]>
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 | ||
44396f4b JB |
346 | /* |
347 | * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag | |
348 | * @dentry: dentry to drop | |
349 | * | |
350 | * This is called when we do a lookup on a placeholder dentry that needed to be | |
351 | * looked up. The dentry should have been hashed in order for it to be found by | |
352 | * the lookup code, but now needs to be unhashed while we do the actual lookup | |
353 | * and clear the DCACHE_NEED_LOOKUP flag. | |
354 | */ | |
355 | void d_clear_need_lookup(struct dentry *dentry) | |
356 | { | |
357 | spin_lock(&dentry->d_lock); | |
358 | __d_drop(dentry); | |
359 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
360 | spin_unlock(&dentry->d_lock); | |
361 | } | |
362 | EXPORT_SYMBOL(d_clear_need_lookup); | |
363 | ||
77812a1e NP |
364 | /* |
365 | * Finish off a dentry we've decided to kill. | |
366 | * dentry->d_lock must be held, returns with it unlocked. | |
367 | * If ref is non-zero, then decrement the refcount too. | |
368 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
369 | */ | |
370 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
371 | __releases(dentry->d_lock) | |
372 | { | |
873feea0 | 373 | struct inode *inode; |
77812a1e NP |
374 | struct dentry *parent; |
375 | ||
873feea0 NP |
376 | inode = dentry->d_inode; |
377 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
378 | relock: |
379 | spin_unlock(&dentry->d_lock); | |
380 | cpu_relax(); | |
381 | return dentry; /* try again with same dentry */ | |
382 | } | |
383 | if (IS_ROOT(dentry)) | |
384 | parent = NULL; | |
385 | else | |
386 | parent = dentry->d_parent; | |
387 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
388 | if (inode) |
389 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
390 | goto relock; |
391 | } | |
31e6b01f | 392 | |
77812a1e NP |
393 | if (ref) |
394 | dentry->d_count--; | |
395 | /* if dentry was on the d_lru list delete it from there */ | |
396 | dentry_lru_del(dentry); | |
397 | /* if it was on the hash then remove it */ | |
398 | __d_drop(dentry); | |
399 | return d_kill(dentry, parent); | |
400 | } | |
401 | ||
1da177e4 LT |
402 | /* |
403 | * This is dput | |
404 | * | |
405 | * This is complicated by the fact that we do not want to put | |
406 | * dentries that are no longer on any hash chain on the unused | |
407 | * list: we'd much rather just get rid of them immediately. | |
408 | * | |
409 | * However, that implies that we have to traverse the dentry | |
410 | * tree upwards to the parents which might _also_ now be | |
411 | * scheduled for deletion (it may have been only waiting for | |
412 | * its last child to go away). | |
413 | * | |
414 | * This tail recursion is done by hand as we don't want to depend | |
415 | * on the compiler to always get this right (gcc generally doesn't). | |
416 | * Real recursion would eat up our stack space. | |
417 | */ | |
418 | ||
419 | /* | |
420 | * dput - release a dentry | |
421 | * @dentry: dentry to release | |
422 | * | |
423 | * Release a dentry. This will drop the usage count and if appropriate | |
424 | * call the dentry unlink method as well as removing it from the queues and | |
425 | * releasing its resources. If the parent dentries were scheduled for release | |
426 | * they too may now get deleted. | |
1da177e4 | 427 | */ |
1da177e4 LT |
428 | void dput(struct dentry *dentry) |
429 | { | |
430 | if (!dentry) | |
431 | return; | |
432 | ||
433 | repeat: | |
b7ab39f6 | 434 | if (dentry->d_count == 1) |
1da177e4 | 435 | might_sleep(); |
1da177e4 | 436 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
437 | BUG_ON(!dentry->d_count); |
438 | if (dentry->d_count > 1) { | |
439 | dentry->d_count--; | |
1da177e4 | 440 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
441 | return; |
442 | } | |
443 | ||
fb045adb | 444 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 445 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 446 | goto kill_it; |
1da177e4 | 447 | } |
265ac902 | 448 | |
1da177e4 LT |
449 | /* Unreachable? Get rid of it */ |
450 | if (d_unhashed(dentry)) | |
451 | goto kill_it; | |
265ac902 | 452 | |
44396f4b JB |
453 | /* |
454 | * If this dentry needs lookup, don't set the referenced flag so that it | |
455 | * is more likely to be cleaned up by the dcache shrinker in case of | |
456 | * memory pressure. | |
457 | */ | |
458 | if (!d_need_lookup(dentry)) | |
459 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 460 | dentry_lru_add(dentry); |
265ac902 | 461 | |
61f3dee4 NP |
462 | dentry->d_count--; |
463 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
464 | return; |
465 | ||
d52b9086 | 466 | kill_it: |
77812a1e | 467 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
468 | if (dentry) |
469 | goto repeat; | |
1da177e4 | 470 | } |
ec4f8605 | 471 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
472 | |
473 | /** | |
474 | * d_invalidate - invalidate a dentry | |
475 | * @dentry: dentry to invalidate | |
476 | * | |
477 | * Try to invalidate the dentry if it turns out to be | |
478 | * possible. If there are other dentries that can be | |
479 | * reached through this one we can't delete it and we | |
480 | * return -EBUSY. On success we return 0. | |
481 | * | |
482 | * no dcache lock. | |
483 | */ | |
484 | ||
485 | int d_invalidate(struct dentry * dentry) | |
486 | { | |
487 | /* | |
488 | * If it's already been dropped, return OK. | |
489 | */ | |
da502956 | 490 | spin_lock(&dentry->d_lock); |
1da177e4 | 491 | if (d_unhashed(dentry)) { |
da502956 | 492 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
493 | return 0; |
494 | } | |
495 | /* | |
496 | * Check whether to do a partial shrink_dcache | |
497 | * to get rid of unused child entries. | |
498 | */ | |
499 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 500 | spin_unlock(&dentry->d_lock); |
1da177e4 | 501 | shrink_dcache_parent(dentry); |
da502956 | 502 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
503 | } |
504 | ||
505 | /* | |
506 | * Somebody else still using it? | |
507 | * | |
508 | * If it's a directory, we can't drop it | |
509 | * for fear of somebody re-populating it | |
510 | * with children (even though dropping it | |
511 | * would make it unreachable from the root, | |
512 | * we might still populate it if it was a | |
513 | * working directory or similar). | |
514 | */ | |
b7ab39f6 | 515 | if (dentry->d_count > 1) { |
1da177e4 LT |
516 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
517 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
518 | return -EBUSY; |
519 | } | |
520 | } | |
521 | ||
522 | __d_drop(dentry); | |
523 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
524 | return 0; |
525 | } | |
ec4f8605 | 526 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 527 | |
b5c84bf6 | 528 | /* This must be called with d_lock held */ |
dc0474be | 529 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 530 | { |
b7ab39f6 | 531 | dentry->d_count++; |
23044507 NP |
532 | } |
533 | ||
dc0474be | 534 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 535 | { |
23044507 | 536 | spin_lock(&dentry->d_lock); |
dc0474be | 537 | __dget_dlock(dentry); |
23044507 | 538 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
539 | } |
540 | ||
b7ab39f6 NP |
541 | struct dentry *dget_parent(struct dentry *dentry) |
542 | { | |
543 | struct dentry *ret; | |
544 | ||
545 | repeat: | |
a734eb45 NP |
546 | /* |
547 | * Don't need rcu_dereference because we re-check it was correct under | |
548 | * the lock. | |
549 | */ | |
550 | rcu_read_lock(); | |
b7ab39f6 | 551 | ret = dentry->d_parent; |
a734eb45 NP |
552 | spin_lock(&ret->d_lock); |
553 | if (unlikely(ret != dentry->d_parent)) { | |
554 | spin_unlock(&ret->d_lock); | |
555 | rcu_read_unlock(); | |
b7ab39f6 NP |
556 | goto repeat; |
557 | } | |
a734eb45 | 558 | rcu_read_unlock(); |
b7ab39f6 NP |
559 | BUG_ON(!ret->d_count); |
560 | ret->d_count++; | |
561 | spin_unlock(&ret->d_lock); | |
b7ab39f6 NP |
562 | return ret; |
563 | } | |
564 | EXPORT_SYMBOL(dget_parent); | |
565 | ||
1da177e4 LT |
566 | /** |
567 | * d_find_alias - grab a hashed alias of inode | |
568 | * @inode: inode in question | |
569 | * @want_discon: flag, used by d_splice_alias, to request | |
570 | * that only a DISCONNECTED alias be returned. | |
571 | * | |
572 | * If inode has a hashed alias, or is a directory and has any alias, | |
573 | * acquire the reference to alias and return it. Otherwise return NULL. | |
574 | * Notice that if inode is a directory there can be only one alias and | |
575 | * it can be unhashed only if it has no children, or if it is the root | |
576 | * of a filesystem. | |
577 | * | |
21c0d8fd | 578 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 579 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 580 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 581 | */ |
da502956 | 582 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 583 | { |
da502956 | 584 | struct dentry *alias, *discon_alias; |
1da177e4 | 585 | |
da502956 NP |
586 | again: |
587 | discon_alias = NULL; | |
588 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
589 | spin_lock(&alias->d_lock); | |
1da177e4 | 590 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 591 | if (IS_ROOT(alias) && |
da502956 | 592 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 593 | discon_alias = alias; |
da502956 | 594 | } else if (!want_discon) { |
dc0474be | 595 | __dget_dlock(alias); |
da502956 NP |
596 | spin_unlock(&alias->d_lock); |
597 | return alias; | |
598 | } | |
599 | } | |
600 | spin_unlock(&alias->d_lock); | |
601 | } | |
602 | if (discon_alias) { | |
603 | alias = discon_alias; | |
604 | spin_lock(&alias->d_lock); | |
605 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
606 | if (IS_ROOT(alias) && | |
607 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 608 | __dget_dlock(alias); |
da502956 | 609 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
610 | return alias; |
611 | } | |
612 | } | |
da502956 NP |
613 | spin_unlock(&alias->d_lock); |
614 | goto again; | |
1da177e4 | 615 | } |
da502956 | 616 | return NULL; |
1da177e4 LT |
617 | } |
618 | ||
da502956 | 619 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 620 | { |
214fda1f DH |
621 | struct dentry *de = NULL; |
622 | ||
623 | if (!list_empty(&inode->i_dentry)) { | |
873feea0 | 624 | spin_lock(&inode->i_lock); |
214fda1f | 625 | de = __d_find_alias(inode, 0); |
873feea0 | 626 | spin_unlock(&inode->i_lock); |
214fda1f | 627 | } |
1da177e4 LT |
628 | return de; |
629 | } | |
ec4f8605 | 630 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
631 | |
632 | /* | |
633 | * Try to kill dentries associated with this inode. | |
634 | * WARNING: you must own a reference to inode. | |
635 | */ | |
636 | void d_prune_aliases(struct inode *inode) | |
637 | { | |
0cdca3f9 | 638 | struct dentry *dentry; |
1da177e4 | 639 | restart: |
873feea0 | 640 | spin_lock(&inode->i_lock); |
0cdca3f9 | 641 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 642 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 643 | if (!dentry->d_count) { |
dc0474be | 644 | __dget_dlock(dentry); |
1da177e4 LT |
645 | __d_drop(dentry); |
646 | spin_unlock(&dentry->d_lock); | |
873feea0 | 647 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
648 | dput(dentry); |
649 | goto restart; | |
650 | } | |
651 | spin_unlock(&dentry->d_lock); | |
652 | } | |
873feea0 | 653 | spin_unlock(&inode->i_lock); |
1da177e4 | 654 | } |
ec4f8605 | 655 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
656 | |
657 | /* | |
77812a1e NP |
658 | * Try to throw away a dentry - free the inode, dput the parent. |
659 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
660 | * Releases dentry->d_lock. | |
d702ccb3 | 661 | * |
77812a1e | 662 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 663 | */ |
77812a1e | 664 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 665 | __releases(dentry->d_lock) |
1da177e4 | 666 | { |
77812a1e | 667 | struct dentry *parent; |
d52b9086 | 668 | |
77812a1e | 669 | parent = dentry_kill(dentry, 0); |
d52b9086 | 670 | /* |
77812a1e NP |
671 | * If dentry_kill returns NULL, we have nothing more to do. |
672 | * if it returns the same dentry, trylocks failed. In either | |
673 | * case, just loop again. | |
674 | * | |
675 | * Otherwise, we need to prune ancestors too. This is necessary | |
676 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
677 | * is also expected to be beneficial in reducing dentry cache | |
678 | * fragmentation. | |
d52b9086 | 679 | */ |
77812a1e NP |
680 | if (!parent) |
681 | return; | |
682 | if (parent == dentry) | |
683 | return; | |
684 | ||
685 | /* Prune ancestors. */ | |
686 | dentry = parent; | |
d52b9086 | 687 | while (dentry) { |
b7ab39f6 | 688 | spin_lock(&dentry->d_lock); |
89e60548 NP |
689 | if (dentry->d_count > 1) { |
690 | dentry->d_count--; | |
691 | spin_unlock(&dentry->d_lock); | |
692 | return; | |
693 | } | |
77812a1e | 694 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 695 | } |
1da177e4 LT |
696 | } |
697 | ||
3049cfe2 | 698 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 699 | { |
da3bbdd4 | 700 | struct dentry *dentry; |
da3bbdd4 | 701 | |
ec33679d NP |
702 | rcu_read_lock(); |
703 | for (;;) { | |
ec33679d NP |
704 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
705 | if (&dentry->d_lru == list) | |
706 | break; /* empty */ | |
707 | spin_lock(&dentry->d_lock); | |
708 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
709 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
710 | continue; |
711 | } | |
712 | ||
1da177e4 LT |
713 | /* |
714 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
715 | * the LRU because of laziness during lookup. Do not free |
716 | * it - just keep it off the LRU list. | |
1da177e4 | 717 | */ |
b7ab39f6 | 718 | if (dentry->d_count) { |
ec33679d | 719 | dentry_lru_del(dentry); |
da3bbdd4 | 720 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
721 | continue; |
722 | } | |
ec33679d | 723 | |
ec33679d | 724 | rcu_read_unlock(); |
77812a1e NP |
725 | |
726 | try_prune_one_dentry(dentry); | |
727 | ||
ec33679d | 728 | rcu_read_lock(); |
da3bbdd4 | 729 | } |
ec33679d | 730 | rcu_read_unlock(); |
3049cfe2 CH |
731 | } |
732 | ||
733 | /** | |
734 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
735 | * @sb: superblock to shrink dentry LRU. | |
736 | * @count: number of entries to prune | |
737 | * @flags: flags to control the dentry processing | |
738 | * | |
739 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
740 | */ | |
b0d40c92 | 741 | static void __shrink_dcache_sb(struct super_block *sb, int count, int flags) |
3049cfe2 | 742 | { |
3049cfe2 CH |
743 | struct dentry *dentry; |
744 | LIST_HEAD(referenced); | |
745 | LIST_HEAD(tmp); | |
3049cfe2 | 746 | |
23044507 NP |
747 | relock: |
748 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
749 | while (!list_empty(&sb->s_dentry_lru)) { |
750 | dentry = list_entry(sb->s_dentry_lru.prev, | |
751 | struct dentry, d_lru); | |
752 | BUG_ON(dentry->d_sb != sb); | |
753 | ||
23044507 NP |
754 | if (!spin_trylock(&dentry->d_lock)) { |
755 | spin_unlock(&dcache_lru_lock); | |
756 | cpu_relax(); | |
757 | goto relock; | |
758 | } | |
759 | ||
3049cfe2 CH |
760 | /* |
761 | * If we are honouring the DCACHE_REFERENCED flag and the | |
762 | * dentry has this flag set, don't free it. Clear the flag | |
763 | * and put it back on the LRU. | |
764 | */ | |
23044507 NP |
765 | if (flags & DCACHE_REFERENCED && |
766 | dentry->d_flags & DCACHE_REFERENCED) { | |
767 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
768 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 769 | spin_unlock(&dentry->d_lock); |
23044507 NP |
770 | } else { |
771 | list_move_tail(&dentry->d_lru, &tmp); | |
772 | spin_unlock(&dentry->d_lock); | |
b0d40c92 | 773 | if (!--count) |
23044507 | 774 | break; |
3049cfe2 | 775 | } |
ec33679d | 776 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 777 | } |
da3bbdd4 KM |
778 | if (!list_empty(&referenced)) |
779 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 780 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
781 | |
782 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
783 | } |
784 | ||
785 | /** | |
b0d40c92 DC |
786 | * prune_dcache_sb - shrink the dcache |
787 | * @nr_to_scan: number of entries to try to free | |
da3bbdd4 | 788 | * |
b0d40c92 DC |
789 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
790 | * done when we need more memory an called from the superblock shrinker | |
791 | * function. | |
da3bbdd4 | 792 | * |
b0d40c92 DC |
793 | * This function may fail to free any resources if all the dentries are in |
794 | * use. | |
da3bbdd4 | 795 | */ |
b0d40c92 | 796 | void prune_dcache_sb(struct super_block *sb, int nr_to_scan) |
da3bbdd4 | 797 | { |
b0d40c92 | 798 | __shrink_dcache_sb(sb, nr_to_scan, DCACHE_REFERENCED); |
1da177e4 LT |
799 | } |
800 | ||
1da177e4 LT |
801 | /** |
802 | * shrink_dcache_sb - shrink dcache for a superblock | |
803 | * @sb: superblock | |
804 | * | |
3049cfe2 CH |
805 | * Shrink the dcache for the specified super block. This is used to free |
806 | * the dcache before unmounting a file system. | |
1da177e4 | 807 | */ |
3049cfe2 | 808 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 809 | { |
3049cfe2 CH |
810 | LIST_HEAD(tmp); |
811 | ||
23044507 | 812 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
813 | while (!list_empty(&sb->s_dentry_lru)) { |
814 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 815 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 816 | shrink_dentry_list(&tmp); |
ec33679d | 817 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 818 | } |
23044507 | 819 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 820 | } |
ec4f8605 | 821 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 822 | |
c636ebdb DH |
823 | /* |
824 | * destroy a single subtree of dentries for unmount | |
825 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
826 | * locking | |
827 | */ | |
828 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
829 | { | |
830 | struct dentry *parent; | |
f8713576 | 831 | unsigned detached = 0; |
c636ebdb DH |
832 | |
833 | BUG_ON(!IS_ROOT(dentry)); | |
834 | ||
835 | /* detach this root from the system */ | |
23044507 | 836 | spin_lock(&dentry->d_lock); |
a4633357 | 837 | dentry_lru_del(dentry); |
c636ebdb | 838 | __d_drop(dentry); |
da502956 | 839 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
840 | |
841 | for (;;) { | |
842 | /* descend to the first leaf in the current subtree */ | |
843 | while (!list_empty(&dentry->d_subdirs)) { | |
844 | struct dentry *loop; | |
845 | ||
846 | /* this is a branch with children - detach all of them | |
847 | * from the system in one go */ | |
2fd6b7f5 | 848 | spin_lock(&dentry->d_lock); |
c636ebdb DH |
849 | list_for_each_entry(loop, &dentry->d_subdirs, |
850 | d_u.d_child) { | |
2fd6b7f5 NP |
851 | spin_lock_nested(&loop->d_lock, |
852 | DENTRY_D_LOCK_NESTED); | |
a4633357 | 853 | dentry_lru_del(loop); |
c636ebdb | 854 | __d_drop(loop); |
da502956 | 855 | spin_unlock(&loop->d_lock); |
c636ebdb | 856 | } |
2fd6b7f5 | 857 | spin_unlock(&dentry->d_lock); |
c636ebdb DH |
858 | |
859 | /* move to the first child */ | |
860 | dentry = list_entry(dentry->d_subdirs.next, | |
861 | struct dentry, d_u.d_child); | |
862 | } | |
863 | ||
864 | /* consume the dentries from this leaf up through its parents | |
865 | * until we find one with children or run out altogether */ | |
866 | do { | |
867 | struct inode *inode; | |
868 | ||
b7ab39f6 | 869 | if (dentry->d_count != 0) { |
c636ebdb DH |
870 | printk(KERN_ERR |
871 | "BUG: Dentry %p{i=%lx,n=%s}" | |
872 | " still in use (%d)" | |
873 | " [unmount of %s %s]\n", | |
874 | dentry, | |
875 | dentry->d_inode ? | |
876 | dentry->d_inode->i_ino : 0UL, | |
877 | dentry->d_name.name, | |
b7ab39f6 | 878 | dentry->d_count, |
c636ebdb DH |
879 | dentry->d_sb->s_type->name, |
880 | dentry->d_sb->s_id); | |
881 | BUG(); | |
882 | } | |
883 | ||
2fd6b7f5 | 884 | if (IS_ROOT(dentry)) { |
c636ebdb | 885 | parent = NULL; |
2fd6b7f5 NP |
886 | list_del(&dentry->d_u.d_child); |
887 | } else { | |
871c0067 | 888 | parent = dentry->d_parent; |
b7ab39f6 NP |
889 | spin_lock(&parent->d_lock); |
890 | parent->d_count--; | |
2fd6b7f5 | 891 | list_del(&dentry->d_u.d_child); |
b7ab39f6 | 892 | spin_unlock(&parent->d_lock); |
871c0067 | 893 | } |
c636ebdb | 894 | |
f8713576 | 895 | detached++; |
c636ebdb DH |
896 | |
897 | inode = dentry->d_inode; | |
898 | if (inode) { | |
899 | dentry->d_inode = NULL; | |
900 | list_del_init(&dentry->d_alias); | |
901 | if (dentry->d_op && dentry->d_op->d_iput) | |
902 | dentry->d_op->d_iput(dentry, inode); | |
903 | else | |
904 | iput(inode); | |
905 | } | |
906 | ||
907 | d_free(dentry); | |
908 | ||
909 | /* finished when we fall off the top of the tree, | |
910 | * otherwise we ascend to the parent and move to the | |
911 | * next sibling if there is one */ | |
912 | if (!parent) | |
312d3ca8 | 913 | return; |
c636ebdb | 914 | dentry = parent; |
c636ebdb DH |
915 | } while (list_empty(&dentry->d_subdirs)); |
916 | ||
917 | dentry = list_entry(dentry->d_subdirs.next, | |
918 | struct dentry, d_u.d_child); | |
919 | } | |
920 | } | |
921 | ||
922 | /* | |
923 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 924 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
925 | * - the superblock is detached from all mountings and open files, so the |
926 | * dentry trees will not be rearranged by the VFS | |
927 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
928 | * any dentries belonging to this superblock that it comes across | |
929 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
930 | * in this superblock | |
931 | */ | |
932 | void shrink_dcache_for_umount(struct super_block *sb) | |
933 | { | |
934 | struct dentry *dentry; | |
935 | ||
936 | if (down_read_trylock(&sb->s_umount)) | |
937 | BUG(); | |
938 | ||
939 | dentry = sb->s_root; | |
940 | sb->s_root = NULL; | |
b7ab39f6 NP |
941 | spin_lock(&dentry->d_lock); |
942 | dentry->d_count--; | |
943 | spin_unlock(&dentry->d_lock); | |
c636ebdb DH |
944 | shrink_dcache_for_umount_subtree(dentry); |
945 | ||
ceb5bdc2 NP |
946 | while (!hlist_bl_empty(&sb->s_anon)) { |
947 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
948 | shrink_dcache_for_umount_subtree(dentry); |
949 | } | |
950 | } | |
951 | ||
c826cb7d LT |
952 | /* |
953 | * This tries to ascend one level of parenthood, but | |
954 | * we can race with renaming, so we need to re-check | |
955 | * the parenthood after dropping the lock and check | |
956 | * that the sequence number still matches. | |
957 | */ | |
958 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
959 | { | |
960 | struct dentry *new = old->d_parent; | |
961 | ||
962 | rcu_read_lock(); | |
963 | spin_unlock(&old->d_lock); | |
964 | spin_lock(&new->d_lock); | |
965 | ||
966 | /* | |
967 | * might go back up the wrong parent if we have had a rename | |
968 | * or deletion | |
969 | */ | |
970 | if (new != old->d_parent || | |
c83ce989 | 971 | (old->d_flags & DCACHE_DISCONNECTED) || |
c826cb7d LT |
972 | (!locked && read_seqretry(&rename_lock, seq))) { |
973 | spin_unlock(&new->d_lock); | |
974 | new = NULL; | |
975 | } | |
976 | rcu_read_unlock(); | |
977 | return new; | |
978 | } | |
979 | ||
980 | ||
1da177e4 LT |
981 | /* |
982 | * Search for at least 1 mount point in the dentry's subdirs. | |
983 | * We descend to the next level whenever the d_subdirs | |
984 | * list is non-empty and continue searching. | |
985 | */ | |
986 | ||
987 | /** | |
988 | * have_submounts - check for mounts over a dentry | |
989 | * @parent: dentry to check. | |
990 | * | |
991 | * Return true if the parent or its subdirectories contain | |
992 | * a mount point | |
993 | */ | |
1da177e4 LT |
994 | int have_submounts(struct dentry *parent) |
995 | { | |
949854d0 | 996 | struct dentry *this_parent; |
1da177e4 | 997 | struct list_head *next; |
949854d0 | 998 | unsigned seq; |
58db63d0 | 999 | int locked = 0; |
949854d0 | 1000 | |
949854d0 | 1001 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1002 | again: |
1003 | this_parent = parent; | |
1da177e4 | 1004 | |
1da177e4 LT |
1005 | if (d_mountpoint(parent)) |
1006 | goto positive; | |
2fd6b7f5 | 1007 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1008 | repeat: |
1009 | next = this_parent->d_subdirs.next; | |
1010 | resume: | |
1011 | while (next != &this_parent->d_subdirs) { | |
1012 | struct list_head *tmp = next; | |
5160ee6f | 1013 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1014 | next = tmp->next; |
2fd6b7f5 NP |
1015 | |
1016 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1017 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1018 | if (d_mountpoint(dentry)) { |
1019 | spin_unlock(&dentry->d_lock); | |
1020 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1021 | goto positive; |
2fd6b7f5 | 1022 | } |
1da177e4 | 1023 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1024 | spin_unlock(&this_parent->d_lock); |
1025 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1026 | this_parent = dentry; |
2fd6b7f5 | 1027 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1028 | goto repeat; |
1029 | } | |
2fd6b7f5 | 1030 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1031 | } |
1032 | /* | |
1033 | * All done at this level ... ascend and resume the search. | |
1034 | */ | |
1035 | if (this_parent != parent) { | |
c826cb7d LT |
1036 | struct dentry *child = this_parent; |
1037 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1038 | if (!this_parent) | |
949854d0 | 1039 | goto rename_retry; |
949854d0 | 1040 | next = child->d_u.d_child.next; |
1da177e4 LT |
1041 | goto resume; |
1042 | } | |
2fd6b7f5 | 1043 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1044 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1045 | goto rename_retry; |
58db63d0 NP |
1046 | if (locked) |
1047 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1048 | return 0; /* No mount points found in tree */ |
1049 | positive: | |
58db63d0 | 1050 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1051 | goto rename_retry; |
58db63d0 NP |
1052 | if (locked) |
1053 | write_sequnlock(&rename_lock); | |
1da177e4 | 1054 | return 1; |
58db63d0 NP |
1055 | |
1056 | rename_retry: | |
1057 | locked = 1; | |
1058 | write_seqlock(&rename_lock); | |
1059 | goto again; | |
1da177e4 | 1060 | } |
ec4f8605 | 1061 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1062 | |
1063 | /* | |
1064 | * Search the dentry child list for the specified parent, | |
1065 | * and move any unused dentries to the end of the unused | |
1066 | * list for prune_dcache(). We descend to the next level | |
1067 | * whenever the d_subdirs list is non-empty and continue | |
1068 | * searching. | |
1069 | * | |
1070 | * It returns zero iff there are no unused children, | |
1071 | * otherwise it returns the number of children moved to | |
1072 | * the end of the unused list. This may not be the total | |
1073 | * number of unused children, because select_parent can | |
1074 | * drop the lock and return early due to latency | |
1075 | * constraints. | |
1076 | */ | |
1077 | static int select_parent(struct dentry * parent) | |
1078 | { | |
949854d0 | 1079 | struct dentry *this_parent; |
1da177e4 | 1080 | struct list_head *next; |
949854d0 | 1081 | unsigned seq; |
1da177e4 | 1082 | int found = 0; |
58db63d0 | 1083 | int locked = 0; |
1da177e4 | 1084 | |
949854d0 | 1085 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1086 | again: |
1087 | this_parent = parent; | |
2fd6b7f5 | 1088 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1089 | repeat: |
1090 | next = this_parent->d_subdirs.next; | |
1091 | resume: | |
1092 | while (next != &this_parent->d_subdirs) { | |
1093 | struct list_head *tmp = next; | |
5160ee6f | 1094 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1095 | next = tmp->next; |
1096 | ||
2fd6b7f5 | 1097 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1098 | |
1da177e4 LT |
1099 | /* |
1100 | * move only zero ref count dentries to the end | |
1101 | * of the unused list for prune_dcache | |
1102 | */ | |
b7ab39f6 | 1103 | if (!dentry->d_count) { |
a4633357 | 1104 | dentry_lru_move_tail(dentry); |
1da177e4 | 1105 | found++; |
a4633357 CH |
1106 | } else { |
1107 | dentry_lru_del(dentry); | |
1da177e4 LT |
1108 | } |
1109 | ||
1110 | /* | |
1111 | * We can return to the caller if we have found some (this | |
1112 | * ensures forward progress). We'll be coming back to find | |
1113 | * the rest. | |
1114 | */ | |
2fd6b7f5 NP |
1115 | if (found && need_resched()) { |
1116 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1117 | goto out; |
2fd6b7f5 | 1118 | } |
1da177e4 LT |
1119 | |
1120 | /* | |
1121 | * Descend a level if the d_subdirs list is non-empty. | |
1122 | */ | |
1123 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1124 | spin_unlock(&this_parent->d_lock); |
1125 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1126 | this_parent = dentry; |
2fd6b7f5 | 1127 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1128 | goto repeat; |
1129 | } | |
2fd6b7f5 NP |
1130 | |
1131 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1132 | } |
1133 | /* | |
1134 | * All done at this level ... ascend and resume the search. | |
1135 | */ | |
1136 | if (this_parent != parent) { | |
c826cb7d LT |
1137 | struct dentry *child = this_parent; |
1138 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1139 | if (!this_parent) | |
949854d0 | 1140 | goto rename_retry; |
949854d0 | 1141 | next = child->d_u.d_child.next; |
1da177e4 LT |
1142 | goto resume; |
1143 | } | |
1144 | out: | |
2fd6b7f5 | 1145 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1146 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1147 | goto rename_retry; |
58db63d0 NP |
1148 | if (locked) |
1149 | write_sequnlock(&rename_lock); | |
1da177e4 | 1150 | return found; |
58db63d0 NP |
1151 | |
1152 | rename_retry: | |
1153 | if (found) | |
1154 | return found; | |
1155 | locked = 1; | |
1156 | write_seqlock(&rename_lock); | |
1157 | goto again; | |
1da177e4 LT |
1158 | } |
1159 | ||
1160 | /** | |
1161 | * shrink_dcache_parent - prune dcache | |
1162 | * @parent: parent of entries to prune | |
1163 | * | |
1164 | * Prune the dcache to remove unused children of the parent dentry. | |
1165 | */ | |
1166 | ||
1167 | void shrink_dcache_parent(struct dentry * parent) | |
1168 | { | |
da3bbdd4 | 1169 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1170 | int found; |
1171 | ||
1172 | while ((found = select_parent(parent)) != 0) | |
b0d40c92 | 1173 | __shrink_dcache_sb(sb, found, 0); |
1da177e4 | 1174 | } |
ec4f8605 | 1175 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1176 | |
1da177e4 | 1177 | /** |
a4464dbc AV |
1178 | * __d_alloc - allocate a dcache entry |
1179 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1180 | * @name: qstr of the name |
1181 | * | |
1182 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1183 | * available. On a success the dentry is returned. The name passed in is | |
1184 | * copied and the copy passed in may be reused after this call. | |
1185 | */ | |
1186 | ||
a4464dbc | 1187 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1188 | { |
1189 | struct dentry *dentry; | |
1190 | char *dname; | |
1191 | ||
e12ba74d | 1192 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1193 | if (!dentry) |
1194 | return NULL; | |
1195 | ||
1196 | if (name->len > DNAME_INLINE_LEN-1) { | |
1197 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1198 | if (!dname) { | |
1199 | kmem_cache_free(dentry_cache, dentry); | |
1200 | return NULL; | |
1201 | } | |
1202 | } else { | |
1203 | dname = dentry->d_iname; | |
1204 | } | |
1205 | dentry->d_name.name = dname; | |
1206 | ||
1207 | dentry->d_name.len = name->len; | |
1208 | dentry->d_name.hash = name->hash; | |
1209 | memcpy(dname, name->name, name->len); | |
1210 | dname[name->len] = 0; | |
1211 | ||
b7ab39f6 | 1212 | dentry->d_count = 1; |
dea3667b | 1213 | dentry->d_flags = 0; |
1da177e4 | 1214 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1215 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1216 | dentry->d_inode = NULL; |
a4464dbc AV |
1217 | dentry->d_parent = dentry; |
1218 | dentry->d_sb = sb; | |
1da177e4 LT |
1219 | dentry->d_op = NULL; |
1220 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1221 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1222 | INIT_LIST_HEAD(&dentry->d_lru); |
1223 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1224 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1225 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1226 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1227 | |
3e880fb5 | 1228 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1229 | |
1da177e4 LT |
1230 | return dentry; |
1231 | } | |
a4464dbc AV |
1232 | |
1233 | /** | |
1234 | * d_alloc - allocate a dcache entry | |
1235 | * @parent: parent of entry to allocate | |
1236 | * @name: qstr of the name | |
1237 | * | |
1238 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1239 | * available. On a success the dentry is returned. The name passed in is | |
1240 | * copied and the copy passed in may be reused after this call. | |
1241 | */ | |
1242 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1243 | { | |
1244 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1245 | if (!dentry) | |
1246 | return NULL; | |
1247 | ||
1248 | spin_lock(&parent->d_lock); | |
1249 | /* | |
1250 | * don't need child lock because it is not subject | |
1251 | * to concurrency here | |
1252 | */ | |
1253 | __dget_dlock(parent); | |
1254 | dentry->d_parent = parent; | |
1255 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1256 | spin_unlock(&parent->d_lock); | |
1257 | ||
1258 | return dentry; | |
1259 | } | |
ec4f8605 | 1260 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1261 | |
4b936885 NP |
1262 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1263 | { | |
a4464dbc AV |
1264 | struct dentry *dentry = __d_alloc(sb, name); |
1265 | if (dentry) | |
4b936885 | 1266 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1267 | return dentry; |
1268 | } | |
1269 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1270 | ||
1da177e4 LT |
1271 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1272 | { | |
1273 | struct qstr q; | |
1274 | ||
1275 | q.name = name; | |
1276 | q.len = strlen(name); | |
1277 | q.hash = full_name_hash(q.name, q.len); | |
1278 | return d_alloc(parent, &q); | |
1279 | } | |
ef26ca97 | 1280 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1281 | |
fb045adb NP |
1282 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1283 | { | |
6f7f7caa LT |
1284 | WARN_ON_ONCE(dentry->d_op); |
1285 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1286 | DCACHE_OP_COMPARE | |
1287 | DCACHE_OP_REVALIDATE | | |
1288 | DCACHE_OP_DELETE )); | |
1289 | dentry->d_op = op; | |
1290 | if (!op) | |
1291 | return; | |
1292 | if (op->d_hash) | |
1293 | dentry->d_flags |= DCACHE_OP_HASH; | |
1294 | if (op->d_compare) | |
1295 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1296 | if (op->d_revalidate) | |
1297 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1298 | if (op->d_delete) | |
1299 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1300 | ||
1301 | } | |
1302 | EXPORT_SYMBOL(d_set_d_op); | |
1303 | ||
360da900 OH |
1304 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1305 | { | |
b23fb0a6 | 1306 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1307 | if (inode) { |
1308 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1309 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
360da900 | 1310 | list_add(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1311 | } |
360da900 | 1312 | dentry->d_inode = inode; |
31e6b01f | 1313 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1314 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1315 | fsnotify_d_instantiate(dentry, inode); |
1316 | } | |
1317 | ||
1da177e4 LT |
1318 | /** |
1319 | * d_instantiate - fill in inode information for a dentry | |
1320 | * @entry: dentry to complete | |
1321 | * @inode: inode to attach to this dentry | |
1322 | * | |
1323 | * Fill in inode information in the entry. | |
1324 | * | |
1325 | * This turns negative dentries into productive full members | |
1326 | * of society. | |
1327 | * | |
1328 | * NOTE! This assumes that the inode count has been incremented | |
1329 | * (or otherwise set) by the caller to indicate that it is now | |
1330 | * in use by the dcache. | |
1331 | */ | |
1332 | ||
1333 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1334 | { | |
28133c7b | 1335 | BUG_ON(!list_empty(&entry->d_alias)); |
873feea0 NP |
1336 | if (inode) |
1337 | spin_lock(&inode->i_lock); | |
360da900 | 1338 | __d_instantiate(entry, inode); |
873feea0 NP |
1339 | if (inode) |
1340 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1341 | security_d_instantiate(entry, inode); |
1342 | } | |
ec4f8605 | 1343 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1344 | |
1345 | /** | |
1346 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1347 | * @entry: dentry to instantiate | |
1348 | * @inode: inode to attach to this dentry | |
1349 | * | |
1350 | * Fill in inode information in the entry. On success, it returns NULL. | |
1351 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1352 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1353 | * |
1354 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1355 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1356 | * |
1357 | * This also assumes that the inode count has been incremented | |
1358 | * (or otherwise set) by the caller to indicate that it is now | |
1359 | * in use by the dcache. | |
1da177e4 | 1360 | */ |
770bfad8 DH |
1361 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1362 | struct inode *inode) | |
1da177e4 LT |
1363 | { |
1364 | struct dentry *alias; | |
1365 | int len = entry->d_name.len; | |
1366 | const char *name = entry->d_name.name; | |
1367 | unsigned int hash = entry->d_name.hash; | |
1368 | ||
770bfad8 | 1369 | if (!inode) { |
360da900 | 1370 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1371 | return NULL; |
1372 | } | |
1373 | ||
1da177e4 LT |
1374 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1375 | struct qstr *qstr = &alias->d_name; | |
1376 | ||
9abca360 NP |
1377 | /* |
1378 | * Don't need alias->d_lock here, because aliases with | |
1379 | * d_parent == entry->d_parent are not subject to name or | |
1380 | * parent changes, because the parent inode i_mutex is held. | |
1381 | */ | |
1da177e4 LT |
1382 | if (qstr->hash != hash) |
1383 | continue; | |
1384 | if (alias->d_parent != entry->d_parent) | |
1385 | continue; | |
9d55c369 | 1386 | if (dentry_cmp(qstr->name, qstr->len, name, len)) |
1da177e4 | 1387 | continue; |
dc0474be | 1388 | __dget(alias); |
1da177e4 LT |
1389 | return alias; |
1390 | } | |
770bfad8 | 1391 | |
360da900 | 1392 | __d_instantiate(entry, inode); |
1da177e4 LT |
1393 | return NULL; |
1394 | } | |
770bfad8 DH |
1395 | |
1396 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1397 | { | |
1398 | struct dentry *result; | |
1399 | ||
1400 | BUG_ON(!list_empty(&entry->d_alias)); | |
1401 | ||
873feea0 NP |
1402 | if (inode) |
1403 | spin_lock(&inode->i_lock); | |
770bfad8 | 1404 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1405 | if (inode) |
1406 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1407 | |
1408 | if (!result) { | |
1409 | security_d_instantiate(entry, inode); | |
1410 | return NULL; | |
1411 | } | |
1412 | ||
1413 | BUG_ON(!d_unhashed(result)); | |
1414 | iput(inode); | |
1415 | return result; | |
1416 | } | |
1417 | ||
1da177e4 LT |
1418 | EXPORT_SYMBOL(d_instantiate_unique); |
1419 | ||
1420 | /** | |
1421 | * d_alloc_root - allocate root dentry | |
1422 | * @root_inode: inode to allocate the root for | |
1423 | * | |
1424 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1425 | * instantiated and returned. %NULL is returned if there is insufficient | |
1426 | * memory or the inode passed is %NULL. | |
1427 | */ | |
1428 | ||
1429 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1430 | { | |
1431 | struct dentry *res = NULL; | |
1432 | ||
1433 | if (root_inode) { | |
1434 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1435 | ||
a4464dbc AV |
1436 | res = __d_alloc(root_inode->i_sb, &name); |
1437 | if (res) | |
1da177e4 | 1438 | d_instantiate(res, root_inode); |
1da177e4 LT |
1439 | } |
1440 | return res; | |
1441 | } | |
ec4f8605 | 1442 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 | 1443 | |
d891eedb BF |
1444 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1445 | { | |
1446 | struct dentry *alias; | |
1447 | ||
1448 | if (list_empty(&inode->i_dentry)) | |
1449 | return NULL; | |
1450 | alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias); | |
1451 | __dget(alias); | |
1452 | return alias; | |
1453 | } | |
1454 | ||
1455 | static struct dentry * d_find_any_alias(struct inode *inode) | |
1456 | { | |
1457 | struct dentry *de; | |
1458 | ||
1459 | spin_lock(&inode->i_lock); | |
1460 | de = __d_find_any_alias(inode); | |
1461 | spin_unlock(&inode->i_lock); | |
1462 | return de; | |
1463 | } | |
1464 | ||
1465 | ||
4ea3ada2 CH |
1466 | /** |
1467 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1468 | * @inode: inode to allocate the dentry for | |
1469 | * | |
1470 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1471 | * similar open by handle operations. The returned dentry may be anonymous, | |
1472 | * or may have a full name (if the inode was already in the cache). | |
1473 | * | |
1474 | * When called on a directory inode, we must ensure that the inode only ever | |
1475 | * has one dentry. If a dentry is found, that is returned instead of | |
1476 | * allocating a new one. | |
1477 | * | |
1478 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1479 | * to the dentry. In case of an error the reference on the inode is released. |
1480 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1481 | * be passed in and will be the error will be propagate to the return value, | |
1482 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1483 | */ |
1484 | struct dentry *d_obtain_alias(struct inode *inode) | |
1485 | { | |
9308a612 CH |
1486 | static const struct qstr anonstring = { .name = "" }; |
1487 | struct dentry *tmp; | |
1488 | struct dentry *res; | |
4ea3ada2 CH |
1489 | |
1490 | if (!inode) | |
44003728 | 1491 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1492 | if (IS_ERR(inode)) |
1493 | return ERR_CAST(inode); | |
1494 | ||
d891eedb | 1495 | res = d_find_any_alias(inode); |
9308a612 CH |
1496 | if (res) |
1497 | goto out_iput; | |
1498 | ||
a4464dbc | 1499 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1500 | if (!tmp) { |
1501 | res = ERR_PTR(-ENOMEM); | |
1502 | goto out_iput; | |
4ea3ada2 | 1503 | } |
b5c84bf6 | 1504 | |
873feea0 | 1505 | spin_lock(&inode->i_lock); |
d891eedb | 1506 | res = __d_find_any_alias(inode); |
9308a612 | 1507 | if (res) { |
873feea0 | 1508 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1509 | dput(tmp); |
1510 | goto out_iput; | |
1511 | } | |
1512 | ||
1513 | /* attach a disconnected dentry */ | |
1514 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1515 | tmp->d_inode = inode; |
1516 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
9308a612 | 1517 | list_add(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1518 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1519 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1520 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1521 | spin_unlock(&tmp->d_lock); |
873feea0 | 1522 | spin_unlock(&inode->i_lock); |
24ff6663 | 1523 | security_d_instantiate(tmp, inode); |
9308a612 | 1524 | |
9308a612 CH |
1525 | return tmp; |
1526 | ||
1527 | out_iput: | |
24ff6663 JB |
1528 | if (res && !IS_ERR(res)) |
1529 | security_d_instantiate(res, inode); | |
9308a612 CH |
1530 | iput(inode); |
1531 | return res; | |
4ea3ada2 | 1532 | } |
adc48720 | 1533 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1534 | |
1535 | /** | |
1536 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1537 | * @inode: the inode which may have a disconnected dentry | |
1538 | * @dentry: a negative dentry which we want to point to the inode. | |
1539 | * | |
1540 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1541 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1542 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1543 | * | |
1544 | * This is needed in the lookup routine of any filesystem that is exportable | |
1545 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1546 | * | |
1547 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1548 | * is returned. This matches the expected return value of ->lookup. | |
1549 | * | |
1550 | */ | |
1551 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1552 | { | |
1553 | struct dentry *new = NULL; | |
1554 | ||
a9049376 AV |
1555 | if (IS_ERR(inode)) |
1556 | return ERR_CAST(inode); | |
1557 | ||
21c0d8fd | 1558 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1559 | spin_lock(&inode->i_lock); |
1da177e4 LT |
1560 | new = __d_find_alias(inode, 1); |
1561 | if (new) { | |
1562 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
873feea0 | 1563 | spin_unlock(&inode->i_lock); |
1da177e4 | 1564 | security_d_instantiate(new, inode); |
1da177e4 LT |
1565 | d_move(new, dentry); |
1566 | iput(inode); | |
1567 | } else { | |
873feea0 | 1568 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1569 | __d_instantiate(dentry, inode); |
873feea0 | 1570 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1571 | security_d_instantiate(dentry, inode); |
1572 | d_rehash(dentry); | |
1573 | } | |
1574 | } else | |
1575 | d_add(dentry, inode); | |
1576 | return new; | |
1577 | } | |
ec4f8605 | 1578 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1579 | |
9403540c BN |
1580 | /** |
1581 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1582 | * @inode: the inode case-insensitive lookup has found | |
1583 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1584 | * @name: the case-exact name to be associated with the returned dentry | |
1585 | * | |
1586 | * This is to avoid filling the dcache with case-insensitive names to the | |
1587 | * same inode, only the actual correct case is stored in the dcache for | |
1588 | * case-insensitive filesystems. | |
1589 | * | |
1590 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1591 | * already exists in in the dcache, use it and return it. | |
1592 | * | |
1593 | * If no entry exists with the exact case name, allocate new dentry with | |
1594 | * the exact case, and return the spliced entry. | |
1595 | */ | |
e45b590b | 1596 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1597 | struct qstr *name) |
1598 | { | |
1599 | int error; | |
1600 | struct dentry *found; | |
1601 | struct dentry *new; | |
1602 | ||
b6520c81 CH |
1603 | /* |
1604 | * First check if a dentry matching the name already exists, | |
1605 | * if not go ahead and create it now. | |
1606 | */ | |
9403540c | 1607 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1608 | if (!found) { |
1609 | new = d_alloc(dentry->d_parent, name); | |
1610 | if (!new) { | |
1611 | error = -ENOMEM; | |
1612 | goto err_out; | |
1613 | } | |
b6520c81 | 1614 | |
9403540c BN |
1615 | found = d_splice_alias(inode, new); |
1616 | if (found) { | |
1617 | dput(new); | |
1618 | return found; | |
1619 | } | |
1620 | return new; | |
1621 | } | |
b6520c81 CH |
1622 | |
1623 | /* | |
1624 | * If a matching dentry exists, and it's not negative use it. | |
1625 | * | |
1626 | * Decrement the reference count to balance the iget() done | |
1627 | * earlier on. | |
1628 | */ | |
9403540c BN |
1629 | if (found->d_inode) { |
1630 | if (unlikely(found->d_inode != inode)) { | |
1631 | /* This can't happen because bad inodes are unhashed. */ | |
1632 | BUG_ON(!is_bad_inode(inode)); | |
1633 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1634 | } | |
9403540c BN |
1635 | iput(inode); |
1636 | return found; | |
1637 | } | |
b6520c81 | 1638 | |
9403540c | 1639 | /* |
44396f4b JB |
1640 | * We are going to instantiate this dentry, unhash it and clear the |
1641 | * lookup flag so we can do that. | |
9403540c | 1642 | */ |
44396f4b JB |
1643 | if (unlikely(d_need_lookup(found))) |
1644 | d_clear_need_lookup(found); | |
b6520c81 | 1645 | |
9403540c | 1646 | /* |
9403540c | 1647 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1648 | * already has a dentry. |
9403540c | 1649 | */ |
4513d899 AV |
1650 | new = d_splice_alias(inode, found); |
1651 | if (new) { | |
1652 | dput(found); | |
1653 | found = new; | |
9403540c | 1654 | } |
4513d899 | 1655 | return found; |
9403540c BN |
1656 | |
1657 | err_out: | |
1658 | iput(inode); | |
1659 | return ERR_PTR(error); | |
1660 | } | |
ec4f8605 | 1661 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1662 | |
31e6b01f NP |
1663 | /** |
1664 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1665 | * @parent: parent dentry | |
1666 | * @name: qstr of name we wish to find | |
1667 | * @seq: returns d_seq value at the point where the dentry was found | |
1668 | * @inode: returns dentry->d_inode when the inode was found valid. | |
1669 | * Returns: dentry, or NULL | |
1670 | * | |
1671 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1672 | * resolution (store-free path walking) design described in | |
1673 | * Documentation/filesystems/path-lookup.txt. | |
1674 | * | |
1675 | * This is not to be used outside core vfs. | |
1676 | * | |
1677 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1678 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1679 | * without taking d_lock and checking d_seq sequence count against @seq | |
1680 | * returned here. | |
1681 | * | |
1682 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1683 | * function. | |
1684 | * | |
1685 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1686 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1687 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1688 | * is formed, giving integrity down the path walk. | |
1689 | */ | |
1690 | struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name, | |
1691 | unsigned *seq, struct inode **inode) | |
1692 | { | |
1693 | unsigned int len = name->len; | |
1694 | unsigned int hash = name->hash; | |
1695 | const unsigned char *str = name->name; | |
b07ad996 | 1696 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1697 | struct hlist_bl_node *node; |
31e6b01f NP |
1698 | struct dentry *dentry; |
1699 | ||
1700 | /* | |
1701 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1702 | * required to prevent single threaded performance regressions | |
1703 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1704 | * Keep the two functions in sync. | |
1705 | */ | |
1706 | ||
1707 | /* | |
1708 | * The hash list is protected using RCU. | |
1709 | * | |
1710 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1711 | * races with d_move(). | |
1712 | * | |
1713 | * It is possible that concurrent renames can mess up our list | |
1714 | * walk here and result in missing our dentry, resulting in the | |
1715 | * false-negative result. d_lookup() protects against concurrent | |
1716 | * renames using rename_lock seqlock. | |
1717 | * | |
b0a4bb83 | 1718 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1719 | */ |
b07ad996 | 1720 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1721 | struct inode *i; |
1722 | const char *tname; | |
1723 | int tlen; | |
1724 | ||
1725 | if (dentry->d_name.hash != hash) | |
1726 | continue; | |
1727 | ||
1728 | seqretry: | |
1729 | *seq = read_seqcount_begin(&dentry->d_seq); | |
1730 | if (dentry->d_parent != parent) | |
1731 | continue; | |
1732 | if (d_unhashed(dentry)) | |
1733 | continue; | |
1734 | tlen = dentry->d_name.len; | |
1735 | tname = dentry->d_name.name; | |
1736 | i = dentry->d_inode; | |
e1bb5782 | 1737 | prefetch(tname); |
31e6b01f NP |
1738 | /* |
1739 | * This seqcount check is required to ensure name and | |
1740 | * len are loaded atomically, so as not to walk off the | |
1741 | * edge of memory when walking. If we could load this | |
1742 | * atomically some other way, we could drop this check. | |
1743 | */ | |
1744 | if (read_seqcount_retry(&dentry->d_seq, *seq)) | |
1745 | goto seqretry; | |
fb045adb | 1746 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
31e6b01f NP |
1747 | if (parent->d_op->d_compare(parent, *inode, |
1748 | dentry, i, | |
1749 | tlen, tname, name)) | |
1750 | continue; | |
1751 | } else { | |
9d55c369 | 1752 | if (dentry_cmp(tname, tlen, str, len)) |
31e6b01f NP |
1753 | continue; |
1754 | } | |
1755 | /* | |
1756 | * No extra seqcount check is required after the name | |
1757 | * compare. The caller must perform a seqcount check in | |
1758 | * order to do anything useful with the returned dentry | |
1759 | * anyway. | |
1760 | */ | |
1761 | *inode = i; | |
1762 | return dentry; | |
1763 | } | |
1764 | return NULL; | |
1765 | } | |
1766 | ||
1da177e4 LT |
1767 | /** |
1768 | * d_lookup - search for a dentry | |
1769 | * @parent: parent dentry | |
1770 | * @name: qstr of name we wish to find | |
b04f784e | 1771 | * Returns: dentry, or NULL |
1da177e4 | 1772 | * |
b04f784e NP |
1773 | * d_lookup searches the children of the parent dentry for the name in |
1774 | * question. If the dentry is found its reference count is incremented and the | |
1775 | * dentry is returned. The caller must use dput to free the entry when it has | |
1776 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1777 | */ |
31e6b01f | 1778 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 | 1779 | { |
31e6b01f | 1780 | struct dentry *dentry; |
949854d0 | 1781 | unsigned seq; |
1da177e4 LT |
1782 | |
1783 | do { | |
1784 | seq = read_seqbegin(&rename_lock); | |
1785 | dentry = __d_lookup(parent, name); | |
1786 | if (dentry) | |
1787 | break; | |
1788 | } while (read_seqretry(&rename_lock, seq)); | |
1789 | return dentry; | |
1790 | } | |
ec4f8605 | 1791 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1792 | |
31e6b01f | 1793 | /** |
b04f784e NP |
1794 | * __d_lookup - search for a dentry (racy) |
1795 | * @parent: parent dentry | |
1796 | * @name: qstr of name we wish to find | |
1797 | * Returns: dentry, or NULL | |
1798 | * | |
1799 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1800 | * false-negative result due to unrelated rename activity. | |
1801 | * | |
1802 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1803 | * however it must be used carefully, eg. with a following d_lookup in | |
1804 | * the case of failure. | |
1805 | * | |
1806 | * __d_lookup callers must be commented. | |
1807 | */ | |
31e6b01f | 1808 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 LT |
1809 | { |
1810 | unsigned int len = name->len; | |
1811 | unsigned int hash = name->hash; | |
1812 | const unsigned char *str = name->name; | |
b07ad996 | 1813 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1814 | struct hlist_bl_node *node; |
31e6b01f | 1815 | struct dentry *found = NULL; |
665a7583 | 1816 | struct dentry *dentry; |
1da177e4 | 1817 | |
31e6b01f NP |
1818 | /* |
1819 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1820 | * required to prevent single threaded performance regressions | |
1821 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1822 | * Keep the two functions in sync. | |
1823 | */ | |
1824 | ||
b04f784e NP |
1825 | /* |
1826 | * The hash list is protected using RCU. | |
1827 | * | |
1828 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1829 | * with d_move(). | |
1830 | * | |
1831 | * It is possible that concurrent renames can mess up our list | |
1832 | * walk here and result in missing our dentry, resulting in the | |
1833 | * false-negative result. d_lookup() protects against concurrent | |
1834 | * renames using rename_lock seqlock. | |
1835 | * | |
b0a4bb83 | 1836 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1837 | */ |
1da177e4 LT |
1838 | rcu_read_lock(); |
1839 | ||
b07ad996 | 1840 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1841 | const char *tname; |
1842 | int tlen; | |
1da177e4 | 1843 | |
1da177e4 LT |
1844 | if (dentry->d_name.hash != hash) |
1845 | continue; | |
1da177e4 LT |
1846 | |
1847 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1848 | if (dentry->d_parent != parent) |
1849 | goto next; | |
d0185c08 LT |
1850 | if (d_unhashed(dentry)) |
1851 | goto next; | |
1852 | ||
1da177e4 LT |
1853 | /* |
1854 | * It is safe to compare names since d_move() cannot | |
1855 | * change the qstr (protected by d_lock). | |
1856 | */ | |
31e6b01f NP |
1857 | tlen = dentry->d_name.len; |
1858 | tname = dentry->d_name.name; | |
fb045adb | 1859 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
621e155a NP |
1860 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1861 | dentry, dentry->d_inode, | |
31e6b01f | 1862 | tlen, tname, name)) |
1da177e4 LT |
1863 | goto next; |
1864 | } else { | |
9d55c369 | 1865 | if (dentry_cmp(tname, tlen, str, len)) |
1da177e4 LT |
1866 | goto next; |
1867 | } | |
1868 | ||
b7ab39f6 | 1869 | dentry->d_count++; |
d0185c08 | 1870 | found = dentry; |
1da177e4 LT |
1871 | spin_unlock(&dentry->d_lock); |
1872 | break; | |
1873 | next: | |
1874 | spin_unlock(&dentry->d_lock); | |
1875 | } | |
1876 | rcu_read_unlock(); | |
1877 | ||
1878 | return found; | |
1879 | } | |
1880 | ||
3e7e241f EB |
1881 | /** |
1882 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1883 | * @dir: Directory to search in | |
1884 | * @name: qstr of name we wish to find | |
1885 | * | |
1886 | * On hash failure or on lookup failure NULL is returned. | |
1887 | */ | |
1888 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1889 | { | |
1890 | struct dentry *dentry = NULL; | |
1891 | ||
1892 | /* | |
1893 | * Check for a fs-specific hash function. Note that we must | |
1894 | * calculate the standard hash first, as the d_op->d_hash() | |
1895 | * routine may choose to leave the hash value unchanged. | |
1896 | */ | |
1897 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1898 | if (dir->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 1899 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1900 | goto out; |
1901 | } | |
1902 | dentry = d_lookup(dir, name); | |
1903 | out: | |
1904 | return dentry; | |
1905 | } | |
1906 | ||
1da177e4 | 1907 | /** |
786a5e15 | 1908 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 1909 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 1910 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
1911 | * |
1912 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1913 | * This is used by ncpfs in its readdir implementation. | |
1914 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1915 | * |
1916 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1917 | */ |
d3a23e16 | 1918 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1919 | { |
786a5e15 | 1920 | struct dentry *child; |
d3a23e16 | 1921 | |
2fd6b7f5 | 1922 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
1923 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
1924 | if (dentry == child) { | |
2fd6b7f5 | 1925 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 1926 | __dget_dlock(dentry); |
2fd6b7f5 NP |
1927 | spin_unlock(&dentry->d_lock); |
1928 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
1929 | return 1; |
1930 | } | |
1931 | } | |
2fd6b7f5 | 1932 | spin_unlock(&dparent->d_lock); |
786a5e15 | 1933 | |
1da177e4 LT |
1934 | return 0; |
1935 | } | |
ec4f8605 | 1936 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1937 | |
1938 | /* | |
1939 | * When a file is deleted, we have two options: | |
1940 | * - turn this dentry into a negative dentry | |
1941 | * - unhash this dentry and free it. | |
1942 | * | |
1943 | * Usually, we want to just turn this into | |
1944 | * a negative dentry, but if anybody else is | |
1945 | * currently using the dentry or the inode | |
1946 | * we can't do that and we fall back on removing | |
1947 | * it from the hash queues and waiting for | |
1948 | * it to be deleted later when it has no users | |
1949 | */ | |
1950 | ||
1951 | /** | |
1952 | * d_delete - delete a dentry | |
1953 | * @dentry: The dentry to delete | |
1954 | * | |
1955 | * Turn the dentry into a negative dentry if possible, otherwise | |
1956 | * remove it from the hash queues so it can be deleted later | |
1957 | */ | |
1958 | ||
1959 | void d_delete(struct dentry * dentry) | |
1960 | { | |
873feea0 | 1961 | struct inode *inode; |
7a91bf7f | 1962 | int isdir = 0; |
1da177e4 LT |
1963 | /* |
1964 | * Are we the only user? | |
1965 | */ | |
357f8e65 | 1966 | again: |
1da177e4 | 1967 | spin_lock(&dentry->d_lock); |
873feea0 NP |
1968 | inode = dentry->d_inode; |
1969 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 1970 | if (dentry->d_count == 1) { |
873feea0 | 1971 | if (inode && !spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
1972 | spin_unlock(&dentry->d_lock); |
1973 | cpu_relax(); | |
1974 | goto again; | |
1975 | } | |
13e3c5e5 | 1976 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 1977 | dentry_unlink_inode(dentry); |
7a91bf7f | 1978 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1979 | return; |
1980 | } | |
1981 | ||
1982 | if (!d_unhashed(dentry)) | |
1983 | __d_drop(dentry); | |
1984 | ||
1985 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
1986 | |
1987 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1988 | } |
ec4f8605 | 1989 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 1990 | |
b07ad996 | 1991 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 1992 | { |
ceb5bdc2 | 1993 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 1994 | hlist_bl_lock(b); |
dea3667b | 1995 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 1996 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 1997 | hlist_bl_unlock(b); |
1da177e4 LT |
1998 | } |
1999 | ||
770bfad8 DH |
2000 | static void _d_rehash(struct dentry * entry) |
2001 | { | |
2002 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2003 | } | |
2004 | ||
1da177e4 LT |
2005 | /** |
2006 | * d_rehash - add an entry back to the hash | |
2007 | * @entry: dentry to add to the hash | |
2008 | * | |
2009 | * Adds a dentry to the hash according to its name. | |
2010 | */ | |
2011 | ||
2012 | void d_rehash(struct dentry * entry) | |
2013 | { | |
1da177e4 | 2014 | spin_lock(&entry->d_lock); |
770bfad8 | 2015 | _d_rehash(entry); |
1da177e4 | 2016 | spin_unlock(&entry->d_lock); |
1da177e4 | 2017 | } |
ec4f8605 | 2018 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2019 | |
fb2d5b86 NP |
2020 | /** |
2021 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2022 | * @dentry: dentry to be updated | |
2023 | * @name: new name | |
2024 | * | |
2025 | * Update a case insensitive dentry with new case of name. | |
2026 | * | |
2027 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2028 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2029 | * lengths). | |
2030 | * | |
2031 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2032 | * keep renames and concurrent inserts, and readdir(2) away). | |
2033 | */ | |
2034 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2035 | { | |
7ebfa57f | 2036 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2037 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2038 | ||
fb2d5b86 | 2039 | spin_lock(&dentry->d_lock); |
31e6b01f | 2040 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2041 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2042 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2043 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2044 | } |
2045 | EXPORT_SYMBOL(dentry_update_name_case); | |
2046 | ||
1da177e4 LT |
2047 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2048 | { | |
2049 | if (dname_external(target)) { | |
2050 | if (dname_external(dentry)) { | |
2051 | /* | |
2052 | * Both external: swap the pointers | |
2053 | */ | |
9a8d5bb4 | 2054 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2055 | } else { |
2056 | /* | |
2057 | * dentry:internal, target:external. Steal target's | |
2058 | * storage and make target internal. | |
2059 | */ | |
321bcf92 BF |
2060 | memcpy(target->d_iname, dentry->d_name.name, |
2061 | dentry->d_name.len + 1); | |
1da177e4 LT |
2062 | dentry->d_name.name = target->d_name.name; |
2063 | target->d_name.name = target->d_iname; | |
2064 | } | |
2065 | } else { | |
2066 | if (dname_external(dentry)) { | |
2067 | /* | |
2068 | * dentry:external, target:internal. Give dentry's | |
2069 | * storage to target and make dentry internal | |
2070 | */ | |
2071 | memcpy(dentry->d_iname, target->d_name.name, | |
2072 | target->d_name.len + 1); | |
2073 | target->d_name.name = dentry->d_name.name; | |
2074 | dentry->d_name.name = dentry->d_iname; | |
2075 | } else { | |
2076 | /* | |
2077 | * Both are internal. Just copy target to dentry | |
2078 | */ | |
2079 | memcpy(dentry->d_iname, target->d_name.name, | |
2080 | target->d_name.len + 1); | |
dc711ca3 AV |
2081 | dentry->d_name.len = target->d_name.len; |
2082 | return; | |
1da177e4 LT |
2083 | } |
2084 | } | |
9a8d5bb4 | 2085 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2086 | } |
2087 | ||
2fd6b7f5 NP |
2088 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2089 | { | |
2090 | /* | |
2091 | * XXXX: do we really need to take target->d_lock? | |
2092 | */ | |
2093 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2094 | spin_lock(&target->d_parent->d_lock); | |
2095 | else { | |
2096 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2097 | spin_lock(&dentry->d_parent->d_lock); | |
2098 | spin_lock_nested(&target->d_parent->d_lock, | |
2099 | DENTRY_D_LOCK_NESTED); | |
2100 | } else { | |
2101 | spin_lock(&target->d_parent->d_lock); | |
2102 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2103 | DENTRY_D_LOCK_NESTED); | |
2104 | } | |
2105 | } | |
2106 | if (target < dentry) { | |
2107 | spin_lock_nested(&target->d_lock, 2); | |
2108 | spin_lock_nested(&dentry->d_lock, 3); | |
2109 | } else { | |
2110 | spin_lock_nested(&dentry->d_lock, 2); | |
2111 | spin_lock_nested(&target->d_lock, 3); | |
2112 | } | |
2113 | } | |
2114 | ||
2115 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2116 | struct dentry *target) | |
2117 | { | |
2118 | if (target->d_parent != dentry->d_parent) | |
2119 | spin_unlock(&dentry->d_parent->d_lock); | |
2120 | if (target->d_parent != target) | |
2121 | spin_unlock(&target->d_parent->d_lock); | |
2122 | } | |
2123 | ||
1da177e4 | 2124 | /* |
2fd6b7f5 NP |
2125 | * When switching names, the actual string doesn't strictly have to |
2126 | * be preserved in the target - because we're dropping the target | |
2127 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2128 | * the new name before we switch. | |
2129 | * | |
2130 | * Note that we have to be a lot more careful about getting the hash | |
2131 | * switched - we have to switch the hash value properly even if it | |
2132 | * then no longer matches the actual (corrupted) string of the target. | |
2133 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2134 | */ |
9eaef27b | 2135 | /* |
18367501 | 2136 | * __d_move - move a dentry |
1da177e4 LT |
2137 | * @dentry: entry to move |
2138 | * @target: new dentry | |
2139 | * | |
2140 | * Update the dcache to reflect the move of a file name. Negative | |
18367501 AV |
2141 | * dcache entries should not be moved in this way. Caller hold |
2142 | * rename_lock. | |
1da177e4 | 2143 | */ |
18367501 | 2144 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2145 | { |
1da177e4 LT |
2146 | if (!dentry->d_inode) |
2147 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2148 | ||
2fd6b7f5 NP |
2149 | BUG_ON(d_ancestor(dentry, target)); |
2150 | BUG_ON(d_ancestor(target, dentry)); | |
2151 | ||
2fd6b7f5 | 2152 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2153 | |
31e6b01f NP |
2154 | write_seqcount_begin(&dentry->d_seq); |
2155 | write_seqcount_begin(&target->d_seq); | |
2156 | ||
ceb5bdc2 NP |
2157 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2158 | ||
2159 | /* | |
2160 | * Move the dentry to the target hash queue. Don't bother checking | |
2161 | * for the same hash queue because of how unlikely it is. | |
2162 | */ | |
2163 | __d_drop(dentry); | |
789680d1 | 2164 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2165 | |
2166 | /* Unhash the target: dput() will then get rid of it */ | |
2167 | __d_drop(target); | |
2168 | ||
5160ee6f ED |
2169 | list_del(&dentry->d_u.d_child); |
2170 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2171 | |
2172 | /* Switch the names.. */ | |
2173 | switch_names(dentry, target); | |
9a8d5bb4 | 2174 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2175 | |
2176 | /* ... and switch the parents */ | |
2177 | if (IS_ROOT(dentry)) { | |
2178 | dentry->d_parent = target->d_parent; | |
2179 | target->d_parent = target; | |
5160ee6f | 2180 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2181 | } else { |
9a8d5bb4 | 2182 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2183 | |
2184 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2185 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2186 | } |
2187 | ||
5160ee6f | 2188 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2189 | |
31e6b01f NP |
2190 | write_seqcount_end(&target->d_seq); |
2191 | write_seqcount_end(&dentry->d_seq); | |
2192 | ||
2fd6b7f5 | 2193 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2194 | spin_unlock(&target->d_lock); |
c32ccd87 | 2195 | fsnotify_d_move(dentry); |
1da177e4 | 2196 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2197 | } |
2198 | ||
2199 | /* | |
2200 | * d_move - move a dentry | |
2201 | * @dentry: entry to move | |
2202 | * @target: new dentry | |
2203 | * | |
2204 | * Update the dcache to reflect the move of a file name. Negative | |
2205 | * dcache entries should not be moved in this way. | |
2206 | */ | |
2207 | void d_move(struct dentry *dentry, struct dentry *target) | |
2208 | { | |
2209 | write_seqlock(&rename_lock); | |
2210 | __d_move(dentry, target); | |
1da177e4 | 2211 | write_sequnlock(&rename_lock); |
9eaef27b | 2212 | } |
ec4f8605 | 2213 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2214 | |
e2761a11 OH |
2215 | /** |
2216 | * d_ancestor - search for an ancestor | |
2217 | * @p1: ancestor dentry | |
2218 | * @p2: child dentry | |
2219 | * | |
2220 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2221 | * an ancestor of p2, else NULL. | |
9eaef27b | 2222 | */ |
e2761a11 | 2223 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2224 | { |
2225 | struct dentry *p; | |
2226 | ||
871c0067 | 2227 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2228 | if (p->d_parent == p1) |
e2761a11 | 2229 | return p; |
9eaef27b | 2230 | } |
e2761a11 | 2231 | return NULL; |
9eaef27b TM |
2232 | } |
2233 | ||
2234 | /* | |
2235 | * This helper attempts to cope with remotely renamed directories | |
2236 | * | |
2237 | * It assumes that the caller is already holding | |
18367501 | 2238 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2239 | * |
2240 | * Note: If ever the locking in lock_rename() changes, then please | |
2241 | * remember to update this too... | |
9eaef27b | 2242 | */ |
873feea0 NP |
2243 | static struct dentry *__d_unalias(struct inode *inode, |
2244 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2245 | { |
2246 | struct mutex *m1 = NULL, *m2 = NULL; | |
2247 | struct dentry *ret; | |
2248 | ||
2249 | /* If alias and dentry share a parent, then no extra locks required */ | |
2250 | if (alias->d_parent == dentry->d_parent) | |
2251 | goto out_unalias; | |
2252 | ||
9eaef27b TM |
2253 | /* See lock_rename() */ |
2254 | ret = ERR_PTR(-EBUSY); | |
2255 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2256 | goto out_err; | |
2257 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2258 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2259 | goto out_err; | |
2260 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2261 | out_unalias: | |
18367501 | 2262 | __d_move(alias, dentry); |
9eaef27b TM |
2263 | ret = alias; |
2264 | out_err: | |
873feea0 | 2265 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2266 | if (m2) |
2267 | mutex_unlock(m2); | |
2268 | if (m1) | |
2269 | mutex_unlock(m1); | |
2270 | return ret; | |
2271 | } | |
2272 | ||
770bfad8 DH |
2273 | /* |
2274 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2275 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2276 | * returns with anon->d_lock held! |
770bfad8 DH |
2277 | */ |
2278 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2279 | { | |
2280 | struct dentry *dparent, *aparent; | |
2281 | ||
2fd6b7f5 | 2282 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2283 | |
31e6b01f NP |
2284 | write_seqcount_begin(&dentry->d_seq); |
2285 | write_seqcount_begin(&anon->d_seq); | |
2286 | ||
770bfad8 DH |
2287 | dparent = dentry->d_parent; |
2288 | aparent = anon->d_parent; | |
2289 | ||
2fd6b7f5 NP |
2290 | switch_names(dentry, anon); |
2291 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2292 | ||
770bfad8 DH |
2293 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2294 | list_del(&dentry->d_u.d_child); | |
2295 | if (!IS_ROOT(dentry)) | |
2296 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2297 | else | |
2298 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2299 | ||
2300 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2301 | list_del(&anon->d_u.d_child); | |
2302 | if (!IS_ROOT(anon)) | |
2303 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2304 | else | |
2305 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2306 | ||
31e6b01f NP |
2307 | write_seqcount_end(&dentry->d_seq); |
2308 | write_seqcount_end(&anon->d_seq); | |
2309 | ||
2fd6b7f5 NP |
2310 | dentry_unlock_parents_for_move(anon, dentry); |
2311 | spin_unlock(&dentry->d_lock); | |
2312 | ||
2313 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2314 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2315 | } | |
2316 | ||
2317 | /** | |
2318 | * d_materialise_unique - introduce an inode into the tree | |
2319 | * @dentry: candidate dentry | |
2320 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2321 | * | |
2322 | * Introduces an dentry into the tree, substituting an extant disconnected | |
2323 | * root directory alias in its place if there is one | |
2324 | */ | |
2325 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2326 | { | |
9eaef27b | 2327 | struct dentry *actual; |
770bfad8 DH |
2328 | |
2329 | BUG_ON(!d_unhashed(dentry)); | |
2330 | ||
770bfad8 DH |
2331 | if (!inode) { |
2332 | actual = dentry; | |
360da900 | 2333 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2334 | d_rehash(actual); |
2335 | goto out_nolock; | |
770bfad8 DH |
2336 | } |
2337 | ||
873feea0 | 2338 | spin_lock(&inode->i_lock); |
357f8e65 | 2339 | |
9eaef27b TM |
2340 | if (S_ISDIR(inode->i_mode)) { |
2341 | struct dentry *alias; | |
2342 | ||
2343 | /* Does an aliased dentry already exist? */ | |
2344 | alias = __d_find_alias(inode, 0); | |
2345 | if (alias) { | |
2346 | actual = alias; | |
18367501 AV |
2347 | write_seqlock(&rename_lock); |
2348 | ||
2349 | if (d_ancestor(alias, dentry)) { | |
2350 | /* Check for loops */ | |
2351 | actual = ERR_PTR(-ELOOP); | |
2352 | } else if (IS_ROOT(alias)) { | |
2353 | /* Is this an anonymous mountpoint that we | |
2354 | * could splice into our tree? */ | |
9eaef27b | 2355 | __d_materialise_dentry(dentry, alias); |
18367501 | 2356 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2357 | __d_drop(alias); |
2358 | goto found; | |
18367501 AV |
2359 | } else { |
2360 | /* Nope, but we must(!) avoid directory | |
2361 | * aliasing */ | |
2362 | actual = __d_unalias(inode, dentry, alias); | |
9eaef27b | 2363 | } |
18367501 | 2364 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2365 | if (IS_ERR(actual)) |
2366 | dput(alias); | |
2367 | goto out_nolock; | |
2368 | } | |
770bfad8 DH |
2369 | } |
2370 | ||
2371 | /* Add a unique reference */ | |
2372 | actual = __d_instantiate_unique(dentry, inode); | |
2373 | if (!actual) | |
2374 | actual = dentry; | |
357f8e65 NP |
2375 | else |
2376 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2377 | |
770bfad8 DH |
2378 | spin_lock(&actual->d_lock); |
2379 | found: | |
2380 | _d_rehash(actual); | |
2381 | spin_unlock(&actual->d_lock); | |
873feea0 | 2382 | spin_unlock(&inode->i_lock); |
9eaef27b | 2383 | out_nolock: |
770bfad8 DH |
2384 | if (actual == dentry) { |
2385 | security_d_instantiate(dentry, inode); | |
2386 | return NULL; | |
2387 | } | |
2388 | ||
2389 | iput(inode); | |
2390 | return actual; | |
770bfad8 | 2391 | } |
ec4f8605 | 2392 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2393 | |
cdd16d02 | 2394 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2395 | { |
2396 | *buflen -= namelen; | |
2397 | if (*buflen < 0) | |
2398 | return -ENAMETOOLONG; | |
2399 | *buffer -= namelen; | |
2400 | memcpy(*buffer, str, namelen); | |
2401 | return 0; | |
2402 | } | |
2403 | ||
cdd16d02 MS |
2404 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2405 | { | |
2406 | return prepend(buffer, buflen, name->name, name->len); | |
2407 | } | |
2408 | ||
1da177e4 | 2409 | /** |
208898c1 | 2410 | * prepend_path - Prepend path string to a buffer |
9d1bc601 MS |
2411 | * @path: the dentry/vfsmount to report |
2412 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2413 | * @buffer: pointer to the end of the buffer |
2414 | * @buflen: pointer to buffer length | |
552ce544 | 2415 | * |
949854d0 | 2416 | * Caller holds the rename_lock. |
9d1bc601 MS |
2417 | * |
2418 | * If path is not reachable from the supplied root, then the value of | |
2419 | * root is changed (without modifying refcounts). | |
1da177e4 | 2420 | */ |
f2eb6575 MS |
2421 | static int prepend_path(const struct path *path, struct path *root, |
2422 | char **buffer, int *buflen) | |
1da177e4 | 2423 | { |
9d1bc601 MS |
2424 | struct dentry *dentry = path->dentry; |
2425 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2426 | bool slash = false; |
2427 | int error = 0; | |
6092d048 | 2428 | |
99b7db7b | 2429 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2430 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2431 | struct dentry * parent; |
2432 | ||
1da177e4 | 2433 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2434 | /* Global root? */ |
1da177e4 | 2435 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2436 | goto global_root; |
2437 | } | |
2438 | dentry = vfsmnt->mnt_mountpoint; | |
2439 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2440 | continue; |
2441 | } | |
2442 | parent = dentry->d_parent; | |
2443 | prefetch(parent); | |
9abca360 | 2444 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2445 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2446 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2447 | if (!error) |
2448 | error = prepend(buffer, buflen, "/", 1); | |
2449 | if (error) | |
2450 | break; | |
2451 | ||
2452 | slash = true; | |
1da177e4 LT |
2453 | dentry = parent; |
2454 | } | |
2455 | ||
be285c71 | 2456 | out: |
f2eb6575 MS |
2457 | if (!error && !slash) |
2458 | error = prepend(buffer, buflen, "/", 1); | |
2459 | ||
99b7db7b | 2460 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2461 | return error; |
1da177e4 LT |
2462 | |
2463 | global_root: | |
98dc568b MS |
2464 | /* |
2465 | * Filesystems needing to implement special "root names" | |
2466 | * should do so with ->d_dname() | |
2467 | */ | |
2468 | if (IS_ROOT(dentry) && | |
2469 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2470 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2471 | (int) dentry->d_name.len, dentry->d_name.name); | |
2472 | } | |
9d1bc601 MS |
2473 | root->mnt = vfsmnt; |
2474 | root->dentry = dentry; | |
be285c71 | 2475 | goto out; |
f2eb6575 | 2476 | } |
be285c71 | 2477 | |
f2eb6575 MS |
2478 | /** |
2479 | * __d_path - return the path of a dentry | |
2480 | * @path: the dentry/vfsmount to report | |
2481 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2482 | * @buf: buffer to return value in |
f2eb6575 MS |
2483 | * @buflen: buffer length |
2484 | * | |
ffd1f4ed | 2485 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2486 | * |
2487 | * Returns a pointer into the buffer or an error code if the | |
2488 | * path was too long. | |
2489 | * | |
be148247 | 2490 | * "buflen" should be positive. |
f2eb6575 MS |
2491 | * |
2492 | * If path is not reachable from the supplied root, then the value of | |
2493 | * root is changed (without modifying refcounts). | |
2494 | */ | |
2495 | char *__d_path(const struct path *path, struct path *root, | |
2496 | char *buf, int buflen) | |
2497 | { | |
2498 | char *res = buf + buflen; | |
2499 | int error; | |
2500 | ||
2501 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2502 | write_seqlock(&rename_lock); |
f2eb6575 | 2503 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2504 | write_sequnlock(&rename_lock); |
be148247 | 2505 | |
f2eb6575 MS |
2506 | if (error) |
2507 | return ERR_PTR(error); | |
f2eb6575 | 2508 | return res; |
1da177e4 LT |
2509 | } |
2510 | ||
ffd1f4ed MS |
2511 | /* |
2512 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2513 | */ | |
2514 | static int path_with_deleted(const struct path *path, struct path *root, | |
2515 | char **buf, int *buflen) | |
2516 | { | |
2517 | prepend(buf, buflen, "\0", 1); | |
2518 | if (d_unlinked(path->dentry)) { | |
2519 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2520 | if (error) | |
2521 | return error; | |
2522 | } | |
2523 | ||
2524 | return prepend_path(path, root, buf, buflen); | |
2525 | } | |
2526 | ||
8df9d1a4 MS |
2527 | static int prepend_unreachable(char **buffer, int *buflen) |
2528 | { | |
2529 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2530 | } | |
2531 | ||
a03a8a70 JB |
2532 | /** |
2533 | * d_path - return the path of a dentry | |
cf28b486 | 2534 | * @path: path to report |
a03a8a70 JB |
2535 | * @buf: buffer to return value in |
2536 | * @buflen: buffer length | |
2537 | * | |
2538 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2539 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2540 | * | |
52afeefb AV |
2541 | * Returns a pointer into the buffer or an error code if the path was |
2542 | * too long. Note: Callers should use the returned pointer, not the passed | |
2543 | * in buffer, to use the name! The implementation often starts at an offset | |
2544 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2545 | * |
31f3e0b3 | 2546 | * "buflen" should be positive. |
a03a8a70 | 2547 | */ |
20d4fdc1 | 2548 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2549 | { |
ffd1f4ed | 2550 | char *res = buf + buflen; |
6ac08c39 | 2551 | struct path root; |
9d1bc601 | 2552 | struct path tmp; |
ffd1f4ed | 2553 | int error; |
1da177e4 | 2554 | |
c23fbb6b ED |
2555 | /* |
2556 | * We have various synthetic filesystems that never get mounted. On | |
2557 | * these filesystems dentries are never used for lookup purposes, and | |
2558 | * thus don't need to be hashed. They also don't need a name until a | |
2559 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2560 | * below allows us to generate a name for these objects on demand: | |
2561 | */ | |
cf28b486 JB |
2562 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2563 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2564 | |
f7ad3c6b | 2565 | get_fs_root(current->fs, &root); |
949854d0 | 2566 | write_seqlock(&rename_lock); |
9d1bc601 | 2567 | tmp = root; |
ffd1f4ed MS |
2568 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2569 | if (error) | |
2570 | res = ERR_PTR(error); | |
949854d0 | 2571 | write_sequnlock(&rename_lock); |
6ac08c39 | 2572 | path_put(&root); |
1da177e4 LT |
2573 | return res; |
2574 | } | |
ec4f8605 | 2575 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2576 | |
8df9d1a4 MS |
2577 | /** |
2578 | * d_path_with_unreachable - return the path of a dentry | |
2579 | * @path: path to report | |
2580 | * @buf: buffer to return value in | |
2581 | * @buflen: buffer length | |
2582 | * | |
2583 | * The difference from d_path() is that this prepends "(unreachable)" | |
2584 | * to paths which are unreachable from the current process' root. | |
2585 | */ | |
2586 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2587 | { | |
2588 | char *res = buf + buflen; | |
2589 | struct path root; | |
2590 | struct path tmp; | |
2591 | int error; | |
2592 | ||
2593 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2594 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2595 | ||
2596 | get_fs_root(current->fs, &root); | |
949854d0 | 2597 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2598 | tmp = root; |
2599 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2600 | if (!error && !path_equal(&tmp, &root)) | |
2601 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2602 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2603 | path_put(&root); |
2604 | if (error) | |
2605 | res = ERR_PTR(error); | |
2606 | ||
2607 | return res; | |
2608 | } | |
2609 | ||
c23fbb6b ED |
2610 | /* |
2611 | * Helper function for dentry_operations.d_dname() members | |
2612 | */ | |
2613 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2614 | const char *fmt, ...) | |
2615 | { | |
2616 | va_list args; | |
2617 | char temp[64]; | |
2618 | int sz; | |
2619 | ||
2620 | va_start(args, fmt); | |
2621 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2622 | va_end(args); | |
2623 | ||
2624 | if (sz > sizeof(temp) || sz > buflen) | |
2625 | return ERR_PTR(-ENAMETOOLONG); | |
2626 | ||
2627 | buffer += buflen - sz; | |
2628 | return memcpy(buffer, temp, sz); | |
2629 | } | |
2630 | ||
6092d048 RP |
2631 | /* |
2632 | * Write full pathname from the root of the filesystem into the buffer. | |
2633 | */ | |
ec2447c2 | 2634 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2635 | { |
2636 | char *end = buf + buflen; | |
2637 | char *retval; | |
2638 | ||
6092d048 | 2639 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2640 | if (buflen < 1) |
2641 | goto Elong; | |
2642 | /* Get '/' right */ | |
2643 | retval = end-1; | |
2644 | *retval = '/'; | |
2645 | ||
cdd16d02 MS |
2646 | while (!IS_ROOT(dentry)) { |
2647 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2648 | int error; |
6092d048 | 2649 | |
6092d048 | 2650 | prefetch(parent); |
9abca360 NP |
2651 | spin_lock(&dentry->d_lock); |
2652 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2653 | spin_unlock(&dentry->d_lock); | |
2654 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2655 | goto Elong; |
2656 | ||
2657 | retval = end; | |
2658 | dentry = parent; | |
2659 | } | |
c103135c AV |
2660 | return retval; |
2661 | Elong: | |
2662 | return ERR_PTR(-ENAMETOOLONG); | |
2663 | } | |
ec2447c2 NP |
2664 | |
2665 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2666 | { | |
2667 | char *retval; | |
2668 | ||
949854d0 | 2669 | write_seqlock(&rename_lock); |
ec2447c2 | 2670 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2671 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2672 | |
2673 | return retval; | |
2674 | } | |
2675 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2676 | |
2677 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2678 | { | |
2679 | char *p = NULL; | |
2680 | char *retval; | |
2681 | ||
949854d0 | 2682 | write_seqlock(&rename_lock); |
c103135c AV |
2683 | if (d_unlinked(dentry)) { |
2684 | p = buf + buflen; | |
2685 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2686 | goto Elong; | |
2687 | buflen++; | |
2688 | } | |
2689 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2690 | write_sequnlock(&rename_lock); |
c103135c AV |
2691 | if (!IS_ERR(retval) && p) |
2692 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2693 | return retval; |
2694 | Elong: | |
6092d048 RP |
2695 | return ERR_PTR(-ENAMETOOLONG); |
2696 | } | |
2697 | ||
1da177e4 LT |
2698 | /* |
2699 | * NOTE! The user-level library version returns a | |
2700 | * character pointer. The kernel system call just | |
2701 | * returns the length of the buffer filled (which | |
2702 | * includes the ending '\0' character), or a negative | |
2703 | * error value. So libc would do something like | |
2704 | * | |
2705 | * char *getcwd(char * buf, size_t size) | |
2706 | * { | |
2707 | * int retval; | |
2708 | * | |
2709 | * retval = sys_getcwd(buf, size); | |
2710 | * if (retval >= 0) | |
2711 | * return buf; | |
2712 | * errno = -retval; | |
2713 | * return NULL; | |
2714 | * } | |
2715 | */ | |
3cdad428 | 2716 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2717 | { |
552ce544 | 2718 | int error; |
6ac08c39 | 2719 | struct path pwd, root; |
552ce544 | 2720 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2721 | |
2722 | if (!page) | |
2723 | return -ENOMEM; | |
2724 | ||
f7ad3c6b | 2725 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2726 | |
552ce544 | 2727 | error = -ENOENT; |
949854d0 | 2728 | write_seqlock(&rename_lock); |
f3da392e | 2729 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2730 | unsigned long len; |
9d1bc601 | 2731 | struct path tmp = root; |
8df9d1a4 MS |
2732 | char *cwd = page + PAGE_SIZE; |
2733 | int buflen = PAGE_SIZE; | |
1da177e4 | 2734 | |
8df9d1a4 MS |
2735 | prepend(&cwd, &buflen, "\0", 1); |
2736 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2737 | write_sequnlock(&rename_lock); |
552ce544 | 2738 | |
8df9d1a4 | 2739 | if (error) |
552ce544 LT |
2740 | goto out; |
2741 | ||
8df9d1a4 MS |
2742 | /* Unreachable from current root */ |
2743 | if (!path_equal(&tmp, &root)) { | |
2744 | error = prepend_unreachable(&cwd, &buflen); | |
2745 | if (error) | |
2746 | goto out; | |
2747 | } | |
2748 | ||
552ce544 LT |
2749 | error = -ERANGE; |
2750 | len = PAGE_SIZE + page - cwd; | |
2751 | if (len <= size) { | |
2752 | error = len; | |
2753 | if (copy_to_user(buf, cwd, len)) | |
2754 | error = -EFAULT; | |
2755 | } | |
949854d0 NP |
2756 | } else { |
2757 | write_sequnlock(&rename_lock); | |
949854d0 | 2758 | } |
1da177e4 LT |
2759 | |
2760 | out: | |
6ac08c39 JB |
2761 | path_put(&pwd); |
2762 | path_put(&root); | |
1da177e4 LT |
2763 | free_page((unsigned long) page); |
2764 | return error; | |
2765 | } | |
2766 | ||
2767 | /* | |
2768 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2769 | * | |
2770 | * Trivially implemented using the dcache structure | |
2771 | */ | |
2772 | ||
2773 | /** | |
2774 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2775 | * @new_dentry: new dentry | |
2776 | * @old_dentry: old dentry | |
2777 | * | |
2778 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2779 | * Returns 0 otherwise. | |
2780 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2781 | */ | |
2782 | ||
e2761a11 | 2783 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2784 | { |
2785 | int result; | |
949854d0 | 2786 | unsigned seq; |
1da177e4 | 2787 | |
e2761a11 OH |
2788 | if (new_dentry == old_dentry) |
2789 | return 1; | |
2790 | ||
e2761a11 | 2791 | do { |
1da177e4 | 2792 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2793 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2794 | /* |
2795 | * Need rcu_readlock to protect against the d_parent trashing | |
2796 | * due to d_move | |
2797 | */ | |
2798 | rcu_read_lock(); | |
e2761a11 | 2799 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2800 | result = 1; |
e2761a11 OH |
2801 | else |
2802 | result = 0; | |
949854d0 | 2803 | rcu_read_unlock(); |
1da177e4 | 2804 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2805 | |
2806 | return result; | |
2807 | } | |
2808 | ||
2096f759 AV |
2809 | int path_is_under(struct path *path1, struct path *path2) |
2810 | { | |
2811 | struct vfsmount *mnt = path1->mnt; | |
2812 | struct dentry *dentry = path1->dentry; | |
2813 | int res; | |
99b7db7b NP |
2814 | |
2815 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2816 | if (mnt != path2->mnt) { |
2817 | for (;;) { | |
2818 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2819 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2820 | return 0; |
2821 | } | |
2822 | if (mnt->mnt_parent == path2->mnt) | |
2823 | break; | |
2824 | mnt = mnt->mnt_parent; | |
2825 | } | |
2826 | dentry = mnt->mnt_mountpoint; | |
2827 | } | |
2828 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2829 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2830 | return res; |
2831 | } | |
2832 | EXPORT_SYMBOL(path_is_under); | |
2833 | ||
1da177e4 LT |
2834 | void d_genocide(struct dentry *root) |
2835 | { | |
949854d0 | 2836 | struct dentry *this_parent; |
1da177e4 | 2837 | struct list_head *next; |
949854d0 | 2838 | unsigned seq; |
58db63d0 | 2839 | int locked = 0; |
1da177e4 | 2840 | |
949854d0 | 2841 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2842 | again: |
2843 | this_parent = root; | |
2fd6b7f5 | 2844 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2845 | repeat: |
2846 | next = this_parent->d_subdirs.next; | |
2847 | resume: | |
2848 | while (next != &this_parent->d_subdirs) { | |
2849 | struct list_head *tmp = next; | |
5160ee6f | 2850 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2851 | next = tmp->next; |
949854d0 | 2852 | |
da502956 NP |
2853 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2854 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2855 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2856 | continue; |
da502956 | 2857 | } |
1da177e4 | 2858 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2859 | spin_unlock(&this_parent->d_lock); |
2860 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2861 | this_parent = dentry; |
2fd6b7f5 | 2862 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2863 | goto repeat; |
2864 | } | |
949854d0 NP |
2865 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2866 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2867 | dentry->d_count--; | |
2868 | } | |
b7ab39f6 | 2869 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2870 | } |
2871 | if (this_parent != root) { | |
c826cb7d | 2872 | struct dentry *child = this_parent; |
949854d0 NP |
2873 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2874 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2875 | this_parent->d_count--; | |
2876 | } | |
c826cb7d LT |
2877 | this_parent = try_to_ascend(this_parent, locked, seq); |
2878 | if (!this_parent) | |
949854d0 | 2879 | goto rename_retry; |
949854d0 | 2880 | next = child->d_u.d_child.next; |
1da177e4 LT |
2881 | goto resume; |
2882 | } | |
2fd6b7f5 | 2883 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2884 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2885 | goto rename_retry; |
58db63d0 NP |
2886 | if (locked) |
2887 | write_sequnlock(&rename_lock); | |
2888 | return; | |
2889 | ||
2890 | rename_retry: | |
2891 | locked = 1; | |
2892 | write_seqlock(&rename_lock); | |
2893 | goto again; | |
1da177e4 LT |
2894 | } |
2895 | ||
2896 | /** | |
2897 | * find_inode_number - check for dentry with name | |
2898 | * @dir: directory to check | |
2899 | * @name: Name to find. | |
2900 | * | |
2901 | * Check whether a dentry already exists for the given name, | |
2902 | * and return the inode number if it has an inode. Otherwise | |
2903 | * 0 is returned. | |
2904 | * | |
2905 | * This routine is used to post-process directory listings for | |
2906 | * filesystems using synthetic inode numbers, and is necessary | |
2907 | * to keep getcwd() working. | |
2908 | */ | |
2909 | ||
2910 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2911 | { | |
2912 | struct dentry * dentry; | |
2913 | ino_t ino = 0; | |
2914 | ||
3e7e241f EB |
2915 | dentry = d_hash_and_lookup(dir, name); |
2916 | if (dentry) { | |
1da177e4 LT |
2917 | if (dentry->d_inode) |
2918 | ino = dentry->d_inode->i_ino; | |
2919 | dput(dentry); | |
2920 | } | |
1da177e4 LT |
2921 | return ino; |
2922 | } | |
ec4f8605 | 2923 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2924 | |
2925 | static __initdata unsigned long dhash_entries; | |
2926 | static int __init set_dhash_entries(char *str) | |
2927 | { | |
2928 | if (!str) | |
2929 | return 0; | |
2930 | dhash_entries = simple_strtoul(str, &str, 0); | |
2931 | return 1; | |
2932 | } | |
2933 | __setup("dhash_entries=", set_dhash_entries); | |
2934 | ||
2935 | static void __init dcache_init_early(void) | |
2936 | { | |
2937 | int loop; | |
2938 | ||
2939 | /* If hashes are distributed across NUMA nodes, defer | |
2940 | * hash allocation until vmalloc space is available. | |
2941 | */ | |
2942 | if (hashdist) | |
2943 | return; | |
2944 | ||
2945 | dentry_hashtable = | |
2946 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 2947 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
2948 | dhash_entries, |
2949 | 13, | |
2950 | HASH_EARLY, | |
2951 | &d_hash_shift, | |
2952 | &d_hash_mask, | |
2953 | 0); | |
2954 | ||
2955 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 2956 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
2957 | } |
2958 | ||
74bf17cf | 2959 | static void __init dcache_init(void) |
1da177e4 LT |
2960 | { |
2961 | int loop; | |
2962 | ||
2963 | /* | |
2964 | * A constructor could be added for stable state like the lists, | |
2965 | * but it is probably not worth it because of the cache nature | |
2966 | * of the dcache. | |
2967 | */ | |
0a31bd5f CL |
2968 | dentry_cache = KMEM_CACHE(dentry, |
2969 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
2970 | |
2971 | /* Hash may have been set up in dcache_init_early */ | |
2972 | if (!hashdist) | |
2973 | return; | |
2974 | ||
2975 | dentry_hashtable = | |
2976 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 2977 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
2978 | dhash_entries, |
2979 | 13, | |
2980 | 0, | |
2981 | &d_hash_shift, | |
2982 | &d_hash_mask, | |
2983 | 0); | |
2984 | ||
2985 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 2986 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
2987 | } |
2988 | ||
2989 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2990 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2991 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2992 | |
1da177e4 LT |
2993 | EXPORT_SYMBOL(d_genocide); |
2994 | ||
1da177e4 LT |
2995 | void __init vfs_caches_init_early(void) |
2996 | { | |
2997 | dcache_init_early(); | |
2998 | inode_init_early(); | |
2999 | } | |
3000 | ||
3001 | void __init vfs_caches_init(unsigned long mempages) | |
3002 | { | |
3003 | unsigned long reserve; | |
3004 | ||
3005 | /* Base hash sizes on available memory, with a reserve equal to | |
3006 | 150% of current kernel size */ | |
3007 | ||
3008 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3009 | mempages -= reserve; | |
3010 | ||
3011 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3012 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3013 | |
74bf17cf DC |
3014 | dcache_init(); |
3015 | inode_init(); | |
1da177e4 | 3016 | files_init(mempages); |
74bf17cf | 3017 | mnt_init(); |
1da177e4 LT |
3018 | bdev_cache_init(); |
3019 | chrdev_init(); | |
3020 | } |