Re: [Ntp] comments on draft-mlichvar-ntp-ntpv5-03 / Message Format

[Dan Drown <[e-mail]>]

Quoting Harlan Stenn <[e-mail]>:
> This is what happens when the choice is made to ignore the design of  
> NTP packet version "evolution".
>
> The original design is that the packet format to the length of a  
> version N packet is ALWAYS the same.  This is why ntpd will respond  
> to a "higher" version number packet with the requested newer version  
> but with a packet length of the responding ntpd's version.
>
> If a v4 NTP packet is 48 bytes, the design and expectation has been  
> that if v5 wants any different data then the packet is made longer  
> and the new data lives AFTER the v4 data.
>
> Assume, for purposes of this thread, that an NTPv5 packet format is  
> now 64 bytes.
>
> Should a v5 client send a request to a v4 server, the v5 client  
> sends a 64-byte v5 request, and assuming the v4 server responds, it  
> will do so with a 48-byte v5 response that is effectively a v4  
> response packet. The client will see this, note that the packet is  
> short, and be in a position to understand that the target server is  
> a v4 server, not a v5 server.
>
> Should a v4 client send a request to a v5 server, the v5 server  
> should respond with a 48-byte v4 response.

The v4 client with a v3/v4/v5 server seems to be a pretty easy case.  
And I agree that the server should respond with the same version that  
it received.

The v4/v5 client with an unknown version server does have challenges,  
and is covered in section 9, "NTPv5 Negotiation in NTPv4"

Sending a v5 packet to a v4-only server is likely going to be dropped.  
It might be nice to document how the most popular NTP server software  
handles this somewhere.