[ntp:questions] Poul-Henning Kamp and re-write of NTP
stenn at ntp.org
Mon Dec 8 03:20:58 UTC 2014
> On Sun, Dec 7, 2014 at 4:35 AM, Rob <nomail at example.com> wrote:
> > It has been discussed before that reference clock drivers should be
> > loadable modules or even separate processes.
> That's unnecessary complexity. refclock_atom is only ~ 200 lines. The
> code just needs to be adjusted so you can build with exactly the
> driver(s) you need plus PPS without (essentially) forcing all-clocks.
> There's little reason for a given refclock build to target multiple
Loadable modules for what is often a root-process are a security threat
I am not comfortable with.
A separate process might be OK, but that adds a communications layer and
other complexities that I'm not fond of. But I think this is probably
better than loadable modules.
I'm still thinking that this could be done with a "refclock definition
language" similar to modemcap or some others, that would produce
threaded code that would be much more secure and would allow folks to
make changes to their refclock processing code without needing to
recompile NTP. Yes, there's a chance that new refclocks might need some
tweaks to the RDL for additional support functions, but by and large
folks would be able to make adjustments or create new refclock drivers
by only compiling some RDL bytecode.
Harlan Stenn <stenn at ntp.org>
http://networktimefoundation.org - be a member!
More information about the questions