Category Archives: Networking

6rd on Free

Today I will discuss how to configure 6rd on the French ISP Free, if you decide not to use the provided Freebox and instead use your own equipment.

Free has been deploying to their customers since 2007. They were one of the first major ISPs to provide customers with IPv6 connectivity. But providing IPv6 for such a long time means they have not always kept up with the latest innovations, and thus Free don’t provide services like DHCPv6 or native IPv6 on some circuits.

If you have FTTH (100/1000MBit), your Freebox will be using the fibre SFP provided during installation. If you instead have xDSL, it will use the included cable to connect directly to the phone line using the DSL port.

With the Freebox, you will have IPv4 and IPv6 connectivity without any effort. But, if you wish to use your equipment after the Freebox you must put it into “bridge” mode and suffer dual NAT. You will also be limited by the features of the Freebox and have to trust Free to keep it updated and safe from vulnerabilities.

Those who choose to use their own equipment must have a device compatible with an SFP adapter, and configure VLAN 836 to receive an IPv4 address via DHCP.

Since IPv6 is provided using IPv6-in-IPv4, further configuration is necessary.

If you are using Mikrotik equipment, detailed documentation exists on how to configure 6rd.

If you are using a Linux-based router (e.g. OpenWrt) the process is slightly different, though the principles remain the same.

Free has an IPv6 prefix of 2a01:e00::/26, with the prefix 2a01:e3a being used for 6rd. The first step to getting working 6rd, is to determine which 6rd gateway Free is using for your IP address. The simplest way to determine this, is to calculate your IPv6 address.

Note: Don’t bother buying a copper SFP and using it in the Freebox to man-in-the-middle the fibre connection with a switch mirror port. It won’t fit physically, and even if you find a way, the Freebox won’t recognize it. ¯\_(ツ)_/¯

Use the prefix 2a01:e3a + your Free IPv4 address in hexadecimal. For example, if your IPv4 address is, the first IPv6 subnet would be 2a01:e3a0:8080:8080::/64

Confirm that your IPv6 address calculation is correct by using an online tool to ping the ::1 IP address in this IPv6 subnet, while running tcpdump on your router and filtering for protocol 41. If you have calculated the IPv6 address correctly, you should see IPv4 encapsulated IPv6 packets reaching your router:

# tcpdump -i eth0.836 proto 41
07:29:54.229910 IP > IP6 2600:3c01::f03c:91ff:fe93:48f8 > 2a01:e3A:ABBB:CCC1::1: ICMP6, echo request, seq 1, length 64

At this point, since we have not configured the 6rd tunnel, you should not expect to see any echo replies from the IPv6 address. Note the source IP address of the packet, this is the 6rd gateway from Free.

Before continuing, you need to add a firewall rule to allow protocol 41 through your firewall to the IP address of the Free 6rd gateway. From the above tcpdump output, the rules to add would be:

iptables -I zone_wan_input -i eth0 -p 41 -s -j ACCEPT
iptables -I zone_wan_output -o eth0 -p 41 -d -j ACCEPT

OpenWrt does include support for 6rd in Luci, but I was never able to have this configuration bring up a working 6rd tunnel. Instead I configured the tunnel manually in /etc/rc.local:

ip tunnel add 6rd mode sit remote local
ip link set 6rd up
ip addr add 2a01:0e3A:ABBB:CCC0::1/64 dev 6rd
ip addr add 2a01:0e3A:ABBB:CCC1::1/64 dev br-lan
ip route add ::/0 dev 6rd

This is almost enough to have IPv6 connectivity working fully. However, your IPv6 routing will be broken, as this interface is manually created and doesn’t belong to the LAN or WAN zones.

To resolve this, go to the OpenWrt web GUI and create a new interface with the Unmanaged protocol, covering the 6rd interface. Assign the new interface to the WAN zone, and restart the firewall. IPv6 routing should now be functional.

You should also configure the LAN interface to have the Router Advertisement-Service and DHCPv6 Service in server mode. This will ensure clients receive an IPv6 address in the IPv6 subnet assigned to the LAN.

I recommend rebooting your OpenWrt router to ensure that your configuration is correctly applied on boot.

You can check that IPv6 is correctly configured correctly by using an online tool such as test-ipv6. If everything has been configured correctly, your test results should be positive!

Debian on WD EX2100

The Western Digital My Cloud EX2100 is a dual-bay NAS based on the Marvell Armada 385 dual core ARMv7 CPU first released in 2015.

In terms of NAS devices available in 2017, it isn’t very special. I would say the only major differences between most other devices in the 2 bay category are:

  • Dual Gigabit Ethernet
  • Screwless and trayless hard drive installation

The dual GigE interfaces are what attracted me to the device over competitors like the Zyxel NAS326 and Western Digital’s own MyCloud EX2 Ultra.

Unlike some other NAS bundles, it’s possible to buy the EX2100 without drives, so you can add your own preferred 3.5″ SATA hard drives. I bought a refurbished unit for 110€, which seems typical for a device with these features. For some reason the resale price of these units has skyrocketed since I bought mine in mid-2017. I personally would not pay more than 150€ for such a device. If you get into the higher price range of these SOHO devices, you’re almost always going to get better value for your money building your own NAS using standard x86 components (such as the HP MicroServer G7/Gen 8/Gen 10) and a distribution like FreeNAS or OpenMediaVault.

Since the vendor supplied firmware is almost always a pile of unsightly hacks, I set to work investigating into how to put a better operating system on the EX2100. If you stick around to the end, you’ll see this particular product also has its share of unfortunate hardware design decisions…

Before we get into anything about u-boot or the operating system, we need to talk about kwboot.

kwboot stands for “Kirkwood boot”

Kirkwood is an ARMv5 SoC from Marvell around 2008-2009 that started out in the SheevaPlug (what single board computers were before the creation of the Raspberry Pi) and sooner or later found its way into a lot of NAS devices like the D-Link DNS320 and the Zyxel NSA320.

Coming back to the near-past (2015), and we have the Western Digital EX2100/4100 which use the Marvell Armada 385/388 CPU, which is a dual-core ARMv7 design. However it was known that the Armada SoC could boot from serial because of the SolidRun ClearFog. But the ClearFog uses DIP switches to set the boot source, and most (all?) consumer devices lack these.

ClearFog Pro boot source selection DIP switches

It turns out it is possible to kwboot consumer devices based on the Armada 38x, however you need to apply this patch to kwboot to parse the response from the Armada CPU, which differs from the Kirkwood response. Unfortunately the patch broke Kirkwood compatibility, and was seemingly never merged into u-boot mainline. However, you can still apply it to the kwboot source in u-boot and compile kwboot for use with Armada CPUs.

Once you have patched kwboot, you can use it to test new versions of u-boot via a USB to uart adapter:
$ ./kwboot -f -t -B 115200 /dev/ttyUSB0 -b u-boot-uart.bin -s 0 -q 1

There are some synchronization issues with the magic sequence, so it often takes several attempts before successfully loading via kwboot. A dead giveaway that you need to power cycle the device and try again is when you immediately see u-boot output in the console instead of “Sending boot image…”

A successful attempt should look similar to the following:
$ ./kwboot -f -t -B 115200 /dev/ttyUSB0 -b u-boot-a38x-Yosemite_2014T3_PQ-nand-uart.bin -s 0 -q 1
Sending boot message. Please reboot the target...-�$�"Ufw�$�"U����$
Dfw�$�"U�\�$�"U����$�DUf�$�"Uw��"U����$4"U���$�"Uw�$�"U���$�DUf|fD�&T���$�"U�E�$�"Df3DD�DU�E7$�"U����$4"U���$�"U�E�4"U�/[email protected]� ��$DUw�$�"U����$�DUff�$�"D��fD$U��
Sending boot image...
0 % [......................................................................]

Once kwboot works, you can safely proceed to testing u-boot modifications without the risk that you brick your device, as kwboot runs code in memory without modifying the contents of NAND.

Unfortunately mainline u-boot doesn’t support this device, although similar devices are supported, such as the Turris Omnia (Armada 385) and Solidrun Clearfog Pro (Armada 388). It’s no surprise that attempting to kwboot a build of mainline u-boot for these targets on the EX2100 doesn’t work. So currently we have no choice but to use the u-boot source from Western Digital’s GPL archive.

The stock u-boot on the device does not support saveenv. Without modifying NAND, it is possible to boot Linux from USB, however this requires using the uart console and manually entering the boot parameters on each boot.

Naively modifying the WD u-boot source to enable the saveenv command results in corruption of the kernel uImage since someone at WD set the environment offset to 5MB and this is beyond the u-boot partition, corrupting the uImage.

However it is possible to modify the WD u-boot source to save environment variables within the 5MB allocated for u-boot. This requires reflashing u-boot to the device. Before you replace the stock u-boot on your device, you should take a backup of the u-boot region of flash. This can be done from within the Western Digital firmware, but requires a USB to UART adapter and a header soldered to the PCB:
# nanddump --noecc --omitoob -f mtd0.bin /dev/mtd0

Remember to copy this file somewhere off-device, such as a USB key, for safe keeping!

The general steps to replace the stock u-boot are:

  1. kwboot a modified u-boot which saves environment variables within u-boot region
  2. Inside u-boot, run saveenv
  3. Boot Debian from USB or SATA
  4. Dump u-boot env to a file (using nandread)
  5. Erase u-boot portion of mtd0 (using flash_erase)
  6. Flash new u-boot (using nandwrite)
  7. Restore u-boot environment variables (using nandwrite)
  8. Reboot

Dump the u-boot env to a file:
# nanddump -s 0x100000 -l 0x80000 -f ubootenv.bin /dev/mtd0

Erase the u-boot portion of mtd0 and flash the new u-boot to NAND:
# flash_erase /dev/mtd0 0 8
# nandwrite -p /dev/mtd0 u-boot-a38x-Yosemite_2014T3_PQ-nand.bin

Restore u-boot environment variables:
# nandwrite -p /dev/mtd0 -s 0x100000 ubootenv.bin

Integrated MCU
Western Digital decided to use an external microcontroller to handle certain system management functions such as fan control, LED control, and power on/off.

Sadly the microcontroller uses a proprietary and undocumented protocol for communication, and as it turns out this protocol can differ even between Western Digital products!

For the Western Digital EX2100 and EX4100, the integrated microcontroller communicates on ttyS1 at 115200n8, unlike other Western Digital NAS products whose microcontroller communicates at 19200n8.

Thankfully, some of the commands are common, so once communication with the microcontroller has been established, fan control and temperature monitoring are functional. Fan control and temperature monitoring are available through a userspace daemon called “mcm-daemon” (MyCloud Mirror daemon). I have forked the mcm-daemon repository on GitHub and made modifications to support the EX2100/4100.

LED control and power on/off are still a work in progress as the reverse engineered commands used on other WD products do not work on the EX2100/4100.

The user bodhi at Doozan forums does a great job of providing pre-built Debian images for a variety of Marvell Armada based NAS devices.

Usually I would link to the excellent instructions bodhi normally writes for installing Debian, but since they don’t have the EX2100, writing the instructions fell to me.

But sadly I haven’t got installation instructions written because I’ve bricked my EX2100.

Weltrend WT61P8
Let’s revisit this mystery microcontroller in charge of so many tasks in the EX2100.

Well, after reading that Western Digital My Cloud products contained a backdoor and multiple vulnerabilities I thought I would go and update the WD firmware before continuing to write the Debian installation instructions. The WD firmware resides on the built in 512MB of EMMC, while Debian lives on a USB device, so the installation of Debian does not replace the original WD firmware.

During the WD update process I noticed that it was also updating the firmware of the Weltrend:

16479 root 6624 S /var/www/cgi-bin/system_mgr.cgi
16480 root 2560 S sh -c cd /usr/local/upload/;upload_firmware -a -n 'nas-new-firmware' >/dev/null;cd /
16481 root 4544 S upload_firmware -a -n nas-new-firmware
19436 root 49120 R mcu_upgrade -r -f /tmp/uP_0.bin
19552 root 2720 R ps ax

How curious! Since the Weltrend is very undocumented I was eager to learn more about the firmware it runs.

I found out a good deal more than I’d expected. Firstly, the mcu_upgrade binary contains some interesting strings. Here is a short sample of strings in the binary:

Enable ISP
Page Erase
Set Address High Byte

The MCU firmware also has some very interesting strings. Here is a short sample of strings in the firmware:

nick 1111
nick 2222
nick 3333
nick 4444
Cannot Copy
from Camera
Cannot Move
Almost Full
Limit Reached
nick pwr on
nick pwr off 1
nick pwr off wol
Welcome to
RTC_ALARM pwr on

Googling the part number “WT61P8” lead to a very interesting datasheet (PDF) describing the microcontroller in detail.

What I found from the datasheet was… not anything I expected to find.

It’s a Turbo 8052 CPU with ~48KB of built-in EEPROM (this is my guess based on the part number and size of firmware mcu_upgrade was sending) and it’s a “Flat Panel Display Control Sub-MCU”

Most information about Weltrend microcontrollers is on Russian language forums dedicated to TV repair. The most common use of this MCU is in Samsung TVs for power management, since it includes an IR receiver and HDMI CEC capabilities.

They do support ISP (In-system Programming) via I2C, if you have the right hardware. There are quite a few Russian articles and YouTube videos on how to program these chips in TVs.


After reading about WD’s numerous firmware vulnerabilities and a back door, which were also present in D-Link NAS products (implying a shared code base or same third party contractor), and then learning that the microcontroller in charge of power management for the EX2100 (and other My Cloud products) was intended for power management in LCD TVs:

My final $0.02: this thing is an utter bodge job in both hardware and software! Don’t buy one of these. It doesn’t matter that can be persuaded to run Debian, it’s terrible value for the price.

You’re far better off getting an older PC and running FreeNAS or OpenMediaVault. Older corporate tower PCs with 2nd or 3rd gen Intel processors like the Dell Optiplex line can easily be purchased for under $150 from places like eBay.

If power consumption is really important to you, then I would recommend something like the Rock64 which has Gigabit Ethernet and excellent USB 3.0 performance with Armbian. It also comes with more RAM than the EX2100 (1/2/4GB while EX2100 has only 1GB) and is a quad core aarch64 instead of dual core armv7!

Best of all, an older PC or Pine64 is going to be cheaper than the EX2100 (or ludicrously more expensive 4 bay EX4100) anyway.

If I ever manage to restore the Weltrend firmware rest assured there will be a follow up article with both the journey of unbricking and instructions to install Debian. Until then, I’m going to take the HDDs I planned to use in the EX2100 and build a FreeNAS in an old PC. ✌️

Allwinner H2+/H3 Ethernet with Linux 4.9-rc8

The Orange Pi PC is not a new single board computer. It’s been released for over a year now, but has mostly been stuck on a heavily patched 3.4 release kernel.

There have been ongoing efforts since the release to have support for the Allwinner H3 in the mainline kernel. In the past weeks there have been new patches released which enable support for the Ethernet MAC on the H3 (and H2+).

Unfortunately this support is not in mainline yet, and won’t make it in the upcoming 4.9 release. However, that doesn’t stop you from taking the patches and applying them against 4.9 yourself.

I wrote a script to compile the kernel from source, applying the necessary patches to the kernel and using a minimal .config file which compiles the sun8i_emac support as a module. You can download the build script from GitHub.

It does try to be somewhat smart: verifying the integrity of the downloaded files, and will bail out if there are errors in patching the source code. But, it doesn’t do toolchain dependency checking because that’s just too complicated. Since the emac support will end up in mainline soon, I doubt it’s worth the time to improve the build script. However if anyone is interested in improving it, the script is released as GPLv2.

The result? If you are patient enough to wait for the kernel to compile, you get a uImage and modules for Linux 4.9-rc8 with Ethernet support:

Allwinner H3 emac performance:

[email protected]:~$ iperf -n 1024M -c
Client connecting to, TCP port 5001
TCP window size: 43.8 KByte (default)
[ 3] local port 42572 connected with port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-100.6 sec 1.00 GBytes 85.4 Mbits/sec

Orange Pi Zero (Allwinner H2+):

I’ve also just received my Orange Pi Zero and confirmed that the same kernel works on the Orange Pi Zero, so you can run Linux 4.9 on the Orange Pi PC (Allwinner H3) or Orange Pi Zero (Allwinner H2+) with Ethernet support.

Allwinner H2+ emac performance:

[email protected]:~$ iperf -n 1024M -c
Client connecting to, TCP port 5001
TCP window size: 43.8 KByte (default)
[ 3] local port 54762 connected with port 5001
[ ID] Interval Transfer Bandwidth
[ 3] 0.0-95.9 sec 1.00 GBytes 89.5 Mbits/sec

If you want to skip compiling the kernel yourself, I’m providing the kernel uImage and modules here.

I am using u-boot v2016.11 from denx.

Note: HDMI isn’t working on my Orange Pi PC, and since I run it headless I’m not interested in investigating why. If you’re using your Orange Pi PC with HDMI output, you may need to modify the kernel .config file to fix HDMI.