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;
283 if (flags & LOOKUP_RCU)
286 sd = dentry->d_fsdata;
287 mutex_lock(&sysfs_mutex);
289 /* The sysfs dirent has been deleted */
290 if (sd->s_flags & SYSFS_FLAG_REMOVED)
293 /* The sysfs dirent has been moved? */
294 if (dentry->d_parent->d_fsdata != sd->s_parent)
297 /* The sysfs dirent has been renamed */
298 if (strcmp(dentry->d_name.name, sd->s_name) != 0)
301 /* The sysfs dirent has been moved to a different namespace */
302 if (sd->s_parent && (sd->s_parent->s_flags & SYSFS_FLAG_NS) &&
303 sysfs_info(dentry->d_sb)->ns != sd->s_ns)
306 mutex_unlock(&sysfs_mutex);
310 /* Remove the dentry from the dcache hashes.
311 * If this is a deleted dentry we use d_drop instead of d_delete
312 * so sysfs doesn't need to cope with negative dentries.
314 * If this is a dentry that has simply been renamed we
315 * use d_drop to remove it from the dcache lookup on its
316 * old parent. If this dentry persists later when a lookup
317 * is performed at its new name the dentry will be readded
318 * to the dcache hashes.
320 mutex_unlock(&sysfs_mutex);
322 /* If we have submounts we must allow the vfs caches
323 * to lie about the state of the filesystem to prevent
324 * leaks and other nasty things.
326 if (check_submounts_and_drop(dentry) != 0)
332 static void sysfs_dentry_release(struct dentry *dentry)
334 sysfs_put(dentry->d_fsdata);
337 const struct dentry_operations sysfs_dentry_ops = {
338 .d_revalidate = sysfs_dentry_revalidate,
339 .d_delete = sysfs_dentry_delete,
340 .d_release = sysfs_dentry_release,
343 struct sysfs_dirent *sysfs_new_dirent(const char *name, umode_t mode, int type)
345 char *dup_name = NULL;
346 struct sysfs_dirent *sd;
348 if (type & SYSFS_COPY_NAME) {
349 name = dup_name = kstrdup(name, GFP_KERNEL);
354 sd = kmem_cache_zalloc(sysfs_dir_cachep, GFP_KERNEL);
358 if (sysfs_alloc_ino(&sd->s_ino))
361 atomic_set(&sd->s_count, 1);
362 atomic_set(&sd->s_active, 0);
366 sd->s_flags = type | SYSFS_FLAG_REMOVED;
371 kmem_cache_free(sysfs_dir_cachep, sd);
378 * sysfs_addrm_start - prepare for sysfs_dirent add/remove
379 * @acxt: pointer to sysfs_addrm_cxt to be used
381 * This function is called when the caller is about to add or remove
382 * sysfs_dirent. This function acquires sysfs_mutex. @acxt is used
383 * to keep and pass context to other addrm functions.
386 * Kernel thread context (may sleep). sysfs_mutex is locked on
389 void sysfs_addrm_start(struct sysfs_addrm_cxt *acxt)
390 __acquires(sysfs_mutex)
392 memset(acxt, 0, sizeof(*acxt));
394 mutex_lock(&sysfs_mutex);
398 * __sysfs_add_one - add sysfs_dirent to parent without warning
399 * @acxt: addrm context to use
400 * @sd: sysfs_dirent to be added
401 * @parent_sd: the parent sysfs_dirent to add @sd to
403 * Get @parent_sd and set @sd->s_parent to it and increment nlink of
404 * the parent inode if @sd is a directory and link into the children
405 * list of the parent.
407 * This function should be called between calls to
408 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
409 * passed the same @acxt as passed to sysfs_addrm_start().
412 * Determined by sysfs_addrm_start().
415 * 0 on success, -EEXIST if entry with the given name already
418 int __sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd,
419 struct sysfs_dirent *parent_sd)
421 bool has_ns = parent_sd->s_flags & SYSFS_FLAG_NS;
422 struct sysfs_inode_attrs *ps_iattr;
425 if (has_ns != (bool)sd->s_ns) {
426 WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n",
427 has_ns ? "required" : "invalid",
428 parent_sd->s_name, sd->s_name);
432 if (sysfs_type(parent_sd) != SYSFS_DIR)
435 sd->s_hash = sysfs_name_hash(sd->s_name, sd->s_ns);
436 sd->s_parent = sysfs_get(parent_sd);
438 ret = sysfs_link_sibling(sd);
442 /* Update timestamps on the parent */
443 ps_iattr = parent_sd->s_iattr;
445 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
446 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
449 /* Mark the entry added into directory tree */
450 sd->s_flags &= ~SYSFS_FLAG_REMOVED;
456 * sysfs_pathname - return full path to sysfs dirent
457 * @sd: sysfs_dirent whose path we want
458 * @path: caller allocated buffer of size PATH_MAX
460 * Gives the name "/" to the sysfs_root entry; any path returned
461 * is relative to wherever sysfs is mounted.
463 static char *sysfs_pathname(struct sysfs_dirent *sd, char *path)
466 sysfs_pathname(sd->s_parent, path);
467 strlcat(path, "/", PATH_MAX);
469 strlcat(path, sd->s_name, PATH_MAX);
473 void sysfs_warn_dup(struct sysfs_dirent *parent, const char *name)
477 path = kzalloc(PATH_MAX, GFP_KERNEL);
479 sysfs_pathname(parent, path);
480 strlcat(path, "/", PATH_MAX);
481 strlcat(path, name, PATH_MAX);
484 WARN(1, KERN_WARNING "sysfs: cannot create duplicate filename '%s'\n",
491 * sysfs_add_one - add sysfs_dirent to parent
492 * @acxt: addrm context to use
493 * @sd: sysfs_dirent to be added
494 * @parent_sd: the parent sysfs_dirent to add @sd to
496 * Get @parent_sd and set @sd->s_parent to it and increment nlink of
497 * the parent inode if @sd is a directory and link into the children
498 * list of the parent.
500 * This function should be called between calls to
501 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
502 * passed the same @acxt as passed to sysfs_addrm_start().
505 * Determined by sysfs_addrm_start().
508 * 0 on success, -EEXIST if entry with the given name already
511 int sysfs_add_one(struct sysfs_addrm_cxt *acxt, struct sysfs_dirent *sd,
512 struct sysfs_dirent *parent_sd)
516 ret = __sysfs_add_one(acxt, sd, parent_sd);
519 sysfs_warn_dup(parent_sd, sd->s_name);
524 * sysfs_remove_one - remove sysfs_dirent from parent
525 * @acxt: addrm context to use
526 * @sd: sysfs_dirent to be removed
528 * Mark @sd removed and drop nlink of parent inode if @sd is a
529 * directory. @sd is unlinked from the children list.
531 * This function should be called between calls to
532 * sysfs_addrm_start() and sysfs_addrm_finish() and should be
533 * passed the same @acxt as passed to sysfs_addrm_start().
536 * Determined by sysfs_addrm_start().
538 static void sysfs_remove_one(struct sysfs_addrm_cxt *acxt,
539 struct sysfs_dirent *sd)
541 struct sysfs_inode_attrs *ps_iattr;
544 * Removal can be called multiple times on the same node. Only the
545 * first invocation is effective and puts the base ref.
547 if (sd->s_flags & SYSFS_FLAG_REMOVED)
550 sysfs_unlink_sibling(sd);
552 /* Update timestamps on the parent */
553 ps_iattr = sd->s_parent->s_iattr;
555 struct iattr *ps_iattrs = &ps_iattr->ia_iattr;
556 ps_iattrs->ia_ctime = ps_iattrs->ia_mtime = CURRENT_TIME;
559 sd->s_flags |= SYSFS_FLAG_REMOVED;
560 sd->u.removed_list = acxt->removed;
565 * sysfs_addrm_finish - finish up sysfs_dirent add/remove
566 * @acxt: addrm context to finish up
568 * Finish up sysfs_dirent add/remove. Resources acquired by
569 * sysfs_addrm_start() are released and removed sysfs_dirents are
573 * sysfs_mutex is released.
575 void sysfs_addrm_finish(struct sysfs_addrm_cxt *acxt)
576 __releases(sysfs_mutex)
578 /* release resources acquired by sysfs_addrm_start() */
579 mutex_unlock(&sysfs_mutex);
581 /* kill removed sysfs_dirents */
582 while (acxt->removed) {
583 struct sysfs_dirent *sd = acxt->removed;
585 acxt->removed = sd->u.removed_list;
587 sysfs_deactivate(sd);
588 sysfs_unmap_bin_file(sd);
594 * sysfs_find_dirent - find sysfs_dirent with the given name
595 * @parent_sd: sysfs_dirent to search under
596 * @name: name to look for
597 * @ns: the namespace tag to use
599 * Look for sysfs_dirent with name @name under @parent_sd.
602 * mutex_lock(sysfs_mutex)
605 * Pointer to sysfs_dirent if found, NULL if not.
607 struct sysfs_dirent *sysfs_find_dirent(struct sysfs_dirent *parent_sd,
608 const unsigned char *name,
611 struct rb_node *node = parent_sd->s_dir.children.rb_node;
612 bool has_ns = parent_sd->s_flags & SYSFS_FLAG_NS;
615 if (has_ns != (bool)ns) {
616 WARN(1, KERN_WARNING "sysfs: ns %s in '%s' for '%s'\n",
617 has_ns ? "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);
670 * kernfs_create_dir_ns - create a directory
671 * @parent: parent in which to create a new directory
672 * @name: name of the new directory
673 * @priv: opaque data associated with the new directory
674 * @ns: optional namespace tag of the directory
676 * Returns the created node on success, ERR_PTR() value on failure.
678 struct sysfs_dirent *kernfs_create_dir_ns(struct sysfs_dirent *parent,
679 const char *name, void *priv,
682 umode_t mode = S_IFDIR | S_IRWXU | S_IRUGO | S_IXUGO;
683 struct sysfs_addrm_cxt acxt;
684 struct sysfs_dirent *sd;
688 sd = sysfs_new_dirent(name, mode, SYSFS_DIR);
690 return ERR_PTR(-ENOMEM);
696 sysfs_addrm_start(&acxt);
697 rc = __sysfs_add_one(&acxt, sd, parent);
698 sysfs_addrm_finish(&acxt);
708 * sysfs_create_dir_ns - create a directory for an object with a namespace tag
709 * @kobj: object we're creating directory for
710 * @ns: the namespace tag to use
712 int sysfs_create_dir_ns(struct kobject *kobj, const void *ns)
714 struct sysfs_dirent *parent_sd, *sd;
719 parent_sd = kobj->parent->sd;
721 parent_sd = &sysfs_root;
726 sd = kernfs_create_dir_ns(parent_sd, kobject_name(kobj), kobj, ns);
728 if (PTR_ERR(sd) == -EEXIST)
729 sysfs_warn_dup(parent_sd, kobject_name(kobj));
737 static struct dentry *sysfs_lookup(struct inode *dir, struct dentry *dentry,
740 struct dentry *ret = NULL;
741 struct dentry *parent = dentry->d_parent;
742 struct sysfs_dirent *parent_sd = parent->d_fsdata;
743 struct sysfs_dirent *sd;
745 const void *ns = NULL;
747 mutex_lock(&sysfs_mutex);
749 if (parent_sd->s_flags & SYSFS_FLAG_NS)
750 ns = sysfs_info(dir->i_sb)->ns;
752 sd = sysfs_find_dirent(parent_sd, dentry->d_name.name, ns);
756 ret = ERR_PTR(-ENOENT);
759 dentry->d_fsdata = sysfs_get(sd);
761 /* attach dentry and inode */
762 inode = sysfs_get_inode(dir->i_sb, sd);
764 ret = ERR_PTR(-ENOMEM);
768 /* instantiate and hash dentry */
769 ret = d_materialise_unique(dentry, inode);
771 mutex_unlock(&sysfs_mutex);
775 const struct inode_operations sysfs_dir_inode_operations = {
776 .lookup = sysfs_lookup,
777 .permission = sysfs_permission,
778 .setattr = sysfs_setattr,
779 .getattr = sysfs_getattr,
780 .setxattr = sysfs_setxattr,
783 static struct sysfs_dirent *sysfs_leftmost_descendant(struct sysfs_dirent *pos)
785 struct sysfs_dirent *last;
792 if (sysfs_type(pos) != SYSFS_DIR)
795 rbn = rb_first(&pos->s_dir.children);
799 pos = to_sysfs_dirent(rbn);
806 * sysfs_next_descendant_post - find the next descendant for post-order walk
807 * @pos: the current position (%NULL to initiate traversal)
808 * @root: sysfs_dirent whose descendants to walk
810 * Find the next descendant to visit for post-order traversal of @root's
811 * descendants. @root is included in the iteration and the last node to be
814 static struct sysfs_dirent *sysfs_next_descendant_post(struct sysfs_dirent *pos,
815 struct sysfs_dirent *root)
819 lockdep_assert_held(&sysfs_mutex);
821 /* if first iteration, visit leftmost descendant which may be root */
823 return sysfs_leftmost_descendant(root);
825 /* if we visited @root, we're done */
829 /* if there's an unvisited sibling, visit its leftmost descendant */
830 rbn = rb_next(&pos->s_rb);
832 return sysfs_leftmost_descendant(to_sysfs_dirent(rbn));
834 /* no sibling left, visit parent */
835 return pos->s_parent;
838 static void __kernfs_remove(struct sysfs_addrm_cxt *acxt,
839 struct sysfs_dirent *sd)
841 struct sysfs_dirent *pos, *next;
846 pr_debug("sysfs %s: removing\n", sd->s_name);
851 next = sysfs_next_descendant_post(pos, sd);
853 sysfs_remove_one(acxt, pos);
858 * kernfs_remove - remove a sysfs_dirent recursively
859 * @sd: the sysfs_dirent to remove
861 * Remove @sd along with all its subdirectories and files.
863 void kernfs_remove(struct sysfs_dirent *sd)
865 struct sysfs_addrm_cxt acxt;
867 sysfs_addrm_start(&acxt);
868 __kernfs_remove(&acxt, sd);
869 sysfs_addrm_finish(&acxt);
873 * kernfs_remove_by_name_ns - find a sysfs_dirent by name and remove it
874 * @dir_sd: parent of the target
875 * @name: name of the sysfs_dirent to remove
876 * @ns: namespace tag of the sysfs_dirent to remove
878 * Look for the sysfs_dirent with @name and @ns under @dir_sd and remove
879 * it. Returns 0 on success, -ENOENT if such entry doesn't exist.
881 int kernfs_remove_by_name_ns(struct sysfs_dirent *dir_sd, const char *name,
884 struct sysfs_addrm_cxt acxt;
885 struct sysfs_dirent *sd;
888 WARN(1, KERN_WARNING "sysfs: can not remove '%s', no directory\n",
893 sysfs_addrm_start(&acxt);
895 sd = sysfs_find_dirent(dir_sd, name, ns);
897 __kernfs_remove(&acxt, sd);
899 sysfs_addrm_finish(&acxt);
908 * sysfs_remove_dir - remove an object's directory.
911 * The only thing special about this is that we remove any files in
912 * the directory before we remove the directory, and we've inlined
913 * what used to be sysfs_rmdir() below, instead of calling separately.
915 void sysfs_remove_dir(struct kobject *kobj)
917 struct sysfs_dirent *sd = kobj->sd;
920 * In general, kboject owner is responsible for ensuring removal
921 * doesn't race with other operations and sysfs doesn't provide any
922 * protection; however, when @kobj is used as a symlink target, the
923 * symlinking entity usually doesn't own @kobj and thus has no
924 * control over removal. @kobj->sd may be removed anytime and
925 * symlink code may end up dereferencing an already freed sd.
927 * sysfs_symlink_target_lock synchronizes @kobj->sd disassociation
928 * against symlink operations so that symlink code can safely
929 * dereference @kobj->sd.
931 spin_lock(&sysfs_symlink_target_lock);
933 spin_unlock(&sysfs_symlink_target_lock);
936 WARN_ON_ONCE(sysfs_type(sd) != SYSFS_DIR);
942 * kernfs_rename_ns - move and rename a kernfs_node
944 * @new_parent: new parent to put @sd under
945 * @new_name: new name
946 * @new_ns: new namespace tag
948 int kernfs_rename_ns(struct sysfs_dirent *sd, struct sysfs_dirent *new_parent,
949 const char *new_name, const void *new_ns)
953 mutex_lock(&sysfs_mutex);
956 if ((sd->s_parent == new_parent) && (sd->s_ns == new_ns) &&
957 (strcmp(sd->s_name, new_name) == 0))
958 goto out; /* nothing to rename */
961 if (sysfs_find_dirent(new_parent, new_name, new_ns))
964 /* rename sysfs_dirent */
965 if (strcmp(sd->s_name, new_name) != 0) {
967 new_name = kstrdup(new_name, GFP_KERNEL);
972 sd->s_name = new_name;
976 * Move to the appropriate place in the appropriate directories rbtree.
978 sysfs_unlink_sibling(sd);
979 sysfs_get(new_parent);
980 sysfs_put(sd->s_parent);
982 sd->s_hash = sysfs_name_hash(sd->s_name, sd->s_ns);
983 sd->s_parent = new_parent;
984 sysfs_link_sibling(sd);
988 mutex_unlock(&sysfs_mutex);
992 int sysfs_rename_dir_ns(struct kobject *kobj, const char *new_name,
995 struct sysfs_dirent *parent_sd = kobj->sd->s_parent;
997 return kernfs_rename_ns(kobj->sd, parent_sd, new_name, new_ns);
1000 int sysfs_move_dir_ns(struct kobject *kobj, struct kobject *new_parent_kobj,
1003 struct sysfs_dirent *sd = kobj->sd;
1004 struct sysfs_dirent *new_parent_sd;
1006 BUG_ON(!sd->s_parent);
1007 new_parent_sd = new_parent_kobj && new_parent_kobj->sd ?
1008 new_parent_kobj->sd : &sysfs_root;
1010 return kernfs_rename_ns(sd, new_parent_sd, sd->s_name, new_ns);
1014 * kernfs_enable_ns - enable namespace under a directory
1015 * @sd: directory of interest, should be empty
1017 * This is to be called right after @sd is created to enable namespace
1018 * under it. All children of @sd must have non-NULL namespace tags and
1019 * only the ones which match the super_block's tag will be visible.
1021 void kernfs_enable_ns(struct sysfs_dirent *sd)
1023 WARN_ON_ONCE(sysfs_type(sd) != SYSFS_DIR);
1024 WARN_ON_ONCE(!RB_EMPTY_ROOT(&sd->s_dir.children));
1025 sd->s_flags |= SYSFS_FLAG_NS;
1028 /* Relationship between s_mode and the DT_xxx types */
1029 static inline unsigned char dt_type(struct sysfs_dirent *sd)
1031 return (sd->s_mode >> 12) & 15;
1034 static int sysfs_dir_release(struct inode *inode, struct file *filp)
1036 sysfs_put(filp->private_data);
1040 static struct sysfs_dirent *sysfs_dir_pos(const void *ns,
1041 struct sysfs_dirent *parent_sd, loff_t hash, struct sysfs_dirent *pos)
1044 int valid = !(pos->s_flags & SYSFS_FLAG_REMOVED) &&
1045 pos->s_parent == parent_sd &&
1046 hash == pos->s_hash;
1051 if (!pos && (hash > 1) && (hash < INT_MAX)) {
1052 struct rb_node *node = parent_sd->s_dir.children.rb_node;
1054 pos = to_sysfs_dirent(node);
1056 if (hash < pos->s_hash)
1057 node = node->rb_left;
1058 else if (hash > pos->s_hash)
1059 node = node->rb_right;
1064 /* Skip over entries in the wrong namespace */
1065 while (pos && pos->s_ns != ns) {
1066 struct rb_node *node = rb_next(&pos->s_rb);
1070 pos = to_sysfs_dirent(node);
1075 static struct sysfs_dirent *sysfs_dir_next_pos(const void *ns,
1076 struct sysfs_dirent *parent_sd, ino_t ino, struct sysfs_dirent *pos)
1078 pos = sysfs_dir_pos(ns, parent_sd, ino, pos);
1081 struct rb_node *node = rb_next(&pos->s_rb);
1085 pos = to_sysfs_dirent(node);
1086 } while (pos && pos->s_ns != ns);
1090 static int sysfs_readdir(struct file *file, struct dir_context *ctx)
1092 struct dentry *dentry = file->f_path.dentry;
1093 struct sysfs_dirent *parent_sd = dentry->d_fsdata;
1094 struct sysfs_dirent *pos = file->private_data;
1095 const void *ns = NULL;
1097 if (!dir_emit_dots(file, ctx))
1099 mutex_lock(&sysfs_mutex);
1101 if (parent_sd->s_flags & SYSFS_FLAG_NS)
1102 ns = sysfs_info(dentry->d_sb)->ns;
1104 for (pos = sysfs_dir_pos(ns, parent_sd, ctx->pos, pos);
1106 pos = sysfs_dir_next_pos(ns, parent_sd, ctx->pos, pos)) {
1107 const char *name = pos->s_name;
1108 unsigned int type = dt_type(pos);
1109 int len = strlen(name);
1110 ino_t ino = pos->s_ino;
1111 ctx->pos = pos->s_hash;
1112 file->private_data = sysfs_get(pos);
1114 mutex_unlock(&sysfs_mutex);
1115 if (!dir_emit(ctx, name, len, ino, type))
1117 mutex_lock(&sysfs_mutex);
1119 mutex_unlock(&sysfs_mutex);
1120 file->private_data = NULL;
1125 static loff_t sysfs_dir_llseek(struct file *file, loff_t offset, int whence)
1127 struct inode *inode = file_inode(file);
1130 mutex_lock(&inode->i_mutex);
1131 ret = generic_file_llseek(file, offset, whence);
1132 mutex_unlock(&inode->i_mutex);
1137 const struct file_operations sysfs_dir_operations = {
1138 .read = generic_read_dir,
1139 .iterate = sysfs_readdir,
1140 .release = sysfs_dir_release,
1141 .llseek = sysfs_dir_llseek,