IGP protocol

Saku Ytti saku at ytti.fi
Sun Nov 18 10:59:19 UTC 2018

On Sun, 18 Nov 2018 at 12:15, Alfie Pates <alfie at fdx.services> wrote:

> There's a school of thought which suggests MD5 security on single-hop BGP is absolute theatre with no security benefit and that MACsec is the route you should be taking.

AFAIK there are no known attacks against HMAC-MD5. eBGP I don't care
about. But for iBGP I consider this a problem:

Someone goes to random forest where fibre is trenched, digs it up,
taps fibre until correct fibre+wave is found, then injects BGP UPDATE
to change L3 MPLS VPN labels, and diverts traffic through their device
while returning it safely. Seems quite cheap attack, maybe <5k, and
entirely compromises MPLS security model. iBGP MD5 should protect well
from this.

Not arguing that MacSec isn't superior feature, it's just cost of
MacSec is non-trivial compared to cost of HMAC-MD5, and it seems
HMAC-MD5 for certain attacks is strong guarantee. Ideally we'd
implement TCP-AO (RFC5925) to replace BGP HMAC-MD5, just to get
derived secret instead of static (how many change their MD5 secret
periodically?)  but it looks like ship may have sailed on that one.


More information about the NANOG mailing list