bind 9.11.3 BIND9_FLATFILE update-policy

[Rowland Penny]

On Thu, 11 Oct 2018 15:40:25 +0300
Sergey Urushkin <urushkin at telros.ru> wrote:

> Rowland Penny писал 2018-10-11 12:38:
> > On Thu, 11 Oct 2018 12:01:22 +0300
> > Sergey Urushkin <urushkin at telros.ru> wrote:
> > 
> >> Rowland Penny писал 2018-10-10 21:40:
> >> > On Thu, 11 Oct 2018 07:23:37 +1300
> >> > Andrew Bartlett <abartlet at samba.org> wrote:
> >> >
> >> >> On Wed, 2018-10-10 at 14:38 +0300, Sergey Urushkin wrote:
> >> >> > ---
> >> >> > Best regards,
> >> >> > Sergey Urushkin
> >> >> >
> >> >> >
> >> >> > Andrew Bartlett писал 2018-10-10 13:04:
> >> >> > >
> >> >> > > Sergey,
> >> >> > >
> >> >> > > Can you fill us in on your use case here?
> >> >> >
> >> >> > Actually, this is just history. When we were migrating from
> >> >> > samba3 NT domain (that was 4.0 alfa-beta times) DLZ backend
> >> >> > was buggy, didn't work with bind views, didn't support zone
> >> >> > types mixing (plain+dlz), dhcp-dns updates (that's what left
> >> >> > in my memory, I could be wrong). Every of these problems
> >> >> > could be solved in some way, but this required additional
> >> >> > configuring, migration and risks, so we decided to do this
> >> >> > later. Now all that problems seems to be solved and dlz is
> >> >> > really stable, so I don't see any reason for not using
> >> >> > DLZ/INTERNAL for new installations. But since we still don't
> >> >> > need features of DLZ, we are still PLAIN, that's why I'm
> >> >> > voting for supporting this feature :).
> >> >>
> >> >> The only part of this that can't be supported is views, but that
> >> >> could be done on another server (with a zone type of forward
> >> >> pointing at Samba) and leave the Samba server with a simple
> >> >> configuration.
> >> 
> >> By view support I meant put dlz inside view (e.g. one view with
> >> dlz, one without; or dlz in both but with different other zones),
> >> what is possible now and, as I remember, was not possible in the
> >> early versions, but I could be wrong.
> >> 
> >> >>
> >> >> > May be someone has a configured/patched bind version, so that
> >> >> > dlz breaks it, but I haven't met such. If someone has plain
> >> >> > backend he knows what he is doing (and can fix config files),
> >> >> > so all we need to support this backend without auto-breaking
> >> >> > it in the future - is removing dns_update.c code. And for new
> >> >> > installations we could describe this backend as DEPRECATED in
> >> >> > docs/tools.
> >> >> >
> >> >> > Fixed patch attached.
> >> >>
> >> >> OK, how about this for a plan:
> >> >>
> >> >> * This patch for master, then backported to 4.9 (also removing
> >> >> the dns_update.c code).
> >> >>
> >> >> The rationale is also that we think that kicking bind with a
> >> >> rndc reload can crash bind (sadly we can't reliably reproduce),
> >> >> we can't do
> >> >
> >> > If you run 'bind9 reload' you get this:
> >> >
> >> > Oct 10 19:28:12 dc3 named[5261]: Loading 'AD DNS Zone' using
> >> > driver dlopen
> >> > Oct 10 19:28:12 dc3 named[5261]: samba_dlz: starting configure
> >> > Oct 10 19:28:12 dc3 named[5261]: samba_dlz: Ignoring duplicate
> >> > zone '0.168.192.in-addr.arpa' from
> >> > 'DC=@,DC=0.168.192.in-addr.arpa,CN=MicrosoftDNS,DC=DomainDnsZones,DC=samdom,DC=example,DC=com'
> >> > Oct 10 19:28:12 dc3 named[5261]: samba_dlz: Ignoring duplicate
> >> > zone 'samdom.example.com' from
> >> > 'DC=@,DC=samdom.example.com,CN=MicrosoftDNS,DC=DomainDnsZones,DC=samdom,DC=example,DC=com'
> >> > Oct 10 19:28:12 dc3 named[5261]: samba_dlz: Ignoring duplicate
> >> > zone '_msdcs.samdom.example.com' from
> >> > 'DC=@,DC=_msdcs.samdom.example.com,CN=MicrosoftDNS,DC=ForestDnsZones,DC=samdom,DC=example,DC=com'
> >> > Oct 10 19:28:12 dc3 named[5261]: samba_dlz: shutting down
> >> >
> >> >> the version detection at runtime, can't tell if we are in
> >> >> BIND9_FLATFILE and shouldn't be doing that on a timer tick for
> >> >> 99% of installs.
> >> 
> >> Agreed.
> >> 
> >> >>
> >> >> * Then mark as DEPRECATED and remove BIND9_FLATFILE from
> >> >> provision in master for 4.10.
> >> >>
> >> >
> >> > Good idea.
> >> 
> >> BIND9_FLATFILE is the only method to use dns server on another
> >> physical server
> > 
> > Yes, but you shouldn't be running the dns server on another server,
> > the dns info is stored in AD.
> 
> This is just the simplest way. Actually I could use another server
> which will be getting info from the AD via LDAP (I know it's crazy
> but this will work).

No the simplest (and the best) is to run Bind9 on the Samba AD DC.

> 
> > 
> >> and/or mix usual bind update keys (e.g.isc-dhcp) with
> >> gssapi update keys for one zone. While I can not tell why someone
> >> would need this with current ready-to-use solutions, this is
> >> possible and fully working configuration (for clients it is
> >> absolutely the same; admins could write it's own update rules, and
> >> use nsupdate instead of samba-tool - I can confirm).
> > 
> > And I can confirm that you can use nsupdate with the dns info
> > stored in AD, even with an isc-dhcp-server
> 
> I know, I was the one who published such dhcp script in the
> forefront :)

No you were not, it was started by Michael Kuron, you just updated his
script, just as I did.

> 
> > 
> >> So why removing?
> > 
> > Because it doesn't work as BIND9_DLZ does and how AD expects.
> 
> As I know, AD (samba) just needs dns, it doesn't use ldap dns-records 
> directly (except for replication and administration purposes 
> (samba-tool)).
> 

It needs dns and it creates the records in AD, so the easiest way is to
use those records.

> > 
> >> After modifying dns_update.c code, this option will
> >> require minimal support,
> > 
> > It will require even less support if it is removed, it was only used
> > because some distro's didn't have a version of bind9 that supported
> > dlz, these distro's are now EOL.
> > 
> > 
> >> it will not break running installations, I
> >> can imagine minimal config template modification if BIND developers
> >> change something again (will it ever happen?).
> >> I'm talking about DEPRECATED -> NOT RECOMMENDED and describing this
> >> in the docs.
> >> 
> >> Or if you really-really want to remove this, we could at least
> >> write some info about possibility of using bind9 with the flat
> >> files in the docs/mans/wiki.
> >> 
> > 
> > Why would we want to tell people about using something that has been
> > removed ?
> 
> We are removing config generation, not an ability to use side dns 
> server.
> Until we stop using nsupdate for records update, we may use another 
> server with tkey-gss support (or even without it).
> 
> > 
> >> Also, I have tested the patch before sending - it works for me.
> > 
> > Yes, it may work in the way you perceive to be correct, but it
> > isn't a supported way and probably isn't working as AD requires.
> 
> I really don't understand what "AD requires" means, except ACLs. 
> Incorrect ACLs in the update-policy could lead to security holes 
> ("administrator" is not dns/domain admin in LDAP) - and that is the 
> reason to not support PLAIN in the base installation (among other 
> reasons you mentioned). But even MS AD could be run with dns security 
> updates disabled (you may see that the first try of updating
> dns-record from windows machine goes without auth).
> 
> I will not be surprised if one of my installations is the only one
> with the flat files :) So, if you think this should be removed, I
> don't mind. But we might mention this possibility in docs (shortly,
> even without examples):
> "Samba AD could potentially be used with any dns-server that supports 
> tkey-gss updates (and even without such support), but samba team
> doesn't recommend this and doesn't provide support. This may bring
> security risks."
> This schema is working, until we remove nsupdate support (actually
> even after it - static records). And by this text, we will shield
> users from wrong decisions, which they could make even without our
> support in the provision script.
> 
> Just MHO, thanks.
> 
> > 
> > Rowland
> > 
> >> 
> >> >
> >> >> We can do this on a more accelerated schedule than normal
> >> >> because existing installations, unchanged, won't notice
> >> >> anything.
> >> >>
> >> >> What do you think?  Alternately it might be simpler to just:
> >> >>
> >> >> * Remove the provision and dns_update.c code from master for
> >> >> 4.10, leave 4.9 alone.  Bypass our normal feature deprecation
> >> >> rules because existing installs will continue to just work, add
> >> >> a note to WHATSNEW.
> >> >>
> >> >
> >> > No, I think we should stick to the deprecate, then remove method,
> >> > just do it faster this time.
> >> >
> >> > Rowland
> >> 
> >> 
> >> ---
> >> Best regards,
> >> Sergey Urushkin

Have you ever heard of the term 'Flogging a dead horse' ???

Rowland