[c-nsp] Cisco 7600, bgp neighbor default-originate breaks

[Christian Bering]

Hi Rodney,

>For the recreate you have, great job btw, can you get the debugs on
>the PE when you do the clear?

I was unable to recreate the exact behaviour in my lab this time. However, I was able to create a different situation in which the default also got withdrawn from the CE.

Another PE (BR1) was injecting a default into the VRF ('default-information originate'). On doing 'clear ip bgp *' on BR1, the PE mentioned before (DR2) saw a withdrawel of the default (as expected):

DR2#
Mar 10 17:38:34.751: BGP(4): 10.10.2.1 rcv UPDATE about 10:10:0.0.0.0/0 -- withdrawn, label 524288
Mar 10 17:38:34.751: BGP(4): no valid path for 10:10:0.0.0.0/0
Mar 10 17:38:34.751: BGP(4): nettable_walker 10:10:0.0.0.0/0 no best path
Mar 10 17:38:34.751: BGP: topo Test:VPNv4 Unicast:base Remove_fwdroute for 10:10:0.0.0.0/0
Mar 10 17:38:34.775: BGP(4): updgrp 2 - 83.136.89.14 updates replicated for neighbors: 83.136.89.6

This was carried over to the CE (not expected):

CPE3:
Mar 10 17:38:34.798: BGP(0): 83.136.89.13 rcv UPDATE about 0.0.0.0/0 -- withdrawn
Mar 10 17:38:34.798: BGP(0): no valid path for 0.0.0.0/0
Mar 10 17:38:34.798: BGP(0): nettable_walker 0.0.0.0/0 no best path
Mar 10 17:38:34.798: BGP: topo global:IPv4 Unicast:base Remove_fwdroute for 0.0.0.0/0

CPE3#show ip bgp
BGP table version is 45, local router ID is 83.136.89.14
Status codes: s suppressed, d damped, h history, * valid, > best, i - internal,
              r RIB-failure, S Stale
Origin codes: i - IGP, e - EGP, ? - incomplete

   Network          Next Hop            Metric LocPrf Weight Path
*> 93.0.0.0/24      0.0.0.0                  0         32768 i



More than a minute later, the default was once again advertised to the CE during the PE's 'import walker':

DR2#
Mar 10 17:39:42.471: BGP(4): Import walker start version 690, end version 697
Mar 10 17:39:42.475: BGP(4): ... start import cfg version = 0
Mar 10 17:39:42.479: BGP(4): Revise route installing 1 of 1 routes for 0.0.0.0/0 -> 10.10.3.1(Test) to Test IP table
Mar 10 17:39:42.519: BGP(4): updgrp 2 - 83.136.89.14 updates replicated for neighbors: 83.136.89.6


CPE3#
Mar 10 17:39:42.493: BGP(0): 83.136.89.13 rcvd UPDATE w/ attr: nexthop 83.136.89.13, origin ?, path 10
Mar 10 17:39:42.497: BGP(0): 83.136.89.13 rcvd 0.0.0.0/0
Mar 10 17:39:42.509: BGP(0): Revise route installing 1 of 1 routes for 0.0.0.0/0 -> 83.136.89.13(global) to main IP table

It must be a bug that the PE withdraws the default from the CE. I realize my latest lab setup is different from my original one but the issue is basically the same: the PE _must not_ withdraw a default from a CE when it's been configured with 'neighbor default-originate'. In my lab, it does. I've seen it happen in three different scenarios now (one of them in a live network):

1) When "default-information originate" was added then removed from the VRF (on a "remote" PE)
2) When a prefix-list was applied outgoing on the PE and the BGP session to the CE was soft cleared
3) When the "remote" PE having the "default-information originate" had all its BGP sessions hard cleared

Ad 1) I'd rather not reproduce
Ad 2) Unable to reproduce currently
Ad 3) Reproduced three times in a row just now

Regards
Christian