I dislike this way either. We need an ability to have an access to container- if (current->fsuid == inode->i_uid)
+ if ((current->fsuid == inode->i_uid) &&
+ (current->nsproxy->user_ns == inode->i_sb->user_ns))
mode >>= 6;
I really don't think assigning a user namespace to a superblock is the
right way to go. Seems to me like the _view_ into the filesystem is
what you want to modify. That would seem to me to mean that each
'struct namespace' (filesystem namespace) or vfsmount would be assigned
a corresponding user namespace, *not* the superblock.