Loading...
1/*
2 * fs/sysfs/dir.c - sysfs core and dir operation implementation
3 *
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007 Tejun Heo <teheo@suse.de>
7 *
8 * This file is released under the GPLv2.
9 *
10 * Please see Documentation/filesystems/sysfs.txt for more information.
11 */
12
13#undef DEBUG
14
15#include <linux/fs.h>
16#include <linux/mount.h>
17#include <linux/module.h>
18#include <linux/kobject.h>
19#include <linux/namei.h>
20#include <linux/idr.h>
21#include <linux/completion.h>
22#include <linux/mutex.h>
23#include <linux/slab.h>
24#include <linux/security.h>
25#include <linux/hash.h>
26#include "sysfs.h"
27
28DEFINE_MUTEX(sysfs_mutex);
29DEFINE_SPINLOCK(sysfs_assoc_lock);
30
31#define to_sysfs_dirent(X) rb_entry((X), struct sysfs_dirent, s_rb);
32
33static DEFINE_SPINLOCK(sysfs_ino_lock);
34static DEFINE_IDA(sysfs_ino_ida);
35
36/**
37 * sysfs_name_hash
38 * @ns: Namespace tag to hash
39 * @name: Null terminated string to hash
40 *
41 * Returns 31 bit hash of ns + name (so it fits in an off_t )
42 */
43static unsigned int sysfs_name_hash(const void *ns, const char *name)
44{
45 unsigned long hash = init_name_hash();
46 unsigned int len = strlen(name);
47 while (len--)
48 hash = partial_name_hash(*name++, hash);
49 hash = ( end_name_hash(hash) ^ hash_ptr( (void *)ns, 31 ) );
50 hash &= 0x7fffffffU;
51 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
52 if (hash < 1)
53 hash += 2;
54 if (hash >= INT_MAX)
55 hash = INT_MAX - 1;
56 return hash;
57}
58
59static int sysfs_name_compare(unsigned int hash, const void *ns,
60 const char *name, const struct sysfs_dirent *sd)
61{
62 if (hash != sd->s_hash)
63 return hash - sd->s_hash;
64 if (ns != sd->s_ns)
65 return ns - sd->s_ns;
66 return strcmp(name, sd->s_name);
67}
68
69static int sysfs_sd_compare(const struct sysfs_dirent *left,
70 const struct sysfs_dirent *right)
71{
72 return sysfs_name_compare(left->s_hash, left->s_ns, left->s_name,
73 right);
74}
75
76/**
77 * sysfs_link_subling - link sysfs_dirent into sibling rbtree
78 * @sd: sysfs_dirent of interest
79 *
80 * Link @sd into its sibling rbtree which starts from
81 * sd->s_parent->s_dir.children.
82 *
83 * Locking:
84 * mutex_lock(sysfs_mutex)
85 *
86 * RETURNS:
87 * 0 on susccess -EEXIST on failure.
88 */
89static int sysfs_link_sibling(struct sysfs_dirent *sd)
90{
91 struct rb_node **node = &sd->s_parent->s_dir.children.rb_node;
92 struct rb_node *parent = NULL;
93
94 if (sysfs_type(sd) == SYSFS_DIR)
95 sd->s_parent->s_dir.subdirs++;
96
97 while (*node) {
98 struct sysfs_dirent *pos;
99 int result;
100
101 pos = to_sysfs_dirent(*node);
102 parent = *node;
103 result = sysfs_sd_compare(sd, pos);
104 if (result < 0)
105 node = &pos->s_rb.rb_left;
106 else if (result > 0)
107 node = &pos->s_rb.rb_right;
108 else
109 return -EEXIST;
110 }
111 /* add new node and rebalance the tree */
112 rb_link_node(&sd->s_rb, parent, node);
113 rb_insert_color(&sd->s_rb, &sd->s_parent->s_dir.children);
114 return 0;
115}
116
117/**
118 * sysfs_unlink_sibling - unlink sysfs_dirent from sibling rbtree
119 * @sd: sysfs_dirent of interest
120 *
121 * Unlink @sd from its sibling rbtree which starts from
122 * sd->s_parent->s_dir.children.
123 *
124 * Locking:
125 * mutex_lock(sysfs_mutex)
126 */
127static void sysfs_unlink_sibling(struct sysfs_dirent *sd)
128{
129 if (sysfs_type(sd) == SYSFS_DIR)
130 sd->s_parent->s_dir.subdirs--;
131
132 rb_erase(&sd->s_rb, &sd->s_parent->s_dir.children);
133}
134
135#ifdef CONFIG_DEBUG_LOCK_ALLOC
136
137/* Test for attributes that want to ignore lockdep for read-locking */
138static bool ignore_lockdep(struct sysfs_dirent *sd)
139{
140 return sysfs_type(sd) == SYSFS_KOBJ_ATTR &&
141 sd->s_attr.attr->ignore_lockdep;
142}
143
144#else
145
146static inline bool ignore_lockdep(struct sysfs_dirent *sd)
147{
148 return true;
149}
150
151#endif
152
153/**
154 * sysfs_get_active - get an active reference to sysfs_dirent
155 * @sd: sysfs_dirent to get an active reference to
156 *
157 * Get an active reference of @sd. This function is noop if @sd
158 * is NULL.
159 *
160 * RETURNS:
161 * Pointer to @sd on success, NULL on failure.
162 */
163struct sysfs_dirent *sysfs_get_active(struct sysfs_dirent *sd)
164{
165 if (unlikely(!sd))
166 return NULL;
167
168 while (1) {
169 int v, t;
170
171 v = atomic_read(&sd->s_active);
172 if (unlikely(v < 0))
173 return NULL;
174
175 t = atomic_cmpxchg(&sd->s_active, v, v + 1);
176 if (likely(t == v))
177 break;
178 if (t < 0)
179 return NULL;
180
181 cpu_relax();
182 }
183
184 if (likely(!ignore_lockdep(sd)))
185 rwsem_acquire_read(&sd->dep_map, 0, 1, _RET_IP_);
186 return sd;
187}
188
189/**
190 * sysfs_put_active - put an active reference to sysfs_dirent
191 * @sd: sysfs_dirent to put an active reference to
192 *
193 * Put an active reference to @sd. This function is noop if @sd
194 * is NULL.
195 */
196void sysfs_put_active(struct sysfs_dirent *sd)
197{
198 int v;
199
200 if (unlikely(!sd))
201 return;
202
203 if (likely(!ignore_lockdep(sd)))
204 rwsem_release(&sd->dep_map, 1, _RET_IP_);
205 v = atomic_dec_return(&sd->s_active);
206 if (likely(v != SD_DEACTIVATED_BIAS))
207 return;
208
209 /* atomic_dec_return() is a mb(), we'll always see the updated
210 * sd->u.completion.
211 */
212 complete(sd->u.completion);
213}
214
215/**
216 * sysfs_deactivate - deactivate sysfs_dirent
217 * @sd: sysfs_dirent to deactivate
218 *
219 * Deny new active references and drain existing ones.
220 */
221static void sysfs_deactivate(struct sysfs_dirent *sd)
222{
223 DECLARE_COMPLETION_ONSTACK(wait);
224 int v;
225
226 BUG_ON(!(sd->s_flags & SYSFS_FLAG_REMOVED));
227
228 if (!(sysfs_type(sd) & SYSFS_ACTIVE_REF))
229 return;
230
231 sd->u.completion = (void *)&wait;
232
233 rwsem_acquire(&sd->dep_map, 0, 0, _RET_IP_);
234 /* atomic_add_return() is a mb(), put_active() will always see
235 * the updated sd->u.completion.
236 */
237 v = atomic_add_return(SD_DEACTIVATED_BIAS, &sd->s_active);
238
239 if (v != SD_DEACTIVATED_BIAS) {
240 lock_contended(&sd->dep_map, _RET_IP_);
241 wait_for_completion(&wait);
242 }
243
244 lock_acquired(&sd->dep_map, _RET_IP_);
245 rwsem_release(&sd->dep_map, 1, _RET_IP_);
246}
247
248static int sysfs_alloc_ino(unsigned int *pino)
249{
250 int ino, rc;
251
252 retry:
253 spin_lock(&sysfs_ino_lock);
254 rc = ida_get_new_above(&sysfs_ino_ida, 2, &ino);
255 spin_unlock(&sysfs_ino_lock);
256
257 if (rc == -EAGAIN) {
258 if (ida_pre_get(&sysfs_ino_ida, GFP_KERNEL))
259 goto retry;
260 rc = -ENOMEM;
261 }
262
263 *pino = ino;
264 return rc;
265}
266
267static void sysfs_free_ino(unsigned int ino)
268{
269 spin_lock(&sysfs_ino_lock);
270 ida_remove(&sysfs_ino_ida, ino);
271 spin_unlock(&sysfs_ino_lock);
272}
273
274void release_sysfs_dirent(struct sysfs_dirent * sd)
275{
276 struct sysfs_dirent *parent_sd;
277
278 repeat:
279 /* Moving/renaming is always done while holding reference.
280 * sd->s_parent won't change beneath us.
281 */
282 parent_sd = sd->s_parent;
283
284 if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
285 sysfs_put(sd->s_symlink.target_sd);
286 if (sysfs_type(sd) & SYSFS_COPY_NAME)
287 kfree(sd->s_name);
288 if (sd->s_iattr && sd->s_iattr->ia_secdata)
289 security_release_secctx(sd->s_iattr->ia_secdata,
290 sd->s_iattr->ia_secdata_len);
291 kfree(sd->s_iattr);
292 sysfs_free_ino(sd->s_ino);
293 kmem_cache_free(sysfs_dir_cachep, sd);
294
295 sd = parent_sd;
296 if (sd && atomic_dec_and_test(&sd->s_count))
297 goto repeat;
298}
299
300static int sysfs_dentry_delete(const struct dentry *dentry)
301{
302 struct sysfs_dirent *sd = dentry->d_fsdata;
303 return !!(sd->s_flags & SYSFS_FLAG_REMOVED);
304}
305
306static int sysfs_dentry_revalidate(struct dentry *dentry, struct nameidata *nd)
307{
308 struct sysfs_dirent *sd;
309 int is_dir;
310
311 if (nd->flags & LOOKUP_RCU)
312 return -ECHILD;
313
314 sd = dentry->d_fsdata;
315 mutex_lock(&sysfs_mutex);
316
317 /* The sysfs dirent has been deleted */
318 if (sd->s_flags & SYSFS_FLAG_REMOVED)
319 goto out_bad;
320
321 /* The sysfs dirent has been moved? */
322 if (dentry->d_parent->d_fsdata != sd->s_parent)
323 goto out_bad;
324
325 /* The sysfs dirent has been renamed */
326 if (strcmp(dentry->d_name.name, sd->s_name) != 0)
327 goto out_bad;
328
329 mutex_unlock(&sysfs_mutex);
330out_valid:
331 return 1;
332out_bad:
333 /* Remove the dentry from the dcache hashes.
334 * If this is a deleted dentry we use d_drop instead of d_delete
335 * so sysfs doesn't need to cope with negative dentries.
336 *
337 * If this is a dentry that has simply been renamed we
338 * use d_drop to remove it from the dcache lookup on its
339 * old parent. If this dentry persists later when a lookup
340 * is performed at its new name the dentry will be readded
341 * to the dcache hashes.
342 */
343 is_dir = (sysfs_type(sd) == SYSFS_DIR);
344 mutex_unlock(&sysfs_mutex);
345 if (is_dir) {
346 /* If we have submounts we must allow the vfs caches
347 * to lie about the state of the filesystem to prevent
348 * leaks and other nasty things.
349 */
350 if (have_submounts(dentry))
351 goto out_valid;
352 shrink_dcache_parent(dentry);
353 }
354 d_drop(dentry);
355 return 0;
356}
357
358static void sysfs_dentry_iput(struct dentry *dentry, struct inode *inode)
359{
360 struct sysfs_dirent * sd = dentry->d_fsdata;
361
362 sysfs_put(sd);
363 iput(inode);
364}
365
366static const struct dentry_operations sysfs_dentry_ops = {
367 .d_revalidate = sysfs_dentry_revalidate,
368 .d_delete = sysfs_dentry_delete,
369 .d_iput = sysfs_dentry_iput,
370};
371
372struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
373{
374 char *dup_name = NULL;
375 struct sysfs_dirent *sd;
376
377 if (type & SYSFS_COPY_NAME) {
378 name = dup_name = kstrdup(name, GFP_KERNEL);
379 if (!name)
380 return NULL;
381 }
382
383 sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
384 if (!sd)
385 goto err_out1;
386
387 if (sysfs_alloc_ino(&sd->s_ino))
388 goto err_out2;
389
390 atomic_set(&sd->s_count, 1);
391 atomic_set(&sd->s_active, 0);
392
393 sd->s_name = name;
394 sd->s_mode = mode;
395 sd->s_flags = type;
396
397 return sd;
398
399 err_out2:
400 kmem_cache_free(sysfs_dir_cachep, sd);
401 err_out1:
402 kfree(dup_name);
403 return NULL;
404}
405
406/**
407 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
408 * @acxt: pointer to sysfs_addrm_cxt to be used
409 * @parent_sd: parent sysfs_dirent
410 *
411 * This function is called when the caller is about to add or
412 * remove sysfs_dirent under @parent_sd. This function acquires
413 * sysfs_mutex. @acxt is used to keep and pass context to
414 * other addrm functions.
415 *
416 * LOCKING:
417 * Kernel thread context (may sleep). sysfs_mutex is locked on
418 * return.
419 */
420void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt,
421 struct sysfs_dirent *parent_sd)
422{
423 memset(acxt, 0, sizeof(*acxt));
424 acxt->parent_sd = parent_sd;
425
426 mutex_lock(&sysfs_mutex);
427}
428
429/**
430 * __sysfs_add_one - add sysfs_dirent to parent without warning
431 * @acxt: addrm context to use
432 * @sd: sysfs_dirent to be added
433 *
434 * Get @acxt->parent_sd and set sd->s_parent to it and increment
435 * nlink of parent inode if @sd is a directory and link into the
436 * children list of the parent.
437 *
438 * This function should be called between calls to
439 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
440 * passed the same @acxt as passed to sysfs_addrm_start().
441 *
442 * LOCKING:
443 * Determined by sysfs_addrm_start().
444 *
445 * RETURNS:
446 * 0 on success, -EEXIST if entry with the given name already
447 * exists.
448 */
449int __sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
450{
451 struct sysfs_inode_attrs *ps_iattr;
452 int ret;
453
454 if (!!sysfs_ns_type(acxt->parent_sd) != !!sd->s_ns) {
455 WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n",
456 sysfs_ns_type(acxt->parent_sd)? "required": "invalid",
457 acxt->parent_sd->s_name, sd->s_name);
458 return -EINVAL;
459 }
460
461 sd->s_hash = sysfs_name_hash(sd->s_ns, sd->s_name);
462 sd->s_parent = sysfs_get(acxt->parent_sd);
463
464 ret = sysfs_link_sibling(sd);
465 if (ret)
466 return ret;
467
468 /* Update timestamps on the parent */
469 ps_iattr = acxt->parent_sd->s_iattr;
470 if (ps_iattr) {
471 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
472 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
473 }
474
475 return 0;
476}
477
478/**
479 * sysfs_pathname - return full path to sysfs dirent
480 * @sd: sysfs_dirent whose path we want
481 * @path: caller allocated buffer
482 *
483 * Gives the name "/" to the sysfs_root entry; any path returned
484 * is relative to wherever sysfs is mounted.
485 *
486 * XXX: does no error checking on @path size
487 */
488static char *sysfs_pathname(struct sysfs_dirent *sd, char *path)
489{
490 if (sd->s_parent) {
491 sysfs_pathname(sd->s_parent, path);
492 strcat(path, "/");
493 }
494 strcat(path, sd->s_name);
495 return path;
496}
497
498/**
499 * sysfs_add_one - add sysfs_dirent to parent
500 * @acxt: addrm context to use
501 * @sd: sysfs_dirent to be added
502 *
503 * Get @acxt->parent_sd and set sd->s_parent to it and increment
504 * nlink of parent inode if @sd is a directory and link into the
505 * children list of the parent.
506 *
507 * This function should be called between calls to
508 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
509 * passed the same @acxt as passed to sysfs_addrm_start().
510 *
511 * LOCKING:
512 * Determined by sysfs_addrm_start().
513 *
514 * RETURNS:
515 * 0 on success, -EEXIST if entry with the given name already
516 * exists.
517 */
518int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
519{
520 int ret;
521
522 ret = __sysfs_add_one(acxt, sd);
523 if (ret == -EEXIST) {
524 char *path = kzalloc(PATH_MAX, GFP_KERNEL);
525 WARN(1, KERN_WARNING
526 "sysfs: cannot create duplicate filename '%s'\n",
527 (path == NULL) ? sd->s_name :
528 strcat(strcat(sysfs_pathname(acxt->parent_sd, path), "/"),
529 sd->s_name));
530 kfree(path);
531 }
532
533 return ret;
534}
535
536/**
537 * sysfs_remove_one - remove sysfs_dirent from parent
538 * @acxt: addrm context to use
539 * @sd: sysfs_dirent to be removed
540 *
541 * Mark @sd removed and drop nlink of parent inode if @sd is a
542 * directory. @sd is unlinked from the children list.
543 *
544 * This function should be called between calls to
545 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
546 * passed the same @acxt as passed to sysfs_addrm_start().
547 *
548 * LOCKING:
549 * Determined by sysfs_addrm_start().
550 */
551void sysfs_remove_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd)
552{
553 struct sysfs_inode_attrs *ps_iattr;
554
555 BUG_ON(sd->s_flags & SYSFS_FLAG_REMOVED);
556
557 sysfs_unlink_sibling(sd);
558
559 /* Update timestamps on the parent */
560 ps_iattr = acxt->parent_sd->s_iattr;
561 if (ps_iattr) {
562 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
563 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
564 }
565
566 sd->s_flags |= SYSFS_FLAG_REMOVED;
567 sd->u.removed_list = acxt->removed;
568 acxt->removed = sd;
569}
570
571/**
572 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
573 * @acxt: addrm context to finish up
574 *
575 * Finish up sysfs_dirent add/remove. Resources acquired by
576 * sysfs_addrm_start() are released and removed sysfs_dirents are
577 * cleaned up.
578 *
579 * LOCKING:
580 * sysfs_mutex is released.
581 */
582void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
583{
584 /* release resources acquired by sysfs_addrm_start() */
585 mutex_unlock(&sysfs_mutex);
586
587 /* kill removed sysfs_dirents */
588 while (acxt->removed) {
589 struct sysfs_dirent *sd = acxt->removed;
590
591 acxt->removed = sd->u.removed_list;
592
593 sysfs_deactivate(sd);
594 unmap_bin_file(sd);
595 sysfs_put(sd);
596 }
597}
598
599/**
600 * sysfs_find_dirent - find sysfs_dirent with the given name
601 * @parent_sd: sysfs_dirent to search under
602 * @name: name to look for
603 *
604 * Look for sysfs_dirent with name @name under @parent_sd.
605 *
606 * LOCKING:
607 * mutex_lock(sysfs_mutex)
608 *
609 * RETURNS:
610 * Pointer to sysfs_dirent if found, NULL if not.
611 */
612struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
613 const void *ns,
614 const unsigned char *name)
615{
616 struct rb_node *node = parent_sd->s_dir.children.rb_node;
617 unsigned int hash;
618
619 if (!!sysfs_ns_type(parent_sd) != !!ns) {
620 WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n",
621 sysfs_ns_type(parent_sd)? "required": "invalid",
622 parent_sd->s_name, name);
623 return NULL;
624 }
625
626 hash = sysfs_name_hash(ns, name);
627 while (node) {
628 struct sysfs_dirent *sd;
629 int result;
630
631 sd = to_sysfs_dirent(node);
632 result = sysfs_name_compare(hash, ns, name, sd);
633 if (result < 0)
634 node = node->rb_left;
635 else if (result > 0)
636 node = node->rb_right;
637 else
638 return sd;
639 }
640 return NULL;
641}
642
643/**
644 * sysfs_get_dirent - find and get sysfs_dirent with the given name
645 * @parent_sd: sysfs_dirent to search under
646 * @name: name to look for
647 *
648 * Look for sysfs_dirent with name @name under @parent_sd and get
649 * it if found.
650 *
651 * LOCKING:
652 * Kernel thread context (may sleep). Grabs sysfs_mutex.
653 *
654 * RETURNS:
655 * Pointer to sysfs_dirent if found, NULL if not.
656 */
657struct sysfs_dirent *sysfs_get_dirent(struct sysfs_dirent *parent_sd,
658 const void *ns,
659 const unsigned char *name)
660{
661 struct sysfs_dirent *sd;
662
663 mutex_lock(&sysfs_mutex);
664 sd = sysfs_find_dirent(parent_sd, ns, name);
665 sysfs_get(sd);
666 mutex_unlock(&sysfs_mutex);
667
668 return sd;
669}
670EXPORT_SYMBOL_GPL(sysfs_get_dirent);
671
672static int create_dir(struct kobject *kobj, struct sysfs_dirent *parent_sd,
673 enum kobj_ns_type type, const void *ns, const char *name,
674 struct sysfs_dirent **p_sd)
675{
676 umode_t mode = S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO;
677 struct sysfs_addrm_cxt acxt;
678 struct sysfs_dirent *sd;
679 int rc;
680
681 /* allocate */
682 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
683 if (!sd)
684 return -ENOMEM;
685
686 sd->s_flags |= (type << SYSFS_NS_TYPE_SHIFT);
687 sd->s_ns = ns;
688 sd->s_dir.kobj = kobj;
689
690 /* link in */
691 sysfs_addrm_start(&acxt, parent_sd);
692 rc = sysfs_add_one(&acxt, sd);
693 sysfs_addrm_finish(&acxt);
694
695 if (rc == 0)
696 *p_sd = sd;
697 else
698 sysfs_put(sd);
699
700 return rc;
701}
702
703int sysfs_create_subdir(struct kobject *kobj, const char *name,
704 struct sysfs_dirent **p_sd)
705{
706 return create_dir(kobj, kobj->sd,
707 KOBJ_NS_TYPE_NONE, NULL, name, p_sd);
708}
709
710/**
711 * sysfs_read_ns_type: return associated ns_type
712 * @kobj: the kobject being queried
713 *
714 * Each kobject can be tagged with exactly one namespace type
715 * (i.e. network or user). Return the ns_type associated with
716 * this object if any
717 */
718static enum kobj_ns_type sysfs_read_ns_type(struct kobject *kobj)
719{
720 const struct kobj_ns_type_operations *ops;
721 enum kobj_ns_type type;
722
723 ops = kobj_child_ns_ops(kobj);
724 if (!ops)
725 return KOBJ_NS_TYPE_NONE;
726
727 type = ops->type;
728 BUG_ON(type <= KOBJ_NS_TYPE_NONE);
729 BUG_ON(type >= KOBJ_NS_TYPES);
730 BUG_ON(!kobj_ns_type_registered(type));
731
732 return type;
733}
734
735/**
736 * sysfs_create_dir - create a directory for an object.
737 * @kobj: object we're creating directory for.
738 */
739int sysfs_create_dir(struct kobject * kobj)
740{
741 enum kobj_ns_type type;
742 struct sysfs_dirent *parent_sd, *sd;
743 const void *ns = NULL;
744 int error = 0;
745
746 BUG_ON(!kobj);
747
748 if (kobj->parent)
749 parent_sd = kobj->parent->sd;
750 else
751 parent_sd = &sysfs_root;
752
753 if (!parent_sd)
754 return -ENOENT;
755
756 if (sysfs_ns_type(parent_sd))
757 ns = kobj->ktype->namespace(kobj);
758 type = sysfs_read_ns_type(kobj);
759
760 error = create_dir(kobj, parent_sd, type, ns, kobject_name(kobj), &sd);
761 if (!error)
762 kobj->sd = sd;
763 return error;
764}
765
766static struct dentry * sysfs_lookup(struct inode *dir, struct dentry *dentry,
767 struct nameidata *nd)
768{
769 struct dentry *ret = NULL;
770 struct dentry *parent = dentry->d_parent;
771 struct sysfs_dirent *parent_sd = parent->d_fsdata;
772 struct sysfs_dirent *sd;
773 struct inode *inode;
774 enum kobj_ns_type type;
775 const void *ns;
776
777 mutex_lock(&sysfs_mutex);
778
779 type = sysfs_ns_type(parent_sd);
780 ns = sysfs_info(dir->i_sb)->ns[type];
781
782 sd = sysfs_find_dirent(parent_sd, ns, dentry->d_name.name);
783
784 /* no such entry */
785 if (!sd) {
786 ret = ERR_PTR(-ENOENT);
787 goto out_unlock;
788 }
789
790 /* attach dentry and inode */
791 inode = sysfs_get_inode(dir->i_sb, sd);
792 if (!inode) {
793 ret = ERR_PTR(-ENOMEM);
794 goto out_unlock;
795 }
796
797 /* instantiate and hash dentry */
798 ret = d_find_alias(inode);
799 if (!ret) {
800 d_set_d_op(dentry, &sysfs_dentry_ops);
801 dentry->d_fsdata = sysfs_get(sd);
802 d_add(dentry, inode);
803 } else {
804 d_move(ret, dentry);
805 iput(inode);
806 }
807
808 out_unlock:
809 mutex_unlock(&sysfs_mutex);
810 return ret;
811}
812
813const struct inode_operations sysfs_dir_inode_operations = {
814 .lookup = sysfs_lookup,
815 .permission = sysfs_permission,
816 .setattr = sysfs_setattr,
817 .getattr = sysfs_getattr,
818 .setxattr = sysfs_setxattr,
819};
820
821static void remove_dir(struct sysfs_dirent *sd)
822{
823 struct sysfs_addrm_cxt acxt;
824
825 sysfs_addrm_start(&acxt, sd->s_parent);
826 sysfs_remove_one(&acxt, sd);
827 sysfs_addrm_finish(&acxt);
828}
829
830void sysfs_remove_subdir(struct sysfs_dirent *sd)
831{
832 remove_dir(sd);
833}
834
835
836static void __sysfs_remove_dir(struct sysfs_dirent *dir_sd)
837{
838 struct sysfs_addrm_cxt acxt;
839 struct rb_node *pos;
840
841 if (!dir_sd)
842 return;
843
844 pr_debug("sysfs %s: removing dir\n", dir_sd->s_name);
845 sysfs_addrm_start(&acxt, dir_sd);
846 pos = rb_first(&dir_sd->s_dir.children);
847 while (pos) {
848 struct sysfs_dirent *sd = to_sysfs_dirent(pos);
849 pos = rb_next(pos);
850 if (sysfs_type(sd) != SYSFS_DIR)
851 sysfs_remove_one(&acxt, sd);
852 }
853 sysfs_addrm_finish(&acxt);
854
855 remove_dir(dir_sd);
856}
857
858/**
859 * sysfs_remove_dir - remove an object's directory.
860 * @kobj: object.
861 *
862 * The only thing special about this is that we remove any files in
863 * the directory before we remove the directory, and we've inlined
864 * what used to be sysfs_rmdir() below, instead of calling separately.
865 */
866
867void sysfs_remove_dir(struct kobject * kobj)
868{
869 struct sysfs_dirent *sd = kobj->sd;
870
871 spin_lock(&sysfs_assoc_lock);
872 kobj->sd = NULL;
873 spin_unlock(&sysfs_assoc_lock);
874
875 __sysfs_remove_dir(sd);
876}
877
878int sysfs_rename(struct sysfs_dirent *sd,
879 struct sysfs_dirent *new_parent_sd, const void *new_ns,
880 const char *new_name)
881{
882 int error;
883
884 mutex_lock(&sysfs_mutex);
885
886 error = 0;
887 if ((sd->s_parent == new_parent_sd) && (sd->s_ns == new_ns) &&
888 (strcmp(sd->s_name, new_name) == 0))
889 goto out; /* nothing to rename */
890
891 error = -EEXIST;
892 if (sysfs_find_dirent(new_parent_sd, new_ns, new_name))
893 goto out;
894
895 /* rename sysfs_dirent */
896 if (strcmp(sd->s_name, new_name) != 0) {
897 error = -ENOMEM;
898 new_name = kstrdup(new_name, GFP_KERNEL);
899 if (!new_name)
900 goto out;
901
902 kfree(sd->s_name);
903 sd->s_name = new_name;
904 }
905
906 /* Move to the appropriate place in the appropriate directories rbtree. */
907 sysfs_unlink_sibling(sd);
908 sysfs_get(new_parent_sd);
909 sysfs_put(sd->s_parent);
910 sd->s_ns = new_ns;
911 sd->s_hash = sysfs_name_hash(sd->s_ns, sd->s_name);
912 sd->s_parent = new_parent_sd;
913 sysfs_link_sibling(sd);
914
915 error = 0;
916 out:
917 mutex_unlock(&sysfs_mutex);
918 return error;
919}
920
921int sysfs_rename_dir(struct kobject *kobj, const char *new_name)
922{
923 struct sysfs_dirent *parent_sd = kobj->sd->s_parent;
924 const void *new_ns = NULL;
925
926 if (sysfs_ns_type(parent_sd))
927 new_ns = kobj->ktype->namespace(kobj);
928
929 return sysfs_rename(kobj->sd, parent_sd, new_ns, new_name);
930}
931
932int sysfs_move_dir(struct kobject *kobj, struct kobject *new_parent_kobj)
933{
934 struct sysfs_dirent *sd = kobj->sd;
935 struct sysfs_dirent *new_parent_sd;
936 const void *new_ns = NULL;
937
938 BUG_ON(!sd->s_parent);
939 if (sysfs_ns_type(sd->s_parent))
940 new_ns = kobj->ktype->namespace(kobj);
941 new_parent_sd = new_parent_kobj && new_parent_kobj->sd ?
942 new_parent_kobj->sd : &sysfs_root;
943
944 return sysfs_rename(sd, new_parent_sd, new_ns, sd->s_name);
945}
946
947/* Relationship between s_mode and the DT_xxx types */
948static inline unsigned char dt_type(struct sysfs_dirent *sd)
949{
950 return (sd->s_mode >> 12) & 15;
951}
952
953static int sysfs_dir_release(struct inode *inode, struct file *filp)
954{
955 sysfs_put(filp->private_data);
956 return 0;
957}
958
959static struct sysfs_dirent *sysfs_dir_pos(const void *ns,
960 struct sysfs_dirent *parent_sd, loff_t hash, struct sysfs_dirent *pos)
961{
962 if (pos) {
963 int valid = !(pos->s_flags & SYSFS_FLAG_REMOVED) &&
964 pos->s_parent == parent_sd &&
965 hash == pos->s_hash;
966 sysfs_put(pos);
967 if (!valid)
968 pos = NULL;
969 }
970 if (!pos && (hash > 1) && (hash < INT_MAX)) {
971 struct rb_node *node = parent_sd->s_dir.children.rb_node;
972 while (node) {
973 pos = to_sysfs_dirent(node);
974
975 if (hash < pos->s_hash)
976 node = node->rb_left;
977 else if (hash > pos->s_hash)
978 node = node->rb_right;
979 else
980 break;
981 }
982 }
983 /* Skip over entries in the wrong namespace */
984 while (pos && pos->s_ns != ns) {
985 struct rb_node *node = rb_next(&pos->s_rb);
986 if (!node)
987 pos = NULL;
988 else
989 pos = to_sysfs_dirent(node);
990 }
991 return pos;
992}
993
994static struct sysfs_dirent *sysfs_dir_next_pos(const void *ns,
995 struct sysfs_dirent *parent_sd, ino_t ino, struct sysfs_dirent *pos)
996{
997 pos = sysfs_dir_pos(ns, parent_sd, ino, pos);
998 if (pos) do {
999 struct rb_node *node = rb_next(&pos->s_rb);
1000 if (!node)
1001 pos = NULL;
1002 else
1003 pos = to_sysfs_dirent(node);
1004 } while (pos && pos->s_ns != ns);
1005 return pos;
1006}
1007
1008static int sysfs_readdir(struct file * filp, void * dirent, filldir_t filldir)
1009{
1010 struct dentry *dentry = filp->f_path.dentry;
1011 struct sysfs_dirent * parent_sd = dentry->d_fsdata;
1012 struct sysfs_dirent *pos = filp->private_data;
1013 enum kobj_ns_type type;
1014 const void *ns;
1015 ino_t ino;
1016
1017 type = sysfs_ns_type(parent_sd);
1018 ns = sysfs_info(dentry->d_sb)->ns[type];
1019
1020 if (filp->f_pos == 0) {
1021 ino = parent_sd->s_ino;
1022 if (filldir(dirent, ".", 1, filp->f_pos, ino, DT_DIR) == 0)
1023 filp->f_pos++;
1024 }
1025 if (filp->f_pos == 1) {
1026 if (parent_sd->s_parent)
1027 ino = parent_sd->s_parent->s_ino;
1028 else
1029 ino = parent_sd->s_ino;
1030 if (filldir(dirent, "..", 2, filp->f_pos, ino, DT_DIR) == 0)
1031 filp->f_pos++;
1032 }
1033 mutex_lock(&sysfs_mutex);
1034 for (pos = sysfs_dir_pos(ns, parent_sd, filp->f_pos, pos);
1035 pos;
1036 pos = sysfs_dir_next_pos(ns, parent_sd, filp->f_pos, pos)) {
1037 const char * name;
1038 unsigned int type;
1039 int len, ret;
1040
1041 name = pos->s_name;
1042 len = strlen(name);
1043 ino = pos->s_ino;
1044 type = dt_type(pos);
1045 filp->f_pos = pos->s_hash;
1046 filp->private_data = sysfs_get(pos);
1047
1048 mutex_unlock(&sysfs_mutex);
1049 ret = filldir(dirent, name, len, filp->f_pos, ino, type);
1050 mutex_lock(&sysfs_mutex);
1051 if (ret < 0)
1052 break;
1053 }
1054 mutex_unlock(&sysfs_mutex);
1055 if ((filp->f_pos > 1) && !pos) { /* EOF */
1056 filp->f_pos = INT_MAX;
1057 filp->private_data = NULL;
1058 }
1059 return 0;
1060}
1061
1062
1063const struct file_operations sysfs_dir_operations = {
1064 .read = generic_read_dir,
1065 .readdir = sysfs_readdir,
1066 .release = sysfs_dir_release,
1067 .llseek = generic_file_llseek,
1068};