[c-nsp] Expose PE subnet to a vrf?

[Jeff Kell]

VRF-lite has baffled me once again.  I have what I thought would be a simple case of a vrf route leak, but I am either overlooking something obvious, or this is going to be harder than I thought.

Scenario:  Moving a "legacy" network to a new multi-VRF core.  The "new" network has the global routes, and VRFs defined for "legacy" and "management" (plus some others that aren't relevant at the moment).

The CEs are linked to the PE with trunks, each VRF having a dedicated vlan like so:

==========
interface Vlan800
 description Global VRF mesh
 ip address 10.100.1.1 255.255.252.0
!
interface Vlan810
 description Legacy network mesh
 ip vrf forwarding legacy
 ip address 10.110.1.1 255.255.252.0
!
interface Vlan820
 description Management network mesh
 ip vrf forwarding management
 ip address 10.120.1.1 255.255.252.0
==========

In order for the "management" vlan to function correctly, it needs visibility to the global vlan800.  I've checked the documentation on "route leaking" between a vrf and the global table, and the suggested method works if the advertised route is in the global table.  But it doesn't want to work if the advertised route is a *connected* subnet:

> New-Core(config)#ip route vrf management 10.100.0.0 255.255.252.0 10.100.1.1 global
> %Invalid next hop address (it's this router)

How do I advertise 10.100 into vrf management, and the return route for 10.120 in the global table?

Jeff