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> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
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> |
dd179946 | 39 | #include <linux/ratelimit.h> |
07f3f05c | 40 | #include "internal.h" |
b2dba1af | 41 | #include "mount.h" |
1da177e4 | 42 | |
789680d1 NP |
43 | /* |
44 | * Usage: | |
873feea0 NP |
45 | * dcache->d_inode->i_lock protects: |
46 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
47 | * dcache_hash_bucket lock protects: |
48 | * - the dcache hash table | |
49 | * s_anon bl list spinlock protects: | |
50 | * - the s_anon list (see __d_drop) | |
23044507 NP |
51 | * dcache_lru_lock protects: |
52 | * - the dcache lru lists and counters | |
53 | * d_lock protects: | |
54 | * - d_flags | |
55 | * - d_name | |
56 | * - d_lru | |
b7ab39f6 | 57 | * - d_count |
da502956 | 58 | * - d_unhashed() |
2fd6b7f5 NP |
59 | * - d_parent and d_subdirs |
60 | * - childrens' d_child and d_parent | |
b23fb0a6 | 61 | * - d_alias, d_inode |
789680d1 NP |
62 | * |
63 | * Ordering: | |
873feea0 | 64 | * dentry->d_inode->i_lock |
b5c84bf6 NP |
65 | * dentry->d_lock |
66 | * dcache_lru_lock | |
ceb5bdc2 NP |
67 | * dcache_hash_bucket lock |
68 | * s_anon lock | |
789680d1 | 69 | * |
da502956 NP |
70 | * If there is an ancestor relationship: |
71 | * dentry->d_parent->...->d_parent->d_lock | |
72 | * ... | |
73 | * dentry->d_parent->d_lock | |
74 | * dentry->d_lock | |
75 | * | |
76 | * If no ancestor relationship: | |
789680d1 NP |
77 | * if (dentry1 < dentry2) |
78 | * dentry1->d_lock | |
79 | * dentry2->d_lock | |
80 | */ | |
fa3536cc | 81 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
82 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
83 | ||
23044507 | 84 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
74c3cbe3 | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 86 | |
949854d0 | 87 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 88 | |
e18b890b | 89 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 90 | |
1da177e4 LT |
91 | /* |
92 | * This is the single most critical data structure when it comes | |
93 | * to the dcache: the hashtable for lookups. Somebody should try | |
94 | * to make this good - I've just made it work. | |
95 | * | |
96 | * This hash-function tries to avoid losing too many bits of hash | |
97 | * information, yet avoid using a prime hash-size or similar. | |
98 | */ | |
1da177e4 | 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 | |
8966be90 | 105 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 106 | unsigned int hash) |
ceb5bdc2 | 107 | { |
6d7d1a0d | 108 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
d4c96061 | 109 | return dentry_hashtable + hash_32(hash, d_hash_shift); |
ceb5bdc2 NP |
110 | } |
111 | ||
1da177e4 LT |
112 | /* Statistics gathering. */ |
113 | struct dentry_stat_t dentry_stat = { | |
114 | .age_limit = 45, | |
115 | }; | |
116 | ||
3e880fb5 | 117 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
118 | |
119 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
120 | static int get_nr_dentry(void) |
121 | { | |
122 | int i; | |
123 | int sum = 0; | |
124 | for_each_possible_cpu(i) | |
125 | sum += per_cpu(nr_dentry, i); | |
126 | return sum < 0 ? 0 : sum; | |
127 | } | |
128 | ||
312d3ca8 CH |
129 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
130 | size_t *lenp, loff_t *ppos) | |
131 | { | |
3e880fb5 | 132 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
133 | return proc_dointvec(table, write, buffer, lenp, ppos); |
134 | } | |
135 | #endif | |
136 | ||
5483f18e LT |
137 | /* |
138 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
139 | * The strings are both count bytes long, and count is non-zero. | |
140 | */ | |
e419b4cc LT |
141 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
142 | ||
143 | #include <asm/word-at-a-time.h> | |
144 | /* | |
145 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
146 | * aligned allocation for this particular component. We don't | |
147 | * strictly need the load_unaligned_zeropad() safety, but it | |
148 | * doesn't hurt either. | |
149 | * | |
150 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
151 | * need the careful unaligned handling. | |
152 | */ | |
94753db5 | 153 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 154 | { |
bfcfaa77 | 155 | unsigned long a,b,mask; |
bfcfaa77 LT |
156 | |
157 | for (;;) { | |
12f8ad4b | 158 | a = *(unsigned long *)cs; |
e419b4cc | 159 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
160 | if (tcount < sizeof(unsigned long)) |
161 | break; | |
162 | if (unlikely(a != b)) | |
163 | return 1; | |
164 | cs += sizeof(unsigned long); | |
165 | ct += sizeof(unsigned long); | |
166 | tcount -= sizeof(unsigned long); | |
167 | if (!tcount) | |
168 | return 0; | |
169 | } | |
170 | mask = ~(~0ul << tcount*8); | |
171 | return unlikely(!!((a ^ b) & mask)); | |
e419b4cc LT |
172 | } |
173 | ||
bfcfaa77 | 174 | #else |
e419b4cc | 175 | |
94753db5 | 176 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 177 | { |
5483f18e LT |
178 | do { |
179 | if (*cs != *ct) | |
180 | return 1; | |
181 | cs++; | |
182 | ct++; | |
183 | tcount--; | |
184 | } while (tcount); | |
185 | return 0; | |
186 | } | |
187 | ||
e419b4cc LT |
188 | #endif |
189 | ||
94753db5 LT |
190 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
191 | { | |
6326c71f | 192 | const unsigned char *cs; |
94753db5 LT |
193 | /* |
194 | * Be careful about RCU walk racing with rename: | |
195 | * use ACCESS_ONCE to fetch the name pointer. | |
196 | * | |
197 | * NOTE! Even if a rename will mean that the length | |
198 | * was not loaded atomically, we don't care. The | |
199 | * RCU walk will check the sequence count eventually, | |
200 | * and catch it. And we won't overrun the buffer, | |
201 | * because we're reading the name pointer atomically, | |
202 | * and a dentry name is guaranteed to be properly | |
203 | * terminated with a NUL byte. | |
204 | * | |
205 | * End result: even if 'len' is wrong, we'll exit | |
206 | * early because the data cannot match (there can | |
207 | * be no NUL in the ct/tcount data) | |
208 | */ | |
6326c71f LT |
209 | cs = ACCESS_ONCE(dentry->d_name.name); |
210 | smp_read_barrier_depends(); | |
211 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
212 | } |
213 | ||
9c82ab9c | 214 | static void __d_free(struct rcu_head *head) |
1da177e4 | 215 | { |
9c82ab9c CH |
216 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
217 | ||
b3d9b7a3 | 218 | WARN_ON(!hlist_unhashed(&dentry->d_alias)); |
1da177e4 LT |
219 | if (dname_external(dentry)) |
220 | kfree(dentry->d_name.name); | |
221 | kmem_cache_free(dentry_cache, dentry); | |
222 | } | |
223 | ||
224 | /* | |
b5c84bf6 | 225 | * no locks, please. |
1da177e4 LT |
226 | */ |
227 | static void d_free(struct dentry *dentry) | |
228 | { | |
b7ab39f6 | 229 | BUG_ON(dentry->d_count); |
3e880fb5 | 230 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
231 | if (dentry->d_op && dentry->d_op->d_release) |
232 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 233 | |
dea3667b LT |
234 | /* if dentry was never visible to RCU, immediate free is OK */ |
235 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
9c82ab9c | 236 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 237 | else |
9c82ab9c | 238 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
239 | } |
240 | ||
31e6b01f NP |
241 | /** |
242 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 243 | * @dentry: the target dentry |
31e6b01f NP |
244 | * After this call, in-progress rcu-walk path lookup will fail. This |
245 | * should be called after unhashing, and after changing d_inode (if | |
246 | * the dentry has not already been unhashed). | |
247 | */ | |
248 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
249 | { | |
250 | assert_spin_locked(&dentry->d_lock); | |
251 | /* Go through a barrier */ | |
252 | write_seqcount_barrier(&dentry->d_seq); | |
253 | } | |
254 | ||
1da177e4 LT |
255 | /* |
256 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
257 | * d_iput() operation if defined. Dentry has no refcount |
258 | * and is unhashed. | |
1da177e4 | 259 | */ |
858119e1 | 260 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 261 | __releases(dentry->d_lock) |
873feea0 | 262 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
263 | { |
264 | struct inode *inode = dentry->d_inode; | |
265 | if (inode) { | |
266 | dentry->d_inode = NULL; | |
b3d9b7a3 | 267 | hlist_del_init(&dentry->d_alias); |
1da177e4 | 268 | spin_unlock(&dentry->d_lock); |
873feea0 | 269 | spin_unlock(&inode->i_lock); |
f805fbda LT |
270 | if (!inode->i_nlink) |
271 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
272 | if (dentry->d_op && dentry->d_op->d_iput) |
273 | dentry->d_op->d_iput(dentry, inode); | |
274 | else | |
275 | iput(inode); | |
276 | } else { | |
277 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
278 | } |
279 | } | |
280 | ||
31e6b01f NP |
281 | /* |
282 | * Release the dentry's inode, using the filesystem | |
283 | * d_iput() operation if defined. dentry remains in-use. | |
284 | */ | |
285 | static void dentry_unlink_inode(struct dentry * dentry) | |
286 | __releases(dentry->d_lock) | |
873feea0 | 287 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
288 | { |
289 | struct inode *inode = dentry->d_inode; | |
290 | dentry->d_inode = NULL; | |
b3d9b7a3 | 291 | hlist_del_init(&dentry->d_alias); |
31e6b01f NP |
292 | dentry_rcuwalk_barrier(dentry); |
293 | spin_unlock(&dentry->d_lock); | |
873feea0 | 294 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
295 | if (!inode->i_nlink) |
296 | fsnotify_inoderemove(inode); | |
297 | if (dentry->d_op && dentry->d_op->d_iput) | |
298 | dentry->d_op->d_iput(dentry, inode); | |
299 | else | |
300 | iput(inode); | |
301 | } | |
302 | ||
da3bbdd4 | 303 | /* |
f0023bc6 | 304 | * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
305 | */ |
306 | static void dentry_lru_add(struct dentry *dentry) | |
307 | { | |
a4633357 | 308 | if (list_empty(&dentry->d_lru)) { |
23044507 | 309 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
310 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
311 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 312 | dentry_stat.nr_unused++; |
23044507 | 313 | spin_unlock(&dcache_lru_lock); |
a4633357 | 314 | } |
da3bbdd4 KM |
315 | } |
316 | ||
23044507 NP |
317 | static void __dentry_lru_del(struct dentry *dentry) |
318 | { | |
319 | list_del_init(&dentry->d_lru); | |
eaf5f907 | 320 | dentry->d_flags &= ~DCACHE_SHRINK_LIST; |
23044507 NP |
321 | dentry->d_sb->s_nr_dentry_unused--; |
322 | dentry_stat.nr_unused--; | |
323 | } | |
324 | ||
f0023bc6 SW |
325 | /* |
326 | * Remove a dentry with references from the LRU. | |
327 | */ | |
da3bbdd4 KM |
328 | static void dentry_lru_del(struct dentry *dentry) |
329 | { | |
330 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
331 | spin_lock(&dcache_lru_lock); |
332 | __dentry_lru_del(dentry); | |
333 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
334 | } |
335 | } | |
336 | ||
b48f03b3 | 337 | static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list) |
da3bbdd4 | 338 | { |
23044507 | 339 | spin_lock(&dcache_lru_lock); |
a4633357 | 340 | if (list_empty(&dentry->d_lru)) { |
b48f03b3 | 341 | list_add_tail(&dentry->d_lru, list); |
a4633357 | 342 | dentry->d_sb->s_nr_dentry_unused++; |
86c8749e | 343 | dentry_stat.nr_unused++; |
a4633357 | 344 | } else { |
b48f03b3 | 345 | list_move_tail(&dentry->d_lru, list); |
da3bbdd4 | 346 | } |
23044507 | 347 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
348 | } |
349 | ||
d52b9086 MS |
350 | /** |
351 | * d_kill - kill dentry and return parent | |
352 | * @dentry: dentry to kill | |
ff5fdb61 | 353 | * @parent: parent dentry |
d52b9086 | 354 | * |
31f3e0b3 | 355 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
356 | * |
357 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 358 | * |
b5c84bf6 NP |
359 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
360 | * d_kill. | |
d52b9086 | 361 | */ |
2fd6b7f5 | 362 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 363 | __releases(dentry->d_lock) |
2fd6b7f5 | 364 | __releases(parent->d_lock) |
873feea0 | 365 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 366 | { |
d52b9086 | 367 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
368 | /* |
369 | * Inform try_to_ascend() that we are no longer attached to the | |
370 | * dentry tree | |
371 | */ | |
b161dfa6 | 372 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
2fd6b7f5 NP |
373 | if (parent) |
374 | spin_unlock(&parent->d_lock); | |
d52b9086 | 375 | dentry_iput(dentry); |
b7ab39f6 NP |
376 | /* |
377 | * dentry_iput drops the locks, at which point nobody (except | |
378 | * transient RCU lookups) can reach this dentry. | |
379 | */ | |
d52b9086 | 380 | d_free(dentry); |
871c0067 | 381 | return parent; |
d52b9086 MS |
382 | } |
383 | ||
c6627c60 DH |
384 | /* |
385 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
386 | * dentry->d_lock is locked. The caller must take care of that, if | |
387 | * appropriate. | |
388 | */ | |
389 | static void __d_shrink(struct dentry *dentry) | |
390 | { | |
391 | if (!d_unhashed(dentry)) { | |
392 | struct hlist_bl_head *b; | |
393 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
394 | b = &dentry->d_sb->s_anon; | |
395 | else | |
396 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
397 | ||
398 | hlist_bl_lock(b); | |
399 | __hlist_bl_del(&dentry->d_hash); | |
400 | dentry->d_hash.pprev = NULL; | |
401 | hlist_bl_unlock(b); | |
402 | } | |
403 | } | |
404 | ||
789680d1 NP |
405 | /** |
406 | * d_drop - drop a dentry | |
407 | * @dentry: dentry to drop | |
408 | * | |
409 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
410 | * be found through a VFS lookup any more. Note that this is different from | |
411 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
412 | * possible, giving a successful _negative_ lookup, while d_drop will | |
413 | * just make the cache lookup fail. | |
414 | * | |
415 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
416 | * reason (NFS timeouts or autofs deletes). | |
417 | * | |
418 | * __d_drop requires dentry->d_lock. | |
419 | */ | |
420 | void __d_drop(struct dentry *dentry) | |
421 | { | |
dea3667b | 422 | if (!d_unhashed(dentry)) { |
c6627c60 | 423 | __d_shrink(dentry); |
dea3667b | 424 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
425 | } |
426 | } | |
427 | EXPORT_SYMBOL(__d_drop); | |
428 | ||
429 | void d_drop(struct dentry *dentry) | |
430 | { | |
789680d1 NP |
431 | spin_lock(&dentry->d_lock); |
432 | __d_drop(dentry); | |
433 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
434 | } |
435 | EXPORT_SYMBOL(d_drop); | |
436 | ||
77812a1e NP |
437 | /* |
438 | * Finish off a dentry we've decided to kill. | |
439 | * dentry->d_lock must be held, returns with it unlocked. | |
440 | * If ref is non-zero, then decrement the refcount too. | |
441 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
442 | */ | |
443 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
444 | __releases(dentry->d_lock) | |
445 | { | |
873feea0 | 446 | struct inode *inode; |
77812a1e NP |
447 | struct dentry *parent; |
448 | ||
873feea0 NP |
449 | inode = dentry->d_inode; |
450 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
451 | relock: |
452 | spin_unlock(&dentry->d_lock); | |
453 | cpu_relax(); | |
454 | return dentry; /* try again with same dentry */ | |
455 | } | |
456 | if (IS_ROOT(dentry)) | |
457 | parent = NULL; | |
458 | else | |
459 | parent = dentry->d_parent; | |
460 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
461 | if (inode) |
462 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
463 | goto relock; |
464 | } | |
31e6b01f | 465 | |
77812a1e NP |
466 | if (ref) |
467 | dentry->d_count--; | |
f0023bc6 | 468 | /* |
f0023bc6 SW |
469 | * inform the fs via d_prune that this dentry is about to be |
470 | * unhashed and destroyed. | |
471 | */ | |
61572bb1 YZ |
472 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
473 | dentry->d_op->d_prune(dentry); | |
474 | ||
475 | dentry_lru_del(dentry); | |
77812a1e NP |
476 | /* if it was on the hash then remove it */ |
477 | __d_drop(dentry); | |
478 | return d_kill(dentry, parent); | |
479 | } | |
480 | ||
1da177e4 LT |
481 | /* |
482 | * This is dput | |
483 | * | |
484 | * This is complicated by the fact that we do not want to put | |
485 | * dentries that are no longer on any hash chain on the unused | |
486 | * list: we'd much rather just get rid of them immediately. | |
487 | * | |
488 | * However, that implies that we have to traverse the dentry | |
489 | * tree upwards to the parents which might _also_ now be | |
490 | * scheduled for deletion (it may have been only waiting for | |
491 | * its last child to go away). | |
492 | * | |
493 | * This tail recursion is done by hand as we don't want to depend | |
494 | * on the compiler to always get this right (gcc generally doesn't). | |
495 | * Real recursion would eat up our stack space. | |
496 | */ | |
497 | ||
498 | /* | |
499 | * dput - release a dentry | |
500 | * @dentry: dentry to release | |
501 | * | |
502 | * Release a dentry. This will drop the usage count and if appropriate | |
503 | * call the dentry unlink method as well as removing it from the queues and | |
504 | * releasing its resources. If the parent dentries were scheduled for release | |
505 | * they too may now get deleted. | |
1da177e4 | 506 | */ |
1da177e4 LT |
507 | void dput(struct dentry *dentry) |
508 | { | |
509 | if (!dentry) | |
510 | return; | |
511 | ||
512 | repeat: | |
b7ab39f6 | 513 | if (dentry->d_count == 1) |
1da177e4 | 514 | might_sleep(); |
1da177e4 | 515 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
516 | BUG_ON(!dentry->d_count); |
517 | if (dentry->d_count > 1) { | |
518 | dentry->d_count--; | |
1da177e4 | 519 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
520 | return; |
521 | } | |
522 | ||
fb045adb | 523 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 524 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 525 | goto kill_it; |
1da177e4 | 526 | } |
265ac902 | 527 | |
1da177e4 LT |
528 | /* Unreachable? Get rid of it */ |
529 | if (d_unhashed(dentry)) | |
530 | goto kill_it; | |
265ac902 | 531 | |
39e3c955 | 532 | dentry->d_flags |= DCACHE_REFERENCED; |
a4633357 | 533 | dentry_lru_add(dentry); |
265ac902 | 534 | |
61f3dee4 NP |
535 | dentry->d_count--; |
536 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
537 | return; |
538 | ||
d52b9086 | 539 | kill_it: |
77812a1e | 540 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
541 | if (dentry) |
542 | goto repeat; | |
1da177e4 | 543 | } |
ec4f8605 | 544 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
545 | |
546 | /** | |
547 | * d_invalidate - invalidate a dentry | |
548 | * @dentry: dentry to invalidate | |
549 | * | |
550 | * Try to invalidate the dentry if it turns out to be | |
551 | * possible. If there are other dentries that can be | |
552 | * reached through this one we can't delete it and we | |
553 | * return -EBUSY. On success we return 0. | |
554 | * | |
555 | * no dcache lock. | |
556 | */ | |
557 | ||
558 | int d_invalidate(struct dentry * dentry) | |
559 | { | |
560 | /* | |
561 | * If it's already been dropped, return OK. | |
562 | */ | |
da502956 | 563 | spin_lock(&dentry->d_lock); |
1da177e4 | 564 | if (d_unhashed(dentry)) { |
da502956 | 565 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
566 | return 0; |
567 | } | |
568 | /* | |
569 | * Check whether to do a partial shrink_dcache | |
570 | * to get rid of unused child entries. | |
571 | */ | |
572 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 573 | spin_unlock(&dentry->d_lock); |
1da177e4 | 574 | shrink_dcache_parent(dentry); |
da502956 | 575 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
576 | } |
577 | ||
578 | /* | |
579 | * Somebody else still using it? | |
580 | * | |
581 | * If it's a directory, we can't drop it | |
582 | * for fear of somebody re-populating it | |
583 | * with children (even though dropping it | |
584 | * would make it unreachable from the root, | |
585 | * we might still populate it if it was a | |
586 | * working directory or similar). | |
50e69630 AV |
587 | * We also need to leave mountpoints alone, |
588 | * directory or not. | |
1da177e4 | 589 | */ |
50e69630 AV |
590 | if (dentry->d_count > 1 && dentry->d_inode) { |
591 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { | |
1da177e4 | 592 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
593 | return -EBUSY; |
594 | } | |
595 | } | |
596 | ||
597 | __d_drop(dentry); | |
598 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
599 | return 0; |
600 | } | |
ec4f8605 | 601 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 602 | |
b5c84bf6 | 603 | /* This must be called with d_lock held */ |
dc0474be | 604 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 605 | { |
b7ab39f6 | 606 | dentry->d_count++; |
23044507 NP |
607 | } |
608 | ||
dc0474be | 609 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 610 | { |
23044507 | 611 | spin_lock(&dentry->d_lock); |
dc0474be | 612 | __dget_dlock(dentry); |
23044507 | 613 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
614 | } |
615 | ||
b7ab39f6 NP |
616 | struct dentry *dget_parent(struct dentry *dentry) |
617 | { | |
618 | struct dentry *ret; | |
619 | ||
620 | repeat: | |
a734eb45 NP |
621 | /* |
622 | * Don't need rcu_dereference because we re-check it was correct under | |
623 | * the lock. | |
624 | */ | |
625 | rcu_read_lock(); | |
b7ab39f6 | 626 | ret = dentry->d_parent; |
a734eb45 NP |
627 | spin_lock(&ret->d_lock); |
628 | if (unlikely(ret != dentry->d_parent)) { | |
629 | spin_unlock(&ret->d_lock); | |
630 | rcu_read_unlock(); | |
b7ab39f6 NP |
631 | goto repeat; |
632 | } | |
a734eb45 | 633 | rcu_read_unlock(); |
b7ab39f6 NP |
634 | BUG_ON(!ret->d_count); |
635 | ret->d_count++; | |
636 | spin_unlock(&ret->d_lock); | |
b7ab39f6 NP |
637 | return ret; |
638 | } | |
639 | EXPORT_SYMBOL(dget_parent); | |
640 | ||
1da177e4 LT |
641 | /** |
642 | * d_find_alias - grab a hashed alias of inode | |
643 | * @inode: inode in question | |
32ba9c3f LT |
644 | * @want_discon: flag, used by d_splice_alias, to request |
645 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
646 | * |
647 | * If inode has a hashed alias, or is a directory and has any alias, | |
648 | * acquire the reference to alias and return it. Otherwise return NULL. | |
649 | * Notice that if inode is a directory there can be only one alias and | |
650 | * it can be unhashed only if it has no children, or if it is the root | |
651 | * of a filesystem. | |
652 | * | |
21c0d8fd | 653 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
654 | * any other hashed alias over that one unless @want_discon is set, |
655 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 656 | */ |
32ba9c3f | 657 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 658 | { |
da502956 | 659 | struct dentry *alias, *discon_alias; |
1da177e4 | 660 | |
da502956 NP |
661 | again: |
662 | discon_alias = NULL; | |
b67bfe0d | 663 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 664 | spin_lock(&alias->d_lock); |
1da177e4 | 665 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 666 | if (IS_ROOT(alias) && |
da502956 | 667 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 668 | discon_alias = alias; |
32ba9c3f | 669 | } else if (!want_discon) { |
dc0474be | 670 | __dget_dlock(alias); |
da502956 NP |
671 | spin_unlock(&alias->d_lock); |
672 | return alias; | |
673 | } | |
674 | } | |
675 | spin_unlock(&alias->d_lock); | |
676 | } | |
677 | if (discon_alias) { | |
678 | alias = discon_alias; | |
679 | spin_lock(&alias->d_lock); | |
680 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
681 | if (IS_ROOT(alias) && | |
682 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 683 | __dget_dlock(alias); |
da502956 | 684 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
685 | return alias; |
686 | } | |
687 | } | |
da502956 NP |
688 | spin_unlock(&alias->d_lock); |
689 | goto again; | |
1da177e4 | 690 | } |
da502956 | 691 | return NULL; |
1da177e4 LT |
692 | } |
693 | ||
da502956 | 694 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 695 | { |
214fda1f DH |
696 | struct dentry *de = NULL; |
697 | ||
b3d9b7a3 | 698 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 699 | spin_lock(&inode->i_lock); |
32ba9c3f | 700 | de = __d_find_alias(inode, 0); |
873feea0 | 701 | spin_unlock(&inode->i_lock); |
214fda1f | 702 | } |
1da177e4 LT |
703 | return de; |
704 | } | |
ec4f8605 | 705 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
706 | |
707 | /* | |
708 | * Try to kill dentries associated with this inode. | |
709 | * WARNING: you must own a reference to inode. | |
710 | */ | |
711 | void d_prune_aliases(struct inode *inode) | |
712 | { | |
0cdca3f9 | 713 | struct dentry *dentry; |
1da177e4 | 714 | restart: |
873feea0 | 715 | spin_lock(&inode->i_lock); |
b67bfe0d | 716 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 717 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 718 | if (!dentry->d_count) { |
dc0474be | 719 | __dget_dlock(dentry); |
1da177e4 LT |
720 | __d_drop(dentry); |
721 | spin_unlock(&dentry->d_lock); | |
873feea0 | 722 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
723 | dput(dentry); |
724 | goto restart; | |
725 | } | |
726 | spin_unlock(&dentry->d_lock); | |
727 | } | |
873feea0 | 728 | spin_unlock(&inode->i_lock); |
1da177e4 | 729 | } |
ec4f8605 | 730 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
731 | |
732 | /* | |
77812a1e NP |
733 | * Try to throw away a dentry - free the inode, dput the parent. |
734 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
735 | * Releases dentry->d_lock. | |
d702ccb3 | 736 | * |
77812a1e | 737 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 738 | */ |
77812a1e | 739 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 740 | __releases(dentry->d_lock) |
1da177e4 | 741 | { |
77812a1e | 742 | struct dentry *parent; |
d52b9086 | 743 | |
77812a1e | 744 | parent = dentry_kill(dentry, 0); |
d52b9086 | 745 | /* |
77812a1e NP |
746 | * If dentry_kill returns NULL, we have nothing more to do. |
747 | * if it returns the same dentry, trylocks failed. In either | |
748 | * case, just loop again. | |
749 | * | |
750 | * Otherwise, we need to prune ancestors too. This is necessary | |
751 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
752 | * is also expected to be beneficial in reducing dentry cache | |
753 | * fragmentation. | |
d52b9086 | 754 | */ |
77812a1e NP |
755 | if (!parent) |
756 | return; | |
757 | if (parent == dentry) | |
758 | return; | |
759 | ||
760 | /* Prune ancestors. */ | |
761 | dentry = parent; | |
d52b9086 | 762 | while (dentry) { |
b7ab39f6 | 763 | spin_lock(&dentry->d_lock); |
89e60548 NP |
764 | if (dentry->d_count > 1) { |
765 | dentry->d_count--; | |
766 | spin_unlock(&dentry->d_lock); | |
767 | return; | |
768 | } | |
77812a1e | 769 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 770 | } |
1da177e4 LT |
771 | } |
772 | ||
3049cfe2 | 773 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 774 | { |
da3bbdd4 | 775 | struct dentry *dentry; |
da3bbdd4 | 776 | |
ec33679d NP |
777 | rcu_read_lock(); |
778 | for (;;) { | |
ec33679d NP |
779 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
780 | if (&dentry->d_lru == list) | |
781 | break; /* empty */ | |
782 | spin_lock(&dentry->d_lock); | |
783 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
784 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
785 | continue; |
786 | } | |
787 | ||
1da177e4 LT |
788 | /* |
789 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
790 | * the LRU because of laziness during lookup. Do not free |
791 | * it - just keep it off the LRU list. | |
1da177e4 | 792 | */ |
b7ab39f6 | 793 | if (dentry->d_count) { |
ec33679d | 794 | dentry_lru_del(dentry); |
da3bbdd4 | 795 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
796 | continue; |
797 | } | |
ec33679d | 798 | |
ec33679d | 799 | rcu_read_unlock(); |
77812a1e NP |
800 | |
801 | try_prune_one_dentry(dentry); | |
802 | ||
ec33679d | 803 | rcu_read_lock(); |
da3bbdd4 | 804 | } |
ec33679d | 805 | rcu_read_unlock(); |
3049cfe2 CH |
806 | } |
807 | ||
808 | /** | |
b48f03b3 DC |
809 | * prune_dcache_sb - shrink the dcache |
810 | * @sb: superblock | |
811 | * @count: number of entries to try to free | |
812 | * | |
813 | * Attempt to shrink the superblock dcache LRU by @count entries. This is | |
814 | * done when we need more memory an called from the superblock shrinker | |
815 | * function. | |
3049cfe2 | 816 | * |
b48f03b3 DC |
817 | * This function may fail to free any resources if all the dentries are in |
818 | * use. | |
3049cfe2 | 819 | */ |
b48f03b3 | 820 | void prune_dcache_sb(struct super_block *sb, int count) |
3049cfe2 | 821 | { |
3049cfe2 CH |
822 | struct dentry *dentry; |
823 | LIST_HEAD(referenced); | |
824 | LIST_HEAD(tmp); | |
3049cfe2 | 825 | |
23044507 NP |
826 | relock: |
827 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
828 | while (!list_empty(&sb->s_dentry_lru)) { |
829 | dentry = list_entry(sb->s_dentry_lru.prev, | |
830 | struct dentry, d_lru); | |
831 | BUG_ON(dentry->d_sb != sb); | |
832 | ||
23044507 NP |
833 | if (!spin_trylock(&dentry->d_lock)) { |
834 | spin_unlock(&dcache_lru_lock); | |
835 | cpu_relax(); | |
836 | goto relock; | |
837 | } | |
838 | ||
b48f03b3 | 839 | if (dentry->d_flags & DCACHE_REFERENCED) { |
23044507 NP |
840 | dentry->d_flags &= ~DCACHE_REFERENCED; |
841 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 842 | spin_unlock(&dentry->d_lock); |
23044507 NP |
843 | } else { |
844 | list_move_tail(&dentry->d_lru, &tmp); | |
eaf5f907 | 845 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
23044507 | 846 | spin_unlock(&dentry->d_lock); |
b0d40c92 | 847 | if (!--count) |
23044507 | 848 | break; |
3049cfe2 | 849 | } |
ec33679d | 850 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 851 | } |
da3bbdd4 KM |
852 | if (!list_empty(&referenced)) |
853 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 854 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
855 | |
856 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
857 | } |
858 | ||
1da177e4 LT |
859 | /** |
860 | * shrink_dcache_sb - shrink dcache for a superblock | |
861 | * @sb: superblock | |
862 | * | |
3049cfe2 CH |
863 | * Shrink the dcache for the specified super block. This is used to free |
864 | * the dcache before unmounting a file system. | |
1da177e4 | 865 | */ |
3049cfe2 | 866 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 867 | { |
3049cfe2 CH |
868 | LIST_HEAD(tmp); |
869 | ||
23044507 | 870 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
871 | while (!list_empty(&sb->s_dentry_lru)) { |
872 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 873 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 874 | shrink_dentry_list(&tmp); |
ec33679d | 875 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 876 | } |
23044507 | 877 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 878 | } |
ec4f8605 | 879 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 880 | |
c636ebdb DH |
881 | /* |
882 | * destroy a single subtree of dentries for unmount | |
883 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
884 | * locking | |
885 | */ | |
886 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
887 | { | |
888 | struct dentry *parent; | |
889 | ||
890 | BUG_ON(!IS_ROOT(dentry)); | |
891 | ||
c636ebdb DH |
892 | for (;;) { |
893 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 894 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
895 | dentry = list_entry(dentry->d_subdirs.next, |
896 | struct dentry, d_u.d_child); | |
c636ebdb DH |
897 | |
898 | /* consume the dentries from this leaf up through its parents | |
899 | * until we find one with children or run out altogether */ | |
900 | do { | |
901 | struct inode *inode; | |
902 | ||
f0023bc6 | 903 | /* |
61572bb1 | 904 | * inform the fs that this dentry is about to be |
f0023bc6 SW |
905 | * unhashed and destroyed. |
906 | */ | |
61572bb1 YZ |
907 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
908 | dentry->d_op->d_prune(dentry); | |
909 | ||
910 | dentry_lru_del(dentry); | |
43c1c9cd DH |
911 | __d_shrink(dentry); |
912 | ||
b7ab39f6 | 913 | if (dentry->d_count != 0) { |
c636ebdb DH |
914 | printk(KERN_ERR |
915 | "BUG: Dentry %p{i=%lx,n=%s}" | |
916 | " still in use (%d)" | |
917 | " [unmount of %s %s]\n", | |
918 | dentry, | |
919 | dentry->d_inode ? | |
920 | dentry->d_inode->i_ino : 0UL, | |
921 | dentry->d_name.name, | |
b7ab39f6 | 922 | dentry->d_count, |
c636ebdb DH |
923 | dentry->d_sb->s_type->name, |
924 | dentry->d_sb->s_id); | |
925 | BUG(); | |
926 | } | |
927 | ||
2fd6b7f5 | 928 | if (IS_ROOT(dentry)) { |
c636ebdb | 929 | parent = NULL; |
2fd6b7f5 NP |
930 | list_del(&dentry->d_u.d_child); |
931 | } else { | |
871c0067 | 932 | parent = dentry->d_parent; |
b7ab39f6 | 933 | parent->d_count--; |
2fd6b7f5 | 934 | list_del(&dentry->d_u.d_child); |
871c0067 | 935 | } |
c636ebdb | 936 | |
c636ebdb DH |
937 | inode = dentry->d_inode; |
938 | if (inode) { | |
939 | dentry->d_inode = NULL; | |
b3d9b7a3 | 940 | hlist_del_init(&dentry->d_alias); |
c636ebdb DH |
941 | if (dentry->d_op && dentry->d_op->d_iput) |
942 | dentry->d_op->d_iput(dentry, inode); | |
943 | else | |
944 | iput(inode); | |
945 | } | |
946 | ||
947 | d_free(dentry); | |
948 | ||
949 | /* finished when we fall off the top of the tree, | |
950 | * otherwise we ascend to the parent and move to the | |
951 | * next sibling if there is one */ | |
952 | if (!parent) | |
312d3ca8 | 953 | return; |
c636ebdb | 954 | dentry = parent; |
c636ebdb DH |
955 | } while (list_empty(&dentry->d_subdirs)); |
956 | ||
957 | dentry = list_entry(dentry->d_subdirs.next, | |
958 | struct dentry, d_u.d_child); | |
959 | } | |
960 | } | |
961 | ||
962 | /* | |
963 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 964 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
965 | * - the superblock is detached from all mountings and open files, so the |
966 | * dentry trees will not be rearranged by the VFS | |
967 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
968 | * any dentries belonging to this superblock that it comes across | |
969 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
970 | * in this superblock | |
971 | */ | |
972 | void shrink_dcache_for_umount(struct super_block *sb) | |
973 | { | |
974 | struct dentry *dentry; | |
975 | ||
976 | if (down_read_trylock(&sb->s_umount)) | |
977 | BUG(); | |
978 | ||
979 | dentry = sb->s_root; | |
980 | sb->s_root = NULL; | |
b7ab39f6 | 981 | dentry->d_count--; |
c636ebdb DH |
982 | shrink_dcache_for_umount_subtree(dentry); |
983 | ||
ceb5bdc2 NP |
984 | while (!hlist_bl_empty(&sb->s_anon)) { |
985 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
986 | shrink_dcache_for_umount_subtree(dentry); |
987 | } | |
988 | } | |
989 | ||
c826cb7d LT |
990 | /* |
991 | * This tries to ascend one level of parenthood, but | |
992 | * we can race with renaming, so we need to re-check | |
993 | * the parenthood after dropping the lock and check | |
994 | * that the sequence number still matches. | |
995 | */ | |
996 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
997 | { | |
998 | struct dentry *new = old->d_parent; | |
999 | ||
1000 | rcu_read_lock(); | |
1001 | spin_unlock(&old->d_lock); | |
1002 | spin_lock(&new->d_lock); | |
1003 | ||
1004 | /* | |
1005 | * might go back up the wrong parent if we have had a rename | |
1006 | * or deletion | |
1007 | */ | |
1008 | if (new != old->d_parent || | |
b161dfa6 | 1009 | (old->d_flags & DCACHE_DENTRY_KILLED) || |
c826cb7d LT |
1010 | (!locked && read_seqretry(&rename_lock, seq))) { |
1011 | spin_unlock(&new->d_lock); | |
1012 | new = NULL; | |
1013 | } | |
1014 | rcu_read_unlock(); | |
1015 | return new; | |
1016 | } | |
1017 | ||
1018 | ||
1da177e4 LT |
1019 | /* |
1020 | * Search for at least 1 mount point in the dentry's subdirs. | |
1021 | * We descend to the next level whenever the d_subdirs | |
1022 | * list is non-empty and continue searching. | |
1023 | */ | |
1024 | ||
1025 | /** | |
1026 | * have_submounts - check for mounts over a dentry | |
1027 | * @parent: dentry to check. | |
1028 | * | |
1029 | * Return true if the parent or its subdirectories contain | |
1030 | * a mount point | |
1031 | */ | |
1da177e4 LT |
1032 | int have_submounts(struct dentry *parent) |
1033 | { | |
949854d0 | 1034 | struct dentry *this_parent; |
1da177e4 | 1035 | struct list_head *next; |
949854d0 | 1036 | unsigned seq; |
58db63d0 | 1037 | int locked = 0; |
949854d0 | 1038 | |
949854d0 | 1039 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1040 | again: |
1041 | this_parent = parent; | |
1da177e4 | 1042 | |
1da177e4 LT |
1043 | if (d_mountpoint(parent)) |
1044 | goto positive; | |
2fd6b7f5 | 1045 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1046 | repeat: |
1047 | next = this_parent->d_subdirs.next; | |
1048 | resume: | |
1049 | while (next != &this_parent->d_subdirs) { | |
1050 | struct list_head *tmp = next; | |
5160ee6f | 1051 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1052 | next = tmp->next; |
2fd6b7f5 NP |
1053 | |
1054 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1055 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1056 | if (d_mountpoint(dentry)) { |
1057 | spin_unlock(&dentry->d_lock); | |
1058 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1059 | goto positive; |
2fd6b7f5 | 1060 | } |
1da177e4 | 1061 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1062 | spin_unlock(&this_parent->d_lock); |
1063 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1064 | this_parent = dentry; |
2fd6b7f5 | 1065 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1066 | goto repeat; |
1067 | } | |
2fd6b7f5 | 1068 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1069 | } |
1070 | /* | |
1071 | * All done at this level ... ascend and resume the search. | |
1072 | */ | |
1073 | if (this_parent != parent) { | |
c826cb7d LT |
1074 | struct dentry *child = this_parent; |
1075 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1076 | if (!this_parent) | |
949854d0 | 1077 | goto rename_retry; |
949854d0 | 1078 | next = child->d_u.d_child.next; |
1da177e4 LT |
1079 | goto resume; |
1080 | } | |
2fd6b7f5 | 1081 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1082 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1083 | goto rename_retry; |
58db63d0 NP |
1084 | if (locked) |
1085 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1086 | return 0; /* No mount points found in tree */ |
1087 | positive: | |
58db63d0 | 1088 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1089 | goto rename_retry; |
58db63d0 NP |
1090 | if (locked) |
1091 | write_sequnlock(&rename_lock); | |
1da177e4 | 1092 | return 1; |
58db63d0 NP |
1093 | |
1094 | rename_retry: | |
8110e16d MS |
1095 | if (locked) |
1096 | goto again; | |
58db63d0 NP |
1097 | locked = 1; |
1098 | write_seqlock(&rename_lock); | |
1099 | goto again; | |
1da177e4 | 1100 | } |
ec4f8605 | 1101 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1102 | |
1103 | /* | |
fd517909 | 1104 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1105 | * and move any unused dentries to the end of the unused |
1106 | * list for prune_dcache(). We descend to the next level | |
1107 | * whenever the d_subdirs list is non-empty and continue | |
1108 | * searching. | |
1109 | * | |
1110 | * It returns zero iff there are no unused children, | |
1111 | * otherwise it returns the number of children moved to | |
1112 | * the end of the unused list. This may not be the total | |
1113 | * number of unused children, because select_parent can | |
1114 | * drop the lock and return early due to latency | |
1115 | * constraints. | |
1116 | */ | |
b48f03b3 | 1117 | static int select_parent(struct dentry *parent, struct list_head *dispose) |
1da177e4 | 1118 | { |
949854d0 | 1119 | struct dentry *this_parent; |
1da177e4 | 1120 | struct list_head *next; |
949854d0 | 1121 | unsigned seq; |
1da177e4 | 1122 | int found = 0; |
58db63d0 | 1123 | int locked = 0; |
1da177e4 | 1124 | |
949854d0 | 1125 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1126 | again: |
1127 | this_parent = parent; | |
2fd6b7f5 | 1128 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1129 | repeat: |
1130 | next = this_parent->d_subdirs.next; | |
1131 | resume: | |
1132 | while (next != &this_parent->d_subdirs) { | |
1133 | struct list_head *tmp = next; | |
5160ee6f | 1134 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1135 | next = tmp->next; |
1136 | ||
2fd6b7f5 | 1137 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1138 | |
b48f03b3 DC |
1139 | /* |
1140 | * move only zero ref count dentries to the dispose list. | |
eaf5f907 MS |
1141 | * |
1142 | * Those which are presently on the shrink list, being processed | |
1143 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1144 | * loop in shrink_dcache_parent() might not make any progress | |
1145 | * and loop forever. | |
1da177e4 | 1146 | */ |
eaf5f907 MS |
1147 | if (dentry->d_count) { |
1148 | dentry_lru_del(dentry); | |
1149 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
b48f03b3 | 1150 | dentry_lru_move_list(dentry, dispose); |
eaf5f907 | 1151 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
1da177e4 LT |
1152 | found++; |
1153 | } | |
1da177e4 LT |
1154 | /* |
1155 | * We can return to the caller if we have found some (this | |
1156 | * ensures forward progress). We'll be coming back to find | |
1157 | * the rest. | |
1158 | */ | |
2fd6b7f5 NP |
1159 | if (found && need_resched()) { |
1160 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1161 | goto out; |
2fd6b7f5 | 1162 | } |
1da177e4 LT |
1163 | |
1164 | /* | |
1165 | * Descend a level if the d_subdirs list is non-empty. | |
1166 | */ | |
1167 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1168 | spin_unlock(&this_parent->d_lock); |
1169 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1170 | this_parent = dentry; |
2fd6b7f5 | 1171 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1172 | goto repeat; |
1173 | } | |
2fd6b7f5 NP |
1174 | |
1175 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1176 | } |
1177 | /* | |
1178 | * All done at this level ... ascend and resume the search. | |
1179 | */ | |
1180 | if (this_parent != parent) { | |
c826cb7d LT |
1181 | struct dentry *child = this_parent; |
1182 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1183 | if (!this_parent) | |
949854d0 | 1184 | goto rename_retry; |
949854d0 | 1185 | next = child->d_u.d_child.next; |
1da177e4 LT |
1186 | goto resume; |
1187 | } | |
1188 | out: | |
2fd6b7f5 | 1189 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1190 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1191 | goto rename_retry; |
58db63d0 NP |
1192 | if (locked) |
1193 | write_sequnlock(&rename_lock); | |
1da177e4 | 1194 | return found; |
58db63d0 NP |
1195 | |
1196 | rename_retry: | |
1197 | if (found) | |
1198 | return found; | |
8110e16d MS |
1199 | if (locked) |
1200 | goto again; | |
58db63d0 NP |
1201 | locked = 1; |
1202 | write_seqlock(&rename_lock); | |
1203 | goto again; | |
1da177e4 LT |
1204 | } |
1205 | ||
1206 | /** | |
1207 | * shrink_dcache_parent - prune dcache | |
1208 | * @parent: parent of entries to prune | |
1209 | * | |
1210 | * Prune the dcache to remove unused children of the parent dentry. | |
1211 | */ | |
1da177e4 LT |
1212 | void shrink_dcache_parent(struct dentry * parent) |
1213 | { | |
b48f03b3 | 1214 | LIST_HEAD(dispose); |
1da177e4 LT |
1215 | int found; |
1216 | ||
421348f1 | 1217 | while ((found = select_parent(parent, &dispose)) != 0) { |
b48f03b3 | 1218 | shrink_dentry_list(&dispose); |
421348f1 GT |
1219 | cond_resched(); |
1220 | } | |
1da177e4 | 1221 | } |
ec4f8605 | 1222 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1223 | |
1da177e4 | 1224 | /** |
a4464dbc AV |
1225 | * __d_alloc - allocate a dcache entry |
1226 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1227 | * @name: qstr of the name |
1228 | * | |
1229 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1230 | * available. On a success the dentry is returned. The name passed in is | |
1231 | * copied and the copy passed in may be reused after this call. | |
1232 | */ | |
1233 | ||
a4464dbc | 1234 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1235 | { |
1236 | struct dentry *dentry; | |
1237 | char *dname; | |
1238 | ||
e12ba74d | 1239 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1240 | if (!dentry) |
1241 | return NULL; | |
1242 | ||
6326c71f LT |
1243 | /* |
1244 | * We guarantee that the inline name is always NUL-terminated. | |
1245 | * This way the memcpy() done by the name switching in rename | |
1246 | * will still always have a NUL at the end, even if we might | |
1247 | * be overwriting an internal NUL character | |
1248 | */ | |
1249 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1250 | if (name->len > DNAME_INLINE_LEN-1) { |
1251 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1252 | if (!dname) { | |
1253 | kmem_cache_free(dentry_cache, dentry); | |
1254 | return NULL; | |
1255 | } | |
1256 | } else { | |
1257 | dname = dentry->d_iname; | |
1258 | } | |
1da177e4 LT |
1259 | |
1260 | dentry->d_name.len = name->len; | |
1261 | dentry->d_name.hash = name->hash; | |
1262 | memcpy(dname, name->name, name->len); | |
1263 | dname[name->len] = 0; | |
1264 | ||
6326c71f LT |
1265 | /* Make sure we always see the terminating NUL character */ |
1266 | smp_wmb(); | |
1267 | dentry->d_name.name = dname; | |
1268 | ||
b7ab39f6 | 1269 | dentry->d_count = 1; |
dea3667b | 1270 | dentry->d_flags = 0; |
1da177e4 | 1271 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1272 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1273 | dentry->d_inode = NULL; |
a4464dbc AV |
1274 | dentry->d_parent = dentry; |
1275 | dentry->d_sb = sb; | |
1da177e4 LT |
1276 | dentry->d_op = NULL; |
1277 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1278 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1279 | INIT_LIST_HEAD(&dentry->d_lru); |
1280 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1281 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1282 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1283 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1284 | |
3e880fb5 | 1285 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1286 | |
1da177e4 LT |
1287 | return dentry; |
1288 | } | |
a4464dbc AV |
1289 | |
1290 | /** | |
1291 | * d_alloc - allocate a dcache entry | |
1292 | * @parent: parent of entry to allocate | |
1293 | * @name: qstr of the name | |
1294 | * | |
1295 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1296 | * available. On a success the dentry is returned. The name passed in is | |
1297 | * copied and the copy passed in may be reused after this call. | |
1298 | */ | |
1299 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1300 | { | |
1301 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1302 | if (!dentry) | |
1303 | return NULL; | |
1304 | ||
1305 | spin_lock(&parent->d_lock); | |
1306 | /* | |
1307 | * don't need child lock because it is not subject | |
1308 | * to concurrency here | |
1309 | */ | |
1310 | __dget_dlock(parent); | |
1311 | dentry->d_parent = parent; | |
1312 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1313 | spin_unlock(&parent->d_lock); | |
1314 | ||
1315 | return dentry; | |
1316 | } | |
ec4f8605 | 1317 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1318 | |
4b936885 NP |
1319 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1320 | { | |
a4464dbc AV |
1321 | struct dentry *dentry = __d_alloc(sb, name); |
1322 | if (dentry) | |
4b936885 | 1323 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1324 | return dentry; |
1325 | } | |
1326 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1327 | ||
1da177e4 LT |
1328 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1329 | { | |
1330 | struct qstr q; | |
1331 | ||
1332 | q.name = name; | |
1333 | q.len = strlen(name); | |
1334 | q.hash = full_name_hash(q.name, q.len); | |
1335 | return d_alloc(parent, &q); | |
1336 | } | |
ef26ca97 | 1337 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1338 | |
fb045adb NP |
1339 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1340 | { | |
6f7f7caa LT |
1341 | WARN_ON_ONCE(dentry->d_op); |
1342 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1343 | DCACHE_OP_COMPARE | |
1344 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1345 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1346 | DCACHE_OP_DELETE )); |
1347 | dentry->d_op = op; | |
1348 | if (!op) | |
1349 | return; | |
1350 | if (op->d_hash) | |
1351 | dentry->d_flags |= DCACHE_OP_HASH; | |
1352 | if (op->d_compare) | |
1353 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1354 | if (op->d_revalidate) | |
1355 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1356 | if (op->d_weak_revalidate) |
1357 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1358 | if (op->d_delete) |
1359 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1360 | if (op->d_prune) |
1361 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1362 | |
1363 | } | |
1364 | EXPORT_SYMBOL(d_set_d_op); | |
1365 | ||
360da900 OH |
1366 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1367 | { | |
b23fb0a6 | 1368 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1369 | if (inode) { |
1370 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1371 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
b3d9b7a3 | 1372 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1373 | } |
360da900 | 1374 | dentry->d_inode = inode; |
31e6b01f | 1375 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1376 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1377 | fsnotify_d_instantiate(dentry, inode); |
1378 | } | |
1379 | ||
1da177e4 LT |
1380 | /** |
1381 | * d_instantiate - fill in inode information for a dentry | |
1382 | * @entry: dentry to complete | |
1383 | * @inode: inode to attach to this dentry | |
1384 | * | |
1385 | * Fill in inode information in the entry. | |
1386 | * | |
1387 | * This turns negative dentries into productive full members | |
1388 | * of society. | |
1389 | * | |
1390 | * NOTE! This assumes that the inode count has been incremented | |
1391 | * (or otherwise set) by the caller to indicate that it is now | |
1392 | * in use by the dcache. | |
1393 | */ | |
1394 | ||
1395 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1396 | { | |
b3d9b7a3 | 1397 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1398 | if (inode) |
1399 | spin_lock(&inode->i_lock); | |
360da900 | 1400 | __d_instantiate(entry, inode); |
873feea0 NP |
1401 | if (inode) |
1402 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1403 | security_d_instantiate(entry, inode); |
1404 | } | |
ec4f8605 | 1405 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1406 | |
1407 | /** | |
1408 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1409 | * @entry: dentry to instantiate | |
1410 | * @inode: inode to attach to this dentry | |
1411 | * | |
1412 | * Fill in inode information in the entry. On success, it returns NULL. | |
1413 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1414 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1415 | * |
1416 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1417 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1418 | * |
1419 | * This also assumes that the inode count has been incremented | |
1420 | * (or otherwise set) by the caller to indicate that it is now | |
1421 | * in use by the dcache. | |
1da177e4 | 1422 | */ |
770bfad8 DH |
1423 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1424 | struct inode *inode) | |
1da177e4 LT |
1425 | { |
1426 | struct dentry *alias; | |
1427 | int len = entry->d_name.len; | |
1428 | const char *name = entry->d_name.name; | |
1429 | unsigned int hash = entry->d_name.hash; | |
1430 | ||
770bfad8 | 1431 | if (!inode) { |
360da900 | 1432 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1433 | return NULL; |
1434 | } | |
1435 | ||
b67bfe0d | 1436 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1437 | /* |
1438 | * Don't need alias->d_lock here, because aliases with | |
1439 | * d_parent == entry->d_parent are not subject to name or | |
1440 | * parent changes, because the parent inode i_mutex is held. | |
1441 | */ | |
12f8ad4b | 1442 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1443 | continue; |
1444 | if (alias->d_parent != entry->d_parent) | |
1445 | continue; | |
ee983e89 LT |
1446 | if (alias->d_name.len != len) |
1447 | continue; | |
12f8ad4b | 1448 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1449 | continue; |
dc0474be | 1450 | __dget(alias); |
1da177e4 LT |
1451 | return alias; |
1452 | } | |
770bfad8 | 1453 | |
360da900 | 1454 | __d_instantiate(entry, inode); |
1da177e4 LT |
1455 | return NULL; |
1456 | } | |
770bfad8 DH |
1457 | |
1458 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1459 | { | |
1460 | struct dentry *result; | |
1461 | ||
b3d9b7a3 | 1462 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1463 | |
873feea0 NP |
1464 | if (inode) |
1465 | spin_lock(&inode->i_lock); | |
770bfad8 | 1466 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1467 | if (inode) |
1468 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1469 | |
1470 | if (!result) { | |
1471 | security_d_instantiate(entry, inode); | |
1472 | return NULL; | |
1473 | } | |
1474 | ||
1475 | BUG_ON(!d_unhashed(result)); | |
1476 | iput(inode); | |
1477 | return result; | |
1478 | } | |
1479 | ||
1da177e4 LT |
1480 | EXPORT_SYMBOL(d_instantiate_unique); |
1481 | ||
adc0e91a AV |
1482 | struct dentry *d_make_root(struct inode *root_inode) |
1483 | { | |
1484 | struct dentry *res = NULL; | |
1485 | ||
1486 | if (root_inode) { | |
26fe5750 | 1487 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1488 | |
1489 | res = __d_alloc(root_inode->i_sb, &name); | |
1490 | if (res) | |
1491 | d_instantiate(res, root_inode); | |
1492 | else | |
1493 | iput(root_inode); | |
1494 | } | |
1495 | return res; | |
1496 | } | |
1497 | EXPORT_SYMBOL(d_make_root); | |
1498 | ||
d891eedb BF |
1499 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1500 | { | |
1501 | struct dentry *alias; | |
1502 | ||
b3d9b7a3 | 1503 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1504 | return NULL; |
b3d9b7a3 | 1505 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1506 | __dget(alias); |
1507 | return alias; | |
1508 | } | |
1509 | ||
46f72b34 SW |
1510 | /** |
1511 | * d_find_any_alias - find any alias for a given inode | |
1512 | * @inode: inode to find an alias for | |
1513 | * | |
1514 | * If any aliases exist for the given inode, take and return a | |
1515 | * reference for one of them. If no aliases exist, return %NULL. | |
1516 | */ | |
1517 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1518 | { |
1519 | struct dentry *de; | |
1520 | ||
1521 | spin_lock(&inode->i_lock); | |
1522 | de = __d_find_any_alias(inode); | |
1523 | spin_unlock(&inode->i_lock); | |
1524 | return de; | |
1525 | } | |
46f72b34 | 1526 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1527 | |
4ea3ada2 CH |
1528 | /** |
1529 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1530 | * @inode: inode to allocate the dentry for | |
1531 | * | |
1532 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1533 | * similar open by handle operations. The returned dentry may be anonymous, | |
1534 | * or may have a full name (if the inode was already in the cache). | |
1535 | * | |
1536 | * When called on a directory inode, we must ensure that the inode only ever | |
1537 | * has one dentry. If a dentry is found, that is returned instead of | |
1538 | * allocating a new one. | |
1539 | * | |
1540 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1541 | * to the dentry. In case of an error the reference on the inode is released. |
1542 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1543 | * be passed in and will be the error will be propagate to the return value, | |
1544 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1545 | */ |
1546 | struct dentry *d_obtain_alias(struct inode *inode) | |
1547 | { | |
b911a6bd | 1548 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1549 | struct dentry *tmp; |
1550 | struct dentry *res; | |
4ea3ada2 CH |
1551 | |
1552 | if (!inode) | |
44003728 | 1553 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1554 | if (IS_ERR(inode)) |
1555 | return ERR_CAST(inode); | |
1556 | ||
d891eedb | 1557 | res = d_find_any_alias(inode); |
9308a612 CH |
1558 | if (res) |
1559 | goto out_iput; | |
1560 | ||
a4464dbc | 1561 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1562 | if (!tmp) { |
1563 | res = ERR_PTR(-ENOMEM); | |
1564 | goto out_iput; | |
4ea3ada2 | 1565 | } |
b5c84bf6 | 1566 | |
873feea0 | 1567 | spin_lock(&inode->i_lock); |
d891eedb | 1568 | res = __d_find_any_alias(inode); |
9308a612 | 1569 | if (res) { |
873feea0 | 1570 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1571 | dput(tmp); |
1572 | goto out_iput; | |
1573 | } | |
1574 | ||
1575 | /* attach a disconnected dentry */ | |
1576 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1577 | tmp->d_inode = inode; |
1578 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
b3d9b7a3 | 1579 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1580 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1581 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1582 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1583 | spin_unlock(&tmp->d_lock); |
873feea0 | 1584 | spin_unlock(&inode->i_lock); |
24ff6663 | 1585 | security_d_instantiate(tmp, inode); |
9308a612 | 1586 | |
9308a612 CH |
1587 | return tmp; |
1588 | ||
1589 | out_iput: | |
24ff6663 JB |
1590 | if (res && !IS_ERR(res)) |
1591 | security_d_instantiate(res, inode); | |
9308a612 CH |
1592 | iput(inode); |
1593 | return res; | |
4ea3ada2 | 1594 | } |
adc48720 | 1595 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1596 | |
1597 | /** | |
1598 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1599 | * @inode: the inode which may have a disconnected dentry | |
1600 | * @dentry: a negative dentry which we want to point to the inode. | |
1601 | * | |
1602 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1603 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1604 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1605 | * | |
1606 | * This is needed in the lookup routine of any filesystem that is exportable | |
1607 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1608 | * | |
1609 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1610 | * is returned. This matches the expected return value of ->lookup. | |
1611 | * | |
1612 | */ | |
1613 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1614 | { | |
1615 | struct dentry *new = NULL; | |
1616 | ||
a9049376 AV |
1617 | if (IS_ERR(inode)) |
1618 | return ERR_CAST(inode); | |
1619 | ||
21c0d8fd | 1620 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1621 | spin_lock(&inode->i_lock); |
32ba9c3f | 1622 | new = __d_find_alias(inode, 1); |
1da177e4 | 1623 | if (new) { |
32ba9c3f | 1624 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1625 | spin_unlock(&inode->i_lock); |
1da177e4 | 1626 | security_d_instantiate(new, inode); |
1da177e4 LT |
1627 | d_move(new, dentry); |
1628 | iput(inode); | |
1629 | } else { | |
873feea0 | 1630 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1631 | __d_instantiate(dentry, inode); |
873feea0 | 1632 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1633 | security_d_instantiate(dentry, inode); |
1634 | d_rehash(dentry); | |
1635 | } | |
1636 | } else | |
1637 | d_add(dentry, inode); | |
1638 | return new; | |
1639 | } | |
ec4f8605 | 1640 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1641 | |
9403540c BN |
1642 | /** |
1643 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1644 | * @inode: the inode case-insensitive lookup has found | |
1645 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1646 | * @name: the case-exact name to be associated with the returned dentry | |
1647 | * | |
1648 | * This is to avoid filling the dcache with case-insensitive names to the | |
1649 | * same inode, only the actual correct case is stored in the dcache for | |
1650 | * case-insensitive filesystems. | |
1651 | * | |
1652 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1653 | * already exists in in the dcache, use it and return it. | |
1654 | * | |
1655 | * If no entry exists with the exact case name, allocate new dentry with | |
1656 | * the exact case, and return the spliced entry. | |
1657 | */ | |
e45b590b | 1658 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1659 | struct qstr *name) |
1660 | { | |
9403540c BN |
1661 | struct dentry *found; |
1662 | struct dentry *new; | |
1663 | ||
b6520c81 CH |
1664 | /* |
1665 | * First check if a dentry matching the name already exists, | |
1666 | * if not go ahead and create it now. | |
1667 | */ | |
9403540c | 1668 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1669 | if (unlikely(IS_ERR(found))) |
1670 | goto err_out; | |
9403540c BN |
1671 | if (!found) { |
1672 | new = d_alloc(dentry->d_parent, name); | |
1673 | if (!new) { | |
4f522a24 | 1674 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1675 | goto err_out; |
1676 | } | |
b6520c81 | 1677 | |
9403540c BN |
1678 | found = d_splice_alias(inode, new); |
1679 | if (found) { | |
1680 | dput(new); | |
1681 | return found; | |
1682 | } | |
1683 | return new; | |
1684 | } | |
b6520c81 CH |
1685 | |
1686 | /* | |
1687 | * If a matching dentry exists, and it's not negative use it. | |
1688 | * | |
1689 | * Decrement the reference count to balance the iget() done | |
1690 | * earlier on. | |
1691 | */ | |
9403540c BN |
1692 | if (found->d_inode) { |
1693 | if (unlikely(found->d_inode != inode)) { | |
1694 | /* This can't happen because bad inodes are unhashed. */ | |
1695 | BUG_ON(!is_bad_inode(inode)); | |
1696 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1697 | } | |
9403540c BN |
1698 | iput(inode); |
1699 | return found; | |
1700 | } | |
b6520c81 | 1701 | |
9403540c | 1702 | /* |
9403540c | 1703 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1704 | * already has a dentry. |
9403540c | 1705 | */ |
4513d899 AV |
1706 | new = d_splice_alias(inode, found); |
1707 | if (new) { | |
1708 | dput(found); | |
1709 | found = new; | |
9403540c | 1710 | } |
4513d899 | 1711 | return found; |
9403540c BN |
1712 | |
1713 | err_out: | |
1714 | iput(inode); | |
4f522a24 | 1715 | return found; |
9403540c | 1716 | } |
ec4f8605 | 1717 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1718 | |
12f8ad4b LT |
1719 | /* |
1720 | * Do the slow-case of the dentry name compare. | |
1721 | * | |
1722 | * Unlike the dentry_cmp() function, we need to atomically | |
1723 | * load the name, length and inode information, so that the | |
1724 | * filesystem can rely on them, and can use the 'name' and | |
1725 | * 'len' information without worrying about walking off the | |
1726 | * end of memory etc. | |
1727 | * | |
1728 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1729 | * in arguments (the low-level filesystem should not look | |
1730 | * at the dentry inode or name contents directly, since | |
1731 | * rename can change them while we're in RCU mode). | |
1732 | */ | |
1733 | enum slow_d_compare { | |
1734 | D_COMP_OK, | |
1735 | D_COMP_NOMATCH, | |
1736 | D_COMP_SEQRETRY, | |
1737 | }; | |
1738 | ||
1739 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1740 | const struct dentry *parent, | |
1741 | struct inode *inode, | |
1742 | struct dentry *dentry, | |
1743 | unsigned int seq, | |
1744 | const struct qstr *name) | |
1745 | { | |
1746 | int tlen = dentry->d_name.len; | |
1747 | const char *tname = dentry->d_name.name; | |
1748 | struct inode *i = dentry->d_inode; | |
1749 | ||
1750 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1751 | cpu_relax(); | |
1752 | return D_COMP_SEQRETRY; | |
1753 | } | |
1754 | if (parent->d_op->d_compare(parent, inode, | |
1755 | dentry, i, | |
1756 | tlen, tname, name)) | |
1757 | return D_COMP_NOMATCH; | |
1758 | return D_COMP_OK; | |
1759 | } | |
1760 | ||
31e6b01f NP |
1761 | /** |
1762 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1763 | * @parent: parent dentry | |
1764 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1765 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1766 | * @inode: returns dentry->d_inode when the inode was found valid. |
1767 | * Returns: dentry, or NULL | |
1768 | * | |
1769 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1770 | * resolution (store-free path walking) design described in | |
1771 | * Documentation/filesystems/path-lookup.txt. | |
1772 | * | |
1773 | * This is not to be used outside core vfs. | |
1774 | * | |
1775 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1776 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1777 | * without taking d_lock and checking d_seq sequence count against @seq | |
1778 | * returned here. | |
1779 | * | |
1780 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1781 | * function. | |
1782 | * | |
1783 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1784 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1785 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1786 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
1787 | * |
1788 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
1789 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 1790 | */ |
8966be90 LT |
1791 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
1792 | const struct qstr *name, | |
12f8ad4b | 1793 | unsigned *seqp, struct inode *inode) |
31e6b01f | 1794 | { |
26fe5750 | 1795 | u64 hashlen = name->hash_len; |
31e6b01f | 1796 | const unsigned char *str = name->name; |
26fe5750 | 1797 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 1798 | struct hlist_bl_node *node; |
31e6b01f NP |
1799 | struct dentry *dentry; |
1800 | ||
1801 | /* | |
1802 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1803 | * required to prevent single threaded performance regressions | |
1804 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1805 | * Keep the two functions in sync. | |
1806 | */ | |
1807 | ||
1808 | /* | |
1809 | * The hash list is protected using RCU. | |
1810 | * | |
1811 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1812 | * races with d_move(). | |
1813 | * | |
1814 | * It is possible that concurrent renames can mess up our list | |
1815 | * walk here and result in missing our dentry, resulting in the | |
1816 | * false-negative result. d_lookup() protects against concurrent | |
1817 | * renames using rename_lock seqlock. | |
1818 | * | |
b0a4bb83 | 1819 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1820 | */ |
b07ad996 | 1821 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 1822 | unsigned seq; |
31e6b01f | 1823 | |
31e6b01f | 1824 | seqretry: |
12f8ad4b LT |
1825 | /* |
1826 | * The dentry sequence count protects us from concurrent | |
1827 | * renames, and thus protects inode, parent and name fields. | |
1828 | * | |
1829 | * The caller must perform a seqcount check in order | |
1830 | * to do anything useful with the returned dentry, | |
1831 | * including using the 'd_inode' pointer. | |
1832 | * | |
1833 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
1834 | * we don't wait for the sequence count to stabilize if it | |
1835 | * is in the middle of a sequence change. If we do the slow | |
1836 | * dentry compare, we will do seqretries until it is stable, | |
1837 | * and if we end up with a successful lookup, we actually | |
1838 | * want to exit RCU lookup anyway. | |
1839 | */ | |
1840 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
1841 | if (dentry->d_parent != parent) |
1842 | continue; | |
2e321806 LT |
1843 | if (d_unhashed(dentry)) |
1844 | continue; | |
12f8ad4b LT |
1845 | *seqp = seq; |
1846 | ||
830c0f0e | 1847 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
1848 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
1849 | continue; | |
12f8ad4b LT |
1850 | switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) { |
1851 | case D_COMP_OK: | |
1852 | return dentry; | |
1853 | case D_COMP_NOMATCH: | |
31e6b01f | 1854 | continue; |
12f8ad4b LT |
1855 | default: |
1856 | goto seqretry; | |
1857 | } | |
31e6b01f | 1858 | } |
12f8ad4b | 1859 | |
26fe5750 | 1860 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 1861 | continue; |
26fe5750 | 1862 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 1863 | return dentry; |
31e6b01f NP |
1864 | } |
1865 | return NULL; | |
1866 | } | |
1867 | ||
1da177e4 LT |
1868 | /** |
1869 | * d_lookup - search for a dentry | |
1870 | * @parent: parent dentry | |
1871 | * @name: qstr of name we wish to find | |
b04f784e | 1872 | * Returns: dentry, or NULL |
1da177e4 | 1873 | * |
b04f784e NP |
1874 | * d_lookup searches the children of the parent dentry for the name in |
1875 | * question. If the dentry is found its reference count is incremented and the | |
1876 | * dentry is returned. The caller must use dput to free the entry when it has | |
1877 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1878 | */ |
da2d8455 | 1879 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 1880 | { |
31e6b01f | 1881 | struct dentry *dentry; |
949854d0 | 1882 | unsigned seq; |
1da177e4 LT |
1883 | |
1884 | do { | |
1885 | seq = read_seqbegin(&rename_lock); | |
1886 | dentry = __d_lookup(parent, name); | |
1887 | if (dentry) | |
1888 | break; | |
1889 | } while (read_seqretry(&rename_lock, seq)); | |
1890 | return dentry; | |
1891 | } | |
ec4f8605 | 1892 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1893 | |
31e6b01f | 1894 | /** |
b04f784e NP |
1895 | * __d_lookup - search for a dentry (racy) |
1896 | * @parent: parent dentry | |
1897 | * @name: qstr of name we wish to find | |
1898 | * Returns: dentry, or NULL | |
1899 | * | |
1900 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1901 | * false-negative result due to unrelated rename activity. | |
1902 | * | |
1903 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1904 | * however it must be used carefully, eg. with a following d_lookup in | |
1905 | * the case of failure. | |
1906 | * | |
1907 | * __d_lookup callers must be commented. | |
1908 | */ | |
a713ca2a | 1909 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
1910 | { |
1911 | unsigned int len = name->len; | |
1912 | unsigned int hash = name->hash; | |
1913 | const unsigned char *str = name->name; | |
b07ad996 | 1914 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1915 | struct hlist_bl_node *node; |
31e6b01f | 1916 | struct dentry *found = NULL; |
665a7583 | 1917 | struct dentry *dentry; |
1da177e4 | 1918 | |
31e6b01f NP |
1919 | /* |
1920 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1921 | * required to prevent single threaded performance regressions | |
1922 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1923 | * Keep the two functions in sync. | |
1924 | */ | |
1925 | ||
b04f784e NP |
1926 | /* |
1927 | * The hash list is protected using RCU. | |
1928 | * | |
1929 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1930 | * with d_move(). | |
1931 | * | |
1932 | * It is possible that concurrent renames can mess up our list | |
1933 | * walk here and result in missing our dentry, resulting in the | |
1934 | * false-negative result. d_lookup() protects against concurrent | |
1935 | * renames using rename_lock seqlock. | |
1936 | * | |
b0a4bb83 | 1937 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1938 | */ |
1da177e4 LT |
1939 | rcu_read_lock(); |
1940 | ||
b07ad996 | 1941 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 1942 | |
1da177e4 LT |
1943 | if (dentry->d_name.hash != hash) |
1944 | continue; | |
1da177e4 LT |
1945 | |
1946 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1947 | if (dentry->d_parent != parent) |
1948 | goto next; | |
d0185c08 LT |
1949 | if (d_unhashed(dentry)) |
1950 | goto next; | |
1951 | ||
1da177e4 LT |
1952 | /* |
1953 | * It is safe to compare names since d_move() cannot | |
1954 | * change the qstr (protected by d_lock). | |
1955 | */ | |
fb045adb | 1956 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
1957 | int tlen = dentry->d_name.len; |
1958 | const char *tname = dentry->d_name.name; | |
621e155a NP |
1959 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1960 | dentry, dentry->d_inode, | |
31e6b01f | 1961 | tlen, tname, name)) |
1da177e4 LT |
1962 | goto next; |
1963 | } else { | |
ee983e89 LT |
1964 | if (dentry->d_name.len != len) |
1965 | goto next; | |
12f8ad4b | 1966 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
1967 | goto next; |
1968 | } | |
1969 | ||
b7ab39f6 | 1970 | dentry->d_count++; |
d0185c08 | 1971 | found = dentry; |
1da177e4 LT |
1972 | spin_unlock(&dentry->d_lock); |
1973 | break; | |
1974 | next: | |
1975 | spin_unlock(&dentry->d_lock); | |
1976 | } | |
1977 | rcu_read_unlock(); | |
1978 | ||
1979 | return found; | |
1980 | } | |
1981 | ||
3e7e241f EB |
1982 | /** |
1983 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1984 | * @dir: Directory to search in | |
1985 | * @name: qstr of name we wish to find | |
1986 | * | |
4f522a24 | 1987 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
1988 | */ |
1989 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1990 | { | |
3e7e241f EB |
1991 | /* |
1992 | * Check for a fs-specific hash function. Note that we must | |
1993 | * calculate the standard hash first, as the d_op->d_hash() | |
1994 | * routine may choose to leave the hash value unchanged. | |
1995 | */ | |
1996 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1997 | if (dir->d_flags & DCACHE_OP_HASH) { |
4f522a24 AV |
1998 | int err = dir->d_op->d_hash(dir, dir->d_inode, name); |
1999 | if (unlikely(err < 0)) | |
2000 | return ERR_PTR(err); | |
3e7e241f | 2001 | } |
4f522a24 | 2002 | return d_lookup(dir, name); |
3e7e241f | 2003 | } |
4f522a24 | 2004 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2005 | |
1da177e4 | 2006 | /** |
786a5e15 | 2007 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2008 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2009 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2010 | * |
2011 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2012 | * This is used by ncpfs in its readdir implementation. | |
2013 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2014 | * |
2015 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2016 | */ |
d3a23e16 | 2017 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2018 | { |
786a5e15 | 2019 | struct dentry *child; |
d3a23e16 | 2020 | |
2fd6b7f5 | 2021 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2022 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2023 | if (dentry == child) { | |
2fd6b7f5 | 2024 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2025 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2026 | spin_unlock(&dentry->d_lock); |
2027 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2028 | return 1; |
2029 | } | |
2030 | } | |
2fd6b7f5 | 2031 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2032 | |
1da177e4 LT |
2033 | return 0; |
2034 | } | |
ec4f8605 | 2035 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2036 | |
2037 | /* | |
2038 | * When a file is deleted, we have two options: | |
2039 | * - turn this dentry into a negative dentry | |
2040 | * - unhash this dentry and free it. | |
2041 | * | |
2042 | * Usually, we want to just turn this into | |
2043 | * a negative dentry, but if anybody else is | |
2044 | * currently using the dentry or the inode | |
2045 | * we can't do that and we fall back on removing | |
2046 | * it from the hash queues and waiting for | |
2047 | * it to be deleted later when it has no users | |
2048 | */ | |
2049 | ||
2050 | /** | |
2051 | * d_delete - delete a dentry | |
2052 | * @dentry: The dentry to delete | |
2053 | * | |
2054 | * Turn the dentry into a negative dentry if possible, otherwise | |
2055 | * remove it from the hash queues so it can be deleted later | |
2056 | */ | |
2057 | ||
2058 | void d_delete(struct dentry * dentry) | |
2059 | { | |
873feea0 | 2060 | struct inode *inode; |
7a91bf7f | 2061 | int isdir = 0; |
1da177e4 LT |
2062 | /* |
2063 | * Are we the only user? | |
2064 | */ | |
357f8e65 | 2065 | again: |
1da177e4 | 2066 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2067 | inode = dentry->d_inode; |
2068 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 2069 | if (dentry->d_count == 1) { |
1fe0c023 | 2070 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2071 | spin_unlock(&dentry->d_lock); |
2072 | cpu_relax(); | |
2073 | goto again; | |
2074 | } | |
13e3c5e5 | 2075 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2076 | dentry_unlink_inode(dentry); |
7a91bf7f | 2077 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2078 | return; |
2079 | } | |
2080 | ||
2081 | if (!d_unhashed(dentry)) | |
2082 | __d_drop(dentry); | |
2083 | ||
2084 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2085 | |
2086 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2087 | } |
ec4f8605 | 2088 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2089 | |
b07ad996 | 2090 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2091 | { |
ceb5bdc2 | 2092 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2093 | hlist_bl_lock(b); |
dea3667b | 2094 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2095 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2096 | hlist_bl_unlock(b); |
1da177e4 LT |
2097 | } |
2098 | ||
770bfad8 DH |
2099 | static void _d_rehash(struct dentry * entry) |
2100 | { | |
2101 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2102 | } | |
2103 | ||
1da177e4 LT |
2104 | /** |
2105 | * d_rehash - add an entry back to the hash | |
2106 | * @entry: dentry to add to the hash | |
2107 | * | |
2108 | * Adds a dentry to the hash according to its name. | |
2109 | */ | |
2110 | ||
2111 | void d_rehash(struct dentry * entry) | |
2112 | { | |
1da177e4 | 2113 | spin_lock(&entry->d_lock); |
770bfad8 | 2114 | _d_rehash(entry); |
1da177e4 | 2115 | spin_unlock(&entry->d_lock); |
1da177e4 | 2116 | } |
ec4f8605 | 2117 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2118 | |
fb2d5b86 NP |
2119 | /** |
2120 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2121 | * @dentry: dentry to be updated | |
2122 | * @name: new name | |
2123 | * | |
2124 | * Update a case insensitive dentry with new case of name. | |
2125 | * | |
2126 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2127 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2128 | * lengths). | |
2129 | * | |
2130 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2131 | * keep renames and concurrent inserts, and readdir(2) away). | |
2132 | */ | |
2133 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2134 | { | |
7ebfa57f | 2135 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2136 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2137 | ||
fb2d5b86 | 2138 | spin_lock(&dentry->d_lock); |
31e6b01f | 2139 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2140 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2141 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2142 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2143 | } |
2144 | EXPORT_SYMBOL(dentry_update_name_case); | |
2145 | ||
1da177e4 LT |
2146 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2147 | { | |
2148 | if (dname_external(target)) { | |
2149 | if (dname_external(dentry)) { | |
2150 | /* | |
2151 | * Both external: swap the pointers | |
2152 | */ | |
9a8d5bb4 | 2153 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2154 | } else { |
2155 | /* | |
2156 | * dentry:internal, target:external. Steal target's | |
2157 | * storage and make target internal. | |
2158 | */ | |
321bcf92 BF |
2159 | memcpy(target->d_iname, dentry->d_name.name, |
2160 | dentry->d_name.len + 1); | |
1da177e4 LT |
2161 | dentry->d_name.name = target->d_name.name; |
2162 | target->d_name.name = target->d_iname; | |
2163 | } | |
2164 | } else { | |
2165 | if (dname_external(dentry)) { | |
2166 | /* | |
2167 | * dentry:external, target:internal. Give dentry's | |
2168 | * storage to target and make dentry internal | |
2169 | */ | |
2170 | memcpy(dentry->d_iname, target->d_name.name, | |
2171 | target->d_name.len + 1); | |
2172 | target->d_name.name = dentry->d_name.name; | |
2173 | dentry->d_name.name = dentry->d_iname; | |
2174 | } else { | |
2175 | /* | |
2176 | * Both are internal. Just copy target to dentry | |
2177 | */ | |
2178 | memcpy(dentry->d_iname, target->d_name.name, | |
2179 | target->d_name.len + 1); | |
dc711ca3 AV |
2180 | dentry->d_name.len = target->d_name.len; |
2181 | return; | |
1da177e4 LT |
2182 | } |
2183 | } | |
9a8d5bb4 | 2184 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2185 | } |
2186 | ||
2fd6b7f5 NP |
2187 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2188 | { | |
2189 | /* | |
2190 | * XXXX: do we really need to take target->d_lock? | |
2191 | */ | |
2192 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2193 | spin_lock(&target->d_parent->d_lock); | |
2194 | else { | |
2195 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2196 | spin_lock(&dentry->d_parent->d_lock); | |
2197 | spin_lock_nested(&target->d_parent->d_lock, | |
2198 | DENTRY_D_LOCK_NESTED); | |
2199 | } else { | |
2200 | spin_lock(&target->d_parent->d_lock); | |
2201 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2202 | DENTRY_D_LOCK_NESTED); | |
2203 | } | |
2204 | } | |
2205 | if (target < dentry) { | |
2206 | spin_lock_nested(&target->d_lock, 2); | |
2207 | spin_lock_nested(&dentry->d_lock, 3); | |
2208 | } else { | |
2209 | spin_lock_nested(&dentry->d_lock, 2); | |
2210 | spin_lock_nested(&target->d_lock, 3); | |
2211 | } | |
2212 | } | |
2213 | ||
2214 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2215 | struct dentry *target) | |
2216 | { | |
2217 | if (target->d_parent != dentry->d_parent) | |
2218 | spin_unlock(&dentry->d_parent->d_lock); | |
2219 | if (target->d_parent != target) | |
2220 | spin_unlock(&target->d_parent->d_lock); | |
2221 | } | |
2222 | ||
1da177e4 | 2223 | /* |
2fd6b7f5 NP |
2224 | * When switching names, the actual string doesn't strictly have to |
2225 | * be preserved in the target - because we're dropping the target | |
2226 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2227 | * the new name before we switch. | |
2228 | * | |
2229 | * Note that we have to be a lot more careful about getting the hash | |
2230 | * switched - we have to switch the hash value properly even if it | |
2231 | * then no longer matches the actual (corrupted) string of the target. | |
2232 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2233 | */ |
9eaef27b | 2234 | /* |
18367501 | 2235 | * __d_move - move a dentry |
1da177e4 LT |
2236 | * @dentry: entry to move |
2237 | * @target: new dentry | |
2238 | * | |
2239 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2240 | * dcache entries should not be moved in this way. Caller must hold |
2241 | * rename_lock, the i_mutex of the source and target directories, | |
2242 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2243 | */ |
18367501 | 2244 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2245 | { |
1da177e4 LT |
2246 | if (!dentry->d_inode) |
2247 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2248 | ||
2fd6b7f5 NP |
2249 | BUG_ON(d_ancestor(dentry, target)); |
2250 | BUG_ON(d_ancestor(target, dentry)); | |
2251 | ||
2fd6b7f5 | 2252 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2253 | |
31e6b01f NP |
2254 | write_seqcount_begin(&dentry->d_seq); |
2255 | write_seqcount_begin(&target->d_seq); | |
2256 | ||
ceb5bdc2 NP |
2257 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2258 | ||
2259 | /* | |
2260 | * Move the dentry to the target hash queue. Don't bother checking | |
2261 | * for the same hash queue because of how unlikely it is. | |
2262 | */ | |
2263 | __d_drop(dentry); | |
789680d1 | 2264 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2265 | |
2266 | /* Unhash the target: dput() will then get rid of it */ | |
2267 | __d_drop(target); | |
2268 | ||
5160ee6f ED |
2269 | list_del(&dentry->d_u.d_child); |
2270 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2271 | |
2272 | /* Switch the names.. */ | |
2273 | switch_names(dentry, target); | |
9a8d5bb4 | 2274 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2275 | |
2276 | /* ... and switch the parents */ | |
2277 | if (IS_ROOT(dentry)) { | |
2278 | dentry->d_parent = target->d_parent; | |
2279 | target->d_parent = target; | |
5160ee6f | 2280 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2281 | } else { |
9a8d5bb4 | 2282 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2283 | |
2284 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2285 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2286 | } |
2287 | ||
5160ee6f | 2288 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2289 | |
31e6b01f NP |
2290 | write_seqcount_end(&target->d_seq); |
2291 | write_seqcount_end(&dentry->d_seq); | |
2292 | ||
2fd6b7f5 | 2293 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2294 | spin_unlock(&target->d_lock); |
c32ccd87 | 2295 | fsnotify_d_move(dentry); |
1da177e4 | 2296 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2297 | } |
2298 | ||
2299 | /* | |
2300 | * d_move - move a dentry | |
2301 | * @dentry: entry to move | |
2302 | * @target: new dentry | |
2303 | * | |
2304 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2305 | * dcache entries should not be moved in this way. See the locking |
2306 | * requirements for __d_move. | |
18367501 AV |
2307 | */ |
2308 | void d_move(struct dentry *dentry, struct dentry *target) | |
2309 | { | |
2310 | write_seqlock(&rename_lock); | |
2311 | __d_move(dentry, target); | |
1da177e4 | 2312 | write_sequnlock(&rename_lock); |
9eaef27b | 2313 | } |
ec4f8605 | 2314 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2315 | |
e2761a11 OH |
2316 | /** |
2317 | * d_ancestor - search for an ancestor | |
2318 | * @p1: ancestor dentry | |
2319 | * @p2: child dentry | |
2320 | * | |
2321 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2322 | * an ancestor of p2, else NULL. | |
9eaef27b | 2323 | */ |
e2761a11 | 2324 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2325 | { |
2326 | struct dentry *p; | |
2327 | ||
871c0067 | 2328 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2329 | if (p->d_parent == p1) |
e2761a11 | 2330 | return p; |
9eaef27b | 2331 | } |
e2761a11 | 2332 | return NULL; |
9eaef27b TM |
2333 | } |
2334 | ||
2335 | /* | |
2336 | * This helper attempts to cope with remotely renamed directories | |
2337 | * | |
2338 | * It assumes that the caller is already holding | |
18367501 | 2339 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2340 | * |
2341 | * Note: If ever the locking in lock_rename() changes, then please | |
2342 | * remember to update this too... | |
9eaef27b | 2343 | */ |
873feea0 NP |
2344 | static struct dentry *__d_unalias(struct inode *inode, |
2345 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2346 | { |
2347 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2348 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2349 | |
2350 | /* If alias and dentry share a parent, then no extra locks required */ | |
2351 | if (alias->d_parent == dentry->d_parent) | |
2352 | goto out_unalias; | |
2353 | ||
9eaef27b | 2354 | /* See lock_rename() */ |
9eaef27b TM |
2355 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2356 | goto out_err; | |
2357 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2358 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2359 | goto out_err; | |
2360 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2361 | out_unalias: | |
ee3efa91 AV |
2362 | if (likely(!d_mountpoint(alias))) { |
2363 | __d_move(alias, dentry); | |
2364 | ret = alias; | |
2365 | } | |
9eaef27b | 2366 | out_err: |
873feea0 | 2367 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2368 | if (m2) |
2369 | mutex_unlock(m2); | |
2370 | if (m1) | |
2371 | mutex_unlock(m1); | |
2372 | return ret; | |
2373 | } | |
2374 | ||
770bfad8 DH |
2375 | /* |
2376 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2377 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2378 | * returns with anon->d_lock held! |
770bfad8 DH |
2379 | */ |
2380 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2381 | { | |
740da42e | 2382 | struct dentry *dparent; |
770bfad8 | 2383 | |
2fd6b7f5 | 2384 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2385 | |
31e6b01f NP |
2386 | write_seqcount_begin(&dentry->d_seq); |
2387 | write_seqcount_begin(&anon->d_seq); | |
2388 | ||
770bfad8 | 2389 | dparent = dentry->d_parent; |
770bfad8 | 2390 | |
2fd6b7f5 NP |
2391 | switch_names(dentry, anon); |
2392 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2393 | ||
740da42e AV |
2394 | dentry->d_parent = dentry; |
2395 | list_del_init(&dentry->d_u.d_child); | |
2396 | anon->d_parent = dparent; | |
9ed53b12 | 2397 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2398 | |
31e6b01f NP |
2399 | write_seqcount_end(&dentry->d_seq); |
2400 | write_seqcount_end(&anon->d_seq); | |
2401 | ||
2fd6b7f5 NP |
2402 | dentry_unlock_parents_for_move(anon, dentry); |
2403 | spin_unlock(&dentry->d_lock); | |
2404 | ||
2405 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2406 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2407 | } | |
2408 | ||
2409 | /** | |
2410 | * d_materialise_unique - introduce an inode into the tree | |
2411 | * @dentry: candidate dentry | |
2412 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2413 | * | |
2414 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2415 | * root directory alias in its place if there is one. Caller must hold the |
2416 | * i_mutex of the parent directory. | |
770bfad8 DH |
2417 | */ |
2418 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2419 | { | |
9eaef27b | 2420 | struct dentry *actual; |
770bfad8 DH |
2421 | |
2422 | BUG_ON(!d_unhashed(dentry)); | |
2423 | ||
770bfad8 DH |
2424 | if (!inode) { |
2425 | actual = dentry; | |
360da900 | 2426 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2427 | d_rehash(actual); |
2428 | goto out_nolock; | |
770bfad8 DH |
2429 | } |
2430 | ||
873feea0 | 2431 | spin_lock(&inode->i_lock); |
357f8e65 | 2432 | |
9eaef27b TM |
2433 | if (S_ISDIR(inode->i_mode)) { |
2434 | struct dentry *alias; | |
2435 | ||
2436 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2437 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2438 | if (alias) { |
2439 | actual = alias; | |
18367501 AV |
2440 | write_seqlock(&rename_lock); |
2441 | ||
2442 | if (d_ancestor(alias, dentry)) { | |
2443 | /* Check for loops */ | |
2444 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2445 | spin_unlock(&inode->i_lock); |
18367501 AV |
2446 | } else if (IS_ROOT(alias)) { |
2447 | /* Is this an anonymous mountpoint that we | |
2448 | * could splice into our tree? */ | |
9eaef27b | 2449 | __d_materialise_dentry(dentry, alias); |
18367501 | 2450 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2451 | __d_drop(alias); |
2452 | goto found; | |
18367501 AV |
2453 | } else { |
2454 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2455 | * aliasing. This drops inode->i_lock */ |
18367501 | 2456 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2457 | } |
18367501 | 2458 | write_sequnlock(&rename_lock); |
dd179946 DH |
2459 | if (IS_ERR(actual)) { |
2460 | if (PTR_ERR(actual) == -ELOOP) | |
2461 | pr_warn_ratelimited( | |
2462 | "VFS: Lookup of '%s' in %s %s" | |
2463 | " would have caused loop\n", | |
2464 | dentry->d_name.name, | |
2465 | inode->i_sb->s_type->name, | |
2466 | inode->i_sb->s_id); | |
9eaef27b | 2467 | dput(alias); |
dd179946 | 2468 | } |
9eaef27b TM |
2469 | goto out_nolock; |
2470 | } | |
770bfad8 DH |
2471 | } |
2472 | ||
2473 | /* Add a unique reference */ | |
2474 | actual = __d_instantiate_unique(dentry, inode); | |
2475 | if (!actual) | |
2476 | actual = dentry; | |
357f8e65 NP |
2477 | else |
2478 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2479 | |
770bfad8 DH |
2480 | spin_lock(&actual->d_lock); |
2481 | found: | |
2482 | _d_rehash(actual); | |
2483 | spin_unlock(&actual->d_lock); | |
873feea0 | 2484 | spin_unlock(&inode->i_lock); |
9eaef27b | 2485 | out_nolock: |
770bfad8 DH |
2486 | if (actual == dentry) { |
2487 | security_d_instantiate(dentry, inode); | |
2488 | return NULL; | |
2489 | } | |
2490 | ||
2491 | iput(inode); | |
2492 | return actual; | |
770bfad8 | 2493 | } |
ec4f8605 | 2494 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2495 | |
cdd16d02 | 2496 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2497 | { |
2498 | *buflen -= namelen; | |
2499 | if (*buflen < 0) | |
2500 | return -ENAMETOOLONG; | |
2501 | *buffer -= namelen; | |
2502 | memcpy(*buffer, str, namelen); | |
2503 | return 0; | |
2504 | } | |
2505 | ||
cdd16d02 MS |
2506 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2507 | { | |
2508 | return prepend(buffer, buflen, name->name, name->len); | |
2509 | } | |
2510 | ||
1da177e4 | 2511 | /** |
208898c1 | 2512 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2513 | * @path: the dentry/vfsmount to report |
02125a82 | 2514 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2515 | * @buffer: pointer to the end of the buffer |
2516 | * @buflen: pointer to buffer length | |
552ce544 | 2517 | * |
949854d0 | 2518 | * Caller holds the rename_lock. |
1da177e4 | 2519 | */ |
02125a82 AV |
2520 | static int prepend_path(const struct path *path, |
2521 | const struct path *root, | |
f2eb6575 | 2522 | char **buffer, int *buflen) |
1da177e4 | 2523 | { |
9d1bc601 MS |
2524 | struct dentry *dentry = path->dentry; |
2525 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2526 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 MS |
2527 | bool slash = false; |
2528 | int error = 0; | |
6092d048 | 2529 | |
f2eb6575 | 2530 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2531 | struct dentry * parent; |
2532 | ||
1da177e4 | 2533 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2534 | /* Global root? */ |
676da58d | 2535 | if (!mnt_has_parent(mnt)) |
1da177e4 | 2536 | goto global_root; |
a73324da | 2537 | dentry = mnt->mnt_mountpoint; |
0714a533 AV |
2538 | mnt = mnt->mnt_parent; |
2539 | vfsmnt = &mnt->mnt; | |
1da177e4 LT |
2540 | continue; |
2541 | } | |
2542 | parent = dentry->d_parent; | |
2543 | prefetch(parent); | |
9abca360 | 2544 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2545 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2546 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2547 | if (!error) |
2548 | error = prepend(buffer, buflen, "/", 1); | |
2549 | if (error) | |
2550 | break; | |
2551 | ||
2552 | slash = true; | |
1da177e4 LT |
2553 | dentry = parent; |
2554 | } | |
2555 | ||
f2eb6575 MS |
2556 | if (!error && !slash) |
2557 | error = prepend(buffer, buflen, "/", 1); | |
2558 | ||
f2eb6575 | 2559 | return error; |
1da177e4 LT |
2560 | |
2561 | global_root: | |
98dc568b MS |
2562 | /* |
2563 | * Filesystems needing to implement special "root names" | |
2564 | * should do so with ->d_dname() | |
2565 | */ | |
2566 | if (IS_ROOT(dentry) && | |
2567 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2568 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2569 | (int) dentry->d_name.len, dentry->d_name.name); | |
2570 | } | |
02125a82 AV |
2571 | if (!slash) |
2572 | error = prepend(buffer, buflen, "/", 1); | |
2573 | if (!error) | |
f7a99c5b | 2574 | error = is_mounted(vfsmnt) ? 1 : 2; |
7ea600b5 | 2575 | return error; |
f2eb6575 | 2576 | } |
be285c71 | 2577 | |
f2eb6575 MS |
2578 | /** |
2579 | * __d_path - return the path of a dentry | |
2580 | * @path: the dentry/vfsmount to report | |
02125a82 | 2581 | * @root: root vfsmnt/dentry |
cd956a1c | 2582 | * @buf: buffer to return value in |
f2eb6575 MS |
2583 | * @buflen: buffer length |
2584 | * | |
ffd1f4ed | 2585 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2586 | * |
2587 | * Returns a pointer into the buffer or an error code if the | |
2588 | * path was too long. | |
2589 | * | |
be148247 | 2590 | * "buflen" should be positive. |
f2eb6575 | 2591 | * |
02125a82 | 2592 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2593 | */ |
02125a82 AV |
2594 | char *__d_path(const struct path *path, |
2595 | const struct path *root, | |
f2eb6575 MS |
2596 | char *buf, int buflen) |
2597 | { | |
2598 | char *res = buf + buflen; | |
2599 | int error; | |
2600 | ||
2601 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2602 | br_read_lock(&vfsmount_lock); |
949854d0 | 2603 | write_seqlock(&rename_lock); |
f2eb6575 | 2604 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2605 | write_sequnlock(&rename_lock); |
7ea600b5 | 2606 | br_read_unlock(&vfsmount_lock); |
be148247 | 2607 | |
02125a82 AV |
2608 | if (error < 0) |
2609 | return ERR_PTR(error); | |
2610 | if (error > 0) | |
2611 | return NULL; | |
2612 | return res; | |
2613 | } | |
2614 | ||
2615 | char *d_absolute_path(const struct path *path, | |
2616 | char *buf, int buflen) | |
2617 | { | |
2618 | struct path root = {}; | |
2619 | char *res = buf + buflen; | |
2620 | int error; | |
2621 | ||
2622 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2623 | br_read_lock(&vfsmount_lock); |
02125a82 AV |
2624 | write_seqlock(&rename_lock); |
2625 | error = prepend_path(path, &root, &res, &buflen); | |
2626 | write_sequnlock(&rename_lock); | |
7ea600b5 | 2627 | br_read_unlock(&vfsmount_lock); |
02125a82 AV |
2628 | |
2629 | if (error > 1) | |
2630 | error = -EINVAL; | |
2631 | if (error < 0) | |
f2eb6575 | 2632 | return ERR_PTR(error); |
f2eb6575 | 2633 | return res; |
1da177e4 LT |
2634 | } |
2635 | ||
ffd1f4ed MS |
2636 | /* |
2637 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2638 | */ | |
02125a82 AV |
2639 | static int path_with_deleted(const struct path *path, |
2640 | const struct path *root, | |
2641 | char **buf, int *buflen) | |
ffd1f4ed MS |
2642 | { |
2643 | prepend(buf, buflen, "\0", 1); | |
2644 | if (d_unlinked(path->dentry)) { | |
2645 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2646 | if (error) | |
2647 | return error; | |
2648 | } | |
2649 | ||
2650 | return prepend_path(path, root, buf, buflen); | |
2651 | } | |
2652 | ||
8df9d1a4 MS |
2653 | static int prepend_unreachable(char **buffer, int *buflen) |
2654 | { | |
2655 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2656 | } | |
2657 | ||
a03a8a70 JB |
2658 | /** |
2659 | * d_path - return the path of a dentry | |
cf28b486 | 2660 | * @path: path to report |
a03a8a70 JB |
2661 | * @buf: buffer to return value in |
2662 | * @buflen: buffer length | |
2663 | * | |
2664 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2665 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2666 | * | |
52afeefb AV |
2667 | * Returns a pointer into the buffer or an error code if the path was |
2668 | * too long. Note: Callers should use the returned pointer, not the passed | |
2669 | * in buffer, to use the name! The implementation often starts at an offset | |
2670 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2671 | * |
31f3e0b3 | 2672 | * "buflen" should be positive. |
a03a8a70 | 2673 | */ |
20d4fdc1 | 2674 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2675 | { |
ffd1f4ed | 2676 | char *res = buf + buflen; |
6ac08c39 | 2677 | struct path root; |
ffd1f4ed | 2678 | int error; |
1da177e4 | 2679 | |
c23fbb6b ED |
2680 | /* |
2681 | * We have various synthetic filesystems that never get mounted. On | |
2682 | * these filesystems dentries are never used for lookup purposes, and | |
2683 | * thus don't need to be hashed. They also don't need a name until a | |
2684 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2685 | * below allows us to generate a name for these objects on demand: | |
3717eee3 EB |
2686 | * |
2687 | * Some pseudo inodes are mountable. When they are mounted | |
2688 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
2689 | * and instead have d_path return the mounted path. | |
c23fbb6b | 2690 | */ |
3717eee3 EB |
2691 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
2692 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 2693 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 2694 | |
f7ad3c6b | 2695 | get_fs_root(current->fs, &root); |
7ea600b5 | 2696 | br_read_lock(&vfsmount_lock); |
949854d0 | 2697 | write_seqlock(&rename_lock); |
02125a82 | 2698 | error = path_with_deleted(path, &root, &res, &buflen); |
7ea600b5 AV |
2699 | write_sequnlock(&rename_lock); |
2700 | br_read_unlock(&vfsmount_lock); | |
02125a82 | 2701 | if (error < 0) |
ffd1f4ed | 2702 | res = ERR_PTR(error); |
6ac08c39 | 2703 | path_put(&root); |
1da177e4 LT |
2704 | return res; |
2705 | } | |
ec4f8605 | 2706 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2707 | |
c23fbb6b ED |
2708 | /* |
2709 | * Helper function for dentry_operations.d_dname() members | |
2710 | */ | |
2711 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2712 | const char *fmt, ...) | |
2713 | { | |
2714 | va_list args; | |
2715 | char temp[64]; | |
2716 | int sz; | |
2717 | ||
2718 | va_start(args, fmt); | |
2719 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2720 | va_end(args); | |
2721 | ||
2722 | if (sz > sizeof(temp) || sz > buflen) | |
2723 | return ERR_PTR(-ENAMETOOLONG); | |
2724 | ||
2725 | buffer += buflen - sz; | |
2726 | return memcpy(buffer, temp, sz); | |
2727 | } | |
2728 | ||
ad4c3cc4 AV |
2729 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
2730 | { | |
2731 | char *end = buffer + buflen; | |
2732 | /* these dentries are never renamed, so d_lock is not needed */ | |
2733 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
2734 | prepend_name(&end, &buflen, &dentry->d_name) || | |
2735 | prepend(&end, &buflen, "/", 1)) | |
2736 | end = ERR_PTR(-ENAMETOOLONG); | |
2737 | return end; | |
2738 | } | |
2739 | ||
6092d048 RP |
2740 | /* |
2741 | * Write full pathname from the root of the filesystem into the buffer. | |
2742 | */ | |
ec2447c2 | 2743 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2744 | { |
2745 | char *end = buf + buflen; | |
2746 | char *retval; | |
2747 | ||
6092d048 | 2748 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2749 | if (buflen < 1) |
2750 | goto Elong; | |
2751 | /* Get '/' right */ | |
2752 | retval = end-1; | |
2753 | *retval = '/'; | |
2754 | ||
cdd16d02 MS |
2755 | while (!IS_ROOT(dentry)) { |
2756 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2757 | int error; |
6092d048 | 2758 | |
6092d048 | 2759 | prefetch(parent); |
9abca360 NP |
2760 | spin_lock(&dentry->d_lock); |
2761 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2762 | spin_unlock(&dentry->d_lock); | |
2763 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2764 | goto Elong; |
2765 | ||
2766 | retval = end; | |
2767 | dentry = parent; | |
2768 | } | |
c103135c AV |
2769 | return retval; |
2770 | Elong: | |
2771 | return ERR_PTR(-ENAMETOOLONG); | |
2772 | } | |
ec2447c2 NP |
2773 | |
2774 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2775 | { | |
2776 | char *retval; | |
2777 | ||
949854d0 | 2778 | write_seqlock(&rename_lock); |
ec2447c2 | 2779 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2780 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2781 | |
2782 | return retval; | |
2783 | } | |
2784 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2785 | |
2786 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2787 | { | |
2788 | char *p = NULL; | |
2789 | char *retval; | |
2790 | ||
949854d0 | 2791 | write_seqlock(&rename_lock); |
c103135c AV |
2792 | if (d_unlinked(dentry)) { |
2793 | p = buf + buflen; | |
2794 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2795 | goto Elong; | |
2796 | buflen++; | |
2797 | } | |
2798 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2799 | write_sequnlock(&rename_lock); |
c103135c AV |
2800 | if (!IS_ERR(retval) && p) |
2801 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2802 | return retval; |
2803 | Elong: | |
6092d048 RP |
2804 | return ERR_PTR(-ENAMETOOLONG); |
2805 | } | |
2806 | ||
1da177e4 LT |
2807 | /* |
2808 | * NOTE! The user-level library version returns a | |
2809 | * character pointer. The kernel system call just | |
2810 | * returns the length of the buffer filled (which | |
2811 | * includes the ending '\0' character), or a negative | |
2812 | * error value. So libc would do something like | |
2813 | * | |
2814 | * char *getcwd(char * buf, size_t size) | |
2815 | * { | |
2816 | * int retval; | |
2817 | * | |
2818 | * retval = sys_getcwd(buf, size); | |
2819 | * if (retval >= 0) | |
2820 | * return buf; | |
2821 | * errno = -retval; | |
2822 | * return NULL; | |
2823 | * } | |
2824 | */ | |
3cdad428 | 2825 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2826 | { |
552ce544 | 2827 | int error; |
6ac08c39 | 2828 | struct path pwd, root; |
552ce544 | 2829 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2830 | |
2831 | if (!page) | |
2832 | return -ENOMEM; | |
2833 | ||
f7ad3c6b | 2834 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2835 | |
552ce544 | 2836 | error = -ENOENT; |
7ea600b5 | 2837 | br_read_lock(&vfsmount_lock); |
949854d0 | 2838 | write_seqlock(&rename_lock); |
f3da392e | 2839 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2840 | unsigned long len; |
8df9d1a4 MS |
2841 | char *cwd = page + PAGE_SIZE; |
2842 | int buflen = PAGE_SIZE; | |
1da177e4 | 2843 | |
8df9d1a4 | 2844 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 2845 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
949854d0 | 2846 | write_sequnlock(&rename_lock); |
7ea600b5 | 2847 | br_read_unlock(&vfsmount_lock); |
552ce544 | 2848 | |
02125a82 | 2849 | if (error < 0) |
552ce544 LT |
2850 | goto out; |
2851 | ||
8df9d1a4 | 2852 | /* Unreachable from current root */ |
02125a82 | 2853 | if (error > 0) { |
8df9d1a4 MS |
2854 | error = prepend_unreachable(&cwd, &buflen); |
2855 | if (error) | |
2856 | goto out; | |
2857 | } | |
2858 | ||
552ce544 LT |
2859 | error = -ERANGE; |
2860 | len = PAGE_SIZE + page - cwd; | |
2861 | if (len <= size) { | |
2862 | error = len; | |
2863 | if (copy_to_user(buf, cwd, len)) | |
2864 | error = -EFAULT; | |
2865 | } | |
949854d0 NP |
2866 | } else { |
2867 | write_sequnlock(&rename_lock); | |
7ea600b5 | 2868 | br_read_unlock(&vfsmount_lock); |
949854d0 | 2869 | } |
1da177e4 LT |
2870 | |
2871 | out: | |
6ac08c39 JB |
2872 | path_put(&pwd); |
2873 | path_put(&root); | |
1da177e4 LT |
2874 | free_page((unsigned long) page); |
2875 | return error; | |
2876 | } | |
2877 | ||
2878 | /* | |
2879 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2880 | * | |
2881 | * Trivially implemented using the dcache structure | |
2882 | */ | |
2883 | ||
2884 | /** | |
2885 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2886 | * @new_dentry: new dentry | |
2887 | * @old_dentry: old dentry | |
2888 | * | |
2889 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2890 | * Returns 0 otherwise. | |
2891 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2892 | */ | |
2893 | ||
e2761a11 | 2894 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2895 | { |
2896 | int result; | |
949854d0 | 2897 | unsigned seq; |
1da177e4 | 2898 | |
e2761a11 OH |
2899 | if (new_dentry == old_dentry) |
2900 | return 1; | |
2901 | ||
e2761a11 | 2902 | do { |
1da177e4 | 2903 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2904 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2905 | /* |
2906 | * Need rcu_readlock to protect against the d_parent trashing | |
2907 | * due to d_move | |
2908 | */ | |
2909 | rcu_read_lock(); | |
e2761a11 | 2910 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2911 | result = 1; |
e2761a11 OH |
2912 | else |
2913 | result = 0; | |
949854d0 | 2914 | rcu_read_unlock(); |
1da177e4 | 2915 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2916 | |
2917 | return result; | |
2918 | } | |
2919 | ||
2920 | void d_genocide(struct dentry *root) | |
2921 | { | |
949854d0 | 2922 | struct dentry *this_parent; |
1da177e4 | 2923 | struct list_head *next; |
949854d0 | 2924 | unsigned seq; |
58db63d0 | 2925 | int locked = 0; |
1da177e4 | 2926 | |
949854d0 | 2927 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2928 | again: |
2929 | this_parent = root; | |
2fd6b7f5 | 2930 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2931 | repeat: |
2932 | next = this_parent->d_subdirs.next; | |
2933 | resume: | |
2934 | while (next != &this_parent->d_subdirs) { | |
2935 | struct list_head *tmp = next; | |
5160ee6f | 2936 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2937 | next = tmp->next; |
949854d0 | 2938 | |
da502956 NP |
2939 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2940 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2941 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2942 | continue; |
da502956 | 2943 | } |
1da177e4 | 2944 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2945 | spin_unlock(&this_parent->d_lock); |
2946 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2947 | this_parent = dentry; |
2fd6b7f5 | 2948 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2949 | goto repeat; |
2950 | } | |
949854d0 NP |
2951 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2952 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2953 | dentry->d_count--; | |
2954 | } | |
b7ab39f6 | 2955 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2956 | } |
2957 | if (this_parent != root) { | |
c826cb7d | 2958 | struct dentry *child = this_parent; |
949854d0 NP |
2959 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2960 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2961 | this_parent->d_count--; | |
2962 | } | |
c826cb7d LT |
2963 | this_parent = try_to_ascend(this_parent, locked, seq); |
2964 | if (!this_parent) | |
949854d0 | 2965 | goto rename_retry; |
949854d0 | 2966 | next = child->d_u.d_child.next; |
1da177e4 LT |
2967 | goto resume; |
2968 | } | |
2fd6b7f5 | 2969 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2970 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2971 | goto rename_retry; |
58db63d0 NP |
2972 | if (locked) |
2973 | write_sequnlock(&rename_lock); | |
2974 | return; | |
2975 | ||
2976 | rename_retry: | |
8110e16d MS |
2977 | if (locked) |
2978 | goto again; | |
58db63d0 NP |
2979 | locked = 1; |
2980 | write_seqlock(&rename_lock); | |
2981 | goto again; | |
1da177e4 LT |
2982 | } |
2983 | ||
2984 | /** | |
2985 | * find_inode_number - check for dentry with name | |
2986 | * @dir: directory to check | |
2987 | * @name: Name to find. | |
2988 | * | |
2989 | * Check whether a dentry already exists for the given name, | |
2990 | * and return the inode number if it has an inode. Otherwise | |
2991 | * 0 is returned. | |
2992 | * | |
2993 | * This routine is used to post-process directory listings for | |
2994 | * filesystems using synthetic inode numbers, and is necessary | |
2995 | * to keep getcwd() working. | |
2996 | */ | |
2997 | ||
2998 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
2999 | { | |
3000 | struct dentry * dentry; | |
3001 | ino_t ino = 0; | |
3002 | ||
3e7e241f | 3003 | dentry = d_hash_and_lookup(dir, name); |
4f522a24 | 3004 | if (!IS_ERR_OR_NULL(dentry)) { |
1da177e4 LT |
3005 | if (dentry->d_inode) |
3006 | ino = dentry->d_inode->i_ino; | |
3007 | dput(dentry); | |
3008 | } | |
1da177e4 LT |
3009 | return ino; |
3010 | } | |
ec4f8605 | 3011 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
3012 | |
3013 | static __initdata unsigned long dhash_entries; | |
3014 | static int __init set_dhash_entries(char *str) | |
3015 | { | |
3016 | if (!str) | |
3017 | return 0; | |
3018 | dhash_entries = simple_strtoul(str, &str, 0); | |
3019 | return 1; | |
3020 | } | |
3021 | __setup("dhash_entries=", set_dhash_entries); | |
3022 | ||
3023 | static void __init dcache_init_early(void) | |
3024 | { | |
074b8517 | 3025 | unsigned int loop; |
1da177e4 LT |
3026 | |
3027 | /* If hashes are distributed across NUMA nodes, defer | |
3028 | * hash allocation until vmalloc space is available. | |
3029 | */ | |
3030 | if (hashdist) | |
3031 | return; | |
3032 | ||
3033 | dentry_hashtable = | |
3034 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3035 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3036 | dhash_entries, |
3037 | 13, | |
3038 | HASH_EARLY, | |
3039 | &d_hash_shift, | |
3040 | &d_hash_mask, | |
31fe62b9 | 3041 | 0, |
1da177e4 LT |
3042 | 0); |
3043 | ||
074b8517 | 3044 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3045 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3046 | } |
3047 | ||
74bf17cf | 3048 | static void __init dcache_init(void) |
1da177e4 | 3049 | { |
074b8517 | 3050 | unsigned int loop; |
1da177e4 LT |
3051 | |
3052 | /* | |
3053 | * A constructor could be added for stable state like the lists, | |
3054 | * but it is probably not worth it because of the cache nature | |
3055 | * of the dcache. | |
3056 | */ | |
0a31bd5f CL |
3057 | dentry_cache = KMEM_CACHE(dentry, |
3058 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3059 | |
3060 | /* Hash may have been set up in dcache_init_early */ | |
3061 | if (!hashdist) | |
3062 | return; | |
3063 | ||
3064 | dentry_hashtable = | |
3065 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3066 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3067 | dhash_entries, |
3068 | 13, | |
3069 | 0, | |
3070 | &d_hash_shift, | |
3071 | &d_hash_mask, | |
31fe62b9 | 3072 | 0, |
1da177e4 LT |
3073 | 0); |
3074 | ||
074b8517 | 3075 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3076 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3077 | } |
3078 | ||
3079 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3080 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3081 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3082 | |
1da177e4 LT |
3083 | EXPORT_SYMBOL(d_genocide); |
3084 | ||
1da177e4 LT |
3085 | void __init vfs_caches_init_early(void) |
3086 | { | |
3087 | dcache_init_early(); | |
3088 | inode_init_early(); | |
3089 | } | |
3090 | ||
3091 | void __init vfs_caches_init(unsigned long mempages) | |
3092 | { | |
3093 | unsigned long reserve; | |
3094 | ||
3095 | /* Base hash sizes on available memory, with a reserve equal to | |
3096 | 150% of current kernel size */ | |
3097 | ||
3098 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3099 | mempages -= reserve; | |
3100 | ||
3101 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3102 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3103 | |
74bf17cf DC |
3104 | dcache_init(); |
3105 | inode_init(); | |
1da177e4 | 3106 | files_init(mempages); |
74bf17cf | 3107 | mnt_init(); |
1da177e4 LT |
3108 | bdev_cache_init(); |
3109 | chrdev_init(); | |
3110 | } |