[Samba] DNS issues after FSMO seize
mathias dufresne
infractory at gmail.com
Tue Apr 5 09:44:48 UTC 2016
Back from downstairs to take a break. I'll try to moderate my words and
explain a bit more.
SOA means Start Of Authority. That is true but that means nothing as itself.
Parents are start of authority facing their children. My boss is start of
authority for my job.
NS are authoritative, but once more, as is, that means nothing.
The question is what means "start of authority" and "NS is authoritative"
in DNS.
In DNS there are two kind of name server: masters and slaves.
In DNS there are two of records to reference name servers: NS and SOA.
The point is to make difference between these two records.
That is simple:
- NS are name servers you can trust the reply (they are authoritative for
replying). This record should be used only by name servers, never by
clients. Clients deal with their resolver which is often a name server, if
the resolver does not know how to contact a name server for a specified
zone, the resolver will ask for NS record of that zone, that request will
be sent to upper level of name server (root servers). Then the resolver get
IP of specified zone name server and can send to that name server the
client's request, and finally send back the reply to the client. The client
never uses NS records or I missed something.
- SOA are name servers you can send them DNS updates, they can modify the
zone. SOA is often declared as "best source of information" but this comes
from the fact SOA would be the first to receive the update, the source of
information the most up to date.
There is no link between SOA and NS except they share the zone.
You can build a DNS zone with 3 name servers, one declared as SOA, two
declared as NS. I expect you can even configure SOA to refuse replying DNS
requests which are not updates (with allow-query parameter I would say).
Let say our SOA allow queries. Our SOA is not declared as NS.
3 clients (c1, c2, c3), each of them is configured with different resolver:
- c1 uses NS1
- c2 uses NS2
- c3 uses SOA
If our three name server allow queries, our three clients are able to
resolve name from our DNS zone. This happens not because we have NS
declared, this happens because our name server allow queries and share the
same zone.
The fact SOA is SOA and not NS does not forbid it to reply, we configured
it to accept queries which means it cans reply.
Update case (Bind with no DLZ or Samba internal DNS)
- c1 use nsupdate to update our zone. nsupdate will look for a name server
of that zone which can modify the zone, as we want to modify the zone and
we know all name server are not _necessarily_ allowed to do that. So first
thing, who is SOA. NS1 reply SOA record refers to machine named SOA (yes I
know the reply is IP and not name, I just try to be understood), c1 send
update request to SOA rather than NS1.
- c2, same process except NS2 replies.
- c3 same again except SOA replies.
Update case (Bind + DLZ or MS AD name server)
- c1 use nsupdate to update our zone. nsupdate will look for a name server
of that zone which can modify the zone, as we want to modify the zone and
we know all name server are not _necessarily_ allowed to do that. So first
thing, who is SOA.Here NS1 replies "I am SOA", c1 send the request to NS1.
- c2 same, with NS2
- c3 same with SOA.
2016-04-05 10:33 GMT+02:00 mathias dufresne <infractory at gmail.com>:
>
>
> 2016-04-04 14:20 GMT+02:00 Rowland penny <rpenny at samba.org>:
>
>> On 04/04/16 10:23, mathias dufresne wrote:
>>
>>> SOA means "this DNS se'rver can modify the zone".
>>>
>>
>> No it doesn't, it stands for 'Start Of Authority' and contains who to
>> contact for the domain records.
>>
>
> Rowland... thank you again Captain Obvious. Yes SOA means Start Of
> Authority, but I took time explain to clarify things which are clear for
> everyone.
> Now why when you want to update DNS you need to find SOA? Because they are
> name server which can write the zone. With bind and no DLZ it's the master,
> with dlz it's all name server you configured to be able to modify the zone.
>
>
>>
>> Using Bind-DLZ all DNS servers can modify the AD zones, they all reply "I
>>> am the SOA" when you ask them about SOA for AD zones.
>>>
>>
>> Sorry, but this, as standard, isn't correct, unless you add the other DC
>> NS records to the SOA, only the first DC is in the SOA.
>>
>
> You don't even tried! You come here to me I'm wrong and you don't even
> took time to verify! Stop answering em if you don't know what you speak
> about. Thank you.
>
> Here is a test I did for you start to understand better DNS and perhaps
> stop telling stupid things:
>
> This server, dns20, uses as a resolver itself.
> When asking for NS, there two: dc200 and dc100.
> When asking SOA there is one: the name server which replied, it replied "I
> am SOA".
>
> In AD DB SOA is dc200 which my FSMO.
>
> dns20:~# dig ad.dgfip.finances.gouv.fr -t NS
>
> ; <<>> DiG 9.9.4-RedHat-9.9.4-29.el7_2.1 <<>> ad.dgfip.finances.gouv.fr
> -t NS
> ;; global options: +cmd
> ;; Got answer:
> ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 2556
> ;; flags: qr aa rd; QUERY: 1, ANSWER: 2, AUTHORITY: 0, ADDITIONAL: 1
> ;; WARNING: recursion requested but not available
>
> ;; OPT PSEUDOSECTION:
> ; EDNS: version: 0, flags:; udp: 4096
> ;; QUESTION SECTION:
> ;ad.dgfip.finances.gouv.fr. IN NS
>
> ;; ANSWER SECTION:
> ad.dgfip.finances.gouv.fr. 900 IN NS
> dc200.ad.dgfip.finances.gouv.fr.
> ad.dgfip.finances.gouv.fr. 900 IN NS
> dc100.ad.dgfip.finances.gouv.fr.
>
> ;; Query time: 1 msec
> ;; SERVER: 10.156.32.99#53(10.156.32.99)
> ;; WHEN: mar. avril 05 10:16:53 CEST 2016
> ;; MSG SIZE rcvd: 94
>
> dns20:~# dig ad.dgfip.finances.gouv.fr -t SOA
>
> ; <<>> DiG 9.9.4-RedHat-9.9.4-29.el7_2.1 <<>> ad.dgfip.finances.gouv.fr
> -t SOA
> ;; global options: +cmd
> ;; Got answer:
> ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 58991
> ;; flags: qr aa rd; QUERY: 1, ANSWER: 1, AUTHORITY: 2, ADDITIONAL: 1
> ;; WARNING: recursion requested but not available
>
> ;; OPT PSEUDOSECTION:
> ; EDNS: version: 0, flags:; udp: 4096
> ;; QUESTION SECTION:
> ;ad.dgfip.finances.gouv.fr. IN SOA
>
> ;; ANSWER SECTION:
> ad.dgfip.finances.gouv.fr. 3600 IN SOA
> dns20.ad.dgfip.finances.gouv.fr. hostmaster.ad.dgfip.finances.gouv.fr.
> 462 900 600 86400 3600
>
> ;; AUTHORITY SECTION:
> ad.dgfip.finances.gouv.fr. 900 IN NS
> dc100.ad.dgfip.finances.gouv.fr.
> ad.dgfip.finances.gouv.fr. 900 IN NS
> dc200.ad.dgfip.finances.gouv.fr.
>
> ;; Query time: 1 msec
> ;; SERVER: 10.156.32.99#53(10.156.32.99)
> ;; WHEN: mar. avril 05 10:16:58 CEST 2016
> ;; MSG SIZE rcvd: 147
>
> Same test on dc102, I don't search again for NS as the reply don't change.
> It changes only for SOA.
>
> dc102:~# dig -t SOA ad.dgfip.finances.gouv.fr
>
> ; <<>> DiG 9.9.4-RedHat-9.9.4-29.el7_2.1 <<>> -t SOA
> ad.dgfip.finances.gouv.fr
> ;; global options: +cmd
> ;; Got answer:
> ;; ->>HEADER<<- opcode: QUERY, status: NOERROR, id: 21947
> ;; flags: qr aa rd; QUERY: 1, ANSWER: 1, AUTHORITY: 2, ADDITIONAL: 1
> ;; WARNING: recursion requested but not available
>
> ;; OPT PSEUDOSECTION:
> ; EDNS: version: 0, flags:; udp: 4096
> ;; QUESTION SECTION:
> ;ad.dgfip.finances.gouv.fr. IN SOA
>
> ;; ANSWER SECTION:
> ad.dgfip.finances.gouv.fr. 3600 IN SOA
> dc102.ad.dgfip.finances.gouv.fr. hostmaster.ad.dgfip.finances.gouv.fr.
> 462 900 600 86400 3600
>
> ;; AUTHORITY SECTION:
> ad.dgfip.finances.gouv.fr. 900 IN NS
> dc100.ad.dgfip.finances.gouv.fr.
> ad.dgfip.finances.gouv.fr. 900 IN NS
> dc200.ad.dgfip.finances.gouv.fr.
>
> ;; Query time: 1 msec
> ;; SERVER: 10.154.102.166#53(10.154.102.166)
> ;; WHEN: mar. avril 05 10:22:23 CEST 2016
> ;; MSG SIZE rcvd: 147
>
>
>
>
>>
>> Using Internal DNS I expect all DNS servers can modify the AD zones also
>>> (that's internal stuff) but even if they can modify the AD zone locally
>>> that's is not the process chosen by Samba Team. Samba Internal DNS relies
>>> on DB content to reply to SOA query and there is only one SOA in the DB.
>>> So with internal DNS you will have always only one SOA.
>>>
>>
>> As standard this is true, but you can add the DC records to the SOA
>> record in AD and then Bind dlz will show each DC as being the owner of the
>> SOA, but as you say, the internal DNS doesn't.
>>
>
> Yes internal does not follow standard.
>
>
>>
>>
>>> This is an issue because if your SOA is down and some DC has DNS updates
>>> to
>>> send, updates will fail because no SOA available.
>>> And when seizing roles because you are about to remove old FSMO, that's
>>> the
>>> same: once the FSMO is removed no more SOA to apply DNS updates on.
>>>
>>
>> The problem isn't if you remove a DC, it is that, as standard, there is
>> only one NS in the SOA.
>>
>
> Missed. SOA as nothing to do with NS. I could write again the difference,
> but you don't trust me. If I don't say same as you, I'm wrong (until I show
> the contrary).
>
> As internal does not follow standard, it must move SOA at some moments
> (hard to define all because conditional behaviour).
>
>
>>
>>
>>> For me, I can be wrong, this behaviour comes from the fact Samba uses
>>> "nsupdate" command to push DNS updates. nsupdate comes from Bind tools
>>> suite, as it is bind tool it follows the protocol. And the protocol says
>>> "updates can be pushed only on SOA". So nsupdate first ask the zone to be
>>> modified what is the SOA to push updates on that server.
>>>
>>
>> I don't think the problem is with 'nsupdate', I think it is with the way
>> the internal DNS uses the SOA.
>>
>
> You who don't seem to test much, do you run a tcpdump when using nsupdate
> -g /path/to/file.generated_by_samba_dnsupdate?
> If you did that you would have seen the first action of nsupdate is to
> find the SOA (because SOA is or are the place(s) to push changes).
>
>
>>
>>
>>> IMHO this should be managed by Samba itself rather than relying on Samba
>>> admins DNS knowledge.
>>> Samba Internal DNS should be able to push update locally and Samba
>>> internal
>>> DNS should answer "I am SOA" as they can push DNS updates locally (they
>>> have access to the DB, they can push updates, even if this needs to write
>>> some code).
>>>
>>
>> Totally agree with you, do you want to write this code ?
>>
>
> No.
>
>
>>
>> Or samba_dnsupdate should not use by default nsupdate from bind tools when
>>> using internal DNS but rather use "samba-tool dns ..." which pushes
>>> updates
>>> locally.
>>>
>>
>> Do you mean like archlinux did with dhcp ?
>>
>
> dhcp on archlinux has some dependancy on samba4 in AD mode? I didn't
> noticed that... Because for dhcp start to use samba_dnsupdate or samba-tool
> there is certainly some work to do in dhcp code...
>
> I'm joking but seriously, I don't see any relation between arch, dhcp and
> samba tools, so I don't see the point of that question.
>
>
>>
>> And before you start thinking I am trying to put you down, I am not, I am
>> just pointing out what *I* see as mistakes. If you have any problems with
>> what I write, provide proof of where I am wrong and I will apologise.
>>
>
> I think I did. For both: being pissed of in the morning and giving you
> explanations.
>
>
>>
>> Rowland
>>
>>
>>>
>>>
>>> 2016-04-01 19:26 GMT+02:00 lingpanda101 at gmail.com <
>>> lingpanda101 at gmail.com>:
>>>
>>> On 3/31/2016 3:12 PM, IT Admin wrote:
>>>>
>>>> Aaaaaaand more problems... Welcome to the continuing saga of FILER.
>>>>>
>>>>> It appears that neither SOA or NS records were updated during the
>>>>> process
>>>>> of moving fsmo roles to CBADC01. SOA entries on all three active DCs
>>>>> point
>>>>> to FILER. There aren't any NS records for any of the new DCs, only
>>>>> FILER.
>>>>> In RSAT each DNS server's properties show filer.cb.cliffbells.com is
>>>>> the
>>>>> primary server. This looks at awful lot like this to me:
>>>>> https://lists.samba.org/archive/samba/2015-October/195352.html
>>>>>
>>>>> This process is killing me.
>>>>>
>>>>> I assume this misconfiguration is in large part responsible for
>>>>> authentication and share access issues I'm now experiencing (I sent a
>>>>> reply
>>>>> to my last thread on failing to join DCs to the domain but received no
>>>>> replies). I'm of the opinion mentioning this potential failure along
>>>>> with
>>>>> the procedure to mitigate the issue would be useful on the wiki page
>>>>> detailing transfer/seize fsmo roles... If I read things right if the
>>>>> original fsmo role holder has been demoted prior to identifying the
>>>>> failure
>>>>> samba-tool won't be able to query the original DC and manual
>>>>> intervention
>>>>> will be required?
>>>>>
>>>>> Please advise. This client is about ready to throw me off the roof and
>>>>> abandon the system for quill and ink by candle light. I'm considering
>>>>> just
>>>>> going with it.
>>>>>
>>>>> JS
>>>>>
>>>>> I should mention the syntax to update the SOA if you do not know. I did
>>>> this back in the Samba 4.0 days. Not sure if things have changed.
>>>>
>>>> samba-tool dns update SOA "fqdn_dns fqdn_email serial refresh retry
>>>> expire
>>>> minimumttl"
>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> -James
>>>>
>>>>
>>>>
>>>> --
>>>> To unsubscribe from this list go to the following URL and read the
>>>> instructions: https://lists.samba.org/mailman/options/samba
>>>>
>>>>
>>
>> --
>> To unsubscribe from this list go to the following URL and read the
>> instructions: https://lists.samba.org/mailman/options/samba
>>
>
>
More information about the samba
mailing list