]>
Commit | Line | Data |
---|---|---|
b8441ed2 TH |
1 | /* |
2 | * fs/kernfs/dir.c - kernfs directory implementation | |
3 | * | |
4 | * Copyright (c) 2001-3 Patrick Mochel | |
5 | * Copyright (c) 2007 SUSE Linux Products GmbH | |
6 | * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org> | |
7 | * | |
8 | * This file is released under the GPLv2. | |
9 | */ | |
fd7b9f7b | 10 | |
abd54f02 | 11 | #include <linux/sched.h> |
fd7b9f7b TH |
12 | #include <linux/fs.h> |
13 | #include <linux/namei.h> | |
14 | #include <linux/idr.h> | |
15 | #include <linux/slab.h> | |
16 | #include <linux/security.h> | |
17 | #include <linux/hash.h> | |
18 | ||
19 | #include "kernfs-internal.h" | |
20 | ||
a797bfc3 | 21 | DEFINE_MUTEX(kernfs_mutex); |
3eef34ad TH |
22 | static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */ |
23 | static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */ | |
fd7b9f7b | 24 | |
adc5e8b5 | 25 | #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb) |
fd7b9f7b | 26 | |
81c173cb TH |
27 | static bool kernfs_active(struct kernfs_node *kn) |
28 | { | |
29 | lockdep_assert_held(&kernfs_mutex); | |
30 | return atomic_read(&kn->active) >= 0; | |
31 | } | |
32 | ||
182fd64b TH |
33 | static bool kernfs_lockdep(struct kernfs_node *kn) |
34 | { | |
35 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
36 | return kn->flags & KERNFS_LOCKDEP; | |
37 | #else | |
38 | return false; | |
39 | #endif | |
40 | } | |
41 | ||
3eef34ad TH |
42 | static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen) |
43 | { | |
44 | return strlcpy(buf, kn->parent ? kn->name : "/", buflen); | |
45 | } | |
46 | ||
47 | static char * __must_check kernfs_path_locked(struct kernfs_node *kn, char *buf, | |
48 | size_t buflen) | |
49 | { | |
50 | char *p = buf + buflen; | |
51 | int len; | |
52 | ||
53 | *--p = '\0'; | |
54 | ||
55 | do { | |
56 | len = strlen(kn->name); | |
57 | if (p - buf < len + 1) { | |
58 | buf[0] = '\0'; | |
59 | p = NULL; | |
60 | break; | |
61 | } | |
62 | p -= len; | |
63 | memcpy(p, kn->name, len); | |
64 | *--p = '/'; | |
65 | kn = kn->parent; | |
66 | } while (kn && kn->parent); | |
67 | ||
68 | return p; | |
69 | } | |
70 | ||
71 | /** | |
72 | * kernfs_name - obtain the name of a given node | |
73 | * @kn: kernfs_node of interest | |
74 | * @buf: buffer to copy @kn's name into | |
75 | * @buflen: size of @buf | |
76 | * | |
77 | * Copies the name of @kn into @buf of @buflen bytes. The behavior is | |
78 | * similar to strlcpy(). It returns the length of @kn's name and if @buf | |
79 | * isn't long enough, it's filled upto @buflen-1 and nul terminated. | |
80 | * | |
81 | * This function can be called from any context. | |
82 | */ | |
83 | int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen) | |
84 | { | |
85 | unsigned long flags; | |
86 | int ret; | |
87 | ||
88 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
89 | ret = kernfs_name_locked(kn, buf, buflen); | |
90 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
91 | return ret; | |
92 | } | |
93 | ||
9acee9c5 TH |
94 | /** |
95 | * kernfs_path_len - determine the length of the full path of a given node | |
96 | * @kn: kernfs_node of interest | |
97 | * | |
98 | * The returned length doesn't include the space for the terminating '\0'. | |
99 | */ | |
100 | size_t kernfs_path_len(struct kernfs_node *kn) | |
101 | { | |
102 | size_t len = 0; | |
103 | unsigned long flags; | |
104 | ||
105 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
106 | ||
107 | do { | |
108 | len += strlen(kn->name) + 1; | |
109 | kn = kn->parent; | |
110 | } while (kn && kn->parent); | |
111 | ||
112 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
113 | ||
114 | return len; | |
115 | } | |
116 | ||
3eef34ad TH |
117 | /** |
118 | * kernfs_path - build full path of a given node | |
119 | * @kn: kernfs_node of interest | |
120 | * @buf: buffer to copy @kn's name into | |
121 | * @buflen: size of @buf | |
122 | * | |
123 | * Builds and returns the full path of @kn in @buf of @buflen bytes. The | |
124 | * path is built from the end of @buf so the returned pointer usually | |
125 | * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated | |
126 | * and %NULL is returned. | |
127 | */ | |
128 | char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen) | |
129 | { | |
130 | unsigned long flags; | |
131 | char *p; | |
132 | ||
133 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
134 | p = kernfs_path_locked(kn, buf, buflen); | |
135 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
136 | return p; | |
137 | } | |
e61734c5 | 138 | EXPORT_SYMBOL_GPL(kernfs_path); |
3eef34ad TH |
139 | |
140 | /** | |
141 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node | |
142 | * @kn: kernfs_node of interest | |
143 | * | |
144 | * This function can be called from any context. | |
145 | */ | |
146 | void pr_cont_kernfs_name(struct kernfs_node *kn) | |
147 | { | |
148 | unsigned long flags; | |
149 | ||
150 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
151 | ||
152 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); | |
153 | pr_cont("%s", kernfs_pr_cont_buf); | |
154 | ||
155 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
156 | } | |
157 | ||
158 | /** | |
159 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node | |
160 | * @kn: kernfs_node of interest | |
161 | * | |
162 | * This function can be called from any context. | |
163 | */ | |
164 | void pr_cont_kernfs_path(struct kernfs_node *kn) | |
165 | { | |
166 | unsigned long flags; | |
167 | char *p; | |
168 | ||
169 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
170 | ||
171 | p = kernfs_path_locked(kn, kernfs_pr_cont_buf, | |
172 | sizeof(kernfs_pr_cont_buf)); | |
173 | if (p) | |
174 | pr_cont("%s", p); | |
175 | else | |
176 | pr_cont("<name too long>"); | |
177 | ||
178 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
179 | } | |
180 | ||
181 | /** | |
182 | * kernfs_get_parent - determine the parent node and pin it | |
183 | * @kn: kernfs_node of interest | |
184 | * | |
185 | * Determines @kn's parent, pins and returns it. This function can be | |
186 | * called from any context. | |
187 | */ | |
188 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) | |
189 | { | |
190 | struct kernfs_node *parent; | |
191 | unsigned long flags; | |
192 | ||
193 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
194 | parent = kn->parent; | |
195 | kernfs_get(parent); | |
196 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
197 | ||
198 | return parent; | |
199 | } | |
200 | ||
fd7b9f7b | 201 | /** |
c637b8ac | 202 | * kernfs_name_hash |
fd7b9f7b TH |
203 | * @name: Null terminated string to hash |
204 | * @ns: Namespace tag to hash | |
205 | * | |
206 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
207 | */ | |
c637b8ac | 208 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b TH |
209 | { |
210 | unsigned long hash = init_name_hash(); | |
211 | unsigned int len = strlen(name); | |
212 | while (len--) | |
213 | hash = partial_name_hash(*name++, hash); | |
214 | hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); | |
215 | hash &= 0x7fffffffU; | |
216 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
88391d49 | 217 | if (hash < 2) |
fd7b9f7b TH |
218 | hash += 2; |
219 | if (hash >= INT_MAX) | |
220 | hash = INT_MAX - 1; | |
221 | return hash; | |
222 | } | |
223 | ||
c637b8ac TH |
224 | static int kernfs_name_compare(unsigned int hash, const char *name, |
225 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 226 | { |
72392ed0 RV |
227 | if (hash < kn->hash) |
228 | return -1; | |
229 | if (hash > kn->hash) | |
230 | return 1; | |
231 | if (ns < kn->ns) | |
232 | return -1; | |
233 | if (ns > kn->ns) | |
234 | return 1; | |
adc5e8b5 | 235 | return strcmp(name, kn->name); |
fd7b9f7b TH |
236 | } |
237 | ||
c637b8ac TH |
238 | static int kernfs_sd_compare(const struct kernfs_node *left, |
239 | const struct kernfs_node *right) | |
fd7b9f7b | 240 | { |
c637b8ac | 241 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
242 | } |
243 | ||
244 | /** | |
c637b8ac | 245 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 246 | * @kn: kernfs_node of interest |
fd7b9f7b | 247 | * |
324a56e1 | 248 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 249 | * @kn->parent->dir.children. |
fd7b9f7b TH |
250 | * |
251 | * Locking: | |
a797bfc3 | 252 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
253 | * |
254 | * RETURNS: | |
255 | * 0 on susccess -EEXIST on failure. | |
256 | */ | |
c637b8ac | 257 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 258 | { |
adc5e8b5 | 259 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
260 | struct rb_node *parent = NULL; |
261 | ||
fd7b9f7b | 262 | while (*node) { |
324a56e1 | 263 | struct kernfs_node *pos; |
fd7b9f7b TH |
264 | int result; |
265 | ||
324a56e1 | 266 | pos = rb_to_kn(*node); |
fd7b9f7b | 267 | parent = *node; |
c637b8ac | 268 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 269 | if (result < 0) |
adc5e8b5 | 270 | node = &pos->rb.rb_left; |
fd7b9f7b | 271 | else if (result > 0) |
adc5e8b5 | 272 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
273 | else |
274 | return -EEXIST; | |
275 | } | |
c1befb88 | 276 | |
fd7b9f7b | 277 | /* add new node and rebalance the tree */ |
adc5e8b5 TH |
278 | rb_link_node(&kn->rb, parent, node); |
279 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
c1befb88 JZ |
280 | |
281 | /* successfully added, account subdir number */ | |
282 | if (kernfs_type(kn) == KERNFS_DIR) | |
283 | kn->parent->dir.subdirs++; | |
284 | ||
fd7b9f7b TH |
285 | return 0; |
286 | } | |
287 | ||
288 | /** | |
c637b8ac | 289 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 290 | * @kn: kernfs_node of interest |
fd7b9f7b | 291 | * |
35beab06 TH |
292 | * Try to unlink @kn from its sibling rbtree which starts from |
293 | * kn->parent->dir.children. Returns %true if @kn was actually | |
294 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
295 | * |
296 | * Locking: | |
a797bfc3 | 297 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 298 | */ |
35beab06 | 299 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 300 | { |
35beab06 TH |
301 | if (RB_EMPTY_NODE(&kn->rb)) |
302 | return false; | |
303 | ||
df23fc39 | 304 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 305 | kn->parent->dir.subdirs--; |
fd7b9f7b | 306 | |
adc5e8b5 | 307 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
308 | RB_CLEAR_NODE(&kn->rb); |
309 | return true; | |
fd7b9f7b TH |
310 | } |
311 | ||
312 | /** | |
c637b8ac | 313 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 314 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 315 | * |
324a56e1 | 316 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
317 | * is NULL. |
318 | * | |
319 | * RETURNS: | |
324a56e1 | 320 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 321 | */ |
c637b8ac | 322 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 323 | { |
324a56e1 | 324 | if (unlikely(!kn)) |
fd7b9f7b TH |
325 | return NULL; |
326 | ||
f4b3e631 GKH |
327 | if (!atomic_inc_unless_negative(&kn->active)) |
328 | return NULL; | |
895a068a | 329 | |
182fd64b | 330 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
331 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
332 | return kn; | |
fd7b9f7b TH |
333 | } |
334 | ||
335 | /** | |
c637b8ac | 336 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 337 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 338 | * |
324a56e1 | 339 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
340 | * is NULL. |
341 | */ | |
c637b8ac | 342 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b | 343 | { |
abd54f02 | 344 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b TH |
345 | int v; |
346 | ||
324a56e1 | 347 | if (unlikely(!kn)) |
fd7b9f7b TH |
348 | return; |
349 | ||
182fd64b | 350 | if (kernfs_lockdep(kn)) |
324a56e1 | 351 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
adc5e8b5 | 352 | v = atomic_dec_return(&kn->active); |
df23fc39 | 353 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
354 | return; |
355 | ||
abd54f02 | 356 | wake_up_all(&root->deactivate_waitq); |
fd7b9f7b TH |
357 | } |
358 | ||
359 | /** | |
81c173cb TH |
360 | * kernfs_drain - drain kernfs_node |
361 | * @kn: kernfs_node to drain | |
fd7b9f7b | 362 | * |
81c173cb TH |
363 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
364 | * removers may invoke this function concurrently on @kn and all will | |
365 | * return after draining is complete. | |
fd7b9f7b | 366 | */ |
81c173cb | 367 | static void kernfs_drain(struct kernfs_node *kn) |
35beab06 | 368 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
fd7b9f7b | 369 | { |
abd54f02 | 370 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b | 371 | |
35beab06 | 372 | lockdep_assert_held(&kernfs_mutex); |
81c173cb | 373 | WARN_ON_ONCE(kernfs_active(kn)); |
ea1c472d | 374 | |
35beab06 | 375 | mutex_unlock(&kernfs_mutex); |
abd54f02 | 376 | |
182fd64b | 377 | if (kernfs_lockdep(kn)) { |
35beab06 TH |
378 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
379 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) | |
380 | lock_contended(&kn->dep_map, _RET_IP_); | |
381 | } | |
abd54f02 | 382 | |
35beab06 | 383 | /* but everyone should wait for draining */ |
abd54f02 TH |
384 | wait_event(root->deactivate_waitq, |
385 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); | |
fd7b9f7b | 386 | |
182fd64b | 387 | if (kernfs_lockdep(kn)) { |
a6607930 TH |
388 | lock_acquired(&kn->dep_map, _RET_IP_); |
389 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | |
390 | } | |
35beab06 | 391 | |
ccf02aaf TH |
392 | kernfs_unmap_bin_file(kn); |
393 | ||
35beab06 | 394 | mutex_lock(&kernfs_mutex); |
fd7b9f7b TH |
395 | } |
396 | ||
fd7b9f7b | 397 | /** |
324a56e1 TH |
398 | * kernfs_get - get a reference count on a kernfs_node |
399 | * @kn: the target kernfs_node | |
fd7b9f7b | 400 | */ |
324a56e1 | 401 | void kernfs_get(struct kernfs_node *kn) |
fd7b9f7b | 402 | { |
324a56e1 | 403 | if (kn) { |
adc5e8b5 TH |
404 | WARN_ON(!atomic_read(&kn->count)); |
405 | atomic_inc(&kn->count); | |
fd7b9f7b TH |
406 | } |
407 | } | |
408 | EXPORT_SYMBOL_GPL(kernfs_get); | |
409 | ||
410 | /** | |
324a56e1 TH |
411 | * kernfs_put - put a reference count on a kernfs_node |
412 | * @kn: the target kernfs_node | |
fd7b9f7b | 413 | * |
324a56e1 | 414 | * Put a reference count of @kn and destroy it if it reached zero. |
fd7b9f7b | 415 | */ |
324a56e1 | 416 | void kernfs_put(struct kernfs_node *kn) |
fd7b9f7b | 417 | { |
324a56e1 | 418 | struct kernfs_node *parent; |
ba7443bc | 419 | struct kernfs_root *root; |
fd7b9f7b | 420 | |
adc5e8b5 | 421 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 422 | return; |
324a56e1 | 423 | root = kernfs_root(kn); |
fd7b9f7b | 424 | repeat: |
81c173cb TH |
425 | /* |
426 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 427 | * kn->parent won't change beneath us. |
fd7b9f7b | 428 | */ |
adc5e8b5 | 429 | parent = kn->parent; |
fd7b9f7b | 430 | |
81c173cb TH |
431 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
432 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
433 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 434 | |
df23fc39 | 435 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 436 | kernfs_put(kn->symlink.target_kn); |
dfeb0750 TH |
437 | |
438 | kfree_const(kn->name); | |
439 | ||
adc5e8b5 TH |
440 | if (kn->iattr) { |
441 | if (kn->iattr->ia_secdata) | |
442 | security_release_secctx(kn->iattr->ia_secdata, | |
443 | kn->iattr->ia_secdata_len); | |
444 | simple_xattrs_free(&kn->iattr->xattrs); | |
2322392b | 445 | } |
adc5e8b5 TH |
446 | kfree(kn->iattr); |
447 | ida_simple_remove(&root->ino_ida, kn->ino); | |
a797bfc3 | 448 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 449 | |
324a56e1 TH |
450 | kn = parent; |
451 | if (kn) { | |
adc5e8b5 | 452 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
453 | goto repeat; |
454 | } else { | |
324a56e1 | 455 | /* just released the root kn, free @root too */ |
bc755553 | 456 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
457 | kfree(root); |
458 | } | |
fd7b9f7b TH |
459 | } |
460 | EXPORT_SYMBOL_GPL(kernfs_put); | |
461 | ||
c637b8ac | 462 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 463 | { |
324a56e1 | 464 | struct kernfs_node *kn; |
fd7b9f7b TH |
465 | |
466 | if (flags & LOOKUP_RCU) | |
467 | return -ECHILD; | |
468 | ||
19bbb926 | 469 | /* Always perform fresh lookup for negatives */ |
2b0143b5 | 470 | if (d_really_is_negative(dentry)) |
19bbb926 TH |
471 | goto out_bad_unlocked; |
472 | ||
324a56e1 | 473 | kn = dentry->d_fsdata; |
a797bfc3 | 474 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 475 | |
81c173cb TH |
476 | /* The kernfs node has been deactivated */ |
477 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
478 | goto out_bad; |
479 | ||
c637b8ac | 480 | /* The kernfs node has been moved? */ |
adc5e8b5 | 481 | if (dentry->d_parent->d_fsdata != kn->parent) |
fd7b9f7b TH |
482 | goto out_bad; |
483 | ||
c637b8ac | 484 | /* The kernfs node has been renamed */ |
adc5e8b5 | 485 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
486 | goto out_bad; |
487 | ||
c637b8ac | 488 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 489 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 490 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
491 | goto out_bad; |
492 | ||
a797bfc3 | 493 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
494 | return 1; |
495 | out_bad: | |
a797bfc3 | 496 | mutex_unlock(&kernfs_mutex); |
19bbb926 | 497 | out_bad_unlocked: |
fd7b9f7b TH |
498 | return 0; |
499 | } | |
500 | ||
c637b8ac | 501 | static void kernfs_dop_release(struct dentry *dentry) |
fd7b9f7b TH |
502 | { |
503 | kernfs_put(dentry->d_fsdata); | |
504 | } | |
505 | ||
a797bfc3 | 506 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 507 | .d_revalidate = kernfs_dop_revalidate, |
c637b8ac | 508 | .d_release = kernfs_dop_release, |
fd7b9f7b TH |
509 | }; |
510 | ||
0c23b225 TH |
511 | /** |
512 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry | |
513 | * @dentry: the dentry in question | |
514 | * | |
515 | * Return the kernfs_node associated with @dentry. If @dentry is not a | |
516 | * kernfs one, %NULL is returned. | |
517 | * | |
518 | * While the returned kernfs_node will stay accessible as long as @dentry | |
519 | * is accessible, the returned node can be in any state and the caller is | |
520 | * fully responsible for determining what's accessible. | |
521 | */ | |
522 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) | |
523 | { | |
f41c5934 | 524 | if (dentry->d_sb->s_op == &kernfs_sops) |
0c23b225 TH |
525 | return dentry->d_fsdata; |
526 | return NULL; | |
527 | } | |
528 | ||
db4aad20 TH |
529 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
530 | const char *name, umode_t mode, | |
531 | unsigned flags) | |
fd7b9f7b | 532 | { |
324a56e1 | 533 | struct kernfs_node *kn; |
bc755553 | 534 | int ret; |
fd7b9f7b | 535 | |
dfeb0750 TH |
536 | name = kstrdup_const(name, GFP_KERNEL); |
537 | if (!name) | |
538 | return NULL; | |
fd7b9f7b | 539 | |
a797bfc3 | 540 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 541 | if (!kn) |
fd7b9f7b TH |
542 | goto err_out1; |
543 | ||
499611ed VD |
544 | /* |
545 | * If the ino of the sysfs entry created for a kmem cache gets | |
546 | * allocated from an ida layer, which is accounted to the memcg that | |
547 | * owns the cache, the memcg will get pinned forever. So do not account | |
548 | * ino ida allocations. | |
549 | */ | |
550 | ret = ida_simple_get(&root->ino_ida, 1, 0, | |
551 | GFP_KERNEL | __GFP_NOACCOUNT); | |
bc755553 | 552 | if (ret < 0) |
fd7b9f7b | 553 | goto err_out2; |
adc5e8b5 | 554 | kn->ino = ret; |
fd7b9f7b | 555 | |
adc5e8b5 | 556 | atomic_set(&kn->count, 1); |
81c173cb | 557 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 558 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 559 | |
adc5e8b5 TH |
560 | kn->name = name; |
561 | kn->mode = mode; | |
81c173cb | 562 | kn->flags = flags; |
fd7b9f7b | 563 | |
324a56e1 | 564 | return kn; |
fd7b9f7b TH |
565 | |
566 | err_out2: | |
a797bfc3 | 567 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 568 | err_out1: |
dfeb0750 | 569 | kfree_const(name); |
fd7b9f7b TH |
570 | return NULL; |
571 | } | |
572 | ||
db4aad20 TH |
573 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
574 | const char *name, umode_t mode, | |
575 | unsigned flags) | |
576 | { | |
577 | struct kernfs_node *kn; | |
578 | ||
579 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); | |
580 | if (kn) { | |
581 | kernfs_get(parent); | |
582 | kn->parent = parent; | |
583 | } | |
584 | return kn; | |
585 | } | |
586 | ||
fd7b9f7b | 587 | /** |
c637b8ac | 588 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 589 | * @kn: kernfs_node to be added |
fd7b9f7b | 590 | * |
db4aad20 TH |
591 | * The caller must already have initialized @kn->parent. This |
592 | * function increments nlink of the parent's inode if @kn is a | |
593 | * directory and link into the children list of the parent. | |
fd7b9f7b | 594 | * |
fd7b9f7b TH |
595 | * RETURNS: |
596 | * 0 on success, -EEXIST if entry with the given name already | |
597 | * exists. | |
598 | */ | |
988cd7af | 599 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 600 | { |
db4aad20 | 601 | struct kernfs_node *parent = kn->parent; |
c525aadd | 602 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 603 | bool has_ns; |
fd7b9f7b TH |
604 | int ret; |
605 | ||
988cd7af TH |
606 | mutex_lock(&kernfs_mutex); |
607 | ||
608 | ret = -EINVAL; | |
609 | has_ns = kernfs_ns_enabled(parent); | |
610 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
611 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
612 | goto out_unlock; | |
fd7b9f7b | 613 | |
df23fc39 | 614 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 615 | goto out_unlock; |
fd7b9f7b | 616 | |
988cd7af | 617 | ret = -ENOENT; |
ea015218 EB |
618 | if (parent->flags & KERNFS_EMPTY_DIR) |
619 | goto out_unlock; | |
620 | ||
d35258ef | 621 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 622 | goto out_unlock; |
798c75a0 | 623 | |
c637b8ac | 624 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 625 | |
c637b8ac | 626 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 627 | if (ret) |
988cd7af | 628 | goto out_unlock; |
fd7b9f7b TH |
629 | |
630 | /* Update timestamps on the parent */ | |
adc5e8b5 | 631 | ps_iattr = parent->iattr; |
fd7b9f7b TH |
632 | if (ps_iattr) { |
633 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
634 | ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME; | |
635 | } | |
636 | ||
d35258ef TH |
637 | mutex_unlock(&kernfs_mutex); |
638 | ||
639 | /* | |
640 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
641 | * If not activated here, the kernfs user is responsible for | |
642 | * activating the node with kernfs_activate(). A node which hasn't | |
643 | * been activated is not visible to userland and its removal won't | |
644 | * trigger deactivation. | |
645 | */ | |
646 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
647 | kernfs_activate(kn); | |
648 | return 0; | |
649 | ||
988cd7af | 650 | out_unlock: |
a797bfc3 | 651 | mutex_unlock(&kernfs_mutex); |
988cd7af | 652 | return ret; |
fd7b9f7b TH |
653 | } |
654 | ||
655 | /** | |
324a56e1 TH |
656 | * kernfs_find_ns - find kernfs_node with the given name |
657 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
658 | * @name: name to look for |
659 | * @ns: the namespace tag to use | |
660 | * | |
324a56e1 TH |
661 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
662 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 663 | */ |
324a56e1 TH |
664 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
665 | const unsigned char *name, | |
666 | const void *ns) | |
fd7b9f7b | 667 | { |
adc5e8b5 | 668 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 669 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
670 | unsigned int hash; |
671 | ||
a797bfc3 | 672 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
673 | |
674 | if (has_ns != (bool)ns) { | |
c637b8ac | 675 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 676 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
677 | return NULL; |
678 | } | |
679 | ||
c637b8ac | 680 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 681 | while (node) { |
324a56e1 | 682 | struct kernfs_node *kn; |
fd7b9f7b TH |
683 | int result; |
684 | ||
324a56e1 | 685 | kn = rb_to_kn(node); |
c637b8ac | 686 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
687 | if (result < 0) |
688 | node = node->rb_left; | |
689 | else if (result > 0) | |
690 | node = node->rb_right; | |
691 | else | |
324a56e1 | 692 | return kn; |
fd7b9f7b TH |
693 | } |
694 | return NULL; | |
695 | } | |
696 | ||
bd96f76a TH |
697 | static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent, |
698 | const unsigned char *path, | |
699 | const void *ns) | |
700 | { | |
701 | static char path_buf[PATH_MAX]; /* protected by kernfs_mutex */ | |
702 | size_t len = strlcpy(path_buf, path, PATH_MAX); | |
703 | char *p = path_buf; | |
704 | char *name; | |
705 | ||
706 | lockdep_assert_held(&kernfs_mutex); | |
707 | ||
708 | if (len >= PATH_MAX) | |
709 | return NULL; | |
710 | ||
711 | while ((name = strsep(&p, "/")) && parent) { | |
712 | if (*name == '\0') | |
713 | continue; | |
714 | parent = kernfs_find_ns(parent, name, ns); | |
715 | } | |
716 | ||
717 | return parent; | |
718 | } | |
719 | ||
fd7b9f7b | 720 | /** |
324a56e1 TH |
721 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
722 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
723 | * @name: name to look for |
724 | * @ns: the namespace tag to use | |
725 | * | |
324a56e1 | 726 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 727 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 728 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 729 | */ |
324a56e1 TH |
730 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
731 | const char *name, const void *ns) | |
fd7b9f7b | 732 | { |
324a56e1 | 733 | struct kernfs_node *kn; |
fd7b9f7b | 734 | |
a797bfc3 | 735 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
736 | kn = kernfs_find_ns(parent, name, ns); |
737 | kernfs_get(kn); | |
a797bfc3 | 738 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 739 | |
324a56e1 | 740 | return kn; |
fd7b9f7b TH |
741 | } |
742 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
743 | ||
bd96f76a TH |
744 | /** |
745 | * kernfs_walk_and_get_ns - find and get kernfs_node with the given path | |
746 | * @parent: kernfs_node to search under | |
747 | * @path: path to look for | |
748 | * @ns: the namespace tag to use | |
749 | * | |
750 | * Look for kernfs_node with path @path under @parent and get a reference | |
751 | * if found. This function may sleep and returns pointer to the found | |
752 | * kernfs_node on success, %NULL on failure. | |
753 | */ | |
754 | struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent, | |
755 | const char *path, const void *ns) | |
756 | { | |
757 | struct kernfs_node *kn; | |
758 | ||
759 | mutex_lock(&kernfs_mutex); | |
760 | kn = kernfs_walk_ns(parent, path, ns); | |
761 | kernfs_get(kn); | |
762 | mutex_unlock(&kernfs_mutex); | |
763 | ||
764 | return kn; | |
765 | } | |
766 | ||
ba7443bc TH |
767 | /** |
768 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 769 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 770 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
771 | * @priv: opaque data associated with the new directory |
772 | * | |
773 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
774 | * failure. | |
775 | */ | |
90c07c89 | 776 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 777 | unsigned int flags, void *priv) |
ba7443bc TH |
778 | { |
779 | struct kernfs_root *root; | |
324a56e1 | 780 | struct kernfs_node *kn; |
ba7443bc TH |
781 | |
782 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
783 | if (!root) | |
784 | return ERR_PTR(-ENOMEM); | |
785 | ||
bc755553 | 786 | ida_init(&root->ino_ida); |
7d568a83 | 787 | INIT_LIST_HEAD(&root->supers); |
bc755553 | 788 | |
db4aad20 TH |
789 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
790 | KERNFS_DIR); | |
324a56e1 | 791 | if (!kn) { |
bc755553 | 792 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
793 | kfree(root); |
794 | return ERR_PTR(-ENOMEM); | |
795 | } | |
796 | ||
324a56e1 | 797 | kn->priv = priv; |
adc5e8b5 | 798 | kn->dir.root = root; |
ba7443bc | 799 | |
90c07c89 | 800 | root->syscall_ops = scops; |
d35258ef | 801 | root->flags = flags; |
324a56e1 | 802 | root->kn = kn; |
abd54f02 | 803 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 804 | |
d35258ef TH |
805 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
806 | kernfs_activate(kn); | |
807 | ||
ba7443bc TH |
808 | return root; |
809 | } | |
810 | ||
811 | /** | |
812 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
813 | * @root: root of the hierarchy to destroy | |
814 | * | |
815 | * Destroy the hierarchy anchored at @root by removing all existing | |
816 | * directories and destroying @root. | |
817 | */ | |
818 | void kernfs_destroy_root(struct kernfs_root *root) | |
819 | { | |
324a56e1 | 820 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
821 | } |
822 | ||
fd7b9f7b TH |
823 | /** |
824 | * kernfs_create_dir_ns - create a directory | |
825 | * @parent: parent in which to create a new directory | |
826 | * @name: name of the new directory | |
bb8b9d09 | 827 | * @mode: mode of the new directory |
fd7b9f7b TH |
828 | * @priv: opaque data associated with the new directory |
829 | * @ns: optional namespace tag of the directory | |
830 | * | |
831 | * Returns the created node on success, ERR_PTR() value on failure. | |
832 | */ | |
324a56e1 | 833 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 TH |
834 | const char *name, umode_t mode, |
835 | void *priv, const void *ns) | |
fd7b9f7b | 836 | { |
324a56e1 | 837 | struct kernfs_node *kn; |
fd7b9f7b TH |
838 | int rc; |
839 | ||
840 | /* allocate */ | |
db4aad20 | 841 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); |
324a56e1 | 842 | if (!kn) |
fd7b9f7b TH |
843 | return ERR_PTR(-ENOMEM); |
844 | ||
adc5e8b5 TH |
845 | kn->dir.root = parent->dir.root; |
846 | kn->ns = ns; | |
324a56e1 | 847 | kn->priv = priv; |
fd7b9f7b TH |
848 | |
849 | /* link in */ | |
988cd7af | 850 | rc = kernfs_add_one(kn); |
fd7b9f7b | 851 | if (!rc) |
324a56e1 | 852 | return kn; |
fd7b9f7b | 853 | |
324a56e1 | 854 | kernfs_put(kn); |
fd7b9f7b TH |
855 | return ERR_PTR(rc); |
856 | } | |
857 | ||
ea015218 EB |
858 | /** |
859 | * kernfs_create_empty_dir - create an always empty directory | |
860 | * @parent: parent in which to create a new directory | |
861 | * @name: name of the new directory | |
862 | * | |
863 | * Returns the created node on success, ERR_PTR() value on failure. | |
864 | */ | |
865 | struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent, | |
866 | const char *name) | |
867 | { | |
868 | struct kernfs_node *kn; | |
869 | int rc; | |
870 | ||
871 | /* allocate */ | |
872 | kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR); | |
873 | if (!kn) | |
874 | return ERR_PTR(-ENOMEM); | |
875 | ||
876 | kn->flags |= KERNFS_EMPTY_DIR; | |
877 | kn->dir.root = parent->dir.root; | |
878 | kn->ns = NULL; | |
879 | kn->priv = NULL; | |
880 | ||
881 | /* link in */ | |
882 | rc = kernfs_add_one(kn); | |
883 | if (!rc) | |
884 | return kn; | |
885 | ||
886 | kernfs_put(kn); | |
887 | return ERR_PTR(rc); | |
888 | } | |
889 | ||
c637b8ac TH |
890 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
891 | struct dentry *dentry, | |
892 | unsigned int flags) | |
fd7b9f7b | 893 | { |
19bbb926 | 894 | struct dentry *ret; |
324a56e1 TH |
895 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; |
896 | struct kernfs_node *kn; | |
fd7b9f7b TH |
897 | struct inode *inode; |
898 | const void *ns = NULL; | |
899 | ||
a797bfc3 | 900 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 901 | |
324a56e1 | 902 | if (kernfs_ns_enabled(parent)) |
c525aadd | 903 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 904 | |
324a56e1 | 905 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
906 | |
907 | /* no such entry */ | |
b9c9dad0 | 908 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 909 | ret = NULL; |
fd7b9f7b TH |
910 | goto out_unlock; |
911 | } | |
324a56e1 TH |
912 | kernfs_get(kn); |
913 | dentry->d_fsdata = kn; | |
fd7b9f7b TH |
914 | |
915 | /* attach dentry and inode */ | |
c637b8ac | 916 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
917 | if (!inode) { |
918 | ret = ERR_PTR(-ENOMEM); | |
919 | goto out_unlock; | |
920 | } | |
921 | ||
922 | /* instantiate and hash dentry */ | |
41d28bca | 923 | ret = d_splice_alias(inode, dentry); |
fd7b9f7b | 924 | out_unlock: |
a797bfc3 | 925 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
926 | return ret; |
927 | } | |
928 | ||
80b9bbef TH |
929 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
930 | umode_t mode) | |
931 | { | |
932 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 933 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 934 | int ret; |
80b9bbef | 935 | |
90c07c89 | 936 | if (!scops || !scops->mkdir) |
80b9bbef TH |
937 | return -EPERM; |
938 | ||
07c7530d TH |
939 | if (!kernfs_get_active(parent)) |
940 | return -ENODEV; | |
941 | ||
90c07c89 | 942 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
943 | |
944 | kernfs_put_active(parent); | |
945 | return ret; | |
80b9bbef TH |
946 | } |
947 | ||
948 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
949 | { | |
950 | struct kernfs_node *kn = dentry->d_fsdata; | |
90c07c89 | 951 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 952 | int ret; |
80b9bbef | 953 | |
90c07c89 | 954 | if (!scops || !scops->rmdir) |
80b9bbef TH |
955 | return -EPERM; |
956 | ||
07c7530d TH |
957 | if (!kernfs_get_active(kn)) |
958 | return -ENODEV; | |
959 | ||
90c07c89 | 960 | ret = scops->rmdir(kn); |
07c7530d TH |
961 | |
962 | kernfs_put_active(kn); | |
963 | return ret; | |
80b9bbef TH |
964 | } |
965 | ||
966 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
967 | struct inode *new_dir, struct dentry *new_dentry) | |
968 | { | |
969 | struct kernfs_node *kn = old_dentry->d_fsdata; | |
970 | struct kernfs_node *new_parent = new_dir->i_private; | |
90c07c89 | 971 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 972 | int ret; |
80b9bbef | 973 | |
90c07c89 | 974 | if (!scops || !scops->rename) |
80b9bbef TH |
975 | return -EPERM; |
976 | ||
07c7530d TH |
977 | if (!kernfs_get_active(kn)) |
978 | return -ENODEV; | |
979 | ||
980 | if (!kernfs_get_active(new_parent)) { | |
981 | kernfs_put_active(kn); | |
982 | return -ENODEV; | |
983 | } | |
984 | ||
90c07c89 | 985 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
986 | |
987 | kernfs_put_active(new_parent); | |
988 | kernfs_put_active(kn); | |
989 | return ret; | |
80b9bbef TH |
990 | } |
991 | ||
a797bfc3 | 992 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
993 | .lookup = kernfs_iop_lookup, |
994 | .permission = kernfs_iop_permission, | |
995 | .setattr = kernfs_iop_setattr, | |
996 | .getattr = kernfs_iop_getattr, | |
997 | .setxattr = kernfs_iop_setxattr, | |
998 | .removexattr = kernfs_iop_removexattr, | |
999 | .getxattr = kernfs_iop_getxattr, | |
1000 | .listxattr = kernfs_iop_listxattr, | |
80b9bbef TH |
1001 | |
1002 | .mkdir = kernfs_iop_mkdir, | |
1003 | .rmdir = kernfs_iop_rmdir, | |
1004 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
1005 | }; |
1006 | ||
c637b8ac | 1007 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 1008 | { |
324a56e1 | 1009 | struct kernfs_node *last; |
fd7b9f7b TH |
1010 | |
1011 | while (true) { | |
1012 | struct rb_node *rbn; | |
1013 | ||
1014 | last = pos; | |
1015 | ||
df23fc39 | 1016 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
1017 | break; |
1018 | ||
adc5e8b5 | 1019 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
1020 | if (!rbn) |
1021 | break; | |
1022 | ||
324a56e1 | 1023 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
1024 | } |
1025 | ||
1026 | return last; | |
1027 | } | |
1028 | ||
1029 | /** | |
c637b8ac | 1030 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 1031 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 1032 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
1033 | * |
1034 | * Find the next descendant to visit for post-order traversal of @root's | |
1035 | * descendants. @root is included in the iteration and the last node to be | |
1036 | * visited. | |
1037 | */ | |
c637b8ac TH |
1038 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
1039 | struct kernfs_node *root) | |
fd7b9f7b TH |
1040 | { |
1041 | struct rb_node *rbn; | |
1042 | ||
a797bfc3 | 1043 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
1044 | |
1045 | /* if first iteration, visit leftmost descendant which may be root */ | |
1046 | if (!pos) | |
c637b8ac | 1047 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
1048 | |
1049 | /* if we visited @root, we're done */ | |
1050 | if (pos == root) | |
1051 | return NULL; | |
1052 | ||
1053 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 1054 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 1055 | if (rbn) |
c637b8ac | 1056 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
1057 | |
1058 | /* no sibling left, visit parent */ | |
adc5e8b5 | 1059 | return pos->parent; |
fd7b9f7b TH |
1060 | } |
1061 | ||
d35258ef TH |
1062 | /** |
1063 | * kernfs_activate - activate a node which started deactivated | |
1064 | * @kn: kernfs_node whose subtree is to be activated | |
1065 | * | |
1066 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
1067 | * needs to be explicitly activated. A node which hasn't been activated | |
1068 | * isn't visible to userland and deactivation is skipped during its | |
1069 | * removal. This is useful to construct atomic init sequences where | |
1070 | * creation of multiple nodes should either succeed or fail atomically. | |
1071 | * | |
1072 | * The caller is responsible for ensuring that this function is not called | |
1073 | * after kernfs_remove*() is invoked on @kn. | |
1074 | */ | |
1075 | void kernfs_activate(struct kernfs_node *kn) | |
1076 | { | |
1077 | struct kernfs_node *pos; | |
1078 | ||
1079 | mutex_lock(&kernfs_mutex); | |
1080 | ||
1081 | pos = NULL; | |
1082 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
1083 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) | |
1084 | continue; | |
1085 | ||
1086 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
1087 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
1088 | ||
1089 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
1090 | pos->flags |= KERNFS_ACTIVATED; | |
1091 | } | |
1092 | ||
1093 | mutex_unlock(&kernfs_mutex); | |
1094 | } | |
1095 | ||
988cd7af | 1096 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1097 | { |
35beab06 TH |
1098 | struct kernfs_node *pos; |
1099 | ||
1100 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 1101 | |
6b0afc2a TH |
1102 | /* |
1103 | * Short-circuit if non-root @kn has already finished removal. | |
1104 | * This is for kernfs_remove_self() which plays with active ref | |
1105 | * after removal. | |
1106 | */ | |
1107 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
1108 | return; |
1109 | ||
c637b8ac | 1110 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 1111 | |
81c173cb | 1112 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
1113 | pos = NULL; |
1114 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
1115 | if (kernfs_active(pos)) |
1116 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
1117 | |
1118 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 1119 | do { |
35beab06 TH |
1120 | pos = kernfs_leftmost_descendant(kn); |
1121 | ||
1122 | /* | |
81c173cb TH |
1123 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
1124 | * base ref could have been put by someone else by the time | |
1125 | * the function returns. Make sure it doesn't go away | |
1126 | * underneath us. | |
35beab06 TH |
1127 | */ |
1128 | kernfs_get(pos); | |
1129 | ||
d35258ef TH |
1130 | /* |
1131 | * Drain iff @kn was activated. This avoids draining and | |
1132 | * its lockdep annotations for nodes which have never been | |
1133 | * activated and allows embedding kernfs_remove() in create | |
1134 | * error paths without worrying about draining. | |
1135 | */ | |
1136 | if (kn->flags & KERNFS_ACTIVATED) | |
1137 | kernfs_drain(pos); | |
1138 | else | |
1139 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
1140 | |
1141 | /* | |
1142 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
1143 | * to decide who's responsible for cleanups. | |
1144 | */ | |
1145 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
1146 | struct kernfs_iattrs *ps_iattr = | |
1147 | pos->parent ? pos->parent->iattr : NULL; | |
1148 | ||
1149 | /* update timestamps on the parent */ | |
1150 | if (ps_iattr) { | |
1151 | ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME; | |
1152 | ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME; | |
1153 | } | |
1154 | ||
988cd7af | 1155 | kernfs_put(pos); |
35beab06 TH |
1156 | } |
1157 | ||
1158 | kernfs_put(pos); | |
1159 | } while (pos != kn); | |
fd7b9f7b TH |
1160 | } |
1161 | ||
1162 | /** | |
324a56e1 TH |
1163 | * kernfs_remove - remove a kernfs_node recursively |
1164 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 1165 | * |
324a56e1 | 1166 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 1167 | */ |
324a56e1 | 1168 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1169 | { |
988cd7af TH |
1170 | mutex_lock(&kernfs_mutex); |
1171 | __kernfs_remove(kn); | |
1172 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
1173 | } |
1174 | ||
6b0afc2a TH |
1175 | /** |
1176 | * kernfs_break_active_protection - break out of active protection | |
1177 | * @kn: the self kernfs_node | |
1178 | * | |
1179 | * The caller must be running off of a kernfs operation which is invoked | |
1180 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
1181 | * this function must also be matched with an invocation of | |
1182 | * kernfs_unbreak_active_protection(). | |
1183 | * | |
1184 | * This function releases the active reference of @kn the caller is | |
1185 | * holding. Once this function is called, @kn may be removed at any point | |
1186 | * and the caller is solely responsible for ensuring that the objects it | |
1187 | * dereferences are accessible. | |
1188 | */ | |
1189 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
1190 | { | |
1191 | /* | |
1192 | * Take out ourself out of the active ref dependency chain. If | |
1193 | * we're called without an active ref, lockdep will complain. | |
1194 | */ | |
1195 | kernfs_put_active(kn); | |
1196 | } | |
1197 | ||
1198 | /** | |
1199 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
1200 | * @kn: the self kernfs_node | |
1201 | * | |
1202 | * If kernfs_break_active_protection() was called, this function must be | |
1203 | * invoked before finishing the kernfs operation. Note that while this | |
1204 | * function restores the active reference, it doesn't and can't actually | |
1205 | * restore the active protection - @kn may already or be in the process of | |
1206 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
1207 | * protection is irreversibly gone for the kernfs operation instance. | |
1208 | * | |
1209 | * While this function may be called at any point after | |
1210 | * kernfs_break_active_protection() is invoked, its most useful location | |
1211 | * would be right before the enclosing kernfs operation returns. | |
1212 | */ | |
1213 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
1214 | { | |
1215 | /* | |
1216 | * @kn->active could be in any state; however, the increment we do | |
1217 | * here will be undone as soon as the enclosing kernfs operation | |
1218 | * finishes and this temporary bump can't break anything. If @kn | |
1219 | * is alive, nothing changes. If @kn is being deactivated, the | |
1220 | * soon-to-follow put will either finish deactivation or restore | |
1221 | * deactivated state. If @kn is already removed, the temporary | |
1222 | * bump is guaranteed to be gone before @kn is released. | |
1223 | */ | |
1224 | atomic_inc(&kn->active); | |
1225 | if (kernfs_lockdep(kn)) | |
1226 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
1227 | } | |
1228 | ||
1229 | /** | |
1230 | * kernfs_remove_self - remove a kernfs_node from its own method | |
1231 | * @kn: the self kernfs_node to remove | |
1232 | * | |
1233 | * The caller must be running off of a kernfs operation which is invoked | |
1234 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
1235 | * implement a file operation which deletes itself. | |
1236 | * | |
1237 | * For example, the "delete" file for a sysfs device directory can be | |
1238 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
1239 | * itself. This function breaks the circular dependency of trying to | |
1240 | * deactivate self while holding an active ref itself. It isn't necessary | |
1241 | * to modify the usual removal path to use kernfs_remove_self(). The | |
1242 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
1243 | * before proceeding with the usual removal path. kernfs will ignore later | |
1244 | * kernfs_remove() on self. | |
1245 | * | |
1246 | * kernfs_remove_self() can be called multiple times concurrently on the | |
1247 | * same kernfs_node. Only the first one actually performs removal and | |
1248 | * returns %true. All others will wait until the kernfs operation which | |
1249 | * won self-removal finishes and return %false. Note that the losers wait | |
1250 | * for the completion of not only the winning kernfs_remove_self() but also | |
1251 | * the whole kernfs_ops which won the arbitration. This can be used to | |
1252 | * guarantee, for example, all concurrent writes to a "delete" file to | |
1253 | * finish only after the whole operation is complete. | |
1254 | */ | |
1255 | bool kernfs_remove_self(struct kernfs_node *kn) | |
1256 | { | |
1257 | bool ret; | |
1258 | ||
1259 | mutex_lock(&kernfs_mutex); | |
1260 | kernfs_break_active_protection(kn); | |
1261 | ||
1262 | /* | |
1263 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1264 | * the first one will actually perform removal. When the removal | |
1265 | * is complete, SUICIDED is set and the active ref is restored | |
1266 | * while holding kernfs_mutex. The ones which lost arbitration | |
1267 | * waits for SUICDED && drained which can happen only after the | |
1268 | * enclosing kernfs operation which executed the winning instance | |
1269 | * of kernfs_remove_self() finished. | |
1270 | */ | |
1271 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1272 | kn->flags |= KERNFS_SUICIDAL; | |
1273 | __kernfs_remove(kn); | |
1274 | kn->flags |= KERNFS_SUICIDED; | |
1275 | ret = true; | |
1276 | } else { | |
1277 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1278 | DEFINE_WAIT(wait); | |
1279 | ||
1280 | while (true) { | |
1281 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1282 | ||
1283 | if ((kn->flags & KERNFS_SUICIDED) && | |
1284 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1285 | break; | |
1286 | ||
1287 | mutex_unlock(&kernfs_mutex); | |
1288 | schedule(); | |
1289 | mutex_lock(&kernfs_mutex); | |
1290 | } | |
1291 | finish_wait(waitq, &wait); | |
1292 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1293 | ret = false; | |
1294 | } | |
1295 | ||
1296 | /* | |
1297 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1298 | * for SUICIDED && deactivated could finish prematurely. | |
1299 | */ | |
1300 | kernfs_unbreak_active_protection(kn); | |
1301 | ||
1302 | mutex_unlock(&kernfs_mutex); | |
1303 | return ret; | |
1304 | } | |
1305 | ||
fd7b9f7b | 1306 | /** |
324a56e1 TH |
1307 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1308 | * @parent: parent of the target | |
1309 | * @name: name of the kernfs_node to remove | |
1310 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1311 | * |
324a56e1 TH |
1312 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1313 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1314 | */ |
324a56e1 | 1315 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1316 | const void *ns) |
1317 | { | |
324a56e1 | 1318 | struct kernfs_node *kn; |
fd7b9f7b | 1319 | |
324a56e1 | 1320 | if (!parent) { |
c637b8ac | 1321 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1322 | name); |
1323 | return -ENOENT; | |
1324 | } | |
1325 | ||
988cd7af | 1326 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1327 | |
324a56e1 TH |
1328 | kn = kernfs_find_ns(parent, name, ns); |
1329 | if (kn) | |
988cd7af | 1330 | __kernfs_remove(kn); |
fd7b9f7b | 1331 | |
988cd7af | 1332 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1333 | |
324a56e1 | 1334 | if (kn) |
fd7b9f7b TH |
1335 | return 0; |
1336 | else | |
1337 | return -ENOENT; | |
1338 | } | |
1339 | ||
1340 | /** | |
1341 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1342 | * @kn: target node |
fd7b9f7b TH |
1343 | * @new_parent: new parent to put @sd under |
1344 | * @new_name: new name | |
1345 | * @new_ns: new namespace tag | |
1346 | */ | |
324a56e1 | 1347 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1348 | const char *new_name, const void *new_ns) |
1349 | { | |
3eef34ad TH |
1350 | struct kernfs_node *old_parent; |
1351 | const char *old_name = NULL; | |
fd7b9f7b TH |
1352 | int error; |
1353 | ||
3eef34ad TH |
1354 | /* can't move or rename root */ |
1355 | if (!kn->parent) | |
1356 | return -EINVAL; | |
1357 | ||
798c75a0 GKH |
1358 | mutex_lock(&kernfs_mutex); |
1359 | ||
d0ae3d43 | 1360 | error = -ENOENT; |
ea015218 EB |
1361 | if (!kernfs_active(kn) || !kernfs_active(new_parent) || |
1362 | (new_parent->flags & KERNFS_EMPTY_DIR)) | |
d0ae3d43 TH |
1363 | goto out; |
1364 | ||
fd7b9f7b | 1365 | error = 0; |
adc5e8b5 TH |
1366 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1367 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1368 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1369 | |
1370 | error = -EEXIST; | |
1371 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1372 | goto out; |
fd7b9f7b | 1373 | |
324a56e1 | 1374 | /* rename kernfs_node */ |
adc5e8b5 | 1375 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b | 1376 | error = -ENOMEM; |
75287a67 | 1377 | new_name = kstrdup_const(new_name, GFP_KERNEL); |
fd7b9f7b | 1378 | if (!new_name) |
798c75a0 | 1379 | goto out; |
3eef34ad TH |
1380 | } else { |
1381 | new_name = NULL; | |
fd7b9f7b TH |
1382 | } |
1383 | ||
1384 | /* | |
1385 | * Move to the appropriate place in the appropriate directories rbtree. | |
1386 | */ | |
c637b8ac | 1387 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1388 | kernfs_get(new_parent); |
3eef34ad TH |
1389 | |
1390 | /* rename_lock protects ->parent and ->name accessors */ | |
1391 | spin_lock_irq(&kernfs_rename_lock); | |
1392 | ||
1393 | old_parent = kn->parent; | |
adc5e8b5 | 1394 | kn->parent = new_parent; |
3eef34ad TH |
1395 | |
1396 | kn->ns = new_ns; | |
1397 | if (new_name) { | |
dfeb0750 | 1398 | old_name = kn->name; |
3eef34ad TH |
1399 | kn->name = new_name; |
1400 | } | |
1401 | ||
1402 | spin_unlock_irq(&kernfs_rename_lock); | |
1403 | ||
9561a896 | 1404 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
c637b8ac | 1405 | kernfs_link_sibling(kn); |
fd7b9f7b | 1406 | |
3eef34ad | 1407 | kernfs_put(old_parent); |
75287a67 | 1408 | kfree_const(old_name); |
3eef34ad | 1409 | |
fd7b9f7b | 1410 | error = 0; |
798c75a0 | 1411 | out: |
a797bfc3 | 1412 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1413 | return error; |
1414 | } | |
1415 | ||
fd7b9f7b | 1416 | /* Relationship between s_mode and the DT_xxx types */ |
324a56e1 | 1417 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1418 | { |
adc5e8b5 | 1419 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1420 | } |
1421 | ||
c637b8ac | 1422 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1423 | { |
1424 | kernfs_put(filp->private_data); | |
1425 | return 0; | |
1426 | } | |
1427 | ||
c637b8ac | 1428 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1429 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1430 | { |
1431 | if (pos) { | |
81c173cb | 1432 | int valid = kernfs_active(pos) && |
798c75a0 | 1433 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1434 | kernfs_put(pos); |
1435 | if (!valid) | |
1436 | pos = NULL; | |
1437 | } | |
1438 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1439 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1440 | while (node) { |
324a56e1 | 1441 | pos = rb_to_kn(node); |
fd7b9f7b | 1442 | |
adc5e8b5 | 1443 | if (hash < pos->hash) |
fd7b9f7b | 1444 | node = node->rb_left; |
adc5e8b5 | 1445 | else if (hash > pos->hash) |
fd7b9f7b TH |
1446 | node = node->rb_right; |
1447 | else | |
1448 | break; | |
1449 | } | |
1450 | } | |
b9c9dad0 TH |
1451 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1452 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1453 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1454 | if (!node) |
1455 | pos = NULL; | |
1456 | else | |
324a56e1 | 1457 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1458 | } |
1459 | return pos; | |
1460 | } | |
1461 | ||
c637b8ac | 1462 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1463 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1464 | { |
c637b8ac | 1465 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1466 | if (pos) { |
fd7b9f7b | 1467 | do { |
adc5e8b5 | 1468 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1469 | if (!node) |
1470 | pos = NULL; | |
1471 | else | |
324a56e1 | 1472 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1473 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1474 | } | |
fd7b9f7b TH |
1475 | return pos; |
1476 | } | |
1477 | ||
c637b8ac | 1478 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1479 | { |
1480 | struct dentry *dentry = file->f_path.dentry; | |
324a56e1 TH |
1481 | struct kernfs_node *parent = dentry->d_fsdata; |
1482 | struct kernfs_node *pos = file->private_data; | |
fd7b9f7b TH |
1483 | const void *ns = NULL; |
1484 | ||
1485 | if (!dir_emit_dots(file, ctx)) | |
1486 | return 0; | |
a797bfc3 | 1487 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1488 | |
324a56e1 | 1489 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1490 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1491 | |
c637b8ac | 1492 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1493 | pos; |
c637b8ac | 1494 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1495 | const char *name = pos->name; |
fd7b9f7b TH |
1496 | unsigned int type = dt_type(pos); |
1497 | int len = strlen(name); | |
adc5e8b5 | 1498 | ino_t ino = pos->ino; |
fd7b9f7b | 1499 | |
adc5e8b5 | 1500 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1501 | file->private_data = pos; |
1502 | kernfs_get(pos); | |
1503 | ||
a797bfc3 | 1504 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1505 | if (!dir_emit(ctx, name, len, ino, type)) |
1506 | return 0; | |
a797bfc3 | 1507 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1508 | } |
a797bfc3 | 1509 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1510 | file->private_data = NULL; |
1511 | ctx->pos = INT_MAX; | |
1512 | return 0; | |
1513 | } | |
1514 | ||
c637b8ac TH |
1515 | static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset, |
1516 | int whence) | |
fd7b9f7b TH |
1517 | { |
1518 | struct inode *inode = file_inode(file); | |
1519 | loff_t ret; | |
1520 | ||
1521 | mutex_lock(&inode->i_mutex); | |
1522 | ret = generic_file_llseek(file, offset, whence); | |
1523 | mutex_unlock(&inode->i_mutex); | |
1524 | ||
1525 | return ret; | |
1526 | } | |
1527 | ||
a797bfc3 | 1528 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1529 | .read = generic_read_dir, |
c637b8ac TH |
1530 | .iterate = kernfs_fop_readdir, |
1531 | .release = kernfs_dir_fop_release, | |
1532 | .llseek = kernfs_dir_fop_llseek, | |
fd7b9f7b | 1533 | }; |