Mailing List Archive: bgp flapping, error: BGP type 2 length 3302 is too large, attribute total length is 2327. attr_endp is 0x7fbb5c0f7125. endp is 0x7fbb5c0f6d4e

Hi,

I have two border routers running quagga 0.99.15 on Centos, each one has
bgp connections with 1 IP transit and 1 internet exchange point:

Border A - 1 IP transit (Carrier Y), 1 IX-NAP;
Border B - 1 IP Transit (Carrier Z), 1 IX-BR;

Plus I have one Juniper MX-5 with 1 IP transit.

Connected to them by iBGP I have three access routers:

Access Router 1 - quagga 0.99.12 on Fedora, iBGP with Border A, iBGP
with Border B, iBGP with MX;
Access Router 2 - quagga 0.99.15 on Centos, iBGP with Border A, iBGP
with Border B, iBGP with MX;
Access Router 3 - quagga 0.99.15 on Centos, iBGP with Border A, iBGP
with Border B, iBGP with MX;

Almost at the same the iBGP connections between the 5 Linux routers
started to flap, no problem between Linux and MX at all
In the borders logs there are only reset errors:

BGP: x.x.x.x [Error] bgp_read_packet error: Connection reset by peer

In the access routers logs there are this errros:

BGP: x.x.x.x BGP type 2 length 3094 is too large, attribute total length
is 2119. attr_endp is 0x7fbb5b9d9735. endp is 0x7fbb5b9d935e
BGP: x.x.x.x BGP type 2 length 3174 is too large, attribute total length
is 2187. attr_endp is 0x7fbb5bf89e05. endp is 0x7fbb5bf89a22

BGP: x.x.x.x BGP type 2 length 3306 is too large, attribute total length
is 2319. attr_endp is 0x7f9444a96e99. endp is 0x7f9444a96ab6
BGP: x.x.x.x BGP type 2 length 3302 is too large, attribute total length
is 2327. attr_endp is 0x7f944c668015. endp is 0x7f944c667c3e

BGP: x.x.x.x BGP type 2 length 3302 is too large, attribute total length
is 2327. attr_endp is 0x7fa33ffc57a5. endp is 0x7fa33ffc53ce
BGP: x.x.x.x BGP type 2 length 3306 is too large, attribute total length
is 2319. attr_endp is 0x7fa33c009ec9. endp is 0x7fa33c009ae6

First I shutted down all ibgp connections (except those with MX), then I
shutted down the transit connections and got up the ibgp connections again.
It's all running fine since then but I'm affraid to get up the IP
transits and the ibgp start to flap again.

I don't see any particular reason for what happened, the access routers
are different machines with different systems and quagga versions, the
borders routers has different carriers and internet exchange points.

Any ideas?

Thanks in advance,
Leonardo

Yes I guess so, the problem is I'm using the distro version so I would
have to compile a new one.

The borders has the same version and doesn't show the problem.

This set up has been working for five or six years, I would like to know
what this length problem is and if there is a workaround or only upgrating.

Thanks.

Em 27-07-2017 18:24, Alexis Rosen escreveu:
> On Jul 27, 2017, at 5:09 PM, Leonardo Porto <leonardo.porto@iw.net.br> wrote:
>> I have two border routers running quagga 0.99.15 on Centos
> You are unlikely to get any help with that version. Try upgrading first.
>
> Of course, deciding what to upgrade to is hard. Upgrading to 0.99.24 should be safe and there's a good chance it'll fix your problem.
>
> Upgrading to 1.anything brings new features, new bugfixes... and also longstanding BGP bugs on which no progress has been made in over a year.
>
> There's also the fork, FRR.
>
> /a
>