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 | 101 | |
b07ad996 | 102 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 103 | |
b07ad996 | 104 | static inline struct hlist_bl_head *d_hash(struct dentry *parent, |
ceb5bdc2 NP |
105 | unsigned long hash) |
106 | { | |
107 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
108 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
109 | return dentry_hashtable + (hash & D_HASHMASK); | |
110 | } | |
111 | ||
b07ad996 | 112 | static inline void spin_lock_bucket(struct hlist_bl_head *b) |
ceb5bdc2 | 113 | { |
b07ad996 | 114 | bit_spin_lock(0, (unsigned long *)&b->first); |
ceb5bdc2 NP |
115 | } |
116 | ||
b07ad996 | 117 | static inline void spin_unlock_bucket(struct hlist_bl_head *b) |
ceb5bdc2 | 118 | { |
b07ad996 | 119 | __bit_spin_unlock(0, (unsigned long *)&b->first); |
ceb5bdc2 | 120 | } |
1da177e4 LT |
121 | |
122 | /* Statistics gathering. */ | |
123 | struct dentry_stat_t dentry_stat = { | |
124 | .age_limit = 45, | |
125 | }; | |
126 | ||
3e880fb5 | 127 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
128 | |
129 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
130 | static int get_nr_dentry(void) |
131 | { | |
132 | int i; | |
133 | int sum = 0; | |
134 | for_each_possible_cpu(i) | |
135 | sum += per_cpu(nr_dentry, i); | |
136 | return sum < 0 ? 0 : sum; | |
137 | } | |
138 | ||
312d3ca8 CH |
139 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
140 | size_t *lenp, loff_t *ppos) | |
141 | { | |
3e880fb5 | 142 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
143 | return proc_dointvec(table, write, buffer, lenp, ppos); |
144 | } | |
145 | #endif | |
146 | ||
9c82ab9c | 147 | static void __d_free(struct rcu_head *head) |
1da177e4 | 148 | { |
9c82ab9c CH |
149 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
150 | ||
fd217f4d | 151 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
152 | if (dname_external(dentry)) |
153 | kfree(dentry->d_name.name); | |
154 | kmem_cache_free(dentry_cache, dentry); | |
155 | } | |
156 | ||
157 | /* | |
b5c84bf6 | 158 | * no locks, please. |
1da177e4 LT |
159 | */ |
160 | static void d_free(struct dentry *dentry) | |
161 | { | |
b7ab39f6 | 162 | BUG_ON(dentry->d_count); |
3e880fb5 | 163 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
164 | if (dentry->d_op && dentry->d_op->d_release) |
165 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 166 | |
dea3667b LT |
167 | /* if dentry was never visible to RCU, immediate free is OK */ |
168 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
9c82ab9c | 169 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 170 | else |
9c82ab9c | 171 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
172 | } |
173 | ||
31e6b01f NP |
174 | /** |
175 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 176 | * @dentry: the target dentry |
31e6b01f NP |
177 | * After this call, in-progress rcu-walk path lookup will fail. This |
178 | * should be called after unhashing, and after changing d_inode (if | |
179 | * the dentry has not already been unhashed). | |
180 | */ | |
181 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
182 | { | |
183 | assert_spin_locked(&dentry->d_lock); | |
184 | /* Go through a barrier */ | |
185 | write_seqcount_barrier(&dentry->d_seq); | |
186 | } | |
187 | ||
1da177e4 LT |
188 | /* |
189 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
190 | * d_iput() operation if defined. Dentry has no refcount |
191 | * and is unhashed. | |
1da177e4 | 192 | */ |
858119e1 | 193 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 194 | __releases(dentry->d_lock) |
873feea0 | 195 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
196 | { |
197 | struct inode *inode = dentry->d_inode; | |
198 | if (inode) { | |
199 | dentry->d_inode = NULL; | |
200 | list_del_init(&dentry->d_alias); | |
201 | spin_unlock(&dentry->d_lock); | |
873feea0 | 202 | spin_unlock(&inode->i_lock); |
f805fbda LT |
203 | if (!inode->i_nlink) |
204 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
205 | if (dentry->d_op && dentry->d_op->d_iput) |
206 | dentry->d_op->d_iput(dentry, inode); | |
207 | else | |
208 | iput(inode); | |
209 | } else { | |
210 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
211 | } |
212 | } | |
213 | ||
31e6b01f NP |
214 | /* |
215 | * Release the dentry's inode, using the filesystem | |
216 | * d_iput() operation if defined. dentry remains in-use. | |
217 | */ | |
218 | static void dentry_unlink_inode(struct dentry * dentry) | |
219 | __releases(dentry->d_lock) | |
873feea0 | 220 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
221 | { |
222 | struct inode *inode = dentry->d_inode; | |
223 | dentry->d_inode = NULL; | |
224 | list_del_init(&dentry->d_alias); | |
225 | dentry_rcuwalk_barrier(dentry); | |
226 | spin_unlock(&dentry->d_lock); | |
873feea0 | 227 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
228 | if (!inode->i_nlink) |
229 | fsnotify_inoderemove(inode); | |
230 | if (dentry->d_op && dentry->d_op->d_iput) | |
231 | dentry->d_op->d_iput(dentry, inode); | |
232 | else | |
233 | iput(inode); | |
234 | } | |
235 | ||
da3bbdd4 | 236 | /* |
23044507 | 237 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
238 | */ |
239 | static void dentry_lru_add(struct dentry *dentry) | |
240 | { | |
a4633357 | 241 | if (list_empty(&dentry->d_lru)) { |
23044507 | 242 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
243 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
244 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 245 | dentry_stat.nr_unused++; |
23044507 | 246 | spin_unlock(&dcache_lru_lock); |
a4633357 | 247 | } |
da3bbdd4 KM |
248 | } |
249 | ||
23044507 NP |
250 | static void __dentry_lru_del(struct dentry *dentry) |
251 | { | |
252 | list_del_init(&dentry->d_lru); | |
253 | dentry->d_sb->s_nr_dentry_unused--; | |
254 | dentry_stat.nr_unused--; | |
255 | } | |
256 | ||
da3bbdd4 KM |
257 | static void dentry_lru_del(struct dentry *dentry) |
258 | { | |
259 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
260 | spin_lock(&dcache_lru_lock); |
261 | __dentry_lru_del(dentry); | |
262 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
263 | } |
264 | } | |
265 | ||
a4633357 | 266 | static void dentry_lru_move_tail(struct dentry *dentry) |
da3bbdd4 | 267 | { |
23044507 | 268 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
269 | if (list_empty(&dentry->d_lru)) { |
270 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
271 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 272 | dentry_stat.nr_unused++; |
a4633357 CH |
273 | } else { |
274 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
da3bbdd4 | 275 | } |
23044507 | 276 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
277 | } |
278 | ||
d52b9086 MS |
279 | /** |
280 | * d_kill - kill dentry and return parent | |
281 | * @dentry: dentry to kill | |
ff5fdb61 | 282 | * @parent: parent dentry |
d52b9086 | 283 | * |
31f3e0b3 | 284 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
285 | * |
286 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 287 | * |
b5c84bf6 NP |
288 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
289 | * d_kill. | |
d52b9086 | 290 | */ |
2fd6b7f5 | 291 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 292 | __releases(dentry->d_lock) |
2fd6b7f5 | 293 | __releases(parent->d_lock) |
873feea0 | 294 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 295 | { |
d52b9086 | 296 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
297 | /* |
298 | * Inform try_to_ascend() that we are no longer attached to the | |
299 | * dentry tree | |
300 | */ | |
301 | dentry->d_flags |= DCACHE_DISCONNECTED; | |
2fd6b7f5 NP |
302 | if (parent) |
303 | spin_unlock(&parent->d_lock); | |
d52b9086 | 304 | dentry_iput(dentry); |
b7ab39f6 NP |
305 | /* |
306 | * dentry_iput drops the locks, at which point nobody (except | |
307 | * transient RCU lookups) can reach this dentry. | |
308 | */ | |
d52b9086 | 309 | d_free(dentry); |
871c0067 | 310 | return parent; |
d52b9086 MS |
311 | } |
312 | ||
789680d1 NP |
313 | /** |
314 | * d_drop - drop a dentry | |
315 | * @dentry: dentry to drop | |
316 | * | |
317 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
318 | * be found through a VFS lookup any more. Note that this is different from | |
319 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
320 | * possible, giving a successful _negative_ lookup, while d_drop will | |
321 | * just make the cache lookup fail. | |
322 | * | |
323 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
324 | * reason (NFS timeouts or autofs deletes). | |
325 | * | |
326 | * __d_drop requires dentry->d_lock. | |
327 | */ | |
328 | void __d_drop(struct dentry *dentry) | |
329 | { | |
dea3667b | 330 | if (!d_unhashed(dentry)) { |
b07ad996 | 331 | struct hlist_bl_head *b; |
dea3667b | 332 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
b07ad996 | 333 | b = &dentry->d_sb->s_anon; |
dea3667b | 334 | else |
ceb5bdc2 | 335 | b = d_hash(dentry->d_parent, dentry->d_name.hash); |
dea3667b LT |
336 | |
337 | spin_lock_bucket(b); | |
338 | __hlist_bl_del(&dentry->d_hash); | |
339 | dentry->d_hash.pprev = NULL; | |
340 | spin_unlock_bucket(b); | |
341 | ||
342 | dentry_rcuwalk_barrier(dentry); | |
789680d1 NP |
343 | } |
344 | } | |
345 | EXPORT_SYMBOL(__d_drop); | |
346 | ||
347 | void d_drop(struct dentry *dentry) | |
348 | { | |
789680d1 NP |
349 | spin_lock(&dentry->d_lock); |
350 | __d_drop(dentry); | |
351 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
352 | } |
353 | EXPORT_SYMBOL(d_drop); | |
354 | ||
77812a1e NP |
355 | /* |
356 | * Finish off a dentry we've decided to kill. | |
357 | * dentry->d_lock must be held, returns with it unlocked. | |
358 | * If ref is non-zero, then decrement the refcount too. | |
359 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
360 | */ | |
361 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
362 | __releases(dentry->d_lock) | |
363 | { | |
873feea0 | 364 | struct inode *inode; |
77812a1e NP |
365 | struct dentry *parent; |
366 | ||
873feea0 NP |
367 | inode = dentry->d_inode; |
368 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
369 | relock: |
370 | spin_unlock(&dentry->d_lock); | |
371 | cpu_relax(); | |
372 | return dentry; /* try again with same dentry */ | |
373 | } | |
374 | if (IS_ROOT(dentry)) | |
375 | parent = NULL; | |
376 | else | |
377 | parent = dentry->d_parent; | |
378 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
379 | if (inode) |
380 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
381 | goto relock; |
382 | } | |
31e6b01f | 383 | |
77812a1e NP |
384 | if (ref) |
385 | dentry->d_count--; | |
386 | /* if dentry was on the d_lru list delete it from there */ | |
387 | dentry_lru_del(dentry); | |
388 | /* if it was on the hash then remove it */ | |
389 | __d_drop(dentry); | |
390 | return d_kill(dentry, parent); | |
391 | } | |
392 | ||
1da177e4 LT |
393 | /* |
394 | * This is dput | |
395 | * | |
396 | * This is complicated by the fact that we do not want to put | |
397 | * dentries that are no longer on any hash chain on the unused | |
398 | * list: we'd much rather just get rid of them immediately. | |
399 | * | |
400 | * However, that implies that we have to traverse the dentry | |
401 | * tree upwards to the parents which might _also_ now be | |
402 | * scheduled for deletion (it may have been only waiting for | |
403 | * its last child to go away). | |
404 | * | |
405 | * This tail recursion is done by hand as we don't want to depend | |
406 | * on the compiler to always get this right (gcc generally doesn't). | |
407 | * Real recursion would eat up our stack space. | |
408 | */ | |
409 | ||
410 | /* | |
411 | * dput - release a dentry | |
412 | * @dentry: dentry to release | |
413 | * | |
414 | * Release a dentry. This will drop the usage count and if appropriate | |
415 | * call the dentry unlink method as well as removing it from the queues and | |
416 | * releasing its resources. If the parent dentries were scheduled for release | |
417 | * they too may now get deleted. | |
1da177e4 | 418 | */ |
1da177e4 LT |
419 | void dput(struct dentry *dentry) |
420 | { | |
421 | if (!dentry) | |
422 | return; | |
423 | ||
424 | repeat: | |
b7ab39f6 | 425 | if (dentry->d_count == 1) |
1da177e4 | 426 | might_sleep(); |
1da177e4 | 427 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
428 | BUG_ON(!dentry->d_count); |
429 | if (dentry->d_count > 1) { | |
430 | dentry->d_count--; | |
1da177e4 | 431 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
432 | return; |
433 | } | |
434 | ||
fb045adb | 435 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 436 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 437 | goto kill_it; |
1da177e4 | 438 | } |
265ac902 | 439 | |
1da177e4 LT |
440 | /* Unreachable? Get rid of it */ |
441 | if (d_unhashed(dentry)) | |
442 | goto kill_it; | |
265ac902 NP |
443 | |
444 | /* Otherwise leave it cached and ensure it's on the LRU */ | |
445 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 446 | dentry_lru_add(dentry); |
265ac902 | 447 | |
61f3dee4 NP |
448 | dentry->d_count--; |
449 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
450 | return; |
451 | ||
d52b9086 | 452 | kill_it: |
77812a1e | 453 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
454 | if (dentry) |
455 | goto repeat; | |
1da177e4 | 456 | } |
ec4f8605 | 457 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
458 | |
459 | /** | |
460 | * d_invalidate - invalidate a dentry | |
461 | * @dentry: dentry to invalidate | |
462 | * | |
463 | * Try to invalidate the dentry if it turns out to be | |
464 | * possible. If there are other dentries that can be | |
465 | * reached through this one we can't delete it and we | |
466 | * return -EBUSY. On success we return 0. | |
467 | * | |
468 | * no dcache lock. | |
469 | */ | |
470 | ||
471 | int d_invalidate(struct dentry * dentry) | |
472 | { | |
473 | /* | |
474 | * If it's already been dropped, return OK. | |
475 | */ | |
da502956 | 476 | spin_lock(&dentry->d_lock); |
1da177e4 | 477 | if (d_unhashed(dentry)) { |
da502956 | 478 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
479 | return 0; |
480 | } | |
481 | /* | |
482 | * Check whether to do a partial shrink_dcache | |
483 | * to get rid of unused child entries. | |
484 | */ | |
485 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 486 | spin_unlock(&dentry->d_lock); |
1da177e4 | 487 | shrink_dcache_parent(dentry); |
da502956 | 488 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
489 | } |
490 | ||
491 | /* | |
492 | * Somebody else still using it? | |
493 | * | |
494 | * If it's a directory, we can't drop it | |
495 | * for fear of somebody re-populating it | |
496 | * with children (even though dropping it | |
497 | * would make it unreachable from the root, | |
498 | * we might still populate it if it was a | |
499 | * working directory or similar). | |
500 | */ | |
b7ab39f6 | 501 | if (dentry->d_count > 1) { |
1da177e4 LT |
502 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
503 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
504 | return -EBUSY; |
505 | } | |
506 | } | |
507 | ||
508 | __d_drop(dentry); | |
509 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
510 | return 0; |
511 | } | |
ec4f8605 | 512 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 513 | |
b5c84bf6 | 514 | /* This must be called with d_lock held */ |
dc0474be | 515 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 516 | { |
b7ab39f6 | 517 | dentry->d_count++; |
23044507 NP |
518 | } |
519 | ||
dc0474be | 520 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 521 | { |
23044507 | 522 | spin_lock(&dentry->d_lock); |
dc0474be | 523 | __dget_dlock(dentry); |
23044507 | 524 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
525 | } |
526 | ||
b7ab39f6 NP |
527 | struct dentry *dget_parent(struct dentry *dentry) |
528 | { | |
529 | struct dentry *ret; | |
530 | ||
531 | repeat: | |
a734eb45 NP |
532 | /* |
533 | * Don't need rcu_dereference because we re-check it was correct under | |
534 | * the lock. | |
535 | */ | |
536 | rcu_read_lock(); | |
b7ab39f6 | 537 | ret = dentry->d_parent; |
a734eb45 NP |
538 | if (!ret) { |
539 | rcu_read_unlock(); | |
b7ab39f6 NP |
540 | goto out; |
541 | } | |
a734eb45 NP |
542 | spin_lock(&ret->d_lock); |
543 | if (unlikely(ret != dentry->d_parent)) { | |
544 | spin_unlock(&ret->d_lock); | |
545 | rcu_read_unlock(); | |
b7ab39f6 NP |
546 | goto repeat; |
547 | } | |
a734eb45 | 548 | rcu_read_unlock(); |
b7ab39f6 NP |
549 | BUG_ON(!ret->d_count); |
550 | ret->d_count++; | |
551 | spin_unlock(&ret->d_lock); | |
552 | out: | |
b7ab39f6 NP |
553 | return ret; |
554 | } | |
555 | EXPORT_SYMBOL(dget_parent); | |
556 | ||
1da177e4 LT |
557 | /** |
558 | * d_find_alias - grab a hashed alias of inode | |
559 | * @inode: inode in question | |
560 | * @want_discon: flag, used by d_splice_alias, to request | |
561 | * that only a DISCONNECTED alias be returned. | |
562 | * | |
563 | * If inode has a hashed alias, or is a directory and has any alias, | |
564 | * acquire the reference to alias and return it. Otherwise return NULL. | |
565 | * Notice that if inode is a directory there can be only one alias and | |
566 | * it can be unhashed only if it has no children, or if it is the root | |
567 | * of a filesystem. | |
568 | * | |
21c0d8fd | 569 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 570 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 571 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 572 | */ |
da502956 | 573 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 574 | { |
da502956 | 575 | struct dentry *alias, *discon_alias; |
1da177e4 | 576 | |
da502956 NP |
577 | again: |
578 | discon_alias = NULL; | |
579 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
580 | spin_lock(&alias->d_lock); | |
1da177e4 | 581 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 582 | if (IS_ROOT(alias) && |
da502956 | 583 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 584 | discon_alias = alias; |
da502956 | 585 | } else if (!want_discon) { |
dc0474be | 586 | __dget_dlock(alias); |
da502956 NP |
587 | spin_unlock(&alias->d_lock); |
588 | return alias; | |
589 | } | |
590 | } | |
591 | spin_unlock(&alias->d_lock); | |
592 | } | |
593 | if (discon_alias) { | |
594 | alias = discon_alias; | |
595 | spin_lock(&alias->d_lock); | |
596 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
597 | if (IS_ROOT(alias) && | |
598 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 599 | __dget_dlock(alias); |
da502956 | 600 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
601 | return alias; |
602 | } | |
603 | } | |
da502956 NP |
604 | spin_unlock(&alias->d_lock); |
605 | goto again; | |
1da177e4 | 606 | } |
da502956 | 607 | return NULL; |
1da177e4 LT |
608 | } |
609 | ||
da502956 | 610 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 611 | { |
214fda1f DH |
612 | struct dentry *de = NULL; |
613 | ||
614 | if (!list_empty(&inode->i_dentry)) { | |
873feea0 | 615 | spin_lock(&inode->i_lock); |
214fda1f | 616 | de = __d_find_alias(inode, 0); |
873feea0 | 617 | spin_unlock(&inode->i_lock); |
214fda1f | 618 | } |
1da177e4 LT |
619 | return de; |
620 | } | |
ec4f8605 | 621 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
622 | |
623 | /* | |
624 | * Try to kill dentries associated with this inode. | |
625 | * WARNING: you must own a reference to inode. | |
626 | */ | |
627 | void d_prune_aliases(struct inode *inode) | |
628 | { | |
0cdca3f9 | 629 | struct dentry *dentry; |
1da177e4 | 630 | restart: |
873feea0 | 631 | spin_lock(&inode->i_lock); |
0cdca3f9 | 632 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 633 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 634 | if (!dentry->d_count) { |
dc0474be | 635 | __dget_dlock(dentry); |
1da177e4 LT |
636 | __d_drop(dentry); |
637 | spin_unlock(&dentry->d_lock); | |
873feea0 | 638 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
639 | dput(dentry); |
640 | goto restart; | |
641 | } | |
642 | spin_unlock(&dentry->d_lock); | |
643 | } | |
873feea0 | 644 | spin_unlock(&inode->i_lock); |
1da177e4 | 645 | } |
ec4f8605 | 646 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
647 | |
648 | /* | |
77812a1e NP |
649 | * Try to throw away a dentry - free the inode, dput the parent. |
650 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
651 | * Releases dentry->d_lock. | |
d702ccb3 | 652 | * |
77812a1e | 653 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 654 | */ |
77812a1e | 655 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 656 | __releases(dentry->d_lock) |
1da177e4 | 657 | { |
77812a1e | 658 | struct dentry *parent; |
d52b9086 | 659 | |
77812a1e | 660 | parent = dentry_kill(dentry, 0); |
d52b9086 | 661 | /* |
77812a1e NP |
662 | * If dentry_kill returns NULL, we have nothing more to do. |
663 | * if it returns the same dentry, trylocks failed. In either | |
664 | * case, just loop again. | |
665 | * | |
666 | * Otherwise, we need to prune ancestors too. This is necessary | |
667 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
668 | * is also expected to be beneficial in reducing dentry cache | |
669 | * fragmentation. | |
d52b9086 | 670 | */ |
77812a1e NP |
671 | if (!parent) |
672 | return; | |
673 | if (parent == dentry) | |
674 | return; | |
675 | ||
676 | /* Prune ancestors. */ | |
677 | dentry = parent; | |
d52b9086 | 678 | while (dentry) { |
b7ab39f6 | 679 | spin_lock(&dentry->d_lock); |
89e60548 NP |
680 | if (dentry->d_count > 1) { |
681 | dentry->d_count--; | |
682 | spin_unlock(&dentry->d_lock); | |
683 | return; | |
684 | } | |
77812a1e | 685 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 686 | } |
1da177e4 LT |
687 | } |
688 | ||
3049cfe2 | 689 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 690 | { |
da3bbdd4 | 691 | struct dentry *dentry; |
da3bbdd4 | 692 | |
ec33679d NP |
693 | rcu_read_lock(); |
694 | for (;;) { | |
ec33679d NP |
695 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
696 | if (&dentry->d_lru == list) | |
697 | break; /* empty */ | |
698 | spin_lock(&dentry->d_lock); | |
699 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
700 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
701 | continue; |
702 | } | |
703 | ||
1da177e4 LT |
704 | /* |
705 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
706 | * the LRU because of laziness during lookup. Do not free |
707 | * it - just keep it off the LRU list. | |
1da177e4 | 708 | */ |
b7ab39f6 | 709 | if (dentry->d_count) { |
ec33679d | 710 | dentry_lru_del(dentry); |
da3bbdd4 | 711 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
712 | continue; |
713 | } | |
ec33679d | 714 | |
ec33679d | 715 | rcu_read_unlock(); |
77812a1e NP |
716 | |
717 | try_prune_one_dentry(dentry); | |
718 | ||
ec33679d | 719 | rcu_read_lock(); |
da3bbdd4 | 720 | } |
ec33679d | 721 | rcu_read_unlock(); |
3049cfe2 CH |
722 | } |
723 | ||
724 | /** | |
725 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
726 | * @sb: superblock to shrink dentry LRU. | |
727 | * @count: number of entries to prune | |
728 | * @flags: flags to control the dentry processing | |
729 | * | |
730 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
731 | */ | |
732 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
733 | { | |
734 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
735 | struct dentry *dentry; | |
736 | LIST_HEAD(referenced); | |
737 | LIST_HEAD(tmp); | |
738 | int cnt = *count; | |
739 | ||
23044507 NP |
740 | relock: |
741 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
742 | while (!list_empty(&sb->s_dentry_lru)) { |
743 | dentry = list_entry(sb->s_dentry_lru.prev, | |
744 | struct dentry, d_lru); | |
745 | BUG_ON(dentry->d_sb != sb); | |
746 | ||
23044507 NP |
747 | if (!spin_trylock(&dentry->d_lock)) { |
748 | spin_unlock(&dcache_lru_lock); | |
749 | cpu_relax(); | |
750 | goto relock; | |
751 | } | |
752 | ||
3049cfe2 CH |
753 | /* |
754 | * If we are honouring the DCACHE_REFERENCED flag and the | |
755 | * dentry has this flag set, don't free it. Clear the flag | |
756 | * and put it back on the LRU. | |
757 | */ | |
23044507 NP |
758 | if (flags & DCACHE_REFERENCED && |
759 | dentry->d_flags & DCACHE_REFERENCED) { | |
760 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
761 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 762 | spin_unlock(&dentry->d_lock); |
23044507 NP |
763 | } else { |
764 | list_move_tail(&dentry->d_lru, &tmp); | |
765 | spin_unlock(&dentry->d_lock); | |
766 | if (!--cnt) | |
767 | break; | |
3049cfe2 | 768 | } |
ec33679d | 769 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 770 | } |
da3bbdd4 KM |
771 | if (!list_empty(&referenced)) |
772 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 773 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
774 | |
775 | shrink_dentry_list(&tmp); | |
776 | ||
777 | *count = cnt; | |
da3bbdd4 KM |
778 | } |
779 | ||
780 | /** | |
781 | * prune_dcache - shrink the dcache | |
782 | * @count: number of entries to try to free | |
783 | * | |
784 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
785 | * need to unmount something (at which point we need to unuse all dentries). | |
786 | * | |
787 | * This function may fail to free any resources if all the dentries are in use. | |
788 | */ | |
789 | static void prune_dcache(int count) | |
790 | { | |
dca33252 | 791 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 792 | int w_count; |
86c8749e | 793 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
794 | int prune_ratio; |
795 | int pruned; | |
796 | ||
797 | if (unused == 0 || count == 0) | |
798 | return; | |
da3bbdd4 KM |
799 | if (count >= unused) |
800 | prune_ratio = 1; | |
801 | else | |
802 | prune_ratio = unused / count; | |
803 | spin_lock(&sb_lock); | |
dca33252 | 804 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
805 | if (list_empty(&sb->s_instances)) |
806 | continue; | |
da3bbdd4 | 807 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 808 | continue; |
da3bbdd4 KM |
809 | sb->s_count++; |
810 | /* Now, we reclaim unused dentrins with fairness. | |
811 | * We reclaim them same percentage from each superblock. | |
812 | * We calculate number of dentries to scan on this sb | |
813 | * as follows, but the implementation is arranged to avoid | |
814 | * overflows: | |
815 | * number of dentries to scan on this sb = | |
816 | * count * (number of dentries on this sb / | |
817 | * number of dentries in the machine) | |
0feae5c4 | 818 | */ |
da3bbdd4 KM |
819 | spin_unlock(&sb_lock); |
820 | if (prune_ratio != 1) | |
821 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
822 | else | |
823 | w_count = sb->s_nr_dentry_unused; | |
824 | pruned = w_count; | |
0feae5c4 | 825 | /* |
da3bbdd4 KM |
826 | * We need to be sure this filesystem isn't being unmounted, |
827 | * otherwise we could race with generic_shutdown_super(), and | |
828 | * end up holding a reference to an inode while the filesystem | |
829 | * is unmounted. So we try to get s_umount, and make sure | |
830 | * s_root isn't NULL. | |
0feae5c4 | 831 | */ |
da3bbdd4 KM |
832 | if (down_read_trylock(&sb->s_umount)) { |
833 | if ((sb->s_root != NULL) && | |
834 | (!list_empty(&sb->s_dentry_lru))) { | |
da3bbdd4 KM |
835 | __shrink_dcache_sb(sb, &w_count, |
836 | DCACHE_REFERENCED); | |
837 | pruned -= w_count; | |
0feae5c4 | 838 | } |
da3bbdd4 | 839 | up_read(&sb->s_umount); |
0feae5c4 | 840 | } |
da3bbdd4 | 841 | spin_lock(&sb_lock); |
dca33252 AV |
842 | if (p) |
843 | __put_super(p); | |
da3bbdd4 | 844 | count -= pruned; |
dca33252 | 845 | p = sb; |
79893c17 AV |
846 | /* more work left to do? */ |
847 | if (count <= 0) | |
848 | break; | |
1da177e4 | 849 | } |
dca33252 AV |
850 | if (p) |
851 | __put_super(p); | |
da3bbdd4 | 852 | spin_unlock(&sb_lock); |
1da177e4 LT |
853 | } |
854 | ||
1da177e4 LT |
855 | /** |
856 | * shrink_dcache_sb - shrink dcache for a superblock | |
857 | * @sb: superblock | |
858 | * | |
3049cfe2 CH |
859 | * Shrink the dcache for the specified super block. This is used to free |
860 | * the dcache before unmounting a file system. | |
1da177e4 | 861 | */ |
3049cfe2 | 862 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 863 | { |
3049cfe2 CH |
864 | LIST_HEAD(tmp); |
865 | ||
23044507 | 866 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
867 | while (!list_empty(&sb->s_dentry_lru)) { |
868 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 869 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 870 | shrink_dentry_list(&tmp); |
ec33679d | 871 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 872 | } |
23044507 | 873 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 874 | } |
ec4f8605 | 875 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 876 | |
c636ebdb DH |
877 | /* |
878 | * destroy a single subtree of dentries for unmount | |
879 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
880 | * locking | |
881 | */ | |
882 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
883 | { | |
884 | struct dentry *parent; | |
f8713576 | 885 | unsigned detached = 0; |
c636ebdb DH |
886 | |
887 | BUG_ON(!IS_ROOT(dentry)); | |
888 | ||
889 | /* detach this root from the system */ | |
23044507 | 890 | spin_lock(&dentry->d_lock); |
a4633357 | 891 | dentry_lru_del(dentry); |
c636ebdb | 892 | __d_drop(dentry); |
da502956 | 893 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
894 | |
895 | for (;;) { | |
896 | /* descend to the first leaf in the current subtree */ | |
897 | while (!list_empty(&dentry->d_subdirs)) { | |
898 | struct dentry *loop; | |
899 | ||
900 | /* this is a branch with children - detach all of them | |
901 | * from the system in one go */ | |
2fd6b7f5 | 902 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
903 | list_for_each_entry(loop, &dentry->d_subdirs, |
904 | d_u.d_child) { | |
2fd6b7f5 NP |
905 | spin_lock_nested(&loop->d_lock, |
906 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 907 | dentry_lru_del(loop); |
c636ebdb | 908 | __d_drop(loop); |
da502956 | 909 | spin_unlock(&loop->d_lock); |
c636ebdb | 910 | } |
2fd6b7f5 | 911 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
912 | |
913 | /* move to the first child */ | |
914 | dentry = list_entry(dentry->d_subdirs.next, | |
915 | struct dentry, d_u.d_child); | |
916 | } | |
917 | ||
918 | /* consume the dentries from this leaf up through its parents | |
919 | * until we find one with children or run out altogether */ | |
920 | do { | |
921 | struct inode *inode; | |
922 | ||
b7ab39f6 | 923 | if (dentry->d_count != 0) { |
c636ebdb DH |
924 | printk(KERN_ERR |
925 | "BUG: Dentry %p{i=%lx,n=%s}" | |
926 | " still in use (%d)" | |
927 | " [unmount of %s %s]\n", | |
928 | dentry, | |
929 | dentry->d_inode ? | |
930 | dentry->d_inode->i_ino : 0UL, | |
931 | dentry->d_name.name, | |
b7ab39f6 | 932 | dentry->d_count, |
c636ebdb DH |
933 | dentry->d_sb->s_type->name, |
934 | dentry->d_sb->s_id); | |
935 | BUG(); | |
936 | } | |
937 | ||
2fd6b7f5 | 938 | if (IS_ROOT(dentry)) { |
c636ebdb | 939 | parent = NULL; |
2fd6b7f5 NP |
940 | list_del(&dentry->d_u.d_child); |
941 | } else { | |
871c0067 | 942 | parent = dentry->d_parent; |
b7ab39f6 NP |
943 | spin_lock(&parent->d_lock); |
944 | parent->d_count--; | |
2fd6b7f5 | 945 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 946 | spin_unlock(&parent->d_lock); |
871c0067 | 947 | } |
c636ebdb | 948 | |
f8713576 | 949 | detached++; |
c636ebdb DH |
950 | |
951 | inode = dentry->d_inode; | |
952 | if (inode) { | |
953 | dentry->d_inode = NULL; | |
954 | list_del_init(&dentry->d_alias); | |
955 | if (dentry->d_op && dentry->d_op->d_iput) | |
956 | dentry->d_op->d_iput(dentry, inode); | |
957 | else | |
958 | iput(inode); | |
959 | } | |
960 | ||
961 | d_free(dentry); | |
962 | ||
963 | /* finished when we fall off the top of the tree, | |
964 | * otherwise we ascend to the parent and move to the | |
965 | * next sibling if there is one */ | |
966 | if (!parent) | |
312d3ca8 | 967 | return; |
c636ebdb | 968 | dentry = parent; |
c636ebdb DH |
969 | } while (list_empty(&dentry->d_subdirs)); |
970 | ||
971 | dentry = list_entry(dentry->d_subdirs.next, | |
972 | struct dentry, d_u.d_child); | |
973 | } | |
974 | } | |
975 | ||
976 | /* | |
977 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 978 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
979 | * - the superblock is detached from all mountings and open files, so the |
980 | * dentry trees will not be rearranged by the VFS | |
981 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
982 | * any dentries belonging to this superblock that it comes across | |
983 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
984 | * in this superblock | |
985 | */ | |
986 | void shrink_dcache_for_umount(struct super_block *sb) | |
987 | { | |
988 | struct dentry *dentry; | |
989 | ||
990 | if (down_read_trylock(&sb->s_umount)) | |
991 | BUG(); | |
992 | ||
993 | dentry = sb->s_root; | |
994 | sb->s_root = NULL; | |
b7ab39f6 NP |
995 | spin_lock(&dentry->d_lock); |
996 | dentry->d_count--; | |
997 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
998 | shrink_dcache_for_umount_subtree(dentry); |
999 | ||
ceb5bdc2 NP |
1000 | while (!hlist_bl_empty(&sb->s_anon)) { |
1001 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
1002 | shrink_dcache_for_umount_subtree(dentry); |
1003 | } | |
1004 | } | |
1005 | ||
c826cb7d LT |
1006 | /* |
1007 | * This tries to ascend one level of parenthood, but | |
1008 | * we can race with renaming, so we need to re-check | |
1009 | * the parenthood after dropping the lock and check | |
1010 | * that the sequence number still matches. | |
1011 | */ | |
1012 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
1013 | { | |
1014 | struct dentry *new = old->d_parent; | |
1015 | ||
1016 | rcu_read_lock(); | |
1017 | spin_unlock(&old->d_lock); | |
1018 | spin_lock(&new->d_lock); | |
1019 | ||
1020 | /* | |
1021 | * might go back up the wrong parent if we have had a rename | |
1022 | * or deletion | |
1023 | */ | |
1024 | if (new != old->d_parent || | |
c83ce989 | 1025 | (old->d_flags & DCACHE_DISCONNECTED) || |
c826cb7d LT |
1026 | (!locked && read_seqretry(&rename_lock, seq))) { |
1027 | spin_unlock(&new->d_lock); | |
1028 | new = NULL; | |
1029 | } | |
1030 | rcu_read_unlock(); | |
1031 | return new; | |
1032 | } | |
1033 | ||
1034 | ||
1da177e4 LT |
1035 | /* |
1036 | * Search for at least 1 mount point in the dentry's subdirs. | |
1037 | * We descend to the next level whenever the d_subdirs | |
1038 | * list is non-empty and continue searching. | |
1039 | */ | |
1040 | ||
1041 | /** | |
1042 | * have_submounts - check for mounts over a dentry | |
1043 | * @parent: dentry to check. | |
1044 | * | |
1045 | * Return true if the parent or its subdirectories contain | |
1046 | * a mount point | |
1047 | */ | |
1da177e4 LT |
1048 | int have_submounts(struct dentry *parent) |
1049 | { | |
949854d0 | 1050 | struct dentry *this_parent; |
1da177e4 | 1051 | struct list_head *next; |
949854d0 | 1052 | unsigned seq; |
58db63d0 | 1053 | int locked = 0; |
949854d0 | 1054 | |
949854d0 | 1055 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1056 | again: |
1057 | this_parent = parent; | |
1da177e4 | 1058 | |
1da177e4 LT |
1059 | if (d_mountpoint(parent)) |
1060 | goto positive; | |
2fd6b7f5 | 1061 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1062 | repeat: |
1063 | next = this_parent->d_subdirs.next; | |
1064 | resume: | |
1065 | while (next != &this_parent->d_subdirs) { | |
1066 | struct list_head *tmp = next; | |
5160ee6f | 1067 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1068 | next = tmp->next; |
2fd6b7f5 NP |
1069 | |
1070 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1071 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1072 | if (d_mountpoint(dentry)) { |
1073 | spin_unlock(&dentry->d_lock); | |
1074 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1075 | goto positive; |
2fd6b7f5 | 1076 | } |
1da177e4 | 1077 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1078 | spin_unlock(&this_parent->d_lock); |
1079 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1080 | this_parent = dentry; |
2fd6b7f5 | 1081 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1082 | goto repeat; |
1083 | } | |
2fd6b7f5 | 1084 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1085 | } |
1086 | /* | |
1087 | * All done at this level ... ascend and resume the search. | |
1088 | */ | |
1089 | if (this_parent != parent) { | |
c826cb7d LT |
1090 | struct dentry *child = this_parent; |
1091 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1092 | if (!this_parent) | |
949854d0 | 1093 | goto rename_retry; |
949854d0 | 1094 | next = child->d_u.d_child.next; |
1da177e4 LT |
1095 | goto resume; |
1096 | } | |
2fd6b7f5 | 1097 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1098 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1099 | goto rename_retry; |
58db63d0 NP |
1100 | if (locked) |
1101 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1102 | return 0; /* No mount points found in tree */ |
1103 | positive: | |
58db63d0 | 1104 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1105 | goto rename_retry; |
58db63d0 NP |
1106 | if (locked) |
1107 | write_sequnlock(&rename_lock); | |
1da177e4 | 1108 | return 1; |
58db63d0 NP |
1109 | |
1110 | rename_retry: | |
1111 | locked = 1; | |
1112 | write_seqlock(&rename_lock); | |
1113 | goto again; | |
1da177e4 | 1114 | } |
ec4f8605 | 1115 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1116 | |
1117 | /* | |
1118 | * Search the dentry child list for the specified parent, | |
1119 | * and move any unused dentries to the end of the unused | |
1120 | * list for prune_dcache(). We descend to the next level | |
1121 | * whenever the d_subdirs list is non-empty and continue | |
1122 | * searching. | |
1123 | * | |
1124 | * It returns zero iff there are no unused children, | |
1125 | * otherwise it returns the number of children moved to | |
1126 | * the end of the unused list. This may not be the total | |
1127 | * number of unused children, because select_parent can | |
1128 | * drop the lock and return early due to latency | |
1129 | * constraints. | |
1130 | */ | |
1131 | static int select_parent(struct dentry * parent) | |
1132 | { | |
949854d0 | 1133 | struct dentry *this_parent; |
1da177e4 | 1134 | struct list_head *next; |
949854d0 | 1135 | unsigned seq; |
1da177e4 | 1136 | int found = 0; |
58db63d0 | 1137 | int locked = 0; |
1da177e4 | 1138 | |
949854d0 | 1139 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1140 | again: |
1141 | this_parent = parent; | |
2fd6b7f5 | 1142 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1143 | repeat: |
1144 | next = this_parent->d_subdirs.next; | |
1145 | resume: | |
1146 | while (next != &this_parent->d_subdirs) { | |
1147 | struct list_head *tmp = next; | |
5160ee6f | 1148 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1149 | next = tmp->next; |
1150 | ||
2fd6b7f5 | 1151 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1152 | |
1da177e4 LT |
1153 | /* |
1154 | * move only zero ref count dentries to the end | |
1155 | * of the unused list for prune_dcache | |
1156 | */ | |
b7ab39f6 | 1157 | if (!dentry->d_count) { |
a4633357 | 1158 | dentry_lru_move_tail(dentry); |
1da177e4 | 1159 | found++; |
a4633357 CH |
1160 | } else { |
1161 | dentry_lru_del(dentry); | |
1da177e4 LT |
1162 | } |
1163 | ||
1164 | /* | |
1165 | * We can return to the caller if we have found some (this | |
1166 | * ensures forward progress). We'll be coming back to find | |
1167 | * the rest. | |
1168 | */ | |
2fd6b7f5 NP |
1169 | if (found && need_resched()) { |
1170 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1171 | goto out; |
2fd6b7f5 | 1172 | } |
1da177e4 LT |
1173 | |
1174 | /* | |
1175 | * Descend a level if the d_subdirs list is non-empty. | |
1176 | */ | |
1177 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1178 | spin_unlock(&this_parent->d_lock); |
1179 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1180 | this_parent = dentry; |
2fd6b7f5 | 1181 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1182 | goto repeat; |
1183 | } | |
2fd6b7f5 NP |
1184 | |
1185 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1186 | } |
1187 | /* | |
1188 | * All done at this level ... ascend and resume the search. | |
1189 | */ | |
1190 | if (this_parent != parent) { | |
c826cb7d LT |
1191 | struct dentry *child = this_parent; |
1192 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1193 | if (!this_parent) | |
949854d0 | 1194 | goto rename_retry; |
949854d0 | 1195 | next = child->d_u.d_child.next; |
1da177e4 LT |
1196 | goto resume; |
1197 | } | |
1198 | out: | |
2fd6b7f5 | 1199 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1200 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1201 | goto rename_retry; |
58db63d0 NP |
1202 | if (locked) |
1203 | write_sequnlock(&rename_lock); | |
1da177e4 | 1204 | return found; |
58db63d0 NP |
1205 | |
1206 | rename_retry: | |
1207 | if (found) | |
1208 | return found; | |
1209 | locked = 1; | |
1210 | write_seqlock(&rename_lock); | |
1211 | goto again; | |
1da177e4 LT |
1212 | } |
1213 | ||
1214 | /** | |
1215 | * shrink_dcache_parent - prune dcache | |
1216 | * @parent: parent of entries to prune | |
1217 | * | |
1218 | * Prune the dcache to remove unused children of the parent dentry. | |
1219 | */ | |
1220 | ||
1221 | void shrink_dcache_parent(struct dentry * parent) | |
1222 | { | |
da3bbdd4 | 1223 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1224 | int found; |
1225 | ||
1226 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 1227 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 1228 | } |
ec4f8605 | 1229 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1230 | |
1da177e4 LT |
1231 | /* |
1232 | * Scan `nr' dentries and return the number which remain. | |
1233 | * | |
1234 | * We need to avoid reentering the filesystem if the caller is performing a | |
1235 | * GFP_NOFS allocation attempt. One example deadlock is: | |
1236 | * | |
1237 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
1238 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
1239 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
1240 | * | |
1241 | * In this case we return -1 to tell the caller that we baled. | |
1242 | */ | |
7f8275d0 | 1243 | static int shrink_dcache_memory(struct shrinker *shrink, int nr, gfp_t gfp_mask) |
1da177e4 LT |
1244 | { |
1245 | if (nr) { | |
1246 | if (!(gfp_mask & __GFP_FS)) | |
1247 | return -1; | |
da3bbdd4 | 1248 | prune_dcache(nr); |
1da177e4 | 1249 | } |
312d3ca8 | 1250 | |
86c8749e | 1251 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
1252 | } |
1253 | ||
8e1f936b RR |
1254 | static struct shrinker dcache_shrinker = { |
1255 | .shrink = shrink_dcache_memory, | |
1256 | .seeks = DEFAULT_SEEKS, | |
1257 | }; | |
1258 | ||
1da177e4 LT |
1259 | /** |
1260 | * d_alloc - allocate a dcache entry | |
1261 | * @parent: parent of entry to allocate | |
1262 | * @name: qstr of the name | |
1263 | * | |
1264 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1265 | * available. On a success the dentry is returned. The name passed in is | |
1266 | * copied and the copy passed in may be reused after this call. | |
1267 | */ | |
1268 | ||
1269 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1270 | { | |
1271 | struct dentry *dentry; | |
1272 | char *dname; | |
1273 | ||
e12ba74d | 1274 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1275 | if (!dentry) |
1276 | return NULL; | |
1277 | ||
1278 | if (name->len > DNAME_INLINE_LEN-1) { | |
1279 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1280 | if (!dname) { | |
1281 | kmem_cache_free(dentry_cache, dentry); | |
1282 | return NULL; | |
1283 | } | |
1284 | } else { | |
1285 | dname = dentry->d_iname; | |
1286 | } | |
1287 | dentry->d_name.name = dname; | |
1288 | ||
1289 | dentry->d_name.len = name->len; | |
1290 | dentry->d_name.hash = name->hash; | |
1291 | memcpy(dname, name->name, name->len); | |
1292 | dname[name->len] = 0; | |
1293 | ||
b7ab39f6 | 1294 | dentry->d_count = 1; |
dea3667b | 1295 | dentry->d_flags = 0; |
1da177e4 | 1296 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1297 | seqcount_init(&dentry->d_seq); |
1da177e4 LT |
1298 | dentry->d_inode = NULL; |
1299 | dentry->d_parent = NULL; | |
1300 | dentry->d_sb = NULL; | |
1301 | dentry->d_op = NULL; | |
1302 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1303 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1304 | INIT_LIST_HEAD(&dentry->d_lru); |
1305 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1306 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1307 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
1308 | |
1309 | if (parent) { | |
2fd6b7f5 | 1310 | spin_lock(&parent->d_lock); |
89ad485f NP |
1311 | /* |
1312 | * don't need child lock because it is not subject | |
1313 | * to concurrency here | |
1314 | */ | |
dc0474be NP |
1315 | __dget_dlock(parent); |
1316 | dentry->d_parent = parent; | |
1da177e4 | 1317 | dentry->d_sb = parent->d_sb; |
c8aebb0c | 1318 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
5160ee6f | 1319 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
2fd6b7f5 | 1320 | spin_unlock(&parent->d_lock); |
2fd6b7f5 | 1321 | } |
1da177e4 | 1322 | |
3e880fb5 | 1323 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1324 | |
1da177e4 LT |
1325 | return dentry; |
1326 | } | |
ec4f8605 | 1327 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1328 | |
4b936885 NP |
1329 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1330 | { | |
1331 | struct dentry *dentry = d_alloc(NULL, name); | |
1332 | if (dentry) { | |
1333 | dentry->d_sb = sb; | |
c8aebb0c | 1334 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
4b936885 NP |
1335 | dentry->d_parent = dentry; |
1336 | dentry->d_flags |= DCACHE_DISCONNECTED; | |
1337 | } | |
1338 | return dentry; | |
1339 | } | |
1340 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1341 | ||
1da177e4 LT |
1342 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1343 | { | |
1344 | struct qstr q; | |
1345 | ||
1346 | q.name = name; | |
1347 | q.len = strlen(name); | |
1348 | q.hash = full_name_hash(q.name, q.len); | |
1349 | return d_alloc(parent, &q); | |
1350 | } | |
ef26ca97 | 1351 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1352 | |
fb045adb NP |
1353 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1354 | { | |
6f7f7caa LT |
1355 | WARN_ON_ONCE(dentry->d_op); |
1356 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1357 | DCACHE_OP_COMPARE | |
1358 | DCACHE_OP_REVALIDATE | | |
1359 | DCACHE_OP_DELETE )); | |
1360 | dentry->d_op = op; | |
1361 | if (!op) | |
1362 | return; | |
1363 | if (op->d_hash) | |
1364 | dentry->d_flags |= DCACHE_OP_HASH; | |
1365 | if (op->d_compare) | |
1366 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1367 | if (op->d_revalidate) | |
1368 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1369 | if (op->d_delete) | |
1370 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1371 | ||
1372 | } | |
1373 | EXPORT_SYMBOL(d_set_d_op); | |
1374 | ||
360da900 OH |
1375 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1376 | { | |
b23fb0a6 | 1377 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1378 | if (inode) { |
1379 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1380 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
360da900 | 1381 | list_add(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1382 | } |
360da900 | 1383 | dentry->d_inode = inode; |
31e6b01f | 1384 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1385 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1386 | fsnotify_d_instantiate(dentry, inode); |
1387 | } | |
1388 | ||
1da177e4 LT |
1389 | /** |
1390 | * d_instantiate - fill in inode information for a dentry | |
1391 | * @entry: dentry to complete | |
1392 | * @inode: inode to attach to this dentry | |
1393 | * | |
1394 | * Fill in inode information in the entry. | |
1395 | * | |
1396 | * This turns negative dentries into productive full members | |
1397 | * of society. | |
1398 | * | |
1399 | * NOTE! This assumes that the inode count has been incremented | |
1400 | * (or otherwise set) by the caller to indicate that it is now | |
1401 | * in use by the dcache. | |
1402 | */ | |
1403 | ||
1404 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1405 | { | |
28133c7b | 1406 | BUG_ON(!list_empty(&entry->d_alias)); |
873feea0 NP |
1407 | if (inode) |
1408 | spin_lock(&inode->i_lock); | |
360da900 | 1409 | __d_instantiate(entry, inode); |
873feea0 NP |
1410 | if (inode) |
1411 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1412 | security_d_instantiate(entry, inode); |
1413 | } | |
ec4f8605 | 1414 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1415 | |
1416 | /** | |
1417 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1418 | * @entry: dentry to instantiate | |
1419 | * @inode: inode to attach to this dentry | |
1420 | * | |
1421 | * Fill in inode information in the entry. On success, it returns NULL. | |
1422 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1423 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1424 | * |
1425 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1426 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1427 | * |
1428 | * This also assumes that the inode count has been incremented | |
1429 | * (or otherwise set) by the caller to indicate that it is now | |
1430 | * in use by the dcache. | |
1da177e4 | 1431 | */ |
770bfad8 DH |
1432 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1433 | struct inode *inode) | |
1da177e4 LT |
1434 | { |
1435 | struct dentry *alias; | |
1436 | int len = entry->d_name.len; | |
1437 | const char *name = entry->d_name.name; | |
1438 | unsigned int hash = entry->d_name.hash; | |
1439 | ||
770bfad8 | 1440 | if (!inode) { |
360da900 | 1441 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1442 | return NULL; |
1443 | } | |
1444 | ||
1da177e4 LT |
1445 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1446 | struct qstr *qstr = &alias->d_name; | |
1447 | ||
9abca360 NP |
1448 | /* |
1449 | * Don't need alias->d_lock here, because aliases with | |
1450 | * d_parent == entry->d_parent are not subject to name or | |
1451 | * parent changes, because the parent inode i_mutex is held. | |
1452 | */ | |
1da177e4 LT |
1453 | if (qstr->hash != hash) |
1454 | continue; | |
1455 | if (alias->d_parent != entry->d_parent) | |
1456 | continue; | |
9d55c369 | 1457 | if (dentry_cmp(qstr->name, qstr->len, name, len)) |
1da177e4 | 1458 | continue; |
dc0474be | 1459 | __dget(alias); |
1da177e4 LT |
1460 | return alias; |
1461 | } | |
770bfad8 | 1462 | |
360da900 | 1463 | __d_instantiate(entry, inode); |
1da177e4 LT |
1464 | return NULL; |
1465 | } | |
770bfad8 DH |
1466 | |
1467 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1468 | { | |
1469 | struct dentry *result; | |
1470 | ||
1471 | BUG_ON(!list_empty(&entry->d_alias)); | |
1472 | ||
873feea0 NP |
1473 | if (inode) |
1474 | spin_lock(&inode->i_lock); | |
770bfad8 | 1475 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1476 | if (inode) |
1477 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1478 | |
1479 | if (!result) { | |
1480 | security_d_instantiate(entry, inode); | |
1481 | return NULL; | |
1482 | } | |
1483 | ||
1484 | BUG_ON(!d_unhashed(result)); | |
1485 | iput(inode); | |
1486 | return result; | |
1487 | } | |
1488 | ||
1da177e4 LT |
1489 | EXPORT_SYMBOL(d_instantiate_unique); |
1490 | ||
1491 | /** | |
1492 | * d_alloc_root - allocate root dentry | |
1493 | * @root_inode: inode to allocate the root for | |
1494 | * | |
1495 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1496 | * instantiated and returned. %NULL is returned if there is insufficient | |
1497 | * memory or the inode passed is %NULL. | |
1498 | */ | |
1499 | ||
1500 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1501 | { | |
1502 | struct dentry *res = NULL; | |
1503 | ||
1504 | if (root_inode) { | |
1505 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1506 | ||
1507 | res = d_alloc(NULL, &name); | |
1508 | if (res) { | |
1509 | res->d_sb = root_inode->i_sb; | |
c8aebb0c | 1510 | d_set_d_op(res, res->d_sb->s_d_op); |
1da177e4 LT |
1511 | res->d_parent = res; |
1512 | d_instantiate(res, root_inode); | |
1513 | } | |
1514 | } | |
1515 | return res; | |
1516 | } | |
ec4f8605 | 1517 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 | 1518 | |
d891eedb BF |
1519 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1520 | { | |
1521 | struct dentry *alias; | |
1522 | ||
1523 | if (list_empty(&inode->i_dentry)) | |
1524 | return NULL; | |
1525 | alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias); | |
1526 | __dget(alias); | |
1527 | return alias; | |
1528 | } | |
1529 | ||
1530 | static struct dentry * d_find_any_alias(struct inode *inode) | |
1531 | { | |
1532 | struct dentry *de; | |
1533 | ||
1534 | spin_lock(&inode->i_lock); | |
1535 | de = __d_find_any_alias(inode); | |
1536 | spin_unlock(&inode->i_lock); | |
1537 | return de; | |
1538 | } | |
1539 | ||
1540 | ||
4ea3ada2 CH |
1541 | /** |
1542 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1543 | * @inode: inode to allocate the dentry for | |
1544 | * | |
1545 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1546 | * similar open by handle operations. The returned dentry may be anonymous, | |
1547 | * or may have a full name (if the inode was already in the cache). | |
1548 | * | |
1549 | * When called on a directory inode, we must ensure that the inode only ever | |
1550 | * has one dentry. If a dentry is found, that is returned instead of | |
1551 | * allocating a new one. | |
1552 | * | |
1553 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1554 | * to the dentry. In case of an error the reference on the inode is released. |
1555 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1556 | * be passed in and will be the error will be propagate to the return value, | |
1557 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1558 | */ |
1559 | struct dentry *d_obtain_alias(struct inode *inode) | |
1560 | { | |
9308a612 CH |
1561 | static const struct qstr anonstring = { .name = "" }; |
1562 | struct dentry *tmp; | |
1563 | struct dentry *res; | |
4ea3ada2 CH |
1564 | |
1565 | if (!inode) | |
44003728 | 1566 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1567 | if (IS_ERR(inode)) |
1568 | return ERR_CAST(inode); | |
1569 | ||
d891eedb | 1570 | res = d_find_any_alias(inode); |
9308a612 CH |
1571 | if (res) |
1572 | goto out_iput; | |
1573 | ||
1574 | tmp = d_alloc(NULL, &anonstring); | |
1575 | if (!tmp) { | |
1576 | res = ERR_PTR(-ENOMEM); | |
1577 | goto out_iput; | |
4ea3ada2 | 1578 | } |
9308a612 CH |
1579 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1580 | ||
b5c84bf6 | 1581 | |
873feea0 | 1582 | spin_lock(&inode->i_lock); |
d891eedb | 1583 | res = __d_find_any_alias(inode); |
9308a612 | 1584 | if (res) { |
873feea0 | 1585 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1586 | dput(tmp); |
1587 | goto out_iput; | |
1588 | } | |
1589 | ||
1590 | /* attach a disconnected dentry */ | |
1591 | spin_lock(&tmp->d_lock); | |
1592 | tmp->d_sb = inode->i_sb; | |
c8aebb0c | 1593 | d_set_d_op(tmp, tmp->d_sb->s_d_op); |
9308a612 CH |
1594 | tmp->d_inode = inode; |
1595 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
9308a612 | 1596 | list_add(&tmp->d_alias, &inode->i_dentry); |
dea3667b | 1597 | spin_lock_bucket(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1598 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
dea3667b | 1599 | spin_unlock_bucket(&tmp->d_sb->s_anon); |
9308a612 | 1600 | spin_unlock(&tmp->d_lock); |
873feea0 | 1601 | spin_unlock(&inode->i_lock); |
24ff6663 | 1602 | security_d_instantiate(tmp, inode); |
9308a612 | 1603 | |
9308a612 CH |
1604 | return tmp; |
1605 | ||
1606 | out_iput: | |
24ff6663 JB |
1607 | if (res && !IS_ERR(res)) |
1608 | security_d_instantiate(res, inode); | |
9308a612 CH |
1609 | iput(inode); |
1610 | return res; | |
4ea3ada2 | 1611 | } |
adc48720 | 1612 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1613 | |
1614 | /** | |
1615 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1616 | * @inode: the inode which may have a disconnected dentry | |
1617 | * @dentry: a negative dentry which we want to point to the inode. | |
1618 | * | |
1619 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1620 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1621 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1622 | * | |
1623 | * This is needed in the lookup routine of any filesystem that is exportable | |
1624 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1625 | * | |
1626 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1627 | * is returned. This matches the expected return value of ->lookup. | |
1628 | * | |
1629 | */ | |
1630 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1631 | { | |
1632 | struct dentry *new = NULL; | |
1633 | ||
21c0d8fd | 1634 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1635 | spin_lock(&inode->i_lock); |
1da177e4 LT |
1636 | new = __d_find_alias(inode, 1); |
1637 | if (new) { | |
1638 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
873feea0 | 1639 | spin_unlock(&inode->i_lock); |
1da177e4 | 1640 | security_d_instantiate(new, inode); |
1da177e4 LT |
1641 | d_move(new, dentry); |
1642 | iput(inode); | |
1643 | } else { | |
873feea0 | 1644 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1645 | __d_instantiate(dentry, inode); |
873feea0 | 1646 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1647 | security_d_instantiate(dentry, inode); |
1648 | d_rehash(dentry); | |
1649 | } | |
1650 | } else | |
1651 | d_add(dentry, inode); | |
1652 | return new; | |
1653 | } | |
ec4f8605 | 1654 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1655 | |
9403540c BN |
1656 | /** |
1657 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1658 | * @inode: the inode case-insensitive lookup has found | |
1659 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1660 | * @name: the case-exact name to be associated with the returned dentry | |
1661 | * | |
1662 | * This is to avoid filling the dcache with case-insensitive names to the | |
1663 | * same inode, only the actual correct case is stored in the dcache for | |
1664 | * case-insensitive filesystems. | |
1665 | * | |
1666 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1667 | * already exists in in the dcache, use it and return it. | |
1668 | * | |
1669 | * If no entry exists with the exact case name, allocate new dentry with | |
1670 | * the exact case, and return the spliced entry. | |
1671 | */ | |
e45b590b | 1672 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1673 | struct qstr *name) |
1674 | { | |
1675 | int error; | |
1676 | struct dentry *found; | |
1677 | struct dentry *new; | |
1678 | ||
b6520c81 CH |
1679 | /* |
1680 | * First check if a dentry matching the name already exists, | |
1681 | * if not go ahead and create it now. | |
1682 | */ | |
9403540c | 1683 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1684 | if (!found) { |
1685 | new = d_alloc(dentry->d_parent, name); | |
1686 | if (!new) { | |
1687 | error = -ENOMEM; | |
1688 | goto err_out; | |
1689 | } | |
b6520c81 | 1690 | |
9403540c BN |
1691 | found = d_splice_alias(inode, new); |
1692 | if (found) { | |
1693 | dput(new); | |
1694 | return found; | |
1695 | } | |
1696 | return new; | |
1697 | } | |
b6520c81 CH |
1698 | |
1699 | /* | |
1700 | * If a matching dentry exists, and it's not negative use it. | |
1701 | * | |
1702 | * Decrement the reference count to balance the iget() done | |
1703 | * earlier on. | |
1704 | */ | |
9403540c BN |
1705 | if (found->d_inode) { |
1706 | if (unlikely(found->d_inode != inode)) { | |
1707 | /* This can't happen because bad inodes are unhashed. */ | |
1708 | BUG_ON(!is_bad_inode(inode)); | |
1709 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1710 | } | |
9403540c BN |
1711 | iput(inode); |
1712 | return found; | |
1713 | } | |
b6520c81 | 1714 | |
9403540c BN |
1715 | /* |
1716 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1717 | * already has a dentry. |
9403540c | 1718 | */ |
873feea0 | 1719 | spin_lock(&inode->i_lock); |
b6520c81 | 1720 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1721 | __d_instantiate(found, inode); |
873feea0 | 1722 | spin_unlock(&inode->i_lock); |
9403540c BN |
1723 | security_d_instantiate(found, inode); |
1724 | return found; | |
1725 | } | |
b6520c81 | 1726 | |
9403540c | 1727 | /* |
b6520c81 CH |
1728 | * In case a directory already has a (disconnected) entry grab a |
1729 | * reference to it, move it in place and use it. | |
9403540c BN |
1730 | */ |
1731 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
dc0474be | 1732 | __dget(new); |
873feea0 | 1733 | spin_unlock(&inode->i_lock); |
9403540c | 1734 | security_d_instantiate(found, inode); |
9403540c | 1735 | d_move(new, found); |
9403540c | 1736 | iput(inode); |
9403540c | 1737 | dput(found); |
9403540c BN |
1738 | return new; |
1739 | ||
1740 | err_out: | |
1741 | iput(inode); | |
1742 | return ERR_PTR(error); | |
1743 | } | |
ec4f8605 | 1744 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1745 | |
31e6b01f NP |
1746 | /** |
1747 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1748 | * @parent: parent dentry | |
1749 | * @name: qstr of name we wish to find | |
1750 | * @seq: returns d_seq value at the point where the dentry was found | |
1751 | * @inode: returns dentry->d_inode when the inode was found valid. | |
1752 | * Returns: dentry, or NULL | |
1753 | * | |
1754 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1755 | * resolution (store-free path walking) design described in | |
1756 | * Documentation/filesystems/path-lookup.txt. | |
1757 | * | |
1758 | * This is not to be used outside core vfs. | |
1759 | * | |
1760 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1761 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1762 | * without taking d_lock and checking d_seq sequence count against @seq | |
1763 | * returned here. | |
1764 | * | |
1765 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1766 | * function. | |
1767 | * | |
1768 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1769 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1770 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1771 | * is formed, giving integrity down the path walk. | |
1772 | */ | |
1773 | struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name, | |
1774 | unsigned *seq, struct inode **inode) | |
1775 | { | |
1776 | unsigned int len = name->len; | |
1777 | unsigned int hash = name->hash; | |
1778 | const unsigned char *str = name->name; | |
b07ad996 | 1779 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1780 | struct hlist_bl_node *node; |
31e6b01f NP |
1781 | struct dentry *dentry; |
1782 | ||
1783 | /* | |
1784 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1785 | * required to prevent single threaded performance regressions | |
1786 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1787 | * Keep the two functions in sync. | |
1788 | */ | |
1789 | ||
1790 | /* | |
1791 | * The hash list is protected using RCU. | |
1792 | * | |
1793 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1794 | * races with d_move(). | |
1795 | * | |
1796 | * It is possible that concurrent renames can mess up our list | |
1797 | * walk here and result in missing our dentry, resulting in the | |
1798 | * false-negative result. d_lookup() protects against concurrent | |
1799 | * renames using rename_lock seqlock. | |
1800 | * | |
b0a4bb83 | 1801 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1802 | */ |
b07ad996 | 1803 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1804 | struct inode *i; |
1805 | const char *tname; | |
1806 | int tlen; | |
1807 | ||
1808 | if (dentry->d_name.hash != hash) | |
1809 | continue; | |
1810 | ||
1811 | seqretry: | |
1812 | *seq = read_seqcount_begin(&dentry->d_seq); | |
1813 | if (dentry->d_parent != parent) | |
1814 | continue; | |
1815 | if (d_unhashed(dentry)) | |
1816 | continue; | |
1817 | tlen = dentry->d_name.len; | |
1818 | tname = dentry->d_name.name; | |
1819 | i = dentry->d_inode; | |
e1bb5782 NP |
1820 | prefetch(tname); |
1821 | if (i) | |
1822 | prefetch(i); | |
31e6b01f NP |
1823 | /* |
1824 | * This seqcount check is required to ensure name and | |
1825 | * len are loaded atomically, so as not to walk off the | |
1826 | * edge of memory when walking. If we could load this | |
1827 | * atomically some other way, we could drop this check. | |
1828 | */ | |
1829 | if (read_seqcount_retry(&dentry->d_seq, *seq)) | |
1830 | goto seqretry; | |
fb045adb | 1831 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
31e6b01f NP |
1832 | if (parent->d_op->d_compare(parent, *inode, |
1833 | dentry, i, | |
1834 | tlen, tname, name)) | |
1835 | continue; | |
1836 | } else { | |
9d55c369 | 1837 | if (dentry_cmp(tname, tlen, str, len)) |
31e6b01f NP |
1838 | continue; |
1839 | } | |
1840 | /* | |
1841 | * No extra seqcount check is required after the name | |
1842 | * compare. The caller must perform a seqcount check in | |
1843 | * order to do anything useful with the returned dentry | |
1844 | * anyway. | |
1845 | */ | |
1846 | *inode = i; | |
1847 | return dentry; | |
1848 | } | |
1849 | return NULL; | |
1850 | } | |
1851 | ||
1da177e4 LT |
1852 | /** |
1853 | * d_lookup - search for a dentry | |
1854 | * @parent: parent dentry | |
1855 | * @name: qstr of name we wish to find | |
b04f784e | 1856 | * Returns: dentry, or NULL |
1da177e4 | 1857 | * |
b04f784e NP |
1858 | * d_lookup searches the children of the parent dentry for the name in |
1859 | * question. If the dentry is found its reference count is incremented and the | |
1860 | * dentry is returned. The caller must use dput to free the entry when it has | |
1861 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1862 | */ |
31e6b01f | 1863 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 | 1864 | { |
31e6b01f | 1865 | struct dentry *dentry; |
949854d0 | 1866 | unsigned seq; |
1da177e4 LT |
1867 | |
1868 | do { | |
1869 | seq = read_seqbegin(&rename_lock); | |
1870 | dentry = __d_lookup(parent, name); | |
1871 | if (dentry) | |
1872 | break; | |
1873 | } while (read_seqretry(&rename_lock, seq)); | |
1874 | return dentry; | |
1875 | } | |
ec4f8605 | 1876 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1877 | |
31e6b01f | 1878 | /** |
b04f784e NP |
1879 | * __d_lookup - search for a dentry (racy) |
1880 | * @parent: parent dentry | |
1881 | * @name: qstr of name we wish to find | |
1882 | * Returns: dentry, or NULL | |
1883 | * | |
1884 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1885 | * false-negative result due to unrelated rename activity. | |
1886 | * | |
1887 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1888 | * however it must be used carefully, eg. with a following d_lookup in | |
1889 | * the case of failure. | |
1890 | * | |
1891 | * __d_lookup callers must be commented. | |
1892 | */ | |
31e6b01f | 1893 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 LT |
1894 | { |
1895 | unsigned int len = name->len; | |
1896 | unsigned int hash = name->hash; | |
1897 | const unsigned char *str = name->name; | |
b07ad996 | 1898 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1899 | struct hlist_bl_node *node; |
31e6b01f | 1900 | struct dentry *found = NULL; |
665a7583 | 1901 | struct dentry *dentry; |
1da177e4 | 1902 | |
31e6b01f NP |
1903 | /* |
1904 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1905 | * required to prevent single threaded performance regressions | |
1906 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1907 | * Keep the two functions in sync. | |
1908 | */ | |
1909 | ||
b04f784e NP |
1910 | /* |
1911 | * The hash list is protected using RCU. | |
1912 | * | |
1913 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1914 | * with d_move(). | |
1915 | * | |
1916 | * It is possible that concurrent renames can mess up our list | |
1917 | * walk here and result in missing our dentry, resulting in the | |
1918 | * false-negative result. d_lookup() protects against concurrent | |
1919 | * renames using rename_lock seqlock. | |
1920 | * | |
b0a4bb83 | 1921 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1922 | */ |
1da177e4 LT |
1923 | rcu_read_lock(); |
1924 | ||
b07ad996 | 1925 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1926 | const char *tname; |
1927 | int tlen; | |
1da177e4 | 1928 | |
1da177e4 LT |
1929 | if (dentry->d_name.hash != hash) |
1930 | continue; | |
1da177e4 LT |
1931 | |
1932 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1933 | if (dentry->d_parent != parent) |
1934 | goto next; | |
d0185c08 LT |
1935 | if (d_unhashed(dentry)) |
1936 | goto next; | |
1937 | ||
1da177e4 LT |
1938 | /* |
1939 | * It is safe to compare names since d_move() cannot | |
1940 | * change the qstr (protected by d_lock). | |
1941 | */ | |
31e6b01f NP |
1942 | tlen = dentry->d_name.len; |
1943 | tname = dentry->d_name.name; | |
fb045adb | 1944 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
621e155a NP |
1945 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1946 | dentry, dentry->d_inode, | |
31e6b01f | 1947 | tlen, tname, name)) |
1da177e4 LT |
1948 | goto next; |
1949 | } else { | |
9d55c369 | 1950 | if (dentry_cmp(tname, tlen, str, len)) |
1da177e4 LT |
1951 | goto next; |
1952 | } | |
1953 | ||
b7ab39f6 | 1954 | dentry->d_count++; |
d0185c08 | 1955 | found = dentry; |
1da177e4 LT |
1956 | spin_unlock(&dentry->d_lock); |
1957 | break; | |
1958 | next: | |
1959 | spin_unlock(&dentry->d_lock); | |
1960 | } | |
1961 | rcu_read_unlock(); | |
1962 | ||
1963 | return found; | |
1964 | } | |
1965 | ||
3e7e241f EB |
1966 | /** |
1967 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1968 | * @dir: Directory to search in | |
1969 | * @name: qstr of name we wish to find | |
1970 | * | |
1971 | * On hash failure or on lookup failure NULL is returned. | |
1972 | */ | |
1973 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1974 | { | |
1975 | struct dentry *dentry = NULL; | |
1976 | ||
1977 | /* | |
1978 | * Check for a fs-specific hash function. Note that we must | |
1979 | * calculate the standard hash first, as the d_op->d_hash() | |
1980 | * routine may choose to leave the hash value unchanged. | |
1981 | */ | |
1982 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1983 | if (dir->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 1984 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1985 | goto out; |
1986 | } | |
1987 | dentry = d_lookup(dir, name); | |
1988 | out: | |
1989 | return dentry; | |
1990 | } | |
1991 | ||
1da177e4 | 1992 | /** |
786a5e15 | 1993 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 1994 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 1995 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
1996 | * |
1997 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1998 | * This is used by ncpfs in its readdir implementation. | |
1999 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2000 | * |
2001 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2002 | */ |
d3a23e16 | 2003 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2004 | { |
786a5e15 | 2005 | struct dentry *child; |
d3a23e16 | 2006 | |
2fd6b7f5 | 2007 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2008 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2009 | if (dentry == child) { | |
2fd6b7f5 | 2010 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2011 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2012 | spin_unlock(&dentry->d_lock); |
2013 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2014 | return 1; |
2015 | } | |
2016 | } | |
2fd6b7f5 | 2017 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2018 | |
1da177e4 LT |
2019 | return 0; |
2020 | } | |
ec4f8605 | 2021 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2022 | |
2023 | /* | |
2024 | * When a file is deleted, we have two options: | |
2025 | * - turn this dentry into a negative dentry | |
2026 | * - unhash this dentry and free it. | |
2027 | * | |
2028 | * Usually, we want to just turn this into | |
2029 | * a negative dentry, but if anybody else is | |
2030 | * currently using the dentry or the inode | |
2031 | * we can't do that and we fall back on removing | |
2032 | * it from the hash queues and waiting for | |
2033 | * it to be deleted later when it has no users | |
2034 | */ | |
2035 | ||
2036 | /** | |
2037 | * d_delete - delete a dentry | |
2038 | * @dentry: The dentry to delete | |
2039 | * | |
2040 | * Turn the dentry into a negative dentry if possible, otherwise | |
2041 | * remove it from the hash queues so it can be deleted later | |
2042 | */ | |
2043 | ||
2044 | void d_delete(struct dentry * dentry) | |
2045 | { | |
873feea0 | 2046 | struct inode *inode; |
7a91bf7f | 2047 | int isdir = 0; |
1da177e4 LT |
2048 | /* |
2049 | * Are we the only user? | |
2050 | */ | |
357f8e65 | 2051 | again: |
1da177e4 | 2052 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2053 | inode = dentry->d_inode; |
2054 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 2055 | if (dentry->d_count == 1) { |
873feea0 | 2056 | if (inode && !spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2057 | spin_unlock(&dentry->d_lock); |
2058 | cpu_relax(); | |
2059 | goto again; | |
2060 | } | |
13e3c5e5 | 2061 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2062 | dentry_unlink_inode(dentry); |
7a91bf7f | 2063 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2064 | return; |
2065 | } | |
2066 | ||
2067 | if (!d_unhashed(dentry)) | |
2068 | __d_drop(dentry); | |
2069 | ||
2070 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2071 | |
2072 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2073 | } |
ec4f8605 | 2074 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2075 | |
b07ad996 | 2076 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2077 | { |
ceb5bdc2 NP |
2078 | BUG_ON(!d_unhashed(entry)); |
2079 | spin_lock_bucket(b); | |
dea3667b | 2080 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2081 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
ceb5bdc2 | 2082 | spin_unlock_bucket(b); |
1da177e4 LT |
2083 | } |
2084 | ||
770bfad8 DH |
2085 | static void _d_rehash(struct dentry * entry) |
2086 | { | |
2087 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2088 | } | |
2089 | ||
1da177e4 LT |
2090 | /** |
2091 | * d_rehash - add an entry back to the hash | |
2092 | * @entry: dentry to add to the hash | |
2093 | * | |
2094 | * Adds a dentry to the hash according to its name. | |
2095 | */ | |
2096 | ||
2097 | void d_rehash(struct dentry * entry) | |
2098 | { | |
1da177e4 | 2099 | spin_lock(&entry->d_lock); |
770bfad8 | 2100 | _d_rehash(entry); |
1da177e4 | 2101 | spin_unlock(&entry->d_lock); |
1da177e4 | 2102 | } |
ec4f8605 | 2103 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2104 | |
fb2d5b86 NP |
2105 | /** |
2106 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2107 | * @dentry: dentry to be updated | |
2108 | * @name: new name | |
2109 | * | |
2110 | * Update a case insensitive dentry with new case of name. | |
2111 | * | |
2112 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2113 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2114 | * lengths). | |
2115 | * | |
2116 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2117 | * keep renames and concurrent inserts, and readdir(2) away). | |
2118 | */ | |
2119 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2120 | { | |
7ebfa57f | 2121 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2122 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2123 | ||
fb2d5b86 | 2124 | spin_lock(&dentry->d_lock); |
31e6b01f | 2125 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2126 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2127 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2128 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2129 | } |
2130 | EXPORT_SYMBOL(dentry_update_name_case); | |
2131 | ||
1da177e4 LT |
2132 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2133 | { | |
2134 | if (dname_external(target)) { | |
2135 | if (dname_external(dentry)) { | |
2136 | /* | |
2137 | * Both external: swap the pointers | |
2138 | */ | |
9a8d5bb4 | 2139 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2140 | } else { |
2141 | /* | |
2142 | * dentry:internal, target:external. Steal target's | |
2143 | * storage and make target internal. | |
2144 | */ | |
321bcf92 BF |
2145 | memcpy(target->d_iname, dentry->d_name.name, |
2146 | dentry->d_name.len + 1); | |
1da177e4 LT |
2147 | dentry->d_name.name = target->d_name.name; |
2148 | target->d_name.name = target->d_iname; | |
2149 | } | |
2150 | } else { | |
2151 | if (dname_external(dentry)) { | |
2152 | /* | |
2153 | * dentry:external, target:internal. Give dentry's | |
2154 | * storage to target and make dentry internal | |
2155 | */ | |
2156 | memcpy(dentry->d_iname, target->d_name.name, | |
2157 | target->d_name.len + 1); | |
2158 | target->d_name.name = dentry->d_name.name; | |
2159 | dentry->d_name.name = dentry->d_iname; | |
2160 | } else { | |
2161 | /* | |
2162 | * Both are internal. Just copy target to dentry | |
2163 | */ | |
2164 | memcpy(dentry->d_iname, target->d_name.name, | |
2165 | target->d_name.len + 1); | |
dc711ca3 AV |
2166 | dentry->d_name.len = target->d_name.len; |
2167 | return; | |
1da177e4 LT |
2168 | } |
2169 | } | |
9a8d5bb4 | 2170 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2171 | } |
2172 | ||
2fd6b7f5 NP |
2173 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2174 | { | |
2175 | /* | |
2176 | * XXXX: do we really need to take target->d_lock? | |
2177 | */ | |
2178 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2179 | spin_lock(&target->d_parent->d_lock); | |
2180 | else { | |
2181 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2182 | spin_lock(&dentry->d_parent->d_lock); | |
2183 | spin_lock_nested(&target->d_parent->d_lock, | |
2184 | DENTRY_D_LOCK_NESTED); | |
2185 | } else { | |
2186 | spin_lock(&target->d_parent->d_lock); | |
2187 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2188 | DENTRY_D_LOCK_NESTED); | |
2189 | } | |
2190 | } | |
2191 | if (target < dentry) { | |
2192 | spin_lock_nested(&target->d_lock, 2); | |
2193 | spin_lock_nested(&dentry->d_lock, 3); | |
2194 | } else { | |
2195 | spin_lock_nested(&dentry->d_lock, 2); | |
2196 | spin_lock_nested(&target->d_lock, 3); | |
2197 | } | |
2198 | } | |
2199 | ||
2200 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2201 | struct dentry *target) | |
2202 | { | |
2203 | if (target->d_parent != dentry->d_parent) | |
2204 | spin_unlock(&dentry->d_parent->d_lock); | |
2205 | if (target->d_parent != target) | |
2206 | spin_unlock(&target->d_parent->d_lock); | |
2207 | } | |
2208 | ||
1da177e4 | 2209 | /* |
2fd6b7f5 NP |
2210 | * When switching names, the actual string doesn't strictly have to |
2211 | * be preserved in the target - because we're dropping the target | |
2212 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2213 | * the new name before we switch. | |
2214 | * | |
2215 | * Note that we have to be a lot more careful about getting the hash | |
2216 | * switched - we have to switch the hash value properly even if it | |
2217 | * then no longer matches the actual (corrupted) string of the target. | |
2218 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2219 | */ |
9eaef27b | 2220 | /* |
b5c84bf6 | 2221 | * d_move - move a dentry |
1da177e4 LT |
2222 | * @dentry: entry to move |
2223 | * @target: new dentry | |
2224 | * | |
2225 | * Update the dcache to reflect the move of a file name. Negative | |
2226 | * dcache entries should not be moved in this way. | |
2227 | */ | |
b5c84bf6 | 2228 | void d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2229 | { |
1da177e4 LT |
2230 | if (!dentry->d_inode) |
2231 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2232 | ||
2fd6b7f5 NP |
2233 | BUG_ON(d_ancestor(dentry, target)); |
2234 | BUG_ON(d_ancestor(target, dentry)); | |
2235 | ||
1da177e4 | 2236 | write_seqlock(&rename_lock); |
2fd6b7f5 NP |
2237 | |
2238 | dentry_lock_for_move(dentry, target); | |
1da177e4 | 2239 | |
31e6b01f NP |
2240 | write_seqcount_begin(&dentry->d_seq); |
2241 | write_seqcount_begin(&target->d_seq); | |
2242 | ||
ceb5bdc2 NP |
2243 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2244 | ||
2245 | /* | |
2246 | * Move the dentry to the target hash queue. Don't bother checking | |
2247 | * for the same hash queue because of how unlikely it is. | |
2248 | */ | |
2249 | __d_drop(dentry); | |
789680d1 | 2250 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2251 | |
2252 | /* Unhash the target: dput() will then get rid of it */ | |
2253 | __d_drop(target); | |
2254 | ||
5160ee6f ED |
2255 | list_del(&dentry->d_u.d_child); |
2256 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2257 | |
2258 | /* Switch the names.. */ | |
2259 | switch_names(dentry, target); | |
9a8d5bb4 | 2260 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2261 | |
2262 | /* ... and switch the parents */ | |
2263 | if (IS_ROOT(dentry)) { | |
2264 | dentry->d_parent = target->d_parent; | |
2265 | target->d_parent = target; | |
5160ee6f | 2266 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2267 | } else { |
9a8d5bb4 | 2268 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2269 | |
2270 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2271 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2272 | } |
2273 | ||
5160ee6f | 2274 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2275 | |
31e6b01f NP |
2276 | write_seqcount_end(&target->d_seq); |
2277 | write_seqcount_end(&dentry->d_seq); | |
2278 | ||
2fd6b7f5 | 2279 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2280 | spin_unlock(&target->d_lock); |
c32ccd87 | 2281 | fsnotify_d_move(dentry); |
1da177e4 LT |
2282 | spin_unlock(&dentry->d_lock); |
2283 | write_sequnlock(&rename_lock); | |
9eaef27b | 2284 | } |
ec4f8605 | 2285 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2286 | |
e2761a11 OH |
2287 | /** |
2288 | * d_ancestor - search for an ancestor | |
2289 | * @p1: ancestor dentry | |
2290 | * @p2: child dentry | |
2291 | * | |
2292 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2293 | * an ancestor of p2, else NULL. | |
9eaef27b | 2294 | */ |
e2761a11 | 2295 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2296 | { |
2297 | struct dentry *p; | |
2298 | ||
871c0067 | 2299 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2300 | if (p->d_parent == p1) |
e2761a11 | 2301 | return p; |
9eaef27b | 2302 | } |
e2761a11 | 2303 | return NULL; |
9eaef27b TM |
2304 | } |
2305 | ||
2306 | /* | |
2307 | * This helper attempts to cope with remotely renamed directories | |
2308 | * | |
2309 | * It assumes that the caller is already holding | |
873feea0 | 2310 | * dentry->d_parent->d_inode->i_mutex and the inode->i_lock |
9eaef27b TM |
2311 | * |
2312 | * Note: If ever the locking in lock_rename() changes, then please | |
2313 | * remember to update this too... | |
9eaef27b | 2314 | */ |
873feea0 NP |
2315 | static struct dentry *__d_unalias(struct inode *inode, |
2316 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2317 | { |
2318 | struct mutex *m1 = NULL, *m2 = NULL; | |
2319 | struct dentry *ret; | |
2320 | ||
2321 | /* If alias and dentry share a parent, then no extra locks required */ | |
2322 | if (alias->d_parent == dentry->d_parent) | |
2323 | goto out_unalias; | |
2324 | ||
2325 | /* Check for loops */ | |
2326 | ret = ERR_PTR(-ELOOP); | |
e2761a11 | 2327 | if (d_ancestor(alias, dentry)) |
9eaef27b TM |
2328 | goto out_err; |
2329 | ||
2330 | /* See lock_rename() */ | |
2331 | ret = ERR_PTR(-EBUSY); | |
2332 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2333 | goto out_err; | |
2334 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2335 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2336 | goto out_err; | |
2337 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2338 | out_unalias: | |
b5c84bf6 | 2339 | d_move(alias, dentry); |
9eaef27b TM |
2340 | ret = alias; |
2341 | out_err: | |
873feea0 | 2342 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2343 | if (m2) |
2344 | mutex_unlock(m2); | |
2345 | if (m1) | |
2346 | mutex_unlock(m1); | |
2347 | return ret; | |
2348 | } | |
2349 | ||
770bfad8 DH |
2350 | /* |
2351 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2352 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2353 | * returns with anon->d_lock held! |
770bfad8 DH |
2354 | */ |
2355 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2356 | { | |
2357 | struct dentry *dparent, *aparent; | |
2358 | ||
2fd6b7f5 | 2359 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2360 | |
31e6b01f NP |
2361 | write_seqcount_begin(&dentry->d_seq); |
2362 | write_seqcount_begin(&anon->d_seq); | |
2363 | ||
770bfad8 DH |
2364 | dparent = dentry->d_parent; |
2365 | aparent = anon->d_parent; | |
2366 | ||
2fd6b7f5 NP |
2367 | switch_names(dentry, anon); |
2368 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2369 | ||
770bfad8 DH |
2370 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2371 | list_del(&dentry->d_u.d_child); | |
2372 | if (!IS_ROOT(dentry)) | |
2373 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2374 | else | |
2375 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2376 | ||
2377 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2378 | list_del(&anon->d_u.d_child); | |
2379 | if (!IS_ROOT(anon)) | |
2380 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2381 | else | |
2382 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2383 | ||
31e6b01f NP |
2384 | write_seqcount_end(&dentry->d_seq); |
2385 | write_seqcount_end(&anon->d_seq); | |
2386 | ||
2fd6b7f5 NP |
2387 | dentry_unlock_parents_for_move(anon, dentry); |
2388 | spin_unlock(&dentry->d_lock); | |
2389 | ||
2390 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2391 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2392 | } | |
2393 | ||
2394 | /** | |
2395 | * d_materialise_unique - introduce an inode into the tree | |
2396 | * @dentry: candidate dentry | |
2397 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2398 | * | |
2399 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2400 | * root directory alias in its place if there is one | |
2401 | */ | |
2402 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2403 | { | |
9eaef27b | 2404 | struct dentry *actual; |
770bfad8 DH |
2405 | |
2406 | BUG_ON(!d_unhashed(dentry)); | |
2407 | ||
770bfad8 DH |
2408 | if (!inode) { |
2409 | actual = dentry; | |
360da900 | 2410 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2411 | d_rehash(actual); |
2412 | goto out_nolock; | |
770bfad8 DH |
2413 | } |
2414 | ||
873feea0 | 2415 | spin_lock(&inode->i_lock); |
357f8e65 | 2416 | |
9eaef27b TM |
2417 | if (S_ISDIR(inode->i_mode)) { |
2418 | struct dentry *alias; | |
2419 | ||
2420 | /* Does an aliased dentry already exist? */ | |
2421 | alias = __d_find_alias(inode, 0); | |
2422 | if (alias) { | |
2423 | actual = alias; | |
2424 | /* Is this an anonymous mountpoint that we could splice | |
2425 | * into our tree? */ | |
2426 | if (IS_ROOT(alias)) { | |
9eaef27b TM |
2427 | __d_materialise_dentry(dentry, alias); |
2428 | __d_drop(alias); | |
2429 | goto found; | |
2430 | } | |
2431 | /* Nope, but we must(!) avoid directory aliasing */ | |
873feea0 | 2432 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b TM |
2433 | if (IS_ERR(actual)) |
2434 | dput(alias); | |
2435 | goto out_nolock; | |
2436 | } | |
770bfad8 DH |
2437 | } |
2438 | ||
2439 | /* Add a unique reference */ | |
2440 | actual = __d_instantiate_unique(dentry, inode); | |
2441 | if (!actual) | |
2442 | actual = dentry; | |
357f8e65 NP |
2443 | else |
2444 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2445 | |
770bfad8 DH |
2446 | spin_lock(&actual->d_lock); |
2447 | found: | |
2448 | _d_rehash(actual); | |
2449 | spin_unlock(&actual->d_lock); | |
873feea0 | 2450 | spin_unlock(&inode->i_lock); |
9eaef27b | 2451 | out_nolock: |
770bfad8 DH |
2452 | if (actual == dentry) { |
2453 | security_d_instantiate(dentry, inode); | |
2454 | return NULL; | |
2455 | } | |
2456 | ||
2457 | iput(inode); | |
2458 | return actual; | |
770bfad8 | 2459 | } |
ec4f8605 | 2460 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2461 | |
cdd16d02 | 2462 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2463 | { |
2464 | *buflen -= namelen; | |
2465 | if (*buflen < 0) | |
2466 | return -ENAMETOOLONG; | |
2467 | *buffer -= namelen; | |
2468 | memcpy(*buffer, str, namelen); | |
2469 | return 0; | |
2470 | } | |
2471 | ||
cdd16d02 MS |
2472 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2473 | { | |
2474 | return prepend(buffer, buflen, name->name, name->len); | |
2475 | } | |
2476 | ||
1da177e4 | 2477 | /** |
208898c1 | 2478 | * prepend_path - Prepend path string to a buffer |
9d1bc601 MS |
2479 | * @path: the dentry/vfsmount to report |
2480 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2481 | * @buffer: pointer to the end of the buffer |
2482 | * @buflen: pointer to buffer length | |
552ce544 | 2483 | * |
949854d0 | 2484 | * Caller holds the rename_lock. |
9d1bc601 MS |
2485 | * |
2486 | * If path is not reachable from the supplied root, then the value of | |
2487 | * root is changed (without modifying refcounts). | |
1da177e4 | 2488 | */ |
f2eb6575 MS |
2489 | static int prepend_path(const struct path *path, struct path *root, |
2490 | char **buffer, int *buflen) | |
1da177e4 | 2491 | { |
9d1bc601 MS |
2492 | struct dentry *dentry = path->dentry; |
2493 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2494 | bool slash = false; |
2495 | int error = 0; | |
6092d048 | 2496 | |
99b7db7b | 2497 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2498 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2499 | struct dentry * parent; |
2500 | ||
1da177e4 | 2501 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2502 | /* Global root? */ |
1da177e4 | 2503 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2504 | goto global_root; |
2505 | } | |
2506 | dentry = vfsmnt->mnt_mountpoint; | |
2507 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2508 | continue; |
2509 | } | |
2510 | parent = dentry->d_parent; | |
2511 | prefetch(parent); | |
9abca360 | 2512 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2513 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2514 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2515 | if (!error) |
2516 | error = prepend(buffer, buflen, "/", 1); | |
2517 | if (error) | |
2518 | break; | |
2519 | ||
2520 | slash = true; | |
1da177e4 LT |
2521 | dentry = parent; |
2522 | } | |
2523 | ||
be285c71 | 2524 | out: |
f2eb6575 MS |
2525 | if (!error && !slash) |
2526 | error = prepend(buffer, buflen, "/", 1); | |
2527 | ||
99b7db7b | 2528 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2529 | return error; |
1da177e4 LT |
2530 | |
2531 | global_root: | |
98dc568b MS |
2532 | /* |
2533 | * Filesystems needing to implement special "root names" | |
2534 | * should do so with ->d_dname() | |
2535 | */ | |
2536 | if (IS_ROOT(dentry) && | |
2537 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2538 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2539 | (int) dentry->d_name.len, dentry->d_name.name); | |
2540 | } | |
9d1bc601 MS |
2541 | root->mnt = vfsmnt; |
2542 | root->dentry = dentry; | |
be285c71 | 2543 | goto out; |
f2eb6575 | 2544 | } |
be285c71 | 2545 | |
f2eb6575 MS |
2546 | /** |
2547 | * __d_path - return the path of a dentry | |
2548 | * @path: the dentry/vfsmount to report | |
2549 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2550 | * @buf: buffer to return value in |
f2eb6575 MS |
2551 | * @buflen: buffer length |
2552 | * | |
ffd1f4ed | 2553 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2554 | * |
2555 | * Returns a pointer into the buffer or an error code if the | |
2556 | * path was too long. | |
2557 | * | |
be148247 | 2558 | * "buflen" should be positive. |
f2eb6575 MS |
2559 | * |
2560 | * If path is not reachable from the supplied root, then the value of | |
2561 | * root is changed (without modifying refcounts). | |
2562 | */ | |
2563 | char *__d_path(const struct path *path, struct path *root, | |
2564 | char *buf, int buflen) | |
2565 | { | |
2566 | char *res = buf + buflen; | |
2567 | int error; | |
2568 | ||
2569 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2570 | write_seqlock(&rename_lock); |
f2eb6575 | 2571 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2572 | write_sequnlock(&rename_lock); |
be148247 | 2573 | |
f2eb6575 MS |
2574 | if (error) |
2575 | return ERR_PTR(error); | |
f2eb6575 | 2576 | return res; |
1da177e4 LT |
2577 | } |
2578 | ||
ffd1f4ed MS |
2579 | /* |
2580 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2581 | */ | |
2582 | static int path_with_deleted(const struct path *path, struct path *root, | |
2583 | char **buf, int *buflen) | |
2584 | { | |
2585 | prepend(buf, buflen, "\0", 1); | |
2586 | if (d_unlinked(path->dentry)) { | |
2587 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2588 | if (error) | |
2589 | return error; | |
2590 | } | |
2591 | ||
2592 | return prepend_path(path, root, buf, buflen); | |
2593 | } | |
2594 | ||
8df9d1a4 MS |
2595 | static int prepend_unreachable(char **buffer, int *buflen) |
2596 | { | |
2597 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2598 | } | |
2599 | ||
a03a8a70 JB |
2600 | /** |
2601 | * d_path - return the path of a dentry | |
cf28b486 | 2602 | * @path: path to report |
a03a8a70 JB |
2603 | * @buf: buffer to return value in |
2604 | * @buflen: buffer length | |
2605 | * | |
2606 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2607 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2608 | * | |
52afeefb AV |
2609 | * Returns a pointer into the buffer or an error code if the path was |
2610 | * too long. Note: Callers should use the returned pointer, not the passed | |
2611 | * in buffer, to use the name! The implementation often starts at an offset | |
2612 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2613 | * |
31f3e0b3 | 2614 | * "buflen" should be positive. |
a03a8a70 | 2615 | */ |
20d4fdc1 | 2616 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2617 | { |
ffd1f4ed | 2618 | char *res = buf + buflen; |
6ac08c39 | 2619 | struct path root; |
9d1bc601 | 2620 | struct path tmp; |
ffd1f4ed | 2621 | int error; |
1da177e4 | 2622 | |
c23fbb6b ED |
2623 | /* |
2624 | * We have various synthetic filesystems that never get mounted. On | |
2625 | * these filesystems dentries are never used for lookup purposes, and | |
2626 | * thus don't need to be hashed. They also don't need a name until a | |
2627 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2628 | * below allows us to generate a name for these objects on demand: | |
2629 | */ | |
cf28b486 JB |
2630 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2631 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2632 | |
f7ad3c6b | 2633 | get_fs_root(current->fs, &root); |
949854d0 | 2634 | write_seqlock(&rename_lock); |
9d1bc601 | 2635 | tmp = root; |
ffd1f4ed MS |
2636 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2637 | if (error) | |
2638 | res = ERR_PTR(error); | |
949854d0 | 2639 | write_sequnlock(&rename_lock); |
6ac08c39 | 2640 | path_put(&root); |
1da177e4 LT |
2641 | return res; |
2642 | } | |
ec4f8605 | 2643 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2644 | |
8df9d1a4 MS |
2645 | /** |
2646 | * d_path_with_unreachable - return the path of a dentry | |
2647 | * @path: path to report | |
2648 | * @buf: buffer to return value in | |
2649 | * @buflen: buffer length | |
2650 | * | |
2651 | * The difference from d_path() is that this prepends "(unreachable)" | |
2652 | * to paths which are unreachable from the current process' root. | |
2653 | */ | |
2654 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2655 | { | |
2656 | char *res = buf + buflen; | |
2657 | struct path root; | |
2658 | struct path tmp; | |
2659 | int error; | |
2660 | ||
2661 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2662 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2663 | ||
2664 | get_fs_root(current->fs, &root); | |
949854d0 | 2665 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2666 | tmp = root; |
2667 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2668 | if (!error && !path_equal(&tmp, &root)) | |
2669 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2670 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2671 | path_put(&root); |
2672 | if (error) | |
2673 | res = ERR_PTR(error); | |
2674 | ||
2675 | return res; | |
2676 | } | |
2677 | ||
c23fbb6b ED |
2678 | /* |
2679 | * Helper function for dentry_operations.d_dname() members | |
2680 | */ | |
2681 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2682 | const char *fmt, ...) | |
2683 | { | |
2684 | va_list args; | |
2685 | char temp[64]; | |
2686 | int sz; | |
2687 | ||
2688 | va_start(args, fmt); | |
2689 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2690 | va_end(args); | |
2691 | ||
2692 | if (sz > sizeof(temp) || sz > buflen) | |
2693 | return ERR_PTR(-ENAMETOOLONG); | |
2694 | ||
2695 | buffer += buflen - sz; | |
2696 | return memcpy(buffer, temp, sz); | |
2697 | } | |
2698 | ||
6092d048 RP |
2699 | /* |
2700 | * Write full pathname from the root of the filesystem into the buffer. | |
2701 | */ | |
ec2447c2 | 2702 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2703 | { |
2704 | char *end = buf + buflen; | |
2705 | char *retval; | |
2706 | ||
6092d048 | 2707 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2708 | if (buflen < 1) |
2709 | goto Elong; | |
2710 | /* Get '/' right */ | |
2711 | retval = end-1; | |
2712 | *retval = '/'; | |
2713 | ||
cdd16d02 MS |
2714 | while (!IS_ROOT(dentry)) { |
2715 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2716 | int error; |
6092d048 | 2717 | |
6092d048 | 2718 | prefetch(parent); |
9abca360 NP |
2719 | spin_lock(&dentry->d_lock); |
2720 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2721 | spin_unlock(&dentry->d_lock); | |
2722 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2723 | goto Elong; |
2724 | ||
2725 | retval = end; | |
2726 | dentry = parent; | |
2727 | } | |
c103135c AV |
2728 | return retval; |
2729 | Elong: | |
2730 | return ERR_PTR(-ENAMETOOLONG); | |
2731 | } | |
ec2447c2 NP |
2732 | |
2733 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2734 | { | |
2735 | char *retval; | |
2736 | ||
949854d0 | 2737 | write_seqlock(&rename_lock); |
ec2447c2 | 2738 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2739 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2740 | |
2741 | return retval; | |
2742 | } | |
2743 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2744 | |
2745 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2746 | { | |
2747 | char *p = NULL; | |
2748 | char *retval; | |
2749 | ||
949854d0 | 2750 | write_seqlock(&rename_lock); |
c103135c AV |
2751 | if (d_unlinked(dentry)) { |
2752 | p = buf + buflen; | |
2753 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2754 | goto Elong; | |
2755 | buflen++; | |
2756 | } | |
2757 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2758 | write_sequnlock(&rename_lock); |
c103135c AV |
2759 | if (!IS_ERR(retval) && p) |
2760 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2761 | return retval; |
2762 | Elong: | |
6092d048 RP |
2763 | return ERR_PTR(-ENAMETOOLONG); |
2764 | } | |
2765 | ||
1da177e4 LT |
2766 | /* |
2767 | * NOTE! The user-level library version returns a | |
2768 | * character pointer. The kernel system call just | |
2769 | * returns the length of the buffer filled (which | |
2770 | * includes the ending '\0' character), or a negative | |
2771 | * error value. So libc would do something like | |
2772 | * | |
2773 | * char *getcwd(char * buf, size_t size) | |
2774 | * { | |
2775 | * int retval; | |
2776 | * | |
2777 | * retval = sys_getcwd(buf, size); | |
2778 | * if (retval >= 0) | |
2779 | * return buf; | |
2780 | * errno = -retval; | |
2781 | * return NULL; | |
2782 | * } | |
2783 | */ | |
3cdad428 | 2784 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2785 | { |
552ce544 | 2786 | int error; |
6ac08c39 | 2787 | struct path pwd, root; |
552ce544 | 2788 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2789 | |
2790 | if (!page) | |
2791 | return -ENOMEM; | |
2792 | ||
f7ad3c6b | 2793 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2794 | |
552ce544 | 2795 | error = -ENOENT; |
949854d0 | 2796 | write_seqlock(&rename_lock); |
f3da392e | 2797 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2798 | unsigned long len; |
9d1bc601 | 2799 | struct path tmp = root; |
8df9d1a4 MS |
2800 | char *cwd = page + PAGE_SIZE; |
2801 | int buflen = PAGE_SIZE; | |
1da177e4 | 2802 | |
8df9d1a4 MS |
2803 | prepend(&cwd, &buflen, "\0", 1); |
2804 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2805 | write_sequnlock(&rename_lock); |
552ce544 | 2806 | |
8df9d1a4 | 2807 | if (error) |
552ce544 LT |
2808 | goto out; |
2809 | ||
8df9d1a4 MS |
2810 | /* Unreachable from current root */ |
2811 | if (!path_equal(&tmp, &root)) { | |
2812 | error = prepend_unreachable(&cwd, &buflen); | |
2813 | if (error) | |
2814 | goto out; | |
2815 | } | |
2816 | ||
552ce544 LT |
2817 | error = -ERANGE; |
2818 | len = PAGE_SIZE + page - cwd; | |
2819 | if (len <= size) { | |
2820 | error = len; | |
2821 | if (copy_to_user(buf, cwd, len)) | |
2822 | error = -EFAULT; | |
2823 | } | |
949854d0 NP |
2824 | } else { |
2825 | write_sequnlock(&rename_lock); | |
949854d0 | 2826 | } |
1da177e4 LT |
2827 | |
2828 | out: | |
6ac08c39 JB |
2829 | path_put(&pwd); |
2830 | path_put(&root); | |
1da177e4 LT |
2831 | free_page((unsigned long) page); |
2832 | return error; | |
2833 | } | |
2834 | ||
2835 | /* | |
2836 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2837 | * | |
2838 | * Trivially implemented using the dcache structure | |
2839 | */ | |
2840 | ||
2841 | /** | |
2842 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2843 | * @new_dentry: new dentry | |
2844 | * @old_dentry: old dentry | |
2845 | * | |
2846 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2847 | * Returns 0 otherwise. | |
2848 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2849 | */ | |
2850 | ||
e2761a11 | 2851 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2852 | { |
2853 | int result; | |
949854d0 | 2854 | unsigned seq; |
1da177e4 | 2855 | |
e2761a11 OH |
2856 | if (new_dentry == old_dentry) |
2857 | return 1; | |
2858 | ||
e2761a11 | 2859 | do { |
1da177e4 | 2860 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2861 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2862 | /* |
2863 | * Need rcu_readlock to protect against the d_parent trashing | |
2864 | * due to d_move | |
2865 | */ | |
2866 | rcu_read_lock(); | |
e2761a11 | 2867 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2868 | result = 1; |
e2761a11 OH |
2869 | else |
2870 | result = 0; | |
949854d0 | 2871 | rcu_read_unlock(); |
1da177e4 | 2872 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2873 | |
2874 | return result; | |
2875 | } | |
2876 | ||
2096f759 AV |
2877 | int path_is_under(struct path *path1, struct path *path2) |
2878 | { | |
2879 | struct vfsmount *mnt = path1->mnt; | |
2880 | struct dentry *dentry = path1->dentry; | |
2881 | int res; | |
99b7db7b NP |
2882 | |
2883 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2884 | if (mnt != path2->mnt) { |
2885 | for (;;) { | |
2886 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2887 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2888 | return 0; |
2889 | } | |
2890 | if (mnt->mnt_parent == path2->mnt) | |
2891 | break; | |
2892 | mnt = mnt->mnt_parent; | |
2893 | } | |
2894 | dentry = mnt->mnt_mountpoint; | |
2895 | } | |
2896 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2897 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2898 | return res; |
2899 | } | |
2900 | EXPORT_SYMBOL(path_is_under); | |
2901 | ||
1da177e4 LT |
2902 | void d_genocide(struct dentry *root) |
2903 | { | |
949854d0 | 2904 | struct dentry *this_parent; |
1da177e4 | 2905 | struct list_head *next; |
949854d0 | 2906 | unsigned seq; |
58db63d0 | 2907 | int locked = 0; |
1da177e4 | 2908 | |
949854d0 | 2909 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2910 | again: |
2911 | this_parent = root; | |
2fd6b7f5 | 2912 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2913 | repeat: |
2914 | next = this_parent->d_subdirs.next; | |
2915 | resume: | |
2916 | while (next != &this_parent->d_subdirs) { | |
2917 | struct list_head *tmp = next; | |
5160ee6f | 2918 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2919 | next = tmp->next; |
949854d0 | 2920 | |
da502956 NP |
2921 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2922 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2923 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2924 | continue; |
da502956 | 2925 | } |
1da177e4 | 2926 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2927 | spin_unlock(&this_parent->d_lock); |
2928 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2929 | this_parent = dentry; |
2fd6b7f5 | 2930 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2931 | goto repeat; |
2932 | } | |
949854d0 NP |
2933 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2934 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2935 | dentry->d_count--; | |
2936 | } | |
b7ab39f6 | 2937 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2938 | } |
2939 | if (this_parent != root) { | |
c826cb7d | 2940 | struct dentry *child = this_parent; |
949854d0 NP |
2941 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2942 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2943 | this_parent->d_count--; | |
2944 | } | |
c826cb7d LT |
2945 | this_parent = try_to_ascend(this_parent, locked, seq); |
2946 | if (!this_parent) | |
949854d0 | 2947 | goto rename_retry; |
949854d0 | 2948 | next = child->d_u.d_child.next; |
1da177e4 LT |
2949 | goto resume; |
2950 | } | |
2fd6b7f5 | 2951 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2952 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2953 | goto rename_retry; |
58db63d0 NP |
2954 | if (locked) |
2955 | write_sequnlock(&rename_lock); | |
2956 | return; | |
2957 | ||
2958 | rename_retry: | |
2959 | locked = 1; | |
2960 | write_seqlock(&rename_lock); | |
2961 | goto again; | |
1da177e4 LT |
2962 | } |
2963 | ||
2964 | /** | |
2965 | * find_inode_number - check for dentry with name | |
2966 | * @dir: directory to check | |
2967 | * @name: Name to find. | |
2968 | * | |
2969 | * Check whether a dentry already exists for the given name, | |
2970 | * and return the inode number if it has an inode. Otherwise | |
2971 | * 0 is returned. | |
2972 | * | |
2973 | * This routine is used to post-process directory listings for | |
2974 | * filesystems using synthetic inode numbers, and is necessary | |
2975 | * to keep getcwd() working. | |
2976 | */ | |
2977 | ||
2978 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2979 | { | |
2980 | struct dentry * dentry; | |
2981 | ino_t ino = 0; | |
2982 | ||
3e7e241f EB |
2983 | dentry = d_hash_and_lookup(dir, name); |
2984 | if (dentry) { | |
1da177e4 LT |
2985 | if (dentry->d_inode) |
2986 | ino = dentry->d_inode->i_ino; | |
2987 | dput(dentry); | |
2988 | } | |
1da177e4 LT |
2989 | return ino; |
2990 | } | |
ec4f8605 | 2991 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2992 | |
2993 | static __initdata unsigned long dhash_entries; | |
2994 | static int __init set_dhash_entries(char *str) | |
2995 | { | |
2996 | if (!str) | |
2997 | return 0; | |
2998 | dhash_entries = simple_strtoul(str, &str, 0); | |
2999 | return 1; | |
3000 | } | |
3001 | __setup("dhash_entries=", set_dhash_entries); | |
3002 | ||
3003 | static void __init dcache_init_early(void) | |
3004 | { | |
3005 | int loop; | |
3006 | ||
3007 | /* If hashes are distributed across NUMA nodes, defer | |
3008 | * hash allocation until vmalloc space is available. | |
3009 | */ | |
3010 | if (hashdist) | |
3011 | return; | |
3012 | ||
3013 | dentry_hashtable = | |
3014 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3015 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3016 | dhash_entries, |
3017 | 13, | |
3018 | HASH_EARLY, | |
3019 | &d_hash_shift, | |
3020 | &d_hash_mask, | |
3021 | 0); | |
3022 | ||
3023 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 3024 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3025 | } |
3026 | ||
74bf17cf | 3027 | static void __init dcache_init(void) |
1da177e4 LT |
3028 | { |
3029 | int loop; | |
3030 | ||
3031 | /* | |
3032 | * A constructor could be added for stable state like the lists, | |
3033 | * but it is probably not worth it because of the cache nature | |
3034 | * of the dcache. | |
3035 | */ | |
0a31bd5f CL |
3036 | dentry_cache = KMEM_CACHE(dentry, |
3037 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 3038 | |
8e1f936b | 3039 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
3040 | |
3041 | /* Hash may have been set up in dcache_init_early */ | |
3042 | if (!hashdist) | |
3043 | return; | |
3044 | ||
3045 | dentry_hashtable = | |
3046 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3047 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3048 | dhash_entries, |
3049 | 13, | |
3050 | 0, | |
3051 | &d_hash_shift, | |
3052 | &d_hash_mask, | |
3053 | 0); | |
3054 | ||
3055 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 3056 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3057 | } |
3058 | ||
3059 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3060 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3061 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3062 | |
1da177e4 LT |
3063 | EXPORT_SYMBOL(d_genocide); |
3064 | ||
1da177e4 LT |
3065 | void __init vfs_caches_init_early(void) |
3066 | { | |
3067 | dcache_init_early(); | |
3068 | inode_init_early(); | |
3069 | } | |
3070 | ||
3071 | void __init vfs_caches_init(unsigned long mempages) | |
3072 | { | |
3073 | unsigned long reserve; | |
3074 | ||
3075 | /* Base hash sizes on available memory, with a reserve equal to | |
3076 | 150% of current kernel size */ | |
3077 | ||
3078 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3079 | mempages -= reserve; | |
3080 | ||
3081 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3082 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3083 | |
74bf17cf DC |
3084 | dcache_init(); |
3085 | inode_init(); | |
1da177e4 | 3086 | files_init(mempages); |
74bf17cf | 3087 | mnt_init(); |
1da177e4 LT |
3088 | bdev_cache_init(); |
3089 | chrdev_init(); | |
3090 | } |