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 | ||
9f6df573 AK |
47 | /* kernfs_node_depth - compute depth from @from to @to */ |
48 | static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to) | |
3eef34ad | 49 | { |
9f6df573 | 50 | size_t depth = 0; |
3eef34ad | 51 | |
9f6df573 AK |
52 | while (to->parent && to != from) { |
53 | depth++; | |
54 | to = to->parent; | |
55 | } | |
56 | return depth; | |
57 | } | |
3eef34ad | 58 | |
9f6df573 AK |
59 | static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a, |
60 | struct kernfs_node *b) | |
61 | { | |
62 | size_t da, db; | |
63 | struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b); | |
64 | ||
65 | if (ra != rb) | |
66 | return NULL; | |
67 | ||
68 | da = kernfs_depth(ra->kn, a); | |
69 | db = kernfs_depth(rb->kn, b); | |
70 | ||
71 | while (da > db) { | |
72 | a = a->parent; | |
73 | da--; | |
74 | } | |
75 | while (db > da) { | |
76 | b = b->parent; | |
77 | db--; | |
78 | } | |
79 | ||
80 | /* worst case b and a will be the same at root */ | |
81 | while (b != a) { | |
82 | b = b->parent; | |
83 | a = a->parent; | |
84 | } | |
85 | ||
86 | return a; | |
87 | } | |
88 | ||
89 | /** | |
90 | * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to, | |
91 | * where kn_from is treated as root of the path. | |
92 | * @kn_from: kernfs node which should be treated as root for the path | |
93 | * @kn_to: kernfs node to which path is needed | |
94 | * @buf: buffer to copy the path into | |
95 | * @buflen: size of @buf | |
96 | * | |
97 | * We need to handle couple of scenarios here: | |
98 | * [1] when @kn_from is an ancestor of @kn_to at some level | |
99 | * kn_from: /n1/n2/n3 | |
100 | * kn_to: /n1/n2/n3/n4/n5 | |
101 | * result: /n4/n5 | |
102 | * | |
103 | * [2] when @kn_from is on a different hierarchy and we need to find common | |
104 | * ancestor between @kn_from and @kn_to. | |
105 | * kn_from: /n1/n2/n3/n4 | |
106 | * kn_to: /n1/n2/n5 | |
107 | * result: /../../n5 | |
108 | * OR | |
109 | * kn_from: /n1/n2/n3/n4/n5 [depth=5] | |
110 | * kn_to: /n1/n2/n3 [depth=3] | |
111 | * result: /../.. | |
112 | * | |
113 | * return value: length of the string. If greater than buflen, | |
114 | * then contents of buf are undefined. On error, -1 is returned. | |
115 | */ | |
116 | static int kernfs_path_from_node_locked(struct kernfs_node *kn_to, | |
117 | struct kernfs_node *kn_from, | |
118 | char *buf, size_t buflen) | |
119 | { | |
120 | struct kernfs_node *kn, *common; | |
121 | const char parent_str[] = "/.."; | |
122 | size_t depth_from, depth_to, len = 0, nlen = 0; | |
123 | char *p; | |
124 | int i; | |
125 | ||
126 | if (!kn_from) | |
127 | kn_from = kernfs_root(kn_to)->kn; | |
128 | ||
129 | if (kn_from == kn_to) | |
130 | return strlcpy(buf, "/", buflen); | |
131 | ||
132 | common = kernfs_common_ancestor(kn_from, kn_to); | |
133 | if (WARN_ON(!common)) | |
134 | return -1; | |
135 | ||
136 | depth_to = kernfs_depth(common, kn_to); | |
137 | depth_from = kernfs_depth(common, kn_from); | |
138 | ||
139 | if (buf) | |
140 | buf[0] = '\0'; | |
141 | ||
142 | for (i = 0; i < depth_from; i++) | |
143 | len += strlcpy(buf + len, parent_str, | |
144 | len < buflen ? buflen - len : 0); | |
145 | ||
146 | /* Calculate how many bytes we need for the rest */ | |
147 | for (kn = kn_to; kn != common; kn = kn->parent) | |
148 | nlen += strlen(kn->name) + 1; | |
149 | ||
150 | if (len + nlen >= buflen) | |
151 | return len + nlen; | |
152 | ||
153 | p = buf + len + nlen; | |
154 | *p = '\0'; | |
155 | for (kn = kn_to; kn != common; kn = kn->parent) { | |
e99ed4de SH |
156 | size_t tmp = strlen(kn->name); |
157 | p -= tmp; | |
158 | memcpy(p, kn->name, tmp); | |
9f6df573 AK |
159 | *(--p) = '/'; |
160 | } | |
3eef34ad | 161 | |
9f6df573 | 162 | return len + nlen; |
3eef34ad TH |
163 | } |
164 | ||
165 | /** | |
166 | * kernfs_name - obtain the name of a given node | |
167 | * @kn: kernfs_node of interest | |
168 | * @buf: buffer to copy @kn's name into | |
169 | * @buflen: size of @buf | |
170 | * | |
171 | * Copies the name of @kn into @buf of @buflen bytes. The behavior is | |
172 | * similar to strlcpy(). It returns the length of @kn's name and if @buf | |
173 | * isn't long enough, it's filled upto @buflen-1 and nul terminated. | |
174 | * | |
175 | * This function can be called from any context. | |
176 | */ | |
177 | int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen) | |
178 | { | |
179 | unsigned long flags; | |
180 | int ret; | |
181 | ||
182 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
183 | ret = kernfs_name_locked(kn, buf, buflen); | |
184 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
185 | return ret; | |
186 | } | |
187 | ||
9acee9c5 TH |
188 | /** |
189 | * kernfs_path_len - determine the length of the full path of a given node | |
190 | * @kn: kernfs_node of interest | |
191 | * | |
192 | * The returned length doesn't include the space for the terminating '\0'. | |
193 | */ | |
194 | size_t kernfs_path_len(struct kernfs_node *kn) | |
195 | { | |
196 | size_t len = 0; | |
197 | unsigned long flags; | |
198 | ||
199 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
200 | ||
201 | do { | |
202 | len += strlen(kn->name) + 1; | |
203 | kn = kn->parent; | |
204 | } while (kn && kn->parent); | |
205 | ||
206 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
207 | ||
208 | return len; | |
209 | } | |
210 | ||
9f6df573 AK |
211 | /** |
212 | * kernfs_path_from_node - build path of node @to relative to @from. | |
213 | * @from: parent kernfs_node relative to which we need to build the path | |
214 | * @to: kernfs_node of interest | |
215 | * @buf: buffer to copy @to's path into | |
216 | * @buflen: size of @buf | |
217 | * | |
218 | * Builds @to's path relative to @from in @buf. @from and @to must | |
219 | * be on the same kernfs-root. If @from is not parent of @to, then a relative | |
220 | * path (which includes '..'s) as needed to reach from @from to @to is | |
221 | * returned. | |
222 | * | |
223 | * If @buf isn't long enough, the return value will be greater than @buflen | |
224 | * and @buf contents are undefined. | |
225 | */ | |
226 | int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from, | |
227 | char *buf, size_t buflen) | |
228 | { | |
229 | unsigned long flags; | |
230 | int ret; | |
231 | ||
232 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
233 | ret = kernfs_path_from_node_locked(to, from, buf, buflen); | |
234 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
235 | return ret; | |
236 | } | |
237 | EXPORT_SYMBOL_GPL(kernfs_path_from_node); | |
238 | ||
3eef34ad TH |
239 | /** |
240 | * kernfs_path - build full path of a given node | |
241 | * @kn: kernfs_node of interest | |
242 | * @buf: buffer to copy @kn's name into | |
243 | * @buflen: size of @buf | |
244 | * | |
245 | * Builds and returns the full path of @kn in @buf of @buflen bytes. The | |
246 | * path is built from the end of @buf so the returned pointer usually | |
247 | * doesn't match @buf. If @buf isn't long enough, @buf is nul terminated | |
248 | * and %NULL is returned. | |
249 | */ | |
250 | char *kernfs_path(struct kernfs_node *kn, char *buf, size_t buflen) | |
251 | { | |
9f6df573 | 252 | int ret; |
3eef34ad | 253 | |
9f6df573 AK |
254 | ret = kernfs_path_from_node(kn, NULL, buf, buflen); |
255 | if (ret < 0 || ret >= buflen) | |
256 | return NULL; | |
257 | return buf; | |
3eef34ad | 258 | } |
e61734c5 | 259 | EXPORT_SYMBOL_GPL(kernfs_path); |
3eef34ad TH |
260 | |
261 | /** | |
262 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node | |
263 | * @kn: kernfs_node of interest | |
264 | * | |
265 | * This function can be called from any context. | |
266 | */ | |
267 | void pr_cont_kernfs_name(struct kernfs_node *kn) | |
268 | { | |
269 | unsigned long flags; | |
270 | ||
271 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
272 | ||
273 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); | |
274 | pr_cont("%s", kernfs_pr_cont_buf); | |
275 | ||
276 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
277 | } | |
278 | ||
279 | /** | |
280 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node | |
281 | * @kn: kernfs_node of interest | |
282 | * | |
283 | * This function can be called from any context. | |
284 | */ | |
285 | void pr_cont_kernfs_path(struct kernfs_node *kn) | |
286 | { | |
287 | unsigned long flags; | |
9f6df573 | 288 | int sz; |
3eef34ad TH |
289 | |
290 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
291 | ||
9f6df573 AK |
292 | sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf, |
293 | sizeof(kernfs_pr_cont_buf)); | |
294 | if (sz < 0) { | |
295 | pr_cont("(error)"); | |
296 | goto out; | |
297 | } | |
298 | ||
299 | if (sz >= sizeof(kernfs_pr_cont_buf)) { | |
300 | pr_cont("(name too long)"); | |
301 | goto out; | |
302 | } | |
303 | ||
304 | pr_cont("%s", kernfs_pr_cont_buf); | |
3eef34ad | 305 | |
9f6df573 | 306 | out: |
3eef34ad TH |
307 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
308 | } | |
309 | ||
310 | /** | |
311 | * kernfs_get_parent - determine the parent node and pin it | |
312 | * @kn: kernfs_node of interest | |
313 | * | |
314 | * Determines @kn's parent, pins and returns it. This function can be | |
315 | * called from any context. | |
316 | */ | |
317 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) | |
318 | { | |
319 | struct kernfs_node *parent; | |
320 | unsigned long flags; | |
321 | ||
322 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
323 | parent = kn->parent; | |
324 | kernfs_get(parent); | |
325 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
326 | ||
327 | return parent; | |
328 | } | |
329 | ||
fd7b9f7b | 330 | /** |
c637b8ac | 331 | * kernfs_name_hash |
fd7b9f7b TH |
332 | * @name: Null terminated string to hash |
333 | * @ns: Namespace tag to hash | |
334 | * | |
335 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
336 | */ | |
c637b8ac | 337 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b TH |
338 | { |
339 | unsigned long hash = init_name_hash(); | |
340 | unsigned int len = strlen(name); | |
341 | while (len--) | |
342 | hash = partial_name_hash(*name++, hash); | |
343 | hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); | |
344 | hash &= 0x7fffffffU; | |
345 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
88391d49 | 346 | if (hash < 2) |
fd7b9f7b TH |
347 | hash += 2; |
348 | if (hash >= INT_MAX) | |
349 | hash = INT_MAX - 1; | |
350 | return hash; | |
351 | } | |
352 | ||
c637b8ac TH |
353 | static int kernfs_name_compare(unsigned int hash, const char *name, |
354 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 355 | { |
72392ed0 RV |
356 | if (hash < kn->hash) |
357 | return -1; | |
358 | if (hash > kn->hash) | |
359 | return 1; | |
360 | if (ns < kn->ns) | |
361 | return -1; | |
362 | if (ns > kn->ns) | |
363 | return 1; | |
adc5e8b5 | 364 | return strcmp(name, kn->name); |
fd7b9f7b TH |
365 | } |
366 | ||
c637b8ac TH |
367 | static int kernfs_sd_compare(const struct kernfs_node *left, |
368 | const struct kernfs_node *right) | |
fd7b9f7b | 369 | { |
c637b8ac | 370 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
371 | } |
372 | ||
373 | /** | |
c637b8ac | 374 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 375 | * @kn: kernfs_node of interest |
fd7b9f7b | 376 | * |
324a56e1 | 377 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 378 | * @kn->parent->dir.children. |
fd7b9f7b TH |
379 | * |
380 | * Locking: | |
a797bfc3 | 381 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
382 | * |
383 | * RETURNS: | |
384 | * 0 on susccess -EEXIST on failure. | |
385 | */ | |
c637b8ac | 386 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 387 | { |
adc5e8b5 | 388 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
389 | struct rb_node *parent = NULL; |
390 | ||
fd7b9f7b | 391 | while (*node) { |
324a56e1 | 392 | struct kernfs_node *pos; |
fd7b9f7b TH |
393 | int result; |
394 | ||
324a56e1 | 395 | pos = rb_to_kn(*node); |
fd7b9f7b | 396 | parent = *node; |
c637b8ac | 397 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 398 | if (result < 0) |
adc5e8b5 | 399 | node = &pos->rb.rb_left; |
fd7b9f7b | 400 | else if (result > 0) |
adc5e8b5 | 401 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
402 | else |
403 | return -EEXIST; | |
404 | } | |
c1befb88 | 405 | |
fd7b9f7b | 406 | /* add new node and rebalance the tree */ |
adc5e8b5 TH |
407 | rb_link_node(&kn->rb, parent, node); |
408 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
c1befb88 JZ |
409 | |
410 | /* successfully added, account subdir number */ | |
411 | if (kernfs_type(kn) == KERNFS_DIR) | |
412 | kn->parent->dir.subdirs++; | |
413 | ||
fd7b9f7b TH |
414 | return 0; |
415 | } | |
416 | ||
417 | /** | |
c637b8ac | 418 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 419 | * @kn: kernfs_node of interest |
fd7b9f7b | 420 | * |
35beab06 TH |
421 | * Try to unlink @kn from its sibling rbtree which starts from |
422 | * kn->parent->dir.children. Returns %true if @kn was actually | |
423 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
424 | * |
425 | * Locking: | |
a797bfc3 | 426 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 427 | */ |
35beab06 | 428 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 429 | { |
35beab06 TH |
430 | if (RB_EMPTY_NODE(&kn->rb)) |
431 | return false; | |
432 | ||
df23fc39 | 433 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 434 | kn->parent->dir.subdirs--; |
fd7b9f7b | 435 | |
adc5e8b5 | 436 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
437 | RB_CLEAR_NODE(&kn->rb); |
438 | return true; | |
fd7b9f7b TH |
439 | } |
440 | ||
441 | /** | |
c637b8ac | 442 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 443 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 444 | * |
324a56e1 | 445 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
446 | * is NULL. |
447 | * | |
448 | * RETURNS: | |
324a56e1 | 449 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 450 | */ |
c637b8ac | 451 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 452 | { |
324a56e1 | 453 | if (unlikely(!kn)) |
fd7b9f7b TH |
454 | return NULL; |
455 | ||
f4b3e631 GKH |
456 | if (!atomic_inc_unless_negative(&kn->active)) |
457 | return NULL; | |
895a068a | 458 | |
182fd64b | 459 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
460 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
461 | return kn; | |
fd7b9f7b TH |
462 | } |
463 | ||
464 | /** | |
c637b8ac | 465 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 466 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 467 | * |
324a56e1 | 468 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
469 | * is NULL. |
470 | */ | |
c637b8ac | 471 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b | 472 | { |
abd54f02 | 473 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b TH |
474 | int v; |
475 | ||
324a56e1 | 476 | if (unlikely(!kn)) |
fd7b9f7b TH |
477 | return; |
478 | ||
182fd64b | 479 | if (kernfs_lockdep(kn)) |
324a56e1 | 480 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
adc5e8b5 | 481 | v = atomic_dec_return(&kn->active); |
df23fc39 | 482 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
483 | return; |
484 | ||
abd54f02 | 485 | wake_up_all(&root->deactivate_waitq); |
fd7b9f7b TH |
486 | } |
487 | ||
488 | /** | |
81c173cb TH |
489 | * kernfs_drain - drain kernfs_node |
490 | * @kn: kernfs_node to drain | |
fd7b9f7b | 491 | * |
81c173cb TH |
492 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
493 | * removers may invoke this function concurrently on @kn and all will | |
494 | * return after draining is complete. | |
fd7b9f7b | 495 | */ |
81c173cb | 496 | static void kernfs_drain(struct kernfs_node *kn) |
35beab06 | 497 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
fd7b9f7b | 498 | { |
abd54f02 | 499 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b | 500 | |
35beab06 | 501 | lockdep_assert_held(&kernfs_mutex); |
81c173cb | 502 | WARN_ON_ONCE(kernfs_active(kn)); |
ea1c472d | 503 | |
35beab06 | 504 | mutex_unlock(&kernfs_mutex); |
abd54f02 | 505 | |
182fd64b | 506 | if (kernfs_lockdep(kn)) { |
35beab06 TH |
507 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
508 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) | |
509 | lock_contended(&kn->dep_map, _RET_IP_); | |
510 | } | |
abd54f02 | 511 | |
35beab06 | 512 | /* but everyone should wait for draining */ |
abd54f02 TH |
513 | wait_event(root->deactivate_waitq, |
514 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); | |
fd7b9f7b | 515 | |
182fd64b | 516 | if (kernfs_lockdep(kn)) { |
a6607930 TH |
517 | lock_acquired(&kn->dep_map, _RET_IP_); |
518 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | |
519 | } | |
35beab06 | 520 | |
ccf02aaf TH |
521 | kernfs_unmap_bin_file(kn); |
522 | ||
35beab06 | 523 | mutex_lock(&kernfs_mutex); |
fd7b9f7b TH |
524 | } |
525 | ||
fd7b9f7b | 526 | /** |
324a56e1 TH |
527 | * kernfs_get - get a reference count on a kernfs_node |
528 | * @kn: the target kernfs_node | |
fd7b9f7b | 529 | */ |
324a56e1 | 530 | void kernfs_get(struct kernfs_node *kn) |
fd7b9f7b | 531 | { |
324a56e1 | 532 | if (kn) { |
adc5e8b5 TH |
533 | WARN_ON(!atomic_read(&kn->count)); |
534 | atomic_inc(&kn->count); | |
fd7b9f7b TH |
535 | } |
536 | } | |
537 | EXPORT_SYMBOL_GPL(kernfs_get); | |
538 | ||
539 | /** | |
324a56e1 TH |
540 | * kernfs_put - put a reference count on a kernfs_node |
541 | * @kn: the target kernfs_node | |
fd7b9f7b | 542 | * |
324a56e1 | 543 | * Put a reference count of @kn and destroy it if it reached zero. |
fd7b9f7b | 544 | */ |
324a56e1 | 545 | void kernfs_put(struct kernfs_node *kn) |
fd7b9f7b | 546 | { |
324a56e1 | 547 | struct kernfs_node *parent; |
ba7443bc | 548 | struct kernfs_root *root; |
fd7b9f7b | 549 | |
adc5e8b5 | 550 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 551 | return; |
324a56e1 | 552 | root = kernfs_root(kn); |
fd7b9f7b | 553 | repeat: |
81c173cb TH |
554 | /* |
555 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 556 | * kn->parent won't change beneath us. |
fd7b9f7b | 557 | */ |
adc5e8b5 | 558 | parent = kn->parent; |
fd7b9f7b | 559 | |
81c173cb TH |
560 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
561 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
562 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 563 | |
df23fc39 | 564 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 565 | kernfs_put(kn->symlink.target_kn); |
dfeb0750 TH |
566 | |
567 | kfree_const(kn->name); | |
568 | ||
adc5e8b5 TH |
569 | if (kn->iattr) { |
570 | if (kn->iattr->ia_secdata) | |
571 | security_release_secctx(kn->iattr->ia_secdata, | |
572 | kn->iattr->ia_secdata_len); | |
573 | simple_xattrs_free(&kn->iattr->xattrs); | |
2322392b | 574 | } |
adc5e8b5 TH |
575 | kfree(kn->iattr); |
576 | ida_simple_remove(&root->ino_ida, kn->ino); | |
a797bfc3 | 577 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 578 | |
324a56e1 TH |
579 | kn = parent; |
580 | if (kn) { | |
adc5e8b5 | 581 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
582 | goto repeat; |
583 | } else { | |
324a56e1 | 584 | /* just released the root kn, free @root too */ |
bc755553 | 585 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
586 | kfree(root); |
587 | } | |
fd7b9f7b TH |
588 | } |
589 | EXPORT_SYMBOL_GPL(kernfs_put); | |
590 | ||
c637b8ac | 591 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 592 | { |
324a56e1 | 593 | struct kernfs_node *kn; |
fd7b9f7b TH |
594 | |
595 | if (flags & LOOKUP_RCU) | |
596 | return -ECHILD; | |
597 | ||
19bbb926 | 598 | /* Always perform fresh lookup for negatives */ |
2b0143b5 | 599 | if (d_really_is_negative(dentry)) |
19bbb926 TH |
600 | goto out_bad_unlocked; |
601 | ||
324a56e1 | 602 | kn = dentry->d_fsdata; |
a797bfc3 | 603 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 604 | |
81c173cb TH |
605 | /* The kernfs node has been deactivated */ |
606 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
607 | goto out_bad; |
608 | ||
c637b8ac | 609 | /* The kernfs node has been moved? */ |
adc5e8b5 | 610 | if (dentry->d_parent->d_fsdata != kn->parent) |
fd7b9f7b TH |
611 | goto out_bad; |
612 | ||
c637b8ac | 613 | /* The kernfs node has been renamed */ |
adc5e8b5 | 614 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
615 | goto out_bad; |
616 | ||
c637b8ac | 617 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 618 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 619 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
620 | goto out_bad; |
621 | ||
a797bfc3 | 622 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
623 | return 1; |
624 | out_bad: | |
a797bfc3 | 625 | mutex_unlock(&kernfs_mutex); |
19bbb926 | 626 | out_bad_unlocked: |
fd7b9f7b TH |
627 | return 0; |
628 | } | |
629 | ||
c637b8ac | 630 | static void kernfs_dop_release(struct dentry *dentry) |
fd7b9f7b TH |
631 | { |
632 | kernfs_put(dentry->d_fsdata); | |
633 | } | |
634 | ||
a797bfc3 | 635 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 636 | .d_revalidate = kernfs_dop_revalidate, |
c637b8ac | 637 | .d_release = kernfs_dop_release, |
fd7b9f7b TH |
638 | }; |
639 | ||
0c23b225 TH |
640 | /** |
641 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry | |
642 | * @dentry: the dentry in question | |
643 | * | |
644 | * Return the kernfs_node associated with @dentry. If @dentry is not a | |
645 | * kernfs one, %NULL is returned. | |
646 | * | |
647 | * While the returned kernfs_node will stay accessible as long as @dentry | |
648 | * is accessible, the returned node can be in any state and the caller is | |
649 | * fully responsible for determining what's accessible. | |
650 | */ | |
651 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) | |
652 | { | |
f41c5934 | 653 | if (dentry->d_sb->s_op == &kernfs_sops) |
0c23b225 TH |
654 | return dentry->d_fsdata; |
655 | return NULL; | |
656 | } | |
657 | ||
db4aad20 TH |
658 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
659 | const char *name, umode_t mode, | |
660 | unsigned flags) | |
fd7b9f7b | 661 | { |
324a56e1 | 662 | struct kernfs_node *kn; |
bc755553 | 663 | int ret; |
fd7b9f7b | 664 | |
dfeb0750 TH |
665 | name = kstrdup_const(name, GFP_KERNEL); |
666 | if (!name) | |
667 | return NULL; | |
fd7b9f7b | 668 | |
a797bfc3 | 669 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 670 | if (!kn) |
fd7b9f7b TH |
671 | goto err_out1; |
672 | ||
b2a209ff | 673 | ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); |
bc755553 | 674 | if (ret < 0) |
fd7b9f7b | 675 | goto err_out2; |
adc5e8b5 | 676 | kn->ino = ret; |
fd7b9f7b | 677 | |
adc5e8b5 | 678 | atomic_set(&kn->count, 1); |
81c173cb | 679 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 680 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 681 | |
adc5e8b5 TH |
682 | kn->name = name; |
683 | kn->mode = mode; | |
81c173cb | 684 | kn->flags = flags; |
fd7b9f7b | 685 | |
324a56e1 | 686 | return kn; |
fd7b9f7b TH |
687 | |
688 | err_out2: | |
a797bfc3 | 689 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 690 | err_out1: |
dfeb0750 | 691 | kfree_const(name); |
fd7b9f7b TH |
692 | return NULL; |
693 | } | |
694 | ||
db4aad20 TH |
695 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
696 | const char *name, umode_t mode, | |
697 | unsigned flags) | |
698 | { | |
699 | struct kernfs_node *kn; | |
700 | ||
701 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); | |
702 | if (kn) { | |
703 | kernfs_get(parent); | |
704 | kn->parent = parent; | |
705 | } | |
706 | return kn; | |
707 | } | |
708 | ||
fd7b9f7b | 709 | /** |
c637b8ac | 710 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 711 | * @kn: kernfs_node to be added |
fd7b9f7b | 712 | * |
db4aad20 TH |
713 | * The caller must already have initialized @kn->parent. This |
714 | * function increments nlink of the parent's inode if @kn is a | |
715 | * directory and link into the children list of the parent. | |
fd7b9f7b | 716 | * |
fd7b9f7b TH |
717 | * RETURNS: |
718 | * 0 on success, -EEXIST if entry with the given name already | |
719 | * exists. | |
720 | */ | |
988cd7af | 721 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 722 | { |
db4aad20 | 723 | struct kernfs_node *parent = kn->parent; |
c525aadd | 724 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 725 | bool has_ns; |
fd7b9f7b TH |
726 | int ret; |
727 | ||
988cd7af TH |
728 | mutex_lock(&kernfs_mutex); |
729 | ||
730 | ret = -EINVAL; | |
731 | has_ns = kernfs_ns_enabled(parent); | |
732 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
733 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
734 | goto out_unlock; | |
fd7b9f7b | 735 | |
df23fc39 | 736 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 737 | goto out_unlock; |
fd7b9f7b | 738 | |
988cd7af | 739 | ret = -ENOENT; |
ea015218 EB |
740 | if (parent->flags & KERNFS_EMPTY_DIR) |
741 | goto out_unlock; | |
742 | ||
d35258ef | 743 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 744 | goto out_unlock; |
798c75a0 | 745 | |
c637b8ac | 746 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 747 | |
c637b8ac | 748 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 749 | if (ret) |
988cd7af | 750 | goto out_unlock; |
fd7b9f7b TH |
751 | |
752 | /* Update timestamps on the parent */ | |
adc5e8b5 | 753 | ps_iattr = parent->iattr; |
fd7b9f7b TH |
754 | if (ps_iattr) { |
755 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
3a3a5fec DD |
756 | ktime_get_real_ts(&ps_iattrs->ia_ctime); |
757 | ps_iattrs->ia_mtime = ps_iattrs->ia_ctime; | |
fd7b9f7b TH |
758 | } |
759 | ||
d35258ef TH |
760 | mutex_unlock(&kernfs_mutex); |
761 | ||
762 | /* | |
763 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
764 | * If not activated here, the kernfs user is responsible for | |
765 | * activating the node with kernfs_activate(). A node which hasn't | |
766 | * been activated is not visible to userland and its removal won't | |
767 | * trigger deactivation. | |
768 | */ | |
769 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
770 | kernfs_activate(kn); | |
771 | return 0; | |
772 | ||
988cd7af | 773 | out_unlock: |
a797bfc3 | 774 | mutex_unlock(&kernfs_mutex); |
988cd7af | 775 | return ret; |
fd7b9f7b TH |
776 | } |
777 | ||
778 | /** | |
324a56e1 TH |
779 | * kernfs_find_ns - find kernfs_node with the given name |
780 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
781 | * @name: name to look for |
782 | * @ns: the namespace tag to use | |
783 | * | |
324a56e1 TH |
784 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
785 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 786 | */ |
324a56e1 TH |
787 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
788 | const unsigned char *name, | |
789 | const void *ns) | |
fd7b9f7b | 790 | { |
adc5e8b5 | 791 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 792 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
793 | unsigned int hash; |
794 | ||
a797bfc3 | 795 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
796 | |
797 | if (has_ns != (bool)ns) { | |
c637b8ac | 798 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 799 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
800 | return NULL; |
801 | } | |
802 | ||
c637b8ac | 803 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 804 | while (node) { |
324a56e1 | 805 | struct kernfs_node *kn; |
fd7b9f7b TH |
806 | int result; |
807 | ||
324a56e1 | 808 | kn = rb_to_kn(node); |
c637b8ac | 809 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
810 | if (result < 0) |
811 | node = node->rb_left; | |
812 | else if (result > 0) | |
813 | node = node->rb_right; | |
814 | else | |
324a56e1 | 815 | return kn; |
fd7b9f7b TH |
816 | } |
817 | return NULL; | |
818 | } | |
819 | ||
bd96f76a TH |
820 | static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent, |
821 | const unsigned char *path, | |
822 | const void *ns) | |
823 | { | |
e56ed358 TH |
824 | size_t len; |
825 | char *p, *name; | |
bd96f76a TH |
826 | |
827 | lockdep_assert_held(&kernfs_mutex); | |
828 | ||
e56ed358 TH |
829 | /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */ |
830 | spin_lock_irq(&kernfs_rename_lock); | |
831 | ||
832 | len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf)); | |
833 | ||
834 | if (len >= sizeof(kernfs_pr_cont_buf)) { | |
835 | spin_unlock_irq(&kernfs_rename_lock); | |
bd96f76a | 836 | return NULL; |
e56ed358 TH |
837 | } |
838 | ||
839 | p = kernfs_pr_cont_buf; | |
bd96f76a TH |
840 | |
841 | while ((name = strsep(&p, "/")) && parent) { | |
842 | if (*name == '\0') | |
843 | continue; | |
844 | parent = kernfs_find_ns(parent, name, ns); | |
845 | } | |
846 | ||
e56ed358 TH |
847 | spin_unlock_irq(&kernfs_rename_lock); |
848 | ||
bd96f76a TH |
849 | return parent; |
850 | } | |
851 | ||
fd7b9f7b | 852 | /** |
324a56e1 TH |
853 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
854 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
855 | * @name: name to look for |
856 | * @ns: the namespace tag to use | |
857 | * | |
324a56e1 | 858 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 859 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 860 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 861 | */ |
324a56e1 TH |
862 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
863 | const char *name, const void *ns) | |
fd7b9f7b | 864 | { |
324a56e1 | 865 | struct kernfs_node *kn; |
fd7b9f7b | 866 | |
a797bfc3 | 867 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
868 | kn = kernfs_find_ns(parent, name, ns); |
869 | kernfs_get(kn); | |
a797bfc3 | 870 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 871 | |
324a56e1 | 872 | return kn; |
fd7b9f7b TH |
873 | } |
874 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
875 | ||
bd96f76a TH |
876 | /** |
877 | * kernfs_walk_and_get_ns - find and get kernfs_node with the given path | |
878 | * @parent: kernfs_node to search under | |
879 | * @path: path to look for | |
880 | * @ns: the namespace tag to use | |
881 | * | |
882 | * Look for kernfs_node with path @path under @parent and get a reference | |
883 | * if found. This function may sleep and returns pointer to the found | |
884 | * kernfs_node on success, %NULL on failure. | |
885 | */ | |
886 | struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent, | |
887 | const char *path, const void *ns) | |
888 | { | |
889 | struct kernfs_node *kn; | |
890 | ||
891 | mutex_lock(&kernfs_mutex); | |
892 | kn = kernfs_walk_ns(parent, path, ns); | |
893 | kernfs_get(kn); | |
894 | mutex_unlock(&kernfs_mutex); | |
895 | ||
896 | return kn; | |
897 | } | |
898 | ||
ba7443bc TH |
899 | /** |
900 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 901 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 902 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
903 | * @priv: opaque data associated with the new directory |
904 | * | |
905 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
906 | * failure. | |
907 | */ | |
90c07c89 | 908 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 909 | unsigned int flags, void *priv) |
ba7443bc TH |
910 | { |
911 | struct kernfs_root *root; | |
324a56e1 | 912 | struct kernfs_node *kn; |
ba7443bc TH |
913 | |
914 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
915 | if (!root) | |
916 | return ERR_PTR(-ENOMEM); | |
917 | ||
bc755553 | 918 | ida_init(&root->ino_ida); |
7d568a83 | 919 | INIT_LIST_HEAD(&root->supers); |
bc755553 | 920 | |
db4aad20 TH |
921 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
922 | KERNFS_DIR); | |
324a56e1 | 923 | if (!kn) { |
bc755553 | 924 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
925 | kfree(root); |
926 | return ERR_PTR(-ENOMEM); | |
927 | } | |
928 | ||
324a56e1 | 929 | kn->priv = priv; |
adc5e8b5 | 930 | kn->dir.root = root; |
ba7443bc | 931 | |
90c07c89 | 932 | root->syscall_ops = scops; |
d35258ef | 933 | root->flags = flags; |
324a56e1 | 934 | root->kn = kn; |
abd54f02 | 935 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 936 | |
d35258ef TH |
937 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
938 | kernfs_activate(kn); | |
939 | ||
ba7443bc TH |
940 | return root; |
941 | } | |
942 | ||
943 | /** | |
944 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
945 | * @root: root of the hierarchy to destroy | |
946 | * | |
947 | * Destroy the hierarchy anchored at @root by removing all existing | |
948 | * directories and destroying @root. | |
949 | */ | |
950 | void kernfs_destroy_root(struct kernfs_root *root) | |
951 | { | |
324a56e1 | 952 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
953 | } |
954 | ||
fd7b9f7b TH |
955 | /** |
956 | * kernfs_create_dir_ns - create a directory | |
957 | * @parent: parent in which to create a new directory | |
958 | * @name: name of the new directory | |
bb8b9d09 | 959 | * @mode: mode of the new directory |
fd7b9f7b TH |
960 | * @priv: opaque data associated with the new directory |
961 | * @ns: optional namespace tag of the directory | |
962 | * | |
963 | * Returns the created node on success, ERR_PTR() value on failure. | |
964 | */ | |
324a56e1 | 965 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 TH |
966 | const char *name, umode_t mode, |
967 | void *priv, const void *ns) | |
fd7b9f7b | 968 | { |
324a56e1 | 969 | struct kernfs_node *kn; |
fd7b9f7b TH |
970 | int rc; |
971 | ||
972 | /* allocate */ | |
db4aad20 | 973 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); |
324a56e1 | 974 | if (!kn) |
fd7b9f7b TH |
975 | return ERR_PTR(-ENOMEM); |
976 | ||
adc5e8b5 TH |
977 | kn->dir.root = parent->dir.root; |
978 | kn->ns = ns; | |
324a56e1 | 979 | kn->priv = priv; |
fd7b9f7b TH |
980 | |
981 | /* link in */ | |
988cd7af | 982 | rc = kernfs_add_one(kn); |
fd7b9f7b | 983 | if (!rc) |
324a56e1 | 984 | return kn; |
fd7b9f7b | 985 | |
324a56e1 | 986 | kernfs_put(kn); |
fd7b9f7b TH |
987 | return ERR_PTR(rc); |
988 | } | |
989 | ||
ea015218 EB |
990 | /** |
991 | * kernfs_create_empty_dir - create an always empty directory | |
992 | * @parent: parent in which to create a new directory | |
993 | * @name: name of the new directory | |
994 | * | |
995 | * Returns the created node on success, ERR_PTR() value on failure. | |
996 | */ | |
997 | struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent, | |
998 | const char *name) | |
999 | { | |
1000 | struct kernfs_node *kn; | |
1001 | int rc; | |
1002 | ||
1003 | /* allocate */ | |
1004 | kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR); | |
1005 | if (!kn) | |
1006 | return ERR_PTR(-ENOMEM); | |
1007 | ||
1008 | kn->flags |= KERNFS_EMPTY_DIR; | |
1009 | kn->dir.root = parent->dir.root; | |
1010 | kn->ns = NULL; | |
1011 | kn->priv = NULL; | |
1012 | ||
1013 | /* link in */ | |
1014 | rc = kernfs_add_one(kn); | |
1015 | if (!rc) | |
1016 | return kn; | |
1017 | ||
1018 | kernfs_put(kn); | |
1019 | return ERR_PTR(rc); | |
1020 | } | |
1021 | ||
c637b8ac TH |
1022 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
1023 | struct dentry *dentry, | |
1024 | unsigned int flags) | |
fd7b9f7b | 1025 | { |
19bbb926 | 1026 | struct dentry *ret; |
324a56e1 TH |
1027 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; |
1028 | struct kernfs_node *kn; | |
fd7b9f7b TH |
1029 | struct inode *inode; |
1030 | const void *ns = NULL; | |
1031 | ||
a797bfc3 | 1032 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1033 | |
324a56e1 | 1034 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1035 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 1036 | |
324a56e1 | 1037 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
1038 | |
1039 | /* no such entry */ | |
b9c9dad0 | 1040 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 1041 | ret = NULL; |
fd7b9f7b TH |
1042 | goto out_unlock; |
1043 | } | |
324a56e1 TH |
1044 | kernfs_get(kn); |
1045 | dentry->d_fsdata = kn; | |
fd7b9f7b TH |
1046 | |
1047 | /* attach dentry and inode */ | |
c637b8ac | 1048 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
1049 | if (!inode) { |
1050 | ret = ERR_PTR(-ENOMEM); | |
1051 | goto out_unlock; | |
1052 | } | |
1053 | ||
1054 | /* instantiate and hash dentry */ | |
41d28bca | 1055 | ret = d_splice_alias(inode, dentry); |
fd7b9f7b | 1056 | out_unlock: |
a797bfc3 | 1057 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1058 | return ret; |
1059 | } | |
1060 | ||
80b9bbef TH |
1061 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
1062 | umode_t mode) | |
1063 | { | |
1064 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 1065 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 1066 | int ret; |
80b9bbef | 1067 | |
90c07c89 | 1068 | if (!scops || !scops->mkdir) |
80b9bbef TH |
1069 | return -EPERM; |
1070 | ||
07c7530d TH |
1071 | if (!kernfs_get_active(parent)) |
1072 | return -ENODEV; | |
1073 | ||
90c07c89 | 1074 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
1075 | |
1076 | kernfs_put_active(parent); | |
1077 | return ret; | |
80b9bbef TH |
1078 | } |
1079 | ||
1080 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
1081 | { | |
1082 | struct kernfs_node *kn = dentry->d_fsdata; | |
90c07c89 | 1083 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1084 | int ret; |
80b9bbef | 1085 | |
90c07c89 | 1086 | if (!scops || !scops->rmdir) |
80b9bbef TH |
1087 | return -EPERM; |
1088 | ||
07c7530d TH |
1089 | if (!kernfs_get_active(kn)) |
1090 | return -ENODEV; | |
1091 | ||
90c07c89 | 1092 | ret = scops->rmdir(kn); |
07c7530d TH |
1093 | |
1094 | kernfs_put_active(kn); | |
1095 | return ret; | |
80b9bbef TH |
1096 | } |
1097 | ||
1098 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1099 | struct inode *new_dir, struct dentry *new_dentry) | |
1100 | { | |
1101 | struct kernfs_node *kn = old_dentry->d_fsdata; | |
1102 | struct kernfs_node *new_parent = new_dir->i_private; | |
90c07c89 | 1103 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1104 | int ret; |
80b9bbef | 1105 | |
90c07c89 | 1106 | if (!scops || !scops->rename) |
80b9bbef TH |
1107 | return -EPERM; |
1108 | ||
07c7530d TH |
1109 | if (!kernfs_get_active(kn)) |
1110 | return -ENODEV; | |
1111 | ||
1112 | if (!kernfs_get_active(new_parent)) { | |
1113 | kernfs_put_active(kn); | |
1114 | return -ENODEV; | |
1115 | } | |
1116 | ||
90c07c89 | 1117 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
1118 | |
1119 | kernfs_put_active(new_parent); | |
1120 | kernfs_put_active(kn); | |
1121 | return ret; | |
80b9bbef TH |
1122 | } |
1123 | ||
a797bfc3 | 1124 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
1125 | .lookup = kernfs_iop_lookup, |
1126 | .permission = kernfs_iop_permission, | |
1127 | .setattr = kernfs_iop_setattr, | |
1128 | .getattr = kernfs_iop_getattr, | |
1129 | .setxattr = kernfs_iop_setxattr, | |
1130 | .removexattr = kernfs_iop_removexattr, | |
1131 | .getxattr = kernfs_iop_getxattr, | |
1132 | .listxattr = kernfs_iop_listxattr, | |
80b9bbef TH |
1133 | |
1134 | .mkdir = kernfs_iop_mkdir, | |
1135 | .rmdir = kernfs_iop_rmdir, | |
1136 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
1137 | }; |
1138 | ||
c637b8ac | 1139 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 1140 | { |
324a56e1 | 1141 | struct kernfs_node *last; |
fd7b9f7b TH |
1142 | |
1143 | while (true) { | |
1144 | struct rb_node *rbn; | |
1145 | ||
1146 | last = pos; | |
1147 | ||
df23fc39 | 1148 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
1149 | break; |
1150 | ||
adc5e8b5 | 1151 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
1152 | if (!rbn) |
1153 | break; | |
1154 | ||
324a56e1 | 1155 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
1156 | } |
1157 | ||
1158 | return last; | |
1159 | } | |
1160 | ||
1161 | /** | |
c637b8ac | 1162 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 1163 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 1164 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
1165 | * |
1166 | * Find the next descendant to visit for post-order traversal of @root's | |
1167 | * descendants. @root is included in the iteration and the last node to be | |
1168 | * visited. | |
1169 | */ | |
c637b8ac TH |
1170 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
1171 | struct kernfs_node *root) | |
fd7b9f7b TH |
1172 | { |
1173 | struct rb_node *rbn; | |
1174 | ||
a797bfc3 | 1175 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
1176 | |
1177 | /* if first iteration, visit leftmost descendant which may be root */ | |
1178 | if (!pos) | |
c637b8ac | 1179 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
1180 | |
1181 | /* if we visited @root, we're done */ | |
1182 | if (pos == root) | |
1183 | return NULL; | |
1184 | ||
1185 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 1186 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 1187 | if (rbn) |
c637b8ac | 1188 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
1189 | |
1190 | /* no sibling left, visit parent */ | |
adc5e8b5 | 1191 | return pos->parent; |
fd7b9f7b TH |
1192 | } |
1193 | ||
d35258ef TH |
1194 | /** |
1195 | * kernfs_activate - activate a node which started deactivated | |
1196 | * @kn: kernfs_node whose subtree is to be activated | |
1197 | * | |
1198 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
1199 | * needs to be explicitly activated. A node which hasn't been activated | |
1200 | * isn't visible to userland and deactivation is skipped during its | |
1201 | * removal. This is useful to construct atomic init sequences where | |
1202 | * creation of multiple nodes should either succeed or fail atomically. | |
1203 | * | |
1204 | * The caller is responsible for ensuring that this function is not called | |
1205 | * after kernfs_remove*() is invoked on @kn. | |
1206 | */ | |
1207 | void kernfs_activate(struct kernfs_node *kn) | |
1208 | { | |
1209 | struct kernfs_node *pos; | |
1210 | ||
1211 | mutex_lock(&kernfs_mutex); | |
1212 | ||
1213 | pos = NULL; | |
1214 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
1215 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) | |
1216 | continue; | |
1217 | ||
1218 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
1219 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
1220 | ||
1221 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
1222 | pos->flags |= KERNFS_ACTIVATED; | |
1223 | } | |
1224 | ||
1225 | mutex_unlock(&kernfs_mutex); | |
1226 | } | |
1227 | ||
988cd7af | 1228 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1229 | { |
35beab06 TH |
1230 | struct kernfs_node *pos; |
1231 | ||
1232 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 1233 | |
6b0afc2a TH |
1234 | /* |
1235 | * Short-circuit if non-root @kn has already finished removal. | |
1236 | * This is for kernfs_remove_self() which plays with active ref | |
1237 | * after removal. | |
1238 | */ | |
1239 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
1240 | return; |
1241 | ||
c637b8ac | 1242 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 1243 | |
81c173cb | 1244 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
1245 | pos = NULL; |
1246 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
1247 | if (kernfs_active(pos)) |
1248 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
1249 | |
1250 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 1251 | do { |
35beab06 TH |
1252 | pos = kernfs_leftmost_descendant(kn); |
1253 | ||
1254 | /* | |
81c173cb TH |
1255 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
1256 | * base ref could have been put by someone else by the time | |
1257 | * the function returns. Make sure it doesn't go away | |
1258 | * underneath us. | |
35beab06 TH |
1259 | */ |
1260 | kernfs_get(pos); | |
1261 | ||
d35258ef TH |
1262 | /* |
1263 | * Drain iff @kn was activated. This avoids draining and | |
1264 | * its lockdep annotations for nodes which have never been | |
1265 | * activated and allows embedding kernfs_remove() in create | |
1266 | * error paths without worrying about draining. | |
1267 | */ | |
1268 | if (kn->flags & KERNFS_ACTIVATED) | |
1269 | kernfs_drain(pos); | |
1270 | else | |
1271 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
1272 | |
1273 | /* | |
1274 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
1275 | * to decide who's responsible for cleanups. | |
1276 | */ | |
1277 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
1278 | struct kernfs_iattrs *ps_iattr = | |
1279 | pos->parent ? pos->parent->iattr : NULL; | |
1280 | ||
1281 | /* update timestamps on the parent */ | |
1282 | if (ps_iattr) { | |
3a3a5fec DD |
1283 | ktime_get_real_ts(&ps_iattr->ia_iattr.ia_ctime); |
1284 | ps_iattr->ia_iattr.ia_mtime = | |
1285 | ps_iattr->ia_iattr.ia_ctime; | |
35beab06 TH |
1286 | } |
1287 | ||
988cd7af | 1288 | kernfs_put(pos); |
35beab06 TH |
1289 | } |
1290 | ||
1291 | kernfs_put(pos); | |
1292 | } while (pos != kn); | |
fd7b9f7b TH |
1293 | } |
1294 | ||
1295 | /** | |
324a56e1 TH |
1296 | * kernfs_remove - remove a kernfs_node recursively |
1297 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 1298 | * |
324a56e1 | 1299 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 1300 | */ |
324a56e1 | 1301 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1302 | { |
988cd7af TH |
1303 | mutex_lock(&kernfs_mutex); |
1304 | __kernfs_remove(kn); | |
1305 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
1306 | } |
1307 | ||
6b0afc2a TH |
1308 | /** |
1309 | * kernfs_break_active_protection - break out of active protection | |
1310 | * @kn: the self kernfs_node | |
1311 | * | |
1312 | * The caller must be running off of a kernfs operation which is invoked | |
1313 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
1314 | * this function must also be matched with an invocation of | |
1315 | * kernfs_unbreak_active_protection(). | |
1316 | * | |
1317 | * This function releases the active reference of @kn the caller is | |
1318 | * holding. Once this function is called, @kn may be removed at any point | |
1319 | * and the caller is solely responsible for ensuring that the objects it | |
1320 | * dereferences are accessible. | |
1321 | */ | |
1322 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
1323 | { | |
1324 | /* | |
1325 | * Take out ourself out of the active ref dependency chain. If | |
1326 | * we're called without an active ref, lockdep will complain. | |
1327 | */ | |
1328 | kernfs_put_active(kn); | |
1329 | } | |
1330 | ||
1331 | /** | |
1332 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
1333 | * @kn: the self kernfs_node | |
1334 | * | |
1335 | * If kernfs_break_active_protection() was called, this function must be | |
1336 | * invoked before finishing the kernfs operation. Note that while this | |
1337 | * function restores the active reference, it doesn't and can't actually | |
1338 | * restore the active protection - @kn may already or be in the process of | |
1339 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
1340 | * protection is irreversibly gone for the kernfs operation instance. | |
1341 | * | |
1342 | * While this function may be called at any point after | |
1343 | * kernfs_break_active_protection() is invoked, its most useful location | |
1344 | * would be right before the enclosing kernfs operation returns. | |
1345 | */ | |
1346 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
1347 | { | |
1348 | /* | |
1349 | * @kn->active could be in any state; however, the increment we do | |
1350 | * here will be undone as soon as the enclosing kernfs operation | |
1351 | * finishes and this temporary bump can't break anything. If @kn | |
1352 | * is alive, nothing changes. If @kn is being deactivated, the | |
1353 | * soon-to-follow put will either finish deactivation or restore | |
1354 | * deactivated state. If @kn is already removed, the temporary | |
1355 | * bump is guaranteed to be gone before @kn is released. | |
1356 | */ | |
1357 | atomic_inc(&kn->active); | |
1358 | if (kernfs_lockdep(kn)) | |
1359 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
1360 | } | |
1361 | ||
1362 | /** | |
1363 | * kernfs_remove_self - remove a kernfs_node from its own method | |
1364 | * @kn: the self kernfs_node to remove | |
1365 | * | |
1366 | * The caller must be running off of a kernfs operation which is invoked | |
1367 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
1368 | * implement a file operation which deletes itself. | |
1369 | * | |
1370 | * For example, the "delete" file for a sysfs device directory can be | |
1371 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
1372 | * itself. This function breaks the circular dependency of trying to | |
1373 | * deactivate self while holding an active ref itself. It isn't necessary | |
1374 | * to modify the usual removal path to use kernfs_remove_self(). The | |
1375 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
1376 | * before proceeding with the usual removal path. kernfs will ignore later | |
1377 | * kernfs_remove() on self. | |
1378 | * | |
1379 | * kernfs_remove_self() can be called multiple times concurrently on the | |
1380 | * same kernfs_node. Only the first one actually performs removal and | |
1381 | * returns %true. All others will wait until the kernfs operation which | |
1382 | * won self-removal finishes and return %false. Note that the losers wait | |
1383 | * for the completion of not only the winning kernfs_remove_self() but also | |
1384 | * the whole kernfs_ops which won the arbitration. This can be used to | |
1385 | * guarantee, for example, all concurrent writes to a "delete" file to | |
1386 | * finish only after the whole operation is complete. | |
1387 | */ | |
1388 | bool kernfs_remove_self(struct kernfs_node *kn) | |
1389 | { | |
1390 | bool ret; | |
1391 | ||
1392 | mutex_lock(&kernfs_mutex); | |
1393 | kernfs_break_active_protection(kn); | |
1394 | ||
1395 | /* | |
1396 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1397 | * the first one will actually perform removal. When the removal | |
1398 | * is complete, SUICIDED is set and the active ref is restored | |
1399 | * while holding kernfs_mutex. The ones which lost arbitration | |
1400 | * waits for SUICDED && drained which can happen only after the | |
1401 | * enclosing kernfs operation which executed the winning instance | |
1402 | * of kernfs_remove_self() finished. | |
1403 | */ | |
1404 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1405 | kn->flags |= KERNFS_SUICIDAL; | |
1406 | __kernfs_remove(kn); | |
1407 | kn->flags |= KERNFS_SUICIDED; | |
1408 | ret = true; | |
1409 | } else { | |
1410 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1411 | DEFINE_WAIT(wait); | |
1412 | ||
1413 | while (true) { | |
1414 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1415 | ||
1416 | if ((kn->flags & KERNFS_SUICIDED) && | |
1417 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1418 | break; | |
1419 | ||
1420 | mutex_unlock(&kernfs_mutex); | |
1421 | schedule(); | |
1422 | mutex_lock(&kernfs_mutex); | |
1423 | } | |
1424 | finish_wait(waitq, &wait); | |
1425 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1426 | ret = false; | |
1427 | } | |
1428 | ||
1429 | /* | |
1430 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1431 | * for SUICIDED && deactivated could finish prematurely. | |
1432 | */ | |
1433 | kernfs_unbreak_active_protection(kn); | |
1434 | ||
1435 | mutex_unlock(&kernfs_mutex); | |
1436 | return ret; | |
1437 | } | |
1438 | ||
fd7b9f7b | 1439 | /** |
324a56e1 TH |
1440 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1441 | * @parent: parent of the target | |
1442 | * @name: name of the kernfs_node to remove | |
1443 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1444 | * |
324a56e1 TH |
1445 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1446 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1447 | */ |
324a56e1 | 1448 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1449 | const void *ns) |
1450 | { | |
324a56e1 | 1451 | struct kernfs_node *kn; |
fd7b9f7b | 1452 | |
324a56e1 | 1453 | if (!parent) { |
c637b8ac | 1454 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1455 | name); |
1456 | return -ENOENT; | |
1457 | } | |
1458 | ||
988cd7af | 1459 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1460 | |
324a56e1 TH |
1461 | kn = kernfs_find_ns(parent, name, ns); |
1462 | if (kn) | |
988cd7af | 1463 | __kernfs_remove(kn); |
fd7b9f7b | 1464 | |
988cd7af | 1465 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1466 | |
324a56e1 | 1467 | if (kn) |
fd7b9f7b TH |
1468 | return 0; |
1469 | else | |
1470 | return -ENOENT; | |
1471 | } | |
1472 | ||
1473 | /** | |
1474 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1475 | * @kn: target node |
fd7b9f7b TH |
1476 | * @new_parent: new parent to put @sd under |
1477 | * @new_name: new name | |
1478 | * @new_ns: new namespace tag | |
1479 | */ | |
324a56e1 | 1480 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1481 | const char *new_name, const void *new_ns) |
1482 | { | |
3eef34ad TH |
1483 | struct kernfs_node *old_parent; |
1484 | const char *old_name = NULL; | |
fd7b9f7b TH |
1485 | int error; |
1486 | ||
3eef34ad TH |
1487 | /* can't move or rename root */ |
1488 | if (!kn->parent) | |
1489 | return -EINVAL; | |
1490 | ||
798c75a0 GKH |
1491 | mutex_lock(&kernfs_mutex); |
1492 | ||
d0ae3d43 | 1493 | error = -ENOENT; |
ea015218 EB |
1494 | if (!kernfs_active(kn) || !kernfs_active(new_parent) || |
1495 | (new_parent->flags & KERNFS_EMPTY_DIR)) | |
d0ae3d43 TH |
1496 | goto out; |
1497 | ||
fd7b9f7b | 1498 | error = 0; |
adc5e8b5 TH |
1499 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1500 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1501 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1502 | |
1503 | error = -EEXIST; | |
1504 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1505 | goto out; |
fd7b9f7b | 1506 | |
324a56e1 | 1507 | /* rename kernfs_node */ |
adc5e8b5 | 1508 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b | 1509 | error = -ENOMEM; |
75287a67 | 1510 | new_name = kstrdup_const(new_name, GFP_KERNEL); |
fd7b9f7b | 1511 | if (!new_name) |
798c75a0 | 1512 | goto out; |
3eef34ad TH |
1513 | } else { |
1514 | new_name = NULL; | |
fd7b9f7b TH |
1515 | } |
1516 | ||
1517 | /* | |
1518 | * Move to the appropriate place in the appropriate directories rbtree. | |
1519 | */ | |
c637b8ac | 1520 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1521 | kernfs_get(new_parent); |
3eef34ad TH |
1522 | |
1523 | /* rename_lock protects ->parent and ->name accessors */ | |
1524 | spin_lock_irq(&kernfs_rename_lock); | |
1525 | ||
1526 | old_parent = kn->parent; | |
adc5e8b5 | 1527 | kn->parent = new_parent; |
3eef34ad TH |
1528 | |
1529 | kn->ns = new_ns; | |
1530 | if (new_name) { | |
dfeb0750 | 1531 | old_name = kn->name; |
3eef34ad TH |
1532 | kn->name = new_name; |
1533 | } | |
1534 | ||
1535 | spin_unlock_irq(&kernfs_rename_lock); | |
1536 | ||
9561a896 | 1537 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
c637b8ac | 1538 | kernfs_link_sibling(kn); |
fd7b9f7b | 1539 | |
3eef34ad | 1540 | kernfs_put(old_parent); |
75287a67 | 1541 | kfree_const(old_name); |
3eef34ad | 1542 | |
fd7b9f7b | 1543 | error = 0; |
798c75a0 | 1544 | out: |
a797bfc3 | 1545 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1546 | return error; |
1547 | } | |
1548 | ||
fd7b9f7b | 1549 | /* Relationship between s_mode and the DT_xxx types */ |
324a56e1 | 1550 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1551 | { |
adc5e8b5 | 1552 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1553 | } |
1554 | ||
c637b8ac | 1555 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1556 | { |
1557 | kernfs_put(filp->private_data); | |
1558 | return 0; | |
1559 | } | |
1560 | ||
c637b8ac | 1561 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1562 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1563 | { |
1564 | if (pos) { | |
81c173cb | 1565 | int valid = kernfs_active(pos) && |
798c75a0 | 1566 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1567 | kernfs_put(pos); |
1568 | if (!valid) | |
1569 | pos = NULL; | |
1570 | } | |
1571 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1572 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1573 | while (node) { |
324a56e1 | 1574 | pos = rb_to_kn(node); |
fd7b9f7b | 1575 | |
adc5e8b5 | 1576 | if (hash < pos->hash) |
fd7b9f7b | 1577 | node = node->rb_left; |
adc5e8b5 | 1578 | else if (hash > pos->hash) |
fd7b9f7b TH |
1579 | node = node->rb_right; |
1580 | else | |
1581 | break; | |
1582 | } | |
1583 | } | |
b9c9dad0 TH |
1584 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1585 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1586 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1587 | if (!node) |
1588 | pos = NULL; | |
1589 | else | |
324a56e1 | 1590 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1591 | } |
1592 | return pos; | |
1593 | } | |
1594 | ||
c637b8ac | 1595 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1596 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1597 | { |
c637b8ac | 1598 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1599 | if (pos) { |
fd7b9f7b | 1600 | do { |
adc5e8b5 | 1601 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1602 | if (!node) |
1603 | pos = NULL; | |
1604 | else | |
324a56e1 | 1605 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1606 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1607 | } | |
fd7b9f7b TH |
1608 | return pos; |
1609 | } | |
1610 | ||
c637b8ac | 1611 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1612 | { |
1613 | struct dentry *dentry = file->f_path.dentry; | |
324a56e1 TH |
1614 | struct kernfs_node *parent = dentry->d_fsdata; |
1615 | struct kernfs_node *pos = file->private_data; | |
fd7b9f7b TH |
1616 | const void *ns = NULL; |
1617 | ||
1618 | if (!dir_emit_dots(file, ctx)) | |
1619 | return 0; | |
a797bfc3 | 1620 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1621 | |
324a56e1 | 1622 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1623 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1624 | |
c637b8ac | 1625 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1626 | pos; |
c637b8ac | 1627 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1628 | const char *name = pos->name; |
fd7b9f7b TH |
1629 | unsigned int type = dt_type(pos); |
1630 | int len = strlen(name); | |
adc5e8b5 | 1631 | ino_t ino = pos->ino; |
fd7b9f7b | 1632 | |
adc5e8b5 | 1633 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1634 | file->private_data = pos; |
1635 | kernfs_get(pos); | |
1636 | ||
a797bfc3 | 1637 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1638 | if (!dir_emit(ctx, name, len, ino, type)) |
1639 | return 0; | |
a797bfc3 | 1640 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1641 | } |
a797bfc3 | 1642 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1643 | file->private_data = NULL; |
1644 | ctx->pos = INT_MAX; | |
1645 | return 0; | |
1646 | } | |
1647 | ||
a797bfc3 | 1648 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1649 | .read = generic_read_dir, |
8cb0d2c1 | 1650 | .iterate_shared = kernfs_fop_readdir, |
c637b8ac | 1651 | .release = kernfs_dir_fop_release, |
8cb0d2c1 | 1652 | .llseek = generic_file_llseek, |
fd7b9f7b | 1653 | }; |