Re: [Fwd: Re: request for info: unbound via https / tls]

[ummeegge <ummeegge@ipfire.org>]

[-- Attachment #1: Type: text/plain, Size: 11521 bytes --]

Hi Michael,


Am Mittwoch, den 12.12.2018, 15:25 +0000 schrieb Michael Tremer:
> Hey,
> 
> > On 12 Dec 2018, at 13:42, ummeegge <ummeegge(a)ipfire.org> wrote:
> > 
> > Hi Michael,
> > 
> > Am Dienstag, den 11.12.2018, 19:54 +0000 schrieb Michael Tremer:
> > > Hey,
> > > 
> > > On 11 Dec 2018, at 19:43, ummeegge <ummeegge(a)ipfire.org> wrote:
> > > > 
> > > > Hi Michael,
> > > > tried that now with this one -->
> > > > 
> > 
> > 
https://people.ipfire.org/~ummeegge/screenshoots/dns-over-tls_wui.png
> > > 
> > > This looks good, but under no circumstances should there be
> > > *another*
> > > place where to configure DNS servers.
> > 
> > Sure. I need to check for myself how this can be accomplished so i
> > take
> > it step-by-step and with a clear CGI it is simply easier for me.
> > 
> > 
> > > We already have three. They need to be unified to one.
> > 
> > You mean dns.cgi and dnsforward.cgi ?
> 
> No. We have the following places where users can configure their DNS
> servers:
> 
> For RED = STATIC: setup
> 
> For RED = DHCP: dns.cgi
> 
> For RED = PPP: The PPP profile
> 
OK i see. This is also a problem for me here to test or to check
further since RED is here static only.

> It can be argued that it makes sense to have different DNS servers in
> place for each PPP profile (I do not think that this has any users at
> all. Either you use your ISP’s DNS or not).
> 
> But overall it is confusing and also quite complicated in the code
> that we are reading the DNS servers from so many different places all
> the time. This needs to be simplified. There is a ticket for it since
> 2015:
> 
>   https://bugzilla.ipfire.org/show_bug.cgi?id=10886

I think this topic might be a good reason to change this but as
mentioned am not able to do this here.

> 
> > 
> > > 
> > > > 
> > > > ... the HTML formatting kills me :D ...
> > > > 
> > > > and it looks now good:
> > > > 
> > > > $ kdig -d @81.3.27.54 +tls-ca=/etc/ssl/certs/ca-bundle.crt
> > > > +tls-
> > > > host=rec1.dns.lightningwirelabs.com google.com
> > > > ;; DEBUG: Querying for owner(google.com.), class(1), type(1),
> > > > server(81.3.27.54), port(853), protocol(TCP)
> > > > ;; DEBUG: TLS, imported 129 certificates from
> > > > '/etc/ssl/certs/ca-
> > > > bundle.crt'
> > > > ;; DEBUG: TLS, received certificate hierarchy:
> > > > ;; DEBUG:  #1, CN=rec1.dns.lightningwirelabs.com
> > > > ;; DEBUG:      SHA-256 PIN:
> > > > pOvVkJSj6rWNPM0vR3hoJr/21kZI6TfImhowIEdcEUQ=
> > > > ;; DEBUG:  #2, C=US,O=Let's Encrypt,CN=Let's Encrypt Authority
> > > > X3
> > > > ;; DEBUG:      SHA-256 PIN:
> > > > YLh1dUR9y6Kja30RrAn7JKnbQG/uEtLMkBgFF2Fuihg=
> > > > ;; DEBUG: TLS, skipping certificate PIN check
> > > > ;; DEBUG: TLS, The certificate is trusted. 
> > > > ;; TLS session (TLS1.2)-(ECDHE-ECDSA-SECP256R1)-(CHACHA20-
> > > > POLY1305)
> > > > ;; ->>HEADER<<- opcode: QUERY; status: NOERROR; id: 1349
> > > > ;; Flags: qr rd ra; QUERY: 1; ANSWER: 1; AUTHORITY: 0;
> > > > ADDITIONAL:
> > > > 1
> > > > 
> > > > ;; EDNS PSEUDOSECTION:
> > > > ;; Version: 0; flags: ; UDP size: 4096 B; ext-rcode: NOERROR
> > > > 
> > > > ;; QUESTION SECTION:
> > > > ;; google.com.         		IN	A
> > > > 
> > > > ;; ANSWER SECTION:
> > > > google.com.         	151	IN	A	216.58.208.46
> > > > 
> > > > ;; Received 55 B
> > > > ;; Time 2018-12-11 20:30:29 CET
> > > > ;; From 81.3.27.54(a)853(TCP) in 25.2 ms
> > > 
> > > 25ms is actually quite good!
> > 
> > Yes, i think so.
> > 
> > > 
> > > > 
> > > > Great, will update my dot.conf. 
> > > > 
> > > > As a beneath one, try it currently with a seperat CGI to have a
> > > > better overview. 
> > > > Patched now as you suggested the 'write_forward_conf()'
> > > > function,
> > > > needed to disable 
> > > > nevertheless update_forwarder() function in initscript if
> > > > forward.conf should be used
> > > > ... (there is more)
> > > > 
> > > 
> > > As we talked about before, I think that we can skip the DNSSEC
> > > tests
> > > entirely. They are more damaging than anything else. 
> > 
> > Yes indeed, i think update_forwarders disables also any forwarder
> > via
> > unbound-control.
> 
> Disables them? It is meant to overwrite them with the current DNS
> servers without restarting unbound.
I meant the update_forwarder() function -->
https://git.ipfire.org/?p=ipfire-2.x.git;a=blob;f=src/initscripts/system/unbound;h=cc46c33c9425cc85d95b1d7412a9db3e146fea4b;hb=refs/heads/next#l154
as far i understand 'unbound-control -q forward off' .

But also for write_forward_conf() where i integrated DoT needed a restart
of unbound as far as i can see/tested since we need different unbound.conf
directives:

server:
  tls-cert-bundle: /etc/ssl/certs/ca-bundle.crt

forward-zone:
  forward-tls-upstream: yes

unbound needed to read out the config file again. The first version
started DoT via 'unbound-control start' but i couldn´t manage this in
this version?!

> 
> > 
> > > That means that we should probably be looking at having a switch
> > > somewhere that enables DNS-over-TLS first and then all configured
> > > name servers are just used without further tests.
> > 
> > Have tried it now in this way -->
> > 
https://git.ipfire.org/?p=ipfire-2.x.git;a=blob;f=src/initscripts/system/unbound;h=cc46c33c9425cc85d95b1d7412a9db3e146fea4b;hb=refs/heads/next#l154
> > . If unbound init finds an 'on' (enabled) in tlsconfig (which will
> > be produced by CGI),
> > it doesn´t execute update_forwarders. Am currently not sure if we
> > need possibly
> > the same for dnsforward config. Have tested it with a dummy entry
> > but an
> > 'unbound-control list_forwarders' shows nothing.
> > If there is no entry or everything is 'off' unbound uses the old 
> > DNS servers configured via setup.
> 
> In case of “off”, unbound is supposed to run in recursor mode.

Haven´t experienced this here. Did You ? unbound switched only back to
the initially configured DNS servers if "off" or not presant.


> 
> > > In the default configuration that cannot be the case because of
> > > the
> > > problems we are trying to overcome by this script.
> > 
> > Isn´t forward.conf not a good place for this ?
> 
> For what exactly?

For extending unbound.conf .

> 
> > > 
> > > But Erik, please let’s find a strategy first because everything
> > > is
> > > being implemented. 
> > 
> > Am happy with this but i really need to know first what´s happen in
> > the
> > existing stuff, also i need to test for myself which ways may be
> > possible to overcome side affects. I need there also some new
> > knowledge
> > causing the whole DNS/unbound thing but also insides how all that
> > has
> > already been implemented.
> > 
> > In here --> 
> > 
https://git.ipfire.org/?p=people/ummeegge/ipfire-2.x.git;a=commit;h=90e45d849e5fa185e4dcf83844e85d68474a09f5
> > a first and also better tested version of DoT can be found whereby
> > i am 
> > happy if someone comes around for some testings/enhancements.
> > 
> > Merging all DNS CGI´s can be one of the following parts (not sure
> > if i´ am the right one for this)
> > but i need a working solution to see how the system is in harmony
> > with all that.
> > Also the dnsforwarding.cgi is in my humble opinion currently not
> > working
> > or i did there really something wrong.
> > 
> > What strategy would you prefer ?
> 
> I do not have one yet. I am just saying that there should be one and
> that it should be discussed before it is being implemented.

I see.

> 
> I can say that I have a couple of things that are not working for me.
> That would be that I do not think that an extra CGI is required. We
> can use the one that we have (although for testing it can be
> implemented where ever you prefer) and that DNS-over-TLS cannot be
> enabled by default.

update_forwarders() was the main problem here in both developmant
stages. Anyways, DoT works for me good some testing scenarios and the
resulting informations are available for further development but sadly
i can not test all that but can offer the already working CGI and
sureley some trial and errors if this is needed.

> 
> I also have a long list of issues that I would like to see tackled
> here as well. Those are that the unbound initscript is not behaving
> as intended. Extending it has to be done very careful to not break it
> even more or we make it shorter first and boil the problems down
> first and then add DNS-over-TLS.

That´s plausible. Since i done also some testings in unbound
initscript, i can may try to do also some work in there but as you
already said, this would really need coordination/description of
what/how to do.


> 
> > > I am absolutely happy that you are doing such good work here, but
> > > keep in mind that this needs to be integrated into IPFire in a
> > > slow
> > > and peer-reviewed way.
> > 
> > Need to think about how we can split things here. Do you have some
> > ideas ?
> 
> What is there to do?
Currently not sure how to find here a way of a release structure in a
peer-reviewed way.

> 
> It looks like you have a working version of the initscript and the UI
> (almost?) done.

Yes, this version works for me. Installed it on a fresh installation
(OpenSSL-1.1.1a ;), tried to kill/test it on an old installation but it
worked so far (a little slow since it restart unboundctrl if things are
changing in there).
Haven´t find bugs in the UI since now may someone have some for me ?

> 
> > Another thing i have in account is the QNAME minimisation -->
> > https://tools.ietf.org/html/rfc7816
> > even in unbound.conf 'qname-minimisation: yes' is active it didn´t
> > worked for me:
> > 
> > $ dig txt qnamemintest.internet.nl +short
> > a.b.qnamemin-test.internet.nl.
> > "NO - QNAME minimisation is NOT enabled on your resolver :("
> > 
> > needed to add also
> > 
> >  qname-minimisation-strict: yes
> >  harden-below-nxdomain: yes
> 
> Please open a ticket for that.

OK, will do that.

> 
> Harden-below-nxdomain is deliberately disabled because it breaks
> loads of things and I do not consider it a correct solution.

Great info! 'harden-below-nxdomain: yes' is also not needed for QNAME
minimization.

> 
> > at earlier tests in my local.d conf to get an
> > 
> > a.b.qnamemin-test.internet.nl.
> > "HOORAY - QNAME minimisation is enabled on your resolver :)!"
> > 
> > . Should we extend unbound.conf or should i add this one in
> > forward.conf if DoT is active ? Or is this may not wanted ?
> > 
> 
> This has nothing to do with DNS-over-TLS. Therefore this should be
> handled independently.

Yes, as mentioned will open a ticket for this.

Another thing has come to mind. A kdig check for the configured DoT
servers might be nice. May like it is managed in ddns.cgi with the
syncronisation function.
For example a kdig command can be used to grep for "is trusted" and the
"rd" flag and if true to set a green colored hostname in (currently)
dnsovertls.cgi, if not a red colored one can help the user for better
understanding if/or_which one is not working so possible
missconfiguration can be detected easier since the number of DNS
servers are not limited and open for randomization ?

As a beneath one, Cloudflair offers TLS1.3 support since a couple of
days/weeks now.


Best,

Erik