2 * fs/sysfs/dir.c - sysfs core and dir operation implementation
4 * Copyright (c) 2001-3 Patrick Mochel
5 * Copyright (c) 2007 SUSE Linux Products GmbH
6 * Copyright (c) 2007 Tejun Heo <teheo@suse.de>
8 * This file is released under the GPLv2.
10 * Please see Documentation/filesystems/sysfs.txt for more information.
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>
28 DEFINE_MUTEX(sysfs_mutex
);
29 DEFINE_SPINLOCK(sysfs_symlink_target_lock
);
31 #define to_sysfs_dirent(X) rb_entry((X), struct sysfs_dirent, s_rb)
33 static DEFINE_SPINLOCK(sysfs_ino_lock
);
34 static DEFINE_IDA(sysfs_ino_ida
);
38 * @name: Null terminated string to hash
39 * @ns: Namespace tag to hash
41 * Returns 31 bit hash of ns + name (so it fits in an off_t )
43 static unsigned int sysfs_name_hash(const char *name
, const void *ns
)
45 unsigned long hash
= init_name_hash();
46 unsigned int len
= strlen(name
);
48 hash
= partial_name_hash(*name
++, hash
);
49 hash
= (end_name_hash(hash
) ^ hash_ptr((void *)ns
, 31));
51 /* Reserve hash numbers 0, 1 and INT_MAX for magic directory entries */
59 static int sysfs_name_compare(unsigned int hash
, const char *name
,
60 const void *ns
, const struct sysfs_dirent
*sd
)
62 if (hash
!= sd
->s_hash
)
63 return hash
- sd
->s_hash
;
66 return strcmp(name
, sd
->s_name
);
69 static int sysfs_sd_compare(const struct sysfs_dirent
*left
,
70 const struct sysfs_dirent
*right
)
72 return sysfs_name_compare(left
->s_hash
, left
->s_name
, left
->s_ns
,
77 * sysfs_link_sibling - link sysfs_dirent into sibling rbtree
78 * @sd: sysfs_dirent of interest
80 * Link @sd into its sibling rbtree which starts from
81 * sd->s_parent->s_dir.children.
84 * mutex_lock(sysfs_mutex)
87 * 0 on susccess -EEXIST on failure.
89 static int sysfs_link_sibling(struct sysfs_dirent
*sd
)
91 struct rb_node
**node
= &sd
->s_parent
->s_dir
.children
.rb_node
;
92 struct rb_node
*parent
= NULL
;
94 if (sysfs_type(sd
) == SYSFS_DIR
)
95 sd
->s_parent
->s_dir
.subdirs
++;
98 struct sysfs_dirent
*pos
;
101 pos
= to_sysfs_dirent(*node
);
103 result
= sysfs_sd_compare(sd
, pos
);
105 node
= &pos
->s_rb
.rb_left
;
107 node
= &pos
->s_rb
.rb_right
;
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
);
118 * sysfs_unlink_sibling - unlink sysfs_dirent from sibling rbtree
119 * @sd: sysfs_dirent of interest
121 * Unlink @sd from its sibling rbtree which starts from
122 * sd->s_parent->s_dir.children.
125 * mutex_lock(sysfs_mutex)
127 static void sysfs_unlink_sibling(struct sysfs_dirent
*sd
)
129 if (sysfs_type(sd
) == SYSFS_DIR
)
130 sd
->s_parent
->s_dir
.subdirs
--;
132 rb_erase(&sd
->s_rb
, &sd
->s_parent
->s_dir
.children
);
136 * sysfs_get_active - get an active reference to sysfs_dirent
137 * @sd: sysfs_dirent to get an active reference to
139 * Get an active reference of @sd. This function is noop if @sd
143 * Pointer to @sd on success, NULL on failure.
145 struct sysfs_dirent
*sysfs_get_active(struct sysfs_dirent
*sd
)
150 if (!atomic_inc_unless_negative(&sd
->s_active
))
153 if (likely(!sysfs_ignore_lockdep(sd
)))
154 rwsem_acquire_read(&sd
->dep_map
, 0, 1, _RET_IP_
);
159 * sysfs_put_active - put an active reference to sysfs_dirent
160 * @sd: sysfs_dirent to put an active reference to
162 * Put an active reference to @sd. This function is noop if @sd
165 void sysfs_put_active(struct sysfs_dirent
*sd
)
172 if (likely(!sysfs_ignore_lockdep(sd
)))
173 rwsem_release(&sd
->dep_map
, 1, _RET_IP_
);
174 v
= atomic_dec_return(&sd
->s_active
);
175 if (likely(v
!= SD_DEACTIVATED_BIAS
))
178 /* atomic_dec_return() is a mb(), we'll always see the updated
181 complete(sd
->u
.completion
);
185 * sysfs_deactivate - deactivate sysfs_dirent
186 * @sd: sysfs_dirent to deactivate
188 * Deny new active references and drain existing ones.
190 static void sysfs_deactivate(struct sysfs_dirent
*sd
)
192 DECLARE_COMPLETION_ONSTACK(wait
);
195 BUG_ON(!(sd
->s_flags
& SYSFS_FLAG_REMOVED
));
197 if (!(sysfs_type(sd
) & SYSFS_ACTIVE_REF
))
200 sd
->u
.completion
= (void *)&wait
;
202 rwsem_acquire(&sd
->dep_map
, 0, 0, _RET_IP_
);
203 /* atomic_add_return() is a mb(), put_active() will always see
204 * the updated sd->u.completion.
206 v
= atomic_add_return(SD_DEACTIVATED_BIAS
, &sd
->s_active
);
208 if (v
!= SD_DEACTIVATED_BIAS
) {
209 lock_contended(&sd
->dep_map
, _RET_IP_
);
210 wait_for_completion(&wait
);
213 lock_acquired(&sd
->dep_map
, _RET_IP_
);
214 rwsem_release(&sd
->dep_map
, 1, _RET_IP_
);
217 static int sysfs_alloc_ino(unsigned int *pino
)
222 spin_lock(&sysfs_ino_lock
);
223 rc
= ida_get_new_above(&sysfs_ino_ida
, 2, &ino
);
224 spin_unlock(&sysfs_ino_lock
);
227 if (ida_pre_get(&sysfs_ino_ida
, GFP_KERNEL
))
236 static void sysfs_free_ino(unsigned int ino
)
238 spin_lock(&sysfs_ino_lock
);
239 ida_remove(&sysfs_ino_ida
, ino
);
240 spin_unlock(&sysfs_ino_lock
);
243 void release_sysfs_dirent(struct sysfs_dirent
*sd
)
245 struct sysfs_dirent
*parent_sd
;
248 /* Moving/renaming is always done while holding reference.
249 * sd->s_parent won't change beneath us.
251 parent_sd
= sd
->s_parent
;
253 WARN(!(sd
->s_flags
& SYSFS_FLAG_REMOVED
),
254 "sysfs: free using entry: %s/%s\n",
255 parent_sd
? parent_sd
->s_name
: "", sd
->s_name
);
257 if (sysfs_type(sd
) == SYSFS_KOBJ_LINK
)
258 sysfs_put(sd
->s_symlink
.target_sd
);
259 if (sysfs_type(sd
) & SYSFS_COPY_NAME
)
261 if (sd
->s_iattr
&& sd
->s_iattr
->ia_secdata
)
262 security_release_secctx(sd
->s_iattr
->ia_secdata
,
263 sd
->s_iattr
->ia_secdata_len
);
265 sysfs_free_ino(sd
->s_ino
);
266 kmem_cache_free(sysfs_dir_cachep
, sd
);
269 if (sd
&& atomic_dec_and_test(&sd
->s_count
))
273 static int sysfs_dentry_delete(const struct dentry
*dentry
)
275 struct sysfs_dirent
*sd
= dentry
->d_fsdata
;
276 return !(sd
&& !(sd
->s_flags
& SYSFS_FLAG_REMOVED
));
279 static int sysfs_dentry_revalidate(struct dentry
*dentry
, unsigned int flags
)
281 struct sysfs_dirent
*sd
;
284 if (flags
& LOOKUP_RCU
)
287 sd
= dentry
->d_fsdata
;
288 mutex_lock(&sysfs_mutex
);
290 /* The sysfs dirent has been deleted */
291 if (sd
->s_flags
& SYSFS_FLAG_REMOVED
)
294 /* The sysfs dirent has been moved? */
295 if (dentry
->d_parent
->d_fsdata
!= sd
->s_parent
)
298 /* The sysfs dirent has been renamed */
299 if (strcmp(dentry
->d_name
.name
, sd
->s_name
) != 0)
302 /* The sysfs dirent has been moved to a different namespace */
303 type
= KOBJ_NS_TYPE_NONE
;
305 type
= sysfs_ns_type(sd
->s_parent
);
306 if (type
!= KOBJ_NS_TYPE_NONE
&&
307 sysfs_info(dentry
->d_sb
)->ns
[type
] != sd
->s_ns
)
311 mutex_unlock(&sysfs_mutex
);
315 /* Remove the dentry from the dcache hashes.
316 * If this is a deleted dentry we use d_drop instead of d_delete
317 * so sysfs doesn't need to cope with negative dentries.
319 * If this is a dentry that has simply been renamed we
320 * use d_drop to remove it from the dcache lookup on its
321 * old parent. If this dentry persists later when a lookup
322 * is performed at its new name the dentry will be readded
323 * to the dcache hashes.
325 mutex_unlock(&sysfs_mutex
);
327 /* If we have submounts we must allow the vfs caches
328 * to lie about the state of the filesystem to prevent
329 * leaks and other nasty things.
331 if (check_submounts_and_drop(dentry
) != 0)
337 static void sysfs_dentry_release(struct dentry
*dentry
)
339 sysfs_put(dentry
->d_fsdata
);
342 const struct dentry_operations sysfs_dentry_ops
= {
343 .d_revalidate
= sysfs_dentry_revalidate
,
344 .d_delete
= sysfs_dentry_delete
,
345 .d_release
= sysfs_dentry_release
,
348 struct sysfs_dirent
*sysfs_new_dirent(const char *name
, umode_t mode
, int type
)
350 char *dup_name
= NULL
;
351 struct sysfs_dirent
*sd
;
353 if (type
& SYSFS_COPY_NAME
) {
354 name
= dup_name
= kstrdup(name
, GFP_KERNEL
);
359 sd
= kmem_cache_zalloc(sysfs_dir_cachep
, GFP_KERNEL
);
363 if (sysfs_alloc_ino(&sd
->s_ino
))
366 atomic_set(&sd
->s_count
, 1);
367 atomic_set(&sd
->s_active
, 0);
371 sd
->s_flags
= type
| SYSFS_FLAG_REMOVED
;
376 kmem_cache_free(sysfs_dir_cachep
, sd
);
383 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
384 * @acxt: pointer to sysfs_addrm_cxt to be used
386 * This function is called when the caller is about to add or remove
387 * sysfs_dirent. This function acquires sysfs_mutex. @acxt is used
388 * to keep and pass context to other addrm functions.
391 * Kernel thread context (may sleep). sysfs_mutex is locked on
394 void sysfs_addrm_start(struct sysfs_addrm_cxt
*acxt
)
395 __acquires(sysfs_mutex
)
397 memset(acxt
, 0, sizeof(*acxt
));
399 mutex_lock(&sysfs_mutex
);
403 * __sysfs_add_one - add sysfs_dirent to parent without warning
404 * @acxt: addrm context to use
405 * @sd: sysfs_dirent to be added
406 * @parent_sd: the parent sysfs_dirent to add @sd to
408 * Get @parent_sd and set @sd->s_parent to it and increment nlink of
409 * the parent inode if @sd is a directory and link into the children
410 * list of the parent.
412 * This function should be called between calls to
413 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
414 * passed the same @acxt as passed to sysfs_addrm_start().
417 * Determined by sysfs_addrm_start().
420 * 0 on success, -EEXIST if entry with the given name already
423 int __sysfs_add_one(struct sysfs_addrm_cxt
*acxt
, struct sysfs_dirent
*sd
,
424 struct sysfs_dirent
*parent_sd
)
426 struct sysfs_inode_attrs
*ps_iattr
;
429 if (!!sysfs_ns_type(parent_sd
) != !!sd
->s_ns
) {
430 WARN(1, KERN_WARNING
"sysfs: ns %s in '%s' for '%s'\n",
431 sysfs_ns_type(parent_sd
) ? "required" : "invalid",
432 parent_sd
->s_name
, sd
->s_name
);
436 sd
->s_hash
= sysfs_name_hash(sd
->s_name
, sd
->s_ns
);
437 sd
->s_parent
= sysfs_get(parent_sd
);
439 ret
= sysfs_link_sibling(sd
);
443 /* Update timestamps on the parent */
444 ps_iattr
= parent_sd
->s_iattr
;
446 struct iattr
*ps_iattrs
= &ps_iattr
->ia_iattr
;
447 ps_iattrs
->ia_ctime
= ps_iattrs
->ia_mtime
= CURRENT_TIME
;
450 /* Mark the entry added into directory tree */
451 sd
->s_flags
&= ~SYSFS_FLAG_REMOVED
;
457 * sysfs_pathname - return full path to sysfs dirent
458 * @sd: sysfs_dirent whose path we want
459 * @path: caller allocated buffer of size PATH_MAX
461 * Gives the name "/" to the sysfs_root entry; any path returned
462 * is relative to wherever sysfs is mounted.
464 static char *sysfs_pathname(struct sysfs_dirent
*sd
, char *path
)
467 sysfs_pathname(sd
->s_parent
, path
);
468 strlcat(path
, "/", PATH_MAX
);
470 strlcat(path
, sd
->s_name
, PATH_MAX
);
474 void sysfs_warn_dup(struct sysfs_dirent
*parent
, const char *name
)
478 path
= kzalloc(PATH_MAX
, GFP_KERNEL
);
480 sysfs_pathname(parent
, path
);
481 strlcat(path
, "/", PATH_MAX
);
482 strlcat(path
, name
, PATH_MAX
);
485 WARN(1, KERN_WARNING
"sysfs: cannot create duplicate filename '%s'\n",
492 * sysfs_add_one - add sysfs_dirent to parent
493 * @acxt: addrm context to use
494 * @sd: sysfs_dirent to be added
495 * @parent_sd: the parent sysfs_dirent to add @sd to
497 * Get @parent_sd and set @sd->s_parent to it and increment nlink of
498 * the parent inode if @sd is a directory and link into the children
499 * list of the parent.
501 * This function should be called between calls to
502 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
503 * passed the same @acxt as passed to sysfs_addrm_start().
506 * Determined by sysfs_addrm_start().
509 * 0 on success, -EEXIST if entry with the given name already
512 int sysfs_add_one(struct sysfs_addrm_cxt
*acxt
, struct sysfs_dirent
*sd
,
513 struct sysfs_dirent
*parent_sd
)
517 ret
= __sysfs_add_one(acxt
, sd
, parent_sd
);
520 sysfs_warn_dup(parent_sd
, sd
->s_name
);
525 * sysfs_remove_one - remove sysfs_dirent from parent
526 * @acxt: addrm context to use
527 * @sd: sysfs_dirent to be removed
529 * Mark @sd removed and drop nlink of parent inode if @sd is a
530 * directory. @sd is unlinked from the children list.
532 * This function should be called between calls to
533 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
534 * passed the same @acxt as passed to sysfs_addrm_start().
537 * Determined by sysfs_addrm_start().
539 static void sysfs_remove_one(struct sysfs_addrm_cxt
*acxt
,
540 struct sysfs_dirent
*sd
)
542 struct sysfs_inode_attrs
*ps_iattr
;
545 * Removal can be called multiple times on the same node. Only the
546 * first invocation is effective and puts the base ref.
548 if (sd
->s_flags
& SYSFS_FLAG_REMOVED
)
551 sysfs_unlink_sibling(sd
);
553 /* Update timestamps on the parent */
554 ps_iattr
= sd
->s_parent
->s_iattr
;
556 struct iattr
*ps_iattrs
= &ps_iattr
->ia_iattr
;
557 ps_iattrs
->ia_ctime
= ps_iattrs
->ia_mtime
= CURRENT_TIME
;
560 sd
->s_flags
|= SYSFS_FLAG_REMOVED
;
561 sd
->u
.removed_list
= acxt
->removed
;
566 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
567 * @acxt: addrm context to finish up
569 * Finish up sysfs_dirent add/remove. Resources acquired by
570 * sysfs_addrm_start() are released and removed sysfs_dirents are
574 * sysfs_mutex is released.
576 void sysfs_addrm_finish(struct sysfs_addrm_cxt
*acxt
)
577 __releases(sysfs_mutex
)
579 /* release resources acquired by sysfs_addrm_start() */
580 mutex_unlock(&sysfs_mutex
);
582 /* kill removed sysfs_dirents */
583 while (acxt
->removed
) {
584 struct sysfs_dirent
*sd
= acxt
->removed
;
586 acxt
->removed
= sd
->u
.removed_list
;
588 sysfs_deactivate(sd
);
589 sysfs_unmap_bin_file(sd
);
595 * sysfs_find_dirent - find sysfs_dirent with the given name
596 * @parent_sd: sysfs_dirent to search under
597 * @name: name to look for
598 * @ns: the namespace tag to use
600 * Look for sysfs_dirent with name @name under @parent_sd.
603 * mutex_lock(sysfs_mutex)
606 * Pointer to sysfs_dirent if found, NULL if not.
608 struct sysfs_dirent
*sysfs_find_dirent(struct sysfs_dirent
*parent_sd
,
609 const unsigned char *name
,
612 struct rb_node
*node
= parent_sd
->s_dir
.children
.rb_node
;
615 if (!!sysfs_ns_type(parent_sd
) != !!ns
) {
616 WARN(1, KERN_WARNING
"sysfs: ns %s in '%s' for '%s'\n",
617 sysfs_ns_type(parent_sd
) ? "required" : "invalid",
618 parent_sd
->s_name
, name
);
622 hash
= sysfs_name_hash(name
, ns
);
624 struct sysfs_dirent
*sd
;
627 sd
= to_sysfs_dirent(node
);
628 result
= sysfs_name_compare(hash
, name
, ns
, sd
);
630 node
= node
->rb_left
;
632 node
= node
->rb_right
;
640 * sysfs_get_dirent_ns - find and get sysfs_dirent with the given name
641 * @parent_sd: sysfs_dirent to search under
642 * @name: name to look for
643 * @ns: the namespace tag to use
645 * Look for sysfs_dirent with name @name under @parent_sd and get
649 * Kernel thread context (may sleep). Grabs sysfs_mutex.
652 * Pointer to sysfs_dirent if found, NULL if not.
654 struct sysfs_dirent
*sysfs_get_dirent_ns(struct sysfs_dirent
*parent_sd
,
655 const unsigned char *name
,
658 struct sysfs_dirent
*sd
;
660 mutex_lock(&sysfs_mutex
);
661 sd
= sysfs_find_dirent(parent_sd
, name
, ns
);
663 mutex_unlock(&sysfs_mutex
);
667 EXPORT_SYMBOL_GPL(sysfs_get_dirent_ns
);
669 static int create_dir(struct kobject
*kobj
, struct sysfs_dirent
*parent_sd
,
670 enum kobj_ns_type type
,
671 const char *name
, const void *ns
,
672 struct sysfs_dirent
**p_sd
)
674 umode_t mode
= S_IFDIR
| S_IRWXU
| S_IRUGO
| S_IXUGO
;
675 struct sysfs_addrm_cxt acxt
;
676 struct sysfs_dirent
*sd
;
680 sd
= sysfs_new_dirent(name
, mode
, SYSFS_DIR
);
684 sd
->s_flags
|= (type
<< SYSFS_NS_TYPE_SHIFT
);
686 sd
->s_dir
.kobj
= kobj
;
689 sysfs_addrm_start(&acxt
);
690 rc
= sysfs_add_one(&acxt
, sd
, parent_sd
);
691 sysfs_addrm_finish(&acxt
);
701 int sysfs_create_subdir(struct kobject
*kobj
, const char *name
,
702 struct sysfs_dirent
**p_sd
)
704 return create_dir(kobj
, kobj
->sd
,
705 KOBJ_NS_TYPE_NONE
, name
, NULL
, p_sd
);
709 * sysfs_read_ns_type: return associated ns_type
710 * @kobj: the kobject being queried
712 * Each kobject can be tagged with exactly one namespace type
713 * (i.e. network or user). Return the ns_type associated with
716 static enum kobj_ns_type
sysfs_read_ns_type(struct kobject
*kobj
)
718 const struct kobj_ns_type_operations
*ops
;
719 enum kobj_ns_type type
;
721 ops
= kobj_child_ns_ops(kobj
);
723 return KOBJ_NS_TYPE_NONE
;
726 BUG_ON(type
<= KOBJ_NS_TYPE_NONE
);
727 BUG_ON(type
>= KOBJ_NS_TYPES
);
728 BUG_ON(!kobj_ns_type_registered(type
));
734 * sysfs_create_dir_ns - create a directory for an object with a namespace tag
735 * @kobj: object we're creating directory for
736 * @ns: the namespace tag to use
738 int sysfs_create_dir_ns(struct kobject
*kobj
, const void *ns
)
740 enum kobj_ns_type type
;
741 struct sysfs_dirent
*parent_sd
, *sd
;
747 parent_sd
= kobj
->parent
->sd
;
749 parent_sd
= &sysfs_root
;
754 type
= sysfs_read_ns_type(kobj
);
756 error
= create_dir(kobj
, parent_sd
, type
, kobject_name(kobj
), ns
, &sd
);
762 static struct dentry
*sysfs_lookup(struct inode
*dir
, struct dentry
*dentry
,
765 struct dentry
*ret
= NULL
;
766 struct dentry
*parent
= dentry
->d_parent
;
767 struct sysfs_dirent
*parent_sd
= parent
->d_fsdata
;
768 struct sysfs_dirent
*sd
;
770 enum kobj_ns_type type
;
773 mutex_lock(&sysfs_mutex
);
775 type
= sysfs_ns_type(parent_sd
);
776 ns
= sysfs_info(dir
->i_sb
)->ns
[type
];
778 sd
= sysfs_find_dirent(parent_sd
, dentry
->d_name
.name
, ns
);
782 ret
= ERR_PTR(-ENOENT
);
785 dentry
->d_fsdata
= sysfs_get(sd
);
787 /* attach dentry and inode */
788 inode
= sysfs_get_inode(dir
->i_sb
, sd
);
790 ret
= ERR_PTR(-ENOMEM
);
794 /* instantiate and hash dentry */
795 ret
= d_materialise_unique(dentry
, inode
);
797 mutex_unlock(&sysfs_mutex
);
801 const struct inode_operations sysfs_dir_inode_operations
= {
802 .lookup
= sysfs_lookup
,
803 .permission
= sysfs_permission
,
804 .setattr
= sysfs_setattr
,
805 .getattr
= sysfs_getattr
,
806 .setxattr
= sysfs_setxattr
,
809 static struct sysfs_dirent
*sysfs_leftmost_descendant(struct sysfs_dirent
*pos
)
811 struct sysfs_dirent
*last
;
818 if (sysfs_type(pos
) != SYSFS_DIR
)
821 rbn
= rb_first(&pos
->s_dir
.children
);
825 pos
= to_sysfs_dirent(rbn
);
832 * sysfs_next_descendant_post - find the next descendant for post-order walk
833 * @pos: the current position (%NULL to initiate traversal)
834 * @root: sysfs_dirent whose descendants to walk
836 * Find the next descendant to visit for post-order traversal of @root's
837 * descendants. @root is included in the iteration and the last node to be
840 static struct sysfs_dirent
*sysfs_next_descendant_post(struct sysfs_dirent
*pos
,
841 struct sysfs_dirent
*root
)
845 lockdep_assert_held(&sysfs_mutex
);
847 /* if first iteration, visit leftmost descendant which may be root */
849 return sysfs_leftmost_descendant(root
);
851 /* if we visited @root, we're done */
855 /* if there's an unvisited sibling, visit its leftmost descendant */
856 rbn
= rb_next(&pos
->s_rb
);
858 return sysfs_leftmost_descendant(to_sysfs_dirent(rbn
));
860 /* no sibling left, visit parent */
861 return pos
->s_parent
;
864 static void __sysfs_remove(struct sysfs_addrm_cxt
*acxt
,
865 struct sysfs_dirent
*sd
)
867 struct sysfs_dirent
*pos
, *next
;
872 pr_debug("sysfs %s: removing\n", sd
->s_name
);
877 next
= sysfs_next_descendant_post(pos
, sd
);
879 sysfs_remove_one(acxt
, pos
);
884 * sysfs_remove - remove a sysfs_dirent recursively
885 * @sd: the sysfs_dirent to remove
887 * Remove @sd along with all its subdirectories and files.
889 void sysfs_remove(struct sysfs_dirent
*sd
)
891 struct sysfs_addrm_cxt acxt
;
893 sysfs_addrm_start(&acxt
);
894 __sysfs_remove(&acxt
, sd
);
895 sysfs_addrm_finish(&acxt
);
899 * sysfs_hash_and_remove - find a sysfs_dirent by name and remove it
900 * @dir_sd: parent of the target
901 * @name: name of the sysfs_dirent to remove
902 * @ns: namespace tag of the sysfs_dirent to remove
904 * Look for the sysfs_dirent with @name and @ns under @dir_sd and remove
905 * it. Returns 0 on success, -ENOENT if such entry doesn't exist.
907 int sysfs_hash_and_remove(struct sysfs_dirent
*dir_sd
, const char *name
,
910 struct sysfs_addrm_cxt acxt
;
911 struct sysfs_dirent
*sd
;
914 WARN(1, KERN_WARNING
"sysfs: can not remove '%s', no directory\n",
919 sysfs_addrm_start(&acxt
);
921 sd
= sysfs_find_dirent(dir_sd
, name
, ns
);
923 __sysfs_remove(&acxt
, sd
);
925 sysfs_addrm_finish(&acxt
);
934 * sysfs_remove_dir - remove an object's directory.
937 * The only thing special about this is that we remove any files in
938 * the directory before we remove the directory, and we've inlined
939 * what used to be sysfs_rmdir() below, instead of calling separately.
941 void sysfs_remove_dir(struct kobject
*kobj
)
943 struct sysfs_dirent
*sd
= kobj
->sd
;
946 * In general, kboject owner is responsible for ensuring removal
947 * doesn't race with other operations and sysfs doesn't provide any
948 * protection; however, when @kobj is used as a symlink target, the
949 * symlinking entity usually doesn't own @kobj and thus has no
950 * control over removal. @kobj->sd may be removed anytime and
951 * symlink code may end up dereferencing an already freed sd.
953 * sysfs_symlink_target_lock synchronizes @kobj->sd disassociation
954 * against symlink operations so that symlink code can safely
955 * dereference @kobj->sd.
957 spin_lock(&sysfs_symlink_target_lock
);
959 spin_unlock(&sysfs_symlink_target_lock
);
962 WARN_ON_ONCE(sysfs_type(sd
) != SYSFS_DIR
);
967 int sysfs_rename(struct sysfs_dirent
*sd
, struct sysfs_dirent
*new_parent_sd
,
968 const char *new_name
, const void *new_ns
)
972 mutex_lock(&sysfs_mutex
);
975 if ((sd
->s_parent
== new_parent_sd
) && (sd
->s_ns
== new_ns
) &&
976 (strcmp(sd
->s_name
, new_name
) == 0))
977 goto out
; /* nothing to rename */
980 if (sysfs_find_dirent(new_parent_sd
, new_name
, new_ns
))
983 /* rename sysfs_dirent */
984 if (strcmp(sd
->s_name
, new_name
) != 0) {
986 new_name
= kstrdup(new_name
, GFP_KERNEL
);
991 sd
->s_name
= new_name
;
995 * Move to the appropriate place in the appropriate directories rbtree.
997 sysfs_unlink_sibling(sd
);
998 sysfs_get(new_parent_sd
);
999 sysfs_put(sd
->s_parent
);
1001 sd
->s_hash
= sysfs_name_hash(sd
->s_name
, sd
->s_ns
);
1002 sd
->s_parent
= new_parent_sd
;
1003 sysfs_link_sibling(sd
);
1007 mutex_unlock(&sysfs_mutex
);
1011 int sysfs_rename_dir_ns(struct kobject
*kobj
, const char *new_name
,
1014 struct sysfs_dirent
*parent_sd
= kobj
->sd
->s_parent
;
1016 return sysfs_rename(kobj
->sd
, parent_sd
, new_name
, new_ns
);
1019 int sysfs_move_dir_ns(struct kobject
*kobj
, struct kobject
*new_parent_kobj
,
1022 struct sysfs_dirent
*sd
= kobj
->sd
;
1023 struct sysfs_dirent
*new_parent_sd
;
1025 BUG_ON(!sd
->s_parent
);
1026 new_parent_sd
= new_parent_kobj
&& new_parent_kobj
->sd
?
1027 new_parent_kobj
->sd
: &sysfs_root
;
1029 return sysfs_rename(sd
, new_parent_sd
, sd
->s_name
, new_ns
);
1032 /* Relationship between s_mode and the DT_xxx types */
1033 static inline unsigned char dt_type(struct sysfs_dirent
*sd
)
1035 return (sd
->s_mode
>> 12) & 15;
1038 static int sysfs_dir_release(struct inode
*inode
, struct file
*filp
)
1040 sysfs_put(filp
->private_data
);
1044 static struct sysfs_dirent
*sysfs_dir_pos(const void *ns
,
1045 struct sysfs_dirent
*parent_sd
, loff_t hash
, struct sysfs_dirent
*pos
)
1048 int valid
= !(pos
->s_flags
& SYSFS_FLAG_REMOVED
) &&
1049 pos
->s_parent
== parent_sd
&&
1050 hash
== pos
->s_hash
;
1055 if (!pos
&& (hash
> 1) && (hash
< INT_MAX
)) {
1056 struct rb_node
*node
= parent_sd
->s_dir
.children
.rb_node
;
1058 pos
= to_sysfs_dirent(node
);
1060 if (hash
< pos
->s_hash
)
1061 node
= node
->rb_left
;
1062 else if (hash
> pos
->s_hash
)
1063 node
= node
->rb_right
;
1068 /* Skip over entries in the wrong namespace */
1069 while (pos
&& pos
->s_ns
!= ns
) {
1070 struct rb_node
*node
= rb_next(&pos
->s_rb
);
1074 pos
= to_sysfs_dirent(node
);
1079 static struct sysfs_dirent
*sysfs_dir_next_pos(const void *ns
,
1080 struct sysfs_dirent
*parent_sd
, ino_t ino
, struct sysfs_dirent
*pos
)
1082 pos
= sysfs_dir_pos(ns
, parent_sd
, ino
, pos
);
1085 struct rb_node
*node
= rb_next(&pos
->s_rb
);
1089 pos
= to_sysfs_dirent(node
);
1090 } while (pos
&& pos
->s_ns
!= ns
);
1094 static int sysfs_readdir(struct file
*file
, struct dir_context
*ctx
)
1096 struct dentry
*dentry
= file
->f_path
.dentry
;
1097 struct sysfs_dirent
*parent_sd
= dentry
->d_fsdata
;
1098 struct sysfs_dirent
*pos
= file
->private_data
;
1099 enum kobj_ns_type type
;
1102 type
= sysfs_ns_type(parent_sd
);
1103 ns
= sysfs_info(dentry
->d_sb
)->ns
[type
];
1105 if (!dir_emit_dots(file
, ctx
))
1107 mutex_lock(&sysfs_mutex
);
1108 for (pos
= sysfs_dir_pos(ns
, parent_sd
, ctx
->pos
, pos
);
1110 pos
= sysfs_dir_next_pos(ns
, parent_sd
, ctx
->pos
, pos
)) {
1111 const char *name
= pos
->s_name
;
1112 unsigned int type
= dt_type(pos
);
1113 int len
= strlen(name
);
1114 ino_t ino
= pos
->s_ino
;
1115 ctx
->pos
= pos
->s_hash
;
1116 file
->private_data
= sysfs_get(pos
);
1118 mutex_unlock(&sysfs_mutex
);
1119 if (!dir_emit(ctx
, name
, len
, ino
, type
))
1121 mutex_lock(&sysfs_mutex
);
1123 mutex_unlock(&sysfs_mutex
);
1124 file
->private_data
= NULL
;
1129 static loff_t
sysfs_dir_llseek(struct file
*file
, loff_t offset
, int whence
)
1131 struct inode
*inode
= file_inode(file
);
1134 mutex_lock(&inode
->i_mutex
);
1135 ret
= generic_file_llseek(file
, offset
, whence
);
1136 mutex_unlock(&inode
->i_mutex
);
1141 const struct file_operations sysfs_dir_operations
= {
1142 .read
= generic_read_dir
,
1143 .iterate
= sysfs_readdir
,
1144 .release
= sysfs_dir_release
,
1145 .llseek
= sysfs_dir_llseek
,