Running a bleeding-edge rolling release Linux distribution like Arch Linux has its challenges and risks, but there is perhaps no greater feeling of absolute control over your own operating system! It also leads to the opportunity to have “adventures” like the one described here. While “adventure” might be tongue-in-cheek, the truth is that there is great educational value in breaking something and thus being forced to fix it!

I recently made the mistake of seeing that a pacman upgrade to icu (International Components for Unicode) would break a few Electron packages, which I needed for the open source build of VS Code, among a few other options. (I generally like to avoid Electron apps, but VS Code is a rare exception!)

I was notified by pacman that there was a dependency issue here that could not be solved, so that the entire transaction was not possible.

My mistake was thinking “oh, I’ll just exclude icu from the upgrade, as it’s causing the dependency issue”. Usually these issues are fixed in a day or two, so I’d upgrade it later.

This was indeed a mistake.

Somehow, apps were now referencing the new icu that wasn’t there already. I rebooted, to discover that the GUI wouldn’t start, and I couldn’t even use pacman in single-user mode to roll back the previous transaction, as pacman itself depended upon icu.

I’d broken the very package manager that I needed to roll back. Oops!

Install media to the rescue

I booted into the Arch install media, and set about to address the issue.

chrooting into the target filesystem and using pacman wasn’t an option, as running the package manager from the target filesystem failed with the same dependency issue.

I had a few false starts with trying to run pacman from the install medium, pointing it at the mounted OS partition with --sysroot.

While this didn’t pan out, I remembered that pacstrap, while designed for install-time, presumably could install packages in the target filesystem without needing pacman itself. I was concerned that this might reinstall some base packages, but it turns out you can specify which packages to install manually on its command line.

So, I gathered the list of packages that needed upgrading by using pacman -Syu --sysroot /mnt against the target filesystem, and then supplied this package list to pacstrap:
pacstrap -G -i -M /mnt icu brltty electron27 electron28 electron29 electron30 freerdp freerdp2 harfbuzz-icu raptor
And… we’re back!

Lessons learned

• The manual isn’t joking about “Partial upgrades are unsupported”.
• Considerable caution is required when using pacman --exclude
• “The OS is ephemeral and I can rebuild it” is a bit too relaxed an attitude when you don’t really want to have to rebuild the OS at short notice. Back up the OS packages and libraries too.
• Really understanding the package manager and install process gives you the tools to pull yourself out of the holes you dig for yourself!

I frequently SSH into various systems from my primary Linux machine. There is an analogous issue to “too many browser tabs” that exists here — having too many SSH sessions open in different terminal tabs!

There is a risk in these cases of accidentally typing a higher-privileged sudo password into a lower security system by typing into the wrong terminal. There are various approaches that can help here; I have used screen banners with different colours before.

A good “last line of defence” approach to this risk that I have settled on is to make use of sudo‘s “lectures”. You will have seen the default:

We trust you have received the usual lecture from the local System Administrator. It usually boils down to these three things:
#1) Respect the privacy of others.
#2) Think before you type.
#3) With great power comes great responsibility.

We can customise this, and also set it to always show, rather than just the first time you ever use sudo on that machine. We’ll create a custom lecture file with our desired text — in my case, the hostname I’m logged into, so I’m sure where I am before I type the password!

Then, use visudo to set these options:

Defaults lecture=always
Defaults lecture_file=/etc/custom_sudo_lecture

I am currently trying to migrate away from VMware Workstation on a Linux host (thanks, Broadcom!) to using KVM virtual machines.

I run many Linux virtual machines, and in many of these I use the GUI, so want well performing video and sound! This was one of the original justifications for VMware Workstation.

The default configuration on Arch was to use the Virtio video driver and a “Display Spice” entry in the VM configuration to support video and sound output.

However, I would experience choppy audio and video output. On the display, it seemed that, roughly vertically, some of the pixels in the framebuffer would not be updated, especially when lots of display changes occurred. This created a sort of irregular “interlaced” look, as you can see here as I have moved the mouse down the menu:

I have been able to work around this issue by enabling OpenGL acceleration in the guest.

Configuration

For an Arch guest, it needs the package qemu-hw-display-virtio-gpu installed.

Check the guest is showing the virtio_gpudrmfb frame buffer device:

# dmesg | grep '\[drm'
...
virtio-pci 0000:00:01.0: [drm] fb0: virtio_gpudrmfb frame buffer device

The guest configuration is as follows:

3D acceleration enabled in the Video Virtio driver. Ensure this is ticked.

On the Display Spice object, the configuration is set to Listen type None and OpenGL is set to on.

References

The idea that enabling OpenGL with the virtio video driver may help was derived from https://www.kraxel.org/blog/2016/09/using-virtio-gpu-with-libvirt-and-spice/.

I bought a used LTO4 tape drive with a 8088 SAS connection. Why?

For fun, for backups that feel like they might be more resilient than the shingled magnetic recording hard drives I accidentally bought (thanks Seagate for disclosing that), and for the enjoyment of something so wonderfully mechanical in a world that is very “solid state”.

This necessitated a SAS card purchase, to give myself the ports necessary to actually plug in the tape drive. It seemed unhappy with one of my PCI Express slots, so I moved it up to the primary PCI Express slot — the one you’d usually use for a graphics card.

Now this Arch Linux machine has no need for fancy graphics. The APU integrated graphics on the Ryzen 7 5700G are perfectly adequate.

However, once the SAS card was in the primary PCI Express slot, X11 would no longer start. My SAS card showed up beautifully with lspci, as did the tape drive with lsscsi, but I had to sacrifice the GUI for it. Seems a little extreme, even for me.

X11 would fail with “no screens found” when the amdgpu driver was enumerating screens.

The integrated graphics moved PCI ID

What had happened is that once something is in that primary PCI Express slot, the integrated graphics moved their PCI bus ID.

I first identified where the “VGA controller” had gone with lspci:

08:00.0 VGA compatible controller: Advanced Micro Devices, Inc. [AMD/ATI] Cezanne [Radeon Vega Series / Radeon Vega Mobile Series] (rev c8)

Then I edited /etc/X11/xorg.conf.d/amdgpu.conf to point the BusID at that new identifier.

For me, it had moved from PCI:7:0:0 to PCI:8:0:0.

And now, I have the delight of a GUI and a SAS card, and a tape drive.

In a number of ways, I have become a dinosaur. One of those ways is an insistence that my password management solution has one of its layers of security being the fact that the data is not in the cloud. There is nothing on someone else’s commodity service, and, thus, no data to potentially be compromised in bulk as a result of an attack on a common service. Those services are probably pretty secure, but they are also huge, concentrated targets.

I have a really high standard to meet for the security of password management for my personal computing. That is how I feel comfortable.

So the march of password managers into the cloud presented me a problem. To move forward and to maintain my investment became incompatible with this principle.

Additionally, I became more and more disillusioned with macOS as a desktop platform. Increasingly, maintaining control over what the system is actually doing became impossible. (Why is the News app refreshing in the background when I have never opened it on this brand new install??) So the new solution needs to be native to the Linux desktop.

So my requirements discounted me from my long-time password management app 1Password (clearly moving towards cloud only) — a migration of some kind was in order.

The Options

I have found myself with a combination of KeePassXC and Strongbox on iOS.

I do sacrifice some convenience on the desktop with browser integration, as I have not yet installed the browser extension for KeePassXC. I would love it to be in Mozilla’s “Recommended” category, where they review the extension on an ongoing basis. I trust the KeePassXC developers are not malicious, but there are lots of risks with browser integration that I don’t fully understand the implications of — the boundary from the external app into the browser context affects the principle of data isolation in ways I haven’t studied.

So, KeePassXC + Strongbox + local network syncing of data is where I have landed.

And because supporting the projects that we depend upon is important, I have paid for the premium version of Strongbox and made an annual donation to the KeePassXC project. I am one of a decreasing number of people who will want to maintain this level of control, so, given that I am fortunate enough to be able to, providing the resources to keep this alive is something I wanted to do.

I recently had some success installing OpenWRT 19.07.2 on my TP-Link WA801ND v5 wireless access point, which I was happy to do since the stock firmware was untouched since 2017.

I’ve documented the process in a video. This isn’t a tutorial with absolutely everything covered — I don’t want to encourage people to brick their devices, so you’ll need competence and proficiency with configuring your network settings in your operating system, and running a TFTP server.

It’s all, obviously, at your own risk. Stay safe and don’t brick your only access point!

Sorry for the inconsistency in audio quality.

UPDATE: This does not work since Raspberry Pi OS based on bullseye. sdm might be a good alternative approach.

I’ve been playing with some Raspberry Pi Zero W machines for a few projects. They are inexpensive, with a form factor that makes for all sorts of interesting possibilities.

They are also a pain to initially provision, because you need adapter cables for mini-HDMI and micro-USB to get a monitor and keyboard connected.

With some help from this post from Linuxconfig.org, here’s what I’ve been doing to get the Pi Zero W up and running, on the network and ready for SSH access without having to plug anything in except power and the SD card!

Mount the image

$ fdisk -l *raspbian.img 

The initial FAT /boot partition lies at 8192 sectors in. A sector is 512 bytes. So, now we’ll mount the /boot partition using the loop pseudo-device.

$ mkdir boot
$ mount -o loop,offset=$(( 512 * 8192)) *raspbian.img boot

Set up SSH

Let’s go into our new boot subfolder where the image’s first partition is now mounted.

All we need to do to enable SSH is create an empty file called ssh.

$ cd boot
$ sudo touch ssh

Configure WiFi

To configure WiFi, we’ll need to drop a wpa-supplicant.conf file in this folder. Upon first boot, Raspbian moves this into the correct config location and the Pi will be able to talk to our network out of the box.

Create a file called wpa-supplicant.conf with these contents:

ctrl_interface=DIR=/var/run/wpa_supplicant GROUP=netdev
update_config=1
country=GB

network={
	ssid="Your network name"
	psk="Your network password"
	key_mgmt=WPA-PSK
}

You’ll need to replace the placeholders with your network name and password, and possibly change country too.

Unmount the image

$ cd ..
$ umount boot

Write the image

Insert the target SD card and verify which device it is using fdisk.

$ sudo fdisk -l

Disk /dev/sdb ...
Model: Card Reader

I am confident that sdb is correct, because I see from fdisk that the model is Card Reader.

$ sudo dd if=*raspbian.img of=/dev/sdb

Security

Remember that once you’ve booted, you’ve just spun up a Raspberry Pi on your network with the default username of pi and the default password of raspberry. Check your router or network logs for the assigned DHCP address of the new Pi, log in promptly over SSH and change that password!

This was a whole month and a bit ago, and I never got around to throwing it here on the site, but I will now announce that I am a Red Hat Certified System Adminstrator!

Even despite the dominance of webmail, I have long used a traditional desktop email client. I like having a local mail archive should “the cloud” have trouble, as well as the ability to exert control over the user interface and user experience. (That might be partly a euphemism for not having to see ads!)

Apple’s Mail.app built into macOS (going to have to get used to not calling it OS X!) has served me pretty well for quite some time now, alongside Thunderbird when I’m on Linux, and while Mail.app offered the most smooth interface for the platform, it didn’t always have all the features I wanted.

For example, being able to run mail rules is more limited than I wanted in Mail.app. I could have rules run automatically as messages arrived in my inbox, or disable them entirely. But actually how I wanted to use rules was to be able to cast my eye over my inbox, and then bulk archive (to a specific folder) all emails of a certain type if I’d decided none needed my fuller attention.

Recently, I moved to Thunderbird on my Mac for managing email and discovered QuickArchiver.

As well as letting you writing rules yourself, QuickArchiver offers the clever feature of learning which emails go where, and then suggesting the right folder to which that message can be archived with a single click.

It’s still early days, but I am enjoying this. Without spending time writing rules, I’m managing email as before, and QuickArchiver is learning in the background what rules should be offered. The extra column I’ve added to my Inbox is now starting to populate with that one-click link to archive the message to the correct folder!

It’s just a nice little add-on if, like me, you (still??) like to operate in this way with your email.

I’m very excited to have put the ideas mentioned in my previous blog post about Cautious Unattended Upgrades into practice!

To quickly recap, the idea is that, on a Debian-based test system (‘the canary’), this is a software package that runs the latest security updates, runs an automated browser-based test suite to make sure these new updates have not broken any critical functionality on our clients’ sites, then ‘pushes’ just these package updates to the production servers.

In keeping with my original plan, the software is written in Ruby and uses Watir/Selenium WebDriver to run a suite of tests that verify, just as a human being would in a live web browser, that client websites work correctly.

Cautious Unattended Upgrades — the canary in the coal mine. Image by stevep2008 on Flickr, licensed under CC-BY 2.0.

I was expecting the biggest challenge would be getting this browser automation side of things working, but actually that proved very easy, which is a testament to the design of those projects.

The software is still a little rough around the edges, as I explain in the README file on GitHub, but I’m very pleased with the project’s progress. We have put it into use on our live systems at Van Patten Media, so we can keep servers promptly up-to-date with security patches without our intervention, but retain a greater peace of mind that our clients’ sites are still working as they should post-upgrade. (This is of course dependent on the quality and breadth of the tests that we write!)

I am particularly excited that this marks the first ‘real’ project in Ruby that I have written. Ruby isn’t a platform I have worked with too extensively before, so I have enjoyed challenging myself to be exposed to a different environment and quickly pick up how to achieve what I want to do. There is definitely more work to do — it really should be organised in a slightly more ‘Ruby-like’ way, and perhaps become a proper Ruby Gem, listed on rubygems.org, so those are things I will be looking at over the longer term for this project.

If you are interested in using Cautious Unattended Upgrades, or contributing to making it better, the project is licensed under a BSD-style licence and the code is available on its GitHub project page.