At Mon, 20 Oct 2008 15:51:47 +1000, Terry Manderson wrote:
> 
> Here is a hypothetical scenario that I think is technically viable.
> 
> Consider two initial parties, Party A and B. Party A is the well  
> mannered ISP who regularly fetches and validates the RPKI repository.  
> Party B who is multi-homed is intent on doing an attack of some form  
> on party A's customers. The entity that allocated the IP address space  
> to party B for some reason retracts the allocation (expired  
> membership, Law enforcement action etc) and issues a BOA.
> 
> Party B somehow sees this happen and puts into play a mechanism close  
> to the repository which identifies the collection of the updated  
> manifest, 3779 certificate, and associated BOA - They then in transit  
> modify one bit of data such that any of the following occurs for  
> relying parties:
>       manifest fails to validate
>       the object's hash doesn't match those of the manifest and the objects  
> fail to validate
> 
> This means that depending on the policy of the relying parties, the  
> bgp announcement from party B will remain in place for as long as the  
> MiTM attack is happening on the repository elements related to party  
> B, and party B's attack on customers of party A (or others) can remain  
> in place.
> 
> Using TLS would make this MiTM attack substantially more difficult to  
> do with the intent of successfully downloading valid data from the  
> repository for RPKI validation.
> 
> .. Am I mistaken, is this not a valid attack on the RPKI infrastructure?

You appear to be describing a selective DoS attack by an on-path
attacker.  Unless I'm missing something, adding TLS to the mix here
changes details but doesn't change the overall situation much, as:

a) The attack was already detectable without TLS (those validation
   failures you mentioned), and

b) An on-path attacker can DoS the relying party even when TLS is in
   use, albeit somewhat more crudely.

So I don't see that TLS buys you much here, and you haven't said
anything about the risks and costs of adding it: new trust
relationships (adding channel trust relationships to what had
previously been an end-to-end object security model) and new potential
attackers (who did you say issues the TLS certificates?).
_______________________________________________
sidr mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/sidr

Reply via email to