Hi Ian, I am rethinking this problem. Can we simply use a global lock? In your original patch 5, you have a global mutex attr_mutex to protect attr, if we change it to a rwsem, is it enough to protect both inode and attr while having the concurrent read ability? like this patch I submitted. ( clearly, I missed __kernfs_iattrs part, but just about that idea ) https://lore.kernel.org/lkml/20201207084333.179132-1-foxhlchen@xxxxxxxxx/ thanks, fox
