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