RE: Preventing BGP Route leak (Hijack) for Management Channel BGP session

[Linda Dunbar <[e-mail]>]

Eric,

Thank you very much for the detailed explanation.

I meant to ask about using BGP UPDATEs of a unique AFI/SAFI to distribute the "Public Keys" and individual "Nonce" to all the CPEs. Just like IKE (where peers exchange public key over internet), now the Public keys are exchanged between peers via RR (Controller). With this approach, the CPEs can offload the Peer Authentication job to Controller.

IPsec's  Diffie-Hellman algorithm use "Public Key" and CPE's Private Key to compute the actual Security Association.

In your draft-rosen-bess-secure-l3vpn-01, you have BLACK BGP session and RED BGP session, and emphasized on how BLACK BGP not leaking routes to RED BGP. I want to know if we can do the same for Management BGP session, i.e. requiring Data Plan BGP session to be completely separate from the Management Plane BGP session, is it Okay?

Thanks, Linda Dunbar



From: Eric C Rosen [mailto:[e-mail]]
Sent: Thursday, August 16, 2018 2:59 PM
To: Linda Dunbar <[e-mail]>; [e-mail]; [e-mail]; Jeff Tantsura <[e-mail]>; RTGWG <[e-mail]>
Subject: Re: Preventing BGP Route leak (Hijack) for Management Channel BGP session

On 8/13/2018 3:26 PM, Linda Dunbar wrote:

One of the comments to https://datatracker.ietf.org/doc/draft-dm-net2cloud-gap-analysis/<https://urldefense.proofpoint.com/v2/url?u=https-3A__datatracker.ietf.org_doc_draft-2Ddm-2Dnet2cloud-2Dgap-2Danalysis_&d=DwMFAg&c=HAkYuh63rsuhr6Scbfh0UjBXeMK-ndb3voDTXcWzoCI&r=-DXB84eU9m4cIlq2OOcCJCQQAwJXQQswyu3F0kG0VNo&m=RToh0UhV7F8cp3q2ud1LmU6GZtypPTJdboL4dgpRzr0&s=9xbKYj5fP6Jv93coe5g-lfzpp0L0bK0GyrlB91Ry3Sw&e=> the In RTGwg session of IETF102 is that using BGP session to pass configuration keys for IPsec can be risky even if the path between RR & node is secure (say via TLS) due to BGP route leak (Hijack).
But the BGP session to carry IPsec configurations is via BGP management session, which is completely isolated form the dataplane BGP sessions.

I'm not sure I have the whole context, but the question seems to be whether it could ever be safe to use BGP to distribute secret keys.

Presumably:

- The keys would be carried in an attribute that can only be attached by UPDATEs of a specified AFI/SAFI, where the specified AFI/SAFI is only used to carry management/configuration information.

- UPDATEs of that AFI/SAFI would only be sent on BGP sessions that are adequately secured so as to provide privacy, integrity and authentication.

- The UPDATEs would carry the NO_ADVERTISE community (to make sure they are not propagated further).

- None of the BGP systems involved would allow any sort of "BGP monitoring" that might expose the unencrypted contents of the UPDATEs.

In this scenario, I don't think it matters whether the secure BGP session also carries other AFI/SAFIs.

The privacy properties of this scenario are pretty good, in theory, but I don't think they are really good enough for distributing secret keys.

- Once you're using BGP to distribute information, it is inevitable that someone will decide to remove the "NO_ADVERTISE" and allow the information to be propagated through intermediate nodes (RRs or ASBRs) to the actual target node.  After all, one of the main values of using BGP to distribute stuff is that you get a big distribution system.  Even if all the intermediate nodes are trusted and all the intermediate BGP sessions have adequate privacy/integrity/authentication, you still wouldn't want to expose the secret keys to those nodes.  You might trust those nodes to see all the routing information, and even to see most of the management information, but you probably don't want them to see all the secret keys.  And you probably don't want the secret keys stored in the clear on those intermediate nodes.

- I would worry about BGP monitoring procedures creating a backdoor through which the secret keys would be exposed.

- No matter how careful you are, when you use BGP you can be pretty sure that your UPDATEs will end up somewhere they're not supposed to go.  It's just too easy to make mistakes.
So I don't think I'd try to do dynamic keying by attaching the actual keys to BGP UPDATEs.  At most I'd use BGP to distribute parameters that could then be used by something like IKEv2 to actually fetch the secret keys.