[c-nsp] IPv6, neighbor detection, BGP and my nerves...

[Michael K. Smith - Adhost]

Hello Elmar:

> -----Original Message-----
> From: cisco-nsp-bounces at puck.nether.net [mailto:cisco-nsp-
> bounces at puck.nether.net] On Behalf Of Elmar K. Bins
> Sent: Thursday, March 04, 2010 5:57 AM
> To: cisco-nsp at puck.nether.net
> Subject: [c-nsp] IPv6, neighbor detection, BGP and my nerves...
> 
> Hello everybody,
> 
> here's the setup:
> 
> Router model is 7301+PA-2FE-TX with IOS 12.3(14)T3. For those
> who shout "upgrade!" already - same problem with 12.4s.
> 
> Router is connected to HKIX via FastEthernet1/1.
> 
> No problems whatsoever in IPv4, but here's what I have in IPv6:
> 
>   1. IPv6 BGP saves routes with a prefix source of <official
>      IPv6 address>, but stores the next hop out of FE80::/8.
> 
> [ipv6 bgp]
> BGP routing table entry for 2001:4F8::/32, version 42465
> ...
>   6939 1280
>     2001:7FA:0:1::CA28:A19E (FE80::212:F2FF:FE9C:2150) from
> 2001:7FA:0:1::CA28:A19E (216.218.252.180)
>       Origin IGP, localpref 100, valid, external, best
> 
> [ipv6 route]
> B   2001:4F8::/32 [20/0]
>      via FE80::212:F2FF:FE9C:2150, FastEthernet1/1
> 
> 
>   2. The trouble is that the router does not "see" some of the
>      FE80::/8 addresses most of the time, meaning that IPv6 nd
>      does not resolve (router gets no answer) and leaves "INCMP"
>      entries in the neighbor table.
> 
>      At other times (quite rarely), the FE80:: address is reachable
>      and IPv6 ND resolves.
> 
>      The official IP address of the partner is always "visible".
> 
> [ipv6 nei]
> 
> 2001:7FA:0:1::CA28:A19E                     0 0012.f29c.2150  REACH
> Fa1/1
> FE80::212:F2FF:FE9C:2150                    0 -               INCMP
> Fa1/1
> 
> 
> Clearing the ipv6 neighbor table does not help...
> 
> Can anyone help me with these following questions:
> 
>   Towards 1. above - can I keep the FE80 addresses out of the equation
>   entirely and have BGP save the neighbor address as the next hop?
> 
>   Towards 2. above - has anyone seen same behaviour and/or have a
>   - workaround
>   - lead to a bug id?
> 
> 
> Any help is much appreciated...
> 
You might want to modify your interface configuration to match the one
below.  You shouldn't need neighbor discovery and the like for a direct
peering connection.  This is the configuration I use at the SIX.

interface GigabitEthernet0/0
 description SIX IP Uplink
 ipv6 address <v6 address>/64
 no ipv6 redirects
 ipv6 nd suppress-ra
 no ipv6 mld router
 no ipv6 pim

Regards,

Mike