On 5/24/19 12:21 PM, Andrija Panic wrote:
In other words - you are hitting an internal interface of a VR?
The VR has two NIC's. I presume that the Guest NIC as vs the Control NIC
is the "internal" NIC?
Type Shared
Traffic Type Guest
Network Name Shared
Netmask 255.255.0.0
IP Address 10.102.199.148
ID 7f59d904-cdc0-43eb-b679-0721077f5bb1
Network ID 924eda5f-9a1f-4a8e-9423-18000dc92093
Isolation URI vlan://102
Broadcast URI vlan://102
Type
Traffic Type Control
Network Name
Netmask 255.255.0.0
IP Address 169.254.2.203
ID 9c3676bc-23e6-48e3-baca-b8cce6511092
Network ID 6eff5bd9-4f4d-48fe-b6ed-f50fc115947b
Isolation URI
Broadcast URI
I would replace (for a test) bind9 with just the default setup of
DNSmasq, while specifying it's uper/ROOT DNS servers to be the VR IP -
i.e. client --> DNSmasq (internal server) --> DNSmasq (VR).
See if that work - so you can draw possibly some conclusions.
That gives me room for some more experiments. I am fairly sure that I am
running into recent changes to bind9 / dnsmasq intended to prevent DNS
amplification and spoofing attacks, but the question of which one
changed and how to work around it is still a question I'm trying to answer.
Andrija
On Fri, 24 May 2019 at 21:12, Eric Lee Green <eric.lee.gr...@gmail.com
<mailto:eric.lee.gr...@gmail.com>> wrote:
On 5/24/19 10:16 AM, Andrija Panic wrote:
> Eric,
>
> your BIND9 servers is on a "Public" network (trying to talk to
the Public
> IP of the VR during forwarding DNS requests) or a VM inside an
Isolated
> network behind VR)?
It's on *a* public network, but not *the* public network. I don't
have
any Isolated networks, though I have them enabled from VLAN
1000-2000. I
am using "Advanced Networking" but for my own purposes -- I have one
"Shared" guest network at VLAN 102, and then several isolated
specialty
physical "Shared" networks like "Security Cameras" (VLAN 103) and
"Storage Network" (VLAN 200) that are attached to virtual machines
that
need access to those things. The "Shared" guest network (VLAN 102) is
routed by my layer 3 switch with the rest of my network's public
VLANs
so if I am on e.g. 10.31.1.2 (VLAN 31), which is similarly a routed
public VLAN (but not one that Cloudstack is allowed to directly
talk to
or manage, it has to go thru the layer 3 switch) or 10.120.0.5 (VLAN
120), I can talk directly to 10.102.199.148 since all are routed into
the common fabric via the layer 3 switch. I only care about the VM's
that are VLAN 102, which are supposed to be publicly available to my
users, thus why my quicky script hack to generate a zone file out
of the
database does
select v.name <http://v.name>, n.ip4_address from vm_instance as
v, nics as n where v.removed is null and n.instance_id = v.id
<http://v.id> and n.ip4_address like '10.102.%' and type =
'User' order by n.ip4_address;
in order to select out the name and IP address of virtual machines
with
NIC's on that VLAN. (Which, if it's a different list from the last
list
that was queried, then gets massaged into a zone file for
name.cloud.mydomain.com <http://name.cloud.mydomain.com> by the
script, which then scp's to my master
domain server and does a reload to reload the zone file from the new
version).
Both of my BIND9 servers can talk directly to 10.102.199.148 (the
IP of
the virtual router for the 10.102.xxx.xxx network, VLAN 102) if I use
'host' to directly query 10.102.199.148 for an API address like, say,
'api-default1.cloud.mydomain.com
<http://api-default1.cloud.mydomain.com>' but when I try to put a
forward domain
there, nope. This was working, but now is not. I suspect it's got
to do
with the recent changes in DNS software, both bind9 and dnsmasq, to
deal with multiple attacks on the domain name system, but I'm having
trouble figuring out why, or what my solution should be.
Note that it's quite reasonable / feasible / viable to put a DNS
server
actually inside the Cloudstack constellation if that's necessary and
then do a two-stage hop if necessary. I'm just trying to figure
out the
"right" way to do this right now so I can retire my hack script.
> On Fri, 24 May 2019 at 02:15, Eric Lee Green
<eric.lee.gr...@gmail.com <mailto:eric.lee.gr...@gmail.com>>
> wrote:
>
>> I had this working under 4.9. All I did was, on my main BIND9
servers,
>> point a forward zone at 'cloud.<mydomain>.com' to the virtual
router
>> associated with all VM's that were publicly available. I could then
>> resolve all foo.cloud.<mydomain>.com names on my global network.
>>
>> Somehow, though, this quit working after I updated to 4.11. I'm not
>> quite sure why.
>>
>> The 'Guest Network' is defined with domain 'cloud.mydomain.com
<http://cloud.mydomain.com>'.
>>
>> Okay, so my router for the 'Guest Network' advanced networking is
>> located at 10.102.199.148. In my master BIND9 DNS server at
10.31.1.2 I
>> have this:
>> zone "cloud.mydomain.com <http://cloud.mydomain.com>" IN {
>> type forward;
>> forward only;
>> forwarders {
>> 10.102.199.148;
>> };
>> };
>>
>> If I send a NAMED request directly to the virtual router while
logged
>> into my main name server, it works:
>>
>> [root@ypbind ~]# host eric-gui.cloud.mydomain.com
<http://eric-gui.cloud.mydomain.com> 10.102.199.148
>> Using domain server:
>> Name: 10.102.199.148
>> Address: 10.102.199.148#53
>> Aliases:
>>
>> eric-gui.cloud.mydomain.com
<http://eric-gui.cloud.mydomain.com> has address 10.102.199.234
>>
>> If I try to use the name server however, it doesn't work:
>>
>> [root@ypbind logs]# host eric-gui.cloud.mydomain.com
<http://eric-gui.cloud.mydomain.com>
>> Host eric-gui.cloud.viakoo.com
<http://eric-gui.cloud.viakoo.com> not found: 3(NXDOMAIN)
>>
>> I'm baffled, because this *was* working.
>>
>> So I disabled any dnssec in the {options} on bind9 and gave all
>> permissions to see if that was the problem (note that this is
internal
>> to my infrastructure, so DNS amplification isn't an issue):
>>
>> dnssec-enable no;
>> dnssec-validation no;
>> dnssec-lookaside auto;
>> recursion yes;
>> allow-recursion { any; };
>> allow-query { any; };
>> allow-query-cache { any; };user
>>
>> Still nope. Still baffled.
>>
>> Anybody got any clues as to what I may be doing wrong? I'm
thinking it
>> has to be on the BIND9 side, because I can resolve the host
name if I
>> talk to the virtual router directly, but for some reason it's not
>> allowing me to get any records from the router.
>>
>> Right now I've temporarily worked around this with a script that
>> directly queries the MySQL database every few minutes and
generates a
>> revised zone file on my master DNS server when the list of virtual
>> machines queried out of the database changes, but that's
clearly not the
>> right way to do it. The question is, what *is* the right way to
do it?
>>
>> -Eric
>>
>>
>>
--
Andrija Panić