Hi,

coming back to Jamals idea of RFC2863 conform operative state, I've made
up a short patch on how an implementation could look like. It also takes
Tims idea of one struct workstruct per device for link state
notification (slightly modified). I know it's incomplete, it doesn't
event contain the needed fix on unregister_netdev I critised on Tim's
patch, but I want to put it to discussion early. May be it breaks the
hotplugging stuff.

If we decide to go this way, we'll also need an extension to rtnetlink
and a new ioctl to transport the information to userspace.

Stefan