VOP_GETWRITEMOUNT() is called on the vn_start_write() path without any
vnode locks held. It's therefore unsafe to blindly access per-mount
and per-vnode data. Instead, follow the approach taken by nullfs and
use the vnode interlock coupled with the hold count to ensure the
mount and the vnode won't be recycled while they are being accessed.