]>
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 TH |
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(sysfs_mutex); | |
21 | ||
22 | #define to_sysfs_dirent(X) rb_entry((X), struct sysfs_dirent, s_rb) | |
23 | ||
fd7b9f7b TH |
24 | /** |
25 | * sysfs_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 sysfs_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 sysfs_name_compare(unsigned int hash, const char *name, | |
48 | const void *ns, const struct sysfs_dirent *sd) | |
49 | { | |
50 | if (hash != sd->s_hash) | |
51 | return hash - sd->s_hash; | |
52 | if (ns != sd->s_ns) | |
53 | return ns - sd->s_ns; | |
54 | return strcmp(name, sd->s_name); | |
55 | } | |
56 | ||
57 | static int sysfs_sd_compare(const struct sysfs_dirent *left, | |
58 | const struct sysfs_dirent *right) | |
59 | { | |
60 | return sysfs_name_compare(left->s_hash, left->s_name, left->s_ns, | |
61 | right); | |
62 | } | |
63 | ||
64 | /** | |
65 | * sysfs_link_sibling - link sysfs_dirent into sibling rbtree | |
66 | * @sd: sysfs_dirent of interest | |
67 | * | |
68 | * Link @sd into its sibling rbtree which starts from | |
69 | * sd->s_parent->s_dir.children. | |
70 | * | |
71 | * Locking: | |
72 | * mutex_lock(sysfs_mutex) | |
73 | * | |
74 | * RETURNS: | |
75 | * 0 on susccess -EEXIST on failure. | |
76 | */ | |
77 | static int sysfs_link_sibling(struct sysfs_dirent *sd) | |
78 | { | |
79 | struct rb_node **node = &sd->s_parent->s_dir.children.rb_node; | |
80 | struct rb_node *parent = NULL; | |
81 | ||
82 | if (sysfs_type(sd) == SYSFS_DIR) | |
83 | sd->s_parent->s_dir.subdirs++; | |
84 | ||
85 | while (*node) { | |
86 | struct sysfs_dirent *pos; | |
87 | int result; | |
88 | ||
89 | pos = to_sysfs_dirent(*node); | |
90 | parent = *node; | |
91 | result = sysfs_sd_compare(sd, pos); | |
92 | if (result < 0) | |
93 | node = &pos->s_rb.rb_left; | |
94 | else if (result > 0) | |
95 | node = &pos->s_rb.rb_right; | |
96 | else | |
97 | return -EEXIST; | |
98 | } | |
99 | /* add new node and rebalance the tree */ | |
100 | rb_link_node(&sd->s_rb, parent, node); | |
101 | rb_insert_color(&sd->s_rb, &sd->s_parent->s_dir.children); | |
102 | return 0; | |
103 | } | |
104 | ||
105 | /** | |
106 | * sysfs_unlink_sibling - unlink sysfs_dirent from sibling rbtree | |
107 | * @sd: sysfs_dirent of interest | |
108 | * | |
109 | * Unlink @sd from its sibling rbtree which starts from | |
110 | * sd->s_parent->s_dir.children. | |
111 | * | |
112 | * Locking: | |
113 | * mutex_lock(sysfs_mutex) | |
114 | */ | |
115 | static void sysfs_unlink_sibling(struct sysfs_dirent *sd) | |
116 | { | |
117 | if (sysfs_type(sd) == SYSFS_DIR) | |
118 | sd->s_parent->s_dir.subdirs--; | |
119 | ||
120 | rb_erase(&sd->s_rb, &sd->s_parent->s_dir.children); | |
121 | } | |
122 | ||
123 | /** | |
124 | * sysfs_get_active - get an active reference to sysfs_dirent | |
125 | * @sd: sysfs_dirent to get an active reference to | |
126 | * | |
127 | * Get an active reference of @sd. This function is noop if @sd | |
128 | * is NULL. | |
129 | * | |
130 | * RETURNS: | |
131 | * Pointer to @sd on success, NULL on failure. | |
132 | */ | |
133 | struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd) | |
134 | { | |
135 | if (unlikely(!sd)) | |
136 | return NULL; | |
137 | ||
138 | if (!atomic_inc_unless_negative(&sd->s_active)) | |
139 | return NULL; | |
140 | ||
141 | if (sd->s_flags & SYSFS_FLAG_LOCKDEP) | |
142 | rwsem_acquire_read(&sd->dep_map, 0, 1, _RET_IP_); | |
143 | return sd; | |
144 | } | |
145 | ||
146 | /** | |
147 | * sysfs_put_active - put an active reference to sysfs_dirent | |
148 | * @sd: sysfs_dirent to put an active reference to | |
149 | * | |
150 | * Put an active reference to @sd. This function is noop if @sd | |
151 | * is NULL. | |
152 | */ | |
153 | void sysfs_put_active(struct sysfs_dirent *sd) | |
154 | { | |
155 | int v; | |
156 | ||
157 | if (unlikely(!sd)) | |
158 | return; | |
159 | ||
160 | if (sd->s_flags & SYSFS_FLAG_LOCKDEP) | |
161 | rwsem_release(&sd->dep_map, 1, _RET_IP_); | |
162 | v = atomic_dec_return(&sd->s_active); | |
163 | if (likely(v != SD_DEACTIVATED_BIAS)) | |
164 | return; | |
165 | ||
166 | /* atomic_dec_return() is a mb(), we'll always see the updated | |
167 | * sd->u.completion. | |
168 | */ | |
169 | complete(sd->u.completion); | |
170 | } | |
171 | ||
172 | /** | |
173 | * sysfs_deactivate - deactivate sysfs_dirent | |
174 | * @sd: sysfs_dirent to deactivate | |
175 | * | |
176 | * Deny new active references and drain existing ones. | |
177 | */ | |
178 | static void sysfs_deactivate(struct sysfs_dirent *sd) | |
179 | { | |
180 | DECLARE_COMPLETION_ONSTACK(wait); | |
181 | int v; | |
182 | ||
183 | BUG_ON(!(sd->s_flags & SYSFS_FLAG_REMOVED)); | |
184 | ||
185 | if (!(sysfs_type(sd) & SYSFS_ACTIVE_REF)) | |
186 | return; | |
187 | ||
188 | sd->u.completion = (void *)&wait; | |
189 | ||
190 | rwsem_acquire(&sd->dep_map, 0, 0, _RET_IP_); | |
191 | /* atomic_add_return() is a mb(), put_active() will always see | |
192 | * the updated sd->u.completion. | |
193 | */ | |
194 | v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active); | |
195 | ||
196 | if (v != SD_DEACTIVATED_BIAS) { | |
197 | lock_contended(&sd->dep_map, _RET_IP_); | |
198 | wait_for_completion(&wait); | |
199 | } | |
200 | ||
201 | lock_acquired(&sd->dep_map, _RET_IP_); | |
202 | rwsem_release(&sd->dep_map, 1, _RET_IP_); | |
203 | } | |
204 | ||
fd7b9f7b TH |
205 | /** |
206 | * kernfs_get - get a reference count on a sysfs_dirent | |
207 | * @sd: the target sysfs_dirent | |
208 | */ | |
209 | void kernfs_get(struct sysfs_dirent *sd) | |
210 | { | |
211 | if (sd) { | |
212 | WARN_ON(!atomic_read(&sd->s_count)); | |
213 | atomic_inc(&sd->s_count); | |
214 | } | |
215 | } | |
216 | EXPORT_SYMBOL_GPL(kernfs_get); | |
217 | ||
218 | /** | |
219 | * kernfs_put - put a reference count on a sysfs_dirent | |
220 | * @sd: the target sysfs_dirent | |
221 | * | |
222 | * Put a reference count of @sd and destroy it if it reached zero. | |
223 | */ | |
224 | void kernfs_put(struct sysfs_dirent *sd) | |
225 | { | |
226 | struct sysfs_dirent *parent_sd; | |
ba7443bc | 227 | struct kernfs_root *root; |
fd7b9f7b TH |
228 | |
229 | if (!sd || !atomic_dec_and_test(&sd->s_count)) | |
230 | return; | |
ba7443bc | 231 | root = kernfs_root(sd); |
fd7b9f7b TH |
232 | repeat: |
233 | /* Moving/renaming is always done while holding reference. | |
234 | * sd->s_parent won't change beneath us. | |
235 | */ | |
236 | parent_sd = sd->s_parent; | |
237 | ||
238 | WARN(!(sd->s_flags & SYSFS_FLAG_REMOVED), | |
239 | "sysfs: free using entry: %s/%s\n", | |
240 | parent_sd ? parent_sd->s_name : "", sd->s_name); | |
241 | ||
242 | if (sysfs_type(sd) == SYSFS_KOBJ_LINK) | |
243 | kernfs_put(sd->s_symlink.target_sd); | |
244 | if (sysfs_type(sd) & SYSFS_COPY_NAME) | |
245 | kfree(sd->s_name); | |
246 | if (sd->s_iattr && sd->s_iattr->ia_secdata) | |
247 | security_release_secctx(sd->s_iattr->ia_secdata, | |
248 | sd->s_iattr->ia_secdata_len); | |
249 | kfree(sd->s_iattr); | |
bc755553 | 250 | ida_simple_remove(&root->ino_ida, sd->s_ino); |
fd7b9f7b TH |
251 | kmem_cache_free(sysfs_dir_cachep, sd); |
252 | ||
253 | sd = parent_sd; | |
ba7443bc TH |
254 | if (sd) { |
255 | if (atomic_dec_and_test(&sd->s_count)) | |
256 | goto repeat; | |
257 | } else { | |
258 | /* just released the root sd, free @root too */ | |
bc755553 | 259 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
260 | kfree(root); |
261 | } | |
fd7b9f7b TH |
262 | } |
263 | EXPORT_SYMBOL_GPL(kernfs_put); | |
264 | ||
265 | static int sysfs_dentry_delete(const struct dentry *dentry) | |
266 | { | |
267 | struct sysfs_dirent *sd = dentry->d_fsdata; | |
268 | return !(sd && !(sd->s_flags & SYSFS_FLAG_REMOVED)); | |
269 | } | |
270 | ||
271 | static int sysfs_dentry_revalidate(struct dentry *dentry, unsigned int flags) | |
272 | { | |
273 | struct sysfs_dirent *sd; | |
274 | ||
275 | if (flags & LOOKUP_RCU) | |
276 | return -ECHILD; | |
277 | ||
278 | sd = dentry->d_fsdata; | |
279 | mutex_lock(&sysfs_mutex); | |
280 | ||
281 | /* The sysfs dirent has been deleted */ | |
282 | if (sd->s_flags & SYSFS_FLAG_REMOVED) | |
283 | goto out_bad; | |
284 | ||
285 | /* The sysfs dirent has been moved? */ | |
286 | if (dentry->d_parent->d_fsdata != sd->s_parent) | |
287 | goto out_bad; | |
288 | ||
289 | /* The sysfs dirent has been renamed */ | |
290 | if (strcmp(dentry->d_name.name, sd->s_name) != 0) | |
291 | goto out_bad; | |
292 | ||
293 | /* The sysfs dirent has been moved to a different namespace */ | |
ac9bba03 | 294 | if (sd->s_parent && kernfs_ns_enabled(sd->s_parent) && |
fd7b9f7b TH |
295 | sysfs_info(dentry->d_sb)->ns != sd->s_ns) |
296 | goto out_bad; | |
297 | ||
298 | mutex_unlock(&sysfs_mutex); | |
299 | out_valid: | |
300 | return 1; | |
301 | out_bad: | |
302 | /* Remove the dentry from the dcache hashes. | |
303 | * If this is a deleted dentry we use d_drop instead of d_delete | |
304 | * so sysfs doesn't need to cope with negative dentries. | |
305 | * | |
306 | * If this is a dentry that has simply been renamed we | |
307 | * use d_drop to remove it from the dcache lookup on its | |
308 | * old parent. If this dentry persists later when a lookup | |
309 | * is performed at its new name the dentry will be readded | |
310 | * to the dcache hashes. | |
311 | */ | |
312 | mutex_unlock(&sysfs_mutex); | |
313 | ||
314 | /* If we have submounts we must allow the vfs caches | |
315 | * to lie about the state of the filesystem to prevent | |
316 | * leaks and other nasty things. | |
317 | */ | |
318 | if (check_submounts_and_drop(dentry) != 0) | |
319 | goto out_valid; | |
320 | ||
321 | return 0; | |
322 | } | |
323 | ||
324 | static void sysfs_dentry_release(struct dentry *dentry) | |
325 | { | |
326 | kernfs_put(dentry->d_fsdata); | |
327 | } | |
328 | ||
329 | const struct dentry_operations sysfs_dentry_ops = { | |
330 | .d_revalidate = sysfs_dentry_revalidate, | |
331 | .d_delete = sysfs_dentry_delete, | |
332 | .d_release = sysfs_dentry_release, | |
333 | }; | |
334 | ||
bc755553 TH |
335 | struct sysfs_dirent *sysfs_new_dirent(struct kernfs_root *root, |
336 | const char *name, umode_t mode, int type) | |
fd7b9f7b TH |
337 | { |
338 | char *dup_name = NULL; | |
339 | struct sysfs_dirent *sd; | |
bc755553 | 340 | int ret; |
fd7b9f7b TH |
341 | |
342 | if (type & SYSFS_COPY_NAME) { | |
343 | name = dup_name = kstrdup(name, GFP_KERNEL); | |
344 | if (!name) | |
345 | return NULL; | |
346 | } | |
347 | ||
348 | sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL); | |
349 | if (!sd) | |
350 | goto err_out1; | |
351 | ||
bc755553 TH |
352 | ret = ida_simple_get(&root->ino_ida, 1, 0, GFP_KERNEL); |
353 | if (ret < 0) | |
fd7b9f7b | 354 | goto err_out2; |
bc755553 | 355 | sd->s_ino = ret; |
fd7b9f7b TH |
356 | |
357 | atomic_set(&sd->s_count, 1); | |
358 | atomic_set(&sd->s_active, 0); | |
359 | ||
360 | sd->s_name = name; | |
361 | sd->s_mode = mode; | |
362 | sd->s_flags = type | SYSFS_FLAG_REMOVED; | |
363 | ||
364 | return sd; | |
365 | ||
366 | err_out2: | |
367 | kmem_cache_free(sysfs_dir_cachep, sd); | |
368 | err_out1: | |
369 | kfree(dup_name); | |
370 | return NULL; | |
371 | } | |
372 | ||
373 | /** | |
374 | * sysfs_addrm_start - prepare for sysfs_dirent add/remove | |
375 | * @acxt: pointer to sysfs_addrm_cxt to be used | |
376 | * | |
377 | * This function is called when the caller is about to add or remove | |
378 | * sysfs_dirent. This function acquires sysfs_mutex. @acxt is used | |
379 | * to keep and pass context to other addrm functions. | |
380 | * | |
381 | * LOCKING: | |
382 | * Kernel thread context (may sleep). sysfs_mutex is locked on | |
383 | * return. | |
384 | */ | |
385 | void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt) | |
386 | __acquires(sysfs_mutex) | |
387 | { | |
388 | memset(acxt, 0, sizeof(*acxt)); | |
389 | ||
390 | mutex_lock(&sysfs_mutex); | |
391 | } | |
392 | ||
393 | /** | |
394 | * sysfs_add_one - add sysfs_dirent to parent without warning | |
395 | * @acxt: addrm context to use | |
396 | * @sd: sysfs_dirent to be added | |
397 | * @parent_sd: the parent sysfs_dirent to add @sd to | |
398 | * | |
399 | * Get @parent_sd and set @sd->s_parent to it and increment nlink of | |
400 | * the parent inode if @sd is a directory and link into the children | |
401 | * list of the parent. | |
402 | * | |
403 | * This function should be called between calls to | |
404 | * sysfs_addrm_start() and sysfs_addrm_finish() and should be | |
405 | * passed the same @acxt as passed to sysfs_addrm_start(). | |
406 | * | |
407 | * LOCKING: | |
408 | * Determined by sysfs_addrm_start(). | |
409 | * | |
410 | * RETURNS: | |
411 | * 0 on success, -EEXIST if entry with the given name already | |
412 | * exists. | |
413 | */ | |
414 | int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd, | |
415 | struct sysfs_dirent *parent_sd) | |
416 | { | |
ac9bba03 | 417 | bool has_ns = kernfs_ns_enabled(parent_sd); |
fd7b9f7b TH |
418 | struct sysfs_inode_attrs *ps_iattr; |
419 | int ret; | |
420 | ||
421 | if (has_ns != (bool)sd->s_ns) { | |
422 | WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n", | |
423 | has_ns ? "required" : "invalid", | |
424 | parent_sd->s_name, sd->s_name); | |
425 | return -EINVAL; | |
426 | } | |
427 | ||
428 | if (sysfs_type(parent_sd) != SYSFS_DIR) | |
429 | return -EINVAL; | |
430 | ||
431 | sd->s_hash = sysfs_name_hash(sd->s_name, sd->s_ns); | |
432 | sd->s_parent = parent_sd; | |
433 | kernfs_get(parent_sd); | |
434 | ||
435 | ret = sysfs_link_sibling(sd); | |
436 | if (ret) | |
437 | return ret; | |
438 | ||
439 | /* Update timestamps on the parent */ | |
440 | ps_iattr = parent_sd->s_iattr; | |
441 | if (ps_iattr) { | |
442 | struct iattr *ps_iattrs = &ps_iattr->ia_iattr; | |
443 | ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME; | |
444 | } | |
445 | ||
446 | /* Mark the entry added into directory tree */ | |
447 | sd->s_flags &= ~SYSFS_FLAG_REMOVED; | |
448 | ||
449 | return 0; | |
450 | } | |
451 | ||
452 | /** | |
453 | * sysfs_remove_one - remove sysfs_dirent from parent | |
454 | * @acxt: addrm context to use | |
455 | * @sd: sysfs_dirent to be removed | |
456 | * | |
457 | * Mark @sd removed and drop nlink of parent inode if @sd is a | |
458 | * directory. @sd is unlinked from the children list. | |
459 | * | |
460 | * This function should be called between calls to | |
461 | * sysfs_addrm_start() and sysfs_addrm_finish() and should be | |
462 | * passed the same @acxt as passed to sysfs_addrm_start(). | |
463 | * | |
464 | * LOCKING: | |
465 | * Determined by sysfs_addrm_start(). | |
466 | */ | |
467 | static void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, | |
468 | struct sysfs_dirent *sd) | |
469 | { | |
470 | struct sysfs_inode_attrs *ps_iattr; | |
471 | ||
472 | /* | |
473 | * Removal can be called multiple times on the same node. Only the | |
474 | * first invocation is effective and puts the base ref. | |
475 | */ | |
476 | if (sd->s_flags & SYSFS_FLAG_REMOVED) | |
477 | return; | |
478 | ||
ba7443bc TH |
479 | if (sd->s_parent) { |
480 | sysfs_unlink_sibling(sd); | |
fd7b9f7b | 481 | |
ba7443bc TH |
482 | /* Update timestamps on the parent */ |
483 | ps_iattr = sd->s_parent->s_iattr; | |
484 | if (ps_iattr) { | |
485 | ps_iattr->ia_iattr.ia_ctime = CURRENT_TIME; | |
486 | ps_iattr->ia_iattr.ia_mtime = CURRENT_TIME; | |
487 | } | |
fd7b9f7b TH |
488 | } |
489 | ||
490 | sd->s_flags |= SYSFS_FLAG_REMOVED; | |
491 | sd->u.removed_list = acxt->removed; | |
492 | acxt->removed = sd; | |
493 | } | |
494 | ||
495 | /** | |
496 | * sysfs_addrm_finish - finish up sysfs_dirent add/remove | |
497 | * @acxt: addrm context to finish up | |
498 | * | |
499 | * Finish up sysfs_dirent add/remove. Resources acquired by | |
500 | * sysfs_addrm_start() are released and removed sysfs_dirents are | |
501 | * cleaned up. | |
502 | * | |
503 | * LOCKING: | |
504 | * sysfs_mutex is released. | |
505 | */ | |
506 | void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt) | |
507 | __releases(sysfs_mutex) | |
508 | { | |
509 | /* release resources acquired by sysfs_addrm_start() */ | |
510 | mutex_unlock(&sysfs_mutex); | |
511 | ||
512 | /* kill removed sysfs_dirents */ | |
513 | while (acxt->removed) { | |
514 | struct sysfs_dirent *sd = acxt->removed; | |
515 | ||
516 | acxt->removed = sd->u.removed_list; | |
517 | ||
518 | sysfs_deactivate(sd); | |
519 | sysfs_unmap_bin_file(sd); | |
520 | kernfs_put(sd); | |
521 | } | |
522 | } | |
523 | ||
524 | /** | |
525 | * kernfs_find_ns - find sysfs_dirent with the given name | |
526 | * @parent: sysfs_dirent to search under | |
527 | * @name: name to look for | |
528 | * @ns: the namespace tag to use | |
529 | * | |
530 | * Look for sysfs_dirent with name @name under @parent. Returns pointer to | |
531 | * the found sysfs_dirent on success, %NULL on failure. | |
532 | */ | |
533 | static struct sysfs_dirent *kernfs_find_ns(struct sysfs_dirent *parent, | |
534 | const unsigned char *name, | |
535 | const void *ns) | |
536 | { | |
537 | struct rb_node *node = parent->s_dir.children.rb_node; | |
ac9bba03 | 538 | bool has_ns = kernfs_ns_enabled(parent); |
fd7b9f7b TH |
539 | unsigned int hash; |
540 | ||
541 | lockdep_assert_held(&sysfs_mutex); | |
542 | ||
543 | if (has_ns != (bool)ns) { | |
544 | WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n", | |
545 | has_ns ? "required" : "invalid", | |
546 | parent->s_name, name); | |
547 | return NULL; | |
548 | } | |
549 | ||
550 | hash = sysfs_name_hash(name, ns); | |
551 | while (node) { | |
552 | struct sysfs_dirent *sd; | |
553 | int result; | |
554 | ||
555 | sd = to_sysfs_dirent(node); | |
556 | result = sysfs_name_compare(hash, name, ns, sd); | |
557 | if (result < 0) | |
558 | node = node->rb_left; | |
559 | else if (result > 0) | |
560 | node = node->rb_right; | |
561 | else | |
562 | return sd; | |
563 | } | |
564 | return NULL; | |
565 | } | |
566 | ||
567 | /** | |
568 | * kernfs_find_and_get_ns - find and get sysfs_dirent with the given name | |
569 | * @parent: sysfs_dirent to search under | |
570 | * @name: name to look for | |
571 | * @ns: the namespace tag to use | |
572 | * | |
573 | * Look for sysfs_dirent with name @name under @parent and get a reference | |
574 | * if found. This function may sleep and returns pointer to the found | |
575 | * sysfs_dirent on success, %NULL on failure. | |
576 | */ | |
577 | struct sysfs_dirent *kernfs_find_and_get_ns(struct sysfs_dirent *parent, | |
578 | const char *name, const void *ns) | |
579 | { | |
580 | struct sysfs_dirent *sd; | |
581 | ||
582 | mutex_lock(&sysfs_mutex); | |
583 | sd = kernfs_find_ns(parent, name, ns); | |
584 | kernfs_get(sd); | |
585 | mutex_unlock(&sysfs_mutex); | |
586 | ||
587 | return sd; | |
588 | } | |
589 | EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns); | |
590 | ||
ba7443bc TH |
591 | /** |
592 | * kernfs_create_root - create a new kernfs hierarchy | |
593 | * @priv: opaque data associated with the new directory | |
594 | * | |
595 | * Returns the root of the new hierarchy on success, ERR_PTR() value on | |
596 | * failure. | |
597 | */ | |
598 | struct kernfs_root *kernfs_create_root(void *priv) | |
599 | { | |
600 | struct kernfs_root *root; | |
601 | struct sysfs_dirent *sd; | |
602 | ||
603 | root = kzalloc(sizeof(*root), GFP_KERNEL); | |
604 | if (!root) | |
605 | return ERR_PTR(-ENOMEM); | |
606 | ||
bc755553 TH |
607 | ida_init(&root->ino_ida); |
608 | ||
609 | sd = sysfs_new_dirent(root, "", S_IFDIR | S_IRUGO | S_IXUGO, SYSFS_DIR); | |
ba7443bc | 610 | if (!sd) { |
bc755553 | 611 | ida_destroy(&root->ino_ida); |
ba7443bc TH |
612 | kfree(root); |
613 | return ERR_PTR(-ENOMEM); | |
614 | } | |
615 | ||
616 | sd->s_flags &= ~SYSFS_FLAG_REMOVED; | |
617 | sd->priv = priv; | |
618 | sd->s_dir.root = root; | |
619 | ||
620 | root->sd = sd; | |
621 | ||
622 | return root; | |
623 | } | |
624 | ||
625 | /** | |
626 | * kernfs_destroy_root - destroy a kernfs hierarchy | |
627 | * @root: root of the hierarchy to destroy | |
628 | * | |
629 | * Destroy the hierarchy anchored at @root by removing all existing | |
630 | * directories and destroying @root. | |
631 | */ | |
632 | void kernfs_destroy_root(struct kernfs_root *root) | |
633 | { | |
634 | kernfs_remove(root->sd); /* will also free @root */ | |
635 | } | |
636 | ||
fd7b9f7b TH |
637 | /** |
638 | * kernfs_create_dir_ns - create a directory | |
639 | * @parent: parent in which to create a new directory | |
640 | * @name: name of the new directory | |
641 | * @priv: opaque data associated with the new directory | |
642 | * @ns: optional namespace tag of the directory | |
643 | * | |
644 | * Returns the created node on success, ERR_PTR() value on failure. | |
645 | */ | |
646 | struct sysfs_dirent *kernfs_create_dir_ns(struct sysfs_dirent *parent, | |
647 | const char *name, void *priv, | |
648 | const void *ns) | |
649 | { | |
650 | umode_t mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO; | |
651 | struct sysfs_addrm_cxt acxt; | |
652 | struct sysfs_dirent *sd; | |
653 | int rc; | |
654 | ||
655 | /* allocate */ | |
bc755553 | 656 | sd = sysfs_new_dirent(kernfs_root(parent), name, mode, SYSFS_DIR); |
fd7b9f7b TH |
657 | if (!sd) |
658 | return ERR_PTR(-ENOMEM); | |
659 | ||
ba7443bc | 660 | sd->s_dir.root = parent->s_dir.root; |
fd7b9f7b TH |
661 | sd->s_ns = ns; |
662 | sd->priv = priv; | |
663 | ||
664 | /* link in */ | |
665 | sysfs_addrm_start(&acxt); | |
666 | rc = sysfs_add_one(&acxt, sd, parent); | |
667 | sysfs_addrm_finish(&acxt); | |
668 | ||
669 | if (!rc) | |
670 | return sd; | |
671 | ||
672 | kernfs_put(sd); | |
673 | return ERR_PTR(rc); | |
674 | } | |
675 | ||
676 | static struct dentry *sysfs_lookup(struct inode *dir, struct dentry *dentry, | |
677 | unsigned int flags) | |
678 | { | |
679 | struct dentry *ret = NULL; | |
680 | struct dentry *parent = dentry->d_parent; | |
681 | struct sysfs_dirent *parent_sd = parent->d_fsdata; | |
682 | struct sysfs_dirent *sd; | |
683 | struct inode *inode; | |
684 | const void *ns = NULL; | |
685 | ||
686 | mutex_lock(&sysfs_mutex); | |
687 | ||
ac9bba03 | 688 | if (kernfs_ns_enabled(parent_sd)) |
fd7b9f7b TH |
689 | ns = sysfs_info(dir->i_sb)->ns; |
690 | ||
691 | sd = kernfs_find_ns(parent_sd, dentry->d_name.name, ns); | |
692 | ||
693 | /* no such entry */ | |
694 | if (!sd) { | |
695 | ret = ERR_PTR(-ENOENT); | |
696 | goto out_unlock; | |
697 | } | |
698 | kernfs_get(sd); | |
699 | dentry->d_fsdata = sd; | |
700 | ||
701 | /* attach dentry and inode */ | |
702 | inode = sysfs_get_inode(dir->i_sb, sd); | |
703 | if (!inode) { | |
704 | ret = ERR_PTR(-ENOMEM); | |
705 | goto out_unlock; | |
706 | } | |
707 | ||
708 | /* instantiate and hash dentry */ | |
709 | ret = d_materialise_unique(dentry, inode); | |
710 | out_unlock: | |
711 | mutex_unlock(&sysfs_mutex); | |
712 | return ret; | |
713 | } | |
714 | ||
715 | const struct inode_operations sysfs_dir_inode_operations = { | |
716 | .lookup = sysfs_lookup, | |
717 | .permission = sysfs_permission, | |
718 | .setattr = sysfs_setattr, | |
719 | .getattr = sysfs_getattr, | |
720 | .setxattr = sysfs_setxattr, | |
721 | }; | |
722 | ||
723 | static struct sysfs_dirent *sysfs_leftmost_descendant(struct sysfs_dirent *pos) | |
724 | { | |
725 | struct sysfs_dirent *last; | |
726 | ||
727 | while (true) { | |
728 | struct rb_node *rbn; | |
729 | ||
730 | last = pos; | |
731 | ||
732 | if (sysfs_type(pos) != SYSFS_DIR) | |
733 | break; | |
734 | ||
735 | rbn = rb_first(&pos->s_dir.children); | |
736 | if (!rbn) | |
737 | break; | |
738 | ||
739 | pos = to_sysfs_dirent(rbn); | |
740 | } | |
741 | ||
742 | return last; | |
743 | } | |
744 | ||
745 | /** | |
746 | * sysfs_next_descendant_post - find the next descendant for post-order walk | |
747 | * @pos: the current position (%NULL to initiate traversal) | |
748 | * @root: sysfs_dirent whose descendants to walk | |
749 | * | |
750 | * Find the next descendant to visit for post-order traversal of @root's | |
751 | * descendants. @root is included in the iteration and the last node to be | |
752 | * visited. | |
753 | */ | |
754 | static struct sysfs_dirent *sysfs_next_descendant_post(struct sysfs_dirent *pos, | |
755 | struct sysfs_dirent *root) | |
756 | { | |
757 | struct rb_node *rbn; | |
758 | ||
759 | lockdep_assert_held(&sysfs_mutex); | |
760 | ||
761 | /* if first iteration, visit leftmost descendant which may be root */ | |
762 | if (!pos) | |
763 | return sysfs_leftmost_descendant(root); | |
764 | ||
765 | /* if we visited @root, we're done */ | |
766 | if (pos == root) | |
767 | return NULL; | |
768 | ||
769 | /* if there's an unvisited sibling, visit its leftmost descendant */ | |
770 | rbn = rb_next(&pos->s_rb); | |
771 | if (rbn) | |
772 | return sysfs_leftmost_descendant(to_sysfs_dirent(rbn)); | |
773 | ||
774 | /* no sibling left, visit parent */ | |
775 | return pos->s_parent; | |
776 | } | |
777 | ||
778 | static void __kernfs_remove(struct sysfs_addrm_cxt *acxt, | |
779 | struct sysfs_dirent *sd) | |
780 | { | |
781 | struct sysfs_dirent *pos, *next; | |
782 | ||
783 | if (!sd) | |
784 | return; | |
785 | ||
786 | pr_debug("sysfs %s: removing\n", sd->s_name); | |
787 | ||
788 | next = NULL; | |
789 | do { | |
790 | pos = next; | |
791 | next = sysfs_next_descendant_post(pos, sd); | |
792 | if (pos) | |
793 | sysfs_remove_one(acxt, pos); | |
794 | } while (next); | |
795 | } | |
796 | ||
797 | /** | |
798 | * kernfs_remove - remove a sysfs_dirent recursively | |
799 | * @sd: the sysfs_dirent to remove | |
800 | * | |
801 | * Remove @sd along with all its subdirectories and files. | |
802 | */ | |
803 | void kernfs_remove(struct sysfs_dirent *sd) | |
804 | { | |
805 | struct sysfs_addrm_cxt acxt; | |
806 | ||
807 | sysfs_addrm_start(&acxt); | |
808 | __kernfs_remove(&acxt, sd); | |
809 | sysfs_addrm_finish(&acxt); | |
810 | } | |
811 | ||
812 | /** | |
813 | * kernfs_remove_by_name_ns - find a sysfs_dirent by name and remove it | |
814 | * @dir_sd: parent of the target | |
815 | * @name: name of the sysfs_dirent to remove | |
816 | * @ns: namespace tag of the sysfs_dirent to remove | |
817 | * | |
818 | * Look for the sysfs_dirent with @name and @ns under @dir_sd and remove | |
819 | * it. Returns 0 on success, -ENOENT if such entry doesn't exist. | |
820 | */ | |
821 | int kernfs_remove_by_name_ns(struct sysfs_dirent *dir_sd, const char *name, | |
822 | const void *ns) | |
823 | { | |
824 | struct sysfs_addrm_cxt acxt; | |
825 | struct sysfs_dirent *sd; | |
826 | ||
827 | if (!dir_sd) { | |
828 | WARN(1, KERN_WARNING "sysfs: can not remove '%s', no directory\n", | |
829 | name); | |
830 | return -ENOENT; | |
831 | } | |
832 | ||
833 | sysfs_addrm_start(&acxt); | |
834 | ||
835 | sd = kernfs_find_ns(dir_sd, name, ns); | |
836 | if (sd) | |
837 | __kernfs_remove(&acxt, sd); | |
838 | ||
839 | sysfs_addrm_finish(&acxt); | |
840 | ||
841 | if (sd) | |
842 | return 0; | |
843 | else | |
844 | return -ENOENT; | |
845 | } | |
846 | ||
847 | /** | |
848 | * kernfs_rename_ns - move and rename a kernfs_node | |
849 | * @sd: target node | |
850 | * @new_parent: new parent to put @sd under | |
851 | * @new_name: new name | |
852 | * @new_ns: new namespace tag | |
853 | */ | |
854 | int kernfs_rename_ns(struct sysfs_dirent *sd, struct sysfs_dirent *new_parent, | |
855 | const char *new_name, const void *new_ns) | |
856 | { | |
857 | int error; | |
858 | ||
859 | mutex_lock(&sysfs_mutex); | |
860 | ||
861 | error = 0; | |
862 | if ((sd->s_parent == new_parent) && (sd->s_ns == new_ns) && | |
863 | (strcmp(sd->s_name, new_name) == 0)) | |
864 | goto out; /* nothing to rename */ | |
865 | ||
866 | error = -EEXIST; | |
867 | if (kernfs_find_ns(new_parent, new_name, new_ns)) | |
868 | goto out; | |
869 | ||
870 | /* rename sysfs_dirent */ | |
871 | if (strcmp(sd->s_name, new_name) != 0) { | |
872 | error = -ENOMEM; | |
873 | new_name = kstrdup(new_name, GFP_KERNEL); | |
874 | if (!new_name) | |
875 | goto out; | |
876 | ||
877 | kfree(sd->s_name); | |
878 | sd->s_name = new_name; | |
879 | } | |
880 | ||
881 | /* | |
882 | * Move to the appropriate place in the appropriate directories rbtree. | |
883 | */ | |
884 | sysfs_unlink_sibling(sd); | |
885 | kernfs_get(new_parent); | |
886 | kernfs_put(sd->s_parent); | |
887 | sd->s_ns = new_ns; | |
888 | sd->s_hash = sysfs_name_hash(sd->s_name, sd->s_ns); | |
889 | sd->s_parent = new_parent; | |
890 | sysfs_link_sibling(sd); | |
891 | ||
892 | error = 0; | |
893 | out: | |
894 | mutex_unlock(&sysfs_mutex); | |
895 | return error; | |
896 | } | |
897 | ||
fd7b9f7b TH |
898 | /* Relationship between s_mode and the DT_xxx types */ |
899 | static inline unsigned char dt_type(struct sysfs_dirent *sd) | |
900 | { | |
901 | return (sd->s_mode >> 12) & 15; | |
902 | } | |
903 | ||
904 | static int sysfs_dir_release(struct inode *inode, struct file *filp) | |
905 | { | |
906 | kernfs_put(filp->private_data); | |
907 | return 0; | |
908 | } | |
909 | ||
910 | static struct sysfs_dirent *sysfs_dir_pos(const void *ns, | |
911 | struct sysfs_dirent *parent_sd, loff_t hash, struct sysfs_dirent *pos) | |
912 | { | |
913 | if (pos) { | |
914 | int valid = !(pos->s_flags & SYSFS_FLAG_REMOVED) && | |
915 | pos->s_parent == parent_sd && | |
916 | hash == pos->s_hash; | |
917 | kernfs_put(pos); | |
918 | if (!valid) | |
919 | pos = NULL; | |
920 | } | |
921 | if (!pos && (hash > 1) && (hash < INT_MAX)) { | |
922 | struct rb_node *node = parent_sd->s_dir.children.rb_node; | |
923 | while (node) { | |
924 | pos = to_sysfs_dirent(node); | |
925 | ||
926 | if (hash < pos->s_hash) | |
927 | node = node->rb_left; | |
928 | else if (hash > pos->s_hash) | |
929 | node = node->rb_right; | |
930 | else | |
931 | break; | |
932 | } | |
933 | } | |
934 | /* Skip over entries in the wrong namespace */ | |
935 | while (pos && pos->s_ns != ns) { | |
936 | struct rb_node *node = rb_next(&pos->s_rb); | |
937 | if (!node) | |
938 | pos = NULL; | |
939 | else | |
940 | pos = to_sysfs_dirent(node); | |
941 | } | |
942 | return pos; | |
943 | } | |
944 | ||
945 | static struct sysfs_dirent *sysfs_dir_next_pos(const void *ns, | |
946 | struct sysfs_dirent *parent_sd, ino_t ino, struct sysfs_dirent *pos) | |
947 | { | |
948 | pos = sysfs_dir_pos(ns, parent_sd, ino, pos); | |
949 | if (pos) | |
950 | do { | |
951 | struct rb_node *node = rb_next(&pos->s_rb); | |
952 | if (!node) | |
953 | pos = NULL; | |
954 | else | |
955 | pos = to_sysfs_dirent(node); | |
956 | } while (pos && pos->s_ns != ns); | |
957 | return pos; | |
958 | } | |
959 | ||
960 | static int sysfs_readdir(struct file *file, struct dir_context *ctx) | |
961 | { | |
962 | struct dentry *dentry = file->f_path.dentry; | |
963 | struct sysfs_dirent *parent_sd = dentry->d_fsdata; | |
964 | struct sysfs_dirent *pos = file->private_data; | |
965 | const void *ns = NULL; | |
966 | ||
967 | if (!dir_emit_dots(file, ctx)) | |
968 | return 0; | |
969 | mutex_lock(&sysfs_mutex); | |
970 | ||
ac9bba03 | 971 | if (kernfs_ns_enabled(parent_sd)) |
fd7b9f7b TH |
972 | ns = sysfs_info(dentry->d_sb)->ns; |
973 | ||
974 | for (pos = sysfs_dir_pos(ns, parent_sd, ctx->pos, pos); | |
975 | pos; | |
976 | pos = sysfs_dir_next_pos(ns, parent_sd, ctx->pos, pos)) { | |
977 | const char *name = pos->s_name; | |
978 | unsigned int type = dt_type(pos); | |
979 | int len = strlen(name); | |
980 | ino_t ino = pos->s_ino; | |
981 | ||
982 | ctx->pos = pos->s_hash; | |
983 | file->private_data = pos; | |
984 | kernfs_get(pos); | |
985 | ||
986 | mutex_unlock(&sysfs_mutex); | |
987 | if (!dir_emit(ctx, name, len, ino, type)) | |
988 | return 0; | |
989 | mutex_lock(&sysfs_mutex); | |
990 | } | |
991 | mutex_unlock(&sysfs_mutex); | |
992 | file->private_data = NULL; | |
993 | ctx->pos = INT_MAX; | |
994 | return 0; | |
995 | } | |
996 | ||
997 | static loff_t sysfs_dir_llseek(struct file *file, loff_t offset, int whence) | |
998 | { | |
999 | struct inode *inode = file_inode(file); | |
1000 | loff_t ret; | |
1001 | ||
1002 | mutex_lock(&inode->i_mutex); | |
1003 | ret = generic_file_llseek(file, offset, whence); | |
1004 | mutex_unlock(&inode->i_mutex); | |
1005 | ||
1006 | return ret; | |
1007 | } | |
1008 | ||
1009 | const struct file_operations sysfs_dir_operations = { | |
1010 | .read = generic_read_dir, | |
1011 | .iterate = sysfs_readdir, | |
1012 | .release = sysfs_dir_release, | |
1013 | .llseek = sysfs_dir_llseek, | |
1014 | }; |