[ntp:questions] Re: uk pool problem

Danny Mayer mayer at ntp.isc.org
Sun Sep 3 00:18:12 UTC 2006


Per Hedeland wrote:
> In article <uj3js3-sv7.ln1 at gateway.py.meinberg.de> Martin Burnicki
> <martin.burnicki at meinberg.de> writes:
>> Ronan Flood wrote:
>>> Harlan Stenn <stenn at ntp.isc.org> wrote:
>>>
>>>> -d is covered, and while there may not be an exact duplicate there is a
>>>> -d flag for ntpd and the sntp command has a way to query the time without
>>>> setting it.  If there is a particular thing you need that is not covered
>>>> open up an enhancement request.
>>>>
>>>> I have not looked at -u.
>>> Perhaps rather than being retired, ntpdate should have the time-setting
>>> code removed and be renamed something like ntpping, with -qu always set.
>>> I for one find it a useful diagnostic tool in query-only and debug modes.
>> Full ack. I very often use it for debugging and testing. The only thing I
>> find deprecated is to use the way it has been used before the -g option had
>> been introduced, namely to set the initial system time.
>>
>> I wouldn't even remove the capabiltiy to send requests via either a
>> priviledged or an unpriviledged port. This is very useful to check whether
>> there's some kind of firewall between the test system and the NTP server
>> which only allows for unpreviledged ports and blocks priviledged, or
>> vice-versa.
> 
> This would actually have to be an enhancement - when wanting to check if
> the discussed server possibly used the source port to determine whether
> to give a bogus answer, I found somewhat to my surprise that it's not
> possible to have ntpdate use source port 123 without setting the clock.
> A patch to make -u independent of -q and -d is trivial of course, but
> where to send it?:-)
> 
> --Per Hedeland
> per at hedeland.org
> 

As I recall, the protocol requires that the source port be 123 but the
ntpd reference server implementation does not enforce that. I don't
recall where I read it. That said it seems useful to differentiate
between an ntpd server requesting time and a client requesting time via
ntpdate. Nevertheless a server shouldn't return time at all unless it's
a KOD packet if it doesn't want to accept packets at all. This one seems
to return a specific packet value if queried via ntpdate. I seems to me
that this is more a WG discussion and probably should be discussed there.

Danny



More information about the questions mailing list