]>
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> | |
268bb0ce | 38 | #include <linux/prefetch.h> |
07f3f05c | 39 | #include "internal.h" |
1da177e4 | 40 | |
789680d1 NP |
41 | /* |
42 | * Usage: | |
873feea0 NP |
43 | * dcache->d_inode->i_lock protects: |
44 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
45 | * dcache_hash_bucket lock protects: |
46 | * - the dcache hash table | |
47 | * s_anon bl list spinlock protects: | |
48 | * - the s_anon list (see __d_drop) | |
23044507 NP |
49 | * dcache_lru_lock protects: |
50 | * - the dcache lru lists and counters | |
51 | * d_lock protects: | |
52 | * - d_flags | |
53 | * - d_name | |
54 | * - d_lru | |
b7ab39f6 | 55 | * - d_count |
da502956 | 56 | * - d_unhashed() |
2fd6b7f5 NP |
57 | * - d_parent and d_subdirs |
58 | * - childrens' d_child and d_parent | |
b23fb0a6 | 59 | * - d_alias, d_inode |
789680d1 NP |
60 | * |
61 | * Ordering: | |
873feea0 | 62 | * dentry->d_inode->i_lock |
b5c84bf6 NP |
63 | * dentry->d_lock |
64 | * dcache_lru_lock | |
ceb5bdc2 NP |
65 | * dcache_hash_bucket lock |
66 | * s_anon lock | |
789680d1 | 67 | * |
da502956 NP |
68 | * If there is an ancestor relationship: |
69 | * dentry->d_parent->...->d_parent->d_lock | |
70 | * ... | |
71 | * dentry->d_parent->d_lock | |
72 | * dentry->d_lock | |
73 | * | |
74 | * If no ancestor relationship: | |
789680d1 NP |
75 | * if (dentry1 < dentry2) |
76 | * dentry1->d_lock | |
77 | * dentry2->d_lock | |
78 | */ | |
fa3536cc | 79 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
80 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
81 | ||
23044507 | 82 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
74c3cbe3 | 83 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 84 | |
949854d0 | 85 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 86 | |
e18b890b | 87 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 88 | |
1da177e4 LT |
89 | /* |
90 | * This is the single most critical data structure when it comes | |
91 | * to the dcache: the hashtable for lookups. Somebody should try | |
92 | * to make this good - I've just made it work. | |
93 | * | |
94 | * This hash-function tries to avoid losing too many bits of hash | |
95 | * information, yet avoid using a prime hash-size or similar. | |
96 | */ | |
97 | #define D_HASHBITS d_hash_shift | |
98 | #define D_HASHMASK d_hash_mask | |
99 | ||
fa3536cc ED |
100 | static unsigned int d_hash_mask __read_mostly; |
101 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 102 | |
b07ad996 | 103 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 104 | |
b07ad996 | 105 | static inline struct hlist_bl_head *d_hash(struct dentry *parent, |
ceb5bdc2 NP |
106 | unsigned long hash) |
107 | { | |
108 | hash += ((unsigned long) parent ^ GOLDEN_RATIO_PRIME) / L1_CACHE_BYTES; | |
109 | hash = hash ^ ((hash ^ GOLDEN_RATIO_PRIME) >> D_HASHBITS); | |
110 | return dentry_hashtable + (hash & D_HASHMASK); | |
111 | } | |
112 | ||
1da177e4 LT |
113 | /* Statistics gathering. */ |
114 | struct dentry_stat_t dentry_stat = { | |
115 | .age_limit = 45, | |
116 | }; | |
117 | ||
3e880fb5 | 118 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
119 | |
120 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
121 | static int get_nr_dentry(void) |
122 | { | |
123 | int i; | |
124 | int sum = 0; | |
125 | for_each_possible_cpu(i) | |
126 | sum += per_cpu(nr_dentry, i); | |
127 | return sum < 0 ? 0 : sum; | |
128 | } | |
129 | ||
312d3ca8 CH |
130 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
131 | size_t *lenp, loff_t *ppos) | |
132 | { | |
3e880fb5 | 133 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
134 | return proc_dointvec(table, write, buffer, lenp, ppos); |
135 | } | |
136 | #endif | |
137 | ||
9c82ab9c | 138 | static void __d_free(struct rcu_head *head) |
1da177e4 | 139 | { |
9c82ab9c CH |
140 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
141 | ||
fd217f4d | 142 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
143 | if (dname_external(dentry)) |
144 | kfree(dentry->d_name.name); | |
145 | kmem_cache_free(dentry_cache, dentry); | |
146 | } | |
147 | ||
148 | /* | |
b5c84bf6 | 149 | * no locks, please. |
1da177e4 LT |
150 | */ |
151 | static void d_free(struct dentry *dentry) | |
152 | { | |
b7ab39f6 | 153 | BUG_ON(dentry->d_count); |
3e880fb5 | 154 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
155 | if (dentry->d_op && dentry->d_op->d_release) |
156 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 157 | |
dea3667b LT |
158 | /* if dentry was never visible to RCU, immediate free is OK */ |
159 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
9c82ab9c | 160 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 161 | else |
9c82ab9c | 162 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
163 | } |
164 | ||
31e6b01f NP |
165 | /** |
166 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 167 | * @dentry: the target dentry |
31e6b01f NP |
168 | * After this call, in-progress rcu-walk path lookup will fail. This |
169 | * should be called after unhashing, and after changing d_inode (if | |
170 | * the dentry has not already been unhashed). | |
171 | */ | |
172 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
173 | { | |
174 | assert_spin_locked(&dentry->d_lock); | |
175 | /* Go through a barrier */ | |
176 | write_seqcount_barrier(&dentry->d_seq); | |
177 | } | |
178 | ||
1da177e4 LT |
179 | /* |
180 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
181 | * d_iput() operation if defined. Dentry has no refcount |
182 | * and is unhashed. | |
1da177e4 | 183 | */ |
858119e1 | 184 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 185 | __releases(dentry->d_lock) |
873feea0 | 186 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
187 | { |
188 | struct inode *inode = dentry->d_inode; | |
189 | if (inode) { | |
190 | dentry->d_inode = NULL; | |
191 | list_del_init(&dentry->d_alias); | |
192 | spin_unlock(&dentry->d_lock); | |
873feea0 | 193 | spin_unlock(&inode->i_lock); |
f805fbda LT |
194 | if (!inode->i_nlink) |
195 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
196 | if (dentry->d_op && dentry->d_op->d_iput) |
197 | dentry->d_op->d_iput(dentry, inode); | |
198 | else | |
199 | iput(inode); | |
200 | } else { | |
201 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
202 | } |
203 | } | |
204 | ||
31e6b01f NP |
205 | /* |
206 | * Release the dentry's inode, using the filesystem | |
207 | * d_iput() operation if defined. dentry remains in-use. | |
208 | */ | |
209 | static void dentry_unlink_inode(struct dentry * dentry) | |
210 | __releases(dentry->d_lock) | |
873feea0 | 211 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
212 | { |
213 | struct inode *inode = dentry->d_inode; | |
214 | dentry->d_inode = NULL; | |
215 | list_del_init(&dentry->d_alias); | |
216 | dentry_rcuwalk_barrier(dentry); | |
217 | spin_unlock(&dentry->d_lock); | |
873feea0 | 218 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
219 | if (!inode->i_nlink) |
220 | fsnotify_inoderemove(inode); | |
221 | if (dentry->d_op && dentry->d_op->d_iput) | |
222 | dentry->d_op->d_iput(dentry, inode); | |
223 | else | |
224 | iput(inode); | |
225 | } | |
226 | ||
da3bbdd4 | 227 | /* |
23044507 | 228 | * dentry_lru_(add|del|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
229 | */ |
230 | static void dentry_lru_add(struct dentry *dentry) | |
231 | { | |
a4633357 | 232 | if (list_empty(&dentry->d_lru)) { |
23044507 | 233 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
234 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
235 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 236 | dentry_stat.nr_unused++; |
23044507 | 237 | spin_unlock(&dcache_lru_lock); |
a4633357 | 238 | } |
da3bbdd4 KM |
239 | } |
240 | ||
23044507 NP |
241 | static void __dentry_lru_del(struct dentry *dentry) |
242 | { | |
243 | list_del_init(&dentry->d_lru); | |
244 | dentry->d_sb->s_nr_dentry_unused--; | |
245 | dentry_stat.nr_unused--; | |
246 | } | |
247 | ||
da3bbdd4 KM |
248 | static void dentry_lru_del(struct dentry *dentry) |
249 | { | |
250 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
251 | spin_lock(&dcache_lru_lock); |
252 | __dentry_lru_del(dentry); | |
253 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
254 | } |
255 | } | |
256 | ||
a4633357 | 257 | static void dentry_lru_move_tail(struct dentry *dentry) |
da3bbdd4 | 258 | { |
23044507 | 259 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
260 | if (list_empty(&dentry->d_lru)) { |
261 | list_add_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
262 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 263 | dentry_stat.nr_unused++; |
a4633357 CH |
264 | } else { |
265 | list_move_tail(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); | |
da3bbdd4 | 266 | } |
23044507 | 267 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
268 | } |
269 | ||
d52b9086 MS |
270 | /** |
271 | * d_kill - kill dentry and return parent | |
272 | * @dentry: dentry to kill | |
ff5fdb61 | 273 | * @parent: parent dentry |
d52b9086 | 274 | * |
31f3e0b3 | 275 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
276 | * |
277 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 278 | * |
b5c84bf6 NP |
279 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
280 | * d_kill. | |
d52b9086 | 281 | */ |
2fd6b7f5 | 282 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 283 | __releases(dentry->d_lock) |
2fd6b7f5 | 284 | __releases(parent->d_lock) |
873feea0 | 285 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 286 | { |
d52b9086 | 287 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
288 | /* |
289 | * Inform try_to_ascend() that we are no longer attached to the | |
290 | * dentry tree | |
291 | */ | |
292 | dentry->d_flags |= DCACHE_DISCONNECTED; | |
2fd6b7f5 NP |
293 | if (parent) |
294 | spin_unlock(&parent->d_lock); | |
d52b9086 | 295 | dentry_iput(dentry); |
b7ab39f6 NP |
296 | /* |
297 | * dentry_iput drops the locks, at which point nobody (except | |
298 | * transient RCU lookups) can reach this dentry. | |
299 | */ | |
d52b9086 | 300 | d_free(dentry); |
871c0067 | 301 | return parent; |
d52b9086 MS |
302 | } |
303 | ||
789680d1 NP |
304 | /** |
305 | * d_drop - drop a dentry | |
306 | * @dentry: dentry to drop | |
307 | * | |
308 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
309 | * be found through a VFS lookup any more. Note that this is different from | |
310 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
311 | * possible, giving a successful _negative_ lookup, while d_drop will | |
312 | * just make the cache lookup fail. | |
313 | * | |
314 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
315 | * reason (NFS timeouts or autofs deletes). | |
316 | * | |
317 | * __d_drop requires dentry->d_lock. | |
318 | */ | |
319 | void __d_drop(struct dentry *dentry) | |
320 | { | |
dea3667b | 321 | if (!d_unhashed(dentry)) { |
b07ad996 | 322 | struct hlist_bl_head *b; |
dea3667b | 323 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
b07ad996 | 324 | b = &dentry->d_sb->s_anon; |
dea3667b | 325 | else |
ceb5bdc2 | 326 | b = d_hash(dentry->d_parent, dentry->d_name.hash); |
dea3667b | 327 | |
1879fd6a | 328 | hlist_bl_lock(b); |
dea3667b LT |
329 | __hlist_bl_del(&dentry->d_hash); |
330 | dentry->d_hash.pprev = NULL; | |
1879fd6a | 331 | hlist_bl_unlock(b); |
dea3667b LT |
332 | |
333 | dentry_rcuwalk_barrier(dentry); | |
789680d1 NP |
334 | } |
335 | } | |
336 | EXPORT_SYMBOL(__d_drop); | |
337 | ||
338 | void d_drop(struct dentry *dentry) | |
339 | { | |
789680d1 NP |
340 | spin_lock(&dentry->d_lock); |
341 | __d_drop(dentry); | |
342 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
343 | } |
344 | EXPORT_SYMBOL(d_drop); | |
345 | ||
77812a1e NP |
346 | /* |
347 | * Finish off a dentry we've decided to kill. | |
348 | * dentry->d_lock must be held, returns with it unlocked. | |
349 | * If ref is non-zero, then decrement the refcount too. | |
350 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
351 | */ | |
352 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
353 | __releases(dentry->d_lock) | |
354 | { | |
873feea0 | 355 | struct inode *inode; |
77812a1e NP |
356 | struct dentry *parent; |
357 | ||
873feea0 NP |
358 | inode = dentry->d_inode; |
359 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
360 | relock: |
361 | spin_unlock(&dentry->d_lock); | |
362 | cpu_relax(); | |
363 | return dentry; /* try again with same dentry */ | |
364 | } | |
365 | if (IS_ROOT(dentry)) | |
366 | parent = NULL; | |
367 | else | |
368 | parent = dentry->d_parent; | |
369 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
370 | if (inode) |
371 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
372 | goto relock; |
373 | } | |
31e6b01f | 374 | |
77812a1e NP |
375 | if (ref) |
376 | dentry->d_count--; | |
377 | /* if dentry was on the d_lru list delete it from there */ | |
378 | dentry_lru_del(dentry); | |
379 | /* if it was on the hash then remove it */ | |
380 | __d_drop(dentry); | |
381 | return d_kill(dentry, parent); | |
382 | } | |
383 | ||
1da177e4 LT |
384 | /* |
385 | * This is dput | |
386 | * | |
387 | * This is complicated by the fact that we do not want to put | |
388 | * dentries that are no longer on any hash chain on the unused | |
389 | * list: we'd much rather just get rid of them immediately. | |
390 | * | |
391 | * However, that implies that we have to traverse the dentry | |
392 | * tree upwards to the parents which might _also_ now be | |
393 | * scheduled for deletion (it may have been only waiting for | |
394 | * its last child to go away). | |
395 | * | |
396 | * This tail recursion is done by hand as we don't want to depend | |
397 | * on the compiler to always get this right (gcc generally doesn't). | |
398 | * Real recursion would eat up our stack space. | |
399 | */ | |
400 | ||
401 | /* | |
402 | * dput - release a dentry | |
403 | * @dentry: dentry to release | |
404 | * | |
405 | * Release a dentry. This will drop the usage count and if appropriate | |
406 | * call the dentry unlink method as well as removing it from the queues and | |
407 | * releasing its resources. If the parent dentries were scheduled for release | |
408 | * they too may now get deleted. | |
1da177e4 | 409 | */ |
1da177e4 LT |
410 | void dput(struct dentry *dentry) |
411 | { | |
412 | if (!dentry) | |
413 | return; | |
414 | ||
415 | repeat: | |
b7ab39f6 | 416 | if (dentry->d_count == 1) |
1da177e4 | 417 | might_sleep(); |
1da177e4 | 418 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
419 | BUG_ON(!dentry->d_count); |
420 | if (dentry->d_count > 1) { | |
421 | dentry->d_count--; | |
1da177e4 | 422 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
423 | return; |
424 | } | |
425 | ||
fb045adb | 426 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 427 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 428 | goto kill_it; |
1da177e4 | 429 | } |
265ac902 | 430 | |
1da177e4 LT |
431 | /* Unreachable? Get rid of it */ |
432 | if (d_unhashed(dentry)) | |
433 | goto kill_it; | |
265ac902 NP |
434 | |
435 | /* Otherwise leave it cached and ensure it's on the LRU */ | |
436 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 437 | dentry_lru_add(dentry); |
265ac902 | 438 | |
61f3dee4 NP |
439 | dentry->d_count--; |
440 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
441 | return; |
442 | ||
d52b9086 | 443 | kill_it: |
77812a1e | 444 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
445 | if (dentry) |
446 | goto repeat; | |
1da177e4 | 447 | } |
ec4f8605 | 448 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
449 | |
450 | /** | |
451 | * d_invalidate - invalidate a dentry | |
452 | * @dentry: dentry to invalidate | |
453 | * | |
454 | * Try to invalidate the dentry if it turns out to be | |
455 | * possible. If there are other dentries that can be | |
456 | * reached through this one we can't delete it and we | |
457 | * return -EBUSY. On success we return 0. | |
458 | * | |
459 | * no dcache lock. | |
460 | */ | |
461 | ||
462 | int d_invalidate(struct dentry * dentry) | |
463 | { | |
464 | /* | |
465 | * If it's already been dropped, return OK. | |
466 | */ | |
da502956 | 467 | spin_lock(&dentry->d_lock); |
1da177e4 | 468 | if (d_unhashed(dentry)) { |
da502956 | 469 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
470 | return 0; |
471 | } | |
472 | /* | |
473 | * Check whether to do a partial shrink_dcache | |
474 | * to get rid of unused child entries. | |
475 | */ | |
476 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 477 | spin_unlock(&dentry->d_lock); |
1da177e4 | 478 | shrink_dcache_parent(dentry); |
da502956 | 479 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
480 | } |
481 | ||
482 | /* | |
483 | * Somebody else still using it? | |
484 | * | |
485 | * If it's a directory, we can't drop it | |
486 | * for fear of somebody re-populating it | |
487 | * with children (even though dropping it | |
488 | * would make it unreachable from the root, | |
489 | * we might still populate it if it was a | |
490 | * working directory or similar). | |
491 | */ | |
b7ab39f6 | 492 | if (dentry->d_count > 1) { |
1da177e4 LT |
493 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
494 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
495 | return -EBUSY; |
496 | } | |
497 | } | |
498 | ||
499 | __d_drop(dentry); | |
500 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
501 | return 0; |
502 | } | |
ec4f8605 | 503 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 504 | |
b5c84bf6 | 505 | /* This must be called with d_lock held */ |
dc0474be | 506 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 507 | { |
b7ab39f6 | 508 | dentry->d_count++; |
23044507 NP |
509 | } |
510 | ||
dc0474be | 511 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 512 | { |
23044507 | 513 | spin_lock(&dentry->d_lock); |
dc0474be | 514 | __dget_dlock(dentry); |
23044507 | 515 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
516 | } |
517 | ||
b7ab39f6 NP |
518 | struct dentry *dget_parent(struct dentry *dentry) |
519 | { | |
520 | struct dentry *ret; | |
521 | ||
522 | repeat: | |
a734eb45 NP |
523 | /* |
524 | * Don't need rcu_dereference because we re-check it was correct under | |
525 | * the lock. | |
526 | */ | |
527 | rcu_read_lock(); | |
b7ab39f6 | 528 | ret = dentry->d_parent; |
a734eb45 NP |
529 | if (!ret) { |
530 | rcu_read_unlock(); | |
b7ab39f6 NP |
531 | goto out; |
532 | } | |
a734eb45 NP |
533 | spin_lock(&ret->d_lock); |
534 | if (unlikely(ret != dentry->d_parent)) { | |
535 | spin_unlock(&ret->d_lock); | |
536 | rcu_read_unlock(); | |
b7ab39f6 NP |
537 | goto repeat; |
538 | } | |
a734eb45 | 539 | rcu_read_unlock(); |
b7ab39f6 NP |
540 | BUG_ON(!ret->d_count); |
541 | ret->d_count++; | |
542 | spin_unlock(&ret->d_lock); | |
543 | out: | |
b7ab39f6 NP |
544 | return ret; |
545 | } | |
546 | EXPORT_SYMBOL(dget_parent); | |
547 | ||
1da177e4 LT |
548 | /** |
549 | * d_find_alias - grab a hashed alias of inode | |
550 | * @inode: inode in question | |
551 | * @want_discon: flag, used by d_splice_alias, to request | |
552 | * that only a DISCONNECTED alias be returned. | |
553 | * | |
554 | * If inode has a hashed alias, or is a directory and has any alias, | |
555 | * acquire the reference to alias and return it. Otherwise return NULL. | |
556 | * Notice that if inode is a directory there can be only one alias and | |
557 | * it can be unhashed only if it has no children, or if it is the root | |
558 | * of a filesystem. | |
559 | * | |
21c0d8fd | 560 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 561 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 562 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 563 | */ |
da502956 | 564 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 565 | { |
da502956 | 566 | struct dentry *alias, *discon_alias; |
1da177e4 | 567 | |
da502956 NP |
568 | again: |
569 | discon_alias = NULL; | |
570 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
571 | spin_lock(&alias->d_lock); | |
1da177e4 | 572 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 573 | if (IS_ROOT(alias) && |
da502956 | 574 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 575 | discon_alias = alias; |
da502956 | 576 | } else if (!want_discon) { |
dc0474be | 577 | __dget_dlock(alias); |
da502956 NP |
578 | spin_unlock(&alias->d_lock); |
579 | return alias; | |
580 | } | |
581 | } | |
582 | spin_unlock(&alias->d_lock); | |
583 | } | |
584 | if (discon_alias) { | |
585 | alias = discon_alias; | |
586 | spin_lock(&alias->d_lock); | |
587 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
588 | if (IS_ROOT(alias) && | |
589 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 590 | __dget_dlock(alias); |
da502956 | 591 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
592 | return alias; |
593 | } | |
594 | } | |
da502956 NP |
595 | spin_unlock(&alias->d_lock); |
596 | goto again; | |
1da177e4 | 597 | } |
da502956 | 598 | return NULL; |
1da177e4 LT |
599 | } |
600 | ||
da502956 | 601 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 602 | { |
214fda1f DH |
603 | struct dentry *de = NULL; |
604 | ||
605 | if (!list_empty(&inode->i_dentry)) { | |
873feea0 | 606 | spin_lock(&inode->i_lock); |
214fda1f | 607 | de = __d_find_alias(inode, 0); |
873feea0 | 608 | spin_unlock(&inode->i_lock); |
214fda1f | 609 | } |
1da177e4 LT |
610 | return de; |
611 | } | |
ec4f8605 | 612 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
613 | |
614 | /* | |
615 | * Try to kill dentries associated with this inode. | |
616 | * WARNING: you must own a reference to inode. | |
617 | */ | |
618 | void d_prune_aliases(struct inode *inode) | |
619 | { | |
0cdca3f9 | 620 | struct dentry *dentry; |
1da177e4 | 621 | restart: |
873feea0 | 622 | spin_lock(&inode->i_lock); |
0cdca3f9 | 623 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 624 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 625 | if (!dentry->d_count) { |
dc0474be | 626 | __dget_dlock(dentry); |
1da177e4 LT |
627 | __d_drop(dentry); |
628 | spin_unlock(&dentry->d_lock); | |
873feea0 | 629 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
630 | dput(dentry); |
631 | goto restart; | |
632 | } | |
633 | spin_unlock(&dentry->d_lock); | |
634 | } | |
873feea0 | 635 | spin_unlock(&inode->i_lock); |
1da177e4 | 636 | } |
ec4f8605 | 637 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
638 | |
639 | /* | |
77812a1e NP |
640 | * Try to throw away a dentry - free the inode, dput the parent. |
641 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
642 | * Releases dentry->d_lock. | |
d702ccb3 | 643 | * |
77812a1e | 644 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 645 | */ |
77812a1e | 646 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 647 | __releases(dentry->d_lock) |
1da177e4 | 648 | { |
77812a1e | 649 | struct dentry *parent; |
d52b9086 | 650 | |
77812a1e | 651 | parent = dentry_kill(dentry, 0); |
d52b9086 | 652 | /* |
77812a1e NP |
653 | * If dentry_kill returns NULL, we have nothing more to do. |
654 | * if it returns the same dentry, trylocks failed. In either | |
655 | * case, just loop again. | |
656 | * | |
657 | * Otherwise, we need to prune ancestors too. This is necessary | |
658 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
659 | * is also expected to be beneficial in reducing dentry cache | |
660 | * fragmentation. | |
d52b9086 | 661 | */ |
77812a1e NP |
662 | if (!parent) |
663 | return; | |
664 | if (parent == dentry) | |
665 | return; | |
666 | ||
667 | /* Prune ancestors. */ | |
668 | dentry = parent; | |
d52b9086 | 669 | while (dentry) { |
b7ab39f6 | 670 | spin_lock(&dentry->d_lock); |
89e60548 NP |
671 | if (dentry->d_count > 1) { |
672 | dentry->d_count--; | |
673 | spin_unlock(&dentry->d_lock); | |
674 | return; | |
675 | } | |
77812a1e | 676 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 677 | } |
1da177e4 LT |
678 | } |
679 | ||
3049cfe2 | 680 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 681 | { |
da3bbdd4 | 682 | struct dentry *dentry; |
da3bbdd4 | 683 | |
ec33679d NP |
684 | rcu_read_lock(); |
685 | for (;;) { | |
ec33679d NP |
686 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
687 | if (&dentry->d_lru == list) | |
688 | break; /* empty */ | |
689 | spin_lock(&dentry->d_lock); | |
690 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
691 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
692 | continue; |
693 | } | |
694 | ||
1da177e4 LT |
695 | /* |
696 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
697 | * the LRU because of laziness during lookup. Do not free |
698 | * it - just keep it off the LRU list. | |
1da177e4 | 699 | */ |
b7ab39f6 | 700 | if (dentry->d_count) { |
ec33679d | 701 | dentry_lru_del(dentry); |
da3bbdd4 | 702 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
703 | continue; |
704 | } | |
ec33679d | 705 | |
ec33679d | 706 | rcu_read_unlock(); |
77812a1e NP |
707 | |
708 | try_prune_one_dentry(dentry); | |
709 | ||
ec33679d | 710 | rcu_read_lock(); |
da3bbdd4 | 711 | } |
ec33679d | 712 | rcu_read_unlock(); |
3049cfe2 CH |
713 | } |
714 | ||
715 | /** | |
716 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
717 | * @sb: superblock to shrink dentry LRU. | |
718 | * @count: number of entries to prune | |
719 | * @flags: flags to control the dentry processing | |
720 | * | |
721 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
722 | */ | |
723 | static void __shrink_dcache_sb(struct super_block *sb, int *count, int flags) | |
724 | { | |
725 | /* called from prune_dcache() and shrink_dcache_parent() */ | |
726 | struct dentry *dentry; | |
727 | LIST_HEAD(referenced); | |
728 | LIST_HEAD(tmp); | |
729 | int cnt = *count; | |
730 | ||
23044507 NP |
731 | relock: |
732 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
733 | while (!list_empty(&sb->s_dentry_lru)) { |
734 | dentry = list_entry(sb->s_dentry_lru.prev, | |
735 | struct dentry, d_lru); | |
736 | BUG_ON(dentry->d_sb != sb); | |
737 | ||
23044507 NP |
738 | if (!spin_trylock(&dentry->d_lock)) { |
739 | spin_unlock(&dcache_lru_lock); | |
740 | cpu_relax(); | |
741 | goto relock; | |
742 | } | |
743 | ||
3049cfe2 CH |
744 | /* |
745 | * If we are honouring the DCACHE_REFERENCED flag and the | |
746 | * dentry has this flag set, don't free it. Clear the flag | |
747 | * and put it back on the LRU. | |
748 | */ | |
23044507 NP |
749 | if (flags & DCACHE_REFERENCED && |
750 | dentry->d_flags & DCACHE_REFERENCED) { | |
751 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
752 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 753 | spin_unlock(&dentry->d_lock); |
23044507 NP |
754 | } else { |
755 | list_move_tail(&dentry->d_lru, &tmp); | |
756 | spin_unlock(&dentry->d_lock); | |
757 | if (!--cnt) | |
758 | break; | |
3049cfe2 | 759 | } |
ec33679d | 760 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 761 | } |
da3bbdd4 KM |
762 | if (!list_empty(&referenced)) |
763 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 764 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
765 | |
766 | shrink_dentry_list(&tmp); | |
767 | ||
768 | *count = cnt; | |
da3bbdd4 KM |
769 | } |
770 | ||
771 | /** | |
772 | * prune_dcache - shrink the dcache | |
773 | * @count: number of entries to try to free | |
774 | * | |
775 | * Shrink the dcache. This is done when we need more memory, or simply when we | |
776 | * need to unmount something (at which point we need to unuse all dentries). | |
777 | * | |
778 | * This function may fail to free any resources if all the dentries are in use. | |
779 | */ | |
780 | static void prune_dcache(int count) | |
781 | { | |
dca33252 | 782 | struct super_block *sb, *p = NULL; |
da3bbdd4 | 783 | int w_count; |
86c8749e | 784 | int unused = dentry_stat.nr_unused; |
da3bbdd4 KM |
785 | int prune_ratio; |
786 | int pruned; | |
787 | ||
788 | if (unused == 0 || count == 0) | |
789 | return; | |
da3bbdd4 KM |
790 | if (count >= unused) |
791 | prune_ratio = 1; | |
792 | else | |
793 | prune_ratio = unused / count; | |
794 | spin_lock(&sb_lock); | |
dca33252 | 795 | list_for_each_entry(sb, &super_blocks, s_list) { |
551de6f3 AV |
796 | if (list_empty(&sb->s_instances)) |
797 | continue; | |
da3bbdd4 | 798 | if (sb->s_nr_dentry_unused == 0) |
1da177e4 | 799 | continue; |
da3bbdd4 KM |
800 | sb->s_count++; |
801 | /* Now, we reclaim unused dentrins with fairness. | |
802 | * We reclaim them same percentage from each superblock. | |
803 | * We calculate number of dentries to scan on this sb | |
804 | * as follows, but the implementation is arranged to avoid | |
805 | * overflows: | |
806 | * number of dentries to scan on this sb = | |
807 | * count * (number of dentries on this sb / | |
808 | * number of dentries in the machine) | |
0feae5c4 | 809 | */ |
da3bbdd4 KM |
810 | spin_unlock(&sb_lock); |
811 | if (prune_ratio != 1) | |
812 | w_count = (sb->s_nr_dentry_unused / prune_ratio) + 1; | |
813 | else | |
814 | w_count = sb->s_nr_dentry_unused; | |
815 | pruned = w_count; | |
0feae5c4 | 816 | /* |
da3bbdd4 KM |
817 | * We need to be sure this filesystem isn't being unmounted, |
818 | * otherwise we could race with generic_shutdown_super(), and | |
819 | * end up holding a reference to an inode while the filesystem | |
820 | * is unmounted. So we try to get s_umount, and make sure | |
821 | * s_root isn't NULL. | |
0feae5c4 | 822 | */ |
da3bbdd4 KM |
823 | if (down_read_trylock(&sb->s_umount)) { |
824 | if ((sb->s_root != NULL) && | |
825 | (!list_empty(&sb->s_dentry_lru))) { | |
da3bbdd4 KM |
826 | __shrink_dcache_sb(sb, &w_count, |
827 | DCACHE_REFERENCED); | |
828 | pruned -= w_count; | |
0feae5c4 | 829 | } |
da3bbdd4 | 830 | up_read(&sb->s_umount); |
0feae5c4 | 831 | } |
da3bbdd4 | 832 | spin_lock(&sb_lock); |
dca33252 AV |
833 | if (p) |
834 | __put_super(p); | |
da3bbdd4 | 835 | count -= pruned; |
dca33252 | 836 | p = sb; |
79893c17 AV |
837 | /* more work left to do? */ |
838 | if (count <= 0) | |
839 | break; | |
1da177e4 | 840 | } |
dca33252 AV |
841 | if (p) |
842 | __put_super(p); | |
da3bbdd4 | 843 | spin_unlock(&sb_lock); |
1da177e4 LT |
844 | } |
845 | ||
1da177e4 LT |
846 | /** |
847 | * shrink_dcache_sb - shrink dcache for a superblock | |
848 | * @sb: superblock | |
849 | * | |
3049cfe2 CH |
850 | * Shrink the dcache for the specified super block. This is used to free |
851 | * the dcache before unmounting a file system. | |
1da177e4 | 852 | */ |
3049cfe2 | 853 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 854 | { |
3049cfe2 CH |
855 | LIST_HEAD(tmp); |
856 | ||
23044507 | 857 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
858 | while (!list_empty(&sb->s_dentry_lru)) { |
859 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 860 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 861 | shrink_dentry_list(&tmp); |
ec33679d | 862 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 863 | } |
23044507 | 864 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 865 | } |
ec4f8605 | 866 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 867 | |
c636ebdb DH |
868 | /* |
869 | * destroy a single subtree of dentries for unmount | |
870 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
871 | * locking | |
872 | */ | |
873 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
874 | { | |
875 | struct dentry *parent; | |
f8713576 | 876 | unsigned detached = 0; |
c636ebdb DH |
877 | |
878 | BUG_ON(!IS_ROOT(dentry)); | |
879 | ||
880 | /* detach this root from the system */ | |
23044507 | 881 | spin_lock(&dentry->d_lock); |
a4633357 | 882 | dentry_lru_del(dentry); |
c636ebdb | 883 | __d_drop(dentry); |
da502956 | 884 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
885 | |
886 | for (;;) { | |
887 | /* descend to the first leaf in the current subtree */ | |
888 | while (!list_empty(&dentry->d_subdirs)) { | |
889 | struct dentry *loop; | |
890 | ||
891 | /* this is a branch with children - detach all of them | |
892 | * from the system in one go */ | |
2fd6b7f5 | 893 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
894 | list_for_each_entry(loop, &dentry->d_subdirs, |
895 | d_u.d_child) { | |
2fd6b7f5 NP |
896 | spin_lock_nested(&loop->d_lock, |
897 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 898 | dentry_lru_del(loop); |
c636ebdb | 899 | __d_drop(loop); |
da502956 | 900 | spin_unlock(&loop->d_lock); |
c636ebdb | 901 | } |
2fd6b7f5 | 902 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
903 | |
904 | /* move to the first child */ | |
905 | dentry = list_entry(dentry->d_subdirs.next, | |
906 | struct dentry, d_u.d_child); | |
907 | } | |
908 | ||
909 | /* consume the dentries from this leaf up through its parents | |
910 | * until we find one with children or run out altogether */ | |
911 | do { | |
912 | struct inode *inode; | |
913 | ||
b7ab39f6 | 914 | if (dentry->d_count != 0) { |
c636ebdb DH |
915 | printk(KERN_ERR |
916 | "BUG: Dentry %p{i=%lx,n=%s}" | |
917 | " still in use (%d)" | |
918 | " [unmount of %s %s]\n", | |
919 | dentry, | |
920 | dentry->d_inode ? | |
921 | dentry->d_inode->i_ino : 0UL, | |
922 | dentry->d_name.name, | |
b7ab39f6 | 923 | dentry->d_count, |
c636ebdb DH |
924 | dentry->d_sb->s_type->name, |
925 | dentry->d_sb->s_id); | |
926 | BUG(); | |
927 | } | |
928 | ||
2fd6b7f5 | 929 | if (IS_ROOT(dentry)) { |
c636ebdb | 930 | parent = NULL; |
2fd6b7f5 NP |
931 | list_del(&dentry->d_u.d_child); |
932 | } else { | |
871c0067 | 933 | parent = dentry->d_parent; |
b7ab39f6 NP |
934 | spin_lock(&parent->d_lock); |
935 | parent->d_count--; | |
2fd6b7f5 | 936 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 937 | spin_unlock(&parent->d_lock); |
871c0067 | 938 | } |
c636ebdb | 939 | |
f8713576 | 940 | detached++; |
c636ebdb DH |
941 | |
942 | inode = dentry->d_inode; | |
943 | if (inode) { | |
944 | dentry->d_inode = NULL; | |
945 | list_del_init(&dentry->d_alias); | |
946 | if (dentry->d_op && dentry->d_op->d_iput) | |
947 | dentry->d_op->d_iput(dentry, inode); | |
948 | else | |
949 | iput(inode); | |
950 | } | |
951 | ||
952 | d_free(dentry); | |
953 | ||
954 | /* finished when we fall off the top of the tree, | |
955 | * otherwise we ascend to the parent and move to the | |
956 | * next sibling if there is one */ | |
957 | if (!parent) | |
312d3ca8 | 958 | return; |
c636ebdb | 959 | dentry = parent; |
c636ebdb DH |
960 | } while (list_empty(&dentry->d_subdirs)); |
961 | ||
962 | dentry = list_entry(dentry->d_subdirs.next, | |
963 | struct dentry, d_u.d_child); | |
964 | } | |
965 | } | |
966 | ||
967 | /* | |
968 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 969 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
970 | * - the superblock is detached from all mountings and open files, so the |
971 | * dentry trees will not be rearranged by the VFS | |
972 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
973 | * any dentries belonging to this superblock that it comes across | |
974 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
975 | * in this superblock | |
976 | */ | |
977 | void shrink_dcache_for_umount(struct super_block *sb) | |
978 | { | |
979 | struct dentry *dentry; | |
980 | ||
981 | if (down_read_trylock(&sb->s_umount)) | |
982 | BUG(); | |
983 | ||
984 | dentry = sb->s_root; | |
985 | sb->s_root = NULL; | |
b7ab39f6 NP |
986 | spin_lock(&dentry->d_lock); |
987 | dentry->d_count--; | |
988 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
989 | shrink_dcache_for_umount_subtree(dentry); |
990 | ||
ceb5bdc2 NP |
991 | while (!hlist_bl_empty(&sb->s_anon)) { |
992 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
993 | shrink_dcache_for_umount_subtree(dentry); |
994 | } | |
995 | } | |
996 | ||
c826cb7d LT |
997 | /* |
998 | * This tries to ascend one level of parenthood, but | |
999 | * we can race with renaming, so we need to re-check | |
1000 | * the parenthood after dropping the lock and check | |
1001 | * that the sequence number still matches. | |
1002 | */ | |
1003 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
1004 | { | |
1005 | struct dentry *new = old->d_parent; | |
1006 | ||
1007 | rcu_read_lock(); | |
1008 | spin_unlock(&old->d_lock); | |
1009 | spin_lock(&new->d_lock); | |
1010 | ||
1011 | /* | |
1012 | * might go back up the wrong parent if we have had a rename | |
1013 | * or deletion | |
1014 | */ | |
1015 | if (new != old->d_parent || | |
c83ce989 | 1016 | (old->d_flags & DCACHE_DISCONNECTED) || |
c826cb7d LT |
1017 | (!locked && read_seqretry(&rename_lock, seq))) { |
1018 | spin_unlock(&new->d_lock); | |
1019 | new = NULL; | |
1020 | } | |
1021 | rcu_read_unlock(); | |
1022 | return new; | |
1023 | } | |
1024 | ||
1025 | ||
1da177e4 LT |
1026 | /* |
1027 | * Search for at least 1 mount point in the dentry's subdirs. | |
1028 | * We descend to the next level whenever the d_subdirs | |
1029 | * list is non-empty and continue searching. | |
1030 | */ | |
1031 | ||
1032 | /** | |
1033 | * have_submounts - check for mounts over a dentry | |
1034 | * @parent: dentry to check. | |
1035 | * | |
1036 | * Return true if the parent or its subdirectories contain | |
1037 | * a mount point | |
1038 | */ | |
1da177e4 LT |
1039 | int have_submounts(struct dentry *parent) |
1040 | { | |
949854d0 | 1041 | struct dentry *this_parent; |
1da177e4 | 1042 | struct list_head *next; |
949854d0 | 1043 | unsigned seq; |
58db63d0 | 1044 | int locked = 0; |
949854d0 | 1045 | |
949854d0 | 1046 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1047 | again: |
1048 | this_parent = parent; | |
1da177e4 | 1049 | |
1da177e4 LT |
1050 | if (d_mountpoint(parent)) |
1051 | goto positive; | |
2fd6b7f5 | 1052 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1053 | repeat: |
1054 | next = this_parent->d_subdirs.next; | |
1055 | resume: | |
1056 | while (next != &this_parent->d_subdirs) { | |
1057 | struct list_head *tmp = next; | |
5160ee6f | 1058 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1059 | next = tmp->next; |
2fd6b7f5 NP |
1060 | |
1061 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1062 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1063 | if (d_mountpoint(dentry)) { |
1064 | spin_unlock(&dentry->d_lock); | |
1065 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1066 | goto positive; |
2fd6b7f5 | 1067 | } |
1da177e4 | 1068 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1069 | spin_unlock(&this_parent->d_lock); |
1070 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1071 | this_parent = dentry; |
2fd6b7f5 | 1072 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1073 | goto repeat; |
1074 | } | |
2fd6b7f5 | 1075 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1076 | } |
1077 | /* | |
1078 | * All done at this level ... ascend and resume the search. | |
1079 | */ | |
1080 | if (this_parent != parent) { | |
c826cb7d LT |
1081 | struct dentry *child = this_parent; |
1082 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1083 | if (!this_parent) | |
949854d0 | 1084 | goto rename_retry; |
949854d0 | 1085 | next = child->d_u.d_child.next; |
1da177e4 LT |
1086 | goto resume; |
1087 | } | |
2fd6b7f5 | 1088 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1089 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1090 | goto rename_retry; |
58db63d0 NP |
1091 | if (locked) |
1092 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1093 | return 0; /* No mount points found in tree */ |
1094 | positive: | |
58db63d0 | 1095 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1096 | goto rename_retry; |
58db63d0 NP |
1097 | if (locked) |
1098 | write_sequnlock(&rename_lock); | |
1da177e4 | 1099 | return 1; |
58db63d0 NP |
1100 | |
1101 | rename_retry: | |
1102 | locked = 1; | |
1103 | write_seqlock(&rename_lock); | |
1104 | goto again; | |
1da177e4 | 1105 | } |
ec4f8605 | 1106 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1107 | |
1108 | /* | |
1109 | * Search the dentry child list for the specified parent, | |
1110 | * and move any unused dentries to the end of the unused | |
1111 | * list for prune_dcache(). We descend to the next level | |
1112 | * whenever the d_subdirs list is non-empty and continue | |
1113 | * searching. | |
1114 | * | |
1115 | * It returns zero iff there are no unused children, | |
1116 | * otherwise it returns the number of children moved to | |
1117 | * the end of the unused list. This may not be the total | |
1118 | * number of unused children, because select_parent can | |
1119 | * drop the lock and return early due to latency | |
1120 | * constraints. | |
1121 | */ | |
1122 | static int select_parent(struct dentry * parent) | |
1123 | { | |
949854d0 | 1124 | struct dentry *this_parent; |
1da177e4 | 1125 | struct list_head *next; |
949854d0 | 1126 | unsigned seq; |
1da177e4 | 1127 | int found = 0; |
58db63d0 | 1128 | int locked = 0; |
1da177e4 | 1129 | |
949854d0 | 1130 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1131 | again: |
1132 | this_parent = parent; | |
2fd6b7f5 | 1133 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1134 | repeat: |
1135 | next = this_parent->d_subdirs.next; | |
1136 | resume: | |
1137 | while (next != &this_parent->d_subdirs) { | |
1138 | struct list_head *tmp = next; | |
5160ee6f | 1139 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1140 | next = tmp->next; |
1141 | ||
2fd6b7f5 | 1142 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1143 | |
1da177e4 LT |
1144 | /* |
1145 | * move only zero ref count dentries to the end | |
1146 | * of the unused list for prune_dcache | |
1147 | */ | |
b7ab39f6 | 1148 | if (!dentry->d_count) { |
a4633357 | 1149 | dentry_lru_move_tail(dentry); |
1da177e4 | 1150 | found++; |
a4633357 CH |
1151 | } else { |
1152 | dentry_lru_del(dentry); | |
1da177e4 LT |
1153 | } |
1154 | ||
1155 | /* | |
1156 | * We can return to the caller if we have found some (this | |
1157 | * ensures forward progress). We'll be coming back to find | |
1158 | * the rest. | |
1159 | */ | |
2fd6b7f5 NP |
1160 | if (found && need_resched()) { |
1161 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1162 | goto out; |
2fd6b7f5 | 1163 | } |
1da177e4 LT |
1164 | |
1165 | /* | |
1166 | * Descend a level if the d_subdirs list is non-empty. | |
1167 | */ | |
1168 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1169 | spin_unlock(&this_parent->d_lock); |
1170 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1171 | this_parent = dentry; |
2fd6b7f5 | 1172 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1173 | goto repeat; |
1174 | } | |
2fd6b7f5 NP |
1175 | |
1176 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1177 | } |
1178 | /* | |
1179 | * All done at this level ... ascend and resume the search. | |
1180 | */ | |
1181 | if (this_parent != parent) { | |
c826cb7d LT |
1182 | struct dentry *child = this_parent; |
1183 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1184 | if (!this_parent) | |
949854d0 | 1185 | goto rename_retry; |
949854d0 | 1186 | next = child->d_u.d_child.next; |
1da177e4 LT |
1187 | goto resume; |
1188 | } | |
1189 | out: | |
2fd6b7f5 | 1190 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1191 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1192 | goto rename_retry; |
58db63d0 NP |
1193 | if (locked) |
1194 | write_sequnlock(&rename_lock); | |
1da177e4 | 1195 | return found; |
58db63d0 NP |
1196 | |
1197 | rename_retry: | |
1198 | if (found) | |
1199 | return found; | |
1200 | locked = 1; | |
1201 | write_seqlock(&rename_lock); | |
1202 | goto again; | |
1da177e4 LT |
1203 | } |
1204 | ||
1205 | /** | |
1206 | * shrink_dcache_parent - prune dcache | |
1207 | * @parent: parent of entries to prune | |
1208 | * | |
1209 | * Prune the dcache to remove unused children of the parent dentry. | |
1210 | */ | |
1211 | ||
1212 | void shrink_dcache_parent(struct dentry * parent) | |
1213 | { | |
da3bbdd4 | 1214 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1215 | int found; |
1216 | ||
1217 | while ((found = select_parent(parent)) != 0) | |
da3bbdd4 | 1218 | __shrink_dcache_sb(sb, &found, 0); |
1da177e4 | 1219 | } |
ec4f8605 | 1220 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1221 | |
1da177e4 | 1222 | /* |
1495f230 | 1223 | * Scan `sc->nr_slab_to_reclaim' dentries and return the number which remain. |
1da177e4 LT |
1224 | * |
1225 | * We need to avoid reentering the filesystem if the caller is performing a | |
1226 | * GFP_NOFS allocation attempt. One example deadlock is: | |
1227 | * | |
1228 | * ext2_new_block->getblk->GFP->shrink_dcache_memory->prune_dcache-> | |
1229 | * prune_one_dentry->dput->dentry_iput->iput->inode->i_sb->s_op->put_inode-> | |
1230 | * ext2_discard_prealloc->ext2_free_blocks->lock_super->DEADLOCK. | |
1231 | * | |
1232 | * In this case we return -1 to tell the caller that we baled. | |
1233 | */ | |
1495f230 YH |
1234 | static int shrink_dcache_memory(struct shrinker *shrink, |
1235 | struct shrink_control *sc) | |
1da177e4 | 1236 | { |
1495f230 YH |
1237 | int nr = sc->nr_to_scan; |
1238 | gfp_t gfp_mask = sc->gfp_mask; | |
1239 | ||
1da177e4 LT |
1240 | if (nr) { |
1241 | if (!(gfp_mask & __GFP_FS)) | |
1242 | return -1; | |
da3bbdd4 | 1243 | prune_dcache(nr); |
1da177e4 | 1244 | } |
312d3ca8 | 1245 | |
86c8749e | 1246 | return (dentry_stat.nr_unused / 100) * sysctl_vfs_cache_pressure; |
1da177e4 LT |
1247 | } |
1248 | ||
8e1f936b RR |
1249 | static struct shrinker dcache_shrinker = { |
1250 | .shrink = shrink_dcache_memory, | |
1251 | .seeks = DEFAULT_SEEKS, | |
1252 | }; | |
1253 | ||
1da177e4 LT |
1254 | /** |
1255 | * d_alloc - allocate a dcache entry | |
1256 | * @parent: parent of entry to allocate | |
1257 | * @name: qstr of the name | |
1258 | * | |
1259 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1260 | * available. On a success the dentry is returned. The name passed in is | |
1261 | * copied and the copy passed in may be reused after this call. | |
1262 | */ | |
1263 | ||
1264 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1265 | { | |
1266 | struct dentry *dentry; | |
1267 | char *dname; | |
1268 | ||
e12ba74d | 1269 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1270 | if (!dentry) |
1271 | return NULL; | |
1272 | ||
1273 | if (name->len > DNAME_INLINE_LEN-1) { | |
1274 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1275 | if (!dname) { | |
1276 | kmem_cache_free(dentry_cache, dentry); | |
1277 | return NULL; | |
1278 | } | |
1279 | } else { | |
1280 | dname = dentry->d_iname; | |
1281 | } | |
1282 | dentry->d_name.name = dname; | |
1283 | ||
1284 | dentry->d_name.len = name->len; | |
1285 | dentry->d_name.hash = name->hash; | |
1286 | memcpy(dname, name->name, name->len); | |
1287 | dname[name->len] = 0; | |
1288 | ||
b7ab39f6 | 1289 | dentry->d_count = 1; |
dea3667b | 1290 | dentry->d_flags = 0; |
1da177e4 | 1291 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1292 | seqcount_init(&dentry->d_seq); |
1da177e4 LT |
1293 | dentry->d_inode = NULL; |
1294 | dentry->d_parent = NULL; | |
1295 | dentry->d_sb = NULL; | |
1296 | dentry->d_op = NULL; | |
1297 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1298 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1299 | INIT_LIST_HEAD(&dentry->d_lru); |
1300 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1301 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1302 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
1da177e4 LT |
1303 | |
1304 | if (parent) { | |
2fd6b7f5 | 1305 | spin_lock(&parent->d_lock); |
89ad485f NP |
1306 | /* |
1307 | * don't need child lock because it is not subject | |
1308 | * to concurrency here | |
1309 | */ | |
dc0474be NP |
1310 | __dget_dlock(parent); |
1311 | dentry->d_parent = parent; | |
1da177e4 | 1312 | dentry->d_sb = parent->d_sb; |
c8aebb0c | 1313 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
5160ee6f | 1314 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); |
2fd6b7f5 | 1315 | spin_unlock(&parent->d_lock); |
2fd6b7f5 | 1316 | } |
1da177e4 | 1317 | |
3e880fb5 | 1318 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1319 | |
1da177e4 LT |
1320 | return dentry; |
1321 | } | |
ec4f8605 | 1322 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1323 | |
4b936885 NP |
1324 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1325 | { | |
1326 | struct dentry *dentry = d_alloc(NULL, name); | |
1327 | if (dentry) { | |
1328 | dentry->d_sb = sb; | |
c8aebb0c | 1329 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
4b936885 NP |
1330 | dentry->d_parent = dentry; |
1331 | dentry->d_flags |= DCACHE_DISCONNECTED; | |
1332 | } | |
1333 | return dentry; | |
1334 | } | |
1335 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1336 | ||
1da177e4 LT |
1337 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1338 | { | |
1339 | struct qstr q; | |
1340 | ||
1341 | q.name = name; | |
1342 | q.len = strlen(name); | |
1343 | q.hash = full_name_hash(q.name, q.len); | |
1344 | return d_alloc(parent, &q); | |
1345 | } | |
ef26ca97 | 1346 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1347 | |
fb045adb NP |
1348 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1349 | { | |
6f7f7caa LT |
1350 | WARN_ON_ONCE(dentry->d_op); |
1351 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1352 | DCACHE_OP_COMPARE | |
1353 | DCACHE_OP_REVALIDATE | | |
1354 | DCACHE_OP_DELETE )); | |
1355 | dentry->d_op = op; | |
1356 | if (!op) | |
1357 | return; | |
1358 | if (op->d_hash) | |
1359 | dentry->d_flags |= DCACHE_OP_HASH; | |
1360 | if (op->d_compare) | |
1361 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1362 | if (op->d_revalidate) | |
1363 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1364 | if (op->d_delete) | |
1365 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1366 | ||
1367 | } | |
1368 | EXPORT_SYMBOL(d_set_d_op); | |
1369 | ||
360da900 OH |
1370 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1371 | { | |
b23fb0a6 | 1372 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1373 | if (inode) { |
1374 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1375 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
360da900 | 1376 | list_add(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1377 | } |
360da900 | 1378 | dentry->d_inode = inode; |
31e6b01f | 1379 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1380 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1381 | fsnotify_d_instantiate(dentry, inode); |
1382 | } | |
1383 | ||
1da177e4 LT |
1384 | /** |
1385 | * d_instantiate - fill in inode information for a dentry | |
1386 | * @entry: dentry to complete | |
1387 | * @inode: inode to attach to this dentry | |
1388 | * | |
1389 | * Fill in inode information in the entry. | |
1390 | * | |
1391 | * This turns negative dentries into productive full members | |
1392 | * of society. | |
1393 | * | |
1394 | * NOTE! This assumes that the inode count has been incremented | |
1395 | * (or otherwise set) by the caller to indicate that it is now | |
1396 | * in use by the dcache. | |
1397 | */ | |
1398 | ||
1399 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1400 | { | |
28133c7b | 1401 | BUG_ON(!list_empty(&entry->d_alias)); |
873feea0 NP |
1402 | if (inode) |
1403 | spin_lock(&inode->i_lock); | |
360da900 | 1404 | __d_instantiate(entry, inode); |
873feea0 NP |
1405 | if (inode) |
1406 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1407 | security_d_instantiate(entry, inode); |
1408 | } | |
ec4f8605 | 1409 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1410 | |
1411 | /** | |
1412 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1413 | * @entry: dentry to instantiate | |
1414 | * @inode: inode to attach to this dentry | |
1415 | * | |
1416 | * Fill in inode information in the entry. On success, it returns NULL. | |
1417 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1418 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1419 | * |
1420 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1421 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1422 | * |
1423 | * This also assumes that the inode count has been incremented | |
1424 | * (or otherwise set) by the caller to indicate that it is now | |
1425 | * in use by the dcache. | |
1da177e4 | 1426 | */ |
770bfad8 DH |
1427 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1428 | struct inode *inode) | |
1da177e4 LT |
1429 | { |
1430 | struct dentry *alias; | |
1431 | int len = entry->d_name.len; | |
1432 | const char *name = entry->d_name.name; | |
1433 | unsigned int hash = entry->d_name.hash; | |
1434 | ||
770bfad8 | 1435 | if (!inode) { |
360da900 | 1436 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1437 | return NULL; |
1438 | } | |
1439 | ||
1da177e4 LT |
1440 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1441 | struct qstr *qstr = &alias->d_name; | |
1442 | ||
9abca360 NP |
1443 | /* |
1444 | * Don't need alias->d_lock here, because aliases with | |
1445 | * d_parent == entry->d_parent are not subject to name or | |
1446 | * parent changes, because the parent inode i_mutex is held. | |
1447 | */ | |
1da177e4 LT |
1448 | if (qstr->hash != hash) |
1449 | continue; | |
1450 | if (alias->d_parent != entry->d_parent) | |
1451 | continue; | |
9d55c369 | 1452 | if (dentry_cmp(qstr->name, qstr->len, name, len)) |
1da177e4 | 1453 | continue; |
dc0474be | 1454 | __dget(alias); |
1da177e4 LT |
1455 | return alias; |
1456 | } | |
770bfad8 | 1457 | |
360da900 | 1458 | __d_instantiate(entry, inode); |
1da177e4 LT |
1459 | return NULL; |
1460 | } | |
770bfad8 DH |
1461 | |
1462 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1463 | { | |
1464 | struct dentry *result; | |
1465 | ||
1466 | BUG_ON(!list_empty(&entry->d_alias)); | |
1467 | ||
873feea0 NP |
1468 | if (inode) |
1469 | spin_lock(&inode->i_lock); | |
770bfad8 | 1470 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1471 | if (inode) |
1472 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1473 | |
1474 | if (!result) { | |
1475 | security_d_instantiate(entry, inode); | |
1476 | return NULL; | |
1477 | } | |
1478 | ||
1479 | BUG_ON(!d_unhashed(result)); | |
1480 | iput(inode); | |
1481 | return result; | |
1482 | } | |
1483 | ||
1da177e4 LT |
1484 | EXPORT_SYMBOL(d_instantiate_unique); |
1485 | ||
1486 | /** | |
1487 | * d_alloc_root - allocate root dentry | |
1488 | * @root_inode: inode to allocate the root for | |
1489 | * | |
1490 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1491 | * instantiated and returned. %NULL is returned if there is insufficient | |
1492 | * memory or the inode passed is %NULL. | |
1493 | */ | |
1494 | ||
1495 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1496 | { | |
1497 | struct dentry *res = NULL; | |
1498 | ||
1499 | if (root_inode) { | |
1500 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1501 | ||
1502 | res = d_alloc(NULL, &name); | |
1503 | if (res) { | |
1504 | res->d_sb = root_inode->i_sb; | |
c8aebb0c | 1505 | d_set_d_op(res, res->d_sb->s_d_op); |
1da177e4 LT |
1506 | res->d_parent = res; |
1507 | d_instantiate(res, root_inode); | |
1508 | } | |
1509 | } | |
1510 | return res; | |
1511 | } | |
ec4f8605 | 1512 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 | 1513 | |
d891eedb BF |
1514 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1515 | { | |
1516 | struct dentry *alias; | |
1517 | ||
1518 | if (list_empty(&inode->i_dentry)) | |
1519 | return NULL; | |
1520 | alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias); | |
1521 | __dget(alias); | |
1522 | return alias; | |
1523 | } | |
1524 | ||
1525 | static struct dentry * d_find_any_alias(struct inode *inode) | |
1526 | { | |
1527 | struct dentry *de; | |
1528 | ||
1529 | spin_lock(&inode->i_lock); | |
1530 | de = __d_find_any_alias(inode); | |
1531 | spin_unlock(&inode->i_lock); | |
1532 | return de; | |
1533 | } | |
1534 | ||
1535 | ||
4ea3ada2 CH |
1536 | /** |
1537 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1538 | * @inode: inode to allocate the dentry for | |
1539 | * | |
1540 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1541 | * similar open by handle operations. The returned dentry may be anonymous, | |
1542 | * or may have a full name (if the inode was already in the cache). | |
1543 | * | |
1544 | * When called on a directory inode, we must ensure that the inode only ever | |
1545 | * has one dentry. If a dentry is found, that is returned instead of | |
1546 | * allocating a new one. | |
1547 | * | |
1548 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1549 | * to the dentry. In case of an error the reference on the inode is released. |
1550 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1551 | * be passed in and will be the error will be propagate to the return value, | |
1552 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1553 | */ |
1554 | struct dentry *d_obtain_alias(struct inode *inode) | |
1555 | { | |
9308a612 CH |
1556 | static const struct qstr anonstring = { .name = "" }; |
1557 | struct dentry *tmp; | |
1558 | struct dentry *res; | |
4ea3ada2 CH |
1559 | |
1560 | if (!inode) | |
44003728 | 1561 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1562 | if (IS_ERR(inode)) |
1563 | return ERR_CAST(inode); | |
1564 | ||
d891eedb | 1565 | res = d_find_any_alias(inode); |
9308a612 CH |
1566 | if (res) |
1567 | goto out_iput; | |
1568 | ||
1569 | tmp = d_alloc(NULL, &anonstring); | |
1570 | if (!tmp) { | |
1571 | res = ERR_PTR(-ENOMEM); | |
1572 | goto out_iput; | |
4ea3ada2 | 1573 | } |
9308a612 CH |
1574 | tmp->d_parent = tmp; /* make sure dput doesn't croak */ |
1575 | ||
b5c84bf6 | 1576 | |
873feea0 | 1577 | spin_lock(&inode->i_lock); |
d891eedb | 1578 | res = __d_find_any_alias(inode); |
9308a612 | 1579 | if (res) { |
873feea0 | 1580 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1581 | dput(tmp); |
1582 | goto out_iput; | |
1583 | } | |
1584 | ||
1585 | /* attach a disconnected dentry */ | |
1586 | spin_lock(&tmp->d_lock); | |
1587 | tmp->d_sb = inode->i_sb; | |
c8aebb0c | 1588 | d_set_d_op(tmp, tmp->d_sb->s_d_op); |
9308a612 CH |
1589 | tmp->d_inode = inode; |
1590 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
9308a612 | 1591 | list_add(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1592 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1593 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1594 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1595 | spin_unlock(&tmp->d_lock); |
873feea0 | 1596 | spin_unlock(&inode->i_lock); |
24ff6663 | 1597 | security_d_instantiate(tmp, inode); |
9308a612 | 1598 | |
9308a612 CH |
1599 | return tmp; |
1600 | ||
1601 | out_iput: | |
24ff6663 JB |
1602 | if (res && !IS_ERR(res)) |
1603 | security_d_instantiate(res, inode); | |
9308a612 CH |
1604 | iput(inode); |
1605 | return res; | |
4ea3ada2 | 1606 | } |
adc48720 | 1607 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1608 | |
1609 | /** | |
1610 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1611 | * @inode: the inode which may have a disconnected dentry | |
1612 | * @dentry: a negative dentry which we want to point to the inode. | |
1613 | * | |
1614 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1615 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1616 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1617 | * | |
1618 | * This is needed in the lookup routine of any filesystem that is exportable | |
1619 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1620 | * | |
1621 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1622 | * is returned. This matches the expected return value of ->lookup. | |
1623 | * | |
1624 | */ | |
1625 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1626 | { | |
1627 | struct dentry *new = NULL; | |
1628 | ||
21c0d8fd | 1629 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1630 | spin_lock(&inode->i_lock); |
1da177e4 LT |
1631 | new = __d_find_alias(inode, 1); |
1632 | if (new) { | |
1633 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
873feea0 | 1634 | spin_unlock(&inode->i_lock); |
1da177e4 | 1635 | security_d_instantiate(new, inode); |
1da177e4 LT |
1636 | d_move(new, dentry); |
1637 | iput(inode); | |
1638 | } else { | |
873feea0 | 1639 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1640 | __d_instantiate(dentry, inode); |
873feea0 | 1641 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1642 | security_d_instantiate(dentry, inode); |
1643 | d_rehash(dentry); | |
1644 | } | |
1645 | } else | |
1646 | d_add(dentry, inode); | |
1647 | return new; | |
1648 | } | |
ec4f8605 | 1649 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1650 | |
9403540c BN |
1651 | /** |
1652 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1653 | * @inode: the inode case-insensitive lookup has found | |
1654 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1655 | * @name: the case-exact name to be associated with the returned dentry | |
1656 | * | |
1657 | * This is to avoid filling the dcache with case-insensitive names to the | |
1658 | * same inode, only the actual correct case is stored in the dcache for | |
1659 | * case-insensitive filesystems. | |
1660 | * | |
1661 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1662 | * already exists in in the dcache, use it and return it. | |
1663 | * | |
1664 | * If no entry exists with the exact case name, allocate new dentry with | |
1665 | * the exact case, and return the spliced entry. | |
1666 | */ | |
e45b590b | 1667 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1668 | struct qstr *name) |
1669 | { | |
1670 | int error; | |
1671 | struct dentry *found; | |
1672 | struct dentry *new; | |
1673 | ||
b6520c81 CH |
1674 | /* |
1675 | * First check if a dentry matching the name already exists, | |
1676 | * if not go ahead and create it now. | |
1677 | */ | |
9403540c | 1678 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1679 | if (!found) { |
1680 | new = d_alloc(dentry->d_parent, name); | |
1681 | if (!new) { | |
1682 | error = -ENOMEM; | |
1683 | goto err_out; | |
1684 | } | |
b6520c81 | 1685 | |
9403540c BN |
1686 | found = d_splice_alias(inode, new); |
1687 | if (found) { | |
1688 | dput(new); | |
1689 | return found; | |
1690 | } | |
1691 | return new; | |
1692 | } | |
b6520c81 CH |
1693 | |
1694 | /* | |
1695 | * If a matching dentry exists, and it's not negative use it. | |
1696 | * | |
1697 | * Decrement the reference count to balance the iget() done | |
1698 | * earlier on. | |
1699 | */ | |
9403540c BN |
1700 | if (found->d_inode) { |
1701 | if (unlikely(found->d_inode != inode)) { | |
1702 | /* This can't happen because bad inodes are unhashed. */ | |
1703 | BUG_ON(!is_bad_inode(inode)); | |
1704 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1705 | } | |
9403540c BN |
1706 | iput(inode); |
1707 | return found; | |
1708 | } | |
b6520c81 | 1709 | |
9403540c BN |
1710 | /* |
1711 | * Negative dentry: instantiate it unless the inode is a directory and | |
b6520c81 | 1712 | * already has a dentry. |
9403540c | 1713 | */ |
873feea0 | 1714 | spin_lock(&inode->i_lock); |
b6520c81 | 1715 | if (!S_ISDIR(inode->i_mode) || list_empty(&inode->i_dentry)) { |
360da900 | 1716 | __d_instantiate(found, inode); |
873feea0 | 1717 | spin_unlock(&inode->i_lock); |
9403540c BN |
1718 | security_d_instantiate(found, inode); |
1719 | return found; | |
1720 | } | |
b6520c81 | 1721 | |
9403540c | 1722 | /* |
b6520c81 CH |
1723 | * In case a directory already has a (disconnected) entry grab a |
1724 | * reference to it, move it in place and use it. | |
9403540c BN |
1725 | */ |
1726 | new = list_entry(inode->i_dentry.next, struct dentry, d_alias); | |
dc0474be | 1727 | __dget(new); |
873feea0 | 1728 | spin_unlock(&inode->i_lock); |
9403540c | 1729 | security_d_instantiate(found, inode); |
9403540c | 1730 | d_move(new, found); |
9403540c | 1731 | iput(inode); |
9403540c | 1732 | dput(found); |
9403540c BN |
1733 | return new; |
1734 | ||
1735 | err_out: | |
1736 | iput(inode); | |
1737 | return ERR_PTR(error); | |
1738 | } | |
ec4f8605 | 1739 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1740 | |
31e6b01f NP |
1741 | /** |
1742 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1743 | * @parent: parent dentry | |
1744 | * @name: qstr of name we wish to find | |
1745 | * @seq: returns d_seq value at the point where the dentry was found | |
1746 | * @inode: returns dentry->d_inode when the inode was found valid. | |
1747 | * Returns: dentry, or NULL | |
1748 | * | |
1749 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1750 | * resolution (store-free path walking) design described in | |
1751 | * Documentation/filesystems/path-lookup.txt. | |
1752 | * | |
1753 | * This is not to be used outside core vfs. | |
1754 | * | |
1755 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1756 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1757 | * without taking d_lock and checking d_seq sequence count against @seq | |
1758 | * returned here. | |
1759 | * | |
1760 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1761 | * function. | |
1762 | * | |
1763 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1764 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1765 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1766 | * is formed, giving integrity down the path walk. | |
1767 | */ | |
1768 | struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name, | |
1769 | unsigned *seq, struct inode **inode) | |
1770 | { | |
1771 | unsigned int len = name->len; | |
1772 | unsigned int hash = name->hash; | |
1773 | const unsigned char *str = name->name; | |
b07ad996 | 1774 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1775 | struct hlist_bl_node *node; |
31e6b01f NP |
1776 | struct dentry *dentry; |
1777 | ||
1778 | /* | |
1779 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1780 | * required to prevent single threaded performance regressions | |
1781 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1782 | * Keep the two functions in sync. | |
1783 | */ | |
1784 | ||
1785 | /* | |
1786 | * The hash list is protected using RCU. | |
1787 | * | |
1788 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1789 | * races with d_move(). | |
1790 | * | |
1791 | * It is possible that concurrent renames can mess up our list | |
1792 | * walk here and result in missing our dentry, resulting in the | |
1793 | * false-negative result. d_lookup() protects against concurrent | |
1794 | * renames using rename_lock seqlock. | |
1795 | * | |
b0a4bb83 | 1796 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1797 | */ |
b07ad996 | 1798 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1799 | struct inode *i; |
1800 | const char *tname; | |
1801 | int tlen; | |
1802 | ||
1803 | if (dentry->d_name.hash != hash) | |
1804 | continue; | |
1805 | ||
1806 | seqretry: | |
1807 | *seq = read_seqcount_begin(&dentry->d_seq); | |
1808 | if (dentry->d_parent != parent) | |
1809 | continue; | |
1810 | if (d_unhashed(dentry)) | |
1811 | continue; | |
1812 | tlen = dentry->d_name.len; | |
1813 | tname = dentry->d_name.name; | |
1814 | i = dentry->d_inode; | |
e1bb5782 NP |
1815 | prefetch(tname); |
1816 | if (i) | |
1817 | prefetch(i); | |
31e6b01f NP |
1818 | /* |
1819 | * This seqcount check is required to ensure name and | |
1820 | * len are loaded atomically, so as not to walk off the | |
1821 | * edge of memory when walking. If we could load this | |
1822 | * atomically some other way, we could drop this check. | |
1823 | */ | |
1824 | if (read_seqcount_retry(&dentry->d_seq, *seq)) | |
1825 | goto seqretry; | |
fb045adb | 1826 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
31e6b01f NP |
1827 | if (parent->d_op->d_compare(parent, *inode, |
1828 | dentry, i, | |
1829 | tlen, tname, name)) | |
1830 | continue; | |
1831 | } else { | |
9d55c369 | 1832 | if (dentry_cmp(tname, tlen, str, len)) |
31e6b01f NP |
1833 | continue; |
1834 | } | |
1835 | /* | |
1836 | * No extra seqcount check is required after the name | |
1837 | * compare. The caller must perform a seqcount check in | |
1838 | * order to do anything useful with the returned dentry | |
1839 | * anyway. | |
1840 | */ | |
1841 | *inode = i; | |
1842 | return dentry; | |
1843 | } | |
1844 | return NULL; | |
1845 | } | |
1846 | ||
1da177e4 LT |
1847 | /** |
1848 | * d_lookup - search for a dentry | |
1849 | * @parent: parent dentry | |
1850 | * @name: qstr of name we wish to find | |
b04f784e | 1851 | * Returns: dentry, or NULL |
1da177e4 | 1852 | * |
b04f784e NP |
1853 | * d_lookup searches the children of the parent dentry for the name in |
1854 | * question. If the dentry is found its reference count is incremented and the | |
1855 | * dentry is returned. The caller must use dput to free the entry when it has | |
1856 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1857 | */ |
31e6b01f | 1858 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 | 1859 | { |
31e6b01f | 1860 | struct dentry *dentry; |
949854d0 | 1861 | unsigned seq; |
1da177e4 LT |
1862 | |
1863 | do { | |
1864 | seq = read_seqbegin(&rename_lock); | |
1865 | dentry = __d_lookup(parent, name); | |
1866 | if (dentry) | |
1867 | break; | |
1868 | } while (read_seqretry(&rename_lock, seq)); | |
1869 | return dentry; | |
1870 | } | |
ec4f8605 | 1871 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1872 | |
31e6b01f | 1873 | /** |
b04f784e NP |
1874 | * __d_lookup - search for a dentry (racy) |
1875 | * @parent: parent dentry | |
1876 | * @name: qstr of name we wish to find | |
1877 | * Returns: dentry, or NULL | |
1878 | * | |
1879 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1880 | * false-negative result due to unrelated rename activity. | |
1881 | * | |
1882 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1883 | * however it must be used carefully, eg. with a following d_lookup in | |
1884 | * the case of failure. | |
1885 | * | |
1886 | * __d_lookup callers must be commented. | |
1887 | */ | |
31e6b01f | 1888 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 LT |
1889 | { |
1890 | unsigned int len = name->len; | |
1891 | unsigned int hash = name->hash; | |
1892 | const unsigned char *str = name->name; | |
b07ad996 | 1893 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1894 | struct hlist_bl_node *node; |
31e6b01f | 1895 | struct dentry *found = NULL; |
665a7583 | 1896 | struct dentry *dentry; |
1da177e4 | 1897 | |
31e6b01f NP |
1898 | /* |
1899 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1900 | * required to prevent single threaded performance regressions | |
1901 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1902 | * Keep the two functions in sync. | |
1903 | */ | |
1904 | ||
b04f784e NP |
1905 | /* |
1906 | * The hash list is protected using RCU. | |
1907 | * | |
1908 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1909 | * with d_move(). | |
1910 | * | |
1911 | * It is possible that concurrent renames can mess up our list | |
1912 | * walk here and result in missing our dentry, resulting in the | |
1913 | * false-negative result. d_lookup() protects against concurrent | |
1914 | * renames using rename_lock seqlock. | |
1915 | * | |
b0a4bb83 | 1916 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1917 | */ |
1da177e4 LT |
1918 | rcu_read_lock(); |
1919 | ||
b07ad996 | 1920 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1921 | const char *tname; |
1922 | int tlen; | |
1da177e4 | 1923 | |
1da177e4 LT |
1924 | if (dentry->d_name.hash != hash) |
1925 | continue; | |
1da177e4 LT |
1926 | |
1927 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1928 | if (dentry->d_parent != parent) |
1929 | goto next; | |
d0185c08 LT |
1930 | if (d_unhashed(dentry)) |
1931 | goto next; | |
1932 | ||
1da177e4 LT |
1933 | /* |
1934 | * It is safe to compare names since d_move() cannot | |
1935 | * change the qstr (protected by d_lock). | |
1936 | */ | |
31e6b01f NP |
1937 | tlen = dentry->d_name.len; |
1938 | tname = dentry->d_name.name; | |
fb045adb | 1939 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
621e155a NP |
1940 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1941 | dentry, dentry->d_inode, | |
31e6b01f | 1942 | tlen, tname, name)) |
1da177e4 LT |
1943 | goto next; |
1944 | } else { | |
9d55c369 | 1945 | if (dentry_cmp(tname, tlen, str, len)) |
1da177e4 LT |
1946 | goto next; |
1947 | } | |
1948 | ||
b7ab39f6 | 1949 | dentry->d_count++; |
d0185c08 | 1950 | found = dentry; |
1da177e4 LT |
1951 | spin_unlock(&dentry->d_lock); |
1952 | break; | |
1953 | next: | |
1954 | spin_unlock(&dentry->d_lock); | |
1955 | } | |
1956 | rcu_read_unlock(); | |
1957 | ||
1958 | return found; | |
1959 | } | |
1960 | ||
3e7e241f EB |
1961 | /** |
1962 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1963 | * @dir: Directory to search in | |
1964 | * @name: qstr of name we wish to find | |
1965 | * | |
1966 | * On hash failure or on lookup failure NULL is returned. | |
1967 | */ | |
1968 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1969 | { | |
1970 | struct dentry *dentry = NULL; | |
1971 | ||
1972 | /* | |
1973 | * Check for a fs-specific hash function. Note that we must | |
1974 | * calculate the standard hash first, as the d_op->d_hash() | |
1975 | * routine may choose to leave the hash value unchanged. | |
1976 | */ | |
1977 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1978 | if (dir->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 1979 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1980 | goto out; |
1981 | } | |
1982 | dentry = d_lookup(dir, name); | |
1983 | out: | |
1984 | return dentry; | |
1985 | } | |
1986 | ||
1da177e4 | 1987 | /** |
786a5e15 | 1988 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 1989 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 1990 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
1991 | * |
1992 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1993 | * This is used by ncpfs in its readdir implementation. | |
1994 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1995 | * |
1996 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1997 | */ |
d3a23e16 | 1998 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1999 | { |
786a5e15 | 2000 | struct dentry *child; |
d3a23e16 | 2001 | |
2fd6b7f5 | 2002 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2003 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2004 | if (dentry == child) { | |
2fd6b7f5 | 2005 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2006 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2007 | spin_unlock(&dentry->d_lock); |
2008 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2009 | return 1; |
2010 | } | |
2011 | } | |
2fd6b7f5 | 2012 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2013 | |
1da177e4 LT |
2014 | return 0; |
2015 | } | |
ec4f8605 | 2016 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2017 | |
2018 | /* | |
2019 | * When a file is deleted, we have two options: | |
2020 | * - turn this dentry into a negative dentry | |
2021 | * - unhash this dentry and free it. | |
2022 | * | |
2023 | * Usually, we want to just turn this into | |
2024 | * a negative dentry, but if anybody else is | |
2025 | * currently using the dentry or the inode | |
2026 | * we can't do that and we fall back on removing | |
2027 | * it from the hash queues and waiting for | |
2028 | * it to be deleted later when it has no users | |
2029 | */ | |
2030 | ||
2031 | /** | |
2032 | * d_delete - delete a dentry | |
2033 | * @dentry: The dentry to delete | |
2034 | * | |
2035 | * Turn the dentry into a negative dentry if possible, otherwise | |
2036 | * remove it from the hash queues so it can be deleted later | |
2037 | */ | |
2038 | ||
2039 | void d_delete(struct dentry * dentry) | |
2040 | { | |
873feea0 | 2041 | struct inode *inode; |
7a91bf7f | 2042 | int isdir = 0; |
1da177e4 LT |
2043 | /* |
2044 | * Are we the only user? | |
2045 | */ | |
357f8e65 | 2046 | again: |
1da177e4 | 2047 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2048 | inode = dentry->d_inode; |
2049 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 2050 | if (dentry->d_count == 1) { |
873feea0 | 2051 | if (inode && !spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2052 | spin_unlock(&dentry->d_lock); |
2053 | cpu_relax(); | |
2054 | goto again; | |
2055 | } | |
13e3c5e5 | 2056 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2057 | dentry_unlink_inode(dentry); |
7a91bf7f | 2058 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2059 | return; |
2060 | } | |
2061 | ||
2062 | if (!d_unhashed(dentry)) | |
2063 | __d_drop(dentry); | |
2064 | ||
2065 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2066 | |
2067 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2068 | } |
ec4f8605 | 2069 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2070 | |
b07ad996 | 2071 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2072 | { |
ceb5bdc2 | 2073 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2074 | hlist_bl_lock(b); |
dea3667b | 2075 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2076 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2077 | hlist_bl_unlock(b); |
1da177e4 LT |
2078 | } |
2079 | ||
770bfad8 DH |
2080 | static void _d_rehash(struct dentry * entry) |
2081 | { | |
2082 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2083 | } | |
2084 | ||
1da177e4 LT |
2085 | /** |
2086 | * d_rehash - add an entry back to the hash | |
2087 | * @entry: dentry to add to the hash | |
2088 | * | |
2089 | * Adds a dentry to the hash according to its name. | |
2090 | */ | |
2091 | ||
2092 | void d_rehash(struct dentry * entry) | |
2093 | { | |
1da177e4 | 2094 | spin_lock(&entry->d_lock); |
770bfad8 | 2095 | _d_rehash(entry); |
1da177e4 | 2096 | spin_unlock(&entry->d_lock); |
1da177e4 | 2097 | } |
ec4f8605 | 2098 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2099 | |
fb2d5b86 NP |
2100 | /** |
2101 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2102 | * @dentry: dentry to be updated | |
2103 | * @name: new name | |
2104 | * | |
2105 | * Update a case insensitive dentry with new case of name. | |
2106 | * | |
2107 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2108 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2109 | * lengths). | |
2110 | * | |
2111 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2112 | * keep renames and concurrent inserts, and readdir(2) away). | |
2113 | */ | |
2114 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2115 | { | |
7ebfa57f | 2116 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2117 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2118 | ||
fb2d5b86 | 2119 | spin_lock(&dentry->d_lock); |
31e6b01f | 2120 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2121 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2122 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2123 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2124 | } |
2125 | EXPORT_SYMBOL(dentry_update_name_case); | |
2126 | ||
1da177e4 LT |
2127 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2128 | { | |
2129 | if (dname_external(target)) { | |
2130 | if (dname_external(dentry)) { | |
2131 | /* | |
2132 | * Both external: swap the pointers | |
2133 | */ | |
9a8d5bb4 | 2134 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2135 | } else { |
2136 | /* | |
2137 | * dentry:internal, target:external. Steal target's | |
2138 | * storage and make target internal. | |
2139 | */ | |
321bcf92 BF |
2140 | memcpy(target->d_iname, dentry->d_name.name, |
2141 | dentry->d_name.len + 1); | |
1da177e4 LT |
2142 | dentry->d_name.name = target->d_name.name; |
2143 | target->d_name.name = target->d_iname; | |
2144 | } | |
2145 | } else { | |
2146 | if (dname_external(dentry)) { | |
2147 | /* | |
2148 | * dentry:external, target:internal. Give dentry's | |
2149 | * storage to target and make dentry internal | |
2150 | */ | |
2151 | memcpy(dentry->d_iname, target->d_name.name, | |
2152 | target->d_name.len + 1); | |
2153 | target->d_name.name = dentry->d_name.name; | |
2154 | dentry->d_name.name = dentry->d_iname; | |
2155 | } else { | |
2156 | /* | |
2157 | * Both are internal. Just copy target to dentry | |
2158 | */ | |
2159 | memcpy(dentry->d_iname, target->d_name.name, | |
2160 | target->d_name.len + 1); | |
dc711ca3 AV |
2161 | dentry->d_name.len = target->d_name.len; |
2162 | return; | |
1da177e4 LT |
2163 | } |
2164 | } | |
9a8d5bb4 | 2165 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2166 | } |
2167 | ||
2fd6b7f5 NP |
2168 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2169 | { | |
2170 | /* | |
2171 | * XXXX: do we really need to take target->d_lock? | |
2172 | */ | |
2173 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2174 | spin_lock(&target->d_parent->d_lock); | |
2175 | else { | |
2176 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2177 | spin_lock(&dentry->d_parent->d_lock); | |
2178 | spin_lock_nested(&target->d_parent->d_lock, | |
2179 | DENTRY_D_LOCK_NESTED); | |
2180 | } else { | |
2181 | spin_lock(&target->d_parent->d_lock); | |
2182 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2183 | DENTRY_D_LOCK_NESTED); | |
2184 | } | |
2185 | } | |
2186 | if (target < dentry) { | |
2187 | spin_lock_nested(&target->d_lock, 2); | |
2188 | spin_lock_nested(&dentry->d_lock, 3); | |
2189 | } else { | |
2190 | spin_lock_nested(&dentry->d_lock, 2); | |
2191 | spin_lock_nested(&target->d_lock, 3); | |
2192 | } | |
2193 | } | |
2194 | ||
2195 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2196 | struct dentry *target) | |
2197 | { | |
2198 | if (target->d_parent != dentry->d_parent) | |
2199 | spin_unlock(&dentry->d_parent->d_lock); | |
2200 | if (target->d_parent != target) | |
2201 | spin_unlock(&target->d_parent->d_lock); | |
2202 | } | |
2203 | ||
1da177e4 | 2204 | /* |
2fd6b7f5 NP |
2205 | * When switching names, the actual string doesn't strictly have to |
2206 | * be preserved in the target - because we're dropping the target | |
2207 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2208 | * the new name before we switch. | |
2209 | * | |
2210 | * Note that we have to be a lot more careful about getting the hash | |
2211 | * switched - we have to switch the hash value properly even if it | |
2212 | * then no longer matches the actual (corrupted) string of the target. | |
2213 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2214 | */ |
9eaef27b | 2215 | /* |
18367501 | 2216 | * __d_move - move a dentry |
1da177e4 LT |
2217 | * @dentry: entry to move |
2218 | * @target: new dentry | |
2219 | * | |
2220 | * Update the dcache to reflect the move of a file name. Negative | |
18367501 AV |
2221 | * dcache entries should not be moved in this way. Caller hold |
2222 | * rename_lock. | |
1da177e4 | 2223 | */ |
18367501 | 2224 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2225 | { |
1da177e4 LT |
2226 | if (!dentry->d_inode) |
2227 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2228 | ||
2fd6b7f5 NP |
2229 | BUG_ON(d_ancestor(dentry, target)); |
2230 | BUG_ON(d_ancestor(target, dentry)); | |
2231 | ||
2fd6b7f5 | 2232 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2233 | |
31e6b01f NP |
2234 | write_seqcount_begin(&dentry->d_seq); |
2235 | write_seqcount_begin(&target->d_seq); | |
2236 | ||
ceb5bdc2 NP |
2237 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2238 | ||
2239 | /* | |
2240 | * Move the dentry to the target hash queue. Don't bother checking | |
2241 | * for the same hash queue because of how unlikely it is. | |
2242 | */ | |
2243 | __d_drop(dentry); | |
789680d1 | 2244 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2245 | |
2246 | /* Unhash the target: dput() will then get rid of it */ | |
2247 | __d_drop(target); | |
2248 | ||
5160ee6f ED |
2249 | list_del(&dentry->d_u.d_child); |
2250 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2251 | |
2252 | /* Switch the names.. */ | |
2253 | switch_names(dentry, target); | |
9a8d5bb4 | 2254 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2255 | |
2256 | /* ... and switch the parents */ | |
2257 | if (IS_ROOT(dentry)) { | |
2258 | dentry->d_parent = target->d_parent; | |
2259 | target->d_parent = target; | |
5160ee6f | 2260 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2261 | } else { |
9a8d5bb4 | 2262 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2263 | |
2264 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2265 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2266 | } |
2267 | ||
5160ee6f | 2268 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2269 | |
31e6b01f NP |
2270 | write_seqcount_end(&target->d_seq); |
2271 | write_seqcount_end(&dentry->d_seq); | |
2272 | ||
2fd6b7f5 | 2273 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2274 | spin_unlock(&target->d_lock); |
c32ccd87 | 2275 | fsnotify_d_move(dentry); |
1da177e4 | 2276 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2277 | } |
2278 | ||
2279 | /* | |
2280 | * d_move - move a dentry | |
2281 | * @dentry: entry to move | |
2282 | * @target: new dentry | |
2283 | * | |
2284 | * Update the dcache to reflect the move of a file name. Negative | |
2285 | * dcache entries should not be moved in this way. | |
2286 | */ | |
2287 | void d_move(struct dentry *dentry, struct dentry *target) | |
2288 | { | |
2289 | write_seqlock(&rename_lock); | |
2290 | __d_move(dentry, target); | |
1da177e4 | 2291 | write_sequnlock(&rename_lock); |
9eaef27b | 2292 | } |
ec4f8605 | 2293 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2294 | |
e2761a11 OH |
2295 | /** |
2296 | * d_ancestor - search for an ancestor | |
2297 | * @p1: ancestor dentry | |
2298 | * @p2: child dentry | |
2299 | * | |
2300 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2301 | * an ancestor of p2, else NULL. | |
9eaef27b | 2302 | */ |
e2761a11 | 2303 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2304 | { |
2305 | struct dentry *p; | |
2306 | ||
871c0067 | 2307 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2308 | if (p->d_parent == p1) |
e2761a11 | 2309 | return p; |
9eaef27b | 2310 | } |
e2761a11 | 2311 | return NULL; |
9eaef27b TM |
2312 | } |
2313 | ||
2314 | /* | |
2315 | * This helper attempts to cope with remotely renamed directories | |
2316 | * | |
2317 | * It assumes that the caller is already holding | |
18367501 | 2318 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2319 | * |
2320 | * Note: If ever the locking in lock_rename() changes, then please | |
2321 | * remember to update this too... | |
9eaef27b | 2322 | */ |
873feea0 NP |
2323 | static struct dentry *__d_unalias(struct inode *inode, |
2324 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2325 | { |
2326 | struct mutex *m1 = NULL, *m2 = NULL; | |
2327 | struct dentry *ret; | |
2328 | ||
2329 | /* If alias and dentry share a parent, then no extra locks required */ | |
2330 | if (alias->d_parent == dentry->d_parent) | |
2331 | goto out_unalias; | |
2332 | ||
9eaef27b TM |
2333 | /* See lock_rename() */ |
2334 | ret = ERR_PTR(-EBUSY); | |
2335 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2336 | goto out_err; | |
2337 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2338 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2339 | goto out_err; | |
2340 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2341 | out_unalias: | |
18367501 | 2342 | __d_move(alias, dentry); |
9eaef27b TM |
2343 | ret = alias; |
2344 | out_err: | |
873feea0 | 2345 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2346 | if (m2) |
2347 | mutex_unlock(m2); | |
2348 | if (m1) | |
2349 | mutex_unlock(m1); | |
2350 | return ret; | |
2351 | } | |
2352 | ||
770bfad8 DH |
2353 | /* |
2354 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2355 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2356 | * returns with anon->d_lock held! |
770bfad8 DH |
2357 | */ |
2358 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2359 | { | |
2360 | struct dentry *dparent, *aparent; | |
2361 | ||
2fd6b7f5 | 2362 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2363 | |
31e6b01f NP |
2364 | write_seqcount_begin(&dentry->d_seq); |
2365 | write_seqcount_begin(&anon->d_seq); | |
2366 | ||
770bfad8 DH |
2367 | dparent = dentry->d_parent; |
2368 | aparent = anon->d_parent; | |
2369 | ||
2fd6b7f5 NP |
2370 | switch_names(dentry, anon); |
2371 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2372 | ||
770bfad8 DH |
2373 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2374 | list_del(&dentry->d_u.d_child); | |
2375 | if (!IS_ROOT(dentry)) | |
2376 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2377 | else | |
2378 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2379 | ||
2380 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2381 | list_del(&anon->d_u.d_child); | |
2382 | if (!IS_ROOT(anon)) | |
2383 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2384 | else | |
2385 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2386 | ||
31e6b01f NP |
2387 | write_seqcount_end(&dentry->d_seq); |
2388 | write_seqcount_end(&anon->d_seq); | |
2389 | ||
2fd6b7f5 NP |
2390 | dentry_unlock_parents_for_move(anon, dentry); |
2391 | spin_unlock(&dentry->d_lock); | |
2392 | ||
2393 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2394 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2395 | } | |
2396 | ||
2397 | /** | |
2398 | * d_materialise_unique - introduce an inode into the tree | |
2399 | * @dentry: candidate dentry | |
2400 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2401 | * | |
2402 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2403 | * root directory alias in its place if there is one | |
2404 | */ | |
2405 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2406 | { | |
9eaef27b | 2407 | struct dentry *actual; |
770bfad8 DH |
2408 | |
2409 | BUG_ON(!d_unhashed(dentry)); | |
2410 | ||
770bfad8 DH |
2411 | if (!inode) { |
2412 | actual = dentry; | |
360da900 | 2413 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2414 | d_rehash(actual); |
2415 | goto out_nolock; | |
770bfad8 DH |
2416 | } |
2417 | ||
873feea0 | 2418 | spin_lock(&inode->i_lock); |
357f8e65 | 2419 | |
9eaef27b TM |
2420 | if (S_ISDIR(inode->i_mode)) { |
2421 | struct dentry *alias; | |
2422 | ||
2423 | /* Does an aliased dentry already exist? */ | |
2424 | alias = __d_find_alias(inode, 0); | |
2425 | if (alias) { | |
2426 | actual = alias; | |
18367501 AV |
2427 | write_seqlock(&rename_lock); |
2428 | ||
2429 | if (d_ancestor(alias, dentry)) { | |
2430 | /* Check for loops */ | |
2431 | actual = ERR_PTR(-ELOOP); | |
2432 | } else if (IS_ROOT(alias)) { | |
2433 | /* Is this an anonymous mountpoint that we | |
2434 | * could splice into our tree? */ | |
9eaef27b | 2435 | __d_materialise_dentry(dentry, alias); |
18367501 | 2436 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2437 | __d_drop(alias); |
2438 | goto found; | |
18367501 AV |
2439 | } else { |
2440 | /* Nope, but we must(!) avoid directory | |
2441 | * aliasing */ | |
2442 | actual = __d_unalias(inode, dentry, alias); | |
9eaef27b | 2443 | } |
18367501 | 2444 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2445 | if (IS_ERR(actual)) |
2446 | dput(alias); | |
2447 | goto out_nolock; | |
2448 | } | |
770bfad8 DH |
2449 | } |
2450 | ||
2451 | /* Add a unique reference */ | |
2452 | actual = __d_instantiate_unique(dentry, inode); | |
2453 | if (!actual) | |
2454 | actual = dentry; | |
357f8e65 NP |
2455 | else |
2456 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2457 | |
770bfad8 DH |
2458 | spin_lock(&actual->d_lock); |
2459 | found: | |
2460 | _d_rehash(actual); | |
2461 | spin_unlock(&actual->d_lock); | |
873feea0 | 2462 | spin_unlock(&inode->i_lock); |
9eaef27b | 2463 | out_nolock: |
770bfad8 DH |
2464 | if (actual == dentry) { |
2465 | security_d_instantiate(dentry, inode); | |
2466 | return NULL; | |
2467 | } | |
2468 | ||
2469 | iput(inode); | |
2470 | return actual; | |
770bfad8 | 2471 | } |
ec4f8605 | 2472 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2473 | |
cdd16d02 | 2474 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2475 | { |
2476 | *buflen -= namelen; | |
2477 | if (*buflen < 0) | |
2478 | return -ENAMETOOLONG; | |
2479 | *buffer -= namelen; | |
2480 | memcpy(*buffer, str, namelen); | |
2481 | return 0; | |
2482 | } | |
2483 | ||
cdd16d02 MS |
2484 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2485 | { | |
2486 | return prepend(buffer, buflen, name->name, name->len); | |
2487 | } | |
2488 | ||
1da177e4 | 2489 | /** |
208898c1 | 2490 | * prepend_path - Prepend path string to a buffer |
9d1bc601 MS |
2491 | * @path: the dentry/vfsmount to report |
2492 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2493 | * @buffer: pointer to the end of the buffer |
2494 | * @buflen: pointer to buffer length | |
552ce544 | 2495 | * |
949854d0 | 2496 | * Caller holds the rename_lock. |
9d1bc601 MS |
2497 | * |
2498 | * If path is not reachable from the supplied root, then the value of | |
2499 | * root is changed (without modifying refcounts). | |
1da177e4 | 2500 | */ |
f2eb6575 MS |
2501 | static int prepend_path(const struct path *path, struct path *root, |
2502 | char **buffer, int *buflen) | |
1da177e4 | 2503 | { |
9d1bc601 MS |
2504 | struct dentry *dentry = path->dentry; |
2505 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2506 | bool slash = false; |
2507 | int error = 0; | |
6092d048 | 2508 | |
99b7db7b | 2509 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2510 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2511 | struct dentry * parent; |
2512 | ||
1da177e4 | 2513 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2514 | /* Global root? */ |
1da177e4 | 2515 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2516 | goto global_root; |
2517 | } | |
2518 | dentry = vfsmnt->mnt_mountpoint; | |
2519 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2520 | continue; |
2521 | } | |
2522 | parent = dentry->d_parent; | |
2523 | prefetch(parent); | |
9abca360 | 2524 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2525 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2526 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2527 | if (!error) |
2528 | error = prepend(buffer, buflen, "/", 1); | |
2529 | if (error) | |
2530 | break; | |
2531 | ||
2532 | slash = true; | |
1da177e4 LT |
2533 | dentry = parent; |
2534 | } | |
2535 | ||
be285c71 | 2536 | out: |
f2eb6575 MS |
2537 | if (!error && !slash) |
2538 | error = prepend(buffer, buflen, "/", 1); | |
2539 | ||
99b7db7b | 2540 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2541 | return error; |
1da177e4 LT |
2542 | |
2543 | global_root: | |
98dc568b MS |
2544 | /* |
2545 | * Filesystems needing to implement special "root names" | |
2546 | * should do so with ->d_dname() | |
2547 | */ | |
2548 | if (IS_ROOT(dentry) && | |
2549 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2550 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2551 | (int) dentry->d_name.len, dentry->d_name.name); | |
2552 | } | |
9d1bc601 MS |
2553 | root->mnt = vfsmnt; |
2554 | root->dentry = dentry; | |
be285c71 | 2555 | goto out; |
f2eb6575 | 2556 | } |
be285c71 | 2557 | |
f2eb6575 MS |
2558 | /** |
2559 | * __d_path - return the path of a dentry | |
2560 | * @path: the dentry/vfsmount to report | |
2561 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2562 | * @buf: buffer to return value in |
f2eb6575 MS |
2563 | * @buflen: buffer length |
2564 | * | |
ffd1f4ed | 2565 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2566 | * |
2567 | * Returns a pointer into the buffer or an error code if the | |
2568 | * path was too long. | |
2569 | * | |
be148247 | 2570 | * "buflen" should be positive. |
f2eb6575 MS |
2571 | * |
2572 | * If path is not reachable from the supplied root, then the value of | |
2573 | * root is changed (without modifying refcounts). | |
2574 | */ | |
2575 | char *__d_path(const struct path *path, struct path *root, | |
2576 | char *buf, int buflen) | |
2577 | { | |
2578 | char *res = buf + buflen; | |
2579 | int error; | |
2580 | ||
2581 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2582 | write_seqlock(&rename_lock); |
f2eb6575 | 2583 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2584 | write_sequnlock(&rename_lock); |
be148247 | 2585 | |
f2eb6575 MS |
2586 | if (error) |
2587 | return ERR_PTR(error); | |
f2eb6575 | 2588 | return res; |
1da177e4 LT |
2589 | } |
2590 | ||
ffd1f4ed MS |
2591 | /* |
2592 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2593 | */ | |
2594 | static int path_with_deleted(const struct path *path, struct path *root, | |
2595 | char **buf, int *buflen) | |
2596 | { | |
2597 | prepend(buf, buflen, "\0", 1); | |
2598 | if (d_unlinked(path->dentry)) { | |
2599 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2600 | if (error) | |
2601 | return error; | |
2602 | } | |
2603 | ||
2604 | return prepend_path(path, root, buf, buflen); | |
2605 | } | |
2606 | ||
8df9d1a4 MS |
2607 | static int prepend_unreachable(char **buffer, int *buflen) |
2608 | { | |
2609 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2610 | } | |
2611 | ||
a03a8a70 JB |
2612 | /** |
2613 | * d_path - return the path of a dentry | |
cf28b486 | 2614 | * @path: path to report |
a03a8a70 JB |
2615 | * @buf: buffer to return value in |
2616 | * @buflen: buffer length | |
2617 | * | |
2618 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2619 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2620 | * | |
52afeefb AV |
2621 | * Returns a pointer into the buffer or an error code if the path was |
2622 | * too long. Note: Callers should use the returned pointer, not the passed | |
2623 | * in buffer, to use the name! The implementation often starts at an offset | |
2624 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2625 | * |
31f3e0b3 | 2626 | * "buflen" should be positive. |
a03a8a70 | 2627 | */ |
20d4fdc1 | 2628 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2629 | { |
ffd1f4ed | 2630 | char *res = buf + buflen; |
6ac08c39 | 2631 | struct path root; |
9d1bc601 | 2632 | struct path tmp; |
ffd1f4ed | 2633 | int error; |
1da177e4 | 2634 | |
c23fbb6b ED |
2635 | /* |
2636 | * We have various synthetic filesystems that never get mounted. On | |
2637 | * these filesystems dentries are never used for lookup purposes, and | |
2638 | * thus don't need to be hashed. They also don't need a name until a | |
2639 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2640 | * below allows us to generate a name for these objects on demand: | |
2641 | */ | |
cf28b486 JB |
2642 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2643 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2644 | |
f7ad3c6b | 2645 | get_fs_root(current->fs, &root); |
949854d0 | 2646 | write_seqlock(&rename_lock); |
9d1bc601 | 2647 | tmp = root; |
ffd1f4ed MS |
2648 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2649 | if (error) | |
2650 | res = ERR_PTR(error); | |
949854d0 | 2651 | write_sequnlock(&rename_lock); |
6ac08c39 | 2652 | path_put(&root); |
1da177e4 LT |
2653 | return res; |
2654 | } | |
ec4f8605 | 2655 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2656 | |
8df9d1a4 MS |
2657 | /** |
2658 | * d_path_with_unreachable - return the path of a dentry | |
2659 | * @path: path to report | |
2660 | * @buf: buffer to return value in | |
2661 | * @buflen: buffer length | |
2662 | * | |
2663 | * The difference from d_path() is that this prepends "(unreachable)" | |
2664 | * to paths which are unreachable from the current process' root. | |
2665 | */ | |
2666 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2667 | { | |
2668 | char *res = buf + buflen; | |
2669 | struct path root; | |
2670 | struct path tmp; | |
2671 | int error; | |
2672 | ||
2673 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2674 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2675 | ||
2676 | get_fs_root(current->fs, &root); | |
949854d0 | 2677 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2678 | tmp = root; |
2679 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2680 | if (!error && !path_equal(&tmp, &root)) | |
2681 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2682 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2683 | path_put(&root); |
2684 | if (error) | |
2685 | res = ERR_PTR(error); | |
2686 | ||
2687 | return res; | |
2688 | } | |
2689 | ||
c23fbb6b ED |
2690 | /* |
2691 | * Helper function for dentry_operations.d_dname() members | |
2692 | */ | |
2693 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2694 | const char *fmt, ...) | |
2695 | { | |
2696 | va_list args; | |
2697 | char temp[64]; | |
2698 | int sz; | |
2699 | ||
2700 | va_start(args, fmt); | |
2701 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2702 | va_end(args); | |
2703 | ||
2704 | if (sz > sizeof(temp) || sz > buflen) | |
2705 | return ERR_PTR(-ENAMETOOLONG); | |
2706 | ||
2707 | buffer += buflen - sz; | |
2708 | return memcpy(buffer, temp, sz); | |
2709 | } | |
2710 | ||
6092d048 RP |
2711 | /* |
2712 | * Write full pathname from the root of the filesystem into the buffer. | |
2713 | */ | |
ec2447c2 | 2714 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2715 | { |
2716 | char *end = buf + buflen; | |
2717 | char *retval; | |
2718 | ||
6092d048 | 2719 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2720 | if (buflen < 1) |
2721 | goto Elong; | |
2722 | /* Get '/' right */ | |
2723 | retval = end-1; | |
2724 | *retval = '/'; | |
2725 | ||
cdd16d02 MS |
2726 | while (!IS_ROOT(dentry)) { |
2727 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2728 | int error; |
6092d048 | 2729 | |
6092d048 | 2730 | prefetch(parent); |
9abca360 NP |
2731 | spin_lock(&dentry->d_lock); |
2732 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2733 | spin_unlock(&dentry->d_lock); | |
2734 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2735 | goto Elong; |
2736 | ||
2737 | retval = end; | |
2738 | dentry = parent; | |
2739 | } | |
c103135c AV |
2740 | return retval; |
2741 | Elong: | |
2742 | return ERR_PTR(-ENAMETOOLONG); | |
2743 | } | |
ec2447c2 NP |
2744 | |
2745 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2746 | { | |
2747 | char *retval; | |
2748 | ||
949854d0 | 2749 | write_seqlock(&rename_lock); |
ec2447c2 | 2750 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2751 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2752 | |
2753 | return retval; | |
2754 | } | |
2755 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2756 | |
2757 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2758 | { | |
2759 | char *p = NULL; | |
2760 | char *retval; | |
2761 | ||
949854d0 | 2762 | write_seqlock(&rename_lock); |
c103135c AV |
2763 | if (d_unlinked(dentry)) { |
2764 | p = buf + buflen; | |
2765 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2766 | goto Elong; | |
2767 | buflen++; | |
2768 | } | |
2769 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2770 | write_sequnlock(&rename_lock); |
c103135c AV |
2771 | if (!IS_ERR(retval) && p) |
2772 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2773 | return retval; |
2774 | Elong: | |
6092d048 RP |
2775 | return ERR_PTR(-ENAMETOOLONG); |
2776 | } | |
2777 | ||
1da177e4 LT |
2778 | /* |
2779 | * NOTE! The user-level library version returns a | |
2780 | * character pointer. The kernel system call just | |
2781 | * returns the length of the buffer filled (which | |
2782 | * includes the ending '\0' character), or a negative | |
2783 | * error value. So libc would do something like | |
2784 | * | |
2785 | * char *getcwd(char * buf, size_t size) | |
2786 | * { | |
2787 | * int retval; | |
2788 | * | |
2789 | * retval = sys_getcwd(buf, size); | |
2790 | * if (retval >= 0) | |
2791 | * return buf; | |
2792 | * errno = -retval; | |
2793 | * return NULL; | |
2794 | * } | |
2795 | */ | |
3cdad428 | 2796 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2797 | { |
552ce544 | 2798 | int error; |
6ac08c39 | 2799 | struct path pwd, root; |
552ce544 | 2800 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2801 | |
2802 | if (!page) | |
2803 | return -ENOMEM; | |
2804 | ||
f7ad3c6b | 2805 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2806 | |
552ce544 | 2807 | error = -ENOENT; |
949854d0 | 2808 | write_seqlock(&rename_lock); |
f3da392e | 2809 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2810 | unsigned long len; |
9d1bc601 | 2811 | struct path tmp = root; |
8df9d1a4 MS |
2812 | char *cwd = page + PAGE_SIZE; |
2813 | int buflen = PAGE_SIZE; | |
1da177e4 | 2814 | |
8df9d1a4 MS |
2815 | prepend(&cwd, &buflen, "\0", 1); |
2816 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2817 | write_sequnlock(&rename_lock); |
552ce544 | 2818 | |
8df9d1a4 | 2819 | if (error) |
552ce544 LT |
2820 | goto out; |
2821 | ||
8df9d1a4 MS |
2822 | /* Unreachable from current root */ |
2823 | if (!path_equal(&tmp, &root)) { | |
2824 | error = prepend_unreachable(&cwd, &buflen); | |
2825 | if (error) | |
2826 | goto out; | |
2827 | } | |
2828 | ||
552ce544 LT |
2829 | error = -ERANGE; |
2830 | len = PAGE_SIZE + page - cwd; | |
2831 | if (len <= size) { | |
2832 | error = len; | |
2833 | if (copy_to_user(buf, cwd, len)) | |
2834 | error = -EFAULT; | |
2835 | } | |
949854d0 NP |
2836 | } else { |
2837 | write_sequnlock(&rename_lock); | |
949854d0 | 2838 | } |
1da177e4 LT |
2839 | |
2840 | out: | |
6ac08c39 JB |
2841 | path_put(&pwd); |
2842 | path_put(&root); | |
1da177e4 LT |
2843 | free_page((unsigned long) page); |
2844 | return error; | |
2845 | } | |
2846 | ||
2847 | /* | |
2848 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2849 | * | |
2850 | * Trivially implemented using the dcache structure | |
2851 | */ | |
2852 | ||
2853 | /** | |
2854 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2855 | * @new_dentry: new dentry | |
2856 | * @old_dentry: old dentry | |
2857 | * | |
2858 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2859 | * Returns 0 otherwise. | |
2860 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2861 | */ | |
2862 | ||
e2761a11 | 2863 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2864 | { |
2865 | int result; | |
949854d0 | 2866 | unsigned seq; |
1da177e4 | 2867 | |
e2761a11 OH |
2868 | if (new_dentry == old_dentry) |
2869 | return 1; | |
2870 | ||
e2761a11 | 2871 | do { |
1da177e4 | 2872 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2873 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2874 | /* |
2875 | * Need rcu_readlock to protect against the d_parent trashing | |
2876 | * due to d_move | |
2877 | */ | |
2878 | rcu_read_lock(); | |
e2761a11 | 2879 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2880 | result = 1; |
e2761a11 OH |
2881 | else |
2882 | result = 0; | |
949854d0 | 2883 | rcu_read_unlock(); |
1da177e4 | 2884 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2885 | |
2886 | return result; | |
2887 | } | |
2888 | ||
2096f759 AV |
2889 | int path_is_under(struct path *path1, struct path *path2) |
2890 | { | |
2891 | struct vfsmount *mnt = path1->mnt; | |
2892 | struct dentry *dentry = path1->dentry; | |
2893 | int res; | |
99b7db7b NP |
2894 | |
2895 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2896 | if (mnt != path2->mnt) { |
2897 | for (;;) { | |
2898 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2899 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2900 | return 0; |
2901 | } | |
2902 | if (mnt->mnt_parent == path2->mnt) | |
2903 | break; | |
2904 | mnt = mnt->mnt_parent; | |
2905 | } | |
2906 | dentry = mnt->mnt_mountpoint; | |
2907 | } | |
2908 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2909 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2910 | return res; |
2911 | } | |
2912 | EXPORT_SYMBOL(path_is_under); | |
2913 | ||
1da177e4 LT |
2914 | void d_genocide(struct dentry *root) |
2915 | { | |
949854d0 | 2916 | struct dentry *this_parent; |
1da177e4 | 2917 | struct list_head *next; |
949854d0 | 2918 | unsigned seq; |
58db63d0 | 2919 | int locked = 0; |
1da177e4 | 2920 | |
949854d0 | 2921 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2922 | again: |
2923 | this_parent = root; | |
2fd6b7f5 | 2924 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2925 | repeat: |
2926 | next = this_parent->d_subdirs.next; | |
2927 | resume: | |
2928 | while (next != &this_parent->d_subdirs) { | |
2929 | struct list_head *tmp = next; | |
5160ee6f | 2930 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2931 | next = tmp->next; |
949854d0 | 2932 | |
da502956 NP |
2933 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2934 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2935 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2936 | continue; |
da502956 | 2937 | } |
1da177e4 | 2938 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2939 | spin_unlock(&this_parent->d_lock); |
2940 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2941 | this_parent = dentry; |
2fd6b7f5 | 2942 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2943 | goto repeat; |
2944 | } | |
949854d0 NP |
2945 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2946 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2947 | dentry->d_count--; | |
2948 | } | |
b7ab39f6 | 2949 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2950 | } |
2951 | if (this_parent != root) { | |
c826cb7d | 2952 | struct dentry *child = this_parent; |
949854d0 NP |
2953 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2954 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2955 | this_parent->d_count--; | |
2956 | } | |
c826cb7d LT |
2957 | this_parent = try_to_ascend(this_parent, locked, seq); |
2958 | if (!this_parent) | |
949854d0 | 2959 | goto rename_retry; |
949854d0 | 2960 | next = child->d_u.d_child.next; |
1da177e4 LT |
2961 | goto resume; |
2962 | } | |
2fd6b7f5 | 2963 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2964 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2965 | goto rename_retry; |
58db63d0 NP |
2966 | if (locked) |
2967 | write_sequnlock(&rename_lock); | |
2968 | return; | |
2969 | ||
2970 | rename_retry: | |
2971 | locked = 1; | |
2972 | write_seqlock(&rename_lock); | |
2973 | goto again; | |
1da177e4 LT |
2974 | } |
2975 | ||
2976 | /** | |
2977 | * find_inode_number - check for dentry with name | |
2978 | * @dir: directory to check | |
2979 | * @name: Name to find. | |
2980 | * | |
2981 | * Check whether a dentry already exists for the given name, | |
2982 | * and return the inode number if it has an inode. Otherwise | |
2983 | * 0 is returned. | |
2984 | * | |
2985 | * This routine is used to post-process directory listings for | |
2986 | * filesystems using synthetic inode numbers, and is necessary | |
2987 | * to keep getcwd() working. | |
2988 | */ | |
2989 | ||
2990 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2991 | { | |
2992 | struct dentry * dentry; | |
2993 | ino_t ino = 0; | |
2994 | ||
3e7e241f EB |
2995 | dentry = d_hash_and_lookup(dir, name); |
2996 | if (dentry) { | |
1da177e4 LT |
2997 | if (dentry->d_inode) |
2998 | ino = dentry->d_inode->i_ino; | |
2999 | dput(dentry); | |
3000 | } | |
1da177e4 LT |
3001 | return ino; |
3002 | } | |
ec4f8605 | 3003 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
3004 | |
3005 | static __initdata unsigned long dhash_entries; | |
3006 | static int __init set_dhash_entries(char *str) | |
3007 | { | |
3008 | if (!str) | |
3009 | return 0; | |
3010 | dhash_entries = simple_strtoul(str, &str, 0); | |
3011 | return 1; | |
3012 | } | |
3013 | __setup("dhash_entries=", set_dhash_entries); | |
3014 | ||
3015 | static void __init dcache_init_early(void) | |
3016 | { | |
3017 | int loop; | |
3018 | ||
3019 | /* If hashes are distributed across NUMA nodes, defer | |
3020 | * hash allocation until vmalloc space is available. | |
3021 | */ | |
3022 | if (hashdist) | |
3023 | return; | |
3024 | ||
3025 | dentry_hashtable = | |
3026 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3027 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3028 | dhash_entries, |
3029 | 13, | |
3030 | HASH_EARLY, | |
3031 | &d_hash_shift, | |
3032 | &d_hash_mask, | |
3033 | 0); | |
3034 | ||
3035 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 3036 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3037 | } |
3038 | ||
74bf17cf | 3039 | static void __init dcache_init(void) |
1da177e4 LT |
3040 | { |
3041 | int loop; | |
3042 | ||
3043 | /* | |
3044 | * A constructor could be added for stable state like the lists, | |
3045 | * but it is probably not worth it because of the cache nature | |
3046 | * of the dcache. | |
3047 | */ | |
0a31bd5f CL |
3048 | dentry_cache = KMEM_CACHE(dentry, |
3049 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 | 3050 | |
8e1f936b | 3051 | register_shrinker(&dcache_shrinker); |
1da177e4 LT |
3052 | |
3053 | /* Hash may have been set up in dcache_init_early */ | |
3054 | if (!hashdist) | |
3055 | return; | |
3056 | ||
3057 | dentry_hashtable = | |
3058 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3059 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3060 | dhash_entries, |
3061 | 13, | |
3062 | 0, | |
3063 | &d_hash_shift, | |
3064 | &d_hash_mask, | |
3065 | 0); | |
3066 | ||
3067 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 3068 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3069 | } |
3070 | ||
3071 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3072 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3073 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3074 | |
1da177e4 LT |
3075 | EXPORT_SYMBOL(d_genocide); |
3076 | ||
1da177e4 LT |
3077 | void __init vfs_caches_init_early(void) |
3078 | { | |
3079 | dcache_init_early(); | |
3080 | inode_init_early(); | |
3081 | } | |
3082 | ||
3083 | void __init vfs_caches_init(unsigned long mempages) | |
3084 | { | |
3085 | unsigned long reserve; | |
3086 | ||
3087 | /* Base hash sizes on available memory, with a reserve equal to | |
3088 | 150% of current kernel size */ | |
3089 | ||
3090 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3091 | mempages -= reserve; | |
3092 | ||
3093 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3094 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3095 | |
74bf17cf DC |
3096 | dcache_init(); |
3097 | inode_init(); | |
1da177e4 | 3098 | files_init(mempages); |
74bf17cf | 3099 | mnt_init(); |
1da177e4 LT |
3100 | bdev_cache_init(); |
3101 | chrdev_init(); | |
3102 | } |