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 (sd->s_flags & SYSFS_FLAG_LOCKDEP)
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 (sd->s_flags & SYSFS_FLAG_LOCKDEP)
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);
244 * kernfs_get - get a reference count on a sysfs_dirent
245 * @sd: the target sysfs_dirent
247 void kernfs_get(struct sysfs_dirent *sd)
250 WARN_ON(!atomic_read(&sd->s_count));
251 atomic_inc(&sd->s_count);
254 EXPORT_SYMBOL_GPL(kernfs_get);
257 * kernfs_put - put a reference count on a sysfs_dirent
258 * @sd: the target sysfs_dirent
260 * Put a reference count of @sd and destroy it if it reached zero.
262 void kernfs_put(struct sysfs_dirent *sd)
264 struct sysfs_dirent *parent_sd;
266 if (!sd || !atomic_dec_and_test(&sd->s_count))
269 /* Moving/renaming is always done while holding reference.
270 * sd->s_parent won't change beneath us.
272 parent_sd = sd->s_parent;
274 WARN(!(sd->s_flags & SYSFS_FLAG_REMOVED),
275 "sysfs: free using entry: %s/%s\n",
276 parent_sd ? parent_sd->s_name : "", sd->s_name);
278 if (sysfs_type(sd) == SYSFS_KOBJ_LINK)
279 kernfs_put(sd->s_symlink.target_sd);
280 if (sysfs_type(sd) & SYSFS_COPY_NAME)
282 if (sd->s_iattr && sd->s_iattr->ia_secdata)
283 security_release_secctx(sd->s_iattr->ia_secdata,
284 sd->s_iattr->ia_secdata_len);
286 sysfs_free_ino(sd->s_ino);
287 kmem_cache_free(sysfs_dir_cachep, sd);
290 if (sd && atomic_dec_and_test(&sd->s_count))
293 EXPORT_SYMBOL_GPL(kernfs_put);
295 static int sysfs_dentry_delete(const struct dentry *dentry)
297 struct sysfs_dirent *sd = dentry->d_fsdata;
298 return !(sd && !(sd->s_flags & SYSFS_FLAG_REMOVED));
301 static int sysfs_dentry_revalidate(struct dentry *dentry, unsigned int flags)
303 struct sysfs_dirent *sd;
305 if (flags & LOOKUP_RCU)
308 sd = dentry->d_fsdata;
309 mutex_lock(&sysfs_mutex);
311 /* The sysfs dirent has been deleted */
312 if (sd->s_flags & SYSFS_FLAG_REMOVED)
315 /* The sysfs dirent has been moved? */
316 if (dentry->d_parent->d_fsdata != sd->s_parent)
319 /* The sysfs dirent has been renamed */
320 if (strcmp(dentry->d_name.name, sd->s_name) != 0)
323 /* The sysfs dirent has been moved to a different namespace */
324 if (sd->s_parent && (sd->s_parent->s_flags & SYSFS_FLAG_NS) &&
325 sysfs_info(dentry->d_sb)->ns != sd->s_ns)
328 mutex_unlock(&sysfs_mutex);
332 /* Remove the dentry from the dcache hashes.
333 * If this is a deleted dentry we use d_drop instead of d_delete
334 * so sysfs doesn't need to cope with negative dentries.
336 * If this is a dentry that has simply been renamed we
337 * use d_drop to remove it from the dcache lookup on its
338 * old parent. If this dentry persists later when a lookup
339 * is performed at its new name the dentry will be readded
340 * to the dcache hashes.
342 mutex_unlock(&sysfs_mutex);
344 /* If we have submounts we must allow the vfs caches
345 * to lie about the state of the filesystem to prevent
346 * leaks and other nasty things.
348 if (check_submounts_and_drop(dentry) != 0)
354 static void sysfs_dentry_release(struct dentry *dentry)
356 kernfs_put(dentry->d_fsdata);
359 const struct dentry_operations sysfs_dentry_ops = {
360 .d_revalidate = sysfs_dentry_revalidate,
361 .d_delete = sysfs_dentry_delete,
362 .d_release = sysfs_dentry_release,
365 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
367 char *dup_name = NULL;
368 struct sysfs_dirent *sd;
370 if (type & SYSFS_COPY_NAME) {
371 name = dup_name = kstrdup(name, GFP_KERNEL);
376 sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
380 if (sysfs_alloc_ino(&sd->s_ino))
383 atomic_set(&sd->s_count, 1);
384 atomic_set(&sd->s_active, 0);
388 sd->s_flags = type | SYSFS_FLAG_REMOVED;
393 kmem_cache_free(sysfs_dir_cachep, sd);
400 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
401 * @acxt: pointer to sysfs_addrm_cxt to be used
403 * This function is called when the caller is about to add or remove
404 * sysfs_dirent. This function acquires sysfs_mutex. @acxt is used
405 * to keep and pass context to other addrm functions.
408 * Kernel thread context (may sleep). sysfs_mutex is locked on
411 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt)
412 __acquires(sysfs_mutex)
414 memset(acxt, 0, sizeof(*acxt));
416 mutex_lock(&sysfs_mutex);
420 * sysfs_add_one - add sysfs_dirent to parent without warning
421 * @acxt: addrm context to use
422 * @sd: sysfs_dirent to be added
423 * @parent_sd: the parent sysfs_dirent to add @sd to
425 * Get @parent_sd and set @sd->s_parent to it and increment nlink of
426 * the parent inode if @sd is a directory and link into the children
427 * list of the parent.
429 * This function should be called between calls to
430 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
431 * passed the same @acxt as passed to sysfs_addrm_start().
434 * Determined by sysfs_addrm_start().
437 * 0 on success, -EEXIST if entry with the given name already
440 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd,
441 struct sysfs_dirent *parent_sd)
443 bool has_ns = parent_sd->s_flags & SYSFS_FLAG_NS;
444 struct sysfs_inode_attrs *ps_iattr;
447 if (has_ns != (bool)sd->s_ns) {
448 WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n",
449 has_ns ? "required" : "invalid",
450 parent_sd->s_name, sd->s_name);
454 if (sysfs_type(parent_sd) != SYSFS_DIR)
457 sd->s_hash = sysfs_name_hash(sd->s_name, sd->s_ns);
458 sd->s_parent = parent_sd;
459 kernfs_get(parent_sd);
461 ret = sysfs_link_sibling(sd);
465 /* Update timestamps on the parent */
466 ps_iattr = parent_sd->s_iattr;
468 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
469 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
472 /* Mark the entry added into directory tree */
473 sd->s_flags &= ~SYSFS_FLAG_REMOVED;
479 * sysfs_pathname - return full path to sysfs dirent
480 * @sd: sysfs_dirent whose path we want
481 * @path: caller allocated buffer of size PATH_MAX
483 * Gives the name "/" to the sysfs_root entry; any path returned
484 * is relative to wherever sysfs is mounted.
486 static char *sysfs_pathname(struct sysfs_dirent *sd, char *path)
489 sysfs_pathname(sd->s_parent, path);
490 strlcat(path, "/", PATH_MAX);
492 strlcat(path, sd->s_name, PATH_MAX);
496 void sysfs_warn_dup(struct sysfs_dirent *parent, const char *name)
500 path = kzalloc(PATH_MAX, GFP_KERNEL);
502 sysfs_pathname(parent, path);
503 strlcat(path, "/", PATH_MAX);
504 strlcat(path, name, PATH_MAX);
507 WARN(1, KERN_WARNING "sysfs: cannot create duplicate filename '%s'\n",
514 * sysfs_remove_one - remove sysfs_dirent from parent
515 * @acxt: addrm context to use
516 * @sd: sysfs_dirent to be removed
518 * Mark @sd removed and drop nlink of parent inode if @sd is a
519 * directory. @sd is unlinked from the children list.
521 * This function should be called between calls to
522 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
523 * passed the same @acxt as passed to sysfs_addrm_start().
526 * Determined by sysfs_addrm_start().
528 static void sysfs_remove_one(struct sysfs_addrm_cxt *acxt,
529 struct sysfs_dirent *sd)
531 struct sysfs_inode_attrs *ps_iattr;
534 * Removal can be called multiple times on the same node. Only the
535 * first invocation is effective and puts the base ref.
537 if (sd->s_flags & SYSFS_FLAG_REMOVED)
540 sysfs_unlink_sibling(sd);
542 /* Update timestamps on the parent */
543 ps_iattr = sd->s_parent->s_iattr;
545 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
546 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
549 sd->s_flags |= SYSFS_FLAG_REMOVED;
550 sd->u.removed_list = acxt->removed;
555 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
556 * @acxt: addrm context to finish up
558 * Finish up sysfs_dirent add/remove. Resources acquired by
559 * sysfs_addrm_start() are released and removed sysfs_dirents are
563 * sysfs_mutex is released.
565 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
566 __releases(sysfs_mutex)
568 /* release resources acquired by sysfs_addrm_start() */
569 mutex_unlock(&sysfs_mutex);
571 /* kill removed sysfs_dirents */
572 while (acxt->removed) {
573 struct sysfs_dirent *sd = acxt->removed;
575 acxt->removed = sd->u.removed_list;
577 sysfs_deactivate(sd);
578 sysfs_unmap_bin_file(sd);
584 * kernfs_find_ns - find sysfs_dirent with the given name
585 * @parent: sysfs_dirent to search under
586 * @name: name to look for
587 * @ns: the namespace tag to use
589 * Look for sysfs_dirent with name @name under @parent. Returns pointer to
590 * the found sysfs_dirent on success, %NULL on failure.
592 static struct sysfs_dirent *kernfs_find_ns(struct sysfs_dirent *parent,
593 const unsigned char *name,
596 struct rb_node *node = parent->s_dir.children.rb_node;
597 bool has_ns = parent->s_flags & SYSFS_FLAG_NS;
600 lockdep_assert_held(&sysfs_mutex);
602 if (has_ns != (bool)ns) {
603 WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n",
604 has_ns ? "required" : "invalid",
605 parent->s_name, name);
609 hash = sysfs_name_hash(name, ns);
611 struct sysfs_dirent *sd;
614 sd = to_sysfs_dirent(node);
615 result = sysfs_name_compare(hash, name, ns, sd);
617 node = node->rb_left;
619 node = node->rb_right;
627 * kernfs_find_and_get_ns - find and get sysfs_dirent with the given name
628 * @parent: sysfs_dirent to search under
629 * @name: name to look for
630 * @ns: the namespace tag to use
632 * Look for sysfs_dirent with name @name under @parent and get a reference
633 * if found. This function may sleep and returns pointer to the found
634 * sysfs_dirent on success, %NULL on failure.
636 struct sysfs_dirent *kernfs_find_and_get_ns(struct sysfs_dirent *parent,
637 const char *name, const void *ns)
639 struct sysfs_dirent *sd;
641 mutex_lock(&sysfs_mutex);
642 sd = kernfs_find_ns(parent, name, ns);
644 mutex_unlock(&sysfs_mutex);
648 EXPORT_SYMBOL_GPL(kernfs_find_and_get_ns);
651 * kernfs_create_dir_ns - create a directory
652 * @parent: parent in which to create a new directory
653 * @name: name of the new directory
654 * @priv: opaque data associated with the new directory
655 * @ns: optional namespace tag of the directory
657 * Returns the created node on success, ERR_PTR() value on failure.
659 struct sysfs_dirent *kernfs_create_dir_ns(struct sysfs_dirent *parent,
660 const char *name, void *priv,
663 umode_t mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
664 struct sysfs_addrm_cxt acxt;
665 struct sysfs_dirent *sd;
669 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
671 return ERR_PTR(-ENOMEM);
677 sysfs_addrm_start(&acxt);
678 rc = sysfs_add_one(&acxt, sd, parent);
679 sysfs_addrm_finish(&acxt);
689 * sysfs_create_dir_ns - create a directory for an object with a namespace tag
690 * @kobj: object we're creating directory for
691 * @ns: the namespace tag to use
693 int sysfs_create_dir_ns(struct kobject *kobj, const void *ns)
695 struct sysfs_dirent *parent_sd, *sd;
700 parent_sd = kobj->parent->sd;
702 parent_sd = &sysfs_root;
707 sd = kernfs_create_dir_ns(parent_sd, kobject_name(kobj), kobj, ns);
709 if (PTR_ERR(sd) == -EEXIST)
710 sysfs_warn_dup(parent_sd, kobject_name(kobj));
718 static struct dentry *sysfs_lookup(struct inode *dir, struct dentry *dentry,
721 struct dentry *ret = NULL;
722 struct dentry *parent = dentry->d_parent;
723 struct sysfs_dirent *parent_sd = parent->d_fsdata;
724 struct sysfs_dirent *sd;
726 const void *ns = NULL;
728 mutex_lock(&sysfs_mutex);
730 if (parent_sd->s_flags & SYSFS_FLAG_NS)
731 ns = sysfs_info(dir->i_sb)->ns;
733 sd = kernfs_find_ns(parent_sd, dentry->d_name.name, ns);
737 ret = ERR_PTR(-ENOENT);
741 dentry->d_fsdata = sd;
743 /* attach dentry and inode */
744 inode = sysfs_get_inode(dir->i_sb, sd);
746 ret = ERR_PTR(-ENOMEM);
750 /* instantiate and hash dentry */
751 ret = d_materialise_unique(dentry, inode);
753 mutex_unlock(&sysfs_mutex);
757 const struct inode_operations sysfs_dir_inode_operations = {
758 .lookup = sysfs_lookup,
759 .permission = sysfs_permission,
760 .setattr = sysfs_setattr,
761 .getattr = sysfs_getattr,
762 .setxattr = sysfs_setxattr,
765 static struct sysfs_dirent *sysfs_leftmost_descendant(struct sysfs_dirent *pos)
767 struct sysfs_dirent *last;
774 if (sysfs_type(pos) != SYSFS_DIR)
777 rbn = rb_first(&pos->s_dir.children);
781 pos = to_sysfs_dirent(rbn);
788 * sysfs_next_descendant_post - find the next descendant for post-order walk
789 * @pos: the current position (%NULL to initiate traversal)
790 * @root: sysfs_dirent whose descendants to walk
792 * Find the next descendant to visit for post-order traversal of @root's
793 * descendants. @root is included in the iteration and the last node to be
796 static struct sysfs_dirent *sysfs_next_descendant_post(struct sysfs_dirent *pos,
797 struct sysfs_dirent *root)
801 lockdep_assert_held(&sysfs_mutex);
803 /* if first iteration, visit leftmost descendant which may be root */
805 return sysfs_leftmost_descendant(root);
807 /* if we visited @root, we're done */
811 /* if there's an unvisited sibling, visit its leftmost descendant */
812 rbn = rb_next(&pos->s_rb);
814 return sysfs_leftmost_descendant(to_sysfs_dirent(rbn));
816 /* no sibling left, visit parent */
817 return pos->s_parent;
820 static void __kernfs_remove(struct sysfs_addrm_cxt *acxt,
821 struct sysfs_dirent *sd)
823 struct sysfs_dirent *pos, *next;
828 pr_debug("sysfs %s: removing\n", sd->s_name);
833 next = sysfs_next_descendant_post(pos, sd);
835 sysfs_remove_one(acxt, pos);
840 * kernfs_remove - remove a sysfs_dirent recursively
841 * @sd: the sysfs_dirent to remove
843 * Remove @sd along with all its subdirectories and files.
845 void kernfs_remove(struct sysfs_dirent *sd)
847 struct sysfs_addrm_cxt acxt;
849 sysfs_addrm_start(&acxt);
850 __kernfs_remove(&acxt, sd);
851 sysfs_addrm_finish(&acxt);
855 * kernfs_remove_by_name_ns - find a sysfs_dirent by name and remove it
856 * @dir_sd: parent of the target
857 * @name: name of the sysfs_dirent to remove
858 * @ns: namespace tag of the sysfs_dirent to remove
860 * Look for the sysfs_dirent with @name and @ns under @dir_sd and remove
861 * it. Returns 0 on success, -ENOENT if such entry doesn't exist.
863 int kernfs_remove_by_name_ns(struct sysfs_dirent *dir_sd, const char *name,
866 struct sysfs_addrm_cxt acxt;
867 struct sysfs_dirent *sd;
870 WARN(1, KERN_WARNING "sysfs: can not remove '%s', no directory\n",
875 sysfs_addrm_start(&acxt);
877 sd = kernfs_find_ns(dir_sd, name, ns);
879 __kernfs_remove(&acxt, sd);
881 sysfs_addrm_finish(&acxt);
890 * sysfs_remove_dir - remove an object's directory.
893 * The only thing special about this is that we remove any files in
894 * the directory before we remove the directory, and we've inlined
895 * what used to be sysfs_rmdir() below, instead of calling separately.
897 void sysfs_remove_dir(struct kobject *kobj)
899 struct sysfs_dirent *sd = kobj->sd;
902 * In general, kboject owner is responsible for ensuring removal
903 * doesn't race with other operations and sysfs doesn't provide any
904 * protection; however, when @kobj is used as a symlink target, the
905 * symlinking entity usually doesn't own @kobj and thus has no
906 * control over removal. @kobj->sd may be removed anytime and
907 * symlink code may end up dereferencing an already freed sd.
909 * sysfs_symlink_target_lock synchronizes @kobj->sd disassociation
910 * against symlink operations so that symlink code can safely
911 * dereference @kobj->sd.
913 spin_lock(&sysfs_symlink_target_lock);
915 spin_unlock(&sysfs_symlink_target_lock);
918 WARN_ON_ONCE(sysfs_type(sd) != SYSFS_DIR);
924 * kernfs_rename_ns - move and rename a kernfs_node
926 * @new_parent: new parent to put @sd under
927 * @new_name: new name
928 * @new_ns: new namespace tag
930 int kernfs_rename_ns(struct sysfs_dirent *sd, struct sysfs_dirent *new_parent,
931 const char *new_name, const void *new_ns)
935 mutex_lock(&sysfs_mutex);
938 if ((sd->s_parent == new_parent) && (sd->s_ns == new_ns) &&
939 (strcmp(sd->s_name, new_name) == 0))
940 goto out; /* nothing to rename */
943 if (kernfs_find_ns(new_parent, new_name, new_ns))
946 /* rename sysfs_dirent */
947 if (strcmp(sd->s_name, new_name) != 0) {
949 new_name = kstrdup(new_name, GFP_KERNEL);
954 sd->s_name = new_name;
958 * Move to the appropriate place in the appropriate directories rbtree.
960 sysfs_unlink_sibling(sd);
961 kernfs_get(new_parent);
962 kernfs_put(sd->s_parent);
964 sd->s_hash = sysfs_name_hash(sd->s_name, sd->s_ns);
965 sd->s_parent = new_parent;
966 sysfs_link_sibling(sd);
970 mutex_unlock(&sysfs_mutex);
974 int sysfs_rename_dir_ns(struct kobject *kobj, const char *new_name,
977 struct sysfs_dirent *parent_sd = kobj->sd->s_parent;
979 return kernfs_rename_ns(kobj->sd, parent_sd, new_name, new_ns);
982 int sysfs_move_dir_ns(struct kobject *kobj, struct kobject *new_parent_kobj,
985 struct sysfs_dirent *sd = kobj->sd;
986 struct sysfs_dirent *new_parent_sd;
988 BUG_ON(!sd->s_parent);
989 new_parent_sd = new_parent_kobj && new_parent_kobj->sd ?
990 new_parent_kobj->sd : &sysfs_root;
992 return kernfs_rename_ns(sd, new_parent_sd, sd->s_name, new_ns);
996 * kernfs_enable_ns - enable namespace under a directory
997 * @sd: directory of interest, should be empty
999 * This is to be called right after @sd is created to enable namespace
1000 * under it. All children of @sd must have non-NULL namespace tags and
1001 * only the ones which match the super_block's tag will be visible.
1003 void kernfs_enable_ns(struct sysfs_dirent *sd)
1005 WARN_ON_ONCE(sysfs_type(sd) != SYSFS_DIR);
1006 WARN_ON_ONCE(!RB_EMPTY_ROOT(&sd->s_dir.children));
1007 sd->s_flags |= SYSFS_FLAG_NS;
1010 /* Relationship between s_mode and the DT_xxx types */
1011 static inline unsigned char dt_type(struct sysfs_dirent *sd)
1013 return (sd->s_mode >> 12) & 15;
1016 static int sysfs_dir_release(struct inode *inode, struct file *filp)
1018 kernfs_put(filp->private_data);
1022 static struct sysfs_dirent *sysfs_dir_pos(const void *ns,
1023 struct sysfs_dirent *parent_sd, loff_t hash, struct sysfs_dirent *pos)
1026 int valid = !(pos->s_flags & SYSFS_FLAG_REMOVED) &&
1027 pos->s_parent == parent_sd &&
1028 hash == pos->s_hash;
1033 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1034 struct rb_node *node = parent_sd->s_dir.children.rb_node;
1036 pos = to_sysfs_dirent(node);
1038 if (hash < pos->s_hash)
1039 node = node->rb_left;
1040 else if (hash > pos->s_hash)
1041 node = node->rb_right;
1046 /* Skip over entries in the wrong namespace */
1047 while (pos && pos->s_ns != ns) {
1048 struct rb_node *node = rb_next(&pos->s_rb);
1052 pos = to_sysfs_dirent(node);
1057 static struct sysfs_dirent *sysfs_dir_next_pos(const void *ns,
1058 struct sysfs_dirent *parent_sd, ino_t ino, struct sysfs_dirent *pos)
1060 pos = sysfs_dir_pos(ns, parent_sd, ino, pos);
1063 struct rb_node *node = rb_next(&pos->s_rb);
1067 pos = to_sysfs_dirent(node);
1068 } while (pos && pos->s_ns != ns);
1072 static int sysfs_readdir(struct file *file, struct dir_context *ctx)
1074 struct dentry *dentry = file->f_path.dentry;
1075 struct sysfs_dirent *parent_sd = dentry->d_fsdata;
1076 struct sysfs_dirent *pos = file->private_data;
1077 const void *ns = NULL;
1079 if (!dir_emit_dots(file, ctx))
1081 mutex_lock(&sysfs_mutex);
1083 if (parent_sd->s_flags & SYSFS_FLAG_NS)
1084 ns = sysfs_info(dentry->d_sb)->ns;
1086 for (pos = sysfs_dir_pos(ns, parent_sd, ctx->pos, pos);
1088 pos = sysfs_dir_next_pos(ns, parent_sd, ctx->pos, pos)) {
1089 const char *name = pos->s_name;
1090 unsigned int type = dt_type(pos);
1091 int len = strlen(name);
1092 ino_t ino = pos->s_ino;
1094 ctx->pos = pos->s_hash;
1095 file->private_data = pos;
1098 mutex_unlock(&sysfs_mutex);
1099 if (!dir_emit(ctx, name, len, ino, type))
1101 mutex_lock(&sysfs_mutex);
1103 mutex_unlock(&sysfs_mutex);
1104 file->private_data = NULL;
1109 static loff_t sysfs_dir_llseek(struct file *file, loff_t offset, int whence)
1111 struct inode *inode = file_inode(file);
1114 mutex_lock(&inode->i_mutex);
1115 ret = generic_file_llseek(file, offset, whence);
1116 mutex_unlock(&inode->i_mutex);
1121 const struct file_operations sysfs_dir_operations = {
1122 .read = generic_read_dir,
1123 .iterate = sysfs_readdir,
1124 .release = sysfs_dir_release,
1125 .llseek = sysfs_dir_llseek,