The goal of this project is to implement the Wi-SUN protocol on Linux devices and allow the use of Linux hosts as Border Routers for Wi-SUN networks. For the time being, the implementation is mostly a port of Silicon Labs' embedded stack on a Linux host. However, the ultimate goal is to replace services currently provided by the stack with native Linux services.
This project provides the wsbrd
daemon, which is responsible for running the
Wi-SUN protocol high-level layers. It is paired with an RF device RCP (Radio
Co-Processor) handling the low-level layers and RF activities. The RCP devices
currently supported are EFR32FG12, EFR32MG12, EFR32FG25, EFR32FG28, and
EFR32ZG28.
The RCP needs to be flashed with a specific firmware to communicate with the daemon. This firmware is provided in binary format. Application Note 1332 explains how to build RCP firmware and flash it.
The communication between the Linux host and the RCP is supported through a
serial link (UART). On Silicon Labs mainboards, this serial link is provided
over USB. The /dev/ttyACMx
device should appear when you connect the
mainboard.
If it is not yet done, start by cloning this repository:
git clone https://github.com/SiliconLabs/wisun-br-linux.git
The build requires mbedTLS
(> 2.18), libnl-3
, libnl-route-3
, and cmake
.
libcap
and libsystemd
are also recommended (note that libsystemd
can be
replaced by elogind
if you do not want to pull systemd
). Optionally, you can
also install Rust/Cargo.
We also encourage the use of Ninja as the cmake
back-end.
On Debian and its derivatives, install the necessary dependencies (except for mbedTLS) with:
sudo apt-get install libnl-3-dev libnl-route-3-dev libcap-dev \
libsystemd-dev cmake ninja-build pkg-config lrzsz
To additionally compile wsbrd_cli
:
sudo apt-get install cargo libdbus-1-dev
cargo fetch --manifest-path=tools/wsbrd_cli/Cargo.toml
Debian does not (yet) package MbedTLS >= 3.0 so you must build it from sources. This project does not support versions < 3.0.
git clone --branch=v3.6.2 --recurse-submodules https://github.com/ARMmbed/mbedtls
cd mbedtls
cmake -G Ninja .
ninja
sudo ninja install
MbedTLS
is highly customizable. The default configuration is sane. However, if
you want a stripped-down version, you can configure it with the configuration
file provided in examples/mbedtls-config.h
:
CFLAGS="-I$FULL_PATH_TO_WSBRD_SRC/examples -DMBEDTLS_CONFIG_FILE='<mbedtls-config.h>'" cmake -G Ninja .
This configuration file has been written for
mbedtls
3.0. Adapt it if necessary.
Optionally, wsbrd
can be compiled with support for Silabs
CPC. To install Silabs CPC library:
git clone https://github.com/SiliconLabs/cpc_daemon.git
cd cpc_daemon
cmake -S . -B build -G Ninja
ninja -C build
sudo ninja -C build install
sudo ldconfig
Caution
The RCP is responsible for encrypting and decrypting IEEE 802.15.4 frames. MAC layer data appears in cleartext in HIF commands so integrators are expected to secure the serial link, for example using CPC with link encpryption, otherwise the system may be exposed to packet injection or eavesdropping. IEEE 802.15.4 encryption keys can even be stolen if a malicious actor intercepts the HIF command which transfers them to the RCP.
Then, you can compile wsbrd
with:
cd wisun-br-linux/
cmake -G Ninja .
ninja
Finally, install the service with:
sudo ninja install
No script for any start-up service is provided for now.
You must provide a configuration file to the Wi-SUN border router. A commented
example is available in /usr/local/share/doc/wsbrd/examples/wsbrd.conf
.
cp -r /usr/local/share/doc/wsbrd/examples .
<edit examples/wsbrd.conf>
You can copy and edit it. You will notice that you need certificates and keys to
authenticate your network's Wi-SUN nodes. The generation of these files is
described in [[Generating the Wi-SUN Public Key Infrastructure]]. For now, you
can use the certificate examples installed in
/usr/local/share/doc/wsbrd/examples/
.
You also must provide the path of the UART representing your RCP device.
Finally, launch wsbrd
with:
sudo wsbrd -F examples/wsbrd.conf -u /dev/ttyACM0
wsbrd
lists the useful options in the output of wsbrd --help
.
A suite of tools is provided for various tasks around wsbrd
and its RCP. For
more detail, refer to the README.md
present in the relevant source folder
(under tools/
), or the --help
message output by the application.
Some of these are not compiled by default and require setting
COMPILE_DEVTOOLS=ON
when configuring the project with CMake.
Application | Description |
---|---|
wsbrd_cli |
A simple application for querying the D-Bus interface |
wsbrd-fwup |
A tool for updating the RCP firmware |
wsbrd-fuzz |
A tool for fuzzing and debugging wsbrd |
wshwping |
A tool for testing the serial link |
wstbu |
An implementation of the Wi-SUN Test Bed Unit REST API |
wsrd |
An experimental Wi-SUN router for Linux (use COMPILE_WSRD ) |
silabs-ws-dc |
Silicon Labs Wi-SUN Direct Connect Linux client |
wsbrd_cli
is a small utility to retrieve the status of the Wi-SUN network. Its
usage is described in the output of wsbrd_cli --help
. The tool relies on the
D-Bus interface provided by wsbrd
, which is described in DBUS.md
.
The certificate generation process is described in section 6.5.1 of the Wi-SUN specification. It uses the standard X.509 certificate format. Some fields and algorithms are enforced.
The process to get official certificates is described on the Wi-SUN alliance Web site (restricted access).
wsbrd
provides a built-in DHCPv6 server. However, it is still possible to use
an external DHCPv6 server. If the DHCP server runs on a remote host, you need to
launch a DHCPv6 relay.
wsbrd
has been tested with ISC DHCP and dnsmasq. Both projects provide DHCP
server and DHCP relay implementations.
First you have to deactivate the internal dhcp server of wsbrd
in wbsrd.conf:
internal_dhcp = false
Obviously, the DHCP server/relay needs a network interface to run. You can
launch the DHCP server/relay just after wsbrd
(you have to ensure the DHCP
service is started before Wi-SUN nodes connect, but they will not connect before
at least several dozen seconds) or create the interface before launching
wsbrd
.
Because of this issue, dnsmasq
is supported from the version 2.87.
dnsmasq
does not need any specific options. A classical invocation can be
used. We suggest increasing the lease time (336h
) and disabling DNS server
(-p 0
):
sudo dnsmasq -d -C /dev/null -p 0 -i tun0 --dhcp-range 2001:db8::,2001:db8::ffff,64,336h
To start the DHCP relay, you have to bind the Wi-SUN
and the upstream network
interfaces (tun0
and eth0
). Then specify the IP addresses of the DHCP server
(2001:db8:a::1
) and of the wsbrd
interface (2001:db8:b::1
) to the
--dhcp-relay
option. You can also disable the DNS server (which is useless in
this case) with -p 0
:
sudo dnsmasq -d -C /dev/null -p 0 -i tun0,eth0 --dhcp-relay 2001:db8:a::1,2001:db8:b::1
Note that ISC DHCP needs to be patched to comply with Wi-SUN specification (for
relay and server). We provide the needed patches in misc/
.
isc-dhcpd
will not start if the lease file does not exist:
sudo mkdir -p /var/lib/dhcpd/
sudo touch /var/lib/dhcpd/dhcpd.leases
We provide a sample configuration file for isc-dhcp. See examples/dhcpd.conf
for details:
sudo dhcpd -6 --no-pid -lf /var/lib/dhcpd/dhcpd.leases -cf examples/dhcpd.conf tun0
To start the DHCP relay, you have to provide the wsbrd
network interface
(tun0
), the address of the DHCP server (2001:db8::1
in the example below),
and the network interface associated with this address (see dhcprelay
manpage):
sudo dhcrelay -6 --no-pid -l tun0 -u 2001:db8::1%eth0
The recommended method to run wsbrd
is to start it as root, let it initialize
necessary resources as root, and then switch to a less privileged user using
the user
and group
parameters, following the principle of least privilege.
If for some reason this method is not desired, the following steps should
enable running wsbrd
without being the root user.
To run wsbrd
without root permissions, first ensure you have permission to
access the UART device (you will have to logout/login after this command):
sudo usermod -aG dialout YOUR_USER
Then, wsbrd
needs some permissions to a initialize network resources
(sockets, virtual interface, addresses...). The following Linux
capabilities are required:
CAP_NET_ADMIN
: Create TUN interface, manage addresses, routes and neighbor proxy entries.CAP_NET_BIND_SERVICE
: Open sockets on privileged ports (DHCPv6).CAP_NET_RAW
: Join multicast groups, open ICMPv6 socket for RPL.
They can be assigned to the wsbrd
binary using:
sudo setcap "cap_net_admin,cap_net_bind_service,cap_net_raw=+ep" $(which wsbrd)
Note that this is comparable to making wsbrd a "setuid" binary, but with
selected permissions. You can now run wsbrd
without root privileges.
While giving those permissions will enable wsbrd
to create its own network
interface, you can also choose to configure it manually. This process can be
useful to set up unusual configurations, or if you need to access tun interface
before wsbrd
is launched. Note that wsbrd
will still need the
aforementioned privileges to function whether you manually configure its
network interface or not.
First, create the network interface to give your user the permission to use it:
sudo ip tuntap add mode tun tun0 user YOUR_USER
The MTU must be set to 1280 bytes to comply with 802.15.4g:
sudo ip link set dev tun0 mtu 1280
Silicon Labs suggests reducing the queue size of the interface to avoid huge latencies:
sudo ip link set dev tun0 txqueuelen 10
The Wi-SUN interface cannot be configured through SLAAC, so do not pollute your network with unnecessary Router Solicitations:
sudo sysctl net.ipv6.conf.tun0.accept_ra=0
Wi-SUN needs a link-address matching the EUI64 of the node. Therefore, Linux should not generate any link-local address by itself.
sudo sysctl net.ipv6.conf.tun0.addr_gen_mode=1
Then, wsbrd
can automatically set up the IP addresses (Global and Link-Local)
of the interface. However, to run without root privileges, you have to do it
yourself.
Disable the tun_autoconf
parameter in wsbrd
's configuration. Then add IP
addresses:
sudo ip addr add dev tun0 fe80::200:5eef:1000:1/64
sudo ip addr add dev tun0 2001:db8::200:5eef:1000:1/64
The 64 least significant bits of these addresses must match with the EUI-64 of
the RCP (you can check logs of wsbrd
to find it).
The network mask of the GUA must match with the ipv6_prefix
parameter.
Finally, bring up the interface:
sudo ip link set dev tun0 up
Finally, you can run wsbrd
.
Transparent IPv6 proxy provides IPv6 connectivity to the Wi-SUN network without changing configuration of existing IPv6 infrastructure. Once enabled:
- The Wi-SUN nodes will appear as classical hosts on the network.
- The other hosts on the network will be able to reach them.
- The Wi-SUN nodes will be able to reach the Internet through the gateway of the host.
- If the upstream gateway provides global addresses and there is no firewall on the way (which is uncommon), hosts on the Internet can reach the Wi-SUN nodes.
To enable this feature:
- The
neighbor_proxy
parameter must be set to the name of the upstream network interface. - The
ipv6_prefix
parameter must be set to the same prefix as the hosting network. - IPv6 forward must be enabled on the host (with
sysctl net.ipv6.conf.all.forwarding=1
). Note that enabling forwarding per interface does not work.
Under the hood, when neighbor_proxy
is in use:
- NDP proxy (
/proc/sys/net/ipv6/conf/*/proxy_ndp
) is enabled. - Wi-SUN nodes are automatically added to the neighbor proxy list (user can
dump them with
ip -6 neigh show proxy
). - IPv6 routes are automatically added for the Wi-SUN nodes (user can dump
them with
ip -6 route show
). - The delay before answering multicast neighbor solicitations
(
/proc/sys/net/ipv6/neigh/*/proxy_delay
) is set to 0.
CPC protocol relies on an external service (CPCd). Therefore, plain UART allows an easier integration for simple setups. However, CPC offers some features:
- Support for SPI bus.
- Support for encrypted link with the RCP.
- Support for Dynamic MultiProtocol (DMP). Thus, CPCd can share the RCP between several network stacks (that is, Bluetooth, Zigbee, OpenThread, and Wi-SUN)
Wi-SUN FAN 1.1 is putting an emphasis on Low Function Nodes (LFNs).
Unfortunately, the way the standard was written does not allow to safely mix
FAN 1.0 devices with LFNs. The default configuration provided allows LFNs but
refuses FAN 1.0 routers. Users need to explicitly set enable_ffn10 = yes
in
their configuration in order to connect legacy devices.
The last update of GitHub seems incompatible with the git version bundled with Rust 1.45. The issue and the workaround are described here and the root cause is solved here.
Before launching cmake
, you can run:
export CARGO_NET_GIT_FETCH_WITH_CLI=true
First, check you have followed the installation process. Especially, check you
have run ninja install
.
There are several DBus instances on your system:
- One system instance
- An instance for each user
By default, wsbrd
tries to use the user
instance and falls back to system
instance.
The DBus session used is shown in the first lines of the log output:
Successfully registered to system DBus
Then, use busctl --system
or busctl --user
accordingly.
Note that if you use sudo
to launch wsbrd
as root user, it will use the
system
instance.
You can enforce the session used with an environment variable
DBUS_STARTER_BUS_TYPE=system
or DBUS_STARTER_BUS_TYPE=user
. If you use
sudo
, you must define this variable inside the sudo
environment:
sudo env DBUS_STARTER_BUS_TYPE=system wsbrd ...
Low function nodes (LFN) that are connected directly to the border router do
not appear as having parent when querrying the D-Bus Nodes
API. This is due
to them being routed differently, and will be fixed in an future version. Tools
like wsbrd_cli
and the web GUI that rely on this D-Bus API are affected by
this limitation.
Upon restart, wsbrd
restores its IPv6 neighbor cache and its RPL routes from
storage. However, it has lost its neighbors' frequency hopping timing
information. Therefore, right after a restart, downstream traffic is not
possible. These frequency hopping timing information will be restored once
wsbrd
receives a PAN Configuration or a Data frame from each neighbor. From
there, downstream traffic within the network will resume. Note that to be able
to send Data Frames upward, wsbrd
's neighbors must have updated wsbrd
's
frequency hopping timing information as well. Without this, upstream traffic
will not be possible either.
Finally, it is important to know that communication with neighboring LFN devices
cannot be resumed the same way after a restart. LFN devices expect to receive an
LFN Time Sync (LTS) Frame every so often depending on your network parameters.
After a restart, since wsbrd
does not have the necessary frequency hopping
timing information to send such frames, LFN devices will consider that their
parent is not present anymore and perform a full join procedure.
Path MTU Discovery works as expected on the Wi-SUN network. The Border Router
replies with ICMPv6/Packet Too Big
if necessary. (Remember that in IPv6,
routers cannot fragment packets, therefore the sender is responsible for the
size of the packet). Direct neighbors of the Border Router can receive frames up
to 1504 bytes, while the other nodes can receive frames up to 1280 bytes.
If you try to send a UDP frame larger than the MTU, there are two options:
- The packet has been sent with
IPV6_DONTFRAG
, and the operating system will return an error. - The packet is not marked with
IPV6_DONTFRAG
, and the operating system will fragment the packet.
On the receiver, the buffer must be large enough (up to 64 kB) to handle the fragmented packet. This feature is sometimes limited on embedded devices. IPv6 requires at least 1500 bytes available during reception, and warns on sending more:
A node must be able to accept a fragmented packet that, after reassembly, is as large as 1500 octets. A node is permitted to accept fragmented packets that reassemble to more than 1500 octets. An upper-layer protocol or application that depends on IPv6 fragmentation to send packets larger than the MTU of a path should not send packets larger than 1500 octets unless it has assurance that the destination is capable of reassembling packets of that larger size.
Typically, on Silicon Labs nodes, the default fragmentation buffer size is 1504 bytes. Therefore, if you send a buffer greater than 1504 bytes (including IP and MAC headers), the packet will be silently dropped.
As another consequence, the commonly used tool nc
cannot be used to stress
Wi-SUN networks with a continuous data stream, as nc
sends 16 kB-long UDP
when fed an endless source such as /dev/urandom
. There is no option to reduce
frame size (or to enable IPV6_DONTFRAG
).
Therefore, sending UDP packets with IPV6_DONTFRAG
is recommended. Use
IPV6_PATHMTU
and IPV6_RECVPATHMTU
to determine the optimal packet size.
Read RFC 8900 for more insights on the question.