Re: [grobj] NATs and scope

[Brian E Carpenter <[e-mail]>]

On 2009-11-20 10:00, Simon Perreault wrote:
> Brian E Carpenter wrote, on 2009-11-19 15:21:
>> On 2009-11-20 03:11, Simon Perreault wrote:
>>
>>> - The "scope" of an address (I'm still not fully understanding the concept) does
>>> not guarantee that that the address is reachable from anyone in that scope. For
>>> example, the address of a NAT binding (called a "reflexive transport address" in
>>> RFC5389) can be reached by a subset of peers in the NAT's scope which depends on
>>> the type of NAT (see e.g. RFC4787 for types of NATs).
>> So, given what I just said about scope (with no reference to NATs), I think
>> Simon needs to tell us what he means by "in the NAT's scope".
> 
> I meant the scope of the interface on the NAT that is usually labeled "WAN".
> That's usually Internet-scope, but it's common to have multiple layers of NAT,
> in which case the scope would be different.
> 
>> To me an address scope is a set of mutually reachable addresses, which
>> means that a NAT automatically forms (part of) the boundary of a scope.
>> I don't understand what is meant by "a subset of peers" - surely all
>> the peers behind a given NAT (more precisely, behind a given interface
>> of a given NAT) are in the same address space?
> 
> Here's an example:
> 
>    LAN --------|NAT|---- WAN
> 10.0.0.0/8             Internet
> 
> A host in the LAN initiates a UDP "connection" to the WAN. The NAT allocates a
> binding on its WAN interface, e.g. 192.0.2.1:1234. The scope of this is
> "Internet" I would presume.
> 
> For some kinds of NAT devices, any host on the Internet can reach the host on
> the LAN by sending packets to 192.0.2.1:1234.
> 
> For other kinds, only hosts to which the host in the LAN has previously talked
> using this binding can reach it.
> 
> So even if hosts are in the same scope as 192.0.2.1:1234, they may not be able
> to use it.

OK. So, in my personal view of the universe, "in scope" means reachable
directly and without translation. We need to be precise about some
of these definitions, clearly.

   Brian