diff options
Diffstat (limited to 'fs/kernfs/dir.c')
-rw-r--r-- | fs/kernfs/dir.c | 1073 |
1 files changed, 1073 insertions, 0 deletions
diff --git a/fs/kernfs/dir.c b/fs/kernfs/dir.c new file mode 100644 index 000000000000..5104cf5d25c5 --- /dev/null +++ b/fs/kernfs/dir.c | |||
@@ -0,0 +1,1073 @@ | |||
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 | */ | ||
10 | |||
11 | #include <linux/fs.h> | ||
12 | #include <linux/namei.h> | ||
13 | #include <linux/idr.h> | ||
14 | #include <linux/slab.h> | ||
15 | #include <linux/security.h> | ||
16 | #include <linux/hash.h> | ||
17 | |||
18 | #include "kernfs-internal.h" | ||
19 | |||
20 | DEFINE_MUTEX(kernfs_mutex); | ||
21 | |||
22 | #define rb_to_kn(X) rb_entry((X), struct kernfs_node, rb) | ||
23 | |||
24 | /** | ||
25 | * kernfs_name_hash | ||
26 | * @name: Null terminated string to hash | ||
27 | * @ns: Namespace tag to hash | ||
28 | * | ||
29 | * Returns 31 bit hash of ns + name (so it fits in an off_t ) | ||
30 | */ | ||
31 | static unsigned int kernfs_name_hash(const char *name, const void *ns) | ||
32 | { | ||
33 | unsigned long hash = init_name_hash(); | ||
34 | unsigned int len = strlen(name); | ||
35 | while (len--) | ||
36 | hash = partial_name_hash(*name++, hash); | ||
37 | hash = (end_name_hash(hash) ^ hash_ptr((void *)ns, 31)); | ||
38 | hash &= 0x7fffffffU; | ||
39 | /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */ | ||
40 | if (hash < 1) | ||
41 | hash += 2; | ||
42 | if (hash >= INT_MAX) | ||
43 | hash = INT_MAX - 1; | ||
44 | return hash; | ||
45 | } | ||
46 | |||
47 | static int kernfs_name_compare(unsigned int hash, const char *name, | ||
48 | const void *ns, const struct kernfs_node *kn) | ||
49 | { | ||
50 | if (hash != kn->hash) | ||
51 | return hash - kn->hash; | ||
52 | if (ns != kn->ns) | ||
53 | return ns - kn->ns; | ||
54 | return strcmp(name, kn->name); | ||
55 | } | ||
56 | |||
57 | static int kernfs_sd_compare(const struct kernfs_node *left, | ||
58 | const struct kernfs_node *right) | ||
59 | { | ||
60 | return kernfs_name_compare(left->hash, left->name, left->ns, right); | ||
61 | } | ||
62 | |||
63 | /** | ||
64 | * kernfs_link_sibling - link kernfs_node into sibling rbtree | ||
65 | * @kn: kernfs_node of interest | ||
66 | * | ||
67 | * Link @kn into its sibling rbtree which starts from | ||
68 | * @kn->parent->dir.children. | ||
69 | * | ||
70 | * Locking: | ||
71 | * mutex_lock(kernfs_mutex) | ||
72 | * | ||
73 | * RETURNS: | ||
74 | * 0 on susccess -EEXIST on failure. | ||
75 | */ | ||
76 | static int kernfs_link_sibling(struct kernfs_node *kn) | ||
77 | { | ||
78 | struct rb_node **node = &kn->parent->dir.children.rb_node; | ||
79 | struct rb_node *parent = NULL; | ||
80 | |||
81 | if (kernfs_type(kn) == KERNFS_DIR) | ||
82 | kn->parent->dir.subdirs++; | ||
83 | |||
84 | while (*node) { | ||
85 | struct kernfs_node *pos; | ||
86 | int result; | ||
87 | |||
88 | pos = rb_to_kn(*node); | ||
89 | parent = *node; | ||
90 | result = kernfs_sd_compare(kn, pos); | ||
91 | if (result < 0) | ||
92 | node = &pos->rb.rb_left; | ||
93 | else if (result > 0) | ||
94 | node = &pos->rb.rb_right; | ||
95 | else | ||
96 | return -EEXIST; | ||
97 | } | ||
98 | /* add new node and rebalance the tree */ | ||
99 | rb_link_node(&kn->rb, parent, node); | ||
100 | rb_insert_color(&kn->rb, &kn->parent->dir.children); | ||
101 | return 0; | ||
102 | } | ||
103 | |||
104 | /** | ||
105 | * kernfs_unlink_sibling - unlink kernfs_node from sibling rbtree | ||
106 | * @kn: kernfs_node of interest | ||
107 | * | ||
108 | * Unlink @kn from its sibling rbtree which starts from | ||
109 | * kn->parent->dir.children. | ||
110 | * | ||
111 | * Locking: | ||
112 | * mutex_lock(kernfs_mutex) | ||
113 | */ | ||
114 | static void kernfs_unlink_sibling(struct kernfs_node *kn) | ||
115 | { | ||
116 | if (kernfs_type(kn) == KERNFS_DIR) | ||
117 | kn->parent->dir.subdirs--; | ||
118 | |||
119 | rb_erase(&kn->rb, &kn->parent->dir.children); | ||
120 | } | ||
121 | |||
122 | /** | ||
123 | * kernfs_get_active - get an active reference to kernfs_node | ||
124 | * @kn: kernfs_node to get an active reference to | ||
125 | * | ||
126 | * Get an active reference of @kn. This function is noop if @kn | ||
127 | * is NULL. | ||
128 | * | ||
129 | * RETURNS: | ||
130 | * Pointer to @kn on success, NULL on failure. | ||
131 | */ | ||
132 | struct kernfs_node *kernfs_get_active(struct kernfs_node *kn) | ||
133 | { | ||
134 | if (unlikely(!kn)) | ||
135 | return NULL; | ||
136 | |||
137 | if (!atomic_inc_unless_negative(&kn->active)) | ||
138 | return NULL; | ||
139 | |||
140 | if (kn->flags & KERNFS_LOCKDEP) | ||
141 | rwsem_acquire_read(&kn->dep_map, 0, 1, _RET_IP_); | ||
142 | return kn; | ||
143 | } | ||
144 | |||
145 | /** | ||
146 | * kernfs_put_active - put an active reference to kernfs_node | ||
147 | * @kn: kernfs_node to put an active reference to | ||
148 | * | ||
149 | * Put an active reference to @kn. This function is noop if @kn | ||
150 | * is NULL. | ||
151 | */ | ||
152 | void kernfs_put_active(struct kernfs_node *kn) | ||
153 | { | ||
154 | int v; | ||
155 | |||
156 | if (unlikely(!kn)) | ||
157 | return; | ||
158 | |||
159 | if (kn->flags & KERNFS_LOCKDEP) | ||
160 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | ||
161 | v = atomic_dec_return(&kn->active); | ||
162 | if (likely(v != KN_DEACTIVATED_BIAS)) | ||
163 | return; | ||
164 | |||
165 | /* | ||
166 | * atomic_dec_return() is a mb(), we'll always see the updated | ||
167 | * kn->u.completion. | ||
168 | */ | ||
169 | complete(kn->u.completion); | ||
170 | } | ||
171 | |||
172 | /** | ||
173 | * kernfs_deactivate - deactivate kernfs_node | ||
174 | * @kn: kernfs_node to deactivate | ||
175 | * | ||
176 | * Deny new active references and drain existing ones. | ||
177 | */ | ||
178 | static void kernfs_deactivate(struct kernfs_node *kn) | ||
179 | { | ||
180 | DECLARE_COMPLETION_ONSTACK(wait); | ||
181 | int v; | ||
182 | |||
183 | BUG_ON(!(kn->flags & KERNFS_REMOVED)); | ||
184 | |||
185 | if (!(kernfs_type(kn) & KERNFS_ACTIVE_REF)) | ||
186 | return; | ||
187 | |||
188 | kn->u.completion = (void *)&wait; | ||
189 | |||
190 | rwsem_acquire(&kn->dep_map, 0, 0, _RET_IP_); | ||
191 | /* atomic_add_return() is a mb(), put_active() will always see | ||
192 | * the updated kn->u.completion. | ||
193 | */ | ||
194 | v = atomic_add_return(KN_DEACTIVATED_BIAS, &kn->active); | ||
195 | |||
196 | if (v != KN_DEACTIVATED_BIAS) { | ||
197 | lock_contended(&kn->dep_map, _RET_IP_); | ||
198 | wait_for_completion(&wait); | ||
199 | } | ||
200 | |||
201 | lock_acquired(&kn->dep_map, _RET_IP_); | ||
202 | rwsem_release(&kn->dep_map, 1, _RET_IP_); | ||
203 | } | ||
204 | |||
205 | /** | ||
206 | * kernfs_get - get a reference count on a kernfs_node | ||
207 | * @kn: the target kernfs_node | ||
208 | */ | ||
209 | void kernfs_get(struct kernfs_node *kn) | ||
210 | { | ||
211 | if (kn) { | ||
212 | WARN_ON(!atomic_read(&kn->count)); | ||
213 | atomic_inc(&kn->count); | ||
214 | } | ||
215 | } | ||
216 | EXPORT_SYMBOL_GPL(kernfs_get); | ||
217 | |||
218 | /** | ||
219 | * kernfs_put - put a reference count on a kernfs_node | ||
220 | * @kn: the target kernfs_node | ||
221 | * | ||
222 | * Put a reference count of @kn and destroy it if it reached zero. | ||
223 | */ | ||
224 | void kernfs_put(struct kernfs_node *kn) | ||
225 | { | ||
226 | struct kernfs_node *parent; | ||
227 | struct kernfs_root *root; | ||
228 | |||
229 | if (!kn || !atomic_dec_and_test(&kn->count)) | ||
230 | return; | ||
231 | root = kernfs_root(kn); | ||
232 | repeat: | ||
233 | /* Moving/renaming is always done while holding reference. | ||
234 | * kn->parent won't change beneath us. | ||
235 | */ | ||
236 | parent = kn->parent; | ||
237 | |||
238 | WARN(!(kn->flags & KERNFS_REMOVED), "kernfs: free using entry: %s/%s\n", | ||
239 | parent ? parent->name : "", kn->name); | ||
240 | |||
241 | if (kernfs_type(kn) == KERNFS_LINK) | ||
242 | kernfs_put(kn->symlink.target_kn); | ||
243 | if (!(kn->flags & KERNFS_STATIC_NAME)) | ||
244 | kfree(kn->name); | ||
245 | if (kn->iattr) { | ||
246 | if (kn->iattr->ia_secdata) | ||
247 | security_release_secctx(kn->iattr->ia_secdata, | ||
248 | kn->iattr->ia_secdata_len); | ||
249 | simple_xattrs_free(&kn->iattr->xattrs); | ||
250 | } | ||
251 | kfree(kn->iattr); | ||
252 | ida_simple_remove(&root->ino_ida, kn->ino); | ||
253 | kmem_cache_free(kernfs_node_cache, kn); | ||
254 | |||
255 | kn = parent; | ||
256 | if (kn) { | ||
257 | if (atomic_dec_and_test(&kn->count)) | ||
258 | goto repeat; | ||
259 | } else { | ||
260 | /* just released the root kn, free @root too */ | ||
261 | ida_destroy(&root->ino_ida); | ||
262 | kfree(root); | ||
263 | } | ||
264 | } | ||
265 | EXPORT_SYMBOL_GPL(kernfs_put); | ||
266 | |||
267 | static int kernfs_dop_revalidate(struct dentry *dentry, unsigned int flags) | ||
268 | { | ||
269 | struct kernfs_node *kn; | ||
270 | |||
271 | if (flags & LOOKUP_RCU) | ||
272 | return -ECHILD; | ||
273 | |||
274 | /* Always perform fresh lookup for negatives */ | ||
275 | if (!dentry->d_inode) | ||
276 | goto out_bad_unlocked; | ||
277 | |||
278 | kn = dentry->d_fsdata; | ||
279 | mutex_lock(&kernfs_mutex); | ||
280 | |||
281 | /* The kernfs node has been deleted */ | ||
282 | if (kn->flags & KERNFS_REMOVED) | ||
283 | goto out_bad; | ||
284 | |||
285 | /* The kernfs node has been moved? */ | ||
286 | if (dentry->d_parent->d_fsdata != kn->parent) | ||
287 | goto out_bad; | ||
288 | |||
289 | /* The kernfs node has been renamed */ | ||
290 | if (strcmp(dentry->d_name.name, kn->name) != 0) | ||
291 | goto out_bad; | ||
292 | |||
293 | /* The kernfs node has been moved to a different namespace */ | ||
294 | if (kn->parent && kernfs_ns_enabled(kn->parent) && | ||
295 | kernfs_info(dentry->d_sb)->ns != kn->ns) | ||
296 | goto out_bad; | ||
297 | |||
298 | mutex_unlock(&kernfs_mutex); | ||
299 | out_valid: | ||
300 | return 1; | ||
301 | out_bad: | ||
302 | mutex_unlock(&kernfs_mutex); | ||
303 | out_bad_unlocked: | ||
304 | /* | ||
305 | * @dentry doesn't match the underlying kernfs node, drop the | ||
306 | * dentry and force lookup. If we have submounts we must allow the | ||
307 | * vfs caches to lie about the state of the filesystem to prevent | ||
308 | * leaks and other nasty things, so use check_submounts_and_drop() | ||
309 | * instead of d_drop(). | ||
310 | */ | ||
311 | if (check_submounts_and_drop(dentry) != 0) | ||
312 | goto out_valid; | ||
313 | |||
314 | return 0; | ||
315 | } | ||
316 | |||
317 | static void kernfs_dop_release(struct dentry *dentry) | ||
318 | { | ||
319 | kernfs_put(dentry->d_fsdata); | ||
320 | } | ||
321 | |||
322 | const struct dentry_operations kernfs_dops = { | ||
323 | .d_revalidate = kernfs_dop_revalidate, | ||
324 | .d_release = kernfs_dop_release, | ||
325 | }; | ||
326 | |||
327 | static struct kernfs_node *__kernfs_new_node(struct kernfs_root *root, | ||
328 | const char *name, umode_t mode, | ||
329 | unsigned flags) | ||
330 | { | ||
331 | char *dup_name = NULL; | ||
332 | struct kernfs_node *kn; | ||
333 | int ret; | ||
334 | |||
335 | if (!(flags & KERNFS_STATIC_NAME)) { | ||
336 | name = dup_name = kstrdup(name, GFP_KERNEL); | ||
337 | if (!name) | ||
338 | return NULL; | ||
339 | } | ||
340 | |||
341 | kn = kmem_cache_zalloc(kernfs_node_cache, GFP_KERNEL); | ||
342 | if (!kn) | ||
343 | goto err_out1; | ||
344 | |||
345 | ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); | ||
346 | if (ret < 0) | ||
347 | goto err_out2; | ||
348 | kn->ino = ret; | ||
349 | |||
350 | atomic_set(&kn->count, 1); | ||
351 | atomic_set(&kn->active, 0); | ||
352 | |||
353 | kn->name = name; | ||
354 | kn->mode = mode; | ||
355 | kn->flags = flags | KERNFS_REMOVED; | ||
356 | |||
357 | return kn; | ||
358 | |||
359 | err_out2: | ||
360 | kmem_cache_free(kernfs_node_cache, kn); | ||
361 | err_out1: | ||
362 | kfree(dup_name); | ||
363 | return NULL; | ||
364 | } | ||
365 | |||
366 | struct kernfs_node *kernfs_new_node(struct kernfs_node *parent, | ||
367 | const char *name, umode_t mode, | ||
368 | unsigned flags) | ||
369 | { | ||
370 | struct kernfs_node *kn; | ||
371 | |||
372 | kn = __kernfs_new_node(kernfs_root(parent), name, mode, flags); | ||
373 | if (kn) { | ||
374 | kernfs_get(parent); | ||
375 | kn->parent = parent; | ||
376 | } | ||
377 | return kn; | ||
378 | } | ||
379 | |||
380 | /** | ||
381 | * kernfs_addrm_start - prepare for kernfs_node add/remove | ||
382 | * @acxt: pointer to kernfs_addrm_cxt to be used | ||
383 | * | ||
384 | * This function is called when the caller is about to add or remove | ||
385 | * kernfs_node. This function acquires kernfs_mutex. @acxt is used | ||
386 | * to keep and pass context to other addrm functions. | ||
387 | * | ||
388 | * LOCKING: | ||
389 | * Kernel thread context (may sleep). kernfs_mutex is locked on | ||
390 | * return. | ||
391 | */ | ||
392 | void kernfs_addrm_start(struct kernfs_addrm_cxt *acxt) | ||
393 | __acquires(kernfs_mutex) | ||
394 | { | ||
395 | memset(acxt, 0, sizeof(*acxt)); | ||
396 | |||
397 | mutex_lock(&kernfs_mutex); | ||
398 | } | ||
399 | |||
400 | /** | ||
401 | * kernfs_add_one - add kernfs_node to parent without warning | ||
402 | * @acxt: addrm context to use | ||
403 | * @kn: kernfs_node to be added | ||
404 | * | ||
405 | * The caller must already have initialized @kn->parent. This | ||
406 | * function increments nlink of the parent's inode if @kn is a | ||
407 | * directory and link into the children list of the parent. | ||
408 | * | ||
409 | * This function should be called between calls to | ||
410 | * kernfs_addrm_start() and kernfs_addrm_finish() and should be passed | ||
411 | * the same @acxt as passed to kernfs_addrm_start(). | ||
412 | * | ||
413 | * LOCKING: | ||
414 | * Determined by kernfs_addrm_start(). | ||
415 | * | ||
416 | * RETURNS: | ||
417 | * 0 on success, -EEXIST if entry with the given name already | ||
418 | * exists. | ||
419 | */ | ||
420 | int kernfs_add_one(struct kernfs_addrm_cxt *acxt, struct kernfs_node *kn) | ||
421 | { | ||
422 | struct kernfs_node *parent = kn->parent; | ||
423 | bool has_ns = kernfs_ns_enabled(parent); | ||
424 | struct kernfs_iattrs *ps_iattr; | ||
425 | int ret; | ||
426 | |||
427 | if (has_ns != (bool)kn->ns) { | ||
428 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | ||
429 | has_ns ? "required" : "invalid", parent->name, kn->name); | ||
430 | return -EINVAL; | ||
431 | } | ||
432 | |||
433 | if (kernfs_type(parent) != KERNFS_DIR) | ||
434 | return -EINVAL; | ||
435 | |||
436 | if (parent->flags & KERNFS_REMOVED) | ||
437 | return -ENOENT; | ||
438 | |||
439 | kn->hash = kernfs_name_hash(kn->name, kn->ns); | ||
440 | |||
441 | ret = kernfs_link_sibling(kn); | ||
442 | if (ret) | ||
443 | return ret; | ||
444 | |||
445 | /* Update timestamps on the parent */ | ||
446 | ps_iattr = parent->iattr; | ||
447 | if (ps_iattr) { | ||
448 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | ||
449 | ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME; | ||
450 | } | ||
451 | |||
452 | /* Mark the entry added into directory tree */ | ||
453 | kn->flags &= ~KERNFS_REMOVED; | ||
454 | |||
455 | return 0; | ||
456 | } | ||
457 | |||
458 | /** | ||
459 | * kernfs_remove_one - remove kernfs_node from parent | ||
460 | * @acxt: addrm context to use | ||
461 | * @kn: kernfs_node to be removed | ||
462 | * | ||
463 | * Mark @kn removed and drop nlink of parent inode if @kn is a | ||
464 | * directory. @kn is unlinked from the children list. | ||
465 | * | ||
466 | * This function should be called between calls to | ||
467 | * kernfs_addrm_start() and kernfs_addrm_finish() and should be | ||
468 | * passed the same @acxt as passed to kernfs_addrm_start(). | ||
469 | * | ||
470 | * LOCKING: | ||
471 | * Determined by kernfs_addrm_start(). | ||
472 | */ | ||
473 | static void kernfs_remove_one(struct kernfs_addrm_cxt *acxt, | ||
474 | struct kernfs_node *kn) | ||
475 | { | ||
476 | struct kernfs_iattrs *ps_iattr; | ||
477 | |||
478 | /* | ||
479 | * Removal can be called multiple times on the same node. Only the | ||
480 | * first invocation is effective and puts the base ref. | ||
481 | */ | ||
482 | if (kn->flags & KERNFS_REMOVED) | ||
483 | return; | ||
484 | |||
485 | if (kn->parent) { | ||
486 | kernfs_unlink_sibling(kn); | ||
487 | |||
488 | /* Update timestamps on the parent */ | ||
489 | ps_iattr = kn->parent->iattr; | ||
490 | if (ps_iattr) { | ||
491 | ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME; | ||
492 | ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME; | ||
493 | } | ||
494 | } | ||
495 | |||
496 | kn->flags |= KERNFS_REMOVED; | ||
497 | kn->u.removed_list = acxt->removed; | ||
498 | acxt->removed = kn; | ||
499 | } | ||
500 | |||
501 | /** | ||
502 | * kernfs_addrm_finish - finish up kernfs_node add/remove | ||
503 | * @acxt: addrm context to finish up | ||
504 | * | ||
505 | * Finish up kernfs_node add/remove. Resources acquired by | ||
506 | * kernfs_addrm_start() are released and removed kernfs_nodes are | ||
507 | * cleaned up. | ||
508 | * | ||
509 | * LOCKING: | ||
510 | * kernfs_mutex is released. | ||
511 | */ | ||
512 | void kernfs_addrm_finish(struct kernfs_addrm_cxt *acxt) | ||
513 | __releases(kernfs_mutex) | ||
514 | { | ||
515 | /* release resources acquired by kernfs_addrm_start() */ | ||
516 | mutex_unlock(&kernfs_mutex); | ||
517 | |||
518 | /* kill removed kernfs_nodes */ | ||
519 | while (acxt->removed) { | ||
520 | struct kernfs_node *kn = acxt->removed; | ||
521 | |||
522 | acxt->removed = kn->u.removed_list; | ||
523 | |||
524 | kernfs_deactivate(kn); | ||
525 | kernfs_unmap_bin_file(kn); | ||
526 | kernfs_put(kn); | ||
527 | } | ||
528 | } | ||
529 | |||
530 | /** | ||
531 | * kernfs_find_ns - find kernfs_node with the given name | ||
532 | * @parent: kernfs_node to search under | ||
533 | * @name: name to look for | ||
534 | * @ns: the namespace tag to use | ||
535 | * | ||
536 | * Look for kernfs_node with name @name under @parent. Returns pointer to | ||
537 | * the found kernfs_node on success, %NULL on failure. | ||
538 | */ | ||
539 | static struct kernfs_node *kernfs_find_ns(struct kernfs_node *parent, | ||
540 | const unsigned char *name, | ||
541 | const void *ns) | ||
542 | { | ||
543 | struct rb_node *node = parent->dir.children.rb_node; | ||
544 | bool has_ns = kernfs_ns_enabled(parent); | ||
545 | unsigned int hash; | ||
546 | |||
547 | lockdep_assert_held(&kernfs_mutex); | ||
548 | |||
549 | if (has_ns != (bool)ns) { | ||
550 | WARN(1, KERN_WARNING "kernfs: ns %s in '%s' for '%s'\n", | ||
551 | has_ns ? "required" : "invalid", parent->name, name); | ||
552 | return NULL; | ||
553 | } | ||
554 | |||
555 | hash = kernfs_name_hash(name, ns); | ||
556 | while (node) { | ||
557 | struct kernfs_node *kn; | ||
558 | int result; | ||
559 | |||
560 | kn = rb_to_kn(node); | ||
561 | result = kernfs_name_compare(hash, name, ns, kn); | ||
562 | if (result < 0) | ||
563 | node = node->rb_left; | ||
564 | else if (result > 0) | ||
565 | node = node->rb_right; | ||
566 | else | ||
567 | return kn; | ||
568 | } | ||
569 | return NULL; | ||
570 | } | ||
571 | |||
572 | /** | ||
573 | * kernfs_find_and_get_ns - find and get kernfs_node with the given name | ||
574 | * @parent: kernfs_node to search under | ||
575 | * @name: name to look for | ||
576 | * @ns: the namespace tag to use | ||
577 | * | ||
578 | * Look for kernfs_node with name @name under @parent and get a reference | ||
579 | * if found. This function may sleep and returns pointer to the found | ||
580 | * kernfs_node on success, %NULL on failure. | ||
581 | */ | ||
582 | struct kernfs_node *kernfs_find_and_get_ns(struct kernfs_node *parent, | ||
583 | const char *name, const void *ns) | ||
584 | { | ||
585 | struct kernfs_node *kn; | ||
586 | |||
587 | mutex_lock(&kernfs_mutex); | ||
588 | kn = kernfs_find_ns(parent, name, ns); | ||
589 | kernfs_get(kn); | ||
590 | mutex_unlock(&kernfs_mutex); | ||
591 | |||
592 | return kn; | ||
593 | } | ||
594 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | ||
595 | |||
596 | /** | ||
597 | * kernfs_create_root - create a new kernfs hierarchy | ||
598 | * @kdops: optional directory syscall operations for the hierarchy | ||
599 | * @priv: opaque data associated with the new directory | ||
600 | * | ||
601 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | ||
602 | * failure. | ||
603 | */ | ||
604 | struct kernfs_root *kernfs_create_root(struct kernfs_dir_ops *kdops, void *priv) | ||
605 | { | ||
606 | struct kernfs_root *root; | ||
607 | struct kernfs_node *kn; | ||
608 | |||
609 | root = kzalloc(sizeof(*root), GFP_KERNEL); | ||
610 | if (!root) | ||
611 | return ERR_PTR(-ENOMEM); | ||
612 | |||
613 | ida_init(&root->ino_ida); | ||
614 | |||
615 | kn = __kernfs_new_node(root, "", S_IFDIR | S_IRUGO | S_IXUGO, | ||
616 | KERNFS_DIR); | ||
617 | if (!kn) { | ||
618 | ida_destroy(&root->ino_ida); | ||
619 | kfree(root); | ||
620 | return ERR_PTR(-ENOMEM); | ||
621 | } | ||
622 | |||
623 | kn->flags &= ~KERNFS_REMOVED; | ||
624 | kn->priv = priv; | ||
625 | kn->dir.root = root; | ||
626 | |||
627 | root->dir_ops = kdops; | ||
628 | root->kn = kn; | ||
629 | |||
630 | return root; | ||
631 | } | ||
632 | |||
633 | /** | ||
634 | * kernfs_destroy_root - destroy a kernfs hierarchy | ||
635 | * @root: root of the hierarchy to destroy | ||
636 | * | ||
637 | * Destroy the hierarchy anchored at @root by removing all existing | ||
638 | * directories and destroying @root. | ||
639 | */ | ||
640 | void kernfs_destroy_root(struct kernfs_root *root) | ||
641 | { | ||
642 | kernfs_remove(root->kn); /* will also free @root */ | ||
643 | } | ||
644 | |||
645 | /** | ||
646 | * kernfs_create_dir_ns - create a directory | ||
647 | * @parent: parent in which to create a new directory | ||
648 | * @name: name of the new directory | ||
649 | * @mode: mode of the new directory | ||
650 | * @priv: opaque data associated with the new directory | ||
651 | * @ns: optional namespace tag of the directory | ||
652 | * | ||
653 | * Returns the created node on success, ERR_PTR() value on failure. | ||
654 | */ | ||
655 | struct kernfs_node *kernfs_create_dir_ns(struct kernfs_node *parent, | ||
656 | const char *name, umode_t mode, | ||
657 | void *priv, const void *ns) | ||
658 | { | ||
659 | struct kernfs_addrm_cxt acxt; | ||
660 | struct kernfs_node *kn; | ||
661 | int rc; | ||
662 | |||
663 | /* allocate */ | ||
664 | kn = kernfs_new_node(parent, name, mode | S_IFDIR, KERNFS_DIR); | ||
665 | if (!kn) | ||
666 | return ERR_PTR(-ENOMEM); | ||
667 | |||
668 | kn->dir.root = parent->dir.root; | ||
669 | kn->ns = ns; | ||
670 | kn->priv = priv; | ||
671 | |||
672 | /* link in */ | ||
673 | kernfs_addrm_start(&acxt); | ||
674 | rc = kernfs_add_one(&acxt, kn); | ||
675 | kernfs_addrm_finish(&acxt); | ||
676 | |||
677 | if (!rc) | ||
678 | return kn; | ||
679 | |||
680 | kernfs_put(kn); | ||
681 | return ERR_PTR(rc); | ||
682 | } | ||
683 | |||
684 | static struct dentry *kernfs_iop_lookup(struct inode *dir, | ||
685 | struct dentry *dentry, | ||
686 | unsigned int flags) | ||
687 | { | ||
688 | struct dentry *ret; | ||
689 | struct kernfs_node *parent = dentry->d_parent->d_fsdata; | ||
690 | struct kernfs_node *kn; | ||
691 | struct inode *inode; | ||
692 | const void *ns = NULL; | ||
693 | |||
694 | mutex_lock(&kernfs_mutex); | ||
695 | |||
696 | if (kernfs_ns_enabled(parent)) | ||
697 | ns = kernfs_info(dir->i_sb)->ns; | ||
698 | |||
699 | kn = kernfs_find_ns(parent, dentry->d_name.name, ns); | ||
700 | |||
701 | /* no such entry */ | ||
702 | if (!kn) { | ||
703 | ret = NULL; | ||
704 | goto out_unlock; | ||
705 | } | ||
706 | kernfs_get(kn); | ||
707 | dentry->d_fsdata = kn; | ||
708 | |||
709 | /* attach dentry and inode */ | ||
710 | inode = kernfs_get_inode(dir->i_sb, kn); | ||
711 | if (!inode) { | ||
712 | ret = ERR_PTR(-ENOMEM); | ||
713 | goto out_unlock; | ||
714 | } | ||
715 | |||
716 | /* instantiate and hash dentry */ | ||
717 | ret = d_materialise_unique(dentry, inode); | ||
718 | out_unlock: | ||
719 | mutex_unlock(&kernfs_mutex); | ||
720 | return ret; | ||
721 | } | ||
722 | |||
723 | static int kernfs_iop_mkdir(struct inode *dir, struct dentry *dentry, | ||
724 | umode_t mode) | ||
725 | { | ||
726 | struct kernfs_node *parent = dir->i_private; | ||
727 | struct kernfs_dir_ops *kdops = kernfs_root(parent)->dir_ops; | ||
728 | |||
729 | if (!kdops || !kdops->mkdir) | ||
730 | return -EPERM; | ||
731 | |||
732 | return kdops->mkdir(parent, dentry->d_name.name, mode); | ||
733 | } | ||
734 | |||
735 | static int kernfs_iop_rmdir(struct inode *dir, struct dentry *dentry) | ||
736 | { | ||
737 | struct kernfs_node *kn = dentry->d_fsdata; | ||
738 | struct kernfs_dir_ops *kdops = kernfs_root(kn)->dir_ops; | ||
739 | |||
740 | if (!kdops || !kdops->rmdir) | ||
741 | return -EPERM; | ||
742 | |||
743 | return kdops->rmdir(kn); | ||
744 | } | ||
745 | |||
746 | static int kernfs_iop_rename(struct inode *old_dir, struct dentry *old_dentry, | ||
747 | struct inode *new_dir, struct dentry *new_dentry) | ||
748 | { | ||
749 | struct kernfs_node *kn = old_dentry->d_fsdata; | ||
750 | struct kernfs_node *new_parent = new_dir->i_private; | ||
751 | struct kernfs_dir_ops *kdops = kernfs_root(kn)->dir_ops; | ||
752 | |||
753 | if (!kdops || !kdops->rename) | ||
754 | return -EPERM; | ||
755 | |||
756 | return kdops->rename(kn, new_parent, new_dentry->d_name.name); | ||
757 | } | ||
758 | |||
759 | const struct inode_operations kernfs_dir_iops = { | ||
760 | .lookup = kernfs_iop_lookup, | ||
761 | .permission = kernfs_iop_permission, | ||
762 | .setattr = kernfs_iop_setattr, | ||
763 | .getattr = kernfs_iop_getattr, | ||
764 | .setxattr = kernfs_iop_setxattr, | ||
765 | .removexattr = kernfs_iop_removexattr, | ||
766 | .getxattr = kernfs_iop_getxattr, | ||
767 | .listxattr = kernfs_iop_listxattr, | ||
768 | |||
769 | .mkdir = kernfs_iop_mkdir, | ||
770 | .rmdir = kernfs_iop_rmdir, | ||
771 | .rename = kernfs_iop_rename, | ||
772 | }; | ||
773 | |||
774 | static struct kernfs_node *kernfs_leftmost_descendant(struct kernfs_node *pos) | ||
775 | { | ||
776 | struct kernfs_node *last; | ||
777 | |||
778 | while (true) { | ||
779 | struct rb_node *rbn; | ||
780 | |||
781 | last = pos; | ||
782 | |||
783 | if (kernfs_type(pos) != KERNFS_DIR) | ||
784 | break; | ||
785 | |||
786 | rbn = rb_first(&pos->dir.children); | ||
787 | if (!rbn) | ||
788 | break; | ||
789 | |||
790 | pos = rb_to_kn(rbn); | ||
791 | } | ||
792 | |||
793 | return last; | ||
794 | } | ||
795 | |||
796 | /** | ||
797 | * kernfs_next_descendant_post - find the next descendant for post-order walk | ||
798 | * @pos: the current position (%NULL to initiate traversal) | ||
799 | * @root: kernfs_node whose descendants to walk | ||
800 | * | ||
801 | * Find the next descendant to visit for post-order traversal of @root's | ||
802 | * descendants. @root is included in the iteration and the last node to be | ||
803 | * visited. | ||
804 | */ | ||
805 | static struct kernfs_node *kernfs_next_descendant_post(struct kernfs_node *pos, | ||
806 | struct kernfs_node *root) | ||
807 | { | ||
808 | struct rb_node *rbn; | ||
809 | |||
810 | lockdep_assert_held(&kernfs_mutex); | ||
811 | |||
812 | /* if first iteration, visit leftmost descendant which may be root */ | ||
813 | if (!pos) | ||
814 | return kernfs_leftmost_descendant(root); | ||
815 | |||
816 | /* if we visited @root, we're done */ | ||
817 | if (pos == root) | ||
818 | return NULL; | ||
819 | |||
820 | /* if there's an unvisited sibling, visit its leftmost descendant */ | ||
821 | rbn = rb_next(&pos->rb); | ||
822 | if (rbn) | ||
823 | return kernfs_leftmost_descendant(rb_to_kn(rbn)); | ||
824 | |||
825 | /* no sibling left, visit parent */ | ||
826 | return pos->parent; | ||
827 | } | ||
828 | |||
829 | static void __kernfs_remove(struct kernfs_addrm_cxt *acxt, | ||
830 | struct kernfs_node *kn) | ||
831 | { | ||
832 | struct kernfs_node *pos, *next; | ||
833 | |||
834 | if (!kn) | ||
835 | return; | ||
836 | |||
837 | pr_debug("kernfs %s: removing\n", kn->name); | ||
838 | |||
839 | next = NULL; | ||
840 | do { | ||
841 | pos = next; | ||
842 | next = kernfs_next_descendant_post(pos, kn); | ||
843 | if (pos) | ||
844 | kernfs_remove_one(acxt, pos); | ||
845 | } while (next); | ||
846 | } | ||
847 | |||
848 | /** | ||
849 | * kernfs_remove - remove a kernfs_node recursively | ||
850 | * @kn: the kernfs_node to remove | ||
851 | * | ||
852 | * Remove @kn along with all its subdirectories and files. | ||
853 | */ | ||
854 | void kernfs_remove(struct kernfs_node *kn) | ||
855 | { | ||
856 | struct kernfs_addrm_cxt acxt; | ||
857 | |||
858 | kernfs_addrm_start(&acxt); | ||
859 | __kernfs_remove(&acxt, kn); | ||
860 | kernfs_addrm_finish(&acxt); | ||
861 | } | ||
862 | |||
863 | /** | ||
864 | * kernfs_remove_by_name_ns - find a kernfs_node by name and remove it | ||
865 | * @parent: parent of the target | ||
866 | * @name: name of the kernfs_node to remove | ||
867 | * @ns: namespace tag of the kernfs_node to remove | ||
868 | * | ||
869 | * Look for the kernfs_node with @name and @ns under @parent and remove it. | ||
870 | * Returns 0 on success, -ENOENT if such entry doesn't exist. | ||
871 | */ | ||
872 | int kernfs_remove_by_name_ns(struct kernfs_node *parent, const char *name, | ||
873 | const void *ns) | ||
874 | { | ||
875 | struct kernfs_addrm_cxt acxt; | ||
876 | struct kernfs_node *kn; | ||
877 | |||
878 | if (!parent) { | ||
879 | WARN(1, KERN_WARNING "kernfs: can not remove '%s', no directory\n", | ||
880 | name); | ||
881 | return -ENOENT; | ||
882 | } | ||
883 | |||
884 | kernfs_addrm_start(&acxt); | ||
885 | |||
886 | kn = kernfs_find_ns(parent, name, ns); | ||
887 | if (kn) | ||
888 | __kernfs_remove(&acxt, kn); | ||
889 | |||
890 | kernfs_addrm_finish(&acxt); | ||
891 | |||
892 | if (kn) | ||
893 | return 0; | ||
894 | else | ||
895 | return -ENOENT; | ||
896 | } | ||
897 | |||
898 | /** | ||
899 | * kernfs_rename_ns - move and rename a kernfs_node | ||
900 | * @kn: target node | ||
901 | * @new_parent: new parent to put @sd under | ||
902 | * @new_name: new name | ||
903 | * @new_ns: new namespace tag | ||
904 | */ | ||
905 | int kernfs_rename_ns(struct kernfs_node *kn, struct kernfs_node *new_parent, | ||
906 | const char *new_name, const void *new_ns) | ||
907 | { | ||
908 | int error; | ||
909 | |||
910 | mutex_lock(&kernfs_mutex); | ||
911 | |||
912 | error = -ENOENT; | ||
913 | if ((kn->flags | new_parent->flags) & KERNFS_REMOVED) | ||
914 | goto out; | ||
915 | |||
916 | error = 0; | ||
917 | if ((kn->parent == new_parent) && (kn->ns == new_ns) && | ||
918 | (strcmp(kn->name, new_name) == 0)) | ||
919 | goto out; /* nothing to rename */ | ||
920 | |||
921 | error = -EEXIST; | ||
922 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | ||
923 | goto out; | ||
924 | |||
925 | /* rename kernfs_node */ | ||
926 | if (strcmp(kn->name, new_name) != 0) { | ||
927 | error = -ENOMEM; | ||
928 | new_name = kstrdup(new_name, GFP_KERNEL); | ||
929 | if (!new_name) | ||
930 | goto out; | ||
931 | |||
932 | if (kn->flags & KERNFS_STATIC_NAME) | ||
933 | kn->flags &= ~KERNFS_STATIC_NAME; | ||
934 | else | ||
935 | kfree(kn->name); | ||
936 | |||
937 | kn->name = new_name; | ||
938 | } | ||
939 | |||
940 | /* | ||
941 | * Move to the appropriate place in the appropriate directories rbtree. | ||
942 | */ | ||
943 | kernfs_unlink_sibling(kn); | ||
944 | kernfs_get(new_parent); | ||
945 | kernfs_put(kn->parent); | ||
946 | kn->ns = new_ns; | ||
947 | kn->hash = kernfs_name_hash(kn->name, kn->ns); | ||
948 | kn->parent = new_parent; | ||
949 | kernfs_link_sibling(kn); | ||
950 | |||
951 | error = 0; | ||
952 | out: | ||
953 | mutex_unlock(&kernfs_mutex); | ||
954 | return error; | ||
955 | } | ||
956 | |||
957 | /* Relationship between s_mode and the DT_xxx types */ | ||
958 | static inline unsigned char dt_type(struct kernfs_node *kn) | ||
959 | { | ||
960 | return (kn->mode >> 12) & 15; | ||
961 | } | ||
962 | |||
963 | static int kernfs_dir_fop_release(struct inode *inode, struct file *filp) | ||
964 | { | ||
965 | kernfs_put(filp->private_data); | ||
966 | return 0; | ||
967 | } | ||
968 | |||
969 | static struct kernfs_node *kernfs_dir_pos(const void *ns, | ||
970 | struct kernfs_node *parent, loff_t hash, struct kernfs_node *pos) | ||
971 | { | ||
972 | if (pos) { | ||
973 | int valid = !(pos->flags & KERNFS_REMOVED) && | ||
974 | pos->parent == parent && hash == pos->hash; | ||
975 | kernfs_put(pos); | ||
976 | if (!valid) | ||
977 | pos = NULL; | ||
978 | } | ||
979 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | ||
980 | struct rb_node *node = parent->dir.children.rb_node; | ||
981 | while (node) { | ||
982 | pos = rb_to_kn(node); | ||
983 | |||
984 | if (hash < pos->hash) | ||
985 | node = node->rb_left; | ||
986 | else if (hash > pos->hash) | ||
987 | node = node->rb_right; | ||
988 | else | ||
989 | break; | ||
990 | } | ||
991 | } | ||
992 | /* Skip over entries in the wrong namespace */ | ||
993 | while (pos && pos->ns != ns) { | ||
994 | struct rb_node *node = rb_next(&pos->rb); | ||
995 | if (!node) | ||
996 | pos = NULL; | ||
997 | else | ||
998 | pos = rb_to_kn(node); | ||
999 | } | ||
1000 | return pos; | ||
1001 | } | ||
1002 | |||
1003 | static struct kernfs_node *kernfs_dir_next_pos(const void *ns, | ||
1004 | struct kernfs_node *parent, ino_t ino, struct kernfs_node *pos) | ||
1005 | { | ||
1006 | pos = kernfs_dir_pos(ns, parent, ino, pos); | ||
1007 | if (pos) | ||
1008 | do { | ||
1009 | struct rb_node *node = rb_next(&pos->rb); | ||
1010 | if (!node) | ||
1011 | pos = NULL; | ||
1012 | else | ||
1013 | pos = rb_to_kn(node); | ||
1014 | } while (pos && pos->ns != ns); | ||
1015 | return pos; | ||
1016 | } | ||
1017 | |||
1018 | static int kernfs_fop_readdir(struct file *file, struct dir_context *ctx) | ||
1019 | { | ||
1020 | struct dentry *dentry = file->f_path.dentry; | ||
1021 | struct kernfs_node *parent = dentry->d_fsdata; | ||
1022 | struct kernfs_node *pos = file->private_data; | ||
1023 | const void *ns = NULL; | ||
1024 | |||
1025 | if (!dir_emit_dots(file, ctx)) | ||
1026 | return 0; | ||
1027 | mutex_lock(&kernfs_mutex); | ||
1028 | |||
1029 | if (kernfs_ns_enabled(parent)) | ||
1030 | ns = kernfs_info(dentry->d_sb)->ns; | ||
1031 | |||
1032 | for (pos = kernfs_dir_pos(ns, parent, ctx->pos, pos); | ||
1033 | pos; | ||
1034 | pos = kernfs_dir_next_pos(ns, parent, ctx->pos, pos)) { | ||
1035 | const char *name = pos->name; | ||
1036 | unsigned int type = dt_type(pos); | ||
1037 | int len = strlen(name); | ||
1038 | ino_t ino = pos->ino; | ||
1039 | |||
1040 | ctx->pos = pos->hash; | ||
1041 | file->private_data = pos; | ||
1042 | kernfs_get(pos); | ||
1043 | |||
1044 | mutex_unlock(&kernfs_mutex); | ||
1045 | if (!dir_emit(ctx, name, len, ino, type)) | ||
1046 | return 0; | ||
1047 | mutex_lock(&kernfs_mutex); | ||
1048 | } | ||
1049 | mutex_unlock(&kernfs_mutex); | ||
1050 | file->private_data = NULL; | ||
1051 | ctx->pos = INT_MAX; | ||
1052 | return 0; | ||
1053 | } | ||
1054 | |||
1055 | static loff_t kernfs_dir_fop_llseek(struct file *file, loff_t offset, | ||
1056 | int whence) | ||
1057 | { | ||
1058 | struct inode *inode = file_inode(file); | ||
1059 | loff_t ret; | ||
1060 | |||
1061 | mutex_lock(&inode->i_mutex); | ||
1062 | ret = generic_file_llseek(file, offset, whence); | ||
1063 | mutex_unlock(&inode->i_mutex); | ||
1064 | |||
1065 | return ret; | ||
1066 | } | ||
1067 | |||
1068 | const struct file_operations kernfs_dir_fops = { | ||
1069 | .read = generic_read_dir, | ||
1070 | .iterate = kernfs_fop_readdir, | ||
1071 | .release = kernfs_dir_fop_release, | ||
1072 | .llseek = kernfs_dir_fop_llseek, | ||
1073 | }; | ||