Open Stack

Hi Aaron,

I missed the latest mail of this thread in my mailbox, so pasting the
content of your response below.


*The issue is that the nova-api says that by default every instance needs
tobe in a default security group that blocks all ingress traffic from
outsideand allows all ingress from instances that belong to the same
securitygroup. If an instance does not have an ip address we are unable to
enforcethe instance->instance communication of members that are part of the
samesecurity group (or at least the iptables implementation in place
doesn't).There is an extension in neutron that implements port_security
that allowsone to disable this so that one can do L2 networking as you want
to. Thoughunfortunately, this does not work correctly at this time with
nova ascurrently it's calling into neutron to create the security group and
attachit to the instance anways. See:
https://bugs.launchpad.net/nova/+bug/1175464
<https://bugs.launchpad.net/nova/+bug/1175464> .I'm hoping to resolve this
issue in the next few weeks.*

*There is an extension in neutron that implements port_security that allows*
*one to disable this so that one can do L2 networking as you want to.
Though*
*unfortunately, this does not work correctly at this time with nova as*
*currently it's calling into neutron to create the security group and
attach*
*it to the instance anways. See:
https://bugs.launchpad.net/nova/+bug/1175464
<https://bugs.launchpad.net/nova/+bug/1175464> .*

*I'm hoping to resolve this issue in the next few weeks.*

Thanks a lot for the information.

As per my understanding of the bug description and the code diff, the
fix would prevent nova from adding 'default' security group if neutron
is not specifying any security group, only for the case where there
are plugins that implement port_security_api and have
port_security_enabled=False

I have two concerns here:

1. If we don't have any such plugins, then port_security is assumed to
be enabled and again default security group would be added by nova,
which mandates subnet and L2 networks does not work as expected.

2. If neutron provides security group(non default) via its own api or
by means of security plugins, L2 networks still fail due to the same
subnet issue.

You may ask why security group is needed for L2 networks as the
security rules are actually converted to iptables. The reason is that
the vNIC may have ipaddress at later point of time and we can add
security rules based on this ip then. I am just wondering why should
the subnet be mandatory for the vNIC during the instance boot time.

Regards,

Kanthi



On Fri, Nov 8, 2013 at 11:45 AM, Kanthi P <pavuluri.kanthi at gmail.com> wrote:

> Hi Aaron,
>
> Thanks for the reply !
> Yes security groups are mapped to L3/L4 headers, these security rules are
> being converted to iptables.
>
> If we remove the subnet check, we will be able to launch instances without
> ipaddress, they just have the mac address.
> Users can configure their own ipaddress after they are booted.
>
> If we use neutron security groups, it provides a default group (on port
> basis) which blocks all ingress and allows all egress traffic.
>
> Later users can configure security groups based on the ip address what
> they provided to the vnics.
>
> I mean to say, ports will have subnet but just that this subnet is not
> known to openstack during the instance boot time.
>
>
>
>
> On Fri, Nov 8, 2013 at 9:42 AM, Aaron Rosen <arosen at nicira.com> wrote:
>
>>
>>
>>
>> On Thu, Nov 7, 2013 at 12:23 PM, Kanthi P <pavuluri.kanthi at gmail.com>wrote:
>>
>>> Hi,
>>>
>>> I am trying to boot a VM which has a network without subnet in Havana,
>>> but it throws an exception saying that subnet is mandatory if quantum
>>> security groups are enabled.
>>>
>>> Here are the commands I used.
>>>
>>> neutron net-create testNet
>>> neutron net-show testNet
>>> +---------------------------+--------------------------------------+
>>> | Field                     | Value                                |
>>> +---------------------------+--------------------------------------+
>>> | admin_state_up            | True                                 |
>>> | id                        | 47208beb-2801-4729-bc7b-6532717232fc |
>>> | name                      | testNet                              |
>>> | provider:network_type     | local                                |
>>> | provider:physical_network |                                      |
>>> | provider:segmentation_id  |                                      |
>>> | router:external           | False                                |
>>> | shared                    | False                                |
>>> | status                    | ACTIVE                               |
>>> | subnets                   |                                      |
>>> | tenant_id                 | b5b591dcda2645cd9d15a4fe3eb1b50d     |
>>> +---------------------------+--------------------------------------+
>>>
>>> nova boot --flavor 2 --image 30c1984c-8a4f-4e3f-8c9a-7de0b321caa2 --nic
>>> net-id=47208beb-2801-4729-bc7b-6532717232fc testServer1
>>>
>>> Here is the piece of code which generated this exception
>>>
>>> nova/network/neutronv2/api.py
>>>
>>>             if (security_groups and not (
>>>                     network['subnets']
>>>                     and network.get('port_security_enabled', True))):
>>>
>>>                 raise exception.SecurityGroupCannotBeApplied()
>>>
>>>
>>> Can someone please explain why do we need this check?
>>>
>>
>> Hi Kanthi,
>>
>> We need this check because because in order to enforce security groups
>> the port needs to have an ip_address (i.e: network needs a subnet) since
>> Security groups only map to L3/4 headers. Today, nova automatically adds a
>> default security group to all instances when booted. Hopefully we can punt
>> this task off to neutron in this release by moving the port-creation up on
>> the stack to nova-api instead of nova-compute though this isn't the case
>> right now.
>>
>> Aaron
>>
>>>
>>> To my understanding subnet is used for two purposes in terms of security
>>> groups
>>> 1. To allow dhcp traffic if dhcp is enabled on the subnet, example given
>>> below
>>>         -A quantum-openvswi-i7bf776d2-b -s 192.168.1.3/32 -p udp -m udp
>>> --sport 67 --dport 68 -j RETURN
>>> 2. To avoid ip spoof
>>>         -A quantum-openvswi-o7bf776d2-b ! -s 192.168.1.2/32 -j DROP
>>>
>>> Can we remove this so that we can have guests which has nic with just
>>> MAC address, guest can configure its own ipaddress. Later if needed they
>>> can enable their own security rules via quantum api.
>>>
>>> _______________________________________________
>>> Openstack-security mailing list
>>> Openstack-security at lists.openstack.org
>>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-security
>>>
>>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20131210/00c8c15d/attachment.html>