]>
Commit | Line | Data |
---|---|---|
55716d26 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b8441ed2 TH |
2 | /* |
3 | * fs/kernfs/dir.c - kernfs directory implementation | |
4 | * | |
5 | * Copyright (c) 2001-3 Patrick Mochel | |
6 | * Copyright (c) 2007 SUSE Linux Products GmbH | |
7 | * Copyright (c) 2007, 2013 Tejun Heo <tj@kernel.org> | |
b8441ed2 | 8 | */ |
fd7b9f7b | 9 | |
abd54f02 | 10 | #include <linux/sched.h> |
fd7b9f7b TH |
11 | #include <linux/fs.h> |
12 | #include <linux/namei.h> | |
13 | #include <linux/idr.h> | |
14 | #include <linux/slab.h> | |
15 | #include <linux/security.h> | |
16 | #include <linux/hash.h> | |
17 | ||
18 | #include "kernfs-internal.h" | |
19 | ||
a797bfc3 | 20 | DEFINE_MUTEX(kernfs_mutex); |
3eef34ad TH |
21 | static DEFINE_SPINLOCK(kernfs_rename_lock); /* kn->parent and ->name */ |
22 | static char kernfs_pr_cont_buf[PATH_MAX]; /* protected by rename_lock */ | |
7d35079f | 23 | static DEFINE_SPINLOCK(kernfs_idr_lock); /* root->ino_idr */ |
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 | { | |
17627157 KK |
44 | if (!kn) |
45 | return strlcpy(buf, "(null)", buflen); | |
46 | ||
3eef34ad TH |
47 | return strlcpy(buf, kn->parent ? kn->name : "/", buflen); |
48 | } | |
49 | ||
9f6df573 AK |
50 | /* kernfs_node_depth - compute depth from @from to @to */ |
51 | static size_t kernfs_depth(struct kernfs_node *from, struct kernfs_node *to) | |
3eef34ad | 52 | { |
9f6df573 | 53 | size_t depth = 0; |
3eef34ad | 54 | |
9f6df573 AK |
55 | while (to->parent && to != from) { |
56 | depth++; | |
57 | to = to->parent; | |
58 | } | |
59 | return depth; | |
60 | } | |
3eef34ad | 61 | |
9f6df573 AK |
62 | static struct kernfs_node *kernfs_common_ancestor(struct kernfs_node *a, |
63 | struct kernfs_node *b) | |
64 | { | |
65 | size_t da, db; | |
66 | struct kernfs_root *ra = kernfs_root(a), *rb = kernfs_root(b); | |
67 | ||
68 | if (ra != rb) | |
69 | return NULL; | |
70 | ||
71 | da = kernfs_depth(ra->kn, a); | |
72 | db = kernfs_depth(rb->kn, b); | |
73 | ||
74 | while (da > db) { | |
75 | a = a->parent; | |
76 | da--; | |
77 | } | |
78 | while (db > da) { | |
79 | b = b->parent; | |
80 | db--; | |
81 | } | |
82 | ||
83 | /* worst case b and a will be the same at root */ | |
84 | while (b != a) { | |
85 | b = b->parent; | |
86 | a = a->parent; | |
87 | } | |
88 | ||
89 | return a; | |
90 | } | |
91 | ||
92 | /** | |
93 | * kernfs_path_from_node_locked - find a pseudo-absolute path to @kn_to, | |
94 | * where kn_from is treated as root of the path. | |
95 | * @kn_from: kernfs node which should be treated as root for the path | |
96 | * @kn_to: kernfs node to which path is needed | |
97 | * @buf: buffer to copy the path into | |
98 | * @buflen: size of @buf | |
99 | * | |
100 | * We need to handle couple of scenarios here: | |
101 | * [1] when @kn_from is an ancestor of @kn_to at some level | |
102 | * kn_from: /n1/n2/n3 | |
103 | * kn_to: /n1/n2/n3/n4/n5 | |
104 | * result: /n4/n5 | |
105 | * | |
106 | * [2] when @kn_from is on a different hierarchy and we need to find common | |
107 | * ancestor between @kn_from and @kn_to. | |
108 | * kn_from: /n1/n2/n3/n4 | |
109 | * kn_to: /n1/n2/n5 | |
110 | * result: /../../n5 | |
111 | * OR | |
112 | * kn_from: /n1/n2/n3/n4/n5 [depth=5] | |
113 | * kn_to: /n1/n2/n3 [depth=3] | |
114 | * result: /../.. | |
115 | * | |
17627157 KK |
116 | * [3] when @kn_to is NULL result will be "(null)" |
117 | * | |
3abb1d90 TH |
118 | * Returns the length of the full path. If the full length is equal to or |
119 | * greater than @buflen, @buf contains the truncated path with the trailing | |
120 | * '\0'. On error, -errno is returned. | |
9f6df573 AK |
121 | */ |
122 | static int kernfs_path_from_node_locked(struct kernfs_node *kn_to, | |
123 | struct kernfs_node *kn_from, | |
124 | char *buf, size_t buflen) | |
125 | { | |
126 | struct kernfs_node *kn, *common; | |
127 | const char parent_str[] = "/.."; | |
3abb1d90 TH |
128 | size_t depth_from, depth_to, len = 0; |
129 | int i, j; | |
9f6df573 | 130 | |
17627157 KK |
131 | if (!kn_to) |
132 | return strlcpy(buf, "(null)", buflen); | |
133 | ||
9f6df573 AK |
134 | if (!kn_from) |
135 | kn_from = kernfs_root(kn_to)->kn; | |
136 | ||
137 | if (kn_from == kn_to) | |
138 | return strlcpy(buf, "/", buflen); | |
139 | ||
bbe70e4e JJB |
140 | if (!buf) |
141 | return -EINVAL; | |
142 | ||
9f6df573 AK |
143 | common = kernfs_common_ancestor(kn_from, kn_to); |
144 | if (WARN_ON(!common)) | |
3abb1d90 | 145 | return -EINVAL; |
9f6df573 AK |
146 | |
147 | depth_to = kernfs_depth(common, kn_to); | |
148 | depth_from = kernfs_depth(common, kn_from); | |
149 | ||
bbe70e4e | 150 | buf[0] = '\0'; |
9f6df573 AK |
151 | |
152 | for (i = 0; i < depth_from; i++) | |
153 | len += strlcpy(buf + len, parent_str, | |
154 | len < buflen ? buflen - len : 0); | |
155 | ||
156 | /* Calculate how many bytes we need for the rest */ | |
3abb1d90 TH |
157 | for (i = depth_to - 1; i >= 0; i--) { |
158 | for (kn = kn_to, j = 0; j < i; j++) | |
159 | kn = kn->parent; | |
160 | len += strlcpy(buf + len, "/", | |
161 | len < buflen ? buflen - len : 0); | |
162 | len += strlcpy(buf + len, kn->name, | |
163 | len < buflen ? buflen - len : 0); | |
9f6df573 | 164 | } |
3eef34ad | 165 | |
3abb1d90 | 166 | return len; |
3eef34ad TH |
167 | } |
168 | ||
169 | /** | |
170 | * kernfs_name - obtain the name of a given node | |
171 | * @kn: kernfs_node of interest | |
172 | * @buf: buffer to copy @kn's name into | |
173 | * @buflen: size of @buf | |
174 | * | |
175 | * Copies the name of @kn into @buf of @buflen bytes. The behavior is | |
176 | * similar to strlcpy(). It returns the length of @kn's name and if @buf | |
177 | * isn't long enough, it's filled upto @buflen-1 and nul terminated. | |
178 | * | |
17627157 KK |
179 | * Fills buffer with "(null)" if @kn is NULL. |
180 | * | |
3eef34ad TH |
181 | * This function can be called from any context. |
182 | */ | |
183 | int kernfs_name(struct kernfs_node *kn, char *buf, size_t buflen) | |
184 | { | |
185 | unsigned long flags; | |
186 | int ret; | |
187 | ||
188 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
189 | ret = kernfs_name_locked(kn, buf, buflen); | |
190 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
191 | return ret; | |
192 | } | |
193 | ||
9f6df573 AK |
194 | /** |
195 | * kernfs_path_from_node - build path of node @to relative to @from. | |
196 | * @from: parent kernfs_node relative to which we need to build the path | |
197 | * @to: kernfs_node of interest | |
198 | * @buf: buffer to copy @to's path into | |
199 | * @buflen: size of @buf | |
200 | * | |
201 | * Builds @to's path relative to @from in @buf. @from and @to must | |
202 | * be on the same kernfs-root. If @from is not parent of @to, then a relative | |
203 | * path (which includes '..'s) as needed to reach from @from to @to is | |
204 | * returned. | |
205 | * | |
3abb1d90 TH |
206 | * Returns the length of the full path. If the full length is equal to or |
207 | * greater than @buflen, @buf contains the truncated path with the trailing | |
208 | * '\0'. On error, -errno is returned. | |
9f6df573 AK |
209 | */ |
210 | int kernfs_path_from_node(struct kernfs_node *to, struct kernfs_node *from, | |
211 | char *buf, size_t buflen) | |
212 | { | |
213 | unsigned long flags; | |
214 | int ret; | |
215 | ||
216 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
217 | ret = kernfs_path_from_node_locked(to, from, buf, buflen); | |
218 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
219 | return ret; | |
220 | } | |
221 | EXPORT_SYMBOL_GPL(kernfs_path_from_node); | |
222 | ||
3eef34ad TH |
223 | /** |
224 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node | |
225 | * @kn: kernfs_node of interest | |
226 | * | |
227 | * This function can be called from any context. | |
228 | */ | |
229 | void pr_cont_kernfs_name(struct kernfs_node *kn) | |
230 | { | |
231 | unsigned long flags; | |
232 | ||
233 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
234 | ||
235 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); | |
236 | pr_cont("%s", kernfs_pr_cont_buf); | |
237 | ||
238 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
239 | } | |
240 | ||
241 | /** | |
242 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node | |
243 | * @kn: kernfs_node of interest | |
244 | * | |
245 | * This function can be called from any context. | |
246 | */ | |
247 | void pr_cont_kernfs_path(struct kernfs_node *kn) | |
248 | { | |
249 | unsigned long flags; | |
9f6df573 | 250 | int sz; |
3eef34ad TH |
251 | |
252 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
253 | ||
9f6df573 AK |
254 | sz = kernfs_path_from_node_locked(kn, NULL, kernfs_pr_cont_buf, |
255 | sizeof(kernfs_pr_cont_buf)); | |
256 | if (sz < 0) { | |
257 | pr_cont("(error)"); | |
258 | goto out; | |
259 | } | |
260 | ||
261 | if (sz >= sizeof(kernfs_pr_cont_buf)) { | |
262 | pr_cont("(name too long)"); | |
263 | goto out; | |
264 | } | |
265 | ||
266 | pr_cont("%s", kernfs_pr_cont_buf); | |
3eef34ad | 267 | |
9f6df573 | 268 | out: |
3eef34ad TH |
269 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); |
270 | } | |
271 | ||
272 | /** | |
273 | * kernfs_get_parent - determine the parent node and pin it | |
274 | * @kn: kernfs_node of interest | |
275 | * | |
276 | * Determines @kn's parent, pins and returns it. This function can be | |
277 | * called from any context. | |
278 | */ | |
279 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) | |
280 | { | |
281 | struct kernfs_node *parent; | |
282 | unsigned long flags; | |
283 | ||
284 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
285 | parent = kn->parent; | |
286 | kernfs_get(parent); | |
287 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
288 | ||
289 | return parent; | |
290 | } | |
291 | ||
fd7b9f7b | 292 | /** |
c637b8ac | 293 | * kernfs_name_hash |
fd7b9f7b TH |
294 | * @name: Null terminated string to hash |
295 | * @ns: Namespace tag to hash | |
296 | * | |
297 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
298 | */ | |
c637b8ac | 299 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b | 300 | { |
8387ff25 | 301 | unsigned long hash = init_name_hash(ns); |
fd7b9f7b TH |
302 | unsigned int len = strlen(name); |
303 | while (len--) | |
304 | hash = partial_name_hash(*name++, hash); | |
8387ff25 | 305 | hash = end_name_hash(hash); |
fd7b9f7b TH |
306 | hash &= 0x7fffffffU; |
307 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
88391d49 | 308 | if (hash < 2) |
fd7b9f7b TH |
309 | hash += 2; |
310 | if (hash >= INT_MAX) | |
311 | hash = INT_MAX - 1; | |
312 | return hash; | |
313 | } | |
314 | ||
c637b8ac TH |
315 | static int kernfs_name_compare(unsigned int hash, const char *name, |
316 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 317 | { |
72392ed0 RV |
318 | if (hash < kn->hash) |
319 | return -1; | |
320 | if (hash > kn->hash) | |
321 | return 1; | |
322 | if (ns < kn->ns) | |
323 | return -1; | |
324 | if (ns > kn->ns) | |
325 | return 1; | |
adc5e8b5 | 326 | return strcmp(name, kn->name); |
fd7b9f7b TH |
327 | } |
328 | ||
c637b8ac TH |
329 | static int kernfs_sd_compare(const struct kernfs_node *left, |
330 | const struct kernfs_node *right) | |
fd7b9f7b | 331 | { |
c637b8ac | 332 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
333 | } |
334 | ||
335 | /** | |
c637b8ac | 336 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 337 | * @kn: kernfs_node of interest |
fd7b9f7b | 338 | * |
324a56e1 | 339 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 340 | * @kn->parent->dir.children. |
fd7b9f7b TH |
341 | * |
342 | * Locking: | |
a797bfc3 | 343 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
344 | * |
345 | * RETURNS: | |
346 | * 0 on susccess -EEXIST on failure. | |
347 | */ | |
c637b8ac | 348 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 349 | { |
adc5e8b5 | 350 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
351 | struct rb_node *parent = NULL; |
352 | ||
fd7b9f7b | 353 | while (*node) { |
324a56e1 | 354 | struct kernfs_node *pos; |
fd7b9f7b TH |
355 | int result; |
356 | ||
324a56e1 | 357 | pos = rb_to_kn(*node); |
fd7b9f7b | 358 | parent = *node; |
c637b8ac | 359 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 360 | if (result < 0) |
adc5e8b5 | 361 | node = &pos->rb.rb_left; |
fd7b9f7b | 362 | else if (result > 0) |
adc5e8b5 | 363 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
364 | else |
365 | return -EEXIST; | |
366 | } | |
c1befb88 | 367 | |
fd7b9f7b | 368 | /* add new node and rebalance the tree */ |
adc5e8b5 TH |
369 | rb_link_node(&kn->rb, parent, node); |
370 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
c1befb88 JZ |
371 | |
372 | /* successfully added, account subdir number */ | |
373 | if (kernfs_type(kn) == KERNFS_DIR) | |
374 | kn->parent->dir.subdirs++; | |
375 | ||
fd7b9f7b TH |
376 | return 0; |
377 | } | |
378 | ||
379 | /** | |
c637b8ac | 380 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 381 | * @kn: kernfs_node of interest |
fd7b9f7b | 382 | * |
35beab06 TH |
383 | * Try to unlink @kn from its sibling rbtree which starts from |
384 | * kn->parent->dir.children. Returns %true if @kn was actually | |
385 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
386 | * |
387 | * Locking: | |
a797bfc3 | 388 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 389 | */ |
35beab06 | 390 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 391 | { |
35beab06 TH |
392 | if (RB_EMPTY_NODE(&kn->rb)) |
393 | return false; | |
394 | ||
df23fc39 | 395 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 396 | kn->parent->dir.subdirs--; |
fd7b9f7b | 397 | |
adc5e8b5 | 398 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
399 | RB_CLEAR_NODE(&kn->rb); |
400 | return true; | |
fd7b9f7b TH |
401 | } |
402 | ||
403 | /** | |
c637b8ac | 404 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 405 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 406 | * |
324a56e1 | 407 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
408 | * is NULL. |
409 | * | |
410 | * RETURNS: | |
324a56e1 | 411 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 412 | */ |
c637b8ac | 413 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 414 | { |
324a56e1 | 415 | if (unlikely(!kn)) |
fd7b9f7b TH |
416 | return NULL; |
417 | ||
f4b3e631 GKH |
418 | if (!atomic_inc_unless_negative(&kn->active)) |
419 | return NULL; | |
895a068a | 420 | |
182fd64b | 421 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
422 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
423 | return kn; | |
fd7b9f7b TH |
424 | } |
425 | ||
426 | /** | |
c637b8ac | 427 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 428 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 429 | * |
324a56e1 | 430 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
431 | * is NULL. |
432 | */ | |
c637b8ac | 433 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b TH |
434 | { |
435 | int v; | |
436 | ||
324a56e1 | 437 | if (unlikely(!kn)) |
fd7b9f7b TH |
438 | return; |
439 | ||
182fd64b | 440 | if (kernfs_lockdep(kn)) |
5facae4f | 441 | rwsem_release(&kn->dep_map, _RET_IP_); |
adc5e8b5 | 442 | v = atomic_dec_return(&kn->active); |
df23fc39 | 443 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
444 | return; |
445 | ||
2fd60da4 | 446 | wake_up_all(&kernfs_root(kn)->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 | 478 | lock_acquired(&kn->dep_map, _RET_IP_); |
5facae4f | 479 | rwsem_release(&kn->dep_map, _RET_IP_); |
a6607930 | 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 | |
adc5e8b5 | 511 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 512 | return; |
324a56e1 | 513 | root = kernfs_root(kn); |
fd7b9f7b | 514 | repeat: |
81c173cb TH |
515 | /* |
516 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 517 | * kn->parent won't change beneath us. |
fd7b9f7b | 518 | */ |
adc5e8b5 | 519 | parent = kn->parent; |
fd7b9f7b | 520 | |
81c173cb TH |
521 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
522 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
523 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 524 | |
df23fc39 | 525 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 526 | kernfs_put(kn->symlink.target_kn); |
dfeb0750 TH |
527 | |
528 | kfree_const(kn->name); | |
529 | ||
adc5e8b5 | 530 | if (kn->iattr) { |
adc5e8b5 | 531 | simple_xattrs_free(&kn->iattr->xattrs); |
26e28d68 | 532 | kmem_cache_free(kernfs_iattrs_cache, kn->iattr); |
2322392b | 533 | } |
7d35079f | 534 | spin_lock(&kernfs_idr_lock); |
40430452 | 535 | idr_remove(&root->ino_idr, (u32)kernfs_ino(kn)); |
7d35079f | 536 | spin_unlock(&kernfs_idr_lock); |
a797bfc3 | 537 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 538 | |
324a56e1 TH |
539 | kn = parent; |
540 | if (kn) { | |
adc5e8b5 | 541 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
542 | goto repeat; |
543 | } else { | |
324a56e1 | 544 | /* just released the root kn, free @root too */ |
7d35079f | 545 | idr_destroy(&root->ino_idr); |
ba7443bc TH |
546 | kfree(root); |
547 | } | |
fd7b9f7b TH |
548 | } |
549 | EXPORT_SYMBOL_GPL(kernfs_put); | |
550 | ||
c637b8ac | 551 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 552 | { |
324a56e1 | 553 | struct kernfs_node *kn; |
fd7b9f7b TH |
554 | |
555 | if (flags & LOOKUP_RCU) | |
556 | return -ECHILD; | |
557 | ||
19bbb926 | 558 | /* Always perform fresh lookup for negatives */ |
2b0143b5 | 559 | if (d_really_is_negative(dentry)) |
19bbb926 TH |
560 | goto out_bad_unlocked; |
561 | ||
319ba91d | 562 | kn = kernfs_dentry_node(dentry); |
a797bfc3 | 563 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 564 | |
81c173cb TH |
565 | /* The kernfs node has been deactivated */ |
566 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
567 | goto out_bad; |
568 | ||
c637b8ac | 569 | /* The kernfs node has been moved? */ |
319ba91d | 570 | if (kernfs_dentry_node(dentry->d_parent) != kn->parent) |
fd7b9f7b TH |
571 | goto out_bad; |
572 | ||
c637b8ac | 573 | /* The kernfs node has been renamed */ |
adc5e8b5 | 574 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
575 | goto out_bad; |
576 | ||
c637b8ac | 577 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 578 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 579 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
580 | goto out_bad; |
581 | ||
a797bfc3 | 582 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
583 | return 1; |
584 | out_bad: | |
a797bfc3 | 585 | mutex_unlock(&kernfs_mutex); |
19bbb926 | 586 | out_bad_unlocked: |
fd7b9f7b TH |
587 | return 0; |
588 | } | |
589 | ||
a797bfc3 | 590 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 591 | .d_revalidate = kernfs_dop_revalidate, |
fd7b9f7b TH |
592 | }; |
593 | ||
0c23b225 TH |
594 | /** |
595 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry | |
596 | * @dentry: the dentry in question | |
597 | * | |
598 | * Return the kernfs_node associated with @dentry. If @dentry is not a | |
599 | * kernfs one, %NULL is returned. | |
600 | * | |
601 | * While the returned kernfs_node will stay accessible as long as @dentry | |
602 | * is accessible, the returned node can be in any state and the caller is | |
603 | * fully responsible for determining what's accessible. | |
604 | */ | |
605 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) | |
606 | { | |
0288e7fa | 607 | if (dentry->d_sb->s_op == &kernfs_sops) |
319ba91d | 608 | return kernfs_dentry_node(dentry); |
0c23b225 TH |
609 | return NULL; |
610 | } | |
611 | ||
db4aad20 | 612 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
e19dfdc8 | 613 | struct kernfs_node *parent, |
db4aad20 | 614 | const char *name, umode_t mode, |
488dee96 | 615 | kuid_t uid, kgid_t gid, |
db4aad20 | 616 | unsigned flags) |
fd7b9f7b | 617 | { |
324a56e1 | 618 | struct kernfs_node *kn; |
40430452 | 619 | u32 id_highbits; |
bc755553 | 620 | int ret; |
fd7b9f7b | 621 | |
dfeb0750 TH |
622 | name = kstrdup_const(name, GFP_KERNEL); |
623 | if (!name) | |
624 | return NULL; | |
fd7b9f7b | 625 | |
a797bfc3 | 626 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 627 | if (!kn) |
fd7b9f7b TH |
628 | goto err_out1; |
629 | ||
7d35079f SL |
630 | idr_preload(GFP_KERNEL); |
631 | spin_lock(&kernfs_idr_lock); | |
4a3ef68a | 632 | ret = idr_alloc_cyclic(&root->ino_idr, kn, 1, 0, GFP_ATOMIC); |
40430452 TH |
633 | if (ret >= 0 && ret < root->last_id_lowbits) |
634 | root->id_highbits++; | |
635 | id_highbits = root->id_highbits; | |
636 | root->last_id_lowbits = ret; | |
7d35079f SL |
637 | spin_unlock(&kernfs_idr_lock); |
638 | idr_preload_end(); | |
bc755553 | 639 | if (ret < 0) |
fd7b9f7b | 640 | goto err_out2; |
67c0496e | 641 | |
40430452 | 642 | kn->id = (u64)id_highbits << 32 | ret; |
fd7b9f7b | 643 | |
b680b081 | 644 | atomic_set(&kn->count, 1); |
81c173cb | 645 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 646 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 647 | |
adc5e8b5 TH |
648 | kn->name = name; |
649 | kn->mode = mode; | |
81c173cb | 650 | kn->flags = flags; |
fd7b9f7b | 651 | |
488dee96 DT |
652 | if (!uid_eq(uid, GLOBAL_ROOT_UID) || !gid_eq(gid, GLOBAL_ROOT_GID)) { |
653 | struct iattr iattr = { | |
654 | .ia_valid = ATTR_UID | ATTR_GID, | |
655 | .ia_uid = uid, | |
656 | .ia_gid = gid, | |
657 | }; | |
658 | ||
659 | ret = __kernfs_setattr(kn, &iattr); | |
660 | if (ret < 0) | |
661 | goto err_out3; | |
662 | } | |
663 | ||
e19dfdc8 OM |
664 | if (parent) { |
665 | ret = security_kernfs_init_security(parent, kn); | |
666 | if (ret) | |
667 | goto err_out3; | |
668 | } | |
669 | ||
324a56e1 | 670 | return kn; |
fd7b9f7b | 671 | |
488dee96 | 672 | err_out3: |
40430452 | 673 | idr_remove(&root->ino_idr, (u32)kernfs_ino(kn)); |
fd7b9f7b | 674 | err_out2: |
a797bfc3 | 675 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 676 | err_out1: |
dfeb0750 | 677 | kfree_const(name); |
fd7b9f7b TH |
678 | return NULL; |
679 | } | |
680 | ||
db4aad20 TH |
681 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
682 | const char *name, umode_t mode, | |
488dee96 | 683 | kuid_t uid, kgid_t gid, |
db4aad20 TH |
684 | unsigned flags) |
685 | { | |
686 | struct kernfs_node *kn; | |
687 | ||
e19dfdc8 | 688 | kn = __kernfs_new_node(kernfs_root(parent), parent, |
488dee96 | 689 | name, mode, uid, gid, flags); |
db4aad20 TH |
690 | if (kn) { |
691 | kernfs_get(parent); | |
692 | kn->parent = parent; | |
693 | } | |
694 | return kn; | |
695 | } | |
696 | ||
ba16b284 | 697 | /* |
fe0f726c | 698 | * kernfs_find_and_get_node_by_id - get kernfs_node from node id |
ba16b284 | 699 | * @root: the kernfs root |
fe0f726c TH |
700 | * @id: the target node id |
701 | * | |
702 | * @id's lower 32bits encode ino and upper gen. If the gen portion is | |
703 | * zero, all generations are matched. | |
ba16b284 SL |
704 | * |
705 | * RETURNS: | |
706 | * NULL on failure. Return a kernfs node with reference counter incremented | |
707 | */ | |
fe0f726c TH |
708 | struct kernfs_node *kernfs_find_and_get_node_by_id(struct kernfs_root *root, |
709 | u64 id) | |
ba16b284 SL |
710 | { |
711 | struct kernfs_node *kn; | |
fe0f726c TH |
712 | ino_t ino = kernfs_id_ino(id); |
713 | u32 gen = kernfs_id_gen(id); | |
ba16b284 | 714 | |
b680b081 TH |
715 | spin_lock(&kernfs_idr_lock); |
716 | ||
40430452 | 717 | kn = idr_find(&root->ino_idr, (u32)ino); |
ba16b284 | 718 | if (!kn) |
b680b081 | 719 | goto err_unlock; |
ba16b284 | 720 | |
40430452 TH |
721 | if (sizeof(ino_t) >= sizeof(u64)) { |
722 | /* we looked up with the low 32bits, compare the whole */ | |
723 | if (kernfs_ino(kn) != ino) | |
724 | goto err_unlock; | |
725 | } else { | |
726 | /* 0 matches all generations */ | |
727 | if (unlikely(gen && kernfs_gen(kn) != gen)) | |
728 | goto err_unlock; | |
729 | } | |
fe0f726c | 730 | |
880df131 TH |
731 | /* |
732 | * ACTIVATED is protected with kernfs_mutex but it was clear when | |
733 | * @kn was added to idr and we just wanna see it set. No need to | |
734 | * grab kernfs_mutex. | |
735 | */ | |
736 | if (unlikely(!(kn->flags & KERNFS_ACTIVATED) || | |
737 | !atomic_inc_not_zero(&kn->count))) | |
b680b081 | 738 | goto err_unlock; |
ba16b284 | 739 | |
b680b081 | 740 | spin_unlock(&kernfs_idr_lock); |
ba16b284 | 741 | return kn; |
b680b081 TH |
742 | err_unlock: |
743 | spin_unlock(&kernfs_idr_lock); | |
ba16b284 SL |
744 | return NULL; |
745 | } | |
746 | ||
fd7b9f7b | 747 | /** |
c637b8ac | 748 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 749 | * @kn: kernfs_node to be added |
fd7b9f7b | 750 | * |
db4aad20 TH |
751 | * The caller must already have initialized @kn->parent. This |
752 | * function increments nlink of the parent's inode if @kn is a | |
753 | * directory and link into the children list of the parent. | |
fd7b9f7b | 754 | * |
fd7b9f7b TH |
755 | * RETURNS: |
756 | * 0 on success, -EEXIST if entry with the given name already | |
757 | * exists. | |
758 | */ | |
988cd7af | 759 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 760 | { |
db4aad20 | 761 | struct kernfs_node *parent = kn->parent; |
c525aadd | 762 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 763 | bool has_ns; |
fd7b9f7b TH |
764 | int ret; |
765 | ||
988cd7af TH |
766 | mutex_lock(&kernfs_mutex); |
767 | ||
768 | ret = -EINVAL; | |
769 | has_ns = kernfs_ns_enabled(parent); | |
770 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
771 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
772 | goto out_unlock; | |
fd7b9f7b | 773 | |
df23fc39 | 774 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 775 | goto out_unlock; |
fd7b9f7b | 776 | |
988cd7af | 777 | ret = -ENOENT; |
ea015218 EB |
778 | if (parent->flags & KERNFS_EMPTY_DIR) |
779 | goto out_unlock; | |
780 | ||
d35258ef | 781 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 782 | goto out_unlock; |
798c75a0 | 783 | |
c637b8ac | 784 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 785 | |
c637b8ac | 786 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 787 | if (ret) |
988cd7af | 788 | goto out_unlock; |
fd7b9f7b TH |
789 | |
790 | /* Update timestamps on the parent */ | |
adc5e8b5 | 791 | ps_iattr = parent->iattr; |
fd7b9f7b | 792 | if (ps_iattr) { |
05895219 OM |
793 | ktime_get_real_ts64(&ps_iattr->ia_ctime); |
794 | ps_iattr->ia_mtime = ps_iattr->ia_ctime; | |
fd7b9f7b TH |
795 | } |
796 | ||
d35258ef TH |
797 | mutex_unlock(&kernfs_mutex); |
798 | ||
799 | /* | |
800 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
801 | * If not activated here, the kernfs user is responsible for | |
802 | * activating the node with kernfs_activate(). A node which hasn't | |
803 | * been activated is not visible to userland and its removal won't | |
804 | * trigger deactivation. | |
805 | */ | |
806 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
807 | kernfs_activate(kn); | |
808 | return 0; | |
809 | ||
988cd7af | 810 | out_unlock: |
a797bfc3 | 811 | mutex_unlock(&kernfs_mutex); |
988cd7af | 812 | return ret; |
fd7b9f7b TH |
813 | } |
814 | ||
815 | /** | |
324a56e1 TH |
816 | * kernfs_find_ns - find kernfs_node with the given name |
817 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
818 | * @name: name to look for |
819 | * @ns: the namespace tag to use | |
820 | * | |
324a56e1 TH |
821 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
822 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 823 | */ |
324a56e1 TH |
824 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
825 | const unsigned char *name, | |
826 | const void *ns) | |
fd7b9f7b | 827 | { |
adc5e8b5 | 828 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 829 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
830 | unsigned int hash; |
831 | ||
a797bfc3 | 832 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
833 | |
834 | if (has_ns != (bool)ns) { | |
c637b8ac | 835 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 836 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
837 | return NULL; |
838 | } | |
839 | ||
c637b8ac | 840 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 841 | while (node) { |
324a56e1 | 842 | struct kernfs_node *kn; |
fd7b9f7b TH |
843 | int result; |
844 | ||
324a56e1 | 845 | kn = rb_to_kn(node); |
c637b8ac | 846 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
847 | if (result < 0) |
848 | node = node->rb_left; | |
849 | else if (result > 0) | |
850 | node = node->rb_right; | |
851 | else | |
324a56e1 | 852 | return kn; |
fd7b9f7b TH |
853 | } |
854 | return NULL; | |
855 | } | |
856 | ||
bd96f76a TH |
857 | static struct kernfs_node *kernfs_walk_ns(struct kernfs_node *parent, |
858 | const unsigned char *path, | |
859 | const void *ns) | |
860 | { | |
e56ed358 TH |
861 | size_t len; |
862 | char *p, *name; | |
bd96f76a TH |
863 | |
864 | lockdep_assert_held(&kernfs_mutex); | |
865 | ||
e56ed358 TH |
866 | /* grab kernfs_rename_lock to piggy back on kernfs_pr_cont_buf */ |
867 | spin_lock_irq(&kernfs_rename_lock); | |
868 | ||
869 | len = strlcpy(kernfs_pr_cont_buf, path, sizeof(kernfs_pr_cont_buf)); | |
870 | ||
871 | if (len >= sizeof(kernfs_pr_cont_buf)) { | |
872 | spin_unlock_irq(&kernfs_rename_lock); | |
bd96f76a | 873 | return NULL; |
e56ed358 TH |
874 | } |
875 | ||
876 | p = kernfs_pr_cont_buf; | |
bd96f76a TH |
877 | |
878 | while ((name = strsep(&p, "/")) && parent) { | |
879 | if (*name == '\0') | |
880 | continue; | |
881 | parent = kernfs_find_ns(parent, name, ns); | |
882 | } | |
883 | ||
e56ed358 TH |
884 | spin_unlock_irq(&kernfs_rename_lock); |
885 | ||
bd96f76a TH |
886 | return parent; |
887 | } | |
888 | ||
fd7b9f7b | 889 | /** |
324a56e1 TH |
890 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
891 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
892 | * @name: name to look for |
893 | * @ns: the namespace tag to use | |
894 | * | |
324a56e1 | 895 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 896 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 897 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 898 | */ |
324a56e1 TH |
899 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
900 | const char *name, const void *ns) | |
fd7b9f7b | 901 | { |
324a56e1 | 902 | struct kernfs_node *kn; |
fd7b9f7b | 903 | |
a797bfc3 | 904 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
905 | kn = kernfs_find_ns(parent, name, ns); |
906 | kernfs_get(kn); | |
a797bfc3 | 907 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 908 | |
324a56e1 | 909 | return kn; |
fd7b9f7b TH |
910 | } |
911 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
912 | ||
bd96f76a TH |
913 | /** |
914 | * kernfs_walk_and_get_ns - find and get kernfs_node with the given path | |
915 | * @parent: kernfs_node to search under | |
916 | * @path: path to look for | |
917 | * @ns: the namespace tag to use | |
918 | * | |
919 | * Look for kernfs_node with path @path under @parent and get a reference | |
920 | * if found. This function may sleep and returns pointer to the found | |
921 | * kernfs_node on success, %NULL on failure. | |
922 | */ | |
923 | struct kernfs_node *kernfs_walk_and_get_ns(struct kernfs_node *parent, | |
924 | const char *path, const void *ns) | |
925 | { | |
926 | struct kernfs_node *kn; | |
927 | ||
928 | mutex_lock(&kernfs_mutex); | |
929 | kn = kernfs_walk_ns(parent, path, ns); | |
930 | kernfs_get(kn); | |
931 | mutex_unlock(&kernfs_mutex); | |
932 | ||
933 | return kn; | |
934 | } | |
935 | ||
ba7443bc TH |
936 | /** |
937 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 938 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 939 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
940 | * @priv: opaque data associated with the new directory |
941 | * | |
942 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
943 | * failure. | |
944 | */ | |
90c07c89 | 945 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 946 | unsigned int flags, void *priv) |
ba7443bc TH |
947 | { |
948 | struct kernfs_root *root; | |
324a56e1 | 949 | struct kernfs_node *kn; |
ba7443bc TH |
950 | |
951 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
952 | if (!root) | |
953 | return ERR_PTR(-ENOMEM); | |
954 | ||
7d35079f | 955 | idr_init(&root->ino_idr); |
7d568a83 | 956 | INIT_LIST_HEAD(&root->supers); |
40430452 TH |
957 | |
958 | /* | |
959 | * On 64bit ino setups, id is ino. On 32bit, low 32bits are ino. | |
960 | * High bits generation. The starting value for both ino and | |
961 | * genenration is 1. Initialize upper 32bit allocation | |
962 | * accordingly. | |
963 | */ | |
964 | if (sizeof(ino_t) >= sizeof(u64)) | |
965 | root->id_highbits = 0; | |
966 | else | |
967 | root->id_highbits = 1; | |
bc755553 | 968 | |
e19dfdc8 | 969 | kn = __kernfs_new_node(root, NULL, "", S_IFDIR | S_IRUGO | S_IXUGO, |
488dee96 | 970 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, |
db4aad20 | 971 | KERNFS_DIR); |
324a56e1 | 972 | if (!kn) { |
7d35079f | 973 | idr_destroy(&root->ino_idr); |
ba7443bc TH |
974 | kfree(root); |
975 | return ERR_PTR(-ENOMEM); | |
976 | } | |
977 | ||
324a56e1 | 978 | kn->priv = priv; |
adc5e8b5 | 979 | kn->dir.root = root; |
ba7443bc | 980 | |
90c07c89 | 981 | root->syscall_ops = scops; |
d35258ef | 982 | root->flags = flags; |
324a56e1 | 983 | root->kn = kn; |
abd54f02 | 984 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 985 | |
d35258ef TH |
986 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
987 | kernfs_activate(kn); | |
988 | ||
ba7443bc TH |
989 | return root; |
990 | } | |
991 | ||
992 | /** | |
993 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
994 | * @root: root of the hierarchy to destroy | |
995 | * | |
996 | * Destroy the hierarchy anchored at @root by removing all existing | |
997 | * directories and destroying @root. | |
998 | */ | |
999 | void kernfs_destroy_root(struct kernfs_root *root) | |
1000 | { | |
324a56e1 | 1001 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
1002 | } |
1003 | ||
fd7b9f7b TH |
1004 | /** |
1005 | * kernfs_create_dir_ns - create a directory | |
1006 | * @parent: parent in which to create a new directory | |
1007 | * @name: name of the new directory | |
bb8b9d09 | 1008 | * @mode: mode of the new directory |
488dee96 DT |
1009 | * @uid: uid of the new directory |
1010 | * @gid: gid of the new directory | |
fd7b9f7b TH |
1011 | * @priv: opaque data associated with the new directory |
1012 | * @ns: optional namespace tag of the directory | |
1013 | * | |
1014 | * Returns the created node on success, ERR_PTR() value on failure. | |
1015 | */ | |
324a56e1 | 1016 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 | 1017 | const char *name, umode_t mode, |
488dee96 | 1018 | kuid_t uid, kgid_t gid, |
bb8b9d09 | 1019 | void *priv, const void *ns) |
fd7b9f7b | 1020 | { |
324a56e1 | 1021 | struct kernfs_node *kn; |
fd7b9f7b TH |
1022 | int rc; |
1023 | ||
1024 | /* allocate */ | |
488dee96 DT |
1025 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, |
1026 | uid, gid, KERNFS_DIR); | |
324a56e1 | 1027 | if (!kn) |
fd7b9f7b TH |
1028 | return ERR_PTR(-ENOMEM); |
1029 | ||
adc5e8b5 TH |
1030 | kn->dir.root = parent->dir.root; |
1031 | kn->ns = ns; | |
324a56e1 | 1032 | kn->priv = priv; |
fd7b9f7b TH |
1033 | |
1034 | /* link in */ | |
988cd7af | 1035 | rc = kernfs_add_one(kn); |
fd7b9f7b | 1036 | if (!rc) |
324a56e1 | 1037 | return kn; |
fd7b9f7b | 1038 | |
324a56e1 | 1039 | kernfs_put(kn); |
fd7b9f7b TH |
1040 | return ERR_PTR(rc); |
1041 | } | |
1042 | ||
ea015218 EB |
1043 | /** |
1044 | * kernfs_create_empty_dir - create an always empty directory | |
1045 | * @parent: parent in which to create a new directory | |
1046 | * @name: name of the new directory | |
1047 | * | |
1048 | * Returns the created node on success, ERR_PTR() value on failure. | |
1049 | */ | |
1050 | struct kernfs_node *kernfs_create_empty_dir(struct kernfs_node *parent, | |
1051 | const char *name) | |
1052 | { | |
1053 | struct kernfs_node *kn; | |
1054 | int rc; | |
1055 | ||
1056 | /* allocate */ | |
488dee96 DT |
1057 | kn = kernfs_new_node(parent, name, S_IRUGO|S_IXUGO|S_IFDIR, |
1058 | GLOBAL_ROOT_UID, GLOBAL_ROOT_GID, KERNFS_DIR); | |
ea015218 EB |
1059 | if (!kn) |
1060 | return ERR_PTR(-ENOMEM); | |
1061 | ||
1062 | kn->flags |= KERNFS_EMPTY_DIR; | |
1063 | kn->dir.root = parent->dir.root; | |
1064 | kn->ns = NULL; | |
1065 | kn->priv = NULL; | |
1066 | ||
1067 | /* link in */ | |
1068 | rc = kernfs_add_one(kn); | |
1069 | if (!rc) | |
1070 | return kn; | |
1071 | ||
1072 | kernfs_put(kn); | |
1073 | return ERR_PTR(rc); | |
1074 | } | |
1075 | ||
c637b8ac TH |
1076 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
1077 | struct dentry *dentry, | |
1078 | unsigned int flags) | |
fd7b9f7b | 1079 | { |
19bbb926 | 1080 | struct dentry *ret; |
319ba91d | 1081 | struct kernfs_node *parent = dir->i_private; |
324a56e1 | 1082 | struct kernfs_node *kn; |
fd7b9f7b TH |
1083 | struct inode *inode; |
1084 | const void *ns = NULL; | |
1085 | ||
a797bfc3 | 1086 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1087 | |
324a56e1 | 1088 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1089 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 1090 | |
324a56e1 | 1091 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
1092 | |
1093 | /* no such entry */ | |
b9c9dad0 | 1094 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 1095 | ret = NULL; |
fd7b9f7b TH |
1096 | goto out_unlock; |
1097 | } | |
fd7b9f7b TH |
1098 | |
1099 | /* attach dentry and inode */ | |
c637b8ac | 1100 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
1101 | if (!inode) { |
1102 | ret = ERR_PTR(-ENOMEM); | |
1103 | goto out_unlock; | |
1104 | } | |
1105 | ||
1106 | /* instantiate and hash dentry */ | |
41d28bca | 1107 | ret = d_splice_alias(inode, dentry); |
fd7b9f7b | 1108 | out_unlock: |
a797bfc3 | 1109 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1110 | return ret; |
1111 | } | |
1112 | ||
80b9bbef TH |
1113 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
1114 | umode_t mode) | |
1115 | { | |
1116 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 1117 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 1118 | int ret; |
80b9bbef | 1119 | |
90c07c89 | 1120 | if (!scops || !scops->mkdir) |
80b9bbef TH |
1121 | return -EPERM; |
1122 | ||
07c7530d TH |
1123 | if (!kernfs_get_active(parent)) |
1124 | return -ENODEV; | |
1125 | ||
90c07c89 | 1126 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
1127 | |
1128 | kernfs_put_active(parent); | |
1129 | return ret; | |
80b9bbef TH |
1130 | } |
1131 | ||
1132 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
1133 | { | |
319ba91d | 1134 | struct kernfs_node *kn = kernfs_dentry_node(dentry); |
90c07c89 | 1135 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1136 | int ret; |
80b9bbef | 1137 | |
90c07c89 | 1138 | if (!scops || !scops->rmdir) |
80b9bbef TH |
1139 | return -EPERM; |
1140 | ||
07c7530d TH |
1141 | if (!kernfs_get_active(kn)) |
1142 | return -ENODEV; | |
1143 | ||
90c07c89 | 1144 | ret = scops->rmdir(kn); |
07c7530d TH |
1145 | |
1146 | kernfs_put_active(kn); | |
1147 | return ret; | |
80b9bbef TH |
1148 | } |
1149 | ||
1150 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
1cd66c93 MS |
1151 | struct inode *new_dir, struct dentry *new_dentry, |
1152 | unsigned int flags) | |
80b9bbef | 1153 | { |
319ba91d | 1154 | struct kernfs_node *kn = kernfs_dentry_node(old_dentry); |
80b9bbef | 1155 | struct kernfs_node *new_parent = new_dir->i_private; |
90c07c89 | 1156 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 1157 | int ret; |
80b9bbef | 1158 | |
1cd66c93 MS |
1159 | if (flags) |
1160 | return -EINVAL; | |
1161 | ||
90c07c89 | 1162 | if (!scops || !scops->rename) |
80b9bbef TH |
1163 | return -EPERM; |
1164 | ||
07c7530d TH |
1165 | if (!kernfs_get_active(kn)) |
1166 | return -ENODEV; | |
1167 | ||
1168 | if (!kernfs_get_active(new_parent)) { | |
1169 | kernfs_put_active(kn); | |
1170 | return -ENODEV; | |
1171 | } | |
1172 | ||
90c07c89 | 1173 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
1174 | |
1175 | kernfs_put_active(new_parent); | |
1176 | kernfs_put_active(kn); | |
1177 | return ret; | |
80b9bbef TH |
1178 | } |
1179 | ||
a797bfc3 | 1180 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
1181 | .lookup = kernfs_iop_lookup, |
1182 | .permission = kernfs_iop_permission, | |
1183 | .setattr = kernfs_iop_setattr, | |
1184 | .getattr = kernfs_iop_getattr, | |
c637b8ac | 1185 | .listxattr = kernfs_iop_listxattr, |
80b9bbef TH |
1186 | |
1187 | .mkdir = kernfs_iop_mkdir, | |
1188 | .rmdir = kernfs_iop_rmdir, | |
1189 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
1190 | }; |
1191 | ||
c637b8ac | 1192 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 1193 | { |
324a56e1 | 1194 | struct kernfs_node *last; |
fd7b9f7b TH |
1195 | |
1196 | while (true) { | |
1197 | struct rb_node *rbn; | |
1198 | ||
1199 | last = pos; | |
1200 | ||
df23fc39 | 1201 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
1202 | break; |
1203 | ||
adc5e8b5 | 1204 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
1205 | if (!rbn) |
1206 | break; | |
1207 | ||
324a56e1 | 1208 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
1209 | } |
1210 | ||
1211 | return last; | |
1212 | } | |
1213 | ||
1214 | /** | |
c637b8ac | 1215 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 1216 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 1217 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
1218 | * |
1219 | * Find the next descendant to visit for post-order traversal of @root's | |
1220 | * descendants. @root is included in the iteration and the last node to be | |
1221 | * visited. | |
1222 | */ | |
c637b8ac TH |
1223 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
1224 | struct kernfs_node *root) | |
fd7b9f7b TH |
1225 | { |
1226 | struct rb_node *rbn; | |
1227 | ||
a797bfc3 | 1228 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
1229 | |
1230 | /* if first iteration, visit leftmost descendant which may be root */ | |
1231 | if (!pos) | |
c637b8ac | 1232 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
1233 | |
1234 | /* if we visited @root, we're done */ | |
1235 | if (pos == root) | |
1236 | return NULL; | |
1237 | ||
1238 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 1239 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 1240 | if (rbn) |
c637b8ac | 1241 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
1242 | |
1243 | /* no sibling left, visit parent */ | |
adc5e8b5 | 1244 | return pos->parent; |
fd7b9f7b TH |
1245 | } |
1246 | ||
d35258ef TH |
1247 | /** |
1248 | * kernfs_activate - activate a node which started deactivated | |
1249 | * @kn: kernfs_node whose subtree is to be activated | |
1250 | * | |
1251 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
1252 | * needs to be explicitly activated. A node which hasn't been activated | |
1253 | * isn't visible to userland and deactivation is skipped during its | |
1254 | * removal. This is useful to construct atomic init sequences where | |
1255 | * creation of multiple nodes should either succeed or fail atomically. | |
1256 | * | |
1257 | * The caller is responsible for ensuring that this function is not called | |
1258 | * after kernfs_remove*() is invoked on @kn. | |
1259 | */ | |
1260 | void kernfs_activate(struct kernfs_node *kn) | |
1261 | { | |
1262 | struct kernfs_node *pos; | |
1263 | ||
1264 | mutex_lock(&kernfs_mutex); | |
1265 | ||
1266 | pos = NULL; | |
1267 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
5bf33f04 | 1268 | if (pos->flags & KERNFS_ACTIVATED) |
d35258ef TH |
1269 | continue; |
1270 | ||
1271 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
1272 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
1273 | ||
1274 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
1275 | pos->flags |= KERNFS_ACTIVATED; | |
1276 | } | |
1277 | ||
1278 | mutex_unlock(&kernfs_mutex); | |
1279 | } | |
1280 | ||
988cd7af | 1281 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1282 | { |
35beab06 TH |
1283 | struct kernfs_node *pos; |
1284 | ||
1285 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 1286 | |
6b0afc2a TH |
1287 | /* |
1288 | * Short-circuit if non-root @kn has already finished removal. | |
1289 | * This is for kernfs_remove_self() which plays with active ref | |
1290 | * after removal. | |
1291 | */ | |
1292 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
1293 | return; |
1294 | ||
c637b8ac | 1295 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 1296 | |
81c173cb | 1297 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
1298 | pos = NULL; |
1299 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
1300 | if (kernfs_active(pos)) |
1301 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
1302 | |
1303 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 1304 | do { |
35beab06 TH |
1305 | pos = kernfs_leftmost_descendant(kn); |
1306 | ||
1307 | /* | |
81c173cb TH |
1308 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
1309 | * base ref could have been put by someone else by the time | |
1310 | * the function returns. Make sure it doesn't go away | |
1311 | * underneath us. | |
35beab06 TH |
1312 | */ |
1313 | kernfs_get(pos); | |
1314 | ||
d35258ef TH |
1315 | /* |
1316 | * Drain iff @kn was activated. This avoids draining and | |
1317 | * its lockdep annotations for nodes which have never been | |
1318 | * activated and allows embedding kernfs_remove() in create | |
1319 | * error paths without worrying about draining. | |
1320 | */ | |
1321 | if (kn->flags & KERNFS_ACTIVATED) | |
1322 | kernfs_drain(pos); | |
1323 | else | |
1324 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
1325 | |
1326 | /* | |
1327 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
1328 | * to decide who's responsible for cleanups. | |
1329 | */ | |
1330 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
1331 | struct kernfs_iattrs *ps_iattr = | |
1332 | pos->parent ? pos->parent->iattr : NULL; | |
1333 | ||
1334 | /* update timestamps on the parent */ | |
1335 | if (ps_iattr) { | |
05895219 OM |
1336 | ktime_get_real_ts64(&ps_iattr->ia_ctime); |
1337 | ps_iattr->ia_mtime = ps_iattr->ia_ctime; | |
35beab06 TH |
1338 | } |
1339 | ||
988cd7af | 1340 | kernfs_put(pos); |
35beab06 TH |
1341 | } |
1342 | ||
1343 | kernfs_put(pos); | |
1344 | } while (pos != kn); | |
fd7b9f7b TH |
1345 | } |
1346 | ||
1347 | /** | |
324a56e1 TH |
1348 | * kernfs_remove - remove a kernfs_node recursively |
1349 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 1350 | * |
324a56e1 | 1351 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 1352 | */ |
324a56e1 | 1353 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1354 | { |
988cd7af TH |
1355 | mutex_lock(&kernfs_mutex); |
1356 | __kernfs_remove(kn); | |
1357 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
1358 | } |
1359 | ||
6b0afc2a TH |
1360 | /** |
1361 | * kernfs_break_active_protection - break out of active protection | |
1362 | * @kn: the self kernfs_node | |
1363 | * | |
1364 | * The caller must be running off of a kernfs operation which is invoked | |
1365 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
1366 | * this function must also be matched with an invocation of | |
1367 | * kernfs_unbreak_active_protection(). | |
1368 | * | |
1369 | * This function releases the active reference of @kn the caller is | |
1370 | * holding. Once this function is called, @kn may be removed at any point | |
1371 | * and the caller is solely responsible for ensuring that the objects it | |
1372 | * dereferences are accessible. | |
1373 | */ | |
1374 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
1375 | { | |
1376 | /* | |
1377 | * Take out ourself out of the active ref dependency chain. If | |
1378 | * we're called without an active ref, lockdep will complain. | |
1379 | */ | |
1380 | kernfs_put_active(kn); | |
1381 | } | |
1382 | ||
1383 | /** | |
1384 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
1385 | * @kn: the self kernfs_node | |
1386 | * | |
1387 | * If kernfs_break_active_protection() was called, this function must be | |
1388 | * invoked before finishing the kernfs operation. Note that while this | |
1389 | * function restores the active reference, it doesn't and can't actually | |
1390 | * restore the active protection - @kn may already or be in the process of | |
1391 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
1392 | * protection is irreversibly gone for the kernfs operation instance. | |
1393 | * | |
1394 | * While this function may be called at any point after | |
1395 | * kernfs_break_active_protection() is invoked, its most useful location | |
1396 | * would be right before the enclosing kernfs operation returns. | |
1397 | */ | |
1398 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
1399 | { | |
1400 | /* | |
1401 | * @kn->active could be in any state; however, the increment we do | |
1402 | * here will be undone as soon as the enclosing kernfs operation | |
1403 | * finishes and this temporary bump can't break anything. If @kn | |
1404 | * is alive, nothing changes. If @kn is being deactivated, the | |
1405 | * soon-to-follow put will either finish deactivation or restore | |
1406 | * deactivated state. If @kn is already removed, the temporary | |
1407 | * bump is guaranteed to be gone before @kn is released. | |
1408 | */ | |
1409 | atomic_inc(&kn->active); | |
1410 | if (kernfs_lockdep(kn)) | |
1411 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
1412 | } | |
1413 | ||
1414 | /** | |
1415 | * kernfs_remove_self - remove a kernfs_node from its own method | |
1416 | * @kn: the self kernfs_node to remove | |
1417 | * | |
1418 | * The caller must be running off of a kernfs operation which is invoked | |
1419 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
1420 | * implement a file operation which deletes itself. | |
1421 | * | |
1422 | * For example, the "delete" file for a sysfs device directory can be | |
1423 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
1424 | * itself. This function breaks the circular dependency of trying to | |
1425 | * deactivate self while holding an active ref itself. It isn't necessary | |
1426 | * to modify the usual removal path to use kernfs_remove_self(). The | |
1427 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
1428 | * before proceeding with the usual removal path. kernfs will ignore later | |
1429 | * kernfs_remove() on self. | |
1430 | * | |
1431 | * kernfs_remove_self() can be called multiple times concurrently on the | |
1432 | * same kernfs_node. Only the first one actually performs removal and | |
1433 | * returns %true. All others will wait until the kernfs operation which | |
1434 | * won self-removal finishes and return %false. Note that the losers wait | |
1435 | * for the completion of not only the winning kernfs_remove_self() but also | |
1436 | * the whole kernfs_ops which won the arbitration. This can be used to | |
1437 | * guarantee, for example, all concurrent writes to a "delete" file to | |
1438 | * finish only after the whole operation is complete. | |
1439 | */ | |
1440 | bool kernfs_remove_self(struct kernfs_node *kn) | |
1441 | { | |
1442 | bool ret; | |
1443 | ||
1444 | mutex_lock(&kernfs_mutex); | |
1445 | kernfs_break_active_protection(kn); | |
1446 | ||
1447 | /* | |
1448 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1449 | * the first one will actually perform removal. When the removal | |
1450 | * is complete, SUICIDED is set and the active ref is restored | |
1451 | * while holding kernfs_mutex. The ones which lost arbitration | |
1452 | * waits for SUICDED && drained which can happen only after the | |
1453 | * enclosing kernfs operation which executed the winning instance | |
1454 | * of kernfs_remove_self() finished. | |
1455 | */ | |
1456 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1457 | kn->flags |= KERNFS_SUICIDAL; | |
1458 | __kernfs_remove(kn); | |
1459 | kn->flags |= KERNFS_SUICIDED; | |
1460 | ret = true; | |
1461 | } else { | |
1462 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1463 | DEFINE_WAIT(wait); | |
1464 | ||
1465 | while (true) { | |
1466 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1467 | ||
1468 | if ((kn->flags & KERNFS_SUICIDED) && | |
1469 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1470 | break; | |
1471 | ||
1472 | mutex_unlock(&kernfs_mutex); | |
1473 | schedule(); | |
1474 | mutex_lock(&kernfs_mutex); | |
1475 | } | |
1476 | finish_wait(waitq, &wait); | |
1477 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1478 | ret = false; | |
1479 | } | |
1480 | ||
1481 | /* | |
1482 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1483 | * for SUICIDED && deactivated could finish prematurely. | |
1484 | */ | |
1485 | kernfs_unbreak_active_protection(kn); | |
1486 | ||
1487 | mutex_unlock(&kernfs_mutex); | |
1488 | return ret; | |
1489 | } | |
1490 | ||
fd7b9f7b | 1491 | /** |
324a56e1 TH |
1492 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1493 | * @parent: parent of the target | |
1494 | * @name: name of the kernfs_node to remove | |
1495 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1496 | * |
324a56e1 TH |
1497 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1498 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1499 | */ |
324a56e1 | 1500 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1501 | const void *ns) |
1502 | { | |
324a56e1 | 1503 | struct kernfs_node *kn; |
fd7b9f7b | 1504 | |
324a56e1 | 1505 | if (!parent) { |
c637b8ac | 1506 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1507 | name); |
1508 | return -ENOENT; | |
1509 | } | |
1510 | ||
988cd7af | 1511 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1512 | |
324a56e1 TH |
1513 | kn = kernfs_find_ns(parent, name, ns); |
1514 | if (kn) | |
988cd7af | 1515 | __kernfs_remove(kn); |
fd7b9f7b | 1516 | |
988cd7af | 1517 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1518 | |
324a56e1 | 1519 | if (kn) |
fd7b9f7b TH |
1520 | return 0; |
1521 | else | |
1522 | return -ENOENT; | |
1523 | } | |
1524 | ||
1525 | /** | |
1526 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1527 | * @kn: target node |
fd7b9f7b TH |
1528 | * @new_parent: new parent to put @sd under |
1529 | * @new_name: new name | |
1530 | * @new_ns: new namespace tag | |
1531 | */ | |
324a56e1 | 1532 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1533 | const char *new_name, const void *new_ns) |
1534 | { | |
3eef34ad TH |
1535 | struct kernfs_node *old_parent; |
1536 | const char *old_name = NULL; | |
fd7b9f7b TH |
1537 | int error; |
1538 | ||
3eef34ad TH |
1539 | /* can't move or rename root */ |
1540 | if (!kn->parent) | |
1541 | return -EINVAL; | |
1542 | ||
798c75a0 GKH |
1543 | mutex_lock(&kernfs_mutex); |
1544 | ||
d0ae3d43 | 1545 | error = -ENOENT; |
ea015218 EB |
1546 | if (!kernfs_active(kn) || !kernfs_active(new_parent) || |
1547 | (new_parent->flags & KERNFS_EMPTY_DIR)) | |
d0ae3d43 TH |
1548 | goto out; |
1549 | ||
fd7b9f7b | 1550 | error = 0; |
adc5e8b5 TH |
1551 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1552 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1553 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1554 | |
1555 | error = -EEXIST; | |
1556 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1557 | goto out; |
fd7b9f7b | 1558 | |
324a56e1 | 1559 | /* rename kernfs_node */ |
adc5e8b5 | 1560 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b | 1561 | error = -ENOMEM; |
75287a67 | 1562 | new_name = kstrdup_const(new_name, GFP_KERNEL); |
fd7b9f7b | 1563 | if (!new_name) |
798c75a0 | 1564 | goto out; |
3eef34ad TH |
1565 | } else { |
1566 | new_name = NULL; | |
fd7b9f7b TH |
1567 | } |
1568 | ||
1569 | /* | |
1570 | * Move to the appropriate place in the appropriate directories rbtree. | |
1571 | */ | |
c637b8ac | 1572 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1573 | kernfs_get(new_parent); |
3eef34ad TH |
1574 | |
1575 | /* rename_lock protects ->parent and ->name accessors */ | |
1576 | spin_lock_irq(&kernfs_rename_lock); | |
1577 | ||
1578 | old_parent = kn->parent; | |
adc5e8b5 | 1579 | kn->parent = new_parent; |
3eef34ad TH |
1580 | |
1581 | kn->ns = new_ns; | |
1582 | if (new_name) { | |
dfeb0750 | 1583 | old_name = kn->name; |
3eef34ad TH |
1584 | kn->name = new_name; |
1585 | } | |
1586 | ||
1587 | spin_unlock_irq(&kernfs_rename_lock); | |
1588 | ||
9561a896 | 1589 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
c637b8ac | 1590 | kernfs_link_sibling(kn); |
fd7b9f7b | 1591 | |
3eef34ad | 1592 | kernfs_put(old_parent); |
75287a67 | 1593 | kfree_const(old_name); |
3eef34ad | 1594 | |
fd7b9f7b | 1595 | error = 0; |
798c75a0 | 1596 | out: |
a797bfc3 | 1597 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1598 | return error; |
1599 | } | |
1600 | ||
21774fd8 | 1601 | /* Relationship between mode and the DT_xxx types */ |
324a56e1 | 1602 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1603 | { |
adc5e8b5 | 1604 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1605 | } |
1606 | ||
c637b8ac | 1607 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1608 | { |
1609 | kernfs_put(filp->private_data); | |
1610 | return 0; | |
1611 | } | |
1612 | ||
c637b8ac | 1613 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1614 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1615 | { |
1616 | if (pos) { | |
81c173cb | 1617 | int valid = kernfs_active(pos) && |
798c75a0 | 1618 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1619 | kernfs_put(pos); |
1620 | if (!valid) | |
1621 | pos = NULL; | |
1622 | } | |
1623 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1624 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1625 | while (node) { |
324a56e1 | 1626 | pos = rb_to_kn(node); |
fd7b9f7b | 1627 | |
adc5e8b5 | 1628 | if (hash < pos->hash) |
fd7b9f7b | 1629 | node = node->rb_left; |
adc5e8b5 | 1630 | else if (hash > pos->hash) |
fd7b9f7b TH |
1631 | node = node->rb_right; |
1632 | else | |
1633 | break; | |
1634 | } | |
1635 | } | |
b9c9dad0 TH |
1636 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1637 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1638 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1639 | if (!node) |
1640 | pos = NULL; | |
1641 | else | |
324a56e1 | 1642 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1643 | } |
1644 | return pos; | |
1645 | } | |
1646 | ||
c637b8ac | 1647 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1648 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1649 | { |
c637b8ac | 1650 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1651 | if (pos) { |
fd7b9f7b | 1652 | do { |
adc5e8b5 | 1653 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1654 | if (!node) |
1655 | pos = NULL; | |
1656 | else | |
324a56e1 | 1657 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1658 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1659 | } | |
fd7b9f7b TH |
1660 | return pos; |
1661 | } | |
1662 | ||
c637b8ac | 1663 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1664 | { |
1665 | struct dentry *dentry = file->f_path.dentry; | |
319ba91d | 1666 | struct kernfs_node *parent = kernfs_dentry_node(dentry); |
324a56e1 | 1667 | struct kernfs_node *pos = file->private_data; |
fd7b9f7b TH |
1668 | const void *ns = NULL; |
1669 | ||
1670 | if (!dir_emit_dots(file, ctx)) | |
1671 | return 0; | |
a797bfc3 | 1672 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1673 | |
324a56e1 | 1674 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1675 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1676 | |
c637b8ac | 1677 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1678 | pos; |
c637b8ac | 1679 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1680 | const char *name = pos->name; |
fd7b9f7b TH |
1681 | unsigned int type = dt_type(pos); |
1682 | int len = strlen(name); | |
67c0496e | 1683 | ino_t ino = kernfs_ino(pos); |
fd7b9f7b | 1684 | |
adc5e8b5 | 1685 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1686 | file->private_data = pos; |
1687 | kernfs_get(pos); | |
1688 | ||
a797bfc3 | 1689 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1690 | if (!dir_emit(ctx, name, len, ino, type)) |
1691 | return 0; | |
a797bfc3 | 1692 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1693 | } |
a797bfc3 | 1694 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1695 | file->private_data = NULL; |
1696 | ctx->pos = INT_MAX; | |
1697 | return 0; | |
1698 | } | |
1699 | ||
a797bfc3 | 1700 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1701 | .read = generic_read_dir, |
8cb0d2c1 | 1702 | .iterate_shared = kernfs_fop_readdir, |
c637b8ac | 1703 | .release = kernfs_dir_fop_release, |
8cb0d2c1 | 1704 | .llseek = generic_file_llseek, |
fd7b9f7b | 1705 | }; |