin_unlock(&entry->d_lock); |
1da177e4 | 2115 | } |
ec4f8605 | 2116 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2117 | |
fb2d5b86 NP |
2118 | /** |
2119 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2120 | * @dentry: dentry to be updated | |
2121 | * @name: new name | |
2122 | * | |
2123 | * Update a case insensitive dentry with new case of name. | |
2124 | * | |
2125 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2126 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2127 | * lengths). | |
2128 | * | |
2129 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2130 | * keep renames and concurrent inserts, and readdir(2) away). | |
2131 | */ | |
2132 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2133 | { | |
2134 | BUG_ON(!mutex_is_locked(&dentry->d_inode->i_mutex)); | |
2135 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ | |
2136 | ||
fb2d5b86 | 2137 | spin_lock(&dentry->d_lock); |
31e6b01f | 2138 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2139 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2140 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2141 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2142 | } |
2143 | EXPORT_SYMBOL(dentry_update_name_case); | |
2144 | ||
1da177e4 LT |
2145 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2146 | { | |
2147 | if (dname_external(target)) { | |
2148 | if (dname_external(dentry)) { | |
2149 | /* | |
2150 | * Both external: swap the pointers | |
2151 | */ | |
9a8d5bb4 | 2152 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2153 | } else { |
2154 | /* | |
2155 | * dentry:internal, target:external. Steal target's | |
2156 | * storage and make target internal. | |
2157 | */ | |
321bcf92 BF |
2158 | memcpy(target->d_iname, dentry->d_name.name, |
2159 | dentry->d_name.len + 1); | |
1da177e4 LT |
2160 | dentry->d_name.name = target->d_name.name; |
2161 | target->d_name.name = target->d_iname; | |
2162 | } | |
2163 | } else { | |
2164 | if (dname_external(dentry)) { | |
2165 | /* | |
2166 | * dentry:external, target:internal. Give dentry's | |
2167 | * storage to target and make dentry internal | |
2168 | */ | |
2169 | memcpy(dentry->d_iname, target->d_name.name, | |
2170 | target->d_name.len + 1); | |
2171 | target->d_name.name = dentry->d_name.name; | |
2172 | dentry->d_name.name = dentry->d_iname; | |
2173 | } else { | |
2174 | /* | |
2175 | * Both are internal. Just copy target to dentry | |
2176 | */ | |
2177 | memcpy(dentry->d_iname, target->d_name.name, | |
2178 | target->d_name.len + 1); | |
dc711ca3 AV |
2179 | dentry->d_name.len = target->d_name.len; |
2180 | return; | |
1da177e4 LT |
2181 | } |
2182 | } | |
9a8d5bb4 | 2183 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2184 | } |
2185 | ||
2fd6b7f5 NP |
2186 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2187 | { | |
2188 | /* | |
2189 | * XXXX: do we really need to take target->d_lock? | |
2190 | */ | |
2191 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2192 | spin_lock(&target->d_parent->d_lock); | |
2193 | else { | |
2194 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2195 | spin_lock(&dentry->d_parent->d_lock); | |
2196 | spin_lock_nested(&target->d_parent->d_lock, | |
2197 | DENTRY_D_LOCK_NESTED); | |
2198 | } else { | |
2199 | spin_lock(&target->d_parent->d_lock); | |
2200 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2201 | DENTRY_D_LOCK_NESTED); | |
2202 | } | |
2203 | } | |
2204 | if (target < dentry) { | |
2205 | spin_lock_nested(&target->d_lock, 2); | |
2206 | spin_lock_nested(&dentry->d_lock, 3); | |
2207 | } else { | |
2208 | spin_lock_nested(&dentry->d_lock, 2); | |
2209 | spin_lock_nested(&target->d_lock, 3); | |
2210 | } | |
2211 | } | |
2212 | ||
2213 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2214 | struct dentry *target) | |
2215 | { | |
2216 | if (target->d_parent != dentry->d_parent) | |
2217 | spin_unlock(&dentry->d_parent->d_lock); | |
2218 | if (target->d_parent != target) | |
2219 | spin_unlock(&target->d_parent->d_lock); | |
2220 | } | |
2221 | ||
1da177e4 | 2222 | /* |
2fd6b7f5 NP |
2223 | * When switching names, the actual string doesn't strictly have to |
2224 | * be preserved in the target - because we're dropping the target | |
2225 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2226 | * the new name before we switch. | |
2227 | * | |
2228 | * Note that we have to be a lot more careful about getting the hash | |
2229 | * switched - we have to switch the hash value properly even if it | |
2230 | * then no longer matches the actual (corrupted) string of the target. | |
2231 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2232 | */ |
9eaef27b | 2233 | /* |
b5c84bf6 | 2234 | * d_move - move a dentry |
1da177e4 LT |
2235 | * @dentry: entry to move |
2236 | * @target: new dentry | |
2237 | * | |
2238 | * Update the dcache to reflect the move of a file name. Negative | |
2239 | * dcache entries should not be moved in this way. | |
2240 | */ | |
b5c84bf6 | 2241 | void d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2242 | { |
1da177e4 LT |
2243 | if (!dentry->d_inode) |
2244 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2245 | ||
2fd6b7f5 NP |
2246 | BUG_ON(d_ancestor(dentry, target)); |
2247 | BUG_ON(d_ancestor(target, dentry)); | |
2248 | ||
1da177e4 | 2249 | write_seqlock(&rename_lock); |
2fd6b7f5 NP |
2250 | |
2251 | dentry_lock_for_move(dentry, target); | |
1da177e4 | 2252 | |
31e6b01f NP |
2253 | write_seqcount_begin(&dentry->d_seq); |
2254 | write_seqcount_begin(&target->d_seq); | |
2255 | ||
ceb5bdc2 NP |
2256 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2257 | ||
2258 | /* | |
2259 | * Move the dentry to the target hash queue. Don't bother checking | |
2260 | * for the same hash queue because of how unlikely it is. | |
2261 | */ | |
2262 | __d_drop(dentry); | |
789680d1 | 2263 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2264 | |
2265 | /* Unhash the target: dput() will then get rid of it */ | |
2266 | __d_drop(target); | |
2267 | ||
5160ee6f ED |
2268 | list_del(&dentry->d_u.d_child); |
2269 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2270 | |
2271 | /* Switch the names.. */ | |
2272 | switch_names(dentry, target); | |
9a8d5bb4 | 2273 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2274 | |
2275 | /* ... and switch the parents */ | |
2276 | if (IS_ROOT(dentry)) { | |
2277 | dentry->d_parent = target->d_parent; | |
2278 | target->d_parent = target; | |
5160ee6f | 2279 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2280 | } else { |
9a8d5bb4 | 2281 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2282 | |
2283 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2284 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2285 | } |
2286 | ||
5160ee6f | 2287 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2288 | |
31e6b01f NP |
2289 | write_seqcount_end(&target->d_seq); |
2290 | write_seqcount_end(&dentry->d_seq); | |
2291 | ||
2fd6b7f5 | 2292 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2293 | spin_unlock(&target->d_lock); |
c32ccd87 | 2294 | fsnotify_d_move(dentry); |
1da177e4 LT |
2295 | spin_unlock(&dentry->d_lock); |
2296 | write_sequnlock(&rename_lock); | |
9eaef27b | 2297 | } |
ec4f8605 | 2298 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2299 | |
e2761a11 OH |
2300 | /** |
2301 | * d_ancestor - search for an ancestor | |
2302 | * @p1: ancestor dentry | |
2303 | * @p2: child dentry | |
2304 | * | |
2305 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2306 | * an ancestor of p2, else NULL. | |
9eaef27b | 2307 | */ |
e2761a11 | 2308 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2309 | { |
2310 | struct dentry *p; | |
2311 | ||
871c0067 | 2312 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2313 | if (p->d_parent == p1) |
e2761a11 | 2314 | return p; |
9eaef27b | 2315 | } |
e2761a11 | 2316 | return NULL; |
9eaef27b TM |
2317 | } |
2318 | ||
2319 | /* | |
2320 | * This helper attempts to cope with remotely renamed directories | |
2321 | * | |
2322 | * It assumes that the caller is already holding | |
873feea0 | 2323 | * dentry->d_parent->d_inode->i_mutex and the inode->i_lock |
9eaef27b TM |
2324 | * |
2325 | * Note: If ever the locking in lock_rename() changes, then please | |
2326 | * remember to update this too... | |
9eaef27b | 2327 | */ |
873feea0 NP |
2328 | static struct dentry *__d_unalias(struct inode *inode, |
2329 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2330 | { |
2331 | struct mutex *m1 = NULL, *m2 = NULL; | |
2332 | struct dentry *ret; | |
2333 | ||
2334 | /* If alias and dentry share a parent, then no extra locks required */ | |
2335 | if (alias->d_parent == dentry->d_parent) | |
2336 | goto out_unalias; | |
2337 | ||
2338 | /* Check for loops */ | |
2339 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 2340 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
2341 | goto out_err; |
2342 | ||
2343 | /* See lock_rename() */ | |
2344 | ret = ERR_PTR(-EBUSY); | |
2345 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2346 | goto out_err; | |
2347 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2348 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2349 | goto out_err; | |
2350 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2351 | out_unalias: | |
b5c84bf6 | 2352 | d_move(alias, dentry); |
9eaef27b TM |
2353 | ret = alias; |
2354 | out_err: | |
873feea0 | 2355 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2356 | if (m2) |
2357 | mutex_unlock(m2); | |
2358 | if (m1) | |
2359 | mutex_unlock(m1); | |
2360 | return ret; | |
2361 | } | |
2362 | ||
770bfad8 DH |
2363 | /* |
2364 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2365 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2366 | * returns with anon->d_lock held! |
770bfad8 DH |
2367 | */ |
2368 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2369 | { | |
2370 | struct dentry *dparent, *aparent; | |
2371 | ||
2fd6b7f5 | 2372 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2373 | |
31e6b01f NP |
2374 | write_seqcount_begin(&dentry->d_seq); |
2375 | write_seqcount_begin(&anon->d_seq); | |
2376 | ||
770bfad8 DH |
2377 | dparent = dentry->d_parent; |
2378 | aparent = anon->d_parent; | |
2379 | ||
2fd6b7f5 NP |
2380 | switch_names(dentry, anon); |
2381 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2382 | ||
770bfad8 DH |
2383 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2384 | list_del(&dentry->d_u.d_child); | |
2385 | if (!IS_ROOT(dentry)) | |
2386 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2387 | else | |
2388 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2389 | ||
2390 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2391 | list_del(&anon->d_u.d_child); | |
2392 | if (!IS_ROOT(anon)) | |
2393 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2394 | else | |
2395 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2396 | ||
31e6b01f NP |
2397 | write_seqcount_end(&dentry->d_seq); |
2398 | write_seqcount_end(&anon->d_seq); | |
2399 | ||
2fd6b7f5 NP |
2400 | dentry_unlock_parents_for_move(anon, dentry); |
2401 | spin_unlock(&dentry->d_lock); | |
2402 | ||
2403 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2404 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2405 | } | |
2406 | ||
2407 | /** | |
2408 | * d_materialise_unique - introduce an inode into the tree | |
2409 | * @dentry: candidate dentry | |
2410 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2411 | * | |
2412 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2413 | * root directory alias in its place if there is one | |
2414 | */ | |
2415 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2416 | { | |
9eaef27b | 2417 | struct dentry *actual; |
770bfad8 DH |
2418 | |
2419 | BUG_ON(!d_unhashed(dentry)); | |
2420 | ||
770bfad8 DH |
2421 | if (!inode) { |
2422 | actual = dentry; | |
360da900 | 2423 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2424 | d_rehash(actual); |
2425 | goto out_nolock; | |
770bfad8 DH |
2426 | } |
2427 | ||
873feea0 | 2428 | spin_lock(&inode->i_lock); |
357f8e65 | 2429 | |
9eaef27b TM |
2430 | if (S_ISDIR(inode->i_mode)) { |
2431 | struct dentry *alias; | |
2432 | ||
2433 | /* Does an aliased dentry already exist? */ | |
2434 | alias = __d_find_alias(inode, 0); | |
2435 | if (alias) { | |
2436 | actual = alias; | |
2437 | /* Is this an anonymous mountpoint that we could splice | |
2438 | * into our tree? */ | |
2439 | if (IS_ROOT(alias)) { | |
9eaef27b TM |
2440 | __d_materialise_dentry(dentry, alias); |
2441 | __d_drop(alias); | |
2442 | goto found; | |
2443 | } | |
2444 | /* Nope, but we must(!) avoid directory aliasing */ | |
873feea0 | 2445 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b TM |
2446 | if (IS_ERR(actual)) |
2447 | dput(alias); | |
2448 | goto out_nolock; | |
2449 | } | |
770bfad8 DH |
2450 | } |
2451 | ||
2452 | /* Add a unique reference */ | |
2453 | actual = __d_instantiate_unique(dentry, inode); | |
2454 | if (!actual) | |
2455 | actual = dentry; | |
357f8e65 NP |
2456 | else |
2457 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2458 | |
770bfad8 DH |
2459 | spin_lock(&actual->d_lock); |
2460 | found: | |
2461 | _d_rehash(actual); | |
2462 | spin_unlock(&actual->d_lock); | |
873feea0 | 2463 | spin_unlock(&inode->i_lock); |
9eaef27b | 2464 | out_nolock: |
770bfad8 DH |
2465 | if (actual == dentry) { |
2466 | security_d_instantiate(dentry, inode); | |
2467 | return NULL; | |
2468 | } | |
2469 | ||
2470 | iput(inode); | |
2471 | return actual; | |
770bfad8 | 2472 | } |
ec4f8605 | 2473 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2474 | |
cdd16d02 | 2475 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2476 | { |
2477 | *buflen -= namelen; | |
2478 | if (*buflen < 0) | |
2479 | return -ENAMETOOLONG; | |
2480 | *buffer -= namelen; | |
2481 | memcpy(*buffer, str, namelen); | |
2482 | return 0; | |
2483 | } | |
2484 | ||
cdd16d02 MS |
2485 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2486 | { | |
2487 | return prepend(buffer, buflen, name->name, name->len); | |
2488 | } | |
2489 | ||
1da177e4 | 2490 | /** |
208898c1 | 2491 | * prepend_path - Prepend path string to a buffer |
9d1bc601 MS |
2492 | * @path: the dentry/vfsmount to report |
2493 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2494 | * @buffer: pointer to the end of the buffer |
2495 | * @buflen: pointer to buffer length | |
552ce544 | 2496 | * |
949854d0 | 2497 | * Caller holds the rename_lock. |
9d1bc601 MS |
2498 | * |
2499 | * If path is not reachable from the supplied root, then the value of | |
2500 | * root is changed (without modifying refcounts). | |
1da177e4 | 2501 | */ |
f2eb6575 MS |
2502 | static int prepend_path(const struct path *path, struct path *root, |
2503 | char **buffer, int *buflen) | |
1da177e4 | 2504 | { |
9d1bc601 MS |
2505 | struct dentry *dentry = path->dentry; |
2506 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2507 | bool slash = false; |
2508 | int error = 0; | |
6092d048 | 2509 | |
99b7db7b | 2510 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2511 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2512 | struct dentry * parent; |
2513 | ||
1da177e4 | 2514 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2515 | /* Global root? */ |
1da177e4 | 2516 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2517 | goto global_root; |
2518 | } | |
2519 | dentry = vfsmnt->mnt_mountpoint; | |
2520 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2521 | continue; |
2522 | } | |
2523 | parent = dentry->d_parent; | |
2524 | prefetch(parent); | |
9abca360 | 2525 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2526 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2527 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2528 | if (!error) |
2529 | error = prepend(buffer, buflen, "/", 1); | |
2530 | if (error) | |
2531 | break; | |
2532 | ||
2533 | slash = true; | |
1da177e4 LT |
2534 | dentry = parent; |
2535 | } | |
2536 | ||
be285c71 | 2537 | out: |
f2eb6575 MS |
2538 | if (!error && !slash) |
2539 | error = prepend(buffer, buflen, "/", 1); | |
2540 | ||
99b7db7b | 2541 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2542 | return error; |
1da177e4 LT |
2543 | |
2544 | global_root: | |
98dc568b MS |
2545 | /* |
2546 | * Filesystems needing to implement special "root names" | |
2547 | * should do so with ->d_dname() | |
2548 | */ | |
2549 | if (IS_ROOT(dentry) && | |
2550 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2551 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2552 | (int) dentry->d_name.len, dentry->d_name.name); | |
2553 | } | |
9d1bc601 MS |
2554 | root->mnt = vfsmnt; |
2555 | root->dentry = dentry; | |
be285c71 | 2556 | goto out; |
f2eb6575 | 2557 | } |
be285c71 | 2558 | |
f2eb6575 MS |
2559 | /** |
2560 | * __d_path - return the path of a dentry | |
2561 | * @path: the dentry/vfsmount to report | |
2562 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2563 | * @buf: buffer to return value in |
f2eb6575 MS |
2564 | * @buflen: buffer length |
2565 | * | |
ffd1f4ed | 2566 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2567 | * |
2568 | * Returns a pointer into the buffer or an error code if the | |
2569 | * path was too long. | |
2570 | * | |
be148247 | 2571 | * "buflen" should be positive. |
f2eb6575 MS |
2572 | * |
2573 | * If path is not reachable from the supplied root, then the value of | |
2574 | * root is changed (without modifying refcounts). | |
2575 | */ | |
2576 | char *__d_path(const struct path *path, struct path *root, | |
2577 | char *buf, int buflen) | |
2578 | { | |
2579 | char *res = buf + buflen; | |
2580 | int error; | |
2581 | ||
2582 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2583 | write_seqlock(&rename_lock); |
f2eb6575 | 2584 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2585 | write_sequnlock(&rename_lock); |
be148247 | 2586 | |
f2eb6575 MS |
2587 | if (error) |
2588 | return ERR_PTR(error); | |
f2eb6575 | 2589 | return res; |
1da177e4 LT |
2590 | } |
2591 | ||
ffd1f4ed MS |
2592 | /* |
2593 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2594 | */ | |
2595 | static int path_with_deleted(const struct path *path, struct path *root, | |
2596 | char **buf, int *buflen) | |
2597 | { | |
2598 | prepend(buf, buflen, "\0", 1); | |
2599 | if (d_unlinked(path->dentry)) { | |
2600 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2601 | if (error) | |
2602 | return error; | |
2603 | } | |
2604 | ||
2605 | return prepend_path(path, root, buf, buflen); | |
2606 | } | |
2607 | ||
8df9d1a4 MS |
2608 | static int prepend_unreachable(char **buffer, int *buflen) |
2609 | { | |
2610 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2611 | } | |
2612 | ||
a03a8a70 JB |
2613 | /** |
2614 | * d_path - return the path of a dentry | |
cf28b486 | 2615 | * @path: path to report |
a03a8a70 JB |
2616 | * @buf: buffer to return value in |
2617 | * @buflen: buffer length | |
2618 | * | |
2619 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2620 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2621 | * | |
52afeefb AV |
2622 | * Returns a pointer into the buffer or an error code if the path was |
2623 | * too long. Note: Callers should use the returned pointer, not the passed | |
2624 | * in buffer, to use the name! The implementation often starts at an offset | |
2625 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2626 | * |
31f3e0b3 | 2627 | * "buflen" should be positive. |
a03a8a70 | 2628 | */ |
20d4fdc1 | 2629 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2630 | { |
ffd1f4ed | 2631 | char *res = buf + buflen; |
6ac08c39 | 2632 | struct path root; |
9d1bc601 | 2633 | struct path tmp; |
ffd1f4ed | 2634 | int error; |
1da177e4 | 2635 | |
c23fbb6b ED |
2636 | /* |
2637 | * We have various synthetic filesystems that never get mounted. On | |
2638 | * these filesystems dentries are never used for lookup purposes, and | |
2639 | * thus don't need to be hashed. They also don't need a name until a | |
2640 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2641 | * below allows us to generate a name for these objects on demand: | |
2642 | */ | |
cf28b486 JB |
2643 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2644 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2645 | |
f7ad3c6b | 2646 | get_fs_root(current->fs, &root); |
949854d0 | 2647 | write_seqlock(&rename_lock); |
9d1bc601 | 2648 | tmp = root; |
ffd1f4ed MS |
2649 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2650 | if (error) | |
2651 | res = ERR_PTR(error); | |
949854d0 | 2652 | write_sequnlock(&rename_lock); |
6ac08c39 | 2653 | path_put(&root); |
1da177e4 LT |
2654 | return res; |
2655 | } | |
ec4f8605 | 2656 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2657 | |
8df9d1a4 MS |
2658 | /** |
2659 | * d_path_with_unreachable - return the path of a dentry | |
2660 | * @path: path to report | |
2661 | * @buf: buffer to return value in | |
2662 | * @buflen: buffer length | |
2663 | * | |
2664 | * The difference from d_path() is that this prepends "(unreachable)" | |
2665 | * to paths which are unreachable from the current process' root. | |
2666 | */ | |
2667 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2668 | { | |
2669 | char *res = buf + buflen; | |
2670 | struct path root; | |
2671 | struct path tmp; | |
2672 | int error; | |
2673 | ||
2674 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2675 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2676 | ||
2677 | get_fs_root(current->fs, &root); | |
949854d0 | 2678 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2679 | tmp = root; |
2680 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2681 | if (!error && !path_equal(&tmp, &root)) | |
2682 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2683 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2684 | path_put(&root); |
2685 | if (error) | |
2686 | res = ERR_PTR(error); | |
2687 | ||
2688 | return res; | |
2689 | } | |
2690 | ||
c23fbb6b ED |
2691 | /* |
2692 | * Helper function for dentry_operations.d_dname() members | |
2693 | */ | |
2694 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2695 | const char *fmt, ...) | |
2696 | { | |
2697 | va_list args; | |
2698 | char temp[64]; | |
2699 | int sz; | |
2700 | ||
2701 | va_start(args, fmt); | |
2702 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2703 | va_end(args); | |
2704 | ||
2705 | if (sz > sizeof(temp) || sz > buflen) | |
2706 | return ERR_PTR(-ENAMETOOLONG); | |
2707 | ||
2708 | buffer += buflen - sz; | |
2709 | return memcpy(buffer, temp, sz); | |
2710 | } | |
2711 | ||
6092d048 RP |
2712 | /* |
2713 | * Write full pathname from the root of the filesystem into the buffer. | |
2714 | */ | |
ec2447c2 | 2715 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2716 | { |
2717 | char *end = buf + buflen; | |
2718 | char *retval; | |
2719 | ||
6092d048 | 2720 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2721 | if (buflen < 1) |
2722 | goto Elong; | |
2723 | /* Get '/' right */ | |
2724 | retval = end-1; | |
2725 | *retval = '/'; | |
2726 | ||
cdd16d02 MS |
2727 | while (!IS_ROOT(dentry)) { |
2728 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2729 | int error; |
6092d048 | 2730 | |
6092d048 | 2731 | prefetch(parent); |
9abca360 NP |
2732 | spin_lock(&dentry->d_lock); |
2733 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2734 | spin_unlock(&dentry->d_lock); | |
2735 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2736 | goto Elong; |
2737 | ||
2738 | retval = end; | |
2739 | dentry = parent; | |
2740 | } | |
c103135c AV |
2741 | return retval; |
2742 | Elong: | |
2743 | return ERR_PTR(-ENAMETOOLONG); | |
2744 | } | |
ec2447c2 NP |
2745 | |
2746 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2747 | { | |
2748 | char *retval; | |
2749 | ||
949854d0 | 2750 | write_seqlock(&rename_lock); |
ec2447c2 | 2751 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2752 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2753 | |
2754 | return retval; | |
2755 | } | |
2756 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2757 | |
2758 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2759 | { | |
2760 | char *p = NULL; | |
2761 | char *retval; | |
2762 | ||
949854d0 | 2763 | write_seqlock(&rename_lock); |
c103135c AV |
2764 | if (d_unlinked(dentry)) { |
2765 | p = buf + buflen; | |
2766 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2767 | goto Elong; | |
2768 | buflen++; | |
2769 | } | |
2770 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2771 | write_sequnlock(&rename_lock); |
c103135c AV |
2772 | if (!IS_ERR(retval) && p) |
2773 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2774 | return retval; |
2775 | Elong: | |
6092d048 RP |
2776 | return ERR_PTR(-ENAMETOOLONG); |
2777 | } | |
2778 | ||
1da177e4 LT |
2779 | /* |
2780 | * NOTE! The user-level library version returns a | |
2781 | * character pointer. The kernel system call just | |
2782 | * returns the length of the buffer filled (which | |
2783 | * includes the ending '\0' character), or a negative | |
2784 | * error value. So libc would do something like | |
2785 | * | |
2786 | * char *getcwd(char * buf, size_t size) | |
2787 | * { | |
2788 | * int retval; | |
2789 | * | |
2790 | * retval = sys_getcwd(buf, size); | |
2791 | * if (retval >= 0) | |
2792 | * return buf; | |
2793 | * errno = -retval; | |
2794 | * return NULL; | |
2795 | * } | |
2796 | */ | |
3cdad428 | 2797 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2798 | { |
552ce544 | 2799 | int error; |
6ac08c39 | 2800 | struct path pwd, root; |
552ce544 | 2801 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2802 | |
2803 | if (!page) | |
2804 | return -ENOMEM; | |
2805 | ||
f7ad3c6b | 2806 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2807 | |
552ce544 | 2808 | error = -ENOENT; |
949854d0 | 2809 | write_seqlock(&rename_lock); |
f3da392e | 2810 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2811 | unsigned long len; |
9d1bc601 | 2812 | struct path tmp = root; |
8df9d1a4 MS |
2813 | char *cwd = page + PAGE_SIZE; |
2814 | int buflen = PAGE_SIZE; | |
1da177e4 | 2815 | |
8df9d1a4 MS |
2816 | prepend(&cwd, &buflen, "\0", 1); |
2817 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2818 | write_sequnlock(&rename_lock); |
552ce544 | 2819 | |
8df9d1a4 | 2820 | if (error) |
552ce544 LT |
2821 | goto out; |
2822 | ||
8df9d1a4 MS |
2823 | /* Unreachable from current root */ |
2824 | if (!path_equal(&tmp, &root)) { | |
2825 | error = prepend_unreachable(&cwd, &buflen); | |
2826 | if (error) | |
2827 | goto out; | |
2828 | } | |
2829 | ||
552ce544 LT |
2830 | error = -ERANGE; |
2831 | len = PAGE_SIZE + page - cwd; | |
2832 | if (len <= size) { | |
2833 | error = len; | |
2834 | if (copy_to_user(buf, cwd, len)) | |
2835 | error = -EFAULT; | |
2836 | } | |
949854d0 NP |
2837 | } else { |
2838 | write_sequnlock(&rename_lock); | |
949854d0 | 2839 | } |
1da177e4 LT |
2840 | |
2841 | out: | |
6ac08c39 JB |
2842 | path_put(&pwd); |
2843 | path_put(&root); | |
1da177e4 LT |
2844 | free_page((unsigned long) page); |
2845 | return error; | |
2846 | } | |
2847 | ||
2848 | /* | |
2849 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2850 | * | |
2851 | * Trivially implemented using the dcache structure | |
2852 | */ | |
2853 | ||
2854 | /** | |
2855 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2856 | * @new_dentry: new dentry | |
2857 | * @old_dentry: old dentry | |
2858 | * | |
2859 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2860 | * Returns 0 otherwise. | |
2861 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2862 | */ | |
2863 | ||
e2761a11 | 2864 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2865 | { |
2866 | int result; | |
949854d0 | 2867 | unsigned seq; |
1da177e4 | 2868 | |
e2761a11 OH |
2869 | if (new_dentry == old_dentry) |
2870 | return 1; | |
2871 | ||
e2761a11 | 2872 | do { |
1da177e4 | 2873 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2874 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2875 | /* |
2876 | * Need rcu_readlock to protect against the d_parent trashing | |
2877 | * due to d_move | |
2878 | */ | |
2879 | rcu_read_lock(); | |
e2761a11 | 2880 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2881 | result = 1; |
e2761a11 OH |
2882 | else |
2883 | result = 0; | |
949854d0 | 2884 | rcu_read_unlock(); |
1da177e4 | 2885 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2886 | |
2887 | return result; | |
2888 | } | |
2889 | ||
2096f759 AV |
2890 | int path_is_under(struct path *path1, struct path *path2) |
2891 | { | |
2892 | struct vfsmount *mnt = path1->mnt; | |
2893 | struct dentry *dentry = path1->dentry; | |
2894 | int res; | |
99b7db7b NP |
2895 | |
2896 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2897 | if (mnt != path2->mnt) { |
2898 | for (;;) { | |
2899 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2900 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2901 | return 0; |
2902 | } | |
2903 | if (mnt->mnt_parent == path2->mnt) | |
2904 | break; | |
2905 | mnt = mnt->mnt_parent; | |
2906 | } | |
2907 | dentry = mnt->mnt_mountpoint; | |
2908 | } | |
2909 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2910 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2911 | return res; |
2912 | } | |
2913 | EXPORT_SYMBOL(path_is_under); | |
2914 | ||
1da177e4 LT |
2915 | void d_genocide(struct dentry *root) |
2916 | { | |
949854d0 | 2917 | struct dentry *this_parent; |
1da177e4 | 2918 | struct list_head *next; |
949854d0 | 2919 | unsigned seq; |
58db63d0 | 2920 | int locked = 0; |
1da177e4 | 2921 | |
949854d0 | 2922 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2923 | again: |
2924 | this_parent = root; | |
2fd6b7f5 | 2925 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2926 | repeat: |
2927 | next = this_parent->d_subdirs.next; | |
2928 | resume: | |
2929 | while (next != &this_parent->d_subdirs) { | |
2930 | struct list_head *tmp = next; | |
5160ee6f | 2931 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2932 | next = tmp->next; |
949854d0 | 2933 | |
da502956 NP |
2934 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2935 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2936 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2937 | continue; |
da502956 | 2938 | } |
1da177e4 | 2939 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2940 | spin_unlock(&this_parent->d_lock); |
2941 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2942 | this_parent = dentry; |
2fd6b7f5 | 2943 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2944 | goto repeat; |
2945 | } | |
949854d0 NP |
2946 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2947 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2948 | dentry->d_count--; | |
2949 | } | |
b7ab39f6 | 2950 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2951 | } |
2952 | if (this_parent != root) { | |
c826cb7d | 2953 | struct dentry *child = this_parent; |
949854d0 NP |
2954 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2955 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2956 | this_parent->d_count--; | |
2957 | } | |
c826cb7d LT |
2958 | this_parent = try_to_ascend(this_parent, locked, seq); |
2959 | if (!this_parent) | |
949854d0 | 2960 | goto rename_retry; |
949854d0 | 2961 | next = child->d_u.d_child.next; |
1da177e4 LT |
2962 | goto resume; |
2963 | } | |
2fd6b7f5 | 2964 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2965 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2966 | goto rename_retry; |
58db63d0 NP |
2967 | if (locked) |
2968 | write_sequnlock(&rename_lock); | |
2969 | return; | |
2970 | ||
2971 | rename_retry: | |
2972 | locked = 1; | |
2973 | write_seqlock(&rename_lock); | |
2974 | goto again; | |
1da177e4 LT |
2975 | } |
2976 | ||
2977 | /** | |
2978 | * find_inode_number - check for dentry with name | |
2979 | * @dir: directory to check | |
2980 | * @name: Name to find. | |
2981 | * | |
2982 | * Check whether a dentry already exists for the given name, | |
2983 | * and return the inode number if it has an inode. Otherwise | |
2984 | * 0 is returned. | |
2985 | * | |
2986 | * This routine is used to post-process directory listings for | |
2987 | * filesystems using synthetic inode numbers, and is necessary | |
2988 | * to keep getcwd() working. | |
2989 | */ | |
2990 | ||
2991 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2992 | { | |
2993 | struct dentry * dentry; | |
2994 | ino_t ino = 0; | |
2995 | ||
3e7e241f EB |
2996 | dentry = d_hash_and_lookup(dir, name); |
2997 | if (dentry) { | |
1da177e4 LT |
2998 | if (dentry->d_inode) |
2999 | ino = dentry->d_inode->i_ino; | |
3000 | dput(dentry); | |
3001 | } | |
1da177e4 LT |
3002 | return ino; |
3003 | } | |
ec4f8605 | 3004 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
3005 | |
3006 | static __initdata unsigned long dhash_entries; | |
3007 | static int __init set_dhash_entries(char *str) | |
3008 | { | |
3009 | if (!str) | |
3010 | return 0; | |
3011 | dhash_entries = simple_strtoul(str, &str, 0); | |
3012 | return 1; | |
3013 | } | |
3014 | __setup("dhash_entries=", set_dhash_entries); | |
3015 | ||
3016 | static void __init dcache_init_early(void) | |
3017 | { | |
3018 | int loop; | |
3019 | ||
3020 | /* If hashes are distributed across NUMA nodes, defer | |
3021 | * hash allocation until vmalloc space is available. | |
3022 | */ | |
3023 | if (hashdist) | |
3024 | return; | |
3025 | ||
3026 | dentry_hashtable = | |
3027 | alloc_large_system_hash("Dentry cache", | |
ceb5bdc2 | 3028 | sizeof(struct dcache_hash_bucket), |
1da177e4 LT |
3029 | dhash_entries, |
3030 | 13, | |
3031 | HASH_EARLY, | |
3032 | &d_hash_shift, | |
3033 | &d_hash_mask, | |
3034 | 0); | |
3035 | ||
3036 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
ceb5bdc2 | 3037 | INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head); |
1da177e4 LT |
3038 | } |
3039 | ||
74bf17cf | 3040 | static void __init dcache_init(void) |
1da177e4 LT |
3041 | { |
3042 | int loop; | |
3043 | ||
3044 | /* | |
3045 | * A constructor could be added for stable state like the lists, | |
3046 | * but it is probably not worth it because of the cache nature | |
3047 | * of the dcache. | |
3048 | */ | |
0a31bd5f CL |
3049 | dentry_cache = KMEM_CACHE(dentry, |
3050 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 3051 | |
8e1f936b | 3052 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
3053 | |
3054 | /* Hash may have been set up in dcache_init_early */ | |
3055 | if (!hashdist) | |
3056 | return; | |
3057 | ||
3058 | dentry_hashtable = | |
3059 | alloc_large_system_hash("Dentry cache", | |
ceb5bdc2 | 3060 | sizeof(struct dcache_hash_bucket), |
1da177e4 LT |
3061 | dhash_entries, |
3062 | 13, | |
3063 | 0, | |
3064 | &d_hash_shift, | |
3065 | &d_hash_mask, | |
3066 | 0); | |
3067 | ||
3068 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
ceb5bdc2 | 3069 | INIT_HLIST_BL_HEAD(&dentry_hashtable[loop].head); |
1da177e4 LT |
3070 | } |
3071 | ||
3072 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3073 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3074 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3075 | |
1da177e4 LT |
3076 | EXPORT_SYMBOL(d_genocide); |
3077 | ||
1da177e4 LT |
3078 | void __init vfs_caches_init_early(void) |
3079 | { | |
3080 | dcache_init_early(); | |
3081 | inode_init_early(); | |
3082 | } | |
3083 | ||
3084 | void __init vfs_caches_init(unsigned long mempages) | |
3085 | { | |
3086 | unsigned long reserve; | |
3087 | ||
3088 | /* Base hash sizes on available memory, with a reserve equal to | |
3089 | 150% of current kernel size */ | |
3090 | ||
3091 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3092 | mempages -= reserve; | |
3093 | ||
3094 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3095 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3096 | |
74bf17cf DC |
3097 | dcache_init(); |
3098 | inode_init(); | |
1da177e4 | 3099 | files_init(mempages); |
74bf17cf | 3100 | mnt_init(); |
1da177e4 LT |
3101 | bdev_cache_init(); |
3102 | chrdev_init(); | |
3103 | } |