Cannot add IPv6 /128 loopback

pimvanpelt

I've configured a TNSR instance tnsr-v20.08-3 on a supermicro baremetal, with OSPF, OSPF6 and BGP in a small network. OSPF and OSPF6 are both forming adjacencies using a /31 and /64 transit network respectively on an interface named xe-0/2/2 (giving away what type of network I usually run):

chbtl1 tnsr# show route dynamic ospf6 neighbor 
Neighbor ID     Pri    DeadTime    State/IfState         Duration I/F[State]
194.1.163.3       1    00:00:32     Full/PointToPoint    00:20:07 xe-0/2/2[PointToPoint]

chbtl1 tnsr# show route dynamic ospf neighbor 
Neighbor ID     Pri State           Dead Time Address         Interface                        RXmtL RqstL DBsmL
194.1.163.3       1 Full/DROther      44.429s 194.1.163.17    xe-0/2/2:194.1.163.16                    0     0     0

I want iBGP sessions to be established between loopback addresses, so I've added a loopback interface loop0 and was able to successfully add an IPv4 /32 to it. That /32 is now visible in the other routers in OSPF. However, adding an IPv6 /128 to loop0 gives me an awkward error:

chbtl1 tnsr(config)# interface loop0 
chbtl1 tnsr(config-interface)# ipv6 address 2001:678:d78::4/128
chbtl1 tnsr(config-interface)# exit
Sep 29 23:52:13: Commit failed.: application operation-failed add:vpp/intf/2001:678:d78::4/128, Address in use, Address conflicts with 2001:678:d78:0:2::1/64 on xe-0/2/2

There's only two interfaces configured on the machine, that xe-0/2/2 and loop0.

chbtl1 tnsr# show  interface xe-0/2/2        
Interface: xe-0/2/2
    Admin status: up
    Link up, link-speed 10 Gbps, full duplex
    Link MTU: 9000 bytes
    MAC address: 9c:69:b4:61:a1:de
    IPv4 MTU: 0 bytes
    IPv4 Route Table: ipv4-VRF:0
    IPv4 addresses:
        194.1.163.16/31
    IPv6 MTU: 0 bytes
    IPv6 Route Table: ipv6-VRF:0
    IPv6 addresses:
        2001:678:d78:0:2::1/64
        fe80::9e69:b4ff:fe61:a1de/64
    VLAN tag rewrite: disable
    Rx-queues
        queue-id 0 : cpu-id 1
    counters:
      received: 40924902 bytes, 36747 packets, 0 errors
      transmitted: 1590880 bytes, 19596 packets, 0 errors
      protocols: 36414 IPv4, 263 IPv6
      206 drops, 0 punts, 0 rx miss, 0 rx no buffer

chbtl1 tnsr# show interface loop0   
Interface: loop0
    Admin status: up
    Link up, unknown duplex
    Link MTU: 9000 bytes
    MAC address: de:ad:00:00:00:00
    IPv4 MTU: 0 bytes
    IPv4 Route Table: ipv4-VRF:0
    IPv4 addresses:
        194.1.163.4/32
    IPv6 MTU: 0 bytes
    IPv6 Route Table: ipv6-VRF:0
    IPv6 addresses:
        fe80::dcad:ff:fe00:0/64
    VLAN tag rewrite: disable
    counters:
      received: 0 bytes, 0 packets, 0 errors
      transmitted: 5964 bytes, 70 packets, 1 errors
      protocols: 0 IPv4, 35 IPv6
      27 drops, 0 punts, 0 rx miss, 0 rx no buffer

I think my OSPF config is reasonably vanilla:

chbtl1 tnsr# show  route dynamic ospf config 
Building configuration...

Current configuration:
!
frr version 7.3.1-29~tnsr_v20.08_3
frr defaults traditional
hostname chbtl1
!
interface loop0
 ip ospf area 0.0.0.0
!
interface xe-0/2/2
 ip ospf area 0.0.0.0
 ip ospf dead-interval 45
 ip ospf hello-interval 9
 ip ospf network point-to-point
!
router ospf
 ospf router-id 194.1.163.4
!
line vty
!
end

chbtl1 tnsr# show route dynamic ospf6 config
Building configuration...

Current configuration:
!
frr version 7.3.1-29~tnsr_v20.08_3
frr defaults traditional
hostname chbtl1
!
interface loop0
 ipv6 ospf6 passive
!
interface xe-0/2/2
 ipv6 ospf6 network point-to-point
!
router ospf6
 interface loop0 area 0.0.0.0
 interface xe-0/2/2 area 0.0.0.0
!
line vty
!
end

I'm wondering what other things I could try to get this loop0 interface to show up in OSPF6. I'm afraid I'm hitting a bug, because show route table ipv6-VRF:0 definitely does not contain that IPv6 address and it's certainly not pingable (neither from the tnsr machine, nor from the rest of my network). Any clues?

pimvanpelt

Sometimes it helps to sleep on it. I found and fixed an issue with this configuration. The issue was PEBCAK. For posterity:

I had used 2001:678:d78:0:2::1/64 on xe-0/2/0 which is a supernet of 2001:678:d78::4/128. Once I changed the transit network into a /112 there was no longer overlap. This does beg a VPP question though -- why wouldn't it be OK to have a supernet and a subnet interface like this? Or, similarly, have identical IPv4 / IPv6 addresses on multiple interfaces?

I scratched my head a little while due to 2001:678:d78::/128 (another router in my network) still not being reachable (ie pinging it from this machine would reach this machine rather than the actual router), but this was because the remote side of the same 2001:678:d78:0:2::2/64 link was still using /64 erratically, so OSPF learned that, in fact, 2001:678:d78:0:2::/64 was reachable over xe-0/2/0 and because ::0 is sometimes reserved for 'the link-local gateway', so the tnsr machine responded to it.

Once I also changed the remote OSPF speaker to announce 2001:678:d78:0:2::2/112, all nodes in the IGP were reachable.

NetFreak

Had the same issue some month ago. In some cases it can be usefull to have IP space overlapping on multiple interfaces. For example if you have a routed /24 which is bound to a loopback interfaces to prevent l3 loops while only having a smaller subnet assigned to a different interface.

eg:

185.121.69.0/24 dev lo
185.121.69.0/26 dev eth0.502
...

However, the developer of TNSR are aware of this, I had a evaluation meeting longer ago where I explained this issue to them.