Container-based network simulation

The “make test” framework provides a good way to test individual features. However, when testing several features at once - or validating nontrivial configurations - it may prove difficult or impossible to use the unit-test framework.

This note explains how to set up lxc/lxd, and a 5-container testbed to test a split-tunnel nat + ikev2 + ipsec + ipv6 prefix-delegation scenario.

OS / Distro test results

This setup has been tested on an Ubuntu 18.04 LTS system. If you’re feeling adventurous, the same scenario also worked on a recent Ubuntu 20.04 “preview” daily build.

Other distros may work fine, or not at all.

Proxy Server

If you need to use a proxy server e.g. from a lab system, you’ll probably need to set HTTP_PROXY, HTTPS_PROXY, http_proxy and https_proxy in /etc/environment. Directly setting variables in the environment doesn’t work. The lxd snap daemon needs the proxy settings, not the user interface.

Something like so:

    HTTP_PROXY=http://my.proxy.server:8080
    HTTPS_PROXY=http://my.proxy.server:4333
    http_proxy=http://my.proxy.server:8080
    https_proxy=http://my.proxy.server:4333

Install and configure lxd

Install the lxd snap. The lxd snap is up to date, as opposed to the results of “sudo apt-get install lxd”.

    # snap install lxd
    # lxd init

“lxd init” asks several questions. With the exception of the storage pool, take the defaults. To match the configs shown below, create a storage pool named “vpp.” Storage pools of type “zfs” and “files” have been tested successfully.

zfs is more space-efficient. “lxc copy” is infinitely faster with zfs. The path for the zfs storage pool is under /var. Do not replace it with a symbolic link, unless you want to rebuild all of your containers from scratch. Ask me how I know that.

Create three network segments

Aka, linux bridges.

    # lxc network create respond
    # lxc network create internet
    # lxc network create initiate

We’ll explain the test topology in a bit. Stay tuned.

Set up the default container profile

Execute “lxc profile edit default”, and install the following configuration. Note that the “shared” directory should mount your vpp workspaces. With that trick, you can edit code from any of the containers, run vpp without installing it, etc.

    config: {}
    description: Default LXD profile
    devices:
      eth0:
        name: eth0
        network: lxdbr0
        type: nic
      eth1:
        name: eth1
        nictype: bridged
        parent: internet
        type: nic
      eth2:
        name: eth2
        nictype: bridged
        parent: respond
        type: nic
      eth3:
        name: eth3
        nictype: bridged
        parent: initiate
        type: nic
      root:
        path: /
        pool: vpp
        type: disk
      shared:
        path: /scratch
        source: /scratch
        type: disk
    name: default

Set up the network configurations

Edit the fake “internet” backbone:

  # lxc network edit internet

Install the ip addresses shown below, to avoid having to rebuild the vpp and host configuration:

    config:
      ipv4.address: 10.26.68.1/24
      ipv4.dhcp.ranges: 10.26.68.10-10.26.68.50
      ipv4.nat: "true"
      ipv6.address: none
      ipv6.nat: "false"
    description: ""
    name: internet
    type: bridge
    used_by:
    managed: true
    status: Created
    locations:
    - none

Repeat the process with the “respond” and “initiate” networks, using these configurations:

respond network configuration

    config:
      ipv4.address: 10.166.14.1/24
      ipv4.dhcp.ranges: 10.166.14.10-10.166.14.50
      ipv4.nat: "true"
      ipv6.address: none
      ipv6.nat: "false"
    description: ""
    name: respond
    type: bridge
    used_by:
    managed: true
    status: Created
    locations:
    - none

initiate network configuration

    config:
      ipv4.address: 10.219.188.1/24
      ipv4.dhcp.ranges: 10.219.188.10-10.219.188.50
      ipv4.nat: "true"
      ipv6.address: none
      ipv6.nat: "false"
    description: ""
    name: initiate
    type: bridge
    used_by:
    managed: true
    status: Created
    locations:
    - none

Create a “master” container image

The master container image should be set up so that you can build vpp, ssh into the container, edit source code, run gdb, etc.

Make sure that e.g. public key auth ssh works.

    # lxd launch ubuntu:18.04 respond
    <spew>
    # lxc exec respond bash
    respond# cd /scratch/my-vpp-workspace
    respond# apt-get install make ssh
    respond# make install-dep
    respond# exit
    # lxc stop respond

Mark the container image privileged. If you forget this step, you’ll trip over a netlink error (-11) aka EAGAIN when you try to roll in the vpp configurations.

    # lxc config set respond security.privileged "true"

Duplicate the “master” container image

To avoid having to configure N containers, be sure that the master container image is fully set up before you help it have children:

    # lxc copy respond respondhost
    # lxc copy respond initiate
    # lxc copy respond initiatehost
    # lxc copy respond dhcpserver    # optional, to test ipv6 prefix delegation

Install handy script

See below for a handly script which executes lxc commands across the current set of running containers. I call it “lxc-foreach,” feel free to call the script Ishmael if you like.

Examples:

    $ lxc-foreach start
    <issues "lxc start" for each container in the list>

After a few seconds, use this one to open an ssh connection to each container. The ssh command parses the output of “lxc info,” which displays container ip addresses.

    $ lxc-foreach ssh

Here’s the script:

    #!/bin/bash

    set -u
    export containers="respond respondhost initiate initiatehost dhcpserver"

    if [ x$1 = "x" ] ; then
        echo missing command
        exit 1
    fi

    if [ $1 = "ssh" ] ; then
        for c in $containers
        do
            inet=`lxc info $c | grep eth0 | grep -v inet6 | head -1 | cut -f 3`
            if [ x$inet = "x" ] ; then
                echo $c not started
            else
                gnome-terminal --command "/usr/bin/ssh $inet"
            fi
        done
    exit 0
    fi

    for c in $containers
    do
        echo lxc $1 $c
        lxc $1 $c
    done

    exit 0

Test topology

Finally, we’re ready to describe a test topology. First, a picture:

    ===+======== management lan/bridge lxdbr0 (dhcp) ===========+===
       |                             |                          |
       |                             |                          |
       |                             |                          |
       v                             |                          v
      eth0                           |                         eth0
    +------+ eth1                                       eth1 +------+
    | respond | 10.26.88.100 <= internet bridge => 10.26.88.101 | initiate |
    +------+                                                 +------+
      eth2 / bvi0 10.166.14.2        |       10.219.188.2 eth3 / bvi0
       |                             |                          |
       | ("respond" bridge)             |          ("initiate" bridge) |
       |                             |                          |
       v                             |                          v
      eth2 10.166.14.3               |           eth3 10.219.188.3
    +----------+                     |                   +----------+
    | respondhost |                     |                   | respondhost |
    +----------+                     |                   +----------+
      eth0 (management lan) <========+========> eth0 (management lan)

Test topology discussion

This topology is suitable for testing almost any tunnel encap/decap scenario. The two containers “respondhost” and “initiatehost” are end-stations connected to two vpp instances running on “respond” and “initiate”.

We leverage the Linux end-station network stacks to generate traffic of all sorts.

The so-called “internet” bridge models the public internet. The “respond” and “initiate” bridges connect vpp instances to local hosts

End station configs

The end-station Linux configurations set up the eth2 and eth3 ip addresses shown above, and add tunnel routes to the opposite end-station networks.

respondhost configuration

    ifconfig eth2 10.166.14.3/24 up
    route add -net 10.219.188.0/24 gw 10.166.14.2

initiatehost configuration

    sudo ifconfig eth3 10.219.188.3/24 up
    sudo route add -net 10.166.14.0/24 gw 10.219.188.2

VPP configs

Split nat44 / ikev2 + ipsec tunneling, with ipv6 prefix delegation in the “respond” config.

respond configuration

    set term pag off

    comment { "internet" }
    create host-interface name eth1
    set int ip address host-eth1 10.26.68.100/24
    set int ip6 table host-eth1 0
    set int state host-eth1 up

    comment { default route via initiate }
    ip route add 0.0.0.0/0 via 10.26.68.101

    comment { "respond-private-net" }
    create host-interface name eth2
    bvi create instance 0
    set int l2 bridge bvi0 1 bvi
    set int ip address bvi0 10.166.14.2/24
    set int state bvi0 up
    set int l2 bridge host-eth2 1
    set int state host-eth2 up


    nat44 add interface address host-eth1
    set interface nat44 in host-eth2 out host-eth1
    nat44 add identity mapping external host-eth1 udp 500
    nat44 add identity mapping external host-eth1 udp 4500
    comment { nat44 untranslated subnet 10.219.188.0/24 }

    comment { responder profile }
    ikev2 profile add initiate
    ikev2 profile set initiate udp-encap
    ikev2 profile set initiate auth rsa-sig cert-file /scratch/setups/respondcert.pem
    set ikev2 local key /scratch/setups/initiatekey.pem
    ikev2 profile set initiate id local fqdn initiator.my.net
    ikev2 profile set initiate id remote fqdn responder.my.net
    ikev2 profile set initiate traffic-selector remote ip-range 10.219.188.0 - 10.219.188.255 port-range 0 - 65535 protocol 0
    ikev2 profile set initiate traffic-selector local ip-range 10.166.14.0 - 10.166.14.255 port-range 0 - 65535 protocol 0
    create ipip tunnel src 10.26.68.100 dst 10.26.68.101
    ikev2 profile set initiate tunnel ipip0

    comment { ipv6 prefix delegation }
    ip6 nd address autoconfig host-eth1 default-route
    dhcp6 client host-eth1
    dhcp6 pd client host-eth1 prefix group hgw
    set ip6 address bvi0 prefix group hgw ::2/56
    ip6 nd address autoconfig bvi0 default-route
    ip6 nd bvi0 ra-interval 5 3 ra-lifetime 180

    set int mtu packet 1390 ipip0
    set int unnum ipip0 use host-eth1
    ip route add 10.219.188.0/24 via ipip0

initiate configuration

    set term pag off

    comment { "internet" }
    create host-interface name eth1
    comment { set dhcp client intfc host-eth1 hostname initiate }
    set int ip address host-eth1 10.26.68.101/24
    set int state host-eth1 up

    comment { default route via "internet gateway" }
    comment { ip route add 0.0.0.0/0 via 10.26.68.1 }

    comment { "initiate-private-net" }
    create host-interface name eth3
    bvi create instance 0
    set int l2 bridge bvi0 1 bvi
    set int ip address bvi0 10.219.188.2/24
    set int state bvi0 up
    set int l2 bridge host-eth3 1
    set int state host-eth3 up

    nat44 add interface address host-eth1
    set interface nat44 in bvi0 out host-eth1
    nat44 add identity mapping external host-eth1 udp 500
    nat44 add identity mapping external host-eth1 udp 4500
    comment { nat44 untranslated subnet 10.166.14.0/24 }

    comment { initiator profile }
    ikev2 profile add respond
    ikev2 profile set respond udp-encap
    ikev2 profile set respond auth rsa-sig cert-file /scratch/setups/initiatecert.pem
    set ikev2 local key /scratch/setups/respondkey.pem
    ikev2 profile set respond id local fqdn responder.my.net
    ikev2 profile set respond id remote fqdn initiator.my.net

    ikev2 profile set respond traffic-selector remote ip-range 10.166.14.0 - 10.166.14.255 port-range 0 - 65535 protocol 0
    ikev2 profile set respond traffic-selector local ip-range 10.219.188.0 - 10.219.188.255 port-range 0 - 65535 protocol 0

    ikev2 profile set respond responder host-eth1 10.26.68.100
    ikev2 profile set respond ike-crypto-alg aes-cbc 256  ike-integ-alg sha1-96  ike-dh modp-2048
    ikev2 profile set respond esp-crypto-alg aes-cbc 256  esp-integ-alg sha1-96  esp-dh ecp-256
    ikev2 profile set respond sa-lifetime 3600 10 5 0

    create ipip tunnel src 10.26.68.101 dst 10.26.68.100
    ikev2 profile set respond tunnel ipip0
    ikev2 initiate sa-init respond

    set int mtu packet 1390 ipip0
    set int unnum ipip0 use host-eth1
    ip route add 10.166.14.0/24 via ipip0

IKEv2 certificate setup

In both of the vpp configurations, you’ll see “/scratch/setups/xxx.pem” mentioned. These certificates are used in the ikev2 key exchange.

Here’s how to generate the certificates:

    openssl req -x509 -nodes -newkey rsa:4096 -keyout respondkey.pem -out respondcert.pem -days 3560
    openssl x509 -text -noout -in respondcert.pem
    openssl req -x509 -nodes -newkey rsa:4096 -keyout initiatekey.pem -out initiatecert.pem -days 3560
    openssl x509 -text -noout -in initiatecert.pem

Make sure that the “respond” and “initiate” configurations point to the certificates.

DHCPv6 server setup

If you need an ipv6 dhcp server to test ipv6 prefix delegation, create the “dhcpserver” container as shown above.

Install the “isc-dhcp-server” Debian package:

    sudo apt-get install isc-dhcp-server

/etc/dhcp/dhcpd6.conf

Edit the dhcpv6 configuration and add an ipv6 subnet with prefix delegation. For example:

    subnet6 2001:db01:0:1::/64 {
            range6 2001:db01:0:1::1 2001:db01:0:1::9;
            prefix6 2001:db01:0:100:: 2001:db01:0:200::/56;
    }

Add an ipv6 address on eth1, which is connected to the “internet” bridge, and start the dhcp server. I use the following trivial bash script, which runs the dhcp6 server in the foreground and produces dhcp traffic spew:

    #!/bin/bash
    ifconfig eth1 inet6 add 2001:db01:0:1::10/64 || true
    dhcpd -6 -d -cf /etc/dhcp/dhcpd6.conf

The “|| true” bit keeps going if eth1 already has the indicated ipv6 address.

Container / Host Interoperation

Host / container interoperation is highly desirable. If the host and a set of containers don’t run the same distro and distro version, it’s reasonably likely that the glibc versions won’t match. That, in turn, makes vpp binaries built in one environment fail in the other.

Trying to install multiple versions of glibc - especially at the host level - often ends very badly and is not recommended. It’s not just glibc, either. The dynamic loader ld-linux-xxx-so.2 is glibc version specific.

Fortunately, it’s reasonable easy to build lxd container images based on specific Ubuntu or Debian versions.

Create a custom root filesystem image

First, install the “debootstrap” tool:

    sudo apt-get install debootstrap

Make a temp directory, and use debootstrap to populate it. In this example, we create an Ubuntu 20.04 (focal fossa) base image:

    # mkdir /tmp/myroot
    # debootstrap focal /tmp/myroot http://archive.ubuntu.com/ubuntu

To tinker with the base image (if desired):

    # chroot /tmp/myroot
    <add packages, etc.>
    # exit

Make a compressed tarball of the base image:

    # tar zcf /tmp/rootfs.tar.gz -C /tmp/myroot .

Create a “metadata.yaml” file which describes the base image:

    architecture: "x86_64"
    # To get current date in Unix time, use `date +%s` command
    creation_date: 1458040200
    properties:
    architecture: "x86_64"
    description: "My custom Focal Fossa image"
    os: "Ubuntu"
    release: "focal"

Make a compressed tarball of metadata.yaml:

    # tar zcf metadata.tar.gz metadata.yaml

Import the image into lxc / lxd:

    $ lxc image import metadata.tar.gz rootfd.tar.gz --alias focal-base

Create a container which uses the customized base image:

    $ lxc launch focal-base focaltest
    $ lxc exec focaltest bash

The next several steps should be executed in the container, in the bash shell spun up by “lxc exec…”

Configure container networking

In the container, create /etc/netplan/50-cloud-init.yaml:

    network:
        version: 2
        ethernets:
            eth0:
                dhcp4: true

Use “cat > /etc/netplan/50-cloud-init.yaml”, and cut-‘n-paste if your favorite text editor is AWOL.

Apply the configuration:

    # netplan apply

At this point, eth0 should have an ip address, and you should see a default route with “route -n”.

Configure apt

Again, in the container, set up /etc/apt/sources.list via cut-‘n-paste from a recently update “focal fossa” host. Something like so:

    deb http://us.archive.ubuntu.com/ubuntu/ focal main restricted
    deb http://us.archive.ubuntu.com/ubuntu/ focal-updates main restricted
    deb http://us.archive.ubuntu.com/ubuntu/ focal universe
    deb http://us.archive.ubuntu.com/ubuntu/ focal-updates universe
    deb http://us.archive.ubuntu.com/ubuntu/ focal multiverse
    deb http://us.archive.ubuntu.com/ubuntu/ focal-updates multiverse
    deb http://us.archive.ubuntu.com/ubuntu/ focal-backports main restricted universe multiverse
    deb http://security.ubuntu.com/ubuntu focal-security main restricted
    deb http://security.ubuntu.com/ubuntu focal-security universe
    deb http://security.ubuntu.com/ubuntu focal-security multiverse

“apt-get update” and “apt-install” should produce reasonable results. Suggest “apt-get install make git”.

At this point, you can use the “/scratch” sharepoint (or similar) to execute “make install-dep install-ext-deps” to set up the container with the vpp toolchain; proceed as desired.