Skip to content

Latest commit

 

History

History
390 lines (288 loc) · 13.4 KB

README.md

File metadata and controls

390 lines (288 loc) · 13.4 KB
Raw

NetBox - Like BusyBox but for Networking

Westermo NetBox is a toolbox for embedded systems based on Buildroot.

NetBox Zero/OS booting up in Qemu

NetBox provides easy access to all Westermo specific custimizations made to Linux and other Open Source projects used in WeOS. You can use it as the base for any application, but is strongly recommended for all use cases for container applications running in WeOS. Official WeOS container applications will be based on NetBox.

NetBox is built using the Buildroot External Tree facility. This is a layered approach which enables customizing without changing Buildroot. You may use NetBox as NetBox use Buildroot, see the App-Demo project for an example -- click Use this template -- to create your own.

To contribute, see the file HACKING for details.

Versioning

NetBox use the same versioning as Buildroot, with an appended -rN to denote the revision of Buildroot with Westermo extensions. E.g., the first release is 2020.02-r1.

Platforms

The NetBox project follows the Westermo product platform naming. This to be able to easily match what container image works on a Westermo device:

Architecture Platform Name Nightly App Nightly OS
arm9 Basis basis.app basis.os
powerpc Coronet coronet.app coronet.os
arm pj4 Dagger dagger.app dagger.os
aarch64 Envoy envoy.app envoy.os
aarch64 Ember [ember.app][] [ember.os][]
x86_64 Zero zero.app zero.os

Note: the Envoy platform includes support also for the Marvell ESPRESSObin (Globalscale) and MACCHIATObin (Solidrun) boards.

Flavor

In addition to various NetBox platforms there are two major flavors available. The current first-class citizen is apps, but it is also possible to build an entire operating system image, including Linux kernel and the same userland already available to apps. To select a pre-configured NetBox flavor for a given platform:

  • netbox_app_$platform
  • netbox_os_$platform

Requirements

The build environment requires the following tools, tested on Ubuntu 21.04 (x86_64): make, gcc, g++, m4, python, and openssl development package.

On Debian/Ubuntu based systems:

~$ sudo apt install build-essential m4 libssl-dev python

To run in Qemu, either enable host-side build in make menuconfig, or for quicker builds you can use the version shipped with your Linux host.

On Debian/Ubuntu based systems:

~$ sudo apt install qemu-system

For smooth sailing, after install, add the following line to the file /etc/qemu/bridge.conf (add file if it does not exist):

allow all

For network access to work out of the box in your Qemu system, install the virt-manager package, this creates a host bridge called virbr0:

~$ sudo apt install virt-manager

Building

First clone the repository, optionally check out the tagged release you want to use. The build system clones the submodule on the first build, but you can also run the command manually:

~$ cd ~/src
~/src$ git clone https://github.com/westermo/netbox.git
~/src$ cd netbox
~/src/netbox$ git submodule update --init

Second, select your target _defconfig, see the configs/ directory, or use make list-defconfigs to see all Buildroot and NetBox configs available. We select the defconfig for Zero (x86-64) NetBox app flavor:

~/src/netbox$ make netbox_app_zero_defconfig

Note: if you want to use the gdbserver on target, this is the point where you have to enable it in make menuconfig. The setting you want is under Toolchain --> "Copy gdb server to the Target". You also want "Build options" --> "build packages with debugging symbols"

Third, type make and fetch a cup of coffee because the first time you build it will take some time:

~/src/netbox$ make

Done. See the output/images/ directory for the resulting SquasFS based root file system: netbox-app-zero.img

Tip: the same source tree can easily be used to build multiple defconfigs, use the Buildroot O= variable to change the default output/... to O=/home/$LOGNAME/src/netbox/zero in one terminal window, and O=/home/$LOGNAME/src/netbox/coronet in another. This way, when working with packages, e.g. editing code, you can build for multiple targets at the same time, without cleaning the tree.

Updating

To update your local copy of NetBox from git, you need to update both NetBox and the Buildroot submodule, like when you first cloned (above):

~/src/netbox$ git pull
~/src/netbox$ git submodule update --init

Running

All NetBox OS builds are supported by Qemu. This is actually a corner stone in NetBox, and principal testing strategy at Westermo. It can be highly useful for quick turnarounds when developing and testing new features. Make sure you have built one of the os images before running, e.g.:

~/src/netbox$ make netbox_os_zero_defconfig

Any feature targeting OSI layer 3, and above, need nothing else to run. For more advanced test setups, with multiple networked Qemu nodes, we highly recommend Qeneth.

To start a single node:

~/src/netbox$ make run

Note: you may need sudo, unless you have set up your system with capabilities https://troglobit.com/2016/12/11/a-life-without-sudo/

Basic Networking in Qemu

By default, this command starts the utils/qemu script and tries to connect one interface to a host bridge called virbr0. That bridge only exists if you installed virt-manager (above), if not, you can have a look at the utils/qemu script arguments and environment variables, or try:

~/src/netbox$ make QEMU_NET=tap run

Persistent Storage in Qemu

Qemu nodes start from the same read-only SquasFS image as built for all targets. For persistent storage a disk image file on the host system is used. This is controlled by the environment variable $QEMU_MNT, which defaults to VENDOR-config-PLATFORM.img, provided ~/.cache exists . E.g., for NetBox Zero OS: ~/.cache/netbox-config-zero.img. See the helper script utils/qemu for more information.

When persistent storage is enabled and working, the /mnt directory on the target system is used to storing an OverlayFS of the target's /etc, /root, and /var, directories. I.e., changing a file in either of these directories (exceptions in /var exist) is persistent across reboots.

Sharing a Host Directory with Qemu

NetBox support 9P file sharing between the host and Qemu targets. Set directory to share, using the absolute path, in QEMU_HOST:

~/src/netbox$ make run QEMU_HOST=/tmp

When booting your target system with make run, the hosts' /tmp directory is available as /host on the target system.

Example

Here is an example run of a Zero OS build, the persistent store for all your configuration (in /etc or /home) is stored in a disk image file named ~/.cache/netbox-config-zero.img:

~/src/netbox$ make distclean
~/src/netbox$ make netbox_os_zero_defconfig
~/src/netbox$ make
~/src/netbox$ make run

Note: you may still need to call sudo make run, see the note on capabilities, above.

Debugging with gdbserver

if you remembered "Copy gdb server to the target", above, we can debug failing programs on our target (Qemu) system. You also need to have the gdb-multiarch program installed on your host system, the regular gdb only supports your host's architecture:

sudo apt install gdb-multiarch

To open the debug port in Qemu we start NetBox with QEMU_GDB=1, this opens localhost:4712 as your debug port (4711 is used for kgdb):

$ make run QEMU_GDB=1

When logged in, start the gdbserver service:

# initctl enable gdbserver
# initctl reload

From your host, in another terminal (with the same $O set!) from the same NetBox directory, you can now connect to your target and the attach to, or remotely start the program you want to debug. NetBox has a few extra tricks up its sleeve when it comes to remote debugging. The below commands are defined in the .gdbinit file:

$ make debug
GNU gdb (Ubuntu 9.2-0ubuntu1~20.04.1) 9.2
Copyright (C) 2020 Free Software Foundation, Inc.

For help, type "help".
(gdb) user-connect
(gdb) user-attach usr/sbin/querierd 488
0x00007f5812afc425 in select () from ./lib64/libc.so.6
(gdb) cont
Continuing.

For more information on how to use GDB, see the manual, or if you want to know a little bit more behind the scenes, see the blog post about Debugging an embedded system.

Running in LXC or LXD

The NetBox app builds can be run in LXC, or LXD, on your PC. With LXD it is even possible to run non-native archs, like Arm64 using the Linux "binfmt misc" mechanism, which runs all binaries through qemu-aarch64. This is only documented in the predecessor to NetBox, myrootfs][].

To run a NetBox app in LXC, first install all dependencies (lxc-utils, libvirt, etc.) and create the necessary directories:

$ sudo mkdir -p /var/lib/lxc/images/
$ sudo mkdir -p /var/lib/lxc/foo/mnt

Since we are playing it safe, we've built the Zero (x86_64) NetBox app, image, so let's install it in the images/ directory. Images can be shared with multiple LXC container apps:

$ sudo cp output/images/netbox-app-zero.img /var/lib/lxc/images/foo.img

The LXC config file might need some tweaking, in particular if you use different path to the .img file. The host bridge you probably want to change as well. Here we have used lxcbr0 only because it's the default in libvirt installs in Debian/Ubuntu and gives us NAT:ed access to the Internet from our app(s) via the host. All this is already set up by libvirt, so we can focus on the LXC container config:

$ sudo sh -c "cat >>/var/lib/lxc/foo/config" <<-EOF
	lxc.uts.name = foo
	lxc.tty.max = 4
	lxc.pty.max=1024
	#lxc.hook.pre-mount = pre-mount.sh /var/lib/lxc/images/foo.img /var/lib/lxc/foo/rootfs
	#lxc.rootfs.path    = overlayfs:/var/lib/lxc/foo/rootfs:/var/lib/lxc/foo/delta0
	#lxc.rootfs.options = -t squashfs
	lxc.rootfs.path = loop:/var/lib/lxc/images/foo.img
	lxc.mount.auto = cgroup:mixed proc:mixed sys:mixed
	#lxc.mount.entry=run run tmpfs rw,nodev,relatime,mode=755 0 0
	#lxc.mount.entry=shm dev/shm tmpfs rw,nodev,noexec,nosuid,relatime,mode=1777,create=dir 0 0
	lxc.mount.entry=/var/lib/lxc/foo/mnt mnt none bind 0 0
	lxc.net.0.type = veth
	lxc.net.0.flags = up
	lxc.net.0.link = lxcbr0
	#lxc.init.cmd = /sbin/init finit.debug
	
	#lxc.seccomp.profile = /usr/share/lxc/config/common.seccomp
	lxc.apparmor.profile = lxc-container-default-with-mounting
EOF

The last two lines are needed on systems with Seccomp and/or AppArmor. Uncomment the one you need, see the host's dmesg when lxc-start fails with mysterious error messages. For convenience the Debian/Ubuntu is uncommented already.

Note1: You may have to create the directory where to mount the container rootfs as it is configured in /var/lib/lxc/foo/config on line 11

$ sudo mkdir -p /var/lib/lxc/foo/mnt

Note2: you may have to add the following two lines to your AppArmor profile to enable writable /etc, /var, /home, and /root directories. The file is in /etc/apparmor.d/lxc/lxc-default-with-mounting:

mount fstype=tmpfs,
mount fstype=overlay,

Reload AppArmor, or restart your system to activate the changes, then we can start the container with:

$ sudo lxc-start -n foo

To see what actually happens when it starts up, append -F. Attach to the container's /dev/console with:

$ sudo lxc-console -n foo -t 0 -e '^p'

The last -e '^p remaps the control key sequence to detach from your container and return to your host: Ctrl-p q