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 */ | |
7d35079f | 24 | static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */ |
fd7b9f7b | 25 | |
adc5e8b5 | 26 | #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb) |
fd7b9f7b | 27 | |
81c173cb TH |
28 | static bool kernfs_active(struct kernfs_node *kn) |
29 | { | |
30 | lockdep_assert_held(&kernfs_mutex); | |
31 | return atomic_read(&kn->active) >= 0; | |
32 | } | |
33 | ||
182fd64b TH |
34 | static bool kernfs_lockdep(struct kernfs_node *kn) |
35 | { | |
36 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
37 | return kn->flags & KERNFS_LOCKDEP; | |
38 | #else | |
39 | return false; | |
40 | #endif | |
41 | } | |
42 | ||
3eef34ad TH |
43 | static int kernfs_name_locked(struct kernfs_node *kn, char *buf, size_t buflen) |
44 | { | |
17627157 KK |
45 | if (!kn) |
46 | return strlcpy(buf, "(null)", buflen); | |
47 | ||
3eef34ad TH |
48 | return strlcpy(buf, kn->parent ? kn->name : "/", buflen); |
49 | } | |
50 | ||
9f6df573 AK |
51 | /* kernfs_node_depth - compute depth from @from to @to */ |
52 | static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to) | |
3eef34ad | 53 | { |
9f6df573 | 54 | size_t depth = 0; |
3eef34ad | 55 | |
9f6df573 AK |
56 | while (to->parent && to != from) { |
57 | depth++; | |
58 | to = to->parent; | |
59 | } | |
60 | return depth; | |
61 | } | |
3eef34ad | 62 | |
9f6df573 AK |
63 | static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a, |
64 | struct kernfs_node *b) | |
65 | { | |
66 | size_t da, db; | |
67 | struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b); | |
68 | ||
69 | if (ra != rb) | |
70 | return NULL; | |
71 | ||
72 | da = kernfs_depth(ra->kn, a); | |
73 | db = kernfs_depth(rb->kn, b); | |
74 | ||
75 | while (da > db) { | |
76 | a = a->parent; | |
77 | da--; | |
78 | } | |
79 | while (db > da) { | |
80 | b = b->parent; | |
81 | db--; | |
82 | } | |
83 | ||
84 | /* worst case b and a will be the same at root */ | |
85 | while (b != a) { | |
86 | b = b->parent; | |
87 | a = a->parent; | |
88 | } | |
89 | ||
90 | return a; | |
91 | } | |
92 | ||
93 | /** | |
94 | * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to, | |
95 | * where kn_from is treated as root of the path. | |
96 | * @kn_from: kernfs node which should be treated as root for the path | |
97 | * @kn_to: kernfs node to which path is needed | |
98 | * @buf: buffer to copy the path into | |
99 | * @buflen: size of @buf | |
100 | * | |
101 | * We need to handle couple of scenarios here: | |
102 | * [1] when @kn_from is an ancestor of @kn_to at some level | |
103 | * kn_from: /n1/n2/n3 | |
104 | * kn_to: /n1/n2/n3/n4/n5 | |
105 | * result: /n4/n5 | |
106 | * | |
107 | * [2] when @kn_from is on a different hierarchy and we need to find common | |
108 | * ancestor between @kn_from and @kn_to. | |
109 | * kn_from: /n1/n2/n3/n4 | |
110 | * kn_to: /n1/n2/n5 | |
111 | * result: /../../n5 | |
112 | * OR | |
113 | * kn_from: /n1/n2/n3/n4/n5 [depth=5] | |
114 | * kn_to: /n1/n2/n3 [depth=3] | |
115 | * result: /../.. | |
116 | * | |
17627157 KK |
117 | * [3] when @kn_to is NULL result will be "(null)" |
118 | * | |
3abb1d90 TH |
119 | * Returns the length of the full path. If the full length is equal to or |
120 | * greater than @buflen, @buf contains the truncated path with the trailing | |
121 | * '\0'. On error, -errno is returned. | |
9f6df573 AK |
122 | */ |
123 | static int kernfs_path_from_node_locked(struct kernfs_node *kn_to, | |
124 | struct kernfs_node *kn_from, | |
125 | char *buf, size_t buflen) | |
126 | { | |
127 | struct kernfs_node *kn, *common; | |
128 | const char parent_str[] = "/.."; | |
3abb1d90 TH |
129 | size_t depth_from, depth_to, len = 0; |
130 | int i, j; | |
9f6df573 | 131 | |
17627157 KK |
132 | if (!kn_to) |
133 | return strlcpy(buf, "(null)", buflen); | |
134 | ||
9f6df573 AK |
135 | if (!kn_from) |
136 | kn_from = kernfs_root(kn_to)->kn; | |
137 | ||
138 | if (kn_from == kn_to) | |
139 | return strlcpy(buf, "/", buflen); | |
140 | ||
141 | common = kernfs_common_ancestor(kn_from, kn_to); | |
142 | if (WARN_ON(!common)) | |
3abb1d90 | 143 | return -EINVAL; |
9f6df573 AK |
144 | |
145 | depth_to = kernfs_depth(common, kn_to); | |
146 | depth_from = kernfs_depth(common, kn_from); | |
147 | ||
148 | if (buf) | |
149 | buf[0] = '\0'; | |
150 | ||
151 | for (i = 0; i < depth_from; i++) | |
152 | len += strlcpy(buf + len, parent_str, | |
153 | len < buflen ? buflen - len : 0); | |
154 | ||
155 | /* Calculate how many bytes we need for the rest */ | |
3abb1d90 TH |
156 | for (i = depth_to - 1; i >= 0; i--) { |
157 | for (kn = kn_to, j = 0; j < i; j++) | |
158 | kn = kn->parent; | |
159 | len += strlcpy(buf + len, "/", | |
160 | len < buflen ? buflen - len : 0); | |
161 | len += strlcpy(buf + len, kn->name, | |
162 | len < buflen ? buflen - len : 0); | |
9f6df573 | 163 | } |
3eef34ad | 164 | |
3abb1d90 | 165 | return len; |
3eef34ad TH |
166 | } |
167 | ||
168 | /** | |
169 | * kernfs_name - obtain the name of a given node | |
170 | * @kn: kernfs_node of interest | |
171 | * @buf: buffer to copy @kn's name into | |
172 | * @buflen: size of @buf | |
173 | * | |
174 | * Copies the name of @kn into @buf of @buflen bytes. The behavior is | |
175 | * similar to strlcpy(). It returns the length of @kn's name and if @buf | |
176 | * isn't long enough, it's filled upto @buflen-1 and nul terminated. | |
177 | * | |
17627157 KK |
178 | * Fills buffer with "(null)" if @kn is NULL. |
179 | * | |
3eef34ad TH |
180 | * This function can be called from any context. |
181 | */ | |
182 | int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen) | |
183 | { | |
184 | unsigned long flags; | |
185 | int ret; | |
186 | ||
187 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
188 | ret = kernfs_name_locked(kn, buf, buflen); | |
189 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
190 | return ret; | |
191 | } | |
192 | ||
9f6df573 AK |
193 | /** |
194 | * kernfs_path_from_node - build path of node @to relative to @from. | |
195 | * @from: parent kernfs_node relative to which we need to build the path | |
196 | * @to: kernfs_node of interest | |
197 | * @buf: buffer to copy @to's path into | |
198 | * @buflen: size of @buf | |
199 | * | |
200 | * Builds @to's path relative to @from in @buf. @from and @to must | |
201 | * be on the same kernfs-root. If @from is not parent of @to, then a relative | |
202 | * path (which includes '..'s) as needed to reach from @from to @to is | |
203 | * returned. | |
204 | * | |
3abb1d90 TH |
205 | * Returns the length of the full path. If the full length is equal to or |
206 | * greater than @buflen, @buf contains the truncated path with the trailing | |
207 | * '\0'. On error, -errno is returned. | |
9f6df573 AK |
208 | */ |
209 | int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from, | |
210 | char *buf, size_t buflen) | |
211 | { | |
212 | unsigned long flags; | |
213 | int ret; | |
214 | ||
215 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
216 | ret = kernfs_path_from_node_locked(to, from, buf, buflen); | |
217 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
218 | return ret; | |
219 | } | |
220 | EXPORT_SYMBOL_GPL(kernfs_path_from_node); | |
221 | ||
3eef34ad TH |
222 | /** |
223 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node | |
224 | * @kn: kernfs_node of interest | |
225 | * | |
226 | * This function can be called from any context. | |
227 | */ | |
228 | void pr_cont_kernfs_name(struct kernfs_node *kn) | |
229 | { | |
230 | unsigned long flags; | |
231 | ||
232 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
233 | ||
234 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); | |
235 | pr_cont("%s", kernfs_pr_cont_buf); | |
236 | ||
237 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
238 | } | |
239 | ||
240 | /** | |
241 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node | |
242 | * @kn: kernfs_node of interest | |
243 | * | |
244 | * This function can be called from any context. | |
245 | */ | |
246 | void pr_cont_kernfs_path(struct kernfs_node *kn) | |
247 | { | |
248 | unsigned long flags; | |
9f6df573 | 249 | int sz; |
3eef34ad TH |
250 | |
251 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
252 | ||
9f6df573 AK |
253 | sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf, |
254 | sizeof(kernfs_pr_cont_buf)); | |
255 | if (sz < 0) { | |
256 | pr_cont("(error)"); | |
257 | goto out; | |
258 | } | |
259 | ||
260 | if (sz >= sizeof(kernfs_pr_cont_buf)) { | |
261 | pr_cont("(name too long)"); | |
262 | goto out; | |
263 | } | |
264 | ||
265 | pr_cont("%s", kernfs_pr_cont_buf); | |
3eef34ad | 266 | |
9f6df573 | 267 | out: |
3eef34ad TH |
268 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
269 | } | |
270 | ||
271 | /** | |
272 | * kernfs_get_parent - determine the parent node and pin it | |
273 | * @kn: kernfs_node of interest | |
274 | * | |
275 | * Determines @kn's parent, pins and returns it. This function can be | |
276 | * called from any context. | |
277 | */ | |
278 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) | |
279 | { | |
280 | struct kernfs_node *parent; | |
281 | unsigned long flags; | |
282 | ||
283 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
284 | parent = kn->parent; | |
285 | kernfs_get(parent); | |
286 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
287 | ||
288 | return parent; | |
289 | } | |
290 | ||
fd7b9f7b | 291 | /** |
c637b8ac | 292 | * kernfs_name_hash |
fd7b9f7b TH |
293 | * @name: Null terminated string to hash |
294 | * @ns: Namespace tag to hash | |
295 | * | |
296 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
297 | */ | |
c637b8ac | 298 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b | 299 | { |
8387ff25 | 300 | unsigned long hash = init_name_hash(ns); |
fd7b9f7b TH |
301 | unsigned int len = strlen(name); |
302 | while (len--) | |
303 | hash = partial_name_hash(*name++, hash); | |
8387ff25 | 304 | hash = end_name_hash(hash); |
fd7b9f7b TH |
305 | hash &= 0x7fffffffU; |
306 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
88391d49 | 307 | if (hash < 2) |
fd7b9f7b TH |
308 | hash += 2; |
309 | if (hash >= INT_MAX) | |
310 | hash = INT_MAX - 1; | |
311 | return hash; | |
312 | } | |
313 | ||
c637b8ac TH |
314 | static int kernfs_name_compare(unsigned int hash, const char *name, |
315 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 316 | { |
72392ed0 RV |
317 | if (hash < kn->hash) |
318 | return -1; | |
319 | if (hash > kn->hash) | |
320 | return 1; | |
321 | if (ns < kn->ns) | |
322 | return -1; | |
323 | if (ns > kn->ns) | |
324 | return 1; | |
adc5e8b5 | 325 | return strcmp(name, kn->name); |
fd7b9f7b TH |
326 | } |
327 | ||
c637b8ac TH |
328 | static int kernfs_sd_compare(const struct kernfs_node *left, |
329 | const struct kernfs_node *right) | |
fd7b9f7b | 330 | { |
c637b8ac | 331 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
332 | } |
333 | ||
334 | /** | |
c637b8ac | 335 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 336 | * @kn: kernfs_node of interest |
fd7b9f7b | 337 | * |
324a56e1 | 338 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 339 | * @kn->parent->dir.children. |
fd7b9f7b TH |
340 | * |
341 | * Locking: | |
a797bfc3 | 342 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
343 | * |
344 | * RETURNS: | |
345 | * 0 on susccess -EEXIST on failure. | |
346 | */ | |
c637b8ac | 347 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 348 | { |
adc5e8b5 | 349 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
350 | struct rb_node *parent = NULL; |
351 | ||
fd7b9f7b | 352 | while (*node) { |
324a56e1 | 353 | struct kernfs_node *pos; |
fd7b9f7b TH |
354 | int result; |
355 | ||
324a56e1 | 356 | pos = rb_to_kn(*node); |
fd7b9f7b | 357 | parent = *node; |
c637b8ac | 358 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 359 | if (result < 0) |
adc5e8b5 | 360 | node = &pos->rb.rb_left; |
fd7b9f7b | 361 | else if (result > 0) |
adc5e8b5 | 362 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
363 | else |
364 | return -EEXIST; | |
365 | } | |
c1befb88 | 366 | |
fd7b9f7b | 367 | /* add new node and rebalance the tree */ |
adc5e8b5 TH |
368 | rb_link_node(&kn->rb, parent, node); |
369 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
c1befb88 JZ |
370 | |
371 | /* successfully added, account subdir number */ | |
372 | if (kernfs_type(kn) == KERNFS_DIR) | |
373 | kn->parent->dir.subdirs++; | |
374 | ||
fd7b9f7b TH |
375 | return 0; |
376 | } | |
377 | ||
378 | /** | |
c637b8ac | 379 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 380 | * @kn: kernfs_node of interest |
fd7b9f7b | 381 | * |
35beab06 TH |
382 | * Try to unlink @kn from its sibling rbtree which starts from |
383 | * kn->parent->dir.children. Returns %true if @kn was actually | |
384 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
385 | * |
386 | * Locking: | |
a797bfc3 | 387 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 388 | */ |
35beab06 | 389 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 390 | { |
35beab06 TH |
391 | if (RB_EMPTY_NODE(&kn->rb)) |
392 | return false; | |
393 | ||
df23fc39 | 394 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 395 | kn->parent->dir.subdirs--; |
fd7b9f7b | 396 | |
adc5e8b5 | 397 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
398 | RB_CLEAR_NODE(&kn->rb); |
399 | return true; | |
fd7b9f7b TH |
400 | } |
401 | ||
402 | /** | |
c637b8ac | 403 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 404 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 405 | * |
324a56e1 | 406 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
407 | * is NULL. |
408 | * | |
409 | * RETURNS: | |
324a56e1 | 410 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 411 | */ |
c637b8ac | 412 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 413 | { |
324a56e1 | 414 | if (unlikely(!kn)) |
fd7b9f7b TH |
415 | return NULL; |
416 | ||
f4b3e631 GKH |
417 | if (!atomic_inc_unless_negative(&kn->active)) |
418 | return NULL; | |
895a068a | 419 | |
182fd64b | 420 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
421 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
422 | return kn; | |
fd7b9f7b TH |
423 | } |
424 | ||
425 | /** | |
c637b8ac | 426 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 427 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 428 | * |
324a56e1 | 429 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
430 | * is NULL. |
431 | */ | |
c637b8ac | 432 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b | 433 | { |
abd54f02 | 434 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b TH |
435 | int v; |
436 | ||
324a56e1 | 437 | if (unlikely(!kn)) |
fd7b9f7b TH |
438 | return; |
439 | ||
182fd64b | 440 | if (kernfs_lockdep(kn)) |
324a56e1 | 441 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
adc5e8b5 | 442 | v = atomic_dec_return(&kn->active); |
df23fc39 | 443 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
444 | return; |
445 | ||
abd54f02 | 446 | wake_up_all(&root->deactivate_waitq); |
fd7b9f7b TH |
447 | } |
448 | ||
449 | /** | |
81c173cb TH |
450 | * kernfs_drain - drain kernfs_node |
451 | * @kn: kernfs_node to drain | |
fd7b9f7b | 452 | * |
81c173cb TH |
453 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
454 | * removers may invoke this function concurrently on @kn and all will | |
455 | * return after draining is complete. | |
fd7b9f7b | 456 | */ |
81c173cb | 457 | static void kernfs_drain(struct kernfs_node *kn) |
35beab06 | 458 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
fd7b9f7b | 459 | { |
abd54f02 | 460 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b | 461 | |
35beab06 | 462 | lockdep_assert_held(&kernfs_mutex); |
81c173cb | 463 | WARN_ON_ONCE(kernfs_active(kn)); |
ea1c472d | 464 | |
35beab06 | 465 | mutex_unlock(&kernfs_mutex); |
abd54f02 | 466 | |
182fd64b | 467 | if (kernfs_lockdep(kn)) { |
35beab06 TH |
468 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
469 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) | |
470 | lock_contended(&kn->dep_map, _RET_IP_); | |
471 | } | |
abd54f02 | 472 | |
35beab06 | 473 | /* but everyone should wait for draining */ |
abd54f02 TH |
474 | wait_event(root->deactivate_waitq, |
475 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); | |
fd7b9f7b | 476 | |
182fd64b | 477 | if (kernfs_lockdep(kn)) { |
a6607930 TH |
478 | lock_acquired(&kn->dep_map, _RET_IP_); |
479 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | |
480 | } | |
35beab06 | 481 | |
0e67db2f | 482 | kernfs_drain_open_files(kn); |
ccf02aaf | 483 | |
35beab06 | 484 | mutex_lock(&kernfs_mutex); |
fd7b9f7b TH |
485 | } |
486 | ||
fd7b9f7b | 487 | /** |
324a56e1 TH |
488 | * kernfs_get - get a reference count on a kernfs_node |
489 | * @kn: the target kernfs_node | |
fd7b9f7b | 490 | */ |
324a56e1 | 491 | void kernfs_get(struct kernfs_node *kn) |
fd7b9f7b | 492 | { |
324a56e1 | 493 | if (kn) { |
adc5e8b5 TH |
494 | WARN_ON(!atomic_read(&kn->count)); |
495 | atomic_inc(&kn->count); | |
fd7b9f7b TH |
496 | } |
497 | } | |
498 | EXPORT_SYMBOL_GPL(kernfs_get); | |
499 | ||
500 | /** | |
324a56e1 TH |
501 | * kernfs_put - put a reference count on a kernfs_node |
502 | * @kn: the target kernfs_node | |
fd7b9f7b | 503 | * |
324a56e1 | 504 | * Put a reference count of @kn and destroy it if it reached zero. |
fd7b9f7b | 505 | */ |
324a56e1 | 506 | void kernfs_put(struct kernfs_node *kn) |
fd7b9f7b | 507 | { |
324a56e1 | 508 | struct kernfs_node *parent; |
ba7443bc | 509 | struct kernfs_root *root; |
fd7b9f7b | 510 | |
ba16b284 SL |
511 | /* |
512 | * kernfs_node is freed with ->count 0, kernfs_find_and_get_node_by_ino | |
513 | * depends on this to filter reused stale node | |
514 | */ | |
adc5e8b5 | 515 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 516 | return; |
324a56e1 | 517 | root = kernfs_root(kn); |
fd7b9f7b | 518 | repeat: |
81c173cb TH |
519 | /* |
520 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 521 | * kn->parent won't change beneath us. |
fd7b9f7b | 522 | */ |
adc5e8b5 | 523 | parent = kn->parent; |
fd7b9f7b | 524 | |
81c173cb TH |
525 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
526 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
527 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 528 | |
df23fc39 | 529 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 530 | kernfs_put(kn->symlink.target_kn); |
dfeb0750 TH |
531 | |
532 | kfree_const(kn->name); | |
533 | ||
adc5e8b5 TH |
534 | if (kn->iattr) { |
535 | if (kn->iattr->ia_secdata) | |
536 | security_release_secctx(kn->iattr->ia_secdata, | |
537 | kn->iattr->ia_secdata_len); | |
538 | simple_xattrs_free(&kn->iattr->xattrs); | |
2322392b | 539 | } |
adc5e8b5 | 540 | kfree(kn->iattr); |
7d35079f | 541 | spin_lock(&kernfs_idr_lock); |
c53cd490 | 542 | idr_remove(&root->ino_idr, kn->id.ino); |
7d35079f | 543 | spin_unlock(&kernfs_idr_lock); |
a797bfc3 | 544 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 545 | |
324a56e1 TH |
546 | kn = parent; |
547 | if (kn) { | |
adc5e8b5 | 548 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
549 | goto repeat; |
550 | } else { | |
324a56e1 | 551 | /* just released the root kn, free @root too */ |
7d35079f | 552 | idr_destroy(&root->ino_idr); |
ba7443bc TH |
553 | kfree(root); |
554 | } | |
fd7b9f7b TH |
555 | } |
556 | EXPORT_SYMBOL_GPL(kernfs_put); | |
557 | ||
c637b8ac | 558 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 559 | { |
324a56e1 | 560 | struct kernfs_node *kn; |
fd7b9f7b TH |
561 | |
562 | if (flags & LOOKUP_RCU) | |
563 | return -ECHILD; | |
564 | ||
19bbb926 | 565 | /* Always perform fresh lookup for negatives */ |
2b0143b5 | 566 | if (d_really_is_negative(dentry)) |
19bbb926 TH |
567 | goto out_bad_unlocked; |
568 | ||
319ba91d | 569 | kn = kernfs_dentry_node(dentry); |
a797bfc3 | 570 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 571 | |
81c173cb TH |
572 | /* The kernfs node has been deactivated */ |
573 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
574 | goto out_bad; |
575 | ||
c637b8ac | 576 | /* The kernfs node has been moved? */ |
319ba91d | 577 | if (kernfs_dentry_node(dentry->d_parent) != kn->parent) |
fd7b9f7b TH |
578 | goto out_bad; |
579 | ||
c637b8ac | 580 | /* The kernfs node has been renamed */ |
adc5e8b5 | 581 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
582 | goto out_bad; |
583 | ||
c637b8ac | 584 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 585 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 586 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
587 | goto out_bad; |
588 | ||
a797bfc3 | 589 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
590 | return 1; |
591 | out_bad: | |
a797bfc3 | 592 | mutex_unlock(&kernfs_mutex); |
19bbb926 | 593 | out_bad_unlocked: |
fd7b9f7b TH |
594 | return 0; |
595 | } | |
596 | ||
a797bfc3 | 597 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 598 | .d_revalidate = kernfs_dop_revalidate, |
fd7b9f7b TH |
599 | }; |
600 | ||
0c23b225 TH |
601 | /** |
602 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry | |
603 | * @dentry: the dentry in question | |
604 | * | |
605 | * Return the kernfs_node associated with @dentry. If @dentry is not a | |
606 | * kernfs one, %NULL is returned. | |
607 | * | |
608 | * While the returned kernfs_node will stay accessible as long as @dentry | |
609 | * is accessible, the returned node can be in any state and the caller is | |
610 | * fully responsible for determining what's accessible. | |
611 | */ | |
612 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) | |
613 | { | |
319ba91d SL |
614 | if (dentry->d_sb->s_op == &kernfs_sops && |
615 | !d_really_is_negative(dentry)) | |
616 | return kernfs_dentry_node(dentry); | |
0c23b225 TH |
617 | return NULL; |
618 | } | |
619 | ||
db4aad20 TH |
620 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
621 | const char *name, umode_t mode, | |
488dee96 | 622 | kuid_t uid, kgid_t gid, |
db4aad20 | 623 | unsigned flags) |
fd7b9f7b | 624 | { |
324a56e1 | 625 | struct kernfs_node *kn; |
4a3ef68a SL |
626 | u32 gen; |
627 | int cursor; | |
bc755553 | 628 | int ret; |
fd7b9f7b | 629 | |
dfeb0750 TH |
630 | name = kstrdup_const(name, GFP_KERNEL); |
631 | if (!name) | |
632 | return NULL; | |
fd7b9f7b | 633 | |
a797bfc3 | 634 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 635 | if (!kn) |
fd7b9f7b TH |
636 | goto err_out1; |
637 | ||
7d35079f SL |
638 | idr_preload(GFP_KERNEL); |
639 | spin_lock(&kernfs_idr_lock); | |
4a3ef68a SL |
640 | cursor = idr_get_cursor(&root->ino_idr); |
641 | ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC); | |
642 | if (ret >= 0 && ret < cursor) | |
643 | root->next_generation++; | |
644 | gen = root->next_generation; | |
7d35079f SL |
645 | spin_unlock(&kernfs_idr_lock); |
646 | idr_preload_end(); | |
bc755553 | 647 | if (ret < 0) |
fd7b9f7b | 648 | goto err_out2; |
c53cd490 SL |
649 | kn->id.ino = ret; |
650 | kn->id.generation = gen; | |
fd7b9f7b | 651 | |
ba16b284 SL |
652 | /* |
653 | * set ino first. This barrier is paired with atomic_inc_not_zero in | |
654 | * kernfs_find_and_get_node_by_ino | |
655 | */ | |
656 | smp_mb__before_atomic(); | |
adc5e8b5 | 657 | atomic_set(&kn->count, 1); |
81c173cb | 658 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 659 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 660 | |
adc5e8b5 TH |
661 | kn->name = name; |
662 | kn->mode = mode; | |
81c173cb | 663 | kn->flags = flags; |
fd7b9f7b | 664 | |
488dee96 DT |
665 | if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) { |
666 | struct iattr iattr = { | |
667 | .ia_valid = ATTR_UID | ATTR_GID, | |
668 | .ia_uid = uid, | |
669 | .ia_gid = gid, | |
670 | }; | |
671 | ||
672 | ret = __kernfs_setattr(kn, &iattr); | |
673 | if (ret < 0) | |
674 | goto err_out3; | |
675 | } | |
676 | ||
324a56e1 | 677 | return kn; |
fd7b9f7b | 678 | |
488dee96 DT |
679 | err_out3: |
680 | idr_remove(&root->ino_idr, kn->id.ino); | |
fd7b9f7b | 681 | err_out2: |
a797bfc3 | 682 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 683 | err_out1: |
dfeb0750 | 684 | kfree_const(name); |
fd7b9f7b TH |
685 | return NULL; |
686 | } | |
687 | ||
db4aad20 TH |
688 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
689 | const char *name, umode_t mode, | |
488dee96 | 690 | kuid_t uid, kgid_t gid, |
db4aad20 TH |
691 | unsigned flags) |
692 | { | |
693 | struct kernfs_node *kn; | |
694 | ||
488dee96 DT |
695 | kn = __kernfs_new_node(kernfs_root(parent), |
696 | name, mode, uid, gid, flags); | |
db4aad20 TH |
697 | if (kn) { |
698 | kernfs_get(parent); | |
699 | kn->parent = parent; | |
700 | } | |
701 | return kn; | |
702 | } | |
703 | ||
ba16b284 SL |
704 | /* |
705 | * kernfs_find_and_get_node_by_ino - get kernfs_node from inode number | |
706 | * @root: the kernfs root | |
707 | * @ino: inode number | |
708 | * | |
709 | * RETURNS: | |
710 | * NULL on failure. Return a kernfs node with reference counter incremented | |
711 | */ | |
712 | struct kernfs_node *kernfs_find_and_get_node_by_ino(struct kernfs_root *root, | |
713 | unsigned int ino) | |
714 | { | |
715 | struct kernfs_node *kn; | |
716 | ||
717 | rcu_read_lock(); | |
718 | kn = idr_find(&root->ino_idr, ino); | |
719 | if (!kn) | |
720 | goto out; | |
721 | ||
722 | /* | |
723 | * Since kernfs_node is freed in RCU, it's possible an old node for ino | |
724 | * is freed, but reused before RCU grace period. But a freed node (see | |
725 | * kernfs_put) or an incompletedly initialized node (see | |
726 | * __kernfs_new_node) should have 'count' 0. We can use this fact to | |
727 | * filter out such node. | |
728 | */ | |
729 | if (!atomic_inc_not_zero(&kn->count)) { | |
730 | kn = NULL; | |
731 | goto out; | |
732 | } | |
733 | ||
734 | /* | |
735 | * The node could be a new node or a reused node. If it's a new node, | |
736 | * we are ok. If it's reused because of RCU (because of | |
737 | * SLAB_TYPESAFE_BY_RCU), the __kernfs_new_node always sets its 'ino' | |
738 | * before 'count'. So if 'count' is uptodate, 'ino' should be uptodate, | |
739 | * hence we can use 'ino' to filter stale node. | |
740 | */ | |
c53cd490 | 741 | if (kn->id.ino != ino) |
ba16b284 SL |
742 | goto out; |
743 | rcu_read_unlock(); | |
744 | ||
745 | return kn; | |
746 | out: | |
747 | rcu_read_unlock(); | |
748 | kernfs_put(kn); | |
749 | return NULL; | |
750 | } | |
751 | ||
fd7b9f7b | 752 | /** |
c637b8ac | 753 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 754 | * @kn: kernfs_node to be added |
fd7b9f7b | 755 | * |
db4aad20 TH |
756 | * The caller must already have initialized @kn->parent. This |
757 | * function increments nlink of the parent's inode if @kn is a | |
758 | * directory and link into the children list of the parent. | |
fd7b9f7b | 759 | * |
fd7b9f7b TH |
760 | * RETURNS: |
761 | * 0 on success, -EEXIST if entry with the given name already | |
762 | * exists. | |
763 | */ | |
988cd7af | 764 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 765 | { |
db4aad20 | 766 | struct kernfs_node *parent = kn->parent; |
c525aadd | 767 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 768 | bool has_ns; |
fd7b9f7b TH |
769 | int ret; |
770 | ||
988cd7af TH |
771 | mutex_lock(&kernfs_mutex); |
772 | ||
773 | ret = -EINVAL; | |
774 | has_ns = kernfs_ns_enabled(parent); | |
775 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
776 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
777 | goto out_unlock; | |
fd7b9f7b | 778 | |
df23fc39 | 779 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 780 | goto out_unlock; |
fd7b9f7b | 781 | |
988cd7af | 782 | ret = -ENOENT; |
ea015218 EB |
783 | if (parent->flags & KERNFS_EMPTY_DIR) |
784 | goto out_unlock; | |
785 | ||
d35258ef | 786 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 787 | goto out_unlock; |
798c75a0 | 788 | |
c637b8ac | 789 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 790 | |
c637b8ac | 791 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 792 | if (ret) |
988cd7af | 793 | goto out_unlock; |
fd7b9f7b TH |
794 | |
795 | /* Update timestamps on the parent */ | |
adc5e8b5 | 796 | ps_iattr = parent->iattr; |
fd7b9f7b TH |
797 | if (ps_iattr) { |
798 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
95582b00 | 799 | ktime_get_real_ts64(&ps_iattrs->ia_ctime); |
3a3a5fec | 800 | ps_iattrs->ia_mtime = ps_iattrs->ia_ctime; |
fd7b9f7b TH |
801 | } |
802 | ||
d35258ef TH |
803 | mutex_unlock(&kernfs_mutex); |
804 | ||
805 | /* | |
806 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
807 | * If not activated here, the kernfs user is responsible for | |
808 | * activating the node with kernfs_activate(). A node which hasn't | |
809 | * been activated is not visible to userland and its removal won't | |
810 | * trigger deactivation. | |
811 | */ | |
812 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
813 | kernfs_activate(kn); | |
814 | return 0; | |
815 | ||
988cd7af | 816 | out_unlock: |
a797bfc3 | 817 | mutex_unlock(&kernfs_mutex); |
988cd7af | 818 | return ret; |
fd7b9f7b TH |
819 | } |
820 | ||
821 | /** | |
324a56e1 TH |
822 | * kernfs_find_ns - find kernfs_node with the given name |
823 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
824 | * @name: name to look for |
825 | * @ns: the namespace tag to use | |
826 | * | |
324a56e1 TH |
827 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
828 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 829 | */ |
324a56e1 TH |
830 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
831 | const unsigned char *name, | |
832 | const void *ns) | |
fd7b9f7b | 833 | { |
adc5e8b5 | 834 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 835 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
836 | unsigned int hash; |
837 | ||
a797bfc3 | 838 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
839 | |
840 | if (has_ns != (bool)ns) { | |
c637b8ac | 841 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 842 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
843 | return NULL; |
844 | } | |
845 | ||
c637b8ac | 846 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 847 | while (node) { |
324a56e1 | 848 | struct kernfs_node *kn; |
fd7b9f7b TH |
849 | int result; |
850 | ||
324a56e1 | 851 | kn = rb_to_kn(node); |
c637b8ac | 852 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
853 | if (result < 0) |
854 | node = node->rb_left; | |
855 | else if (result > 0) | |
856 | node = node->rb_right; | |
857 | else | |
324a56e1 | 858 | return kn; |
fd7b9f7b TH |
859 | } |
860 | return NULL; | |
861 | } | |
862 | ||
bd96f76a TH |
863 | static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent, |
864 | const unsigned char *path, | |
865 | const void *ns) | |
866 | { | |
e56ed358 TH |
867 | size_t len; |
868 | char *p, *name; | |
bd96f76a TH |
869 | |
870 | lockdep_assert_held(&kernfs_mutex); | |
871 | ||
e56ed358 TH |
872 | /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */ |
873 | spin_lock_irq(&kernfs_rename_lock); | |
874 | ||
875 | len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf)); | |
876 | ||
877 | if (len >= sizeof(kernfs_pr_cont_buf)) { | |
878 | spin_unlock_irq(&kernfs_rename_lock); | |
bd96f76a | 879 | return NULL; |
e56ed358 TH |
880 | } |
881 | ||
882 | p = kernfs_pr_cont_buf; | |
bd96f76a TH |
883 | |
884 | while ((name = strsep(&p, "/")) && parent) { | |
885 | if (*name == '\0') | |
886 | continue; | |
887 | parent = kernfs_find_ns(parent, name, ns); | |
888 | } | |
889 | ||
e56ed358 TH |
890 | spin_unlock_irq(&kernfs_rename_lock); |
891 | ||
bd96f76a TH |
892 | return parent; |
893 | } | |
894 | ||
fd7b9f7b | 895 | /** |
324a56e1 TH |
896 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
897 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
898 | * @name: name to look for |
899 | * @ns: the namespace tag to use | |
900 | * | |
324a56e1 | 901 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 902 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 903 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 904 | */ |
324a56e1 TH |
905 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
906 | const char *name, const void *ns) | |
fd7b9f7b | 907 | { |
324a56e1 | 908 | struct kernfs_node *kn; |
fd7b9f7b | 909 | |
a797bfc3 | 910 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
911 | kn = kernfs_find_ns(parent, name, ns); |
912 | kernfs_get(kn); | |
a797bfc3 | 913 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 914 | |
324a56e1 | 915 | return kn; |
fd7b9f7b TH |
916 | } |
917 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
918 | ||
bd96f76a TH |
919 | /** |
920 | * kernfs_walk_and_get_ns - find and get kernfs_node with the given path | |
921 | * @parent: kernfs_node to search under | |
922 | * @path: path to look for | |
923 | * @ns: the namespace tag to use | |
924 | * | |
925 | * Look for kernfs_node with path @path under @parent and get a reference | |
926 | * if found. This function may sleep and returns pointer to the found | |
927 | * kernfs_node on success, %NULL on failure. | |
928 | */ | |
929 | struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent, | |
930 | const char *path, const void *ns) | |
931 | { | |
932 | struct kernfs_node *kn; | |
933 | ||
934 | mutex_lock(&kernfs_mutex); | |
935 | kn = kernfs_walk_ns(parent, path, ns); | |
936 | kernfs_get(kn); | |
937 | mutex_unlock(&kernfs_mutex); | |
938 | ||
939 | return kn; | |
940 | } | |
941 | ||
ba7443bc TH |
942 | /** |
943 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 944 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 945 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
946 | * @priv: opaque data associated with the new directory |
947 | * | |
948 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
949 | * failure. | |
950 | */ | |
90c07c89 | 951 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 952 | unsigned int flags, void *priv) |
ba7443bc TH |
953 | { |
954 | struct kernfs_root *root; | |
324a56e1 | 955 | struct kernfs_node *kn; |
ba7443bc TH |
956 | |
957 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
958 | if (!root) | |
959 | return ERR_PTR(-ENOMEM); | |
960 | ||
7d35079f | 961 | idr_init(&root->ino_idr); |
7d568a83 | 962 | INIT_LIST_HEAD(&root->supers); |
4a3ef68a | 963 | root->next_generation = 1; |
bc755553 | 964 | |
db4aad20 | 965 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
488dee96 | 966 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, |
db4aad20 | 967 | KERNFS_DIR); |
324a56e1 | 968 | if (!kn) { |
7d35079f | 969 | idr_destroy(&root->ino_idr); |
ba7443bc TH |
970 | kfree(root); |
971 | return ERR_PTR(-ENOMEM); | |
972 | } | |
973 | ||
324a56e1 | 974 | kn->priv = priv; |
adc5e8b5 | 975 | kn->dir.root = root; |
ba7443bc | 976 | |
90c07c89 | 977 | root->syscall_ops = scops; |
d35258ef | 978 | root->flags = flags; |
324a56e1 | 979 | root->kn = kn; |
abd54f02 | 980 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 981 | |
d35258ef TH |
982 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
983 | kernfs_activate(kn); | |
984 | ||
ba7443bc TH |
985 | return root; |
986 | } | |
987 | ||
988 | /** | |
989 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
990 | * @root: root of the hierarchy to destroy | |
991 | * | |
992 | * Destroy the hierarchy anchored at @root by removing all existing | |
993 | * directories and destroying @root. | |
994 | */ | |
995 | void kernfs_destroy_root(struct kernfs_root *root) | |
996 | { | |
324a56e1 | 997 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
998 | } |
999 | ||
fd7b9f7b TH |
1000 | /** |
1001 | * kernfs_create_dir_ns - create a directory | |
1002 | * @parent: parent in which to create a new directory | |
1003 | * @name: name of the new directory | |
bb8b9d09 | 1004 | * @mode: mode of the new directory |
488dee96 DT |
1005 | * @uid: uid of the new directory |
1006 | * @gid: gid of the new directory | |
fd7b9f7b TH |
1007 | * @priv: opaque data associated with the new directory |
1008 | * @ns: optional namespace tag of the directory | |
1009 | * | |
1010 | * Returns the created node on success, ERR_PTR() value on failure. | |
1011 | */ | |
324a56e1 | 1012 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 | 1013 | const char *name, umode_t mode, |
488dee96 | 1014 | kuid_t uid, kgid_t gid, |
bb8b9d09 | 1015 | void *priv, const void *ns) |
fd7b9f7b | 1016 | { |
324a56e1 | 1017 | struct kernfs_node *kn; |
fd7b9f7b TH |
1018 | int rc; |
1019 | ||
1020 | /* allocate */ | |
488dee96 DT |
1021 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, |
1022 | uid, gid, KERNFS_DIR); | |
324a56e1 | 1023 | if (!kn) |
fd7b9f7b TH |
1024 | return ERR_PTR(-ENOMEM); |
1025 | ||
adc5e8b5 TH |
1026 | kn->dir.root = parent->dir.root; |
1027 | kn->ns = ns; | |
324a56e1 | 1028 | kn->priv = priv; |
fd7b9f7b TH |
1029 | |
1030 | /* link in */ | |
988cd7af | 1031 | rc = kernfs_add_one(kn); |
fd7b9f7b | 1032 | if (!rc) |
324a56e1 | 1033 | return kn; |
fd7b9f7b | 1034 | |
324a56e1 | 1035 | kernfs_put(kn); |
fd7b9f7b TH |
1036 | return ERR_PTR(rc); |
1037 | } | |
1038 | ||
ea015218 EB |
1039 | /** |
1040 | * kernfs_create_empty_dir - create an always empty directory | |
1041 | * @parent: parent in which to create a new directory | |
1042 | * @name: name of the new directory | |
1043 | * | |
1044 | * Returns the created node on success, ERR_PTR() value on failure. | |
1045 | */ | |
1046 | struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent, | |
1047 | const char *name) | |
1048 | { | |
1049 | struct kernfs_node *kn; | |
1050 | int rc; | |
1051 | ||
1052 | /* allocate */ | |
488dee96 DT |
1053 | kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, |
1054 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR); | |
ea015218 EB |
1055 | if (!kn) |
1056 | return ERR_PTR(-ENOMEM); | |
1057 | ||
1058 | kn->flags |= KERNFS_EMPTY_DIR; | |
1059 | kn->dir.root = parent->dir.root; | |
1060 | kn->ns = NULL; | |
1061 | kn->priv = NULL; | |
1062 | ||
1063 | /* link in */ | |
1064 | rc = kernfs_add_one(kn); | |
1065 | if (!rc) | |
1066 | return kn; | |
1067 | ||
1068 | kernfs_put(kn); | |
1069 | return ERR_PTR(rc); | |
1070 | } | |
1071 | ||
c637b8ac TH |
1072 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
1073 | struct dentry *dentry, | |
1074 | unsigned int flags) | |
fd7b9f7b | 1075 | { |
19bbb926 | 1076 | struct dentry *ret; |
319ba91d | 1077 | struct kernfs_node *parent = dir->i_private; |
324a56e1 | 1078 | struct kernfs_node *kn; |
fd7b9f7b TH |
1079 | struct inode *inode; |
1080 | const void *ns = NULL; | |
1081 | ||
a797bfc3 | 1082 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1083 | |
324a56e1 | 1084 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1085 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 1086 | |
324a56e1 | 1087 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
1088 | |
1089 | /* no such entry */ | |
b9c9dad0 | 1090 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 1091 | ret = NULL; |
fd7b9f7b TH |
1092 | goto out_unlock; |
1093 | } | |
fd7b9f7b TH |
1094 | |
1095 | /* attach dentry and inode */ | |
c637b8ac | 1096 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
1097 | if (!inode) { |
1098 | ret = ERR_PTR(-ENOMEM); | |
1099 | goto out_unlock; | |
1100 | } | |
1101 | ||
1102 | /* instantiate and hash dentry */ | |
41d28bca | 1103 | ret = d_splice_alias(inode, dentry); |
fd7b9f7b | 1104 | out_unlock: |
a797bfc3 | 1105 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1106 | return ret; |
1107 | } | |
1108 | ||
80b9bbef TH |
1109 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
1110 | umode_t mode) | |
1111 | { | |
1112 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 1113 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 1114 | int ret; |
80b9bbef | 1115 | |
90c07c89 | 1116 | if (!scops || !scops->mkdir) |
80b9bbef TH |
1117 | return -EPERM; |
1118 | ||
07c7530d TH |
1119 | if (!kernfs_get_active(parent)) |
1120 | return -ENODEV; | |
1121 | ||
90c07c89 | 1122 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
1123 | |
1124 | kernfs_put_active(parent); | |
1125 | return ret; | |
80b9bbef TH |
1126 | } |
1127 | ||
1128 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
1129 | { | |
319ba91d | 1130 | struct kernfs_node *kn = kernfs_dentry_node(dentry); |
90c07c89 | 1131 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1132 | int ret; |
80b9bbef | 1133 | |
90c07c89 | 1134 | if (!scops || !scops->rmdir) |
80b9bbef TH |
1135 | return -EPERM; |
1136 | ||
07c7530d TH |
1137 | if (!kernfs_get_active(kn)) |
1138 | return -ENODEV; | |
1139 | ||
90c07c89 | 1140 | ret = scops->rmdir(kn); |
07c7530d TH |
1141 | |
1142 | kernfs_put_active(kn); | |
1143 | return ret; | |
80b9bbef TH |
1144 | } |
1145 | ||
1146 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1cd66c93 MS |
1147 | struct inode *new_dir, struct dentry *new_dentry, |
1148 | unsigned int flags) | |
80b9bbef | 1149 | { |
319ba91d | 1150 | struct kernfs_node *kn = kernfs_dentry_node(old_dentry); |
80b9bbef | 1151 | struct kernfs_node *new_parent = new_dir->i_private; |
90c07c89 | 1152 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1153 | int ret; |
80b9bbef | 1154 | |
1cd66c93 MS |
1155 | if (flags) |
1156 | return -EINVAL; | |
1157 | ||
90c07c89 | 1158 | if (!scops || !scops->rename) |
80b9bbef TH |
1159 | return -EPERM; |
1160 | ||
07c7530d TH |
1161 | if (!kernfs_get_active(kn)) |
1162 | return -ENODEV; | |
1163 | ||
1164 | if (!kernfs_get_active(new_parent)) { | |
1165 | kernfs_put_active(kn); | |
1166 | return -ENODEV; | |
1167 | } | |
1168 | ||
90c07c89 | 1169 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
1170 | |
1171 | kernfs_put_active(new_parent); | |
1172 | kernfs_put_active(kn); | |
1173 | return ret; | |
80b9bbef TH |
1174 | } |
1175 | ||
a797bfc3 | 1176 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
1177 | .lookup = kernfs_iop_lookup, |
1178 | .permission = kernfs_iop_permission, | |
1179 | .setattr = kernfs_iop_setattr, | |
1180 | .getattr = kernfs_iop_getattr, | |
c637b8ac | 1181 | .listxattr = kernfs_iop_listxattr, |
80b9bbef TH |
1182 | |
1183 | .mkdir = kernfs_iop_mkdir, | |
1184 | .rmdir = kernfs_iop_rmdir, | |
1185 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
1186 | }; |
1187 | ||
c637b8ac | 1188 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 1189 | { |
324a56e1 | 1190 | struct kernfs_node *last; |
fd7b9f7b TH |
1191 | |
1192 | while (true) { | |
1193 | struct rb_node *rbn; | |
1194 | ||
1195 | last = pos; | |
1196 | ||
df23fc39 | 1197 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
1198 | break; |
1199 | ||
adc5e8b5 | 1200 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
1201 | if (!rbn) |
1202 | break; | |
1203 | ||
324a56e1 | 1204 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
1205 | } |
1206 | ||
1207 | return last; | |
1208 | } | |
1209 | ||
1210 | /** | |
c637b8ac | 1211 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 1212 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 1213 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
1214 | * |
1215 | * Find the next descendant to visit for post-order traversal of @root's | |
1216 | * descendants. @root is included in the iteration and the last node to be | |
1217 | * visited. | |
1218 | */ | |
c637b8ac TH |
1219 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
1220 | struct kernfs_node *root) | |
fd7b9f7b TH |
1221 | { |
1222 | struct rb_node *rbn; | |
1223 | ||
a797bfc3 | 1224 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
1225 | |
1226 | /* if first iteration, visit leftmost descendant which may be root */ | |
1227 | if (!pos) | |
c637b8ac | 1228 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
1229 | |
1230 | /* if we visited @root, we're done */ | |
1231 | if (pos == root) | |
1232 | return NULL; | |
1233 | ||
1234 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 1235 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 1236 | if (rbn) |
c637b8ac | 1237 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
1238 | |
1239 | /* no sibling left, visit parent */ | |
adc5e8b5 | 1240 | return pos->parent; |
fd7b9f7b TH |
1241 | } |
1242 | ||
d35258ef TH |
1243 | /** |
1244 | * kernfs_activate - activate a node which started deactivated | |
1245 | * @kn: kernfs_node whose subtree is to be activated | |
1246 | * | |
1247 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
1248 | * needs to be explicitly activated. A node which hasn't been activated | |
1249 | * isn't visible to userland and deactivation is skipped during its | |
1250 | * removal. This is useful to construct atomic init sequences where | |
1251 | * creation of multiple nodes should either succeed or fail atomically. | |
1252 | * | |
1253 | * The caller is responsible for ensuring that this function is not called | |
1254 | * after kernfs_remove*() is invoked on @kn. | |
1255 | */ | |
1256 | void kernfs_activate(struct kernfs_node *kn) | |
1257 | { | |
1258 | struct kernfs_node *pos; | |
1259 | ||
1260 | mutex_lock(&kernfs_mutex); | |
1261 | ||
1262 | pos = NULL; | |
1263 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
1264 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) | |
1265 | continue; | |
1266 | ||
1267 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
1268 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
1269 | ||
1270 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
1271 | pos->flags |= KERNFS_ACTIVATED; | |
1272 | } | |
1273 | ||
1274 | mutex_unlock(&kernfs_mutex); | |
1275 | } | |
1276 | ||
988cd7af | 1277 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1278 | { |
35beab06 TH |
1279 | struct kernfs_node *pos; |
1280 | ||
1281 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 1282 | |
6b0afc2a TH |
1283 | /* |
1284 | * Short-circuit if non-root @kn has already finished removal. | |
1285 | * This is for kernfs_remove_self() which plays with active ref | |
1286 | * after removal. | |
1287 | */ | |
1288 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
1289 | return; |
1290 | ||
c637b8ac | 1291 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 1292 | |
81c173cb | 1293 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
1294 | pos = NULL; |
1295 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
1296 | if (kernfs_active(pos)) |
1297 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
1298 | |
1299 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 1300 | do { |
35beab06 TH |
1301 | pos = kernfs_leftmost_descendant(kn); |
1302 | ||
1303 | /* | |
81c173cb TH |
1304 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
1305 | * base ref could have been put by someone else by the time | |
1306 | * the function returns. Make sure it doesn't go away | |
1307 | * underneath us. | |
35beab06 TH |
1308 | */ |
1309 | kernfs_get(pos); | |
1310 | ||
d35258ef TH |
1311 | /* |
1312 | * Drain iff @kn was activated. This avoids draining and | |
1313 | * its lockdep annotations for nodes which have never been | |
1314 | * activated and allows embedding kernfs_remove() in create | |
1315 | * error paths without worrying about draining. | |
1316 | */ | |
1317 | if (kn->flags & KERNFS_ACTIVATED) | |
1318 | kernfs_drain(pos); | |
1319 | else | |
1320 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
1321 | |
1322 | /* | |
1323 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
1324 | * to decide who's responsible for cleanups. | |
1325 | */ | |
1326 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
1327 | struct kernfs_iattrs *ps_iattr = | |
1328 | pos->parent ? pos->parent->iattr : NULL; | |
1329 | ||
1330 | /* update timestamps on the parent */ | |
1331 | if (ps_iattr) { | |
95582b00 | 1332 | ktime_get_real_ts64(&ps_iattr->ia_iattr.ia_ctime); |
3a3a5fec DD |
1333 | ps_iattr->ia_iattr.ia_mtime = |
1334 | ps_iattr->ia_iattr.ia_ctime; | |
35beab06 TH |
1335 | } |
1336 | ||
988cd7af | 1337 | kernfs_put(pos); |
35beab06 TH |
1338 | } |
1339 | ||
1340 | kernfs_put(pos); | |
1341 | } while (pos != kn); | |
fd7b9f7b TH |
1342 | } |
1343 | ||
1344 | /** | |
324a56e1 TH |
1345 | * kernfs_remove - remove a kernfs_node recursively |
1346 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 1347 | * |
324a56e1 | 1348 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 1349 | */ |
324a56e1 | 1350 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1351 | { |
988cd7af TH |
1352 | mutex_lock(&kernfs_mutex); |
1353 | __kernfs_remove(kn); | |
1354 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
1355 | } |
1356 | ||
6b0afc2a TH |
1357 | /** |
1358 | * kernfs_break_active_protection - break out of active protection | |
1359 | * @kn: the self kernfs_node | |
1360 | * | |
1361 | * The caller must be running off of a kernfs operation which is invoked | |
1362 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
1363 | * this function must also be matched with an invocation of | |
1364 | * kernfs_unbreak_active_protection(). | |
1365 | * | |
1366 | * This function releases the active reference of @kn the caller is | |
1367 | * holding. Once this function is called, @kn may be removed at any point | |
1368 | * and the caller is solely responsible for ensuring that the objects it | |
1369 | * dereferences are accessible. | |
1370 | */ | |
1371 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
1372 | { | |
1373 | /* | |
1374 | * Take out ourself out of the active ref dependency chain. If | |
1375 | * we're called without an active ref, lockdep will complain. | |
1376 | */ | |
1377 | kernfs_put_active(kn); | |
1378 | } | |
1379 | ||
1380 | /** | |
1381 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
1382 | * @kn: the self kernfs_node | |
1383 | * | |
1384 | * If kernfs_break_active_protection() was called, this function must be | |
1385 | * invoked before finishing the kernfs operation. Note that while this | |
1386 | * function restores the active reference, it doesn't and can't actually | |
1387 | * restore the active protection - @kn may already or be in the process of | |
1388 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
1389 | * protection is irreversibly gone for the kernfs operation instance. | |
1390 | * | |
1391 | * While this function may be called at any point after | |
1392 | * kernfs_break_active_protection() is invoked, its most useful location | |
1393 | * would be right before the enclosing kernfs operation returns. | |
1394 | */ | |
1395 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
1396 | { | |
1397 | /* | |
1398 | * @kn->active could be in any state; however, the increment we do | |
1399 | * here will be undone as soon as the enclosing kernfs operation | |
1400 | * finishes and this temporary bump can't break anything. If @kn | |
1401 | * is alive, nothing changes. If @kn is being deactivated, the | |
1402 | * soon-to-follow put will either finish deactivation or restore | |
1403 | * deactivated state. If @kn is already removed, the temporary | |
1404 | * bump is guaranteed to be gone before @kn is released. | |
1405 | */ | |
1406 | atomic_inc(&kn->active); | |
1407 | if (kernfs_lockdep(kn)) | |
1408 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
1409 | } | |
1410 | ||
1411 | /** | |
1412 | * kernfs_remove_self - remove a kernfs_node from its own method | |
1413 | * @kn: the self kernfs_node to remove | |
1414 | * | |
1415 | * The caller must be running off of a kernfs operation which is invoked | |
1416 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
1417 | * implement a file operation which deletes itself. | |
1418 | * | |
1419 | * For example, the "delete" file for a sysfs device directory can be | |
1420 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
1421 | * itself. This function breaks the circular dependency of trying to | |
1422 | * deactivate self while holding an active ref itself. It isn't necessary | |
1423 | * to modify the usual removal path to use kernfs_remove_self(). The | |
1424 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
1425 | * before proceeding with the usual removal path. kernfs will ignore later | |
1426 | * kernfs_remove() on self. | |
1427 | * | |
1428 | * kernfs_remove_self() can be called multiple times concurrently on the | |
1429 | * same kernfs_node. Only the first one actually performs removal and | |
1430 | * returns %true. All others will wait until the kernfs operation which | |
1431 | * won self-removal finishes and return %false. Note that the losers wait | |
1432 | * for the completion of not only the winning kernfs_remove_self() but also | |
1433 | * the whole kernfs_ops which won the arbitration. This can be used to | |
1434 | * guarantee, for example, all concurrent writes to a "delete" file to | |
1435 | * finish only after the whole operation is complete. | |
1436 | */ | |
1437 | bool kernfs_remove_self(struct kernfs_node *kn) | |
1438 | { | |
1439 | bool ret; | |
1440 | ||
1441 | mutex_lock(&kernfs_mutex); | |
1442 | kernfs_break_active_protection(kn); | |
1443 | ||
1444 | /* | |
1445 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1446 | * the first one will actually perform removal. When the removal | |
1447 | * is complete, SUICIDED is set and the active ref is restored | |
1448 | * while holding kernfs_mutex. The ones which lost arbitration | |
1449 | * waits for SUICDED && drained which can happen only after the | |
1450 | * enclosing kernfs operation which executed the winning instance | |
1451 | * of kernfs_remove_self() finished. | |
1452 | */ | |
1453 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1454 | kn->flags |= KERNFS_SUICIDAL; | |
1455 | __kernfs_remove(kn); | |
1456 | kn->flags |= KERNFS_SUICIDED; | |
1457 | ret = true; | |
1458 | } else { | |
1459 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1460 | DEFINE_WAIT(wait); | |
1461 | ||
1462 | while (true) { | |
1463 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1464 | ||
1465 | if ((kn->flags & KERNFS_SUICIDED) && | |
1466 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1467 | break; | |
1468 | ||
1469 | mutex_unlock(&kernfs_mutex); | |
1470 | schedule(); | |
1471 | mutex_lock(&kernfs_mutex); | |
1472 | } | |
1473 | finish_wait(waitq, &wait); | |
1474 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1475 | ret = false; | |
1476 | } | |
1477 | ||
1478 | /* | |
1479 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1480 | * for SUICIDED && deactivated could finish prematurely. | |
1481 | */ | |
1482 | kernfs_unbreak_active_protection(kn); | |
1483 | ||
1484 | mutex_unlock(&kernfs_mutex); | |
1485 | return ret; | |
1486 | } | |
1487 | ||
fd7b9f7b | 1488 | /** |
324a56e1 TH |
1489 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1490 | * @parent: parent of the target | |
1491 | * @name: name of the kernfs_node to remove | |
1492 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1493 | * |
324a56e1 TH |
1494 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1495 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1496 | */ |
324a56e1 | 1497 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1498 | const void *ns) |
1499 | { | |
324a56e1 | 1500 | struct kernfs_node *kn; |
fd7b9f7b | 1501 | |
324a56e1 | 1502 | if (!parent) { |
c637b8ac | 1503 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1504 | name); |
1505 | return -ENOENT; | |
1506 | } | |
1507 | ||
988cd7af | 1508 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1509 | |
324a56e1 TH |
1510 | kn = kernfs_find_ns(parent, name, ns); |
1511 | if (kn) | |
988cd7af | 1512 | __kernfs_remove(kn); |
fd7b9f7b | 1513 | |
988cd7af | 1514 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1515 | |
324a56e1 | 1516 | if (kn) |
fd7b9f7b TH |
1517 | return 0; |
1518 | else | |
1519 | return -ENOENT; | |
1520 | } | |
1521 | ||
1522 | /** | |
1523 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1524 | * @kn: target node |
fd7b9f7b TH |
1525 | * @new_parent: new parent to put @sd under |
1526 | * @new_name: new name | |
1527 | * @new_ns: new namespace tag | |
1528 | */ | |
324a56e1 | 1529 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1530 | const char *new_name, const void *new_ns) |
1531 | { | |
3eef34ad TH |
1532 | struct kernfs_node *old_parent; |
1533 | const char *old_name = NULL; | |
fd7b9f7b TH |
1534 | int error; |
1535 | ||
3eef34ad TH |
1536 | /* can't move or rename root */ |
1537 | if (!kn->parent) | |
1538 | return -EINVAL; | |
1539 | ||
798c75a0 GKH |
1540 | mutex_lock(&kernfs_mutex); |
1541 | ||
d0ae3d43 | 1542 | error = -ENOENT; |
ea015218 EB |
1543 | if (!kernfs_active(kn) || !kernfs_active(new_parent) || |
1544 | (new_parent->flags & KERNFS_EMPTY_DIR)) | |
d0ae3d43 TH |
1545 | goto out; |
1546 | ||
fd7b9f7b | 1547 | error = 0; |
adc5e8b5 TH |
1548 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1549 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1550 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1551 | |
1552 | error = -EEXIST; | |
1553 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1554 | goto out; |
fd7b9f7b | 1555 | |
324a56e1 | 1556 | /* rename kernfs_node */ |
adc5e8b5 | 1557 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b | 1558 | error = -ENOMEM; |
75287a67 | 1559 | new_name = kstrdup_const(new_name, GFP_KERNEL); |
fd7b9f7b | 1560 | if (!new_name) |
798c75a0 | 1561 | goto out; |
3eef34ad TH |
1562 | } else { |
1563 | new_name = NULL; | |
fd7b9f7b TH |
1564 | } |
1565 | ||
1566 | /* | |
1567 | * Move to the appropriate place in the appropriate directories rbtree. | |
1568 | */ | |
c637b8ac | 1569 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1570 | kernfs_get(new_parent); |
3eef34ad TH |
1571 | |
1572 | /* rename_lock protects ->parent and ->name accessors */ | |
1573 | spin_lock_irq(&kernfs_rename_lock); | |
1574 | ||
1575 | old_parent = kn->parent; | |
adc5e8b5 | 1576 | kn->parent = new_parent; |
3eef34ad TH |
1577 | |
1578 | kn->ns = new_ns; | |
1579 | if (new_name) { | |
dfeb0750 | 1580 | old_name = kn->name; |
3eef34ad TH |
1581 | kn->name = new_name; |
1582 | } | |
1583 | ||
1584 | spin_unlock_irq(&kernfs_rename_lock); | |
1585 | ||
9561a896 | 1586 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
c637b8ac | 1587 | kernfs_link_sibling(kn); |
fd7b9f7b | 1588 | |
3eef34ad | 1589 | kernfs_put(old_parent); |
75287a67 | 1590 | kfree_const(old_name); |
3eef34ad | 1591 | |
fd7b9f7b | 1592 | error = 0; |
798c75a0 | 1593 | out: |
a797bfc3 | 1594 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1595 | return error; |
1596 | } | |
1597 | ||
fd7b9f7b | 1598 | /* Relationship between s_mode and the DT_xxx types */ |
324a56e1 | 1599 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1600 | { |
adc5e8b5 | 1601 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1602 | } |
1603 | ||
c637b8ac | 1604 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1605 | { |
1606 | kernfs_put(filp->private_data); | |
1607 | return 0; | |
1608 | } | |
1609 | ||
c637b8ac | 1610 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1611 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1612 | { |
1613 | if (pos) { | |
81c173cb | 1614 | int valid = kernfs_active(pos) && |
798c75a0 | 1615 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1616 | kernfs_put(pos); |
1617 | if (!valid) | |
1618 | pos = NULL; | |
1619 | } | |
1620 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1621 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1622 | while (node) { |
324a56e1 | 1623 | pos = rb_to_kn(node); |
fd7b9f7b | 1624 | |
adc5e8b5 | 1625 | if (hash < pos->hash) |
fd7b9f7b | 1626 | node = node->rb_left; |
adc5e8b5 | 1627 | else if (hash > pos->hash) |
fd7b9f7b TH |
1628 | node = node->rb_right; |
1629 | else | |
1630 | break; | |
1631 | } | |
1632 | } | |
b9c9dad0 TH |
1633 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1634 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1635 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1636 | if (!node) |
1637 | pos = NULL; | |
1638 | else | |
324a56e1 | 1639 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1640 | } |
1641 | return pos; | |
1642 | } | |
1643 | ||
c637b8ac | 1644 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1645 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1646 | { |
c637b8ac | 1647 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1648 | if (pos) { |
fd7b9f7b | 1649 | do { |
adc5e8b5 | 1650 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1651 | if (!node) |
1652 | pos = NULL; | |
1653 | else | |
324a56e1 | 1654 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1655 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1656 | } | |
fd7b9f7b TH |
1657 | return pos; |
1658 | } | |
1659 | ||
c637b8ac | 1660 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1661 | { |
1662 | struct dentry *dentry = file->f_path.dentry; | |
319ba91d | 1663 | struct kernfs_node *parent = kernfs_dentry_node(dentry); |
324a56e1 | 1664 | struct kernfs_node *pos = file->private_data; |
fd7b9f7b TH |
1665 | const void *ns = NULL; |
1666 | ||
1667 | if (!dir_emit_dots(file, ctx)) | |
1668 | return 0; | |
a797bfc3 | 1669 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1670 | |
324a56e1 | 1671 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1672 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1673 | |
c637b8ac | 1674 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1675 | pos; |
c637b8ac | 1676 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1677 | const char *name = pos->name; |
fd7b9f7b TH |
1678 | unsigned int type = dt_type(pos); |
1679 | int len = strlen(name); | |
c53cd490 | 1680 | ino_t ino = pos->id.ino; |
fd7b9f7b | 1681 | |
adc5e8b5 | 1682 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1683 | file->private_data = pos; |
1684 | kernfs_get(pos); | |
1685 | ||
a797bfc3 | 1686 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1687 | if (!dir_emit(ctx, name, len, ino, type)) |
1688 | return 0; | |
a797bfc3 | 1689 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1690 | } |
a797bfc3 | 1691 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1692 | file->private_data = NULL; |
1693 | ctx->pos = INT_MAX; | |
1694 | return 0; | |
1695 | } | |
1696 | ||
a797bfc3 | 1697 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1698 | .read = generic_read_dir, |
8cb0d2c1 | 1699 | .iterate_shared = kernfs_fop_readdir, |
c637b8ac | 1700 | .release = kernfs_dir_fop_release, |
8cb0d2c1 | 1701 | .llseek = generic_file_llseek, |
fd7b9f7b | 1702 | }; |