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 | ||
c6627c60 DH |
304 | /* |
305 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
306 | * dentry->d_lock is locked. The caller must take care of that, if | |
307 | * appropriate. | |
308 | */ | |
309 | static void __d_shrink(struct dentry *dentry) | |
310 | { | |
311 | if (!d_unhashed(dentry)) { | |
312 | struct hlist_bl_head *b; | |
313 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
314 | b = &dentry->d_sb->s_anon; | |
315 | else | |
316 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
317 | ||
318 | hlist_bl_lock(b); | |
319 | __hlist_bl_del(&dentry->d_hash); | |
320 | dentry->d_hash.pprev = NULL; | |
321 | hlist_bl_unlock(b); | |
322 | } | |
323 | } | |
324 | ||
789680d1 NP |
325 | /** |
326 | * d_drop - drop a dentry | |
327 | * @dentry: dentry to drop | |
328 | * | |
329 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
330 | * be found through a VFS lookup any more. Note that this is different from | |
331 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
332 | * possible, giving a successful _negative_ lookup, while d_drop will | |
333 | * just make the cache lookup fail. | |
334 | * | |
335 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
336 | * reason (NFS timeouts or autofs deletes). | |
337 | * | |
338 | * __d_drop requires dentry->d_lock. | |
339 | */ | |
340 | void __d_drop(struct dentry *dentry) | |
341 | { | |
dea3667b | 342 | if (!d_unhashed(dentry)) { |
c6627c60 | 343 | __d_shrink(dentry); |
dea3667b | 344 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
345 | } |
346 | } | |
347 | EXPORT_SYMBOL(__d_drop); | |
348 | ||
349 | void d_drop(struct dentry *dentry) | |
350 | { | |
789680d1 NP |
351 | spin_lock(&dentry->d_lock); |
352 | __d_drop(dentry); | |
353 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
354 | } |
355 | EXPORT_SYMBOL(d_drop); | |
356 | ||
44396f4b JB |
357 | /* |
358 | * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag | |
359 | * @dentry: dentry to drop | |
360 | * | |
361 | * This is called when we do a lookup on a placeholder dentry that needed to be | |
362 | * looked up. The dentry should have been hashed in order for it to be found by | |
363 | * the lookup code, but now needs to be unhashed while we do the actual lookup | |
364 | * and clear the DCACHE_NEED_LOOKUP flag. | |
365 | */ | |
366 | void d_clear_need_lookup(struct dentry *dentry) | |
367 | { | |
368 | spin_lock(&dentry->d_lock); | |
369 | __d_drop(dentry); | |
370 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
371 | spin_unlock(&dentry->d_lock); | |
372 | } | |
373 | EXPORT_SYMBOL(d_clear_need_lookup); | |
374 | ||
77812a1e NP |
375 | /* |
376 | * Finish off a dentry we've decided to kill. | |
377 | * dentry->d_lock must be held, returns with it unlocked. | |
378 | * If ref is non-zero, then decrement the refcount too. | |
379 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
380 | */ | |
381 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
382 | __releases(dentry->d_lock) | |
383 | { | |
873feea0 | 384 | struct inode *inode; |
77812a1e NP |
385 | struct dentry *parent; |
386 | ||
873feea0 NP |
387 | inode = dentry->d_inode; |
388 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
389 | relock: |
390 | spin_unlock(&dentry->d_lock); | |
391 | cpu_relax(); | |
392 | return dentry; /* try again with same dentry */ | |
393 | } | |
394 | if (IS_ROOT(dentry)) | |
395 | parent = NULL; | |
396 | else | |
397 | parent = dentry->d_parent; | |
398 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
399 | if (inode) |
400 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
401 | goto relock; |
402 | } | |
31e6b01f | 403 | |
77812a1e NP |
404 | if (ref) |
405 | dentry->d_count--; | |
406 | /* if dentry was on the d_lru list delete it from there */ | |
407 | dentry_lru_del(dentry); | |
408 | /* if it was on the hash then remove it */ | |
409 | __d_drop(dentry); | |
410 | return d_kill(dentry, parent); | |
411 | } | |
412 | ||
1da177e4 LT |
413 | /* |
414 | * This is dput | |
415 | * | |
416 | * This is complicated by the fact that we do not want to put | |
417 | * dentries that are no longer on any hash chain on the unused | |
418 | * list: we'd much rather just get rid of them immediately. | |
419 | * | |
420 | * However, that implies that we have to traverse the dentry | |
421 | * tree upwards to the parents which might _also_ now be | |
422 | * scheduled for deletion (it may have been only waiting for | |
423 | * its last child to go away). | |
424 | * | |
425 | * This tail recursion is done by hand as we don't want to depend | |
426 | * on the compiler to always get this right (gcc generally doesn't). | |
427 | * Real recursion would eat up our stack space. | |
428 | */ | |
429 | ||
430 | /* | |
431 | * dput - release a dentry | |
432 | * @dentry: dentry to release | |
433 | * | |
434 | * Release a dentry. This will drop the usage count and if appropriate | |
435 | * call the dentry unlink method as well as removing it from the queues and | |
436 | * releasing its resources. If the parent dentries were scheduled for release | |
437 | * they too may now get deleted. | |
1da177e4 | 438 | */ |
1da177e4 LT |
439 | void dput(struct dentry *dentry) |
440 | { | |
441 | if (!dentry) | |
442 | return; | |
443 | ||
444 | repeat: | |
b7ab39f6 | 445 | if (dentry->d_count == 1) |
1da177e4 | 446 | might_sleep(); |
1da177e4 | 447 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
448 | BUG_ON(!dentry->d_count); |
449 | if (dentry->d_count > 1) { | |
450 | dentry->d_count--; | |
1da177e4 | 451 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
452 | return; |
453 | } | |
454 | ||
fb045adb | 455 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 456 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 457 | goto kill_it; |
1da177e4 | 458 | } |
265ac902 | 459 | |
1da177e4 LT |
460 | /* Unreachable? Get rid of it */ |
461 | if (d_unhashed(dentry)) | |
462 | goto kill_it; | |
265ac902 | 463 | |
44396f4b JB |
464 | /* |
465 | * If this dentry needs lookup, don't set the referenced flag so that it | |
466 | * is more likely to be cleaned up by the dcache shrinker in case of | |
467 | * memory pressure. | |
468 | */ | |
469 | if (!d_need_lookup(dentry)) | |
470 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 471 | dentry_lru_add(dentry); |
265ac902 | 472 | |
61f3dee4 NP |
473 | dentry->d_count--; |
474 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
475 | return; |
476 | ||
d52b9086 | 477 | kill_it: |
77812a1e | 478 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
479 | if (dentry) |
480 | goto repeat; | |
1da177e4 | 481 | } |
ec4f8605 | 482 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
483 | |
484 | /** | |
485 | * d_invalidate - invalidate a dentry | |
486 | * @dentry: dentry to invalidate | |
487 | * | |
488 | * Try to invalidate the dentry if it turns out to be | |
489 | * possible. If there are other dentries that can be | |
490 | * reached through this one we can't delete it and we | |
491 | * return -EBUSY. On success we return 0. | |
492 | * | |
493 | * no dcache lock. | |
494 | */ | |
495 | ||
496 | int d_invalidate(struct dentry * dentry) | |
497 | { | |
498 | /* | |
499 | * If it's already been dropped, return OK. | |
500 | */ | |
da502956 | 501 | spin_lock(&dentry->d_lock); |
1da177e4 | 502 | if (d_unhashed(dentry)) { |
da502956 | 503 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
504 | return 0; |
505 | } | |
506 | /* | |
507 | * Check whether to do a partial shrink_dcache | |
508 | * to get rid of unused child entries. | |
509 | */ | |
510 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 511 | spin_unlock(&dentry->d_lock); |
1da177e4 | 512 | shrink_dcache_parent(dentry); |
da502956 | 513 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
514 | } |
515 | ||
516 | /* | |
517 | * Somebody else still using it? | |
518 | * | |
519 | * If it's a directory, we can't drop it | |
520 | * for fear of somebody re-populating it | |
521 | * with children (even though dropping it | |
522 | * would make it unreachable from the root, | |
523 | * we might still populate it if it was a | |
524 | * working directory or similar). | |
525 | */ | |
b7ab39f6 | 526 | if (dentry->d_count > 1) { |
1da177e4 LT |
527 | if (dentry->d_inode && S_ISDIR(dentry->d_inode->i_mode)) { |
528 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
529 | return -EBUSY; |
530 | } | |
531 | } | |
532 | ||
533 | __d_drop(dentry); | |
534 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
535 | return 0; |
536 | } | |
ec4f8605 | 537 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 538 | |
b5c84bf6 | 539 | /* This must be called with d_lock held */ |
dc0474be | 540 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 541 | { |
b7ab39f6 | 542 | dentry->d_count++; |
23044507 NP |
543 | } |
544 | ||
dc0474be | 545 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 546 | { |
23044507 | 547 | spin_lock(&dentry->d_lock); |
dc0474be | 548 | __dget_dlock(dentry); |
23044507 | 549 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
550 | } |
551 | ||
b7ab39f6 NP |
552 | struct dentry *dget_parent(struct dentry *dentry) |
553 | { | |
554 | struct dentry *ret; | |
555 | ||
556 | repeat: | |
a734eb45 NP |
557 | /* |
558 | * Don't need rcu_dereference because we re-check it was correct under | |
559 | * the lock. | |
560 | */ | |
561 | rcu_read_lock(); | |
b7ab39f6 | 562 | ret = dentry->d_parent; |
a734eb45 NP |
563 | spin_lock(&ret->d_lock); |
564 | if (unlikely(ret != dentry->d_parent)) { | |
565 | spin_unlock(&ret->d_lock); | |
566 | rcu_read_unlock(); | |
b7ab39f6 NP |
567 | goto repeat; |
568 | } | |
a734eb45 | 569 | rcu_read_unlock(); |
b7ab39f6 NP |
570 | BUG_ON(!ret->d_count); |
571 | ret->d_count++; | |
572 | spin_unlock(&ret->d_lock); | |
b7ab39f6 NP |
573 | return ret; |
574 | } | |
575 | EXPORT_SYMBOL(dget_parent); | |
576 | ||
1da177e4 LT |
577 | /** |
578 | * d_find_alias - grab a hashed alias of inode | |
579 | * @inode: inode in question | |
580 | * @want_discon: flag, used by d_splice_alias, to request | |
581 | * that only a DISCONNECTED alias be returned. | |
582 | * | |
583 | * If inode has a hashed alias, or is a directory and has any alias, | |
584 | * acquire the reference to alias and return it. Otherwise return NULL. | |
585 | * Notice that if inode is a directory there can be only one alias and | |
586 | * it can be unhashed only if it has no children, or if it is the root | |
587 | * of a filesystem. | |
588 | * | |
21c0d8fd | 589 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
1da177e4 | 590 | * any other hashed alias over that one unless @want_discon is set, |
21c0d8fd | 591 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. |
1da177e4 | 592 | */ |
da502956 | 593 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 594 | { |
da502956 | 595 | struct dentry *alias, *discon_alias; |
1da177e4 | 596 | |
da502956 NP |
597 | again: |
598 | discon_alias = NULL; | |
599 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
600 | spin_lock(&alias->d_lock); | |
1da177e4 | 601 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 602 | if (IS_ROOT(alias) && |
da502956 | 603 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 604 | discon_alias = alias; |
da502956 | 605 | } else if (!want_discon) { |
dc0474be | 606 | __dget_dlock(alias); |
da502956 NP |
607 | spin_unlock(&alias->d_lock); |
608 | return alias; | |
609 | } | |
610 | } | |
611 | spin_unlock(&alias->d_lock); | |
612 | } | |
613 | if (discon_alias) { | |
614 | alias = discon_alias; | |
615 | spin_lock(&alias->d_lock); | |
616 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
617 | if (IS_ROOT(alias) && | |
618 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 619 | __dget_dlock(alias); |
da502956 | 620 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
621 | return alias; |
622 | } | |
623 | } | |
da502956 NP |
624 | spin_unlock(&alias->d_lock); |
625 | goto again; | |
1da177e4 | 626 | } |
da502956 | 627 | return NULL; |
1da177e4 LT |
628 | } |
629 | ||
da502956 | 630 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 631 | { |
214fda1f DH |
632 | struct dentry *de = NULL; |
633 | ||
634 | if (!list_empty(&inode->i_dentry)) { | |
873feea0 | 635 | spin_lock(&inode->i_lock); |
214fda1f | 636 | de = __d_find_alias(inode, 0); |
873feea0 | 637 | spin_unlock(&inode->i_lock); |
214fda1f | 638 | } |
1da177e4 LT |
639 | return de; |
640 | } | |
ec4f8605 | 641 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
642 | |
643 | /* | |
644 | * Try to kill dentries associated with this inode. | |
645 | * WARNING: you must own a reference to inode. | |
646 | */ | |
647 | void d_prune_aliases(struct inode *inode) | |
648 | { | |
0cdca3f9 | 649 | struct dentry *dentry; |
1da177e4 | 650 | restart: |
873feea0 | 651 | spin_lock(&inode->i_lock); |
0cdca3f9 | 652 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 653 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 654 | if (!dentry->d_count) { |
dc0474be | 655 | __dget_dlock(dentry); |
1da177e4 LT |
656 | __d_drop(dentry); |
657 | spin_unlock(&dentry->d_lock); | |
873feea0 | 658 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
659 | dput(dentry); |
660 | goto restart; | |
661 | } | |
662 | spin_unlock(&dentry->d_lock); | |
663 | } | |
873feea0 | 664 | spin_unlock(&inode->i_lock); |
1da177e4 | 665 | } |
ec4f8605 | 666 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
667 | |
668 | /* | |
77812a1e NP |
669 | * Try to throw away a dentry - free the inode, dput the parent. |
670 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
671 | * Releases dentry->d_lock. | |
d702ccb3 | 672 | * |
77812a1e | 673 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 674 | */ |
77812a1e | 675 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 676 | __releases(dentry->d_lock) |
1da177e4 | 677 | { |
77812a1e | 678 | struct dentry *parent; |
d52b9086 | 679 | |
77812a1e | 680 | parent = dentry_kill(dentry, 0); |
d52b9086 | 681 | /* |
77812a1e NP |
682 | * If dentry_kill returns NULL, we have nothing more to do. |
683 | * if it returns the same dentry, trylocks failed. In either | |
684 | * case, just loop again. | |
685 | * | |
686 | * Otherwise, we need to prune ancestors too. This is necessary | |
687 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
688 | * is also expected to be beneficial in reducing dentry cache | |
689 | * fragmentation. | |
d52b9086 | 690 | */ |
77812a1e NP |
691 | if (!parent) |
692 | return; | |
693 | if (parent == dentry) | |
694 | return; | |
695 | ||
696 | /* Prune ancestors. */ | |
697 | dentry = parent; | |
d52b9086 | 698 | while (dentry) { |
b7ab39f6 | 699 | spin_lock(&dentry->d_lock); |
89e60548 NP |
700 | if (dentry->d_count > 1) { |
701 | dentry->d_count--; | |
702 | spin_unlock(&dentry->d_lock); | |
703 | return; | |
704 | } | |
77812a1e | 705 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 706 | } |
1da177e4 LT |
707 | } |
708 | ||
3049cfe2 | 709 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 710 | { |
da3bbdd4 | 711 | struct dentry *dentry; |
da3bbdd4 | 712 | |
ec33679d NP |
713 | rcu_read_lock(); |
714 | for (;;) { | |
ec33679d NP |
715 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
716 | if (&dentry->d_lru == list) | |
717 | break; /* empty */ | |
718 | spin_lock(&dentry->d_lock); | |
719 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
720 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
721 | continue; |
722 | } | |
723 | ||
1da177e4 LT |
724 | /* |
725 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
726 | * the LRU because of laziness during lookup. Do not free |
727 | * it - just keep it off the LRU list. | |
1da177e4 | 728 | */ |
b7ab39f6 | 729 | if (dentry->d_count) { |
ec33679d | 730 | dentry_lru_del(dentry); |
da3bbdd4 | 731 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
732 | continue; |
733 | } | |
ec33679d | 734 | |
ec33679d | 735 | rcu_read_unlock(); |
77812a1e NP |
736 | |
737 | try_prune_one_dentry(dentry); | |
738 | ||
ec33679d | 739 | rcu_read_lock(); |
da3bbdd4 | 740 | } |
ec33679d | 741 | rcu_read_unlock(); |
3049cfe2 CH |
742 | } |
743 | ||
744 | /** | |
745 | * __shrink_dcache_sb - shrink the dentry LRU on a given superblock | |
746 | * @sb: superblock to shrink dentry LRU. | |
747 | * @count: number of entries to prune | |
748 | * @flags: flags to control the dentry processing | |
749 | * | |
750 | * If flags contains DCACHE_REFERENCED reference dentries will not be pruned. | |
751 | */ | |
b0d40c92 | 752 | static void __shrink_dcache_sb(struct super_block *sb, int count, int flags) |
3049cfe2 | 753 | { |
3049cfe2 CH |
754 | struct dentry *dentry; |
755 | LIST_HEAD(referenced); | |
756 | LIST_HEAD(tmp); | |
3049cfe2 | 757 | |
23044507 NP |
758 | relock: |
759 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
760 | while (!list_empty(&sb->s_dentry_lru)) { |
761 | dentry = list_entry(sb->s_dentry_lru.prev, | |
762 | struct dentry, d_lru); | |
763 | BUG_ON(dentry->d_sb != sb); | |
764 | ||
23044507 NP |
765 | if (!spin_trylock(&dentry->d_lock)) { |
766 | spin_unlock(&dcache_lru_lock); | |
767 | cpu_relax(); | |
768 | goto relock; | |
769 | } | |
770 | ||
3049cfe2 CH |
771 | /* |
772 | * If we are honouring the DCACHE_REFERENCED flag and the | |
773 | * dentry has this flag set, don't free it. Clear the flag | |
774 | * and put it back on the LRU. | |
775 | */ | |
23044507 NP |
776 | if (flags & DCACHE_REFERENCED && |
777 | dentry->d_flags & DCACHE_REFERENCED) { | |
778 | dentry->d_flags &= ~DCACHE_REFERENCED; | |
779 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 780 | spin_unlock(&dentry->d_lock); |
23044507 NP |
781 | } else { |
782 | list_move_tail(&dentry->d_lru, &tmp); | |
783 | spin_unlock(&dentry->d_lock); | |
b0d40c92 | 784 | if (!--count) |
23044507 | 785 | break; |
3049cfe2 | 786 | } |
ec33679d | 787 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 788 | } |
da3bbdd4 KM |
789 | if (!list_empty(&referenced)) |
790 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 791 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
792 | |
793 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
794 | } |
795 | ||
796 | /** | |
b0d40c92 | 797 | * prune_dcache_sb - shrink the dcache |
2af14162 | 798 | * @sb: superblock |
b0d40c92 | 799 | * @nr_to_scan: number of entries to try to free |
da3bbdd4 | 800 | * |
b0d40c92 DC |
801 | * Attempt to shrink the superblock dcache LRU by @nr_to_scan entries. This is |
802 | * done when we need more memory an called from the superblock shrinker | |
803 | * function. | |
da3bbdd4 | 804 | * |
b0d40c92 DC |
805 | * This function may fail to free any resources if all the dentries are in |
806 | * use. | |
da3bbdd4 | 807 | */ |
b0d40c92 | 808 | void prune_dcache_sb(struct super_block *sb, int nr_to_scan) |
da3bbdd4 | 809 | { |
b0d40c92 | 810 | __shrink_dcache_sb(sb, nr_to_scan, DCACHE_REFERENCED); |
1da177e4 LT |
811 | } |
812 | ||
1da177e4 LT |
813 | /** |
814 | * shrink_dcache_sb - shrink dcache for a superblock | |
815 | * @sb: superblock | |
816 | * | |
3049cfe2 CH |
817 | * Shrink the dcache for the specified super block. This is used to free |
818 | * the dcache before unmounting a file system. | |
1da177e4 | 819 | */ |
3049cfe2 | 820 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 821 | { |
3049cfe2 CH |
822 | LIST_HEAD(tmp); |
823 | ||
23044507 | 824 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
825 | while (!list_empty(&sb->s_dentry_lru)) { |
826 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 827 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 828 | shrink_dentry_list(&tmp); |
ec33679d | 829 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 830 | } |
23044507 | 831 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 832 | } |
ec4f8605 | 833 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 834 | |
c636ebdb DH |
835 | /* |
836 | * destroy a single subtree of dentries for unmount | |
837 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
838 | * locking | |
839 | */ | |
840 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
841 | { | |
842 | struct dentry *parent; | |
843 | ||
844 | BUG_ON(!IS_ROOT(dentry)); | |
845 | ||
c636ebdb DH |
846 | for (;;) { |
847 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 848 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
849 | dentry = list_entry(dentry->d_subdirs.next, |
850 | struct dentry, d_u.d_child); | |
c636ebdb DH |
851 | |
852 | /* consume the dentries from this leaf up through its parents | |
853 | * until we find one with children or run out altogether */ | |
854 | do { | |
855 | struct inode *inode; | |
856 | ||
43c1c9cd DH |
857 | /* detach from the system */ |
858 | dentry_lru_del(dentry); | |
859 | __d_shrink(dentry); | |
860 | ||
b7ab39f6 | 861 | if (dentry->d_count != 0) { |
c636ebdb DH |
862 | printk(KERN_ERR |
863 | "BUG: Dentry %p{i=%lx,n=%s}" | |
864 | " still in use (%d)" | |
865 | " [unmount of %s %s]\n", | |
866 | dentry, | |
867 | dentry->d_inode ? | |
868 | dentry->d_inode->i_ino : 0UL, | |
869 | dentry->d_name.name, | |
b7ab39f6 | 870 | dentry->d_count, |
c636ebdb DH |
871 | dentry->d_sb->s_type->name, |
872 | dentry->d_sb->s_id); | |
873 | BUG(); | |
874 | } | |
875 | ||
2fd6b7f5 | 876 | if (IS_ROOT(dentry)) { |
c636ebdb | 877 | parent = NULL; |
2fd6b7f5 NP |
878 | list_del(&dentry->d_u.d_child); |
879 | } else { | |
871c0067 | 880 | parent = dentry->d_parent; |
b7ab39f6 | 881 | parent->d_count--; |
2fd6b7f5 | 882 | list_del(&dentry->d_u.d_child); |
871c0067 | 883 | } |
c636ebdb | 884 | |
c636ebdb DH |
885 | inode = dentry->d_inode; |
886 | if (inode) { | |
887 | dentry->d_inode = NULL; | |
888 | list_del_init(&dentry->d_alias); | |
889 | if (dentry->d_op && dentry->d_op->d_iput) | |
890 | dentry->d_op->d_iput(dentry, inode); | |
891 | else | |
892 | iput(inode); | |
893 | } | |
894 | ||
895 | d_free(dentry); | |
896 | ||
897 | /* finished when we fall off the top of the tree, | |
898 | * otherwise we ascend to the parent and move to the | |
899 | * next sibling if there is one */ | |
900 | if (!parent) | |
312d3ca8 | 901 | return; |
c636ebdb | 902 | dentry = parent; |
c636ebdb DH |
903 | } while (list_empty(&dentry->d_subdirs)); |
904 | ||
905 | dentry = list_entry(dentry->d_subdirs.next, | |
906 | struct dentry, d_u.d_child); | |
907 | } | |
908 | } | |
909 | ||
910 | /* | |
911 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 912 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
913 | * - the superblock is detached from all mountings and open files, so the |
914 | * dentry trees will not be rearranged by the VFS | |
915 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
916 | * any dentries belonging to this superblock that it comes across | |
917 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
918 | * in this superblock | |
919 | */ | |
920 | void shrink_dcache_for_umount(struct super_block *sb) | |
921 | { | |
922 | struct dentry *dentry; | |
923 | ||
924 | if (down_read_trylock(&sb->s_umount)) | |
925 | BUG(); | |
926 | ||
927 | dentry = sb->s_root; | |
928 | sb->s_root = NULL; | |
b7ab39f6 | 929 | dentry->d_count--; |
c636ebdb DH |
930 | shrink_dcache_for_umount_subtree(dentry); |
931 | ||
ceb5bdc2 NP |
932 | while (!hlist_bl_empty(&sb->s_anon)) { |
933 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
934 | shrink_dcache_for_umount_subtree(dentry); |
935 | } | |
936 | } | |
937 | ||
c826cb7d LT |
938 | /* |
939 | * This tries to ascend one level of parenthood, but | |
940 | * we can race with renaming, so we need to re-check | |
941 | * the parenthood after dropping the lock and check | |
942 | * that the sequence number still matches. | |
943 | */ | |
944 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
945 | { | |
946 | struct dentry *new = old->d_parent; | |
947 | ||
948 | rcu_read_lock(); | |
949 | spin_unlock(&old->d_lock); | |
950 | spin_lock(&new->d_lock); | |
951 | ||
952 | /* | |
953 | * might go back up the wrong parent if we have had a rename | |
954 | * or deletion | |
955 | */ | |
956 | if (new != old->d_parent || | |
c83ce989 | 957 | (old->d_flags & DCACHE_DISCONNECTED) || |
c826cb7d LT |
958 | (!locked && read_seqretry(&rename_lock, seq))) { |
959 | spin_unlock(&new->d_lock); | |
960 | new = NULL; | |
961 | } | |
962 | rcu_read_unlock(); | |
963 | return new; | |
964 | } | |
965 | ||
966 | ||
1da177e4 LT |
967 | /* |
968 | * Search for at least 1 mount point in the dentry's subdirs. | |
969 | * We descend to the next level whenever the d_subdirs | |
970 | * list is non-empty and continue searching. | |
971 | */ | |
972 | ||
973 | /** | |
974 | * have_submounts - check for mounts over a dentry | |
975 | * @parent: dentry to check. | |
976 | * | |
977 | * Return true if the parent or its subdirectories contain | |
978 | * a mount point | |
979 | */ | |
1da177e4 LT |
980 | int have_submounts(struct dentry *parent) |
981 | { | |
949854d0 | 982 | struct dentry *this_parent; |
1da177e4 | 983 | struct list_head *next; |
949854d0 | 984 | unsigned seq; |
58db63d0 | 985 | int locked = 0; |
949854d0 | 986 | |
949854d0 | 987 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
988 | again: |
989 | this_parent = parent; | |
1da177e4 | 990 | |
1da177e4 LT |
991 | if (d_mountpoint(parent)) |
992 | goto positive; | |
2fd6b7f5 | 993 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
994 | repeat: |
995 | next = this_parent->d_subdirs.next; | |
996 | resume: | |
997 | while (next != &this_parent->d_subdirs) { | |
998 | struct list_head *tmp = next; | |
5160ee6f | 999 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1000 | next = tmp->next; |
2fd6b7f5 NP |
1001 | |
1002 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1003 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1004 | if (d_mountpoint(dentry)) { |
1005 | spin_unlock(&dentry->d_lock); | |
1006 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1007 | goto positive; |
2fd6b7f5 | 1008 | } |
1da177e4 | 1009 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1010 | spin_unlock(&this_parent->d_lock); |
1011 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1012 | this_parent = dentry; |
2fd6b7f5 | 1013 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1014 | goto repeat; |
1015 | } | |
2fd6b7f5 | 1016 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1017 | } |
1018 | /* | |
1019 | * All done at this level ... ascend and resume the search. | |
1020 | */ | |
1021 | if (this_parent != parent) { | |
c826cb7d LT |
1022 | struct dentry *child = this_parent; |
1023 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1024 | if (!this_parent) | |
949854d0 | 1025 | goto rename_retry; |
949854d0 | 1026 | next = child->d_u.d_child.next; |
1da177e4 LT |
1027 | goto resume; |
1028 | } | |
2fd6b7f5 | 1029 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1030 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1031 | goto rename_retry; |
58db63d0 NP |
1032 | if (locked) |
1033 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1034 | return 0; /* No mount points found in tree */ |
1035 | positive: | |
58db63d0 | 1036 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1037 | goto rename_retry; |
58db63d0 NP |
1038 | if (locked) |
1039 | write_sequnlock(&rename_lock); | |
1da177e4 | 1040 | return 1; |
58db63d0 NP |
1041 | |
1042 | rename_retry: | |
1043 | locked = 1; | |
1044 | write_seqlock(&rename_lock); | |
1045 | goto again; | |
1da177e4 | 1046 | } |
ec4f8605 | 1047 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1048 | |
1049 | /* | |
1050 | * Search the dentry child list for the specified parent, | |
1051 | * and move any unused dentries to the end of the unused | |
1052 | * list for prune_dcache(). We descend to the next level | |
1053 | * whenever the d_subdirs list is non-empty and continue | |
1054 | * searching. | |
1055 | * | |
1056 | * It returns zero iff there are no unused children, | |
1057 | * otherwise it returns the number of children moved to | |
1058 | * the end of the unused list. This may not be the total | |
1059 | * number of unused children, because select_parent can | |
1060 | * drop the lock and return early due to latency | |
1061 | * constraints. | |
1062 | */ | |
1063 | static int select_parent(struct dentry * parent) | |
1064 | { | |
949854d0 | 1065 | struct dentry *this_parent; |
1da177e4 | 1066 | struct list_head *next; |
949854d0 | 1067 | unsigned seq; |
1da177e4 | 1068 | int found = 0; |
58db63d0 | 1069 | int locked = 0; |
1da177e4 | 1070 | |
949854d0 | 1071 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1072 | again: |
1073 | this_parent = parent; | |
2fd6b7f5 | 1074 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1075 | repeat: |
1076 | next = this_parent->d_subdirs.next; | |
1077 | resume: | |
1078 | while (next != &this_parent->d_subdirs) { | |
1079 | struct list_head *tmp = next; | |
5160ee6f | 1080 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1081 | next = tmp->next; |
1082 | ||
2fd6b7f5 | 1083 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1084 | |
1da177e4 LT |
1085 | /* |
1086 | * move only zero ref count dentries to the end | |
1087 | * of the unused list for prune_dcache | |
1088 | */ | |
b7ab39f6 | 1089 | if (!dentry->d_count) { |
a4633357 | 1090 | dentry_lru_move_tail(dentry); |
1da177e4 | 1091 | found++; |
a4633357 CH |
1092 | } else { |
1093 | dentry_lru_del(dentry); | |
1da177e4 LT |
1094 | } |
1095 | ||
1096 | /* | |
1097 | * We can return to the caller if we have found some (this | |
1098 | * ensures forward progress). We'll be coming back to find | |
1099 | * the rest. | |
1100 | */ | |
2fd6b7f5 NP |
1101 | if (found && need_resched()) { |
1102 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1103 | goto out; |
2fd6b7f5 | 1104 | } |
1da177e4 LT |
1105 | |
1106 | /* | |
1107 | * Descend a level if the d_subdirs list is non-empty. | |
1108 | */ | |
1109 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1110 | spin_unlock(&this_parent->d_lock); |
1111 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1112 | this_parent = dentry; |
2fd6b7f5 | 1113 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1114 | goto repeat; |
1115 | } | |
2fd6b7f5 NP |
1116 | |
1117 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1118 | } |
1119 | /* | |
1120 | * All done at this level ... ascend and resume the search. | |
1121 | */ | |
1122 | if (this_parent != parent) { | |
c826cb7d LT |
1123 | struct dentry *child = this_parent; |
1124 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1125 | if (!this_parent) | |
949854d0 | 1126 | goto rename_retry; |
949854d0 | 1127 | next = child->d_u.d_child.next; |
1da177e4 LT |
1128 | goto resume; |
1129 | } | |
1130 | out: | |
2fd6b7f5 | 1131 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1132 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1133 | goto rename_retry; |
58db63d0 NP |
1134 | if (locked) |
1135 | write_sequnlock(&rename_lock); | |
1da177e4 | 1136 | return found; |
58db63d0 NP |
1137 | |
1138 | rename_retry: | |
1139 | if (found) | |
1140 | return found; | |
1141 | locked = 1; | |
1142 | write_seqlock(&rename_lock); | |
1143 | goto again; | |
1da177e4 LT |
1144 | } |
1145 | ||
1146 | /** | |
1147 | * shrink_dcache_parent - prune dcache | |
1148 | * @parent: parent of entries to prune | |
1149 | * | |
1150 | * Prune the dcache to remove unused children of the parent dentry. | |
1151 | */ | |
1152 | ||
1153 | void shrink_dcache_parent(struct dentry * parent) | |
1154 | { | |
da3bbdd4 | 1155 | struct super_block *sb = parent->d_sb; |
1da177e4 LT |
1156 | int found; |
1157 | ||
1158 | while ((found = select_parent(parent)) != 0) | |
b0d40c92 | 1159 | __shrink_dcache_sb(sb, found, 0); |
1da177e4 | 1160 | } |
ec4f8605 | 1161 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1162 | |
1da177e4 | 1163 | /** |
a4464dbc AV |
1164 | * __d_alloc - allocate a dcache entry |
1165 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1166 | * @name: qstr of the name |
1167 | * | |
1168 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1169 | * available. On a success the dentry is returned. The name passed in is | |
1170 | * copied and the copy passed in may be reused after this call. | |
1171 | */ | |
1172 | ||
a4464dbc | 1173 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1174 | { |
1175 | struct dentry *dentry; | |
1176 | char *dname; | |
1177 | ||
e12ba74d | 1178 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1179 | if (!dentry) |
1180 | return NULL; | |
1181 | ||
1182 | if (name->len > DNAME_INLINE_LEN-1) { | |
1183 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1184 | if (!dname) { | |
1185 | kmem_cache_free(dentry_cache, dentry); | |
1186 | return NULL; | |
1187 | } | |
1188 | } else { | |
1189 | dname = dentry->d_iname; | |
1190 | } | |
1191 | dentry->d_name.name = dname; | |
1192 | ||
1193 | dentry->d_name.len = name->len; | |
1194 | dentry->d_name.hash = name->hash; | |
1195 | memcpy(dname, name->name, name->len); | |
1196 | dname[name->len] = 0; | |
1197 | ||
b7ab39f6 | 1198 | dentry->d_count = 1; |
dea3667b | 1199 | dentry->d_flags = 0; |
1da177e4 | 1200 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1201 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1202 | dentry->d_inode = NULL; |
a4464dbc AV |
1203 | dentry->d_parent = dentry; |
1204 | dentry->d_sb = sb; | |
1da177e4 LT |
1205 | dentry->d_op = NULL; |
1206 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1207 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1208 | INIT_LIST_HEAD(&dentry->d_lru); |
1209 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1210 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1211 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1212 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1213 | |
3e880fb5 | 1214 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1215 | |
1da177e4 LT |
1216 | return dentry; |
1217 | } | |
a4464dbc AV |
1218 | |
1219 | /** | |
1220 | * d_alloc - allocate a dcache entry | |
1221 | * @parent: parent of entry to allocate | |
1222 | * @name: qstr of the name | |
1223 | * | |
1224 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1225 | * available. On a success the dentry is returned. The name passed in is | |
1226 | * copied and the copy passed in may be reused after this call. | |
1227 | */ | |
1228 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1229 | { | |
1230 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1231 | if (!dentry) | |
1232 | return NULL; | |
1233 | ||
1234 | spin_lock(&parent->d_lock); | |
1235 | /* | |
1236 | * don't need child lock because it is not subject | |
1237 | * to concurrency here | |
1238 | */ | |
1239 | __dget_dlock(parent); | |
1240 | dentry->d_parent = parent; | |
1241 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1242 | spin_unlock(&parent->d_lock); | |
1243 | ||
1244 | return dentry; | |
1245 | } | |
ec4f8605 | 1246 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1247 | |
4b936885 NP |
1248 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1249 | { | |
a4464dbc AV |
1250 | struct dentry *dentry = __d_alloc(sb, name); |
1251 | if (dentry) | |
4b936885 | 1252 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1253 | return dentry; |
1254 | } | |
1255 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1256 | ||
1da177e4 LT |
1257 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1258 | { | |
1259 | struct qstr q; | |
1260 | ||
1261 | q.name = name; | |
1262 | q.len = strlen(name); | |
1263 | q.hash = full_name_hash(q.name, q.len); | |
1264 | return d_alloc(parent, &q); | |
1265 | } | |
ef26ca97 | 1266 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1267 | |
fb045adb NP |
1268 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1269 | { | |
6f7f7caa LT |
1270 | WARN_ON_ONCE(dentry->d_op); |
1271 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1272 | DCACHE_OP_COMPARE | |
1273 | DCACHE_OP_REVALIDATE | | |
1274 | DCACHE_OP_DELETE )); | |
1275 | dentry->d_op = op; | |
1276 | if (!op) | |
1277 | return; | |
1278 | if (op->d_hash) | |
1279 | dentry->d_flags |= DCACHE_OP_HASH; | |
1280 | if (op->d_compare) | |
1281 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1282 | if (op->d_revalidate) | |
1283 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1284 | if (op->d_delete) | |
1285 | dentry->d_flags |= DCACHE_OP_DELETE; | |
1286 | ||
1287 | } | |
1288 | EXPORT_SYMBOL(d_set_d_op); | |
1289 | ||
360da900 OH |
1290 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1291 | { | |
b23fb0a6 | 1292 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1293 | if (inode) { |
1294 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1295 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
360da900 | 1296 | list_add(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1297 | } |
360da900 | 1298 | dentry->d_inode = inode; |
31e6b01f | 1299 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1300 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1301 | fsnotify_d_instantiate(dentry, inode); |
1302 | } | |
1303 | ||
1da177e4 LT |
1304 | /** |
1305 | * d_instantiate - fill in inode information for a dentry | |
1306 | * @entry: dentry to complete | |
1307 | * @inode: inode to attach to this dentry | |
1308 | * | |
1309 | * Fill in inode information in the entry. | |
1310 | * | |
1311 | * This turns negative dentries into productive full members | |
1312 | * of society. | |
1313 | * | |
1314 | * NOTE! This assumes that the inode count has been incremented | |
1315 | * (or otherwise set) by the caller to indicate that it is now | |
1316 | * in use by the dcache. | |
1317 | */ | |
1318 | ||
1319 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1320 | { | |
28133c7b | 1321 | BUG_ON(!list_empty(&entry->d_alias)); |
873feea0 NP |
1322 | if (inode) |
1323 | spin_lock(&inode->i_lock); | |
360da900 | 1324 | __d_instantiate(entry, inode); |
873feea0 NP |
1325 | if (inode) |
1326 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1327 | security_d_instantiate(entry, inode); |
1328 | } | |
ec4f8605 | 1329 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1330 | |
1331 | /** | |
1332 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1333 | * @entry: dentry to instantiate | |
1334 | * @inode: inode to attach to this dentry | |
1335 | * | |
1336 | * Fill in inode information in the entry. On success, it returns NULL. | |
1337 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1338 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1339 | * |
1340 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1341 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1342 | * |
1343 | * This also assumes that the inode count has been incremented | |
1344 | * (or otherwise set) by the caller to indicate that it is now | |
1345 | * in use by the dcache. | |
1da177e4 | 1346 | */ |
770bfad8 DH |
1347 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1348 | struct inode *inode) | |
1da177e4 LT |
1349 | { |
1350 | struct dentry *alias; | |
1351 | int len = entry->d_name.len; | |
1352 | const char *name = entry->d_name.name; | |
1353 | unsigned int hash = entry->d_name.hash; | |
1354 | ||
770bfad8 | 1355 | if (!inode) { |
360da900 | 1356 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1357 | return NULL; |
1358 | } | |
1359 | ||
1da177e4 LT |
1360 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
1361 | struct qstr *qstr = &alias->d_name; | |
1362 | ||
9abca360 NP |
1363 | /* |
1364 | * Don't need alias->d_lock here, because aliases with | |
1365 | * d_parent == entry->d_parent are not subject to name or | |
1366 | * parent changes, because the parent inode i_mutex is held. | |
1367 | */ | |
1da177e4 LT |
1368 | if (qstr->hash != hash) |
1369 | continue; | |
1370 | if (alias->d_parent != entry->d_parent) | |
1371 | continue; | |
9d55c369 | 1372 | if (dentry_cmp(qstr->name, qstr->len, name, len)) |
1da177e4 | 1373 | continue; |
dc0474be | 1374 | __dget(alias); |
1da177e4 LT |
1375 | return alias; |
1376 | } | |
770bfad8 | 1377 | |
360da900 | 1378 | __d_instantiate(entry, inode); |
1da177e4 LT |
1379 | return NULL; |
1380 | } | |
770bfad8 DH |
1381 | |
1382 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1383 | { | |
1384 | struct dentry *result; | |
1385 | ||
1386 | BUG_ON(!list_empty(&entry->d_alias)); | |
1387 | ||
873feea0 NP |
1388 | if (inode) |
1389 | spin_lock(&inode->i_lock); | |
770bfad8 | 1390 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1391 | if (inode) |
1392 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1393 | |
1394 | if (!result) { | |
1395 | security_d_instantiate(entry, inode); | |
1396 | return NULL; | |
1397 | } | |
1398 | ||
1399 | BUG_ON(!d_unhashed(result)); | |
1400 | iput(inode); | |
1401 | return result; | |
1402 | } | |
1403 | ||
1da177e4 LT |
1404 | EXPORT_SYMBOL(d_instantiate_unique); |
1405 | ||
1406 | /** | |
1407 | * d_alloc_root - allocate root dentry | |
1408 | * @root_inode: inode to allocate the root for | |
1409 | * | |
1410 | * Allocate a root ("/") dentry for the inode given. The inode is | |
1411 | * instantiated and returned. %NULL is returned if there is insufficient | |
1412 | * memory or the inode passed is %NULL. | |
1413 | */ | |
1414 | ||
1415 | struct dentry * d_alloc_root(struct inode * root_inode) | |
1416 | { | |
1417 | struct dentry *res = NULL; | |
1418 | ||
1419 | if (root_inode) { | |
1420 | static const struct qstr name = { .name = "/", .len = 1 }; | |
1421 | ||
a4464dbc AV |
1422 | res = __d_alloc(root_inode->i_sb, &name); |
1423 | if (res) | |
1da177e4 | 1424 | d_instantiate(res, root_inode); |
1da177e4 LT |
1425 | } |
1426 | return res; | |
1427 | } | |
ec4f8605 | 1428 | EXPORT_SYMBOL(d_alloc_root); |
1da177e4 | 1429 | |
d891eedb BF |
1430 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1431 | { | |
1432 | struct dentry *alias; | |
1433 | ||
1434 | if (list_empty(&inode->i_dentry)) | |
1435 | return NULL; | |
1436 | alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias); | |
1437 | __dget(alias); | |
1438 | return alias; | |
1439 | } | |
1440 | ||
1441 | static struct dentry * d_find_any_alias(struct inode *inode) | |
1442 | { | |
1443 | struct dentry *de; | |
1444 | ||
1445 | spin_lock(&inode->i_lock); | |
1446 | de = __d_find_any_alias(inode); | |
1447 | spin_unlock(&inode->i_lock); | |
1448 | return de; | |
1449 | } | |
1450 | ||
1451 | ||
4ea3ada2 CH |
1452 | /** |
1453 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1454 | * @inode: inode to allocate the dentry for | |
1455 | * | |
1456 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1457 | * similar open by handle operations. The returned dentry may be anonymous, | |
1458 | * or may have a full name (if the inode was already in the cache). | |
1459 | * | |
1460 | * When called on a directory inode, we must ensure that the inode only ever | |
1461 | * has one dentry. If a dentry is found, that is returned instead of | |
1462 | * allocating a new one. | |
1463 | * | |
1464 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1465 | * to the dentry. In case of an error the reference on the inode is released. |
1466 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1467 | * be passed in and will be the error will be propagate to the return value, | |
1468 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1469 | */ |
1470 | struct dentry *d_obtain_alias(struct inode *inode) | |
1471 | { | |
9308a612 CH |
1472 | static const struct qstr anonstring = { .name = "" }; |
1473 | struct dentry *tmp; | |
1474 | struct dentry *res; | |
4ea3ada2 CH |
1475 | |
1476 | if (!inode) | |
44003728 | 1477 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1478 | if (IS_ERR(inode)) |
1479 | return ERR_CAST(inode); | |
1480 | ||
d891eedb | 1481 | res = d_find_any_alias(inode); |
9308a612 CH |
1482 | if (res) |
1483 | goto out_iput; | |
1484 | ||
a4464dbc | 1485 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1486 | if (!tmp) { |
1487 | res = ERR_PTR(-ENOMEM); | |
1488 | goto out_iput; | |
4ea3ada2 | 1489 | } |
b5c84bf6 | 1490 | |
873feea0 | 1491 | spin_lock(&inode->i_lock); |
d891eedb | 1492 | res = __d_find_any_alias(inode); |
9308a612 | 1493 | if (res) { |
873feea0 | 1494 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1495 | dput(tmp); |
1496 | goto out_iput; | |
1497 | } | |
1498 | ||
1499 | /* attach a disconnected dentry */ | |
1500 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1501 | tmp->d_inode = inode; |
1502 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
9308a612 | 1503 | list_add(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1504 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1505 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1506 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1507 | spin_unlock(&tmp->d_lock); |
873feea0 | 1508 | spin_unlock(&inode->i_lock); |
24ff6663 | 1509 | security_d_instantiate(tmp, inode); |
9308a612 | 1510 | |
9308a612 CH |
1511 | return tmp; |
1512 | ||
1513 | out_iput: | |
24ff6663 JB |
1514 | if (res && !IS_ERR(res)) |
1515 | security_d_instantiate(res, inode); | |
9308a612 CH |
1516 | iput(inode); |
1517 | return res; | |
4ea3ada2 | 1518 | } |
adc48720 | 1519 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1520 | |
1521 | /** | |
1522 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1523 | * @inode: the inode which may have a disconnected dentry | |
1524 | * @dentry: a negative dentry which we want to point to the inode. | |
1525 | * | |
1526 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1527 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1528 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1529 | * | |
1530 | * This is needed in the lookup routine of any filesystem that is exportable | |
1531 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1532 | * | |
1533 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1534 | * is returned. This matches the expected return value of ->lookup. | |
1535 | * | |
1536 | */ | |
1537 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1538 | { | |
1539 | struct dentry *new = NULL; | |
1540 | ||
a9049376 AV |
1541 | if (IS_ERR(inode)) |
1542 | return ERR_CAST(inode); | |
1543 | ||
21c0d8fd | 1544 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1545 | spin_lock(&inode->i_lock); |
1da177e4 LT |
1546 | new = __d_find_alias(inode, 1); |
1547 | if (new) { | |
1548 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); | |
873feea0 | 1549 | spin_unlock(&inode->i_lock); |
1da177e4 | 1550 | security_d_instantiate(new, inode); |
1da177e4 LT |
1551 | d_move(new, dentry); |
1552 | iput(inode); | |
1553 | } else { | |
873feea0 | 1554 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1555 | __d_instantiate(dentry, inode); |
873feea0 | 1556 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1557 | security_d_instantiate(dentry, inode); |
1558 | d_rehash(dentry); | |
1559 | } | |
1560 | } else | |
1561 | d_add(dentry, inode); | |
1562 | return new; | |
1563 | } | |
ec4f8605 | 1564 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1565 | |
9403540c BN |
1566 | /** |
1567 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1568 | * @inode: the inode case-insensitive lookup has found | |
1569 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1570 | * @name: the case-exact name to be associated with the returned dentry | |
1571 | * | |
1572 | * This is to avoid filling the dcache with case-insensitive names to the | |
1573 | * same inode, only the actual correct case is stored in the dcache for | |
1574 | * case-insensitive filesystems. | |
1575 | * | |
1576 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1577 | * already exists in in the dcache, use it and return it. | |
1578 | * | |
1579 | * If no entry exists with the exact case name, allocate new dentry with | |
1580 | * the exact case, and return the spliced entry. | |
1581 | */ | |
e45b590b | 1582 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1583 | struct qstr *name) |
1584 | { | |
1585 | int error; | |
1586 | struct dentry *found; | |
1587 | struct dentry *new; | |
1588 | ||
b6520c81 CH |
1589 | /* |
1590 | * First check if a dentry matching the name already exists, | |
1591 | * if not go ahead and create it now. | |
1592 | */ | |
9403540c | 1593 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1594 | if (!found) { |
1595 | new = d_alloc(dentry->d_parent, name); | |
1596 | if (!new) { | |
1597 | error = -ENOMEM; | |
1598 | goto err_out; | |
1599 | } | |
b6520c81 | 1600 | |
9403540c BN |
1601 | found = d_splice_alias(inode, new); |
1602 | if (found) { | |
1603 | dput(new); | |
1604 | return found; | |
1605 | } | |
1606 | return new; | |
1607 | } | |
b6520c81 CH |
1608 | |
1609 | /* | |
1610 | * If a matching dentry exists, and it's not negative use it. | |
1611 | * | |
1612 | * Decrement the reference count to balance the iget() done | |
1613 | * earlier on. | |
1614 | */ | |
9403540c BN |
1615 | if (found->d_inode) { |
1616 | if (unlikely(found->d_inode != inode)) { | |
1617 | /* This can't happen because bad inodes are unhashed. */ | |
1618 | BUG_ON(!is_bad_inode(inode)); | |
1619 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1620 | } | |
9403540c BN |
1621 | iput(inode); |
1622 | return found; | |
1623 | } | |
b6520c81 | 1624 | |
9403540c | 1625 | /* |
44396f4b JB |
1626 | * We are going to instantiate this dentry, unhash it and clear the |
1627 | * lookup flag so we can do that. | |
9403540c | 1628 | */ |
44396f4b JB |
1629 | if (unlikely(d_need_lookup(found))) |
1630 | d_clear_need_lookup(found); | |
b6520c81 | 1631 | |
9403540c | 1632 | /* |
9403540c | 1633 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1634 | * already has a dentry. |
9403540c | 1635 | */ |
4513d899 AV |
1636 | new = d_splice_alias(inode, found); |
1637 | if (new) { | |
1638 | dput(found); | |
1639 | found = new; | |
9403540c | 1640 | } |
4513d899 | 1641 | return found; |
9403540c BN |
1642 | |
1643 | err_out: | |
1644 | iput(inode); | |
1645 | return ERR_PTR(error); | |
1646 | } | |
ec4f8605 | 1647 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1648 | |
31e6b01f NP |
1649 | /** |
1650 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1651 | * @parent: parent dentry | |
1652 | * @name: qstr of name we wish to find | |
1653 | * @seq: returns d_seq value at the point where the dentry was found | |
1654 | * @inode: returns dentry->d_inode when the inode was found valid. | |
1655 | * Returns: dentry, or NULL | |
1656 | * | |
1657 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1658 | * resolution (store-free path walking) design described in | |
1659 | * Documentation/filesystems/path-lookup.txt. | |
1660 | * | |
1661 | * This is not to be used outside core vfs. | |
1662 | * | |
1663 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1664 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1665 | * without taking d_lock and checking d_seq sequence count against @seq | |
1666 | * returned here. | |
1667 | * | |
1668 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1669 | * function. | |
1670 | * | |
1671 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1672 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1673 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1674 | * is formed, giving integrity down the path walk. | |
1675 | */ | |
1676 | struct dentry *__d_lookup_rcu(struct dentry *parent, struct qstr *name, | |
1677 | unsigned *seq, struct inode **inode) | |
1678 | { | |
1679 | unsigned int len = name->len; | |
1680 | unsigned int hash = name->hash; | |
1681 | const unsigned char *str = name->name; | |
b07ad996 | 1682 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1683 | struct hlist_bl_node *node; |
31e6b01f NP |
1684 | struct dentry *dentry; |
1685 | ||
1686 | /* | |
1687 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1688 | * required to prevent single threaded performance regressions | |
1689 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1690 | * Keep the two functions in sync. | |
1691 | */ | |
1692 | ||
1693 | /* | |
1694 | * The hash list is protected using RCU. | |
1695 | * | |
1696 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1697 | * races with d_move(). | |
1698 | * | |
1699 | * It is possible that concurrent renames can mess up our list | |
1700 | * walk here and result in missing our dentry, resulting in the | |
1701 | * false-negative result. d_lookup() protects against concurrent | |
1702 | * renames using rename_lock seqlock. | |
1703 | * | |
b0a4bb83 | 1704 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1705 | */ |
b07ad996 | 1706 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1707 | struct inode *i; |
1708 | const char *tname; | |
1709 | int tlen; | |
1710 | ||
1711 | if (dentry->d_name.hash != hash) | |
1712 | continue; | |
1713 | ||
1714 | seqretry: | |
1715 | *seq = read_seqcount_begin(&dentry->d_seq); | |
1716 | if (dentry->d_parent != parent) | |
1717 | continue; | |
1718 | if (d_unhashed(dentry)) | |
1719 | continue; | |
1720 | tlen = dentry->d_name.len; | |
1721 | tname = dentry->d_name.name; | |
1722 | i = dentry->d_inode; | |
e1bb5782 | 1723 | prefetch(tname); |
31e6b01f NP |
1724 | /* |
1725 | * This seqcount check is required to ensure name and | |
1726 | * len are loaded atomically, so as not to walk off the | |
1727 | * edge of memory when walking. If we could load this | |
1728 | * atomically some other way, we could drop this check. | |
1729 | */ | |
1730 | if (read_seqcount_retry(&dentry->d_seq, *seq)) | |
1731 | goto seqretry; | |
830c0f0e | 1732 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
31e6b01f NP |
1733 | if (parent->d_op->d_compare(parent, *inode, |
1734 | dentry, i, | |
1735 | tlen, tname, name)) | |
1736 | continue; | |
1737 | } else { | |
9d55c369 | 1738 | if (dentry_cmp(tname, tlen, str, len)) |
31e6b01f NP |
1739 | continue; |
1740 | } | |
1741 | /* | |
1742 | * No extra seqcount check is required after the name | |
1743 | * compare. The caller must perform a seqcount check in | |
1744 | * order to do anything useful with the returned dentry | |
1745 | * anyway. | |
1746 | */ | |
1747 | *inode = i; | |
1748 | return dentry; | |
1749 | } | |
1750 | return NULL; | |
1751 | } | |
1752 | ||
1da177e4 LT |
1753 | /** |
1754 | * d_lookup - search for a dentry | |
1755 | * @parent: parent dentry | |
1756 | * @name: qstr of name we wish to find | |
b04f784e | 1757 | * Returns: dentry, or NULL |
1da177e4 | 1758 | * |
b04f784e NP |
1759 | * d_lookup searches the children of the parent dentry for the name in |
1760 | * question. If the dentry is found its reference count is incremented and the | |
1761 | * dentry is returned. The caller must use dput to free the entry when it has | |
1762 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1763 | */ |
31e6b01f | 1764 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 | 1765 | { |
31e6b01f | 1766 | struct dentry *dentry; |
949854d0 | 1767 | unsigned seq; |
1da177e4 LT |
1768 | |
1769 | do { | |
1770 | seq = read_seqbegin(&rename_lock); | |
1771 | dentry = __d_lookup(parent, name); | |
1772 | if (dentry) | |
1773 | break; | |
1774 | } while (read_seqretry(&rename_lock, seq)); | |
1775 | return dentry; | |
1776 | } | |
ec4f8605 | 1777 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1778 | |
31e6b01f | 1779 | /** |
b04f784e NP |
1780 | * __d_lookup - search for a dentry (racy) |
1781 | * @parent: parent dentry | |
1782 | * @name: qstr of name we wish to find | |
1783 | * Returns: dentry, or NULL | |
1784 | * | |
1785 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1786 | * false-negative result due to unrelated rename activity. | |
1787 | * | |
1788 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1789 | * however it must be used carefully, eg. with a following d_lookup in | |
1790 | * the case of failure. | |
1791 | * | |
1792 | * __d_lookup callers must be commented. | |
1793 | */ | |
31e6b01f | 1794 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 LT |
1795 | { |
1796 | unsigned int len = name->len; | |
1797 | unsigned int hash = name->hash; | |
1798 | const unsigned char *str = name->name; | |
b07ad996 | 1799 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1800 | struct hlist_bl_node *node; |
31e6b01f | 1801 | struct dentry *found = NULL; |
665a7583 | 1802 | struct dentry *dentry; |
1da177e4 | 1803 | |
31e6b01f NP |
1804 | /* |
1805 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1806 | * required to prevent single threaded performance regressions | |
1807 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1808 | * Keep the two functions in sync. | |
1809 | */ | |
1810 | ||
b04f784e NP |
1811 | /* |
1812 | * The hash list is protected using RCU. | |
1813 | * | |
1814 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1815 | * with d_move(). | |
1816 | * | |
1817 | * It is possible that concurrent renames can mess up our list | |
1818 | * walk here and result in missing our dentry, resulting in the | |
1819 | * false-negative result. d_lookup() protects against concurrent | |
1820 | * renames using rename_lock seqlock. | |
1821 | * | |
b0a4bb83 | 1822 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1823 | */ |
1da177e4 LT |
1824 | rcu_read_lock(); |
1825 | ||
b07ad996 | 1826 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
31e6b01f NP |
1827 | const char *tname; |
1828 | int tlen; | |
1da177e4 | 1829 | |
1da177e4 LT |
1830 | if (dentry->d_name.hash != hash) |
1831 | continue; | |
1da177e4 LT |
1832 | |
1833 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1834 | if (dentry->d_parent != parent) |
1835 | goto next; | |
d0185c08 LT |
1836 | if (d_unhashed(dentry)) |
1837 | goto next; | |
1838 | ||
1da177e4 LT |
1839 | /* |
1840 | * It is safe to compare names since d_move() cannot | |
1841 | * change the qstr (protected by d_lock). | |
1842 | */ | |
31e6b01f NP |
1843 | tlen = dentry->d_name.len; |
1844 | tname = dentry->d_name.name; | |
fb045adb | 1845 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
621e155a NP |
1846 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1847 | dentry, dentry->d_inode, | |
31e6b01f | 1848 | tlen, tname, name)) |
1da177e4 LT |
1849 | goto next; |
1850 | } else { | |
9d55c369 | 1851 | if (dentry_cmp(tname, tlen, str, len)) |
1da177e4 LT |
1852 | goto next; |
1853 | } | |
1854 | ||
b7ab39f6 | 1855 | dentry->d_count++; |
d0185c08 | 1856 | found = dentry; |
1da177e4 LT |
1857 | spin_unlock(&dentry->d_lock); |
1858 | break; | |
1859 | next: | |
1860 | spin_unlock(&dentry->d_lock); | |
1861 | } | |
1862 | rcu_read_unlock(); | |
1863 | ||
1864 | return found; | |
1865 | } | |
1866 | ||
3e7e241f EB |
1867 | /** |
1868 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1869 | * @dir: Directory to search in | |
1870 | * @name: qstr of name we wish to find | |
1871 | * | |
1872 | * On hash failure or on lookup failure NULL is returned. | |
1873 | */ | |
1874 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1875 | { | |
1876 | struct dentry *dentry = NULL; | |
1877 | ||
1878 | /* | |
1879 | * Check for a fs-specific hash function. Note that we must | |
1880 | * calculate the standard hash first, as the d_op->d_hash() | |
1881 | * routine may choose to leave the hash value unchanged. | |
1882 | */ | |
1883 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1884 | if (dir->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 1885 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
1886 | goto out; |
1887 | } | |
1888 | dentry = d_lookup(dir, name); | |
1889 | out: | |
1890 | return dentry; | |
1891 | } | |
1892 | ||
1da177e4 | 1893 | /** |
786a5e15 | 1894 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 1895 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 1896 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
1897 | * |
1898 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
1899 | * This is used by ncpfs in its readdir implementation. | |
1900 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
1901 | * |
1902 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 1903 | */ |
d3a23e16 | 1904 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 1905 | { |
786a5e15 | 1906 | struct dentry *child; |
d3a23e16 | 1907 | |
2fd6b7f5 | 1908 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
1909 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
1910 | if (dentry == child) { | |
2fd6b7f5 | 1911 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 1912 | __dget_dlock(dentry); |
2fd6b7f5 NP |
1913 | spin_unlock(&dentry->d_lock); |
1914 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
1915 | return 1; |
1916 | } | |
1917 | } | |
2fd6b7f5 | 1918 | spin_unlock(&dparent->d_lock); |
786a5e15 | 1919 | |
1da177e4 LT |
1920 | return 0; |
1921 | } | |
ec4f8605 | 1922 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
1923 | |
1924 | /* | |
1925 | * When a file is deleted, we have two options: | |
1926 | * - turn this dentry into a negative dentry | |
1927 | * - unhash this dentry and free it. | |
1928 | * | |
1929 | * Usually, we want to just turn this into | |
1930 | * a negative dentry, but if anybody else is | |
1931 | * currently using the dentry or the inode | |
1932 | * we can't do that and we fall back on removing | |
1933 | * it from the hash queues and waiting for | |
1934 | * it to be deleted later when it has no users | |
1935 | */ | |
1936 | ||
1937 | /** | |
1938 | * d_delete - delete a dentry | |
1939 | * @dentry: The dentry to delete | |
1940 | * | |
1941 | * Turn the dentry into a negative dentry if possible, otherwise | |
1942 | * remove it from the hash queues so it can be deleted later | |
1943 | */ | |
1944 | ||
1945 | void d_delete(struct dentry * dentry) | |
1946 | { | |
873feea0 | 1947 | struct inode *inode; |
7a91bf7f | 1948 | int isdir = 0; |
1da177e4 LT |
1949 | /* |
1950 | * Are we the only user? | |
1951 | */ | |
357f8e65 | 1952 | again: |
1da177e4 | 1953 | spin_lock(&dentry->d_lock); |
873feea0 NP |
1954 | inode = dentry->d_inode; |
1955 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 1956 | if (dentry->d_count == 1) { |
873feea0 | 1957 | if (inode && !spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
1958 | spin_unlock(&dentry->d_lock); |
1959 | cpu_relax(); | |
1960 | goto again; | |
1961 | } | |
13e3c5e5 | 1962 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 1963 | dentry_unlink_inode(dentry); |
7a91bf7f | 1964 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
1965 | return; |
1966 | } | |
1967 | ||
1968 | if (!d_unhashed(dentry)) | |
1969 | __d_drop(dentry); | |
1970 | ||
1971 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
1972 | |
1973 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 1974 | } |
ec4f8605 | 1975 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 1976 | |
b07ad996 | 1977 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 1978 | { |
ceb5bdc2 | 1979 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 1980 | hlist_bl_lock(b); |
dea3667b | 1981 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 1982 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 1983 | hlist_bl_unlock(b); |
1da177e4 LT |
1984 | } |
1985 | ||
770bfad8 DH |
1986 | static void _d_rehash(struct dentry * entry) |
1987 | { | |
1988 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
1989 | } | |
1990 | ||
1da177e4 LT |
1991 | /** |
1992 | * d_rehash - add an entry back to the hash | |
1993 | * @entry: dentry to add to the hash | |
1994 | * | |
1995 | * Adds a dentry to the hash according to its name. | |
1996 | */ | |
1997 | ||
1998 | void d_rehash(struct dentry * entry) | |
1999 | { | |
1da177e4 | 2000 | spin_lock(&entry->d_lock); |
770bfad8 | 2001 | _d_rehash(entry); |
1da177e4 | 2002 | spin_unlock(&entry->d_lock); |
1da177e4 | 2003 | } |
ec4f8605 | 2004 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2005 | |
fb2d5b86 NP |
2006 | /** |
2007 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2008 | * @dentry: dentry to be updated | |
2009 | * @name: new name | |
2010 | * | |
2011 | * Update a case insensitive dentry with new case of name. | |
2012 | * | |
2013 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2014 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2015 | * lengths). | |
2016 | * | |
2017 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2018 | * keep renames and concurrent inserts, and readdir(2) away). | |
2019 | */ | |
2020 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2021 | { | |
7ebfa57f | 2022 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2023 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2024 | ||
fb2d5b86 | 2025 | spin_lock(&dentry->d_lock); |
31e6b01f | 2026 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2027 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2028 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2029 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2030 | } |
2031 | EXPORT_SYMBOL(dentry_update_name_case); | |
2032 | ||
1da177e4 LT |
2033 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2034 | { | |
2035 | if (dname_external(target)) { | |
2036 | if (dname_external(dentry)) { | |
2037 | /* | |
2038 | * Both external: swap the pointers | |
2039 | */ | |
9a8d5bb4 | 2040 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2041 | } else { |
2042 | /* | |
2043 | * dentry:internal, target:external. Steal target's | |
2044 | * storage and make target internal. | |
2045 | */ | |
321bcf92 BF |
2046 | memcpy(target->d_iname, dentry->d_name.name, |
2047 | dentry->d_name.len + 1); | |
1da177e4 LT |
2048 | dentry->d_name.name = target->d_name.name; |
2049 | target->d_name.name = target->d_iname; | |
2050 | } | |
2051 | } else { | |
2052 | if (dname_external(dentry)) { | |
2053 | /* | |
2054 | * dentry:external, target:internal. Give dentry's | |
2055 | * storage to target and make dentry internal | |
2056 | */ | |
2057 | memcpy(dentry->d_iname, target->d_name.name, | |
2058 | target->d_name.len + 1); | |
2059 | target->d_name.name = dentry->d_name.name; | |
2060 | dentry->d_name.name = dentry->d_iname; | |
2061 | } else { | |
2062 | /* | |
2063 | * Both are internal. Just copy target to dentry | |
2064 | */ | |
2065 | memcpy(dentry->d_iname, target->d_name.name, | |
2066 | target->d_name.len + 1); | |
dc711ca3 AV |
2067 | dentry->d_name.len = target->d_name.len; |
2068 | return; | |
1da177e4 LT |
2069 | } |
2070 | } | |
9a8d5bb4 | 2071 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2072 | } |
2073 | ||
2fd6b7f5 NP |
2074 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2075 | { | |
2076 | /* | |
2077 | * XXXX: do we really need to take target->d_lock? | |
2078 | */ | |
2079 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2080 | spin_lock(&target->d_parent->d_lock); | |
2081 | else { | |
2082 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2083 | spin_lock(&dentry->d_parent->d_lock); | |
2084 | spin_lock_nested(&target->d_parent->d_lock, | |
2085 | DENTRY_D_LOCK_NESTED); | |
2086 | } else { | |
2087 | spin_lock(&target->d_parent->d_lock); | |
2088 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2089 | DENTRY_D_LOCK_NESTED); | |
2090 | } | |
2091 | } | |
2092 | if (target < dentry) { | |
2093 | spin_lock_nested(&target->d_lock, 2); | |
2094 | spin_lock_nested(&dentry->d_lock, 3); | |
2095 | } else { | |
2096 | spin_lock_nested(&dentry->d_lock, 2); | |
2097 | spin_lock_nested(&target->d_lock, 3); | |
2098 | } | |
2099 | } | |
2100 | ||
2101 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2102 | struct dentry *target) | |
2103 | { | |
2104 | if (target->d_parent != dentry->d_parent) | |
2105 | spin_unlock(&dentry->d_parent->d_lock); | |
2106 | if (target->d_parent != target) | |
2107 | spin_unlock(&target->d_parent->d_lock); | |
2108 | } | |
2109 | ||
1da177e4 | 2110 | /* |
2fd6b7f5 NP |
2111 | * When switching names, the actual string doesn't strictly have to |
2112 | * be preserved in the target - because we're dropping the target | |
2113 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2114 | * the new name before we switch. | |
2115 | * | |
2116 | * Note that we have to be a lot more careful about getting the hash | |
2117 | * switched - we have to switch the hash value properly even if it | |
2118 | * then no longer matches the actual (corrupted) string of the target. | |
2119 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2120 | */ |
9eaef27b | 2121 | /* |
18367501 | 2122 | * __d_move - move a dentry |
1da177e4 LT |
2123 | * @dentry: entry to move |
2124 | * @target: new dentry | |
2125 | * | |
2126 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2127 | * dcache entries should not be moved in this way. Caller must hold |
2128 | * rename_lock, the i_mutex of the source and target directories, | |
2129 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2130 | */ |
18367501 | 2131 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2132 | { |
1da177e4 LT |
2133 | if (!dentry->d_inode) |
2134 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2135 | ||
2fd6b7f5 NP |
2136 | BUG_ON(d_ancestor(dentry, target)); |
2137 | BUG_ON(d_ancestor(target, dentry)); | |
2138 | ||
2fd6b7f5 | 2139 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2140 | |
31e6b01f NP |
2141 | write_seqcount_begin(&dentry->d_seq); |
2142 | write_seqcount_begin(&target->d_seq); | |
2143 | ||
ceb5bdc2 NP |
2144 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2145 | ||
2146 | /* | |
2147 | * Move the dentry to the target hash queue. Don't bother checking | |
2148 | * for the same hash queue because of how unlikely it is. | |
2149 | */ | |
2150 | __d_drop(dentry); | |
789680d1 | 2151 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2152 | |
2153 | /* Unhash the target: dput() will then get rid of it */ | |
2154 | __d_drop(target); | |
2155 | ||
5160ee6f ED |
2156 | list_del(&dentry->d_u.d_child); |
2157 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2158 | |
2159 | /* Switch the names.. */ | |
2160 | switch_names(dentry, target); | |
9a8d5bb4 | 2161 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2162 | |
2163 | /* ... and switch the parents */ | |
2164 | if (IS_ROOT(dentry)) { | |
2165 | dentry->d_parent = target->d_parent; | |
2166 | target->d_parent = target; | |
5160ee6f | 2167 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2168 | } else { |
9a8d5bb4 | 2169 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2170 | |
2171 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2172 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2173 | } |
2174 | ||
5160ee6f | 2175 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2176 | |
31e6b01f NP |
2177 | write_seqcount_end(&target->d_seq); |
2178 | write_seqcount_end(&dentry->d_seq); | |
2179 | ||
2fd6b7f5 | 2180 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2181 | spin_unlock(&target->d_lock); |
c32ccd87 | 2182 | fsnotify_d_move(dentry); |
1da177e4 | 2183 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2184 | } |
2185 | ||
2186 | /* | |
2187 | * d_move - move a dentry | |
2188 | * @dentry: entry to move | |
2189 | * @target: new dentry | |
2190 | * | |
2191 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2192 | * dcache entries should not be moved in this way. See the locking |
2193 | * requirements for __d_move. | |
18367501 AV |
2194 | */ |
2195 | void d_move(struct dentry *dentry, struct dentry *target) | |
2196 | { | |
2197 | write_seqlock(&rename_lock); | |
2198 | __d_move(dentry, target); | |
1da177e4 | 2199 | write_sequnlock(&rename_lock); |
9eaef27b | 2200 | } |
ec4f8605 | 2201 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2202 | |
e2761a11 OH |
2203 | /** |
2204 | * d_ancestor - search for an ancestor | |
2205 | * @p1: ancestor dentry | |
2206 | * @p2: child dentry | |
2207 | * | |
2208 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2209 | * an ancestor of p2, else NULL. | |
9eaef27b | 2210 | */ |
e2761a11 | 2211 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2212 | { |
2213 | struct dentry *p; | |
2214 | ||
871c0067 | 2215 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2216 | if (p->d_parent == p1) |
e2761a11 | 2217 | return p; |
9eaef27b | 2218 | } |
e2761a11 | 2219 | return NULL; |
9eaef27b TM |
2220 | } |
2221 | ||
2222 | /* | |
2223 | * This helper attempts to cope with remotely renamed directories | |
2224 | * | |
2225 | * It assumes that the caller is already holding | |
18367501 | 2226 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2227 | * |
2228 | * Note: If ever the locking in lock_rename() changes, then please | |
2229 | * remember to update this too... | |
9eaef27b | 2230 | */ |
873feea0 NP |
2231 | static struct dentry *__d_unalias(struct inode *inode, |
2232 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2233 | { |
2234 | struct mutex *m1 = NULL, *m2 = NULL; | |
2235 | struct dentry *ret; | |
2236 | ||
2237 | /* If alias and dentry share a parent, then no extra locks required */ | |
2238 | if (alias->d_parent == dentry->d_parent) | |
2239 | goto out_unalias; | |
2240 | ||
9eaef27b TM |
2241 | /* See lock_rename() */ |
2242 | ret = ERR_PTR(-EBUSY); | |
2243 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2244 | goto out_err; | |
2245 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2246 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2247 | goto out_err; | |
2248 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2249 | out_unalias: | |
18367501 | 2250 | __d_move(alias, dentry); |
9eaef27b TM |
2251 | ret = alias; |
2252 | out_err: | |
873feea0 | 2253 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2254 | if (m2) |
2255 | mutex_unlock(m2); | |
2256 | if (m1) | |
2257 | mutex_unlock(m1); | |
2258 | return ret; | |
2259 | } | |
2260 | ||
770bfad8 DH |
2261 | /* |
2262 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2263 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2264 | * returns with anon->d_lock held! |
770bfad8 DH |
2265 | */ |
2266 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2267 | { | |
2268 | struct dentry *dparent, *aparent; | |
2269 | ||
2fd6b7f5 | 2270 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2271 | |
31e6b01f NP |
2272 | write_seqcount_begin(&dentry->d_seq); |
2273 | write_seqcount_begin(&anon->d_seq); | |
2274 | ||
770bfad8 DH |
2275 | dparent = dentry->d_parent; |
2276 | aparent = anon->d_parent; | |
2277 | ||
2fd6b7f5 NP |
2278 | switch_names(dentry, anon); |
2279 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2280 | ||
770bfad8 DH |
2281 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2282 | list_del(&dentry->d_u.d_child); | |
2283 | if (!IS_ROOT(dentry)) | |
2284 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2285 | else | |
2286 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2287 | ||
2288 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2289 | list_del(&anon->d_u.d_child); | |
2290 | if (!IS_ROOT(anon)) | |
2291 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2292 | else | |
2293 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2294 | ||
31e6b01f NP |
2295 | write_seqcount_end(&dentry->d_seq); |
2296 | write_seqcount_end(&anon->d_seq); | |
2297 | ||
2fd6b7f5 NP |
2298 | dentry_unlock_parents_for_move(anon, dentry); |
2299 | spin_unlock(&dentry->d_lock); | |
2300 | ||
2301 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2302 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2303 | } | |
2304 | ||
2305 | /** | |
2306 | * d_materialise_unique - introduce an inode into the tree | |
2307 | * @dentry: candidate dentry | |
2308 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2309 | * | |
2310 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2311 | * root directory alias in its place if there is one. Caller must hold the |
2312 | * i_mutex of the parent directory. | |
770bfad8 DH |
2313 | */ |
2314 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2315 | { | |
9eaef27b | 2316 | struct dentry *actual; |
770bfad8 DH |
2317 | |
2318 | BUG_ON(!d_unhashed(dentry)); | |
2319 | ||
770bfad8 DH |
2320 | if (!inode) { |
2321 | actual = dentry; | |
360da900 | 2322 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2323 | d_rehash(actual); |
2324 | goto out_nolock; | |
770bfad8 DH |
2325 | } |
2326 | ||
873feea0 | 2327 | spin_lock(&inode->i_lock); |
357f8e65 | 2328 | |
9eaef27b TM |
2329 | if (S_ISDIR(inode->i_mode)) { |
2330 | struct dentry *alias; | |
2331 | ||
2332 | /* Does an aliased dentry already exist? */ | |
2333 | alias = __d_find_alias(inode, 0); | |
2334 | if (alias) { | |
2335 | actual = alias; | |
18367501 AV |
2336 | write_seqlock(&rename_lock); |
2337 | ||
2338 | if (d_ancestor(alias, dentry)) { | |
2339 | /* Check for loops */ | |
2340 | actual = ERR_PTR(-ELOOP); | |
2341 | } else if (IS_ROOT(alias)) { | |
2342 | /* Is this an anonymous mountpoint that we | |
2343 | * could splice into our tree? */ | |
9eaef27b | 2344 | __d_materialise_dentry(dentry, alias); |
18367501 | 2345 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2346 | __d_drop(alias); |
2347 | goto found; | |
18367501 AV |
2348 | } else { |
2349 | /* Nope, but we must(!) avoid directory | |
2350 | * aliasing */ | |
2351 | actual = __d_unalias(inode, dentry, alias); | |
9eaef27b | 2352 | } |
18367501 | 2353 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2354 | if (IS_ERR(actual)) |
2355 | dput(alias); | |
2356 | goto out_nolock; | |
2357 | } | |
770bfad8 DH |
2358 | } |
2359 | ||
2360 | /* Add a unique reference */ | |
2361 | actual = __d_instantiate_unique(dentry, inode); | |
2362 | if (!actual) | |
2363 | actual = dentry; | |
357f8e65 NP |
2364 | else |
2365 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2366 | |
770bfad8 DH |
2367 | spin_lock(&actual->d_lock); |
2368 | found: | |
2369 | _d_rehash(actual); | |
2370 | spin_unlock(&actual->d_lock); | |
873feea0 | 2371 | spin_unlock(&inode->i_lock); |
9eaef27b | 2372 | out_nolock: |
770bfad8 DH |
2373 | if (actual == dentry) { |
2374 | security_d_instantiate(dentry, inode); | |
2375 | return NULL; | |
2376 | } | |
2377 | ||
2378 | iput(inode); | |
2379 | return actual; | |
770bfad8 | 2380 | } |
ec4f8605 | 2381 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2382 | |
cdd16d02 | 2383 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2384 | { |
2385 | *buflen -= namelen; | |
2386 | if (*buflen < 0) | |
2387 | return -ENAMETOOLONG; | |
2388 | *buffer -= namelen; | |
2389 | memcpy(*buffer, str, namelen); | |
2390 | return 0; | |
2391 | } | |
2392 | ||
cdd16d02 MS |
2393 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2394 | { | |
2395 | return prepend(buffer, buflen, name->name, name->len); | |
2396 | } | |
2397 | ||
1da177e4 | 2398 | /** |
208898c1 | 2399 | * prepend_path - Prepend path string to a buffer |
9d1bc601 MS |
2400 | * @path: the dentry/vfsmount to report |
2401 | * @root: root vfsmnt/dentry (may be modified by this function) | |
f2eb6575 MS |
2402 | * @buffer: pointer to the end of the buffer |
2403 | * @buflen: pointer to buffer length | |
552ce544 | 2404 | * |
949854d0 | 2405 | * Caller holds the rename_lock. |
9d1bc601 MS |
2406 | * |
2407 | * If path is not reachable from the supplied root, then the value of | |
2408 | * root is changed (without modifying refcounts). | |
1da177e4 | 2409 | */ |
f2eb6575 MS |
2410 | static int prepend_path(const struct path *path, struct path *root, |
2411 | char **buffer, int *buflen) | |
1da177e4 | 2412 | { |
9d1bc601 MS |
2413 | struct dentry *dentry = path->dentry; |
2414 | struct vfsmount *vfsmnt = path->mnt; | |
f2eb6575 MS |
2415 | bool slash = false; |
2416 | int error = 0; | |
6092d048 | 2417 | |
99b7db7b | 2418 | br_read_lock(vfsmount_lock); |
f2eb6575 | 2419 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2420 | struct dentry * parent; |
2421 | ||
1da177e4 | 2422 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2423 | /* Global root? */ |
1da177e4 | 2424 | if (vfsmnt->mnt_parent == vfsmnt) { |
1da177e4 LT |
2425 | goto global_root; |
2426 | } | |
2427 | dentry = vfsmnt->mnt_mountpoint; | |
2428 | vfsmnt = vfsmnt->mnt_parent; | |
1da177e4 LT |
2429 | continue; |
2430 | } | |
2431 | parent = dentry->d_parent; | |
2432 | prefetch(parent); | |
9abca360 | 2433 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2434 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2435 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2436 | if (!error) |
2437 | error = prepend(buffer, buflen, "/", 1); | |
2438 | if (error) | |
2439 | break; | |
2440 | ||
2441 | slash = true; | |
1da177e4 LT |
2442 | dentry = parent; |
2443 | } | |
2444 | ||
be285c71 | 2445 | out: |
f2eb6575 MS |
2446 | if (!error && !slash) |
2447 | error = prepend(buffer, buflen, "/", 1); | |
2448 | ||
99b7db7b | 2449 | br_read_unlock(vfsmount_lock); |
f2eb6575 | 2450 | return error; |
1da177e4 LT |
2451 | |
2452 | global_root: | |
98dc568b MS |
2453 | /* |
2454 | * Filesystems needing to implement special "root names" | |
2455 | * should do so with ->d_dname() | |
2456 | */ | |
2457 | if (IS_ROOT(dentry) && | |
2458 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2459 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2460 | (int) dentry->d_name.len, dentry->d_name.name); | |
2461 | } | |
9d1bc601 MS |
2462 | root->mnt = vfsmnt; |
2463 | root->dentry = dentry; | |
be285c71 | 2464 | goto out; |
f2eb6575 | 2465 | } |
be285c71 | 2466 | |
f2eb6575 MS |
2467 | /** |
2468 | * __d_path - return the path of a dentry | |
2469 | * @path: the dentry/vfsmount to report | |
2470 | * @root: root vfsmnt/dentry (may be modified by this function) | |
cd956a1c | 2471 | * @buf: buffer to return value in |
f2eb6575 MS |
2472 | * @buflen: buffer length |
2473 | * | |
ffd1f4ed | 2474 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2475 | * |
2476 | * Returns a pointer into the buffer or an error code if the | |
2477 | * path was too long. | |
2478 | * | |
be148247 | 2479 | * "buflen" should be positive. |
f2eb6575 MS |
2480 | * |
2481 | * If path is not reachable from the supplied root, then the value of | |
2482 | * root is changed (without modifying refcounts). | |
2483 | */ | |
2484 | char *__d_path(const struct path *path, struct path *root, | |
2485 | char *buf, int buflen) | |
2486 | { | |
2487 | char *res = buf + buflen; | |
2488 | int error; | |
2489 | ||
2490 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2491 | write_seqlock(&rename_lock); |
f2eb6575 | 2492 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2493 | write_sequnlock(&rename_lock); |
be148247 | 2494 | |
f2eb6575 MS |
2495 | if (error) |
2496 | return ERR_PTR(error); | |
f2eb6575 | 2497 | return res; |
1da177e4 LT |
2498 | } |
2499 | ||
ffd1f4ed MS |
2500 | /* |
2501 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2502 | */ | |
2503 | static int path_with_deleted(const struct path *path, struct path *root, | |
2504 | char **buf, int *buflen) | |
2505 | { | |
2506 | prepend(buf, buflen, "\0", 1); | |
2507 | if (d_unlinked(path->dentry)) { | |
2508 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2509 | if (error) | |
2510 | return error; | |
2511 | } | |
2512 | ||
2513 | return prepend_path(path, root, buf, buflen); | |
2514 | } | |
2515 | ||
8df9d1a4 MS |
2516 | static int prepend_unreachable(char **buffer, int *buflen) |
2517 | { | |
2518 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2519 | } | |
2520 | ||
a03a8a70 JB |
2521 | /** |
2522 | * d_path - return the path of a dentry | |
cf28b486 | 2523 | * @path: path to report |
a03a8a70 JB |
2524 | * @buf: buffer to return value in |
2525 | * @buflen: buffer length | |
2526 | * | |
2527 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2528 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2529 | * | |
52afeefb AV |
2530 | * Returns a pointer into the buffer or an error code if the path was |
2531 | * too long. Note: Callers should use the returned pointer, not the passed | |
2532 | * in buffer, to use the name! The implementation often starts at an offset | |
2533 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2534 | * |
31f3e0b3 | 2535 | * "buflen" should be positive. |
a03a8a70 | 2536 | */ |
20d4fdc1 | 2537 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2538 | { |
ffd1f4ed | 2539 | char *res = buf + buflen; |
6ac08c39 | 2540 | struct path root; |
9d1bc601 | 2541 | struct path tmp; |
ffd1f4ed | 2542 | int error; |
1da177e4 | 2543 | |
c23fbb6b ED |
2544 | /* |
2545 | * We have various synthetic filesystems that never get mounted. On | |
2546 | * these filesystems dentries are never used for lookup purposes, and | |
2547 | * thus don't need to be hashed. They also don't need a name until a | |
2548 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2549 | * below allows us to generate a name for these objects on demand: | |
2550 | */ | |
cf28b486 JB |
2551 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2552 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2553 | |
f7ad3c6b | 2554 | get_fs_root(current->fs, &root); |
949854d0 | 2555 | write_seqlock(&rename_lock); |
9d1bc601 | 2556 | tmp = root; |
ffd1f4ed MS |
2557 | error = path_with_deleted(path, &tmp, &res, &buflen); |
2558 | if (error) | |
2559 | res = ERR_PTR(error); | |
949854d0 | 2560 | write_sequnlock(&rename_lock); |
6ac08c39 | 2561 | path_put(&root); |
1da177e4 LT |
2562 | return res; |
2563 | } | |
ec4f8605 | 2564 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2565 | |
8df9d1a4 MS |
2566 | /** |
2567 | * d_path_with_unreachable - return the path of a dentry | |
2568 | * @path: path to report | |
2569 | * @buf: buffer to return value in | |
2570 | * @buflen: buffer length | |
2571 | * | |
2572 | * The difference from d_path() is that this prepends "(unreachable)" | |
2573 | * to paths which are unreachable from the current process' root. | |
2574 | */ | |
2575 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2576 | { | |
2577 | char *res = buf + buflen; | |
2578 | struct path root; | |
2579 | struct path tmp; | |
2580 | int error; | |
2581 | ||
2582 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2583 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2584 | ||
2585 | get_fs_root(current->fs, &root); | |
949854d0 | 2586 | write_seqlock(&rename_lock); |
8df9d1a4 MS |
2587 | tmp = root; |
2588 | error = path_with_deleted(path, &tmp, &res, &buflen); | |
2589 | if (!error && !path_equal(&tmp, &root)) | |
2590 | error = prepend_unreachable(&res, &buflen); | |
949854d0 | 2591 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2592 | path_put(&root); |
2593 | if (error) | |
2594 | res = ERR_PTR(error); | |
2595 | ||
2596 | return res; | |
2597 | } | |
2598 | ||
c23fbb6b ED |
2599 | /* |
2600 | * Helper function for dentry_operations.d_dname() members | |
2601 | */ | |
2602 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2603 | const char *fmt, ...) | |
2604 | { | |
2605 | va_list args; | |
2606 | char temp[64]; | |
2607 | int sz; | |
2608 | ||
2609 | va_start(args, fmt); | |
2610 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2611 | va_end(args); | |
2612 | ||
2613 | if (sz > sizeof(temp) || sz > buflen) | |
2614 | return ERR_PTR(-ENAMETOOLONG); | |
2615 | ||
2616 | buffer += buflen - sz; | |
2617 | return memcpy(buffer, temp, sz); | |
2618 | } | |
2619 | ||
6092d048 RP |
2620 | /* |
2621 | * Write full pathname from the root of the filesystem into the buffer. | |
2622 | */ | |
ec2447c2 | 2623 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2624 | { |
2625 | char *end = buf + buflen; | |
2626 | char *retval; | |
2627 | ||
6092d048 | 2628 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2629 | if (buflen < 1) |
2630 | goto Elong; | |
2631 | /* Get '/' right */ | |
2632 | retval = end-1; | |
2633 | *retval = '/'; | |
2634 | ||
cdd16d02 MS |
2635 | while (!IS_ROOT(dentry)) { |
2636 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2637 | int error; |
6092d048 | 2638 | |
6092d048 | 2639 | prefetch(parent); |
9abca360 NP |
2640 | spin_lock(&dentry->d_lock); |
2641 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2642 | spin_unlock(&dentry->d_lock); | |
2643 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2644 | goto Elong; |
2645 | ||
2646 | retval = end; | |
2647 | dentry = parent; | |
2648 | } | |
c103135c AV |
2649 | return retval; |
2650 | Elong: | |
2651 | return ERR_PTR(-ENAMETOOLONG); | |
2652 | } | |
ec2447c2 NP |
2653 | |
2654 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2655 | { | |
2656 | char *retval; | |
2657 | ||
949854d0 | 2658 | write_seqlock(&rename_lock); |
ec2447c2 | 2659 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2660 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2661 | |
2662 | return retval; | |
2663 | } | |
2664 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2665 | |
2666 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2667 | { | |
2668 | char *p = NULL; | |
2669 | char *retval; | |
2670 | ||
949854d0 | 2671 | write_seqlock(&rename_lock); |
c103135c AV |
2672 | if (d_unlinked(dentry)) { |
2673 | p = buf + buflen; | |
2674 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2675 | goto Elong; | |
2676 | buflen++; | |
2677 | } | |
2678 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2679 | write_sequnlock(&rename_lock); |
c103135c AV |
2680 | if (!IS_ERR(retval) && p) |
2681 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2682 | return retval; |
2683 | Elong: | |
6092d048 RP |
2684 | return ERR_PTR(-ENAMETOOLONG); |
2685 | } | |
2686 | ||
1da177e4 LT |
2687 | /* |
2688 | * NOTE! The user-level library version returns a | |
2689 | * character pointer. The kernel system call just | |
2690 | * returns the length of the buffer filled (which | |
2691 | * includes the ending '\0' character), or a negative | |
2692 | * error value. So libc would do something like | |
2693 | * | |
2694 | * char *getcwd(char * buf, size_t size) | |
2695 | * { | |
2696 | * int retval; | |
2697 | * | |
2698 | * retval = sys_getcwd(buf, size); | |
2699 | * if (retval >= 0) | |
2700 | * return buf; | |
2701 | * errno = -retval; | |
2702 | * return NULL; | |
2703 | * } | |
2704 | */ | |
3cdad428 | 2705 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2706 | { |
552ce544 | 2707 | int error; |
6ac08c39 | 2708 | struct path pwd, root; |
552ce544 | 2709 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2710 | |
2711 | if (!page) | |
2712 | return -ENOMEM; | |
2713 | ||
f7ad3c6b | 2714 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2715 | |
552ce544 | 2716 | error = -ENOENT; |
949854d0 | 2717 | write_seqlock(&rename_lock); |
f3da392e | 2718 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2719 | unsigned long len; |
9d1bc601 | 2720 | struct path tmp = root; |
8df9d1a4 MS |
2721 | char *cwd = page + PAGE_SIZE; |
2722 | int buflen = PAGE_SIZE; | |
1da177e4 | 2723 | |
8df9d1a4 MS |
2724 | prepend(&cwd, &buflen, "\0", 1); |
2725 | error = prepend_path(&pwd, &tmp, &cwd, &buflen); | |
949854d0 | 2726 | write_sequnlock(&rename_lock); |
552ce544 | 2727 | |
8df9d1a4 | 2728 | if (error) |
552ce544 LT |
2729 | goto out; |
2730 | ||
8df9d1a4 MS |
2731 | /* Unreachable from current root */ |
2732 | if (!path_equal(&tmp, &root)) { | |
2733 | error = prepend_unreachable(&cwd, &buflen); | |
2734 | if (error) | |
2735 | goto out; | |
2736 | } | |
2737 | ||
552ce544 LT |
2738 | error = -ERANGE; |
2739 | len = PAGE_SIZE + page - cwd; | |
2740 | if (len <= size) { | |
2741 | error = len; | |
2742 | if (copy_to_user(buf, cwd, len)) | |
2743 | error = -EFAULT; | |
2744 | } | |
949854d0 NP |
2745 | } else { |
2746 | write_sequnlock(&rename_lock); | |
949854d0 | 2747 | } |
1da177e4 LT |
2748 | |
2749 | out: | |
6ac08c39 JB |
2750 | path_put(&pwd); |
2751 | path_put(&root); | |
1da177e4 LT |
2752 | free_page((unsigned long) page); |
2753 | return error; | |
2754 | } | |
2755 | ||
2756 | /* | |
2757 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2758 | * | |
2759 | * Trivially implemented using the dcache structure | |
2760 | */ | |
2761 | ||
2762 | /** | |
2763 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2764 | * @new_dentry: new dentry | |
2765 | * @old_dentry: old dentry | |
2766 | * | |
2767 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2768 | * Returns 0 otherwise. | |
2769 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2770 | */ | |
2771 | ||
e2761a11 | 2772 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2773 | { |
2774 | int result; | |
949854d0 | 2775 | unsigned seq; |
1da177e4 | 2776 | |
e2761a11 OH |
2777 | if (new_dentry == old_dentry) |
2778 | return 1; | |
2779 | ||
e2761a11 | 2780 | do { |
1da177e4 | 2781 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2782 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2783 | /* |
2784 | * Need rcu_readlock to protect against the d_parent trashing | |
2785 | * due to d_move | |
2786 | */ | |
2787 | rcu_read_lock(); | |
e2761a11 | 2788 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2789 | result = 1; |
e2761a11 OH |
2790 | else |
2791 | result = 0; | |
949854d0 | 2792 | rcu_read_unlock(); |
1da177e4 | 2793 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2794 | |
2795 | return result; | |
2796 | } | |
2797 | ||
2096f759 AV |
2798 | int path_is_under(struct path *path1, struct path *path2) |
2799 | { | |
2800 | struct vfsmount *mnt = path1->mnt; | |
2801 | struct dentry *dentry = path1->dentry; | |
2802 | int res; | |
99b7db7b NP |
2803 | |
2804 | br_read_lock(vfsmount_lock); | |
2096f759 AV |
2805 | if (mnt != path2->mnt) { |
2806 | for (;;) { | |
2807 | if (mnt->mnt_parent == mnt) { | |
99b7db7b | 2808 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2809 | return 0; |
2810 | } | |
2811 | if (mnt->mnt_parent == path2->mnt) | |
2812 | break; | |
2813 | mnt = mnt->mnt_parent; | |
2814 | } | |
2815 | dentry = mnt->mnt_mountpoint; | |
2816 | } | |
2817 | res = is_subdir(dentry, path2->dentry); | |
99b7db7b | 2818 | br_read_unlock(vfsmount_lock); |
2096f759 AV |
2819 | return res; |
2820 | } | |
2821 | EXPORT_SYMBOL(path_is_under); | |
2822 | ||
1da177e4 LT |
2823 | void d_genocide(struct dentry *root) |
2824 | { | |
949854d0 | 2825 | struct dentry *this_parent; |
1da177e4 | 2826 | struct list_head *next; |
949854d0 | 2827 | unsigned seq; |
58db63d0 | 2828 | int locked = 0; |
1da177e4 | 2829 | |
949854d0 | 2830 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2831 | again: |
2832 | this_parent = root; | |
2fd6b7f5 | 2833 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2834 | repeat: |
2835 | next = this_parent->d_subdirs.next; | |
2836 | resume: | |
2837 | while (next != &this_parent->d_subdirs) { | |
2838 | struct list_head *tmp = next; | |
5160ee6f | 2839 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2840 | next = tmp->next; |
949854d0 | 2841 | |
da502956 NP |
2842 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2843 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2844 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2845 | continue; |
da502956 | 2846 | } |
1da177e4 | 2847 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2848 | spin_unlock(&this_parent->d_lock); |
2849 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2850 | this_parent = dentry; |
2fd6b7f5 | 2851 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2852 | goto repeat; |
2853 | } | |
949854d0 NP |
2854 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2855 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2856 | dentry->d_count--; | |
2857 | } | |
b7ab39f6 | 2858 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2859 | } |
2860 | if (this_parent != root) { | |
c826cb7d | 2861 | struct dentry *child = this_parent; |
949854d0 NP |
2862 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2863 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2864 | this_parent->d_count--; | |
2865 | } | |
c826cb7d LT |
2866 | this_parent = try_to_ascend(this_parent, locked, seq); |
2867 | if (!this_parent) | |
949854d0 | 2868 | goto rename_retry; |
949854d0 | 2869 | next = child->d_u.d_child.next; |
1da177e4 LT |
2870 | goto resume; |
2871 | } | |
2fd6b7f5 | 2872 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2873 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2874 | goto rename_retry; |
58db63d0 NP |
2875 | if (locked) |
2876 | write_sequnlock(&rename_lock); | |
2877 | return; | |
2878 | ||
2879 | rename_retry: | |
2880 | locked = 1; | |
2881 | write_seqlock(&rename_lock); | |
2882 | goto again; | |
1da177e4 LT |
2883 | } |
2884 | ||
2885 | /** | |
2886 | * find_inode_number - check for dentry with name | |
2887 | * @dir: directory to check | |
2888 | * @name: Name to find. | |
2889 | * | |
2890 | * Check whether a dentry already exists for the given name, | |
2891 | * and return the inode number if it has an inode. Otherwise | |
2892 | * 0 is returned. | |
2893 | * | |
2894 | * This routine is used to post-process directory listings for | |
2895 | * filesystems using synthetic inode numbers, and is necessary | |
2896 | * to keep getcwd() working. | |
2897 | */ | |
2898 | ||
2899 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2900 | { | |
2901 | struct dentry * dentry; | |
2902 | ino_t ino = 0; | |
2903 | ||
3e7e241f EB |
2904 | dentry = d_hash_and_lookup(dir, name); |
2905 | if (dentry) { | |
1da177e4 LT |
2906 | if (dentry->d_inode) |
2907 | ino = dentry->d_inode->i_ino; | |
2908 | dput(dentry); | |
2909 | } | |
1da177e4 LT |
2910 | return ino; |
2911 | } | |
ec4f8605 | 2912 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
2913 | |
2914 | static __initdata unsigned long dhash_entries; | |
2915 | static int __init set_dhash_entries(char *str) | |
2916 | { | |
2917 | if (!str) | |
2918 | return 0; | |
2919 | dhash_entries = simple_strtoul(str, &str, 0); | |
2920 | return 1; | |
2921 | } | |
2922 | __setup("dhash_entries=", set_dhash_entries); | |
2923 | ||
2924 | static void __init dcache_init_early(void) | |
2925 | { | |
2926 | int loop; | |
2927 | ||
2928 | /* If hashes are distributed across NUMA nodes, defer | |
2929 | * hash allocation until vmalloc space is available. | |
2930 | */ | |
2931 | if (hashdist) | |
2932 | return; | |
2933 | ||
2934 | dentry_hashtable = | |
2935 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 2936 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
2937 | dhash_entries, |
2938 | 13, | |
2939 | HASH_EARLY, | |
2940 | &d_hash_shift, | |
2941 | &d_hash_mask, | |
2942 | 0); | |
2943 | ||
2944 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 2945 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
2946 | } |
2947 | ||
74bf17cf | 2948 | static void __init dcache_init(void) |
1da177e4 LT |
2949 | { |
2950 | int loop; | |
2951 | ||
2952 | /* | |
2953 | * A constructor could be added for stable state like the lists, | |
2954 | * but it is probably not worth it because of the cache nature | |
2955 | * of the dcache. | |
2956 | */ | |
0a31bd5f CL |
2957 | dentry_cache = KMEM_CACHE(dentry, |
2958 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
2959 | |
2960 | /* Hash may have been set up in dcache_init_early */ | |
2961 | if (!hashdist) | |
2962 | return; | |
2963 | ||
2964 | dentry_hashtable = | |
2965 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 2966 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
2967 | dhash_entries, |
2968 | 13, | |
2969 | 0, | |
2970 | &d_hash_shift, | |
2971 | &d_hash_mask, | |
2972 | 0); | |
2973 | ||
2974 | for (loop = 0; loop < (1 << d_hash_shift); loop++) | |
b07ad996 | 2975 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
2976 | } |
2977 | ||
2978 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 2979 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 2980 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 2981 | |
1da177e4 LT |
2982 | EXPORT_SYMBOL(d_genocide); |
2983 | ||
1da177e4 LT |
2984 | void __init vfs_caches_init_early(void) |
2985 | { | |
2986 | dcache_init_early(); | |
2987 | inode_init_early(); | |
2988 | } | |
2989 | ||
2990 | void __init vfs_caches_init(unsigned long mempages) | |
2991 | { | |
2992 | unsigned long reserve; | |
2993 | ||
2994 | /* Base hash sizes on available memory, with a reserve equal to | |
2995 | 150% of current kernel size */ | |
2996 | ||
2997 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
2998 | mempages -= reserve; | |
2999 | ||
3000 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3001 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3002 | |
74bf17cf DC |
3003 | dcache_init(); |
3004 | inode_init(); | |
1da177e4 | 3005 | files_init(mempages); |
74bf17cf | 3006 | mnt_init(); |
1da177e4 LT |
3007 | bdev_cache_init(); |
3008 | chrdev_init(); | |
3009 | } |