]>
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 | */ | |
590fb51f | 475 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && !d_unhashed(dentry)) |
61572bb1 YZ |
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 | { | |
df3d0bbc | 614 | int gotref; |
b7ab39f6 NP |
615 | struct dentry *ret; |
616 | ||
df3d0bbc WL |
617 | /* |
618 | * Do optimistic parent lookup without any | |
619 | * locking. | |
620 | */ | |
621 | rcu_read_lock(); | |
622 | ret = ACCESS_ONCE(dentry->d_parent); | |
623 | gotref = lockref_get_not_zero(&ret->d_lockref); | |
624 | rcu_read_unlock(); | |
625 | if (likely(gotref)) { | |
626 | if (likely(ret == ACCESS_ONCE(dentry->d_parent))) | |
627 | return ret; | |
628 | dput(ret); | |
629 | } | |
630 | ||
b7ab39f6 | 631 | repeat: |
a734eb45 NP |
632 | /* |
633 | * Don't need rcu_dereference because we re-check it was correct under | |
634 | * the lock. | |
635 | */ | |
636 | rcu_read_lock(); | |
b7ab39f6 | 637 | ret = dentry->d_parent; |
a734eb45 NP |
638 | spin_lock(&ret->d_lock); |
639 | if (unlikely(ret != dentry->d_parent)) { | |
640 | spin_unlock(&ret->d_lock); | |
641 | rcu_read_unlock(); | |
b7ab39f6 NP |
642 | goto repeat; |
643 | } | |
a734eb45 | 644 | rcu_read_unlock(); |
98474236 WL |
645 | BUG_ON(!ret->d_lockref.count); |
646 | ret->d_lockref.count++; | |
b7ab39f6 | 647 | spin_unlock(&ret->d_lock); |
b7ab39f6 NP |
648 | return ret; |
649 | } | |
650 | EXPORT_SYMBOL(dget_parent); | |
651 | ||
1da177e4 LT |
652 | /** |
653 | * d_find_alias - grab a hashed alias of inode | |
654 | * @inode: inode in question | |
32ba9c3f LT |
655 | * @want_discon: flag, used by d_splice_alias, to request |
656 | * that only a DISCONNECTED alias be returned. | |
1da177e4 LT |
657 | * |
658 | * If inode has a hashed alias, or is a directory and has any alias, | |
659 | * acquire the reference to alias and return it. Otherwise return NULL. | |
660 | * Notice that if inode is a directory there can be only one alias and | |
661 | * it can be unhashed only if it has no children, or if it is the root | |
662 | * of a filesystem. | |
663 | * | |
21c0d8fd | 664 | * If the inode has an IS_ROOT, DCACHE_DISCONNECTED alias, then prefer |
32ba9c3f LT |
665 | * any other hashed alias over that one unless @want_discon is set, |
666 | * in which case only return an IS_ROOT, DCACHE_DISCONNECTED alias. | |
1da177e4 | 667 | */ |
32ba9c3f | 668 | static struct dentry *__d_find_alias(struct inode *inode, int want_discon) |
1da177e4 | 669 | { |
da502956 | 670 | struct dentry *alias, *discon_alias; |
1da177e4 | 671 | |
da502956 NP |
672 | again: |
673 | discon_alias = NULL; | |
b67bfe0d | 674 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
da502956 | 675 | spin_lock(&alias->d_lock); |
1da177e4 | 676 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { |
21c0d8fd | 677 | if (IS_ROOT(alias) && |
da502956 | 678 | (alias->d_flags & DCACHE_DISCONNECTED)) { |
1da177e4 | 679 | discon_alias = alias; |
32ba9c3f | 680 | } else if (!want_discon) { |
dc0474be | 681 | __dget_dlock(alias); |
da502956 NP |
682 | spin_unlock(&alias->d_lock); |
683 | return alias; | |
684 | } | |
685 | } | |
686 | spin_unlock(&alias->d_lock); | |
687 | } | |
688 | if (discon_alias) { | |
689 | alias = discon_alias; | |
690 | spin_lock(&alias->d_lock); | |
691 | if (S_ISDIR(inode->i_mode) || !d_unhashed(alias)) { | |
692 | if (IS_ROOT(alias) && | |
693 | (alias->d_flags & DCACHE_DISCONNECTED)) { | |
dc0474be | 694 | __dget_dlock(alias); |
da502956 | 695 | spin_unlock(&alias->d_lock); |
1da177e4 LT |
696 | return alias; |
697 | } | |
698 | } | |
da502956 NP |
699 | spin_unlock(&alias->d_lock); |
700 | goto again; | |
1da177e4 | 701 | } |
da502956 | 702 | return NULL; |
1da177e4 LT |
703 | } |
704 | ||
da502956 | 705 | struct dentry *d_find_alias(struct inode *inode) |
1da177e4 | 706 | { |
214fda1f DH |
707 | struct dentry *de = NULL; |
708 | ||
b3d9b7a3 | 709 | if (!hlist_empty(&inode->i_dentry)) { |
873feea0 | 710 | spin_lock(&inode->i_lock); |
32ba9c3f | 711 | de = __d_find_alias(inode, 0); |
873feea0 | 712 | spin_unlock(&inode->i_lock); |
214fda1f | 713 | } |
1da177e4 LT |
714 | return de; |
715 | } | |
ec4f8605 | 716 | EXPORT_SYMBOL(d_find_alias); |
1da177e4 LT |
717 | |
718 | /* | |
719 | * Try to kill dentries associated with this inode. | |
720 | * WARNING: you must own a reference to inode. | |
721 | */ | |
722 | void d_prune_aliases(struct inode *inode) | |
723 | { | |
0cdca3f9 | 724 | struct dentry *dentry; |
1da177e4 | 725 | restart: |
873feea0 | 726 | spin_lock(&inode->i_lock); |
b67bfe0d | 727 | hlist_for_each_entry(dentry, &inode->i_dentry, d_alias) { |
1da177e4 | 728 | spin_lock(&dentry->d_lock); |
98474236 | 729 | if (!dentry->d_lockref.count) { |
590fb51f YZ |
730 | /* |
731 | * inform the fs via d_prune that this dentry | |
732 | * is about to be unhashed and destroyed. | |
733 | */ | |
734 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && | |
735 | !d_unhashed(dentry)) | |
736 | dentry->d_op->d_prune(dentry); | |
737 | ||
dc0474be | 738 | __dget_dlock(dentry); |
1da177e4 LT |
739 | __d_drop(dentry); |
740 | spin_unlock(&dentry->d_lock); | |
873feea0 | 741 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
742 | dput(dentry); |
743 | goto restart; | |
744 | } | |
745 | spin_unlock(&dentry->d_lock); | |
746 | } | |
873feea0 | 747 | spin_unlock(&inode->i_lock); |
1da177e4 | 748 | } |
ec4f8605 | 749 | EXPORT_SYMBOL(d_prune_aliases); |
1da177e4 LT |
750 | |
751 | /* | |
77812a1e NP |
752 | * Try to throw away a dentry - free the inode, dput the parent. |
753 | * Requires dentry->d_lock is held, and dentry->d_count == 0. | |
754 | * Releases dentry->d_lock. | |
d702ccb3 | 755 | * |
77812a1e | 756 | * This may fail if locks cannot be acquired no problem, just try again. |
1da177e4 | 757 | */ |
77812a1e | 758 | static void try_prune_one_dentry(struct dentry *dentry) |
31f3e0b3 | 759 | __releases(dentry->d_lock) |
1da177e4 | 760 | { |
77812a1e | 761 | struct dentry *parent; |
d52b9086 | 762 | |
77812a1e | 763 | parent = dentry_kill(dentry, 0); |
d52b9086 | 764 | /* |
77812a1e NP |
765 | * If dentry_kill returns NULL, we have nothing more to do. |
766 | * if it returns the same dentry, trylocks failed. In either | |
767 | * case, just loop again. | |
768 | * | |
769 | * Otherwise, we need to prune ancestors too. This is necessary | |
770 | * to prevent quadratic behavior of shrink_dcache_parent(), but | |
771 | * is also expected to be beneficial in reducing dentry cache | |
772 | * fragmentation. | |
d52b9086 | 773 | */ |
77812a1e NP |
774 | if (!parent) |
775 | return; | |
776 | if (parent == dentry) | |
777 | return; | |
778 | ||
779 | /* Prune ancestors. */ | |
780 | dentry = parent; | |
d52b9086 | 781 | while (dentry) { |
98474236 | 782 | if (lockref_put_or_lock(&dentry->d_lockref)) |
89e60548 | 783 | return; |
77812a1e | 784 | dentry = dentry_kill(dentry, 1); |
d52b9086 | 785 | } |
1da177e4 LT |
786 | } |
787 | ||
3049cfe2 | 788 | static void shrink_dentry_list(struct list_head *list) |
1da177e4 | 789 | { |
da3bbdd4 | 790 | struct dentry *dentry; |
da3bbdd4 | 791 | |
ec33679d NP |
792 | rcu_read_lock(); |
793 | for (;;) { | |
ec33679d NP |
794 | dentry = list_entry_rcu(list->prev, struct dentry, d_lru); |
795 | if (&dentry->d_lru == list) | |
796 | break; /* empty */ | |
797 | spin_lock(&dentry->d_lock); | |
798 | if (dentry != list_entry(list->prev, struct dentry, d_lru)) { | |
799 | spin_unlock(&dentry->d_lock); | |
23044507 NP |
800 | continue; |
801 | } | |
802 | ||
1da177e4 LT |
803 | /* |
804 | * We found an inuse dentry which was not removed from | |
da3bbdd4 KM |
805 | * the LRU because of laziness during lookup. Do not free |
806 | * it - just keep it off the LRU list. | |
1da177e4 | 807 | */ |
98474236 | 808 | if (dentry->d_lockref.count) { |
ec33679d | 809 | dentry_lru_del(dentry); |
da3bbdd4 | 810 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
811 | continue; |
812 | } | |
ec33679d | 813 | |
ec33679d | 814 | rcu_read_unlock(); |
77812a1e NP |
815 | |
816 | try_prune_one_dentry(dentry); | |
817 | ||
ec33679d | 818 | rcu_read_lock(); |
da3bbdd4 | 819 | } |
ec33679d | 820 | rcu_read_unlock(); |
3049cfe2 CH |
821 | } |
822 | ||
823 | /** | |
b48f03b3 DC |
824 | * prune_dcache_sb - shrink the dcache |
825 | * @sb: superblock | |
826 | * @count: number of entries to try to free | |
827 | * | |
828 | * Attempt to shrink the superblock dcache LRU by @count entries. This is | |
829 | * done when we need more memory an called from the superblock shrinker | |
830 | * function. | |
3049cfe2 | 831 | * |
b48f03b3 DC |
832 | * This function may fail to free any resources if all the dentries are in |
833 | * use. | |
3049cfe2 | 834 | */ |
b48f03b3 | 835 | void prune_dcache_sb(struct super_block *sb, int count) |
3049cfe2 | 836 | { |
3049cfe2 CH |
837 | struct dentry *dentry; |
838 | LIST_HEAD(referenced); | |
839 | LIST_HEAD(tmp); | |
3049cfe2 | 840 | |
23044507 NP |
841 | relock: |
842 | spin_lock(&dcache_lru_lock); | |
3049cfe2 CH |
843 | while (!list_empty(&sb->s_dentry_lru)) { |
844 | dentry = list_entry(sb->s_dentry_lru.prev, | |
845 | struct dentry, d_lru); | |
846 | BUG_ON(dentry->d_sb != sb); | |
847 | ||
23044507 NP |
848 | if (!spin_trylock(&dentry->d_lock)) { |
849 | spin_unlock(&dcache_lru_lock); | |
850 | cpu_relax(); | |
851 | goto relock; | |
852 | } | |
853 | ||
b48f03b3 | 854 | if (dentry->d_flags & DCACHE_REFERENCED) { |
23044507 NP |
855 | dentry->d_flags &= ~DCACHE_REFERENCED; |
856 | list_move(&dentry->d_lru, &referenced); | |
3049cfe2 | 857 | spin_unlock(&dentry->d_lock); |
23044507 NP |
858 | } else { |
859 | list_move_tail(&dentry->d_lru, &tmp); | |
eaf5f907 | 860 | dentry->d_flags |= DCACHE_SHRINK_LIST; |
23044507 | 861 | spin_unlock(&dentry->d_lock); |
b0d40c92 | 862 | if (!--count) |
23044507 | 863 | break; |
3049cfe2 | 864 | } |
ec33679d | 865 | cond_resched_lock(&dcache_lru_lock); |
3049cfe2 | 866 | } |
da3bbdd4 KM |
867 | if (!list_empty(&referenced)) |
868 | list_splice(&referenced, &sb->s_dentry_lru); | |
23044507 | 869 | spin_unlock(&dcache_lru_lock); |
ec33679d NP |
870 | |
871 | shrink_dentry_list(&tmp); | |
da3bbdd4 KM |
872 | } |
873 | ||
1da177e4 LT |
874 | /** |
875 | * shrink_dcache_sb - shrink dcache for a superblock | |
876 | * @sb: superblock | |
877 | * | |
3049cfe2 CH |
878 | * Shrink the dcache for the specified super block. This is used to free |
879 | * the dcache before unmounting a file system. | |
1da177e4 | 880 | */ |
3049cfe2 | 881 | void shrink_dcache_sb(struct super_block *sb) |
1da177e4 | 882 | { |
3049cfe2 CH |
883 | LIST_HEAD(tmp); |
884 | ||
23044507 | 885 | spin_lock(&dcache_lru_lock); |
3049cfe2 CH |
886 | while (!list_empty(&sb->s_dentry_lru)) { |
887 | list_splice_init(&sb->s_dentry_lru, &tmp); | |
ec33679d | 888 | spin_unlock(&dcache_lru_lock); |
3049cfe2 | 889 | shrink_dentry_list(&tmp); |
ec33679d | 890 | spin_lock(&dcache_lru_lock); |
3049cfe2 | 891 | } |
23044507 | 892 | spin_unlock(&dcache_lru_lock); |
1da177e4 | 893 | } |
ec4f8605 | 894 | EXPORT_SYMBOL(shrink_dcache_sb); |
1da177e4 | 895 | |
c636ebdb DH |
896 | /* |
897 | * destroy a single subtree of dentries for unmount | |
898 | * - see the comments on shrink_dcache_for_umount() for a description of the | |
899 | * locking | |
900 | */ | |
901 | static void shrink_dcache_for_umount_subtree(struct dentry *dentry) | |
902 | { | |
903 | struct dentry *parent; | |
904 | ||
905 | BUG_ON(!IS_ROOT(dentry)); | |
906 | ||
c636ebdb DH |
907 | for (;;) { |
908 | /* descend to the first leaf in the current subtree */ | |
43c1c9cd | 909 | while (!list_empty(&dentry->d_subdirs)) |
c636ebdb DH |
910 | dentry = list_entry(dentry->d_subdirs.next, |
911 | struct dentry, d_u.d_child); | |
c636ebdb DH |
912 | |
913 | /* consume the dentries from this leaf up through its parents | |
914 | * until we find one with children or run out altogether */ | |
915 | do { | |
916 | struct inode *inode; | |
917 | ||
f0023bc6 | 918 | /* |
61572bb1 | 919 | * inform the fs that this dentry is about to be |
f0023bc6 SW |
920 | * unhashed and destroyed. |
921 | */ | |
590fb51f YZ |
922 | if ((dentry->d_flags & DCACHE_OP_PRUNE) && |
923 | !d_unhashed(dentry)) | |
61572bb1 YZ |
924 | dentry->d_op->d_prune(dentry); |
925 | ||
926 | dentry_lru_del(dentry); | |
43c1c9cd DH |
927 | __d_shrink(dentry); |
928 | ||
98474236 | 929 | if (dentry->d_lockref.count != 0) { |
c636ebdb DH |
930 | printk(KERN_ERR |
931 | "BUG: Dentry %p{i=%lx,n=%s}" | |
932 | " still in use (%d)" | |
933 | " [unmount of %s %s]\n", | |
934 | dentry, | |
935 | dentry->d_inode ? | |
936 | dentry->d_inode->i_ino : 0UL, | |
937 | dentry->d_name.name, | |
98474236 | 938 | dentry->d_lockref.count, |
c636ebdb DH |
939 | dentry->d_sb->s_type->name, |
940 | dentry->d_sb->s_id); | |
941 | BUG(); | |
942 | } | |
943 | ||
2fd6b7f5 | 944 | if (IS_ROOT(dentry)) { |
c636ebdb | 945 | parent = NULL; |
2fd6b7f5 NP |
946 | list_del(&dentry->d_u.d_child); |
947 | } else { | |
871c0067 | 948 | parent = dentry->d_parent; |
98474236 | 949 | parent->d_lockref.count--; |
2fd6b7f5 | 950 | list_del(&dentry->d_u.d_child); |
871c0067 | 951 | } |
c636ebdb | 952 | |
c636ebdb DH |
953 | inode = dentry->d_inode; |
954 | if (inode) { | |
955 | dentry->d_inode = NULL; | |
b3d9b7a3 | 956 | hlist_del_init(&dentry->d_alias); |
c636ebdb DH |
957 | if (dentry->d_op && dentry->d_op->d_iput) |
958 | dentry->d_op->d_iput(dentry, inode); | |
959 | else | |
960 | iput(inode); | |
961 | } | |
962 | ||
963 | d_free(dentry); | |
964 | ||
965 | /* finished when we fall off the top of the tree, | |
966 | * otherwise we ascend to the parent and move to the | |
967 | * next sibling if there is one */ | |
968 | if (!parent) | |
312d3ca8 | 969 | return; |
c636ebdb | 970 | dentry = parent; |
c636ebdb DH |
971 | } while (list_empty(&dentry->d_subdirs)); |
972 | ||
973 | dentry = list_entry(dentry->d_subdirs.next, | |
974 | struct dentry, d_u.d_child); | |
975 | } | |
976 | } | |
977 | ||
978 | /* | |
979 | * destroy the dentries attached to a superblock on unmounting | |
b5c84bf6 | 980 | * - we don't need to use dentry->d_lock because: |
c636ebdb DH |
981 | * - the superblock is detached from all mountings and open files, so the |
982 | * dentry trees will not be rearranged by the VFS | |
983 | * - s_umount is write-locked, so the memory pressure shrinker will ignore | |
984 | * any dentries belonging to this superblock that it comes across | |
985 | * - the filesystem itself is no longer permitted to rearrange the dentries | |
986 | * in this superblock | |
987 | */ | |
988 | void shrink_dcache_for_umount(struct super_block *sb) | |
989 | { | |
990 | struct dentry *dentry; | |
991 | ||
992 | if (down_read_trylock(&sb->s_umount)) | |
993 | BUG(); | |
994 | ||
995 | dentry = sb->s_root; | |
996 | sb->s_root = NULL; | |
98474236 | 997 | dentry->d_lockref.count--; |
c636ebdb DH |
998 | shrink_dcache_for_umount_subtree(dentry); |
999 | ||
ceb5bdc2 NP |
1000 | while (!hlist_bl_empty(&sb->s_anon)) { |
1001 | dentry = hlist_bl_entry(hlist_bl_first(&sb->s_anon), struct dentry, d_hash); | |
c636ebdb DH |
1002 | shrink_dcache_for_umount_subtree(dentry); |
1003 | } | |
1004 | } | |
1005 | ||
c826cb7d LT |
1006 | /* |
1007 | * This tries to ascend one level of parenthood, but | |
1008 | * we can race with renaming, so we need to re-check | |
1009 | * the parenthood after dropping the lock and check | |
1010 | * that the sequence number still matches. | |
1011 | */ | |
1012 | static struct dentry *try_to_ascend(struct dentry *old, int locked, unsigned seq) | |
1013 | { | |
1014 | struct dentry *new = old->d_parent; | |
1015 | ||
1016 | rcu_read_lock(); | |
1017 | spin_unlock(&old->d_lock); | |
1018 | spin_lock(&new->d_lock); | |
1019 | ||
1020 | /* | |
1021 | * might go back up the wrong parent if we have had a rename | |
1022 | * or deletion | |
1023 | */ | |
1024 | if (new != old->d_parent || | |
b161dfa6 | 1025 | (old->d_flags & DCACHE_DENTRY_KILLED) || |
c826cb7d LT |
1026 | (!locked && read_seqretry(&rename_lock, seq))) { |
1027 | spin_unlock(&new->d_lock); | |
1028 | new = NULL; | |
1029 | } | |
1030 | rcu_read_unlock(); | |
1031 | return new; | |
1032 | } | |
1033 | ||
db14fc3a MS |
1034 | /** |
1035 | * enum d_walk_ret - action to talke during tree walk | |
1036 | * @D_WALK_CONTINUE: contrinue walk | |
1037 | * @D_WALK_QUIT: quit walk | |
1038 | * @D_WALK_NORETRY: quit when retry is needed | |
1039 | * @D_WALK_SKIP: skip this dentry and its children | |
1040 | */ | |
1041 | enum d_walk_ret { | |
1042 | D_WALK_CONTINUE, | |
1043 | D_WALK_QUIT, | |
1044 | D_WALK_NORETRY, | |
1045 | D_WALK_SKIP, | |
1046 | }; | |
c826cb7d | 1047 | |
1da177e4 | 1048 | /** |
db14fc3a MS |
1049 | * d_walk - walk the dentry tree |
1050 | * @parent: start of walk | |
1051 | * @data: data passed to @enter() and @finish() | |
1052 | * @enter: callback when first entering the dentry | |
1053 | * @finish: callback when successfully finished the walk | |
1da177e4 | 1054 | * |
db14fc3a | 1055 | * The @enter() and @finish() callbacks are called with d_lock held. |
1da177e4 | 1056 | */ |
db14fc3a MS |
1057 | static void d_walk(struct dentry *parent, void *data, |
1058 | enum d_walk_ret (*enter)(void *, struct dentry *), | |
1059 | void (*finish)(void *)) | |
1da177e4 | 1060 | { |
949854d0 | 1061 | struct dentry *this_parent; |
1da177e4 | 1062 | struct list_head *next; |
949854d0 | 1063 | unsigned seq; |
58db63d0 | 1064 | int locked = 0; |
db14fc3a MS |
1065 | enum d_walk_ret ret; |
1066 | bool retry = true; | |
949854d0 | 1067 | |
949854d0 | 1068 | seq = read_seqbegin(&rename_lock); |
58db63d0 NP |
1069 | again: |
1070 | this_parent = parent; | |
2fd6b7f5 | 1071 | spin_lock(&this_parent->d_lock); |
db14fc3a MS |
1072 | |
1073 | ret = enter(data, this_parent); | |
1074 | switch (ret) { | |
1075 | case D_WALK_CONTINUE: | |
1076 | break; | |
1077 | case D_WALK_QUIT: | |
1078 | case D_WALK_SKIP: | |
1079 | goto out_unlock; | |
1080 | case D_WALK_NORETRY: | |
1081 | retry = false; | |
1082 | break; | |
1083 | } | |
1da177e4 LT |
1084 | repeat: |
1085 | next = this_parent->d_subdirs.next; | |
1086 | resume: | |
1087 | while (next != &this_parent->d_subdirs) { | |
1088 | struct list_head *tmp = next; | |
5160ee6f | 1089 | struct dentry *dentry = list_entry(tmp, struct dentry, d_u.d_child); |
1da177e4 | 1090 | next = tmp->next; |
2fd6b7f5 NP |
1091 | |
1092 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
db14fc3a MS |
1093 | |
1094 | ret = enter(data, dentry); | |
1095 | switch (ret) { | |
1096 | case D_WALK_CONTINUE: | |
1097 | break; | |
1098 | case D_WALK_QUIT: | |
2fd6b7f5 | 1099 | spin_unlock(&dentry->d_lock); |
db14fc3a MS |
1100 | goto out_unlock; |
1101 | case D_WALK_NORETRY: | |
1102 | retry = false; | |
1103 | break; | |
1104 | case D_WALK_SKIP: | |
1105 | spin_unlock(&dentry->d_lock); | |
1106 | continue; | |
2fd6b7f5 | 1107 | } |
db14fc3a | 1108 | |
1da177e4 | 1109 | if (!list_empty(&dentry->d_subdirs)) { |
2fd6b7f5 NP |
1110 | spin_unlock(&this_parent->d_lock); |
1111 | spin_release(&dentry->d_lock.dep_map, 1, _RET_IP_); | |
1da177e4 | 1112 | this_parent = dentry; |
2fd6b7f5 | 1113 | spin_acquire(&this_parent->d_lock.dep_map, 0, 1, _RET_IP_); |
1da177e4 LT |
1114 | goto repeat; |
1115 | } | |
2fd6b7f5 | 1116 | spin_unlock(&dentry->d_lock); |
1da177e4 LT |
1117 | } |
1118 | /* | |
1119 | * All done at this level ... ascend and resume the search. | |
1120 | */ | |
1121 | if (this_parent != parent) { | |
c826cb7d LT |
1122 | struct dentry *child = this_parent; |
1123 | this_parent = try_to_ascend(this_parent, locked, seq); | |
1124 | if (!this_parent) | |
949854d0 | 1125 | goto rename_retry; |
949854d0 | 1126 | next = child->d_u.d_child.next; |
1da177e4 LT |
1127 | goto resume; |
1128 | } | |
db14fc3a MS |
1129 | if (!locked && read_seqretry(&rename_lock, seq)) { |
1130 | spin_unlock(&this_parent->d_lock); | |
949854d0 | 1131 | goto rename_retry; |
db14fc3a MS |
1132 | } |
1133 | if (finish) | |
1134 | finish(data); | |
1135 | ||
1136 | out_unlock: | |
1137 | spin_unlock(&this_parent->d_lock); | |
58db63d0 NP |
1138 | if (locked) |
1139 | write_sequnlock(&rename_lock); | |
db14fc3a | 1140 | return; |
58db63d0 NP |
1141 | |
1142 | rename_retry: | |
db14fc3a MS |
1143 | if (!retry) |
1144 | return; | |
8110e16d MS |
1145 | if (locked) |
1146 | goto again; | |
58db63d0 NP |
1147 | locked = 1; |
1148 | write_seqlock(&rename_lock); | |
1149 | goto again; | |
1da177e4 | 1150 | } |
db14fc3a MS |
1151 | |
1152 | /* | |
1153 | * Search for at least 1 mount point in the dentry's subdirs. | |
1154 | * We descend to the next level whenever the d_subdirs | |
1155 | * list is non-empty and continue searching. | |
1156 | */ | |
1157 | ||
1158 | /** | |
1159 | * have_submounts - check for mounts over a dentry | |
1160 | * @parent: dentry to check. | |
1161 | * | |
1162 | * Return true if the parent or its subdirectories contain | |
1163 | * a mount point | |
1164 | */ | |
1165 | ||
1166 | static enum d_walk_ret check_mount(void *data, struct dentry *dentry) | |
1167 | { | |
1168 | int *ret = data; | |
1169 | if (d_mountpoint(dentry)) { | |
1170 | *ret = 1; | |
1171 | return D_WALK_QUIT; | |
1172 | } | |
1173 | return D_WALK_CONTINUE; | |
1174 | } | |
1175 | ||
1176 | int have_submounts(struct dentry *parent) | |
1177 | { | |
1178 | int ret = 0; | |
1179 | ||
1180 | d_walk(parent, &ret, check_mount, NULL); | |
1181 | ||
1182 | return ret; | |
1183 | } | |
ec4f8605 | 1184 | EXPORT_SYMBOL(have_submounts); |
1da177e4 | 1185 | |
eed81007 MS |
1186 | /* |
1187 | * Called by mount code to set a mountpoint and check if the mountpoint is | |
1188 | * reachable (e.g. NFS can unhash a directory dentry and then the complete | |
1189 | * subtree can become unreachable). | |
1190 | * | |
1191 | * Only one of check_submounts_and_drop() and d_set_mounted() must succeed. For | |
1192 | * this reason take rename_lock and d_lock on dentry and ancestors. | |
1193 | */ | |
1194 | int d_set_mounted(struct dentry *dentry) | |
1195 | { | |
1196 | struct dentry *p; | |
1197 | int ret = -ENOENT; | |
1198 | write_seqlock(&rename_lock); | |
1199 | for (p = dentry->d_parent; !IS_ROOT(p); p = p->d_parent) { | |
1200 | /* Need exclusion wrt. check_submounts_and_drop() */ | |
1201 | spin_lock(&p->d_lock); | |
1202 | if (unlikely(d_unhashed(p))) { | |
1203 | spin_unlock(&p->d_lock); | |
1204 | goto out; | |
1205 | } | |
1206 | spin_unlock(&p->d_lock); | |
1207 | } | |
1208 | spin_lock(&dentry->d_lock); | |
1209 | if (!d_unlinked(dentry)) { | |
1210 | dentry->d_flags |= DCACHE_MOUNTED; | |
1211 | ret = 0; | |
1212 | } | |
1213 | spin_unlock(&dentry->d_lock); | |
1214 | out: | |
1215 | write_sequnlock(&rename_lock); | |
1216 | return ret; | |
1217 | } | |
1218 | ||
1da177e4 | 1219 | /* |
fd517909 | 1220 | * Search the dentry child list of the specified parent, |
1da177e4 LT |
1221 | * and move any unused dentries to the end of the unused |
1222 | * list for prune_dcache(). We descend to the next level | |
1223 | * whenever the d_subdirs list is non-empty and continue | |
1224 | * searching. | |
1225 | * | |
1226 | * It returns zero iff there are no unused children, | |
1227 | * otherwise it returns the number of children moved to | |
1228 | * the end of the unused list. This may not be the total | |
1229 | * number of unused children, because select_parent can | |
1230 | * drop the lock and return early due to latency | |
1231 | * constraints. | |
1232 | */ | |
1da177e4 | 1233 | |
db14fc3a MS |
1234 | struct select_data { |
1235 | struct dentry *start; | |
1236 | struct list_head dispose; | |
1237 | int found; | |
1238 | }; | |
23044507 | 1239 | |
db14fc3a MS |
1240 | static enum d_walk_ret select_collect(void *_data, struct dentry *dentry) |
1241 | { | |
1242 | struct select_data *data = _data; | |
1243 | enum d_walk_ret ret = D_WALK_CONTINUE; | |
1da177e4 | 1244 | |
db14fc3a MS |
1245 | if (data->start == dentry) |
1246 | goto out; | |
2fd6b7f5 | 1247 | |
1da177e4 | 1248 | /* |
db14fc3a MS |
1249 | * move only zero ref count dentries to the dispose list. |
1250 | * | |
1251 | * Those which are presently on the shrink list, being processed | |
1252 | * by shrink_dentry_list(), shouldn't be moved. Otherwise the | |
1253 | * loop in shrink_dcache_parent() might not make any progress | |
1254 | * and loop forever. | |
1da177e4 | 1255 | */ |
db14fc3a MS |
1256 | if (dentry->d_lockref.count) { |
1257 | dentry_lru_del(dentry); | |
1258 | } else if (!(dentry->d_flags & DCACHE_SHRINK_LIST)) { | |
1259 | dentry_lru_move_list(dentry, &data->dispose); | |
1260 | dentry->d_flags |= DCACHE_SHRINK_LIST; | |
1261 | data->found++; | |
1262 | ret = D_WALK_NORETRY; | |
1da177e4 | 1263 | } |
db14fc3a MS |
1264 | /* |
1265 | * We can return to the caller if we have found some (this | |
1266 | * ensures forward progress). We'll be coming back to find | |
1267 | * the rest. | |
1268 | */ | |
1269 | if (data->found && need_resched()) | |
1270 | ret = D_WALK_QUIT; | |
1da177e4 | 1271 | out: |
db14fc3a | 1272 | return ret; |
1da177e4 LT |
1273 | } |
1274 | ||
1275 | /** | |
1276 | * shrink_dcache_parent - prune dcache | |
1277 | * @parent: parent of entries to prune | |
1278 | * | |
1279 | * Prune the dcache to remove unused children of the parent dentry. | |
1280 | */ | |
db14fc3a | 1281 | void shrink_dcache_parent(struct dentry *parent) |
1da177e4 | 1282 | { |
db14fc3a MS |
1283 | for (;;) { |
1284 | struct select_data data; | |
1da177e4 | 1285 | |
db14fc3a MS |
1286 | INIT_LIST_HEAD(&data.dispose); |
1287 | data.start = parent; | |
1288 | data.found = 0; | |
1289 | ||
1290 | d_walk(parent, &data, select_collect, NULL); | |
1291 | if (!data.found) | |
1292 | break; | |
1293 | ||
1294 | shrink_dentry_list(&data.dispose); | |
421348f1 GT |
1295 | cond_resched(); |
1296 | } | |
1da177e4 | 1297 | } |
ec4f8605 | 1298 | EXPORT_SYMBOL(shrink_dcache_parent); |
1da177e4 | 1299 | |
848ac114 MS |
1300 | static enum d_walk_ret check_and_collect(void *_data, struct dentry *dentry) |
1301 | { | |
1302 | struct select_data *data = _data; | |
1303 | ||
1304 | if (d_mountpoint(dentry)) { | |
1305 | data->found = -EBUSY; | |
1306 | return D_WALK_QUIT; | |
1307 | } | |
1308 | ||
1309 | return select_collect(_data, dentry); | |
1310 | } | |
1311 | ||
1312 | static void check_and_drop(void *_data) | |
1313 | { | |
1314 | struct select_data *data = _data; | |
1315 | ||
1316 | if (d_mountpoint(data->start)) | |
1317 | data->found = -EBUSY; | |
1318 | if (!data->found) | |
1319 | __d_drop(data->start); | |
1320 | } | |
1321 | ||
1322 | /** | |
1323 | * check_submounts_and_drop - prune dcache, check for submounts and drop | |
1324 | * | |
1325 | * All done as a single atomic operation relative to has_unlinked_ancestor(). | |
1326 | * Returns 0 if successfully unhashed @parent. If there were submounts then | |
1327 | * return -EBUSY. | |
1328 | * | |
1329 | * @dentry: dentry to prune and drop | |
1330 | */ | |
1331 | int check_submounts_and_drop(struct dentry *dentry) | |
1332 | { | |
1333 | int ret = 0; | |
1334 | ||
1335 | /* Negative dentries can be dropped without further checks */ | |
1336 | if (!dentry->d_inode) { | |
1337 | d_drop(dentry); | |
1338 | goto out; | |
1339 | } | |
1340 | ||
1341 | for (;;) { | |
1342 | struct select_data data; | |
1343 | ||
1344 | INIT_LIST_HEAD(&data.dispose); | |
1345 | data.start = dentry; | |
1346 | data.found = 0; | |
1347 | ||
1348 | d_walk(dentry, &data, check_and_collect, check_and_drop); | |
1349 | ret = data.found; | |
1350 | ||
1351 | if (!list_empty(&data.dispose)) | |
1352 | shrink_dentry_list(&data.dispose); | |
1353 | ||
1354 | if (ret <= 0) | |
1355 | break; | |
1356 | ||
1357 | cond_resched(); | |
1358 | } | |
1359 | ||
1360 | out: | |
1361 | return ret; | |
1362 | } | |
1363 | EXPORT_SYMBOL(check_submounts_and_drop); | |
1364 | ||
1da177e4 | 1365 | /** |
a4464dbc AV |
1366 | * __d_alloc - allocate a dcache entry |
1367 | * @sb: filesystem it will belong to | |
1da177e4 LT |
1368 | * @name: qstr of the name |
1369 | * | |
1370 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1371 | * available. On a success the dentry is returned. The name passed in is | |
1372 | * copied and the copy passed in may be reused after this call. | |
1373 | */ | |
1374 | ||
a4464dbc | 1375 | struct dentry *__d_alloc(struct super_block *sb, const struct qstr *name) |
1da177e4 LT |
1376 | { |
1377 | struct dentry *dentry; | |
1378 | char *dname; | |
1379 | ||
e12ba74d | 1380 | dentry = kmem_cache_alloc(dentry_cache, GFP_KERNEL); |
1da177e4 LT |
1381 | if (!dentry) |
1382 | return NULL; | |
1383 | ||
6326c71f LT |
1384 | /* |
1385 | * We guarantee that the inline name is always NUL-terminated. | |
1386 | * This way the memcpy() done by the name switching in rename | |
1387 | * will still always have a NUL at the end, even if we might | |
1388 | * be overwriting an internal NUL character | |
1389 | */ | |
1390 | dentry->d_iname[DNAME_INLINE_LEN-1] = 0; | |
1da177e4 LT |
1391 | if (name->len > DNAME_INLINE_LEN-1) { |
1392 | dname = kmalloc(name->len + 1, GFP_KERNEL); | |
1393 | if (!dname) { | |
1394 | kmem_cache_free(dentry_cache, dentry); | |
1395 | return NULL; | |
1396 | } | |
1397 | } else { | |
1398 | dname = dentry->d_iname; | |
1399 | } | |
1da177e4 LT |
1400 | |
1401 | dentry->d_name.len = name->len; | |
1402 | dentry->d_name.hash = name->hash; | |
1403 | memcpy(dname, name->name, name->len); | |
1404 | dname[name->len] = 0; | |
1405 | ||
6326c71f LT |
1406 | /* Make sure we always see the terminating NUL character */ |
1407 | smp_wmb(); | |
1408 | dentry->d_name.name = dname; | |
1409 | ||
98474236 | 1410 | dentry->d_lockref.count = 1; |
dea3667b | 1411 | dentry->d_flags = 0; |
1da177e4 | 1412 | spin_lock_init(&dentry->d_lock); |
31e6b01f | 1413 | seqcount_init(&dentry->d_seq); |
1da177e4 | 1414 | dentry->d_inode = NULL; |
a4464dbc AV |
1415 | dentry->d_parent = dentry; |
1416 | dentry->d_sb = sb; | |
1da177e4 LT |
1417 | dentry->d_op = NULL; |
1418 | dentry->d_fsdata = NULL; | |
ceb5bdc2 | 1419 | INIT_HLIST_BL_NODE(&dentry->d_hash); |
1da177e4 LT |
1420 | INIT_LIST_HEAD(&dentry->d_lru); |
1421 | INIT_LIST_HEAD(&dentry->d_subdirs); | |
b3d9b7a3 | 1422 | INIT_HLIST_NODE(&dentry->d_alias); |
2fd6b7f5 | 1423 | INIT_LIST_HEAD(&dentry->d_u.d_child); |
a4464dbc | 1424 | d_set_d_op(dentry, dentry->d_sb->s_d_op); |
1da177e4 | 1425 | |
3e880fb5 | 1426 | this_cpu_inc(nr_dentry); |
312d3ca8 | 1427 | |
1da177e4 LT |
1428 | return dentry; |
1429 | } | |
a4464dbc AV |
1430 | |
1431 | /** | |
1432 | * d_alloc - allocate a dcache entry | |
1433 | * @parent: parent of entry to allocate | |
1434 | * @name: qstr of the name | |
1435 | * | |
1436 | * Allocates a dentry. It returns %NULL if there is insufficient memory | |
1437 | * available. On a success the dentry is returned. The name passed in is | |
1438 | * copied and the copy passed in may be reused after this call. | |
1439 | */ | |
1440 | struct dentry *d_alloc(struct dentry * parent, const struct qstr *name) | |
1441 | { | |
1442 | struct dentry *dentry = __d_alloc(parent->d_sb, name); | |
1443 | if (!dentry) | |
1444 | return NULL; | |
1445 | ||
1446 | spin_lock(&parent->d_lock); | |
1447 | /* | |
1448 | * don't need child lock because it is not subject | |
1449 | * to concurrency here | |
1450 | */ | |
1451 | __dget_dlock(parent); | |
1452 | dentry->d_parent = parent; | |
1453 | list_add(&dentry->d_u.d_child, &parent->d_subdirs); | |
1454 | spin_unlock(&parent->d_lock); | |
1455 | ||
1456 | return dentry; | |
1457 | } | |
ec4f8605 | 1458 | EXPORT_SYMBOL(d_alloc); |
1da177e4 | 1459 | |
4b936885 NP |
1460 | struct dentry *d_alloc_pseudo(struct super_block *sb, const struct qstr *name) |
1461 | { | |
a4464dbc AV |
1462 | struct dentry *dentry = __d_alloc(sb, name); |
1463 | if (dentry) | |
4b936885 | 1464 | dentry->d_flags |= DCACHE_DISCONNECTED; |
4b936885 NP |
1465 | return dentry; |
1466 | } | |
1467 | EXPORT_SYMBOL(d_alloc_pseudo); | |
1468 | ||
1da177e4 LT |
1469 | struct dentry *d_alloc_name(struct dentry *parent, const char *name) |
1470 | { | |
1471 | struct qstr q; | |
1472 | ||
1473 | q.name = name; | |
1474 | q.len = strlen(name); | |
1475 | q.hash = full_name_hash(q.name, q.len); | |
1476 | return d_alloc(parent, &q); | |
1477 | } | |
ef26ca97 | 1478 | EXPORT_SYMBOL(d_alloc_name); |
1da177e4 | 1479 | |
fb045adb NP |
1480 | void d_set_d_op(struct dentry *dentry, const struct dentry_operations *op) |
1481 | { | |
6f7f7caa LT |
1482 | WARN_ON_ONCE(dentry->d_op); |
1483 | WARN_ON_ONCE(dentry->d_flags & (DCACHE_OP_HASH | | |
fb045adb NP |
1484 | DCACHE_OP_COMPARE | |
1485 | DCACHE_OP_REVALIDATE | | |
ecf3d1f1 | 1486 | DCACHE_OP_WEAK_REVALIDATE | |
fb045adb NP |
1487 | DCACHE_OP_DELETE )); |
1488 | dentry->d_op = op; | |
1489 | if (!op) | |
1490 | return; | |
1491 | if (op->d_hash) | |
1492 | dentry->d_flags |= DCACHE_OP_HASH; | |
1493 | if (op->d_compare) | |
1494 | dentry->d_flags |= DCACHE_OP_COMPARE; | |
1495 | if (op->d_revalidate) | |
1496 | dentry->d_flags |= DCACHE_OP_REVALIDATE; | |
ecf3d1f1 JL |
1497 | if (op->d_weak_revalidate) |
1498 | dentry->d_flags |= DCACHE_OP_WEAK_REVALIDATE; | |
fb045adb NP |
1499 | if (op->d_delete) |
1500 | dentry->d_flags |= DCACHE_OP_DELETE; | |
f0023bc6 SW |
1501 | if (op->d_prune) |
1502 | dentry->d_flags |= DCACHE_OP_PRUNE; | |
fb045adb NP |
1503 | |
1504 | } | |
1505 | EXPORT_SYMBOL(d_set_d_op); | |
1506 | ||
360da900 OH |
1507 | static void __d_instantiate(struct dentry *dentry, struct inode *inode) |
1508 | { | |
b23fb0a6 | 1509 | spin_lock(&dentry->d_lock); |
9875cf80 DH |
1510 | if (inode) { |
1511 | if (unlikely(IS_AUTOMOUNT(inode))) | |
1512 | dentry->d_flags |= DCACHE_NEED_AUTOMOUNT; | |
b3d9b7a3 | 1513 | hlist_add_head(&dentry->d_alias, &inode->i_dentry); |
9875cf80 | 1514 | } |
360da900 | 1515 | dentry->d_inode = inode; |
31e6b01f | 1516 | dentry_rcuwalk_barrier(dentry); |
b23fb0a6 | 1517 | spin_unlock(&dentry->d_lock); |
360da900 OH |
1518 | fsnotify_d_instantiate(dentry, inode); |
1519 | } | |
1520 | ||
1da177e4 LT |
1521 | /** |
1522 | * d_instantiate - fill in inode information for a dentry | |
1523 | * @entry: dentry to complete | |
1524 | * @inode: inode to attach to this dentry | |
1525 | * | |
1526 | * Fill in inode information in the entry. | |
1527 | * | |
1528 | * This turns negative dentries into productive full members | |
1529 | * of society. | |
1530 | * | |
1531 | * NOTE! This assumes that the inode count has been incremented | |
1532 | * (or otherwise set) by the caller to indicate that it is now | |
1533 | * in use by the dcache. | |
1534 | */ | |
1535 | ||
1536 | void d_instantiate(struct dentry *entry, struct inode * inode) | |
1537 | { | |
b3d9b7a3 | 1538 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
873feea0 NP |
1539 | if (inode) |
1540 | spin_lock(&inode->i_lock); | |
360da900 | 1541 | __d_instantiate(entry, inode); |
873feea0 NP |
1542 | if (inode) |
1543 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
1544 | security_d_instantiate(entry, inode); |
1545 | } | |
ec4f8605 | 1546 | EXPORT_SYMBOL(d_instantiate); |
1da177e4 LT |
1547 | |
1548 | /** | |
1549 | * d_instantiate_unique - instantiate a non-aliased dentry | |
1550 | * @entry: dentry to instantiate | |
1551 | * @inode: inode to attach to this dentry | |
1552 | * | |
1553 | * Fill in inode information in the entry. On success, it returns NULL. | |
1554 | * If an unhashed alias of "entry" already exists, then we return the | |
e866cfa9 | 1555 | * aliased dentry instead and drop one reference to inode. |
1da177e4 LT |
1556 | * |
1557 | * Note that in order to avoid conflicts with rename() etc, the caller | |
1558 | * had better be holding the parent directory semaphore. | |
e866cfa9 OD |
1559 | * |
1560 | * This also assumes that the inode count has been incremented | |
1561 | * (or otherwise set) by the caller to indicate that it is now | |
1562 | * in use by the dcache. | |
1da177e4 | 1563 | */ |
770bfad8 DH |
1564 | static struct dentry *__d_instantiate_unique(struct dentry *entry, |
1565 | struct inode *inode) | |
1da177e4 LT |
1566 | { |
1567 | struct dentry *alias; | |
1568 | int len = entry->d_name.len; | |
1569 | const char *name = entry->d_name.name; | |
1570 | unsigned int hash = entry->d_name.hash; | |
1571 | ||
770bfad8 | 1572 | if (!inode) { |
360da900 | 1573 | __d_instantiate(entry, NULL); |
770bfad8 DH |
1574 | return NULL; |
1575 | } | |
1576 | ||
b67bfe0d | 1577 | hlist_for_each_entry(alias, &inode->i_dentry, d_alias) { |
9abca360 NP |
1578 | /* |
1579 | * Don't need alias->d_lock here, because aliases with | |
1580 | * d_parent == entry->d_parent are not subject to name or | |
1581 | * parent changes, because the parent inode i_mutex is held. | |
1582 | */ | |
12f8ad4b | 1583 | if (alias->d_name.hash != hash) |
1da177e4 LT |
1584 | continue; |
1585 | if (alias->d_parent != entry->d_parent) | |
1586 | continue; | |
ee983e89 LT |
1587 | if (alias->d_name.len != len) |
1588 | continue; | |
12f8ad4b | 1589 | if (dentry_cmp(alias, name, len)) |
1da177e4 | 1590 | continue; |
dc0474be | 1591 | __dget(alias); |
1da177e4 LT |
1592 | return alias; |
1593 | } | |
770bfad8 | 1594 | |
360da900 | 1595 | __d_instantiate(entry, inode); |
1da177e4 LT |
1596 | return NULL; |
1597 | } | |
770bfad8 DH |
1598 | |
1599 | struct dentry *d_instantiate_unique(struct dentry *entry, struct inode *inode) | |
1600 | { | |
1601 | struct dentry *result; | |
1602 | ||
b3d9b7a3 | 1603 | BUG_ON(!hlist_unhashed(&entry->d_alias)); |
770bfad8 | 1604 | |
873feea0 NP |
1605 | if (inode) |
1606 | spin_lock(&inode->i_lock); | |
770bfad8 | 1607 | result = __d_instantiate_unique(entry, inode); |
873feea0 NP |
1608 | if (inode) |
1609 | spin_unlock(&inode->i_lock); | |
770bfad8 DH |
1610 | |
1611 | if (!result) { | |
1612 | security_d_instantiate(entry, inode); | |
1613 | return NULL; | |
1614 | } | |
1615 | ||
1616 | BUG_ON(!d_unhashed(result)); | |
1617 | iput(inode); | |
1618 | return result; | |
1619 | } | |
1620 | ||
1da177e4 LT |
1621 | EXPORT_SYMBOL(d_instantiate_unique); |
1622 | ||
adc0e91a AV |
1623 | struct dentry *d_make_root(struct inode *root_inode) |
1624 | { | |
1625 | struct dentry *res = NULL; | |
1626 | ||
1627 | if (root_inode) { | |
26fe5750 | 1628 | static const struct qstr name = QSTR_INIT("/", 1); |
adc0e91a AV |
1629 | |
1630 | res = __d_alloc(root_inode->i_sb, &name); | |
1631 | if (res) | |
1632 | d_instantiate(res, root_inode); | |
1633 | else | |
1634 | iput(root_inode); | |
1635 | } | |
1636 | return res; | |
1637 | } | |
1638 | EXPORT_SYMBOL(d_make_root); | |
1639 | ||
d891eedb BF |
1640 | static struct dentry * __d_find_any_alias(struct inode *inode) |
1641 | { | |
1642 | struct dentry *alias; | |
1643 | ||
b3d9b7a3 | 1644 | if (hlist_empty(&inode->i_dentry)) |
d891eedb | 1645 | return NULL; |
b3d9b7a3 | 1646 | alias = hlist_entry(inode->i_dentry.first, struct dentry, d_alias); |
d891eedb BF |
1647 | __dget(alias); |
1648 | return alias; | |
1649 | } | |
1650 | ||
46f72b34 SW |
1651 | /** |
1652 | * d_find_any_alias - find any alias for a given inode | |
1653 | * @inode: inode to find an alias for | |
1654 | * | |
1655 | * If any aliases exist for the given inode, take and return a | |
1656 | * reference for one of them. If no aliases exist, return %NULL. | |
1657 | */ | |
1658 | struct dentry *d_find_any_alias(struct inode *inode) | |
d891eedb BF |
1659 | { |
1660 | struct dentry *de; | |
1661 | ||
1662 | spin_lock(&inode->i_lock); | |
1663 | de = __d_find_any_alias(inode); | |
1664 | spin_unlock(&inode->i_lock); | |
1665 | return de; | |
1666 | } | |
46f72b34 | 1667 | EXPORT_SYMBOL(d_find_any_alias); |
d891eedb | 1668 | |
4ea3ada2 CH |
1669 | /** |
1670 | * d_obtain_alias - find or allocate a dentry for a given inode | |
1671 | * @inode: inode to allocate the dentry for | |
1672 | * | |
1673 | * Obtain a dentry for an inode resulting from NFS filehandle conversion or | |
1674 | * similar open by handle operations. The returned dentry may be anonymous, | |
1675 | * or may have a full name (if the inode was already in the cache). | |
1676 | * | |
1677 | * When called on a directory inode, we must ensure that the inode only ever | |
1678 | * has one dentry. If a dentry is found, that is returned instead of | |
1679 | * allocating a new one. | |
1680 | * | |
1681 | * On successful return, the reference to the inode has been transferred | |
44003728 CH |
1682 | * to the dentry. In case of an error the reference on the inode is released. |
1683 | * To make it easier to use in export operations a %NULL or IS_ERR inode may | |
1684 | * be passed in and will be the error will be propagate to the return value, | |
1685 | * with a %NULL @inode replaced by ERR_PTR(-ESTALE). | |
4ea3ada2 CH |
1686 | */ |
1687 | struct dentry *d_obtain_alias(struct inode *inode) | |
1688 | { | |
b911a6bd | 1689 | static const struct qstr anonstring = QSTR_INIT("/", 1); |
9308a612 CH |
1690 | struct dentry *tmp; |
1691 | struct dentry *res; | |
4ea3ada2 CH |
1692 | |
1693 | if (!inode) | |
44003728 | 1694 | return ERR_PTR(-ESTALE); |
4ea3ada2 CH |
1695 | if (IS_ERR(inode)) |
1696 | return ERR_CAST(inode); | |
1697 | ||
d891eedb | 1698 | res = d_find_any_alias(inode); |
9308a612 CH |
1699 | if (res) |
1700 | goto out_iput; | |
1701 | ||
a4464dbc | 1702 | tmp = __d_alloc(inode->i_sb, &anonstring); |
9308a612 CH |
1703 | if (!tmp) { |
1704 | res = ERR_PTR(-ENOMEM); | |
1705 | goto out_iput; | |
4ea3ada2 | 1706 | } |
b5c84bf6 | 1707 | |
873feea0 | 1708 | spin_lock(&inode->i_lock); |
d891eedb | 1709 | res = __d_find_any_alias(inode); |
9308a612 | 1710 | if (res) { |
873feea0 | 1711 | spin_unlock(&inode->i_lock); |
9308a612 CH |
1712 | dput(tmp); |
1713 | goto out_iput; | |
1714 | } | |
1715 | ||
1716 | /* attach a disconnected dentry */ | |
1717 | spin_lock(&tmp->d_lock); | |
9308a612 CH |
1718 | tmp->d_inode = inode; |
1719 | tmp->d_flags |= DCACHE_DISCONNECTED; | |
b3d9b7a3 | 1720 | hlist_add_head(&tmp->d_alias, &inode->i_dentry); |
1879fd6a | 1721 | hlist_bl_lock(&tmp->d_sb->s_anon); |
ceb5bdc2 | 1722 | hlist_bl_add_head(&tmp->d_hash, &tmp->d_sb->s_anon); |
1879fd6a | 1723 | hlist_bl_unlock(&tmp->d_sb->s_anon); |
9308a612 | 1724 | spin_unlock(&tmp->d_lock); |
873feea0 | 1725 | spin_unlock(&inode->i_lock); |
24ff6663 | 1726 | security_d_instantiate(tmp, inode); |
9308a612 | 1727 | |
9308a612 CH |
1728 | return tmp; |
1729 | ||
1730 | out_iput: | |
24ff6663 JB |
1731 | if (res && !IS_ERR(res)) |
1732 | security_d_instantiate(res, inode); | |
9308a612 CH |
1733 | iput(inode); |
1734 | return res; | |
4ea3ada2 | 1735 | } |
adc48720 | 1736 | EXPORT_SYMBOL(d_obtain_alias); |
1da177e4 LT |
1737 | |
1738 | /** | |
1739 | * d_splice_alias - splice a disconnected dentry into the tree if one exists | |
1740 | * @inode: the inode which may have a disconnected dentry | |
1741 | * @dentry: a negative dentry which we want to point to the inode. | |
1742 | * | |
1743 | * If inode is a directory and has a 'disconnected' dentry (i.e. IS_ROOT and | |
1744 | * DCACHE_DISCONNECTED), then d_move that in place of the given dentry | |
1745 | * and return it, else simply d_add the inode to the dentry and return NULL. | |
1746 | * | |
1747 | * This is needed in the lookup routine of any filesystem that is exportable | |
1748 | * (via knfsd) so that we can build dcache paths to directories effectively. | |
1749 | * | |
1750 | * If a dentry was found and moved, then it is returned. Otherwise NULL | |
1751 | * is returned. This matches the expected return value of ->lookup. | |
1752 | * | |
6d4ade98 SW |
1753 | * Cluster filesystems may call this function with a negative, hashed dentry. |
1754 | * In that case, we know that the inode will be a regular file, and also this | |
1755 | * will only occur during atomic_open. So we need to check for the dentry | |
1756 | * being already hashed only in the final case. | |
1da177e4 LT |
1757 | */ |
1758 | struct dentry *d_splice_alias(struct inode *inode, struct dentry *dentry) | |
1759 | { | |
1760 | struct dentry *new = NULL; | |
1761 | ||
a9049376 AV |
1762 | if (IS_ERR(inode)) |
1763 | return ERR_CAST(inode); | |
1764 | ||
21c0d8fd | 1765 | if (inode && S_ISDIR(inode->i_mode)) { |
873feea0 | 1766 | spin_lock(&inode->i_lock); |
32ba9c3f | 1767 | new = __d_find_alias(inode, 1); |
1da177e4 | 1768 | if (new) { |
32ba9c3f | 1769 | BUG_ON(!(new->d_flags & DCACHE_DISCONNECTED)); |
873feea0 | 1770 | spin_unlock(&inode->i_lock); |
1da177e4 | 1771 | security_d_instantiate(new, inode); |
1da177e4 LT |
1772 | d_move(new, dentry); |
1773 | iput(inode); | |
1774 | } else { | |
873feea0 | 1775 | /* already taking inode->i_lock, so d_add() by hand */ |
360da900 | 1776 | __d_instantiate(dentry, inode); |
873feea0 | 1777 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
1778 | security_d_instantiate(dentry, inode); |
1779 | d_rehash(dentry); | |
1780 | } | |
6d4ade98 SW |
1781 | } else { |
1782 | d_instantiate(dentry, inode); | |
1783 | if (d_unhashed(dentry)) | |
1784 | d_rehash(dentry); | |
1785 | } | |
1da177e4 LT |
1786 | return new; |
1787 | } | |
ec4f8605 | 1788 | EXPORT_SYMBOL(d_splice_alias); |
1da177e4 | 1789 | |
9403540c BN |
1790 | /** |
1791 | * d_add_ci - lookup or allocate new dentry with case-exact name | |
1792 | * @inode: the inode case-insensitive lookup has found | |
1793 | * @dentry: the negative dentry that was passed to the parent's lookup func | |
1794 | * @name: the case-exact name to be associated with the returned dentry | |
1795 | * | |
1796 | * This is to avoid filling the dcache with case-insensitive names to the | |
1797 | * same inode, only the actual correct case is stored in the dcache for | |
1798 | * case-insensitive filesystems. | |
1799 | * | |
1800 | * For a case-insensitive lookup match and if the the case-exact dentry | |
1801 | * already exists in in the dcache, use it and return it. | |
1802 | * | |
1803 | * If no entry exists with the exact case name, allocate new dentry with | |
1804 | * the exact case, and return the spliced entry. | |
1805 | */ | |
e45b590b | 1806 | struct dentry *d_add_ci(struct dentry *dentry, struct inode *inode, |
9403540c BN |
1807 | struct qstr *name) |
1808 | { | |
9403540c BN |
1809 | struct dentry *found; |
1810 | struct dentry *new; | |
1811 | ||
b6520c81 CH |
1812 | /* |
1813 | * First check if a dentry matching the name already exists, | |
1814 | * if not go ahead and create it now. | |
1815 | */ | |
9403540c | 1816 | found = d_hash_and_lookup(dentry->d_parent, name); |
4f522a24 AV |
1817 | if (unlikely(IS_ERR(found))) |
1818 | goto err_out; | |
9403540c BN |
1819 | if (!found) { |
1820 | new = d_alloc(dentry->d_parent, name); | |
1821 | if (!new) { | |
4f522a24 | 1822 | found = ERR_PTR(-ENOMEM); |
9403540c BN |
1823 | goto err_out; |
1824 | } | |
b6520c81 | 1825 | |
9403540c BN |
1826 | found = d_splice_alias(inode, new); |
1827 | if (found) { | |
1828 | dput(new); | |
1829 | return found; | |
1830 | } | |
1831 | return new; | |
1832 | } | |
b6520c81 CH |
1833 | |
1834 | /* | |
1835 | * If a matching dentry exists, and it's not negative use it. | |
1836 | * | |
1837 | * Decrement the reference count to balance the iget() done | |
1838 | * earlier on. | |
1839 | */ | |
9403540c BN |
1840 | if (found->d_inode) { |
1841 | if (unlikely(found->d_inode != inode)) { | |
1842 | /* This can't happen because bad inodes are unhashed. */ | |
1843 | BUG_ON(!is_bad_inode(inode)); | |
1844 | BUG_ON(!is_bad_inode(found->d_inode)); | |
1845 | } | |
9403540c BN |
1846 | iput(inode); |
1847 | return found; | |
1848 | } | |
b6520c81 | 1849 | |
9403540c | 1850 | /* |
9403540c | 1851 | * Negative dentry: instantiate it unless the inode is a directory and |
b6520c81 | 1852 | * already has a dentry. |
9403540c | 1853 | */ |
4513d899 AV |
1854 | new = d_splice_alias(inode, found); |
1855 | if (new) { | |
1856 | dput(found); | |
1857 | found = new; | |
9403540c | 1858 | } |
4513d899 | 1859 | return found; |
9403540c BN |
1860 | |
1861 | err_out: | |
1862 | iput(inode); | |
4f522a24 | 1863 | return found; |
9403540c | 1864 | } |
ec4f8605 | 1865 | EXPORT_SYMBOL(d_add_ci); |
1da177e4 | 1866 | |
12f8ad4b LT |
1867 | /* |
1868 | * Do the slow-case of the dentry name compare. | |
1869 | * | |
1870 | * Unlike the dentry_cmp() function, we need to atomically | |
da53be12 | 1871 | * load the name and length information, so that the |
12f8ad4b LT |
1872 | * filesystem can rely on them, and can use the 'name' and |
1873 | * 'len' information without worrying about walking off the | |
1874 | * end of memory etc. | |
1875 | * | |
1876 | * Thus the read_seqcount_retry() and the "duplicate" info | |
1877 | * in arguments (the low-level filesystem should not look | |
1878 | * at the dentry inode or name contents directly, since | |
1879 | * rename can change them while we're in RCU mode). | |
1880 | */ | |
1881 | enum slow_d_compare { | |
1882 | D_COMP_OK, | |
1883 | D_COMP_NOMATCH, | |
1884 | D_COMP_SEQRETRY, | |
1885 | }; | |
1886 | ||
1887 | static noinline enum slow_d_compare slow_dentry_cmp( | |
1888 | const struct dentry *parent, | |
12f8ad4b LT |
1889 | struct dentry *dentry, |
1890 | unsigned int seq, | |
1891 | const struct qstr *name) | |
1892 | { | |
1893 | int tlen = dentry->d_name.len; | |
1894 | const char *tname = dentry->d_name.name; | |
12f8ad4b LT |
1895 | |
1896 | if (read_seqcount_retry(&dentry->d_seq, seq)) { | |
1897 | cpu_relax(); | |
1898 | return D_COMP_SEQRETRY; | |
1899 | } | |
da53be12 | 1900 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
12f8ad4b LT |
1901 | return D_COMP_NOMATCH; |
1902 | return D_COMP_OK; | |
1903 | } | |
1904 | ||
31e6b01f NP |
1905 | /** |
1906 | * __d_lookup_rcu - search for a dentry (racy, store-free) | |
1907 | * @parent: parent dentry | |
1908 | * @name: qstr of name we wish to find | |
1f1e6e52 | 1909 | * @seqp: returns d_seq value at the point where the dentry was found |
31e6b01f NP |
1910 | * Returns: dentry, or NULL |
1911 | * | |
1912 | * __d_lookup_rcu is the dcache lookup function for rcu-walk name | |
1913 | * resolution (store-free path walking) design described in | |
1914 | * Documentation/filesystems/path-lookup.txt. | |
1915 | * | |
1916 | * This is not to be used outside core vfs. | |
1917 | * | |
1918 | * __d_lookup_rcu must only be used in rcu-walk mode, ie. with vfsmount lock | |
1919 | * held, and rcu_read_lock held. The returned dentry must not be stored into | |
1920 | * without taking d_lock and checking d_seq sequence count against @seq | |
1921 | * returned here. | |
1922 | * | |
15570086 | 1923 | * A refcount may be taken on the found dentry with the d_rcu_to_refcount |
31e6b01f NP |
1924 | * function. |
1925 | * | |
1926 | * Alternatively, __d_lookup_rcu may be called again to look up the child of | |
1927 | * the returned dentry, so long as its parent's seqlock is checked after the | |
1928 | * child is looked up. Thus, an interlocking stepping of sequence lock checks | |
1929 | * is formed, giving integrity down the path walk. | |
12f8ad4b LT |
1930 | * |
1931 | * NOTE! The caller *has* to check the resulting dentry against the sequence | |
1932 | * number we've returned before using any of the resulting dentry state! | |
31e6b01f | 1933 | */ |
8966be90 LT |
1934 | struct dentry *__d_lookup_rcu(const struct dentry *parent, |
1935 | const struct qstr *name, | |
da53be12 | 1936 | unsigned *seqp) |
31e6b01f | 1937 | { |
26fe5750 | 1938 | u64 hashlen = name->hash_len; |
31e6b01f | 1939 | const unsigned char *str = name->name; |
26fe5750 | 1940 | struct hlist_bl_head *b = d_hash(parent, hashlen_hash(hashlen)); |
ceb5bdc2 | 1941 | struct hlist_bl_node *node; |
31e6b01f NP |
1942 | struct dentry *dentry; |
1943 | ||
1944 | /* | |
1945 | * Note: There is significant duplication with __d_lookup_rcu which is | |
1946 | * required to prevent single threaded performance regressions | |
1947 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
1948 | * Keep the two functions in sync. | |
1949 | */ | |
1950 | ||
1951 | /* | |
1952 | * The hash list is protected using RCU. | |
1953 | * | |
1954 | * Carefully use d_seq when comparing a candidate dentry, to avoid | |
1955 | * races with d_move(). | |
1956 | * | |
1957 | * It is possible that concurrent renames can mess up our list | |
1958 | * walk here and result in missing our dentry, resulting in the | |
1959 | * false-negative result. d_lookup() protects against concurrent | |
1960 | * renames using rename_lock seqlock. | |
1961 | * | |
b0a4bb83 | 1962 | * See Documentation/filesystems/path-lookup.txt for more details. |
31e6b01f | 1963 | */ |
b07ad996 | 1964 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
8966be90 | 1965 | unsigned seq; |
31e6b01f | 1966 | |
31e6b01f | 1967 | seqretry: |
12f8ad4b LT |
1968 | /* |
1969 | * The dentry sequence count protects us from concurrent | |
da53be12 | 1970 | * renames, and thus protects parent and name fields. |
12f8ad4b LT |
1971 | * |
1972 | * The caller must perform a seqcount check in order | |
da53be12 | 1973 | * to do anything useful with the returned dentry. |
12f8ad4b LT |
1974 | * |
1975 | * NOTE! We do a "raw" seqcount_begin here. That means that | |
1976 | * we don't wait for the sequence count to stabilize if it | |
1977 | * is in the middle of a sequence change. If we do the slow | |
1978 | * dentry compare, we will do seqretries until it is stable, | |
1979 | * and if we end up with a successful lookup, we actually | |
1980 | * want to exit RCU lookup anyway. | |
1981 | */ | |
1982 | seq = raw_seqcount_begin(&dentry->d_seq); | |
31e6b01f NP |
1983 | if (dentry->d_parent != parent) |
1984 | continue; | |
2e321806 LT |
1985 | if (d_unhashed(dentry)) |
1986 | continue; | |
12f8ad4b | 1987 | |
830c0f0e | 1988 | if (unlikely(parent->d_flags & DCACHE_OP_COMPARE)) { |
26fe5750 LT |
1989 | if (dentry->d_name.hash != hashlen_hash(hashlen)) |
1990 | continue; | |
da53be12 LT |
1991 | *seqp = seq; |
1992 | switch (slow_dentry_cmp(parent, dentry, seq, name)) { | |
12f8ad4b LT |
1993 | case D_COMP_OK: |
1994 | return dentry; | |
1995 | case D_COMP_NOMATCH: | |
31e6b01f | 1996 | continue; |
12f8ad4b LT |
1997 | default: |
1998 | goto seqretry; | |
1999 | } | |
31e6b01f | 2000 | } |
12f8ad4b | 2001 | |
26fe5750 | 2002 | if (dentry->d_name.hash_len != hashlen) |
ee983e89 | 2003 | continue; |
da53be12 | 2004 | *seqp = seq; |
26fe5750 | 2005 | if (!dentry_cmp(dentry, str, hashlen_len(hashlen))) |
12f8ad4b | 2006 | return dentry; |
31e6b01f NP |
2007 | } |
2008 | return NULL; | |
2009 | } | |
2010 | ||
1da177e4 LT |
2011 | /** |
2012 | * d_lookup - search for a dentry | |
2013 | * @parent: parent dentry | |
2014 | * @name: qstr of name we wish to find | |
b04f784e | 2015 | * Returns: dentry, or NULL |
1da177e4 | 2016 | * |
b04f784e NP |
2017 | * d_lookup searches the children of the parent dentry for the name in |
2018 | * question. If the dentry is found its reference count is incremented and the | |
2019 | * dentry is returned. The caller must use dput to free the entry when it has | |
2020 | * finished using it. %NULL is returned if the dentry does not exist. | |
1da177e4 | 2021 | */ |
da2d8455 | 2022 | struct dentry *d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 | 2023 | { |
31e6b01f | 2024 | struct dentry *dentry; |
949854d0 | 2025 | unsigned seq; |
1da177e4 LT |
2026 | |
2027 | do { | |
2028 | seq = read_seqbegin(&rename_lock); | |
2029 | dentry = __d_lookup(parent, name); | |
2030 | if (dentry) | |
2031 | break; | |
2032 | } while (read_seqretry(&rename_lock, seq)); | |
2033 | return dentry; | |
2034 | } | |
ec4f8605 | 2035 | EXPORT_SYMBOL(d_lookup); |
1da177e4 | 2036 | |
31e6b01f | 2037 | /** |
b04f784e NP |
2038 | * __d_lookup - search for a dentry (racy) |
2039 | * @parent: parent dentry | |
2040 | * @name: qstr of name we wish to find | |
2041 | * Returns: dentry, or NULL | |
2042 | * | |
2043 | * __d_lookup is like d_lookup, however it may (rarely) return a | |
2044 | * false-negative result due to unrelated rename activity. | |
2045 | * | |
2046 | * __d_lookup is slightly faster by avoiding rename_lock read seqlock, | |
2047 | * however it must be used carefully, eg. with a following d_lookup in | |
2048 | * the case of failure. | |
2049 | * | |
2050 | * __d_lookup callers must be commented. | |
2051 | */ | |
a713ca2a | 2052 | struct dentry *__d_lookup(const struct dentry *parent, const struct qstr *name) |
1da177e4 LT |
2053 | { |
2054 | unsigned int len = name->len; | |
2055 | unsigned int hash = name->hash; | |
2056 | const unsigned char *str = name->name; | |
b07ad996 | 2057 | struct hlist_bl_head *b = d_hash(parent, hash); |
ceb5bdc2 | 2058 | struct hlist_bl_node *node; |
31e6b01f | 2059 | struct dentry *found = NULL; |
665a7583 | 2060 | struct dentry *dentry; |
1da177e4 | 2061 | |
31e6b01f NP |
2062 | /* |
2063 | * Note: There is significant duplication with __d_lookup_rcu which is | |
2064 | * required to prevent single threaded performance regressions | |
2065 | * especially on architectures where smp_rmb (in seqcounts) are costly. | |
2066 | * Keep the two functions in sync. | |
2067 | */ | |
2068 | ||
b04f784e NP |
2069 | /* |
2070 | * The hash list is protected using RCU. | |
2071 | * | |
2072 | * Take d_lock when comparing a candidate dentry, to avoid races | |
2073 | * with d_move(). | |
2074 | * | |
2075 | * It is possible that concurrent renames can mess up our list | |
2076 | * walk here and result in missing our dentry, resulting in the | |
2077 | * false-negative result. d_lookup() protects against concurrent | |
2078 | * renames using rename_lock seqlock. | |
2079 | * | |
b0a4bb83 | 2080 | * See Documentation/filesystems/path-lookup.txt for more details. |
b04f784e | 2081 | */ |
1da177e4 LT |
2082 | rcu_read_lock(); |
2083 | ||
b07ad996 | 2084 | hlist_bl_for_each_entry_rcu(dentry, node, b, d_hash) { |
1da177e4 | 2085 | |
1da177e4 LT |
2086 | if (dentry->d_name.hash != hash) |
2087 | continue; | |
1da177e4 LT |
2088 | |
2089 | spin_lock(&dentry->d_lock); | |
1da177e4 LT |
2090 | if (dentry->d_parent != parent) |
2091 | goto next; | |
d0185c08 LT |
2092 | if (d_unhashed(dentry)) |
2093 | goto next; | |
2094 | ||
1da177e4 LT |
2095 | /* |
2096 | * It is safe to compare names since d_move() cannot | |
2097 | * change the qstr (protected by d_lock). | |
2098 | */ | |
fb045adb | 2099 | if (parent->d_flags & DCACHE_OP_COMPARE) { |
12f8ad4b LT |
2100 | int tlen = dentry->d_name.len; |
2101 | const char *tname = dentry->d_name.name; | |
da53be12 | 2102 | if (parent->d_op->d_compare(parent, dentry, tlen, tname, name)) |
1da177e4 LT |
2103 | goto next; |
2104 | } else { | |
ee983e89 LT |
2105 | if (dentry->d_name.len != len) |
2106 | goto next; | |
12f8ad4b | 2107 | if (dentry_cmp(dentry, str, len)) |
1da177e4 LT |
2108 | goto next; |
2109 | } | |
2110 | ||
98474236 | 2111 | dentry->d_lockref.count++; |
d0185c08 | 2112 | found = dentry; |
1da177e4 LT |
2113 | spin_unlock(&dentry->d_lock); |
2114 | break; | |
2115 | next: | |
2116 | spin_unlock(&dentry->d_lock); | |
2117 | } | |
2118 | rcu_read_unlock(); | |
2119 | ||
2120 | return found; | |
2121 | } | |
2122 | ||
3e7e241f EB |
2123 | /** |
2124 | * d_hash_and_lookup - hash the qstr then search for a dentry | |
2125 | * @dir: Directory to search in | |
2126 | * @name: qstr of name we wish to find | |
2127 | * | |
4f522a24 | 2128 | * On lookup failure NULL is returned; on bad name - ERR_PTR(-error) |
3e7e241f EB |
2129 | */ |
2130 | struct dentry *d_hash_and_lookup(struct dentry *dir, struct qstr *name) | |
2131 | { | |
3e7e241f EB |
2132 | /* |
2133 | * Check for a fs-specific hash function. Note that we must | |
2134 | * calculate the standard hash first, as the d_op->d_hash() | |
2135 | * routine may choose to leave the hash value unchanged. | |
2136 | */ | |
2137 | name->hash = full_name_hash(name->name, name->len); | |
fb045adb | 2138 | if (dir->d_flags & DCACHE_OP_HASH) { |
da53be12 | 2139 | int err = dir->d_op->d_hash(dir, name); |
4f522a24 AV |
2140 | if (unlikely(err < 0)) |
2141 | return ERR_PTR(err); | |
3e7e241f | 2142 | } |
4f522a24 | 2143 | return d_lookup(dir, name); |
3e7e241f | 2144 | } |
4f522a24 | 2145 | EXPORT_SYMBOL(d_hash_and_lookup); |
3e7e241f | 2146 | |
1da177e4 | 2147 | /** |
786a5e15 | 2148 | * d_validate - verify dentry provided from insecure source (deprecated) |
1da177e4 | 2149 | * @dentry: The dentry alleged to be valid child of @dparent |
ff5fdb61 | 2150 | * @dparent: The parent dentry (known to be valid) |
1da177e4 LT |
2151 | * |
2152 | * An insecure source has sent us a dentry, here we verify it and dget() it. | |
2153 | * This is used by ncpfs in its readdir implementation. | |
2154 | * Zero is returned in the dentry is invalid. | |
786a5e15 NP |
2155 | * |
2156 | * This function is slow for big directories, and deprecated, do not use it. | |
1da177e4 | 2157 | */ |
d3a23e16 | 2158 | int d_validate(struct dentry *dentry, struct dentry *dparent) |
1da177e4 | 2159 | { |
786a5e15 | 2160 | struct dentry *child; |
d3a23e16 | 2161 | |
2fd6b7f5 | 2162 | spin_lock(&dparent->d_lock); |
786a5e15 NP |
2163 | list_for_each_entry(child, &dparent->d_subdirs, d_u.d_child) { |
2164 | if (dentry == child) { | |
2fd6b7f5 | 2165 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); |
dc0474be | 2166 | __dget_dlock(dentry); |
2fd6b7f5 NP |
2167 | spin_unlock(&dentry->d_lock); |
2168 | spin_unlock(&dparent->d_lock); | |
1da177e4 LT |
2169 | return 1; |
2170 | } | |
2171 | } | |
2fd6b7f5 | 2172 | spin_unlock(&dparent->d_lock); |
786a5e15 | 2173 | |
1da177e4 LT |
2174 | return 0; |
2175 | } | |
ec4f8605 | 2176 | EXPORT_SYMBOL(d_validate); |
1da177e4 LT |
2177 | |
2178 | /* | |
2179 | * When a file is deleted, we have two options: | |
2180 | * - turn this dentry into a negative dentry | |
2181 | * - unhash this dentry and free it. | |
2182 | * | |
2183 | * Usually, we want to just turn this into | |
2184 | * a negative dentry, but if anybody else is | |
2185 | * currently using the dentry or the inode | |
2186 | * we can't do that and we fall back on removing | |
2187 | * it from the hash queues and waiting for | |
2188 | * it to be deleted later when it has no users | |
2189 | */ | |
2190 | ||
2191 | /** | |
2192 | * d_delete - delete a dentry | |
2193 | * @dentry: The dentry to delete | |
2194 | * | |
2195 | * Turn the dentry into a negative dentry if possible, otherwise | |
2196 | * remove it from the hash queues so it can be deleted later | |
2197 | */ | |
2198 | ||
2199 | void d_delete(struct dentry * dentry) | |
2200 | { | |
873feea0 | 2201 | struct inode *inode; |
7a91bf7f | 2202 | int isdir = 0; |
1da177e4 LT |
2203 | /* |
2204 | * Are we the only user? | |
2205 | */ | |
357f8e65 | 2206 | again: |
1da177e4 | 2207 | spin_lock(&dentry->d_lock); |
873feea0 NP |
2208 | inode = dentry->d_inode; |
2209 | isdir = S_ISDIR(inode->i_mode); | |
98474236 | 2210 | if (dentry->d_lockref.count == 1) { |
1fe0c023 | 2211 | if (!spin_trylock(&inode->i_lock)) { |
357f8e65 NP |
2212 | spin_unlock(&dentry->d_lock); |
2213 | cpu_relax(); | |
2214 | goto again; | |
2215 | } | |
13e3c5e5 | 2216 | dentry->d_flags &= ~DCACHE_CANT_MOUNT; |
31e6b01f | 2217 | dentry_unlink_inode(dentry); |
7a91bf7f | 2218 | fsnotify_nameremove(dentry, isdir); |
1da177e4 LT |
2219 | return; |
2220 | } | |
2221 | ||
2222 | if (!d_unhashed(dentry)) | |
2223 | __d_drop(dentry); | |
2224 | ||
2225 | spin_unlock(&dentry->d_lock); | |
7a91bf7f JM |
2226 | |
2227 | fsnotify_nameremove(dentry, isdir); | |
1da177e4 | 2228 | } |
ec4f8605 | 2229 | EXPORT_SYMBOL(d_delete); |
1da177e4 | 2230 | |
b07ad996 | 2231 | static void __d_rehash(struct dentry * entry, struct hlist_bl_head *b) |
1da177e4 | 2232 | { |
ceb5bdc2 | 2233 | BUG_ON(!d_unhashed(entry)); |
1879fd6a | 2234 | hlist_bl_lock(b); |
dea3667b | 2235 | entry->d_flags |= DCACHE_RCUACCESS; |
b07ad996 | 2236 | hlist_bl_add_head_rcu(&entry->d_hash, b); |
1879fd6a | 2237 | hlist_bl_unlock(b); |
1da177e4 LT |
2238 | } |
2239 | ||
770bfad8 DH |
2240 | static void _d_rehash(struct dentry * entry) |
2241 | { | |
2242 | __d_rehash(entry, d_hash(entry->d_parent, entry->d_name.hash)); | |
2243 | } | |
2244 | ||
1da177e4 LT |
2245 | /** |
2246 | * d_rehash - add an entry back to the hash | |
2247 | * @entry: dentry to add to the hash | |
2248 | * | |
2249 | * Adds a dentry to the hash according to its name. | |
2250 | */ | |
2251 | ||
2252 | void d_rehash(struct dentry * entry) | |
2253 | { | |
1da177e4 | 2254 | spin_lock(&entry->d_lock); |
770bfad8 | 2255 | _d_rehash(entry); |
1da177e4 | 2256 | spin_unlock(&entry->d_lock); |
1da177e4 | 2257 | } |
ec4f8605 | 2258 | EXPORT_SYMBOL(d_rehash); |
1da177e4 | 2259 | |
fb2d5b86 NP |
2260 | /** |
2261 | * dentry_update_name_case - update case insensitive dentry with a new name | |
2262 | * @dentry: dentry to be updated | |
2263 | * @name: new name | |
2264 | * | |
2265 | * Update a case insensitive dentry with new case of name. | |
2266 | * | |
2267 | * dentry must have been returned by d_lookup with name @name. Old and new | |
2268 | * name lengths must match (ie. no d_compare which allows mismatched name | |
2269 | * lengths). | |
2270 | * | |
2271 | * Parent inode i_mutex must be held over d_lookup and into this call (to | |
2272 | * keep renames and concurrent inserts, and readdir(2) away). | |
2273 | */ | |
2274 | void dentry_update_name_case(struct dentry *dentry, struct qstr *name) | |
2275 | { | |
7ebfa57f | 2276 | BUG_ON(!mutex_is_locked(&dentry->d_parent->d_inode->i_mutex)); |
fb2d5b86 NP |
2277 | BUG_ON(dentry->d_name.len != name->len); /* d_lookup gives this */ |
2278 | ||
fb2d5b86 | 2279 | spin_lock(&dentry->d_lock); |
31e6b01f | 2280 | write_seqcount_begin(&dentry->d_seq); |
fb2d5b86 | 2281 | memcpy((unsigned char *)dentry->d_name.name, name->name, name->len); |
31e6b01f | 2282 | write_seqcount_end(&dentry->d_seq); |
fb2d5b86 | 2283 | spin_unlock(&dentry->d_lock); |
fb2d5b86 NP |
2284 | } |
2285 | EXPORT_SYMBOL(dentry_update_name_case); | |
2286 | ||
1da177e4 LT |
2287 | static void switch_names(struct dentry *dentry, struct dentry *target) |
2288 | { | |
2289 | if (dname_external(target)) { | |
2290 | if (dname_external(dentry)) { | |
2291 | /* | |
2292 | * Both external: swap the pointers | |
2293 | */ | |
9a8d5bb4 | 2294 | swap(target->d_name.name, dentry->d_name.name); |
1da177e4 LT |
2295 | } else { |
2296 | /* | |
2297 | * dentry:internal, target:external. Steal target's | |
2298 | * storage and make target internal. | |
2299 | */ | |
321bcf92 BF |
2300 | memcpy(target->d_iname, dentry->d_name.name, |
2301 | dentry->d_name.len + 1); | |
1da177e4 LT |
2302 | dentry->d_name.name = target->d_name.name; |
2303 | target->d_name.name = target->d_iname; | |
2304 | } | |
2305 | } else { | |
2306 | if (dname_external(dentry)) { | |
2307 | /* | |
2308 | * dentry:external, target:internal. Give dentry's | |
2309 | * storage to target and make dentry internal | |
2310 | */ | |
2311 | memcpy(dentry->d_iname, target->d_name.name, | |
2312 | target->d_name.len + 1); | |
2313 | target->d_name.name = dentry->d_name.name; | |
2314 | dentry->d_name.name = dentry->d_iname; | |
2315 | } else { | |
2316 | /* | |
2317 | * Both are internal. Just copy target to dentry | |
2318 | */ | |
2319 | memcpy(dentry->d_iname, target->d_name.name, | |
2320 | target->d_name.len + 1); | |
dc711ca3 AV |
2321 | dentry->d_name.len = target->d_name.len; |
2322 | return; | |
1da177e4 LT |
2323 | } |
2324 | } | |
9a8d5bb4 | 2325 | swap(dentry->d_name.len, target->d_name.len); |
1da177e4 LT |
2326 | } |
2327 | ||
2fd6b7f5 NP |
2328 | static void dentry_lock_for_move(struct dentry *dentry, struct dentry *target) |
2329 | { | |
2330 | /* | |
2331 | * XXXX: do we really need to take target->d_lock? | |
2332 | */ | |
2333 | if (IS_ROOT(dentry) || dentry->d_parent == target->d_parent) | |
2334 | spin_lock(&target->d_parent->d_lock); | |
2335 | else { | |
2336 | if (d_ancestor(dentry->d_parent, target->d_parent)) { | |
2337 | spin_lock(&dentry->d_parent->d_lock); | |
2338 | spin_lock_nested(&target->d_parent->d_lock, | |
2339 | DENTRY_D_LOCK_NESTED); | |
2340 | } else { | |
2341 | spin_lock(&target->d_parent->d_lock); | |
2342 | spin_lock_nested(&dentry->d_parent->d_lock, | |
2343 | DENTRY_D_LOCK_NESTED); | |
2344 | } | |
2345 | } | |
2346 | if (target < dentry) { | |
2347 | spin_lock_nested(&target->d_lock, 2); | |
2348 | spin_lock_nested(&dentry->d_lock, 3); | |
2349 | } else { | |
2350 | spin_lock_nested(&dentry->d_lock, 2); | |
2351 | spin_lock_nested(&target->d_lock, 3); | |
2352 | } | |
2353 | } | |
2354 | ||
2355 | static void dentry_unlock_parents_for_move(struct dentry *dentry, | |
2356 | struct dentry *target) | |
2357 | { | |
2358 | if (target->d_parent != dentry->d_parent) | |
2359 | spin_unlock(&dentry->d_parent->d_lock); | |
2360 | if (target->d_parent != target) | |
2361 | spin_unlock(&target->d_parent->d_lock); | |
2362 | } | |
2363 | ||
1da177e4 | 2364 | /* |
2fd6b7f5 NP |
2365 | * When switching names, the actual string doesn't strictly have to |
2366 | * be preserved in the target - because we're dropping the target | |
2367 | * anyway. As such, we can just do a simple memcpy() to copy over | |
2368 | * the new name before we switch. | |
2369 | * | |
2370 | * Note that we have to be a lot more careful about getting the hash | |
2371 | * switched - we have to switch the hash value properly even if it | |
2372 | * then no longer matches the actual (corrupted) string of the target. | |
2373 | * The hash value has to match the hash queue that the dentry is on.. | |
1da177e4 | 2374 | */ |
9eaef27b | 2375 | /* |
18367501 | 2376 | * __d_move - move a dentry |
1da177e4 LT |
2377 | * @dentry: entry to move |
2378 | * @target: new dentry | |
2379 | * | |
2380 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2381 | * dcache entries should not be moved in this way. Caller must hold |
2382 | * rename_lock, the i_mutex of the source and target directories, | |
2383 | * and the sb->s_vfs_rename_mutex if they differ. See lock_rename(). | |
1da177e4 | 2384 | */ |
18367501 | 2385 | static void __d_move(struct dentry * dentry, struct dentry * target) |
1da177e4 | 2386 | { |
1da177e4 LT |
2387 | if (!dentry->d_inode) |
2388 | printk(KERN_WARNING "VFS: moving negative dcache entry\n"); | |
2389 | ||
2fd6b7f5 NP |
2390 | BUG_ON(d_ancestor(dentry, target)); |
2391 | BUG_ON(d_ancestor(target, dentry)); | |
2392 | ||
2fd6b7f5 | 2393 | dentry_lock_for_move(dentry, target); |
1da177e4 | 2394 | |
31e6b01f NP |
2395 | write_seqcount_begin(&dentry->d_seq); |
2396 | write_seqcount_begin(&target->d_seq); | |
2397 | ||
ceb5bdc2 NP |
2398 | /* __d_drop does write_seqcount_barrier, but they're OK to nest. */ |
2399 | ||
2400 | /* | |
2401 | * Move the dentry to the target hash queue. Don't bother checking | |
2402 | * for the same hash queue because of how unlikely it is. | |
2403 | */ | |
2404 | __d_drop(dentry); | |
789680d1 | 2405 | __d_rehash(dentry, d_hash(target->d_parent, target->d_name.hash)); |
1da177e4 LT |
2406 | |
2407 | /* Unhash the target: dput() will then get rid of it */ | |
2408 | __d_drop(target); | |
2409 | ||
5160ee6f ED |
2410 | list_del(&dentry->d_u.d_child); |
2411 | list_del(&target->d_u.d_child); | |
1da177e4 LT |
2412 | |
2413 | /* Switch the names.. */ | |
2414 | switch_names(dentry, target); | |
9a8d5bb4 | 2415 | swap(dentry->d_name.hash, target->d_name.hash); |
1da177e4 LT |
2416 | |
2417 | /* ... and switch the parents */ | |
2418 | if (IS_ROOT(dentry)) { | |
2419 | dentry->d_parent = target->d_parent; | |
2420 | target->d_parent = target; | |
5160ee6f | 2421 | INIT_LIST_HEAD(&target->d_u.d_child); |
1da177e4 | 2422 | } else { |
9a8d5bb4 | 2423 | swap(dentry->d_parent, target->d_parent); |
1da177e4 LT |
2424 | |
2425 | /* And add them back to the (new) parent lists */ | |
5160ee6f | 2426 | list_add(&target->d_u.d_child, &target->d_parent->d_subdirs); |
1da177e4 LT |
2427 | } |
2428 | ||
5160ee6f | 2429 | list_add(&dentry->d_u.d_child, &dentry->d_parent->d_subdirs); |
2fd6b7f5 | 2430 | |
31e6b01f NP |
2431 | write_seqcount_end(&target->d_seq); |
2432 | write_seqcount_end(&dentry->d_seq); | |
2433 | ||
2fd6b7f5 | 2434 | dentry_unlock_parents_for_move(dentry, target); |
1da177e4 | 2435 | spin_unlock(&target->d_lock); |
c32ccd87 | 2436 | fsnotify_d_move(dentry); |
1da177e4 | 2437 | spin_unlock(&dentry->d_lock); |
18367501 AV |
2438 | } |
2439 | ||
2440 | /* | |
2441 | * d_move - move a dentry | |
2442 | * @dentry: entry to move | |
2443 | * @target: new dentry | |
2444 | * | |
2445 | * Update the dcache to reflect the move of a file name. Negative | |
c46c8877 JL |
2446 | * dcache entries should not be moved in this way. See the locking |
2447 | * requirements for __d_move. | |
18367501 AV |
2448 | */ |
2449 | void d_move(struct dentry *dentry, struct dentry *target) | |
2450 | { | |
2451 | write_seqlock(&rename_lock); | |
2452 | __d_move(dentry, target); | |
1da177e4 | 2453 | write_sequnlock(&rename_lock); |
9eaef27b | 2454 | } |
ec4f8605 | 2455 | EXPORT_SYMBOL(d_move); |
1da177e4 | 2456 | |
e2761a11 OH |
2457 | /** |
2458 | * d_ancestor - search for an ancestor | |
2459 | * @p1: ancestor dentry | |
2460 | * @p2: child dentry | |
2461 | * | |
2462 | * Returns the ancestor dentry of p2 which is a child of p1, if p1 is | |
2463 | * an ancestor of p2, else NULL. | |
9eaef27b | 2464 | */ |
e2761a11 | 2465 | struct dentry *d_ancestor(struct dentry *p1, struct dentry *p2) |
9eaef27b TM |
2466 | { |
2467 | struct dentry *p; | |
2468 | ||
871c0067 | 2469 | for (p = p2; !IS_ROOT(p); p = p->d_parent) { |
9eaef27b | 2470 | if (p->d_parent == p1) |
e2761a11 | 2471 | return p; |
9eaef27b | 2472 | } |
e2761a11 | 2473 | return NULL; |
9eaef27b TM |
2474 | } |
2475 | ||
2476 | /* | |
2477 | * This helper attempts to cope with remotely renamed directories | |
2478 | * | |
2479 | * It assumes that the caller is already holding | |
18367501 | 2480 | * dentry->d_parent->d_inode->i_mutex, inode->i_lock and rename_lock |
9eaef27b TM |
2481 | * |
2482 | * Note: If ever the locking in lock_rename() changes, then please | |
2483 | * remember to update this too... | |
9eaef27b | 2484 | */ |
873feea0 NP |
2485 | static struct dentry *__d_unalias(struct inode *inode, |
2486 | struct dentry *dentry, struct dentry *alias) | |
9eaef27b TM |
2487 | { |
2488 | struct mutex *m1 = NULL, *m2 = NULL; | |
ee3efa91 | 2489 | struct dentry *ret = ERR_PTR(-EBUSY); |
9eaef27b TM |
2490 | |
2491 | /* If alias and dentry share a parent, then no extra locks required */ | |
2492 | if (alias->d_parent == dentry->d_parent) | |
2493 | goto out_unalias; | |
2494 | ||
9eaef27b | 2495 | /* See lock_rename() */ |
9eaef27b TM |
2496 | if (!mutex_trylock(&dentry->d_sb->s_vfs_rename_mutex)) |
2497 | goto out_err; | |
2498 | m1 = &dentry->d_sb->s_vfs_rename_mutex; | |
2499 | if (!mutex_trylock(&alias->d_parent->d_inode->i_mutex)) | |
2500 | goto out_err; | |
2501 | m2 = &alias->d_parent->d_inode->i_mutex; | |
2502 | out_unalias: | |
ee3efa91 AV |
2503 | if (likely(!d_mountpoint(alias))) { |
2504 | __d_move(alias, dentry); | |
2505 | ret = alias; | |
2506 | } | |
9eaef27b | 2507 | out_err: |
873feea0 | 2508 | spin_unlock(&inode->i_lock); |
9eaef27b TM |
2509 | if (m2) |
2510 | mutex_unlock(m2); | |
2511 | if (m1) | |
2512 | mutex_unlock(m1); | |
2513 | return ret; | |
2514 | } | |
2515 | ||
770bfad8 DH |
2516 | /* |
2517 | * Prepare an anonymous dentry for life in the superblock's dentry tree as a | |
2518 | * named dentry in place of the dentry to be replaced. | |
2fd6b7f5 | 2519 | * returns with anon->d_lock held! |
770bfad8 DH |
2520 | */ |
2521 | static void __d_materialise_dentry(struct dentry *dentry, struct dentry *anon) | |
2522 | { | |
740da42e | 2523 | struct dentry *dparent; |
770bfad8 | 2524 | |
2fd6b7f5 | 2525 | dentry_lock_for_move(anon, dentry); |
770bfad8 | 2526 | |
31e6b01f NP |
2527 | write_seqcount_begin(&dentry->d_seq); |
2528 | write_seqcount_begin(&anon->d_seq); | |
2529 | ||
770bfad8 | 2530 | dparent = dentry->d_parent; |
770bfad8 | 2531 | |
2fd6b7f5 NP |
2532 | switch_names(dentry, anon); |
2533 | swap(dentry->d_name.hash, anon->d_name.hash); | |
2534 | ||
740da42e AV |
2535 | dentry->d_parent = dentry; |
2536 | list_del_init(&dentry->d_u.d_child); | |
2537 | anon->d_parent = dparent; | |
9ed53b12 | 2538 | list_move(&anon->d_u.d_child, &dparent->d_subdirs); |
770bfad8 | 2539 | |
31e6b01f NP |
2540 | write_seqcount_end(&dentry->d_seq); |
2541 | write_seqcount_end(&anon->d_seq); | |
2542 | ||
2fd6b7f5 NP |
2543 | dentry_unlock_parents_for_move(anon, dentry); |
2544 | spin_unlock(&dentry->d_lock); | |
2545 | ||
2546 | /* anon->d_lock still locked, returns locked */ | |
770bfad8 DH |
2547 | anon->d_flags &= ~DCACHE_DISCONNECTED; |
2548 | } | |
2549 | ||
2550 | /** | |
2551 | * d_materialise_unique - introduce an inode into the tree | |
2552 | * @dentry: candidate dentry | |
2553 | * @inode: inode to bind to the dentry, to which aliases may be attached | |
2554 | * | |
2555 | * Introduces an dentry into the tree, substituting an extant disconnected | |
c46c8877 JL |
2556 | * root directory alias in its place if there is one. Caller must hold the |
2557 | * i_mutex of the parent directory. | |
770bfad8 DH |
2558 | */ |
2559 | struct dentry *d_materialise_unique(struct dentry *dentry, struct inode *inode) | |
2560 | { | |
9eaef27b | 2561 | struct dentry *actual; |
770bfad8 DH |
2562 | |
2563 | BUG_ON(!d_unhashed(dentry)); | |
2564 | ||
770bfad8 DH |
2565 | if (!inode) { |
2566 | actual = dentry; | |
360da900 | 2567 | __d_instantiate(dentry, NULL); |
357f8e65 NP |
2568 | d_rehash(actual); |
2569 | goto out_nolock; | |
770bfad8 DH |
2570 | } |
2571 | ||
873feea0 | 2572 | spin_lock(&inode->i_lock); |
357f8e65 | 2573 | |
9eaef27b TM |
2574 | if (S_ISDIR(inode->i_mode)) { |
2575 | struct dentry *alias; | |
2576 | ||
2577 | /* Does an aliased dentry already exist? */ | |
32ba9c3f | 2578 | alias = __d_find_alias(inode, 0); |
9eaef27b TM |
2579 | if (alias) { |
2580 | actual = alias; | |
18367501 AV |
2581 | write_seqlock(&rename_lock); |
2582 | ||
2583 | if (d_ancestor(alias, dentry)) { | |
2584 | /* Check for loops */ | |
2585 | actual = ERR_PTR(-ELOOP); | |
b18dafc8 | 2586 | spin_unlock(&inode->i_lock); |
18367501 AV |
2587 | } else if (IS_ROOT(alias)) { |
2588 | /* Is this an anonymous mountpoint that we | |
2589 | * could splice into our tree? */ | |
9eaef27b | 2590 | __d_materialise_dentry(dentry, alias); |
18367501 | 2591 | write_sequnlock(&rename_lock); |
9eaef27b TM |
2592 | __d_drop(alias); |
2593 | goto found; | |
18367501 AV |
2594 | } else { |
2595 | /* Nope, but we must(!) avoid directory | |
b18dafc8 | 2596 | * aliasing. This drops inode->i_lock */ |
18367501 | 2597 | actual = __d_unalias(inode, dentry, alias); |
9eaef27b | 2598 | } |
18367501 | 2599 | write_sequnlock(&rename_lock); |
dd179946 DH |
2600 | if (IS_ERR(actual)) { |
2601 | if (PTR_ERR(actual) == -ELOOP) | |
2602 | pr_warn_ratelimited( | |
2603 | "VFS: Lookup of '%s' in %s %s" | |
2604 | " would have caused loop\n", | |
2605 | dentry->d_name.name, | |
2606 | inode->i_sb->s_type->name, | |
2607 | inode->i_sb->s_id); | |
9eaef27b | 2608 | dput(alias); |
dd179946 | 2609 | } |
9eaef27b TM |
2610 | goto out_nolock; |
2611 | } | |
770bfad8 DH |
2612 | } |
2613 | ||
2614 | /* Add a unique reference */ | |
2615 | actual = __d_instantiate_unique(dentry, inode); | |
2616 | if (!actual) | |
2617 | actual = dentry; | |
357f8e65 NP |
2618 | else |
2619 | BUG_ON(!d_unhashed(actual)); | |
770bfad8 | 2620 | |
770bfad8 DH |
2621 | spin_lock(&actual->d_lock); |
2622 | found: | |
2623 | _d_rehash(actual); | |
2624 | spin_unlock(&actual->d_lock); | |
873feea0 | 2625 | spin_unlock(&inode->i_lock); |
9eaef27b | 2626 | out_nolock: |
770bfad8 DH |
2627 | if (actual == dentry) { |
2628 | security_d_instantiate(dentry, inode); | |
2629 | return NULL; | |
2630 | } | |
2631 | ||
2632 | iput(inode); | |
2633 | return actual; | |
770bfad8 | 2634 | } |
ec4f8605 | 2635 | EXPORT_SYMBOL_GPL(d_materialise_unique); |
770bfad8 | 2636 | |
cdd16d02 | 2637 | static int prepend(char **buffer, int *buflen, const char *str, int namelen) |
6092d048 RP |
2638 | { |
2639 | *buflen -= namelen; | |
2640 | if (*buflen < 0) | |
2641 | return -ENAMETOOLONG; | |
2642 | *buffer -= namelen; | |
2643 | memcpy(*buffer, str, namelen); | |
2644 | return 0; | |
2645 | } | |
2646 | ||
cdd16d02 MS |
2647 | static int prepend_name(char **buffer, int *buflen, struct qstr *name) |
2648 | { | |
2649 | return prepend(buffer, buflen, name->name, name->len); | |
2650 | } | |
2651 | ||
1da177e4 | 2652 | /** |
208898c1 | 2653 | * prepend_path - Prepend path string to a buffer |
9d1bc601 | 2654 | * @path: the dentry/vfsmount to report |
02125a82 | 2655 | * @root: root vfsmnt/dentry |
f2eb6575 MS |
2656 | * @buffer: pointer to the end of the buffer |
2657 | * @buflen: pointer to buffer length | |
552ce544 | 2658 | * |
949854d0 | 2659 | * Caller holds the rename_lock. |
1da177e4 | 2660 | */ |
02125a82 AV |
2661 | static int prepend_path(const struct path *path, |
2662 | const struct path *root, | |
f2eb6575 | 2663 | char **buffer, int *buflen) |
1da177e4 | 2664 | { |
9d1bc601 MS |
2665 | struct dentry *dentry = path->dentry; |
2666 | struct vfsmount *vfsmnt = path->mnt; | |
0714a533 | 2667 | struct mount *mnt = real_mount(vfsmnt); |
f2eb6575 MS |
2668 | bool slash = false; |
2669 | int error = 0; | |
6092d048 | 2670 | |
f2eb6575 | 2671 | while (dentry != root->dentry || vfsmnt != root->mnt) { |
1da177e4 LT |
2672 | struct dentry * parent; |
2673 | ||
1da177e4 | 2674 | if (dentry == vfsmnt->mnt_root || IS_ROOT(dentry)) { |
552ce544 | 2675 | /* Global root? */ |
676da58d | 2676 | if (!mnt_has_parent(mnt)) |
1da177e4 | 2677 | goto global_root; |
a73324da | 2678 | dentry = mnt->mnt_mountpoint; |
0714a533 AV |
2679 | mnt = mnt->mnt_parent; |
2680 | vfsmnt = &mnt->mnt; | |
1da177e4 LT |
2681 | continue; |
2682 | } | |
2683 | parent = dentry->d_parent; | |
2684 | prefetch(parent); | |
9abca360 | 2685 | spin_lock(&dentry->d_lock); |
f2eb6575 | 2686 | error = prepend_name(buffer, buflen, &dentry->d_name); |
9abca360 | 2687 | spin_unlock(&dentry->d_lock); |
f2eb6575 MS |
2688 | if (!error) |
2689 | error = prepend(buffer, buflen, "/", 1); | |
2690 | if (error) | |
2691 | break; | |
2692 | ||
2693 | slash = true; | |
1da177e4 LT |
2694 | dentry = parent; |
2695 | } | |
2696 | ||
f2eb6575 MS |
2697 | if (!error && !slash) |
2698 | error = prepend(buffer, buflen, "/", 1); | |
2699 | ||
f2eb6575 | 2700 | return error; |
1da177e4 LT |
2701 | |
2702 | global_root: | |
98dc568b MS |
2703 | /* |
2704 | * Filesystems needing to implement special "root names" | |
2705 | * should do so with ->d_dname() | |
2706 | */ | |
2707 | if (IS_ROOT(dentry) && | |
2708 | (dentry->d_name.len != 1 || dentry->d_name.name[0] != '/')) { | |
2709 | WARN(1, "Root dentry has weird name <%.*s>\n", | |
2710 | (int) dentry->d_name.len, dentry->d_name.name); | |
2711 | } | |
02125a82 AV |
2712 | if (!slash) |
2713 | error = prepend(buffer, buflen, "/", 1); | |
2714 | if (!error) | |
f7a99c5b | 2715 | error = is_mounted(vfsmnt) ? 1 : 2; |
7ea600b5 | 2716 | return error; |
f2eb6575 | 2717 | } |
be285c71 | 2718 | |
f2eb6575 MS |
2719 | /** |
2720 | * __d_path - return the path of a dentry | |
2721 | * @path: the dentry/vfsmount to report | |
02125a82 | 2722 | * @root: root vfsmnt/dentry |
cd956a1c | 2723 | * @buf: buffer to return value in |
f2eb6575 MS |
2724 | * @buflen: buffer length |
2725 | * | |
ffd1f4ed | 2726 | * Convert a dentry into an ASCII path name. |
f2eb6575 MS |
2727 | * |
2728 | * Returns a pointer into the buffer or an error code if the | |
2729 | * path was too long. | |
2730 | * | |
be148247 | 2731 | * "buflen" should be positive. |
f2eb6575 | 2732 | * |
02125a82 | 2733 | * If the path is not reachable from the supplied root, return %NULL. |
f2eb6575 | 2734 | */ |
02125a82 AV |
2735 | char *__d_path(const struct path *path, |
2736 | const struct path *root, | |
f2eb6575 MS |
2737 | char *buf, int buflen) |
2738 | { | |
2739 | char *res = buf + buflen; | |
2740 | int error; | |
2741 | ||
2742 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2743 | br_read_lock(&vfsmount_lock); |
949854d0 | 2744 | write_seqlock(&rename_lock); |
f2eb6575 | 2745 | error = prepend_path(path, root, &res, &buflen); |
949854d0 | 2746 | write_sequnlock(&rename_lock); |
7ea600b5 | 2747 | br_read_unlock(&vfsmount_lock); |
be148247 | 2748 | |
02125a82 AV |
2749 | if (error < 0) |
2750 | return ERR_PTR(error); | |
2751 | if (error > 0) | |
2752 | return NULL; | |
2753 | return res; | |
2754 | } | |
2755 | ||
2756 | char *d_absolute_path(const struct path *path, | |
2757 | char *buf, int buflen) | |
2758 | { | |
2759 | struct path root = {}; | |
2760 | char *res = buf + buflen; | |
2761 | int error; | |
2762 | ||
2763 | prepend(&res, &buflen, "\0", 1); | |
7ea600b5 | 2764 | br_read_lock(&vfsmount_lock); |
02125a82 AV |
2765 | write_seqlock(&rename_lock); |
2766 | error = prepend_path(path, &root, &res, &buflen); | |
2767 | write_sequnlock(&rename_lock); | |
7ea600b5 | 2768 | br_read_unlock(&vfsmount_lock); |
02125a82 AV |
2769 | |
2770 | if (error > 1) | |
2771 | error = -EINVAL; | |
2772 | if (error < 0) | |
f2eb6575 | 2773 | return ERR_PTR(error); |
f2eb6575 | 2774 | return res; |
1da177e4 LT |
2775 | } |
2776 | ||
ffd1f4ed MS |
2777 | /* |
2778 | * same as __d_path but appends "(deleted)" for unlinked files. | |
2779 | */ | |
02125a82 AV |
2780 | static int path_with_deleted(const struct path *path, |
2781 | const struct path *root, | |
2782 | char **buf, int *buflen) | |
ffd1f4ed MS |
2783 | { |
2784 | prepend(buf, buflen, "\0", 1); | |
2785 | if (d_unlinked(path->dentry)) { | |
2786 | int error = prepend(buf, buflen, " (deleted)", 10); | |
2787 | if (error) | |
2788 | return error; | |
2789 | } | |
2790 | ||
2791 | return prepend_path(path, root, buf, buflen); | |
2792 | } | |
2793 | ||
8df9d1a4 MS |
2794 | static int prepend_unreachable(char **buffer, int *buflen) |
2795 | { | |
2796 | return prepend(buffer, buflen, "(unreachable)", 13); | |
2797 | } | |
2798 | ||
a03a8a70 JB |
2799 | /** |
2800 | * d_path - return the path of a dentry | |
cf28b486 | 2801 | * @path: path to report |
a03a8a70 JB |
2802 | * @buf: buffer to return value in |
2803 | * @buflen: buffer length | |
2804 | * | |
2805 | * Convert a dentry into an ASCII path name. If the entry has been deleted | |
2806 | * the string " (deleted)" is appended. Note that this is ambiguous. | |
2807 | * | |
52afeefb AV |
2808 | * Returns a pointer into the buffer or an error code if the path was |
2809 | * too long. Note: Callers should use the returned pointer, not the passed | |
2810 | * in buffer, to use the name! The implementation often starts at an offset | |
2811 | * into the buffer, and may leave 0 bytes at the start. | |
a03a8a70 | 2812 | * |
31f3e0b3 | 2813 | * "buflen" should be positive. |
a03a8a70 | 2814 | */ |
20d4fdc1 | 2815 | char *d_path(const struct path *path, char *buf, int buflen) |
1da177e4 | 2816 | { |
ffd1f4ed | 2817 | char *res = buf + buflen; |
6ac08c39 | 2818 | struct path root; |
ffd1f4ed | 2819 | int error; |
1da177e4 | 2820 | |
c23fbb6b ED |
2821 | /* |
2822 | * We have various synthetic filesystems that never get mounted. On | |
2823 | * these filesystems dentries are never used for lookup purposes, and | |
2824 | * thus don't need to be hashed. They also don't need a name until a | |
2825 | * user wants to identify the object in /proc/pid/fd/. The little hack | |
2826 | * below allows us to generate a name for these objects on demand: | |
2827 | */ | |
cf28b486 JB |
2828 | if (path->dentry->d_op && path->dentry->d_op->d_dname) |
2829 | return path->dentry->d_op->d_dname(path->dentry, buf, buflen); | |
c23fbb6b | 2830 | |
f7ad3c6b | 2831 | get_fs_root(current->fs, &root); |
7ea600b5 | 2832 | br_read_lock(&vfsmount_lock); |
949854d0 | 2833 | write_seqlock(&rename_lock); |
02125a82 | 2834 | error = path_with_deleted(path, &root, &res, &buflen); |
7ea600b5 AV |
2835 | write_sequnlock(&rename_lock); |
2836 | br_read_unlock(&vfsmount_lock); | |
02125a82 | 2837 | if (error < 0) |
ffd1f4ed | 2838 | res = ERR_PTR(error); |
6ac08c39 | 2839 | path_put(&root); |
1da177e4 LT |
2840 | return res; |
2841 | } | |
ec4f8605 | 2842 | EXPORT_SYMBOL(d_path); |
1da177e4 | 2843 | |
c23fbb6b ED |
2844 | /* |
2845 | * Helper function for dentry_operations.d_dname() members | |
2846 | */ | |
2847 | char *dynamic_dname(struct dentry *dentry, char *buffer, int buflen, | |
2848 | const char *fmt, ...) | |
2849 | { | |
2850 | va_list args; | |
2851 | char temp[64]; | |
2852 | int sz; | |
2853 | ||
2854 | va_start(args, fmt); | |
2855 | sz = vsnprintf(temp, sizeof(temp), fmt, args) + 1; | |
2856 | va_end(args); | |
2857 | ||
2858 | if (sz > sizeof(temp) || sz > buflen) | |
2859 | return ERR_PTR(-ENAMETOOLONG); | |
2860 | ||
2861 | buffer += buflen - sz; | |
2862 | return memcpy(buffer, temp, sz); | |
2863 | } | |
2864 | ||
118b2302 AV |
2865 | char *simple_dname(struct dentry *dentry, char *buffer, int buflen) |
2866 | { | |
2867 | char *end = buffer + buflen; | |
2868 | /* these dentries are never renamed, so d_lock is not needed */ | |
2869 | if (prepend(&end, &buflen, " (deleted)", 11) || | |
2870 | prepend_name(&end, &buflen, &dentry->d_name) || | |
2871 | prepend(&end, &buflen, "/", 1)) | |
2872 | end = ERR_PTR(-ENAMETOOLONG); | |
2873 | return end; | |
2874 | } | |
2875 | ||
6092d048 RP |
2876 | /* |
2877 | * Write full pathname from the root of the filesystem into the buffer. | |
2878 | */ | |
ec2447c2 | 2879 | static char *__dentry_path(struct dentry *dentry, char *buf, int buflen) |
6092d048 RP |
2880 | { |
2881 | char *end = buf + buflen; | |
2882 | char *retval; | |
2883 | ||
6092d048 | 2884 | prepend(&end, &buflen, "\0", 1); |
6092d048 RP |
2885 | if (buflen < 1) |
2886 | goto Elong; | |
2887 | /* Get '/' right */ | |
2888 | retval = end-1; | |
2889 | *retval = '/'; | |
2890 | ||
cdd16d02 MS |
2891 | while (!IS_ROOT(dentry)) { |
2892 | struct dentry *parent = dentry->d_parent; | |
9abca360 | 2893 | int error; |
6092d048 | 2894 | |
6092d048 | 2895 | prefetch(parent); |
9abca360 NP |
2896 | spin_lock(&dentry->d_lock); |
2897 | error = prepend_name(&end, &buflen, &dentry->d_name); | |
2898 | spin_unlock(&dentry->d_lock); | |
2899 | if (error != 0 || prepend(&end, &buflen, "/", 1) != 0) | |
6092d048 RP |
2900 | goto Elong; |
2901 | ||
2902 | retval = end; | |
2903 | dentry = parent; | |
2904 | } | |
c103135c AV |
2905 | return retval; |
2906 | Elong: | |
2907 | return ERR_PTR(-ENAMETOOLONG); | |
2908 | } | |
ec2447c2 NP |
2909 | |
2910 | char *dentry_path_raw(struct dentry *dentry, char *buf, int buflen) | |
2911 | { | |
2912 | char *retval; | |
2913 | ||
949854d0 | 2914 | write_seqlock(&rename_lock); |
ec2447c2 | 2915 | retval = __dentry_path(dentry, buf, buflen); |
949854d0 | 2916 | write_sequnlock(&rename_lock); |
ec2447c2 NP |
2917 | |
2918 | return retval; | |
2919 | } | |
2920 | EXPORT_SYMBOL(dentry_path_raw); | |
c103135c AV |
2921 | |
2922 | char *dentry_path(struct dentry *dentry, char *buf, int buflen) | |
2923 | { | |
2924 | char *p = NULL; | |
2925 | char *retval; | |
2926 | ||
949854d0 | 2927 | write_seqlock(&rename_lock); |
c103135c AV |
2928 | if (d_unlinked(dentry)) { |
2929 | p = buf + buflen; | |
2930 | if (prepend(&p, &buflen, "//deleted", 10) != 0) | |
2931 | goto Elong; | |
2932 | buflen++; | |
2933 | } | |
2934 | retval = __dentry_path(dentry, buf, buflen); | |
949854d0 | 2935 | write_sequnlock(&rename_lock); |
c103135c AV |
2936 | if (!IS_ERR(retval) && p) |
2937 | *p = '/'; /* restore '/' overriden with '\0' */ | |
6092d048 RP |
2938 | return retval; |
2939 | Elong: | |
6092d048 RP |
2940 | return ERR_PTR(-ENAMETOOLONG); |
2941 | } | |
2942 | ||
1da177e4 LT |
2943 | /* |
2944 | * NOTE! The user-level library version returns a | |
2945 | * character pointer. The kernel system call just | |
2946 | * returns the length of the buffer filled (which | |
2947 | * includes the ending '\0' character), or a negative | |
2948 | * error value. So libc would do something like | |
2949 | * | |
2950 | * char *getcwd(char * buf, size_t size) | |
2951 | * { | |
2952 | * int retval; | |
2953 | * | |
2954 | * retval = sys_getcwd(buf, size); | |
2955 | * if (retval >= 0) | |
2956 | * return buf; | |
2957 | * errno = -retval; | |
2958 | * return NULL; | |
2959 | * } | |
2960 | */ | |
3cdad428 | 2961 | SYSCALL_DEFINE2(getcwd, char __user *, buf, unsigned long, size) |
1da177e4 | 2962 | { |
552ce544 | 2963 | int error; |
6ac08c39 | 2964 | struct path pwd, root; |
552ce544 | 2965 | char *page = (char *) __get_free_page(GFP_USER); |
1da177e4 LT |
2966 | |
2967 | if (!page) | |
2968 | return -ENOMEM; | |
2969 | ||
f7ad3c6b | 2970 | get_fs_root_and_pwd(current->fs, &root, &pwd); |
1da177e4 | 2971 | |
552ce544 | 2972 | error = -ENOENT; |
7ea600b5 | 2973 | br_read_lock(&vfsmount_lock); |
949854d0 | 2974 | write_seqlock(&rename_lock); |
f3da392e | 2975 | if (!d_unlinked(pwd.dentry)) { |
552ce544 | 2976 | unsigned long len; |
8df9d1a4 MS |
2977 | char *cwd = page + PAGE_SIZE; |
2978 | int buflen = PAGE_SIZE; | |
1da177e4 | 2979 | |
8df9d1a4 | 2980 | prepend(&cwd, &buflen, "\0", 1); |
02125a82 | 2981 | error = prepend_path(&pwd, &root, &cwd, &buflen); |
949854d0 | 2982 | write_sequnlock(&rename_lock); |
7ea600b5 | 2983 | br_read_unlock(&vfsmount_lock); |
552ce544 | 2984 | |
02125a82 | 2985 | if (error < 0) |
552ce544 LT |
2986 | goto out; |
2987 | ||
8df9d1a4 | 2988 | /* Unreachable from current root */ |
02125a82 | 2989 | if (error > 0) { |
8df9d1a4 MS |
2990 | error = prepend_unreachable(&cwd, &buflen); |
2991 | if (error) | |
2992 | goto out; | |
2993 | } | |
2994 | ||
552ce544 LT |
2995 | error = -ERANGE; |
2996 | len = PAGE_SIZE + page - cwd; | |
2997 | if (len <= size) { | |
2998 | error = len; | |
2999 | if (copy_to_user(buf, cwd, len)) | |
3000 | error = -EFAULT; | |
3001 | } | |
949854d0 NP |
3002 | } else { |
3003 | write_sequnlock(&rename_lock); | |
7ea600b5 | 3004 | br_read_unlock(&vfsmount_lock); |
949854d0 | 3005 | } |
1da177e4 LT |
3006 | |
3007 | out: | |
6ac08c39 JB |
3008 | path_put(&pwd); |
3009 | path_put(&root); | |
1da177e4 LT |
3010 | free_page((unsigned long) page); |
3011 | return error; | |
3012 | } | |
3013 | ||
3014 | /* | |
3015 | * Test whether new_dentry is a subdirectory of old_dentry. | |
3016 | * | |
3017 | * Trivially implemented using the dcache structure | |
3018 | */ | |
3019 | ||
3020 | /** | |
3021 | * is_subdir - is new dentry a subdirectory of old_dentry | |
3022 | * @new_dentry: new dentry | |
3023 | * @old_dentry: old dentry | |
3024 | * | |
3025 | * Returns 1 if new_dentry is a subdirectory of the parent (at any depth). | |
3026 | * Returns 0 otherwise. | |
3027 | * Caller must ensure that "new_dentry" is pinned before calling is_subdir() | |
3028 | */ | |
3029 | ||
e2761a11 | 3030 | int is_subdir(struct dentry *new_dentry, struct dentry *old_dentry) |
1da177e4 LT |
3031 | { |
3032 | int result; | |
949854d0 | 3033 | unsigned seq; |
1da177e4 | 3034 | |
e2761a11 OH |
3035 | if (new_dentry == old_dentry) |
3036 | return 1; | |
3037 | ||
e2761a11 | 3038 | do { |
1da177e4 | 3039 | /* for restarting inner loop in case of seq retry */ |
1da177e4 | 3040 | seq = read_seqbegin(&rename_lock); |
949854d0 NP |
3041 | /* |
3042 | * Need rcu_readlock to protect against the d_parent trashing | |
3043 | * due to d_move | |
3044 | */ | |
3045 | rcu_read_lock(); | |
e2761a11 | 3046 | if (d_ancestor(old_dentry, new_dentry)) |
1da177e4 | 3047 | result = 1; |
e2761a11 OH |
3048 | else |
3049 | result = 0; | |
949854d0 | 3050 | rcu_read_unlock(); |
1da177e4 | 3051 | } while (read_seqretry(&rename_lock, seq)); |
1da177e4 LT |
3052 | |
3053 | return result; | |
3054 | } | |
3055 | ||
db14fc3a | 3056 | static enum d_walk_ret d_genocide_kill(void *data, struct dentry *dentry) |
1da177e4 | 3057 | { |
db14fc3a MS |
3058 | struct dentry *root = data; |
3059 | if (dentry != root) { | |
3060 | if (d_unhashed(dentry) || !dentry->d_inode) | |
3061 | return D_WALK_SKIP; | |
1da177e4 | 3062 | |
01ddc4ed MS |
3063 | if (!(dentry->d_flags & DCACHE_GENOCIDE)) { |
3064 | dentry->d_flags |= DCACHE_GENOCIDE; | |
3065 | dentry->d_lockref.count--; | |
3066 | } | |
1da177e4 | 3067 | } |
db14fc3a MS |
3068 | return D_WALK_CONTINUE; |
3069 | } | |
58db63d0 | 3070 | |
db14fc3a MS |
3071 | void d_genocide(struct dentry *parent) |
3072 | { | |
3073 | d_walk(parent, parent, d_genocide_kill, NULL); | |
1da177e4 LT |
3074 | } |
3075 | ||
60545d0d | 3076 | void d_tmpfile(struct dentry *dentry, struct inode *inode) |
1da177e4 | 3077 | { |
60545d0d AV |
3078 | inode_dec_link_count(inode); |
3079 | BUG_ON(dentry->d_name.name != dentry->d_iname || | |
3080 | !hlist_unhashed(&dentry->d_alias) || | |
3081 | !d_unlinked(dentry)); | |
3082 | spin_lock(&dentry->d_parent->d_lock); | |
3083 | spin_lock_nested(&dentry->d_lock, DENTRY_D_LOCK_NESTED); | |
3084 | dentry->d_name.len = sprintf(dentry->d_iname, "#%llu", | |
3085 | (unsigned long long)inode->i_ino); | |
3086 | spin_unlock(&dentry->d_lock); | |
3087 | spin_unlock(&dentry->d_parent->d_lock); | |
3088 | d_instantiate(dentry, inode); | |
1da177e4 | 3089 | } |
60545d0d | 3090 | EXPORT_SYMBOL(d_tmpfile); |
1da177e4 LT |
3091 | |
3092 | static __initdata unsigned long dhash_entries; | |
3093 | static int __init set_dhash_entries(char *str) | |
3094 | { | |
3095 | if (!str) | |
3096 | return 0; | |
3097 | dhash_entries = simple_strtoul(str, &str, 0); | |
3098 | return 1; | |
3099 | } | |
3100 | __setup("dhash_entries=", set_dhash_entries); | |
3101 | ||
3102 | static void __init dcache_init_early(void) | |
3103 | { | |
074b8517 | 3104 | unsigned int loop; |
1da177e4 LT |
3105 | |
3106 | /* If hashes are distributed across NUMA nodes, defer | |
3107 | * hash allocation until vmalloc space is available. | |
3108 | */ | |
3109 | if (hashdist) | |
3110 | return; | |
3111 | ||
3112 | dentry_hashtable = | |
3113 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3114 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3115 | dhash_entries, |
3116 | 13, | |
3117 | HASH_EARLY, | |
3118 | &d_hash_shift, | |
3119 | &d_hash_mask, | |
31fe62b9 | 3120 | 0, |
1da177e4 LT |
3121 | 0); |
3122 | ||
074b8517 | 3123 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3124 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3125 | } |
3126 | ||
74bf17cf | 3127 | static void __init dcache_init(void) |
1da177e4 | 3128 | { |
074b8517 | 3129 | unsigned int loop; |
1da177e4 LT |
3130 | |
3131 | /* | |
3132 | * A constructor could be added for stable state like the lists, | |
3133 | * but it is probably not worth it because of the cache nature | |
3134 | * of the dcache. | |
3135 | */ | |
0a31bd5f CL |
3136 | dentry_cache = KMEM_CACHE(dentry, |
3137 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|SLAB_MEM_SPREAD); | |
1da177e4 LT |
3138 | |
3139 | /* Hash may have been set up in dcache_init_early */ | |
3140 | if (!hashdist) | |
3141 | return; | |
3142 | ||
3143 | dentry_hashtable = | |
3144 | alloc_large_system_hash("Dentry cache", | |
b07ad996 | 3145 | sizeof(struct hlist_bl_head), |
1da177e4 LT |
3146 | dhash_entries, |
3147 | 13, | |
3148 | 0, | |
3149 | &d_hash_shift, | |
3150 | &d_hash_mask, | |
31fe62b9 | 3151 | 0, |
1da177e4 LT |
3152 | 0); |
3153 | ||
074b8517 | 3154 | for (loop = 0; loop < (1U << d_hash_shift); loop++) |
b07ad996 | 3155 | INIT_HLIST_BL_HEAD(dentry_hashtable + loop); |
1da177e4 LT |
3156 | } |
3157 | ||
3158 | /* SLAB cache for __getname() consumers */ | |
e18b890b | 3159 | struct kmem_cache *names_cachep __read_mostly; |
ec4f8605 | 3160 | EXPORT_SYMBOL(names_cachep); |
1da177e4 | 3161 | |
1da177e4 LT |
3162 | EXPORT_SYMBOL(d_genocide); |
3163 | ||
1da177e4 LT |
3164 | void __init vfs_caches_init_early(void) |
3165 | { | |
3166 | dcache_init_early(); | |
3167 | inode_init_early(); | |
3168 | } | |
3169 | ||
3170 | void __init vfs_caches_init(unsigned long mempages) | |
3171 | { | |
3172 | unsigned long reserve; | |
3173 | ||
3174 | /* Base hash sizes on available memory, with a reserve equal to | |
3175 | 150% of current kernel size */ | |
3176 | ||
3177 | reserve = min((mempages - nr_free_pages()) * 3/2, mempages - 1); | |
3178 | mempages -= reserve; | |
3179 | ||
3180 | names_cachep = kmem_cache_create("names_cache", PATH_MAX, 0, | |
20c2df83 | 3181 | SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); |
1da177e4 | 3182 | |
74bf17cf DC |
3183 | dcache_init(); |
3184 | inode_init(); | |
1da177e4 | 3185 | files_init(mempages); |
74bf17cf | 3186 | mnt_init(); |
1da177e4 LT |
3187 | bdev_cache_init(); |
3188 | chrdev_init(); | |
3189 | } |