EUserv is a virtual private server (VPS) provider in Germany. Notably, they offer a container-based Linux server, VS2-free, free of charge. VS2-free comes with one 1GHz CPU core, 1GB memory, and 10GB storage. Although I already have more than enough servers to play with, who doesn't like some more computing resources for free?
There's one catch: the VS2-free is IPv6-only. It neither has a public IPv4 address, nor offers NAT-based IPv4 access. All you can have is a single /128 IPv6 address.
$ ip addr 1: lo: <LOOPBACK,UP,LOWER_UP> mtu 65536 qdisc noqueue state UNKNOWN group default qlen 1000 link/loopback 00:00:00:00:00:00 brd 00:00:00:00:00:00 inet 127.0.0.1/8 scope host lo valid_lft forever preferred_lft forever inet6 ::1/128 scope host valid_lft forever preferred_lft forever 546: eth0@if547: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UP group default qlen 1000 link/ether b2:77:4b:c0:eb:0b brd ff:ff:ff:ff:ff:ff link-netnsid 0 inet6 2001:db8:6:1::6dae/128 scope global valid_lft forever preferred_lft forever inet6 fe80::5ed4:d66f:bd01:6936/64 scope link valid_lft forever preferred_lft forever
If I attempt to access an IPv4-only destination, a "Network is unreachable" error appears:
$ host lgger.nexusbytes.com lgger.nexusbytes.com has address 184.108.40.206 $ ping -n -c 4 lgger.nexusbytes.com connect: Network is unreachable
Not having IPv4 access severely restricts the usefulness of the VS2-free, because I would be unable to access many external resources that are not yet IPv6-enabled. Is there a way to get some IPv4 access in the IPv6-only VS2-free vServer?
Stateful NAT64 translation is a network protocol that allows IPv6-only clients to contact IPv4 servers using unicast UDP, TCP, or ICMP. It relies on a dual-stack server, known as a NAT64 translator, to proxy packets between IPv6 and IPv4 networks.
There are a number of public NAT64 services in Europe that would enable IPv4 access from my server. To use NAT64, all I need to do is changing the DNS settings in my server:
$ sudoedit /etc/resolvconf/resolv.conf.d/base nameserver 2a01:4f9:c010:3f02::1 nameserver 2a00:1098:2c::1 nameserver 2a00:1098:2b::1 $ sudo resolvconf -u
Note that on a Debian 10 system with
resolveconf package, the proper way to change DNS servers is editing
/etc/resolvconf/resolv.conf.d/base and then executing
resolvconf -u to regenerate
If you modify
/etc/resolv.conf directly, the changes will be overwritten during the next reboot.
After making the changing, DNS responses for IPv4-only destinations would contain additional IPv6 addresses that belong to the NAT64 translator, which would facilitate the connection:
$ host lgger.nexusbytes.com lgger.nexusbytes.com has address 220.127.116.11 lgger.nexusbytes.com has IPv6 address 2a00:1098:2c::1:2e04:c7e1 lgger.nexusbytes.com has IPv6 address 2a01:4f9:c010:3f02:64:0:2e04:c7e1 lgger.nexusbytes.com has IPv6 address 2a00:1098:2b::2e04:c7e1 $ ping -n -c 4 lgger.nexusbytes.com PING lgger.nexusbytes.com(2a00:1098:2c::1:2e04:c7e1) 56 data bytes 64 bytes from 2a00:1098:2c::1:2e04:c7e1: icmp_seq=1 ttl=41 time=39.9 ms 64 bytes from 2a00:1098:2c::1:2e04:c7e1: icmp_seq=2 ttl=41 time=39.7 ms 64 bytes from 2a00:1098:2c::1:2e04:c7e1: icmp_seq=3 ttl=41 time=39.6 ms 64 bytes from 2a00:1098:2c::1:2e04:c7e1: icmp_seq=4 ttl=41 time=39.8 ms
It is easy to gain IPv4 access on the EUserv VS2-free container by using a public NAT64 service, but there are several drawbacks:
The IPv4 addresses of public NAT64 services are shared by many users. If any other user misbehaves, the shared IPv4 address of the NAT64 translator could be blocklisted by the destination IPv4 service.
The NAT64 translator could apply rate limits if it gets busy.
While we can contact an IPv4-only destination by its hostname, it is still not possible to contact an IPv4 address:
$ ping 18.104.22.168 connect: Network is unreachable
IPv4 NAT over VXLAN
To get true IPv4 access on an IPv6-only server, we need to create a tunnel between the IPv6-only server and a dual-stack server, and then configure Network Address Translation (NAT) on the dual stack server. Many people would think about using a VPN software, such as OpenVPN or WireGuard. However, VPN is overkill, because there is a lighter weight solution: VXLAN.
VXLAN, or Virtual eXtensible Local Area Network, is a framework for overlaying virtualized layer 2 networks over layer 3 networks. In our case, I can create a virtualized Ethernet (layer 2) network over an IPv6 (layer 3) network. Then, I can assign IPv4 addresses to the virtual Ethernet adapters, in order to give IPv4 access to the previously IPv6-only VS2-free vServer.
I have a small dual-stack server in Germany, offered by Gullo's Hosting. It is an OpenVZ 7 container. It runs Debian 10, the same operating system as my VS2-free. I will be using this server to share IPv4 to the VS2-free.
In the examples below:
2001:db8:473a:723d:276e::2is the public IPv6 address of the dual-stack server.
2001:db8:6:1::6daeis the public IPv6 address of the IPv6-only server.
192.0.2.1is the public IPv4 address of the dual-stack server.
After reverting the DNS changes from the previous section, I execute the following commands on the EUserv vServer to setup a VXLAN tunnel:
sudo ip link add vx84 type vxlan id 0 remote 2001:db8:473a:723d:276e::2 local 2001:db8:6:1::6dae dstport 4789 sudo ip link set vx84 mtu 1420 sudo ip link set vx84 up sudo ip addr add 192.168.84.2/24 dev vx84 sudo ip route add 0.0.0.0/0 via 192.168.84.1
Note that I reduced the MTU of the VXLAN tunnel interface to 1420 from the default 1500. This is necessary to accommodate the overhead of VXLAN headers, so that the encapsulated IPv6 packets can fit into the normal MTU.
On the dual-stack server, I execute these commands to setup its end of the tunnel and enable NAT:
sudo ip link add vx84 type vxlan id 0 remote 2001:db8:6:1::6dae local 2001:db8:473a:723d:276e::2 dstport 4789 sudo ip link set vx84 mtu 1420 sudo ip link set vx84 up sudo ip addr add 192.168.84.1/24 dev vx84 sudo iptables-legacy -t nat -A POSTROUTING -s 192.168.84.0/24 ! -d 192.168.84.0/24 -j SNAT --to 192.0.2.1
It's worth noting that the command for enabling NAT is iptables-legacy instead of iptables. Apparently, there are two variants of iptables that access different kernel APIs. Although both commands would succeed, only iptables-legacy is effective in an OpenVZ 7 container. This had me scratching my head for a while.
After these setup, I'm able to access IPv4 from the IPv6-only server:
$ traceroute -n -q1 lgger.nexusbytes.com traceroute to lgger.nexusbytes.com (22.214.171.124), 30 hops max, 60 byte packets 1 192.168.84.1 23.566 ms 2 * 3 126.96.36.199 34.058 ms 4 188.8.131.52 23.615 ms 5 184.108.40.206 24.077 ms 6 220.127.116.11 23.955 ms
In Wireshark, these packets would look like this:
Frame 5: 146 bytes on wire (1168 bits), 146 bytes captured (1168 bits) Linux cooked capture v1 Internet Protocol Version 6, Src: 2001:db8:6:1::6dae, Dst: 2001:db8:473a:723d:276e::2 User Datagram Protocol, Src Port: 53037, Dst Port: 4789 Virtual eXtensible Local Area Network Ethernet II, Src: b6:ab:7c:af:51:d1 (b6:ab:7c:af:51:d1), Dst: be:ce:c9:cf:a7:f3 (be:ce:c9:cf:a7:f3) Internet Protocol Version 4, Src: 192.168.84.2, Dst: 18.104.22.168 User Datagram Protocol, Src Port: 50047, Dst Port: 33439 Data (32 bytes)
Make Them Persistent
ip commands will be lost after a reboot.
Normally the VXLAN tunnel should be written into the ifupdown configuration file, but as I discovered earlier, OpenVZ 7 would revert any modifications to the
Thus, I have to apply these changes dynamically using a systemd service.
The systemd service unit for the IPv6-only server is:
[Unit] Description=VXLAN tunnel to vps9 After=network-online.target Wants=network-online.target [Service] ExecStartPre=ip link add vx84 type vxlan id 0 remote 2001:db8:473a:723d:276e::2 local 2001:db8:6:1::6dae dstport 4789 ExecStartPre=ip link set vx84 mtu 1420 ExecStartPre=ip link set vx84 up ExecStartPre=ip addr add 192.168.84.2/24 dev vx84 ExecStartPre=ip route add 0.0.0.0/0 via 192.168.84.1 ExecStart=true RemainAfterExit=yes ExecStopPost=ip link del vx84 [Install] WantedBy=multi-user.target
The systemd service unit for the dual-stack server is:
[Unit] Description=VXLAN tunnel to vps2 After=network-online.target Wants=network-online.target [Service] ExecStartPre=ip link add vx84 type vxlan id 0 remote 2001:db8:6:1::6dae local 2001:db8:473a:723d:276e::2 dstport 4789 ExecStartPre=ip link set vx84 mtu 1420 ExecStartPre=ip link set vx84 up ExecStartPre=ip addr add 192.168.84.1/24 dev vx84 ExecStartPre=iptables-legacy -t nat -A POSTROUTING -s 192.168.84.0/24 ! -d 192.168.84.0/24 -j SNAT --to 192.0.2.1 ExecStart=true RemainAfterExit=yes ExecStopPost=iptables-legacy -t nat -D POSTROUTING -s 192.168.84.0/24 ! -d 192.168.84.0/24 -j SNAT --to 192.0.2.1 ExecStopPost=ip link del vx84 [Install] WantedBy=multi-user.target
On both servers, this service unit file should be uploaded to
Then, I can enable the service unit with these commands:
sudo systemctl daemon-reload sudo systemctl enable vx84
They will take effect after a reboot:
$ ip addr show vx84 4: vx84: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc noqueue state UNKNOWN group default qlen 1000 link/ether f2:4c:5d:6c:4b:25 brd ff:ff:ff:ff:ff:ff inet 192.168.84.2/24 scope global vx84 valid_lft forever preferred_lft forever inet6 fe80::f04c:5dff:fe6c:4b25/64 scope link valid_lft forever preferred_lft forever $ ping -c 4 22.214.171.124 PING 126.96.36.199 (188.8.131.52) 56(84) bytes of data. 64 bytes from 184.108.40.206: icmp_seq=1 ttl=57 time=28.9 ms 64 bytes from 220.127.116.11: icmp_seq=2 ttl=57 time=28.7 ms 64 bytes from 18.104.22.168: icmp_seq=3 ttl=57 time=28.9 ms 64 bytes from 22.214.171.124: icmp_seq=4 ttl=57 time=28.10 ms
This article describes two methods of gaining IPv4 access on an IPv6-only server such as the EUserv VS2-free.
- Use a public NAT64 translator.
- Establish a VXLAN tunnel to a dual-stack server, and then configure IPv4 addresses and NAT on the virtual Ethernet interfaces.
To workaround OpenVZ 7 limitation of not being able to modify
/etc/network/interfaces, we use a systemd service unit to dynamically establish and teardown the VXLAN tunnel and related configuration.