Re: [BEHAVE] [v6ops] protocols without need for ALG ?

[Joe Touch <[e-mail]>]

Hi, Mark,

On 7/31/2015 4:45 PM, Mark Smith wrote:
> On 31 July 2015 at 23:21, Joe Touch <[e-mail]> wrote:
>> TFTP servers are typically reached at UDP port 69.
>>
>> It does not use ports or addresses in-band and thus should not need an ALG.
> 
> Hmm, to my mind, an "ALG" is necessary if something about the protocol
> needs to be understood e.g., look for/change in-band ports or
> addresses, and possibly set up corresponding state or temporary access
> list/firewall permissions for related traffic.
> 
> In the case of TFTP, it is the "TID"s:
> 
> "The transfer identifiers (TID's) used by
>    TFTP are passed to the Datagram layer to be used as ports; therefore
>    they must be between 0 and 65,535.  The initialization of TID's is
>    discussed in the section on initial connection protocol."
> 
> " A
>    requesting host chooses its source TID as described above, and sends
>    its initial request to the known TID 69 decimal (105 octal) on the
>    serving host.  The response to the request, under normal operation,
>    uses a TID chosen by the server as its source TID and the TID chosen
>    for the previous message by the requestor as its destination TID.
>    The two chosen TID's are then used for the remainder of the transfer."

Ahh, right - I forgot that TFTP doesn't stick to the same UDP port pair.
It works a bit like FTP in that it "hands off" the transfer to the
server port indicated by the server-side TID after the initial request.

> I think a server could choose to continue to use 69 as its TID for the
> full transfer, however in my case it didn't.

It never will. The first packet is supposed to have a destination TID of
69 but the rest of the session uses TIDs chosen by each end.

	client sends sTID=X, dTID=69 on UDP sPORT=X dPORT=69
	
	server replies with sTID=Y, dTID=X on UDP sPORT=Y dPORT=X,
	which is where the rest of the transfer stays until a new TID
	is selected.

> I still remember it today
> because I was only able to get around the unpredictable TID selection
> on both ends by using just host IP addresses, which had some risks
> because it was a very coarse way of selecting "interesting"
> dial-on-demand traffic to hold the link up.

I'm not sure what you were doing; "using just host IP addresses" doesn't
fix the issue because TFTP still wants to pick TIDs, and the UDP ports
vary when the TIDs do.

> So if in Toerless's scenario it is stateless 1:1 translation between
> IPv4 and IPv6, then I don't think an ALG would be necessary for TFTP.
> However, if the translation is stateful because translation between
> IPv4 and IPv6 isn't 1:1, then I think an ALG is necessary to set up a
> mapping of some form.

So first let me backtrack - TFTP requires an ALG to work through a NAT
even when the client is on the private side. The return traffic content
needs to be inspected to be able to update the private:public
translation entry.

I.e., this won't work with stateless 1:1 *or* stateful 1:1 without an
ALG. I can't imagine how to make sense of it safely in a non-1:1 case.

Note that the whole point of TFTP is to be used on a LAN only. It has
zero security, and it's easily possible that the translation entry could
inadvertently affect some other UDP translation entry.

Joe