[reSIProcate] FlowId Class Questions

[Alan Hawrylyshen]

Oops, posting to the list too.

On Jun 8, 2005, at 16:50, Derek MacDonald wrote:

> Dlb & I talked about this; if that pointer isn't in a set of valid  
> pointers
> it will be treated as bad. It really doesn't matter if we use a map  
> token or
> an existence check by a set in this case.
>
> Once the GruuMonkey is more written FlowId can be tweaked to work  
> the other
> way.
>
>


I disagree -- pointers will follow a particular pattern and a  
malicious client will be able to convince you to use someone else's  
response context or connection by guessing a flowid. I would argue  
that a map, with random keys is a lightweight approach that mitigates  
this attack.

You don't want to answer the question "is this pointer valid?" but  
"is this pointer valid for this  SIP transaction / context?".  
Therefore, in order to prevent a trivial attack mechanism, there  
needs to be some way of preventing the 'wire-space' people from  
suggesting a flowid.  This can be done with randomization and a  
porous key-space or by incorporating some sort of message  
authentication technique for the flowid.  I get the shivers thinking  
about taking  a pointer value or index from the wire without a way to  
qualify it to the appropriate scope.

Thoughts?

A