iBGP Peer not updated after eBGP routing loss

[yves.haemmerli]

Hi,

I have a strange behaviour on 3 catalyst 6500 running BGP. In summary, when a VRF looses its eBGP peering with another router, it does not inform its iBGP peer of the withdrawn routes. The difficulty is that the problem does not always occur...

Here is the simplified layout :

+----------+ +----------+ +----------+ 10.56.0.0 /23

| FR01RT01 |------ iBGP ------| FR01RT02 |------ eBGP ------| CH01RT01 | 10.56.5.0 /24

+----------+ +----------+ +----------+ 10.56.6.0 /24

10.56.23.2 10.56.23.3 10.56.3.26 10.56.3.25 10.56.7.9 /24

10.56.8.0 /24

When everything works, if we loose the eBGP peering due to connection failure, the router (VRF) FR01RT02 sends the following updates to its iBGP peer FR01RT01 (debug messages) :

Sep 17 16:15:18.223 CET: %BGP-5-ADJCHANGE: neighbor 10.56.3.25 vpn vrf FR01RT02 Down BGP Notification sent

Sep 17 16:15:18.223 CET: %BGP-3-NOTIFICATION: sent to neighbor 10.56.3.25 4/0 (hold time expired) 0 bytes

Sep 17 16:15:18.231 CET: BGP(0): 10.56.23.2 send unreachable (format) 10.56.8.0/24

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 send UPDATE 10.56.8.0/24 -- unreachable

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 Format UPDATE -- unreachable :

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 send UPDATE 10.56.7.0/24 -- unreachable

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 Format UPDATE -- unreachable :

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 send UPDATE 10.56.6.0/24 -- unreachable

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 Format UPDATE -- unreachable :

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 send UPDATE 10.56.5.0/24 -- unreachable

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 Format UPDATE -- unreachable :

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 send UPDATE 10.56.0.0/23 -- unreachable

Sep 17 16:15:18.231 CET: BGP(4): 10.56.23.2 Format UPDATE -- unreachable :

Sep 17 16:15:23.827 CET: BGP(4): Import walker start version 45, end version 51

Sep 17 16:15:23.827 CET: BGP(4): ... start import cfg version = 1

After these events, the router FR01RT01 BGP table is correctly updated.

But sometimes, for example after a BGP reinitialization (clear ip bgp *), the UPDATE messages are not sent to the iBGP peer :

Sep 17 16:29:49.679 CET: %BGP-5-ADJCHANGE: neighbor 10.56.3.25 vpn vrf FR01RT02 Down BGP Notification sent

Sep 17 16:29:49.679 CET: %BGP-3-NOTIFICATION: sent to neighbor 10.56.3.25 4/0 (hold time expired) 0 bytes

Sep 17 16:29:54.003 CET: BGP(4): Import walker start version 19, end version 25

Sep 17 16:29:54.003 CET: BGP(4): ... start import cfg version = 1

--> we see that no UPDATE messages are sent to FR01RT01 !!!

Here is the IOS version : Version 12.2(33)SXH3

What could be the problem ? I suppose a bug...

Thank you for any help

Yves

[Giuseppe Larosa]

Hello Yves,

I think you are hitting the following bug

CSCsu03167 Bug Details

IPv4/v6 BGP routes not cleared when source routes is gone

It should be fixed in 12.2(33)SXH3a or better.

It is classified as 1 catastrophic.

Hope to help

Giuseppe

[yves.haemmerli]

Hello Giuseppe,

Thank you for your answer, I really appreciate. Yes, it is exactly the symptom I have. This is especially uncomfortable as the bug arrives randomly..and you know as myself, it always come at the worst moment, due to the MRCP (Murphy Rule Control Protocol) ;-)

So I will upgrade the switches,

Thanks again,

Yves