Mon 15 February 2021

Filed under Networking

Tags networking wifi vpn


Winter vacation means that most of my time is spent at home, instead of the dorm room. The consequence of this is that those dumb devices that require access to Gravity to properly function is now basically offline, as they're not capable of dialing in via IKEv2 or WireGuard for remote access by themselves. These are some examples of such devices (and where they're trying to communicate with):

  • Kindle (
  • Chromebooks (
  • Note that while Chromebooks, when set up, are capable of dialing a VPN, they are not in the Welcome screen
  • Android phones (
  • Same as above

With the help of a NUC borrowed from @Catofes, we can set up a WiFi access point to Gravity for these devices. While it is true that the performance is going to be somewhat limited, we're not doing much other than syncing e-books, looking up Wikipedia definitions, and signing into Google accounts, so that should suffice. Here's a list of what we're going to set up:

  • Platform: VM, WiFi card passthrough, OS
  • Backhaul: Gravity address allocation, WireGuard site-to-site
  • Access: 802.11 - Hostapd, ISC DHCP - DHCP & DHCPv6, RADVD - SLAAC
  • Profit!

Let's get started!

The platform

The NUC itself runs ESXi 7.0 to support some other tasks (mainly IoT stuff) that are not covered in this article. We're using a Debian 10 VM with the Intel 7265D WiFi card passthrough. A single core and 256 MB of RAM is enough to run the site-to-site VPN, hostapd, and all the address handout stuff.

A quick check with iw list shows that the 7265D is recognized and passthrough correctly, that it supports AP mode, but only in 2.4GHz:

Wiphy phy0
        Supported interface modes:
                 * IBSS
                 * managed
                 * AP
                 * AP/VLAN
                 * monitor
                 * P2P-client
                 * P2P-GO
                 * P2P-device
        Band 1:
                        * 2412 MHz [1] (22.0 dBm)
                        * 2417 MHz [2] (22.0 dBm)
                        * 2422 MHz [3] (22.0 dBm)
                        * 2427 MHz [4] (22.0 dBm)
                        * 2432 MHz [5] (22.0 dBm)
                        * 2437 MHz [6] (22.0 dBm)
                        * 2442 MHz [7] (22.0 dBm)
                        * 2447 MHz [8] (22.0 dBm)
                        * 2452 MHz [9] (22.0 dBm)
                        * 2457 MHz [10] (22.0 dBm)
                        * 2462 MHz [11] (22.0 dBm)
                        * 2467 MHz [12] (22.0 dBm) (no IR)
                        * 2472 MHz [13] (22.0 dBm) (no IR)
                        * 2484 MHz [14] (disabled)
        Band 2:
                        * 5180 MHz [36] (22.0 dBm) (no IR)
                        * 5200 MHz [40] (22.0 dBm) (no IR)
                        * 5220 MHz [44] (22.0 dBm) (no IR)
                        * 5240 MHz [48] (22.0 dBm) (no IR)
                        * 5260 MHz [52] (22.0 dBm) (no IR, radar detection)
                        * 5280 MHz [56] (22.0 dBm) (no IR, radar detection)
                        * 5300 MHz [60] (22.0 dBm) (no IR, radar detection)
                        * 5320 MHz [64] (22.0 dBm) (no IR, radar detection)
                        * 5500 MHz [100] (22.0 dBm) (no IR, radar detection)
                        * 5520 MHz [104] (22.0 dBm) (no IR, radar detection)
                        * 5540 MHz [108] (22.0 dBm) (no IR, radar detection)
                        * 5560 MHz [112] (22.0 dBm) (no IR, radar detection)
                        * 5580 MHz [116] (22.0 dBm) (no IR, radar detection)
                        * 5600 MHz [120] (22.0 dBm) (no IR, radar detection)
                        * 5620 MHz [124] (22.0 dBm) (no IR, radar detection)
                        * 5640 MHz [128] (22.0 dBm) (no IR, radar detection)
                        * 5660 MHz [132] (22.0 dBm) (no IR, radar detection)
                        * 5680 MHz [136] (22.0 dBm) (no IR, radar detection)
                        * 5700 MHz [140] (22.0 dBm) (no IR, radar detection)
                        * 5720 MHz [144] (22.0 dBm) (no IR, radar detection)
                        * 5745 MHz [149] (22.0 dBm) (no IR)
                        * 5765 MHz [153] (22.0 dBm) (no IR)
                        * 5785 MHz [157] (22.0 dBm) (no IR)
                        * 5805 MHz [161] (22.0 dBm) (no IR)
                        * 5825 MHz [165] (22.0 dBm) (no IR)

Note how every frequency in the 5GHz band is marked with no IR, which means no initiating radiation apparently. While this is most likely a regulatory limitation, soft or hard, I don't feel bothered to bypass this. We're sticking with 2.4GHz for now.

The backhaul

Backhaul in the current situation means bridging the incoming devices into Gravity. While it is possible to run a full node on the NUC VM, I don't really feel the need to do so: we're behind multiple layers of NAT (3 layers if including the ISP's), no good IPv6, and not stable enough to relay traffic for the rest of the network. Instead, a site-to-site VPN through WireGuard to a full node will serve the purpose. Note that this is very similar to the Gravity AP deployed at my dorm, which uses GRE for its lightweight and sharing the campus network with the full node. WireGuard should help with traversing NATs this time.

Note that no address is assigned to the WireGuard interface on the NUC VM; the gateway address sits on the wlan adapter. WireGuard works just fine forwarding these packets with the fwmark-based default route handling (from wg-quick).

Gravity address allocation

As we're planning full dual-stack access, the following two blocks are alloted to this new site:

  • IPv4: (private)
  • IPv6: <prefix redacted>:cc20::/60

The first host address of each network is the gateway and sits on the new site. These should not collide with anything else currently...

The AP

hostapd - L2

Hostapd runs the AP in 802.11n mode. The configuration is pasted below for future reference.

# /etc/hostapd/hostapd.wls192.conf

ssid=JSteward Access

We also need to edit /etc/hostapd/ to activate the run-parts hook of ifupdown to start and stop hostapd automatically. Insert this at the beginning:

if [ -f /etc/hostapd/hostapd.${IFACE}.conf ]; then

ifupdown - L3

As stated in the previous section, the gateway address sits on the wlan interface (for ARP/RA). The configuration is pasted below for future reference.

# /etc/network/interfaces
# Ethernet/lo configuration elided
# wifi AP
auto wls192
iface wls192 inet static
        pre-up systemctl start --no-block wg-quick@seki
        post-down systemctl stop --no-block wg-quick@seki

iface wls192 inet6 static
        dad-attempts 0
        address <prefix redacted>:cc20::1/60

Note the --no-block option to systemctl; this is necessary, otherwise the attempt to ifup -a at boot time will timeout due to name resolution issues (the Ethernet connection is not up yet). Having it delay a little bit solves all the problems.

ISC DHCP & RADVD - address distribution

radvd and isc-dhcp-server works together to automatically hand out addresses to clients dual-stack. DHCPv4 is relatively simple:

# /etc/dhcp/dhcpd.conf
option domain-name "";
option domain-name-servers,;
default-lease-time 600;
max-lease-time 7200;
ddns-update-style none;

subnet netmask {
  option routers;

DHCPv6 is just as simple. Note that there's no routers option for DHCPv6 though; the gateway (usually as an LL address) needs to be handed out though router advertisement. For consistency, DNS name server and search domain will be handed out via router advertisement as well.

# /etc/dhcp/dhcpd6.conf
default-lease-time 2592000;
preferred-lifetime 604800;
option dhcp-renewal-time 3600;
option dhcp-rebinding-time 7200;
allow leasequery;
option 21600;

subnet6 <prefix redacted>:cc20::/60 {
        range6 <prefix redacted>:cc20:b::1 <prefix redacted>:cc20:b::ff:ffff;

RADVD here works in the so-called Assisted mode:

Assisted: Stateful configuration, address configuration provided by DHCPv6, although advertised routes can also be used on Stateless Address Autoconfiguration setups (SLAAC).

# /etc/radvd.conf
interface wls192 {
        AdvSendAdvert on;
        AdvManagedFlag on;
        AdvOtherConfigFlag on;
        MinRtrAdvInterval 3;
        MaxRtrAdvInterval 10;
        prefix <prefix redacted>:cc20::/60 {
                AdvOnLink on;
                AdvAutonomous off;
        prefix <prefix redacted>:cc2a::/64 {
                AdvOnLink on;
                AdvAutonomous on;

        RDNSS 2001:4860:4860::8888 2001:4860:4860::8844 { };
        DNSSL { };

The SLAAC block (<prefix redacted>:cc2a::/64) uses a different prefix for easy distinguishing. Note that for unknown reasons using just one single AdvAutonomous /60 prefix for both SLAAC and DHCPv6 didn't work; please leave a comment below if you have any clues.


The system can support about 50Mbps downlink and 20Mbps uplink with an 2020 Macbook Air, translating to about 15% utility (all kernel time) on a single core. Considering that we're running this on 2.4GHz band, the bottleneck is most likely due to 7265D's Tx power limits and/or bad channel selection. In addition, the coverage is pretty bad as well, with signals barely usable outside the bedroom (the NUC is placed on the desk with plenty of cords on top). But, just as described in the preface, this should be sufficient for most of the use cases. Happy hacking!


Sat 18 November 2017

Filed under Networking

Tags networking


With a bunch of servers, it's necessary that they're named in a consistent manner. This article describes how the machines should be named.

Basic rule

A machine name should consist of the following three parts:

  • a unique identifier for the machine (Romaji of Japanese words in one Kanji, examples …
Read More

Wed 15 November 2017

Filed under Networking

Tags networking gre mpls iproute2


MPLS can be used for prioritizing traffic, which will be used in JSteward Tech's routing policy. This article works out how to label packets with MPLS and switch according to packets between two host connected by a GRE tunnel.

Before you start

Full MPLS support wasn't present in Linux …

Read More

Sat 11 November 2017

Filed under Networking

Tags networking infrastructure


As the number of machines I manage increase gradually, it has become a necessity to connect the machines together through secure, reliable tunnels, thus forming an internal network. A private network can have the following advantages:

  • Ease of configuring services that involve multiple machines, such as a databases and …
Read More

Fri 10 November 2017

Filed under Networking

Tags networking gentoo


On a machine with native IPv6 that's working quite well (e.g. CERNET2), it's usually desirable to set the host's default IPv6 route to go through native IPv6. Yet the cheap /48 address block offered by HE Tunnelbroker can be useful if one wishes to assign IPv6 addresses through …

Read More

Pengcheng Xu's Place © Pengcheng Xu Powered by Pelican and Twitter Bootstrap. Icons by Font Awesome and Font Awesome More