Wow. Real problem. I check it - it allows one tenant to interrupt
traffic on other. I was not able to intercept TCP traffic, victim lost
connection and TCP was struggling with retransmissions, but it was not
good.
But idea of 'no routers in software' is too appealing. I think I'll
stick with this idea (and wait for fix). Plus, I think, OVS can provide
some kind of excellent antispoofing mechanism (OVS can use nw_src/nw_dst
for arp filtering) - if no fix will be available, I'll do some crude fix
on neutron/nova with OVS (I don't really like etables).
On 01/09/2015 09:25 PM, Kris G. Lindgren wrote:
Also, If you are running this configuration you should be aware of the
following bug:
https://bugs.launchpad.net/neutron/+bug/1274034
And the corresponding fix: https://review.openstack.org/#/c/141130/
Basically - Neutron security group rules do nothing to protect against arp
spoofing/poisoning from vm's. So its possible under a shared network
configuration for a vm to arp for another vm's ip address and temporarily
knock that vm offline. The above commit - which is still a WIP adds
ebtable rules to allow neutron to filter protocols other than IP (eg arp).
____________________________________________
Kris Lindgren
Senior Linux Systems Engineer
GoDaddy, LLC.
On 1/8/15, 9:50 AM, "Kris G. Lindgren" <klindg...@godaddy.com> wrote:
On 1/8/15, 4:34 AM, "Antonio Messina" <antonio.s.mess...@gmail.com> wrote:
On Thu, Jan 8, 2015 at 12:12 PM, gustavo panizzo (gfa) <g...@zumbi.com.ar>
wrote:
On 01/08/2015 07:01 PM, Antonio Messina wrote:
On Thu, Jan 8, 2015 at 11:53 AM, gustavo panizzo (gfa)
<g...@zumbi.com.ar>
wrote:
i may be wrong as i haven't tested that on juno, but in icehouse and
havana
i've setup external/provider networks one for each tenant
Ah, ok, this is the point. What I would like to have instead is
1) one big external network with routable, private IPs, to be used by
*any*
tenant (where any tenant can plug ports)
neutron net-create --shared should do the trick
I guess the problem is that I was creating *external* _and_ *shared*
network, but if I don't want to use floating IPs from that network I
probably don't need the network to be external, right?
Correct. We have our "external networks" (private ip) only configured as
shared so any vm from any tenant can create a port on it. External means
it's meant to be plugged into a router/l3 agent.
3) small external networks dedicated to a tenant
neutron net-create --tenant-id XXXXX-XXXXXX
i've made that mix (also add tenantn networks) in my lab running
icehouse
(2014.1.2) worked fine, i've upgraded it to juno but i haven't test
that yet
Thank you, I will test it further.
do you run more than one l3 agent? are you floating ips configured on
br-ex?
i think if you fix the policy.json on nova you should get it working
I currently have only 1 l3-agent running, but it's suboptimal. I would
like to have L3-HA for the floating IPs and DVR for the shared and
tenant networks, but as far as I know it's not currently supported. I
plan to deploy the production cloud after Kilo release, and crossing
my fingers...
Floating IPs are currently configured on br-ex.
You have floating ips working under this configuration?
If so then that means you are using the neutron router as the gateway for
all of your vm's and not the gateway provided by your network device. As
soon as you created a router and attached the the shared network to it the
router got configured with the gateway address that you configured in your
network. You may want to make sure that you don¹t have traffic flapping
between your real network gateway and your neutron gateway on your l3
router.
The reason why floating ip's won't work under this configuration (using a
real network device for the gateway) is the fact that the floating ip is
applied at the router as a nat so the traffic flows like: Client (4.2.2.1)
-> floating ip (8.8.8.8) -> external port on router -> The router changes
the destination IP from the floating ip to the private IP of the vm
(172.23.2.40) and send the traffic out to the vm via the routers
connection on the shared network. The VM responds to the client traffic
which is not in its subnet directly to the default gateway (172.23.0.1).
The gateway (if it is a neutron gateway) will then undo the nat and send
the traffic back to the client.
Where this breaks down with a physical network gateway is when the traffic
>from the VM is sent to the real network gateway - the gateway doesn't know
anything about the nat and does not re-map the vm ip to the floating ip.
IE The client ip never gets changed to the floating ip and sent back to
the client IP. The Response would be seen as coming directly from
172.23.2.40, instead of 8.8.8.8. So you will never establish a tcp
connection.
If you have this working somehow - please let me know. When we attempted
to run this configuration we ran into the above issues with floating ips
when using a real gateway on a network device.
Note: this is also why you need to create your shared network with a
specific router to the metadata service. Most likely on your real network
gateway you haven't added a route for 169.254.169.254 to a specific ip.
.a.
--
antonio.s.mess...@gmail.com
antonio.mess...@uzh.ch +41 (0)44 635 42 22
S3IT: Service and Support for Science IT http://www.s3it.uzh.ch/
University of Zurich
Winterthurerstrasse 190
CH-8057 Zurich Switzerland
_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
_______________________________________________
OpenStack-operators mailing list
OpenStack-operators@lists.openstack.org
http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
