Merge bk://ldm.bkbits.net/linux-2.5-kobject

into penguin.transmeta.com:/home/penguin/torvalds/repositories/kernel/linux

Documentation/kobject.txt

+kobjects - Simple, Generic Kernel Objects
+
+Patrick Mochel <mochel@osdl.org>
+
+30 October 2002
+
+
+kobjects
+
+struct kobject introduces a simple, intregral datatype and a simple
+set of semantics for operating on the device. kobjects are intended to
+be embedded in larger data structures and replace fields it
+duplicates. A set of library functions has been developed to assist in
+the manipulation of kobjects.
+
+struct kobject looks like this:
+
+
+struct kobject {
+	char			name[16];
+	atomic_t		refcount;
+	struct list_head	entry;
+	struct kobject		* parent;
+	struct subsystem	* subsys;
+	struct dentry		* dentry;
+};
+
+void kobject_init(struct kobject *);
+int kobject_register(struct kobject *);
+void kobject_unregister(struct kobject *);
+struct kobject * kobject_get(struct kobject *);
+void kobject_put(struct kobject *);
+
+
+
+subsystems
+
+struct subsystem is introduced to describe a collection of objects of
+a certain type. subsystems are kobjects themselves, though they
+contain lists of kobjects that belong to that subsystem. Objects of a
+subsystem (the embedder objects in which kobjects live) are all of the
+same type. The interface looks like:
+
+
+struct subsystem {
+	struct kobject		kobj;
+	struct list_head	list;
+	struct rw_semaphore	rwsem;
+	struct subsystem	* parent;
+	void (*release)(struct kobject *);
+	struct sysfs_ops	* sysfs_ops;
+	struct attribute	** default_attrs;
+};
+
+void subsystem_init(struct subsystem *);
+int subsystem_register(struct subsystem *);
+void subsystem_unregister(struct subsystem *);
+
+struct subsystem * subsys_get(struct subsystem * s);
+void subsys_put(struct subsystem * s);
+
+
+
+Familial Relations
+
+kobjects and subsystems intersect and intertwine in several ways. Each
+is well-defined (though maybe they could be made simpler). Each kobject
+belongs to a subsystem. Since subsystems are kobjects themselves, they
+also belong to a controlling subsystem. This implies that subsystems
+are hierarchial. 
+
+Many kobjects are hierarchial in nature, which is represented by
+including a pointer to its parent kobject in struct kobject. Many
+different types of kobject-embedding objects may all point to the same
+parent. 
+
+The ancestral hierarchy of kobjects should not be confused with
+membership in a subsystem, or the ancestral relationship of
+subsystems. A set of kobjects may all belong to a subsystem, but all
+have different parents. 
+
+kobjects may be orphans and have no explicit parent. In that case, the
+subsystem to which the object belongs becomes its parent. 
+
+
+Sysfs
+
+These rules force a complete kobject hierarchy, which Suprise! maps
+very well onto a filesystem.
+
+driverfs was recently cloned, and there now exists sysfs. All driverfs
+operations operate on a separate data type: struct driver_dir_entry,
+which all objects that are represented in driverfs must have. driverfs
+also allowed rogue directory creation that had no explicit objects
+associated with them.
+
+struct kobject is intended to be the common data type which sysfs
+operates on. This gives the filesystem the ability to directly access
+more fields of the object, including the reference count. This also
+forces each directory in the filesystem to be tied directly to a
+kobject. 
+
+
+Directory Placement
+
+Parental relationships are determined in the kobject/subsystem layer,
+and the kobject is then passed off to the sysfs layer. kobjects with
+no parent have directories created for them in the sysfs root
+directory. Per the rules above, the only kobjects that remain orphans
+are subsystems without parent subsystems (since leaf objects either
+have an explicit parent, or are assigned their controlling subsystem
+as their foster parent). 
+
+
+File Callbacks
+
+Previously, each driverfs directory contained a pointer to a list of file
+operations for reading and writing driverfs files. These callbacks
+received a struct driver_dir_entry, when they performed a
+container_of() transform on to receive the specific object type for
+which the call was meant. 
+
+These callbacks have been converted to accept a struct kobject instead
+of struct driver_dir_entry. Since all kobjects belong to a subsystem
+that contains kobjects all of the same type, the sysfs operations
+have been moved to reside in the subsystem, since they are common for
+all kobjects.
+
+
+Default Attributes
+
+Most subsystems have a set of default attributes associated with an
+object that registers with them. A subsystem definition may contain a
+NULL-terminated array of attributes that will be exported when an
+object is registered with the subsystem. 
+
+
+Reference Counting
+
+All objects contain reference counts. All functions accessing objects
+should increment the reference count until they are finished, and
+decrement the reference count. When an object is initialized, it
+receives a reference count of 1. When a device is unregistered, the
+reference is decremented. When the reference counts reaches 0, the
+subsystem's ->release() callback for that object type (remember
+subsystems control only one type of device each) is called; and the
+reference counts of the kobject's subsystem and parent are
+decremented. 
+
+The ->release() callback is the opportunity for the subsystem to free
+memory allocated for the object. It is the notification that
+absolutely no one is using the structure any more (and can't acquire a
+reference to it), so it is safe to free it.
+

fs/sysfs/inode.c

@@ -33,7 +33,7 @@
 #include <linux/module.h>
 #include <linux/slab.h>
 #include <linux/backing-dev.h>
-#include <linux/sysfs.h>
+#include <linux/kobject.h>
 
 #include <asm/uaccess.h>
 
@@ -42,7 +42,6 @@
 
 static struct super_operations sysfs_ops;
 static struct file_operations sysfs_file_operations;
-static struct inode_operations sysfs_dir_inode_operations;
 static struct address_space_operations sysfs_aops;
 
 static struct vfsmount *sysfs_mount;
@@ -52,45 +51,6 @@ static struct backing_dev_info sysfs_backing_dev_info = {
 	.memory_backed	= 1,	/* Does not contribute to dirty memory */
 };
 
-static int sysfs_readpage(struct file *file, struct page * page)
-{
-	if (!PageUptodate(page)) {
-		void *kaddr = kmap_atomic(page, KM_USER0);
-
-		memset(kaddr, 0, PAGE_CACHE_SIZE);
-		flush_dcache_page(page);
-		kunmap_atomic(kaddr, KM_USER0);
-		SetPageUptodate(page);
-	}
-	unlock_page(page);
-	return 0;
-}
-
-static int sysfs_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
-{
-	if (!PageUptodate(page)) {
-		void *kaddr = kmap_atomic(page, KM_USER0);
-
-		memset(kaddr, 0, PAGE_CACHE_SIZE);
-		flush_dcache_page(page);
-		kunmap_atomic(kaddr, KM_USER0);
-		SetPageUptodate(page);
-	}
-	return 0;
-}
-
-static int sysfs_commit_write(struct file *file, struct page *page, unsigned offset, unsigned to)
-{
-	struct inode *inode = page->mapping->host;
-	loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
-
-	set_page_dirty(page);
-	if (pos > inode->i_size)
-		inode->i_size = pos;
-	return 0;
-}
-
-
 static struct inode *sysfs_get_inode(struct super_block *sb, int mode, int dev)
 {
 	struct inode *inode = new_inode(sb);
@@ -113,7 +73,7 @@ static struct inode *sysfs_get_inode(struct super_block *sb, int mode, int dev)
 			inode->i_fop = &sysfs_file_operations;
 			break;
 		case S_IFDIR:
-			inode->i_op = &sysfs_dir_inode_operations;
+			inode->i_op = &simple_dir_inode_operations;
 			inode->i_fop = &simple_dir_operations;
 
 			/* directory inodes start off with i_nlink == 2 (for "." entry) */
@@ -182,29 +142,6 @@ static int sysfs_symlink(struct inode * dir, struct dentry *dentry, const char *
 	return error;
 }
 
-static inline int sysfs_positive(struct dentry *dentry)
-{
-	return (dentry->d_inode && !d_unhashed(dentry));
-}
-
-static int sysfs_empty(struct dentry *dentry)
-{
-	struct list_head *list;
-
-	spin_lock(&dcache_lock);
-
-	list_for_each(list, &dentry->d_subdirs) {
-		struct dentry *de = list_entry(list, struct dentry, d_child);
-		if (sysfs_positive(de)) {
-			spin_unlock(&dcache_lock);
-			return 0;
-		}
-	}
-
-	spin_unlock(&dcache_lock);
-	return 1;
-}
-
 static int sysfs_unlink(struct inode *dir, struct dentry *dentry)
 {
 	struct inode *inode = dentry->d_inode;
@@ -217,30 +154,34 @@ static int sysfs_unlink(struct inode *dir, struct dentry *dentry)
 }
 
 /**
- * sysfs_read_file - "read" data from a file.
- * @file:	file pointer
- * @buf:	buffer to fill
- * @count:	number of bytes to read
- * @ppos:	starting offset in file
+ *	sysfs_read_file - read an attribute. 
+ *	@file:	file pointer.
+ *	@buf:	buffer to fill.
+ *	@count:	number of bytes to read.
+ *	@ppos:	starting offset in file.
+ *
+ *	Userspace wants to read an attribute file. The attribute descriptor
+ *	is in the file's ->d_fsdata. The target object is in the directory's
+ *	->d_fsdata. 
  *
- * Userspace wants data from a file. It is up to the creator of the file to
- * provide that data.
- * There is a struct device_attribute embedded in file->private_data. We
- * obtain that and check if the read callback is implemented. If so, we call
- * it, passing the data field of the file entry.
- * Said callback is responsible for filling the buffer and returning the number
- * of bytes it put in it. We update @ppos correctly.
+ *	We allocate a %PAGE_SIZE buffer, and pass it to the object's ->show()
+ *	method (along with the object). We loop doing this until @count is 
+ *	satisfied, or ->show() returns %0. 
  */
+
 static ssize_t
 sysfs_read_file(struct file *file, char *buf, size_t count, loff_t *ppos)
 {
 	struct attribute * attr = file->f_dentry->d_fsdata;
-	struct driver_dir_entry * dir;
+	struct sysfs_ops * ops = NULL;
+	struct kobject * kobj;
 	unsigned char *page;
 	ssize_t retval = 0;
 
-	dir = file->f_dentry->d_parent->d_fsdata;
-	if (!dir->ops->show)
+	kobj = file->f_dentry->d_parent->d_fsdata;
+	if (kobj && kobj->subsys)
+		ops = kobj->subsys->sysfs_ops;
+	if (!ops || !ops->show)
 		return 0;
 
 	if (count > PAGE_SIZE)
@@ -253,7 +194,7 @@ sysfs_read_file(struct file *file, char *buf, size_t count, loff_t *ppos)
 	while (count > 0) {
 		ssize_t len;
 
-		len = dir->ops->show(dir,attr,page,count,*ppos);
+		len = ops->show(kobj,attr,page,count,*ppos);
 
 		if (len <= 0) {
 			if (len < 0)
@@ -277,27 +218,32 @@ sysfs_read_file(struct file *file, char *buf, size_t count, loff_t *ppos)
 }
 
 /**
- * sysfs_write_file - "write" to a file
- * @file:	file pointer
- * @buf:	data to write
- * @count:	number of bytes
- * @ppos:	starting offset
+ *	sysfs_write_file - write an attribute.
+ *	@file:	file pointer
+ *	@buf:	data to write
+ *	@count:	number of bytes
+ *	@ppos:	starting offset
  *
- * Similarly to sysfs_read_file, we act essentially as a bit pipe.
- * We check for a "write" callback in file->private_data, and pass
- * @buffer, @count, @ppos, and the file entry's data to the callback.
- * The number of bytes written is returned, and we handle updating
- * @ppos properly.
+ *	Identical to sysfs_read_file(), though going the opposite direction.
+ *	We allocate a %PAGE_SIZE buffer and copy in the userspace buffer. We
+ *	pass that to the object's ->store() method until we reach @count or 
+ *	->store() returns %0 or less.
  */
+
 static ssize_t
 sysfs_write_file(struct file *file, const char *buf, size_t count, loff_t *ppos)
 {
 	struct attribute * attr = file->f_dentry->d_fsdata;
-	struct driver_dir_entry * dir;
+	struct sysfs_ops * ops = NULL;
+	struct kobject * kobj;
 	ssize_t retval = 0;
 	char * page;
 
-	dir = file->f_dentry->d_parent->d_fsdata;
+	kobj = file->f_dentry->d_parent->d_fsdata;
+	if (kobj && kobj->subsys)
+		ops = kobj->subsys->sysfs_ops;
+	if (!ops || !ops->store)
+		return 0;
 
 	page = (char *)__get_free_page(GFP_KERNEL);
 	if (!page)
@@ -312,7 +258,7 @@ sysfs_write_file(struct file *file, const char *buf, size_t count, loff_t *ppos)
 	while (count > 0) {
 		ssize_t len;
 
-		len = dir->ops->store(dir,attr,page + retval,count,*ppos);
+		len = ops->store(kobj,attr,page + retval,count,*ppos);
 
 		if (len <= 0) {
 			if (len < 0)
@@ -329,77 +275,42 @@ sysfs_write_file(struct file *file, const char *buf, size_t count, loff_t *ppos)
 	return retval;
 }
 
-static loff_t
-sysfs_file_lseek(struct file *file, loff_t offset, int orig)
-{
-	loff_t retval = -EINVAL;
-
-	down(&file->f_dentry->d_inode->i_sem);
-	switch(orig) {
-	case 0:
-		if (offset > 0) {
-			file->f_pos = offset;
-			retval = file->f_pos;
-		}
-		break;
-	case 1:
-		if ((offset + file->f_pos) > 0) {
-			file->f_pos += offset;
-			retval = file->f_pos;
-		}
-		break;
-	default:
-		break;
-	}
-	up(&file->f_dentry->d_inode->i_sem);
-	return retval;
-}
-
 static int sysfs_open_file(struct inode * inode, struct file * filp)
 {
-	struct driver_dir_entry * dir;
+	struct kobject * kobj;
 	int error = 0;
 
-	dir = (struct driver_dir_entry *)filp->f_dentry->d_parent->d_fsdata;
-	if (dir) {
+	kobj = filp->f_dentry->d_parent->d_fsdata;
+	if ((kobj = kobject_get(kobj))) {
 		struct attribute * attr = filp->f_dentry->d_fsdata;
-		if (attr && dir->ops) {
-			if (dir->ops->open)
-				error = dir->ops->open(dir);
-			goto Done;
-		}
-	}
-	error = -EINVAL;
- Done:
+		if (!attr)
+			error = -EINVAL;
+	} else
+		error = -EINVAL;
 	return error;
 }
 
 static int sysfs_release(struct inode * inode, struct file * filp)
 {
-	struct driver_dir_entry * dir;
-	dir = (struct driver_dir_entry *)filp->f_dentry->d_parent->d_fsdata;
-	if (dir->ops->close)
-		dir->ops->close(dir);
+	struct kobject * kobj = filp->f_dentry->d_parent->d_fsdata;
+	if (kobj) 
+		kobject_put(kobj);
 	return 0;
 }
 
 static struct file_operations sysfs_file_operations = {
 	.read		= sysfs_read_file,
 	.write		= sysfs_write_file,
-	.llseek		= sysfs_file_lseek,
+	.llseek		= generic_file_llseek,
 	.open		= sysfs_open_file,
 	.release	= sysfs_release,
 };
 
-static struct inode_operations sysfs_dir_inode_operations = {
-	.lookup		= simple_lookup,
-};
-
 static struct address_space_operations sysfs_aops = {
-	.readpage	= sysfs_readpage,
+	.readpage	= simple_readpage,
 	.writepage	= fail_writepage,
-	.prepare_write	= sysfs_prepare_write,
-	.commit_write	= sysfs_commit_write
+	.prepare_write	= simple_prepare_write,
+	.commit_write	= simple_commit_write
 };
 
 static struct super_operations sysfs_ops = {
@@ -464,6 +375,7 @@ static int __init sysfs_init(void)
 
 core_initcall(sysfs_init);
 
+
 static struct dentry * get_dentry(struct dentry * parent, const char * name)
 {
 	struct qstr qstr;
@@ -474,137 +386,207 @@ static struct dentry * get_dentry(struct dentry * parent, const char * name)
 	return lookup_hash(&qstr,parent);
 }
 
+
 /**
- * sysfs_create_dir - create a directory in the filesystem
- * @entry:	directory entry
- * @parent:	parent directory entry
+ *	sysfs_create_dir - create a directory for an object.
+ *	@parent:	parent parent object.
+ *	@kobj:		object we're creating directory for. 
  */
-int
-sysfs_create_dir(struct driver_dir_entry * entry,
-		    struct driver_dir_entry * parent)
+
+int sysfs_create_dir(struct kobject * kobj)
 {
 	struct dentry * dentry = NULL;
-	struct dentry * parent_dentry;
+	struct dentry * parent;
 	int error = 0;
 
-	if (!entry)
+	if (!kobj)
 		return -EINVAL;
 
-	parent_dentry = parent ? parent->dentry : NULL;
-
-	if (!parent_dentry)
-		if (sysfs_mount && sysfs_mount->mnt_sb)
-			parent_dentry = sysfs_mount->mnt_sb->s_root;
-
-	if (!parent_dentry)
+	if (kobj->parent)
+		parent = kobj->parent->dentry;
+	else if (sysfs_mount && sysfs_mount->mnt_sb)
+		parent = sysfs_mount->mnt_sb->s_root;
+	else
 		return -EFAULT;
 
-	down(&parent_dentry->d_inode->i_sem);
-	dentry = get_dentry(parent_dentry,entry->name);
+	down(&parent->d_inode->i_sem);
+	dentry = get_dentry(parent,kobj->name);
 	if (!IS_ERR(dentry)) {
-		dentry->d_fsdata = (void *) entry;
-		entry->dentry = dentry;
-		error = sysfs_mkdir(parent_dentry->d_inode,dentry,entry->mode);
+		dentry->d_fsdata = (void *)kobj;
+		kobj->dentry = dentry;
+		error = sysfs_mkdir(parent->d_inode,dentry,
+				    (S_IFDIR| S_IRWXU | S_IRUGO | S_IXUGO));
 	} else
 		error = PTR_ERR(dentry);
-	up(&parent_dentry->d_inode->i_sem);
+	up(&parent->d_inode->i_sem);
 
 	return error;
 }
 
+
 /**
- * sysfs_create_file - create a file
- * @entry:	structure describing the file
- * @parent:	directory to create it in
+ *	sysfs_create_file - create an attribute file for an object.
+ *	@kobj:	object we're creating for. 
+ *	@attr:	atrribute descriptor.
  */
-int
-sysfs_create_file(struct attribute * entry,
-		     struct driver_dir_entry * parent)
+
+int sysfs_create_file(struct kobject * kobj, struct attribute * attr)
 {
 	struct dentry * dentry;
+	struct dentry * parent;
 	int error = 0;
 
-	if (!entry || !parent)
+	if (!kobj || !attr)
 		return -EINVAL;
 
-	if (!parent->dentry)
-		return -EINVAL;
+	parent = kobj->dentry;
 
-	down(&parent->dentry->d_inode->i_sem);
-	dentry = get_dentry(parent->dentry,entry->name);
+	down(&parent->d_inode->i_sem);
+	dentry = get_dentry(parent,attr->name);
 	if (!IS_ERR(dentry)) {
-		dentry->d_fsdata = (void *)entry;
-		error = sysfs_create(parent->dentry->d_inode,dentry,entry->mode);
+		dentry->d_fsdata = (void *)attr;
+		error = sysfs_create(parent->d_inode,dentry,attr->mode);
 	} else
 		error = PTR_ERR(dentry);
-	up(&parent->dentry->d_inode->i_sem);
+	up(&parent->d_inode->i_sem);
 	return error;
 }
 
+
+static int object_depth(struct kobject * kobj)
+{
+	struct kobject * p = kobj;
+	int depth = 0;
+	do { depth++; } while ((p = p->parent));
+	return depth;
+}
+
+static int object_path_length(struct kobject * kobj)
+{
+	struct kobject * p = kobj;
+	int length = 1;
+	do {
+		length += strlen(p->name) + 1;
+		p = p->parent;
+	} while (p);
+	return length;
+}
+
+static void fill_object_path(struct kobject * kobj, char * buffer, int length)
+{
+	struct kobject * p;
+
+	--length;
+	for (p = kobj; p; p = p->parent) {
+		int cur = strlen(p->name);
+
+		/* back up enough to print this bus id with '/' */
+		length -= cur;
+		strncpy(buffer + length,p->name,cur);
+		*(buffer + --length) = '/';
+	}
+}
+
 /**
- * sysfs_create_symlink - make a symlink
- * @parent:	directory we're creating in 
- * @entry:	entry describing link
- * @target:	place we're symlinking to
- * 
+ *	sysfs_create_link - create symlink between two objects.
+ *	@kobj:	object whose directory we're creating the link in.
+ *	@target:	object we're pointing to.
+ *	@name:		name of the symlink.
  */
-int sysfs_create_symlink(struct driver_dir_entry * parent, 
-			    char * name, char * target)
+int sysfs_create_link(struct kobject * kobj, struct kobject * target, char * name)
 {
-	struct dentry * dentry;
+	struct dentry * dentry = kobj->dentry;
+	struct dentry * d;
 	int error = 0;
+	int size;
+	int depth;
+	char * path;
+	char * s;
+
+	depth = object_depth(kobj);
+	size = object_path_length(target) + depth * 3 - 1;
+	if (size > PATH_MAX)
+		return -ENAMETOOLONG;
+	pr_debug("%s: depth = %d, size = %d\n",__FUNCTION__,depth,size);
+
+	path = kmalloc(size,GFP_KERNEL);
+	if (!path)
+		return -ENOMEM;
+	memset(path,0,size);
 
-	if (!parent || !parent->dentry)
-		return -EINVAL;
+	for (s = path; depth--; s += 3)
+		strcpy(s,"../");
+
+	fill_object_path(target,path,size);
+	pr_debug("%s: path = '%s'\n",__FUNCTION__,path);
 
-	down(&parent->dentry->d_inode->i_sem);
-	dentry = get_dentry(parent->dentry,name);
-	if (!IS_ERR(dentry))
-		error = sysfs_symlink(parent->dentry->d_inode,dentry,target);
+	down(&dentry->d_inode->i_sem);
+	d = get_dentry(dentry,name);
+	if (!IS_ERR(d))
+		error = sysfs_symlink(dentry->d_inode,d,path);
 	else
-		error = PTR_ERR(dentry);
-	up(&parent->dentry->d_inode->i_sem);
+		error = PTR_ERR(d);
+	up(&dentry->d_inode->i_sem);
+	kfree(path);
 	return error;
 }
 
+static void hash_and_remove(struct dentry * dir, const char * name)
+{
+	struct dentry * victim;
+
+	down(&dir->d_inode->i_sem);
+	victim = get_dentry(dir,name);
+	if (!IS_ERR(victim)) {
+		/* make sure dentry is really there */
+		if (victim->d_inode && 
+		    (victim->d_parent->d_inode == dir->d_inode)) {
+			sysfs_unlink(dir->d_inode,victim);
+		}
+	}
+	up(&dir->d_inode->i_sem);
+}
+
+
 /**
- * sysfs_remove_file - exported file removal
- * @dir:	directory the file supposedly resides in
- * @name:	name of the file
+ *	sysfs_remove_file - remove an object attribute.
+ *	@kobj:	object we're acting for.
+ *	@attr:	attribute descriptor.
  *
- * Try and find the file in the dir's list.
- * If it's there, call __remove_file() (above) for the dentry.
+ *	Hash the attribute name and kill the victim.
  */
-void sysfs_remove_file(struct driver_dir_entry * dir, const char * name)
+
+void sysfs_remove_file(struct kobject * kobj, struct attribute * attr)
 {
-	struct dentry * dentry;
+	hash_and_remove(kobj->dentry,attr->name);
+}
 
-	if (!dir->dentry)
-		return;
 
-	down(&dir->dentry->d_inode->i_sem);
-	dentry = get_dentry(dir->dentry,name);
-	if (!IS_ERR(dentry)) {
-		/* make sure dentry is really there */
-		if (dentry->d_inode && 
-		    (dentry->d_parent->d_inode == dir->dentry->d_inode)) {
-			sysfs_unlink(dir->dentry->d_inode,dentry);
-		}
-	}
-	up(&dir->dentry->d_inode->i_sem);
+/**
+ *	sysfs_remove_link - remove symlink in object's directory.
+ *	@kobj:	object we're acting for.
+ *	@name:	name of the symlink to remove.
+ */
+
+void sysfs_remove_link(struct kobject * kobj, char * name)
+{
+	hash_and_remove(kobj->dentry,name);
 }
 
+
 /**
- * sysfs_remove_dir - exportable directory removal
- * @dir:	directory to remove
+ *	sysfs_remove_dir - remove an object's directory.
+ *	@kobj:	object. 
  *
- * To make sure we don't orphan anyone, first remove
- * all the children in the list, then do clean up the directory.
+ *	The only thing special about this is that we remove any files in 
+ *	the directory before we remove the directory, and we've inlined
+ *	what used to be sysfs_rmdir() below, instead of calling separately.
  */
-void sysfs_remove_dir(struct driver_dir_entry * dir)
+
+void sysfs_remove_dir(struct kobject * kobj)
 {
 	struct list_head * node, * next;
-	struct dentry * dentry = dir->dentry;
+	struct dentry * dentry = kobj->dentry;
 	struct dentry * parent;
 
 	if (!dentry)
@@ -622,7 +604,7 @@ void sysfs_remove_dir(struct driver_dir_entry * dir)
 	}
 
 	d_invalidate(dentry);
-	if (sysfs_empty(dentry)) {
+	if (simple_empty(dentry)) {
 		dentry->d_inode->i_nlink -= 2;
 		dentry->d_inode->i_flags |= S_DEAD;
 		parent->d_inode->i_nlink--;
@@ -635,8 +617,9 @@ void sysfs_remove_dir(struct driver_dir_entry * dir)
 }
 
 EXPORT_SYMBOL(sysfs_create_file);
-EXPORT_SYMBOL(sysfs_create_symlink);
-EXPORT_SYMBOL(sysfs_create_dir);
 EXPORT_SYMBOL(sysfs_remove_file);
+EXPORT_SYMBOL(sysfs_create_link);
+EXPORT_SYMBOL(sysfs_remove_link);
+EXPORT_SYMBOL(sysfs_create_dir);
 EXPORT_SYMBOL(sysfs_remove_dir);
 MODULE_LICENSE("GPL");

include/linux/kobject.h

+/*
+ * kobject.h - generic kernel object infrastructure.
+ *
+ */
+
+#ifndef _KOBJECT_H_
+#define _KOBJECT_H_
+
+#include <linux/types.h>
+#include <linux/list.h>
+#include <linux/sysfs.h>
+#include <linux/rwsem.h>
+#include <asm/atomic.h>
+
+struct kobject {
+	char			name[16];
+	atomic_t		refcount;
+	struct list_head	entry;
+	struct kobject		* parent;
+	struct subsystem	* subsys;
+	struct dentry		* dentry;
+};
+
+extern void kobject_init(struct kobject *);
+
+extern int kobject_register(struct kobject *);
+extern void kobject_unregister(struct kobject *);
+
+extern struct kobject * kobject_get(struct kobject *);
+extern void kobject_put(struct kobject *);
+
+
+struct subsystem {
+	struct kobject		kobj;
+	struct list_head	list;
+	struct rw_semaphore	rwsem;
+	struct subsystem	* parent;
+	void (*release)(struct kobject *);
+	struct sysfs_ops	* sysfs_ops;
+	struct attribute	** default_attrs;
+};
+
+extern void subsystem_init(struct subsystem *);
+extern int subsystem_register(struct subsystem *);
+extern void subsystem_unregister(struct subsystem *);
+
+static inline struct subsystem * subsys_get(struct subsystem * s)
+{
+	return container_of(kobject_get(&s->kobj),struct subsystem,kobj);
+}
+
+static inline void subsys_put(struct subsystem * s)
+{
+	kobject_put(&s->kobj);
+}
+
+#endif /* _KOBJECT_H_ */

include/linux/sysfs.h

@@ -11,19 +11,11 @@
 
 struct driver_dir_entry;
 struct attribute;
+struct kobject;
 
 struct sysfs_ops {
-	int	(*open)(struct driver_dir_entry *);
-	int	(*close)(struct driver_dir_entry *);
-	ssize_t	(*show)(struct driver_dir_entry *, struct attribute *,char *, size_t, loff_t);
-	ssize_t	(*store)(struct driver_dir_entry *,struct attribute *,const char *, size_t, loff_t);
-};
-
-struct driver_dir_entry {
-	char			* name;
-	struct dentry		* dentry;
-	mode_t			mode;
-	struct sysfs_ops	* ops;
+	ssize_t	(*show)(struct kobject *, struct attribute *,char *, size_t, loff_t);
+	ssize_t	(*store)(struct kobject *,struct attribute *,const char *, size_t, loff_t);
 };
 
 struct attribute {
@@ -32,20 +24,21 @@ struct attribute {
 };
 
 extern int
-sysfs_create_dir(struct driver_dir_entry *, struct driver_dir_entry *);
+sysfs_create_dir(struct kobject *);
 
 extern void
-sysfs_remove_dir(struct driver_dir_entry * entry);
+sysfs_remove_dir(struct kobject *);
 
 extern int
-sysfs_create_file(struct attribute * attr,
-		     struct driver_dir_entry * parent);
+sysfs_create_file(struct kobject *, struct attribute *);
+
+extern void
+sysfs_remove_file(struct kobject *, struct attribute *);
 
 extern int 
-sysfs_create_symlink(struct driver_dir_entry * parent, 
-			char * name, char * target);
+sysfs_create_link(struct kobject * kobj, struct kobject * target, char * name);
 
 extern void
-sysfs_remove_file(struct driver_dir_entry *, const char * name);
+sysfs_remove_link(struct kobject *, char * name);
 
 #endif /* _SYSFS_H_ */

lib/Makefile

@@ -9,10 +9,11 @@
 L_TARGET := lib.a
 
 export-objs := cmdline.o dec_and_lock.o rwsem-spinlock.o rwsem.o \
-	       crc32.o rbtree.o radix-tree.o
+	       crc32.o rbtree.o radix-tree.o kobject.o
 
 obj-y := errno.o ctype.o string.o vsprintf.o brlock.o cmdline.o \
-	 bust_spinlocks.o rbtree.o radix-tree.o dump_stack.o
+	 bust_spinlocks.o rbtree.o radix-tree.o dump_stack.o \
+	 kobject.o
 
 obj-$(CONFIG_RWSEM_GENERIC_SPINLOCK) += rwsem-spinlock.o
 obj-$(CONFIG_RWSEM_XCHGADD_ALGORITHM) += rwsem.o

lib/kobject.c

+/*
+ * kobject.c - library routines for handling generic kernel objects
+ */
+
+#define DEBUG 1
+
+#include <linux/kobject.h>
+#include <linux/module.h>
+#include <linux/stat.h>
+
+/**
+ *	kobject_populate_dir - populate directory with attributes.
+ *	@kobj:	object we're working on.
+ *
+ *	Most subsystems have a set of default attributes that 
+ *	are associated with an object that registers with them.
+ *	This is a helper called during object registration that 
+ *	loops through the default attributes of the subsystem 
+ *	and creates attributes files for them in sysfs.
+ *
+ */
+
+static int kobject_populate_dir(struct kobject * kobj)
+{
+	struct subsystem * s = kobj->subsys;
+	struct attribute * attr;
+	int error = 0;
+	int i;
+	
+	if (s && s->default_attrs) {
+		for (i = 0; (attr = s->default_attrs[i]); i++) {
+			if ((error = sysfs_create_file(kobj,attr)))
+				break;
+		}
+	}
+	return error;
+}
+
+/**
+ *	kobject_init - initialize object.
+ *	@kobj:	object in question.
+ */
+
+void kobject_init(struct kobject * kobj)
+{
+	atomic_set(&kobj->refcount,1);
+	INIT_LIST_HEAD(&kobj->entry);
+}
+
+/**
+ *	kobject_register - register an object.
+ *	@kobj:	object in question.
+ *
+ *	For now, fill in the replicated fields in the object's
+ *	directory entry, and create a dir in sysfs. 
+ *	This stuff should go away in the future, as we move
+ *	more implicit things to sysfs.
+ */
+
+int kobject_register(struct kobject * kobj)
+{
+	int error = 0;
+	struct subsystem * s = subsys_get(kobj->subsys);
+	struct kobject * parent = kobject_get(kobj->parent);
+
+	pr_debug("kobject %s: registering\n",kobj->name);
+	if (parent)
+		pr_debug("  parent is %s\n",parent->name);
+	if (s) {
+		down_write(&s->rwsem);
+		if (parent) 
+			list_add_tail(&kobj->entry,&parent->entry);
+		else {
+			list_add_tail(&kobj->entry,&s->list);
+			kobj->parent = &s->kobj;
+		}
+		up_write(&s->rwsem);
+	}
+	error = sysfs_create_dir(kobj);
+	if (!error) {
+		error = kobject_populate_dir(kobj);
+		if (error)
+			sysfs_remove_dir(kobj);
+	}
+	return error;
+}
+
+/**
+ *	kobject_unregister - unlink an object.
+ *	@kobj:	object going away.
+ *
+ *	The device has been told to be removed, but may
+ *	not necessarily be disappearing from the kernel.
+ *	So, we remove the directory and decrement the refcount
+ *	that we set with kobject_register().
+ *
+ *	Eventually (maybe now), the refcount will hit 0, and 
+ *	put_device() will clean the device up.
+ */
+
+void kobject_unregister(struct kobject * kobj)
+{
+	pr_debug("kobject %s: unregistering\n",kobj->name);
+	sysfs_remove_dir(kobj);
+	if (kobj->subsys) {
+		down_write(&kobj->subsys->rwsem);
+		list_del_init(&kobj->entry);
+		up_write(&kobj->subsys->rwsem);
+	}
+	kobject_put(kobj);
+}
+
+/**
+ *	kobject_get - increment refcount for object.
+ *	@kobj:	object.
+ */
+
+struct kobject * kobject_get(struct kobject * kobj)
+{
+	struct kobject * ret = kobj;
+	if (kobj && atomic_read(&kobj->refcount) > 0)
+		atomic_inc(&kobj->refcount);
+	else
+		ret = NULL;
+	return ret;
+}
+
+/**
+ *	kobject_put - decrement refcount for object.
+ *	@kobj:	object.
+ *
+ *	Decrement the refcount, and check if 0. If it is, then 
+ *	we're gonna need to clean it up, and decrement the refcount
+ *	of its parent.
+ *
+ *	@kobj->parent could point to its subsystem, which we also 
+ *	want to decrement the reference count for. We always dec 
+ *	the refcount for the parent, but only do so for the subsystem
+ *	if it points to a different place than the parent.
+ */
+
+void kobject_put(struct kobject * kobj)
+{
+	struct kobject * parent = kobj->parent;
+	struct subsystem * s = kobj->subsys;
+
+	if (!atomic_dec_and_test(&kobj->refcount))
+		return;
+
+	pr_debug("kobject %s: cleaning up\n",kobj->name);
+	if (s) {
+		if (s->release)
+			s->release(kobj);
+		if (&s->kobj != parent)
+			subsys_put(s);
+	} 
+
+	if (parent) 
+		kobject_put(parent);
+}
+
+
+void subsystem_init(struct subsystem * s)
+{
+	kobject_init(&s->kobj);
+	init_rwsem(&s->rwsem);
+	INIT_LIST_HEAD(&s->list);
+}
+
+/**
+ *	subsystem_register - register a subsystem.
+ *	@s:	the subsystem we're registering.
+ */
+
+int subsystem_register(struct subsystem * s)
+{
+	subsystem_init(s);
+	if (s->parent)
+		s->kobj.parent = &s->parent->kobj;
+	pr_debug("subsystem %s: registering\n",s->kobj.name);
+	if (s->parent)
+		pr_debug("  parent is %s\n",s->parent->kobj.name);
+	return kobject_register(&s->kobj);
+}
+
+void subsystem_unregister(struct subsystem * s)
+{
+	pr_debug("subsystem %s: unregistering\n",s->kobj.name);
+	kobject_unregister(&s->kobj);
+}
+
+
+EXPORT_SYMBOL(kobject_init);
+EXPORT_SYMBOL(kobject_register);
+EXPORT_SYMBOL(kobject_unregister);
+EXPORT_SYMBOL(kobject_get);
+EXPORT_SYMBOL(kobject_put);
+
+EXPORT_SYMBOL(subsystem_init);
+EXPORT_SYMBOL(subsystem_register);
+EXPORT_SYMBOL(subsystem_unregister);