linux - How to assign multiple public IP-Adresses for 2 KVM-Guests

I am new to this whole topic and I try for days now to figure out how to assign multiple public ip-addresses to KVM-guests through a KVM host. I found tons of examples howto get such a setup with 1 public IP running.

Here is my setup:
The Server has only one NIC/MAC and runs 2 KVM-Guests with apache(and other stuff). Both guest-environments are ubuntu server 11.10 and must run in separate VMs. The 5 public ip-addresses are used to handle SSL-certificates and other stuff. The first VM should use 3 of the 5 addresses/certificates. The second VM gets the rest. The apache-stuff is configured correctly.

I have tried a number of differend ways via iptables to route the traffic from the hosts NIC to the guest-NICs. In spite of the fact that one way was the right one but only wrong implemented, I leave the details untold to leave you unprepossessed. The question is: Whats the ideal way it should be done?

The following conditions should be met:

• Apache must get the original ip-address of the visitor


• Apache must know, what public-ip address was ultilized to use the right ssl-vhost


• The traffic must not be routed through a (reverse-)proxy on the host, since there are 2 other non-http-services, on other VM-guests, that should be accessible from public. And: Only sshd should listen directly on the host - nothing else


• Each VM should be able to access the internet directly.


• The network in the data-center is switched MAC-based. As I figured out, the only way to comunicate with the internet is through eth0 and its MAC-address.

If I would discard all the virtualization-stuff, this would be perfectly easy, as apache get the request directly from a specific ip-address.

I am open for any working solution.

Answer

Use a bridge on your dom0 (e.g. KVM Host) WAN interface. This requires installing bridge-utils package. Since this is Debian-based distro, you may configure it in /etc/network/interfaces:

iface eth0 inet manual

auto br_wan

iface br_wan inet dhcp
    # Assuming DHCP to get address, otherwise migrate all WAN connection options here
    #address 192.168.122.0
    bridge_ports eth0 tap_guest1
    bridge_stp      off
    bridge_maxwait  0
    bridge_fd       0
    pre-up ip tuntap add dev tap_guest1 user guest1 mode tap
    # This command is required if your ISP allocates static IPs depending on MAC address
    # You shouldn't use this but might be handy some time

    #pre-up sysctl -q -w net/ipv4/conf/tap_guest1/proxy_arp=1
    post-down ip tuntap del tap_guest1 mode tap

Pre-up commands set up TAP interface to connect your KVM guest to a bridge. Note that this setup allows to run kvm from non-privileged user guest1. Note that setting net.ipv4.ip_forward = 1 with sysctl might be usefull as well.

I have used ip tuntap command from iproute2 package. It's not yet documented in the Debian package but soon will be available in upstream's manual page. Since this package is installed on every Debian-based server, you won't need to install uml-utilities or openvpn package to just create these interfaces.

This approach sure lacks some elegance to manage lots of tap interfaces, because you'll need to create similar pre-up and post-down lines as for tap_guest1 interface. This can be fixed by writing additional scripts in /etc/network/pre-up.d and /etc/network/post-down.d. It is also a problem if you want to reconfigure br_wan interface with ifdown/ifup scripts while KVM guests are still running — you'll need either to remove all interfaces except eth0 from bridge configuration and detach them from bridge manually (don't forget to attach them back after bridge reconfiguration then) or shutdown all KVM instances running on a bridge.

Another way, perhaps more clean, is to write custom ifup script for KVM itself and use it in script option for your NIC. You can get an example in /etc/qemu-ifup. See kvm manual page for details.

Then you can run your KVM box like this:

kvm -net nic,model=virtio,macaddr=12:34:56:78:9a:bc \
    -net tap,ifname=tap_guest1,script=no,downscript=no \
    -boot c -nographic -display none -daemonize \
    guest1-drive.qcow2

Setting several IP addresses on one interface for your KVM guest can be done manually with command

ip address add aaa.bbb.ccc.101/24 dev eth0

Or permanently in /etc/network/interfaces like this:

auto eth0 eth0:1
iface eth0 inet static
    address aaa.bbb.ccc.100

    network aaa.bbb.ccc.0
    netmask 255.255.255.0
    broadcast aaa.bbb.ccc.255
    gateway aaa.bbb.ccc.1

iface eth0:1 inet static
    address aaa.bbb.ccc.101
    network aaa.bbb.ccc.0
    netmask 255.255.255.0
    broadcast aaa.bbb.ccc.255

    gateway aaa.bbb.ccc.1

Note that if your datacenter/provider does not expect you to reveal additional boxes on the same net he might not configure them and they will be unavailable. In this case you might want to create internal bridge and use iptables to forward packets between your WAN interface and this bridge using DNAT and SNAT. Assuming your local virtual bridge network is 10.0.0.0/8, your guest1 is 10.0.0.2 you'll need this:

iptables -t nat -A PREROUTING --dst aaa.bbb.ccc.100 -p tcp --dport 80 -j DNAT --to-destination 10.0.0.2
iptables -t nat -A PREROUTING --dst aaa.bbb.ccc.101 -p tcp --dport 80 -j DNAT --to-destination 10.0.0.2
...
iptables -t nat -A POSTROUTING -p tcp --dst 10.0.0.2 -j SNAT --to-source aaa.bbb.ccc.100

Note that you'll need as much DNAT commands as external IPs per KVM guest you have, but only one SNAT rule to give access to the internet. Also you can allow only HTTP/HTTPS/SSH traffic by allowing only desired ports. If you omit the --dport statement then all ports will be forwarded. Your KVM guest should have static network settings with KVM host as default gateway unless you're willing to host DHCP server.