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 | 45 | * dcache->d_inode->i_lock protects: |
6637ecd3 | 46 | * - i_dentry, d_u.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 | |
6637ecd3 | 61 | * - d_u.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 | ||
1da177e4 LT |
218 | if (dname_external(dentry)) |
219 | kfree(dentry->d_name.name); | |
220 | kmem_cache_free(dentry_cache, dentry); | |
221 | } | |
222 | ||
223 | /* | |
b5c84bf6 | 224 | * no locks, please. |
1da177e4 LT |
225 | */ |
226 | static void d_free(struct dentry *dentry) | |
227 | { | |
6637ecd3 | 228 | WARN_ON(!hlist_unhashed(&dentry->d_u.d_alias)); |
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; | |
6637ecd3 | 267 | hlist_del_init(&dentry->d_u.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; | |
6637ecd3 | 291 | hlist_del_init(&dentry->d_u.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 | { |
5f03ac13 | 367 | __list_del_entry(&dentry->d_child); |
c83ce989 | 368 | /* |
5f03ac13 | 369 | * Inform ascending readers that we are no longer attached to the |
c83ce989 TM |
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 | ||
518309e0 AV |
523 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) |
524 | goto kill_it; | |
525 | ||
fb045adb | 526 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 527 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 528 | goto kill_it; |
1da177e4 | 529 | } |
265ac902 | 530 | |
1da177e4 LT |
531 | /* Unreachable? Get rid of it */ |
532 | if (d_unhashed(dentry)) | |
533 | goto kill_it; | |
265ac902 | 534 | |
39e3c955 | 535 | dentry->d_flags |= DCACHE_REFERENCED; |
a4633357 | 536 | dentry_lru_add(dentry); |
265ac902 | 537 | |
61f3dee4 NP |
538 | dentry->d_count--; |
539 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
540 | return; |
541 | ||
d52b9086 | 542 | kill_it: |
77812a1e | 543 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
544 | if (dentry) |
545 | goto repeat; | |
1da177e4 | 546 | } |
ec4f8605 | 547 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
548 | |
549 | /** | |
550 | * d_invalidate - invalidate a dentry | |
551 | * @dentry: dentry to invalidate | |
552 | * | |
553 | * Try to invalidate the dentry if it turns out to be | |
554 | * possible. If there are other dentries that can be | |
555 | * reached through this one we can't delete it and we | |
556 | * return -EBUSY. On success we return 0. | |
557 | * | |
558 | * no dcache lock. | |
559 | */ | |
560 | ||
561 | int d_invalidate(struct dentry * dentry) | |
562 | { | |
563 | /* | |
564 | * If it's already been dropped, return OK. | |
565 | */ | |
da502956 | 566 | spin_lock(&dentry->d_lock); |
1da177e4 | 567 | if (d_unhashed(dentry)) { |
da502956 | 568 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
569 | return 0; |
570 | } | |
571 | /* | |
572 | * Check whether to do a partial shrink_dcache | |
573 | * to get rid of unused child entries. | |
574 | */ | |
575 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 576 | spin_unlock(&dentry->d_lock); |
1da177e4 | 577 | shrink_dcache_parent(dentry); |
da502956 | 578 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
579 | } |
580 | ||
581 | /* | |
582 | * Somebody else still using it? | |
583 | * | |
584 | * If it's a directory, we can't drop it | |
585 | * for fear of somebody re-populating it | |
586 | * with children (even though dropping it | |
587 | * would make it unreachable from the root, | |
588 | * we might still populate it if it was a | |
589 | * working directory or similar). | |
50e69630 AV |
590 | * We also need to leave mountpoints alone, |
591 | * directory or not. | |
1da177e4 | 592 | */ |
50e69630 AV |
593 | if (dentry->d_count > 1 && dentry->d_inode) { |
594 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { | |
1da177e4 | 595 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
596 | return -EBUSY; |
597 | } | |
598 | } | |
599 | ||
600 | __d_drop(dentry); | |
601 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
602 | return 0; |
603 | } | |
ec4f8605 | 604 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 605 | |
b5c84bf6 | 606 | /* This must be called with d_lock held */ |
dc0474be | 607 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 608 | { |
b7ab39f6 | 609 | dentry->d_count++; |
23044507 NP |
610 | } |
611 | ||
dc0474be | 612 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 613 | { |
23044507 | 614 | spin_lock(&dentry->d_lock); |
dc0474be | 615 | __dget_dlock(dentry); |
23044507 | 616 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
617 | } |
618 | ||
b7ab39f6 NP |
619 | struct dentry *dget_parent(struct dentry *dentry) |
620 | { | |
621 | struct dentry *ret; | |
622 | ||
623 | repeat: | |
a734eb45 NP |
624 | /* |
625 | * Don't need rcu_dereference because we re-check it was correct under | |
626 | * the lock. | |
627 | */ | |
628 | rcu_read_lock(); | |
b7ab39f6 | 629 | ret = dentry->d_parent; |
a734eb45 NP |
630 | spin_lock(&ret->d_lock); |
631 | if (unlikely(ret != dentry->d_parent)) { | |
632 | spin_unlock(&ret->d_lock); | |
633 | rcu_read_unlock(); | |
b7ab39f6 NP |
634 | goto repeat; |
635 | } | |
a734eb45 | 636 | rcu_read_unlock(); |
b7ab39f6 NP |
637 | BUG_ON(!ret->d_count); |
638 | ret->d_count++; | |
639 | spin_unlock(&ret->d_lock); | |
b7ab39f6 NP |
640 | return ret; |
641 | } | |
642 | EXPORT_SYMBOL(dget_parent); | |
643 | ||
1da177e4 LT |
644 | /** |
645 | * d_find_alias - grab a hashed alias of inode | |
646 | * @inode: inode in question | |
32ba9c3f LT |
647 | * @want_discon: flag, used by d_splice_alias, to request |
648 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
649 | * |
650 | * If inode has a hashed alias, or is a directory and has any alias, | |
651 | * acquire the reference to alias and return it. Otherwise return NULL. | |
652 | * Notice that if inode is a directory there can be only one alias and | |
653 | * it can be unhashed only if it has no children, or if it is the root | |
654 | * of a filesystem. | |
655 | * | |
21c0d8fd | 656 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
657 | * any other hashed alias over that one unless @want_discon is set, |
658 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 659 | */ |
32ba9c3f | 660 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 661 | { |
da502956 | 662 | struct dentry *alias, *discon_alias; |
1da177e4 | 663 | |
da502956 NP |
664 | again: |
665 | discon_alias = NULL; | |
6637ecd3 | 666 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
da502956 | 667 | spin_lock(&alias->d_lock); |
1da177e4 | 668 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 669 | if (IS_ROOT(alias) && |
da502956 | 670 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 671 | discon_alias = alias; |
32ba9c3f | 672 | } else if (!want_discon) { |
dc0474be | 673 | __dget_dlock(alias); |
da502956 NP |
674 | spin_unlock(&alias->d_lock); |
675 | return alias; | |
676 | } | |
677 | } | |
678 | spin_unlock(&alias->d_lock); | |
679 | } | |
680 | if (discon_alias) { | |
681 | alias = discon_alias; | |
682 | spin_lock(&alias->d_lock); | |
683 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
684 | if (IS_ROOT(alias) && | |
685 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 686 | __dget_dlock(alias); |
da502956 | 687 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
688 | return alias; |
689 | } | |
690 | } | |
da502956 NP |
691 | spin_unlock(&alias->d_lock); |
692 | goto again; | |
1da177e4 | 693 | } |
da502956 | 694 | return NULL; |
1da177e4 LT |
695 | } |
696 | ||
da502956 | 697 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 698 | { |
214fda1f DH |
699 | struct dentry *de = NULL; |
700 | ||
b3d9b7a3 | 701 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 702 | spin_lock(&inode->i_lock); |
32ba9c3f | 703 | de = __d_find_alias(inode, 0); |
873feea0 | 704 | spin_unlock(&inode->i_lock); |
214fda1f | 705 | } |
1da177e4 LT |
706 | return de; |
707 | } | |
ec4f8605 | 708 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
709 | |
710 | /* | |
711 | * Try to kill dentries associated with this inode. | |
712 | * WARNING: you must own a reference to inode. | |
713 | */ | |
714 | void d_prune_aliases(struct inode *inode) | |
715 | { | |
0cdca3f9 | 716 | struct dentry *dentry; |
1da177e4 | 717 | restart: |
873feea0 | 718 | spin_lock(&inode->i_lock); |
6637ecd3 | 719 | hlist_for_each_entry(dentry, &inode->i_dentry, d_u.d_alias) { |
1da177e4 | 720 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 721 | if (!dentry->d_count) { |
dc0474be | 722 | __dget_dlock(dentry); |
1da177e4 LT |
723 | __d_drop(dentry); |
724 | spin_unlock(&dentry->d_lock); | |
873feea0 | 725 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
726 | dput(dentry); |
727 | goto restart; | |
728 | } | |
729 | spin_unlock(&dentry->d_lock); | |
730 | } | |
873feea0 | 731 | spin_unlock(&inode->i_lock); |
1da177e4 | 732 | } |
ec4f8605 | 733 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
734 | |
735 | /* | |
77812a1e NP |
736 | * Try to throw away a dentry - free the inode, dput the parent. |
737 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
738 | * Releases dentry->d_lock. | |
d702ccb3 | 739 | * |
77812a1e | 740 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 741 | */ |
77812a1e | 742 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 743 | __releases(dentry->d_lock) |
1da177e4 | 744 | { |
77812a1e | 745 | struct dentry *parent; |
d52b9086 | 746 | |
77812a1e | 747 | parent = dentry_kill(dentry, 0); |
d52b9086 | 748 | /* |
77812a1e NP |
749 | * If dentry_kill returns NULL, we have nothing more to do. |
750 | * if it returns the same dentry, trylocks failed. In either | |
751 | * case, just loop again. | |
752 | * | |
753 | * Otherwise, we need to prune ancestors too. This is necessary | |
754 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
755 | * is also expected to be beneficial in reducing dentry cache | |
756 | * fragmentation. | |
d52b9086 | 757 | */ |
77812a1e NP |
758 | if (!parent) |
759 | return; | |
760 | if (parent == dentry) | |
761 | return; | |
762 | ||
763 | /* Prune ancestors. */ | |
764 | dentry = parent; | |
d52b9086 | 765 | while (dentry) { |
b7ab39f6 | 766 | spin_lock(&dentry->d_lock); |
89e60548 NP |
767 | if (dentry->d_count > 1) { |
768 | dentry->d_count--; | |
769 | spin_unlock(&dentry->d_lock); | |
770 | return; | |
771 | } | |
77812a1e | 772 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 773 | } |
1da177e4 LT |
774 | } |
775 | ||
3049cfe2 | 776 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 777 | { |
da3bbdd4 | 778 | struct dentry *dentry; |
da3bbdd4 | 779 | |
ec33679d NP |
780 | rcu_read_lock(); |
781 | for (;;) { | |
ec33679d NP |
782 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
783 | if (&dentry->d_lru == list) | |
784 | break; /* empty */ | |
785 | spin_lock(&dentry->d_lock); | |
786 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
787 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
788 | continue; |
789 | } | |
790 | ||
1da177e4 LT |
791 | /* |
792 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
793 | * the LRU because of laziness during lookup. Do not free |
794 | * it - just keep it off the LRU list. | |
1da177e4 | 795 | */ |
b7ab39f6 | 796 | if (dentry->d_count) { |
ec33679d | 797 | dentry_lru_del(dentry); |
da3bbdd4 | 798 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
799 | continue; |
800 | } | |
ec33679d | 801 | |
ec33679d | 802 | rcu_read_unlock(); |
77812a1e NP |
803 | |
804 | try_prune_one_dentry(dentry); | |
805 | ||
ec33679d | 806 | rcu_read_lock(); |
da3bbdd4 | 807 | } |
ec33679d | 808 | rcu_read_unlock(); |
3049cfe2 CH |
809 | } |
810 | ||
811 | /** | |
b48f03b3 DC |
812 | * prune_dcache_sb - shrink the dcache |
813 | * @sb: superblock | |
814 | * @count: number of entries to try to free | |
815 | * | |
816 | * Attempt to shrink the superblock dcache LRU by @count entries. This is | |
817 | * done when we need more memory an called from the superblock shrinker | |
818 | * function. | |
3049cfe2 | 819 | * |
b48f03b3 DC |
820 | * This function may fail to free any resources if all the dentries are in |
821 | * use. | |
3049cfe2 | 822 | */ |
b48f03b3 | 823 | void prune_dcache_sb(struct super_block *sb, int count) |
3049cfe2 | 824 | { |
3049cfe2 CH |
825 | struct dentry *dentry; |
826 | LIST_HEAD(referenced); | |
827 | LIST_HEAD(tmp); | |
3049cfe2 | 828 | |
23044507 NP |
829 | relock: |
830 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
831 | while (!list_empty(&sb->s_dentry_lru)) { |
832 | dentry = list_entry(sb->s_dentry_lru.prev, | |
833 | struct dentry, d_lru); | |
834 | BUG_ON(dentry->d_sb != sb); | |
835 | ||
23044507 NP |
836 | if (!spin_trylock(&dentry->d_lock)) { |
837 | spin_unlock(&dcache_lru_lock); | |
838 | cpu_relax(); | |
839 | goto relock; | |
840 | } | |
841 | ||
b48f03b3 | 842 | if (dentry->d_flags & DCACHE_REFERENCED) { |
23044507 NP |
843 | dentry->d_flags &= ~DCACHE_REFERENCED; |
844 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 845 | spin_unlock(&dentry->d_lock); |
23044507 NP |
846 | } else { |
847 | list_move_tail(&dentry->d_lru, &tmp); | |
eaf5f907 | 848 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
23044507 | 849 | spin_unlock(&dentry->d_lock); |
b0d40c92 | 850 | if (!--count) |
23044507 | 851 | break; |
3049cfe2 | 852 | } |
ec33679d | 853 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 854 | } |
da3bbdd4 KM |
855 | if (!list_empty(&referenced)) |
856 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 857 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
858 | |
859 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
860 | } |
861 | ||
1da177e4 LT |
862 | /** |
863 | * shrink_dcache_sb - shrink dcache for a superblock | |
864 | * @sb: superblock | |
865 | * | |
3049cfe2 CH |
866 | * Shrink the dcache for the specified super block. This is used to free |
867 | * the dcache before unmounting a file system. | |
1da177e4 | 868 | */ |
3049cfe2 | 869 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 870 | { |
3049cfe2 CH |
871 | LIST_HEAD(tmp); |
872 | ||
23044507 | 873 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
874 | while (!list_empty(&sb->s_dentry_lru)) { |
875 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 876 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 877 | shrink_dentry_list(&tmp); |
ec33679d | 878 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 879 | } |
23044507 | 880 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 881 | } |
ec4f8605 | 882 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 883 | |
c636ebdb DH |
884 | /* |
885 | * destroy a single subtree of dentries for unmount | |
886 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
887 | * locking | |
888 | */ | |
889 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
890 | { | |
891 | struct dentry *parent; | |
892 | ||
893 | BUG_ON(!IS_ROOT(dentry)); | |
894 | ||
c636ebdb DH |
895 | for (;;) { |
896 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 897 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb | 898 | dentry = list_entry(dentry->d_subdirs.next, |
6637ecd3 | 899 | struct dentry, d_child); |
c636ebdb DH |
900 | |
901 | /* consume the dentries from this leaf up through its parents | |
902 | * until we find one with children or run out altogether */ | |
903 | do { | |
904 | struct inode *inode; | |
905 | ||
f0023bc6 | 906 | /* |
61572bb1 | 907 | * inform the fs that this dentry is about to be |
f0023bc6 SW |
908 | * unhashed and destroyed. |
909 | */ | |
61572bb1 YZ |
910 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
911 | dentry->d_op->d_prune(dentry); | |
912 | ||
913 | dentry_lru_del(dentry); | |
43c1c9cd DH |
914 | __d_shrink(dentry); |
915 | ||
b7ab39f6 | 916 | if (dentry->d_count != 0) { |
c636ebdb DH |
917 | printk(KERN_ERR |
918 | "BUG: Dentry %p{i=%lx,n=%s}" | |
919 | " still in use (%d)" | |
920 | " [unmount of %s %s]\n", | |
921 | dentry, | |
922 | dentry->d_inode ? | |
923 | dentry->d_inode->i_ino : 0UL, | |
924 | dentry->d_name.name, | |
b7ab39f6 | 925 | dentry->d_count, |
c636ebdb DH |
926 | dentry->d_sb->s_type->name, |
927 | dentry->d_sb->s_id); | |
928 | BUG(); | |
929 | } | |
930 | ||
2fd6b7f5 | 931 | if (IS_ROOT(dentry)) { |
c636ebdb | 932 | parent = NULL; |
6637ecd3 | 933 | list_del(&dentry->d_child); |
2fd6b7f5 | 934 | } else { |
871c0067 | 935 | parent = dentry->d_parent; |
b7ab39f6 | 936 | parent->d_count--; |
6637ecd3 | 937 | list_del(&dentry->d_child); |
871c0067 | 938 | } |
c636ebdb | 939 | |
c636ebdb DH |
940 | inode = dentry->d_inode; |
941 | if (inode) { | |
942 | dentry->d_inode = NULL; | |
6637ecd3 | 943 | hlist_del_init(&dentry->d_u.d_alias); |
c636ebdb DH |
944 | if (dentry->d_op && dentry->d_op->d_iput) |
945 | dentry->d_op->d_iput(dentry, inode); | |
946 | else | |
947 | iput(inode); | |
948 | } | |
949 | ||
950 | d_free(dentry); | |
951 | ||
952 | /* finished when we fall off the top of the tree, | |
953 | * otherwise we ascend to the parent and move to the | |
954 | * next sibling if there is one */ | |
955 | if (!parent) | |
312d3ca8 | 956 | return; |
c636ebdb | 957 | dentry = parent; |
c636ebdb DH |
958 | } while (list_empty(&dentry->d_subdirs)); |
959 | ||
960 | dentry = list_entry(dentry->d_subdirs.next, | |
6637ecd3 | 961 | struct dentry, d_child); |
c636ebdb DH |
962 | } |
963 | } | |
964 | ||
965 | /* | |
966 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 967 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
968 | * - the superblock is detached from all mountings and open files, so the |
969 | * dentry trees will not be rearranged by the VFS | |
970 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
971 | * any dentries belonging to this superblock that it comes across | |
972 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
973 | * in this superblock | |
974 | */ | |
975 | void shrink_dcache_for_umount(struct super_block *sb) | |
976 | { | |
977 | struct dentry *dentry; | |
978 | ||
979 | if (down_read_trylock(&sb->s_umount)) | |
980 | BUG(); | |
981 | ||
982 | dentry = sb->s_root; | |
983 | sb->s_root = NULL; | |
b7ab39f6 | 984 | dentry->d_count--; |
c636ebdb DH |
985 | shrink_dcache_for_umount_subtree(dentry); |
986 | ||
ceb5bdc2 NP |
987 | while (!hlist_bl_empty(&sb->s_anon)) { |
988 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
989 | shrink_dcache_for_umount_subtree(dentry); |
990 | } | |
991 | } | |
992 | ||
1da177e4 LT |
993 | /* |
994 | * Search for at least 1 mount point in the dentry's subdirs. | |
995 | * We descend to the next level whenever the d_subdirs | |
996 | * list is non-empty and continue searching. | |
997 | */ | |
998 | ||
999 | /** | |
1000 | * have_submounts - check for mounts over a dentry | |
1001 | * @parent: dentry to check. | |
1002 | * | |
1003 | * Return true if the parent or its subdirectories contain | |
1004 | * a mount point | |
1005 | */ | |
1da177e4 LT |
1006 | int have_submounts(struct dentry *parent) |
1007 | { | |
949854d0 | 1008 | struct dentry *this_parent; |
1da177e4 | 1009 | struct list_head *next; |
949854d0 | 1010 | unsigned seq; |
58db63d0 | 1011 | int locked = 0; |
949854d0 | 1012 | |
949854d0 | 1013 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1014 | again: |
1015 | this_parent = parent; | |
1da177e4 | 1016 | |
1da177e4 LT |
1017 | if (d_mountpoint(parent)) |
1018 | goto positive; | |
2fd6b7f5 | 1019 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1020 | repeat: |
1021 | next = this_parent->d_subdirs.next; | |
1022 | resume: | |
1023 | while (next != &this_parent->d_subdirs) { | |
1024 | struct list_head *tmp = next; | |
6637ecd3 | 1025 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 1026 | next = tmp->next; |
2fd6b7f5 NP |
1027 | |
1028 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1029 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1030 | if (d_mountpoint(dentry)) { |
1031 | spin_unlock(&dentry->d_lock); | |
1032 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1033 | goto positive; |
2fd6b7f5 | 1034 | } |
1da177e4 | 1035 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1036 | spin_unlock(&this_parent->d_lock); |
1037 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1038 | this_parent = dentry; |
2fd6b7f5 | 1039 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1040 | goto repeat; |
1041 | } | |
2fd6b7f5 | 1042 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1043 | } |
1044 | /* | |
1045 | * All done at this level ... ascend and resume the search. | |
1046 | */ | |
5f03ac13 AV |
1047 | rcu_read_lock(); |
1048 | ascend: | |
1da177e4 | 1049 | if (this_parent != parent) { |
c826cb7d | 1050 | struct dentry *child = this_parent; |
5f03ac13 AV |
1051 | this_parent = child->d_parent; |
1052 | ||
1053 | spin_unlock(&child->d_lock); | |
1054 | spin_lock(&this_parent->d_lock); | |
1055 | ||
1056 | /* might go back up the wrong parent if we have had a rename. */ | |
1057 | if (!locked && read_seqretry(&rename_lock, seq)) | |
949854d0 | 1058 | goto rename_retry; |
b93e0acc JR |
1059 | /* go into the first sibling still alive */ |
1060 | do { | |
1061 | next = child->d_child.next; | |
5f03ac13 AV |
1062 | if (next == &this_parent->d_subdirs) |
1063 | goto ascend; | |
1064 | child = list_entry(next, struct dentry, d_child); | |
b93e0acc | 1065 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
5f03ac13 | 1066 | rcu_read_unlock(); |
1da177e4 LT |
1067 | goto resume; |
1068 | } | |
58db63d0 | 1069 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1070 | goto rename_retry; |
5f03ac13 AV |
1071 | spin_unlock(&this_parent->d_lock); |
1072 | rcu_read_unlock(); | |
58db63d0 NP |
1073 | if (locked) |
1074 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1075 | return 0; /* No mount points found in tree */ |
1076 | positive: | |
58db63d0 | 1077 | if (!locked && read_seqretry(&rename_lock, seq)) |
c190d250 | 1078 | goto rename_retry_unlocked; |
58db63d0 NP |
1079 | if (locked) |
1080 | write_sequnlock(&rename_lock); | |
1da177e4 | 1081 | return 1; |
58db63d0 NP |
1082 | |
1083 | rename_retry: | |
5f03ac13 AV |
1084 | spin_unlock(&this_parent->d_lock); |
1085 | rcu_read_unlock(); | |
8110e16d MS |
1086 | if (locked) |
1087 | goto again; | |
c190d250 | 1088 | rename_retry_unlocked: |
58db63d0 NP |
1089 | locked = 1; |
1090 | write_seqlock(&rename_lock); | |
1091 | goto again; | |
1da177e4 | 1092 | } |
ec4f8605 | 1093 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1094 | |
1095 | /* | |
fd517909 | 1096 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1097 | * and move any unused dentries to the end of the unused |
1098 | * list for prune_dcache(). We descend to the next level | |
1099 | * whenever the d_subdirs list is non-empty and continue | |
1100 | * searching. | |
1101 | * | |
1102 | * It returns zero iff there are no unused children, | |
1103 | * otherwise it returns the number of children moved to | |
1104 | * the end of the unused list. This may not be the total | |
1105 | * number of unused children, because select_parent can | |
1106 | * drop the lock and return early due to latency | |
1107 | * constraints. | |
1108 | */ | |
b48f03b3 | 1109 | static int select_parent(struct dentry *parent, struct list_head *dispose) |
1da177e4 | 1110 | { |
949854d0 | 1111 | struct dentry *this_parent; |
1da177e4 | 1112 | struct list_head *next; |
949854d0 | 1113 | unsigned seq; |
1da177e4 | 1114 | int found = 0; |
58db63d0 | 1115 | int locked = 0; |
1da177e4 | 1116 | |
949854d0 | 1117 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1118 | again: |
1119 | this_parent = parent; | |
2fd6b7f5 | 1120 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1121 | repeat: |
1122 | next = this_parent->d_subdirs.next; | |
1123 | resume: | |
1124 | while (next != &this_parent->d_subdirs) { | |
1125 | struct list_head *tmp = next; | |
6637ecd3 | 1126 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 LT |
1127 | next = tmp->next; |
1128 | ||
2fd6b7f5 | 1129 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1130 | |
b48f03b3 DC |
1131 | /* |
1132 | * move only zero ref count dentries to the dispose list. | |
eaf5f907 MS |
1133 | * |
1134 | * Those which are presently on the shrink list, being processed | |
1135 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1136 | * loop in shrink_dcache_parent() might not make any progress | |
1137 | * and loop forever. | |
1da177e4 | 1138 | */ |
eaf5f907 MS |
1139 | if (dentry->d_count) { |
1140 | dentry_lru_del(dentry); | |
1141 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
b48f03b3 | 1142 | dentry_lru_move_list(dentry, dispose); |
eaf5f907 | 1143 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
1da177e4 LT |
1144 | found++; |
1145 | } | |
1da177e4 LT |
1146 | /* |
1147 | * We can return to the caller if we have found some (this | |
1148 | * ensures forward progress). We'll be coming back to find | |
1149 | * the rest. | |
1150 | */ | |
2fd6b7f5 NP |
1151 | if (found && need_resched()) { |
1152 | spin_unlock(&dentry->d_lock); | |
c190d250 | 1153 | rcu_read_lock(); |
1da177e4 | 1154 | goto out; |
2fd6b7f5 | 1155 | } |
1da177e4 LT |
1156 | |
1157 | /* | |
1158 | * Descend a level if the d_subdirs list is non-empty. | |
1159 | */ | |
1160 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1161 | spin_unlock(&this_parent->d_lock); |
1162 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1163 | this_parent = dentry; |
2fd6b7f5 | 1164 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1165 | goto repeat; |
1166 | } | |
2fd6b7f5 NP |
1167 | |
1168 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1169 | } |
1170 | /* | |
1171 | * All done at this level ... ascend and resume the search. | |
1172 | */ | |
5f03ac13 AV |
1173 | rcu_read_lock(); |
1174 | ascend: | |
1da177e4 | 1175 | if (this_parent != parent) { |
c826cb7d | 1176 | struct dentry *child = this_parent; |
5f03ac13 AV |
1177 | this_parent = child->d_parent; |
1178 | ||
1179 | spin_unlock(&child->d_lock); | |
1180 | spin_lock(&this_parent->d_lock); | |
1181 | ||
1182 | /* might go back up the wrong parent if we have had a rename. */ | |
1183 | if (!locked && read_seqretry(&rename_lock, seq)) | |
949854d0 | 1184 | goto rename_retry; |
bf6aa269 AV |
1185 | /* go into the first sibling still alive */ |
1186 | do { | |
1187 | next = child->d_child.next; | |
5f03ac13 AV |
1188 | if (next == &this_parent->d_subdirs) |
1189 | goto ascend; | |
1190 | child = list_entry(next, struct dentry, d_child); | |
bf6aa269 | 1191 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
5f03ac13 | 1192 | rcu_read_unlock(); |
1da177e4 LT |
1193 | goto resume; |
1194 | } | |
1195 | out: | |
58db63d0 | 1196 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1197 | goto rename_retry; |
5f03ac13 AV |
1198 | spin_unlock(&this_parent->d_lock); |
1199 | rcu_read_unlock(); | |
58db63d0 NP |
1200 | if (locked) |
1201 | write_sequnlock(&rename_lock); | |
1da177e4 | 1202 | return found; |
58db63d0 NP |
1203 | |
1204 | rename_retry: | |
5f03ac13 AV |
1205 | spin_unlock(&this_parent->d_lock); |
1206 | rcu_read_unlock(); | |
58db63d0 NP |
1207 | if (found) |
1208 | return found; | |
8110e16d MS |
1209 | if (locked) |
1210 | goto again; | |
58db63d0 NP |
1211 | locked = 1; |
1212 | write_seqlock(&rename_lock); | |
1213 | goto again; | |
1da177e4 LT |
1214 | } |
1215 | ||
1216 | /** | |
1217 | * shrink_dcache_parent - prune dcache | |
1218 | * @parent: parent of entries to prune | |
1219 | * | |
1220 | * Prune the dcache to remove unused children of the parent dentry. | |
1221 | */ | |
1da177e4 LT |
1222 | void shrink_dcache_parent(struct dentry * parent) |
1223 | { | |
b48f03b3 | 1224 | LIST_HEAD(dispose); |
1da177e4 LT |
1225 | int found; |
1226 | ||
421348f1 | 1227 | while ((found = select_parent(parent, &dispose)) != 0) { |
b48f03b3 | 1228 | shrink_dentry_list(&dispose); |
421348f1 GT |
1229 | cond_resched(); |
1230 | } | |
1da177e4 | 1231 | } |
ec4f8605 | 1232 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1233 | |
1da177e4 | 1234 | /** |
a4464dbc AV |
1235 | * __d_alloc - allocate a dcache entry |
1236 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1237 | * @name: qstr of the name |
1238 | * | |
1239 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1240 | * available. On a success the dentry is returned. The name passed in is | |
1241 | * copied and the copy passed in may be reused after this call. | |
1242 | */ | |
1243 | ||
a4464dbc | 1244 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1245 | { |
1246 | struct dentry *dentry; | |
1247 | char *dname; | |
1248 | ||
e12ba74d | 1249 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1250 | if (!dentry) |
1251 | return NULL; | |
1252 | ||
6326c71f LT |
1253 | /* |
1254 | * We guarantee that the inline name is always NUL-terminated. | |
1255 | * This way the memcpy() done by the name switching in rename | |
1256 | * will still always have a NUL at the end, even if we might | |
1257 | * be overwriting an internal NUL character | |
1258 | */ | |
1259 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1260 | if (name->len > DNAME_INLINE_LEN-1) { |
1261 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1262 | if (!dname) { | |
1263 | kmem_cache_free(dentry_cache, dentry); | |
1264 | return NULL; | |
1265 | } | |
1266 | } else { | |
1267 | dname = dentry->d_iname; | |
1268 | } | |
1da177e4 LT |
1269 | |
1270 | dentry->d_name.len = name->len; | |
1271 | dentry->d_name.hash = name->hash; | |
1272 | memcpy(dname, name->name, name->len); | |
1273 | dname[name->len] = 0; | |
1274 | ||
6326c71f LT |
1275 | /* Make sure we always see the terminating NUL character */ |
1276 | smp_wmb(); | |
1277 | dentry->d_name.name = dname; | |
1278 | ||
b7ab39f6 | 1279 | dentry->d_count = 1; |
dea3667b | 1280 | dentry->d_flags = 0; |
1da177e4 | 1281 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1282 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1283 | dentry->d_inode = NULL; |
a4464dbc AV |
1284 | dentry->d_parent = dentry; |
1285 | dentry->d_sb = sb; | |
1da177e4 LT |
1286 | dentry->d_op = NULL; |
1287 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1288 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1289 | INIT_LIST_HEAD(&dentry->d_lru); |
1290 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
6637ecd3 AV |
1291 | INIT_HLIST_NODE(&dentry->d_u.d_alias); |
1292 | INIT_LIST_HEAD(&dentry->d_child); | |
a4464dbc | 1293 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1294 | |
3e880fb5 | 1295 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1296 | |
1da177e4 LT |
1297 | return dentry; |
1298 | } | |
a4464dbc AV |
1299 | |
1300 | /** | |
1301 | * d_alloc - allocate a dcache entry | |
1302 | * @parent: parent of entry to allocate | |
1303 | * @name: qstr of the name | |
1304 | * | |
1305 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1306 | * available. On a success the dentry is returned. The name passed in is | |
1307 | * copied and the copy passed in may be reused after this call. | |
1308 | */ | |
1309 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1310 | { | |
1311 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1312 | if (!dentry) | |
1313 | return NULL; | |
742d555a | 1314 | dentry->d_flags |= DCACHE_RCUACCESS; |
a4464dbc AV |
1315 | spin_lock(&parent->d_lock); |
1316 | /* | |
1317 | * don't need child lock because it is not subject | |
1318 | * to concurrency here | |
1319 | */ | |
1320 | __dget_dlock(parent); | |
1321 | dentry->d_parent = parent; | |
6637ecd3 | 1322 | list_add(&dentry->d_child, &parent->d_subdirs); |
a4464dbc AV |
1323 | spin_unlock(&parent->d_lock); |
1324 | ||
1325 | return dentry; | |
1326 | } | |
ec4f8605 | 1327 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1328 | |
4b936885 NP |
1329 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1330 | { | |
a4464dbc AV |
1331 | struct dentry *dentry = __d_alloc(sb, name); |
1332 | if (dentry) | |
4b936885 | 1333 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1334 | return dentry; |
1335 | } | |
1336 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1337 | ||
1da177e4 LT |
1338 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1339 | { | |
1340 | struct qstr q; | |
1341 | ||
1342 | q.name = name; | |
1343 | q.len = strlen(name); | |
1344 | q.hash = full_name_hash(q.name, q.len); | |
1345 | return d_alloc(parent, &q); | |
1346 | } | |
ef26ca97 | 1347 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1348 | |
fb045adb NP |
1349 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1350 | { | |
6f7f7caa LT |
1351 | WARN_ON_ONCE(dentry->d_op); |
1352 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1353 | DCACHE_OP_COMPARE | |
1354 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1355 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1356 | DCACHE_OP_DELETE )); |
1357 | dentry->d_op = op; | |
1358 | if (!op) | |
1359 | return; | |
1360 | if (op->d_hash) | |
1361 | dentry->d_flags |= DCACHE_OP_HASH; | |
1362 | if (op->d_compare) | |
1363 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1364 | if (op->d_revalidate) | |
1365 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1366 | if (op->d_weak_revalidate) |
1367 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1368 | if (op->d_delete) |
1369 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1370 | if (op->d_prune) |
1371 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1372 | |
1373 | } | |
1374 | EXPORT_SYMBOL(d_set_d_op); | |
1375 | ||
360da900 OH |
1376 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1377 | { | |
b23fb0a6 | 1378 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1379 | if (inode) { |
1380 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1381 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
6637ecd3 | 1382 | hlist_add_head(&dentry->d_u.d_alias, &inode->i_dentry); |
9875cf80 | 1383 | } |
360da900 | 1384 | dentry->d_inode = inode; |
31e6b01f | 1385 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1386 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1387 | fsnotify_d_instantiate(dentry, inode); |
1388 | } | |
1389 | ||
1da177e4 LT |
1390 | /** |
1391 | * d_instantiate - fill in inode information for a dentry | |
1392 | * @entry: dentry to complete | |
1393 | * @inode: inode to attach to this dentry | |
1394 | * | |
1395 | * Fill in inode information in the entry. | |
1396 | * | |
1397 | * This turns negative dentries into productive full members | |
1398 | * of society. | |
1399 | * | |
1400 | * NOTE! This assumes that the inode count has been incremented | |
1401 | * (or otherwise set) by the caller to indicate that it is now | |
1402 | * in use by the dcache. | |
1403 | */ | |
1404 | ||
1405 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1406 | { | |
6637ecd3 | 1407 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
873feea0 NP |
1408 | if (inode) |
1409 | spin_lock(&inode->i_lock); | |
360da900 | 1410 | __d_instantiate(entry, inode); |
873feea0 NP |
1411 | if (inode) |
1412 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1413 | security_d_instantiate(entry, inode); |
1414 | } | |
ec4f8605 | 1415 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1416 | |
1417 | /** | |
1418 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1419 | * @entry: dentry to instantiate | |
1420 | * @inode: inode to attach to this dentry | |
1421 | * | |
1422 | * Fill in inode information in the entry. On success, it returns NULL. | |
1423 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1424 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1425 | * |
1426 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1427 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1428 | * |
1429 | * This also assumes that the inode count has been incremented | |
1430 | * (or otherwise set) by the caller to indicate that it is now | |
1431 | * in use by the dcache. | |
1da177e4 | 1432 | */ |
770bfad8 DH |
1433 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1434 | struct inode *inode) | |
1da177e4 LT |
1435 | { |
1436 | struct dentry *alias; | |
1437 | int len = entry->d_name.len; | |
1438 | const char *name = entry->d_name.name; | |
1439 | unsigned int hash = entry->d_name.hash; | |
1440 | ||
770bfad8 | 1441 | if (!inode) { |
360da900 | 1442 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1443 | return NULL; |
1444 | } | |
1445 | ||
6637ecd3 | 1446 | hlist_for_each_entry(alias, &inode->i_dentry, d_u.d_alias) { |
9abca360 NP |
1447 | /* |
1448 | * Don't need alias->d_lock here, because aliases with | |
1449 | * d_parent == entry->d_parent are not subject to name or | |
1450 | * parent changes, because the parent inode i_mutex is held. | |
1451 | */ | |
12f8ad4b | 1452 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1453 | continue; |
1454 | if (alias->d_parent != entry->d_parent) | |
1455 | continue; | |
ee983e89 LT |
1456 | if (alias->d_name.len != len) |
1457 | continue; | |
12f8ad4b | 1458 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1459 | continue; |
dc0474be | 1460 | __dget(alias); |
1da177e4 LT |
1461 | return alias; |
1462 | } | |
770bfad8 | 1463 | |
360da900 | 1464 | __d_instantiate(entry, inode); |
1da177e4 LT |
1465 | return NULL; |
1466 | } | |
770bfad8 DH |
1467 | |
1468 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1469 | { | |
1470 | struct dentry *result; | |
1471 | ||
6637ecd3 | 1472 | BUG_ON(!hlist_unhashed(&entry->d_u.d_alias)); |
770bfad8 | 1473 | |
873feea0 NP |
1474 | if (inode) |
1475 | spin_lock(&inode->i_lock); | |
770bfad8 | 1476 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1477 | if (inode) |
1478 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1479 | |
1480 | if (!result) { | |
1481 | security_d_instantiate(entry, inode); | |
1482 | return NULL; | |
1483 | } | |
1484 | ||
1485 | BUG_ON(!d_unhashed(result)); | |
1486 | iput(inode); | |
1487 | return result; | |
1488 | } | |
1489 | ||
1da177e4 LT |
1490 | EXPORT_SYMBOL(d_instantiate_unique); |
1491 | ||
adc0e91a AV |
1492 | struct dentry *d_make_root(struct inode *root_inode) |
1493 | { | |
1494 | struct dentry *res = NULL; | |
1495 | ||
1496 | if (root_inode) { | |
26fe5750 | 1497 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1498 | |
1499 | res = __d_alloc(root_inode->i_sb, &name); | |
1500 | if (res) | |
1501 | d_instantiate(res, root_inode); | |
1502 | else | |
1503 | iput(root_inode); | |
1504 | } | |
1505 | return res; | |
1506 | } | |
1507 | EXPORT_SYMBOL(d_make_root); | |
1508 | ||
d891eedb BF |
1509 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1510 | { | |
1511 | struct dentry *alias; | |
1512 | ||
b3d9b7a3 | 1513 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1514 | return NULL; |
6637ecd3 | 1515 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_u.d_alias); |
d891eedb BF |
1516 | __dget(alias); |
1517 | return alias; | |
1518 | } | |
1519 | ||
46f72b34 SW |
1520 | /** |
1521 | * d_find_any_alias - find any alias for a given inode | |
1522 | * @inode: inode to find an alias for | |
1523 | * | |
1524 | * If any aliases exist for the given inode, take and return a | |
1525 | * reference for one of them. If no aliases exist, return %NULL. | |
1526 | */ | |
1527 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1528 | { |
1529 | struct dentry *de; | |
1530 | ||
1531 | spin_lock(&inode->i_lock); | |
1532 | de = __d_find_any_alias(inode); | |
1533 | spin_unlock(&inode->i_lock); | |
1534 | return de; | |
1535 | } | |
46f72b34 | 1536 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1537 | |
4ea3ada2 CH |
1538 | /** |
1539 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1540 | * @inode: inode to allocate the dentry for | |
1541 | * | |
1542 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1543 | * similar open by handle operations. The returned dentry may be anonymous, | |
1544 | * or may have a full name (if the inode was already in the cache). | |
1545 | * | |
1546 | * When called on a directory inode, we must ensure that the inode only ever | |
1547 | * has one dentry. If a dentry is found, that is returned instead of | |
1548 | * allocating a new one. | |
1549 | * | |
1550 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1551 | * to the dentry. In case of an error the reference on the inode is released. |
1552 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1553 | * be passed in and will be the error will be propagate to the return value, | |
1554 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1555 | */ |
1556 | struct dentry *d_obtain_alias(struct inode *inode) | |
1557 | { | |
b911a6bd | 1558 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1559 | struct dentry *tmp; |
1560 | struct dentry *res; | |
4ea3ada2 CH |
1561 | |
1562 | if (!inode) | |
44003728 | 1563 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1564 | if (IS_ERR(inode)) |
1565 | return ERR_CAST(inode); | |
1566 | ||
d891eedb | 1567 | res = d_find_any_alias(inode); |
9308a612 CH |
1568 | if (res) |
1569 | goto out_iput; | |
1570 | ||
a4464dbc | 1571 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1572 | if (!tmp) { |
1573 | res = ERR_PTR(-ENOMEM); | |
1574 | goto out_iput; | |
4ea3ada2 | 1575 | } |
b5c84bf6 | 1576 | |
873feea0 | 1577 | spin_lock(&inode->i_lock); |
d891eedb | 1578 | res = __d_find_any_alias(inode); |
9308a612 | 1579 | if (res) { |
873feea0 | 1580 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1581 | dput(tmp); |
1582 | goto out_iput; | |
1583 | } | |
1584 | ||
1585 | /* attach a disconnected dentry */ | |
1586 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1587 | tmp->d_inode = inode; |
1588 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
6637ecd3 | 1589 | hlist_add_head(&tmp->d_u.d_alias, &inode->i_dentry); |
1879fd6a | 1590 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1591 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1592 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1593 | spin_unlock(&tmp->d_lock); |
873feea0 | 1594 | spin_unlock(&inode->i_lock); |
24ff6663 | 1595 | security_d_instantiate(tmp, inode); |
9308a612 | 1596 | |
9308a612 CH |
1597 | return tmp; |
1598 | ||
1599 | out_iput: | |
24ff6663 JB |
1600 | if (res && !IS_ERR(res)) |
1601 | security_d_instantiate(res, inode); | |
9308a612 CH |
1602 | iput(inode); |
1603 | return res; | |
4ea3ada2 | 1604 | } |
adc48720 | 1605 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1606 | |
1607 | /** | |
1608 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1609 | * @inode: the inode which may have a disconnected dentry | |
1610 | * @dentry: a negative dentry which we want to point to the inode. | |
1611 | * | |
1612 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1613 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1614 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1615 | * | |
1616 | * This is needed in the lookup routine of any filesystem that is exportable | |
1617 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1618 | * | |
1619 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1620 | * is returned. This matches the expected return value of ->lookup. | |
1621 | * | |
1622 | */ | |
1623 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1624 | { | |
1625 | struct dentry *new = NULL; | |
1626 | ||
a9049376 AV |
1627 | if (IS_ERR(inode)) |
1628 | return ERR_CAST(inode); | |
1629 | ||
21c0d8fd | 1630 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1631 | spin_lock(&inode->i_lock); |
32ba9c3f | 1632 | new = __d_find_alias(inode, 1); |
1da177e4 | 1633 | if (new) { |
32ba9c3f | 1634 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1635 | spin_unlock(&inode->i_lock); |
1da177e4 | 1636 | security_d_instantiate(new, inode); |
1da177e4 LT |
1637 | d_move(new, dentry); |
1638 | iput(inode); | |
1639 | } else { | |
873feea0 | 1640 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1641 | __d_instantiate(dentry, inode); |
873feea0 | 1642 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1643 | security_d_instantiate(dentry, inode); |
1644 | d_rehash(dentry); | |
1645 | } | |
1646 | } else | |
1647 | d_add(dentry, inode); | |
1648 | return new; | |
1649 | } | |
ec4f8605 | 1650 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1651 | |
9403540c BN |
1652 | /** |
1653 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1654 | * @inode: the inode case-insensitive lookup has found | |
1655 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1656 | * @name: the case-exact name to be associated with the returned dentry | |
1657 | * | |
1658 | * This is to avoid filling the dcache with case-insensitive names to the | |
1659 | * same inode, only the actual correct case is stored in the dcache for | |
1660 | * case-insensitive filesystems. | |
1661 | * | |
1662 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1663 | * already exists in in the dcache, use it and return it. | |
1664 | * | |
1665 | * If no entry exists with the exact case name, allocate new dentry with | |
1666 | * the exact case, and return the spliced entry. | |
1667 | */ | |
e45b590b | 1668 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1669 | struct qstr *name) |
1670 | { | |
9403540c BN |
1671 | struct dentry *found; |
1672 | struct dentry *new; | |
1673 | ||
b6520c81 CH |
1674 | /* |
1675 | * First check if a dentry matching the name already exists, | |
1676 | * if not go ahead and create it now. | |
1677 | */ | |
9403540c | 1678 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1679 | if (unlikely(IS_ERR(found))) |
1680 | goto err_out; | |
9403540c BN |
1681 | if (!found) { |
1682 | new = d_alloc(dentry->d_parent, name); | |
1683 | if (!new) { | |
4f522a24 | 1684 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1685 | goto err_out; |
1686 | } | |
b6520c81 | 1687 | |
9403540c BN |
1688 | found = d_splice_alias(inode, new); |
1689 | if (found) { | |
1690 | dput(new); | |
1691 | return found; | |
1692 | } | |
1693 | return new; | |
1694 | } | |
b6520c81 CH |
1695 | |
1696 | /* | |
1697 | * If a matching dentry exists, and it's not negative use it. | |
1698 | * | |
1699 | * Decrement the reference count to balance the iget() done | |
1700 | * earlier on. | |
1701 | */ | |
9403540c BN |
1702 | if (found->d_inode) { |
1703 | if (unlikely(found->d_inode != inode)) { | |
1704 | /* This can't happen because bad inodes are unhashed. */ | |
1705 | BUG_ON(!is_bad_inode(inode)); | |
1706 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1707 | } | |
9403540c BN |
1708 | iput(inode); |
1709 | return found; | |
1710 | } | |
b6520c81 | 1711 | |
9403540c | 1712 | /* |
9403540c | 1713 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1714 | * already has a dentry. |
9403540c | 1715 | */ |
4513d899 AV |
1716 | new = d_splice_alias(inode, found); |
1717 | if (new) { | |
1718 | dput(found); | |
1719 | found = new; | |
9403540c | 1720 | } |
4513d899 | 1721 | return found; |
9403540c BN |
1722 | |
1723 | err_out: | |
1724 | iput(inode); | |
4f522a24 | 1725 | return found; |
9403540c | 1726 | } |
ec4f8605 | 1727 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1728 | |
12f8ad4b LT |
1729 | /* |
1730 | * Do the slow-case of the dentry name compare. | |
1731 | * | |
1732 | * Unlike the dentry_cmp() function, we need to atomically | |
1733 | * load the name, length and inode information, so that the | |
1734 | * filesystem can rely on them, and can use the 'name' and | |
1735 | * 'len' information without worrying about walking off the | |
1736 | * end of memory etc. | |
1737 | * | |
1738 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1739 | * in arguments (the low-level filesystem should not look | |
1740 | * at the dentry inode or name contents directly, since | |
1741 | * rename can change them while we're in RCU mode). | |
1742 | */ | |
1743 | enum slow_d_compare { | |
1744 | D_COMP_OK, | |
1745 | D_COMP_NOMATCH, | |
1746 | D_COMP_SEQRETRY, | |
1747 | }; | |
1748 | ||
1749 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1750 | const struct dentry *parent, | |
1751 | struct inode *inode, | |
1752 | struct dentry *dentry, | |
1753 | unsigned int seq, | |
1754 | const struct qstr *name) | |
1755 | { | |
1756 | int tlen = dentry->d_name.len; | |
1757 | const char *tname = dentry->d_name.name; | |
1758 | struct inode *i = dentry->d_inode; | |
1759 | ||
1760 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1761 | cpu_relax(); | |
1762 | return D_COMP_SEQRETRY; | |
1763 | } | |
1764 | if (parent->d_op->d_compare(parent, inode, | |
1765 | dentry, i, | |
1766 | tlen, tname, name)) | |
1767 | return D_COMP_NOMATCH; | |
1768 | return D_COMP_OK; | |
1769 | } | |
1770 | ||
31e6b01f NP |
1771 | /** |
1772 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1773 | * @parent: parent dentry | |
1774 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1775 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1776 | * @inode: returns dentry->d_inode when the inode was found valid. |
1777 | * Returns: dentry, or NULL | |
1778 | * | |
1779 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1780 | * resolution (store-free path walking) design described in | |
1781 | * Documentation/filesystems/path-lookup.txt. | |
1782 | * | |
1783 | * This is not to be used outside core vfs. | |
1784 | * | |
1785 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1786 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1787 | * without taking d_lock and checking d_seq sequence count against @seq | |
1788 | * returned here. | |
1789 | * | |
1790 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1791 | * function. | |
1792 | * | |
1793 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1794 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1795 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1796 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
1797 | * |
1798 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
1799 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 1800 | */ |
8966be90 LT |
1801 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
1802 | const struct qstr *name, | |
12f8ad4b | 1803 | unsigned *seqp, struct inode *inode) |
31e6b01f | 1804 | { |
26fe5750 | 1805 | u64 hashlen = name->hash_len; |
31e6b01f | 1806 | const unsigned char *str = name->name; |
26fe5750 | 1807 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 1808 | struct hlist_bl_node *node; |
31e6b01f NP |
1809 | struct dentry *dentry; |
1810 | ||
1811 | /* | |
1812 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1813 | * required to prevent single threaded performance regressions | |
1814 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1815 | * Keep the two functions in sync. | |
1816 | */ | |
1817 | ||
1818 | /* | |
1819 | * The hash list is protected using RCU. | |
1820 | * | |
1821 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1822 | * races with d_move(). | |
1823 | * | |
1824 | * It is possible that concurrent renames can mess up our list | |
1825 | * walk here and result in missing our dentry, resulting in the | |
1826 | * false-negative result. d_lookup() protects against concurrent | |
1827 | * renames using rename_lock seqlock. | |
1828 | * | |
b0a4bb83 | 1829 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1830 | */ |
b07ad996 | 1831 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 1832 | unsigned seq; |
31e6b01f | 1833 | |
31e6b01f | 1834 | seqretry: |
12f8ad4b LT |
1835 | /* |
1836 | * The dentry sequence count protects us from concurrent | |
1837 | * renames, and thus protects inode, parent and name fields. | |
1838 | * | |
1839 | * The caller must perform a seqcount check in order | |
1840 | * to do anything useful with the returned dentry, | |
1841 | * including using the 'd_inode' pointer. | |
1842 | * | |
1843 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
1844 | * we don't wait for the sequence count to stabilize if it | |
1845 | * is in the middle of a sequence change. If we do the slow | |
1846 | * dentry compare, we will do seqretries until it is stable, | |
1847 | * and if we end up with a successful lookup, we actually | |
1848 | * want to exit RCU lookup anyway. | |
1849 | */ | |
1850 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
1851 | if (dentry->d_parent != parent) |
1852 | continue; | |
2e321806 LT |
1853 | if (d_unhashed(dentry)) |
1854 | continue; | |
12f8ad4b LT |
1855 | *seqp = seq; |
1856 | ||
830c0f0e | 1857 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
1858 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
1859 | continue; | |
12f8ad4b LT |
1860 | switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) { |
1861 | case D_COMP_OK: | |
1862 | return dentry; | |
1863 | case D_COMP_NOMATCH: | |
31e6b01f | 1864 | continue; |
12f8ad4b LT |
1865 | default: |
1866 | goto seqretry; | |
1867 | } | |
31e6b01f | 1868 | } |
12f8ad4b | 1869 | |
26fe5750 | 1870 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 1871 | continue; |
26fe5750 | 1872 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 1873 | return dentry; |
31e6b01f NP |
1874 | } |
1875 | return NULL; | |
1876 | } | |
1877 | ||
1da177e4 LT |
1878 | /** |
1879 | * d_lookup - search for a dentry | |
1880 | * @parent: parent dentry | |
1881 | * @name: qstr of name we wish to find | |
b04f784e | 1882 | * Returns: dentry, or NULL |
1da177e4 | 1883 | * |
b04f784e NP |
1884 | * d_lookup searches the children of the parent dentry for the name in |
1885 | * question. If the dentry is found its reference count is incremented and the | |
1886 | * dentry is returned. The caller must use dput to free the entry when it has | |
1887 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1888 | */ |
da2d8455 | 1889 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 1890 | { |
31e6b01f | 1891 | struct dentry *dentry; |
949854d0 | 1892 | unsigned seq; |
1da177e4 LT |
1893 | |
1894 | do { | |
1895 | seq = read_seqbegin(&rename_lock); | |
1896 | dentry = __d_lookup(parent, name); | |
1897 | if (dentry) | |
1898 | break; | |
1899 | } while (read_seqretry(&rename_lock, seq)); | |
1900 | return dentry; | |
1901 | } | |
ec4f8605 | 1902 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1903 | |
31e6b01f | 1904 | /** |
b04f784e NP |
1905 | * __d_lookup - search for a dentry (racy) |
1906 | * @parent: parent dentry | |
1907 | * @name: qstr of name we wish to find | |
1908 | * Returns: dentry, or NULL | |
1909 | * | |
1910 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1911 | * false-negative result due to unrelated rename activity. | |
1912 | * | |
1913 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1914 | * however it must be used carefully, eg. with a following d_lookup in | |
1915 | * the case of failure. | |
1916 | * | |
1917 | * __d_lookup callers must be commented. | |
1918 | */ | |
a713ca2a | 1919 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
1920 | { |
1921 | unsigned int len = name->len; | |
1922 | unsigned int hash = name->hash; | |
1923 | const unsigned char *str = name->name; | |
b07ad996 | 1924 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1925 | struct hlist_bl_node *node; |
31e6b01f | 1926 | struct dentry *found = NULL; |
665a7583 | 1927 | struct dentry *dentry; |
1da177e4 | 1928 | |
31e6b01f NP |
1929 | /* |
1930 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1931 | * required to prevent single threaded performance regressions | |
1932 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1933 | * Keep the two functions in sync. | |
1934 | */ | |
1935 | ||
b04f784e NP |
1936 | /* |
1937 | * The hash list is protected using RCU. | |
1938 | * | |
1939 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1940 | * with d_move(). | |
1941 | * | |
1942 | * It is possible that concurrent renames can mess up our list | |
1943 | * walk here and result in missing our dentry, resulting in the | |
1944 | * false-negative result. d_lookup() protects against concurrent | |
1945 | * renames using rename_lock seqlock. | |
1946 | * | |
b0a4bb83 | 1947 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1948 | */ |
1da177e4 LT |
1949 | rcu_read_lock(); |
1950 | ||
b07ad996 | 1951 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 1952 | |
1da177e4 LT |
1953 | if (dentry->d_name.hash != hash) |
1954 | continue; | |
1da177e4 LT |
1955 | |
1956 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1957 | if (dentry->d_parent != parent) |
1958 | goto next; | |
d0185c08 LT |
1959 | if (d_unhashed(dentry)) |
1960 | goto next; | |
1961 | ||
1da177e4 LT |
1962 | /* |
1963 | * It is safe to compare names since d_move() cannot | |
1964 | * change the qstr (protected by d_lock). | |
1965 | */ | |
fb045adb | 1966 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
1967 | int tlen = dentry->d_name.len; |
1968 | const char *tname = dentry->d_name.name; | |
621e155a NP |
1969 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1970 | dentry, dentry->d_inode, | |
31e6b01f | 1971 | tlen, tname, name)) |
1da177e4 LT |
1972 | goto next; |
1973 | } else { | |
ee983e89 LT |
1974 | if (dentry->d_name.len != len) |
1975 | goto next; | |
12f8ad4b | 1976 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
1977 | goto next; |
1978 | } | |
1979 | ||
b7ab39f6 | 1980 | dentry->d_count++; |
d0185c08 | 1981 | found = dentry; |
1da177e4 LT |
1982 | spin_unlock(&dentry->d_lock); |
1983 | break; | |
1984 | next: | |
1985 | spin_unlock(&dentry->d_lock); | |
1986 | } | |
1987 | rcu_read_unlock(); | |
1988 | ||
1989 | return found; | |
1990 | } | |
1991 | ||
3e7e241f EB |
1992 | /** |
1993 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1994 | * @dir: Directory to search in | |
1995 | * @name: qstr of name we wish to find | |
1996 | * | |
4f522a24 | 1997 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
1998 | */ |
1999 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2000 | { | |
3e7e241f EB |
2001 | /* |
2002 | * Check for a fs-specific hash function. Note that we must | |
2003 | * calculate the standard hash first, as the d_op->d_hash() | |
2004 | * routine may choose to leave the hash value unchanged. | |
2005 | */ | |
2006 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2007 | if (dir->d_flags & DCACHE_OP_HASH) { |
4f522a24 AV |
2008 | int err = dir->d_op->d_hash(dir, dir->d_inode, name); |
2009 | if (unlikely(err < 0)) | |
2010 | return ERR_PTR(err); | |
3e7e241f | 2011 | } |
4f522a24 | 2012 | return d_lookup(dir, name); |
3e7e241f | 2013 | } |
4f522a24 | 2014 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2015 | |
1da177e4 | 2016 | /** |
786a5e15 | 2017 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2018 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2019 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2020 | * |
2021 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2022 | * This is used by ncpfs in its readdir implementation. | |
2023 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2024 | * |
2025 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2026 | */ |
d3a23e16 | 2027 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2028 | { |
786a5e15 | 2029 | struct dentry *child; |
d3a23e16 | 2030 | |
2fd6b7f5 | 2031 | spin_lock(&dparent->d_lock); |
6637ecd3 | 2032 | list_for_each_entry(child, &dparent->d_subdirs, d_child) { |
786a5e15 | 2033 | if (dentry == child) { |
2fd6b7f5 | 2034 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2035 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2036 | spin_unlock(&dentry->d_lock); |
2037 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2038 | return 1; |
2039 | } | |
2040 | } | |
2fd6b7f5 | 2041 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2042 | |
1da177e4 LT |
2043 | return 0; |
2044 | } | |
ec4f8605 | 2045 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2046 | |
2047 | /* | |
2048 | * When a file is deleted, we have two options: | |
2049 | * - turn this dentry into a negative dentry | |
2050 | * - unhash this dentry and free it. | |
2051 | * | |
2052 | * Usually, we want to just turn this into | |
2053 | * a negative dentry, but if anybody else is | |
2054 | * currently using the dentry or the inode | |
2055 | * we can't do that and we fall back on removing | |
2056 | * it from the hash queues and waiting for | |
2057 | * it to be deleted later when it has no users | |
2058 | */ | |
2059 | ||
2060 | /** | |
2061 | * d_delete - delete a dentry | |
2062 | * @dentry: The dentry to delete | |
2063 | * | |
2064 | * Turn the dentry into a negative dentry if possible, otherwise | |
2065 | * remove it from the hash queues so it can be deleted later | |
2066 | */ | |
2067 | ||
2068 | void d_delete(struct dentry * dentry) | |
2069 | { | |
873feea0 | 2070 | struct inode *inode; |
7a91bf7f | 2071 | int isdir = 0; |
1da177e4 LT |
2072 | /* |
2073 | * Are we the only user? | |
2074 | */ | |
357f8e65 | 2075 | again: |
1da177e4 | 2076 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2077 | inode = dentry->d_inode; |
2078 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 2079 | if (dentry->d_count == 1) { |
1fe0c023 | 2080 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2081 | spin_unlock(&dentry->d_lock); |
2082 | cpu_relax(); | |
2083 | goto again; | |
2084 | } | |
13e3c5e5 | 2085 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2086 | dentry_unlink_inode(dentry); |
7a91bf7f | 2087 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2088 | return; |
2089 | } | |
2090 | ||
2091 | if (!d_unhashed(dentry)) | |
2092 | __d_drop(dentry); | |
2093 | ||
2094 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2095 | |
2096 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2097 | } |
ec4f8605 | 2098 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2099 | |
b07ad996 | 2100 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2101 | { |
ceb5bdc2 | 2102 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2103 | hlist_bl_lock(b); |
b07ad996 | 2104 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2105 | hlist_bl_unlock(b); |
1da177e4 LT |
2106 | } |
2107 | ||
770bfad8 DH |
2108 | static void _d_rehash(struct dentry * entry) |
2109 | { | |
2110 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2111 | } | |
2112 | ||
1da177e4 LT |
2113 | /** |
2114 | * d_rehash - add an entry back to the hash | |
2115 | * @entry: dentry to add to the hash | |
2116 | * | |
2117 | * Adds a dentry to the hash according to its name. | |
2118 | */ | |
2119 | ||
2120 | void d_rehash(struct dentry * entry) | |
2121 | { | |
1da177e4 | 2122 | spin_lock(&entry->d_lock); |
770bfad8 | 2123 | _d_rehash(entry); |
1da177e4 | 2124 | spin_unlock(&entry->d_lock); |
1da177e4 | 2125 | } |
ec4f8605 | 2126 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2127 | |
fb2d5b86 NP |
2128 | /** |
2129 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2130 | * @dentry: dentry to be updated | |
2131 | * @name: new name | |
2132 | * | |
2133 | * Update a case insensitive dentry with new case of name. | |
2134 | * | |
2135 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2136 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2137 | * lengths). | |
2138 | * | |
2139 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2140 | * keep renames and concurrent inserts, and readdir(2) away). | |
2141 | */ | |
2142 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2143 | { | |
7ebfa57f | 2144 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2145 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2146 | ||
fb2d5b86 | 2147 | spin_lock(&dentry->d_lock); |
31e6b01f | 2148 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2149 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2150 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2151 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2152 | } |
2153 | EXPORT_SYMBOL(dentry_update_name_case); | |
2154 | ||
1da177e4 LT |
2155 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2156 | { | |
2157 | if (dname_external(target)) { | |
2158 | if (dname_external(dentry)) { | |
2159 | /* | |
2160 | * Both external: swap the pointers | |
2161 | */ | |
9a8d5bb4 | 2162 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2163 | } else { |
2164 | /* | |
2165 | * dentry:internal, target:external. Steal target's | |
2166 | * storage and make target internal. | |
2167 | */ | |
321bcf92 BF |
2168 | memcpy(target->d_iname, dentry->d_name.name, |
2169 | dentry->d_name.len + 1); | |
1da177e4 LT |
2170 | dentry->d_name.name = target->d_name.name; |
2171 | target->d_name.name = target->d_iname; | |
2172 | } | |
2173 | } else { | |
2174 | if (dname_external(dentry)) { | |
2175 | /* | |
2176 | * dentry:external, target:internal. Give dentry's | |
2177 | * storage to target and make dentry internal | |
2178 | */ | |
2179 | memcpy(dentry->d_iname, target->d_name.name, | |
2180 | target->d_name.len + 1); | |
2181 | target->d_name.name = dentry->d_name.name; | |
2182 | dentry->d_name.name = dentry->d_iname; | |
2183 | } else { | |
2184 | /* | |
2185 | * Both are internal. Just copy target to dentry | |
2186 | */ | |
2187 | memcpy(dentry->d_iname, target->d_name.name, | |
2188 | target->d_name.len + 1); | |
dc711ca3 AV |
2189 | dentry->d_name.len = target->d_name.len; |
2190 | return; | |
1da177e4 LT |
2191 | } |
2192 | } | |
9a8d5bb4 | 2193 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2194 | } |
2195 | ||
2fd6b7f5 NP |
2196 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2197 | { | |
2198 | /* | |
2199 | * XXXX: do we really need to take target->d_lock? | |
2200 | */ | |
2201 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2202 | spin_lock(&target->d_parent->d_lock); | |
2203 | else { | |
2204 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2205 | spin_lock(&dentry->d_parent->d_lock); | |
2206 | spin_lock_nested(&target->d_parent->d_lock, | |
2207 | DENTRY_D_LOCK_NESTED); | |
2208 | } else { | |
2209 | spin_lock(&target->d_parent->d_lock); | |
2210 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2211 | DENTRY_D_LOCK_NESTED); | |
2212 | } | |
2213 | } | |
2214 | if (target < dentry) { | |
2215 | spin_lock_nested(&target->d_lock, 2); | |
2216 | spin_lock_nested(&dentry->d_lock, 3); | |
2217 | } else { | |
2218 | spin_lock_nested(&dentry->d_lock, 2); | |
2219 | spin_lock_nested(&target->d_lock, 3); | |
2220 | } | |
2221 | } | |
2222 | ||
2223 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2224 | struct dentry *target) | |
2225 | { | |
2226 | if (target->d_parent != dentry->d_parent) | |
2227 | spin_unlock(&dentry->d_parent->d_lock); | |
2228 | if (target->d_parent != target) | |
2229 | spin_unlock(&target->d_parent->d_lock); | |
2230 | } | |
2231 | ||
1da177e4 | 2232 | /* |
2fd6b7f5 NP |
2233 | * When switching names, the actual string doesn't strictly have to |
2234 | * be preserved in the target - because we're dropping the target | |
2235 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2236 | * the new name before we switch. | |
2237 | * | |
2238 | * Note that we have to be a lot more careful about getting the hash | |
2239 | * switched - we have to switch the hash value properly even if it | |
2240 | * then no longer matches the actual (corrupted) string of the target. | |
2241 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2242 | */ |
9eaef27b | 2243 | /* |
18367501 | 2244 | * __d_move - move a dentry |
1da177e4 LT |
2245 | * @dentry: entry to move |
2246 | * @target: new dentry | |
2247 | * | |
2248 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2249 | * dcache entries should not be moved in this way. Caller must hold |
2250 | * rename_lock, the i_mutex of the source and target directories, | |
2251 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2252 | */ |
18367501 | 2253 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2254 | { |
1da177e4 LT |
2255 | if (!dentry->d_inode) |
2256 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2257 | ||
2fd6b7f5 NP |
2258 | BUG_ON(d_ancestor(dentry, target)); |
2259 | BUG_ON(d_ancestor(target, dentry)); | |
2260 | ||
2fd6b7f5 | 2261 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2262 | |
31e6b01f NP |
2263 | write_seqcount_begin(&dentry->d_seq); |
2264 | write_seqcount_begin(&target->d_seq); | |
2265 | ||
ceb5bdc2 NP |
2266 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2267 | ||
2268 | /* | |
2269 | * Move the dentry to the target hash queue. Don't bother checking | |
2270 | * for the same hash queue because of how unlikely it is. | |
2271 | */ | |
2272 | __d_drop(dentry); | |
789680d1 | 2273 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2274 | |
2275 | /* Unhash the target: dput() will then get rid of it */ | |
2276 | __d_drop(target); | |
2277 | ||
6637ecd3 AV |
2278 | list_del(&dentry->d_child); |
2279 | list_del(&target->d_child); | |
1da177e4 LT |
2280 | |
2281 | /* Switch the names.. */ | |
2282 | switch_names(dentry, target); | |
9a8d5bb4 | 2283 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2284 | |
2285 | /* ... and switch the parents */ | |
2286 | if (IS_ROOT(dentry)) { | |
742d555a | 2287 | dentry->d_flags |= DCACHE_RCUACCESS; |
1da177e4 LT |
2288 | dentry->d_parent = target->d_parent; |
2289 | target->d_parent = target; | |
6637ecd3 | 2290 | INIT_LIST_HEAD(&target->d_child); |
1da177e4 | 2291 | } else { |
9a8d5bb4 | 2292 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2293 | |
2294 | /* And add them back to the (new) parent lists */ | |
6637ecd3 | 2295 | list_add(&target->d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2296 | } |
2297 | ||
6637ecd3 | 2298 | list_add(&dentry->d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2299 | |
31e6b01f NP |
2300 | write_seqcount_end(&target->d_seq); |
2301 | write_seqcount_end(&dentry->d_seq); | |
2302 | ||
2fd6b7f5 | 2303 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2304 | spin_unlock(&target->d_lock); |
c32ccd87 | 2305 | fsnotify_d_move(dentry); |
1da177e4 | 2306 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2307 | } |
2308 | ||
2309 | /* | |
2310 | * d_move - move a dentry | |
2311 | * @dentry: entry to move | |
2312 | * @target: new dentry | |
2313 | * | |
2314 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2315 | * dcache entries should not be moved in this way. See the locking |
2316 | * requirements for __d_move. | |
18367501 AV |
2317 | */ |
2318 | void d_move(struct dentry *dentry, struct dentry *target) | |
2319 | { | |
2320 | write_seqlock(&rename_lock); | |
2321 | __d_move(dentry, target); | |
1da177e4 | 2322 | write_sequnlock(&rename_lock); |
9eaef27b | 2323 | } |
ec4f8605 | 2324 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2325 | |
e2761a11 OH |
2326 | /** |
2327 | * d_ancestor - search for an ancestor | |
2328 | * @p1: ancestor dentry | |
2329 | * @p2: child dentry | |
2330 | * | |
2331 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2332 | * an ancestor of p2, else NULL. | |
9eaef27b | 2333 | */ |
e2761a11 | 2334 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2335 | { |
2336 | struct dentry *p; | |
2337 | ||
871c0067 | 2338 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2339 | if (p->d_parent == p1) |
e2761a11 | 2340 | return p; |
9eaef27b | 2341 | } |
e2761a11 | 2342 | return NULL; |
9eaef27b TM |
2343 | } |
2344 | ||
2345 | /* | |
2346 | * This helper attempts to cope with remotely renamed directories | |
2347 | * | |
2348 | * It assumes that the caller is already holding | |
18367501 | 2349 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2350 | * |
2351 | * Note: If ever the locking in lock_rename() changes, then please | |
2352 | * remember to update this too... | |
9eaef27b | 2353 | */ |
873feea0 NP |
2354 | static struct dentry *__d_unalias(struct inode *inode, |
2355 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2356 | { |
2357 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2358 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2359 | |
2360 | /* If alias and dentry share a parent, then no extra locks required */ | |
2361 | if (alias->d_parent == dentry->d_parent) | |
2362 | goto out_unalias; | |
2363 | ||
9eaef27b | 2364 | /* See lock_rename() */ |
9eaef27b TM |
2365 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2366 | goto out_err; | |
2367 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2368 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2369 | goto out_err; | |
2370 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2371 | out_unalias: | |
ee3efa91 AV |
2372 | if (likely(!d_mountpoint(alias))) { |
2373 | __d_move(alias, dentry); | |
2374 | ret = alias; | |
2375 | } | |
9eaef27b | 2376 | out_err: |
873feea0 | 2377 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2378 | if (m2) |
2379 | mutex_unlock(m2); | |
2380 | if (m1) | |
2381 | mutex_unlock(m1); | |
2382 | return ret; | |
2383 | } | |
2384 | ||
770bfad8 DH |
2385 | /* |
2386 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2387 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2388 | * returns with anon->d_lock held! |
770bfad8 DH |
2389 | */ |
2390 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2391 | { | |
740da42e | 2392 | struct dentry *dparent; |
770bfad8 | 2393 | |
2fd6b7f5 | 2394 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2395 | |
31e6b01f NP |
2396 | write_seqcount_begin(&dentry->d_seq); |
2397 | write_seqcount_begin(&anon->d_seq); | |
2398 | ||
770bfad8 | 2399 | dparent = dentry->d_parent; |
770bfad8 | 2400 | |
2fd6b7f5 NP |
2401 | switch_names(dentry, anon); |
2402 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2403 | ||
742d555a | 2404 | dentry->d_flags |= DCACHE_RCUACCESS; |
740da42e | 2405 | dentry->d_parent = dentry; |
6637ecd3 | 2406 | list_del_init(&dentry->d_child); |
740da42e | 2407 | anon->d_parent = dparent; |
64df8d3b AV |
2408 | if (likely(!d_unhashed(anon))) { |
2409 | hlist_bl_lock(&anon->d_sb->s_anon); | |
2410 | __hlist_bl_del(&anon->d_hash); | |
2411 | anon->d_hash.pprev = NULL; | |
2412 | hlist_bl_unlock(&anon->d_sb->s_anon); | |
2413 | } | |
6637ecd3 | 2414 | list_move(&anon->d_child, &dparent->d_subdirs); |
770bfad8 | 2415 | |
31e6b01f NP |
2416 | write_seqcount_end(&dentry->d_seq); |
2417 | write_seqcount_end(&anon->d_seq); | |
2418 | ||
2fd6b7f5 NP |
2419 | dentry_unlock_parents_for_move(anon, dentry); |
2420 | spin_unlock(&dentry->d_lock); | |
2421 | ||
2422 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2423 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2424 | } | |
2425 | ||
2426 | /** | |
2427 | * d_materialise_unique - introduce an inode into the tree | |
2428 | * @dentry: candidate dentry | |
2429 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2430 | * | |
2431 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2432 | * root directory alias in its place if there is one. Caller must hold the |
2433 | * i_mutex of the parent directory. | |
770bfad8 DH |
2434 | */ |
2435 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2436 | { | |
9eaef27b | 2437 | struct dentry *actual; |
770bfad8 DH |
2438 | |
2439 | BUG_ON(!d_unhashed(dentry)); | |
2440 | ||
770bfad8 DH |
2441 | if (!inode) { |
2442 | actual = dentry; | |
360da900 | 2443 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2444 | d_rehash(actual); |
2445 | goto out_nolock; | |
770bfad8 DH |
2446 | } |
2447 | ||
873feea0 | 2448 | spin_lock(&inode->i_lock); |
357f8e65 | 2449 | |
9eaef27b TM |
2450 | if (S_ISDIR(inode->i_mode)) { |
2451 | struct dentry *alias; | |
2452 | ||
2453 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2454 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2455 | if (alias) { |
2456 | actual = alias; | |
18367501 AV |
2457 | write_seqlock(&rename_lock); |
2458 | ||
2459 | if (d_ancestor(alias, dentry)) { | |
2460 | /* Check for loops */ | |
2461 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2462 | spin_unlock(&inode->i_lock); |
18367501 AV |
2463 | } else if (IS_ROOT(alias)) { |
2464 | /* Is this an anonymous mountpoint that we | |
2465 | * could splice into our tree? */ | |
9eaef27b | 2466 | __d_materialise_dentry(dentry, alias); |
18367501 | 2467 | write_sequnlock(&rename_lock); |
9eaef27b | 2468 | goto found; |
18367501 AV |
2469 | } else { |
2470 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2471 | * aliasing. This drops inode->i_lock */ |
18367501 | 2472 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2473 | } |
18367501 | 2474 | write_sequnlock(&rename_lock); |
dd179946 DH |
2475 | if (IS_ERR(actual)) { |
2476 | if (PTR_ERR(actual) == -ELOOP) | |
2477 | pr_warn_ratelimited( | |
2478 | "VFS: Lookup of '%s' in %s %s" | |
2479 | " would have caused loop\n", | |
2480 | dentry->d_name.name, | |
2481 | inode->i_sb->s_type->name, | |
2482 | inode->i_sb->s_id); | |
9eaef27b | 2483 | dput(alias); |
dd179946 | 2484 | } |
9eaef27b TM |
2485 | goto out_nolock; |
2486 | } | |
770bfad8 DH |
2487 | } |
2488 | ||
2489 | /* Add a unique reference */ | |
2490 | actual = __d_instantiate_unique(dentry, inode); | |
2491 | if (!actual) | |
2492 | actual = dentry; | |
357f8e65 NP |
2493 | else |
2494 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2495 | |
770bfad8 DH |
2496 | spin_lock(&actual->d_lock); |
2497 | found: | |
2498 | _d_rehash(actual); | |
2499 | spin_unlock(&actual->d_lock); | |
873feea0 | 2500 | spin_unlock(&inode->i_lock); |
9eaef27b | 2501 | out_nolock: |
770bfad8 DH |
2502 | if (actual == dentry) { |
2503 | security_d_instantiate(dentry, inode); | |
2504 | return NULL; | |
2505 | } | |
2506 | ||
2507 | iput(inode); | |
2508 | return actual; | |
770bfad8 | 2509 | } |
ec4f8605 | 2510 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2511 | |
cdd16d02 | 2512 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2513 | { |
2514 | *buflen -= namelen; | |
2515 | if (*buflen < 0) | |
2516 | return -ENAMETOOLONG; | |
2517 | *buffer -= namelen; | |
2518 | memcpy(*buffer, str, namelen); | |
2519 | return 0; | |
2520 | } | |
2521 | ||
cdd16d02 MS |
2522 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2523 | { | |
2524 | return prepend(buffer, buflen, name->name, name->len); | |
2525 | } | |
2526 | ||
1da177e4 | 2527 | /** |
208898c1 | 2528 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2529 | * @path: the dentry/vfsmount to report |
02125a82 | 2530 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2531 | * @buffer: pointer to the end of the buffer |
2532 | * @buflen: pointer to buffer length | |
552ce544 | 2533 | * |
949854d0 | 2534 | * Caller holds the rename_lock. |
1da177e4 | 2535 | */ |
02125a82 AV |
2536 | static int prepend_path(const struct path *path, |
2537 | const struct path *root, | |
f2eb6575 | 2538 | char **buffer, int *buflen) |
1da177e4 | 2539 | { |
9d1bc601 MS |
2540 | struct dentry *dentry = path->dentry; |
2541 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2542 | struct mount *mnt = real_mount(vfsmnt); |
91397d56 EB |
2543 | char *orig_buffer = *buffer; |
2544 | int orig_len = *buflen; | |
f2eb6575 MS |
2545 | bool slash = false; |
2546 | int error = 0; | |
6092d048 | 2547 | |
f2eb6575 | 2548 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2549 | struct dentry * parent; |
2550 | ||
1da177e4 | 2551 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
91397d56 EB |
2552 | /* Escaped? */ |
2553 | if (dentry != vfsmnt->mnt_root) { | |
2554 | *buffer = orig_buffer; | |
2555 | *buflen = orig_len; | |
2556 | slash = false; | |
2557 | error = 3; | |
2558 | goto global_root; | |
2559 | } | |
552ce544 | 2560 | /* Global root? */ |
676da58d | 2561 | if (!mnt_has_parent(mnt)) |
1da177e4 | 2562 | goto global_root; |
a73324da | 2563 | dentry = mnt->mnt_mountpoint; |
0714a533 AV |
2564 | mnt = mnt->mnt_parent; |
2565 | vfsmnt = &mnt->mnt; | |
1da177e4 LT |
2566 | continue; |
2567 | } | |
2568 | parent = dentry->d_parent; | |
2569 | prefetch(parent); | |
9abca360 | 2570 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2571 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2572 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2573 | if (!error) |
2574 | error = prepend(buffer, buflen, "/", 1); | |
2575 | if (error) | |
2576 | break; | |
2577 | ||
2578 | slash = true; | |
1da177e4 LT |
2579 | dentry = parent; |
2580 | } | |
2581 | ||
f2eb6575 MS |
2582 | if (!error && !slash) |
2583 | error = prepend(buffer, buflen, "/", 1); | |
2584 | ||
f2eb6575 | 2585 | return error; |
1da177e4 LT |
2586 | |
2587 | global_root: | |
02125a82 AV |
2588 | if (!slash) |
2589 | error = prepend(buffer, buflen, "/", 1); | |
2590 | if (!error) | |
f7a99c5b | 2591 | error = is_mounted(vfsmnt) ? 1 : 2; |
7ea600b5 | 2592 | return error; |
f2eb6575 | 2593 | } |
be285c71 | 2594 | |
f2eb6575 MS |
2595 | /** |
2596 | * __d_path - return the path of a dentry | |
2597 | * @path: the dentry/vfsmount to report | |
02125a82 | 2598 | * @root: root vfsmnt/dentry |
cd956a1c | 2599 | * @buf: buffer to return value in |
f2eb6575 MS |
2600 | * @buflen: buffer length |
2601 | * | |
ffd1f4ed | 2602 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2603 | * |
2604 | * Returns a pointer into the buffer or an error code if the | |
2605 | * path was too long. | |
2606 | * | |
be148247 | 2607 | * "buflen" should be positive. |
f2eb6575 | 2608 | * |
02125a82 | 2609 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2610 | */ |
02125a82 AV |
2611 | char *__d_path(const struct path *path, |
2612 | const struct path *root, | |
f2eb6575 MS |
2613 | char *buf, int buflen) |
2614 | { | |
2615 | char *res = buf + buflen; | |
2616 | int error; | |
2617 | ||
2618 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2619 | br_read_lock(&vfsmount_lock); |
949854d0 | 2620 | write_seqlock(&rename_lock); |
f2eb6575 | 2621 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2622 | write_sequnlock(&rename_lock); |
7ea600b5 | 2623 | br_read_unlock(&vfsmount_lock); |
be148247 | 2624 | |
02125a82 AV |
2625 | if (error < 0) |
2626 | return ERR_PTR(error); | |
2627 | if (error > 0) | |
2628 | return NULL; | |
2629 | return res; | |
2630 | } | |
2631 | ||
2632 | char *d_absolute_path(const struct path *path, | |
2633 | char *buf, int buflen) | |
2634 | { | |
2635 | struct path root = {}; | |
2636 | char *res = buf + buflen; | |
2637 | int error; | |
2638 | ||
2639 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2640 | br_read_lock(&vfsmount_lock); |
02125a82 AV |
2641 | write_seqlock(&rename_lock); |
2642 | error = prepend_path(path, &root, &res, &buflen); | |
2643 | write_sequnlock(&rename_lock); | |
7ea600b5 | 2644 | br_read_unlock(&vfsmount_lock); |
02125a82 AV |
2645 | |
2646 | if (error > 1) | |
2647 | error = -EINVAL; | |
2648 | if (error < 0) | |
f2eb6575 | 2649 | return ERR_PTR(error); |
f2eb6575 | 2650 | return res; |
1da177e4 LT |
2651 | } |
2652 | ||
ffd1f4ed MS |
2653 | /* |
2654 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2655 | */ | |
02125a82 AV |
2656 | static int path_with_deleted(const struct path *path, |
2657 | const struct path *root, | |
2658 | char **buf, int *buflen) | |
ffd1f4ed MS |
2659 | { |
2660 | prepend(buf, buflen, "\0", 1); | |
2661 | if (d_unlinked(path->dentry)) { | |
2662 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2663 | if (error) | |
2664 | return error; | |
2665 | } | |
2666 | ||
2667 | return prepend_path(path, root, buf, buflen); | |
2668 | } | |
2669 | ||
8df9d1a4 MS |
2670 | static int prepend_unreachable(char **buffer, int *buflen) |
2671 | { | |
2672 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2673 | } | |
2674 | ||
a03a8a70 JB |
2675 | /** |
2676 | * d_path - return the path of a dentry | |
cf28b486 | 2677 | * @path: path to report |
a03a8a70 JB |
2678 | * @buf: buffer to return value in |
2679 | * @buflen: buffer length | |
2680 | * | |
2681 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2682 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2683 | * | |
52afeefb AV |
2684 | * Returns a pointer into the buffer or an error code if the path was |
2685 | * too long. Note: Callers should use the returned pointer, not the passed | |
2686 | * in buffer, to use the name! The implementation often starts at an offset | |
2687 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2688 | * |
31f3e0b3 | 2689 | * "buflen" should be positive. |
a03a8a70 | 2690 | */ |
20d4fdc1 | 2691 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2692 | { |
ffd1f4ed | 2693 | char *res = buf + buflen; |
6ac08c39 | 2694 | struct path root; |
ffd1f4ed | 2695 | int error; |
1da177e4 | 2696 | |
c23fbb6b ED |
2697 | /* |
2698 | * We have various synthetic filesystems that never get mounted. On | |
2699 | * these filesystems dentries are never used for lookup purposes, and | |
2700 | * thus don't need to be hashed. They also don't need a name until a | |
2701 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2702 | * below allows us to generate a name for these objects on demand: | |
3717eee3 EB |
2703 | * |
2704 | * Some pseudo inodes are mountable. When they are mounted | |
2705 | * path->dentry == path->mnt->mnt_root. In that case don't call d_dname | |
2706 | * and instead have d_path return the mounted path. | |
c23fbb6b | 2707 | */ |
3717eee3 EB |
2708 | if (path->dentry->d_op && path->dentry->d_op->d_dname && |
2709 | (!IS_ROOT(path->dentry) || path->dentry != path->mnt->mnt_root)) | |
cf28b486 | 2710 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); |
c23fbb6b | 2711 | |
f7ad3c6b | 2712 | get_fs_root(current->fs, &root); |
7ea600b5 | 2713 | br_read_lock(&vfsmount_lock); |
949854d0 | 2714 | write_seqlock(&rename_lock); |
02125a82 | 2715 | error = path_with_deleted(path, &root, &res, &buflen); |
7ea600b5 AV |
2716 | write_sequnlock(&rename_lock); |
2717 | br_read_unlock(&vfsmount_lock); | |
02125a82 | 2718 | if (error < 0) |
ffd1f4ed | 2719 | res = ERR_PTR(error); |
6ac08c39 | 2720 | path_put(&root); |
1da177e4 LT |
2721 | return res; |
2722 | } | |
ec4f8605 | 2723 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2724 | |
c23fbb6b ED |
2725 | /* |
2726 | * Helper function for dentry_operations.d_dname() members | |
2727 | */ | |
2728 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2729 | const char *fmt, ...) | |
2730 | { | |
2731 | va_list args; | |
2732 | char temp[64]; | |
2733 | int sz; | |
2734 | ||
2735 | va_start(args, fmt); | |
2736 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2737 | va_end(args); | |
2738 | ||
2739 | if (sz > sizeof(temp) || sz > buflen) | |
2740 | return ERR_PTR(-ENAMETOOLONG); | |
2741 | ||
2742 | buffer += buflen - sz; | |
2743 | return memcpy(buffer, temp, sz); | |
2744 | } | |
2745 | ||
ad4c3cc4 AV |
2746 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
2747 | { | |
2748 | char *end = buffer + buflen; | |
2749 | /* these dentries are never renamed, so d_lock is not needed */ | |
2750 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
2751 | prepend_name(&end, &buflen, &dentry->d_name) || | |
2752 | prepend(&end, &buflen, "/", 1)) | |
2753 | end = ERR_PTR(-ENAMETOOLONG); | |
2754 | return end; | |
2755 | } | |
2756 | ||
6092d048 RP |
2757 | /* |
2758 | * Write full pathname from the root of the filesystem into the buffer. | |
2759 | */ | |
ec2447c2 | 2760 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2761 | { |
2762 | char *end = buf + buflen; | |
2763 | char *retval; | |
2764 | ||
6092d048 | 2765 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2766 | if (buflen < 1) |
2767 | goto Elong; | |
2768 | /* Get '/' right */ | |
2769 | retval = end-1; | |
2770 | *retval = '/'; | |
2771 | ||
cdd16d02 MS |
2772 | while (!IS_ROOT(dentry)) { |
2773 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2774 | int error; |
6092d048 | 2775 | |
6092d048 | 2776 | prefetch(parent); |
9abca360 NP |
2777 | spin_lock(&dentry->d_lock); |
2778 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2779 | spin_unlock(&dentry->d_lock); | |
2780 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2781 | goto Elong; |
2782 | ||
2783 | retval = end; | |
2784 | dentry = parent; | |
2785 | } | |
c103135c AV |
2786 | return retval; |
2787 | Elong: | |
2788 | return ERR_PTR(-ENAMETOOLONG); | |
2789 | } | |
ec2447c2 NP |
2790 | |
2791 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2792 | { | |
2793 | char *retval; | |
2794 | ||
949854d0 | 2795 | write_seqlock(&rename_lock); |
ec2447c2 | 2796 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2797 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2798 | |
2799 | return retval; | |
2800 | } | |
2801 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2802 | |
2803 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2804 | { | |
2805 | char *p = NULL; | |
2806 | char *retval; | |
2807 | ||
949854d0 | 2808 | write_seqlock(&rename_lock); |
c103135c AV |
2809 | if (d_unlinked(dentry)) { |
2810 | p = buf + buflen; | |
2811 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2812 | goto Elong; | |
2813 | buflen++; | |
2814 | } | |
2815 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2816 | write_sequnlock(&rename_lock); |
c103135c AV |
2817 | if (!IS_ERR(retval) && p) |
2818 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2819 | return retval; |
2820 | Elong: | |
6092d048 RP |
2821 | return ERR_PTR(-ENAMETOOLONG); |
2822 | } | |
2823 | ||
1da177e4 LT |
2824 | /* |
2825 | * NOTE! The user-level library version returns a | |
2826 | * character pointer. The kernel system call just | |
2827 | * returns the length of the buffer filled (which | |
2828 | * includes the ending '\0' character), or a negative | |
2829 | * error value. So libc would do something like | |
2830 | * | |
2831 | * char *getcwd(char * buf, size_t size) | |
2832 | * { | |
2833 | * int retval; | |
2834 | * | |
2835 | * retval = sys_getcwd(buf, size); | |
2836 | * if (retval >= 0) | |
2837 | * return buf; | |
2838 | * errno = -retval; | |
2839 | * return NULL; | |
2840 | * } | |
2841 | */ | |
3cdad428 | 2842 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2843 | { |
552ce544 | 2844 | int error; |
6ac08c39 | 2845 | struct path pwd, root; |
552ce544 | 2846 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2847 | |
2848 | if (!page) | |
2849 | return -ENOMEM; | |
2850 | ||
f7ad3c6b | 2851 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2852 | |
552ce544 | 2853 | error = -ENOENT; |
7ea600b5 | 2854 | br_read_lock(&vfsmount_lock); |
949854d0 | 2855 | write_seqlock(&rename_lock); |
f3da392e | 2856 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2857 | unsigned long len; |
8df9d1a4 MS |
2858 | char *cwd = page + PAGE_SIZE; |
2859 | int buflen = PAGE_SIZE; | |
1da177e4 | 2860 | |
8df9d1a4 | 2861 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 2862 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
949854d0 | 2863 | write_sequnlock(&rename_lock); |
7ea600b5 | 2864 | br_read_unlock(&vfsmount_lock); |
552ce544 | 2865 | |
02125a82 | 2866 | if (error < 0) |
552ce544 LT |
2867 | goto out; |
2868 | ||
8df9d1a4 | 2869 | /* Unreachable from current root */ |
02125a82 | 2870 | if (error > 0) { |
8df9d1a4 MS |
2871 | error = prepend_unreachable(&cwd, &buflen); |
2872 | if (error) | |
2873 | goto out; | |
2874 | } | |
2875 | ||
552ce544 LT |
2876 | error = -ERANGE; |
2877 | len = PAGE_SIZE + page - cwd; | |
2878 | if (len <= size) { | |
2879 | error = len; | |
2880 | if (copy_to_user(buf, cwd, len)) | |
2881 | error = -EFAULT; | |
2882 | } | |
949854d0 NP |
2883 | } else { |
2884 | write_sequnlock(&rename_lock); | |
7ea600b5 | 2885 | br_read_unlock(&vfsmount_lock); |
949854d0 | 2886 | } |
1da177e4 LT |
2887 | |
2888 | out: | |
6ac08c39 JB |
2889 | path_put(&pwd); |
2890 | path_put(&root); | |
1da177e4 LT |
2891 | free_page((unsigned long) page); |
2892 | return error; | |
2893 | } | |
2894 | ||
2895 | /* | |
2896 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2897 | * | |
2898 | * Trivially implemented using the dcache structure | |
2899 | */ | |
2900 | ||
2901 | /** | |
2902 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2903 | * @new_dentry: new dentry | |
2904 | * @old_dentry: old dentry | |
2905 | * | |
2906 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2907 | * Returns 0 otherwise. | |
2908 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2909 | */ | |
2910 | ||
e2761a11 | 2911 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2912 | { |
2913 | int result; | |
949854d0 | 2914 | unsigned seq; |
1da177e4 | 2915 | |
e2761a11 OH |
2916 | if (new_dentry == old_dentry) |
2917 | return 1; | |
2918 | ||
e2761a11 | 2919 | do { |
1da177e4 | 2920 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2921 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2922 | /* |
2923 | * Need rcu_readlock to protect against the d_parent trashing | |
2924 | * due to d_move | |
2925 | */ | |
2926 | rcu_read_lock(); | |
e2761a11 | 2927 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2928 | result = 1; |
e2761a11 OH |
2929 | else |
2930 | result = 0; | |
949854d0 | 2931 | rcu_read_unlock(); |
1da177e4 | 2932 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2933 | |
2934 | return result; | |
2935 | } | |
2936 | ||
2937 | void d_genocide(struct dentry *root) | |
2938 | { | |
949854d0 | 2939 | struct dentry *this_parent; |
1da177e4 | 2940 | struct list_head *next; |
949854d0 | 2941 | unsigned seq; |
58db63d0 | 2942 | int locked = 0; |
1da177e4 | 2943 | |
949854d0 | 2944 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2945 | again: |
2946 | this_parent = root; | |
2fd6b7f5 | 2947 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2948 | repeat: |
2949 | next = this_parent->d_subdirs.next; | |
2950 | resume: | |
2951 | while (next != &this_parent->d_subdirs) { | |
2952 | struct list_head *tmp = next; | |
6637ecd3 | 2953 | struct dentry *dentry = list_entry(tmp, struct dentry, d_child); |
1da177e4 | 2954 | next = tmp->next; |
949854d0 | 2955 | |
da502956 NP |
2956 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2957 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2958 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2959 | continue; |
da502956 | 2960 | } |
1da177e4 | 2961 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2962 | spin_unlock(&this_parent->d_lock); |
2963 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2964 | this_parent = dentry; |
2fd6b7f5 | 2965 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2966 | goto repeat; |
2967 | } | |
949854d0 NP |
2968 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2969 | dentry->d_flags |= DCACHE_GENOCIDE; | |
2970 | dentry->d_count--; | |
2971 | } | |
b7ab39f6 | 2972 | spin_unlock(&dentry->d_lock); |
1da177e4 | 2973 | } |
5f03ac13 AV |
2974 | rcu_read_lock(); |
2975 | ascend: | |
1da177e4 | 2976 | if (this_parent != root) { |
c826cb7d | 2977 | struct dentry *child = this_parent; |
949854d0 NP |
2978 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2979 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
2980 | this_parent->d_count--; | |
2981 | } | |
5f03ac13 AV |
2982 | this_parent = child->d_parent; |
2983 | ||
2984 | spin_unlock(&child->d_lock); | |
2985 | spin_lock(&this_parent->d_lock); | |
2986 | ||
2987 | /* might go back up the wrong parent if we have had a rename. */ | |
2988 | if (!locked && read_seqretry(&rename_lock, seq)) | |
949854d0 | 2989 | goto rename_retry; |
b93e0acc JR |
2990 | /* go into the first sibling still alive */ |
2991 | do { | |
2992 | next = child->d_child.next; | |
5f03ac13 AV |
2993 | if (next == &this_parent->d_subdirs) |
2994 | goto ascend; | |
2995 | child = list_entry(next, struct dentry, d_child); | |
b93e0acc | 2996 | } while (unlikely(child->d_flags & DCACHE_DENTRY_KILLED)); |
5f03ac13 | 2997 | rcu_read_unlock(); |
1da177e4 LT |
2998 | goto resume; |
2999 | } | |
58db63d0 | 3000 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 3001 | goto rename_retry; |
5f03ac13 AV |
3002 | spin_unlock(&this_parent->d_lock); |
3003 | rcu_read_unlock(); | |
58db63d0 NP |
3004 | if (locked) |
3005 | write_sequnlock(&rename_lock); | |
3006 | return; | |
3007 | ||
3008 | rename_retry: | |
5f03ac13 AV |
3009 | spin_unlock(&this_parent->d_lock); |
3010 | rcu_read_unlock(); | |
8110e16d MS |
3011 | if (locked) |
3012 | goto again; | |
58db63d0 NP |
3013 | locked = 1; |
3014 | write_seqlock(&rename_lock); | |
3015 | goto again; | |
1da177e4 LT |
3016 | } |
3017 | ||
3018 | /** | |
3019 | * find_inode_number - check for dentry with name | |
3020 | * @dir: directory to check | |
3021 | * @name: Name to find. | |
3022 | * | |
3023 | * Check whether a dentry already exists for the given name, | |
3024 | * and return the inode number if it has an inode. Otherwise | |
3025 | * 0 is returned. | |
3026 | * | |
3027 | * This routine is used to post-process directory listings for | |
3028 | * filesystems using synthetic inode numbers, and is necessary | |
3029 | * to keep getcwd() working. | |
3030 | */ | |
3031 | ||
3032 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
3033 | { | |
3034 | struct dentry * dentry; | |
3035 | ino_t ino = 0; | |
3036 | ||
3e7e241f | 3037 | dentry = d_hash_and_lookup(dir, name); |
4f522a24 | 3038 | if (!IS_ERR_OR_NULL(dentry)) { |
1da177e4 LT |
3039 | if (dentry->d_inode) |
3040 | ino = dentry->d_inode->i_ino; | |
3041 | dput(dentry); | |
3042 | } | |
1da177e4 LT |
3043 | return ino; |
3044 | } | |
ec4f8605 | 3045 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
3046 | |
3047 | static __initdata unsigned long dhash_entries; | |
3048 | static int __init set_dhash_entries(char *str) | |
3049 | { | |
3050 | if (!str) | |
3051 | return 0; | |
3052 | dhash_entries = simple_strtoul(str, &str, 0); | |
3053 | return 1; | |
3054 | } | |
3055 | __setup("dhash_entries=", set_dhash_entries); | |
3056 | ||
3057 | static void __init dcache_init_early(void) | |
3058 | { | |
074b8517 | 3059 | unsigned int loop; |
1da177e4 LT |
3060 | |
3061 | /* If hashes are distributed across NUMA nodes, defer | |
3062 | * hash allocation until vmalloc space is available. | |
3063 | */ | |
3064 | if (hashdist) | |
3065 | return; | |
3066 | ||
3067 | dentry_hashtable = | |
3068 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3069 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3070 | dhash_entries, |
3071 | 13, | |
3072 | HASH_EARLY, | |
3073 | &d_hash_shift, | |
3074 | &d_hash_mask, | |
31fe62b9 | 3075 | 0, |
1da177e4 LT |
3076 | 0); |
3077 | ||
074b8517 | 3078 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3079 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3080 | } |
3081 | ||
74bf17cf | 3082 | static void __init dcache_init(void) |
1da177e4 | 3083 | { |
074b8517 | 3084 | unsigned int loop; |
1da177e4 LT |
3085 | |
3086 | /* | |
3087 | * A constructor could be added for stable state like the lists, | |
3088 | * but it is probably not worth it because of the cache nature | |
3089 | * of the dcache. | |
3090 | */ | |
0a31bd5f CL |
3091 | dentry_cache = KMEM_CACHE(dentry, |
3092 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3093 | |
3094 | /* Hash may have been set up in dcache_init_early */ | |
3095 | if (!hashdist) | |
3096 | return; | |
3097 | ||
3098 | dentry_hashtable = | |
3099 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3100 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3101 | dhash_entries, |
3102 | 13, | |
3103 | 0, | |
3104 | &d_hash_shift, | |
3105 | &d_hash_mask, | |
31fe62b9 | 3106 | 0, |
1da177e4 LT |
3107 | 0); |
3108 | ||
074b8517 | 3109 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3110 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3111 | } |
3112 | ||
3113 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3114 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3115 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3116 | |
1da177e4 LT |
3117 | EXPORT_SYMBOL(d_genocide); |
3118 | ||
1da177e4 LT |
3119 | void __init vfs_caches_init_early(void) |
3120 | { | |
3121 | dcache_init_early(); | |
3122 | inode_init_early(); | |
3123 | } | |
3124 | ||
3125 | void __init vfs_caches_init(unsigned long mempages) | |
3126 | { | |
3127 | unsigned long reserve; | |
3128 | ||
3129 | /* Base hash sizes on available memory, with a reserve equal to | |
3130 | 150% of current kernel size */ | |
3131 | ||
3132 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3133 | mempages -= reserve; | |
3134 | ||
3135 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3136 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3137 | |
74bf17cf DC |
3138 | dcache_init(); |
3139 | inode_init(); | |
1da177e4 | 3140 | files_init(mempages); |
74bf17cf | 3141 | mnt_init(); |
1da177e4 LT |
3142 | bdev_cache_init(); |
3143 | chrdev_init(); | |
3144 | } |