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 | } | |
115 | ||
116 | /** | |
117 | * pr_cont_kernfs_name - pr_cont name of a kernfs_node | |
118 | * @kn: kernfs_node of interest | |
119 | * | |
120 | * This function can be called from any context. | |
121 | */ | |
122 | void pr_cont_kernfs_name(struct kernfs_node *kn) | |
123 | { | |
124 | unsigned long flags; | |
125 | ||
126 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
127 | ||
128 | kernfs_name_locked(kn, kernfs_pr_cont_buf, sizeof(kernfs_pr_cont_buf)); | |
129 | pr_cont("%s", kernfs_pr_cont_buf); | |
130 | ||
131 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
132 | } | |
133 | ||
134 | /** | |
135 | * pr_cont_kernfs_path - pr_cont path of a kernfs_node | |
136 | * @kn: kernfs_node of interest | |
137 | * | |
138 | * This function can be called from any context. | |
139 | */ | |
140 | void pr_cont_kernfs_path(struct kernfs_node *kn) | |
141 | { | |
142 | unsigned long flags; | |
143 | char *p; | |
144 | ||
145 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
146 | ||
147 | p = kernfs_path_locked(kn, kernfs_pr_cont_buf, | |
148 | sizeof(kernfs_pr_cont_buf)); | |
149 | if (p) | |
150 | pr_cont("%s", p); | |
151 | else | |
152 | pr_cont("<name too long>"); | |
153 | ||
154 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
155 | } | |
156 | ||
157 | /** | |
158 | * kernfs_get_parent - determine the parent node and pin it | |
159 | * @kn: kernfs_node of interest | |
160 | * | |
161 | * Determines @kn's parent, pins and returns it. This function can be | |
162 | * called from any context. | |
163 | */ | |
164 | struct kernfs_node *kernfs_get_parent(struct kernfs_node *kn) | |
165 | { | |
166 | struct kernfs_node *parent; | |
167 | unsigned long flags; | |
168 | ||
169 | spin_lock_irqsave(&kernfs_rename_lock, flags); | |
170 | parent = kn->parent; | |
171 | kernfs_get(parent); | |
172 | spin_unlock_irqrestore(&kernfs_rename_lock, flags); | |
173 | ||
174 | return parent; | |
175 | } | |
176 | ||
fd7b9f7b | 177 | /** |
c637b8ac | 178 | * kernfs_name_hash |
fd7b9f7b TH |
179 | * @name: Null terminated string to hash |
180 | * @ns: Namespace tag to hash | |
181 | * | |
182 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | |
183 | */ | |
c637b8ac | 184 | static unsigned int kernfs_name_hash(const char *name, const void *ns) |
fd7b9f7b TH |
185 | { |
186 | unsigned long hash = init_name_hash(); | |
187 | unsigned int len = strlen(name); | |
188 | while (len--) | |
189 | hash = partial_name_hash(*name++, hash); | |
190 | hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); | |
191 | hash &= 0x7fffffffU; | |
192 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | |
193 | if (hash < 1) | |
194 | hash += 2; | |
195 | if (hash >= INT_MAX) | |
196 | hash = INT_MAX - 1; | |
197 | return hash; | |
198 | } | |
199 | ||
c637b8ac TH |
200 | static int kernfs_name_compare(unsigned int hash, const char *name, |
201 | const void *ns, const struct kernfs_node *kn) | |
fd7b9f7b | 202 | { |
adc5e8b5 TH |
203 | if (hash != kn->hash) |
204 | return hash - kn->hash; | |
205 | if (ns != kn->ns) | |
206 | return ns - kn->ns; | |
207 | return strcmp(name, kn->name); | |
fd7b9f7b TH |
208 | } |
209 | ||
c637b8ac TH |
210 | static int kernfs_sd_compare(const struct kernfs_node *left, |
211 | const struct kernfs_node *right) | |
fd7b9f7b | 212 | { |
c637b8ac | 213 | return kernfs_name_compare(left->hash, left->name, left->ns, right); |
fd7b9f7b TH |
214 | } |
215 | ||
216 | /** | |
c637b8ac | 217 | * kernfs_link_sibling - link kernfs_node into sibling rbtree |
324a56e1 | 218 | * @kn: kernfs_node of interest |
fd7b9f7b | 219 | * |
324a56e1 | 220 | * Link @kn into its sibling rbtree which starts from |
adc5e8b5 | 221 | * @kn->parent->dir.children. |
fd7b9f7b TH |
222 | * |
223 | * Locking: | |
a797bfc3 | 224 | * mutex_lock(kernfs_mutex) |
fd7b9f7b TH |
225 | * |
226 | * RETURNS: | |
227 | * 0 on susccess -EEXIST on failure. | |
228 | */ | |
c637b8ac | 229 | static int kernfs_link_sibling(struct kernfs_node *kn) |
fd7b9f7b | 230 | { |
adc5e8b5 | 231 | struct rb_node **node = &kn->parent->dir.children.rb_node; |
fd7b9f7b TH |
232 | struct rb_node *parent = NULL; |
233 | ||
df23fc39 | 234 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 235 | kn->parent->dir.subdirs++; |
fd7b9f7b TH |
236 | |
237 | while (*node) { | |
324a56e1 | 238 | struct kernfs_node *pos; |
fd7b9f7b TH |
239 | int result; |
240 | ||
324a56e1 | 241 | pos = rb_to_kn(*node); |
fd7b9f7b | 242 | parent = *node; |
c637b8ac | 243 | result = kernfs_sd_compare(kn, pos); |
fd7b9f7b | 244 | if (result < 0) |
adc5e8b5 | 245 | node = &pos->rb.rb_left; |
fd7b9f7b | 246 | else if (result > 0) |
adc5e8b5 | 247 | node = &pos->rb.rb_right; |
fd7b9f7b TH |
248 | else |
249 | return -EEXIST; | |
250 | } | |
251 | /* add new node and rebalance the tree */ | |
adc5e8b5 TH |
252 | rb_link_node(&kn->rb, parent, node); |
253 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | |
fd7b9f7b TH |
254 | return 0; |
255 | } | |
256 | ||
257 | /** | |
c637b8ac | 258 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree |
324a56e1 | 259 | * @kn: kernfs_node of interest |
fd7b9f7b | 260 | * |
35beab06 TH |
261 | * Try to unlink @kn from its sibling rbtree which starts from |
262 | * kn->parent->dir.children. Returns %true if @kn was actually | |
263 | * removed, %false if @kn wasn't on the rbtree. | |
fd7b9f7b TH |
264 | * |
265 | * Locking: | |
a797bfc3 | 266 | * mutex_lock(kernfs_mutex) |
fd7b9f7b | 267 | */ |
35beab06 | 268 | static bool kernfs_unlink_sibling(struct kernfs_node *kn) |
fd7b9f7b | 269 | { |
35beab06 TH |
270 | if (RB_EMPTY_NODE(&kn->rb)) |
271 | return false; | |
272 | ||
df23fc39 | 273 | if (kernfs_type(kn) == KERNFS_DIR) |
adc5e8b5 | 274 | kn->parent->dir.subdirs--; |
fd7b9f7b | 275 | |
adc5e8b5 | 276 | rb_erase(&kn->rb, &kn->parent->dir.children); |
35beab06 TH |
277 | RB_CLEAR_NODE(&kn->rb); |
278 | return true; | |
fd7b9f7b TH |
279 | } |
280 | ||
281 | /** | |
c637b8ac | 282 | * kernfs_get_active - get an active reference to kernfs_node |
324a56e1 | 283 | * @kn: kernfs_node to get an active reference to |
fd7b9f7b | 284 | * |
324a56e1 | 285 | * Get an active reference of @kn. This function is noop if @kn |
fd7b9f7b TH |
286 | * is NULL. |
287 | * | |
288 | * RETURNS: | |
324a56e1 | 289 | * Pointer to @kn on success, NULL on failure. |
fd7b9f7b | 290 | */ |
c637b8ac | 291 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) |
fd7b9f7b | 292 | { |
324a56e1 | 293 | if (unlikely(!kn)) |
fd7b9f7b TH |
294 | return NULL; |
295 | ||
f4b3e631 GKH |
296 | if (!atomic_inc_unless_negative(&kn->active)) |
297 | return NULL; | |
895a068a | 298 | |
182fd64b | 299 | if (kernfs_lockdep(kn)) |
f4b3e631 GKH |
300 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); |
301 | return kn; | |
fd7b9f7b TH |
302 | } |
303 | ||
304 | /** | |
c637b8ac | 305 | * kernfs_put_active - put an active reference to kernfs_node |
324a56e1 | 306 | * @kn: kernfs_node to put an active reference to |
fd7b9f7b | 307 | * |
324a56e1 | 308 | * Put an active reference to @kn. This function is noop if @kn |
fd7b9f7b TH |
309 | * is NULL. |
310 | */ | |
c637b8ac | 311 | void kernfs_put_active(struct kernfs_node *kn) |
fd7b9f7b | 312 | { |
abd54f02 | 313 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b TH |
314 | int v; |
315 | ||
324a56e1 | 316 | if (unlikely(!kn)) |
fd7b9f7b TH |
317 | return; |
318 | ||
182fd64b | 319 | if (kernfs_lockdep(kn)) |
324a56e1 | 320 | rwsem_release(&kn->dep_map, 1, _RET_IP_); |
adc5e8b5 | 321 | v = atomic_dec_return(&kn->active); |
df23fc39 | 322 | if (likely(v != KN_DEACTIVATED_BIAS)) |
fd7b9f7b TH |
323 | return; |
324 | ||
abd54f02 | 325 | wake_up_all(&root->deactivate_waitq); |
fd7b9f7b TH |
326 | } |
327 | ||
328 | /** | |
81c173cb TH |
329 | * kernfs_drain - drain kernfs_node |
330 | * @kn: kernfs_node to drain | |
fd7b9f7b | 331 | * |
81c173cb TH |
332 | * Drain existing usages and nuke all existing mmaps of @kn. Mutiple |
333 | * removers may invoke this function concurrently on @kn and all will | |
334 | * return after draining is complete. | |
fd7b9f7b | 335 | */ |
81c173cb | 336 | static void kernfs_drain(struct kernfs_node *kn) |
35beab06 | 337 | __releases(&kernfs_mutex) __acquires(&kernfs_mutex) |
fd7b9f7b | 338 | { |
abd54f02 | 339 | struct kernfs_root *root = kernfs_root(kn); |
fd7b9f7b | 340 | |
35beab06 | 341 | lockdep_assert_held(&kernfs_mutex); |
81c173cb | 342 | WARN_ON_ONCE(kernfs_active(kn)); |
ea1c472d | 343 | |
35beab06 | 344 | mutex_unlock(&kernfs_mutex); |
abd54f02 | 345 | |
182fd64b | 346 | if (kernfs_lockdep(kn)) { |
35beab06 TH |
347 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); |
348 | if (atomic_read(&kn->active) != KN_DEACTIVATED_BIAS) | |
349 | lock_contended(&kn->dep_map, _RET_IP_); | |
350 | } | |
abd54f02 | 351 | |
35beab06 | 352 | /* but everyone should wait for draining */ |
abd54f02 TH |
353 | wait_event(root->deactivate_waitq, |
354 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS); | |
fd7b9f7b | 355 | |
182fd64b | 356 | if (kernfs_lockdep(kn)) { |
a6607930 TH |
357 | lock_acquired(&kn->dep_map, _RET_IP_); |
358 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | |
359 | } | |
35beab06 | 360 | |
ccf02aaf TH |
361 | kernfs_unmap_bin_file(kn); |
362 | ||
35beab06 | 363 | mutex_lock(&kernfs_mutex); |
fd7b9f7b TH |
364 | } |
365 | ||
fd7b9f7b | 366 | /** |
324a56e1 TH |
367 | * kernfs_get - get a reference count on a kernfs_node |
368 | * @kn: the target kernfs_node | |
fd7b9f7b | 369 | */ |
324a56e1 | 370 | void kernfs_get(struct kernfs_node *kn) |
fd7b9f7b | 371 | { |
324a56e1 | 372 | if (kn) { |
adc5e8b5 TH |
373 | WARN_ON(!atomic_read(&kn->count)); |
374 | atomic_inc(&kn->count); | |
fd7b9f7b TH |
375 | } |
376 | } | |
377 | EXPORT_SYMBOL_GPL(kernfs_get); | |
378 | ||
379 | /** | |
324a56e1 TH |
380 | * kernfs_put - put a reference count on a kernfs_node |
381 | * @kn: the target kernfs_node | |
fd7b9f7b | 382 | * |
324a56e1 | 383 | * Put a reference count of @kn and destroy it if it reached zero. |
fd7b9f7b | 384 | */ |
324a56e1 | 385 | void kernfs_put(struct kernfs_node *kn) |
fd7b9f7b | 386 | { |
324a56e1 | 387 | struct kernfs_node *parent; |
ba7443bc | 388 | struct kernfs_root *root; |
fd7b9f7b | 389 | |
adc5e8b5 | 390 | if (!kn || !atomic_dec_and_test(&kn->count)) |
fd7b9f7b | 391 | return; |
324a56e1 | 392 | root = kernfs_root(kn); |
fd7b9f7b | 393 | repeat: |
81c173cb TH |
394 | /* |
395 | * Moving/renaming is always done while holding reference. | |
adc5e8b5 | 396 | * kn->parent won't change beneath us. |
fd7b9f7b | 397 | */ |
adc5e8b5 | 398 | parent = kn->parent; |
fd7b9f7b | 399 | |
81c173cb TH |
400 | WARN_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS, |
401 | "kernfs_put: %s/%s: released with incorrect active_ref %d\n", | |
402 | parent ? parent->name : "", kn->name, atomic_read(&kn->active)); | |
324a56e1 | 403 | |
df23fc39 | 404 | if (kernfs_type(kn) == KERNFS_LINK) |
adc5e8b5 | 405 | kernfs_put(kn->symlink.target_kn); |
2063d608 | 406 | if (!(kn->flags & KERNFS_STATIC_NAME)) |
adc5e8b5 TH |
407 | kfree(kn->name); |
408 | if (kn->iattr) { | |
409 | if (kn->iattr->ia_secdata) | |
410 | security_release_secctx(kn->iattr->ia_secdata, | |
411 | kn->iattr->ia_secdata_len); | |
412 | simple_xattrs_free(&kn->iattr->xattrs); | |
2322392b | 413 | } |
adc5e8b5 TH |
414 | kfree(kn->iattr); |
415 | ida_simple_remove(&root->ino_ida, kn->ino); | |
a797bfc3 | 416 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b | 417 | |
324a56e1 TH |
418 | kn = parent; |
419 | if (kn) { | |
adc5e8b5 | 420 | if (atomic_dec_and_test(&kn->count)) |
ba7443bc TH |
421 | goto repeat; |
422 | } else { | |
324a56e1 | 423 | /* just released the root kn, free @root too */ |
bc755553 | 424 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
425 | kfree(root); |
426 | } | |
fd7b9f7b TH |
427 | } |
428 | EXPORT_SYMBOL_GPL(kernfs_put); | |
429 | ||
c637b8ac | 430 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) |
fd7b9f7b | 431 | { |
324a56e1 | 432 | struct kernfs_node *kn; |
fd7b9f7b TH |
433 | |
434 | if (flags & LOOKUP_RCU) | |
435 | return -ECHILD; | |
436 | ||
19bbb926 TH |
437 | /* Always perform fresh lookup for negatives */ |
438 | if (!dentry->d_inode) | |
439 | goto out_bad_unlocked; | |
440 | ||
324a56e1 | 441 | kn = dentry->d_fsdata; |
a797bfc3 | 442 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 443 | |
81c173cb TH |
444 | /* The kernfs node has been deactivated */ |
445 | if (!kernfs_active(kn)) | |
fd7b9f7b TH |
446 | goto out_bad; |
447 | ||
c637b8ac | 448 | /* The kernfs node has been moved? */ |
adc5e8b5 | 449 | if (dentry->d_parent->d_fsdata != kn->parent) |
fd7b9f7b TH |
450 | goto out_bad; |
451 | ||
c637b8ac | 452 | /* The kernfs node has been renamed */ |
adc5e8b5 | 453 | if (strcmp(dentry->d_name.name, kn->name) != 0) |
fd7b9f7b TH |
454 | goto out_bad; |
455 | ||
c637b8ac | 456 | /* The kernfs node has been moved to a different namespace */ |
adc5e8b5 | 457 | if (kn->parent && kernfs_ns_enabled(kn->parent) && |
c525aadd | 458 | kernfs_info(dentry->d_sb)->ns != kn->ns) |
fd7b9f7b TH |
459 | goto out_bad; |
460 | ||
a797bfc3 | 461 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
462 | out_valid: |
463 | return 1; | |
464 | out_bad: | |
a797bfc3 | 465 | mutex_unlock(&kernfs_mutex); |
19bbb926 TH |
466 | out_bad_unlocked: |
467 | /* | |
468 | * @dentry doesn't match the underlying kernfs node, drop the | |
469 | * dentry and force lookup. If we have submounts we must allow the | |
470 | * vfs caches to lie about the state of the filesystem to prevent | |
471 | * leaks and other nasty things, so use check_submounts_and_drop() | |
472 | * instead of d_drop(). | |
fd7b9f7b TH |
473 | */ |
474 | if (check_submounts_and_drop(dentry) != 0) | |
475 | goto out_valid; | |
476 | ||
477 | return 0; | |
478 | } | |
479 | ||
c637b8ac | 480 | static void kernfs_dop_release(struct dentry *dentry) |
fd7b9f7b TH |
481 | { |
482 | kernfs_put(dentry->d_fsdata); | |
483 | } | |
484 | ||
a797bfc3 | 485 | const struct dentry_operations kernfs_dops = { |
c637b8ac | 486 | .d_revalidate = kernfs_dop_revalidate, |
c637b8ac | 487 | .d_release = kernfs_dop_release, |
fd7b9f7b TH |
488 | }; |
489 | ||
0c23b225 TH |
490 | /** |
491 | * kernfs_node_from_dentry - determine kernfs_node associated with a dentry | |
492 | * @dentry: the dentry in question | |
493 | * | |
494 | * Return the kernfs_node associated with @dentry. If @dentry is not a | |
495 | * kernfs one, %NULL is returned. | |
496 | * | |
497 | * While the returned kernfs_node will stay accessible as long as @dentry | |
498 | * is accessible, the returned node can be in any state and the caller is | |
499 | * fully responsible for determining what's accessible. | |
500 | */ | |
501 | struct kernfs_node *kernfs_node_from_dentry(struct dentry *dentry) | |
502 | { | |
503 | if (dentry->d_op == &kernfs_dops) | |
504 | return dentry->d_fsdata; | |
505 | return NULL; | |
506 | } | |
507 | ||
db4aad20 TH |
508 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, |
509 | const char *name, umode_t mode, | |
510 | unsigned flags) | |
fd7b9f7b TH |
511 | { |
512 | char *dup_name = NULL; | |
324a56e1 | 513 | struct kernfs_node *kn; |
bc755553 | 514 | int ret; |
fd7b9f7b | 515 | |
2063d608 | 516 | if (!(flags & KERNFS_STATIC_NAME)) { |
fd7b9f7b TH |
517 | name = dup_name = kstrdup(name, GFP_KERNEL); |
518 | if (!name) | |
519 | return NULL; | |
520 | } | |
521 | ||
a797bfc3 | 522 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); |
324a56e1 | 523 | if (!kn) |
fd7b9f7b TH |
524 | goto err_out1; |
525 | ||
bc755553 TH |
526 | ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); |
527 | if (ret < 0) | |
fd7b9f7b | 528 | goto err_out2; |
adc5e8b5 | 529 | kn->ino = ret; |
fd7b9f7b | 530 | |
adc5e8b5 | 531 | atomic_set(&kn->count, 1); |
81c173cb | 532 | atomic_set(&kn->active, KN_DEACTIVATED_BIAS); |
35beab06 | 533 | RB_CLEAR_NODE(&kn->rb); |
fd7b9f7b | 534 | |
adc5e8b5 TH |
535 | kn->name = name; |
536 | kn->mode = mode; | |
81c173cb | 537 | kn->flags = flags; |
fd7b9f7b | 538 | |
324a56e1 | 539 | return kn; |
fd7b9f7b TH |
540 | |
541 | err_out2: | |
a797bfc3 | 542 | kmem_cache_free(kernfs_node_cache, kn); |
fd7b9f7b TH |
543 | err_out1: |
544 | kfree(dup_name); | |
545 | return NULL; | |
546 | } | |
547 | ||
db4aad20 TH |
548 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, |
549 | const char *name, umode_t mode, | |
550 | unsigned flags) | |
551 | { | |
552 | struct kernfs_node *kn; | |
553 | ||
554 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); | |
555 | if (kn) { | |
556 | kernfs_get(parent); | |
557 | kn->parent = parent; | |
558 | } | |
559 | return kn; | |
560 | } | |
561 | ||
fd7b9f7b | 562 | /** |
c637b8ac | 563 | * kernfs_add_one - add kernfs_node to parent without warning |
324a56e1 | 564 | * @kn: kernfs_node to be added |
fd7b9f7b | 565 | * |
db4aad20 TH |
566 | * The caller must already have initialized @kn->parent. This |
567 | * function increments nlink of the parent's inode if @kn is a | |
568 | * directory and link into the children list of the parent. | |
fd7b9f7b | 569 | * |
fd7b9f7b TH |
570 | * RETURNS: |
571 | * 0 on success, -EEXIST if entry with the given name already | |
572 | * exists. | |
573 | */ | |
988cd7af | 574 | int kernfs_add_one(struct kernfs_node *kn) |
fd7b9f7b | 575 | { |
db4aad20 | 576 | struct kernfs_node *parent = kn->parent; |
c525aadd | 577 | struct kernfs_iattrs *ps_iattr; |
988cd7af | 578 | bool has_ns; |
fd7b9f7b TH |
579 | int ret; |
580 | ||
988cd7af TH |
581 | mutex_lock(&kernfs_mutex); |
582 | ||
583 | ret = -EINVAL; | |
584 | has_ns = kernfs_ns_enabled(parent); | |
585 | if (WARN(has_ns != (bool)kn->ns, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | |
586 | has_ns ? "required" : "invalid", parent->name, kn->name)) | |
587 | goto out_unlock; | |
fd7b9f7b | 588 | |
df23fc39 | 589 | if (kernfs_type(parent) != KERNFS_DIR) |
988cd7af | 590 | goto out_unlock; |
fd7b9f7b | 591 | |
988cd7af | 592 | ret = -ENOENT; |
d35258ef | 593 | if ((parent->flags & KERNFS_ACTIVATED) && !kernfs_active(parent)) |
988cd7af | 594 | goto out_unlock; |
798c75a0 | 595 | |
c637b8ac | 596 | kn->hash = kernfs_name_hash(kn->name, kn->ns); |
fd7b9f7b | 597 | |
c637b8ac | 598 | ret = kernfs_link_sibling(kn); |
fd7b9f7b | 599 | if (ret) |
988cd7af | 600 | goto out_unlock; |
fd7b9f7b TH |
601 | |
602 | /* Update timestamps on the parent */ | |
adc5e8b5 | 603 | ps_iattr = parent->iattr; |
fd7b9f7b TH |
604 | if (ps_iattr) { |
605 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
606 | ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME; | |
607 | } | |
608 | ||
d35258ef TH |
609 | mutex_unlock(&kernfs_mutex); |
610 | ||
611 | /* | |
612 | * Activate the new node unless CREATE_DEACTIVATED is requested. | |
613 | * If not activated here, the kernfs user is responsible for | |
614 | * activating the node with kernfs_activate(). A node which hasn't | |
615 | * been activated is not visible to userland and its removal won't | |
616 | * trigger deactivation. | |
617 | */ | |
618 | if (!(kernfs_root(kn)->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) | |
619 | kernfs_activate(kn); | |
620 | return 0; | |
621 | ||
988cd7af | 622 | out_unlock: |
a797bfc3 | 623 | mutex_unlock(&kernfs_mutex); |
988cd7af | 624 | return ret; |
fd7b9f7b TH |
625 | } |
626 | ||
627 | /** | |
324a56e1 TH |
628 | * kernfs_find_ns - find kernfs_node with the given name |
629 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
630 | * @name: name to look for |
631 | * @ns: the namespace tag to use | |
632 | * | |
324a56e1 TH |
633 | * Look for kernfs_node with name @name under @parent. Returns pointer to |
634 | * the found kernfs_node on success, %NULL on failure. | |
fd7b9f7b | 635 | */ |
324a56e1 TH |
636 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, |
637 | const unsigned char *name, | |
638 | const void *ns) | |
fd7b9f7b | 639 | { |
adc5e8b5 | 640 | struct rb_node *node = parent->dir.children.rb_node; |
ac9bba03 | 641 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
642 | unsigned int hash; |
643 | ||
a797bfc3 | 644 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
645 | |
646 | if (has_ns != (bool)ns) { | |
c637b8ac | 647 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", |
adc5e8b5 | 648 | has_ns ? "required" : "invalid", parent->name, name); |
fd7b9f7b TH |
649 | return NULL; |
650 | } | |
651 | ||
c637b8ac | 652 | hash = kernfs_name_hash(name, ns); |
fd7b9f7b | 653 | while (node) { |
324a56e1 | 654 | struct kernfs_node *kn; |
fd7b9f7b TH |
655 | int result; |
656 | ||
324a56e1 | 657 | kn = rb_to_kn(node); |
c637b8ac | 658 | result = kernfs_name_compare(hash, name, ns, kn); |
fd7b9f7b TH |
659 | if (result < 0) |
660 | node = node->rb_left; | |
661 | else if (result > 0) | |
662 | node = node->rb_right; | |
663 | else | |
324a56e1 | 664 | return kn; |
fd7b9f7b TH |
665 | } |
666 | return NULL; | |
667 | } | |
668 | ||
669 | /** | |
324a56e1 TH |
670 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name |
671 | * @parent: kernfs_node to search under | |
fd7b9f7b TH |
672 | * @name: name to look for |
673 | * @ns: the namespace tag to use | |
674 | * | |
324a56e1 | 675 | * Look for kernfs_node with name @name under @parent and get a reference |
fd7b9f7b | 676 | * if found. This function may sleep and returns pointer to the found |
324a56e1 | 677 | * kernfs_node on success, %NULL on failure. |
fd7b9f7b | 678 | */ |
324a56e1 TH |
679 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, |
680 | const char *name, const void *ns) | |
fd7b9f7b | 681 | { |
324a56e1 | 682 | struct kernfs_node *kn; |
fd7b9f7b | 683 | |
a797bfc3 | 684 | mutex_lock(&kernfs_mutex); |
324a56e1 TH |
685 | kn = kernfs_find_ns(parent, name, ns); |
686 | kernfs_get(kn); | |
a797bfc3 | 687 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 688 | |
324a56e1 | 689 | return kn; |
fd7b9f7b TH |
690 | } |
691 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
692 | ||
ba7443bc TH |
693 | /** |
694 | * kernfs_create_root - create a new kernfs hierarchy | |
90c07c89 | 695 | * @scops: optional syscall operations for the hierarchy |
d35258ef | 696 | * @flags: KERNFS_ROOT_* flags |
ba7443bc TH |
697 | * @priv: opaque data associated with the new directory |
698 | * | |
699 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
700 | * failure. | |
701 | */ | |
90c07c89 | 702 | struct kernfs_root *kernfs_create_root(struct kernfs_syscall_ops *scops, |
d35258ef | 703 | unsigned int flags, void *priv) |
ba7443bc TH |
704 | { |
705 | struct kernfs_root *root; | |
324a56e1 | 706 | struct kernfs_node *kn; |
ba7443bc TH |
707 | |
708 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
709 | if (!root) | |
710 | return ERR_PTR(-ENOMEM); | |
711 | ||
bc755553 TH |
712 | ida_init(&root->ino_ida); |
713 | ||
db4aad20 TH |
714 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, |
715 | KERNFS_DIR); | |
324a56e1 | 716 | if (!kn) { |
bc755553 | 717 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
718 | kfree(root); |
719 | return ERR_PTR(-ENOMEM); | |
720 | } | |
721 | ||
324a56e1 | 722 | kn->priv = priv; |
adc5e8b5 | 723 | kn->dir.root = root; |
ba7443bc | 724 | |
90c07c89 | 725 | root->syscall_ops = scops; |
d35258ef | 726 | root->flags = flags; |
324a56e1 | 727 | root->kn = kn; |
abd54f02 | 728 | init_waitqueue_head(&root->deactivate_waitq); |
ba7443bc | 729 | |
d35258ef TH |
730 | if (!(root->flags & KERNFS_ROOT_CREATE_DEACTIVATED)) |
731 | kernfs_activate(kn); | |
732 | ||
ba7443bc TH |
733 | return root; |
734 | } | |
735 | ||
736 | /** | |
737 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
738 | * @root: root of the hierarchy to destroy | |
739 | * | |
740 | * Destroy the hierarchy anchored at @root by removing all existing | |
741 | * directories and destroying @root. | |
742 | */ | |
743 | void kernfs_destroy_root(struct kernfs_root *root) | |
744 | { | |
324a56e1 | 745 | kernfs_remove(root->kn); /* will also free @root */ |
ba7443bc TH |
746 | } |
747 | ||
fd7b9f7b TH |
748 | /** |
749 | * kernfs_create_dir_ns - create a directory | |
750 | * @parent: parent in which to create a new directory | |
751 | * @name: name of the new directory | |
bb8b9d09 | 752 | * @mode: mode of the new directory |
fd7b9f7b TH |
753 | * @priv: opaque data associated with the new directory |
754 | * @ns: optional namespace tag of the directory | |
755 | * | |
756 | * Returns the created node on success, ERR_PTR() value on failure. | |
757 | */ | |
324a56e1 | 758 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, |
bb8b9d09 TH |
759 | const char *name, umode_t mode, |
760 | void *priv, const void *ns) | |
fd7b9f7b | 761 | { |
324a56e1 | 762 | struct kernfs_node *kn; |
fd7b9f7b TH |
763 | int rc; |
764 | ||
765 | /* allocate */ | |
db4aad20 | 766 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); |
324a56e1 | 767 | if (!kn) |
fd7b9f7b TH |
768 | return ERR_PTR(-ENOMEM); |
769 | ||
adc5e8b5 TH |
770 | kn->dir.root = parent->dir.root; |
771 | kn->ns = ns; | |
324a56e1 | 772 | kn->priv = priv; |
fd7b9f7b TH |
773 | |
774 | /* link in */ | |
988cd7af | 775 | rc = kernfs_add_one(kn); |
fd7b9f7b | 776 | if (!rc) |
324a56e1 | 777 | return kn; |
fd7b9f7b | 778 | |
324a56e1 | 779 | kernfs_put(kn); |
fd7b9f7b TH |
780 | return ERR_PTR(rc); |
781 | } | |
782 | ||
c637b8ac TH |
783 | static struct dentry *kernfs_iop_lookup(struct inode *dir, |
784 | struct dentry *dentry, | |
785 | unsigned int flags) | |
fd7b9f7b | 786 | { |
19bbb926 | 787 | struct dentry *ret; |
324a56e1 TH |
788 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; |
789 | struct kernfs_node *kn; | |
fd7b9f7b TH |
790 | struct inode *inode; |
791 | const void *ns = NULL; | |
792 | ||
a797bfc3 | 793 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 794 | |
324a56e1 | 795 | if (kernfs_ns_enabled(parent)) |
c525aadd | 796 | ns = kernfs_info(dir->i_sb)->ns; |
fd7b9f7b | 797 | |
324a56e1 | 798 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); |
fd7b9f7b TH |
799 | |
800 | /* no such entry */ | |
b9c9dad0 | 801 | if (!kn || !kernfs_active(kn)) { |
19bbb926 | 802 | ret = NULL; |
fd7b9f7b TH |
803 | goto out_unlock; |
804 | } | |
324a56e1 TH |
805 | kernfs_get(kn); |
806 | dentry->d_fsdata = kn; | |
fd7b9f7b TH |
807 | |
808 | /* attach dentry and inode */ | |
c637b8ac | 809 | inode = kernfs_get_inode(dir->i_sb, kn); |
fd7b9f7b TH |
810 | if (!inode) { |
811 | ret = ERR_PTR(-ENOMEM); | |
812 | goto out_unlock; | |
813 | } | |
814 | ||
815 | /* instantiate and hash dentry */ | |
816 | ret = d_materialise_unique(dentry, inode); | |
817 | out_unlock: | |
a797bfc3 | 818 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
819 | return ret; |
820 | } | |
821 | ||
80b9bbef TH |
822 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, |
823 | umode_t mode) | |
824 | { | |
825 | struct kernfs_node *parent = dir->i_private; | |
90c07c89 | 826 | struct kernfs_syscall_ops *scops = kernfs_root(parent)->syscall_ops; |
07c7530d | 827 | int ret; |
80b9bbef | 828 | |
90c07c89 | 829 | if (!scops || !scops->mkdir) |
80b9bbef TH |
830 | return -EPERM; |
831 | ||
07c7530d TH |
832 | if (!kernfs_get_active(parent)) |
833 | return -ENODEV; | |
834 | ||
90c07c89 | 835 | ret = scops->mkdir(parent, dentry->d_name.name, mode); |
07c7530d TH |
836 | |
837 | kernfs_put_active(parent); | |
838 | return ret; | |
80b9bbef TH |
839 | } |
840 | ||
841 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | |
842 | { | |
843 | struct kernfs_node *kn = dentry->d_fsdata; | |
90c07c89 | 844 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 845 | int ret; |
80b9bbef | 846 | |
90c07c89 | 847 | if (!scops || !scops->rmdir) |
80b9bbef TH |
848 | return -EPERM; |
849 | ||
07c7530d TH |
850 | if (!kernfs_get_active(kn)) |
851 | return -ENODEV; | |
852 | ||
90c07c89 | 853 | ret = scops->rmdir(kn); |
07c7530d TH |
854 | |
855 | kernfs_put_active(kn); | |
856 | return ret; | |
80b9bbef TH |
857 | } |
858 | ||
859 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | |
860 | struct inode *new_dir, struct dentry *new_dentry) | |
861 | { | |
862 | struct kernfs_node *kn = old_dentry->d_fsdata; | |
863 | struct kernfs_node *new_parent = new_dir->i_private; | |
90c07c89 | 864 | struct kernfs_syscall_ops *scops = kernfs_root(kn)->syscall_ops; |
07c7530d | 865 | int ret; |
80b9bbef | 866 | |
90c07c89 | 867 | if (!scops || !scops->rename) |
80b9bbef TH |
868 | return -EPERM; |
869 | ||
07c7530d TH |
870 | if (!kernfs_get_active(kn)) |
871 | return -ENODEV; | |
872 | ||
873 | if (!kernfs_get_active(new_parent)) { | |
874 | kernfs_put_active(kn); | |
875 | return -ENODEV; | |
876 | } | |
877 | ||
90c07c89 | 878 | ret = scops->rename(kn, new_parent, new_dentry->d_name.name); |
07c7530d TH |
879 | |
880 | kernfs_put_active(new_parent); | |
881 | kernfs_put_active(kn); | |
882 | return ret; | |
80b9bbef TH |
883 | } |
884 | ||
a797bfc3 | 885 | const struct inode_operations kernfs_dir_iops = { |
c637b8ac TH |
886 | .lookup = kernfs_iop_lookup, |
887 | .permission = kernfs_iop_permission, | |
888 | .setattr = kernfs_iop_setattr, | |
889 | .getattr = kernfs_iop_getattr, | |
890 | .setxattr = kernfs_iop_setxattr, | |
891 | .removexattr = kernfs_iop_removexattr, | |
892 | .getxattr = kernfs_iop_getxattr, | |
893 | .listxattr = kernfs_iop_listxattr, | |
80b9bbef TH |
894 | |
895 | .mkdir = kernfs_iop_mkdir, | |
896 | .rmdir = kernfs_iop_rmdir, | |
897 | .rename = kernfs_iop_rename, | |
fd7b9f7b TH |
898 | }; |
899 | ||
c637b8ac | 900 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) |
fd7b9f7b | 901 | { |
324a56e1 | 902 | struct kernfs_node *last; |
fd7b9f7b TH |
903 | |
904 | while (true) { | |
905 | struct rb_node *rbn; | |
906 | ||
907 | last = pos; | |
908 | ||
df23fc39 | 909 | if (kernfs_type(pos) != KERNFS_DIR) |
fd7b9f7b TH |
910 | break; |
911 | ||
adc5e8b5 | 912 | rbn = rb_first(&pos->dir.children); |
fd7b9f7b TH |
913 | if (!rbn) |
914 | break; | |
915 | ||
324a56e1 | 916 | pos = rb_to_kn(rbn); |
fd7b9f7b TH |
917 | } |
918 | ||
919 | return last; | |
920 | } | |
921 | ||
922 | /** | |
c637b8ac | 923 | * kernfs_next_descendant_post - find the next descendant for post-order walk |
fd7b9f7b | 924 | * @pos: the current position (%NULL to initiate traversal) |
324a56e1 | 925 | * @root: kernfs_node whose descendants to walk |
fd7b9f7b TH |
926 | * |
927 | * Find the next descendant to visit for post-order traversal of @root's | |
928 | * descendants. @root is included in the iteration and the last node to be | |
929 | * visited. | |
930 | */ | |
c637b8ac TH |
931 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, |
932 | struct kernfs_node *root) | |
fd7b9f7b TH |
933 | { |
934 | struct rb_node *rbn; | |
935 | ||
a797bfc3 | 936 | lockdep_assert_held(&kernfs_mutex); |
fd7b9f7b TH |
937 | |
938 | /* if first iteration, visit leftmost descendant which may be root */ | |
939 | if (!pos) | |
c637b8ac | 940 | return kernfs_leftmost_descendant(root); |
fd7b9f7b TH |
941 | |
942 | /* if we visited @root, we're done */ | |
943 | if (pos == root) | |
944 | return NULL; | |
945 | ||
946 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
adc5e8b5 | 947 | rbn = rb_next(&pos->rb); |
fd7b9f7b | 948 | if (rbn) |
c637b8ac | 949 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); |
fd7b9f7b TH |
950 | |
951 | /* no sibling left, visit parent */ | |
adc5e8b5 | 952 | return pos->parent; |
fd7b9f7b TH |
953 | } |
954 | ||
d35258ef TH |
955 | /** |
956 | * kernfs_activate - activate a node which started deactivated | |
957 | * @kn: kernfs_node whose subtree is to be activated | |
958 | * | |
959 | * If the root has KERNFS_ROOT_CREATE_DEACTIVATED set, a newly created node | |
960 | * needs to be explicitly activated. A node which hasn't been activated | |
961 | * isn't visible to userland and deactivation is skipped during its | |
962 | * removal. This is useful to construct atomic init sequences where | |
963 | * creation of multiple nodes should either succeed or fail atomically. | |
964 | * | |
965 | * The caller is responsible for ensuring that this function is not called | |
966 | * after kernfs_remove*() is invoked on @kn. | |
967 | */ | |
968 | void kernfs_activate(struct kernfs_node *kn) | |
969 | { | |
970 | struct kernfs_node *pos; | |
971 | ||
972 | mutex_lock(&kernfs_mutex); | |
973 | ||
974 | pos = NULL; | |
975 | while ((pos = kernfs_next_descendant_post(pos, kn))) { | |
976 | if (!pos || (pos->flags & KERNFS_ACTIVATED)) | |
977 | continue; | |
978 | ||
979 | WARN_ON_ONCE(pos->parent && RB_EMPTY_NODE(&pos->rb)); | |
980 | WARN_ON_ONCE(atomic_read(&pos->active) != KN_DEACTIVATED_BIAS); | |
981 | ||
982 | atomic_sub(KN_DEACTIVATED_BIAS, &pos->active); | |
983 | pos->flags |= KERNFS_ACTIVATED; | |
984 | } | |
985 | ||
986 | mutex_unlock(&kernfs_mutex); | |
987 | } | |
988 | ||
988cd7af | 989 | static void __kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 990 | { |
35beab06 TH |
991 | struct kernfs_node *pos; |
992 | ||
993 | lockdep_assert_held(&kernfs_mutex); | |
fd7b9f7b | 994 | |
6b0afc2a TH |
995 | /* |
996 | * Short-circuit if non-root @kn has already finished removal. | |
997 | * This is for kernfs_remove_self() which plays with active ref | |
998 | * after removal. | |
999 | */ | |
1000 | if (!kn || (kn->parent && RB_EMPTY_NODE(&kn->rb))) | |
ce9b499c GKH |
1001 | return; |
1002 | ||
c637b8ac | 1003 | pr_debug("kernfs %s: removing\n", kn->name); |
fd7b9f7b | 1004 | |
81c173cb | 1005 | /* prevent any new usage under @kn by deactivating all nodes */ |
35beab06 TH |
1006 | pos = NULL; |
1007 | while ((pos = kernfs_next_descendant_post(pos, kn))) | |
81c173cb TH |
1008 | if (kernfs_active(pos)) |
1009 | atomic_add(KN_DEACTIVATED_BIAS, &pos->active); | |
35beab06 TH |
1010 | |
1011 | /* deactivate and unlink the subtree node-by-node */ | |
fd7b9f7b | 1012 | do { |
35beab06 TH |
1013 | pos = kernfs_leftmost_descendant(kn); |
1014 | ||
1015 | /* | |
81c173cb TH |
1016 | * kernfs_drain() drops kernfs_mutex temporarily and @pos's |
1017 | * base ref could have been put by someone else by the time | |
1018 | * the function returns. Make sure it doesn't go away | |
1019 | * underneath us. | |
35beab06 TH |
1020 | */ |
1021 | kernfs_get(pos); | |
1022 | ||
d35258ef TH |
1023 | /* |
1024 | * Drain iff @kn was activated. This avoids draining and | |
1025 | * its lockdep annotations for nodes which have never been | |
1026 | * activated and allows embedding kernfs_remove() in create | |
1027 | * error paths without worrying about draining. | |
1028 | */ | |
1029 | if (kn->flags & KERNFS_ACTIVATED) | |
1030 | kernfs_drain(pos); | |
1031 | else | |
1032 | WARN_ON_ONCE(atomic_read(&kn->active) != KN_DEACTIVATED_BIAS); | |
35beab06 TH |
1033 | |
1034 | /* | |
1035 | * kernfs_unlink_sibling() succeeds once per node. Use it | |
1036 | * to decide who's responsible for cleanups. | |
1037 | */ | |
1038 | if (!pos->parent || kernfs_unlink_sibling(pos)) { | |
1039 | struct kernfs_iattrs *ps_iattr = | |
1040 | pos->parent ? pos->parent->iattr : NULL; | |
1041 | ||
1042 | /* update timestamps on the parent */ | |
1043 | if (ps_iattr) { | |
1044 | ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME; | |
1045 | ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME; | |
1046 | } | |
1047 | ||
988cd7af | 1048 | kernfs_put(pos); |
35beab06 TH |
1049 | } |
1050 | ||
1051 | kernfs_put(pos); | |
1052 | } while (pos != kn); | |
fd7b9f7b TH |
1053 | } |
1054 | ||
1055 | /** | |
324a56e1 TH |
1056 | * kernfs_remove - remove a kernfs_node recursively |
1057 | * @kn: the kernfs_node to remove | |
fd7b9f7b | 1058 | * |
324a56e1 | 1059 | * Remove @kn along with all its subdirectories and files. |
fd7b9f7b | 1060 | */ |
324a56e1 | 1061 | void kernfs_remove(struct kernfs_node *kn) |
fd7b9f7b | 1062 | { |
988cd7af TH |
1063 | mutex_lock(&kernfs_mutex); |
1064 | __kernfs_remove(kn); | |
1065 | mutex_unlock(&kernfs_mutex); | |
fd7b9f7b TH |
1066 | } |
1067 | ||
6b0afc2a TH |
1068 | /** |
1069 | * kernfs_break_active_protection - break out of active protection | |
1070 | * @kn: the self kernfs_node | |
1071 | * | |
1072 | * The caller must be running off of a kernfs operation which is invoked | |
1073 | * with an active reference - e.g. one of kernfs_ops. Each invocation of | |
1074 | * this function must also be matched with an invocation of | |
1075 | * kernfs_unbreak_active_protection(). | |
1076 | * | |
1077 | * This function releases the active reference of @kn the caller is | |
1078 | * holding. Once this function is called, @kn may be removed at any point | |
1079 | * and the caller is solely responsible for ensuring that the objects it | |
1080 | * dereferences are accessible. | |
1081 | */ | |
1082 | void kernfs_break_active_protection(struct kernfs_node *kn) | |
1083 | { | |
1084 | /* | |
1085 | * Take out ourself out of the active ref dependency chain. If | |
1086 | * we're called without an active ref, lockdep will complain. | |
1087 | */ | |
1088 | kernfs_put_active(kn); | |
1089 | } | |
1090 | ||
1091 | /** | |
1092 | * kernfs_unbreak_active_protection - undo kernfs_break_active_protection() | |
1093 | * @kn: the self kernfs_node | |
1094 | * | |
1095 | * If kernfs_break_active_protection() was called, this function must be | |
1096 | * invoked before finishing the kernfs operation. Note that while this | |
1097 | * function restores the active reference, it doesn't and can't actually | |
1098 | * restore the active protection - @kn may already or be in the process of | |
1099 | * being removed. Once kernfs_break_active_protection() is invoked, that | |
1100 | * protection is irreversibly gone for the kernfs operation instance. | |
1101 | * | |
1102 | * While this function may be called at any point after | |
1103 | * kernfs_break_active_protection() is invoked, its most useful location | |
1104 | * would be right before the enclosing kernfs operation returns. | |
1105 | */ | |
1106 | void kernfs_unbreak_active_protection(struct kernfs_node *kn) | |
1107 | { | |
1108 | /* | |
1109 | * @kn->active could be in any state; however, the increment we do | |
1110 | * here will be undone as soon as the enclosing kernfs operation | |
1111 | * finishes and this temporary bump can't break anything. If @kn | |
1112 | * is alive, nothing changes. If @kn is being deactivated, the | |
1113 | * soon-to-follow put will either finish deactivation or restore | |
1114 | * deactivated state. If @kn is already removed, the temporary | |
1115 | * bump is guaranteed to be gone before @kn is released. | |
1116 | */ | |
1117 | atomic_inc(&kn->active); | |
1118 | if (kernfs_lockdep(kn)) | |
1119 | rwsem_acquire(&kn->dep_map, 0, 1, _RET_IP_); | |
1120 | } | |
1121 | ||
1122 | /** | |
1123 | * kernfs_remove_self - remove a kernfs_node from its own method | |
1124 | * @kn: the self kernfs_node to remove | |
1125 | * | |
1126 | * The caller must be running off of a kernfs operation which is invoked | |
1127 | * with an active reference - e.g. one of kernfs_ops. This can be used to | |
1128 | * implement a file operation which deletes itself. | |
1129 | * | |
1130 | * For example, the "delete" file for a sysfs device directory can be | |
1131 | * implemented by invoking kernfs_remove_self() on the "delete" file | |
1132 | * itself. This function breaks the circular dependency of trying to | |
1133 | * deactivate self while holding an active ref itself. It isn't necessary | |
1134 | * to modify the usual removal path to use kernfs_remove_self(). The | |
1135 | * "delete" implementation can simply invoke kernfs_remove_self() on self | |
1136 | * before proceeding with the usual removal path. kernfs will ignore later | |
1137 | * kernfs_remove() on self. | |
1138 | * | |
1139 | * kernfs_remove_self() can be called multiple times concurrently on the | |
1140 | * same kernfs_node. Only the first one actually performs removal and | |
1141 | * returns %true. All others will wait until the kernfs operation which | |
1142 | * won self-removal finishes and return %false. Note that the losers wait | |
1143 | * for the completion of not only the winning kernfs_remove_self() but also | |
1144 | * the whole kernfs_ops which won the arbitration. This can be used to | |
1145 | * guarantee, for example, all concurrent writes to a "delete" file to | |
1146 | * finish only after the whole operation is complete. | |
1147 | */ | |
1148 | bool kernfs_remove_self(struct kernfs_node *kn) | |
1149 | { | |
1150 | bool ret; | |
1151 | ||
1152 | mutex_lock(&kernfs_mutex); | |
1153 | kernfs_break_active_protection(kn); | |
1154 | ||
1155 | /* | |
1156 | * SUICIDAL is used to arbitrate among competing invocations. Only | |
1157 | * the first one will actually perform removal. When the removal | |
1158 | * is complete, SUICIDED is set and the active ref is restored | |
1159 | * while holding kernfs_mutex. The ones which lost arbitration | |
1160 | * waits for SUICDED && drained which can happen only after the | |
1161 | * enclosing kernfs operation which executed the winning instance | |
1162 | * of kernfs_remove_self() finished. | |
1163 | */ | |
1164 | if (!(kn->flags & KERNFS_SUICIDAL)) { | |
1165 | kn->flags |= KERNFS_SUICIDAL; | |
1166 | __kernfs_remove(kn); | |
1167 | kn->flags |= KERNFS_SUICIDED; | |
1168 | ret = true; | |
1169 | } else { | |
1170 | wait_queue_head_t *waitq = &kernfs_root(kn)->deactivate_waitq; | |
1171 | DEFINE_WAIT(wait); | |
1172 | ||
1173 | while (true) { | |
1174 | prepare_to_wait(waitq, &wait, TASK_UNINTERRUPTIBLE); | |
1175 | ||
1176 | if ((kn->flags & KERNFS_SUICIDED) && | |
1177 | atomic_read(&kn->active) == KN_DEACTIVATED_BIAS) | |
1178 | break; | |
1179 | ||
1180 | mutex_unlock(&kernfs_mutex); | |
1181 | schedule(); | |
1182 | mutex_lock(&kernfs_mutex); | |
1183 | } | |
1184 | finish_wait(waitq, &wait); | |
1185 | WARN_ON_ONCE(!RB_EMPTY_NODE(&kn->rb)); | |
1186 | ret = false; | |
1187 | } | |
1188 | ||
1189 | /* | |
1190 | * This must be done while holding kernfs_mutex; otherwise, waiting | |
1191 | * for SUICIDED && deactivated could finish prematurely. | |
1192 | */ | |
1193 | kernfs_unbreak_active_protection(kn); | |
1194 | ||
1195 | mutex_unlock(&kernfs_mutex); | |
1196 | return ret; | |
1197 | } | |
1198 | ||
fd7b9f7b | 1199 | /** |
324a56e1 TH |
1200 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it |
1201 | * @parent: parent of the target | |
1202 | * @name: name of the kernfs_node to remove | |
1203 | * @ns: namespace tag of the kernfs_node to remove | |
fd7b9f7b | 1204 | * |
324a56e1 TH |
1205 | * Look for the kernfs_node with @name and @ns under @parent and remove it. |
1206 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | |
fd7b9f7b | 1207 | */ |
324a56e1 | 1208 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, |
fd7b9f7b TH |
1209 | const void *ns) |
1210 | { | |
324a56e1 | 1211 | struct kernfs_node *kn; |
fd7b9f7b | 1212 | |
324a56e1 | 1213 | if (!parent) { |
c637b8ac | 1214 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", |
fd7b9f7b TH |
1215 | name); |
1216 | return -ENOENT; | |
1217 | } | |
1218 | ||
988cd7af | 1219 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1220 | |
324a56e1 TH |
1221 | kn = kernfs_find_ns(parent, name, ns); |
1222 | if (kn) | |
988cd7af | 1223 | __kernfs_remove(kn); |
fd7b9f7b | 1224 | |
988cd7af | 1225 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b | 1226 | |
324a56e1 | 1227 | if (kn) |
fd7b9f7b TH |
1228 | return 0; |
1229 | else | |
1230 | return -ENOENT; | |
1231 | } | |
1232 | ||
1233 | /** | |
1234 | * kernfs_rename_ns - move and rename a kernfs_node | |
324a56e1 | 1235 | * @kn: target node |
fd7b9f7b TH |
1236 | * @new_parent: new parent to put @sd under |
1237 | * @new_name: new name | |
1238 | * @new_ns: new namespace tag | |
1239 | */ | |
324a56e1 | 1240 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, |
fd7b9f7b TH |
1241 | const char *new_name, const void *new_ns) |
1242 | { | |
3eef34ad TH |
1243 | struct kernfs_node *old_parent; |
1244 | const char *old_name = NULL; | |
fd7b9f7b TH |
1245 | int error; |
1246 | ||
3eef34ad TH |
1247 | /* can't move or rename root */ |
1248 | if (!kn->parent) | |
1249 | return -EINVAL; | |
1250 | ||
798c75a0 GKH |
1251 | mutex_lock(&kernfs_mutex); |
1252 | ||
d0ae3d43 | 1253 | error = -ENOENT; |
81c173cb | 1254 | if (!kernfs_active(kn) || !kernfs_active(new_parent)) |
d0ae3d43 TH |
1255 | goto out; |
1256 | ||
fd7b9f7b | 1257 | error = 0; |
adc5e8b5 TH |
1258 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && |
1259 | (strcmp(kn->name, new_name) == 0)) | |
798c75a0 | 1260 | goto out; /* nothing to rename */ |
fd7b9f7b TH |
1261 | |
1262 | error = -EEXIST; | |
1263 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
798c75a0 | 1264 | goto out; |
fd7b9f7b | 1265 | |
324a56e1 | 1266 | /* rename kernfs_node */ |
adc5e8b5 | 1267 | if (strcmp(kn->name, new_name) != 0) { |
fd7b9f7b TH |
1268 | error = -ENOMEM; |
1269 | new_name = kstrdup(new_name, GFP_KERNEL); | |
1270 | if (!new_name) | |
798c75a0 | 1271 | goto out; |
3eef34ad TH |
1272 | } else { |
1273 | new_name = NULL; | |
fd7b9f7b TH |
1274 | } |
1275 | ||
1276 | /* | |
1277 | * Move to the appropriate place in the appropriate directories rbtree. | |
1278 | */ | |
c637b8ac | 1279 | kernfs_unlink_sibling(kn); |
fd7b9f7b | 1280 | kernfs_get(new_parent); |
3eef34ad TH |
1281 | |
1282 | /* rename_lock protects ->parent and ->name accessors */ | |
1283 | spin_lock_irq(&kernfs_rename_lock); | |
1284 | ||
1285 | old_parent = kn->parent; | |
adc5e8b5 | 1286 | kn->parent = new_parent; |
3eef34ad TH |
1287 | |
1288 | kn->ns = new_ns; | |
1289 | if (new_name) { | |
1290 | if (!(kn->flags & KERNFS_STATIC_NAME)) | |
1291 | old_name = kn->name; | |
1292 | kn->flags &= ~KERNFS_STATIC_NAME; | |
1293 | kn->name = new_name; | |
1294 | } | |
1295 | ||
1296 | spin_unlock_irq(&kernfs_rename_lock); | |
1297 | ||
1298 | kn->hash = kernfs_name_hash(new_name, new_ns); | |
c637b8ac | 1299 | kernfs_link_sibling(kn); |
fd7b9f7b | 1300 | |
3eef34ad TH |
1301 | kernfs_put(old_parent); |
1302 | kfree(old_name); | |
1303 | ||
fd7b9f7b | 1304 | error = 0; |
798c75a0 | 1305 | out: |
a797bfc3 | 1306 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1307 | return error; |
1308 | } | |
1309 | ||
fd7b9f7b | 1310 | /* Relationship between s_mode and the DT_xxx types */ |
324a56e1 | 1311 | static inline unsigned char dt_type(struct kernfs_node *kn) |
fd7b9f7b | 1312 | { |
adc5e8b5 | 1313 | return (kn->mode >> 12) & 15; |
fd7b9f7b TH |
1314 | } |
1315 | ||
c637b8ac | 1316 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) |
fd7b9f7b TH |
1317 | { |
1318 | kernfs_put(filp->private_data); | |
1319 | return 0; | |
1320 | } | |
1321 | ||
c637b8ac | 1322 | static struct kernfs_node *kernfs_dir_pos(const void *ns, |
324a56e1 | 1323 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) |
fd7b9f7b TH |
1324 | { |
1325 | if (pos) { | |
81c173cb | 1326 | int valid = kernfs_active(pos) && |
798c75a0 | 1327 | pos->parent == parent && hash == pos->hash; |
fd7b9f7b TH |
1328 | kernfs_put(pos); |
1329 | if (!valid) | |
1330 | pos = NULL; | |
1331 | } | |
1332 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
adc5e8b5 | 1333 | struct rb_node *node = parent->dir.children.rb_node; |
fd7b9f7b | 1334 | while (node) { |
324a56e1 | 1335 | pos = rb_to_kn(node); |
fd7b9f7b | 1336 | |
adc5e8b5 | 1337 | if (hash < pos->hash) |
fd7b9f7b | 1338 | node = node->rb_left; |
adc5e8b5 | 1339 | else if (hash > pos->hash) |
fd7b9f7b TH |
1340 | node = node->rb_right; |
1341 | else | |
1342 | break; | |
1343 | } | |
1344 | } | |
b9c9dad0 TH |
1345 | /* Skip over entries which are dying/dead or in the wrong namespace */ |
1346 | while (pos && (!kernfs_active(pos) || pos->ns != ns)) { | |
adc5e8b5 | 1347 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1348 | if (!node) |
1349 | pos = NULL; | |
1350 | else | |
324a56e1 | 1351 | pos = rb_to_kn(node); |
fd7b9f7b TH |
1352 | } |
1353 | return pos; | |
1354 | } | |
1355 | ||
c637b8ac | 1356 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, |
324a56e1 | 1357 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) |
fd7b9f7b | 1358 | { |
c637b8ac | 1359 | pos = kernfs_dir_pos(ns, parent, ino, pos); |
b9c9dad0 | 1360 | if (pos) { |
fd7b9f7b | 1361 | do { |
adc5e8b5 | 1362 | struct rb_node *node = rb_next(&pos->rb); |
fd7b9f7b TH |
1363 | if (!node) |
1364 | pos = NULL; | |
1365 | else | |
324a56e1 | 1366 | pos = rb_to_kn(node); |
b9c9dad0 TH |
1367 | } while (pos && (!kernfs_active(pos) || pos->ns != ns)); |
1368 | } | |
fd7b9f7b TH |
1369 | return pos; |
1370 | } | |
1371 | ||
c637b8ac | 1372 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) |
fd7b9f7b TH |
1373 | { |
1374 | struct dentry *dentry = file->f_path.dentry; | |
324a56e1 TH |
1375 | struct kernfs_node *parent = dentry->d_fsdata; |
1376 | struct kernfs_node *pos = file->private_data; | |
fd7b9f7b TH |
1377 | const void *ns = NULL; |
1378 | ||
1379 | if (!dir_emit_dots(file, ctx)) | |
1380 | return 0; | |
a797bfc3 | 1381 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1382 | |
324a56e1 | 1383 | if (kernfs_ns_enabled(parent)) |
c525aadd | 1384 | ns = kernfs_info(dentry->d_sb)->ns; |
fd7b9f7b | 1385 | |
c637b8ac | 1386 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); |
fd7b9f7b | 1387 | pos; |
c637b8ac | 1388 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { |
adc5e8b5 | 1389 | const char *name = pos->name; |
fd7b9f7b TH |
1390 | unsigned int type = dt_type(pos); |
1391 | int len = strlen(name); | |
adc5e8b5 | 1392 | ino_t ino = pos->ino; |
fd7b9f7b | 1393 | |
adc5e8b5 | 1394 | ctx->pos = pos->hash; |
fd7b9f7b TH |
1395 | file->private_data = pos; |
1396 | kernfs_get(pos); | |
1397 | ||
a797bfc3 | 1398 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1399 | if (!dir_emit(ctx, name, len, ino, type)) |
1400 | return 0; | |
a797bfc3 | 1401 | mutex_lock(&kernfs_mutex); |
fd7b9f7b | 1402 | } |
a797bfc3 | 1403 | mutex_unlock(&kernfs_mutex); |
fd7b9f7b TH |
1404 | file->private_data = NULL; |
1405 | ctx->pos = INT_MAX; | |
1406 | return 0; | |
1407 | } | |
1408 | ||
c637b8ac TH |
1409 | static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset, |
1410 | int whence) | |
fd7b9f7b TH |
1411 | { |
1412 | struct inode *inode = file_inode(file); | |
1413 | loff_t ret; | |
1414 | ||
1415 | mutex_lock(&inode->i_mutex); | |
1416 | ret = generic_file_llseek(file, offset, whence); | |
1417 | mutex_unlock(&inode->i_mutex); | |
1418 | ||
1419 | return ret; | |
1420 | } | |
1421 | ||
a797bfc3 | 1422 | const struct file_operations kernfs_dir_fops = { |
fd7b9f7b | 1423 | .read = generic_read_dir, |
c637b8ac TH |
1424 | .iterate = kernfs_fop_readdir, |
1425 | .release = kernfs_dir_fop_release, | |
1426 | .llseek = kernfs_dir_fop_llseek, | |
fd7b9f7b | 1427 | }; |