[Ntp] Antw: [EXT] Handling of future versions in historic implementations

[Ulrich Windl <[e-mail]>]

>>> Miroslav Lichvar <[e-mail]> schrieb am 22.09.2022 um 17:25 in
Nachricht <Yyx+VOkuFCAuBBYb@localhost>:
> I did a bit of archeology to see how the older NTP implementations
> handled future versions, now that there are some concerns about NTPv5
> making incompatible changes.

Thanks for the work!

...
> The NTPv1 implementation (ntpd) was written by Louis A. Mamakos and Mike
> Petry in C. In the code I found (from May 1989) in the main() function
> there is:
> 
> 	if ((pkt‑>status & VERSIONMASK) != NTPVERSION_1)
> 		continue;
> 
> 	receive(dst, pkt, &tv, i);
> 
> That is, future versions were ignored.

...and "older versions"...

> 
> The NTPv2 implementation seems to be a new implementation written from
...
> Again, future versions were ignored.
> 
> The NTPv3 implementation was based on the NTPv2 xntpd and the NTPv3
...
> This seems to be intentionally handling future versions as compatible
> with NTPv3.

A bad idea as the version was not defined at the time of writing
...

> 
> The NTPv4 support was implemented in xntpd by David Mills and it was
> renamed to ntpd. In the current version of the code there is:
> 
> 	if (hisversion == NTP_VERSION) {
> 		sys_newversion++;		/* new version */
> 	} else if (   !(restrict_mask & RES_VERSION)
> 		   && hisversion >= NTP_OLDVERSION) {
> 		sys_oldversion++;		/* previous version */

I think the assumption was (at the time of writing) that there won't be a
version > NTP_VERSION, so the oldversion test would apply to the interval
[NTP_OLDVERSION...NTP_VERSION-1].

As obvious in all example code, a comment explaining what the code is supposed
to do would have been helpful.

...

> The future versions are accepted as "previous version". This might
> seem like a bug, but from what we have heard on this list recently,
> it was intended.

I think that's just a "postmortem interpretation", but it never was the
intention.

> 
> So, it seems this idea of future versions being compatible with older
> versions started with NTPv3, but for some reason it wasn't specified
> in RFC 1305 or RFC 5905.

As stated before, I think "it's a bad idea".

Regards,
Ulrich