Re: [RFC 2/5] sysfs-dir.patch

[viro@parcelfarce.linux.theplanet.co.uk]

On Wed, Nov 12, 2003 at 05:55:29PM +0530, Maneesh Soni wrote:
> @@ -110,10 +231,15 @@ void sysfs_remove_subdir(struct dentry *
>  void sysfs_remove_dir(struct kobject * kobj)
>  {
>  	struct list_head * node;
> -	struct dentry * dentry = dget(kobj->dentry);
> +	struct dentry * dentry = kobj->s_dirent->s_dentry;
> +	struct sysfs_dirent * parent_sd;
>  
>  	if (!dentry)
> -		return;
> +		goto exit;
> +		
> +	spin_lock(&dcache_lock);
> +	dentry = dget_locked(dentry);
> +	spin_unlock(&dcache_lock);

Racy.  Directory might've been looked up just as you've decided that it
had no dentry.

>  void sysfs_rename_dir(struct kobject * kobj, const char *new_name)
> @@ -162,14 +292,170 @@ void sysfs_rename_dir(struct kobject * k
>  	if (!kobj->parent)
>  		return;
>  
> -	parent = kobj->parent->dentry;
> +	parent = kobj->parent->s_dirent->s_dentry;
> +	if (parent) {

Ditto.

Look, the *only* benefit of ramfs as a backing store for sysfs was that we
could easily get locking right.  You want second tree - you get to fight
for coherency.