While trying to convince the driver that I am not, in fact, an American, I discovered that Linux’s wireless support is moving toward a new framework for handling the regulatory domain information. Drivers use this to enable the correct (and legal) channels, RF power limits and other behavior for the country you’re in…

The first thing I tried was to create a file (/etc/modprobe.d/cfg80211) containing the string:

options cfg80211 ieee80211_regdom=ZA

This was a step in the right direction, but it turns out that only US, EU and JP regulatory information is hard-coded into the kernel. If one of those won’t work for you, you need the new Wireless Central Regulatory Domain Agent user-space utility. This little program is called by the cfg80211 module through udev and provides the missing regulatory info. So I grabbed and installed the wireless-crda package from Ubuntu, and put the following in /etc/udev/rules.d/regulatory.rules:

KERNEL=="regulatory*", ACTION=="change", SUBSYSTEM=="platform", RUN+="/sbin/crda"

Reloading all the wireless-related modules I observed CRDA doing it’s magic:

charlie kernel: [ 4801.596578] cfg80211: Using static regulatory domain info
charlie kernel: [ 4801.596581] cfg80211: Regulatory domain: US
charlie kernel: [ 4801.596583] (start_freq - end_freq @ bandwidth), (max_antenna_gain, max_eirp)
charlie kernel: [ 4801.596585] (2402000 KHz - 2472000 KHz @ 40000 KHz), (600 mBi, 2700 mBm)
charlie kernel: [ 4801.596587] (5170000 KHz - 5190000 KHz @ 40000 KHz), (600 mBi, 2300 mBm)
charlie kernel: [ 4801.596589] (5190000 KHz - 5210000 KHz @ 40000 KHz), (600 mBi, 2300 mBm)
charlie kernel: [ 4801.596591] (5210000 KHz - 5230000 KHz @ 40000 KHz), (600 mBi, 2300 mBm)
charlie kernel: [ 4801.596593] (5230000 KHz - 5330000 KHz @ 40000 KHz), (600 mBi, 2300 mBm)
charlie kernel: [ 4801.596595] (5735000 KHz - 5835000 KHz @ 40000 KHz), (600 mBi, 3000 mBm)
charlie kernel: [ 4801.596597] cfg80211: Calling CRDA for country: ZA
charlie kernel: [ 4801.617628] iwlagn: Intel(R) Wireless WiFi Link AGN driver for Linux, 1.3.27ks
charlie kernel: [ 4801.617631] iwlagn: Copyright(c) 2003-2008 Intel Corporation
charlie kernel: [ 4801.617706] iwlagn 0000:03:00.0: PCI INT A -> GSI 18 (level, low) -> IRQ 18
charlie kernel: [ 4801.617734] iwlagn 0000:03:00.0: setting latency timer to 64
charlie kernel: [ 4801.617908] iwlagn: Detected Intel Wireless WiFi Link 5100AGN REV=0x54
charlie kernel: [ 4801.639541] iwlagn: Tunable channels: 13 802.11bg, 24 802.11a channels
charlie kernel: [ 4801.640028] wmaster0 (iwlagn): not using net_device_ops yet
charlie kernel: [ 4801.640575] phy0: Selected rate control algorithm 'iwl-agn-rs'
charlie kernel: [ 4801.640591] wlan0 (iwlagn): not using net_device_ops yet
charlie kernel: [ 4801.644881] cfg80211: Regulatory domain changed to country: ZA
charlie kernel: [ 4801.644883] (start_freq - end_freq @ bandwidth), (max_antenna_gain, max_eirp)
charlie kernel: [ 4801.644885] (2402000 KHz - 2482000 KHz @ 40000 KHz), (N/A, 2000 mBm)
charlie kernel: [ 4801.644888] (5170000 KHz - 5250000 KHz @ 20000 KHz), (300 mBi, 1700 mBm)
charlie kernel: [ 4801.644890] (5250000 KHz - 5330000 KHz @ 20000 KHz), (300 mBi, 2000 mBm)
charlie kernel: [ 4801.644892] (5490000 KHz - 5710000 KHz @ 20000 KHz), (300 mBi, 2000 mBm)
charlie kernel: [ 4801.644894] (5735000 KHz - 5835000 KHz @ 20000 KHz), (300 mBi, 3000 mBm)

Notice that cfg80211 loads with the US regulatiory domain, then calls CRDA, and while it’s waiting, iwlagn initializes the wireless device. Finally, cfg80211 switches to ZA, using the info obtained from user-space.

Unfortunately at this point iw dev wlan0 info showed the card was still using US channels. I discovered through experimentation that ZA channels were enabled if cfg80211 had enough time to switch from US to ZA before the iwlagn driver is loaded.

I filed a bug report and Luis Rodriguez promptly responded with a patch deleting 3 lines from net/wireless/reg.c, fixing the problem.

Update:
The patch seems to have been merged into kernel 2.6.30. I’ve upgraded to 2.6.30 from backports.org and now the CRDA magic Just Works (TM).

To make this work from any host on the network, I installed quagga on each machine, and configured OSPF to distribute the default routes. On one of the N possibly-Internet-connected hosts, my ospfd.conf looks like this (with the router-id set to the static IP of the host):

password secretABC
!
interface eth0
ip ospf message-digest-key 1 md5 secretXYZ
!
router ospf
ospf router-id 10.0.0.1
network 10.0.0.0/24 area 0.0.0.0
area 0.0.0.0 authentication message-digest
default-information originate metric 100
!


On the other hosts (set M minus set N), I have a similar configuration, but without the "default-information" line.

One could do this with a set of static routes, but I like this setup because it is dynamic and if I ever needed to do something fancy, OSPF has the flexibility. Now the Internet-connected hosts will advertise their default routes, and withdraw them when they disconnect.

I need this capability, because my home network doesn’t have a router-like device to act as a DHCP server, but I would like to be able to take my notebook to work (or elsewhere), and use DHCP with minimal fuss.

It turns out this fall-back capability is supported in dhcp3-client (and perhaps other DHCP clients), by using a pre-defined lease in the config file (/etc/dhcp3/dhclient.conf on my Ubuntu 8.10 install):

request subnet-mask, broadcast-address, time-offset, routers,
domain-name, domain-name-servers, domain-search, host-name,
netbios-name-servers, netbios-scope, interface-mtu;

lease {
interface "eth0";
fixed-address 10.0.0.1;
option subnet-mask 255.255.255.0;
renew 2 2030/1/12 00:00:01;
rebind 2 2030/1/12 00:00:01;
expire 2 2030/1/12 00:00:01;
}


dhcp3-client will now try to get a lease from DHCP, and if it fails, it will configure the IP given in the pre-defined lease, effectively making this a static fall-back IP address.

Unfortunately when NetworkManager gets the news from the DHCP client that we could not get a lease from a DHCP server, it nukes the static configuration on the interface and takes us back to square one. I wanted a work-around that would allow me to keep NetworkManager (since I use it for my 3G modem too), so I came up with this:

I added the following stanza to my /etc/network/interfaces file:

allow-hotplug eth0
iface eth0 inet dhcp


This enables the standard Debian ifup/ifdown magic, and it also stops NetworkManager from managing the interface (which I don’t mind, since I never use 802.1x on wired networks).

In order to get back plug-and-play functionality (like NetworkManager provided by configuring the interface when it detected the link going up), I installed ifplugd, and configured it to watch my wired interface by changing INTERFACES="" to INTERFACES="eth0" in /etc/default/ifplugd, and restarting the daemon.

An alternative to this setup would be to use IPv4LL addressing provided by avahi-autoip, but I prefer to manually allocate a static IP to each host.