[Dnsmasq-discuss] Serial loosed after restart

[Simon Kelley]

On 26/09/14 09:34, Christian Ruppert wrote:
> Hi Simon,
> 
> it's a VM with no real HW-Clock, that is correct. Using seconds since 1970 would
> be ok but the serial starts with "1" when dnsmasq has been (re-)started:
> 
> # ps aux | grep dnsmasq
> dnsmasq  17460  0.0  0.0  30296   956 ?        S    Sep25   0:06
> /usr/sbin/dnsmasq -x /var/run/dnsmasq/dnsmasq.pid -u dnsmasq --local-service
> 
> So it's running since yesterday (2014/09/25) and has currently a serial of:
> # dig -t SOA +short ipmi.rz.babiel.com @127.0.0.1 -p 5353
> . . 4947 1200 180 1209600 600
> 
> Now if I restart dnsmasq:
> # /etc/init.d/dnsmasq restart
> [ ok ] Restarting DNS forwarder and DHCP server: dnsmasq.
> # dig -t SOA +short ipmi.rz.babiel.com @127.0.0.1 -p 5353
> . . 1 1200 180 1209600 600
> 
> It starts from 1 again.

This is a mystery. The code is quite simple, it's around line 212 in
src/dnsmasq.c

now = dnsmasq_time();

/* Create a serial at startup if not configured. */
  if (daemon->authinterface && daemon->soa_sn == 0)
#ifdef HAVE_BROKEN_RTC
    die(_("zone serial must be configured in --auth-soa"), NULL,
EC_BADCONF);
#else
  daemon->soa_sn = now;
#endif

If you can see how that code can end up with daemon->soa_sn having value
1 in your VM, that would solve the mystery.

> 
> Re "--auth-soa":
> I tried "auth-soa=20140926001" but:
> # dig -t SOA +short ipmi.rz.babiel.com @127.0.0.1 -p 5353
> . . 2961056818 1200 180 1209600 600
> So that's not the serial that I specified. Anyway, after some time it increased
> to 19, 20, ... and after a restart it was again at "2961056818". So using
> "auth-soa" won't help :(

The serial gets incremented every time the DHCP lease database changes
(because the domain includes names derived from the leases, so changing
a lease can change the domain.) So that explains what you're seeing
there. I concede it does make my suggestion not viable.

I guess that's why the seconds-since-epoch value is used.  It's a
reasonable assumption that over a reasonable period, DHCP leases won't
change more than once a second.


Cheers,

Simon.



> 
> On 09/25/2014 11:04 PM, Simon Kelley wrote:
>> On 25/09/14 10:39, Christian Ruppert wrote:
>>> Hey Guys,
>>>
>>> I use the auth-zone, auth-sec, auth-peer features and I noticed that dnsmasq
>>> looses its actual SOA resp. serial during restarts and thus it started again
>>> from the beginning (1). All slaves were rejecting the changes because of that
>>> serial mismatch. E.g. the slaves all had "34286" and dnsmasq started from "1" again.
>>> It would be really good to save the serial in the lease file or somewhere else
>>> and re-use it afterwards to avoid such problems. Is that a bug or was it on purpose?
>>>
>>
>> It's supposed to work like this:
>>
>> The serial number starts as the time (seconds since 1970) when dnsmasq
>> is started. Therefore stopping and restarting dnsmasq should _increase_
>> the serial.
>>
>> My guess is that you're using a platform which doesn't have a hardware
>> real-time-clock, and so it's idea of the time gets reset whenever it's
>> rebooted. Even if it uses NTP to get a good value of the time, this will
>> happen after dnsmasq has started.
>>
>>
>> You can set the initial serial number when starting dnsmasq using the
>> command-line argument (or config option.)
>>
>> --auth-soa=<serial number>
>>
>> so you could implement something like the behaviour you want by keeping
>> the serial in a file, incrementing it each time dnsmasq starts, and
>> feeding it to dnsmasq via the command line. Shell scripting to do this
>> left as an excercise....
>>
>>
>> Cheers,
>>
>> Simon.
>>
>>
>>
>>
>> _______________________________________________
>> Dnsmasq-discuss mailing list
>> Dnsmasq-discuss at lists.thekelleys.org.uk
>> http://lists.thekelleys.org.uk/mailman/listinfo/dnsmasq-discuss
>>
>