Hi Robert,

First, We have analyzed the adversary models of SAV messages as follows. 

1) Message alteration: A malicious AS alters any part of a DSAV message, such 
as source prefix field or propagation scope field.
2) Message injection: A malicious AS injects a “valid” DSAV message and sends 
it to the corresponding next-hop AS, such as replay attacks.
3) Path deviation: A malicious AS sends a DSAV message to a wrong next-hop AS 
which conflicts with the propagation scope field.
4) Combination attack: A malicious AS alters a DSAV message to prevent path 
deviation from causing obvious conflicts, which is the most challenging attack.

Generally, we are facing the same problem as in BGP routing. So BGPsec and 
RPKI-like approaches can be taken to prevent nodes from advertising false 
prefixes. Existing symmetric cryptography mechanisms for origin and path 
validation can also be used address these adversary models. Other security 
problems and mechanisms are to be explored. 

Can you explain more about "A lot of harm can be done if you just try to copy 
BGP state of the art in security"?

Second, at this stage I am not saying that we cannot leverage BGP FlowSpec to 
help distribute the SAV messages. We are also thinking about the feasibility. 
But even so, what I want to mean is that it is not all of the story in our 
solution. We need a way to scalably notify the source prefixes of each AS to 
other ASes along the real forwarding path, which is the core of our 
architecture. 

Again, can you explain more about "But I guess/hope analysis has been done on 
this already "?

Backup path is also one of our design choices to handle the convergence problem 
during failure, which can be found from the slides we presented in the last 
SAVNET BOF. In this case, we just regard the backup path as a valid path as 
well when building the SAV tables. Given our solution supports multi-path 
routing, there is no big difference in the design architecture.

Best,
Dan

发件人: Robert Raszuk <[email protected]> 
发送时间: 2022年5月11日 23:30
收件人: Dan Li <[email protected]>
抄送: [email protected]; RTGWG <[email protected]>; Lizhenbin <[email protected]>
主题: Re: [savnet] IPv6 and/or IPv4//RE: SAVNET WG charter

Hi Dan,

> So the security approaches adopted to BGP routing can also be applied to 
> secure 
> SAV messages, such as BGPsec, RPKI, etc.

If you want to apply SAV messages for any filtering the security aspect must be 
solved first. A lot of harm can be done if you just try to copy BGP state of 
the art in security.  

> The idea of SAV protocol is to let each AS advertise it’s prefixes along the 
> “real” data-plane forwarding path.
+
> path should make a decision on which neighboring AS to propagate this 
> message, based on its AS routing path 
> it chooses in the data plane.

I am not seeing what is missing in FlowSpec encoding to do just that. 

Of course advertising SAV to selected peers unlike advertising reachability to 
all peers will have a drastic connectivity restoration effect. But I guess/hope 
analysis has been done on this already.  Waiting for any control plane 
convergence during link or node failure is not the correct model to build 
robust networks. Backup state (including filters if applicable)  should be in 
place ahead of any failures. 

Best,
Robert


On Wed, May 11, 2022 at 4:23 PM Dan Li <mailto:[email protected]> wrote:
Hi Robert,
 
First, in the charter we do not specifically mention how to secure the new 
protocol. We want to leave it to the architecture design document. But we have 
widely discussed the security issue of protocol messages in the mailing list 
before. Generally, the SAV message propagation direction is the inverse of BGP 
routing advertisement messages,  and SAV faces similar security problems as BGP 
routing messages. So the security approaches adopted to BGP routing can also be 
applied to secure SAV messages, such as BGPsec, RPKI, etc.
 
Second, RFC 8955 does not solve the problem of SAV. If we talk about BGP flow 
spec, let’s focus on the inter-domain scenario of SAV. The idea of SAV protocol 
is to let each AS advertise it’s prefixes along the “real” data-plane 
forwarding path. Each AS in the path should make a decision on which 
neighboring AS to propagate this message, based on its AS routing path it 
chooses in the data plane. It’s not “broadcasting” to all neighboring ASes. BGP 
flow spec protocol does not handle these issues.
 
Best,
Dan
 
 
发件人: Robert Raszuk <mailto:[email protected]> 
发送时间: 2022年5月11日 21:39
收件人: Dan Li <mailto:[email protected]>; mailto:[email protected]; RTGWG 
<mailto:[email protected]>
抄送: Lizhenbin <lizhenbin=mailto:[email protected]>
主题: Re: [savnet] IPv6 and/or IPv4//RE: SAVNET WG charter
 
Hi,
 
Could someone point me to the text (or charter section) describing how this new 
protocol is going to be secured ? 
 
In RFC5575/8955 we used BGP to automatically validate if the peer advertising 
the DDoS vector is the same as the peer advertising IP reachability to those 
src addresses subject to additional filtering. 
 
How is SAVNET going to solve this ? 
 
Also let me repeat one question which I asked before and have not seen reply to 
.. What is missing in vanilla RFC8955 for SAVNET ? It is supported today by all 
major vendors, deployed and in fact in use. Do we need a mirror of existing 
solution ? 
 
There is also ongoing work for Flowspec v2 if there is something missing in v1 
- but for SAV I am not sure v2 would be needed. 
 
Thx a lot,
Robert.

_______________________________________________
rtgwg mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/rtgwg

Reply via email to