Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/dcache.c | |
3 | * | |
4 | * Complete reimplementation | |
5 | * (C) 1997 Thomas Schoebel-Theuer, | |
6 | * with heavy changes by Linus Torvalds | |
7 | */ | |
8 | ||
9 | /* | |
10 | * Notes on the allocation strategy: | |
11 | * | |
12 | * The dcache is a master of the icache - whenever a dcache entry | |
13 | * exists, the inode will always exist. "iput()" is done either when | |
14 | * the dcache entry is deleted or garbage collected. | |
15 | */ | |
16 | ||
1da177e4 LT |
17 | #include <linux/syscalls.h> |
18 | #include <linux/string.h> | |
19 | #include <linux/mm.h> | |
20 | #include <linux/fs.h> | |
7a91bf7f | 21 | #include <linux/fsnotify.h> |
1da177e4 LT |
22 | #include <linux/slab.h> |
23 | #include <linux/init.h> | |
1da177e4 LT |
24 | #include <linux/hash.h> |
25 | #include <linux/cache.h> | |
630d9c47 | 26 | #include <linux/export.h> |
1da177e4 LT |
27 | #include <linux/mount.h> |
28 | #include <linux/file.h> | |
29 | #include <asm/uaccess.h> | |
30 | #include <linux/security.h> | |
31 | #include <linux/seqlock.h> | |
32 | #include <linux/swap.h> | |
33 | #include <linux/bootmem.h> | |
5ad4e53b | 34 | #include <linux/fs_struct.h> |
613afbf8 | 35 | #include <linux/hardirq.h> |
ceb5bdc2 NP |
36 | #include <linux/bit_spinlock.h> |
37 | #include <linux/rculist_bl.h> | |
268bb0ce | 38 | #include <linux/prefetch.h> |
dd179946 | 39 | #include <linux/ratelimit.h> |
07f3f05c | 40 | #include "internal.h" |
b2dba1af | 41 | #include "mount.h" |
1da177e4 | 42 | |
789680d1 NP |
43 | /* |
44 | * Usage: | |
873feea0 NP |
45 | * dcache->d_inode->i_lock protects: |
46 | * - i_dentry, d_alias, d_inode of aliases | |
ceb5bdc2 NP |
47 | * dcache_hash_bucket lock protects: |
48 | * - the dcache hash table | |
49 | * s_anon bl list spinlock protects: | |
50 | * - the s_anon list (see __d_drop) | |
23044507 NP |
51 | * dcache_lru_lock protects: |
52 | * - the dcache lru lists and counters | |
53 | * d_lock protects: | |
54 | * - d_flags | |
55 | * - d_name | |
56 | * - d_lru | |
b7ab39f6 | 57 | * - d_count |
da502956 | 58 | * - d_unhashed() |
2fd6b7f5 NP |
59 | * - d_parent and d_subdirs |
60 | * - childrens' d_child and d_parent | |
b23fb0a6 | 61 | * - d_alias, d_inode |
789680d1 NP |
62 | * |
63 | * Ordering: | |
873feea0 | 64 | * dentry->d_inode->i_lock |
b5c84bf6 NP |
65 | * dentry->d_lock |
66 | * dcache_lru_lock | |
ceb5bdc2 NP |
67 | * dcache_hash_bucket lock |
68 | * s_anon lock | |
789680d1 | 69 | * |
da502956 NP |
70 | * If there is an ancestor relationship: |
71 | * dentry->d_parent->...->d_parent->d_lock | |
72 | * ... | |
73 | * dentry->d_parent->d_lock | |
74 | * dentry->d_lock | |
75 | * | |
76 | * If no ancestor relationship: | |
789680d1 NP |
77 | * if (dentry1 < dentry2) |
78 | * dentry1->d_lock | |
79 | * dentry2->d_lock | |
80 | */ | |
fa3536cc | 81 | int sysctl_vfs_cache_pressure __read_mostly = 100; |
1da177e4 LT |
82 | EXPORT_SYMBOL_GPL(sysctl_vfs_cache_pressure); |
83 | ||
23044507 | 84 | static __cacheline_aligned_in_smp DEFINE_SPINLOCK(dcache_lru_lock); |
74c3cbe3 | 85 | __cacheline_aligned_in_smp DEFINE_SEQLOCK(rename_lock); |
1da177e4 | 86 | |
949854d0 | 87 | EXPORT_SYMBOL(rename_lock); |
1da177e4 | 88 | |
e18b890b | 89 | static struct kmem_cache *dentry_cache __read_mostly; |
1da177e4 | 90 | |
1da177e4 LT |
91 | /* |
92 | * This is the single most critical data structure when it comes | |
93 | * to the dcache: the hashtable for lookups. Somebody should try | |
94 | * to make this good - I've just made it work. | |
95 | * | |
96 | * This hash-function tries to avoid losing too many bits of hash | |
97 | * information, yet avoid using a prime hash-size or similar. | |
98 | */ | |
99 | #define D_HASHBITS d_hash_shift | |
100 | #define D_HASHMASK d_hash_mask | |
101 | ||
fa3536cc ED |
102 | static unsigned int d_hash_mask __read_mostly; |
103 | static unsigned int d_hash_shift __read_mostly; | |
ceb5bdc2 | 104 | |
b07ad996 | 105 | static struct hlist_bl_head *dentry_hashtable __read_mostly; |
ceb5bdc2 | 106 | |
8966be90 | 107 | static inline struct hlist_bl_head *d_hash(const struct dentry *parent, |
6d7d1a0d | 108 | unsigned int hash) |
ceb5bdc2 | 109 | { |
6d7d1a0d LT |
110 | hash += (unsigned long) parent / L1_CACHE_BYTES; |
111 | hash = hash + (hash >> D_HASHBITS); | |
ceb5bdc2 NP |
112 | return dentry_hashtable + (hash & D_HASHMASK); |
113 | } | |
114 | ||
1da177e4 LT |
115 | /* Statistics gathering. */ |
116 | struct dentry_stat_t dentry_stat = { | |
117 | .age_limit = 45, | |
118 | }; | |
119 | ||
3e880fb5 | 120 | static DEFINE_PER_CPU(unsigned int, nr_dentry); |
312d3ca8 CH |
121 | |
122 | #if defined(CONFIG_SYSCTL) && defined(CONFIG_PROC_FS) | |
3e880fb5 NP |
123 | static int get_nr_dentry(void) |
124 | { | |
125 | int i; | |
126 | int sum = 0; | |
127 | for_each_possible_cpu(i) | |
128 | sum += per_cpu(nr_dentry, i); | |
129 | return sum < 0 ? 0 : sum; | |
130 | } | |
131 | ||
312d3ca8 CH |
132 | int proc_nr_dentry(ctl_table *table, int write, void __user *buffer, |
133 | size_t *lenp, loff_t *ppos) | |
134 | { | |
3e880fb5 | 135 | dentry_stat.nr_dentry = get_nr_dentry(); |
312d3ca8 CH |
136 | return proc_dointvec(table, write, buffer, lenp, ppos); |
137 | } | |
138 | #endif | |
139 | ||
5483f18e LT |
140 | /* |
141 | * Compare 2 name strings, return 0 if they match, otherwise non-zero. | |
142 | * The strings are both count bytes long, and count is non-zero. | |
143 | */ | |
e419b4cc LT |
144 | #ifdef CONFIG_DCACHE_WORD_ACCESS |
145 | ||
146 | #include <asm/word-at-a-time.h> | |
147 | /* | |
148 | * NOTE! 'cs' and 'scount' come from a dentry, so it has a | |
149 | * aligned allocation for this particular component. We don't | |
150 | * strictly need the load_unaligned_zeropad() safety, but it | |
151 | * doesn't hurt either. | |
152 | * | |
153 | * In contrast, 'ct' and 'tcount' can be from a pathname, and do | |
154 | * need the careful unaligned handling. | |
155 | */ | |
94753db5 | 156 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
5483f18e | 157 | { |
bfcfaa77 | 158 | unsigned long a,b,mask; |
bfcfaa77 LT |
159 | |
160 | for (;;) { | |
12f8ad4b | 161 | a = *(unsigned long *)cs; |
e419b4cc | 162 | b = load_unaligned_zeropad(ct); |
bfcfaa77 LT |
163 | if (tcount < sizeof(unsigned long)) |
164 | break; | |
165 | if (unlikely(a != b)) | |
166 | return 1; | |
167 | cs += sizeof(unsigned long); | |
168 | ct += sizeof(unsigned long); | |
169 | tcount -= sizeof(unsigned long); | |
170 | if (!tcount) | |
171 | return 0; | |
172 | } | |
173 | mask = ~(~0ul << tcount*8); | |
174 | return unlikely(!!((a ^ b) & mask)); | |
e419b4cc LT |
175 | } |
176 | ||
bfcfaa77 | 177 | #else |
e419b4cc | 178 | |
94753db5 | 179 | static inline int dentry_string_cmp(const unsigned char *cs, const unsigned char *ct, unsigned tcount) |
e419b4cc | 180 | { |
5483f18e LT |
181 | do { |
182 | if (*cs != *ct) | |
183 | return 1; | |
184 | cs++; | |
185 | ct++; | |
186 | tcount--; | |
187 | } while (tcount); | |
188 | return 0; | |
189 | } | |
190 | ||
e419b4cc LT |
191 | #endif |
192 | ||
94753db5 LT |
193 | static inline int dentry_cmp(const struct dentry *dentry, const unsigned char *ct, unsigned tcount) |
194 | { | |
6326c71f | 195 | const unsigned char *cs; |
94753db5 LT |
196 | /* |
197 | * Be careful about RCU walk racing with rename: | |
198 | * use ACCESS_ONCE to fetch the name pointer. | |
199 | * | |
200 | * NOTE! Even if a rename will mean that the length | |
201 | * was not loaded atomically, we don't care. The | |
202 | * RCU walk will check the sequence count eventually, | |
203 | * and catch it. And we won't overrun the buffer, | |
204 | * because we're reading the name pointer atomically, | |
205 | * and a dentry name is guaranteed to be properly | |
206 | * terminated with a NUL byte. | |
207 | * | |
208 | * End result: even if 'len' is wrong, we'll exit | |
209 | * early because the data cannot match (there can | |
210 | * be no NUL in the ct/tcount data) | |
211 | */ | |
6326c71f LT |
212 | cs = ACCESS_ONCE(dentry->d_name.name); |
213 | smp_read_barrier_depends(); | |
214 | return dentry_string_cmp(cs, ct, tcount); | |
94753db5 LT |
215 | } |
216 | ||
9c82ab9c | 217 | static void __d_free(struct rcu_head *head) |
1da177e4 | 218 | { |
9c82ab9c CH |
219 | struct dentry *dentry = container_of(head, struct dentry, d_u.d_rcu); |
220 | ||
fd217f4d | 221 | WARN_ON(!list_empty(&dentry->d_alias)); |
1da177e4 LT |
222 | if (dname_external(dentry)) |
223 | kfree(dentry->d_name.name); | |
224 | kmem_cache_free(dentry_cache, dentry); | |
225 | } | |
226 | ||
227 | /* | |
b5c84bf6 | 228 | * no locks, please. |
1da177e4 LT |
229 | */ |
230 | static void d_free(struct dentry *dentry) | |
231 | { | |
b7ab39f6 | 232 | BUG_ON(dentry->d_count); |
3e880fb5 | 233 | this_cpu_dec(nr_dentry); |
1da177e4 LT |
234 | if (dentry->d_op && dentry->d_op->d_release) |
235 | dentry->d_op->d_release(dentry); | |
312d3ca8 | 236 | |
dea3667b LT |
237 | /* if dentry was never visible to RCU, immediate free is OK */ |
238 | if (!(dentry->d_flags & DCACHE_RCUACCESS)) | |
9c82ab9c | 239 | __d_free(&dentry->d_u.d_rcu); |
b3423415 | 240 | else |
9c82ab9c | 241 | call_rcu(&dentry->d_u.d_rcu, __d_free); |
1da177e4 LT |
242 | } |
243 | ||
31e6b01f NP |
244 | /** |
245 | * dentry_rcuwalk_barrier - invalidate in-progress rcu-walk lookups | |
ff5fdb61 | 246 | * @dentry: the target dentry |
31e6b01f NP |
247 | * After this call, in-progress rcu-walk path lookup will fail. This |
248 | * should be called after unhashing, and after changing d_inode (if | |
249 | * the dentry has not already been unhashed). | |
250 | */ | |
251 | static inline void dentry_rcuwalk_barrier(struct dentry *dentry) | |
252 | { | |
253 | assert_spin_locked(&dentry->d_lock); | |
254 | /* Go through a barrier */ | |
255 | write_seqcount_barrier(&dentry->d_seq); | |
256 | } | |
257 | ||
1da177e4 LT |
258 | /* |
259 | * Release the dentry's inode, using the filesystem | |
31e6b01f NP |
260 | * d_iput() operation if defined. Dentry has no refcount |
261 | * and is unhashed. | |
1da177e4 | 262 | */ |
858119e1 | 263 | static void dentry_iput(struct dentry * dentry) |
31f3e0b3 | 264 | __releases(dentry->d_lock) |
873feea0 | 265 | __releases(dentry->d_inode->i_lock) |
1da177e4 LT |
266 | { |
267 | struct inode *inode = dentry->d_inode; | |
268 | if (inode) { | |
269 | dentry->d_inode = NULL; | |
270 | list_del_init(&dentry->d_alias); | |
271 | spin_unlock(&dentry->d_lock); | |
873feea0 | 272 | spin_unlock(&inode->i_lock); |
f805fbda LT |
273 | if (!inode->i_nlink) |
274 | fsnotify_inoderemove(inode); | |
1da177e4 LT |
275 | if (dentry->d_op && dentry->d_op->d_iput) |
276 | dentry->d_op->d_iput(dentry, inode); | |
277 | else | |
278 | iput(inode); | |
279 | } else { | |
280 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
281 | } |
282 | } | |
283 | ||
31e6b01f NP |
284 | /* |
285 | * Release the dentry's inode, using the filesystem | |
286 | * d_iput() operation if defined. dentry remains in-use. | |
287 | */ | |
288 | static void dentry_unlink_inode(struct dentry * dentry) | |
289 | __releases(dentry->d_lock) | |
873feea0 | 290 | __releases(dentry->d_inode->i_lock) |
31e6b01f NP |
291 | { |
292 | struct inode *inode = dentry->d_inode; | |
293 | dentry->d_inode = NULL; | |
294 | list_del_init(&dentry->d_alias); | |
295 | dentry_rcuwalk_barrier(dentry); | |
296 | spin_unlock(&dentry->d_lock); | |
873feea0 | 297 | spin_unlock(&inode->i_lock); |
31e6b01f NP |
298 | if (!inode->i_nlink) |
299 | fsnotify_inoderemove(inode); | |
300 | if (dentry->d_op && dentry->d_op->d_iput) | |
301 | dentry->d_op->d_iput(dentry, inode); | |
302 | else | |
303 | iput(inode); | |
304 | } | |
305 | ||
da3bbdd4 | 306 | /* |
f0023bc6 | 307 | * dentry_lru_(add|del|prune|move_tail) must be called with d_lock held. |
da3bbdd4 KM |
308 | */ |
309 | static void dentry_lru_add(struct dentry *dentry) | |
310 | { | |
a4633357 | 311 | if (list_empty(&dentry->d_lru)) { |
23044507 | 312 | spin_lock(&dcache_lru_lock); |
a4633357 CH |
313 | list_add(&dentry->d_lru, &dentry->d_sb->s_dentry_lru); |
314 | dentry->d_sb->s_nr_dentry_unused++; | |
86c8749e | 315 | dentry_stat.nr_unused++; |
23044507 | 316 | spin_unlock(&dcache_lru_lock); |
a4633357 | 317 | } |
da3bbdd4 KM |
318 | } |
319 | ||
23044507 NP |
320 | static void __dentry_lru_del(struct dentry *dentry) |
321 | { | |
322 | list_del_init(&dentry->d_lru); | |
eaf5f907 | 323 | dentry->d_flags &= ~DCACHE_SHRINK_LIST; |
23044507 NP |
324 | dentry->d_sb->s_nr_dentry_unused--; |
325 | dentry_stat.nr_unused--; | |
326 | } | |
327 | ||
f0023bc6 SW |
328 | /* |
329 | * Remove a dentry with references from the LRU. | |
330 | */ | |
da3bbdd4 KM |
331 | static void dentry_lru_del(struct dentry *dentry) |
332 | { | |
333 | if (!list_empty(&dentry->d_lru)) { | |
23044507 NP |
334 | spin_lock(&dcache_lru_lock); |
335 | __dentry_lru_del(dentry); | |
336 | spin_unlock(&dcache_lru_lock); | |
da3bbdd4 KM |
337 | } |
338 | } | |
339 | ||
f0023bc6 SW |
340 | /* |
341 | * Remove a dentry that is unreferenced and about to be pruned | |
342 | * (unhashed and destroyed) from the LRU, and inform the file system. | |
343 | * This wrapper should be called _prior_ to unhashing a victim dentry. | |
344 | */ | |
345 | static void dentry_lru_prune(struct dentry *dentry) | |
346 | { | |
347 | if (!list_empty(&dentry->d_lru)) { | |
348 | if (dentry->d_flags & DCACHE_OP_PRUNE) | |
349 | dentry->d_op->d_prune(dentry); | |
350 | ||
351 | spin_lock(&dcache_lru_lock); | |
352 | __dentry_lru_del(dentry); | |
353 | spin_unlock(&dcache_lru_lock); | |
354 | } | |
355 | } | |
356 | ||
b48f03b3 | 357 | static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list) |
da3bbdd4 | 358 | { |
23044507 | 359 | spin_lock(&dcache_lru_lock); |
a4633357 | 360 | if (list_empty(&dentry->d_lru)) { |
b48f03b3 | 361 | list_add_tail(&dentry->d_lru, list); |
a4633357 | 362 | dentry->d_sb->s_nr_dentry_unused++; |
86c8749e | 363 | dentry_stat.nr_unused++; |
a4633357 | 364 | } else { |
b48f03b3 | 365 | list_move_tail(&dentry->d_lru, list); |
da3bbdd4 | 366 | } |
23044507 | 367 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
368 | } |
369 | ||
d52b9086 MS |
370 | /** |
371 | * d_kill - kill dentry and return parent | |
372 | * @dentry: dentry to kill | |
ff5fdb61 | 373 | * @parent: parent dentry |
d52b9086 | 374 | * |
31f3e0b3 | 375 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
376 | * |
377 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 378 | * |
b5c84bf6 NP |
379 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
380 | * d_kill. | |
d52b9086 | 381 | */ |
2fd6b7f5 | 382 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 383 | __releases(dentry->d_lock) |
2fd6b7f5 | 384 | __releases(parent->d_lock) |
873feea0 | 385 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 386 | { |
d52b9086 | 387 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
388 | /* |
389 | * Inform try_to_ascend() that we are no longer attached to the | |
390 | * dentry tree | |
391 | */ | |
392 | dentry->d_flags |= DCACHE_DISCONNECTED; | |
2fd6b7f5 NP |
393 | if (parent) |
394 | spin_unlock(&parent->d_lock); | |
d52b9086 | 395 | dentry_iput(dentry); |
b7ab39f6 NP |
396 | /* |
397 | * dentry_iput drops the locks, at which point nobody (except | |
398 | * transient RCU lookups) can reach this dentry. | |
399 | */ | |
d52b9086 | 400 | d_free(dentry); |
871c0067 | 401 | return parent; |
d52b9086 MS |
402 | } |
403 | ||
c6627c60 DH |
404 | /* |
405 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
406 | * dentry->d_lock is locked. The caller must take care of that, if | |
407 | * appropriate. | |
408 | */ | |
409 | static void __d_shrink(struct dentry *dentry) | |
410 | { | |
411 | if (!d_unhashed(dentry)) { | |
412 | struct hlist_bl_head *b; | |
413 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
414 | b = &dentry->d_sb->s_anon; | |
415 | else | |
416 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
417 | ||
418 | hlist_bl_lock(b); | |
419 | __hlist_bl_del(&dentry->d_hash); | |
420 | dentry->d_hash.pprev = NULL; | |
421 | hlist_bl_unlock(b); | |
422 | } | |
423 | } | |
424 | ||
789680d1 NP |
425 | /** |
426 | * d_drop - drop a dentry | |
427 | * @dentry: dentry to drop | |
428 | * | |
429 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
430 | * be found through a VFS lookup any more. Note that this is different from | |
431 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
432 | * possible, giving a successful _negative_ lookup, while d_drop will | |
433 | * just make the cache lookup fail. | |
434 | * | |
435 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
436 | * reason (NFS timeouts or autofs deletes). | |
437 | * | |
438 | * __d_drop requires dentry->d_lock. | |
439 | */ | |
440 | void __d_drop(struct dentry *dentry) | |
441 | { | |
dea3667b | 442 | if (!d_unhashed(dentry)) { |
c6627c60 | 443 | __d_shrink(dentry); |
dea3667b | 444 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
445 | } |
446 | } | |
447 | EXPORT_SYMBOL(__d_drop); | |
448 | ||
449 | void d_drop(struct dentry *dentry) | |
450 | { | |
789680d1 NP |
451 | spin_lock(&dentry->d_lock); |
452 | __d_drop(dentry); | |
453 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
454 | } |
455 | EXPORT_SYMBOL(d_drop); | |
456 | ||
44396f4b JB |
457 | /* |
458 | * d_clear_need_lookup - drop a dentry from cache and clear the need lookup flag | |
459 | * @dentry: dentry to drop | |
460 | * | |
461 | * This is called when we do a lookup on a placeholder dentry that needed to be | |
462 | * looked up. The dentry should have been hashed in order for it to be found by | |
463 | * the lookup code, but now needs to be unhashed while we do the actual lookup | |
464 | * and clear the DCACHE_NEED_LOOKUP flag. | |
465 | */ | |
466 | void d_clear_need_lookup(struct dentry *dentry) | |
467 | { | |
468 | spin_lock(&dentry->d_lock); | |
469 | __d_drop(dentry); | |
470 | dentry->d_flags &= ~DCACHE_NEED_LOOKUP; | |
471 | spin_unlock(&dentry->d_lock); | |
472 | } | |
473 | EXPORT_SYMBOL(d_clear_need_lookup); | |
474 | ||
77812a1e NP |
475 | /* |
476 | * Finish off a dentry we've decided to kill. | |
477 | * dentry->d_lock must be held, returns with it unlocked. | |
478 | * If ref is non-zero, then decrement the refcount too. | |
479 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
480 | */ | |
481 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
482 | __releases(dentry->d_lock) | |
483 | { | |
873feea0 | 484 | struct inode *inode; |
77812a1e NP |
485 | struct dentry *parent; |
486 | ||
873feea0 NP |
487 | inode = dentry->d_inode; |
488 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
489 | relock: |
490 | spin_unlock(&dentry->d_lock); | |
491 | cpu_relax(); | |
492 | return dentry; /* try again with same dentry */ | |
493 | } | |
494 | if (IS_ROOT(dentry)) | |
495 | parent = NULL; | |
496 | else | |
497 | parent = dentry->d_parent; | |
498 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
499 | if (inode) |
500 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
501 | goto relock; |
502 | } | |
31e6b01f | 503 | |
77812a1e NP |
504 | if (ref) |
505 | dentry->d_count--; | |
f0023bc6 SW |
506 | /* |
507 | * if dentry was on the d_lru list delete it from there. | |
508 | * inform the fs via d_prune that this dentry is about to be | |
509 | * unhashed and destroyed. | |
510 | */ | |
511 | dentry_lru_prune(dentry); | |
77812a1e NP |
512 | /* if it was on the hash then remove it */ |
513 | __d_drop(dentry); | |
514 | return d_kill(dentry, parent); | |
515 | } | |
516 | ||
1da177e4 LT |
517 | /* |
518 | * This is dput | |
519 | * | |
520 | * This is complicated by the fact that we do not want to put | |
521 | * dentries that are no longer on any hash chain on the unused | |
522 | * list: we'd much rather just get rid of them immediately. | |
523 | * | |
524 | * However, that implies that we have to traverse the dentry | |
525 | * tree upwards to the parents which might _also_ now be | |
526 | * scheduled for deletion (it may have been only waiting for | |
527 | * its last child to go away). | |
528 | * | |
529 | * This tail recursion is done by hand as we don't want to depend | |
530 | * on the compiler to always get this right (gcc generally doesn't). | |
531 | * Real recursion would eat up our stack space. | |
532 | */ | |
533 | ||
534 | /* | |
535 | * dput - release a dentry | |
536 | * @dentry: dentry to release | |
537 | * | |
538 | * Release a dentry. This will drop the usage count and if appropriate | |
539 | * call the dentry unlink method as well as removing it from the queues and | |
540 | * releasing its resources. If the parent dentries were scheduled for release | |
541 | * they too may now get deleted. | |
1da177e4 | 542 | */ |
1da177e4 LT |
543 | void dput(struct dentry *dentry) |
544 | { | |
545 | if (!dentry) | |
546 | return; | |
547 | ||
548 | repeat: | |
b7ab39f6 | 549 | if (dentry->d_count == 1) |
1da177e4 | 550 | might_sleep(); |
1da177e4 | 551 | spin_lock(&dentry->d_lock); |
61f3dee4 NP |
552 | BUG_ON(!dentry->d_count); |
553 | if (dentry->d_count > 1) { | |
554 | dentry->d_count--; | |
1da177e4 | 555 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
556 | return; |
557 | } | |
558 | ||
fb045adb | 559 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 560 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 561 | goto kill_it; |
1da177e4 | 562 | } |
265ac902 | 563 | |
1da177e4 LT |
564 | /* Unreachable? Get rid of it */ |
565 | if (d_unhashed(dentry)) | |
566 | goto kill_it; | |
265ac902 | 567 | |
44396f4b JB |
568 | /* |
569 | * If this dentry needs lookup, don't set the referenced flag so that it | |
570 | * is more likely to be cleaned up by the dcache shrinker in case of | |
571 | * memory pressure. | |
572 | */ | |
573 | if (!d_need_lookup(dentry)) | |
574 | dentry->d_flags |= DCACHE_REFERENCED; | |
a4633357 | 575 | dentry_lru_add(dentry); |
265ac902 | 576 | |
61f3dee4 NP |
577 | dentry->d_count--; |
578 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
579 | return; |
580 | ||
d52b9086 | 581 | kill_it: |
77812a1e | 582 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
583 | if (dentry) |
584 | goto repeat; | |
1da177e4 | 585 | } |
ec4f8605 | 586 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
587 | |
588 | /** | |
589 | * d_invalidate - invalidate a dentry | |
590 | * @dentry: dentry to invalidate | |
591 | * | |
592 | * Try to invalidate the dentry if it turns out to be | |
593 | * possible. If there are other dentries that can be | |
594 | * reached through this one we can't delete it and we | |
595 | * return -EBUSY. On success we return 0. | |
596 | * | |
597 | * no dcache lock. | |
598 | */ | |
599 | ||
600 | int d_invalidate(struct dentry * dentry) | |
601 | { | |
602 | /* | |
603 | * If it's already been dropped, return OK. | |
604 | */ | |
da502956 | 605 | spin_lock(&dentry->d_lock); |
1da177e4 | 606 | if (d_unhashed(dentry)) { |
da502956 | 607 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
608 | return 0; |
609 | } | |
610 | /* | |
611 | * Check whether to do a partial shrink_dcache | |
612 | * to get rid of unused child entries. | |
613 | */ | |
614 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 615 | spin_unlock(&dentry->d_lock); |
1da177e4 | 616 | shrink_dcache_parent(dentry); |
da502956 | 617 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
618 | } |
619 | ||
620 | /* | |
621 | * Somebody else still using it? | |
622 | * | |
623 | * If it's a directory, we can't drop it | |
624 | * for fear of somebody re-populating it | |
625 | * with children (even though dropping it | |
626 | * would make it unreachable from the root, | |
627 | * we might still populate it if it was a | |
628 | * working directory or similar). | |
50e69630 AV |
629 | * We also need to leave mountpoints alone, |
630 | * directory or not. | |
1da177e4 | 631 | */ |
50e69630 AV |
632 | if (dentry->d_count > 1 && dentry->d_inode) { |
633 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { | |
1da177e4 | 634 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
635 | return -EBUSY; |
636 | } | |
637 | } | |
638 | ||
639 | __d_drop(dentry); | |
640 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
641 | return 0; |
642 | } | |
ec4f8605 | 643 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 644 | |
b5c84bf6 | 645 | /* This must be called with d_lock held */ |
dc0474be | 646 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 647 | { |
b7ab39f6 | 648 | dentry->d_count++; |
23044507 NP |
649 | } |
650 | ||
dc0474be | 651 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 652 | { |
23044507 | 653 | spin_lock(&dentry->d_lock); |
dc0474be | 654 | __dget_dlock(dentry); |
23044507 | 655 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
656 | } |
657 | ||
b7ab39f6 NP |
658 | struct dentry *dget_parent(struct dentry *dentry) |
659 | { | |
660 | struct dentry *ret; | |
661 | ||
662 | repeat: | |
a734eb45 NP |
663 | /* |
664 | * Don't need rcu_dereference because we re-check it was correct under | |
665 | * the lock. | |
666 | */ | |
667 | rcu_read_lock(); | |
b7ab39f6 | 668 | ret = dentry->d_parent; |
a734eb45 NP |
669 | spin_lock(&ret->d_lock); |
670 | if (unlikely(ret != dentry->d_parent)) { | |
671 | spin_unlock(&ret->d_lock); | |
672 | rcu_read_unlock(); | |
b7ab39f6 NP |
673 | goto repeat; |
674 | } | |
a734eb45 | 675 | rcu_read_unlock(); |
b7ab39f6 NP |
676 | BUG_ON(!ret->d_count); |
677 | ret->d_count++; | |
678 | spin_unlock(&ret->d_lock); | |
b7ab39f6 NP |
679 | return ret; |
680 | } | |
681 | EXPORT_SYMBOL(dget_parent); | |
682 | ||
1da177e4 LT |
683 | /** |
684 | * d_find_alias - grab a hashed alias of inode | |
685 | * @inode: inode in question | |
1da177e4 LT |
686 | * |
687 | * If inode has a hashed alias, or is a directory and has any alias, | |
688 | * acquire the reference to alias and return it. Otherwise return NULL. | |
689 | * Notice that if inode is a directory there can be only one alias and | |
690 | * it can be unhashed only if it has no children, or if it is the root | |
691 | * of a filesystem. | |
692 | * | |
21c0d8fd | 693 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
3f50fff4 | 694 | * any other hashed alias over that. |
1da177e4 | 695 | */ |
3f50fff4 | 696 | static struct dentry *__d_find_alias(struct inode *inode) |
1da177e4 | 697 | { |
da502956 | 698 | struct dentry *alias, *discon_alias; |
1da177e4 | 699 | |
da502956 NP |
700 | again: |
701 | discon_alias = NULL; | |
702 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { | |
703 | spin_lock(&alias->d_lock); | |
1da177e4 | 704 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 705 | if (IS_ROOT(alias) && |
da502956 | 706 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 707 | discon_alias = alias; |
3f50fff4 | 708 | } else { |
dc0474be | 709 | __dget_dlock(alias); |
da502956 NP |
710 | spin_unlock(&alias->d_lock); |
711 | return alias; | |
712 | } | |
713 | } | |
714 | spin_unlock(&alias->d_lock); | |
715 | } | |
716 | if (discon_alias) { | |
717 | alias = discon_alias; | |
718 | spin_lock(&alias->d_lock); | |
719 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
720 | if (IS_ROOT(alias) && | |
721 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 722 | __dget_dlock(alias); |
da502956 | 723 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
724 | return alias; |
725 | } | |
726 | } | |
da502956 NP |
727 | spin_unlock(&alias->d_lock); |
728 | goto again; | |
1da177e4 | 729 | } |
da502956 | 730 | return NULL; |
1da177e4 LT |
731 | } |
732 | ||
da502956 | 733 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 734 | { |
214fda1f DH |
735 | struct dentry *de = NULL; |
736 | ||
737 | if (!list_empty(&inode->i_dentry)) { | |
873feea0 | 738 | spin_lock(&inode->i_lock); |
3f50fff4 | 739 | de = __d_find_alias(inode); |
873feea0 | 740 | spin_unlock(&inode->i_lock); |
214fda1f | 741 | } |
1da177e4 LT |
742 | return de; |
743 | } | |
ec4f8605 | 744 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
745 | |
746 | /* | |
747 | * Try to kill dentries associated with this inode. | |
748 | * WARNING: you must own a reference to inode. | |
749 | */ | |
750 | void d_prune_aliases(struct inode *inode) | |
751 | { | |
0cdca3f9 | 752 | struct dentry *dentry; |
1da177e4 | 753 | restart: |
873feea0 | 754 | spin_lock(&inode->i_lock); |
0cdca3f9 | 755 | list_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 756 | spin_lock(&dentry->d_lock); |
b7ab39f6 | 757 | if (!dentry->d_count) { |
dc0474be | 758 | __dget_dlock(dentry); |
1da177e4 LT |
759 | __d_drop(dentry); |
760 | spin_unlock(&dentry->d_lock); | |
873feea0 | 761 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
762 | dput(dentry); |
763 | goto restart; | |
764 | } | |
765 | spin_unlock(&dentry->d_lock); | |
766 | } | |
873feea0 | 767 | spin_unlock(&inode->i_lock); |
1da177e4 | 768 | } |
ec4f8605 | 769 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
770 | |
771 | /* | |
77812a1e NP |
772 | * Try to throw away a dentry - free the inode, dput the parent. |
773 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
774 | * Releases dentry->d_lock. | |
d702ccb3 | 775 | * |
77812a1e | 776 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 777 | */ |
77812a1e | 778 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 779 | __releases(dentry->d_lock) |
1da177e4 | 780 | { |
77812a1e | 781 | struct dentry *parent; |
d52b9086 | 782 | |
77812a1e | 783 | parent = dentry_kill(dentry, 0); |
d52b9086 | 784 | /* |
77812a1e NP |
785 | * If dentry_kill returns NULL, we have nothing more to do. |
786 | * if it returns the same dentry, trylocks failed. In either | |
787 | * case, just loop again. | |
788 | * | |
789 | * Otherwise, we need to prune ancestors too. This is necessary | |
790 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
791 | * is also expected to be beneficial in reducing dentry cache | |
792 | * fragmentation. | |
d52b9086 | 793 | */ |
77812a1e NP |
794 | if (!parent) |
795 | return; | |
796 | if (parent == dentry) | |
797 | return; | |
798 | ||
799 | /* Prune ancestors. */ | |
800 | dentry = parent; | |
d52b9086 | 801 | while (dentry) { |
b7ab39f6 | 802 | spin_lock(&dentry->d_lock); |
89e60548 NP |
803 | if (dentry->d_count > 1) { |
804 | dentry->d_count--; | |
805 | spin_unlock(&dentry->d_lock); | |
806 | return; | |
807 | } | |
77812a1e | 808 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 809 | } |
1da177e4 LT |
810 | } |
811 | ||
3049cfe2 | 812 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 813 | { |
da3bbdd4 | 814 | struct dentry *dentry; |
da3bbdd4 | 815 | |
ec33679d NP |
816 | rcu_read_lock(); |
817 | for (;;) { | |
ec33679d NP |
818 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
819 | if (&dentry->d_lru == list) | |
820 | break; /* empty */ | |
821 | spin_lock(&dentry->d_lock); | |
822 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
823 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
824 | continue; |
825 | } | |
826 | ||
1da177e4 LT |
827 | /* |
828 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
829 | * the LRU because of laziness during lookup. Do not free |
830 | * it - just keep it off the LRU list. | |
1da177e4 | 831 | */ |
b7ab39f6 | 832 | if (dentry->d_count) { |
ec33679d | 833 | dentry_lru_del(dentry); |
da3bbdd4 | 834 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
835 | continue; |
836 | } | |
ec33679d | 837 | |
ec33679d | 838 | rcu_read_unlock(); |
77812a1e NP |
839 | |
840 | try_prune_one_dentry(dentry); | |
841 | ||
ec33679d | 842 | rcu_read_lock(); |
da3bbdd4 | 843 | } |
ec33679d | 844 | rcu_read_unlock(); |
3049cfe2 CH |
845 | } |
846 | ||
847 | /** | |
b48f03b3 DC |
848 | * prune_dcache_sb - shrink the dcache |
849 | * @sb: superblock | |
850 | * @count: number of entries to try to free | |
851 | * | |
852 | * Attempt to shrink the superblock dcache LRU by @count entries. This is | |
853 | * done when we need more memory an called from the superblock shrinker | |
854 | * function. | |
3049cfe2 | 855 | * |
b48f03b3 DC |
856 | * This function may fail to free any resources if all the dentries are in |
857 | * use. | |
3049cfe2 | 858 | */ |
b48f03b3 | 859 | void prune_dcache_sb(struct super_block *sb, int count) |
3049cfe2 | 860 | { |
3049cfe2 CH |
861 | struct dentry *dentry; |
862 | LIST_HEAD(referenced); | |
863 | LIST_HEAD(tmp); | |
3049cfe2 | 864 | |
23044507 NP |
865 | relock: |
866 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
867 | while (!list_empty(&sb->s_dentry_lru)) { |
868 | dentry = list_entry(sb->s_dentry_lru.prev, | |
869 | struct dentry, d_lru); | |
870 | BUG_ON(dentry->d_sb != sb); | |
871 | ||
23044507 NP |
872 | if (!spin_trylock(&dentry->d_lock)) { |
873 | spin_unlock(&dcache_lru_lock); | |
874 | cpu_relax(); | |
875 | goto relock; | |
876 | } | |
877 | ||
b48f03b3 | 878 | if (dentry->d_flags & DCACHE_REFERENCED) { |
23044507 NP |
879 | dentry->d_flags &= ~DCACHE_REFERENCED; |
880 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 881 | spin_unlock(&dentry->d_lock); |
23044507 NP |
882 | } else { |
883 | list_move_tail(&dentry->d_lru, &tmp); | |
eaf5f907 | 884 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
23044507 | 885 | spin_unlock(&dentry->d_lock); |
b0d40c92 | 886 | if (!--count) |
23044507 | 887 | break; |
3049cfe2 | 888 | } |
ec33679d | 889 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 890 | } |
da3bbdd4 KM |
891 | if (!list_empty(&referenced)) |
892 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 893 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
894 | |
895 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
896 | } |
897 | ||
1da177e4 LT |
898 | /** |
899 | * shrink_dcache_sb - shrink dcache for a superblock | |
900 | * @sb: superblock | |
901 | * | |
3049cfe2 CH |
902 | * Shrink the dcache for the specified super block. This is used to free |
903 | * the dcache before unmounting a file system. | |
1da177e4 | 904 | */ |
3049cfe2 | 905 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 906 | { |
3049cfe2 CH |
907 | LIST_HEAD(tmp); |
908 | ||
23044507 | 909 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
910 | while (!list_empty(&sb->s_dentry_lru)) { |
911 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 912 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 913 | shrink_dentry_list(&tmp); |
ec33679d | 914 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 915 | } |
23044507 | 916 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 917 | } |
ec4f8605 | 918 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 919 | |
c636ebdb DH |
920 | /* |
921 | * destroy a single subtree of dentries for unmount | |
922 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
923 | * locking | |
924 | */ | |
925 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
926 | { | |
927 | struct dentry *parent; | |
928 | ||
929 | BUG_ON(!IS_ROOT(dentry)); | |
930 | ||
c636ebdb DH |
931 | for (;;) { |
932 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 933 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
934 | dentry = list_entry(dentry->d_subdirs.next, |
935 | struct dentry, d_u.d_child); | |
c636ebdb DH |
936 | |
937 | /* consume the dentries from this leaf up through its parents | |
938 | * until we find one with children or run out altogether */ | |
939 | do { | |
940 | struct inode *inode; | |
941 | ||
f0023bc6 SW |
942 | /* |
943 | * remove the dentry from the lru, and inform | |
944 | * the fs that this dentry is about to be | |
945 | * unhashed and destroyed. | |
946 | */ | |
947 | dentry_lru_prune(dentry); | |
43c1c9cd DH |
948 | __d_shrink(dentry); |
949 | ||
b7ab39f6 | 950 | if (dentry->d_count != 0) { |
c636ebdb DH |
951 | printk(KERN_ERR |
952 | "BUG: Dentry %p{i=%lx,n=%s}" | |
953 | " still in use (%d)" | |
954 | " [unmount of %s %s]\n", | |
955 | dentry, | |
956 | dentry->d_inode ? | |
957 | dentry->d_inode->i_ino : 0UL, | |
958 | dentry->d_name.name, | |
b7ab39f6 | 959 | dentry->d_count, |
c636ebdb DH |
960 | dentry->d_sb->s_type->name, |
961 | dentry->d_sb->s_id); | |
962 | BUG(); | |
963 | } | |
964 | ||
2fd6b7f5 | 965 | if (IS_ROOT(dentry)) { |
c636ebdb | 966 | parent = NULL; |
2fd6b7f5 NP |
967 | list_del(&dentry->d_u.d_child); |
968 | } else { | |
871c0067 | 969 | parent = dentry->d_parent; |
b7ab39f6 | 970 | parent->d_count--; |
2fd6b7f5 | 971 | list_del(&dentry->d_u.d_child); |
871c0067 | 972 | } |
c636ebdb | 973 | |
c636ebdb DH |
974 | inode = dentry->d_inode; |
975 | if (inode) { | |
976 | dentry->d_inode = NULL; | |
977 | list_del_init(&dentry->d_alias); | |
978 | if (dentry->d_op && dentry->d_op->d_iput) | |
979 | dentry->d_op->d_iput(dentry, inode); | |
980 | else | |
981 | iput(inode); | |
982 | } | |
983 | ||
984 | d_free(dentry); | |
985 | ||
986 | /* finished when we fall off the top of the tree, | |
987 | * otherwise we ascend to the parent and move to the | |
988 | * next sibling if there is one */ | |
989 | if (!parent) | |
312d3ca8 | 990 | return; |
c636ebdb | 991 | dentry = parent; |
c636ebdb DH |
992 | } while (list_empty(&dentry->d_subdirs)); |
993 | ||
994 | dentry = list_entry(dentry->d_subdirs.next, | |
995 | struct dentry, d_u.d_child); | |
996 | } | |
997 | } | |
998 | ||
999 | /* | |
1000 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 1001 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
1002 | * - the superblock is detached from all mountings and open files, so the |
1003 | * dentry trees will not be rearranged by the VFS | |
1004 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
1005 | * any dentries belonging to this superblock that it comes across | |
1006 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
1007 | * in this superblock | |
1008 | */ | |
1009 | void shrink_dcache_for_umount(struct super_block *sb) | |
1010 | { | |
1011 | struct dentry *dentry; | |
1012 | ||
1013 | if (down_read_trylock(&sb->s_umount)) | |
1014 | BUG(); | |
1015 | ||
1016 | dentry = sb->s_root; | |
1017 | sb->s_root = NULL; | |
b7ab39f6 | 1018 | dentry->d_count--; |
c636ebdb DH |
1019 | shrink_dcache_for_umount_subtree(dentry); |
1020 | ||
ceb5bdc2 NP |
1021 | while (!hlist_bl_empty(&sb->s_anon)) { |
1022 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
1023 | shrink_dcache_for_umount_subtree(dentry); |
1024 | } | |
1025 | } | |
1026 | ||
c826cb7d LT |
1027 | /* |
1028 | * This tries to ascend one level of parenthood, but | |
1029 | * we can race with renaming, so we need to re-check | |
1030 | * the parenthood after dropping the lock and check | |
1031 | * that the sequence number still matches. | |
1032 | */ | |
1033 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
1034 | { | |
1035 | struct dentry *new = old->d_parent; | |
1036 | ||
1037 | rcu_read_lock(); | |
1038 | spin_unlock(&old->d_lock); | |
1039 | spin_lock(&new->d_lock); | |
1040 | ||
1041 | /* | |
1042 | * might go back up the wrong parent if we have had a rename | |
1043 | * or deletion | |
1044 | */ | |
1045 | if (new != old->d_parent || | |
c83ce989 | 1046 | (old->d_flags & DCACHE_DISCONNECTED) || |
c826cb7d LT |
1047 | (!locked && read_seqretry(&rename_lock, seq))) { |
1048 | spin_unlock(&new->d_lock); | |
1049 | new = NULL; | |
1050 | } | |
1051 | rcu_read_unlock(); | |
1052 | return new; | |
1053 | } | |
1054 | ||
1055 | ||
1da177e4 LT |
1056 | /* |
1057 | * Search for at least 1 mount point in the dentry's subdirs. | |
1058 | * We descend to the next level whenever the d_subdirs | |
1059 | * list is non-empty and continue searching. | |
1060 | */ | |
1061 | ||
1062 | /** | |
1063 | * have_submounts - check for mounts over a dentry | |
1064 | * @parent: dentry to check. | |
1065 | * | |
1066 | * Return true if the parent or its subdirectories contain | |
1067 | * a mount point | |
1068 | */ | |
1da177e4 LT |
1069 | int have_submounts(struct dentry *parent) |
1070 | { | |
949854d0 | 1071 | struct dentry *this_parent; |
1da177e4 | 1072 | struct list_head *next; |
949854d0 | 1073 | unsigned seq; |
58db63d0 | 1074 | int locked = 0; |
949854d0 | 1075 | |
949854d0 | 1076 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1077 | again: |
1078 | this_parent = parent; | |
1da177e4 | 1079 | |
1da177e4 LT |
1080 | if (d_mountpoint(parent)) |
1081 | goto positive; | |
2fd6b7f5 | 1082 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1083 | repeat: |
1084 | next = this_parent->d_subdirs.next; | |
1085 | resume: | |
1086 | while (next != &this_parent->d_subdirs) { | |
1087 | struct list_head *tmp = next; | |
5160ee6f | 1088 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1089 | next = tmp->next; |
2fd6b7f5 NP |
1090 | |
1091 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1092 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1093 | if (d_mountpoint(dentry)) { |
1094 | spin_unlock(&dentry->d_lock); | |
1095 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1096 | goto positive; |
2fd6b7f5 | 1097 | } |
1da177e4 | 1098 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1099 | spin_unlock(&this_parent->d_lock); |
1100 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1101 | this_parent = dentry; |
2fd6b7f5 | 1102 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1103 | goto repeat; |
1104 | } | |
2fd6b7f5 | 1105 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1106 | } |
1107 | /* | |
1108 | * All done at this level ... ascend and resume the search. | |
1109 | */ | |
1110 | if (this_parent != parent) { | |
c826cb7d LT |
1111 | struct dentry *child = this_parent; |
1112 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1113 | if (!this_parent) | |
949854d0 | 1114 | goto rename_retry; |
949854d0 | 1115 | next = child->d_u.d_child.next; |
1da177e4 LT |
1116 | goto resume; |
1117 | } | |
2fd6b7f5 | 1118 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1119 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1120 | goto rename_retry; |
58db63d0 NP |
1121 | if (locked) |
1122 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1123 | return 0; /* No mount points found in tree */ |
1124 | positive: | |
58db63d0 | 1125 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1126 | goto rename_retry; |
58db63d0 NP |
1127 | if (locked) |
1128 | write_sequnlock(&rename_lock); | |
1da177e4 | 1129 | return 1; |
58db63d0 NP |
1130 | |
1131 | rename_retry: | |
1132 | locked = 1; | |
1133 | write_seqlock(&rename_lock); | |
1134 | goto again; | |
1da177e4 | 1135 | } |
ec4f8605 | 1136 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1137 | |
1138 | /* | |
1139 | * Search the dentry child list for the specified parent, | |
1140 | * and move any unused dentries to the end of the unused | |
1141 | * list for prune_dcache(). We descend to the next level | |
1142 | * whenever the d_subdirs list is non-empty and continue | |
1143 | * searching. | |
1144 | * | |
1145 | * It returns zero iff there are no unused children, | |
1146 | * otherwise it returns the number of children moved to | |
1147 | * the end of the unused list. This may not be the total | |
1148 | * number of unused children, because select_parent can | |
1149 | * drop the lock and return early due to latency | |
1150 | * constraints. | |
1151 | */ | |
b48f03b3 | 1152 | static int select_parent(struct dentry *parent, struct list_head *dispose) |
1da177e4 | 1153 | { |
949854d0 | 1154 | struct dentry *this_parent; |
1da177e4 | 1155 | struct list_head *next; |
949854d0 | 1156 | unsigned seq; |
1da177e4 | 1157 | int found = 0; |
58db63d0 | 1158 | int locked = 0; |
1da177e4 | 1159 | |
949854d0 | 1160 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1161 | again: |
1162 | this_parent = parent; | |
2fd6b7f5 | 1163 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1164 | repeat: |
1165 | next = this_parent->d_subdirs.next; | |
1166 | resume: | |
1167 | while (next != &this_parent->d_subdirs) { | |
1168 | struct list_head *tmp = next; | |
5160ee6f | 1169 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1170 | next = tmp->next; |
1171 | ||
2fd6b7f5 | 1172 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1173 | |
b48f03b3 DC |
1174 | /* |
1175 | * move only zero ref count dentries to the dispose list. | |
eaf5f907 MS |
1176 | * |
1177 | * Those which are presently on the shrink list, being processed | |
1178 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1179 | * loop in shrink_dcache_parent() might not make any progress | |
1180 | * and loop forever. | |
1da177e4 | 1181 | */ |
eaf5f907 MS |
1182 | if (dentry->d_count) { |
1183 | dentry_lru_del(dentry); | |
1184 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
b48f03b3 | 1185 | dentry_lru_move_list(dentry, dispose); |
eaf5f907 | 1186 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
1da177e4 LT |
1187 | found++; |
1188 | } | |
1da177e4 LT |
1189 | /* |
1190 | * We can return to the caller if we have found some (this | |
1191 | * ensures forward progress). We'll be coming back to find | |
1192 | * the rest. | |
1193 | */ | |
2fd6b7f5 NP |
1194 | if (found && need_resched()) { |
1195 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1196 | goto out; |
2fd6b7f5 | 1197 | } |
1da177e4 LT |
1198 | |
1199 | /* | |
1200 | * Descend a level if the d_subdirs list is non-empty. | |
1201 | */ | |
1202 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1203 | spin_unlock(&this_parent->d_lock); |
1204 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1205 | this_parent = dentry; |
2fd6b7f5 | 1206 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1207 | goto repeat; |
1208 | } | |
2fd6b7f5 NP |
1209 | |
1210 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1211 | } |
1212 | /* | |
1213 | * All done at this level ... ascend and resume the search. | |
1214 | */ | |
1215 | if (this_parent != parent) { | |
c826cb7d LT |
1216 | struct dentry *child = this_parent; |
1217 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1218 | if (!this_parent) | |
949854d0 | 1219 | goto rename_retry; |
949854d0 | 1220 | next = child->d_u.d_child.next; |
1da177e4 LT |
1221 | goto resume; |
1222 | } | |
1223 | out: | |
2fd6b7f5 | 1224 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1225 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1226 | goto rename_retry; |
58db63d0 NP |
1227 | if (locked) |
1228 | write_sequnlock(&rename_lock); | |
1da177e4 | 1229 | return found; |
58db63d0 NP |
1230 | |
1231 | rename_retry: | |
1232 | if (found) | |
1233 | return found; | |
1234 | locked = 1; | |
1235 | write_seqlock(&rename_lock); | |
1236 | goto again; | |
1da177e4 LT |
1237 | } |
1238 | ||
1239 | /** | |
1240 | * shrink_dcache_parent - prune dcache | |
1241 | * @parent: parent of entries to prune | |
1242 | * | |
1243 | * Prune the dcache to remove unused children of the parent dentry. | |
1244 | */ | |
1da177e4 LT |
1245 | void shrink_dcache_parent(struct dentry * parent) |
1246 | { | |
b48f03b3 | 1247 | LIST_HEAD(dispose); |
1da177e4 LT |
1248 | int found; |
1249 | ||
b48f03b3 DC |
1250 | while ((found = select_parent(parent, &dispose)) != 0) |
1251 | shrink_dentry_list(&dispose); | |
1da177e4 | 1252 | } |
ec4f8605 | 1253 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1254 | |
1da177e4 | 1255 | /** |
a4464dbc AV |
1256 | * __d_alloc - allocate a dcache entry |
1257 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1258 | * @name: qstr of the name |
1259 | * | |
1260 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1261 | * available. On a success the dentry is returned. The name passed in is | |
1262 | * copied and the copy passed in may be reused after this call. | |
1263 | */ | |
1264 | ||
a4464dbc | 1265 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1266 | { |
1267 | struct dentry *dentry; | |
1268 | char *dname; | |
1269 | ||
e12ba74d | 1270 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1271 | if (!dentry) |
1272 | return NULL; | |
1273 | ||
6326c71f LT |
1274 | /* |
1275 | * We guarantee that the inline name is always NUL-terminated. | |
1276 | * This way the memcpy() done by the name switching in rename | |
1277 | * will still always have a NUL at the end, even if we might | |
1278 | * be overwriting an internal NUL character | |
1279 | */ | |
1280 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1281 | if (name->len > DNAME_INLINE_LEN-1) { |
1282 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1283 | if (!dname) { | |
1284 | kmem_cache_free(dentry_cache, dentry); | |
1285 | return NULL; | |
1286 | } | |
1287 | } else { | |
1288 | dname = dentry->d_iname; | |
1289 | } | |
1da177e4 LT |
1290 | |
1291 | dentry->d_name.len = name->len; | |
1292 | dentry->d_name.hash = name->hash; | |
1293 | memcpy(dname, name->name, name->len); | |
1294 | dname[name->len] = 0; | |
1295 | ||
6326c71f LT |
1296 | /* Make sure we always see the terminating NUL character */ |
1297 | smp_wmb(); | |
1298 | dentry->d_name.name = dname; | |
1299 | ||
b7ab39f6 | 1300 | dentry->d_count = 1; |
dea3667b | 1301 | dentry->d_flags = 0; |
1da177e4 | 1302 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1303 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1304 | dentry->d_inode = NULL; |
a4464dbc AV |
1305 | dentry->d_parent = dentry; |
1306 | dentry->d_sb = sb; | |
1da177e4 LT |
1307 | dentry->d_op = NULL; |
1308 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1309 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1310 | INIT_LIST_HEAD(&dentry->d_lru); |
1311 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
1312 | INIT_LIST_HEAD(&dentry->d_alias); | |
2fd6b7f5 | 1313 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1314 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1315 | |
3e880fb5 | 1316 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1317 | |
1da177e4 LT |
1318 | return dentry; |
1319 | } | |
a4464dbc AV |
1320 | |
1321 | /** | |
1322 | * d_alloc - allocate a dcache entry | |
1323 | * @parent: parent of entry to allocate | |
1324 | * @name: qstr of the name | |
1325 | * | |
1326 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1327 | * available. On a success the dentry is returned. The name passed in is | |
1328 | * copied and the copy passed in may be reused after this call. | |
1329 | */ | |
1330 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1331 | { | |
1332 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1333 | if (!dentry) | |
1334 | return NULL; | |
1335 | ||
1336 | spin_lock(&parent->d_lock); | |
1337 | /* | |
1338 | * don't need child lock because it is not subject | |
1339 | * to concurrency here | |
1340 | */ | |
1341 | __dget_dlock(parent); | |
1342 | dentry->d_parent = parent; | |
1343 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1344 | spin_unlock(&parent->d_lock); | |
1345 | ||
1346 | return dentry; | |
1347 | } | |
ec4f8605 | 1348 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1349 | |
4b936885 NP |
1350 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1351 | { | |
a4464dbc AV |
1352 | struct dentry *dentry = __d_alloc(sb, name); |
1353 | if (dentry) | |
4b936885 | 1354 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1355 | return dentry; |
1356 | } | |
1357 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1358 | ||
1da177e4 LT |
1359 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1360 | { | |
1361 | struct qstr q; | |
1362 | ||
1363 | q.name = name; | |
1364 | q.len = strlen(name); | |
1365 | q.hash = full_name_hash(q.name, q.len); | |
1366 | return d_alloc(parent, &q); | |
1367 | } | |
ef26ca97 | 1368 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1369 | |
fb045adb NP |
1370 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1371 | { | |
6f7f7caa LT |
1372 | WARN_ON_ONCE(dentry->d_op); |
1373 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1374 | DCACHE_OP_COMPARE | |
1375 | DCACHE_OP_REVALIDATE | | |
1376 | DCACHE_OP_DELETE )); | |
1377 | dentry->d_op = op; | |
1378 | if (!op) | |
1379 | return; | |
1380 | if (op->d_hash) | |
1381 | dentry->d_flags |= DCACHE_OP_HASH; | |
1382 | if (op->d_compare) | |
1383 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1384 | if (op->d_revalidate) | |
1385 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
1386 | if (op->d_delete) | |
1387 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1388 | if (op->d_prune) |
1389 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1390 | |
1391 | } | |
1392 | EXPORT_SYMBOL(d_set_d_op); | |
1393 | ||
360da900 OH |
1394 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1395 | { | |
b23fb0a6 | 1396 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1397 | if (inode) { |
1398 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1399 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
360da900 | 1400 | list_add(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1401 | } |
360da900 | 1402 | dentry->d_inode = inode; |
31e6b01f | 1403 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1404 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1405 | fsnotify_d_instantiate(dentry, inode); |
1406 | } | |
1407 | ||
1da177e4 LT |
1408 | /** |
1409 | * d_instantiate - fill in inode information for a dentry | |
1410 | * @entry: dentry to complete | |
1411 | * @inode: inode to attach to this dentry | |
1412 | * | |
1413 | * Fill in inode information in the entry. | |
1414 | * | |
1415 | * This turns negative dentries into productive full members | |
1416 | * of society. | |
1417 | * | |
1418 | * NOTE! This assumes that the inode count has been incremented | |
1419 | * (or otherwise set) by the caller to indicate that it is now | |
1420 | * in use by the dcache. | |
1421 | */ | |
1422 | ||
1423 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1424 | { | |
28133c7b | 1425 | BUG_ON(!list_empty(&entry->d_alias)); |
873feea0 NP |
1426 | if (inode) |
1427 | spin_lock(&inode->i_lock); | |
360da900 | 1428 | __d_instantiate(entry, inode); |
873feea0 NP |
1429 | if (inode) |
1430 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1431 | security_d_instantiate(entry, inode); |
1432 | } | |
ec4f8605 | 1433 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1434 | |
1435 | /** | |
1436 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1437 | * @entry: dentry to instantiate | |
1438 | * @inode: inode to attach to this dentry | |
1439 | * | |
1440 | * Fill in inode information in the entry. On success, it returns NULL. | |
1441 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1442 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1443 | * |
1444 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1445 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1446 | * |
1447 | * This also assumes that the inode count has been incremented | |
1448 | * (or otherwise set) by the caller to indicate that it is now | |
1449 | * in use by the dcache. | |
1da177e4 | 1450 | */ |
770bfad8 DH |
1451 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1452 | struct inode *inode) | |
1da177e4 LT |
1453 | { |
1454 | struct dentry *alias; | |
1455 | int len = entry->d_name.len; | |
1456 | const char *name = entry->d_name.name; | |
1457 | unsigned int hash = entry->d_name.hash; | |
1458 | ||
770bfad8 | 1459 | if (!inode) { |
360da900 | 1460 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1461 | return NULL; |
1462 | } | |
1463 | ||
1da177e4 | 1464 | list_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1465 | /* |
1466 | * Don't need alias->d_lock here, because aliases with | |
1467 | * d_parent == entry->d_parent are not subject to name or | |
1468 | * parent changes, because the parent inode i_mutex is held. | |
1469 | */ | |
12f8ad4b | 1470 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1471 | continue; |
1472 | if (alias->d_parent != entry->d_parent) | |
1473 | continue; | |
ee983e89 LT |
1474 | if (alias->d_name.len != len) |
1475 | continue; | |
12f8ad4b | 1476 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1477 | continue; |
dc0474be | 1478 | __dget(alias); |
1da177e4 LT |
1479 | return alias; |
1480 | } | |
770bfad8 | 1481 | |
360da900 | 1482 | __d_instantiate(entry, inode); |
1da177e4 LT |
1483 | return NULL; |
1484 | } | |
770bfad8 DH |
1485 | |
1486 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1487 | { | |
1488 | struct dentry *result; | |
1489 | ||
1490 | BUG_ON(!list_empty(&entry->d_alias)); | |
1491 | ||
873feea0 NP |
1492 | if (inode) |
1493 | spin_lock(&inode->i_lock); | |
770bfad8 | 1494 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1495 | if (inode) |
1496 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1497 | |
1498 | if (!result) { | |
1499 | security_d_instantiate(entry, inode); | |
1500 | return NULL; | |
1501 | } | |
1502 | ||
1503 | BUG_ON(!d_unhashed(result)); | |
1504 | iput(inode); | |
1505 | return result; | |
1506 | } | |
1507 | ||
1da177e4 LT |
1508 | EXPORT_SYMBOL(d_instantiate_unique); |
1509 | ||
adc0e91a AV |
1510 | struct dentry *d_make_root(struct inode *root_inode) |
1511 | { | |
1512 | struct dentry *res = NULL; | |
1513 | ||
1514 | if (root_inode) { | |
26fe5750 | 1515 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1516 | |
1517 | res = __d_alloc(root_inode->i_sb, &name); | |
1518 | if (res) | |
1519 | d_instantiate(res, root_inode); | |
1520 | else | |
1521 | iput(root_inode); | |
1522 | } | |
1523 | return res; | |
1524 | } | |
1525 | EXPORT_SYMBOL(d_make_root); | |
1526 | ||
d891eedb BF |
1527 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1528 | { | |
1529 | struct dentry *alias; | |
1530 | ||
1531 | if (list_empty(&inode->i_dentry)) | |
1532 | return NULL; | |
1533 | alias = list_first_entry(&inode->i_dentry, struct dentry, d_alias); | |
1534 | __dget(alias); | |
1535 | return alias; | |
1536 | } | |
1537 | ||
46f72b34 SW |
1538 | /** |
1539 | * d_find_any_alias - find any alias for a given inode | |
1540 | * @inode: inode to find an alias for | |
1541 | * | |
1542 | * If any aliases exist for the given inode, take and return a | |
1543 | * reference for one of them. If no aliases exist, return %NULL. | |
1544 | */ | |
1545 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1546 | { |
1547 | struct dentry *de; | |
1548 | ||
1549 | spin_lock(&inode->i_lock); | |
1550 | de = __d_find_any_alias(inode); | |
1551 | spin_unlock(&inode->i_lock); | |
1552 | return de; | |
1553 | } | |
46f72b34 | 1554 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1555 | |
4ea3ada2 CH |
1556 | /** |
1557 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1558 | * @inode: inode to allocate the dentry for | |
1559 | * | |
1560 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1561 | * similar open by handle operations. The returned dentry may be anonymous, | |
1562 | * or may have a full name (if the inode was already in the cache). | |
1563 | * | |
1564 | * When called on a directory inode, we must ensure that the inode only ever | |
1565 | * has one dentry. If a dentry is found, that is returned instead of | |
1566 | * allocating a new one. | |
1567 | * | |
1568 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1569 | * to the dentry. In case of an error the reference on the inode is released. |
1570 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1571 | * be passed in and will be the error will be propagate to the return value, | |
1572 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1573 | */ |
1574 | struct dentry *d_obtain_alias(struct inode *inode) | |
1575 | { | |
9308a612 CH |
1576 | static const struct qstr anonstring = { .name = "" }; |
1577 | struct dentry *tmp; | |
1578 | struct dentry *res; | |
4ea3ada2 CH |
1579 | |
1580 | if (!inode) | |
44003728 | 1581 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1582 | if (IS_ERR(inode)) |
1583 | return ERR_CAST(inode); | |
1584 | ||
d891eedb | 1585 | res = d_find_any_alias(inode); |
9308a612 CH |
1586 | if (res) |
1587 | goto out_iput; | |
1588 | ||
a4464dbc | 1589 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1590 | if (!tmp) { |
1591 | res = ERR_PTR(-ENOMEM); | |
1592 | goto out_iput; | |
4ea3ada2 | 1593 | } |
b5c84bf6 | 1594 | |
873feea0 | 1595 | spin_lock(&inode->i_lock); |
d891eedb | 1596 | res = __d_find_any_alias(inode); |
9308a612 | 1597 | if (res) { |
873feea0 | 1598 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1599 | dput(tmp); |
1600 | goto out_iput; | |
1601 | } | |
1602 | ||
1603 | /* attach a disconnected dentry */ | |
1604 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1605 | tmp->d_inode = inode; |
1606 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
9308a612 | 1607 | list_add(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1608 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1609 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1610 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1611 | spin_unlock(&tmp->d_lock); |
873feea0 | 1612 | spin_unlock(&inode->i_lock); |
24ff6663 | 1613 | security_d_instantiate(tmp, inode); |
9308a612 | 1614 | |
9308a612 CH |
1615 | return tmp; |
1616 | ||
1617 | out_iput: | |
24ff6663 JB |
1618 | if (res && !IS_ERR(res)) |
1619 | security_d_instantiate(res, inode); | |
9308a612 CH |
1620 | iput(inode); |
1621 | return res; | |
4ea3ada2 | 1622 | } |
adc48720 | 1623 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1624 | |
1625 | /** | |
1626 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1627 | * @inode: the inode which may have a disconnected dentry | |
1628 | * @dentry: a negative dentry which we want to point to the inode. | |
1629 | * | |
1630 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1631 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1632 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1633 | * | |
1634 | * This is needed in the lookup routine of any filesystem that is exportable | |
1635 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1636 | * | |
1637 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1638 | * is returned. This matches the expected return value of ->lookup. | |
1639 | * | |
1640 | */ | |
1641 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1642 | { | |
1643 | struct dentry *new = NULL; | |
1644 | ||
a9049376 AV |
1645 | if (IS_ERR(inode)) |
1646 | return ERR_CAST(inode); | |
1647 | ||
21c0d8fd | 1648 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1649 | spin_lock(&inode->i_lock); |
7732a557 | 1650 | new = __d_find_any_alias(inode); |
1da177e4 | 1651 | if (new) { |
873feea0 | 1652 | spin_unlock(&inode->i_lock); |
1da177e4 | 1653 | security_d_instantiate(new, inode); |
1da177e4 LT |
1654 | d_move(new, dentry); |
1655 | iput(inode); | |
1656 | } else { | |
873feea0 | 1657 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1658 | __d_instantiate(dentry, inode); |
873feea0 | 1659 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1660 | security_d_instantiate(dentry, inode); |
1661 | d_rehash(dentry); | |
1662 | } | |
1663 | } else | |
1664 | d_add(dentry, inode); | |
1665 | return new; | |
1666 | } | |
ec4f8605 | 1667 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1668 | |
9403540c BN |
1669 | /** |
1670 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1671 | * @inode: the inode case-insensitive lookup has found | |
1672 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1673 | * @name: the case-exact name to be associated with the returned dentry | |
1674 | * | |
1675 | * This is to avoid filling the dcache with case-insensitive names to the | |
1676 | * same inode, only the actual correct case is stored in the dcache for | |
1677 | * case-insensitive filesystems. | |
1678 | * | |
1679 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1680 | * already exists in in the dcache, use it and return it. | |
1681 | * | |
1682 | * If no entry exists with the exact case name, allocate new dentry with | |
1683 | * the exact case, and return the spliced entry. | |
1684 | */ | |
e45b590b | 1685 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1686 | struct qstr *name) |
1687 | { | |
1688 | int error; | |
1689 | struct dentry *found; | |
1690 | struct dentry *new; | |
1691 | ||
b6520c81 CH |
1692 | /* |
1693 | * First check if a dentry matching the name already exists, | |
1694 | * if not go ahead and create it now. | |
1695 | */ | |
9403540c | 1696 | found = d_hash_and_lookup(dentry->d_parent, name); |
9403540c BN |
1697 | if (!found) { |
1698 | new = d_alloc(dentry->d_parent, name); | |
1699 | if (!new) { | |
1700 | error = -ENOMEM; | |
1701 | goto err_out; | |
1702 | } | |
b6520c81 | 1703 | |
9403540c BN |
1704 | found = d_splice_alias(inode, new); |
1705 | if (found) { | |
1706 | dput(new); | |
1707 | return found; | |
1708 | } | |
1709 | return new; | |
1710 | } | |
b6520c81 CH |
1711 | |
1712 | /* | |
1713 | * If a matching dentry exists, and it's not negative use it. | |
1714 | * | |
1715 | * Decrement the reference count to balance the iget() done | |
1716 | * earlier on. | |
1717 | */ | |
9403540c BN |
1718 | if (found->d_inode) { |
1719 | if (unlikely(found->d_inode != inode)) { | |
1720 | /* This can't happen because bad inodes are unhashed. */ | |
1721 | BUG_ON(!is_bad_inode(inode)); | |
1722 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1723 | } | |
9403540c BN |
1724 | iput(inode); |
1725 | return found; | |
1726 | } | |
b6520c81 | 1727 | |
9403540c | 1728 | /* |
44396f4b JB |
1729 | * We are going to instantiate this dentry, unhash it and clear the |
1730 | * lookup flag so we can do that. | |
9403540c | 1731 | */ |
44396f4b JB |
1732 | if (unlikely(d_need_lookup(found))) |
1733 | d_clear_need_lookup(found); | |
b6520c81 | 1734 | |
9403540c | 1735 | /* |
9403540c | 1736 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1737 | * already has a dentry. |
9403540c | 1738 | */ |
4513d899 AV |
1739 | new = d_splice_alias(inode, found); |
1740 | if (new) { | |
1741 | dput(found); | |
1742 | found = new; | |
9403540c | 1743 | } |
4513d899 | 1744 | return found; |
9403540c BN |
1745 | |
1746 | err_out: | |
1747 | iput(inode); | |
1748 | return ERR_PTR(error); | |
1749 | } | |
ec4f8605 | 1750 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1751 | |
12f8ad4b LT |
1752 | /* |
1753 | * Do the slow-case of the dentry name compare. | |
1754 | * | |
1755 | * Unlike the dentry_cmp() function, we need to atomically | |
1756 | * load the name, length and inode information, so that the | |
1757 | * filesystem can rely on them, and can use the 'name' and | |
1758 | * 'len' information without worrying about walking off the | |
1759 | * end of memory etc. | |
1760 | * | |
1761 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1762 | * in arguments (the low-level filesystem should not look | |
1763 | * at the dentry inode or name contents directly, since | |
1764 | * rename can change them while we're in RCU mode). | |
1765 | */ | |
1766 | enum slow_d_compare { | |
1767 | D_COMP_OK, | |
1768 | D_COMP_NOMATCH, | |
1769 | D_COMP_SEQRETRY, | |
1770 | }; | |
1771 | ||
1772 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1773 | const struct dentry *parent, | |
1774 | struct inode *inode, | |
1775 | struct dentry *dentry, | |
1776 | unsigned int seq, | |
1777 | const struct qstr *name) | |
1778 | { | |
1779 | int tlen = dentry->d_name.len; | |
1780 | const char *tname = dentry->d_name.name; | |
1781 | struct inode *i = dentry->d_inode; | |
1782 | ||
1783 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1784 | cpu_relax(); | |
1785 | return D_COMP_SEQRETRY; | |
1786 | } | |
1787 | if (parent->d_op->d_compare(parent, inode, | |
1788 | dentry, i, | |
1789 | tlen, tname, name)) | |
1790 | return D_COMP_NOMATCH; | |
1791 | return D_COMP_OK; | |
1792 | } | |
1793 | ||
31e6b01f NP |
1794 | /** |
1795 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1796 | * @parent: parent dentry | |
1797 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1798 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1799 | * @inode: returns dentry->d_inode when the inode was found valid. |
1800 | * Returns: dentry, or NULL | |
1801 | * | |
1802 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1803 | * resolution (store-free path walking) design described in | |
1804 | * Documentation/filesystems/path-lookup.txt. | |
1805 | * | |
1806 | * This is not to be used outside core vfs. | |
1807 | * | |
1808 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1809 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1810 | * without taking d_lock and checking d_seq sequence count against @seq | |
1811 | * returned here. | |
1812 | * | |
1813 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1814 | * function. | |
1815 | * | |
1816 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1817 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1818 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1819 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
1820 | * |
1821 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
1822 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 1823 | */ |
8966be90 LT |
1824 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
1825 | const struct qstr *name, | |
12f8ad4b | 1826 | unsigned *seqp, struct inode *inode) |
31e6b01f | 1827 | { |
26fe5750 | 1828 | u64 hashlen = name->hash_len; |
31e6b01f | 1829 | const unsigned char *str = name->name; |
26fe5750 | 1830 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 1831 | struct hlist_bl_node *node; |
31e6b01f NP |
1832 | struct dentry *dentry; |
1833 | ||
1834 | /* | |
1835 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1836 | * required to prevent single threaded performance regressions | |
1837 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1838 | * Keep the two functions in sync. | |
1839 | */ | |
1840 | ||
1841 | /* | |
1842 | * The hash list is protected using RCU. | |
1843 | * | |
1844 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1845 | * races with d_move(). | |
1846 | * | |
1847 | * It is possible that concurrent renames can mess up our list | |
1848 | * walk here and result in missing our dentry, resulting in the | |
1849 | * false-negative result. d_lookup() protects against concurrent | |
1850 | * renames using rename_lock seqlock. | |
1851 | * | |
b0a4bb83 | 1852 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1853 | */ |
b07ad996 | 1854 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 1855 | unsigned seq; |
31e6b01f | 1856 | |
31e6b01f | 1857 | seqretry: |
12f8ad4b LT |
1858 | /* |
1859 | * The dentry sequence count protects us from concurrent | |
1860 | * renames, and thus protects inode, parent and name fields. | |
1861 | * | |
1862 | * The caller must perform a seqcount check in order | |
1863 | * to do anything useful with the returned dentry, | |
1864 | * including using the 'd_inode' pointer. | |
1865 | * | |
1866 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
1867 | * we don't wait for the sequence count to stabilize if it | |
1868 | * is in the middle of a sequence change. If we do the slow | |
1869 | * dentry compare, we will do seqretries until it is stable, | |
1870 | * and if we end up with a successful lookup, we actually | |
1871 | * want to exit RCU lookup anyway. | |
1872 | */ | |
1873 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
1874 | if (dentry->d_parent != parent) |
1875 | continue; | |
2e321806 LT |
1876 | if (d_unhashed(dentry)) |
1877 | continue; | |
12f8ad4b LT |
1878 | *seqp = seq; |
1879 | ||
830c0f0e | 1880 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
1881 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
1882 | continue; | |
12f8ad4b LT |
1883 | switch (slow_dentry_cmp(parent, inode, dentry, seq, name)) { |
1884 | case D_COMP_OK: | |
1885 | return dentry; | |
1886 | case D_COMP_NOMATCH: | |
31e6b01f | 1887 | continue; |
12f8ad4b LT |
1888 | default: |
1889 | goto seqretry; | |
1890 | } | |
31e6b01f | 1891 | } |
12f8ad4b | 1892 | |
26fe5750 | 1893 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 1894 | continue; |
26fe5750 | 1895 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 1896 | return dentry; |
31e6b01f NP |
1897 | } |
1898 | return NULL; | |
1899 | } | |
1900 | ||
1da177e4 LT |
1901 | /** |
1902 | * d_lookup - search for a dentry | |
1903 | * @parent: parent dentry | |
1904 | * @name: qstr of name we wish to find | |
b04f784e | 1905 | * Returns: dentry, or NULL |
1da177e4 | 1906 | * |
b04f784e NP |
1907 | * d_lookup searches the children of the parent dentry for the name in |
1908 | * question. If the dentry is found its reference count is incremented and the | |
1909 | * dentry is returned. The caller must use dput to free the entry when it has | |
1910 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1911 | */ |
31e6b01f | 1912 | struct dentry *d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 | 1913 | { |
31e6b01f | 1914 | struct dentry *dentry; |
949854d0 | 1915 | unsigned seq; |
1da177e4 LT |
1916 | |
1917 | do { | |
1918 | seq = read_seqbegin(&rename_lock); | |
1919 | dentry = __d_lookup(parent, name); | |
1920 | if (dentry) | |
1921 | break; | |
1922 | } while (read_seqretry(&rename_lock, seq)); | |
1923 | return dentry; | |
1924 | } | |
ec4f8605 | 1925 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1926 | |
31e6b01f | 1927 | /** |
b04f784e NP |
1928 | * __d_lookup - search for a dentry (racy) |
1929 | * @parent: parent dentry | |
1930 | * @name: qstr of name we wish to find | |
1931 | * Returns: dentry, or NULL | |
1932 | * | |
1933 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1934 | * false-negative result due to unrelated rename activity. | |
1935 | * | |
1936 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1937 | * however it must be used carefully, eg. with a following d_lookup in | |
1938 | * the case of failure. | |
1939 | * | |
1940 | * __d_lookup callers must be commented. | |
1941 | */ | |
31e6b01f | 1942 | struct dentry *__d_lookup(struct dentry *parent, struct qstr *name) |
1da177e4 LT |
1943 | { |
1944 | unsigned int len = name->len; | |
1945 | unsigned int hash = name->hash; | |
1946 | const unsigned char *str = name->name; | |
b07ad996 | 1947 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1948 | struct hlist_bl_node *node; |
31e6b01f | 1949 | struct dentry *found = NULL; |
665a7583 | 1950 | struct dentry *dentry; |
1da177e4 | 1951 | |
31e6b01f NP |
1952 | /* |
1953 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1954 | * required to prevent single threaded performance regressions | |
1955 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1956 | * Keep the two functions in sync. | |
1957 | */ | |
1958 | ||
b04f784e NP |
1959 | /* |
1960 | * The hash list is protected using RCU. | |
1961 | * | |
1962 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1963 | * with d_move(). | |
1964 | * | |
1965 | * It is possible that concurrent renames can mess up our list | |
1966 | * walk here and result in missing our dentry, resulting in the | |
1967 | * false-negative result. d_lookup() protects against concurrent | |
1968 | * renames using rename_lock seqlock. | |
1969 | * | |
b0a4bb83 | 1970 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1971 | */ |
1da177e4 LT |
1972 | rcu_read_lock(); |
1973 | ||
b07ad996 | 1974 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 1975 | |
1da177e4 LT |
1976 | if (dentry->d_name.hash != hash) |
1977 | continue; | |
1da177e4 LT |
1978 | |
1979 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1980 | if (dentry->d_parent != parent) |
1981 | goto next; | |
d0185c08 LT |
1982 | if (d_unhashed(dentry)) |
1983 | goto next; | |
1984 | ||
1da177e4 LT |
1985 | /* |
1986 | * It is safe to compare names since d_move() cannot | |
1987 | * change the qstr (protected by d_lock). | |
1988 | */ | |
fb045adb | 1989 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
1990 | int tlen = dentry->d_name.len; |
1991 | const char *tname = dentry->d_name.name; | |
621e155a NP |
1992 | if (parent->d_op->d_compare(parent, parent->d_inode, |
1993 | dentry, dentry->d_inode, | |
31e6b01f | 1994 | tlen, tname, name)) |
1da177e4 LT |
1995 | goto next; |
1996 | } else { | |
ee983e89 LT |
1997 | if (dentry->d_name.len != len) |
1998 | goto next; | |
12f8ad4b | 1999 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2000 | goto next; |
2001 | } | |
2002 | ||
b7ab39f6 | 2003 | dentry->d_count++; |
d0185c08 | 2004 | found = dentry; |
1da177e4 LT |
2005 | spin_unlock(&dentry->d_lock); |
2006 | break; | |
2007 | next: | |
2008 | spin_unlock(&dentry->d_lock); | |
2009 | } | |
2010 | rcu_read_unlock(); | |
2011 | ||
2012 | return found; | |
2013 | } | |
2014 | ||
3e7e241f EB |
2015 | /** |
2016 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2017 | * @dir: Directory to search in | |
2018 | * @name: qstr of name we wish to find | |
2019 | * | |
2020 | * On hash failure or on lookup failure NULL is returned. | |
2021 | */ | |
2022 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2023 | { | |
2024 | struct dentry *dentry = NULL; | |
2025 | ||
2026 | /* | |
2027 | * Check for a fs-specific hash function. Note that we must | |
2028 | * calculate the standard hash first, as the d_op->d_hash() | |
2029 | * routine may choose to leave the hash value unchanged. | |
2030 | */ | |
2031 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2032 | if (dir->d_flags & DCACHE_OP_HASH) { |
b1e6a015 | 2033 | if (dir->d_op->d_hash(dir, dir->d_inode, name) < 0) |
3e7e241f EB |
2034 | goto out; |
2035 | } | |
2036 | dentry = d_lookup(dir, name); | |
2037 | out: | |
2038 | return dentry; | |
2039 | } | |
2040 | ||
1da177e4 | 2041 | /** |
786a5e15 | 2042 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2043 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2044 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2045 | * |
2046 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2047 | * This is used by ncpfs in its readdir implementation. | |
2048 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2049 | * |
2050 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2051 | */ |
d3a23e16 | 2052 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2053 | { |
786a5e15 | 2054 | struct dentry *child; |
d3a23e16 | 2055 | |
2fd6b7f5 | 2056 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2057 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2058 | if (dentry == child) { | |
2fd6b7f5 | 2059 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2060 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2061 | spin_unlock(&dentry->d_lock); |
2062 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2063 | return 1; |
2064 | } | |
2065 | } | |
2fd6b7f5 | 2066 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2067 | |
1da177e4 LT |
2068 | return 0; |
2069 | } | |
ec4f8605 | 2070 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2071 | |
2072 | /* | |
2073 | * When a file is deleted, we have two options: | |
2074 | * - turn this dentry into a negative dentry | |
2075 | * - unhash this dentry and free it. | |
2076 | * | |
2077 | * Usually, we want to just turn this into | |
2078 | * a negative dentry, but if anybody else is | |
2079 | * currently using the dentry or the inode | |
2080 | * we can't do that and we fall back on removing | |
2081 | * it from the hash queues and waiting for | |
2082 | * it to be deleted later when it has no users | |
2083 | */ | |
2084 | ||
2085 | /** | |
2086 | * d_delete - delete a dentry | |
2087 | * @dentry: The dentry to delete | |
2088 | * | |
2089 | * Turn the dentry into a negative dentry if possible, otherwise | |
2090 | * remove it from the hash queues so it can be deleted later | |
2091 | */ | |
2092 | ||
2093 | void d_delete(struct dentry * dentry) | |
2094 | { | |
873feea0 | 2095 | struct inode *inode; |
7a91bf7f | 2096 | int isdir = 0; |
1da177e4 LT |
2097 | /* |
2098 | * Are we the only user? | |
2099 | */ | |
357f8e65 | 2100 | again: |
1da177e4 | 2101 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2102 | inode = dentry->d_inode; |
2103 | isdir = S_ISDIR(inode->i_mode); | |
b7ab39f6 | 2104 | if (dentry->d_count == 1) { |
873feea0 | 2105 | if (inode && !spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2106 | spin_unlock(&dentry->d_lock); |
2107 | cpu_relax(); | |
2108 | goto again; | |
2109 | } | |
13e3c5e5 | 2110 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2111 | dentry_unlink_inode(dentry); |
7a91bf7f | 2112 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2113 | return; |
2114 | } | |
2115 | ||
2116 | if (!d_unhashed(dentry)) | |
2117 | __d_drop(dentry); | |
2118 | ||
2119 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2120 | |
2121 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2122 | } |
ec4f8605 | 2123 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2124 | |
b07ad996 | 2125 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2126 | { |
ceb5bdc2 | 2127 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2128 | hlist_bl_lock(b); |
dea3667b | 2129 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2130 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2131 | hlist_bl_unlock(b); |
1da177e4 LT |
2132 | } |
2133 | ||
770bfad8 DH |
2134 | static void _d_rehash(struct dentry * entry) |
2135 | { | |
2136 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2137 | } | |
2138 | ||
1da177e4 LT |
2139 | /** |
2140 | * d_rehash - add an entry back to the hash | |
2141 | * @entry: dentry to add to the hash | |
2142 | * | |
2143 | * Adds a dentry to the hash according to its name. | |
2144 | */ | |
2145 | ||
2146 | void d_rehash(struct dentry * entry) | |
2147 | { | |
1da177e4 | 2148 | spin_lock(&entry->d_lock); |
770bfad8 | 2149 | _d_rehash(entry); |
1da177e4 | 2150 | spin_unlock(&entry->d_lock); |
1da177e4 | 2151 | } |
ec4f8605 | 2152 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2153 | |
fb2d5b86 NP |
2154 | /** |
2155 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2156 | * @dentry: dentry to be updated | |
2157 | * @name: new name | |
2158 | * | |
2159 | * Update a case insensitive dentry with new case of name. | |
2160 | * | |
2161 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2162 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2163 | * lengths). | |
2164 | * | |
2165 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2166 | * keep renames and concurrent inserts, and readdir(2) away). | |
2167 | */ | |
2168 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2169 | { | |
7ebfa57f | 2170 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2171 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2172 | ||
fb2d5b86 | 2173 | spin_lock(&dentry->d_lock); |
31e6b01f | 2174 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2175 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2176 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2177 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2178 | } |
2179 | EXPORT_SYMBOL(dentry_update_name_case); | |
2180 | ||
1da177e4 LT |
2181 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2182 | { | |
2183 | if (dname_external(target)) { | |
2184 | if (dname_external(dentry)) { | |
2185 | /* | |
2186 | * Both external: swap the pointers | |
2187 | */ | |
9a8d5bb4 | 2188 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2189 | } else { |
2190 | /* | |
2191 | * dentry:internal, target:external. Steal target's | |
2192 | * storage and make target internal. | |
2193 | */ | |
321bcf92 BF |
2194 | memcpy(target->d_iname, dentry->d_name.name, |
2195 | dentry->d_name.len + 1); | |
1da177e4 LT |
2196 | dentry->d_name.name = target->d_name.name; |
2197 | target->d_name.name = target->d_iname; | |
2198 | } | |
2199 | } else { | |
2200 | if (dname_external(dentry)) { | |
2201 | /* | |
2202 | * dentry:external, target:internal. Give dentry's | |
2203 | * storage to target and make dentry internal | |
2204 | */ | |
2205 | memcpy(dentry->d_iname, target->d_name.name, | |
2206 | target->d_name.len + 1); | |
2207 | target->d_name.name = dentry->d_name.name; | |
2208 | dentry->d_name.name = dentry->d_iname; | |
2209 | } else { | |
2210 | /* | |
2211 | * Both are internal. Just copy target to dentry | |
2212 | */ | |
2213 | memcpy(dentry->d_iname, target->d_name.name, | |
2214 | target->d_name.len + 1); | |
dc711ca3 AV |
2215 | dentry->d_name.len = target->d_name.len; |
2216 | return; | |
1da177e4 LT |
2217 | } |
2218 | } | |
9a8d5bb4 | 2219 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2220 | } |
2221 | ||
2fd6b7f5 NP |
2222 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2223 | { | |
2224 | /* | |
2225 | * XXXX: do we really need to take target->d_lock? | |
2226 | */ | |
2227 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2228 | spin_lock(&target->d_parent->d_lock); | |
2229 | else { | |
2230 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2231 | spin_lock(&dentry->d_parent->d_lock); | |
2232 | spin_lock_nested(&target->d_parent->d_lock, | |
2233 | DENTRY_D_LOCK_NESTED); | |
2234 | } else { | |
2235 | spin_lock(&target->d_parent->d_lock); | |
2236 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2237 | DENTRY_D_LOCK_NESTED); | |
2238 | } | |
2239 | } | |
2240 | if (target < dentry) { | |
2241 | spin_lock_nested(&target->d_lock, 2); | |
2242 | spin_lock_nested(&dentry->d_lock, 3); | |
2243 | } else { | |
2244 | spin_lock_nested(&dentry->d_lock, 2); | |
2245 | spin_lock_nested(&target->d_lock, 3); | |
2246 | } | |
2247 | } | |
2248 | ||
2249 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2250 | struct dentry *target) | |
2251 | { | |
2252 | if (target->d_parent != dentry->d_parent) | |
2253 | spin_unlock(&dentry->d_parent->d_lock); | |
2254 | if (target->d_parent != target) | |
2255 | spin_unlock(&target->d_parent->d_lock); | |
2256 | } | |
2257 | ||
1da177e4 | 2258 | /* |
2fd6b7f5 NP |
2259 | * When switching names, the actual string doesn't strictly have to |
2260 | * be preserved in the target - because we're dropping the target | |
2261 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2262 | * the new name before we switch. | |
2263 | * | |
2264 | * Note that we have to be a lot more careful about getting the hash | |
2265 | * switched - we have to switch the hash value properly even if it | |
2266 | * then no longer matches the actual (corrupted) string of the target. | |
2267 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2268 | */ |
9eaef27b | 2269 | /* |
18367501 | 2270 | * __d_move - move a dentry |
1da177e4 LT |
2271 | * @dentry: entry to move |
2272 | * @target: new dentry | |
2273 | * | |
2274 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2275 | * dcache entries should not be moved in this way. Caller must hold |
2276 | * rename_lock, the i_mutex of the source and target directories, | |
2277 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2278 | */ |
18367501 | 2279 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2280 | { |
1da177e4 LT |
2281 | if (!dentry->d_inode) |
2282 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2283 | ||
2fd6b7f5 NP |
2284 | BUG_ON(d_ancestor(dentry, target)); |
2285 | BUG_ON(d_ancestor(target, dentry)); | |
2286 | ||
2fd6b7f5 | 2287 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2288 | |
31e6b01f NP |
2289 | write_seqcount_begin(&dentry->d_seq); |
2290 | write_seqcount_begin(&target->d_seq); | |
2291 | ||
ceb5bdc2 NP |
2292 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2293 | ||
2294 | /* | |
2295 | * Move the dentry to the target hash queue. Don't bother checking | |
2296 | * for the same hash queue because of how unlikely it is. | |
2297 | */ | |
2298 | __d_drop(dentry); | |
789680d1 | 2299 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2300 | |
2301 | /* Unhash the target: dput() will then get rid of it */ | |
2302 | __d_drop(target); | |
2303 | ||
5160ee6f ED |
2304 | list_del(&dentry->d_u.d_child); |
2305 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2306 | |
2307 | /* Switch the names.. */ | |
2308 | switch_names(dentry, target); | |
9a8d5bb4 | 2309 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2310 | |
2311 | /* ... and switch the parents */ | |
2312 | if (IS_ROOT(dentry)) { | |
2313 | dentry->d_parent = target->d_parent; | |
2314 | target->d_parent = target; | |
5160ee6f | 2315 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2316 | } else { |
9a8d5bb4 | 2317 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2318 | |
2319 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2320 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2321 | } |
2322 | ||
5160ee6f | 2323 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2324 | |
31e6b01f NP |
2325 | write_seqcount_end(&target->d_seq); |
2326 | write_seqcount_end(&dentry->d_seq); | |
2327 | ||
2fd6b7f5 | 2328 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2329 | spin_unlock(&target->d_lock); |
c32ccd87 | 2330 | fsnotify_d_move(dentry); |
1da177e4 | 2331 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2332 | } |
2333 | ||
2334 | /* | |
2335 | * d_move - move a dentry | |
2336 | * @dentry: entry to move | |
2337 | * @target: new dentry | |
2338 | * | |
2339 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2340 | * dcache entries should not be moved in this way. See the locking |
2341 | * requirements for __d_move. | |
18367501 AV |
2342 | */ |
2343 | void d_move(struct dentry *dentry, struct dentry *target) | |
2344 | { | |
2345 | write_seqlock(&rename_lock); | |
2346 | __d_move(dentry, target); | |
1da177e4 | 2347 | write_sequnlock(&rename_lock); |
9eaef27b | 2348 | } |
ec4f8605 | 2349 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2350 | |
e2761a11 OH |
2351 | /** |
2352 | * d_ancestor - search for an ancestor | |
2353 | * @p1: ancestor dentry | |
2354 | * @p2: child dentry | |
2355 | * | |
2356 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2357 | * an ancestor of p2, else NULL. | |
9eaef27b | 2358 | */ |
e2761a11 | 2359 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2360 | { |
2361 | struct dentry *p; | |
2362 | ||
871c0067 | 2363 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2364 | if (p->d_parent == p1) |
e2761a11 | 2365 | return p; |
9eaef27b | 2366 | } |
e2761a11 | 2367 | return NULL; |
9eaef27b TM |
2368 | } |
2369 | ||
2370 | /* | |
2371 | * This helper attempts to cope with remotely renamed directories | |
2372 | * | |
2373 | * It assumes that the caller is already holding | |
18367501 | 2374 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2375 | * |
2376 | * Note: If ever the locking in lock_rename() changes, then please | |
2377 | * remember to update this too... | |
9eaef27b | 2378 | */ |
873feea0 NP |
2379 | static struct dentry *__d_unalias(struct inode *inode, |
2380 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2381 | { |
2382 | struct mutex *m1 = NULL, *m2 = NULL; | |
2383 | struct dentry *ret; | |
2384 | ||
2385 | /* If alias and dentry share a parent, then no extra locks required */ | |
2386 | if (alias->d_parent == dentry->d_parent) | |
2387 | goto out_unalias; | |
2388 | ||
9eaef27b TM |
2389 | /* See lock_rename() */ |
2390 | ret = ERR_PTR(-EBUSY); | |
2391 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) | |
2392 | goto out_err; | |
2393 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2394 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2395 | goto out_err; | |
2396 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2397 | out_unalias: | |
18367501 | 2398 | __d_move(alias, dentry); |
9eaef27b TM |
2399 | ret = alias; |
2400 | out_err: | |
873feea0 | 2401 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2402 | if (m2) |
2403 | mutex_unlock(m2); | |
2404 | if (m1) | |
2405 | mutex_unlock(m1); | |
2406 | return ret; | |
2407 | } | |
2408 | ||
770bfad8 DH |
2409 | /* |
2410 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2411 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2412 | * returns with anon->d_lock held! |
770bfad8 DH |
2413 | */ |
2414 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2415 | { | |
2416 | struct dentry *dparent, *aparent; | |
2417 | ||
2fd6b7f5 | 2418 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2419 | |
31e6b01f NP |
2420 | write_seqcount_begin(&dentry->d_seq); |
2421 | write_seqcount_begin(&anon->d_seq); | |
2422 | ||
770bfad8 DH |
2423 | dparent = dentry->d_parent; |
2424 | aparent = anon->d_parent; | |
2425 | ||
2fd6b7f5 NP |
2426 | switch_names(dentry, anon); |
2427 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2428 | ||
770bfad8 DH |
2429 | dentry->d_parent = (aparent == anon) ? dentry : aparent; |
2430 | list_del(&dentry->d_u.d_child); | |
2431 | if (!IS_ROOT(dentry)) | |
2432 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); | |
2433 | else | |
2434 | INIT_LIST_HEAD(&dentry->d_u.d_child); | |
2435 | ||
2436 | anon->d_parent = (dparent == dentry) ? anon : dparent; | |
2437 | list_del(&anon->d_u.d_child); | |
2438 | if (!IS_ROOT(anon)) | |
2439 | list_add(&anon->d_u.d_child, &anon->d_parent->d_subdirs); | |
2440 | else | |
2441 | INIT_LIST_HEAD(&anon->d_u.d_child); | |
2442 | ||
31e6b01f NP |
2443 | write_seqcount_end(&dentry->d_seq); |
2444 | write_seqcount_end(&anon->d_seq); | |
2445 | ||
2fd6b7f5 NP |
2446 | dentry_unlock_parents_for_move(anon, dentry); |
2447 | spin_unlock(&dentry->d_lock); | |
2448 | ||
2449 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2450 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2451 | } | |
2452 | ||
2453 | /** | |
2454 | * d_materialise_unique - introduce an inode into the tree | |
2455 | * @dentry: candidate dentry | |
2456 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2457 | * | |
2458 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2459 | * root directory alias in its place if there is one. Caller must hold the |
2460 | * i_mutex of the parent directory. | |
770bfad8 DH |
2461 | */ |
2462 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2463 | { | |
9eaef27b | 2464 | struct dentry *actual; |
770bfad8 DH |
2465 | |
2466 | BUG_ON(!d_unhashed(dentry)); | |
2467 | ||
770bfad8 DH |
2468 | if (!inode) { |
2469 | actual = dentry; | |
360da900 | 2470 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2471 | d_rehash(actual); |
2472 | goto out_nolock; | |
770bfad8 DH |
2473 | } |
2474 | ||
873feea0 | 2475 | spin_lock(&inode->i_lock); |
357f8e65 | 2476 | |
9eaef27b TM |
2477 | if (S_ISDIR(inode->i_mode)) { |
2478 | struct dentry *alias; | |
2479 | ||
2480 | /* Does an aliased dentry already exist? */ | |
3f50fff4 | 2481 | alias = __d_find_alias(inode); |
9eaef27b TM |
2482 | if (alias) { |
2483 | actual = alias; | |
18367501 AV |
2484 | write_seqlock(&rename_lock); |
2485 | ||
2486 | if (d_ancestor(alias, dentry)) { | |
2487 | /* Check for loops */ | |
2488 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2489 | spin_unlock(&inode->i_lock); |
18367501 AV |
2490 | } else if (IS_ROOT(alias)) { |
2491 | /* Is this an anonymous mountpoint that we | |
2492 | * could splice into our tree? */ | |
9eaef27b | 2493 | __d_materialise_dentry(dentry, alias); |
18367501 | 2494 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2495 | __d_drop(alias); |
2496 | goto found; | |
18367501 AV |
2497 | } else { |
2498 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2499 | * aliasing. This drops inode->i_lock */ |
18367501 | 2500 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2501 | } |
18367501 | 2502 | write_sequnlock(&rename_lock); |
dd179946 DH |
2503 | if (IS_ERR(actual)) { |
2504 | if (PTR_ERR(actual) == -ELOOP) | |
2505 | pr_warn_ratelimited( | |
2506 | "VFS: Lookup of '%s' in %s %s" | |
2507 | " would have caused loop\n", | |
2508 | dentry->d_name.name, | |
2509 | inode->i_sb->s_type->name, | |
2510 | inode->i_sb->s_id); | |
9eaef27b | 2511 | dput(alias); |
dd179946 | 2512 | } |
9eaef27b TM |
2513 | goto out_nolock; |
2514 | } | |
770bfad8 DH |
2515 | } |
2516 | ||
2517 | /* Add a unique reference */ | |
2518 | actual = __d_instantiate_unique(dentry, inode); | |
2519 | if (!actual) | |
2520 | actual = dentry; | |
357f8e65 NP |
2521 | else |
2522 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2523 | |
770bfad8 DH |
2524 | spin_lock(&actual->d_lock); |
2525 | found: | |
2526 | _d_rehash(actual); | |
2527 | spin_unlock(&actual->d_lock); | |
873feea0 | 2528 | spin_unlock(&inode->i_lock); |
9eaef27b | 2529 | out_nolock: |
770bfad8 DH |
2530 | if (actual == dentry) { |
2531 | security_d_instantiate(dentry, inode); | |
2532 | return NULL; | |
2533 | } | |
2534 | ||
2535 | iput(inode); | |
2536 | return actual; | |
770bfad8 | 2537 | } |
ec4f8605 | 2538 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2539 | |
cdd16d02 | 2540 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2541 | { |
2542 | *buflen -= namelen; | |
2543 | if (*buflen < 0) | |
2544 | return -ENAMETOOLONG; | |
2545 | *buffer -= namelen; | |
2546 | memcpy(*buffer, str, namelen); | |
2547 | return 0; | |
2548 | } | |
2549 | ||
cdd16d02 MS |
2550 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2551 | { | |
2552 | return prepend(buffer, buflen, name->name, name->len); | |
2553 | } | |
2554 | ||
1da177e4 | 2555 | /** |
208898c1 | 2556 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2557 | * @path: the dentry/vfsmount to report |
02125a82 | 2558 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2559 | * @buffer: pointer to the end of the buffer |
2560 | * @buflen: pointer to buffer length | |
552ce544 | 2561 | * |
949854d0 | 2562 | * Caller holds the rename_lock. |
1da177e4 | 2563 | */ |
02125a82 AV |
2564 | static int prepend_path(const struct path *path, |
2565 | const struct path *root, | |
f2eb6575 | 2566 | char **buffer, int *buflen) |
1da177e4 | 2567 | { |
9d1bc601 MS |
2568 | struct dentry *dentry = path->dentry; |
2569 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2570 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 MS |
2571 | bool slash = false; |
2572 | int error = 0; | |
6092d048 | 2573 | |
962830df | 2574 | br_read_lock(&vfsmount_lock); |
f2eb6575 | 2575 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2576 | struct dentry * parent; |
2577 | ||
1da177e4 | 2578 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2579 | /* Global root? */ |
676da58d | 2580 | if (!mnt_has_parent(mnt)) |
1da177e4 | 2581 | goto global_root; |
a73324da | 2582 | dentry = mnt->mnt_mountpoint; |
0714a533 AV |
2583 | mnt = mnt->mnt_parent; |
2584 | vfsmnt = &mnt->mnt; | |
1da177e4 LT |
2585 | continue; |
2586 | } | |
2587 | parent = dentry->d_parent; | |
2588 | prefetch(parent); | |
9abca360 | 2589 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2590 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2591 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2592 | if (!error) |
2593 | error = prepend(buffer, buflen, "/", 1); | |
2594 | if (error) | |
2595 | break; | |
2596 | ||
2597 | slash = true; | |
1da177e4 LT |
2598 | dentry = parent; |
2599 | } | |
2600 | ||
f2eb6575 MS |
2601 | if (!error && !slash) |
2602 | error = prepend(buffer, buflen, "/", 1); | |
2603 | ||
02125a82 | 2604 | out: |
962830df | 2605 | br_read_unlock(&vfsmount_lock); |
f2eb6575 | 2606 | return error; |
1da177e4 LT |
2607 | |
2608 | global_root: | |
98dc568b MS |
2609 | /* |
2610 | * Filesystems needing to implement special "root names" | |
2611 | * should do so with ->d_dname() | |
2612 | */ | |
2613 | if (IS_ROOT(dentry) && | |
2614 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2615 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2616 | (int) dentry->d_name.len, dentry->d_name.name); | |
2617 | } | |
02125a82 AV |
2618 | if (!slash) |
2619 | error = prepend(buffer, buflen, "/", 1); | |
2620 | if (!error) | |
143c8c91 | 2621 | error = real_mount(vfsmnt)->mnt_ns ? 1 : 2; |
be285c71 | 2622 | goto out; |
f2eb6575 | 2623 | } |
be285c71 | 2624 | |
f2eb6575 MS |
2625 | /** |
2626 | * __d_path - return the path of a dentry | |
2627 | * @path: the dentry/vfsmount to report | |
02125a82 | 2628 | * @root: root vfsmnt/dentry |
cd956a1c | 2629 | * @buf: buffer to return value in |
f2eb6575 MS |
2630 | * @buflen: buffer length |
2631 | * | |
ffd1f4ed | 2632 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2633 | * |
2634 | * Returns a pointer into the buffer or an error code if the | |
2635 | * path was too long. | |
2636 | * | |
be148247 | 2637 | * "buflen" should be positive. |
f2eb6575 | 2638 | * |
02125a82 | 2639 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2640 | */ |
02125a82 AV |
2641 | char *__d_path(const struct path *path, |
2642 | const struct path *root, | |
f2eb6575 MS |
2643 | char *buf, int buflen) |
2644 | { | |
2645 | char *res = buf + buflen; | |
2646 | int error; | |
2647 | ||
2648 | prepend(&res, &buflen, "\0", 1); | |
949854d0 | 2649 | write_seqlock(&rename_lock); |
f2eb6575 | 2650 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2651 | write_sequnlock(&rename_lock); |
be148247 | 2652 | |
02125a82 AV |
2653 | if (error < 0) |
2654 | return ERR_PTR(error); | |
2655 | if (error > 0) | |
2656 | return NULL; | |
2657 | return res; | |
2658 | } | |
2659 | ||
2660 | char *d_absolute_path(const struct path *path, | |
2661 | char *buf, int buflen) | |
2662 | { | |
2663 | struct path root = {}; | |
2664 | char *res = buf + buflen; | |
2665 | int error; | |
2666 | ||
2667 | prepend(&res, &buflen, "\0", 1); | |
2668 | write_seqlock(&rename_lock); | |
2669 | error = prepend_path(path, &root, &res, &buflen); | |
2670 | write_sequnlock(&rename_lock); | |
2671 | ||
2672 | if (error > 1) | |
2673 | error = -EINVAL; | |
2674 | if (error < 0) | |
f2eb6575 | 2675 | return ERR_PTR(error); |
f2eb6575 | 2676 | return res; |
1da177e4 LT |
2677 | } |
2678 | ||
ffd1f4ed MS |
2679 | /* |
2680 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2681 | */ | |
02125a82 AV |
2682 | static int path_with_deleted(const struct path *path, |
2683 | const struct path *root, | |
2684 | char **buf, int *buflen) | |
ffd1f4ed MS |
2685 | { |
2686 | prepend(buf, buflen, "\0", 1); | |
2687 | if (d_unlinked(path->dentry)) { | |
2688 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2689 | if (error) | |
2690 | return error; | |
2691 | } | |
2692 | ||
2693 | return prepend_path(path, root, buf, buflen); | |
2694 | } | |
2695 | ||
8df9d1a4 MS |
2696 | static int prepend_unreachable(char **buffer, int *buflen) |
2697 | { | |
2698 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2699 | } | |
2700 | ||
a03a8a70 JB |
2701 | /** |
2702 | * d_path - return the path of a dentry | |
cf28b486 | 2703 | * @path: path to report |
a03a8a70 JB |
2704 | * @buf: buffer to return value in |
2705 | * @buflen: buffer length | |
2706 | * | |
2707 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2708 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2709 | * | |
52afeefb AV |
2710 | * Returns a pointer into the buffer or an error code if the path was |
2711 | * too long. Note: Callers should use the returned pointer, not the passed | |
2712 | * in buffer, to use the name! The implementation often starts at an offset | |
2713 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2714 | * |
31f3e0b3 | 2715 | * "buflen" should be positive. |
a03a8a70 | 2716 | */ |
20d4fdc1 | 2717 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2718 | { |
ffd1f4ed | 2719 | char *res = buf + buflen; |
6ac08c39 | 2720 | struct path root; |
ffd1f4ed | 2721 | int error; |
1da177e4 | 2722 | |
c23fbb6b ED |
2723 | /* |
2724 | * We have various synthetic filesystems that never get mounted. On | |
2725 | * these filesystems dentries are never used for lookup purposes, and | |
2726 | * thus don't need to be hashed. They also don't need a name until a | |
2727 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2728 | * below allows us to generate a name for these objects on demand: | |
2729 | */ | |
cf28b486 JB |
2730 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2731 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2732 | |
f7ad3c6b | 2733 | get_fs_root(current->fs, &root); |
949854d0 | 2734 | write_seqlock(&rename_lock); |
02125a82 AV |
2735 | error = path_with_deleted(path, &root, &res, &buflen); |
2736 | if (error < 0) | |
ffd1f4ed | 2737 | res = ERR_PTR(error); |
949854d0 | 2738 | write_sequnlock(&rename_lock); |
6ac08c39 | 2739 | path_put(&root); |
1da177e4 LT |
2740 | return res; |
2741 | } | |
ec4f8605 | 2742 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2743 | |
8df9d1a4 MS |
2744 | /** |
2745 | * d_path_with_unreachable - return the path of a dentry | |
2746 | * @path: path to report | |
2747 | * @buf: buffer to return value in | |
2748 | * @buflen: buffer length | |
2749 | * | |
2750 | * The difference from d_path() is that this prepends "(unreachable)" | |
2751 | * to paths which are unreachable from the current process' root. | |
2752 | */ | |
2753 | char *d_path_with_unreachable(const struct path *path, char *buf, int buflen) | |
2754 | { | |
2755 | char *res = buf + buflen; | |
2756 | struct path root; | |
8df9d1a4 MS |
2757 | int error; |
2758 | ||
2759 | if (path->dentry->d_op && path->dentry->d_op->d_dname) | |
2760 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
2761 | ||
2762 | get_fs_root(current->fs, &root); | |
949854d0 | 2763 | write_seqlock(&rename_lock); |
02125a82 AV |
2764 | error = path_with_deleted(path, &root, &res, &buflen); |
2765 | if (error > 0) | |
8df9d1a4 | 2766 | error = prepend_unreachable(&res, &buflen); |
949854d0 | 2767 | write_sequnlock(&rename_lock); |
8df9d1a4 MS |
2768 | path_put(&root); |
2769 | if (error) | |
2770 | res = ERR_PTR(error); | |
2771 | ||
2772 | return res; | |
2773 | } | |
2774 | ||
c23fbb6b ED |
2775 | /* |
2776 | * Helper function for dentry_operations.d_dname() members | |
2777 | */ | |
2778 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2779 | const char *fmt, ...) | |
2780 | { | |
2781 | va_list args; | |
2782 | char temp[64]; | |
2783 | int sz; | |
2784 | ||
2785 | va_start(args, fmt); | |
2786 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2787 | va_end(args); | |
2788 | ||
2789 | if (sz > sizeof(temp) || sz > buflen) | |
2790 | return ERR_PTR(-ENAMETOOLONG); | |
2791 | ||
2792 | buffer += buflen - sz; | |
2793 | return memcpy(buffer, temp, sz); | |
2794 | } | |
2795 | ||
6092d048 RP |
2796 | /* |
2797 | * Write full pathname from the root of the filesystem into the buffer. | |
2798 | */ | |
ec2447c2 | 2799 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2800 | { |
2801 | char *end = buf + buflen; | |
2802 | char *retval; | |
2803 | ||
6092d048 | 2804 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2805 | if (buflen < 1) |
2806 | goto Elong; | |
2807 | /* Get '/' right */ | |
2808 | retval = end-1; | |
2809 | *retval = '/'; | |
2810 | ||
cdd16d02 MS |
2811 | while (!IS_ROOT(dentry)) { |
2812 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2813 | int error; |
6092d048 | 2814 | |
6092d048 | 2815 | prefetch(parent); |
9abca360 NP |
2816 | spin_lock(&dentry->d_lock); |
2817 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2818 | spin_unlock(&dentry->d_lock); | |
2819 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2820 | goto Elong; |
2821 | ||
2822 | retval = end; | |
2823 | dentry = parent; | |
2824 | } | |
c103135c AV |
2825 | return retval; |
2826 | Elong: | |
2827 | return ERR_PTR(-ENAMETOOLONG); | |
2828 | } | |
ec2447c2 NP |
2829 | |
2830 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2831 | { | |
2832 | char *retval; | |
2833 | ||
949854d0 | 2834 | write_seqlock(&rename_lock); |
ec2447c2 | 2835 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2836 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2837 | |
2838 | return retval; | |
2839 | } | |
2840 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2841 | |
2842 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2843 | { | |
2844 | char *p = NULL; | |
2845 | char *retval; | |
2846 | ||
949854d0 | 2847 | write_seqlock(&rename_lock); |
c103135c AV |
2848 | if (d_unlinked(dentry)) { |
2849 | p = buf + buflen; | |
2850 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2851 | goto Elong; | |
2852 | buflen++; | |
2853 | } | |
2854 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2855 | write_sequnlock(&rename_lock); |
c103135c AV |
2856 | if (!IS_ERR(retval) && p) |
2857 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2858 | return retval; |
2859 | Elong: | |
6092d048 RP |
2860 | return ERR_PTR(-ENAMETOOLONG); |
2861 | } | |
2862 | ||
1da177e4 LT |
2863 | /* |
2864 | * NOTE! The user-level library version returns a | |
2865 | * character pointer. The kernel system call just | |
2866 | * returns the length of the buffer filled (which | |
2867 | * includes the ending '\0' character), or a negative | |
2868 | * error value. So libc would do something like | |
2869 | * | |
2870 | * char *getcwd(char * buf, size_t size) | |
2871 | * { | |
2872 | * int retval; | |
2873 | * | |
2874 | * retval = sys_getcwd(buf, size); | |
2875 | * if (retval >= 0) | |
2876 | * return buf; | |
2877 | * errno = -retval; | |
2878 | * return NULL; | |
2879 | * } | |
2880 | */ | |
3cdad428 | 2881 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2882 | { |
552ce544 | 2883 | int error; |
6ac08c39 | 2884 | struct path pwd, root; |
552ce544 | 2885 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2886 | |
2887 | if (!page) | |
2888 | return -ENOMEM; | |
2889 | ||
f7ad3c6b | 2890 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2891 | |
552ce544 | 2892 | error = -ENOENT; |
949854d0 | 2893 | write_seqlock(&rename_lock); |
f3da392e | 2894 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2895 | unsigned long len; |
8df9d1a4 MS |
2896 | char *cwd = page + PAGE_SIZE; |
2897 | int buflen = PAGE_SIZE; | |
1da177e4 | 2898 | |
8df9d1a4 | 2899 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 2900 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
949854d0 | 2901 | write_sequnlock(&rename_lock); |
552ce544 | 2902 | |
02125a82 | 2903 | if (error < 0) |
552ce544 LT |
2904 | goto out; |
2905 | ||
8df9d1a4 | 2906 | /* Unreachable from current root */ |
02125a82 | 2907 | if (error > 0) { |
8df9d1a4 MS |
2908 | error = prepend_unreachable(&cwd, &buflen); |
2909 | if (error) | |
2910 | goto out; | |
2911 | } | |
2912 | ||
552ce544 LT |
2913 | error = -ERANGE; |
2914 | len = PAGE_SIZE + page - cwd; | |
2915 | if (len <= size) { | |
2916 | error = len; | |
2917 | if (copy_to_user(buf, cwd, len)) | |
2918 | error = -EFAULT; | |
2919 | } | |
949854d0 NP |
2920 | } else { |
2921 | write_sequnlock(&rename_lock); | |
949854d0 | 2922 | } |
1da177e4 LT |
2923 | |
2924 | out: | |
6ac08c39 JB |
2925 | path_put(&pwd); |
2926 | path_put(&root); | |
1da177e4 LT |
2927 | free_page((unsigned long) page); |
2928 | return error; | |
2929 | } | |
2930 | ||
2931 | /* | |
2932 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2933 | * | |
2934 | * Trivially implemented using the dcache structure | |
2935 | */ | |
2936 | ||
2937 | /** | |
2938 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2939 | * @new_dentry: new dentry | |
2940 | * @old_dentry: old dentry | |
2941 | * | |
2942 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2943 | * Returns 0 otherwise. | |
2944 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2945 | */ | |
2946 | ||
e2761a11 | 2947 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2948 | { |
2949 | int result; | |
949854d0 | 2950 | unsigned seq; |
1da177e4 | 2951 | |
e2761a11 OH |
2952 | if (new_dentry == old_dentry) |
2953 | return 1; | |
2954 | ||
e2761a11 | 2955 | do { |
1da177e4 | 2956 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2957 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2958 | /* |
2959 | * Need rcu_readlock to protect against the d_parent trashing | |
2960 | * due to d_move | |
2961 | */ | |
2962 | rcu_read_lock(); | |
e2761a11 | 2963 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2964 | result = 1; |
e2761a11 OH |
2965 | else |
2966 | result = 0; | |
949854d0 | 2967 | rcu_read_unlock(); |
1da177e4 | 2968 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2969 | |
2970 | return result; | |
2971 | } | |
2972 | ||
2973 | void d_genocide(struct dentry *root) | |
2974 | { | |
949854d0 | 2975 | struct dentry *this_parent; |
1da177e4 | 2976 | struct list_head *next; |
949854d0 | 2977 | unsigned seq; |
58db63d0 | 2978 | int locked = 0; |
1da177e4 | 2979 | |
949854d0 | 2980 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2981 | again: |
2982 | this_parent = root; | |
2fd6b7f5 | 2983 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2984 | repeat: |
2985 | next = this_parent->d_subdirs.next; | |
2986 | resume: | |
2987 | while (next != &this_parent->d_subdirs) { | |
2988 | struct list_head *tmp = next; | |
5160ee6f | 2989 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2990 | next = tmp->next; |
949854d0 | 2991 | |
da502956 NP |
2992 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2993 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2994 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2995 | continue; |
da502956 | 2996 | } |
1da177e4 | 2997 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2998 | spin_unlock(&this_parent->d_lock); |
2999 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 3000 | this_parent = dentry; |
2fd6b7f5 | 3001 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
3002 | goto repeat; |
3003 | } | |
949854d0 NP |
3004 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3005 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3006 | dentry->d_count--; | |
3007 | } | |
b7ab39f6 | 3008 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
3009 | } |
3010 | if (this_parent != root) { | |
c826cb7d | 3011 | struct dentry *child = this_parent; |
949854d0 NP |
3012 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
3013 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
3014 | this_parent->d_count--; | |
3015 | } | |
c826cb7d LT |
3016 | this_parent = try_to_ascend(this_parent, locked, seq); |
3017 | if (!this_parent) | |
949854d0 | 3018 | goto rename_retry; |
949854d0 | 3019 | next = child->d_u.d_child.next; |
1da177e4 LT |
3020 | goto resume; |
3021 | } | |
2fd6b7f5 | 3022 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 3023 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 3024 | goto rename_retry; |
58db63d0 NP |
3025 | if (locked) |
3026 | write_sequnlock(&rename_lock); | |
3027 | return; | |
3028 | ||
3029 | rename_retry: | |
3030 | locked = 1; | |
3031 | write_seqlock(&rename_lock); | |
3032 | goto again; | |
1da177e4 LT |
3033 | } |
3034 | ||
3035 | /** | |
3036 | * find_inode_number - check for dentry with name | |
3037 | * @dir: directory to check | |
3038 | * @name: Name to find. | |
3039 | * | |
3040 | * Check whether a dentry already exists for the given name, | |
3041 | * and return the inode number if it has an inode. Otherwise | |
3042 | * 0 is returned. | |
3043 | * | |
3044 | * This routine is used to post-process directory listings for | |
3045 | * filesystems using synthetic inode numbers, and is necessary | |
3046 | * to keep getcwd() working. | |
3047 | */ | |
3048 | ||
3049 | ino_t find_inode_number(struct dentry *dir, struct qstr *name) | |
3050 | { | |
3051 | struct dentry * dentry; | |
3052 | ino_t ino = 0; | |
3053 | ||
3e7e241f EB |
3054 | dentry = d_hash_and_lookup(dir, name); |
3055 | if (dentry) { | |
1da177e4 LT |
3056 | if (dentry->d_inode) |
3057 | ino = dentry->d_inode->i_ino; | |
3058 | dput(dentry); | |
3059 | } | |
1da177e4 LT |
3060 | return ino; |
3061 | } | |
ec4f8605 | 3062 | EXPORT_SYMBOL(find_inode_number); |
1da177e4 LT |
3063 | |
3064 | static __initdata unsigned long dhash_entries; | |
3065 | static int __init set_dhash_entries(char *str) | |
3066 | { | |
3067 | if (!str) | |
3068 | return 0; | |
3069 | dhash_entries = simple_strtoul(str, &str, 0); | |
3070 | return 1; | |
3071 | } | |
3072 | __setup("dhash_entries=", set_dhash_entries); | |
3073 | ||
3074 | static void __init dcache_init_early(void) | |
3075 | { | |
074b8517 | 3076 | unsigned int loop; |
1da177e4 LT |
3077 | |
3078 | /* If hashes are distributed across NUMA nodes, defer | |
3079 | * hash allocation until vmalloc space is available. | |
3080 | */ | |
3081 | if (hashdist) | |
3082 | return; | |
3083 | ||
3084 | dentry_hashtable = | |
3085 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3086 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3087 | dhash_entries, |
3088 | 13, | |
3089 | HASH_EARLY, | |
3090 | &d_hash_shift, | |
3091 | &d_hash_mask, | |
31fe62b9 | 3092 | 0, |
1da177e4 LT |
3093 | 0); |
3094 | ||
074b8517 | 3095 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3096 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3097 | } |
3098 | ||
74bf17cf | 3099 | static void __init dcache_init(void) |
1da177e4 | 3100 | { |
074b8517 | 3101 | unsigned int loop; |
1da177e4 LT |
3102 | |
3103 | /* | |
3104 | * A constructor could be added for stable state like the lists, | |
3105 | * but it is probably not worth it because of the cache nature | |
3106 | * of the dcache. | |
3107 | */ | |
0a31bd5f CL |
3108 | dentry_cache = KMEM_CACHE(dentry, |
3109 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3110 | |
3111 | /* Hash may have been set up in dcache_init_early */ | |
3112 | if (!hashdist) | |
3113 | return; | |
3114 | ||
3115 | dentry_hashtable = | |
3116 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3117 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3118 | dhash_entries, |
3119 | 13, | |
3120 | 0, | |
3121 | &d_hash_shift, | |
3122 | &d_hash_mask, | |
31fe62b9 | 3123 | 0, |
1da177e4 LT |
3124 | 0); |
3125 | ||
074b8517 | 3126 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3127 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3128 | } |
3129 | ||
3130 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3131 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3132 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3133 | |
1da177e4 LT |
3134 | EXPORT_SYMBOL(d_genocide); |
3135 | ||
1da177e4 LT |
3136 | void __init vfs_caches_init_early(void) |
3137 | { | |
3138 | dcache_init_early(); | |
3139 | inode_init_early(); | |
3140 | } | |
3141 | ||
3142 | void __init vfs_caches_init(unsigned long mempages) | |
3143 | { | |
3144 | unsigned long reserve; | |
3145 | ||
3146 | /* Base hash sizes on available memory, with a reserve equal to | |
3147 | 150% of current kernel size */ | |
3148 | ||
3149 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3150 | mempages -= reserve; | |
3151 | ||
3152 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3153 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3154 | |
74bf17cf DC |
3155 | dcache_init(); |
3156 | inode_init(); | |
1da177e4 | 3157 | files_init(mempages); |
74bf17cf | 3158 | mnt_init(); |
1da177e4 LT |
3159 | bdev_cache_init(); |
3160 | chrdev_init(); | |
3161 | } |