]>
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 | ||
b3d9b7a3 | 221 | WARN_ON(!hlist_unhashed(&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 | { | |
98474236 | 232 | BUG_ON(dentry->d_lockref.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; | |
b3d9b7a3 | 270 | hlist_del_init(&dentry->d_alias); |
1da177e4 | 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; | |
b3d9b7a3 | 294 | hlist_del_init(&dentry->d_alias); |
31e6b01f NP |
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 | ||
b48f03b3 | 340 | static void dentry_lru_move_list(struct dentry *dentry, struct list_head *list) |
da3bbdd4 | 341 | { |
23044507 | 342 | spin_lock(&dcache_lru_lock); |
a4633357 | 343 | if (list_empty(&dentry->d_lru)) { |
b48f03b3 | 344 | list_add_tail(&dentry->d_lru, list); |
a4633357 | 345 | dentry->d_sb->s_nr_dentry_unused++; |
86c8749e | 346 | dentry_stat.nr_unused++; |
a4633357 | 347 | } else { |
b48f03b3 | 348 | list_move_tail(&dentry->d_lru, list); |
da3bbdd4 | 349 | } |
23044507 | 350 | spin_unlock(&dcache_lru_lock); |
da3bbdd4 KM |
351 | } |
352 | ||
d52b9086 MS |
353 | /** |
354 | * d_kill - kill dentry and return parent | |
355 | * @dentry: dentry to kill | |
ff5fdb61 | 356 | * @parent: parent dentry |
d52b9086 | 357 | * |
31f3e0b3 | 358 | * The dentry must already be unhashed and removed from the LRU. |
d52b9086 MS |
359 | * |
360 | * If this is the root of the dentry tree, return NULL. | |
23044507 | 361 | * |
b5c84bf6 NP |
362 | * dentry->d_lock and parent->d_lock must be held by caller, and are dropped by |
363 | * d_kill. | |
d52b9086 | 364 | */ |
2fd6b7f5 | 365 | static struct dentry *d_kill(struct dentry *dentry, struct dentry *parent) |
31f3e0b3 | 366 | __releases(dentry->d_lock) |
2fd6b7f5 | 367 | __releases(parent->d_lock) |
873feea0 | 368 | __releases(dentry->d_inode->i_lock) |
d52b9086 | 369 | { |
d52b9086 | 370 | list_del(&dentry->d_u.d_child); |
c83ce989 TM |
371 | /* |
372 | * Inform try_to_ascend() that we are no longer attached to the | |
373 | * dentry tree | |
374 | */ | |
b161dfa6 | 375 | dentry->d_flags |= DCACHE_DENTRY_KILLED; |
2fd6b7f5 NP |
376 | if (parent) |
377 | spin_unlock(&parent->d_lock); | |
d52b9086 | 378 | dentry_iput(dentry); |
b7ab39f6 NP |
379 | /* |
380 | * dentry_iput drops the locks, at which point nobody (except | |
381 | * transient RCU lookups) can reach this dentry. | |
382 | */ | |
d52b9086 | 383 | d_free(dentry); |
871c0067 | 384 | return parent; |
d52b9086 MS |
385 | } |
386 | ||
c6627c60 DH |
387 | /* |
388 | * Unhash a dentry without inserting an RCU walk barrier or checking that | |
389 | * dentry->d_lock is locked. The caller must take care of that, if | |
390 | * appropriate. | |
391 | */ | |
392 | static void __d_shrink(struct dentry *dentry) | |
393 | { | |
394 | if (!d_unhashed(dentry)) { | |
395 | struct hlist_bl_head *b; | |
396 | if (unlikely(dentry->d_flags & DCACHE_DISCONNECTED)) | |
397 | b = &dentry->d_sb->s_anon; | |
398 | else | |
399 | b = d_hash(dentry->d_parent, dentry->d_name.hash); | |
400 | ||
401 | hlist_bl_lock(b); | |
402 | __hlist_bl_del(&dentry->d_hash); | |
403 | dentry->d_hash.pprev = NULL; | |
404 | hlist_bl_unlock(b); | |
405 | } | |
406 | } | |
407 | ||
789680d1 NP |
408 | /** |
409 | * d_drop - drop a dentry | |
410 | * @dentry: dentry to drop | |
411 | * | |
412 | * d_drop() unhashes the entry from the parent dentry hashes, so that it won't | |
413 | * be found through a VFS lookup any more. Note that this is different from | |
414 | * deleting the dentry - d_delete will try to mark the dentry negative if | |
415 | * possible, giving a successful _negative_ lookup, while d_drop will | |
416 | * just make the cache lookup fail. | |
417 | * | |
418 | * d_drop() is used mainly for stuff that wants to invalidate a dentry for some | |
419 | * reason (NFS timeouts or autofs deletes). | |
420 | * | |
421 | * __d_drop requires dentry->d_lock. | |
422 | */ | |
423 | void __d_drop(struct dentry *dentry) | |
424 | { | |
dea3667b | 425 | if (!d_unhashed(dentry)) { |
c6627c60 | 426 | __d_shrink(dentry); |
dea3667b | 427 | dentry_rcuwalk_barrier(dentry); |
789680d1 NP |
428 | } |
429 | } | |
430 | EXPORT_SYMBOL(__d_drop); | |
431 | ||
432 | void d_drop(struct dentry *dentry) | |
433 | { | |
789680d1 NP |
434 | spin_lock(&dentry->d_lock); |
435 | __d_drop(dentry); | |
436 | spin_unlock(&dentry->d_lock); | |
789680d1 NP |
437 | } |
438 | EXPORT_SYMBOL(d_drop); | |
439 | ||
77812a1e NP |
440 | /* |
441 | * Finish off a dentry we've decided to kill. | |
442 | * dentry->d_lock must be held, returns with it unlocked. | |
443 | * If ref is non-zero, then decrement the refcount too. | |
444 | * Returns dentry requiring refcount drop, or NULL if we're done. | |
445 | */ | |
446 | static inline struct dentry *dentry_kill(struct dentry *dentry, int ref) | |
447 | __releases(dentry->d_lock) | |
448 | { | |
873feea0 | 449 | struct inode *inode; |
77812a1e NP |
450 | struct dentry *parent; |
451 | ||
873feea0 NP |
452 | inode = dentry->d_inode; |
453 | if (inode && !spin_trylock(&inode->i_lock)) { | |
77812a1e NP |
454 | relock: |
455 | spin_unlock(&dentry->d_lock); | |
456 | cpu_relax(); | |
457 | return dentry; /* try again with same dentry */ | |
458 | } | |
459 | if (IS_ROOT(dentry)) | |
460 | parent = NULL; | |
461 | else | |
462 | parent = dentry->d_parent; | |
463 | if (parent && !spin_trylock(&parent->d_lock)) { | |
873feea0 NP |
464 | if (inode) |
465 | spin_unlock(&inode->i_lock); | |
77812a1e NP |
466 | goto relock; |
467 | } | |
31e6b01f | 468 | |
77812a1e | 469 | if (ref) |
98474236 | 470 | dentry->d_lockref.count--; |
f0023bc6 | 471 | /* |
f0023bc6 SW |
472 | * inform the fs via d_prune that this dentry is about to be |
473 | * unhashed and destroyed. | |
474 | */ | |
61572bb1 YZ |
475 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
476 | dentry->d_op->d_prune(dentry); | |
477 | ||
478 | dentry_lru_del(dentry); | |
77812a1e NP |
479 | /* if it was on the hash then remove it */ |
480 | __d_drop(dentry); | |
481 | return d_kill(dentry, parent); | |
482 | } | |
483 | ||
1da177e4 LT |
484 | /* |
485 | * This is dput | |
486 | * | |
487 | * This is complicated by the fact that we do not want to put | |
488 | * dentries that are no longer on any hash chain on the unused | |
489 | * list: we'd much rather just get rid of them immediately. | |
490 | * | |
491 | * However, that implies that we have to traverse the dentry | |
492 | * tree upwards to the parents which might _also_ now be | |
493 | * scheduled for deletion (it may have been only waiting for | |
494 | * its last child to go away). | |
495 | * | |
496 | * This tail recursion is done by hand as we don't want to depend | |
497 | * on the compiler to always get this right (gcc generally doesn't). | |
498 | * Real recursion would eat up our stack space. | |
499 | */ | |
500 | ||
501 | /* | |
502 | * dput - release a dentry | |
503 | * @dentry: dentry to release | |
504 | * | |
505 | * Release a dentry. This will drop the usage count and if appropriate | |
506 | * call the dentry unlink method as well as removing it from the queues and | |
507 | * releasing its resources. If the parent dentries were scheduled for release | |
508 | * they too may now get deleted. | |
1da177e4 | 509 | */ |
1da177e4 LT |
510 | void dput(struct dentry *dentry) |
511 | { | |
512 | if (!dentry) | |
513 | return; | |
514 | ||
515 | repeat: | |
98474236 | 516 | if (dentry->d_lockref.count == 1) |
1da177e4 | 517 | might_sleep(); |
98474236 | 518 | if (lockref_put_or_lock(&dentry->d_lockref)) |
1da177e4 | 519 | return; |
1da177e4 | 520 | |
fb045adb | 521 | if (dentry->d_flags & DCACHE_OP_DELETE) { |
1da177e4 | 522 | if (dentry->d_op->d_delete(dentry)) |
61f3dee4 | 523 | goto kill_it; |
1da177e4 | 524 | } |
265ac902 | 525 | |
1da177e4 LT |
526 | /* Unreachable? Get rid of it */ |
527 | if (d_unhashed(dentry)) | |
528 | goto kill_it; | |
265ac902 | 529 | |
39e3c955 | 530 | dentry->d_flags |= DCACHE_REFERENCED; |
a4633357 | 531 | dentry_lru_add(dentry); |
265ac902 | 532 | |
98474236 | 533 | dentry->d_lockref.count--; |
61f3dee4 | 534 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
535 | return; |
536 | ||
d52b9086 | 537 | kill_it: |
77812a1e | 538 | dentry = dentry_kill(dentry, 1); |
d52b9086 MS |
539 | if (dentry) |
540 | goto repeat; | |
1da177e4 | 541 | } |
ec4f8605 | 542 | EXPORT_SYMBOL(dput); |
1da177e4 LT |
543 | |
544 | /** | |
545 | * d_invalidate - invalidate a dentry | |
546 | * @dentry: dentry to invalidate | |
547 | * | |
548 | * Try to invalidate the dentry if it turns out to be | |
549 | * possible. If there are other dentries that can be | |
550 | * reached through this one we can't delete it and we | |
551 | * return -EBUSY. On success we return 0. | |
552 | * | |
553 | * no dcache lock. | |
554 | */ | |
555 | ||
556 | int d_invalidate(struct dentry * dentry) | |
557 | { | |
558 | /* | |
559 | * If it's already been dropped, return OK. | |
560 | */ | |
da502956 | 561 | spin_lock(&dentry->d_lock); |
1da177e4 | 562 | if (d_unhashed(dentry)) { |
da502956 | 563 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
564 | return 0; |
565 | } | |
566 | /* | |
567 | * Check whether to do a partial shrink_dcache | |
568 | * to get rid of unused child entries. | |
569 | */ | |
570 | if (!list_empty(&dentry->d_subdirs)) { | |
da502956 | 571 | spin_unlock(&dentry->d_lock); |
1da177e4 | 572 | shrink_dcache_parent(dentry); |
da502956 | 573 | spin_lock(&dentry->d_lock); |
1da177e4 LT |
574 | } |
575 | ||
576 | /* | |
577 | * Somebody else still using it? | |
578 | * | |
579 | * If it's a directory, we can't drop it | |
580 | * for fear of somebody re-populating it | |
581 | * with children (even though dropping it | |
582 | * would make it unreachable from the root, | |
583 | * we might still populate it if it was a | |
584 | * working directory or similar). | |
50e69630 AV |
585 | * We also need to leave mountpoints alone, |
586 | * directory or not. | |
1da177e4 | 587 | */ |
98474236 | 588 | if (dentry->d_lockref.count > 1 && dentry->d_inode) { |
50e69630 | 589 | if (S_ISDIR(dentry->d_inode->i_mode) || d_mountpoint(dentry)) { |
1da177e4 | 590 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
591 | return -EBUSY; |
592 | } | |
593 | } | |
594 | ||
595 | __d_drop(dentry); | |
596 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
597 | return 0; |
598 | } | |
ec4f8605 | 599 | EXPORT_SYMBOL(d_invalidate); |
1da177e4 | 600 | |
b5c84bf6 | 601 | /* This must be called with d_lock held */ |
dc0474be | 602 | static inline void __dget_dlock(struct dentry *dentry) |
23044507 | 603 | { |
98474236 | 604 | dentry->d_lockref.count++; |
23044507 NP |
605 | } |
606 | ||
dc0474be | 607 | static inline void __dget(struct dentry *dentry) |
1da177e4 | 608 | { |
98474236 | 609 | lockref_get(&dentry->d_lockref); |
1da177e4 LT |
610 | } |
611 | ||
b7ab39f6 NP |
612 | struct dentry *dget_parent(struct dentry *dentry) |
613 | { | |
614 | struct dentry *ret; | |
615 | ||
616 | repeat: | |
a734eb45 NP |
617 | /* |
618 | * Don't need rcu_dereference because we re-check it was correct under | |
619 | * the lock. | |
620 | */ | |
621 | rcu_read_lock(); | |
b7ab39f6 | 622 | ret = dentry->d_parent; |
a734eb45 NP |
623 | spin_lock(&ret->d_lock); |
624 | if (unlikely(ret != dentry->d_parent)) { | |
625 | spin_unlock(&ret->d_lock); | |
626 | rcu_read_unlock(); | |
b7ab39f6 NP |
627 | goto repeat; |
628 | } | |
a734eb45 | 629 | rcu_read_unlock(); |
98474236 WL |
630 | BUG_ON(!ret->d_lockref.count); |
631 | ret->d_lockref.count++; | |
b7ab39f6 | 632 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
633 | return ret; |
634 | } | |
635 | EXPORT_SYMBOL(dget_parent); | |
636 | ||
1da177e4 LT |
637 | /** |
638 | * d_find_alias - grab a hashed alias of inode | |
639 | * @inode: inode in question | |
32ba9c3f LT |
640 | * @want_discon: flag, used by d_splice_alias, to request |
641 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
642 | * |
643 | * If inode has a hashed alias, or is a directory and has any alias, | |
644 | * acquire the reference to alias and return it. Otherwise return NULL. | |
645 | * Notice that if inode is a directory there can be only one alias and | |
646 | * it can be unhashed only if it has no children, or if it is the root | |
647 | * of a filesystem. | |
648 | * | |
21c0d8fd | 649 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
650 | * any other hashed alias over that one unless @want_discon is set, |
651 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 652 | */ |
32ba9c3f | 653 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 654 | { |
da502956 | 655 | struct dentry *alias, *discon_alias; |
1da177e4 | 656 | |
da502956 NP |
657 | again: |
658 | discon_alias = NULL; | |
b67bfe0d | 659 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 660 | spin_lock(&alias->d_lock); |
1da177e4 | 661 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 662 | if (IS_ROOT(alias) && |
da502956 | 663 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 664 | discon_alias = alias; |
32ba9c3f | 665 | } else if (!want_discon) { |
dc0474be | 666 | __dget_dlock(alias); |
da502956 NP |
667 | spin_unlock(&alias->d_lock); |
668 | return alias; | |
669 | } | |
670 | } | |
671 | spin_unlock(&alias->d_lock); | |
672 | } | |
673 | if (discon_alias) { | |
674 | alias = discon_alias; | |
675 | spin_lock(&alias->d_lock); | |
676 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
677 | if (IS_ROOT(alias) && | |
678 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 679 | __dget_dlock(alias); |
da502956 | 680 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
681 | return alias; |
682 | } | |
683 | } | |
da502956 NP |
684 | spin_unlock(&alias->d_lock); |
685 | goto again; | |
1da177e4 | 686 | } |
da502956 | 687 | return NULL; |
1da177e4 LT |
688 | } |
689 | ||
da502956 | 690 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 691 | { |
214fda1f DH |
692 | struct dentry *de = NULL; |
693 | ||
b3d9b7a3 | 694 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 695 | spin_lock(&inode->i_lock); |
32ba9c3f | 696 | de = __d_find_alias(inode, 0); |
873feea0 | 697 | spin_unlock(&inode->i_lock); |
214fda1f | 698 | } |
1da177e4 LT |
699 | return de; |
700 | } | |
ec4f8605 | 701 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
702 | |
703 | /* | |
704 | * Try to kill dentries associated with this inode. | |
705 | * WARNING: you must own a reference to inode. | |
706 | */ | |
707 | void d_prune_aliases(struct inode *inode) | |
708 | { | |
0cdca3f9 | 709 | struct dentry *dentry; |
1da177e4 | 710 | restart: |
873feea0 | 711 | spin_lock(&inode->i_lock); |
b67bfe0d | 712 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 713 | spin_lock(&dentry->d_lock); |
98474236 | 714 | if (!dentry->d_lockref.count) { |
dc0474be | 715 | __dget_dlock(dentry); |
1da177e4 LT |
716 | __d_drop(dentry); |
717 | spin_unlock(&dentry->d_lock); | |
873feea0 | 718 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
719 | dput(dentry); |
720 | goto restart; | |
721 | } | |
722 | spin_unlock(&dentry->d_lock); | |
723 | } | |
873feea0 | 724 | spin_unlock(&inode->i_lock); |
1da177e4 | 725 | } |
ec4f8605 | 726 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
727 | |
728 | /* | |
77812a1e NP |
729 | * Try to throw away a dentry - free the inode, dput the parent. |
730 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
731 | * Releases dentry->d_lock. | |
d702ccb3 | 732 | * |
77812a1e | 733 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 734 | */ |
77812a1e | 735 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 736 | __releases(dentry->d_lock) |
1da177e4 | 737 | { |
77812a1e | 738 | struct dentry *parent; |
d52b9086 | 739 | |
77812a1e | 740 | parent = dentry_kill(dentry, 0); |
d52b9086 | 741 | /* |
77812a1e NP |
742 | * If dentry_kill returns NULL, we have nothing more to do. |
743 | * if it returns the same dentry, trylocks failed. In either | |
744 | * case, just loop again. | |
745 | * | |
746 | * Otherwise, we need to prune ancestors too. This is necessary | |
747 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
748 | * is also expected to be beneficial in reducing dentry cache | |
749 | * fragmentation. | |
d52b9086 | 750 | */ |
77812a1e NP |
751 | if (!parent) |
752 | return; | |
753 | if (parent == dentry) | |
754 | return; | |
755 | ||
756 | /* Prune ancestors. */ | |
757 | dentry = parent; | |
d52b9086 | 758 | while (dentry) { |
98474236 | 759 | if (lockref_put_or_lock(&dentry->d_lockref)) |
89e60548 | 760 | return; |
77812a1e | 761 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 762 | } |
1da177e4 LT |
763 | } |
764 | ||
3049cfe2 | 765 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 766 | { |
da3bbdd4 | 767 | struct dentry *dentry; |
da3bbdd4 | 768 | |
ec33679d NP |
769 | rcu_read_lock(); |
770 | for (;;) { | |
ec33679d NP |
771 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
772 | if (&dentry->d_lru == list) | |
773 | break; /* empty */ | |
774 | spin_lock(&dentry->d_lock); | |
775 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
776 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
777 | continue; |
778 | } | |
779 | ||
1da177e4 LT |
780 | /* |
781 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
782 | * the LRU because of laziness during lookup. Do not free |
783 | * it - just keep it off the LRU list. | |
1da177e4 | 784 | */ |
98474236 | 785 | if (dentry->d_lockref.count) { |
ec33679d | 786 | dentry_lru_del(dentry); |
da3bbdd4 | 787 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
788 | continue; |
789 | } | |
ec33679d | 790 | |
ec33679d | 791 | rcu_read_unlock(); |
77812a1e NP |
792 | |
793 | try_prune_one_dentry(dentry); | |
794 | ||
ec33679d | 795 | rcu_read_lock(); |
da3bbdd4 | 796 | } |
ec33679d | 797 | rcu_read_unlock(); |
3049cfe2 CH |
798 | } |
799 | ||
800 | /** | |
b48f03b3 DC |
801 | * prune_dcache_sb - shrink the dcache |
802 | * @sb: superblock | |
803 | * @count: number of entries to try to free | |
804 | * | |
805 | * Attempt to shrink the superblock dcache LRU by @count entries. This is | |
806 | * done when we need more memory an called from the superblock shrinker | |
807 | * function. | |
3049cfe2 | 808 | * |
b48f03b3 DC |
809 | * This function may fail to free any resources if all the dentries are in |
810 | * use. | |
3049cfe2 | 811 | */ |
b48f03b3 | 812 | void prune_dcache_sb(struct super_block *sb, int count) |
3049cfe2 | 813 | { |
3049cfe2 CH |
814 | struct dentry *dentry; |
815 | LIST_HEAD(referenced); | |
816 | LIST_HEAD(tmp); | |
3049cfe2 | 817 | |
23044507 NP |
818 | relock: |
819 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
820 | while (!list_empty(&sb->s_dentry_lru)) { |
821 | dentry = list_entry(sb->s_dentry_lru.prev, | |
822 | struct dentry, d_lru); | |
823 | BUG_ON(dentry->d_sb != sb); | |
824 | ||
23044507 NP |
825 | if (!spin_trylock(&dentry->d_lock)) { |
826 | spin_unlock(&dcache_lru_lock); | |
827 | cpu_relax(); | |
828 | goto relock; | |
829 | } | |
830 | ||
b48f03b3 | 831 | if (dentry->d_flags & DCACHE_REFERENCED) { |
23044507 NP |
832 | dentry->d_flags &= ~DCACHE_REFERENCED; |
833 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 834 | spin_unlock(&dentry->d_lock); |
23044507 NP |
835 | } else { |
836 | list_move_tail(&dentry->d_lru, &tmp); | |
eaf5f907 | 837 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
23044507 | 838 | spin_unlock(&dentry->d_lock); |
b0d40c92 | 839 | if (!--count) |
23044507 | 840 | break; |
3049cfe2 | 841 | } |
ec33679d | 842 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 843 | } |
da3bbdd4 KM |
844 | if (!list_empty(&referenced)) |
845 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 846 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
847 | |
848 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
849 | } |
850 | ||
1da177e4 LT |
851 | /** |
852 | * shrink_dcache_sb - shrink dcache for a superblock | |
853 | * @sb: superblock | |
854 | * | |
3049cfe2 CH |
855 | * Shrink the dcache for the specified super block. This is used to free |
856 | * the dcache before unmounting a file system. | |
1da177e4 | 857 | */ |
3049cfe2 | 858 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 859 | { |
3049cfe2 CH |
860 | LIST_HEAD(tmp); |
861 | ||
23044507 | 862 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
863 | while (!list_empty(&sb->s_dentry_lru)) { |
864 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 865 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 866 | shrink_dentry_list(&tmp); |
ec33679d | 867 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 868 | } |
23044507 | 869 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 870 | } |
ec4f8605 | 871 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 872 | |
c636ebdb DH |
873 | /* |
874 | * destroy a single subtree of dentries for unmount | |
875 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
876 | * locking | |
877 | */ | |
878 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
879 | { | |
880 | struct dentry *parent; | |
881 | ||
882 | BUG_ON(!IS_ROOT(dentry)); | |
883 | ||
c636ebdb DH |
884 | for (;;) { |
885 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 886 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
887 | dentry = list_entry(dentry->d_subdirs.next, |
888 | struct dentry, d_u.d_child); | |
c636ebdb DH |
889 | |
890 | /* consume the dentries from this leaf up through its parents | |
891 | * until we find one with children or run out altogether */ | |
892 | do { | |
893 | struct inode *inode; | |
894 | ||
f0023bc6 | 895 | /* |
61572bb1 | 896 | * inform the fs that this dentry is about to be |
f0023bc6 SW |
897 | * unhashed and destroyed. |
898 | */ | |
61572bb1 YZ |
899 | if (dentry->d_flags & DCACHE_OP_PRUNE) |
900 | dentry->d_op->d_prune(dentry); | |
901 | ||
902 | dentry_lru_del(dentry); | |
43c1c9cd DH |
903 | __d_shrink(dentry); |
904 | ||
98474236 | 905 | if (dentry->d_lockref.count != 0) { |
c636ebdb DH |
906 | printk(KERN_ERR |
907 | "BUG: Dentry %p{i=%lx,n=%s}" | |
908 | " still in use (%d)" | |
909 | " [unmount of %s %s]\n", | |
910 | dentry, | |
911 | dentry->d_inode ? | |
912 | dentry->d_inode->i_ino : 0UL, | |
913 | dentry->d_name.name, | |
98474236 | 914 | dentry->d_lockref.count, |
c636ebdb DH |
915 | dentry->d_sb->s_type->name, |
916 | dentry->d_sb->s_id); | |
917 | BUG(); | |
918 | } | |
919 | ||
2fd6b7f5 | 920 | if (IS_ROOT(dentry)) { |
c636ebdb | 921 | parent = NULL; |
2fd6b7f5 NP |
922 | list_del(&dentry->d_u.d_child); |
923 | } else { | |
871c0067 | 924 | parent = dentry->d_parent; |
98474236 | 925 | parent->d_lockref.count--; |
2fd6b7f5 | 926 | list_del(&dentry->d_u.d_child); |
871c0067 | 927 | } |
c636ebdb | 928 | |
c636ebdb DH |
929 | inode = dentry->d_inode; |
930 | if (inode) { | |
931 | dentry->d_inode = NULL; | |
b3d9b7a3 | 932 | hlist_del_init(&dentry->d_alias); |
c636ebdb DH |
933 | if (dentry->d_op && dentry->d_op->d_iput) |
934 | dentry->d_op->d_iput(dentry, inode); | |
935 | else | |
936 | iput(inode); | |
937 | } | |
938 | ||
939 | d_free(dentry); | |
940 | ||
941 | /* finished when we fall off the top of the tree, | |
942 | * otherwise we ascend to the parent and move to the | |
943 | * next sibling if there is one */ | |
944 | if (!parent) | |
312d3ca8 | 945 | return; |
c636ebdb | 946 | dentry = parent; |
c636ebdb DH |
947 | } while (list_empty(&dentry->d_subdirs)); |
948 | ||
949 | dentry = list_entry(dentry->d_subdirs.next, | |
950 | struct dentry, d_u.d_child); | |
951 | } | |
952 | } | |
953 | ||
954 | /* | |
955 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 956 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
957 | * - the superblock is detached from all mountings and open files, so the |
958 | * dentry trees will not be rearranged by the VFS | |
959 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
960 | * any dentries belonging to this superblock that it comes across | |
961 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
962 | * in this superblock | |
963 | */ | |
964 | void shrink_dcache_for_umount(struct super_block *sb) | |
965 | { | |
966 | struct dentry *dentry; | |
967 | ||
968 | if (down_read_trylock(&sb->s_umount)) | |
969 | BUG(); | |
970 | ||
971 | dentry = sb->s_root; | |
972 | sb->s_root = NULL; | |
98474236 | 973 | dentry->d_lockref.count--; |
c636ebdb DH |
974 | shrink_dcache_for_umount_subtree(dentry); |
975 | ||
ceb5bdc2 NP |
976 | while (!hlist_bl_empty(&sb->s_anon)) { |
977 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
978 | shrink_dcache_for_umount_subtree(dentry); |
979 | } | |
980 | } | |
981 | ||
c826cb7d LT |
982 | /* |
983 | * This tries to ascend one level of parenthood, but | |
984 | * we can race with renaming, so we need to re-check | |
985 | * the parenthood after dropping the lock and check | |
986 | * that the sequence number still matches. | |
987 | */ | |
988 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
989 | { | |
990 | struct dentry *new = old->d_parent; | |
991 | ||
992 | rcu_read_lock(); | |
993 | spin_unlock(&old->d_lock); | |
994 | spin_lock(&new->d_lock); | |
995 | ||
996 | /* | |
997 | * might go back up the wrong parent if we have had a rename | |
998 | * or deletion | |
999 | */ | |
1000 | if (new != old->d_parent || | |
b161dfa6 | 1001 | (old->d_flags & DCACHE_DENTRY_KILLED) || |
c826cb7d LT |
1002 | (!locked && read_seqretry(&rename_lock, seq))) { |
1003 | spin_unlock(&new->d_lock); | |
1004 | new = NULL; | |
1005 | } | |
1006 | rcu_read_unlock(); | |
1007 | return new; | |
1008 | } | |
1009 | ||
1010 | ||
1da177e4 LT |
1011 | /* |
1012 | * Search for at least 1 mount point in the dentry's subdirs. | |
1013 | * We descend to the next level whenever the d_subdirs | |
1014 | * list is non-empty and continue searching. | |
1015 | */ | |
1016 | ||
1017 | /** | |
1018 | * have_submounts - check for mounts over a dentry | |
1019 | * @parent: dentry to check. | |
1020 | * | |
1021 | * Return true if the parent or its subdirectories contain | |
1022 | * a mount point | |
1023 | */ | |
1da177e4 LT |
1024 | int have_submounts(struct dentry *parent) |
1025 | { | |
949854d0 | 1026 | struct dentry *this_parent; |
1da177e4 | 1027 | struct list_head *next; |
949854d0 | 1028 | unsigned seq; |
58db63d0 | 1029 | int locked = 0; |
949854d0 | 1030 | |
949854d0 | 1031 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1032 | again: |
1033 | this_parent = parent; | |
1da177e4 | 1034 | |
1da177e4 LT |
1035 | if (d_mountpoint(parent)) |
1036 | goto positive; | |
2fd6b7f5 | 1037 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1038 | repeat: |
1039 | next = this_parent->d_subdirs.next; | |
1040 | resume: | |
1041 | while (next != &this_parent->d_subdirs) { | |
1042 | struct list_head *tmp = next; | |
5160ee6f | 1043 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1044 | next = tmp->next; |
2fd6b7f5 NP |
1045 | |
1046 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
1da177e4 | 1047 | /* Have we found a mount point ? */ |
2fd6b7f5 NP |
1048 | if (d_mountpoint(dentry)) { |
1049 | spin_unlock(&dentry->d_lock); | |
1050 | spin_unlock(&this_parent->d_lock); | |
1da177e4 | 1051 | goto positive; |
2fd6b7f5 | 1052 | } |
1da177e4 | 1053 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1054 | spin_unlock(&this_parent->d_lock); |
1055 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1056 | this_parent = dentry; |
2fd6b7f5 | 1057 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1058 | goto repeat; |
1059 | } | |
2fd6b7f5 | 1060 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1061 | } |
1062 | /* | |
1063 | * All done at this level ... ascend and resume the search. | |
1064 | */ | |
1065 | if (this_parent != parent) { | |
c826cb7d LT |
1066 | struct dentry *child = this_parent; |
1067 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1068 | if (!this_parent) | |
949854d0 | 1069 | goto rename_retry; |
949854d0 | 1070 | next = child->d_u.d_child.next; |
1da177e4 LT |
1071 | goto resume; |
1072 | } | |
2fd6b7f5 | 1073 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1074 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1075 | goto rename_retry; |
58db63d0 NP |
1076 | if (locked) |
1077 | write_sequnlock(&rename_lock); | |
1da177e4 LT |
1078 | return 0; /* No mount points found in tree */ |
1079 | positive: | |
58db63d0 | 1080 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1081 | goto rename_retry; |
58db63d0 NP |
1082 | if (locked) |
1083 | write_sequnlock(&rename_lock); | |
1da177e4 | 1084 | return 1; |
58db63d0 NP |
1085 | |
1086 | rename_retry: | |
8110e16d MS |
1087 | if (locked) |
1088 | goto again; | |
58db63d0 NP |
1089 | locked = 1; |
1090 | write_seqlock(&rename_lock); | |
1091 | goto again; | |
1da177e4 | 1092 | } |
ec4f8605 | 1093 | EXPORT_SYMBOL(have_submounts); |
1da177e4 LT |
1094 | |
1095 | /* | |
fd517909 | 1096 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1097 | * and move any unused dentries to the end of the unused |
1098 | * list for prune_dcache(). We descend to the next level | |
1099 | * whenever the d_subdirs list is non-empty and continue | |
1100 | * searching. | |
1101 | * | |
1102 | * It returns zero iff there are no unused children, | |
1103 | * otherwise it returns the number of children moved to | |
1104 | * the end of the unused list. This may not be the total | |
1105 | * number of unused children, because select_parent can | |
1106 | * drop the lock and return early due to latency | |
1107 | * constraints. | |
1108 | */ | |
b48f03b3 | 1109 | static int select_parent(struct dentry *parent, struct list_head *dispose) |
1da177e4 | 1110 | { |
949854d0 | 1111 | struct dentry *this_parent; |
1da177e4 | 1112 | struct list_head *next; |
949854d0 | 1113 | unsigned seq; |
1da177e4 | 1114 | int found = 0; |
58db63d0 | 1115 | int locked = 0; |
1da177e4 | 1116 | |
949854d0 | 1117 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1118 | again: |
1119 | this_parent = parent; | |
2fd6b7f5 | 1120 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
1121 | repeat: |
1122 | next = this_parent->d_subdirs.next; | |
1123 | resume: | |
1124 | while (next != &this_parent->d_subdirs) { | |
1125 | struct list_head *tmp = next; | |
5160ee6f | 1126 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 LT |
1127 | next = tmp->next; |
1128 | ||
2fd6b7f5 | 1129 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
23044507 | 1130 | |
b48f03b3 DC |
1131 | /* |
1132 | * move only zero ref count dentries to the dispose list. | |
eaf5f907 MS |
1133 | * |
1134 | * Those which are presently on the shrink list, being processed | |
1135 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1136 | * loop in shrink_dcache_parent() might not make any progress | |
1137 | * and loop forever. | |
1da177e4 | 1138 | */ |
98474236 | 1139 | if (dentry->d_lockref.count) { |
eaf5f907 MS |
1140 | dentry_lru_del(dentry); |
1141 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
b48f03b3 | 1142 | dentry_lru_move_list(dentry, dispose); |
eaf5f907 | 1143 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
1da177e4 LT |
1144 | found++; |
1145 | } | |
1da177e4 LT |
1146 | /* |
1147 | * We can return to the caller if we have found some (this | |
1148 | * ensures forward progress). We'll be coming back to find | |
1149 | * the rest. | |
1150 | */ | |
2fd6b7f5 NP |
1151 | if (found && need_resched()) { |
1152 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 1153 | goto out; |
2fd6b7f5 | 1154 | } |
1da177e4 LT |
1155 | |
1156 | /* | |
1157 | * Descend a level if the d_subdirs list is non-empty. | |
1158 | */ | |
1159 | if (!list_empty(&dentry->d_subdirs)) { | |
2fd6b7f5 NP |
1160 | spin_unlock(&this_parent->d_lock); |
1161 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1162 | this_parent = dentry; |
2fd6b7f5 | 1163 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1164 | goto repeat; |
1165 | } | |
2fd6b7f5 NP |
1166 | |
1167 | spin_unlock(&dentry->d_lock); | |
1da177e4 LT |
1168 | } |
1169 | /* | |
1170 | * All done at this level ... ascend and resume the search. | |
1171 | */ | |
1172 | if (this_parent != parent) { | |
c826cb7d LT |
1173 | struct dentry *child = this_parent; |
1174 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1175 | if (!this_parent) | |
949854d0 | 1176 | goto rename_retry; |
949854d0 | 1177 | next = child->d_u.d_child.next; |
1da177e4 LT |
1178 | goto resume; |
1179 | } | |
1180 | out: | |
2fd6b7f5 | 1181 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 1182 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 1183 | goto rename_retry; |
58db63d0 NP |
1184 | if (locked) |
1185 | write_sequnlock(&rename_lock); | |
1da177e4 | 1186 | return found; |
58db63d0 NP |
1187 | |
1188 | rename_retry: | |
1189 | if (found) | |
1190 | return found; | |
8110e16d MS |
1191 | if (locked) |
1192 | goto again; | |
58db63d0 NP |
1193 | locked = 1; |
1194 | write_seqlock(&rename_lock); | |
1195 | goto again; | |
1da177e4 LT |
1196 | } |
1197 | ||
1198 | /** | |
1199 | * shrink_dcache_parent - prune dcache | |
1200 | * @parent: parent of entries to prune | |
1201 | * | |
1202 | * Prune the dcache to remove unused children of the parent dentry. | |
1203 | */ | |
1da177e4 LT |
1204 | void shrink_dcache_parent(struct dentry * parent) |
1205 | { | |
b48f03b3 | 1206 | LIST_HEAD(dispose); |
1da177e4 LT |
1207 | int found; |
1208 | ||
421348f1 | 1209 | while ((found = select_parent(parent, &dispose)) != 0) { |
b48f03b3 | 1210 | shrink_dentry_list(&dispose); |
421348f1 GT |
1211 | cond_resched(); |
1212 | } | |
1da177e4 | 1213 | } |
ec4f8605 | 1214 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1215 | |
1da177e4 | 1216 | /** |
a4464dbc AV |
1217 | * __d_alloc - allocate a dcache entry |
1218 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1219 | * @name: qstr of the name |
1220 | * | |
1221 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1222 | * available. On a success the dentry is returned. The name passed in is | |
1223 | * copied and the copy passed in may be reused after this call. | |
1224 | */ | |
1225 | ||
a4464dbc | 1226 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1227 | { |
1228 | struct dentry *dentry; | |
1229 | char *dname; | |
1230 | ||
e12ba74d | 1231 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1232 | if (!dentry) |
1233 | return NULL; | |
1234 | ||
6326c71f LT |
1235 | /* |
1236 | * We guarantee that the inline name is always NUL-terminated. | |
1237 | * This way the memcpy() done by the name switching in rename | |
1238 | * will still always have a NUL at the end, even if we might | |
1239 | * be overwriting an internal NUL character | |
1240 | */ | |
1241 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1242 | if (name->len > DNAME_INLINE_LEN-1) { |
1243 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1244 | if (!dname) { | |
1245 | kmem_cache_free(dentry_cache, dentry); | |
1246 | return NULL; | |
1247 | } | |
1248 | } else { | |
1249 | dname = dentry->d_iname; | |
1250 | } | |
1da177e4 LT |
1251 | |
1252 | dentry->d_name.len = name->len; | |
1253 | dentry->d_name.hash = name->hash; | |
1254 | memcpy(dname, name->name, name->len); | |
1255 | dname[name->len] = 0; | |
1256 | ||
6326c71f LT |
1257 | /* Make sure we always see the terminating NUL character */ |
1258 | smp_wmb(); | |
1259 | dentry->d_name.name = dname; | |
1260 | ||
98474236 | 1261 | dentry->d_lockref.count = 1; |
dea3667b | 1262 | dentry->d_flags = 0; |
1da177e4 | 1263 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1264 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1265 | dentry->d_inode = NULL; |
a4464dbc AV |
1266 | dentry->d_parent = dentry; |
1267 | dentry->d_sb = sb; | |
1da177e4 LT |
1268 | dentry->d_op = NULL; |
1269 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1270 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1271 | INIT_LIST_HEAD(&dentry->d_lru); |
1272 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1273 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1274 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1275 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1276 | |
3e880fb5 | 1277 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1278 | |
1da177e4 LT |
1279 | return dentry; |
1280 | } | |
a4464dbc AV |
1281 | |
1282 | /** | |
1283 | * d_alloc - allocate a dcache entry | |
1284 | * @parent: parent of entry to allocate | |
1285 | * @name: qstr of the name | |
1286 | * | |
1287 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1288 | * available. On a success the dentry is returned. The name passed in is | |
1289 | * copied and the copy passed in may be reused after this call. | |
1290 | */ | |
1291 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1292 | { | |
1293 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1294 | if (!dentry) | |
1295 | return NULL; | |
1296 | ||
1297 | spin_lock(&parent->d_lock); | |
1298 | /* | |
1299 | * don't need child lock because it is not subject | |
1300 | * to concurrency here | |
1301 | */ | |
1302 | __dget_dlock(parent); | |
1303 | dentry->d_parent = parent; | |
1304 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1305 | spin_unlock(&parent->d_lock); | |
1306 | ||
1307 | return dentry; | |
1308 | } | |
ec4f8605 | 1309 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1310 | |
4b936885 NP |
1311 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1312 | { | |
a4464dbc AV |
1313 | struct dentry *dentry = __d_alloc(sb, name); |
1314 | if (dentry) | |
4b936885 | 1315 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1316 | return dentry; |
1317 | } | |
1318 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1319 | ||
1da177e4 LT |
1320 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1321 | { | |
1322 | struct qstr q; | |
1323 | ||
1324 | q.name = name; | |
1325 | q.len = strlen(name); | |
1326 | q.hash = full_name_hash(q.name, q.len); | |
1327 | return d_alloc(parent, &q); | |
1328 | } | |
ef26ca97 | 1329 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1330 | |
fb045adb NP |
1331 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1332 | { | |
6f7f7caa LT |
1333 | WARN_ON_ONCE(dentry->d_op); |
1334 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1335 | DCACHE_OP_COMPARE | |
1336 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1337 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1338 | DCACHE_OP_DELETE )); |
1339 | dentry->d_op = op; | |
1340 | if (!op) | |
1341 | return; | |
1342 | if (op->d_hash) | |
1343 | dentry->d_flags |= DCACHE_OP_HASH; | |
1344 | if (op->d_compare) | |
1345 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1346 | if (op->d_revalidate) | |
1347 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1348 | if (op->d_weak_revalidate) |
1349 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1350 | if (op->d_delete) |
1351 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1352 | if (op->d_prune) |
1353 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1354 | |
1355 | } | |
1356 | EXPORT_SYMBOL(d_set_d_op); | |
1357 | ||
360da900 OH |
1358 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1359 | { | |
b23fb0a6 | 1360 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1361 | if (inode) { |
1362 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1363 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
b3d9b7a3 | 1364 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1365 | } |
360da900 | 1366 | dentry->d_inode = inode; |
31e6b01f | 1367 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1368 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1369 | fsnotify_d_instantiate(dentry, inode); |
1370 | } | |
1371 | ||
1da177e4 LT |
1372 | /** |
1373 | * d_instantiate - fill in inode information for a dentry | |
1374 | * @entry: dentry to complete | |
1375 | * @inode: inode to attach to this dentry | |
1376 | * | |
1377 | * Fill in inode information in the entry. | |
1378 | * | |
1379 | * This turns negative dentries into productive full members | |
1380 | * of society. | |
1381 | * | |
1382 | * NOTE! This assumes that the inode count has been incremented | |
1383 | * (or otherwise set) by the caller to indicate that it is now | |
1384 | * in use by the dcache. | |
1385 | */ | |
1386 | ||
1387 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1388 | { | |
b3d9b7a3 | 1389 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1390 | if (inode) |
1391 | spin_lock(&inode->i_lock); | |
360da900 | 1392 | __d_instantiate(entry, inode); |
873feea0 NP |
1393 | if (inode) |
1394 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1395 | security_d_instantiate(entry, inode); |
1396 | } | |
ec4f8605 | 1397 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1398 | |
1399 | /** | |
1400 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1401 | * @entry: dentry to instantiate | |
1402 | * @inode: inode to attach to this dentry | |
1403 | * | |
1404 | * Fill in inode information in the entry. On success, it returns NULL. | |
1405 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1406 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1407 | * |
1408 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1409 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1410 | * |
1411 | * This also assumes that the inode count has been incremented | |
1412 | * (or otherwise set) by the caller to indicate that it is now | |
1413 | * in use by the dcache. | |
1da177e4 | 1414 | */ |
770bfad8 DH |
1415 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1416 | struct inode *inode) | |
1da177e4 LT |
1417 | { |
1418 | struct dentry *alias; | |
1419 | int len = entry->d_name.len; | |
1420 | const char *name = entry->d_name.name; | |
1421 | unsigned int hash = entry->d_name.hash; | |
1422 | ||
770bfad8 | 1423 | if (!inode) { |
360da900 | 1424 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1425 | return NULL; |
1426 | } | |
1427 | ||
b67bfe0d | 1428 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1429 | /* |
1430 | * Don't need alias->d_lock here, because aliases with | |
1431 | * d_parent == entry->d_parent are not subject to name or | |
1432 | * parent changes, because the parent inode i_mutex is held. | |
1433 | */ | |
12f8ad4b | 1434 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1435 | continue; |
1436 | if (alias->d_parent != entry->d_parent) | |
1437 | continue; | |
ee983e89 LT |
1438 | if (alias->d_name.len != len) |
1439 | continue; | |
12f8ad4b | 1440 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1441 | continue; |
dc0474be | 1442 | __dget(alias); |
1da177e4 LT |
1443 | return alias; |
1444 | } | |
770bfad8 | 1445 | |
360da900 | 1446 | __d_instantiate(entry, inode); |
1da177e4 LT |
1447 | return NULL; |
1448 | } | |
770bfad8 DH |
1449 | |
1450 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1451 | { | |
1452 | struct dentry *result; | |
1453 | ||
b3d9b7a3 | 1454 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1455 | |
873feea0 NP |
1456 | if (inode) |
1457 | spin_lock(&inode->i_lock); | |
770bfad8 | 1458 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1459 | if (inode) |
1460 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1461 | |
1462 | if (!result) { | |
1463 | security_d_instantiate(entry, inode); | |
1464 | return NULL; | |
1465 | } | |
1466 | ||
1467 | BUG_ON(!d_unhashed(result)); | |
1468 | iput(inode); | |
1469 | return result; | |
1470 | } | |
1471 | ||
1da177e4 LT |
1472 | EXPORT_SYMBOL(d_instantiate_unique); |
1473 | ||
adc0e91a AV |
1474 | struct dentry *d_make_root(struct inode *root_inode) |
1475 | { | |
1476 | struct dentry *res = NULL; | |
1477 | ||
1478 | if (root_inode) { | |
26fe5750 | 1479 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1480 | |
1481 | res = __d_alloc(root_inode->i_sb, &name); | |
1482 | if (res) | |
1483 | d_instantiate(res, root_inode); | |
1484 | else | |
1485 | iput(root_inode); | |
1486 | } | |
1487 | return res; | |
1488 | } | |
1489 | EXPORT_SYMBOL(d_make_root); | |
1490 | ||
d891eedb BF |
1491 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1492 | { | |
1493 | struct dentry *alias; | |
1494 | ||
b3d9b7a3 | 1495 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1496 | return NULL; |
b3d9b7a3 | 1497 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1498 | __dget(alias); |
1499 | return alias; | |
1500 | } | |
1501 | ||
46f72b34 SW |
1502 | /** |
1503 | * d_find_any_alias - find any alias for a given inode | |
1504 | * @inode: inode to find an alias for | |
1505 | * | |
1506 | * If any aliases exist for the given inode, take and return a | |
1507 | * reference for one of them. If no aliases exist, return %NULL. | |
1508 | */ | |
1509 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1510 | { |
1511 | struct dentry *de; | |
1512 | ||
1513 | spin_lock(&inode->i_lock); | |
1514 | de = __d_find_any_alias(inode); | |
1515 | spin_unlock(&inode->i_lock); | |
1516 | return de; | |
1517 | } | |
46f72b34 | 1518 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1519 | |
4ea3ada2 CH |
1520 | /** |
1521 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1522 | * @inode: inode to allocate the dentry for | |
1523 | * | |
1524 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1525 | * similar open by handle operations. The returned dentry may be anonymous, | |
1526 | * or may have a full name (if the inode was already in the cache). | |
1527 | * | |
1528 | * When called on a directory inode, we must ensure that the inode only ever | |
1529 | * has one dentry. If a dentry is found, that is returned instead of | |
1530 | * allocating a new one. | |
1531 | * | |
1532 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1533 | * to the dentry. In case of an error the reference on the inode is released. |
1534 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1535 | * be passed in and will be the error will be propagate to the return value, | |
1536 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1537 | */ |
1538 | struct dentry *d_obtain_alias(struct inode *inode) | |
1539 | { | |
b911a6bd | 1540 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1541 | struct dentry *tmp; |
1542 | struct dentry *res; | |
4ea3ada2 CH |
1543 | |
1544 | if (!inode) | |
44003728 | 1545 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1546 | if (IS_ERR(inode)) |
1547 | return ERR_CAST(inode); | |
1548 | ||
d891eedb | 1549 | res = d_find_any_alias(inode); |
9308a612 CH |
1550 | if (res) |
1551 | goto out_iput; | |
1552 | ||
a4464dbc | 1553 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1554 | if (!tmp) { |
1555 | res = ERR_PTR(-ENOMEM); | |
1556 | goto out_iput; | |
4ea3ada2 | 1557 | } |
b5c84bf6 | 1558 | |
873feea0 | 1559 | spin_lock(&inode->i_lock); |
d891eedb | 1560 | res = __d_find_any_alias(inode); |
9308a612 | 1561 | if (res) { |
873feea0 | 1562 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1563 | dput(tmp); |
1564 | goto out_iput; | |
1565 | } | |
1566 | ||
1567 | /* attach a disconnected dentry */ | |
1568 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1569 | tmp->d_inode = inode; |
1570 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
b3d9b7a3 | 1571 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1572 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1573 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1574 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1575 | spin_unlock(&tmp->d_lock); |
873feea0 | 1576 | spin_unlock(&inode->i_lock); |
24ff6663 | 1577 | security_d_instantiate(tmp, inode); |
9308a612 | 1578 | |
9308a612 CH |
1579 | return tmp; |
1580 | ||
1581 | out_iput: | |
24ff6663 JB |
1582 | if (res && !IS_ERR(res)) |
1583 | security_d_instantiate(res, inode); | |
9308a612 CH |
1584 | iput(inode); |
1585 | return res; | |
4ea3ada2 | 1586 | } |
adc48720 | 1587 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1588 | |
1589 | /** | |
1590 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1591 | * @inode: the inode which may have a disconnected dentry | |
1592 | * @dentry: a negative dentry which we want to point to the inode. | |
1593 | * | |
1594 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1595 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1596 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1597 | * | |
1598 | * This is needed in the lookup routine of any filesystem that is exportable | |
1599 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1600 | * | |
1601 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1602 | * is returned. This matches the expected return value of ->lookup. | |
1603 | * | |
6d4ade98 SW |
1604 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1605 | * In that case, we know that the inode will be a regular file, and also this | |
1606 | * will only occur during atomic_open. So we need to check for the dentry | |
1607 | * being already hashed only in the final case. | |
1da177e4 LT |
1608 | */ |
1609 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1610 | { | |
1611 | struct dentry *new = NULL; | |
1612 | ||
a9049376 AV |
1613 | if (IS_ERR(inode)) |
1614 | return ERR_CAST(inode); | |
1615 | ||
21c0d8fd | 1616 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1617 | spin_lock(&inode->i_lock); |
32ba9c3f | 1618 | new = __d_find_alias(inode, 1); |
1da177e4 | 1619 | if (new) { |
32ba9c3f | 1620 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1621 | spin_unlock(&inode->i_lock); |
1da177e4 | 1622 | security_d_instantiate(new, inode); |
1da177e4 LT |
1623 | d_move(new, dentry); |
1624 | iput(inode); | |
1625 | } else { | |
873feea0 | 1626 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1627 | __d_instantiate(dentry, inode); |
873feea0 | 1628 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1629 | security_d_instantiate(dentry, inode); |
1630 | d_rehash(dentry); | |
1631 | } | |
6d4ade98 SW |
1632 | } else { |
1633 | d_instantiate(dentry, inode); | |
1634 | if (d_unhashed(dentry)) | |
1635 | d_rehash(dentry); | |
1636 | } | |
1da177e4 LT |
1637 | return new; |
1638 | } | |
ec4f8605 | 1639 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1640 | |
9403540c BN |
1641 | /** |
1642 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1643 | * @inode: the inode case-insensitive lookup has found | |
1644 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1645 | * @name: the case-exact name to be associated with the returned dentry | |
1646 | * | |
1647 | * This is to avoid filling the dcache with case-insensitive names to the | |
1648 | * same inode, only the actual correct case is stored in the dcache for | |
1649 | * case-insensitive filesystems. | |
1650 | * | |
1651 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1652 | * already exists in in the dcache, use it and return it. | |
1653 | * | |
1654 | * If no entry exists with the exact case name, allocate new dentry with | |
1655 | * the exact case, and return the spliced entry. | |
1656 | */ | |
e45b590b | 1657 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1658 | struct qstr *name) |
1659 | { | |
9403540c BN |
1660 | struct dentry *found; |
1661 | struct dentry *new; | |
1662 | ||
b6520c81 CH |
1663 | /* |
1664 | * First check if a dentry matching the name already exists, | |
1665 | * if not go ahead and create it now. | |
1666 | */ | |
9403540c | 1667 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1668 | if (unlikely(IS_ERR(found))) |
1669 | goto err_out; | |
9403540c BN |
1670 | if (!found) { |
1671 | new = d_alloc(dentry->d_parent, name); | |
1672 | if (!new) { | |
4f522a24 | 1673 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1674 | goto err_out; |
1675 | } | |
b6520c81 | 1676 | |
9403540c BN |
1677 | found = d_splice_alias(inode, new); |
1678 | if (found) { | |
1679 | dput(new); | |
1680 | return found; | |
1681 | } | |
1682 | return new; | |
1683 | } | |
b6520c81 CH |
1684 | |
1685 | /* | |
1686 | * If a matching dentry exists, and it's not negative use it. | |
1687 | * | |
1688 | * Decrement the reference count to balance the iget() done | |
1689 | * earlier on. | |
1690 | */ | |
9403540c BN |
1691 | if (found->d_inode) { |
1692 | if (unlikely(found->d_inode != inode)) { | |
1693 | /* This can't happen because bad inodes are unhashed. */ | |
1694 | BUG_ON(!is_bad_inode(inode)); | |
1695 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1696 | } | |
9403540c BN |
1697 | iput(inode); |
1698 | return found; | |
1699 | } | |
b6520c81 | 1700 | |
9403540c | 1701 | /* |
9403540c | 1702 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1703 | * already has a dentry. |
9403540c | 1704 | */ |
4513d899 AV |
1705 | new = d_splice_alias(inode, found); |
1706 | if (new) { | |
1707 | dput(found); | |
1708 | found = new; | |
9403540c | 1709 | } |
4513d899 | 1710 | return found; |
9403540c BN |
1711 | |
1712 | err_out: | |
1713 | iput(inode); | |
4f522a24 | 1714 | return found; |
9403540c | 1715 | } |
ec4f8605 | 1716 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1717 | |
12f8ad4b LT |
1718 | /* |
1719 | * Do the slow-case of the dentry name compare. | |
1720 | * | |
1721 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1722 | * load the name and length information, so that the |
12f8ad4b LT |
1723 | * filesystem can rely on them, and can use the 'name' and |
1724 | * 'len' information without worrying about walking off the | |
1725 | * end of memory etc. | |
1726 | * | |
1727 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1728 | * in arguments (the low-level filesystem should not look | |
1729 | * at the dentry inode or name contents directly, since | |
1730 | * rename can change them while we're in RCU mode). | |
1731 | */ | |
1732 | enum slow_d_compare { | |
1733 | D_COMP_OK, | |
1734 | D_COMP_NOMATCH, | |
1735 | D_COMP_SEQRETRY, | |
1736 | }; | |
1737 | ||
1738 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1739 | const struct dentry *parent, | |
12f8ad4b LT |
1740 | struct dentry *dentry, |
1741 | unsigned int seq, | |
1742 | const struct qstr *name) | |
1743 | { | |
1744 | int tlen = dentry->d_name.len; | |
1745 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
1746 | |
1747 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1748 | cpu_relax(); | |
1749 | return D_COMP_SEQRETRY; | |
1750 | } | |
da53be12 | 1751 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
1752 | return D_COMP_NOMATCH; |
1753 | return D_COMP_OK; | |
1754 | } | |
1755 | ||
31e6b01f NP |
1756 | /** |
1757 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1758 | * @parent: parent dentry | |
1759 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1760 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1761 | * Returns: dentry, or NULL |
1762 | * | |
1763 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1764 | * resolution (store-free path walking) design described in | |
1765 | * Documentation/filesystems/path-lookup.txt. | |
1766 | * | |
1767 | * This is not to be used outside core vfs. | |
1768 | * | |
1769 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1770 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1771 | * without taking d_lock and checking d_seq sequence count against @seq | |
1772 | * returned here. | |
1773 | * | |
1774 | * A refcount may be taken on the found dentry with the __d_rcu_to_refcount | |
1775 | * function. | |
1776 | * | |
1777 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1778 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1779 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1780 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
1781 | * |
1782 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
1783 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 1784 | */ |
8966be90 LT |
1785 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
1786 | const struct qstr *name, | |
da53be12 | 1787 | unsigned *seqp) |
31e6b01f | 1788 | { |
26fe5750 | 1789 | u64 hashlen = name->hash_len; |
31e6b01f | 1790 | const unsigned char *str = name->name; |
26fe5750 | 1791 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 1792 | struct hlist_bl_node *node; |
31e6b01f NP |
1793 | struct dentry *dentry; |
1794 | ||
1795 | /* | |
1796 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1797 | * required to prevent single threaded performance regressions | |
1798 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1799 | * Keep the two functions in sync. | |
1800 | */ | |
1801 | ||
1802 | /* | |
1803 | * The hash list is protected using RCU. | |
1804 | * | |
1805 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1806 | * races with d_move(). | |
1807 | * | |
1808 | * It is possible that concurrent renames can mess up our list | |
1809 | * walk here and result in missing our dentry, resulting in the | |
1810 | * false-negative result. d_lookup() protects against concurrent | |
1811 | * renames using rename_lock seqlock. | |
1812 | * | |
b0a4bb83 | 1813 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1814 | */ |
b07ad996 | 1815 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 1816 | unsigned seq; |
31e6b01f | 1817 | |
31e6b01f | 1818 | seqretry: |
12f8ad4b LT |
1819 | /* |
1820 | * The dentry sequence count protects us from concurrent | |
da53be12 | 1821 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
1822 | * |
1823 | * The caller must perform a seqcount check in order | |
da53be12 | 1824 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
1825 | * |
1826 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
1827 | * we don't wait for the sequence count to stabilize if it | |
1828 | * is in the middle of a sequence change. If we do the slow | |
1829 | * dentry compare, we will do seqretries until it is stable, | |
1830 | * and if we end up with a successful lookup, we actually | |
1831 | * want to exit RCU lookup anyway. | |
1832 | */ | |
1833 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
1834 | if (dentry->d_parent != parent) |
1835 | continue; | |
2e321806 LT |
1836 | if (d_unhashed(dentry)) |
1837 | continue; | |
12f8ad4b | 1838 | |
830c0f0e | 1839 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
1840 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
1841 | continue; | |
da53be12 LT |
1842 | *seqp = seq; |
1843 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
1844 | case D_COMP_OK: |
1845 | return dentry; | |
1846 | case D_COMP_NOMATCH: | |
31e6b01f | 1847 | continue; |
12f8ad4b LT |
1848 | default: |
1849 | goto seqretry; | |
1850 | } | |
31e6b01f | 1851 | } |
12f8ad4b | 1852 | |
26fe5750 | 1853 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 1854 | continue; |
da53be12 | 1855 | *seqp = seq; |
26fe5750 | 1856 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 1857 | return dentry; |
31e6b01f NP |
1858 | } |
1859 | return NULL; | |
1860 | } | |
1861 | ||
1da177e4 LT |
1862 | /** |
1863 | * d_lookup - search for a dentry | |
1864 | * @parent: parent dentry | |
1865 | * @name: qstr of name we wish to find | |
b04f784e | 1866 | * Returns: dentry, or NULL |
1da177e4 | 1867 | * |
b04f784e NP |
1868 | * d_lookup searches the children of the parent dentry for the name in |
1869 | * question. If the dentry is found its reference count is incremented and the | |
1870 | * dentry is returned. The caller must use dput to free the entry when it has | |
1871 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 1872 | */ |
da2d8455 | 1873 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 1874 | { |
31e6b01f | 1875 | struct dentry *dentry; |
949854d0 | 1876 | unsigned seq; |
1da177e4 LT |
1877 | |
1878 | do { | |
1879 | seq = read_seqbegin(&rename_lock); | |
1880 | dentry = __d_lookup(parent, name); | |
1881 | if (dentry) | |
1882 | break; | |
1883 | } while (read_seqretry(&rename_lock, seq)); | |
1884 | return dentry; | |
1885 | } | |
ec4f8605 | 1886 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 1887 | |
31e6b01f | 1888 | /** |
b04f784e NP |
1889 | * __d_lookup - search for a dentry (racy) |
1890 | * @parent: parent dentry | |
1891 | * @name: qstr of name we wish to find | |
1892 | * Returns: dentry, or NULL | |
1893 | * | |
1894 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
1895 | * false-negative result due to unrelated rename activity. | |
1896 | * | |
1897 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
1898 | * however it must be used carefully, eg. with a following d_lookup in | |
1899 | * the case of failure. | |
1900 | * | |
1901 | * __d_lookup callers must be commented. | |
1902 | */ | |
a713ca2a | 1903 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
1904 | { |
1905 | unsigned int len = name->len; | |
1906 | unsigned int hash = name->hash; | |
1907 | const unsigned char *str = name->name; | |
b07ad996 | 1908 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 1909 | struct hlist_bl_node *node; |
31e6b01f | 1910 | struct dentry *found = NULL; |
665a7583 | 1911 | struct dentry *dentry; |
1da177e4 | 1912 | |
31e6b01f NP |
1913 | /* |
1914 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1915 | * required to prevent single threaded performance regressions | |
1916 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1917 | * Keep the two functions in sync. | |
1918 | */ | |
1919 | ||
b04f784e NP |
1920 | /* |
1921 | * The hash list is protected using RCU. | |
1922 | * | |
1923 | * Take d_lock when comparing a candidate dentry, to avoid races | |
1924 | * with d_move(). | |
1925 | * | |
1926 | * It is possible that concurrent renames can mess up our list | |
1927 | * walk here and result in missing our dentry, resulting in the | |
1928 | * false-negative result. d_lookup() protects against concurrent | |
1929 | * renames using rename_lock seqlock. | |
1930 | * | |
b0a4bb83 | 1931 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 1932 | */ |
1da177e4 LT |
1933 | rcu_read_lock(); |
1934 | ||
b07ad996 | 1935 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 1936 | |
1da177e4 LT |
1937 | if (dentry->d_name.hash != hash) |
1938 | continue; | |
1da177e4 LT |
1939 | |
1940 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
1941 | if (dentry->d_parent != parent) |
1942 | goto next; | |
d0185c08 LT |
1943 | if (d_unhashed(dentry)) |
1944 | goto next; | |
1945 | ||
1da177e4 LT |
1946 | /* |
1947 | * It is safe to compare names since d_move() cannot | |
1948 | * change the qstr (protected by d_lock). | |
1949 | */ | |
fb045adb | 1950 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
1951 | int tlen = dentry->d_name.len; |
1952 | const char *tname = dentry->d_name.name; | |
da53be12 | 1953 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
1954 | goto next; |
1955 | } else { | |
ee983e89 LT |
1956 | if (dentry->d_name.len != len) |
1957 | goto next; | |
12f8ad4b | 1958 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
1959 | goto next; |
1960 | } | |
1961 | ||
98474236 | 1962 | dentry->d_lockref.count++; |
d0185c08 | 1963 | found = dentry; |
1da177e4 LT |
1964 | spin_unlock(&dentry->d_lock); |
1965 | break; | |
1966 | next: | |
1967 | spin_unlock(&dentry->d_lock); | |
1968 | } | |
1969 | rcu_read_unlock(); | |
1970 | ||
1971 | return found; | |
1972 | } | |
1973 | ||
3e7e241f EB |
1974 | /** |
1975 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
1976 | * @dir: Directory to search in | |
1977 | * @name: qstr of name we wish to find | |
1978 | * | |
4f522a24 | 1979 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
1980 | */ |
1981 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
1982 | { | |
3e7e241f EB |
1983 | /* |
1984 | * Check for a fs-specific hash function. Note that we must | |
1985 | * calculate the standard hash first, as the d_op->d_hash() | |
1986 | * routine may choose to leave the hash value unchanged. | |
1987 | */ | |
1988 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 1989 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 1990 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
1991 | if (unlikely(err < 0)) |
1992 | return ERR_PTR(err); | |
3e7e241f | 1993 | } |
4f522a24 | 1994 | return d_lookup(dir, name); |
3e7e241f | 1995 | } |
4f522a24 | 1996 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 1997 | |
1da177e4 | 1998 | /** |
786a5e15 | 1999 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2000 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2001 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2002 | * |
2003 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2004 | * This is used by ncpfs in its readdir implementation. | |
2005 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2006 | * |
2007 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2008 | */ |
d3a23e16 | 2009 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2010 | { |
786a5e15 | 2011 | struct dentry *child; |
d3a23e16 | 2012 | |
2fd6b7f5 | 2013 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2014 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2015 | if (dentry == child) { | |
2fd6b7f5 | 2016 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2017 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2018 | spin_unlock(&dentry->d_lock); |
2019 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2020 | return 1; |
2021 | } | |
2022 | } | |
2fd6b7f5 | 2023 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2024 | |
1da177e4 LT |
2025 | return 0; |
2026 | } | |
ec4f8605 | 2027 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2028 | |
2029 | /* | |
2030 | * When a file is deleted, we have two options: | |
2031 | * - turn this dentry into a negative dentry | |
2032 | * - unhash this dentry and free it. | |
2033 | * | |
2034 | * Usually, we want to just turn this into | |
2035 | * a negative dentry, but if anybody else is | |
2036 | * currently using the dentry or the inode | |
2037 | * we can't do that and we fall back on removing | |
2038 | * it from the hash queues and waiting for | |
2039 | * it to be deleted later when it has no users | |
2040 | */ | |
2041 | ||
2042 | /** | |
2043 | * d_delete - delete a dentry | |
2044 | * @dentry: The dentry to delete | |
2045 | * | |
2046 | * Turn the dentry into a negative dentry if possible, otherwise | |
2047 | * remove it from the hash queues so it can be deleted later | |
2048 | */ | |
2049 | ||
2050 | void d_delete(struct dentry * dentry) | |
2051 | { | |
873feea0 | 2052 | struct inode *inode; |
7a91bf7f | 2053 | int isdir = 0; |
1da177e4 LT |
2054 | /* |
2055 | * Are we the only user? | |
2056 | */ | |
357f8e65 | 2057 | again: |
1da177e4 | 2058 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2059 | inode = dentry->d_inode; |
2060 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2061 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2062 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2063 | spin_unlock(&dentry->d_lock); |
2064 | cpu_relax(); | |
2065 | goto again; | |
2066 | } | |
13e3c5e5 | 2067 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2068 | dentry_unlink_inode(dentry); |
7a91bf7f | 2069 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2070 | return; |
2071 | } | |
2072 | ||
2073 | if (!d_unhashed(dentry)) | |
2074 | __d_drop(dentry); | |
2075 | ||
2076 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2077 | |
2078 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2079 | } |
ec4f8605 | 2080 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2081 | |
b07ad996 | 2082 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2083 | { |
ceb5bdc2 | 2084 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2085 | hlist_bl_lock(b); |
dea3667b | 2086 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2087 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2088 | hlist_bl_unlock(b); |
1da177e4 LT |
2089 | } |
2090 | ||
770bfad8 DH |
2091 | static void _d_rehash(struct dentry * entry) |
2092 | { | |
2093 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2094 | } | |
2095 | ||
1da177e4 LT |
2096 | /** |
2097 | * d_rehash - add an entry back to the hash | |
2098 | * @entry: dentry to add to the hash | |
2099 | * | |
2100 | * Adds a dentry to the hash according to its name. | |
2101 | */ | |
2102 | ||
2103 | void d_rehash(struct dentry * entry) | |
2104 | { | |
1da177e4 | 2105 | spin_lock(&entry->d_lock); |
770bfad8 | 2106 | _d_rehash(entry); |
1da177e4 | 2107 | spin_unlock(&entry->d_lock); |
1da177e4 | 2108 | } |
ec4f8605 | 2109 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2110 | |
fb2d5b86 NP |
2111 | /** |
2112 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2113 | * @dentry: dentry to be updated | |
2114 | * @name: new name | |
2115 | * | |
2116 | * Update a case insensitive dentry with new case of name. | |
2117 | * | |
2118 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2119 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2120 | * lengths). | |
2121 | * | |
2122 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2123 | * keep renames and concurrent inserts, and readdir(2) away). | |
2124 | */ | |
2125 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2126 | { | |
7ebfa57f | 2127 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2128 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2129 | ||
fb2d5b86 | 2130 | spin_lock(&dentry->d_lock); |
31e6b01f | 2131 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2132 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2133 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2134 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2135 | } |
2136 | EXPORT_SYMBOL(dentry_update_name_case); | |
2137 | ||
1da177e4 LT |
2138 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2139 | { | |
2140 | if (dname_external(target)) { | |
2141 | if (dname_external(dentry)) { | |
2142 | /* | |
2143 | * Both external: swap the pointers | |
2144 | */ | |
9a8d5bb4 | 2145 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2146 | } else { |
2147 | /* | |
2148 | * dentry:internal, target:external. Steal target's | |
2149 | * storage and make target internal. | |
2150 | */ | |
321bcf92 BF |
2151 | memcpy(target->d_iname, dentry->d_name.name, |
2152 | dentry->d_name.len + 1); | |
1da177e4 LT |
2153 | dentry->d_name.name = target->d_name.name; |
2154 | target->d_name.name = target->d_iname; | |
2155 | } | |
2156 | } else { | |
2157 | if (dname_external(dentry)) { | |
2158 | /* | |
2159 | * dentry:external, target:internal. Give dentry's | |
2160 | * storage to target and make dentry internal | |
2161 | */ | |
2162 | memcpy(dentry->d_iname, target->d_name.name, | |
2163 | target->d_name.len + 1); | |
2164 | target->d_name.name = dentry->d_name.name; | |
2165 | dentry->d_name.name = dentry->d_iname; | |
2166 | } else { | |
2167 | /* | |
2168 | * Both are internal. Just copy target to dentry | |
2169 | */ | |
2170 | memcpy(dentry->d_iname, target->d_name.name, | |
2171 | target->d_name.len + 1); | |
dc711ca3 AV |
2172 | dentry->d_name.len = target->d_name.len; |
2173 | return; | |
1da177e4 LT |
2174 | } |
2175 | } | |
9a8d5bb4 | 2176 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2177 | } |
2178 | ||
2fd6b7f5 NP |
2179 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2180 | { | |
2181 | /* | |
2182 | * XXXX: do we really need to take target->d_lock? | |
2183 | */ | |
2184 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2185 | spin_lock(&target->d_parent->d_lock); | |
2186 | else { | |
2187 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2188 | spin_lock(&dentry->d_parent->d_lock); | |
2189 | spin_lock_nested(&target->d_parent->d_lock, | |
2190 | DENTRY_D_LOCK_NESTED); | |
2191 | } else { | |
2192 | spin_lock(&target->d_parent->d_lock); | |
2193 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2194 | DENTRY_D_LOCK_NESTED); | |
2195 | } | |
2196 | } | |
2197 | if (target < dentry) { | |
2198 | spin_lock_nested(&target->d_lock, 2); | |
2199 | spin_lock_nested(&dentry->d_lock, 3); | |
2200 | } else { | |
2201 | spin_lock_nested(&dentry->d_lock, 2); | |
2202 | spin_lock_nested(&target->d_lock, 3); | |
2203 | } | |
2204 | } | |
2205 | ||
2206 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2207 | struct dentry *target) | |
2208 | { | |
2209 | if (target->d_parent != dentry->d_parent) | |
2210 | spin_unlock(&dentry->d_parent->d_lock); | |
2211 | if (target->d_parent != target) | |
2212 | spin_unlock(&target->d_parent->d_lock); | |
2213 | } | |
2214 | ||
1da177e4 | 2215 | /* |
2fd6b7f5 NP |
2216 | * When switching names, the actual string doesn't strictly have to |
2217 | * be preserved in the target - because we're dropping the target | |
2218 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2219 | * the new name before we switch. | |
2220 | * | |
2221 | * Note that we have to be a lot more careful about getting the hash | |
2222 | * switched - we have to switch the hash value properly even if it | |
2223 | * then no longer matches the actual (corrupted) string of the target. | |
2224 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2225 | */ |
9eaef27b | 2226 | /* |
18367501 | 2227 | * __d_move - move a dentry |
1da177e4 LT |
2228 | * @dentry: entry to move |
2229 | * @target: new dentry | |
2230 | * | |
2231 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2232 | * dcache entries should not be moved in this way. Caller must hold |
2233 | * rename_lock, the i_mutex of the source and target directories, | |
2234 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2235 | */ |
18367501 | 2236 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2237 | { |
1da177e4 LT |
2238 | if (!dentry->d_inode) |
2239 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2240 | ||
2fd6b7f5 NP |
2241 | BUG_ON(d_ancestor(dentry, target)); |
2242 | BUG_ON(d_ancestor(target, dentry)); | |
2243 | ||
2fd6b7f5 | 2244 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2245 | |
31e6b01f NP |
2246 | write_seqcount_begin(&dentry->d_seq); |
2247 | write_seqcount_begin(&target->d_seq); | |
2248 | ||
ceb5bdc2 NP |
2249 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2250 | ||
2251 | /* | |
2252 | * Move the dentry to the target hash queue. Don't bother checking | |
2253 | * for the same hash queue because of how unlikely it is. | |
2254 | */ | |
2255 | __d_drop(dentry); | |
789680d1 | 2256 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2257 | |
2258 | /* Unhash the target: dput() will then get rid of it */ | |
2259 | __d_drop(target); | |
2260 | ||
5160ee6f ED |
2261 | list_del(&dentry->d_u.d_child); |
2262 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2263 | |
2264 | /* Switch the names.. */ | |
2265 | switch_names(dentry, target); | |
9a8d5bb4 | 2266 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2267 | |
2268 | /* ... and switch the parents */ | |
2269 | if (IS_ROOT(dentry)) { | |
2270 | dentry->d_parent = target->d_parent; | |
2271 | target->d_parent = target; | |
5160ee6f | 2272 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2273 | } else { |
9a8d5bb4 | 2274 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2275 | |
2276 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2277 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2278 | } |
2279 | ||
5160ee6f | 2280 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2281 | |
31e6b01f NP |
2282 | write_seqcount_end(&target->d_seq); |
2283 | write_seqcount_end(&dentry->d_seq); | |
2284 | ||
2fd6b7f5 | 2285 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2286 | spin_unlock(&target->d_lock); |
c32ccd87 | 2287 | fsnotify_d_move(dentry); |
1da177e4 | 2288 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2289 | } |
2290 | ||
2291 | /* | |
2292 | * d_move - move a dentry | |
2293 | * @dentry: entry to move | |
2294 | * @target: new dentry | |
2295 | * | |
2296 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2297 | * dcache entries should not be moved in this way. See the locking |
2298 | * requirements for __d_move. | |
18367501 AV |
2299 | */ |
2300 | void d_move(struct dentry *dentry, struct dentry *target) | |
2301 | { | |
2302 | write_seqlock(&rename_lock); | |
2303 | __d_move(dentry, target); | |
1da177e4 | 2304 | write_sequnlock(&rename_lock); |
9eaef27b | 2305 | } |
ec4f8605 | 2306 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2307 | |
e2761a11 OH |
2308 | /** |
2309 | * d_ancestor - search for an ancestor | |
2310 | * @p1: ancestor dentry | |
2311 | * @p2: child dentry | |
2312 | * | |
2313 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2314 | * an ancestor of p2, else NULL. | |
9eaef27b | 2315 | */ |
e2761a11 | 2316 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2317 | { |
2318 | struct dentry *p; | |
2319 | ||
871c0067 | 2320 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2321 | if (p->d_parent == p1) |
e2761a11 | 2322 | return p; |
9eaef27b | 2323 | } |
e2761a11 | 2324 | return NULL; |
9eaef27b TM |
2325 | } |
2326 | ||
2327 | /* | |
2328 | * This helper attempts to cope with remotely renamed directories | |
2329 | * | |
2330 | * It assumes that the caller is already holding | |
18367501 | 2331 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2332 | * |
2333 | * Note: If ever the locking in lock_rename() changes, then please | |
2334 | * remember to update this too... | |
9eaef27b | 2335 | */ |
873feea0 NP |
2336 | static struct dentry *__d_unalias(struct inode *inode, |
2337 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2338 | { |
2339 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2340 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2341 | |
2342 | /* If alias and dentry share a parent, then no extra locks required */ | |
2343 | if (alias->d_parent == dentry->d_parent) | |
2344 | goto out_unalias; | |
2345 | ||
9eaef27b | 2346 | /* See lock_rename() */ |
9eaef27b TM |
2347 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2348 | goto out_err; | |
2349 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2350 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2351 | goto out_err; | |
2352 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2353 | out_unalias: | |
ee3efa91 AV |
2354 | if (likely(!d_mountpoint(alias))) { |
2355 | __d_move(alias, dentry); | |
2356 | ret = alias; | |
2357 | } | |
9eaef27b | 2358 | out_err: |
873feea0 | 2359 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2360 | if (m2) |
2361 | mutex_unlock(m2); | |
2362 | if (m1) | |
2363 | mutex_unlock(m1); | |
2364 | return ret; | |
2365 | } | |
2366 | ||
770bfad8 DH |
2367 | /* |
2368 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2369 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2370 | * returns with anon->d_lock held! |
770bfad8 DH |
2371 | */ |
2372 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2373 | { | |
740da42e | 2374 | struct dentry *dparent; |
770bfad8 | 2375 | |
2fd6b7f5 | 2376 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2377 | |
31e6b01f NP |
2378 | write_seqcount_begin(&dentry->d_seq); |
2379 | write_seqcount_begin(&anon->d_seq); | |
2380 | ||
770bfad8 | 2381 | dparent = dentry->d_parent; |
770bfad8 | 2382 | |
2fd6b7f5 NP |
2383 | switch_names(dentry, anon); |
2384 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2385 | ||
740da42e AV |
2386 | dentry->d_parent = dentry; |
2387 | list_del_init(&dentry->d_u.d_child); | |
2388 | anon->d_parent = dparent; | |
9ed53b12 | 2389 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2390 | |
31e6b01f NP |
2391 | write_seqcount_end(&dentry->d_seq); |
2392 | write_seqcount_end(&anon->d_seq); | |
2393 | ||
2fd6b7f5 NP |
2394 | dentry_unlock_parents_for_move(anon, dentry); |
2395 | spin_unlock(&dentry->d_lock); | |
2396 | ||
2397 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2398 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2399 | } | |
2400 | ||
2401 | /** | |
2402 | * d_materialise_unique - introduce an inode into the tree | |
2403 | * @dentry: candidate dentry | |
2404 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2405 | * | |
2406 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2407 | * root directory alias in its place if there is one. Caller must hold the |
2408 | * i_mutex of the parent directory. | |
770bfad8 DH |
2409 | */ |
2410 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2411 | { | |
9eaef27b | 2412 | struct dentry *actual; |
770bfad8 DH |
2413 | |
2414 | BUG_ON(!d_unhashed(dentry)); | |
2415 | ||
770bfad8 DH |
2416 | if (!inode) { |
2417 | actual = dentry; | |
360da900 | 2418 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2419 | d_rehash(actual); |
2420 | goto out_nolock; | |
770bfad8 DH |
2421 | } |
2422 | ||
873feea0 | 2423 | spin_lock(&inode->i_lock); |
357f8e65 | 2424 | |
9eaef27b TM |
2425 | if (S_ISDIR(inode->i_mode)) { |
2426 | struct dentry *alias; | |
2427 | ||
2428 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2429 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2430 | if (alias) { |
2431 | actual = alias; | |
18367501 AV |
2432 | write_seqlock(&rename_lock); |
2433 | ||
2434 | if (d_ancestor(alias, dentry)) { | |
2435 | /* Check for loops */ | |
2436 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2437 | spin_unlock(&inode->i_lock); |
18367501 AV |
2438 | } else if (IS_ROOT(alias)) { |
2439 | /* Is this an anonymous mountpoint that we | |
2440 | * could splice into our tree? */ | |
9eaef27b | 2441 | __d_materialise_dentry(dentry, alias); |
18367501 | 2442 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2443 | __d_drop(alias); |
2444 | goto found; | |
18367501 AV |
2445 | } else { |
2446 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2447 | * aliasing. This drops inode->i_lock */ |
18367501 | 2448 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2449 | } |
18367501 | 2450 | write_sequnlock(&rename_lock); |
dd179946 DH |
2451 | if (IS_ERR(actual)) { |
2452 | if (PTR_ERR(actual) == -ELOOP) | |
2453 | pr_warn_ratelimited( | |
2454 | "VFS: Lookup of '%s' in %s %s" | |
2455 | " would have caused loop\n", | |
2456 | dentry->d_name.name, | |
2457 | inode->i_sb->s_type->name, | |
2458 | inode->i_sb->s_id); | |
9eaef27b | 2459 | dput(alias); |
dd179946 | 2460 | } |
9eaef27b TM |
2461 | goto out_nolock; |
2462 | } | |
770bfad8 DH |
2463 | } |
2464 | ||
2465 | /* Add a unique reference */ | |
2466 | actual = __d_instantiate_unique(dentry, inode); | |
2467 | if (!actual) | |
2468 | actual = dentry; | |
357f8e65 NP |
2469 | else |
2470 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2471 | |
770bfad8 DH |
2472 | spin_lock(&actual->d_lock); |
2473 | found: | |
2474 | _d_rehash(actual); | |
2475 | spin_unlock(&actual->d_lock); | |
873feea0 | 2476 | spin_unlock(&inode->i_lock); |
9eaef27b | 2477 | out_nolock: |
770bfad8 DH |
2478 | if (actual == dentry) { |
2479 | security_d_instantiate(dentry, inode); | |
2480 | return NULL; | |
2481 | } | |
2482 | ||
2483 | iput(inode); | |
2484 | return actual; | |
770bfad8 | 2485 | } |
ec4f8605 | 2486 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2487 | |
cdd16d02 | 2488 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2489 | { |
2490 | *buflen -= namelen; | |
2491 | if (*buflen < 0) | |
2492 | return -ENAMETOOLONG; | |
2493 | *buffer -= namelen; | |
2494 | memcpy(*buffer, str, namelen); | |
2495 | return 0; | |
2496 | } | |
2497 | ||
cdd16d02 MS |
2498 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2499 | { | |
2500 | return prepend(buffer, buflen, name->name, name->len); | |
2501 | } | |
2502 | ||
1da177e4 | 2503 | /** |
208898c1 | 2504 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2505 | * @path: the dentry/vfsmount to report |
02125a82 | 2506 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2507 | * @buffer: pointer to the end of the buffer |
2508 | * @buflen: pointer to buffer length | |
552ce544 | 2509 | * |
949854d0 | 2510 | * Caller holds the rename_lock. |
1da177e4 | 2511 | */ |
02125a82 AV |
2512 | static int prepend_path(const struct path *path, |
2513 | const struct path *root, | |
f2eb6575 | 2514 | char **buffer, int *buflen) |
1da177e4 | 2515 | { |
9d1bc601 MS |
2516 | struct dentry *dentry = path->dentry; |
2517 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2518 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 MS |
2519 | bool slash = false; |
2520 | int error = 0; | |
6092d048 | 2521 | |
f2eb6575 | 2522 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2523 | struct dentry * parent; |
2524 | ||
1da177e4 | 2525 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2526 | /* Global root? */ |
676da58d | 2527 | if (!mnt_has_parent(mnt)) |
1da177e4 | 2528 | goto global_root; |
a73324da | 2529 | dentry = mnt->mnt_mountpoint; |
0714a533 AV |
2530 | mnt = mnt->mnt_parent; |
2531 | vfsmnt = &mnt->mnt; | |
1da177e4 LT |
2532 | continue; |
2533 | } | |
2534 | parent = dentry->d_parent; | |
2535 | prefetch(parent); | |
9abca360 | 2536 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2537 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2538 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2539 | if (!error) |
2540 | error = prepend(buffer, buflen, "/", 1); | |
2541 | if (error) | |
2542 | break; | |
2543 | ||
2544 | slash = true; | |
1da177e4 LT |
2545 | dentry = parent; |
2546 | } | |
2547 | ||
f2eb6575 MS |
2548 | if (!error && !slash) |
2549 | error = prepend(buffer, buflen, "/", 1); | |
2550 | ||
f2eb6575 | 2551 | return error; |
1da177e4 LT |
2552 | |
2553 | global_root: | |
98dc568b MS |
2554 | /* |
2555 | * Filesystems needing to implement special "root names" | |
2556 | * should do so with ->d_dname() | |
2557 | */ | |
2558 | if (IS_ROOT(dentry) && | |
2559 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2560 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2561 | (int) dentry->d_name.len, dentry->d_name.name); | |
2562 | } | |
02125a82 AV |
2563 | if (!slash) |
2564 | error = prepend(buffer, buflen, "/", 1); | |
2565 | if (!error) | |
f7a99c5b | 2566 | error = is_mounted(vfsmnt) ? 1 : 2; |
7ea600b5 | 2567 | return error; |
f2eb6575 | 2568 | } |
be285c71 | 2569 | |
f2eb6575 MS |
2570 | /** |
2571 | * __d_path - return the path of a dentry | |
2572 | * @path: the dentry/vfsmount to report | |
02125a82 | 2573 | * @root: root vfsmnt/dentry |
cd956a1c | 2574 | * @buf: buffer to return value in |
f2eb6575 MS |
2575 | * @buflen: buffer length |
2576 | * | |
ffd1f4ed | 2577 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2578 | * |
2579 | * Returns a pointer into the buffer or an error code if the | |
2580 | * path was too long. | |
2581 | * | |
be148247 | 2582 | * "buflen" should be positive. |
f2eb6575 | 2583 | * |
02125a82 | 2584 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2585 | */ |
02125a82 AV |
2586 | char *__d_path(const struct path *path, |
2587 | const struct path *root, | |
f2eb6575 MS |
2588 | char *buf, int buflen) |
2589 | { | |
2590 | char *res = buf + buflen; | |
2591 | int error; | |
2592 | ||
2593 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2594 | br_read_lock(&vfsmount_lock); |
949854d0 | 2595 | write_seqlock(&rename_lock); |
f2eb6575 | 2596 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2597 | write_sequnlock(&rename_lock); |
7ea600b5 | 2598 | br_read_unlock(&vfsmount_lock); |
be148247 | 2599 | |
02125a82 AV |
2600 | if (error < 0) |
2601 | return ERR_PTR(error); | |
2602 | if (error > 0) | |
2603 | return NULL; | |
2604 | return res; | |
2605 | } | |
2606 | ||
2607 | char *d_absolute_path(const struct path *path, | |
2608 | char *buf, int buflen) | |
2609 | { | |
2610 | struct path root = {}; | |
2611 | char *res = buf + buflen; | |
2612 | int error; | |
2613 | ||
2614 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2615 | br_read_lock(&vfsmount_lock); |
02125a82 AV |
2616 | write_seqlock(&rename_lock); |
2617 | error = prepend_path(path, &root, &res, &buflen); | |
2618 | write_sequnlock(&rename_lock); | |
7ea600b5 | 2619 | br_read_unlock(&vfsmount_lock); |
02125a82 AV |
2620 | |
2621 | if (error > 1) | |
2622 | error = -EINVAL; | |
2623 | if (error < 0) | |
f2eb6575 | 2624 | return ERR_PTR(error); |
f2eb6575 | 2625 | return res; |
1da177e4 LT |
2626 | } |
2627 | ||
ffd1f4ed MS |
2628 | /* |
2629 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2630 | */ | |
02125a82 AV |
2631 | static int path_with_deleted(const struct path *path, |
2632 | const struct path *root, | |
2633 | char **buf, int *buflen) | |
ffd1f4ed MS |
2634 | { |
2635 | prepend(buf, buflen, "\0", 1); | |
2636 | if (d_unlinked(path->dentry)) { | |
2637 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2638 | if (error) | |
2639 | return error; | |
2640 | } | |
2641 | ||
2642 | return prepend_path(path, root, buf, buflen); | |
2643 | } | |
2644 | ||
8df9d1a4 MS |
2645 | static int prepend_unreachable(char **buffer, int *buflen) |
2646 | { | |
2647 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2648 | } | |
2649 | ||
a03a8a70 JB |
2650 | /** |
2651 | * d_path - return the path of a dentry | |
cf28b486 | 2652 | * @path: path to report |
a03a8a70 JB |
2653 | * @buf: buffer to return value in |
2654 | * @buflen: buffer length | |
2655 | * | |
2656 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2657 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2658 | * | |
52afeefb AV |
2659 | * Returns a pointer into the buffer or an error code if the path was |
2660 | * too long. Note: Callers should use the returned pointer, not the passed | |
2661 | * in buffer, to use the name! The implementation often starts at an offset | |
2662 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2663 | * |
31f3e0b3 | 2664 | * "buflen" should be positive. |
a03a8a70 | 2665 | */ |
20d4fdc1 | 2666 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2667 | { |
ffd1f4ed | 2668 | char *res = buf + buflen; |
6ac08c39 | 2669 | struct path root; |
ffd1f4ed | 2670 | int error; |
1da177e4 | 2671 | |
c23fbb6b ED |
2672 | /* |
2673 | * We have various synthetic filesystems that never get mounted. On | |
2674 | * these filesystems dentries are never used for lookup purposes, and | |
2675 | * thus don't need to be hashed. They also don't need a name until a | |
2676 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2677 | * below allows us to generate a name for these objects on demand: | |
2678 | */ | |
cf28b486 JB |
2679 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2680 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2681 | |
f7ad3c6b | 2682 | get_fs_root(current->fs, &root); |
7ea600b5 | 2683 | br_read_lock(&vfsmount_lock); |
949854d0 | 2684 | write_seqlock(&rename_lock); |
02125a82 | 2685 | error = path_with_deleted(path, &root, &res, &buflen); |
7ea600b5 AV |
2686 | write_sequnlock(&rename_lock); |
2687 | br_read_unlock(&vfsmount_lock); | |
02125a82 | 2688 | if (error < 0) |
ffd1f4ed | 2689 | res = ERR_PTR(error); |
6ac08c39 | 2690 | path_put(&root); |
1da177e4 LT |
2691 | return res; |
2692 | } | |
ec4f8605 | 2693 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2694 | |
c23fbb6b ED |
2695 | /* |
2696 | * Helper function for dentry_operations.d_dname() members | |
2697 | */ | |
2698 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2699 | const char *fmt, ...) | |
2700 | { | |
2701 | va_list args; | |
2702 | char temp[64]; | |
2703 | int sz; | |
2704 | ||
2705 | va_start(args, fmt); | |
2706 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2707 | va_end(args); | |
2708 | ||
2709 | if (sz > sizeof(temp) || sz > buflen) | |
2710 | return ERR_PTR(-ENAMETOOLONG); | |
2711 | ||
2712 | buffer += buflen - sz; | |
2713 | return memcpy(buffer, temp, sz); | |
2714 | } | |
2715 | ||
118b2302 AV |
2716 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
2717 | { | |
2718 | char *end = buffer + buflen; | |
2719 | /* these dentries are never renamed, so d_lock is not needed */ | |
2720 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
2721 | prepend_name(&end, &buflen, &dentry->d_name) || | |
2722 | prepend(&end, &buflen, "/", 1)) | |
2723 | end = ERR_PTR(-ENAMETOOLONG); | |
2724 | return end; | |
2725 | } | |
2726 | ||
6092d048 RP |
2727 | /* |
2728 | * Write full pathname from the root of the filesystem into the buffer. | |
2729 | */ | |
ec2447c2 | 2730 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2731 | { |
2732 | char *end = buf + buflen; | |
2733 | char *retval; | |
2734 | ||
6092d048 | 2735 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2736 | if (buflen < 1) |
2737 | goto Elong; | |
2738 | /* Get '/' right */ | |
2739 | retval = end-1; | |
2740 | *retval = '/'; | |
2741 | ||
cdd16d02 MS |
2742 | while (!IS_ROOT(dentry)) { |
2743 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2744 | int error; |
6092d048 | 2745 | |
6092d048 | 2746 | prefetch(parent); |
9abca360 NP |
2747 | spin_lock(&dentry->d_lock); |
2748 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2749 | spin_unlock(&dentry->d_lock); | |
2750 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2751 | goto Elong; |
2752 | ||
2753 | retval = end; | |
2754 | dentry = parent; | |
2755 | } | |
c103135c AV |
2756 | return retval; |
2757 | Elong: | |
2758 | return ERR_PTR(-ENAMETOOLONG); | |
2759 | } | |
ec2447c2 NP |
2760 | |
2761 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2762 | { | |
2763 | char *retval; | |
2764 | ||
949854d0 | 2765 | write_seqlock(&rename_lock); |
ec2447c2 | 2766 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2767 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2768 | |
2769 | return retval; | |
2770 | } | |
2771 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2772 | |
2773 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2774 | { | |
2775 | char *p = NULL; | |
2776 | char *retval; | |
2777 | ||
949854d0 | 2778 | write_seqlock(&rename_lock); |
c103135c AV |
2779 | if (d_unlinked(dentry)) { |
2780 | p = buf + buflen; | |
2781 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2782 | goto Elong; | |
2783 | buflen++; | |
2784 | } | |
2785 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2786 | write_sequnlock(&rename_lock); |
c103135c AV |
2787 | if (!IS_ERR(retval) && p) |
2788 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2789 | return retval; |
2790 | Elong: | |
6092d048 RP |
2791 | return ERR_PTR(-ENAMETOOLONG); |
2792 | } | |
2793 | ||
1da177e4 LT |
2794 | /* |
2795 | * NOTE! The user-level library version returns a | |
2796 | * character pointer. The kernel system call just | |
2797 | * returns the length of the buffer filled (which | |
2798 | * includes the ending '\0' character), or a negative | |
2799 | * error value. So libc would do something like | |
2800 | * | |
2801 | * char *getcwd(char * buf, size_t size) | |
2802 | * { | |
2803 | * int retval; | |
2804 | * | |
2805 | * retval = sys_getcwd(buf, size); | |
2806 | * if (retval >= 0) | |
2807 | * return buf; | |
2808 | * errno = -retval; | |
2809 | * return NULL; | |
2810 | * } | |
2811 | */ | |
3cdad428 | 2812 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2813 | { |
552ce544 | 2814 | int error; |
6ac08c39 | 2815 | struct path pwd, root; |
552ce544 | 2816 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2817 | |
2818 | if (!page) | |
2819 | return -ENOMEM; | |
2820 | ||
f7ad3c6b | 2821 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2822 | |
552ce544 | 2823 | error = -ENOENT; |
7ea600b5 | 2824 | br_read_lock(&vfsmount_lock); |
949854d0 | 2825 | write_seqlock(&rename_lock); |
f3da392e | 2826 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2827 | unsigned long len; |
8df9d1a4 MS |
2828 | char *cwd = page + PAGE_SIZE; |
2829 | int buflen = PAGE_SIZE; | |
1da177e4 | 2830 | |
8df9d1a4 | 2831 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 2832 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
949854d0 | 2833 | write_sequnlock(&rename_lock); |
7ea600b5 | 2834 | br_read_unlock(&vfsmount_lock); |
552ce544 | 2835 | |
02125a82 | 2836 | if (error < 0) |
552ce544 LT |
2837 | goto out; |
2838 | ||
8df9d1a4 | 2839 | /* Unreachable from current root */ |
02125a82 | 2840 | if (error > 0) { |
8df9d1a4 MS |
2841 | error = prepend_unreachable(&cwd, &buflen); |
2842 | if (error) | |
2843 | goto out; | |
2844 | } | |
2845 | ||
552ce544 LT |
2846 | error = -ERANGE; |
2847 | len = PAGE_SIZE + page - cwd; | |
2848 | if (len <= size) { | |
2849 | error = len; | |
2850 | if (copy_to_user(buf, cwd, len)) | |
2851 | error = -EFAULT; | |
2852 | } | |
949854d0 NP |
2853 | } else { |
2854 | write_sequnlock(&rename_lock); | |
7ea600b5 | 2855 | br_read_unlock(&vfsmount_lock); |
949854d0 | 2856 | } |
1da177e4 LT |
2857 | |
2858 | out: | |
6ac08c39 JB |
2859 | path_put(&pwd); |
2860 | path_put(&root); | |
1da177e4 LT |
2861 | free_page((unsigned long) page); |
2862 | return error; | |
2863 | } | |
2864 | ||
2865 | /* | |
2866 | * Test whether new_dentry is a subdirectory of old_dentry. | |
2867 | * | |
2868 | * Trivially implemented using the dcache structure | |
2869 | */ | |
2870 | ||
2871 | /** | |
2872 | * is_subdir - is new dentry a subdirectory of old_dentry | |
2873 | * @new_dentry: new dentry | |
2874 | * @old_dentry: old dentry | |
2875 | * | |
2876 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
2877 | * Returns 0 otherwise. | |
2878 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
2879 | */ | |
2880 | ||
e2761a11 | 2881 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
2882 | { |
2883 | int result; | |
949854d0 | 2884 | unsigned seq; |
1da177e4 | 2885 | |
e2761a11 OH |
2886 | if (new_dentry == old_dentry) |
2887 | return 1; | |
2888 | ||
e2761a11 | 2889 | do { |
1da177e4 | 2890 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 2891 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
2892 | /* |
2893 | * Need rcu_readlock to protect against the d_parent trashing | |
2894 | * due to d_move | |
2895 | */ | |
2896 | rcu_read_lock(); | |
e2761a11 | 2897 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 2898 | result = 1; |
e2761a11 OH |
2899 | else |
2900 | result = 0; | |
949854d0 | 2901 | rcu_read_unlock(); |
1da177e4 | 2902 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
2903 | |
2904 | return result; | |
2905 | } | |
2906 | ||
2907 | void d_genocide(struct dentry *root) | |
2908 | { | |
949854d0 | 2909 | struct dentry *this_parent; |
1da177e4 | 2910 | struct list_head *next; |
949854d0 | 2911 | unsigned seq; |
58db63d0 | 2912 | int locked = 0; |
1da177e4 | 2913 | |
949854d0 | 2914 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
2915 | again: |
2916 | this_parent = root; | |
2fd6b7f5 | 2917 | spin_lock(&this_parent->d_lock); |
1da177e4 LT |
2918 | repeat: |
2919 | next = this_parent->d_subdirs.next; | |
2920 | resume: | |
2921 | while (next != &this_parent->d_subdirs) { | |
2922 | struct list_head *tmp = next; | |
5160ee6f | 2923 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 2924 | next = tmp->next; |
949854d0 | 2925 | |
da502956 NP |
2926 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
2927 | if (d_unhashed(dentry) || !dentry->d_inode) { | |
2928 | spin_unlock(&dentry->d_lock); | |
1da177e4 | 2929 | continue; |
da502956 | 2930 | } |
1da177e4 | 2931 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
2932 | spin_unlock(&this_parent->d_lock); |
2933 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 2934 | this_parent = dentry; |
2fd6b7f5 | 2935 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
2936 | goto repeat; |
2937 | } | |
949854d0 NP |
2938 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
2939 | dentry->d_flags |= DCACHE_GENOCIDE; | |
98474236 | 2940 | dentry->d_lockref.count--; |
949854d0 | 2941 | } |
b7ab39f6 | 2942 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
2943 | } |
2944 | if (this_parent != root) { | |
c826cb7d | 2945 | struct dentry *child = this_parent; |
949854d0 NP |
2946 | if (!(this_parent->d_flags & DCACHE_GENOCIDE)) { |
2947 | this_parent->d_flags |= DCACHE_GENOCIDE; | |
98474236 | 2948 | this_parent->d_lockref.count--; |
949854d0 | 2949 | } |
c826cb7d LT |
2950 | this_parent = try_to_ascend(this_parent, locked, seq); |
2951 | if (!this_parent) | |
949854d0 | 2952 | goto rename_retry; |
949854d0 | 2953 | next = child->d_u.d_child.next; |
1da177e4 LT |
2954 | goto resume; |
2955 | } | |
2fd6b7f5 | 2956 | spin_unlock(&this_parent->d_lock); |
58db63d0 | 2957 | if (!locked && read_seqretry(&rename_lock, seq)) |
949854d0 | 2958 | goto rename_retry; |
58db63d0 NP |
2959 | if (locked) |
2960 | write_sequnlock(&rename_lock); | |
2961 | return; | |
2962 | ||
2963 | rename_retry: | |
8110e16d MS |
2964 | if (locked) |
2965 | goto again; | |
58db63d0 NP |
2966 | locked = 1; |
2967 | write_seqlock(&rename_lock); | |
2968 | goto again; | |
1da177e4 LT |
2969 | } |
2970 | ||
60545d0d | 2971 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 2972 | { |
60545d0d AV |
2973 | inode_dec_link_count(inode); |
2974 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
2975 | !hlist_unhashed(&dentry->d_alias) || | |
2976 | !d_unlinked(dentry)); | |
2977 | spin_lock(&dentry->d_parent->d_lock); | |
2978 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
2979 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
2980 | (unsigned long long)inode->i_ino); | |
2981 | spin_unlock(&dentry->d_lock); | |
2982 | spin_unlock(&dentry->d_parent->d_lock); | |
2983 | d_instantiate(dentry, inode); | |
1da177e4 | 2984 | } |
60545d0d | 2985 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
2986 | |
2987 | static __initdata unsigned long dhash_entries; | |
2988 | static int __init set_dhash_entries(char *str) | |
2989 | { | |
2990 | if (!str) | |
2991 | return 0; | |
2992 | dhash_entries = simple_strtoul(str, &str, 0); | |
2993 | return 1; | |
2994 | } | |
2995 | __setup("dhash_entries=", set_dhash_entries); | |
2996 | ||
2997 | static void __init dcache_init_early(void) | |
2998 | { | |
074b8517 | 2999 | unsigned int loop; |
1da177e4 LT |
3000 | |
3001 | /* If hashes are distributed across NUMA nodes, defer | |
3002 | * hash allocation until vmalloc space is available. | |
3003 | */ | |
3004 | if (hashdist) | |
3005 | return; | |
3006 | ||
3007 | dentry_hashtable = | |
3008 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3009 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3010 | dhash_entries, |
3011 | 13, | |
3012 | HASH_EARLY, | |
3013 | &d_hash_shift, | |
3014 | &d_hash_mask, | |
31fe62b9 | 3015 | 0, |
1da177e4 LT |
3016 | 0); |
3017 | ||
074b8517 | 3018 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3019 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3020 | } |
3021 | ||
74bf17cf | 3022 | static void __init dcache_init(void) |
1da177e4 | 3023 | { |
074b8517 | 3024 | unsigned int loop; |
1da177e4 LT |
3025 | |
3026 | /* | |
3027 | * A constructor could be added for stable state like the lists, | |
3028 | * but it is probably not worth it because of the cache nature | |
3029 | * of the dcache. | |
3030 | */ | |
0a31bd5f CL |
3031 | dentry_cache = KMEM_CACHE(dentry, |
3032 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3033 | |
3034 | /* Hash may have been set up in dcache_init_early */ | |
3035 | if (!hashdist) | |
3036 | return; | |
3037 | ||
3038 | dentry_hashtable = | |
3039 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3040 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3041 | dhash_entries, |
3042 | 13, | |
3043 | 0, | |
3044 | &d_hash_shift, | |
3045 | &d_hash_mask, | |
31fe62b9 | 3046 | 0, |
1da177e4 LT |
3047 | 0); |
3048 | ||
074b8517 | 3049 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3050 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3051 | } |
3052 | ||
3053 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3054 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3055 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3056 | |
1da177e4 LT |
3057 | EXPORT_SYMBOL(d_genocide); |
3058 | ||
1da177e4 LT |
3059 | void __init vfs_caches_init_early(void) |
3060 | { | |
3061 | dcache_init_early(); | |
3062 | inode_init_early(); | |
3063 | } | |
3064 | ||
3065 | void __init vfs_caches_init(unsigned long mempages) | |
3066 | { | |
3067 | unsigned long reserve; | |
3068 | ||
3069 | /* Base hash sizes on available memory, with a reserve equal to | |
3070 | 150% of current kernel size */ | |
3071 | ||
3072 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3073 | mempages -= reserve; | |
3074 | ||
3075 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3076 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3077 | |
74bf17cf DC |
3078 | dcache_init(); |
3079 | inode_init(); | |
1da177e4 | 3080 | files_init(mempages); |
74bf17cf | 3081 | mnt_init(); |
1da177e4 LT |
3082 | bdev_cache_init(); |
3083 | chrdev_init(); | |
3084 | } |