]>
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 | * | |
3abb1d90 TH |
113 | * Returns the length of the full path. If the full length is equal to or |
114 | * greater than @buflen, @buf contains the truncated path with the trailing | |
115 | * '\0'. On error, -errno is returned. | |
9f6df573 AK |
116 | */ |
117 | static int kernfs_path_from_node_locked(struct kernfs_node *kn_to, | |
118 | struct kernfs_node *kn_from, | |
119 | char *buf, size_t buflen) | |
120 | { | |
121 | struct kernfs_node *kn, *common; | |
122 | const char parent_str[] = "/.."; | |
3abb1d90 TH |
123 | size_t depth_from, depth_to, len = 0; |
124 | int i, j; | |
9f6df573 AK |
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)) | |
3abb1d90 | 134 | return -EINVAL; |
9f6df573 AK |
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 */ | |
3abb1d90 TH |
147 | for (i = depth_to - 1; i >= 0; i--) { |
148 | for (kn = kn_to, j = 0; j < i; j++) | |
149 | kn = kn->parent; | |
150 | len += strlcpy(buf + len, "/", | |
151 | len < buflen ? buflen - len : 0); | |
152 | len += strlcpy(buf + len, kn->name, | |
153 | len < buflen ? buflen - len : 0); | |
9f6df573 | 154 | } |
3eef34ad | 155 | |
3abb1d90 | 156 | return len; |
3eef34ad TH |
157 | } |
158 | ||
159 | /** | |
160 | * kernfs_name - obtain the name of a given node | |
161 | * @kn: kernfs_node of interest | |
162 | * @buf: buffer to copy @kn's name into | |
163 | * @buflen: size of @buf | |
164 | * | |
165 | * Copies the name of @kn into @buf of @buflen bytes. The behavior is | |
166 | * similar to strlcpy(). It returns the length of @kn's name and if @buf | |
167 | * isn't long enough, it's filled upto @buflen-1 and nul terminated. | |
168 | * | |
169 | * This function can be called from any context. | |
170 | */ | |
171 | int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen) | |
172 | { | |
173 | unsigned long flags; | |
174 | int ret; | |
175 | ||
176 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
177 | ret = kernfs_name_locked(kn, buf, buflen); | |
178 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
179 | return ret; | |
180 | } | |
181 | ||
9acee9c5 TH |
182 | /** |
183 | * kernfs_path_len - determine the length of the full path of a given node | |
184 | * @kn: kernfs_node of interest | |
185 | * | |
186 | * The returned length doesn't include the space for the terminating '\0'. | |
187 | */ | |
188 | size_t kernfs_path_len(struct kernfs_node *kn) | |
189 | { | |
190 | size_t len = 0; | |
191 | unsigned long flags; | |
192 | ||
193 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
194 | ||
195 | do { | |
196 | len += strlen(kn->name) + 1; | |
197 | kn = kn->parent; | |
198 | } while (kn && kn->parent); | |
199 | ||
200 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
201 | ||
202 | return len; | |
203 | } | |
204 | ||
9f6df573 AK |
205 | /** |
206 | * kernfs_path_from_node - build path of node @to relative to @from. | |
207 | * @from: parent kernfs_node relative to which we need to build the path | |
208 | * @to: kernfs_node of interest | |
209 | * @buf: buffer to copy @to's path into | |
210 | * @buflen: size of @buf | |
211 | * | |
212 | * Builds @to's path relative to @from in @buf. @from and @to must | |
213 | * be on the same kernfs-root. If @from is not parent of @to, then a relative | |
214 | * path (which includes '..'s) as needed to reach from @from to @to is | |
215 | * returned. | |
216 | * | |
3abb1d90 TH |
217 | * Returns the length of the full path. If the full length is equal to or |
218 | * greater than @buflen, @buf contains the truncated path with the trailing | |
219 | * '\0'. On error, -errno is returned. | |
9f6df573 AK |
220 | */ |
221 | int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from, | |
222 | char *buf, size_t buflen) | |
223 | { | |
224 | unsigned long flags; | |
225 | int ret; | |
226 | ||
227 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
228 | ret = kernfs_path_from_node_locked(to, from, buf, buflen); | |
229 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
230 | return ret; | |
231 | } | |
232 | EXPORT_SYMBOL_GPL(kernfs_path_from_node); | |
233 | ||
3eef34ad TH |
234 | /** |
235 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node | |
236 | * @kn: kernfs_node of interest | |
237 | * | |
238 | * This function can be called from any context. | |
239 | */ | |
240 | void pr_cont_kernfs_name(struct kernfs_node *kn) | |
241 | { | |
242 | unsigned long flags; | |
243 | ||
244 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
245 | ||
246 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); | |
247 | pr_cont("%s", kernfs_pr_cont_buf); | |
248 | ||
249 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
250 | } | |
251 | ||
252 | /** | |
253 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node | |
254 | * @kn: kernfs_node of interest | |
255 | * | |
256 | * This function can be called from any context. | |
257 | */ | |
258 | void pr_cont_kernfs_path(struct kernfs_node *kn) | |
259 | { | |
260 | unsigned long flags; | |
9f6df573 | 261 | int sz; |
3eef34ad TH |
262 | |
263 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
264 | ||
9f6df573 AK |
265 | sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf, |
266 | sizeof(kernfs_pr_cont_buf)); | |
267 | if (sz < 0) { | |
268 | pr_cont("(error)"); | |
269 | goto out; | |
270 | } | |
271 | ||
272 | if (sz >= sizeof(kernfs_pr_cont_buf)) { | |
273 | pr_cont("(name too long)"); | |
274 | goto out; | |
275 | } | |
276 | ||
277 | pr_cont("%s", kernfs_pr_cont_buf); | |
3eef34ad | 278 | |
9f6df573 | 279 | out: |
3eef34ad TH |
280 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
281 | } | |
282 | ||
283 | /** | |
284 | * kernfs_get_parent - determine the parent node and pin it | |
285 | * @kn: kernfs_node of interest | |
286 | * | |
287 | * Determines @kn's parent, pins and returns it. This function can be | |
288 | * called from any context. | |
289 | */ | |
290 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) | |
291 | { | |
292 | struct kernfs_node *parent; | |
293 | unsigned long flags; | |
294 | ||
295 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
296 | parent = kn->parent; | |
297 | kernfs_get(parent); | |
298 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
299 | ||
300 | return parent; | |
301 | } | |
302 | ||
fd7b9f7b | 303 | /** |
c637b8ac | 304 | * kernfs_name_hash |
fd7b9f7b TH |
305 | * @name: Null terminated string to hash |
306 | * @ns: Namespace tag to hash | |
307 | * | |
308 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
309 | */ | |
c637b8ac | 310 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b | 311 | { |
8387ff25 | 312 | unsigned long hash = init_name_hash(ns); |
fd7b9f7b TH |
313 | unsigned int len = strlen(name); |
314 | while (len--) | |
315 | hash = partial_name_hash(*name++, hash); | |
8387ff25 | 316 | hash = end_name_hash(hash); |
fd7b9f7b TH |
317 | hash &= 0x7fffffffU; |
318 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
88391d49 | 319 | if (hash < 2) |
fd7b9f7b TH |
320 | hash += 2; |
321 | if (hash >= INT_MAX) | |
322 | hash = INT_MAX - 1; | |
323 | return hash; | |
324 | } | |
325 | ||
c637b8ac TH |
326 | static int kernfs_name_compare(unsigned int hash, const char *name, |
327 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 328 | { |
72392ed0 RV |
329 | if (hash < kn->hash) |
330 | return -1; | |
331 | if (hash > kn->hash) | |
332 | return 1; | |
333 | if (ns < kn->ns) | |
334 | return -1; | |
335 | if (ns > kn->ns) | |
336 | return 1; | |
adc5e8b5 | 337 | return strcmp(name, kn->name); |
fd7b9f7b TH |
338 | } |
339 | ||
c637b8ac TH |
340 | static int kernfs_sd_compare(const struct kernfs_node *left, |
341 | const struct kernfs_node *right) | |
fd7b9f7b | 342 | { |
c637b8ac | 343 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
344 | } |
345 | ||
346 | /** | |
c637b8ac | 347 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 348 | * @kn: kernfs_node of interest |
fd7b9f7b | 349 | * |
324a56e1 | 350 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 351 | * @kn->parent->dir.children. |
fd7b9f7b TH |
352 | * |
353 | * Locking: | |
a797bfc3 | 354 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
355 | * |
356 | * RETURNS: | |
357 | * 0 on susccess -EEXIST on failure. | |
358 | */ | |
c637b8ac | 359 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 360 | { |
adc5e8b5 | 361 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
362 | struct rb_node *parent = NULL; |
363 | ||
fd7b9f7b | 364 | while (*node) { |
324a56e1 | 365 | struct kernfs_node *pos; |
fd7b9f7b TH |
366 | int result; |
367 | ||
324a56e1 | 368 | pos = rb_to_kn(*node); |
fd7b9f7b | 369 | parent = *node; |
c637b8ac | 370 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 371 | if (result < 0) |
adc5e8b5 | 372 | node = &pos->rb.rb_left; |
fd7b9f7b | 373 | else if (result > 0) |
adc5e8b5 | 374 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
375 | else |
376 | return -EEXIST; | |
377 | } | |
c1befb88 | 378 | |
fd7b9f7b | 379 | /* add new node and rebalance the tree */ |
adc5e8b5 TH |
380 | rb_link_node(&kn->rb, parent, node); |
381 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
c1befb88 JZ |
382 | |
383 | /* successfully added, account subdir number */ | |
384 | if (kernfs_type(kn) == KERNFS_DIR) | |
385 | kn->parent->dir.subdirs++; | |
386 | ||
fd7b9f7b TH |
387 | return 0; |
388 | } | |
389 | ||
390 | /** | |
c637b8ac | 391 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 392 | * @kn: kernfs_node of interest |
fd7b9f7b | 393 | * |
35beab06 TH |
394 | * Try to unlink @kn from its sibling rbtree which starts from |
395 | * kn->parent->dir.children. Returns %true if @kn was actually | |
396 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
397 | * |
398 | * Locking: | |
a797bfc3 | 399 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 400 | */ |
35beab06 | 401 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 402 | { |
35beab06 TH |
403 | if (RB_EMPTY_NODE(&kn->rb)) |
404 | return false; | |
405 | ||
df23fc39 | 406 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 407 | kn->parent->dir.subdirs--; |
fd7b9f7b | 408 | |
adc5e8b5 | 409 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
410 | RB_CLEAR_NODE(&kn->rb); |
411 | return true; | |
fd7b9f7b TH |
412 | } |
413 | ||
414 | /** | |
c637b8ac | 415 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 416 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 417 | * |
324a56e1 | 418 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
419 | * is NULL. |
420 | * | |
421 | * RETURNS: | |
324a56e1 | 422 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 423 | */ |
c637b8ac | 424 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 425 | { |
324a56e1 | 426 | if (unlikely(!kn)) |
fd7b9f7b TH |
427 | return NULL; |
428 | ||
f4b3e631 GKH |
429 | if (!atomic_inc_unless_negative(&kn->active)) |
430 | return NULL; | |
895a068a | 431 | |
182fd64b | 432 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
433 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
434 | return kn; | |
fd7b9f7b TH |
435 | } |
436 | ||
437 | /** | |
c637b8ac | 438 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 439 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 440 | * |
324a56e1 | 441 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
442 | * is NULL. |
443 | */ | |
c637b8ac | 444 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b | 445 | { |
abd54f02 | 446 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b TH |
447 | int v; |
448 | ||
324a56e1 | 449 | if (unlikely(!kn)) |
fd7b9f7b TH |
450 | return; |
451 | ||
182fd64b | 452 | if (kernfs_lockdep(kn)) |
324a56e1 | 453 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
adc5e8b5 | 454 | v = atomic_dec_return(&kn->active); |
df23fc39 | 455 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
456 | return; |
457 | ||
abd54f02 | 458 | wake_up_all(&root->deactivate_waitq); |
fd7b9f7b TH |
459 | } |
460 | ||
461 | /** | |
81c173cb TH |
462 | * kernfs_drain - drain kernfs_node |
463 | * @kn: kernfs_node to drain | |
fd7b9f7b | 464 | * |
81c173cb TH |
465 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
466 | * removers may invoke this function concurrently on @kn and all will | |
467 | * return after draining is complete. | |
fd7b9f7b | 468 | */ |
81c173cb | 469 | static void kernfs_drain(struct kernfs_node *kn) |
35beab06 | 470 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
fd7b9f7b | 471 | { |
abd54f02 | 472 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b | 473 | |
35beab06 | 474 | lockdep_assert_held(&kernfs_mutex); |
81c173cb | 475 | WARN_ON_ONCE(kernfs_active(kn)); |
ea1c472d | 476 | |
35beab06 | 477 | mutex_unlock(&kernfs_mutex); |
abd54f02 | 478 | |
182fd64b | 479 | if (kernfs_lockdep(kn)) { |
35beab06 TH |
480 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
481 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) | |
482 | lock_contended(&kn->dep_map, _RET_IP_); | |
483 | } | |
abd54f02 | 484 | |
35beab06 | 485 | /* but everyone should wait for draining */ |
abd54f02 TH |
486 | wait_event(root->deactivate_waitq, |
487 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); | |
fd7b9f7b | 488 | |
182fd64b | 489 | if (kernfs_lockdep(kn)) { |
a6607930 TH |
490 | lock_acquired(&kn->dep_map, _RET_IP_); |
491 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | |
492 | } | |
35beab06 | 493 | |
ccf02aaf TH |
494 | kernfs_unmap_bin_file(kn); |
495 | ||
35beab06 | 496 | mutex_lock(&kernfs_mutex); |
fd7b9f7b TH |
497 | } |
498 | ||
fd7b9f7b | 499 | /** |
324a56e1 TH |
500 | * kernfs_get - get a reference count on a kernfs_node |
501 | * @kn: the target kernfs_node | |
fd7b9f7b | 502 | */ |
324a56e1 | 503 | void kernfs_get(struct kernfs_node *kn) |
fd7b9f7b | 504 | { |
324a56e1 | 505 | if (kn) { |
adc5e8b5 TH |
506 | WARN_ON(!atomic_read(&kn->count)); |
507 | atomic_inc(&kn->count); | |
fd7b9f7b TH |
508 | } |
509 | } | |
510 | EXPORT_SYMBOL_GPL(kernfs_get); | |
511 | ||
512 | /** | |
324a56e1 TH |
513 | * kernfs_put - put a reference count on a kernfs_node |
514 | * @kn: the target kernfs_node | |
fd7b9f7b | 515 | * |
324a56e1 | 516 | * Put a reference count of @kn and destroy it if it reached zero. |
fd7b9f7b | 517 | */ |
324a56e1 | 518 | void kernfs_put(struct kernfs_node *kn) |
fd7b9f7b | 519 | { |
324a56e1 | 520 | struct kernfs_node *parent; |
ba7443bc | 521 | struct kernfs_root *root; |
fd7b9f7b | 522 | |
adc5e8b5 | 523 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 524 | return; |
324a56e1 | 525 | root = kernfs_root(kn); |
fd7b9f7b | 526 | repeat: |
81c173cb TH |
527 | /* |
528 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 529 | * kn->parent won't change beneath us. |
fd7b9f7b | 530 | */ |
adc5e8b5 | 531 | parent = kn->parent; |
fd7b9f7b | 532 | |
81c173cb TH |
533 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
534 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
535 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 536 | |
df23fc39 | 537 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 538 | kernfs_put(kn->symlink.target_kn); |
dfeb0750 TH |
539 | |
540 | kfree_const(kn->name); | |
541 | ||
adc5e8b5 TH |
542 | if (kn->iattr) { |
543 | if (kn->iattr->ia_secdata) | |
544 | security_release_secctx(kn->iattr->ia_secdata, | |
545 | kn->iattr->ia_secdata_len); | |
546 | simple_xattrs_free(&kn->iattr->xattrs); | |
2322392b | 547 | } |
adc5e8b5 TH |
548 | kfree(kn->iattr); |
549 | ida_simple_remove(&root->ino_ida, kn->ino); | |
a797bfc3 | 550 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 551 | |
324a56e1 TH |
552 | kn = parent; |
553 | if (kn) { | |
adc5e8b5 | 554 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
555 | goto repeat; |
556 | } else { | |
324a56e1 | 557 | /* just released the root kn, free @root too */ |
bc755553 | 558 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
559 | kfree(root); |
560 | } | |
fd7b9f7b TH |
561 | } |
562 | EXPORT_SYMBOL_GPL(kernfs_put); | |
563 | ||
c637b8ac | 564 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 565 | { |
324a56e1 | 566 | struct kernfs_node *kn; |
fd7b9f7b TH |
567 | |
568 | if (flags & LOOKUP_RCU) | |
569 | return -ECHILD; | |
570 | ||
19bbb926 | 571 | /* Always perform fresh lookup for negatives */ |
2b0143b5 | 572 | if (d_really_is_negative(dentry)) |
19bbb926 TH |
573 | goto out_bad_unlocked; |
574 | ||
324a56e1 | 575 | kn = dentry->d_fsdata; |
a797bfc3 | 576 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 577 | |
81c173cb TH |
578 | /* The kernfs node has been deactivated */ |
579 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
580 | goto out_bad; |
581 | ||
c637b8ac | 582 | /* The kernfs node has been moved? */ |
adc5e8b5 | 583 | if (dentry->d_parent->d_fsdata != kn->parent) |
fd7b9f7b TH |
584 | goto out_bad; |
585 | ||
c637b8ac | 586 | /* The kernfs node has been renamed */ |
adc5e8b5 | 587 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
588 | goto out_bad; |
589 | ||
c637b8ac | 590 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 591 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 592 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
593 | goto out_bad; |
594 | ||
a797bfc3 | 595 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
596 | return 1; |
597 | out_bad: | |
a797bfc3 | 598 | mutex_unlock(&kernfs_mutex); |
19bbb926 | 599 | out_bad_unlocked: |
fd7b9f7b TH |
600 | return 0; |
601 | } | |
602 | ||
c637b8ac | 603 | static void kernfs_dop_release(struct dentry *dentry) |
fd7b9f7b TH |
604 | { |
605 | kernfs_put(dentry->d_fsdata); | |
606 | } | |
607 | ||
a797bfc3 | 608 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 609 | .d_revalidate = kernfs_dop_revalidate, |
c637b8ac | 610 | .d_release = kernfs_dop_release, |
fd7b9f7b TH |
611 | }; |
612 | ||
0c23b225 TH |
613 | /** |
614 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry | |
615 | * @dentry: the dentry in question | |
616 | * | |
617 | * Return the kernfs_node associated with @dentry. If @dentry is not a | |
618 | * kernfs one, %NULL is returned. | |
619 | * | |
620 | * While the returned kernfs_node will stay accessible as long as @dentry | |
621 | * is accessible, the returned node can be in any state and the caller is | |
622 | * fully responsible for determining what's accessible. | |
623 | */ | |
624 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) | |
625 | { | |
f41c5934 | 626 | if (dentry->d_sb->s_op == &kernfs_sops) |
0c23b225 TH |
627 | return dentry->d_fsdata; |
628 | return NULL; | |
629 | } | |
630 | ||
db4aad20 TH |
631 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
632 | const char *name, umode_t mode, | |
633 | unsigned flags) | |
fd7b9f7b | 634 | { |
324a56e1 | 635 | struct kernfs_node *kn; |
bc755553 | 636 | int ret; |
fd7b9f7b | 637 | |
dfeb0750 TH |
638 | name = kstrdup_const(name, GFP_KERNEL); |
639 | if (!name) | |
640 | return NULL; | |
fd7b9f7b | 641 | |
a797bfc3 | 642 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 643 | if (!kn) |
fd7b9f7b TH |
644 | goto err_out1; |
645 | ||
b2a209ff | 646 | ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); |
bc755553 | 647 | if (ret < 0) |
fd7b9f7b | 648 | goto err_out2; |
adc5e8b5 | 649 | kn->ino = ret; |
fd7b9f7b | 650 | |
adc5e8b5 | 651 | atomic_set(&kn->count, 1); |
81c173cb | 652 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 653 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 654 | |
adc5e8b5 TH |
655 | kn->name = name; |
656 | kn->mode = mode; | |
81c173cb | 657 | kn->flags = flags; |
fd7b9f7b | 658 | |
324a56e1 | 659 | return kn; |
fd7b9f7b TH |
660 | |
661 | err_out2: | |
a797bfc3 | 662 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 663 | err_out1: |
dfeb0750 | 664 | kfree_const(name); |
fd7b9f7b TH |
665 | return NULL; |
666 | } | |
667 | ||
db4aad20 TH |
668 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
669 | const char *name, umode_t mode, | |
670 | unsigned flags) | |
671 | { | |
672 | struct kernfs_node *kn; | |
673 | ||
674 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); | |
675 | if (kn) { | |
676 | kernfs_get(parent); | |
677 | kn->parent = parent; | |
678 | } | |
679 | return kn; | |
680 | } | |
681 | ||
fd7b9f7b | 682 | /** |
c637b8ac | 683 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 684 | * @kn: kernfs_node to be added |
fd7b9f7b | 685 | * |
db4aad20 TH |
686 | * The caller must already have initialized @kn->parent. This |
687 | * function increments nlink of the parent's inode if @kn is a | |
688 | * directory and link into the children list of the parent. | |
fd7b9f7b | 689 | * |
fd7b9f7b TH |
690 | * RETURNS: |
691 | * 0 on success, -EEXIST if entry with the given name already | |
692 | * exists. | |
693 | */ | |
988cd7af | 694 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 695 | { |
db4aad20 | 696 | struct kernfs_node *parent = kn->parent; |
c525aadd | 697 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 698 | bool has_ns; |
fd7b9f7b TH |
699 | int ret; |
700 | ||
988cd7af TH |
701 | mutex_lock(&kernfs_mutex); |
702 | ||
703 | ret = -EINVAL; | |
704 | has_ns = kernfs_ns_enabled(parent); | |
705 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
706 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
707 | goto out_unlock; | |
fd7b9f7b | 708 | |
df23fc39 | 709 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 710 | goto out_unlock; |
fd7b9f7b | 711 | |
988cd7af | 712 | ret = -ENOENT; |
ea015218 EB |
713 | if (parent->flags & KERNFS_EMPTY_DIR) |
714 | goto out_unlock; | |
715 | ||
d35258ef | 716 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 717 | goto out_unlock; |
798c75a0 | 718 | |
c637b8ac | 719 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 720 | |
c637b8ac | 721 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 722 | if (ret) |
988cd7af | 723 | goto out_unlock; |
fd7b9f7b TH |
724 | |
725 | /* Update timestamps on the parent */ | |
adc5e8b5 | 726 | ps_iattr = parent->iattr; |
fd7b9f7b TH |
727 | if (ps_iattr) { |
728 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
3a3a5fec DD |
729 | ktime_get_real_ts(&ps_iattrs->ia_ctime); |
730 | ps_iattrs->ia_mtime = ps_iattrs->ia_ctime; | |
fd7b9f7b TH |
731 | } |
732 | ||
d35258ef TH |
733 | mutex_unlock(&kernfs_mutex); |
734 | ||
735 | /* | |
736 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
737 | * If not activated here, the kernfs user is responsible for | |
738 | * activating the node with kernfs_activate(). A node which hasn't | |
739 | * been activated is not visible to userland and its removal won't | |
740 | * trigger deactivation. | |
741 | */ | |
742 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
743 | kernfs_activate(kn); | |
744 | return 0; | |
745 | ||
988cd7af | 746 | out_unlock: |
a797bfc3 | 747 | mutex_unlock(&kernfs_mutex); |
988cd7af | 748 | return ret; |
fd7b9f7b TH |
749 | } |
750 | ||
751 | /** | |
324a56e1 TH |
752 | * kernfs_find_ns - find kernfs_node with the given name |
753 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
754 | * @name: name to look for |
755 | * @ns: the namespace tag to use | |
756 | * | |
324a56e1 TH |
757 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
758 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 759 | */ |
324a56e1 TH |
760 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
761 | const unsigned char *name, | |
762 | const void *ns) | |
fd7b9f7b | 763 | { |
adc5e8b5 | 764 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 765 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
766 | unsigned int hash; |
767 | ||
a797bfc3 | 768 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
769 | |
770 | if (has_ns != (bool)ns) { | |
c637b8ac | 771 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 772 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
773 | return NULL; |
774 | } | |
775 | ||
c637b8ac | 776 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 777 | while (node) { |
324a56e1 | 778 | struct kernfs_node *kn; |
fd7b9f7b TH |
779 | int result; |
780 | ||
324a56e1 | 781 | kn = rb_to_kn(node); |
c637b8ac | 782 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
783 | if (result < 0) |
784 | node = node->rb_left; | |
785 | else if (result > 0) | |
786 | node = node->rb_right; | |
787 | else | |
324a56e1 | 788 | return kn; |
fd7b9f7b TH |
789 | } |
790 | return NULL; | |
791 | } | |
792 | ||
bd96f76a TH |
793 | static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent, |
794 | const unsigned char *path, | |
795 | const void *ns) | |
796 | { | |
e56ed358 TH |
797 | size_t len; |
798 | char *p, *name; | |
bd96f76a TH |
799 | |
800 | lockdep_assert_held(&kernfs_mutex); | |
801 | ||
e56ed358 TH |
802 | /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */ |
803 | spin_lock_irq(&kernfs_rename_lock); | |
804 | ||
805 | len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf)); | |
806 | ||
807 | if (len >= sizeof(kernfs_pr_cont_buf)) { | |
808 | spin_unlock_irq(&kernfs_rename_lock); | |
bd96f76a | 809 | return NULL; |
e56ed358 TH |
810 | } |
811 | ||
812 | p = kernfs_pr_cont_buf; | |
bd96f76a TH |
813 | |
814 | while ((name = strsep(&p, "/")) && parent) { | |
815 | if (*name == '\0') | |
816 | continue; | |
817 | parent = kernfs_find_ns(parent, name, ns); | |
818 | } | |
819 | ||
e56ed358 TH |
820 | spin_unlock_irq(&kernfs_rename_lock); |
821 | ||
bd96f76a TH |
822 | return parent; |
823 | } | |
824 | ||
fd7b9f7b | 825 | /** |
324a56e1 TH |
826 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
827 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
828 | * @name: name to look for |
829 | * @ns: the namespace tag to use | |
830 | * | |
324a56e1 | 831 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 832 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 833 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 834 | */ |
324a56e1 TH |
835 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
836 | const char *name, const void *ns) | |
fd7b9f7b | 837 | { |
324a56e1 | 838 | struct kernfs_node *kn; |
fd7b9f7b | 839 | |
a797bfc3 | 840 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
841 | kn = kernfs_find_ns(parent, name, ns); |
842 | kernfs_get(kn); | |
a797bfc3 | 843 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 844 | |
324a56e1 | 845 | return kn; |
fd7b9f7b TH |
846 | } |
847 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
848 | ||
bd96f76a TH |
849 | /** |
850 | * kernfs_walk_and_get_ns - find and get kernfs_node with the given path | |
851 | * @parent: kernfs_node to search under | |
852 | * @path: path to look for | |
853 | * @ns: the namespace tag to use | |
854 | * | |
855 | * Look for kernfs_node with path @path under @parent and get a reference | |
856 | * if found. This function may sleep and returns pointer to the found | |
857 | * kernfs_node on success, %NULL on failure. | |
858 | */ | |
859 | struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent, | |
860 | const char *path, const void *ns) | |
861 | { | |
862 | struct kernfs_node *kn; | |
863 | ||
864 | mutex_lock(&kernfs_mutex); | |
865 | kn = kernfs_walk_ns(parent, path, ns); | |
866 | kernfs_get(kn); | |
867 | mutex_unlock(&kernfs_mutex); | |
868 | ||
869 | return kn; | |
870 | } | |
871 | ||
ba7443bc TH |
872 | /** |
873 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 874 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 875 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
876 | * @priv: opaque data associated with the new directory |
877 | * | |
878 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
879 | * failure. | |
880 | */ | |
90c07c89 | 881 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 882 | unsigned int flags, void *priv) |
ba7443bc TH |
883 | { |
884 | struct kernfs_root *root; | |
324a56e1 | 885 | struct kernfs_node *kn; |
ba7443bc TH |
886 | |
887 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
888 | if (!root) | |
889 | return ERR_PTR(-ENOMEM); | |
890 | ||
bc755553 | 891 | ida_init(&root->ino_ida); |
7d568a83 | 892 | INIT_LIST_HEAD(&root->supers); |
bc755553 | 893 | |
db4aad20 TH |
894 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
895 | KERNFS_DIR); | |
324a56e1 | 896 | if (!kn) { |
bc755553 | 897 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
898 | kfree(root); |
899 | return ERR_PTR(-ENOMEM); | |
900 | } | |
901 | ||
324a56e1 | 902 | kn->priv = priv; |
adc5e8b5 | 903 | kn->dir.root = root; |
ba7443bc | 904 | |
90c07c89 | 905 | root->syscall_ops = scops; |
d35258ef | 906 | root->flags = flags; |
324a56e1 | 907 | root->kn = kn; |
abd54f02 | 908 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 909 | |
d35258ef TH |
910 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
911 | kernfs_activate(kn); | |
912 | ||
ba7443bc TH |
913 | return root; |
914 | } | |
915 | ||
916 | /** | |
917 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
918 | * @root: root of the hierarchy to destroy | |
919 | * | |
920 | * Destroy the hierarchy anchored at @root by removing all existing | |
921 | * directories and destroying @root. | |
922 | */ | |
923 | void kernfs_destroy_root(struct kernfs_root *root) | |
924 | { | |
324a56e1 | 925 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
926 | } |
927 | ||
fd7b9f7b TH |
928 | /** |
929 | * kernfs_create_dir_ns - create a directory | |
930 | * @parent: parent in which to create a new directory | |
931 | * @name: name of the new directory | |
bb8b9d09 | 932 | * @mode: mode of the new directory |
fd7b9f7b TH |
933 | * @priv: opaque data associated with the new directory |
934 | * @ns: optional namespace tag of the directory | |
935 | * | |
936 | * Returns the created node on success, ERR_PTR() value on failure. | |
937 | */ | |
324a56e1 | 938 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 TH |
939 | const char *name, umode_t mode, |
940 | void *priv, const void *ns) | |
fd7b9f7b | 941 | { |
324a56e1 | 942 | struct kernfs_node *kn; |
fd7b9f7b TH |
943 | int rc; |
944 | ||
945 | /* allocate */ | |
db4aad20 | 946 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); |
324a56e1 | 947 | if (!kn) |
fd7b9f7b TH |
948 | return ERR_PTR(-ENOMEM); |
949 | ||
adc5e8b5 TH |
950 | kn->dir.root = parent->dir.root; |
951 | kn->ns = ns; | |
324a56e1 | 952 | kn->priv = priv; |
fd7b9f7b TH |
953 | |
954 | /* link in */ | |
988cd7af | 955 | rc = kernfs_add_one(kn); |
fd7b9f7b | 956 | if (!rc) |
324a56e1 | 957 | return kn; |
fd7b9f7b | 958 | |
324a56e1 | 959 | kernfs_put(kn); |
fd7b9f7b TH |
960 | return ERR_PTR(rc); |
961 | } | |
962 | ||
ea015218 EB |
963 | /** |
964 | * kernfs_create_empty_dir - create an always empty directory | |
965 | * @parent: parent in which to create a new directory | |
966 | * @name: name of the new directory | |
967 | * | |
968 | * Returns the created node on success, ERR_PTR() value on failure. | |
969 | */ | |
970 | struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent, | |
971 | const char *name) | |
972 | { | |
973 | struct kernfs_node *kn; | |
974 | int rc; | |
975 | ||
976 | /* allocate */ | |
977 | kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, KERNFS_DIR); | |
978 | if (!kn) | |
979 | return ERR_PTR(-ENOMEM); | |
980 | ||
981 | kn->flags |= KERNFS_EMPTY_DIR; | |
982 | kn->dir.root = parent->dir.root; | |
983 | kn->ns = NULL; | |
984 | kn->priv = NULL; | |
985 | ||
986 | /* link in */ | |
987 | rc = kernfs_add_one(kn); | |
988 | if (!rc) | |
989 | return kn; | |
990 | ||
991 | kernfs_put(kn); | |
992 | return ERR_PTR(rc); | |
993 | } | |
994 | ||
c637b8ac TH |
995 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
996 | struct dentry *dentry, | |
997 | unsigned int flags) | |
fd7b9f7b | 998 | { |
19bbb926 | 999 | struct dentry *ret; |
324a56e1 TH |
1000 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; |
1001 | struct kernfs_node *kn; | |
fd7b9f7b TH |
1002 | struct inode *inode; |
1003 | const void *ns = NULL; | |
1004 | ||
a797bfc3 | 1005 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1006 | |
324a56e1 | 1007 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1008 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 1009 | |
324a56e1 | 1010 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
1011 | |
1012 | /* no such entry */ | |
b9c9dad0 | 1013 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 1014 | ret = NULL; |
fd7b9f7b TH |
1015 | goto out_unlock; |
1016 | } | |
324a56e1 TH |
1017 | kernfs_get(kn); |
1018 | dentry->d_fsdata = kn; | |
fd7b9f7b TH |
1019 | |
1020 | /* attach dentry and inode */ | |
c637b8ac | 1021 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
1022 | if (!inode) { |
1023 | ret = ERR_PTR(-ENOMEM); | |
1024 | goto out_unlock; | |
1025 | } | |
1026 | ||
1027 | /* instantiate and hash dentry */ | |
41d28bca | 1028 | ret = d_splice_alias(inode, dentry); |
fd7b9f7b | 1029 | out_unlock: |
a797bfc3 | 1030 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1031 | return ret; |
1032 | } | |
1033 | ||
80b9bbef TH |
1034 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
1035 | umode_t mode) | |
1036 | { | |
1037 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 1038 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 1039 | int ret; |
80b9bbef | 1040 | |
90c07c89 | 1041 | if (!scops || !scops->mkdir) |
80b9bbef TH |
1042 | return -EPERM; |
1043 | ||
07c7530d TH |
1044 | if (!kernfs_get_active(parent)) |
1045 | return -ENODEV; | |
1046 | ||
90c07c89 | 1047 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
1048 | |
1049 | kernfs_put_active(parent); | |
1050 | return ret; | |
80b9bbef TH |
1051 | } |
1052 | ||
1053 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
1054 | { | |
1055 | struct kernfs_node *kn = dentry->d_fsdata; | |
90c07c89 | 1056 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1057 | int ret; |
80b9bbef | 1058 | |
90c07c89 | 1059 | if (!scops || !scops->rmdir) |
80b9bbef TH |
1060 | return -EPERM; |
1061 | ||
07c7530d TH |
1062 | if (!kernfs_get_active(kn)) |
1063 | return -ENODEV; | |
1064 | ||
90c07c89 | 1065 | ret = scops->rmdir(kn); |
07c7530d TH |
1066 | |
1067 | kernfs_put_active(kn); | |
1068 | return ret; | |
80b9bbef TH |
1069 | } |
1070 | ||
1071 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1072 | struct inode *new_dir, struct dentry *new_dentry) | |
1073 | { | |
1074 | struct kernfs_node *kn = old_dentry->d_fsdata; | |
1075 | struct kernfs_node *new_parent = new_dir->i_private; | |
90c07c89 | 1076 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1077 | int ret; |
80b9bbef | 1078 | |
90c07c89 | 1079 | if (!scops || !scops->rename) |
80b9bbef TH |
1080 | return -EPERM; |
1081 | ||
07c7530d TH |
1082 | if (!kernfs_get_active(kn)) |
1083 | return -ENODEV; | |
1084 | ||
1085 | if (!kernfs_get_active(new_parent)) { | |
1086 | kernfs_put_active(kn); | |
1087 | return -ENODEV; | |
1088 | } | |
1089 | ||
90c07c89 | 1090 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
1091 | |
1092 | kernfs_put_active(new_parent); | |
1093 | kernfs_put_active(kn); | |
1094 | return ret; | |
80b9bbef TH |
1095 | } |
1096 | ||
a797bfc3 | 1097 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
1098 | .lookup = kernfs_iop_lookup, |
1099 | .permission = kernfs_iop_permission, | |
1100 | .setattr = kernfs_iop_setattr, | |
1101 | .getattr = kernfs_iop_getattr, | |
1102 | .setxattr = kernfs_iop_setxattr, | |
1103 | .removexattr = kernfs_iop_removexattr, | |
1104 | .getxattr = kernfs_iop_getxattr, | |
1105 | .listxattr = kernfs_iop_listxattr, | |
80b9bbef TH |
1106 | |
1107 | .mkdir = kernfs_iop_mkdir, | |
1108 | .rmdir = kernfs_iop_rmdir, | |
1109 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
1110 | }; |
1111 | ||
c637b8ac | 1112 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 1113 | { |
324a56e1 | 1114 | struct kernfs_node *last; |
fd7b9f7b TH |
1115 | |
1116 | while (true) { | |
1117 | struct rb_node *rbn; | |
1118 | ||
1119 | last = pos; | |
1120 | ||
df23fc39 | 1121 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
1122 | break; |
1123 | ||
adc5e8b5 | 1124 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
1125 | if (!rbn) |
1126 | break; | |
1127 | ||
324a56e1 | 1128 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
1129 | } |
1130 | ||
1131 | return last; | |
1132 | } | |
1133 | ||
1134 | /** | |
c637b8ac | 1135 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 1136 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 1137 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
1138 | * |
1139 | * Find the next descendant to visit for post-order traversal of @root's | |
1140 | * descendants. @root is included in the iteration and the last node to be | |
1141 | * visited. | |
1142 | */ | |
c637b8ac TH |
1143 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
1144 | struct kernfs_node *root) | |
fd7b9f7b TH |
1145 | { |
1146 | struct rb_node *rbn; | |
1147 | ||
a797bfc3 | 1148 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
1149 | |
1150 | /* if first iteration, visit leftmost descendant which may be root */ | |
1151 | if (!pos) | |
c637b8ac | 1152 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
1153 | |
1154 | /* if we visited @root, we're done */ | |
1155 | if (pos == root) | |
1156 | return NULL; | |
1157 | ||
1158 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 1159 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 1160 | if (rbn) |
c637b8ac | 1161 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
1162 | |
1163 | /* no sibling left, visit parent */ | |
adc5e8b5 | 1164 | return pos->parent; |
fd7b9f7b TH |
1165 | } |
1166 | ||
d35258ef TH |
1167 | /** |
1168 | * kernfs_activate - activate a node which started deactivated | |
1169 | * @kn: kernfs_node whose subtree is to be activated | |
1170 | * | |
1171 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
1172 | * needs to be explicitly activated. A node which hasn't been activated | |
1173 | * isn't visible to userland and deactivation is skipped during its | |
1174 | * removal. This is useful to construct atomic init sequences where | |
1175 | * creation of multiple nodes should either succeed or fail atomically. | |
1176 | * | |
1177 | * The caller is responsible for ensuring that this function is not called | |
1178 | * after kernfs_remove*() is invoked on @kn. | |
1179 | */ | |
1180 | void kernfs_activate(struct kernfs_node *kn) | |
1181 | { | |
1182 | struct kernfs_node *pos; | |
1183 | ||
1184 | mutex_lock(&kernfs_mutex); | |
1185 | ||
1186 | pos = NULL; | |
1187 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
1188 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) | |
1189 | continue; | |
1190 | ||
1191 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
1192 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
1193 | ||
1194 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
1195 | pos->flags |= KERNFS_ACTIVATED; | |
1196 | } | |
1197 | ||
1198 | mutex_unlock(&kernfs_mutex); | |
1199 | } | |
1200 | ||
988cd7af | 1201 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1202 | { |
35beab06 TH |
1203 | struct kernfs_node *pos; |
1204 | ||
1205 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 1206 | |
6b0afc2a TH |
1207 | /* |
1208 | * Short-circuit if non-root @kn has already finished removal. | |
1209 | * This is for kernfs_remove_self() which plays with active ref | |
1210 | * after removal. | |
1211 | */ | |
1212 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
1213 | return; |
1214 | ||
c637b8ac | 1215 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 1216 | |
81c173cb | 1217 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
1218 | pos = NULL; |
1219 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
1220 | if (kernfs_active(pos)) |
1221 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
1222 | |
1223 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 1224 | do { |
35beab06 TH |
1225 | pos = kernfs_leftmost_descendant(kn); |
1226 | ||
1227 | /* | |
81c173cb TH |
1228 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
1229 | * base ref could have been put by someone else by the time | |
1230 | * the function returns. Make sure it doesn't go away | |
1231 | * underneath us. | |
35beab06 TH |
1232 | */ |
1233 | kernfs_get(pos); | |
1234 | ||
d35258ef TH |
1235 | /* |
1236 | * Drain iff @kn was activated. This avoids draining and | |
1237 | * its lockdep annotations for nodes which have never been | |
1238 | * activated and allows embedding kernfs_remove() in create | |
1239 | * error paths without worrying about draining. | |
1240 | */ | |
1241 | if (kn->flags & KERNFS_ACTIVATED) | |
1242 | kernfs_drain(pos); | |
1243 | else | |
1244 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
1245 | |
1246 | /* | |
1247 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
1248 | * to decide who's responsible for cleanups. | |
1249 | */ | |
1250 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
1251 | struct kernfs_iattrs *ps_iattr = | |
1252 | pos->parent ? pos->parent->iattr : NULL; | |
1253 | ||
1254 | /* update timestamps on the parent */ | |
1255 | if (ps_iattr) { | |
3a3a5fec DD |
1256 | ktime_get_real_ts(&ps_iattr->ia_iattr.ia_ctime); |
1257 | ps_iattr->ia_iattr.ia_mtime = | |
1258 | ps_iattr->ia_iattr.ia_ctime; | |
35beab06 TH |
1259 | } |
1260 | ||
988cd7af | 1261 | kernfs_put(pos); |
35beab06 TH |
1262 | } |
1263 | ||
1264 | kernfs_put(pos); | |
1265 | } while (pos != kn); | |
fd7b9f7b TH |
1266 | } |
1267 | ||
1268 | /** | |
324a56e1 TH |
1269 | * kernfs_remove - remove a kernfs_node recursively |
1270 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 1271 | * |
324a56e1 | 1272 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 1273 | */ |
324a56e1 | 1274 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1275 | { |
988cd7af TH |
1276 | mutex_lock(&kernfs_mutex); |
1277 | __kernfs_remove(kn); | |
1278 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
1279 | } |
1280 | ||
6b0afc2a TH |
1281 | /** |
1282 | * kernfs_break_active_protection - break out of active protection | |
1283 | * @kn: the self kernfs_node | |
1284 | * | |
1285 | * The caller must be running off of a kernfs operation which is invoked | |
1286 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
1287 | * this function must also be matched with an invocation of | |
1288 | * kernfs_unbreak_active_protection(). | |
1289 | * | |
1290 | * This function releases the active reference of @kn the caller is | |
1291 | * holding. Once this function is called, @kn may be removed at any point | |
1292 | * and the caller is solely responsible for ensuring that the objects it | |
1293 | * dereferences are accessible. | |
1294 | */ | |
1295 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
1296 | { | |
1297 | /* | |
1298 | * Take out ourself out of the active ref dependency chain. If | |
1299 | * we're called without an active ref, lockdep will complain. | |
1300 | */ | |
1301 | kernfs_put_active(kn); | |
1302 | } | |
1303 | ||
1304 | /** | |
1305 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
1306 | * @kn: the self kernfs_node | |
1307 | * | |
1308 | * If kernfs_break_active_protection() was called, this function must be | |
1309 | * invoked before finishing the kernfs operation. Note that while this | |
1310 | * function restores the active reference, it doesn't and can't actually | |
1311 | * restore the active protection - @kn may already or be in the process of | |
1312 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
1313 | * protection is irreversibly gone for the kernfs operation instance. | |
1314 | * | |
1315 | * While this function may be called at any point after | |
1316 | * kernfs_break_active_protection() is invoked, its most useful location | |
1317 | * would be right before the enclosing kernfs operation returns. | |
1318 | */ | |
1319 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
1320 | { | |
1321 | /* | |
1322 | * @kn->active could be in any state; however, the increment we do | |
1323 | * here will be undone as soon as the enclosing kernfs operation | |
1324 | * finishes and this temporary bump can't break anything. If @kn | |
1325 | * is alive, nothing changes. If @kn is being deactivated, the | |
1326 | * soon-to-follow put will either finish deactivation or restore | |
1327 | * deactivated state. If @kn is already removed, the temporary | |
1328 | * bump is guaranteed to be gone before @kn is released. | |
1329 | */ | |
1330 | atomic_inc(&kn->active); | |
1331 | if (kernfs_lockdep(kn)) | |
1332 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
1333 | } | |
1334 | ||
1335 | /** | |
1336 | * kernfs_remove_self - remove a kernfs_node from its own method | |
1337 | * @kn: the self kernfs_node to remove | |
1338 | * | |
1339 | * The caller must be running off of a kernfs operation which is invoked | |
1340 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
1341 | * implement a file operation which deletes itself. | |
1342 | * | |
1343 | * For example, the "delete" file for a sysfs device directory can be | |
1344 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
1345 | * itself. This function breaks the circular dependency of trying to | |
1346 | * deactivate self while holding an active ref itself. It isn't necessary | |
1347 | * to modify the usual removal path to use kernfs_remove_self(). The | |
1348 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
1349 | * before proceeding with the usual removal path. kernfs will ignore later | |
1350 | * kernfs_remove() on self. | |
1351 | * | |
1352 | * kernfs_remove_self() can be called multiple times concurrently on the | |
1353 | * same kernfs_node. Only the first one actually performs removal and | |
1354 | * returns %true. All others will wait until the kernfs operation which | |
1355 | * won self-removal finishes and return %false. Note that the losers wait | |
1356 | * for the completion of not only the winning kernfs_remove_self() but also | |
1357 | * the whole kernfs_ops which won the arbitration. This can be used to | |
1358 | * guarantee, for example, all concurrent writes to a "delete" file to | |
1359 | * finish only after the whole operation is complete. | |
1360 | */ | |
1361 | bool kernfs_remove_self(struct kernfs_node *kn) | |
1362 | { | |
1363 | bool ret; | |
1364 | ||
1365 | mutex_lock(&kernfs_mutex); | |
1366 | kernfs_break_active_protection(kn); | |
1367 | ||
1368 | /* | |
1369 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1370 | * the first one will actually perform removal. When the removal | |
1371 | * is complete, SUICIDED is set and the active ref is restored | |
1372 | * while holding kernfs_mutex. The ones which lost arbitration | |
1373 | * waits for SUICDED && drained which can happen only after the | |
1374 | * enclosing kernfs operation which executed the winning instance | |
1375 | * of kernfs_remove_self() finished. | |
1376 | */ | |
1377 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1378 | kn->flags |= KERNFS_SUICIDAL; | |
1379 | __kernfs_remove(kn); | |
1380 | kn->flags |= KERNFS_SUICIDED; | |
1381 | ret = true; | |
1382 | } else { | |
1383 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1384 | DEFINE_WAIT(wait); | |
1385 | ||
1386 | while (true) { | |
1387 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1388 | ||
1389 | if ((kn->flags & KERNFS_SUICIDED) && | |
1390 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1391 | break; | |
1392 | ||
1393 | mutex_unlock(&kernfs_mutex); | |
1394 | schedule(); | |
1395 | mutex_lock(&kernfs_mutex); | |
1396 | } | |
1397 | finish_wait(waitq, &wait); | |
1398 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1399 | ret = false; | |
1400 | } | |
1401 | ||
1402 | /* | |
1403 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1404 | * for SUICIDED && deactivated could finish prematurely. | |
1405 | */ | |
1406 | kernfs_unbreak_active_protection(kn); | |
1407 | ||
1408 | mutex_unlock(&kernfs_mutex); | |
1409 | return ret; | |
1410 | } | |
1411 | ||
fd7b9f7b | 1412 | /** |
324a56e1 TH |
1413 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1414 | * @parent: parent of the target | |
1415 | * @name: name of the kernfs_node to remove | |
1416 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1417 | * |
324a56e1 TH |
1418 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1419 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1420 | */ |
324a56e1 | 1421 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1422 | const void *ns) |
1423 | { | |
324a56e1 | 1424 | struct kernfs_node *kn; |
fd7b9f7b | 1425 | |
324a56e1 | 1426 | if (!parent) { |
c637b8ac | 1427 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1428 | name); |
1429 | return -ENOENT; | |
1430 | } | |
1431 | ||
988cd7af | 1432 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1433 | |
324a56e1 TH |
1434 | kn = kernfs_find_ns(parent, name, ns); |
1435 | if (kn) | |
988cd7af | 1436 | __kernfs_remove(kn); |
fd7b9f7b | 1437 | |
988cd7af | 1438 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1439 | |
324a56e1 | 1440 | if (kn) |
fd7b9f7b TH |
1441 | return 0; |
1442 | else | |
1443 | return -ENOENT; | |
1444 | } | |
1445 | ||
1446 | /** | |
1447 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1448 | * @kn: target node |
fd7b9f7b TH |
1449 | * @new_parent: new parent to put @sd under |
1450 | * @new_name: new name | |
1451 | * @new_ns: new namespace tag | |
1452 | */ | |
324a56e1 | 1453 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1454 | const char *new_name, const void *new_ns) |
1455 | { | |
3eef34ad TH |
1456 | struct kernfs_node *old_parent; |
1457 | const char *old_name = NULL; | |
fd7b9f7b TH |
1458 | int error; |
1459 | ||
3eef34ad TH |
1460 | /* can't move or rename root */ |
1461 | if (!kn->parent) | |
1462 | return -EINVAL; | |
1463 | ||
798c75a0 GKH |
1464 | mutex_lock(&kernfs_mutex); |
1465 | ||
d0ae3d43 | 1466 | error = -ENOENT; |
ea015218 EB |
1467 | if (!kernfs_active(kn) || !kernfs_active(new_parent) || |
1468 | (new_parent->flags & KERNFS_EMPTY_DIR)) | |
d0ae3d43 TH |
1469 | goto out; |
1470 | ||
fd7b9f7b | 1471 | error = 0; |
adc5e8b5 TH |
1472 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1473 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1474 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1475 | |
1476 | error = -EEXIST; | |
1477 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1478 | goto out; |
fd7b9f7b | 1479 | |
324a56e1 | 1480 | /* rename kernfs_node */ |
adc5e8b5 | 1481 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b | 1482 | error = -ENOMEM; |
75287a67 | 1483 | new_name = kstrdup_const(new_name, GFP_KERNEL); |
fd7b9f7b | 1484 | if (!new_name) |
798c75a0 | 1485 | goto out; |
3eef34ad TH |
1486 | } else { |
1487 | new_name = NULL; | |
fd7b9f7b TH |
1488 | } |
1489 | ||
1490 | /* | |
1491 | * Move to the appropriate place in the appropriate directories rbtree. | |
1492 | */ | |
c637b8ac | 1493 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1494 | kernfs_get(new_parent); |
3eef34ad TH |
1495 | |
1496 | /* rename_lock protects ->parent and ->name accessors */ | |
1497 | spin_lock_irq(&kernfs_rename_lock); | |
1498 | ||
1499 | old_parent = kn->parent; | |
adc5e8b5 | 1500 | kn->parent = new_parent; |
3eef34ad TH |
1501 | |
1502 | kn->ns = new_ns; | |
1503 | if (new_name) { | |
dfeb0750 | 1504 | old_name = kn->name; |
3eef34ad TH |
1505 | kn->name = new_name; |
1506 | } | |
1507 | ||
1508 | spin_unlock_irq(&kernfs_rename_lock); | |
1509 | ||
9561a896 | 1510 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
c637b8ac | 1511 | kernfs_link_sibling(kn); |
fd7b9f7b | 1512 | |
3eef34ad | 1513 | kernfs_put(old_parent); |
75287a67 | 1514 | kfree_const(old_name); |
3eef34ad | 1515 | |
fd7b9f7b | 1516 | error = 0; |
798c75a0 | 1517 | out: |
a797bfc3 | 1518 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1519 | return error; |
1520 | } | |
1521 | ||
fd7b9f7b | 1522 | /* Relationship between s_mode and the DT_xxx types */ |
324a56e1 | 1523 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1524 | { |
adc5e8b5 | 1525 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1526 | } |
1527 | ||
c637b8ac | 1528 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1529 | { |
1530 | kernfs_put(filp->private_data); | |
1531 | return 0; | |
1532 | } | |
1533 | ||
c637b8ac | 1534 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1535 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1536 | { |
1537 | if (pos) { | |
81c173cb | 1538 | int valid = kernfs_active(pos) && |
798c75a0 | 1539 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1540 | kernfs_put(pos); |
1541 | if (!valid) | |
1542 | pos = NULL; | |
1543 | } | |
1544 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1545 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1546 | while (node) { |
324a56e1 | 1547 | pos = rb_to_kn(node); |
fd7b9f7b | 1548 | |
adc5e8b5 | 1549 | if (hash < pos->hash) |
fd7b9f7b | 1550 | node = node->rb_left; |
adc5e8b5 | 1551 | else if (hash > pos->hash) |
fd7b9f7b TH |
1552 | node = node->rb_right; |
1553 | else | |
1554 | break; | |
1555 | } | |
1556 | } | |
b9c9dad0 TH |
1557 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1558 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1559 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1560 | if (!node) |
1561 | pos = NULL; | |
1562 | else | |
324a56e1 | 1563 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1564 | } |
1565 | return pos; | |
1566 | } | |
1567 | ||
c637b8ac | 1568 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1569 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1570 | { |
c637b8ac | 1571 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1572 | if (pos) { |
fd7b9f7b | 1573 | do { |
adc5e8b5 | 1574 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1575 | if (!node) |
1576 | pos = NULL; | |
1577 | else | |
324a56e1 | 1578 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1579 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1580 | } | |
fd7b9f7b TH |
1581 | return pos; |
1582 | } | |
1583 | ||
c637b8ac | 1584 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1585 | { |
1586 | struct dentry *dentry = file->f_path.dentry; | |
324a56e1 TH |
1587 | struct kernfs_node *parent = dentry->d_fsdata; |
1588 | struct kernfs_node *pos = file->private_data; | |
fd7b9f7b TH |
1589 | const void *ns = NULL; |
1590 | ||
1591 | if (!dir_emit_dots(file, ctx)) | |
1592 | return 0; | |
a797bfc3 | 1593 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1594 | |
324a56e1 | 1595 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1596 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1597 | |
c637b8ac | 1598 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1599 | pos; |
c637b8ac | 1600 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1601 | const char *name = pos->name; |
fd7b9f7b TH |
1602 | unsigned int type = dt_type(pos); |
1603 | int len = strlen(name); | |
adc5e8b5 | 1604 | ino_t ino = pos->ino; |
fd7b9f7b | 1605 | |
adc5e8b5 | 1606 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1607 | file->private_data = pos; |
1608 | kernfs_get(pos); | |
1609 | ||
a797bfc3 | 1610 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1611 | if (!dir_emit(ctx, name, len, ino, type)) |
1612 | return 0; | |
a797bfc3 | 1613 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1614 | } |
a797bfc3 | 1615 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1616 | file->private_data = NULL; |
1617 | ctx->pos = INT_MAX; | |
1618 | return 0; | |
1619 | } | |
1620 | ||
a797bfc3 | 1621 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1622 | .read = generic_read_dir, |
8cb0d2c1 | 1623 | .iterate_shared = kernfs_fop_readdir, |
c637b8ac | 1624 | .release = kernfs_dir_fop_release, |
8cb0d2c1 | 1625 | .llseek = generic_file_llseek, |
fd7b9f7b | 1626 | }; |