VEs and HNs in same subnets
This describes a method of setting up networking for a host and its VEs such that the networking configuration for the VEs can be configured exactly as if the VEs were standalone hosts of their own in the same subnets or VLAN as the host. This method makes use of the Virtual Ethernet device and bridges between the host and its containers. This technique has the advantage of allowing IPv6 network configurations to work on both VEs and hosts as they normally would. In particular, both hosts and VEs can use IPv6 autoconfiguration. The network configuration of a VE can be identical to that of a non-VE system.
In the following example the host has two physical interfaces and we are setting up the network configuration for VE 100. The host IP configuration is moved out of the ethN interface configs and into the brN interface config scripts (ifcfg-br0 and ifcfg-br1). Ie. the host IP configuration will now reside on the brN interfaces instead of the ethN interfaces. The example also assumes IPv4 is configured statically, whereas IPv6 is auto-configured.
Configure host bridge interfaces
Steps 1 through 4 are done only once on the host.
1. (Optional) Verify that you can create a bridge interfaces for each physical interface on the host.
/usr/sbin/brctl addbr br0 /usr/sbin/brctl addbr br1
If the above commands do not work you may need to install the bridge-utils package.
2. Make note of the existing IP configuration in the hosts ifcfg-ethN files. Also, record the hardware MAC addresses of the ethernet interfaces from the output of 'ifconfig'.
/sbin/ifconfig eth0 /sbin/ifconfig eth1
Then, modify the ifcfg-ethN files on the host so that they ONLY bridge to the corresponding brN interface. /etc/sysconfig/network-scripts/ifcfg-eth0 should look like:
DEVICE=eth0 BOOTPROTO=none ONBOOT=yes BRIDGE=br0
Similarly ifcfg-eth1 will look like:
DEVICE=eth1 BOOTPROTO=none ONBOOT=yes BRIDGE=br1
Note that the ifcfg-ethN files on the host do not contain any IP information anymore.
3. Create ifcfg-brN files and copy the IP configuration that was previously in the ifcfg-ethN files into ifcfg-brN. Here's what host:/etc/sysconfig/network-scripts/ifcfg-br0 would look like assuming the IPv4 address is assigned statically and IPv6 auto-configuration (SLAAC) is used:
DEVICE=br0 BOOTPROTO=static IPADDR=xxx.xxx.xxx.xxx NETMASK=aaa.aaa.aaa.aaa ONBOOT=yes TYPE=Bridge MACADDR=mm:mm:mm:mm:mm:mm
Similarly, ifcfg-br1 should look like:
DEVICE=br1 BOOTPROTO=static IPADDR=yyy.yyy.yyy.yyy NETMASK=bbb.bbb.bbb.bbb ONBOOT=yes TYPE=Bridge MACADDR=nn:nn:nn:nn:nn:nn
Note that TYPE 'Bridge' is case-sensitive. Otherwise, the bridge interfaces will not initialize correctly during boot.
The bridge MACADDR should be hard-coded to match the corresponding hardware MAC address of the ethernet interface. Otherwise the default behaviour is to use the lowest MAC address of all the interfaces in the bridge. This is to prevent the bridge MAC and any auto-configured IPv6 address on the bridge interface from changing as VEs are created, started, or stopped.
4. On the host, do a 'service network restart' and verify the host has both IPv4 and IPv6 connectivity to its brN interfaces.
Create the VE veth interfaces
5. Create the VE as you normally would, except do NOT specify any IP address, just the hostname. Specifying an IP address during VE creation creates an unwanted venet interface which is not used in this configuration.
/usr/sbin/vzctl create 100 --ostemplate name --hostname name
However, if the VE already exists, use vzctl to remove any venet devices - they will not be used:
/usr/sbin/vzctl set 100 --ipdel all --save
6. For each VE, create ethN devices (ignore warnings about "Container does not have configured veth") on the host:
/usr/sbin/vzctl set 100 --netif_add eth0 --save /usr/sbin/vzctl set 100 --netif_add eth1 --save
The above updates the host /etc/vz/conf/100.conf file with generated MAC addresses for the veth devices. When the VE is started, the veth100.0 and veth100.1 devices will be automatically created on the host.
Bridge the host and VE
7. Next we add the host vethN interfaces to the host bridged interfaces (brN).
DEVICE=veth100.0 ONBOOT=no BRIDGE=br0
DEVICE=veth100.1 ONBOOT=no BRIDGE=br1
To make the above take effect, either start the VE,
/usr/sbin/vzctl start 100
Or if it's already started then manually add each VE interface to its corresponding bridge using:
/usr/sbin/brctl addif br0 veth100.0 /usr/sbin/brctl addif br1 veth100.1
8. Verify each bridge includes the host interface and the veth interfaces for each VE:
Configure the VE networking
9. Enter the VE from the host:
/usr/sbin/vzctl enter 100
In the container create the ifcfg network scripts for each interface eth0 and eth1. The ifcfg-ethN files should look like standard ifcfg network scripts for a non-VE host.
vi /etc/sysconfig/network-scripts/ifcfg-eth0 vi /etc/sysconfig/network-scripts/ifcfg-eth1
As noted above, the ifcfg-ethN files in the VE should be created to be identical to standard ifcfg-eth* files from a non-virtualized host. A minimum ifcfg-eth0 file using a static IPv4 address would have the following entries:
DEVICE=eth0 BOOTPROTO=static IPADDR=xxx.xxx.xxx.xxx NETMASK=yyy.yyy.yyy.yyy ONBOOT=yes GATEWAY=zzz.zzz.zzz.zzz
10. Initialize the interfaces and restart the network service on the container.
/sbin/ifconfig eth0 0 /sbin/ifconfig eth1 0 /sbin/service network restart
Alternatively, just restart the VE from the host.
11. Verify the host and VE have connectivity to each other as well as to the rest of the network.
12. For each additional VE, start at step #5.
Notes on IPv6 autoconfiguration
If your CT0 is also performing routing duties, you might chance upon the problem that IPv6 stateless autoconfiguration using radvd is not working for the CT's. The description below is specific for a Debian (Lenny) CT0 and Debian (Squeeze) CT.
First check if your CT is actually receiving any router advertisements (RA's). This can be done by installing radvd (apt-get install radvd) and running radvdump. Simply wait for the next round of RA's from radvd, or trigger it (by restarting radvd on CT0, for example). If you do not receive any RA's there is a more fundamental problem. The following only concerns the scenario where the CT receives RA's but does not configure the network interfaces accordingly. Do not forget to remove the radvd package after checking.
Because CT0 is performing routing services, all or some values under /proc/sys/net/ipv6/conf/*/forwarding and under /proc/sys/net/ipv6/conf/*/mc_forwarding are set to 1. This appears to override the defaults for these values in the /proc filesystems for the CT's. Unless you explicitly disable forwarding in /etc/sysctl.conf, your CT will also use these values. This means that, to the IPv6 Neighbour Discovery Protocol (NDP) responsible for router advertisements and autoconfiguration, your CT is a router and therefore not allowed to use the RA's to configure the interfaces. To fix this, add or change the following lines to /etc/sysctl.conf on the CT, not on CT0:
You may also want to explicitly disable IPv4 forwarding since the CT is not a router. To do this, also change the line:
Now reload sysctl on the CT by executing
and you will be good to go. The CT will now autoconfigure the network interfaces the next time it sees an RA.
NOTE: Due to bug 1723 this setup might not work: Enabling the routing on CT0 can effectively kill all IPv6 connectivity for the CT, depending on the setup. (This bug is reported to be solved since 2011-06-07, so this shouldn't be an issue anymore.)