Search Results: "luk"

18 July 2023

Lukas M rdian: A declarative approach to Linux networking with Netplan

Photo by Taylor Vick (Unsplash)
Linux networking can be confusing due to the wide range of technology stacks and tools in use, in addition to the complexity of the surrounding network environment. The configuration of bridges, bonds, VRFs or routes can be done programmatically, declaratively, manually or with automated with tools like ifupdown, ifupdown2, ifupdown-ng, iproute2, NetworkManager, systemd-networkd and others. Each of these tools use different formats and locations to store their configuration files. Netplan, a utility for easily configuring networking on a Linux system, is designed to unify and standardise how administrators interact with these underlying technologies. Starting from a YAML description of the required network interfaces and what each should be configured to do, Netplan will generate all the necessary configuration for your chosen tool. In this article, we will provide an overview of how Ubuntu uses Netplan to manage Linux networking in a unified way. By creating a common interface across two disparate technology stacks, IT administrators benefit from a unified experience across both desktops and servers whilst retaining the unique advantages of the underlying tech. But first, let s start with a bit of history and show where we are today.

The history of Netplan in Ubuntu Starting with Ubuntu 16.10 and driven by the need to express network configuration in a common way across cloud metadata and other installer systems, we had the opportunity to switch to a network stack that integrates better with our dependency-based boot model. We chose systemd-networkd on server installations for its active upstream community and because it was already part of Systemd and therefore included in any Ubuntu base installation. It has a much better outlook for the future, using modern development techniques, good test coverage and CI integration, compared to the ifupdown tool we used previously. On desktop installations, we kept using NetworkManager due to its very good integration with the user interface. Having to manage and configure two separate network stacks, depending on the Ubuntu variant in use, can be confusing, and we wanted to provide a streamlined user experience across any flavour of Ubuntu. Therefore, we introduced Netplan.io as a control layer above systemd-networkd and NetworkManager. Netplan takes declarative YAML files from /etc/netplan/ as an input and generates corresponding network configuration for the relevant network stack backend in /run/systemd/network/ or /run/NetworkManager/ depending on the system configuration. All while keeping full flexibility to control the underlying network stack in its native way if need be.
Design overview (netplan.io)

Who is using Netplan? Recent versions of Netplan are available and ready to be installed on many distributions, such as Ubuntu, Fedora, RedHat Enterprise Linux, Debian and Arch Linux.

Ubuntu As stated above, Netplan has been installed by default on Ubuntu systems since 2016 and is therefore being used by millions of users across multiple long-term support versions of Ubuntu (18.04, 20.04, 22.04) on a day-to-day basis. This covers Ubuntu server scenarios primarily, such as bridges, bonding, VLANs, VXLANs, VRFs, IP tunnels or WireGuard tunnels, using systemd-networkd as the backend renderer. On Ubuntu desktop systems, Netplan can be used manually through its declarative YAML configuration files, and it will handle those to configure the NetworkManager stack. Keep reading to get a glimpse of how this will be improved through automation and integration with the desktop stack in the future.

Cloud It might not be as obvious, but many people have been using Netplan without knowing about it when configuring a public cloud instance on AWS, Google Cloud or elsewhere through cloud-init. This is because cloud-init s Networking Config Version 2 is a passthrough configuration to Netplan, which will then set up the underlying network stack on the given cloud instance. This is why Netplan is also a key package on the Debian distribution, for example, as it s being used by default on Debian cloud images, too.

Our vision for Linux networking We know that Linux networking can be a beast, and we want to keep simple things simple. But also allow for custom setups of any complexity. With Netplan, the day-to-day networking needs are covered through easily comprehensible and nicely documented YAML files, that describe the desired state of the local network interfaces, which will be rendered into corresponding configuration files for the relevant network stack and applied at (re-)boot or at runtime, using the netplan apply CLI. For example /etc/netplan/lan.yaml:
network:
  version: 2
  renderer: networkd
  ethernets:
    enp3s0:
      dhcp4: true
Having a single source of truth for network configuration is also important for administrators, so they do not need to understand multiple network stacks, but can rely on the declarative data given in /etc/netplan/ to configure a system, independent of the underlying network configuration backend. This is also very helpful to seed the initial network configuration for new Linux installations, for example through installation systems such as Subiquity, Ubuntu s desktop installer or cloud-init across the public and private clouds. In addition to describing and applying network configuration, the netplan status CLI can be used to query relevant data from the underlying network stack(s), such as systemd-networkd, NetworkManager or iproute2, and present them in a unified way.
Netplan status (Debian)
At the Netplan project we strive for very high test automation and coverage with plenty of unit tests, integration tests and linting steps, across multiple Linux distros, which gives high confidence in also supporting more advanced networking use cases, such as Open vSwitch or SR-IOV network virtualization, in addition to normal wired (static IP, DHCP, routing), wireless (e.g. wwan modems, WPA2/3 connections, WiFi hotspot, controlling the regulatory domain, ) and common server scenarios. Should there ever be a scenario that is not covered by Netplan natively, it allows for full flexibility to control the underlying network stack directly through systemd override configurations or NetworkManager passthrough settings in addition to having manual configuration side-by-side with interfaces controlled through Netplan.

The future of Netplan desktop integration On workstations, the most common scenario is for end users to configure NetworkManager through its user interface tools, instead of driving it through Netplan s declarative YAML files, which makes use of NetworkManager s native configuration files. To avoid Netplan just handing over control to NetworkManager on such systems, we re working on a bidirectional integration between NetworkManager and Netplan to further improve the single source of truth use case on Ubuntu desktop installations. Netplan is shipping a libnetplan library that provides an API to access Netplan s parser and validation internals, that can be used by NetworkManager to write back a network interface configuration. For instance, configuration given through NetworkManager s UI tools or D-Bus API can be exported to Netplan s native YAML format in the common location at /etc/netplan/. This way, administrators just need to care about Netplan when managing a fleet of Desktop installations. This solution is currently being used in more confined environments, like Ubuntu Core, when using the NetworkManager snap, and we will deliver it to generic Ubuntu desktop systems in 24.04 LTS. In addition to NetworkManager, libnetplan can also be used to integrate with other tools in the networking space, such as cloud-init for improved validation of user data or installation systems when seeding new Linux images.

Conclusion Overall, Netplan can be considered to be a good citizen within a network environment that plays hand-in-hand with other networking tools and makes it easy to control modern network stacks, such as systemd-networkd or NetworkManager in a common, streamlined and declarative way. It provides a single source of truth to network administrators about the network state, while keeping simple things simple, but allowing for arbitrarily complex custom setups.
If you want to learn more, feel free to follow our activities on Netplan.io, GitHub, Launchpad, IRC or our Netplan Developer Diaries blog on discourse.

10 July 2023

Lukas M rdian: Netplan and systemd-networkd on Debian Bookworm

Debian s cloud-images are using systemd-networkd as their default network stack in Bookworm. A slim and feature rich networking daemon that comes included with Systemd itself. Debian s cloud-images are deploying Netplan on top of this as an easy-to-use, declarative control layer. If you want to experiment with systemd-networkd and Netplan on Debian, this can be done easily in QEMU using the official images. To start, you need to download the relevant .qcow2 Debian cloud-image from: https://cloud.debian.org/images/cloud/bookworm/latest/
$ wget https://cloud.debian.org/images/cloud/bookworm/latest/debian-12-generic-amd64.qcow2

Prepare a cloud image Next, you need to prepare some configuration files for cloud-init and Netplan, to prepare a data-source (seed.img) for your local cloud-image.
$ cat > meta.yaml <<EOF
instance-id: debian01
local-hostname: cloudimg
EOF
$ cat > user.yaml <<EOF
#cloud-config
ssh_pwauth: true
password: test
chpasswd:
  expire: false
EOF
$ cat > netplan.yaml <<EOF
network:
  version: 2
  ethernets:
    id0:
      match:
        macaddress: "ca:fe:ca:fe:00:aa"
      dhcp4: true
      dhcp6: true
      set-name: lan0
EOF
Once all configuration is prepared, you can create the local data-source image, using the cloud-localds tool from the cloud-image-utils package:
$ cloud-localds --network-config=netplan.yaml seed.img user.yaml meta.yaml

Launch the local VM Now, everything is prepared to launch a QEMU VM with two NICs and do some experimentation! The following command will launch an ephemeral environment for you, keeping the original Debian cloud-image untouched. If you want to preserve any changes on disk, you can remove the trailing -snapshot parameter.
$ qemu-system-x86_64 \
  -machine accel=kvm,type=q35 \
  -cpu host \
  -m 2G \
  -device virtio-net-pci,netdev=net0,mac=ca:fe:ca:fe:00:aa \
  -netdev user,id=net0,hostfwd=tcp::2222-:22 \
  -nic user,model=virtio-net-pci,mac=f0:0d:ca:fe:00:bb \
  -drive if=virtio,format=qcow2,file=debian-12-generic-amd64.qcow2 \
  -drive if=virtio,format=raw,file=seed.img -snapshot
We set up the default debian user account through cloud-init s user-data configuration above, so you can now login to the system, using that user with the (very unsafe!) password test .
$ ssh -o "StrictHostKeyChecking=no" -o "UserKnownHostsFile=/dev/null" -p 2222 debian@localhost # password: test

Experience Netplan and systemd-networkd Once logged in successfully, you can execute the netplan status command to check the system s network configuration, as configured through cloud-init s netplan.yaml passthrough. So you ve already used Netplan at this point implicitly and it did all the configuration of systemd-networkd for you in the background!
debian@cloudimg:~$ sudo netplan status -a
     Online state: online
    DNS Addresses: 10.0.2.3 (compat)
       DNS Search: .
   1: lo ethernet UNKNOWN/UP (unmanaged)
      MAC Address: 00:00:00:00:00:00
        Addresses: 127.0.0.1/8
                   ::1/128
           Routes: ::1 metric 256
   2: enp0s2 ethernet DOWN (unmanaged)
      MAC Address: f0:0d:ca:fe:00:bb (Red Hat, Inc.)
   3: lan0 ethernet UP (networkd: id0)
      MAC Address: ca:fe:ca:fe:00:aa (Red Hat, Inc.)
        Addresses: 10.0.2.15/24 (dhcp)
                   fec0::c8fe:caff:fefe:aa/64
                   fe80::c8fe:caff:fefe:aa/64 (link)
    DNS Addresses: 10.0.2.3
           Routes: default via 10.0.2.2 from 10.0.2.15 metric 100 (dhcp)
                   10.0.2.0/24 from 10.0.2.15 metric 100 (link)
                   10.0.2.2 from 10.0.2.15 metric 100 (dhcp, link)
                   10.0.2.3 from 10.0.2.15 metric 100 (dhcp, link)
                   fe80::/64 metric 256
                   fec0::/64 metric 100 (ra)
                   default via fe80::2 metric 100 (ra)
As you can see from this output, the lan0 interface is configured via the id0 Netplan ID to be managed by systemd-networkd. Compare this data to the netplan.yaml file above, the networkctl output, the local Netplan configuration in /etc/netplan/ and the auto-generated systemd-networkd configuration.
debian@cloudimg:~$ networkctl 
IDX LINK   TYPE     OPERATIONAL SETUP     
  1 lo     loopback carrier     unmanaged
  2 enp0s2 ether    off         unmanaged
  3 lan0   ether    routable    configured
3 links listed.
debian@cloudimg:~$ cat /etc/netplan/50-cloud-init.yaml 
# [...]
network:
    ethernets:
        id0:
            dhcp4: true
            dhcp6: true
            match:
                macaddress: ca:fe:ca:fe:00:aa
            set-name: lan0
    version: 2

debian@cloudimg:~$ ls -l /run/systemd/network/
total 8
-rw-r--r-- 1 root root  78 Jul  5 15:23 10-netplan-id0.link
-rw-r--r-- 1 root root 137 Jul  5 15:23 10-netplan-id0.network
Now you can go ahead and try something more advanced, like link aggregation, using the second NIC that you configured for this QEMU VM and explore all the possibilities of Netplan on Debian, by checking the Netplan YAML documentation.

9 July 2023

Vasudev Kamath: Using LUKS-Encrypted USB Stick with TPM2 Integration

I use a LUKS-encrypted USB stick to store my GPG and SSH keys, which acts as a backup and portable key setup when working on different laptops. One inconvenience with LUKS-encrypted USB sticks is that you need to enter the password every time you want to mount the device, either through a Window Manager like KDE or using the cryptsetup luksOpen command. Fortunately, many laptops nowadays come equipped with TPM2 modules, which can be utilized to automatically decrypt the device and subsequently mount it. In this post, we'll explore the usage of systemd-cryptenroll for this purpose, along with udev rules and a set of scripts to automate the mounting of the encrypted USB. First, ensure that your device has a TPM2 module. You can run the following command to check:
sudo journalctl -k --grep=tpm2
The output should resemble the following:
Jul 08 18:57:32 bhairava kernel: ACPI: SSDT 0x00000000BBEFC000 0003C6 (v02
LENOVO Tpm2Tabl 00001000 INTL 20160422) Jul 08 18:57:32 bhairava kernel:
ACPI: TPM2 0x00000000BBEFB000 000034 (v03 LENOVO TP-R0D 00000830
PTEC 00000002) Jul 08 18:57:32 bhairava kernel: ACPI: Reserving TPM2 table
memory at [mem 0xbbefb000-0xbbefb033]
You can also use the systemd-cryptenroll command to check for the availability of a TPM2 device on your laptop:
systemd-cryptenroll --tpm2-device=list
The output will be something like following:
blog git:(master) systemd-cryptenroll --tpm2-device=list
PATH        DEVICE      DRIVER
/dev/tpmrm0 MSFT0101:00 tpm_tis
   blog git:(master)
Next, ensure that you have connected your encrypted USB device. Note that systemd-cryptenroll only works with LUKS2 and not LUKS1. If your device is LUKS1-encrypted, you may encounter an error while enrolling the device, complaining about the LUKS2 superblock not found. To determine if your device uses a LUKS1 header or LUKS2, use the cryptsetup luksDump <device> command. If it is LUKS1, the header will begin with:
LUKS header information for /dev/sdb1
Version:        1
Cipher name:    aes
Cipher mode:    xts-plain64
Hash spec:      sha256
Payload offset: 4096
Converting from LUKS1 to LUKS2 is a simple process, but for safety, ensure that you backup the header using the cryptsetup luksHeaderBackup command. Once backed up, use the following command to convert the header to LUKS2:
sudo cryptsetup convert --type luks2 /dev/sdb1
After conversion, the header will look like this:
Version:        2
Epoch:          4
Metadata area:  16384 [bytes]
Keyslots area:  2064384 [bytes]
UUID:           000b2670-be4a-41b4-98eb-9adbd12a7616
Label:          (no label)
Subsystem:      (no subsystem)
Flags:          (no flags)
The next step is to enroll the new LUKS key for the encrypted device using systemd-cryptenroll. Run the following command:
sudo systemd-cryptenroll --tpm2-device=/dev/tpmrm0 --tpm2-pcrs="0+7" /dev/sdb1
This command will prompt you to provide the existing key to unseal the device. It will then add a new random key to the volume, allowing it to be unlocked in addition to the existing keys. Additionally, it will bind this new key to PCRs 0 and 7, representing the system firmware and Secure Boot state. If there is only one TPM device on the system, you can use --tpm2-device=auto to automatically select the device. To confirm that the new key has been enrolled, you can dump the LUKS configuration and look for a systemd-tpm2 token entry, as well as an additional entry in the Keyslots section. To test the setup, you can use the /usr/lib/systemd/systemd-cryptsetup command. Additionally, you can check if the device is unsealed by using lsblk:
sudo /usr/lib/systemd/systemd-cryptsetup attach GPG_USB "/dev/sdb1" - tpm2-device=auto
lsblk
The lsblk command should display the unsealed and mounted device, like this:
NAME        MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINTS
sda           8:0    0 223.6G  0 disk
 sda1        8:1    0   976M  0 part  /boot/efi
 sda2        8:2    0 222.6G  0 part
   root    254:0    0 222.6G  0 crypt /
sdb           8:16   1   7.5G  0 disk
 sdb1        8:17   1   7.5G  0 part
   GPG_USB 254:1    0   7.5G  0 crypt /media/vasudev/GPG_USB
Auto Mounting the device Now that we have solved the initial problem of unsealing the USB device using TPM2 instead of manually entering the key, the next step is to automatically mount the device upon insertion and remove the mapping when the device is removed. This can be achieved using the following udev rules:
ACTION=="add", KERNEL=="sd*", ENV DEVTYPE =="partition", ENV ID_BUS =="usb", ENV SYSTEMD_WANTS ="mount-gpg-usb@$env DEVNAME .service"
ACTION=="remove", KERNEL=="sd*", ENV DEVTYPE =="partition", ENV ID_BUS =="usb", RUN+="/usr/local/bin/umount_enc_usb.sh '%E ID_FS_UUID '"
When a device is added, a systemd service is triggered to mount the device at a specific location. Initially, I used a script with the RUN directive, but it resulted in an exit code of 32. This might be due to systemd-cryptsetup taking some time to return, causing udev to time out. To address this, I opted to use a systemd service instead. For device removal, even though the physical device is no longer present, the mapping may still remain, causing issues upon reinsertion. To resolve this, I created a script to close the luks mapping upon device removal. Below are the systemd service and script files: mount_enc_usb.sh:
#!/bin/bash
set -x
if [[ "$#" -ne 1 ]]; then
    echo "$(basename $0) <device>"
    exit 1
fi
device_uuid="$(blkid --output udev $1   grep ID_FS_UUID=   cut -d= -f2)"
if [[ "$device_uuid" == 000b2670-be4a-41b4-98eb-9adbd12a7616 ]]; then
    # Found our device, let's trigger systemd-cryptsetup
    /usr/lib/systemd/systemd-cryptsetup attach GPG_USB "$1" - tpm2-device=auto
    [[ -d /media/vasudev/GPG_USB ]]   (mkdir -p /media/vasudev/GPG_USB/ && chown vasudev:vasudev /media/vasudev/GPG_USB)
    mount /dev/mapper/GPG_USB /media/vasudev/GPG_USB
else
    echo "Not the interested device. Ignoring."
    exit 0
fi
umount_enc_usb.sh:
#!/bin/bash
if [[ "$#" -ne 1 ]]; then
  echo "$(basename $0) <fsuuid>"
  exit 1
fi
if [[ "$1" == "000b2670-be4a-41b4-98eb-9adbd12a7616" ]]; then
  # Our device is removed, let's close the luks mapping
  [[ -e /dev/mapper/GPG_USB ]] && cryptsetup luksClose /dev/mapper/GPG_USB
else
  echo "Not our device."
  exit 0
fi
mount-gpg-usb@.service:
[Unit]
Description=Mount the encrypted USB device service
[Service]
Type=simple
ExecStart=/usr/local/bin/mount_enc_usb.sh
With this setup, plugging in the USB device will automatically unseal and mount it, and upon removal, the luks mapping will be closed.

Note

This can be even done for LUKS2 encrypted root disk but will need some tweaking in initramfs.

7 July 2023

Dirk Eddelbuettel: Rcpp 1.0.11 on CRAN: Updates and Maintenance

rcpp logo The Rcpp Core Team is delighted to announce that the newest release 1.0.11 of the Rcpp package arrived on CRAN and in Debian earlier today. Windows and macOS builds should appear at CRAN in the next few days, as will builds in different Linux distribution and of course at r2u. The release was finalized three days ago, but given the widespread use and extended reverse dependencies at CRAN it usually takes a few days to be processed. This release continues with the six-months January-July cycle started with release 1.0.5 in July 2020. As a reminder, we do of course make interim snapshot dev or rc releases available via the Rcpp drat repo and strongly encourage their use and testing I run my systems with these versions which tend to work just as well, and are also fully tested against all reverse-dependencies. Rcpp has long established itself as the most popular way of enhancing R with C or C++ code. Right now, 2720 packages on CRAN depend on Rcpp for making analytical code go faster and further, along with 251 in BioConductor. On CRAN, 13.7% of all packages depend (directly) on Rcpp, and 59.6% of all compiled packages do. From the cloud mirror of CRAN (which is but a subset of all CRAN downloads), Rcpp has been downloaded 72.5 million times. The two published papers (also included in the package as preprint vignettes) have, respectively, 1678 (JSS, 2011) and 259 (TAS, 2018) citations, while the the book (Springer useR!, 2013) has another 588. This release is incremental as usual, generally preserving existing capabilities faithfully while smoothing our corners and / or extending slightly, sometimes in response to changing and tightened demands from CRAN or R standards. The full list below details all changes, their respective PRs and, if applicable, issue tickets. Big thanks from all of us to all contributors!

Changes in Rcpp version 1.0.11 (2023-07-03)
  • Changes in Rcpp API:
    • Rcpp:::CxxFlags() now quotes only non-standard include path on linux (Lukasz in #1243 closing #1242).
    • Two unit tests no longer accidentally bark on stdout (Dirk and I aki in #1245).
    • Compilation under C++20 using clang++ and its standard library is enabled (Dirk in #1248 closing #1244).
    • Use backticks in a generated .Call() statement in RcppExports.R (Dirk #1256 closing #1255).
    • Switch to system2() to capture standard error messages in error cases (I aki in #1259 and #1261 fixing #1257).
  • Changes in Rcpp Documentation:
    • The CITATION file format has been updated (Dirk in #1250 fixing #1249).
  • Changes in Rcpp Deployment:
    • A test for qnorm now uses the more accurate value from R 4.3.0 (Dirk in #1252 and #1260 fixing #1251).
    • Skip tests with path issues on Windows (I aki in #1258).
    • Container deployment in continuous integrations was improved. (I aki and Dirk in #1264, Dirk in #1269).
    • Several files receives minor edits to please R CMD check from r-devel (Dirk in #1267).

Thanks to my CRANberries, you can also look at a diff to the previous release. Questions, comments etc should go to the rcpp-devel mailing list off the R-Forge page. Bugs reports are welcome at the GitHub issue tracker as well (where one can also search among open or closed issues); questions are also welcome under rcpp tag at StackOverflow which also allows searching among the (currently) 2994 previous questions. If you like this or other open-source work I do, you can sponsor me at GitHub.

This post by Dirk Eddelbuettel originated on his Thinking inside the box blog. Please report excessive re-aggregation in third-party for-profit settings.

29 June 2023

C.J. Collier: Converting a windows install to a libvirt VM

Reduce the size of your c: partition to the smallest it can be and then turn off windows with the understanding that you will never boot this system on the iron ever again.
Boot into a netinst installer image (no GUI). hold alt and press left arrow a few times until you get to a prompt to press enter. Press enter. In this example /dev/sda is your windows disk which contains the c: partition
and /dev/disk/by-id/usb0 is the USB-3 attached SATA controller that you have your SSD attached to (please find an example attached). This SSD should be equal to or larger than the windows disk for best compatability. A photo of a USB-3 attached SATA controller To find the literal path names of your detected drives you can run fdisk -l. Pay attention to the names of the partitions and the sizes of the drives to help determine which is which. Once you have a shell in the netinst installer, you should maybe be able to run a command like the following. This will duplicate the disk located at if (in file) to the disk located at of (out file) while showing progress as the status.
dd if=/dev/sda of=/dev/disk/by-id/usb0 status=progress
If you confirm that dd is available on the netinst image and the previous command runs successfully, test that your windows partition is visible in the new disk s partition table. The start block of the windows partition on each should match, as should the partition size.
fdisk -l /dev/disk/by-id/usb0
fdisk -l /dev/sda
If the output from the first is the same as the output from the second, then you are probably safe to proceed. Once you confirm that you have made and tested a full copy of the blocks from your windows drive saved on your usb disk, nuke your windows partition table from orbit.
dd if=/dev/zero of=/dev/sda bs=1M count=42
You can press alt-f1 to return to the Debian installer now. Follow the instructions to install Debian. Don t forget to remove all attached USB drives. Once you install Debian, press ctrl-alt-f3 to get a root shell. Add your user to the sudoers group:
# adduser cjac sudoers
log out
# exit
log in as your user and confirm that you have sudo
$ sudo ls
Don t forget to read the spider man advice enter your password you ll need to install virt-manager. I think this should help:
$ sudo apt-get install virt-manager libvirt-daemon-driver-qemu qemu-system-x86
insert the USB drive. You can now create a qcow2 file for your virtual machine.
$ sudo qemu-img convert -O qcow2 \
/dev/disk/by-id/usb0 \
/var/lib/libvirt/images/windows.qcow2
I personally create a volume group called /dev/vg00 for the stuff I want to run raw and instead of converting to qcow2 like all of the other users do, I instead write it to a new logical volume.
sudo lvcreate /dev/vg00 -n windows -L 42G # or however large your drive was
sudo dd if=/dev/disk/by-id/usb0 of=/dev/vg00/windows status=progress
Now that you ve got the qcow2 file created, press alt-left until you return to your GDM session. The apt-get install command above installed virt-manager, so log in to your system if you haven t already and open up gnome-terminal by pressing the windows key or moving your mouse/gesture to the top left of your screen. Type in gnome-terminal and either press enter or click/tap on the icon. I like to run this full screen so that I feel like I m in a space ship. If you like to feel like you re in a spaceship, too, press F11. You can start virt-manager from this shell or you can press the windows key and type in virt-manager and press enter. You ll want the shell to run commands such as virsh console windows or virsh list When virt-manager starts, right click on QEMU/KVM and select New.
In the New VM window, select Import existing disk image
When prompted for the path to the image, use the one we created with sudo qemu-img convert above.
Select the version of Windows you want.
Select memory and CPUs to allocate to the VM.
Tick the Customize configuration before install box
If you re prompted to enable the default network, do so now.
The default hardware layout should probably suffice. Get it as close to the underlying hardware as it is convenient to do. But Windows is pretty lenient these days about virtualizing licensed windows instances so long as they re not running in more than one place at a time. Good luck! Leave comments if you have questions.

26 June 2023

Lukas M rdian: Netplan 0.106.1 stable release

We are happy to announce that Netplan 0.106.1 is available for download on Ubuntu Mantic Minotaur and Debian testing. This release includes some improvements in our documentation and CI infrastructure and a number of bug fixes.

What s new in Netplan 0.106.1?

Documentation

Infrastructure
  • canonical/setup-lxd GitHub action. The autopkgtest environment creation was standardized to use Canonical s setup-lxd action.
  • Snapd integrations tests with spread. A new test set for the Snapd integration with Netplan was introduced using the spread tool.
  • DBus. A number of DBus integration tests were added to the Debian package.

New features
  • Keyfile parser improvements. Our Network Manager keyfile parser (the capability of loading Network Manager configuration to Netplan YAML) was expanded to support all the types of tunnels supported by Netplan.

Misc
  • Ubuntu s Code of Conduct 2.0 was added to the code repository.
  • We added a new bash autocompletion script with all the Netplan s subcommands.
  • The new release package was synchronized with Debian.

Bug fixes
  • Keyfile parser. This release contains a couple of important fixes for the NetworkManager integration stability: 1) adding WPA enterprise connections is now working fine and new test cases were added to the package; 2) a WireGuard peer with allowed IPs that don t include the network prefix are now accepted.
  • Netplan parser. A number of memory leaks and stability issues were fixed.
  • DBus. An issue related to how directory paths are built in the Netplan DBus service was causing issues in the Snapd integration and was fixed.
For the complete list of changes please consult the debian/changelog file in https://launchpad.net/ubuntu/+source/netplan.io/+changelog

29 May 2023

Jonathan Carter: MiniDebConf Germany 2023

This year I attended Debian Reunion Hamburg (aka MiniDebConf Germany) for the second time. My goal for this MiniDebConf was just to talk to people and make the most of the time I have there. No other specific plans or goals. Despite this simple goal, it was a very productive and successful event for me. Tuesday 23rd:
Wednesday 24th:
Thursday 25th:
Friday 26th:
Saturday 27th: Sunday 28th: Monday 29th:
Das is nicht gut.
Tuesday 30th:

Thank you to Holger for organising this event yet again!

18 April 2023

Matthew Garrett: PSA: upgrade your LUKS key derivation function

Here's an article from a French anarchist describing how his (encrypted) laptop was seized after he was arrested, and material from the encrypted partition has since been entered as evidence against him. His encryption password was supposedly greater than 20 characters and included a mixture of cases, numbers, and punctuation, so in the absence of any sort of opsec failures this implies that even relatively complex passwords can now be brute forced, and we should be transitioning to even more secure passphrases.

Or does it? Let's go into what LUKS is doing in the first place. The actual data is typically encrypted with AES, an extremely popular and well-tested encryption algorithm. AES has no known major weaknesses and is not considered to be practically brute-forceable - at least, assuming you have a random key. Unfortunately it's not really practical to ask a user to type in 128 bits of binary every time they want to unlock their drive, so another approach has to be taken.

This is handled using something called a "key derivation function", or KDF. A KDF is a function that takes some input (in this case the user's password) and generates a key. As an extremely simple example, think of MD5 - it takes an input and generates a 128-bit output, so we could simply MD5 the user's password and use the output as an AES key. While this could technically be considered a KDF, it would be an extremely bad one! MD5s can be calculated extremely quickly, so someone attempting to brute-force a disk encryption key could simply generate the MD5 of every plausible password (probably on a lot of machines in parallel, likely using GPUs) and test each of them to see whether it decrypts the drive.

(things are actually slightly more complicated than this - your password is used to generate a key that is then used to encrypt and decrypt the actual encryption key. This is necessary in order to allow you to change your password without having to re-encrypt the entire drive - instead you simply re-encrypt the encryption key with the new password-derived key. This also allows you to have multiple passwords or unlock mechanisms per drive)

Good KDFs reduce this risk by being what's technically referred to as "expensive". Rather than performing one simple calculation to turn a password into a key, they perform a lot of calculations. The number of calculations performed is generally configurable, in order to let you trade off between the amount of security (the number of calculations you'll force an attacker to perform when attempting to generate a key from a potential password) and performance (the amount of time you're willing to wait for your laptop to generate the key after you type in your password so it can actually boot). But, obviously, this tradeoff changes over time - defaults that made sense 10 years ago are not necessarily good defaults now. If you set up your encrypted partition some time ago, the number of calculations required may no longer be considered up to scratch.

And, well, some of these assumptions are kind of bad in the first place! Just making things computationally expensive doesn't help a lot if your adversary has the ability to test a large number of passwords in parallel. GPUs are extremely good at performing the sort of calculations that KDFs generally use, so an attacker can "just" get a whole pile of GPUs and throw them at the problem. KDFs that are computationally expensive don't do a great deal to protect against this. However, there's another axis of expense that can be considered - memory. If the KDF algorithm requires a significant amount of RAM, the degree to which it can be performed in parallel on a GPU is massively reduced. A Geforce 4090 may have 16,384 execution units, but if each password attempt requires 1GB of RAM and the card only has 24GB on board, the attacker is restricted to running 24 attempts in parallel.

So, in these days of attackers with access to a pile of GPUs, a purely computationally expensive KDF is just not a good choice. And, unfortunately, the subject of this story was almost certainly using one of those. Ubuntu 18.04 used the LUKS1 header format, and the only KDF supported in this format is PBKDF2. This is not a memory expensive KDF, and so is vulnerable to GPU-based attacks. But even so, systems using the LUKS2 header format used to default to argon2i, again not a memory expensive KDFwhich is memory strong, but not designed to be resistant to GPU attack (thanks to the comments pointing out my misunderstanding here). New versions default to argon2id, which is. You want to be using argon2id.

What makes this worse is that distributions generally don't update this in any way. If you installed your system and it gave you pbkdf2 as your KDF, you're probably still using pbkdf2 even if you've upgraded to a system that would use argon2id on a fresh install. Thankfully, this can all be fixed-up in place. But note that if anything goes wrong here you could lose access to all your encrypted data, so before doing anything make sure it's all backed up (and figure out how to keep said backup secure so you don't just have your data seized that way).

First, make sure you're running as up-to-date a version of your distribution as possible. Having tools that support the LUKS2 format doesn't mean that your distribution has all of that integrated, and old distribution versions may allow you to update your LUKS setup without actually supporting booting from it. Also, if you're using an encrypted /boot, stop now - very recent versions of grub2 support LUKS2, but they don't support argon2id, and this will render your system unbootable.

Next, figure out which device under /dev corresponds to your encrypted partition. Run

lsblk

and look for entries that have a type of "crypt". The device above that in the tree is the actual encrypted device. Record that name, and run

sudo cryptsetup luksHeaderBackup /dev/whatever --header-backup-file /tmp/luksheader

and copy that to a USB stick or something. If something goes wrong here you'll be able to boot a live image and run

sudo cryptsetup luksHeaderRestore /dev/whatever --header-backup-file luksheader

to restore it.

(Edit to add: Once everything is working, delete this backup! It contains the old weak key, and someone with it can potentially use that to brute force your disk encryption key using the old KDF even if you've updated the on-disk KDF.)

Next, run

sudo cryptsetup luksDump /dev/whatever

and look for the Version: line. If it's version 1, you need to update the header to LUKS2. Run

sudo cryptsetup convert /dev/whatever --type luks2

and follow the prompts. Make sure your system still boots, and if not go back and restore the backup of your header. Assuming everything is ok at this point, run

sudo cryptsetup luksDump /dev/whatever

again and look for the PBKDF: line in each keyslot (pay attention only to the keyslots, ignore any references to pbkdf2 that come after the Digests: line). If the PBKDF is either "pbkdf2" or "argon2i" you should convert to argon2id. Run the following:

sudo cryptsetup luksConvertKey /dev/whatever --pbkdf argon2id

and follow the prompts. If you have multiple passwords associated with your drive you'll have multiple keyslots, and you'll need to repeat this for each password.

Distributions! You should really be handling this sort of thing on upgrade. People who installed their systems with your encryption defaults several years ago are now much less secure than people who perform a fresh install today. Please please please do something about this.

comment count unavailable comments

22 March 2023

Michael Prokop: Automatically unlocking a LUKS encrypted root filesystem during boot

Update on 2023-03-23: thanks to Daniel Roschka for mentioning the Mandos and TPM approaches, which might be better alternatives, depending on your options and needs. Peter Palfrader furthermore pointed me towards clevis-initramfs and tang. A customer of mine runs dedicated servers inside a foreign data-center, remote hands only. In such an environment you might need a disk replacement because you need bigger or faster disks, though also a disk might (start to) fail and you need a replacement. One has to be prepared for such a scenario, but fully wiping your used disk then might not always be an option, especially once disks (start to) fail. On the other hand you don t want to end up with (partial) data on your disk handed over to someone unexpected. By encrypting the data on your disks upfront you can prevent against this scenario. But if you have a fleet of servers you might not want to manually jump on servers during boot and unlock crypto volumes manually. It s especially annoying if it s about the root filesystem where a solution like dropbear-initramfs needs to be used for remote access during initramfs boot stage. So my task for the customer was to adjust encrypted LUKS devices such that no one needs to manually unlock the encrypted device during server boot (with some specific assumptions about possible attack vectors one has to live with, see the disclaimer at the end). The documentation about this use-case was rather inconsistent, especially because special rules apply for the root filesystem (no key file usage), we see different behavior between what s supported by systemd (hello key file again), initramfs-tools and dracut, not to mention the changes between different distributions. Since tests with this tend to be rather annoying (better make sure to have a Grml live system available :)), I m hereby documenting what worked for us (Debian/bullseye with initramfs-tools and cryptsetup-initramfs). The system was installed with LVM on-top of an encrypted Software-RAID device, only the /boot partition is unencrypted. But even if you don t use Software-RAID nor LVM the same instructions apply. The system looks like this:
% mount -t ext4 -l
/dev/mapper/foobar-root_1 on / type ext4 (rw,relatime,errors=remount-ro)
% sudo pvs
  PV                    VG     Fmt  Attr PSize   PFree
  /dev/mapper/md1_crypt foobar lvm2 a--  445.95g 430.12g
% sudo vgs
  VG     #PV #LV #SN Attr   VSize   VFree
  foobar   1   2   0 wz--n- 445.95g 430.12g
% sudo lvs
  LV     VG     Attr       LSize   Pool Origin Data%  Meta%  Move Log Cpy%Sync Convert
  root_1 foobar -wi-ao---- <14.90g
% lsblk
NAME                  MAJ:MIN RM   SIZE RO TYPE  MOUNTPOINT
[...]
sdd                     8:48   0 447.1G  0 disk
 sdd1                  8:49   0   571M  0 part  /boot/efi
 sdd2                  8:50   0   488M  0 part
   md0                 9:0    0   487M  0 raid1 /boot
 sdd3                  8:51   0 446.1G  0 part
   md1                 9:1    0   446G  0 raid1
     md1_crypt       253:0    0   446G  0 crypt
       foobar-root_1 253:1    0  14.9G  0 lvm   /
[...]
sdf                     8:80   0 447.1G  0 disk
 sdf1                  8:81   0   571M  0 part
 sdf2                  8:82   0   488M  0 part
   md0                 9:0    0   487M  0 raid1 /boot
 sdf3                  8:83   0 446.1G  0 part
   md1                 9:1    0   446G  0 raid1
     md1_crypt       253:0    0   446G  0 crypt
       foobar-root_1 253:1    0  14.9G  0 lvm   /
The actual crypsetup configuration is:
% cat /etc/crypttab
md1_crypt UUID=77246138-b666-4151-b01c-5a12db54b28b none luks,discard
Now, to automatically open the crypto device during boot we can instead use:
% cat /etc/crypttab 
md1_crypt UUID=77246138-b666-4151-b01c-5a12db54b28b none luks,discard,keyscript=/etc/initramfs-tools/unlock.sh
# touch /etc/initramfs-tools/unlock.sh
# chmod 0700 /etc/initramfs-tools/unlock.sh
# $EDITOR etc/initramfs-tools/unlock.sh
# cat /etc/initramfs-tools/unlock.sh
#!/bin/sh
echo -n "provide_the_actual_password_here"
# update-initramfs -k all -u
[...]
The server will then boot without prompting for a crypto password. Note that initramfs-tools by default uses an insecure umask of 0022, resulting in the initrd being accessible to everyone. But if you have the dropbear-initramfs package installed, its /usr/share/initramfs-tools/conf-hooks.d/dropbear sets UMASK=0077 , so the resulting /boot/initrd* file should automatically have proper permissions (0600). The cryptsetup hook warns about a permissive umask configuration during update-initramfs runs, but if you want to be sure, explicitly set it via e.g.:
# cat > /etc/initramfs-tools/conf.d/umask << EOF
# restrictive umask to avoid non-root access to initrd:
UMASK=0077
EOF
# update-initramfs -k all -u
Disclaimer: Of course you need to trust users with access to /etc/initramfs-tools/unlock.sh as well as the initramfs/initrd on your system. Furthermore you should wipe the boot partition (to destroy the keyfile information) before handing over such a disk. But that is a risk my customer can live with, YMMV.

15 February 2023

Lukas M rdian: Netplan v0.106 is now available

I m happy to announce that Netplan version 0.106 is now available on GitHub and is soon to be deployed into an Ubuntu/Debian/Fedora installation near you! Six months and 65 commits after the previous version, this release is brought to you by 4 free software contributors from around the globe. Highlights Highlights of this release include the new netplan status command, which queries your system for IP addresses, routes, DNS information, etc in addition to the Netplan backend renderer (NetworkManager/networkd) in use and the relevant Netplan YAML configuration ID. It displays all this in a nicely formatted way (or alternatively in machine readable YAML/JSON format).
Furthermore, we implemented a clean libnetplan API which can be used by external tools to parse Netplan configuration, migrated away from non-inclusive language (PR#303) and improved the overall Netplan documentation. Another change that should be noted, is that the match.macaddress stanza now only matches on PermanentMACAddress= on the systemd-networkd backend, as has been the case on the NetworkManager backend ever since (see PR#278 for background information on this slight change in behavior). Changelog Bug fixes:

13 February 2023

Jonathan Dowland: A visit to Prusa Labs

.
.

In September I was in Czechia for a Red Hat event. I ended up travelling via Prague, and had an unexpected extra day due to an airline strike causing my flight home to be cancelled. I took the opportunity to visit Prusa's offices/factory/Lab, and it was amazing! The Prusa team were all busy getting ready for the Prague Maker Faire that was happening the day afterwards.1 On arriving at the street which houses Prusa's Lab and Office buildings, the first thing that hit me was the smell. I find the melted-plastic smell of FDM printing (with PLA, at least) quite pleasant, and this was a super-condensed version of that, pumping out of their ground-floor windows. I started at the reception area on the ground floor. Outside reception there's a lovely sculpture representing the history of the development of the MK3S+.
The Reception The Reception
SLS Farm in the former Hack lab SLS Farm in the former Hack lab
History of the MK3S+ History of the MK3S+

At the reception you have a small waiting area with shelves of demonstration prints and some spools of Prusament. From here, our kind guide first took us to a region on the ground floor that used to be (prior to COVID times) the public maker/hack lab. The lab contained two modest farms of printers: one of their flagship FDM printer, the MK3S+, and another of their SLS resin printers. A close up of some example resin models is pictured above. The bicycle was tiny: about the size of a thumbnail.
A Historic display A Historic display
Bespoke QE equipment Bespoke QE equipment
The MK3S+ Farm The MK3S+ Farm

. Moi in the Farm
2KG orange PLA spools for the Farm 2KG orange PLA spools for the Farm
The rest of the ground floor area was full of heavy machinery and prototyping equipment. Onwards we went to the upstairs floors. Upstairs, past a nice graphic of Prusa's historic products, we visited the assembly and QE testing areas. They have a very organised system of parts buckets and some thorough QE processes, including some bespoke equipment that produces the "receipt" of tests and calibration that they provide for you in the box when you buy a 3D printer. After that, we visited the production Farm: a large room full of MK3S+ printers churning out parts for other printers. The noise was remarkable. The printers were running custom firmware to continually print the part they had been set for. Some of the printers were designated for printing with ASA: they were colour-coded (yellow controller surround) and within boxed regions to prevent the fumes causing problems. (picture at the top) Outside the room sat palettes of orange PETG plastic for the printer farm on 2KG spools (not a size they sell to the public just yet) The final part of the trip was outside, to the real farm: Prusa have a smallholding with Alpacas at the rear of their estate. Whilst we visited it, Josef Prusa himself turned up (in a snazzy looking custom colour Tesla) to feed the animals, say hello and pose for a picture. Overall, it was a fantastic visit. I'm very grateful to Air France for cancelling my flight home, and to Prusa Labs (in particular Luk ) for allowing me to come and say hello!

  1. I managed to squeeze that in too on my way to my rescheduled flight, although it was a rush visit and I don't have much to say or show from it. Suffice to say that it was lovely and bittersweet since the UK Maker Faire used to be hosted in my fair home city of Newcastle before they stopped.

12 December 2022

Vasudev Kamath: Installing Debian from GRML Live CD

I had bought a Thinkpad E470 laptop back in 2018 which was lying unused for quite some time. Recently when I wanted to use it, I found that the keyboard is not working, especially some keys and after some time the laptop will hang in Lenovo boot screen. I came back to Bangalore almost after 2 years from my hometown (WFH due to Covid) and thought it was the right time to get my laptop back to normal working state. After getting the keyboard replaced I noticed that 1TB HDD is no longer fast enough for my taste!. I've to admit I never thought I would start disliking HDD so quickly thanks to modern SSD based work laptops. So as a second upgrade I got the HDD removed from my laptop and got a 240G SSD. Yeah I know its reduction from my original size but I intend to continue using my old HDD via USB SATA enclosure as an external HDD which can house the extra data which I need to save. So now that I've a SSD I need to install Debian Unstable again on it and this is where I tried something new. My colleague (name redacted on request) suggested to me use GRML live CD and install Debian via debootstrap. And after giving a thought I decided to try this out. Some reason for going ahead with this are listed below
  1. Debian Installer does not support a proper BTRFS based root file system. It just allows btrfs as root but no subvolume support. Also I'm not sure about the luks support with btrfs as root.
  2. I also wanted to give a try to systemd-boot as my laptop is UEFI capable and I've slowly started disliking Grub.
  3. I really hate installing task-kde-desktop (Yeah you read it right, I've switched to be a KDE user for quite some time) which will pull tons of unwanted stuff and bloat. Well it's not just task-kde-desktop but any other task-desktop package does similar and I don't want to have too much of unused stuff and services running.
Disk Preparation As a first step I went to GRML website and downloaded current pre-release. Frankly, I'm using GRML for first time and I was not sure what to expect. When I booted it up I was bit taken a back to see its console based and I did not have a wired lan just a plain wireless dongle (Jiofi device) and was wondering what it will take to connect. But surprisingly curses based UI was pretty much straight forward to allow me to connect to Wifi AP. Another thing was the rescue CD had non-free firmware as the laptop was using ath10k device and needed non-free blobs to operate. Once I got shell prompt in rescue CD first thing I did was to reconfigure console-setup to increase font size which was very very small on default boot. Once that is done I did the following to create a 1G (FAT32) partition for EFI.
parted -a optimal -s /dev/sda mklabel gpt
parted -a optimal -s /dev/sda mkpart primary vfat 0% 1G
parted -a optimal -s /dev/sda set 1 esp on
mkfs.vfat -n boot_disk -F 32 /dev/sda1
So here is what I did: created a 1G vfat type partition and set the esp flag on it. This will be mounted to /boot/efi for systemd-boot. Next I created a single partition on the rest of the available free disk which will be used as the root file system. Next I encrypted the root parition using LUKS and then created the BTRFS file system on top of it.
cryptsetup luksFormat /dev/sda2
cryptsetup luksOpen /dev/sda2 ENC
mkfs.btrfs -L root_disk /dev/mapper/ENC
Next is to create subvolumes in BTRFS. I followed suggestion by colleague and created a top-level @ as subvolume below which created @/home @/var/log @/opt . Also enabled compression with zstd and level of 1 to avoid battery drain. Finally marked the @ as default subvolume to avoid adding it to fstab entry.
mount -o compress=zstd:1 /dev/mapper/ENC /mnt
btrfs subvol create /mnt/@
cd /mnt/@
btrfs subvol create ./home
btrfs subvol create ./opt
mkdir -p var
btrfs subvol create ./var/log
btrfs suvol set-default /mnt/@
Bootstrapping Debian Now that root disk is prepared next step was to bootstrap the root file system. I used debootstrap for this job. One thing I missed here from installer was ability to preseed. I tried looking around to figure out if we can preseed debootstrap but did not find much. If you know the procedure do point it to me.
cd /mnt/
debootstrap --include=dbus,locales,tzdata unstable @/ http://deb.debian.org/debian
Well this just gets a bare minimal installation of Debian I need to install rest of the things post this step manually by chroot into target folder @/. I like the grml-chroot command for chroot purpose, it does most of the job of mounting all required directory like /dev/ /proc /sys etc. But before entering chroot I need to mount the ESP partition we created to /boot/efi so that I can finalize the installation of kernel and systemd-boot.
umount /mnt
mount -o compress=zstd:1 /dev/mapper/ENC /mnt
mkdir -p /mnt/boot/efi
mount /dev/sda1 /mnt/boot/efi
grml-chroot /mnt /bin/bash
I remounted the root subvolume @ directly to /mnt now, remember I made @ as default subvolume before. I also mounted ESP partition with FAT32 file system to /boot/efi. Finally I used grml-chroot to get into chroot of newly bootstrapped file system. Now I will install the kernel and minimal KDE desktop installation and configure locales and time zone data for the new system. I wanted to use dracut instead of default initramfs-tools for initrd. I also need to install cryptsetup and btrfs-progs so I can decrypt and really boot into my new system.
apt-get update
apt-get install linux-image-amd64 dracut openssh-client \
                        kde-plasma-desktop plasma-workspace-wayland \
                        plasma-nm cryptsetup btrfs-progs sudo
Next is setting up crypttab and fstab entries for new system. Following entry is added to fstab
LABEL="root_disk" / btrfs defaults,compress=zstd:1 0 0
And the crypttab entry
ENCRYPTED_ROOT UUID=xxxx none discard,x-initrd.attach
I've not written actual UUID above this is just for the purpose of showing the content of /etc/crypttab. Once these entries are added we need to recreate initrd. I just reconfigured the installed kernel package for retriggerring the recreation of initrd using dracut. .. Reconfiguration was locales is done by editing /etc/locales.gen to uncomment en_US.UTF-8 and writing /etc/timezone with Asia/Kolkata. I used DEBIAN_FRONTEND=noninteractive to avoid another prompt asking for locale and timezone information.
export DEBIAN_FRONTEND=noninteractive
dpkg-reconfigure locales
dpkg-reconfigure tzdata
Added my user using adduser command and also set the root password as well. Added my user to sudo group so I can use sudo to elevate privileges.
Setting up systemd-boot So now basic usable system is ready last part is enabling the systemd-boot configuration as I'm not gonna use grub. I did following to install systemd-boot. Frankly I'm not expert of this it was colleague's suggestion. Before installing the systemd-boot I had to setup kernel command line. This can be done by writing command line to /etc/kernel/cmdline with following contents.
systemd.gpt_auto=no quiet root=LABEL=root_disk
I'm disabling systemd-gpt-generator to avoid race condition between crypttab entry and auto generated entry by systemd. I faced this mainly because of my stupidity of not adding entry root=LABEL=root_disk
apt-get install -y systemd-boot
bootctl install --make-entry-directory=yes --entry-token=machine-id
dpkg-reconfigure linux-image-6.0.0-5-amd64
Finally exit from the chroot and reboot into the freshly installed system. systemd-boot already ships a hook file zz-systemd-boot under /etc/kernel so its pretty much usable without any manual intervention. Previously after kernel installation we had to manually update kernel image in efi partitions using bootctl
Conclussion Though installing from live image is not new and debian-installer also does the same only difference is more control over installation and doing things which is installer is not letting you do (or should I say is not part of default installation?). If properly automated using scripts we can leverage this to do custom installation in large scale environments. I know there is FAI but I've not explored it and felt there is too much to setup for a simple installations with specific requirements. So finally I've a system with Debian which differs from default Debian installation :-). I should thank my colleague for rekindling nerd inside me who had stopped experimenting quite a long time back.

16 November 2022

Antoine Beaupr : A ZFS migration

In my tubman setup, I started using ZFS on an old server I had lying around. The machine is really old though (2011!) and it "feels" pretty slow. I want to see how much of that is ZFS and how much is the machine. Synthetic benchmarks show that ZFS may be slower than mdadm in RAID-10 or RAID-6 configuration, so I want to confirm that on a live workload: my workstation. Plus, I want easy, regular, high performance backups (with send/receive snapshots) and there's no way I'm going to use BTRFS because I find it too confusing and unreliable. So off we go.

Installation Since this is a conversion (and not a new install), our procedure is slightly different than the official documentation but otherwise it's pretty much in the same spirit: we're going to use ZFS for everything, including the root filesystem. So, install the required packages, on the current system:
apt install --yes gdisk zfs-dkms zfs zfs-initramfs zfsutils-linux
We also tell DKMS that we need to rebuild the initrd when upgrading:
echo REMAKE_INITRD=yes > /etc/dkms/zfs.conf

Partitioning This is going to partition /dev/sdc with:
  • 1MB MBR / BIOS legacy boot
  • 512MB EFI boot
  • 1GB bpool, unencrypted pool for /boot
  • rest of the disk for zpool, the rest of the data
     sgdisk --zap-all /dev/sdc
     sgdisk -a1 -n1:24K:+1000K -t1:EF02 /dev/sdc
     sgdisk     -n2:1M:+512M   -t2:EF00 /dev/sdc
     sgdisk     -n3:0:+1G      -t3:BF01 /dev/sdc
     sgdisk     -n4:0:0        -t4:BF00 /dev/sdc
    
That will look something like this:
    root@curie:/home/anarcat# sgdisk -p /dev/sdc
    Disk /dev/sdc: 1953525168 sectors, 931.5 GiB
    Model: ESD-S1C         
    Sector size (logical/physical): 512/512 bytes
    Disk identifier (GUID): [REDACTED]
    Partition table holds up to 128 entries
    Main partition table begins at sector 2 and ends at sector 33
    First usable sector is 34, last usable sector is 1953525134
    Partitions will be aligned on 16-sector boundaries
    Total free space is 14 sectors (7.0 KiB)
    Number  Start (sector)    End (sector)  Size       Code  Name
       1              48            2047   1000.0 KiB  EF02  
       2            2048         1050623   512.0 MiB   EF00  
       3         1050624         3147775   1024.0 MiB  BF01  
       4         3147776      1953525134   930.0 GiB   BF00
Unfortunately, we can't be sure of the sector size here, because the USB controller is probably lying to us about it. Normally, this smartctl command should tell us the sector size as well:
root@curie:~# smartctl -i /dev/sdb -qnoserial
smartctl 7.2 2020-12-30 r5155 [x86_64-linux-5.10.0-14-amd64] (local build)
Copyright (C) 2002-20, Bruce Allen, Christian Franke, www.smartmontools.org
=== START OF INFORMATION SECTION ===
Model Family:     Western Digital Black Mobile
Device Model:     WDC WD10JPLX-00MBPT0
Firmware Version: 01.01H01
User Capacity:    1 000 204 886 016 bytes [1,00 TB]
Sector Sizes:     512 bytes logical, 4096 bytes physical
Rotation Rate:    7200 rpm
Form Factor:      2.5 inches
Device is:        In smartctl database [for details use: -P show]
ATA Version is:   ATA8-ACS T13/1699-D revision 6
SATA Version is:  SATA 3.0, 6.0 Gb/s (current: 6.0 Gb/s)
Local Time is:    Tue May 17 13:33:04 2022 EDT
SMART support is: Available - device has SMART capability.
SMART support is: Enabled
Above is the example of the builtin HDD drive. But the SSD device enclosed in that USB controller doesn't support SMART commands, so we can't trust that it really has 512 bytes sectors. This matters because we need to tweak the ashift value correctly. We're going to go ahead the SSD drive has the common 4KB settings, which means ashift=12. Note here that we are not creating a separate partition for swap. Swap on ZFS volumes (AKA "swap on ZVOL") can trigger lockups and that issue is still not fixed upstream. Ubuntu recommends using a separate partition for swap instead. But since this is "just" a workstation, we're betting that we will not suffer from this problem, after hearing a report from another Debian developer running this setup on their workstation successfully. We do not recommend this setup though. In fact, if I were to redo this partition scheme, I would probably use LUKS encryption and setup a dedicated swap partition, as I had problems with ZFS encryption as well.

Creating pools ZFS pools are somewhat like "volume groups" if you are familiar with LVM, except they obviously also do things like RAID-10. (Even though LVM can technically also do RAID, people typically use mdadm instead.) In any case, the guide suggests creating two different pools here: one, in cleartext, for boot, and a separate, encrypted one, for the rest. Technically, the boot partition is required because the Grub bootloader only supports readonly ZFS pools, from what I understand. But I'm a little out of my depth here and just following the guide.

Boot pool creation This creates the boot pool in readonly mode with features that grub supports:
    zpool create \
        -o cachefile=/etc/zfs/zpool.cache \
        -o ashift=12 -d \
        -o feature@async_destroy=enabled \
        -o feature@bookmarks=enabled \
        -o feature@embedded_data=enabled \
        -o feature@empty_bpobj=enabled \
        -o feature@enabled_txg=enabled \
        -o feature@extensible_dataset=enabled \
        -o feature@filesystem_limits=enabled \
        -o feature@hole_birth=enabled \
        -o feature@large_blocks=enabled \
        -o feature@lz4_compress=enabled \
        -o feature@spacemap_histogram=enabled \
        -o feature@zpool_checkpoint=enabled \
        -O acltype=posixacl -O canmount=off \
        -O compression=lz4 \
        -O devices=off -O normalization=formD -O relatime=on -O xattr=sa \
        -O mountpoint=/boot -R /mnt \
        bpool /dev/sdc3
I haven't investigated all those settings and just trust the upstream guide on the above.

Main pool creation This is a more typical pool creation.
    zpool create \
        -o ashift=12 \
        -O encryption=on -O keylocation=prompt -O keyformat=passphrase \
        -O acltype=posixacl -O xattr=sa -O dnodesize=auto \
        -O compression=zstd \
        -O relatime=on \
        -O canmount=off \
        -O mountpoint=/ -R /mnt \
        rpool /dev/sdc4
Breaking this down:
  • -o ashift=12: mentioned above, 4k sector size
  • -O encryption=on -O keylocation=prompt -O keyformat=passphrase: encryption, prompt for a password, default algorithm is aes-256-gcm, explicit in the guide, made implicit here
  • -O acltype=posixacl -O xattr=sa: enable ACLs, with better performance (not enabled by default)
  • -O dnodesize=auto: related to extended attributes, less compatibility with other implementations
  • -O compression=zstd: enable zstd compression, can be disabled/enabled by dataset to with zfs set compression=off rpool/example
  • -O relatime=on: classic atime optimisation, another that could be used on a busy server is atime=off
  • -O canmount=off: do not make the pool mount automatically with mount -a?
  • -O mountpoint=/ -R /mnt: mount pool on / in the future, but /mnt for now
Those settings are all available in zfsprops(8). Other flags are defined in zpool-create(8). The reasoning behind them is also explained in the upstream guide and some also in [the Debian wiki][]. Those flags were actually not used:
  • -O normalization=formD: normalize file names on comparisons (not storage), implies utf8only=on, which is a bad idea (and effectively meant my first sync failed to copy some files, including this folder from a supysonic checkout). and this cannot be changed after the filesystem is created. bad, bad, bad.
[the Debian wiki]: https://wiki.debian.org/ZFS#Advanced_Topics

Side note about single-disk pools Also note that we're living dangerously here: single-disk ZFS pools are rumoured to be more dangerous than not running ZFS at all. The choice quote from this article is:
[...] any error can be detected, but cannot be corrected. This sounds like an acceptable compromise, but its actually not. The reason its not is that ZFS' metadata cannot be allowed to be corrupted. If it is it is likely the zpool will be impossible to mount (and will probably crash the system once the corruption is found). So a couple of bad sectors in the right place will mean that all data on the zpool will be lost. Not some, all. Also there's no ZFS recovery tools, so you cannot recover any data on the drives.
Compared with (say) ext4, where a single disk error can recovered, this is pretty bad. But we are ready to live with this with the idea that we'll have hourly offline snapshots that we can easily recover from. It's trade-off. Also, we're running this on a NVMe/M.2 drive which typically just blinks out of existence completely, and doesn't "bit rot" the way a HDD would. Also, the FreeBSD handbook quick start doesn't have any warnings about their first example, which is with a single disk. So I am reassured at least.

Creating mount points Next we create the actual filesystems, known as "datasets" which are the things that get mounted on mountpoint and hold the actual files.
  • this creates two containers, for ROOT and BOOT
     zfs create -o canmount=off -o mountpoint=none rpool/ROOT &&
     zfs create -o canmount=off -o mountpoint=none bpool/BOOT
    
    Note that it's unclear to me why those datasets are necessary, but they seem common practice, also used in this FreeBSD example. The OpenZFS guide mentions the Solaris upgrades and Ubuntu's zsys that use that container for upgrades and rollbacks. This blog post seems to explain a bit the layout behind the installer.
  • this creates the actual boot and root filesystems:
     zfs create -o canmount=noauto -o mountpoint=/ rpool/ROOT/debian &&
     zfs mount rpool/ROOT/debian &&
     zfs create -o mountpoint=/boot bpool/BOOT/debian
    
    I guess the debian name here is because we could technically have multiple operating systems with the same underlying datasets.
  • then the main datasets:
     zfs create                                 rpool/home &&
     zfs create -o mountpoint=/root             rpool/home/root &&
     chmod 700 /mnt/root &&
     zfs create                                 rpool/var
    
  • exclude temporary files from snapshots:
     zfs create -o com.sun:auto-snapshot=false  rpool/var/cache &&
     zfs create -o com.sun:auto-snapshot=false  rpool/var/tmp &&
     chmod 1777 /mnt/var/tmp
    
  • and skip automatic snapshots in Docker:
     zfs create -o canmount=off                 rpool/var/lib &&
     zfs create -o com.sun:auto-snapshot=false  rpool/var/lib/docker
    
    Notice here a peculiarity: we must create rpool/var/lib to create rpool/var/lib/docker otherwise we get this error:
     cannot create 'rpool/var/lib/docker': parent does not exist
    
    ... and no, just creating /mnt/var/lib doesn't fix that problem. In fact, it makes things even more confusing because an existing directory shadows a mountpoint, which is the opposite of how things normally work. Also note that you will probably need to change storage driver in Docker, see the zfs-driver documentation for details but, basically, I did:
    echo '  "storage-driver": "zfs"  ' > /etc/docker/daemon.json
    
    Note that podman has the same problem (and similar solution):
    printf '[storage]\ndriver = "zfs"\n' > /etc/containers/storage.conf
    
  • make a tmpfs for /run:
     mkdir /mnt/run &&
     mount -t tmpfs tmpfs /mnt/run &&
     mkdir /mnt/run/lock
    
We don't create a /srv, as that's the HDD stuff. Also mount the EFI partition:
mkfs.fat -F 32 /dev/sdc2 &&
mount /dev/sdc2 /mnt/boot/efi/
At this point, everything should be mounted in /mnt. It should look like this:
root@curie:~# LANG=C df -h -t zfs -t vfat
Filesystem            Size  Used Avail Use% Mounted on
rpool/ROOT/debian     899G  384K  899G   1% /mnt
bpool/BOOT/debian     832M  123M  709M  15% /mnt/boot
rpool/home            899G  256K  899G   1% /mnt/home
rpool/home/root       899G  256K  899G   1% /mnt/root
rpool/var             899G  384K  899G   1% /mnt/var
rpool/var/cache       899G  256K  899G   1% /mnt/var/cache
rpool/var/tmp         899G  256K  899G   1% /mnt/var/tmp
rpool/var/lib/docker  899G  256K  899G   1% /mnt/var/lib/docker
/dev/sdc2             511M  4.0K  511M   1% /mnt/boot/efi
Now that we have everything setup and mounted, let's copy all files over.

Copying files This is a list of all the mounted filesystems
for fs in /boot/ /boot/efi/ / /home/; do
    echo "syncing $fs to /mnt$fs..." && 
    rsync -aSHAXx --info=progress2 --delete $fs /mnt$fs
done
You can check that the list is correct with:
mount -l -t ext4,btrfs,vfat   awk ' print $3 '
Note that we skip /srv as it's on a different disk. On the first run, we had:
root@curie:~# for fs in /boot/ /boot/efi/ / /home/; do
        echo "syncing $fs to /mnt$fs..." && 
        rsync -aSHAXx --info=progress2 $fs /mnt$fs
    done
syncing /boot/ to /mnt/boot/...
              0   0%    0.00kB/s    0:00:00 (xfr#0, to-chk=0/299)  
syncing /boot/efi/ to /mnt/boot/efi/...
     16,831,437 100%  184.14MB/s    0:00:00 (xfr#101, to-chk=0/110)
syncing / to /mnt/...
 28,019,293,280  94%   47.63MB/s    0:09:21 (xfr#703710, ir-chk=6748/839220)rsync: [generator] delete_file: rmdir(var/lib/docker) failed: Device or resource busy (16)
could not make way for new symlink: var/lib/docker
 34,081,267,990  98%   50.71MB/s    0:10:40 (xfr#736577, to-chk=0/867732)    
rsync error: some files/attrs were not transferred (see previous errors) (code 23) at main.c(1333) [sender=3.2.3]
syncing /home/ to /mnt/home/...
rsync: [sender] readlink_stat("/home/anarcat/.fuse") failed: Permission denied (13)
 24,456,268,098  98%   68.03MB/s    0:05:42 (xfr#159867, ir-chk=6875/172377) 
file has vanished: "/home/anarcat/.cache/mozilla/firefox/s2hwvqbu.quantum/cache2/entries/B3AB0CDA9C4454B3C1197E5A22669DF8EE849D90"
199,762,528,125  93%   74.82MB/s    0:42:26 (xfr#1437846, ir-chk=1018/1983979)rsync: [generator] recv_generator: mkdir "/mnt/home/anarcat/dist/supysonic/tests/assets/\#346" failed: Invalid or incomplete multibyte or wide character (84)
*** Skipping any contents from this failed directory ***
315,384,723,978  96%   76.82MB/s    1:05:15 (xfr#2256473, to-chk=0/2993950)    
rsync error: some files/attrs were not transferred (see previous errors) (code 23) at main.c(1333) [sender=3.2.3]
Note the failure to transfer that supysonic file? It turns out they had a weird filename in their source tree, since then removed, but still it showed how the utf8only feature might not be such a bad idea. At this point, the procedure was restarted all the way back to "Creating pools", after unmounting all ZFS filesystems (umount /mnt/run /mnt/boot/efi && umount -t zfs -a) and destroying the pool, which, surprisingly, doesn't require any confirmation (zpool destroy rpool). The second run was cleaner:
root@curie:~# for fs in /boot/ /boot/efi/ / /home/; do
        echo "syncing $fs to /mnt$fs..." && 
        rsync -aSHAXx --info=progress2 --delete $fs /mnt$fs
    done
syncing /boot/ to /mnt/boot/...
              0   0%    0.00kB/s    0:00:00 (xfr#0, to-chk=0/299)  
syncing /boot/efi/ to /mnt/boot/efi/...
              0   0%    0.00kB/s    0:00:00 (xfr#0, to-chk=0/110)  
syncing / to /mnt/...
 28,019,033,070  97%   42.03MB/s    0:10:35 (xfr#703671, ir-chk=1093/833515)rsync: [generator] delete_file: rmdir(var/lib/docker) failed: Device or resource busy (16)
could not make way for new symlink: var/lib/docker
 34,081,807,102  98%   44.84MB/s    0:12:04 (xfr#736580, to-chk=0/867723)    
rsync error: some files/attrs were not transferred (see previous errors) (code 23) at main.c(1333) [sender=3.2.3]
syncing /home/ to /mnt/home/...
rsync: [sender] readlink_stat("/home/anarcat/.fuse") failed: Permission denied (13)
IO error encountered -- skipping file deletion
 24,043,086,450  96%   62.03MB/s    0:06:09 (xfr#151819, ir-chk=15117/172571)
file has vanished: "/home/anarcat/.cache/mozilla/firefox/s2hwvqbu.quantum/cache2/entries/4C1FDBFEA976FF924D062FB990B24B897A77B84B"
315,423,626,507  96%   67.09MB/s    1:14:43 (xfr#2256845, to-chk=0/2994364)    
rsync error: some files/attrs were not transferred (see previous errors) (code 23) at main.c(1333) [sender=3.2.3]
Also note the transfer speed: we seem capped at 76MB/s, or 608Mbit/s. This is not as fast as I was expecting: the USB connection seems to be at around 5Gbps:
anarcat@curie:~$ lsusb -tv   head -4
/:  Bus 02.Port 1: Dev 1, Class=root_hub, Driver=xhci_hcd/6p, 5000M
    ID 1d6b:0003 Linux Foundation 3.0 root hub
     __ Port 1: Dev 4, If 0, Class=Mass Storage, Driver=uas, 5000M
        ID 0b05:1932 ASUSTek Computer, Inc.
So it shouldn't cap at that speed. It's possible the USB adapter is failing to give me the full speed though. It's not the M.2 SSD drive either, as that has a ~500MB/s bandwidth, acccording to its spec. At this point, we're about ready to do the final configuration. We drop to single user mode and do the rest of the procedure. That used to be shutdown now, but it seems like the systemd switch broke that, so now you can reboot into grub and pick the "recovery" option. Alternatively, you might try systemctl rescue, as I found out. I also wanted to copy the drive over to another new NVMe drive, but that failed: it looks like the USB controller I have doesn't work with older, non-NVME drives.

Boot configuration Now we need to enter the new system to rebuild the boot loader and initrd and so on. First, we bind mounts and chroot into the ZFS disk:
mount --rbind /dev  /mnt/dev &&
mount --rbind /proc /mnt/proc &&
mount --rbind /sys  /mnt/sys &&
chroot /mnt /bin/bash
Next we add an extra service that imports the bpool on boot, to make sure it survives a zpool.cache destruction:
cat > /etc/systemd/system/zfs-import-bpool.service <<EOF
[Unit]
DefaultDependencies=no
Before=zfs-import-scan.service
Before=zfs-import-cache.service
[Service]
Type=oneshot
RemainAfterExit=yes
ExecStart=/sbin/zpool import -N -o cachefile=none bpool
# Work-around to preserve zpool cache:
ExecStartPre=-/bin/mv /etc/zfs/zpool.cache /etc/zfs/preboot_zpool.cache
ExecStartPost=-/bin/mv /etc/zfs/preboot_zpool.cache /etc/zfs/zpool.cache
[Install]
WantedBy=zfs-import.target
EOF
Enable the service:
systemctl enable zfs-import-bpool.service
I had to trim down /etc/fstab and /etc/crypttab to only contain references to the legacy filesystems (/srv is still BTRFS!). If we don't already have a tmpfs defined in /etc/fstab:
ln -s /usr/share/systemd/tmp.mount /etc/systemd/system/ &&
systemctl enable tmp.mount
Rebuild boot loader with support for ZFS, but also to workaround GRUB's missing zpool-features support:
grub-probe /boot   grep -q zfs &&
update-initramfs -c -k all &&
sed -i 's,GRUB_CMDLINE_LINUX.*,GRUB_CMDLINE_LINUX="root=ZFS=rpool/ROOT/debian",' /etc/default/grub &&
update-grub
For good measure, make sure the right disk is configured here, for example you might want to tag both drives in a RAID array:
dpkg-reconfigure grub-pc
Install grub to EFI while you're there:
grub-install --target=x86_64-efi --efi-directory=/boot/efi --bootloader-id=debian --recheck --no-floppy
Filesystem mount ordering. The rationale here in the OpenZFS guide is a little strange, but I don't dare ignore that.
mkdir /etc/zfs/zfs-list.cache
touch /etc/zfs/zfs-list.cache/bpool
touch /etc/zfs/zfs-list.cache/rpool
zed -F &
Verify that zed updated the cache by making sure these are not empty:
cat /etc/zfs/zfs-list.cache/bpool
cat /etc/zfs/zfs-list.cache/rpool
Once the files have data, stop zed:
fg
Press Ctrl-C.
Fix the paths to eliminate /mnt:
sed -Ei "s /mnt/? / " /etc/zfs/zfs-list.cache/*
Snapshot initial install:
zfs snapshot bpool/BOOT/debian@install
zfs snapshot rpool/ROOT/debian@install
Exit chroot:
exit

Finalizing One last sync was done in rescue mode:
for fs in /boot/ /boot/efi/ / /home/; do
    echo "syncing $fs to /mnt$fs..." && 
    rsync -aSHAXx --info=progress2 --delete $fs /mnt$fs
done
Then we unmount all filesystems:
mount   grep -v zfs   tac   awk '/\/mnt/  print $3 '   xargs -i  umount -lf  
zpool export -a
Reboot, swap the drives, and boot in ZFS. Hurray!

Benchmarks This is a test that was ran in single-user mode using fio and the Ars Technica recommended tests, which are:
  • Single 4KiB random write process:
     fio --name=randwrite4k1x --ioengine=posixaio --rw=randwrite --bs=4k --size=4g --numjobs=1 --iodepth=1 --runtime=60 --time_based --end_fsync=1
    
  • 16 parallel 64KiB random write processes:
     fio --name=randwrite64k16x --ioengine=posixaio --rw=randwrite --bs=64k --size=256m --numjobs=16 --iodepth=16 --runtime=60 --time_based --end_fsync=1
    
  • Single 1MiB random write process:
     fio --name=randwrite1m1x --ioengine=posixaio --rw=randwrite --bs=1m --size=16g --numjobs=1 --iodepth=1 --runtime=60 --time_based --end_fsync=1
    
Strangely, that's not exactly what the author, Jim Salter, did in his actual test bench used in the ZFS benchmarking article. The first thing is there's no read test at all, which is already pretty strange. But also it doesn't include stuff like dropping caches or repeating results. So here's my variation, which i called fio-ars-bench.sh for now. It just batches a bunch of fio tests, one by one, 60 seconds each. It should take about 12 minutes to run, as there are 3 pair of tests, read/write, with and without async. My bias, before building, running and analysing those results is that ZFS should outperform the traditional stack on writes, but possibly not on reads. It's also possible it outperforms it on both, because it's a newer drive. A new test might be possible with a new external USB drive as well, although I doubt I will find the time to do this.

Results All tests were done on WD blue SN550 drives, which claims to be able to push 2400MB/s read and 1750MB/s write. An extra drive was bought to move the LVM setup from a WDC WDS500G1B0B-00AS40 SSD, a WD blue M.2 2280 SSD that was at least 5 years old, spec'd at 560MB/s read, 530MB/s write. Benchmarks were done on the M.2 SSD drive but discarded so that the drive difference is not a factor in the test. In practice, I'm going to assume we'll never reach those numbers because we're not actually NVMe (this is an old workstation!) so the bottleneck isn't the disk itself. For our purposes, it might still give us useful results.

Rescue test, LUKS/LVM/ext4 Those tests were performed with everything shutdown, after either entering the system in rescue mode, or by reaching that target with:
systemctl rescue
The network might have been started before or after the test as well:
systemctl start systemd-networkd
So it should be fairly reliable as basically nothing else is running. Raw numbers, from the ?job-curie-lvm.log, converted to MiB/s and manually merged:
test read I/O read IOPS write I/O write IOPS
rand4k4g1x 39.27 10052 212.15 54310
rand4k4g1x--fsync=1 39.29 10057 2.73 699
rand64k256m16x 1297.00 20751 1068.57 17097
rand64k256m16x--fsync=1 1290.90 20654 353.82 5661
rand1m16g1x 315.15 315 563.77 563
rand1m16g1x--fsync=1 345.88 345 157.01 157
Peaks are at about 20k IOPS and ~1.3GiB/s read, 1GiB/s write in the 64KB blocks with 16 jobs. Slowest is the random 4k block sync write at an abysmal 3MB/s and 700 IOPS The 1MB read/write tests have lower IOPS, but that is expected.

Rescue test, ZFS This test was also performed in rescue mode. Raw numbers, from the ?job-curie-zfs.log, converted to MiB/s and manually merged:
test read I/O read IOPS write I/O write IOPS
rand4k4g1x 77.20 19763 27.13 6944
rand4k4g1x--fsync=1 76.16 19495 6.53 1673
rand64k256m16x 1882.40 30118 70.58 1129
rand64k256m16x--fsync=1 1865.13 29842 71.98 1151
rand1m16g1x 921.62 921 102.21 102
rand1m16g1x--fsync=1 908.37 908 64.30 64
Peaks are at 1.8GiB/s read, also in the 64k job like above, but much faster. The write is, as expected, much slower at 70MiB/s (compared to 1GiB/s!), but it should be noted the sync write doesn't degrade performance compared to async writes (although it's still below the LVM 300MB/s).

Conclusions Really, ZFS has trouble performing in all write conditions. The random 4k sync write test is the only place where ZFS outperforms LVM in writes, and barely (7MiB/s vs 3MiB/s). Everywhere else, writes are much slower, sometimes by an order of magnitude. And before some ZFS zealot jumps in talking about the SLOG or some other cache that could be added to improved performance, I'll remind you that those numbers are on a bare bones NVMe drive, pretty much as fast storage as you can find on this machine. Adding another NVMe drive as a cache probably will not improve write performance here. Still, those are very different results than the tests performed by Salter which shows ZFS beating traditional configurations in all categories but uncached 4k reads (not writes!). That said, those tests are very different from the tests I performed here, where I test writes on a single disk, not a RAID array, which might explain the discrepancy. Also, note that neither LVM or ZFS manage to reach the 2400MB/s read and 1750MB/s write performance specification. ZFS does manage to reach 82% of the read performance (1973MB/s) and LVM 64% of the write performance (1120MB/s). LVM hits 57% of the read performance and ZFS hits barely 6% of the write performance. Overall, I'm a bit disappointed in the ZFS write performance here, I must say. Maybe I need to tweak the record size or some other ZFS voodoo, but I'll note that I didn't have to do any such configuration on the other side to kick ZFS in the pants...

Real world experience This section document not synthetic backups, but actual real world workloads, comparing before and after I switched my workstation to ZFS.

Docker performance I had the feeling that running some git hook (which was firing a Docker container) was "slower" somehow. It seems that, at runtime, ZFS backends are significant slower than their overlayfs/ext4 equivalent:
May 16 14:42:52 curie systemd[1]: home-docker-overlay2-17e4d24228decc2d2d493efc401dbfb7ac29739da0e46775e122078d9daf3e87\x2dinit-merged.mount: Succeeded.
May 16 14:42:52 curie systemd[5161]: home-docker-overlay2-17e4d24228decc2d2d493efc401dbfb7ac29739da0e46775e122078d9daf3e87\x2dinit-merged.mount: Succeeded.
May 16 14:42:52 curie systemd[1]: home-docker-overlay2-17e4d24228decc2d2d493efc401dbfb7ac29739da0e46775e122078d9daf3e87-merged.mount: Succeeded.
May 16 14:42:53 curie dockerd[1723]: time="2022-05-16T14:42:53.087219426-04:00" level=info msg="starting signal loop" namespace=moby path=/run/docker/containerd/daemon/io.containerd.runtime.v2.task/moby/af22586fba07014a4d10ab19da10cf280db7a43cad804d6c1e9f2682f12b5f10 pid=151170
May 16 14:42:53 curie systemd[1]: Started libcontainer container af22586fba07014a4d10ab19da10cf280db7a43cad804d6c1e9f2682f12b5f10.
May 16 14:42:54 curie systemd[1]: docker-af22586fba07014a4d10ab19da10cf280db7a43cad804d6c1e9f2682f12b5f10.scope: Succeeded.
May 16 14:42:54 curie dockerd[1723]: time="2022-05-16T14:42:54.047297800-04:00" level=info msg="shim disconnected" id=af22586fba07014a4d10ab19da10cf280db7a43cad804d6c1e9f2682f12b5f10
May 16 14:42:54 curie dockerd[998]: time="2022-05-16T14:42:54.051365015-04:00" level=info msg="ignoring event" container=af22586fba07014a4d10ab19da10cf280db7a43cad804d6c1e9f2682f12b5f10 module=libcontainerd namespace=moby topic=/tasks/delete type="*events.TaskDelete"
May 16 14:42:54 curie systemd[2444]: run-docker-netns-f5453c87c879.mount: Succeeded.
May 16 14:42:54 curie systemd[5161]: run-docker-netns-f5453c87c879.mount: Succeeded.
May 16 14:42:54 curie systemd[2444]: home-docker-overlay2-17e4d24228decc2d2d493efc401dbfb7ac29739da0e46775e122078d9daf3e87-merged.mount: Succeeded.
May 16 14:42:54 curie systemd[5161]: home-docker-overlay2-17e4d24228decc2d2d493efc401dbfb7ac29739da0e46775e122078d9daf3e87-merged.mount: Succeeded.
May 16 14:42:54 curie systemd[1]: run-docker-netns-f5453c87c879.mount: Succeeded.
May 16 14:42:54 curie systemd[1]: home-docker-overlay2-17e4d24228decc2d2d493efc401dbfb7ac29739da0e46775e122078d9daf3e87-merged.mount: Succeeded.
Translating this:
  • container setup: ~1 second
  • container runtime: ~1 second
  • container teardown: ~1 second
  • total runtime: 2-3 seconds
Obviously, those timestamps are not quite accurate enough to make precise measurements... After I switched to ZFS:
mai 30 15:31:39 curie systemd[1]: var-lib-docker-zfs-graph-41ce08fb7a1d3a9c101694b82722f5621c0b4819bd1d9f070933fd1e00543cdf\x2dinit.mount: Succeeded. 
mai 30 15:31:39 curie systemd[5287]: var-lib-docker-zfs-graph-41ce08fb7a1d3a9c101694b82722f5621c0b4819bd1d9f070933fd1e00543cdf\x2dinit.mount: Succeeded. 
mai 30 15:31:40 curie systemd[1]: var-lib-docker-zfs-graph-41ce08fb7a1d3a9c101694b82722f5621c0b4819bd1d9f070933fd1e00543cdf.mount: Succeeded. 
mai 30 15:31:40 curie systemd[5287]: var-lib-docker-zfs-graph-41ce08fb7a1d3a9c101694b82722f5621c0b4819bd1d9f070933fd1e00543cdf.mount: Succeeded. 
mai 30 15:31:41 curie dockerd[3199]: time="2022-05-30T15:31:41.551403693-04:00" level=info msg="starting signal loop" namespace=moby path=/run/docker/containerd/daemon/io.containerd.runtime.v2.task/moby/42a1a1ed5912a7227148e997f442e7ab2e5cc3558aa3471548223c5888c9b142 pid=141080 
mai 30 15:31:41 curie systemd[1]: run-docker-runtime\x2drunc-moby-42a1a1ed5912a7227148e997f442e7ab2e5cc3558aa3471548223c5888c9b142-runc.ZVcjvl.mount: Succeeded. 
mai 30 15:31:41 curie systemd[5287]: run-docker-runtime\x2drunc-moby-42a1a1ed5912a7227148e997f442e7ab2e5cc3558aa3471548223c5888c9b142-runc.ZVcjvl.mount: Succeeded. 
mai 30 15:31:41 curie systemd[1]: Started libcontainer container 42a1a1ed5912a7227148e997f442e7ab2e5cc3558aa3471548223c5888c9b142. 
mai 30 15:31:45 curie systemd[1]: docker-42a1a1ed5912a7227148e997f442e7ab2e5cc3558aa3471548223c5888c9b142.scope: Succeeded. 
mai 30 15:31:45 curie dockerd[3199]: time="2022-05-30T15:31:45.883019128-04:00" level=info msg="shim disconnected" id=42a1a1ed5912a7227148e997f442e7ab2e5cc3558aa3471548223c5888c9b142 
mai 30 15:31:45 curie dockerd[1726]: time="2022-05-30T15:31:45.883064491-04:00" level=info msg="ignoring event" container=42a1a1ed5912a7227148e997f442e7ab2e5cc3558aa3471548223c5888c9b142 module=libcontainerd namespace=moby topic=/tasks/delete type="*events.TaskDelete" 
mai 30 15:31:45 curie systemd[1]: run-docker-netns-e45f5cf5f465.mount: Succeeded. 
mai 30 15:31:45 curie systemd[5287]: run-docker-netns-e45f5cf5f465.mount: Succeeded. 
mai 30 15:31:45 curie systemd[1]: var-lib-docker-zfs-graph-41ce08fb7a1d3a9c101694b82722f5621c0b4819bd1d9f070933fd1e00543cdf.mount: Succeeded. 
mai 30 15:31:45 curie systemd[5287]: var-lib-docker-zfs-graph-41ce08fb7a1d3a9c101694b82722f5621c0b4819bd1d9f070933fd1e00543cdf.mount: Succeeded.
That's double or triple the run time, from 2 seconds to 6 seconds. Most of the time is spent in run time, inside the container. Here's the breakdown:
  • container setup: ~2 seconds
  • container run: ~4 seconds
  • container teardown: ~1 second
  • total run time: about ~6-7 seconds
That's a two- to three-fold increase! Clearly something is going on here that I should tweak. It's possible that code path is less optimized in Docker. I also worry about podman, but apparently it also supports ZFS backends. Possibly it would perform better, but at this stage I wouldn't have a good comparison: maybe it would have performed better on non-ZFS as well...

Interactivity While doing the offsite backups (below), the system became somewhat "sluggish". I felt everything was slow, and I estimate it introduced ~50ms latency in any input device. Arguably, those are all USB and the external drive was connected through USB, but I suspect the ZFS drivers are not as well tuned with the scheduler as the regular filesystem drivers...

Recovery procedures For test purposes, I unmounted all systems during the procedure:
umount /mnt/boot/efi /mnt/boot/run
umount -a -t zfs
zpool export -a
And disconnected the drive, to see how I would recover this system from another Linux system in case of a total motherboard failure. To import an existing pool, plug the device, then import the pool with an alternate root, so it doesn't mount over your existing filesystems, then you mount the root filesystem and all the others:
zpool import -l -a -R /mnt &&
zfs mount rpool/ROOT/debian &&
zfs mount -a &&
mount /dev/sdc2 /mnt/boot/efi &&
mount -t tmpfs tmpfs /mnt/run &&
mkdir /mnt/run/lock

Offsite backup Part of the goal of using ZFS is to simplify and harden backups. I wanted to experiment with shorter recovery times specifically both point in time recovery objective and recovery time objective and faster incremental backups. This is, therefore, part of my backup services. This section documents how an external NVMe enclosure was setup in a pool to mirror the datasets from my workstation. The final setup should include syncoid copying datasets to the backup server regularly, but I haven't finished that configuration yet.

Partitioning The above partitioning procedure used sgdisk, but I couldn't figure out how to do this with sgdisk, so this uses sfdisk to dump the partition from the first disk to an external, identical drive:
sfdisk -d /dev/nvme0n1   sfdisk --no-reread /dev/sda --force

Pool creation This is similar to the main pool creation, except we tweaked a few bits after changing the upstream procedure:
zpool create \
        -o cachefile=/etc/zfs/zpool.cache \
        -o ashift=12 -d \
        -o feature@async_destroy=enabled \
        -o feature@bookmarks=enabled \
        -o feature@embedded_data=enabled \
        -o feature@empty_bpobj=enabled \
        -o feature@enabled_txg=enabled \
        -o feature@extensible_dataset=enabled \
        -o feature@filesystem_limits=enabled \
        -o feature@hole_birth=enabled \
        -o feature@large_blocks=enabled \
        -o feature@lz4_compress=enabled \
        -o feature@spacemap_histogram=enabled \
        -o feature@zpool_checkpoint=enabled \
        -O acltype=posixacl -O xattr=sa \
        -O compression=lz4 \
        -O devices=off \
        -O relatime=on \
        -O canmount=off \
        -O mountpoint=/boot -R /mnt \
        bpool-tubman /dev/sdb3
The change from the main boot pool are: Main pool creation is:
zpool create \
        -o ashift=12 \
        -O encryption=on -O keylocation=prompt -O keyformat=passphrase \
        -O acltype=posixacl -O xattr=sa -O dnodesize=auto \
        -O compression=zstd \
        -O relatime=on \
        -O canmount=off \
        -O mountpoint=/ -R /mnt \
        rpool-tubman /dev/sdb4

First sync I used syncoid to copy all pools over to the external device. syncoid is a thing that's part of the sanoid project which is specifically designed to sync snapshots between pool, typically over SSH links but it can also operate locally. The sanoid command had a --readonly argument to simulate changes, but syncoid didn't so I tried to fix that with an upstream PR. It seems it would be better to do this by hand, but this was much easier. The full first sync was:
root@curie:/home/anarcat# ./bin/syncoid -r  bpool bpool-tubman
CRITICAL ERROR: Target bpool-tubman exists but has no snapshots matching with bpool!
                Replication to target would require destroying existing
                target. Cowardly refusing to destroy your existing target.
          NOTE: Target bpool-tubman dataset is < 64MB used - did you mistakenly run
                 zfs create bpool-tubman  on the target? ZFS initial
                replication must be to a NON EXISTENT DATASET, which will
                then be CREATED BY the initial replication process.
INFO: Sending oldest full snapshot bpool/BOOT@test (~ 42 KB) to new target filesystem:
44.2KiB 0:00:00 [4.19MiB/s] [========================================================================================================================] 103%            
INFO: Updating new target filesystem with incremental bpool/BOOT@test ... syncoid_curie_2022-05-30:12:50:39 (~ 4 KB):
2.13KiB 0:00:00 [ 114KiB/s] [===============================================================>                                                         ] 53%            
INFO: Sending oldest full snapshot bpool/BOOT/debian@install (~ 126.0 MB) to new target filesystem:
 126MiB 0:00:00 [ 308MiB/s] [=======================================================================================================================>] 100%            
INFO: Updating new target filesystem with incremental bpool/BOOT/debian@install ... syncoid_curie_2022-05-30:12:50:39 (~ 113.4 MB):
 113MiB 0:00:00 [ 315MiB/s] [=======================================================================================================================>] 100%
root@curie:/home/anarcat# ./bin/syncoid -r  rpool rpool-tubman
CRITICAL ERROR: Target rpool-tubman exists but has no snapshots matching with rpool!
                Replication to target would require destroying existing
                target. Cowardly refusing to destroy your existing target.
          NOTE: Target rpool-tubman dataset is < 64MB used - did you mistakenly run
                 zfs create rpool-tubman  on the target? ZFS initial
                replication must be to a NON EXISTENT DATASET, which will
                then be CREATED BY the initial replication process.
INFO: Sending oldest full snapshot rpool/ROOT@syncoid_curie_2022-05-30:12:50:51 (~ 69 KB) to new target filesystem:
44.2KiB 0:00:00 [2.44MiB/s] [===========================================================================>                                             ] 63%            
INFO: Sending oldest full snapshot rpool/ROOT/debian@install (~ 25.9 GB) to new target filesystem:
25.9GiB 0:03:33 [ 124MiB/s] [=======================================================================================================================>] 100%            
INFO: Updating new target filesystem with incremental rpool/ROOT/debian@install ... syncoid_curie_2022-05-30:12:50:52 (~ 3.9 GB):
3.92GiB 0:00:33 [ 119MiB/s] [======================================================================================================================>  ] 99%            
INFO: Sending oldest full snapshot rpool/home@syncoid_curie_2022-05-30:12:55:04 (~ 276.8 GB) to new target filesystem:
 277GiB 0:27:13 [ 174MiB/s] [=======================================================================================================================>] 100%            
INFO: Sending oldest full snapshot rpool/home/root@syncoid_curie_2022-05-30:13:22:19 (~ 2.2 GB) to new target filesystem:
2.22GiB 0:00:25 [90.2MiB/s] [=======================================================================================================================>] 100%            
INFO: Sending oldest full snapshot rpool/var@syncoid_curie_2022-05-30:13:22:47 (~ 5.6 GB) to new target filesystem:
5.56GiB 0:00:32 [ 176MiB/s] [=======================================================================================================================>] 100%            
INFO: Sending oldest full snapshot rpool/var/cache@syncoid_curie_2022-05-30:13:23:22 (~ 627.3 MB) to new target filesystem:
 627MiB 0:00:03 [ 169MiB/s] [=======================================================================================================================>] 100%            
INFO: Sending oldest full snapshot rpool/var/lib@syncoid_curie_2022-05-30:13:23:28 (~ 69 KB) to new target filesystem:
44.2KiB 0:00:00 [1.40MiB/s] [===========================================================================>                                             ] 63%            
INFO: Sending oldest full snapshot rpool/var/lib/docker@syncoid_curie_2022-05-30:13:23:28 (~ 442.6 MB) to new target filesystem:
 443MiB 0:00:04 [ 103MiB/s] [=======================================================================================================================>] 100%            
INFO: Sending oldest full snapshot rpool/var/lib/docker/05c0de7fabbea60500eaa495d0d82038249f6faa63b12914737c4d71520e62c5@266253254 (~ 6.3 MB) to new target filesystem:
6.49MiB 0:00:00 [12.9MiB/s] [========================================================================================================================] 102%            
INFO: Updating new target filesystem with incremental rpool/var/lib/docker/05c0de7fabbea60500eaa495d0d82038249f6faa63b12914737c4d71520e62c5@266253254 ... syncoid_curie_2022-05-30:13:23:34 (~ 4 KB):
1.52KiB 0:00:00 [27.6KiB/s] [============================================>                                                                            ] 38%            
INFO: Sending oldest full snapshot rpool/var/lib/flatpak@syncoid_curie_2022-05-30:13:23:36 (~ 2.0 GB) to new target filesystem:
2.00GiB 0:00:17 [ 115MiB/s] [=======================================================================================================================>] 100%            
INFO: Sending oldest full snapshot rpool/var/tmp@syncoid_curie_2022-05-30:13:23:55 (~ 57.0 MB) to new target filesystem:
61.8MiB 0:00:01 [45.0MiB/s] [========================================================================================================================] 108%            
INFO: Clone is recreated on target rpool-tubman/var/lib/docker/ed71ddd563a779ba6fb37b3b1d0cc2c11eca9b594e77b4b234867ebcb162b205 based on rpool/var/lib/docker/05c0de7fabbea60500eaa495d0d82038249f6faa63b12914737c4d71520e62c5@266253254
INFO: Sending oldest full snapshot rpool/var/lib/docker/ed71ddd563a779ba6fb37b3b1d0cc2c11eca9b594e77b4b234867ebcb162b205@syncoid_curie_2022-05-30:13:23:58 (~ 218.6 MB) to new target filesystem:
 219MiB 0:00:01 [ 151MiB/s] [=======================================================================================================================>] 100%
Funny how the CRITICAL ERROR doesn't actually stop syncoid and it just carries on merrily doing when it's telling you it's "cowardly refusing to destroy your existing target"... Maybe that's because my pull request broke something though... During the transfer, the computer was very sluggish: everything feels like it has ~30-50ms latency extra:
anarcat@curie:sanoid$ LANG=C top -b  -n 1   head -20
top - 13:07:05 up 6 days,  4:01,  1 user,  load average: 16.13, 16.55, 11.83
Tasks: 606 total,   6 running, 598 sleeping,   0 stopped,   2 zombie
%Cpu(s): 18.8 us, 72.5 sy,  1.2 ni,  5.0 id,  1.2 wa,  0.0 hi,  1.2 si,  0.0 st
MiB Mem :  15898.4 total,   1387.6 free,  13170.0 used,   1340.8 buff/cache
MiB Swap:      0.0 total,      0.0 free,      0.0 used.   1319.8 avail Mem 
    PID USER      PR  NI    VIRT    RES    SHR S  %CPU  %MEM     TIME+ COMMAND
     70 root      20   0       0      0      0 S  83.3   0.0   6:12.67 kswapd0
4024878 root      20   0  282644  96432  10288 S  44.4   0.6   0:11.43 puppet
3896136 root      20   0   35328  16528     48 S  22.2   0.1   2:08.04 mbuffer
3896135 root      20   0   10328    776    168 R  16.7   0.0   1:22.93 zfs
3896138 root      20   0   10588    788    156 R  16.7   0.0   1:49.30 zfs
    350 root       0 -20       0      0      0 R  11.1   0.0   1:03.53 z_rd_int
    351 root       0 -20       0      0      0 S  11.1   0.0   1:04.15 z_rd_int
3896137 root      20   0    4384    352    244 R  11.1   0.0   0:44.73 pv
4034094 anarcat   30  10   20028  13960   2428 S  11.1   0.1   0:00.70 mbsync
4036539 anarcat   20   0    9604   3464   2408 R  11.1   0.0   0:00.04 top
    352 root       0 -20       0      0      0 S   5.6   0.0   1:03.64 z_rd_int
    353 root       0 -20       0      0      0 S   5.6   0.0   1:03.64 z_rd_int
    354 root       0 -20       0      0      0 S   5.6   0.0   1:04.01 z_rd_int
I wonder how much of that is due to syncoid, particularly because I often saw mbuffer and pv in there which are not strictly necessary to do those kind of operations, as far as I understand. Once that's done, export the pools to disconnect the drive:
zpool export bpool-tubman
zpool export rpool-tubman

Raw disk benchmark Copied the 512GB SSD/M.2 device to another 1024GB NVMe/M.2 device:
anarcat@curie:~$ sudo dd if=/dev/sdb of=/dev/sdc bs=4M status=progress conv=fdatasync
499944259584 octets (500 GB, 466 GiB) copi s, 1713 s, 292 MB/s
119235+1 enregistrements lus
119235+1 enregistrements  crits
500107862016 octets (500 GB, 466 GiB) copi s, 1719,93 s, 291 MB/s
... while both over USB, whoohoo 300MB/s!

Monitoring ZFS should be monitoring your pools regularly. Normally, the [[!debman zed]] daemon monitors all ZFS events. It is the thing that will report when a scrub failed, for example. See this configuration guide. Scrubs should be regularly scheduled to ensure consistency of the pool. This can be done in newer zfsutils-linux versions (bullseye-backports or bookworm) with one of those, depending on the desired frequency:
systemctl enable zfs-scrub-weekly@rpool.timer --now
systemctl enable zfs-scrub-monthly@rpool.timer --now
When the scrub runs, if it finds anything it will send an event which will get picked up by the zed daemon which will then send a notification, see below for an example. TODO: deploy on curie, if possible (probably not because no RAID) TODO: this should be in Puppet

Scrub warning example So what happens when problems are found? Here's an example of how I dealt with an error I received. After setting up another server (tubman) with ZFS, I eventually ended up getting a warning from the ZFS toolchain.
Date: Sun, 09 Oct 2022 00:58:08 -0400
From: root <root@anarc.at>
To: root@anarc.at
Subject: ZFS scrub_finish event for rpool on tubman
ZFS has finished a scrub:
   eid: 39536
 class: scrub_finish
  host: tubman
  time: 2022-10-09 00:58:07-0400
  pool: rpool
 state: ONLINE
status: One or more devices has experienced an unrecoverable error.  An
        attempt was made to correct the error.  Applications are unaffected.
action: Determine if the device needs to be replaced, and clear the errors
        using 'zpool clear' or replace the device with 'zpool replace'.
   see: https://openzfs.github.io/openzfs-docs/msg/ZFS-8000-9P
  scan: scrub repaired 0B in 00:33:57 with 0 errors on Sun Oct  9 00:58:07 2022
config:
        NAME        STATE     READ WRITE CKSUM
        rpool       ONLINE       0     0     0
          mirror-0  ONLINE       0     0     0
            sdb4    ONLINE       0     1     0
            sdc4    ONLINE       0     0     0
        cache
          sda3      ONLINE       0     0     0
errors: No known data errors
This, in itself, is a little worrisome. But it helpfully links to this more detailed documentation (and props up there: the link still works) which explains this is a "minor" problem (something that could be included in the report). In this case, this happened on a server setup on 2021-04-28, but the disks and server hardware are much older. The server itself (marcos v1) was built around 2011, over 10 years ago now. The hard drive in question is:
root@tubman:~# smartctl -i -qnoserial /dev/sdb
smartctl 7.2 2020-12-30 r5155 [x86_64-linux-5.10.0-15-amd64] (local build)
Copyright (C) 2002-20, Bruce Allen, Christian Franke, www.smartmontools.org
=== START OF INFORMATION SECTION ===
Model Family:     Seagate BarraCuda 3.5
Device Model:     ST4000DM004-2CV104
Firmware Version: 0001
User Capacity:    4,000,787,030,016 bytes [4.00 TB]
Sector Sizes:     512 bytes logical, 4096 bytes physical
Rotation Rate:    5425 rpm
Form Factor:      3.5 inches
Device is:        In smartctl database [for details use: -P show]
ATA Version is:   ACS-3 T13/2161-D revision 5
SATA Version is:  SATA 3.1, 6.0 Gb/s (current: 3.0 Gb/s)
Local Time is:    Tue Oct 11 11:02:32 2022 EDT
SMART support is: Available - device has SMART capability.
SMART support is: Enabled
Some more SMART stats:
root@tubman:~# smartctl -a -qnoserial /dev/sdb   grep -e  Head_Flying_Hours -e Power_On_Hours -e Total_LBA -e 'Sector Sizes'
Sector Sizes:     512 bytes logical, 4096 bytes physical
  9 Power_On_Hours          0x0032   086   086   000    Old_age   Always       -       12464 (206 202 0)
240 Head_Flying_Hours       0x0000   100   253   000    Old_age   Offline      -       10966h+55m+23.757s
241 Total_LBAs_Written      0x0000   100   253   000    Old_age   Offline      -       21107792664
242 Total_LBAs_Read         0x0000   100   253   000    Old_age   Offline      -       3201579750
That's over a year of power on, which shouldn't be so bad. It has written about 10TB of data (21107792664 LBAs * 512 byte/LBA), which is about two full writes. According to its specification, this device is supposed to support 55 TB/year of writes, so we're far below spec. Note that are still far from the "non-recoverable read error per bits" spec (1 per 10E15), as we've basically read 13E12 bits (3201579750 LBAs * 512 byte/LBA = 13E12 bits). It's likely this disk was made in 2018, so it is in its fourth year. Interestingly, /dev/sdc is also a Seagate drive, but of a different series:
root@tubman:~# smartctl -qnoserial  -i /dev/sdb
smartctl 7.2 2020-12-30 r5155 [x86_64-linux-5.10.0-15-amd64] (local build)
Copyright (C) 2002-20, Bruce Allen, Christian Franke, www.smartmontools.org
=== START OF INFORMATION SECTION ===
Model Family:     Seagate BarraCuda 3.5
Device Model:     ST4000DM004-2CV104
Firmware Version: 0001
User Capacity:    4,000,787,030,016 bytes [4.00 TB]
Sector Sizes:     512 bytes logical, 4096 bytes physical
Rotation Rate:    5425 rpm
Form Factor:      3.5 inches
Device is:        In smartctl database [for details use: -P show]
ATA Version is:   ACS-3 T13/2161-D revision 5
SATA Version is:  SATA 3.1, 6.0 Gb/s (current: 3.0 Gb/s)
Local Time is:    Tue Oct 11 11:21:35 2022 EDT
SMART support is: Available - device has SMART capability.
SMART support is: Enabled
It has seen much more reads than the other disk which is also interesting:
root@tubman:~# smartctl -a -qnoserial /dev/sdc   grep -e  Head_Flying_Hours -e Power_On_Hours -e Total_LBA -e 'Sector Sizes'
Sector Sizes:     512 bytes logical, 4096 bytes physical
  9 Power_On_Hours          0x0032   059   059   000    Old_age   Always       -       36240
240 Head_Flying_Hours       0x0000   100   253   000    Old_age   Offline      -       33994h+10m+52.118s
241 Total_LBAs_Written      0x0000   100   253   000    Old_age   Offline      -       30730174438
242 Total_LBAs_Read         0x0000   100   253   000    Old_age   Offline      -       51894566538
That's 4 years of Head_Flying_Hours, and over 4 years (4 years and 48 days) of Power_On_Hours. The copyright date on that drive's specs goes back to 2016, so it's a much older drive. SMART self-test succeeded.

Remaining issues
  • TODO: move send/receive backups to offsite host, see also zfs for alternatives to syncoid/sanoid there
  • TODO: setup backup cron job (or timer?)
  • TODO: swap still not setup on curie, see zfs
  • TODO: document this somewhere: bpool and rpool are both pools and datasets. that's pretty confusing, but also very useful because it allows for pool-wide recursive snapshots, which are used for the backup system

fio improvements I really want to improve my experience with fio. Right now, I'm just cargo-culting stuff from other folks and I don't really like it. stressant is a good example of my struggles, in the sense that it doesn't really work that well for disk tests. I would love to have just a single .fio job file that lists multiple jobs to run serially. For example, this file describes the above workload pretty well:
[global]
# cargo-culting Salter
fallocate=none
ioengine=posixaio
runtime=60
time_based=1
end_fsync=1
stonewall=1
group_reporting=1
# no need to drop caches, done by default
# invalidate=1
# Single 4KiB random read/write process
[randread-4k-4g-1x]
rw=randread
bs=4k
size=4g
numjobs=1
iodepth=1
[randwrite-4k-4g-1x]
rw=randwrite
bs=4k
size=4g
numjobs=1
iodepth=1
# 16 parallel 64KiB random read/write processes:
[randread-64k-256m-16x]
rw=randread
bs=64k
size=256m
numjobs=16
iodepth=16
[randwrite-64k-256m-16x]
rw=randwrite
bs=64k
size=256m
numjobs=16
iodepth=16
# Single 1MiB random read/write process
[randread-1m-16g-1x]
rw=randread
bs=1m
size=16g
numjobs=1
iodepth=1
[randwrite-1m-16g-1x]
rw=randwrite
bs=1m
size=16g
numjobs=1
iodepth=1
... except the jobs are actually started in parallel, even though they are stonewall'd, as far as I can tell by the reports. I sent a mail to the fio mailing list for clarification. It looks like the jobs are started in parallel, but actual (correctly) run serially. It seems like this might just be a matter of reporting the right timestamps in the end, although it does feel like starting all the processes (even if not doing any work yet) could skew the results.

Hangs during procedure During the procedure, it happened a few times where any ZFS command would completely hang. It seems that using an external USB drive to sync stuff didn't work so well: sometimes it would reconnect under a different device (from sdc to sdd, for example), and this would greatly confuse ZFS. Here, for example, is sdd reappearing out of the blue:
May 19 11:22:53 curie kernel: [  699.820301] scsi host4: uas
May 19 11:22:53 curie kernel: [  699.820544] usb 2-1: authorized to connect
May 19 11:22:53 curie kernel: [  699.922433] scsi 4:0:0:0: Direct-Access     ROG      ESD-S1C          0    PQ: 0 ANSI: 6
May 19 11:22:53 curie kernel: [  699.923235] sd 4:0:0:0: Attached scsi generic sg2 type 0
May 19 11:22:53 curie kernel: [  699.923676] sd 4:0:0:0: [sdd] 1953525168 512-byte logical blocks: (1.00 TB/932 GiB)
May 19 11:22:53 curie kernel: [  699.923788] sd 4:0:0:0: [sdd] Write Protect is off
May 19 11:22:53 curie kernel: [  699.923949] sd 4:0:0:0: [sdd] Write cache: enabled, read cache: enabled, doesn't support DPO or FUA
May 19 11:22:53 curie kernel: [  699.924149] sd 4:0:0:0: [sdd] Optimal transfer size 33553920 bytes
May 19 11:22:53 curie kernel: [  699.961602]  sdd: sdd1 sdd2 sdd3 sdd4
May 19 11:22:53 curie kernel: [  699.996083] sd 4:0:0:0: [sdd] Attached SCSI disk
Next time I run a ZFS command (say zpool list), the command completely hangs (D state) and this comes up in the logs:
May 19 11:34:21 curie kernel: [ 1387.914843] zio pool=bpool vdev=/dev/sdc3 error=5 type=2 offset=71344128 size=4096 flags=184880
May 19 11:34:21 curie kernel: [ 1387.914859] zio pool=bpool vdev=/dev/sdc3 error=5 type=2 offset=205565952 size=4096 flags=184880
May 19 11:34:21 curie kernel: [ 1387.914874] zio pool=bpool vdev=/dev/sdc3 error=5 type=2 offset=272789504 size=4096 flags=184880
May 19 11:34:21 curie kernel: [ 1387.914906] zio pool=bpool vdev=/dev/sdc3 error=5 type=1 offset=270336 size=8192 flags=b08c1
May 19 11:34:21 curie kernel: [ 1387.914932] zio pool=bpool vdev=/dev/sdc3 error=5 type=1 offset=1073225728 size=8192 flags=b08c1
May 19 11:34:21 curie kernel: [ 1387.914948] zio pool=bpool vdev=/dev/sdc3 error=5 type=1 offset=1073487872 size=8192 flags=b08c1
May 19 11:34:21 curie kernel: [ 1387.915165] zio pool=bpool vdev=/dev/sdc3 error=5 type=2 offset=272793600 size=4096 flags=184880
May 19 11:34:21 curie kernel: [ 1387.915183] zio pool=bpool vdev=/dev/sdc3 error=5 type=2 offset=339853312 size=4096 flags=184880
May 19 11:34:21 curie kernel: [ 1387.915648] WARNING: Pool 'bpool' has encountered an uncorrectable I/O failure and has been suspended.
May 19 11:34:21 curie kernel: [ 1387.915648] 
May 19 11:37:25 curie kernel: [ 1571.558614] task:txg_sync        state:D stack:    0 pid:  997 ppid:     2 flags:0x00004000
May 19 11:37:25 curie kernel: [ 1571.558623] Call Trace:
May 19 11:37:25 curie kernel: [ 1571.558640]  __schedule+0x282/0x870
May 19 11:37:25 curie kernel: [ 1571.558650]  schedule+0x46/0xb0
May 19 11:37:25 curie kernel: [ 1571.558670]  schedule_timeout+0x8b/0x140
May 19 11:37:25 curie kernel: [ 1571.558675]  ? __next_timer_interrupt+0x110/0x110
May 19 11:37:25 curie kernel: [ 1571.558678]  io_schedule_timeout+0x4c/0x80
May 19 11:37:25 curie kernel: [ 1571.558689]  __cv_timedwait_common+0x12b/0x160 [spl]
May 19 11:37:25 curie kernel: [ 1571.558694]  ? add_wait_queue_exclusive+0x70/0x70
May 19 11:37:25 curie kernel: [ 1571.558702]  __cv_timedwait_io+0x15/0x20 [spl]
May 19 11:37:25 curie kernel: [ 1571.558816]  zio_wait+0x129/0x2b0 [zfs]
May 19 11:37:25 curie kernel: [ 1571.558929]  dsl_pool_sync+0x461/0x4f0 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559032]  spa_sync+0x575/0xfa0 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559138]  ? spa_txg_history_init_io+0x101/0x110 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559245]  txg_sync_thread+0x2e0/0x4a0 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559354]  ? txg_fini+0x240/0x240 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559366]  thread_generic_wrapper+0x6f/0x80 [spl]
May 19 11:37:25 curie kernel: [ 1571.559376]  ? __thread_exit+0x20/0x20 [spl]
May 19 11:37:25 curie kernel: [ 1571.559379]  kthread+0x11b/0x140
May 19 11:37:25 curie kernel: [ 1571.559382]  ? __kthread_bind_mask+0x60/0x60
May 19 11:37:25 curie kernel: [ 1571.559386]  ret_from_fork+0x22/0x30
May 19 11:37:25 curie kernel: [ 1571.559401] task:zed             state:D stack:    0 pid: 1564 ppid:     1 flags:0x00000000
May 19 11:37:25 curie kernel: [ 1571.559404] Call Trace:
May 19 11:37:25 curie kernel: [ 1571.559409]  __schedule+0x282/0x870
May 19 11:37:25 curie kernel: [ 1571.559412]  ? __kmalloc_node+0x141/0x2b0
May 19 11:37:25 curie kernel: [ 1571.559417]  schedule+0x46/0xb0
May 19 11:37:25 curie kernel: [ 1571.559420]  schedule_preempt_disabled+0xa/0x10
May 19 11:37:25 curie kernel: [ 1571.559424]  __mutex_lock.constprop.0+0x133/0x460
May 19 11:37:25 curie kernel: [ 1571.559435]  ? nvlist_xalloc.part.0+0x68/0xc0 [znvpair]
May 19 11:37:25 curie kernel: [ 1571.559537]  spa_all_configs+0x41/0x120 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559644]  zfs_ioc_pool_configs+0x17/0x70 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559752]  zfsdev_ioctl_common+0x697/0x870 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559758]  ? _copy_from_user+0x28/0x60
May 19 11:37:25 curie kernel: [ 1571.559860]  zfsdev_ioctl+0x53/0xe0 [zfs]
May 19 11:37:25 curie kernel: [ 1571.559866]  __x64_sys_ioctl+0x83/0xb0
May 19 11:37:25 curie kernel: [ 1571.559869]  do_syscall_64+0x33/0x80
May 19 11:37:25 curie kernel: [ 1571.559873]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
May 19 11:37:25 curie kernel: [ 1571.559876] RIP: 0033:0x7fcf0ef32cc7
May 19 11:37:25 curie kernel: [ 1571.559878] RSP: 002b:00007fcf0e181618 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
May 19 11:37:25 curie kernel: [ 1571.559881] RAX: ffffffffffffffda RBX: 000055b212f972a0 RCX: 00007fcf0ef32cc7
May 19 11:37:25 curie kernel: [ 1571.559883] RDX: 00007fcf0e181640 RSI: 0000000000005a04 RDI: 000000000000000b
May 19 11:37:25 curie kernel: [ 1571.559885] RBP: 00007fcf0e184c30 R08: 00007fcf08016810 R09: 00007fcf08000080
May 19 11:37:25 curie kernel: [ 1571.559886] R10: 0000000000080000 R11: 0000000000000246 R12: 000055b212f972a0
May 19 11:37:25 curie kernel: [ 1571.559888] R13: 0000000000000000 R14: 00007fcf0e181640 R15: 0000000000000000
May 19 11:37:25 curie kernel: [ 1571.559980] task:zpool           state:D stack:    0 pid:11815 ppid:  3816 flags:0x00004000
May 19 11:37:25 curie kernel: [ 1571.559983] Call Trace:
May 19 11:37:25 curie kernel: [ 1571.559988]  __schedule+0x282/0x870
May 19 11:37:25 curie kernel: [ 1571.559992]  schedule+0x46/0xb0
May 19 11:37:25 curie kernel: [ 1571.559995]  io_schedule+0x42/0x70
May 19 11:37:25 curie kernel: [ 1571.560004]  cv_wait_common+0xac/0x130 [spl]
May 19 11:37:25 curie kernel: [ 1571.560008]  ? add_wait_queue_exclusive+0x70/0x70
May 19 11:37:25 curie kernel: [ 1571.560118]  txg_wait_synced_impl+0xc9/0x110 [zfs]
May 19 11:37:25 curie kernel: [ 1571.560223]  txg_wait_synced+0xc/0x40 [zfs]
May 19 11:37:25 curie kernel: [ 1571.560325]  spa_export_common+0x4cd/0x590 [zfs]
May 19 11:37:25 curie kernel: [ 1571.560430]  ? zfs_log_history+0x9c/0xf0 [zfs]
May 19 11:37:25 curie kernel: [ 1571.560537]  zfsdev_ioctl_common+0x697/0x870 [zfs]
May 19 11:37:25 curie kernel: [ 1571.560543]  ? _copy_from_user+0x28/0x60
May 19 11:37:25 curie kernel: [ 1571.560644]  zfsdev_ioctl+0x53/0xe0 [zfs]
May 19 11:37:25 curie kernel: [ 1571.560649]  __x64_sys_ioctl+0x83/0xb0
May 19 11:37:25 curie kernel: [ 1571.560653]  do_syscall_64+0x33/0x80
May 19 11:37:25 curie kernel: [ 1571.560656]  entry_SYSCALL_64_after_hwframe+0x44/0xa9
May 19 11:37:25 curie kernel: [ 1571.560659] RIP: 0033:0x7fdc23be2cc7
May 19 11:37:25 curie kernel: [ 1571.560661] RSP: 002b:00007ffc8c792478 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
May 19 11:37:25 curie kernel: [ 1571.560664] RAX: ffffffffffffffda RBX: 000055942ca49e20 RCX: 00007fdc23be2cc7
May 19 11:37:25 curie kernel: [ 1571.560666] RDX: 00007ffc8c792490 RSI: 0000000000005a03 RDI: 0000000000000003
May 19 11:37:25 curie kernel: [ 1571.560667] RBP: 00007ffc8c795e80 R08: 00000000ffffffff R09: 00007ffc8c792310
May 19 11:37:25 curie kernel: [ 1571.560669] R10: 000055942ca49e30 R11: 0000000000000246 R12: 00007ffc8c792490
May 19 11:37:25 curie kernel: [ 1571.560671] R13: 000055942ca49e30 R14: 000055942aed2c20 R15: 00007ffc8c795a40
Here's another example, where you see the USB controller bleeping out and back into existence:
mai 19 11:38:39 curie kernel: usb 2-1: USB disconnect, device number 2
mai 19 11:38:39 curie kernel: sd 4:0:0:0: [sdd] Synchronizing SCSI cache
mai 19 11:38:39 curie kernel: sd 4:0:0:0: [sdd] Synchronize Cache(10) failed: Result: hostbyte=DID_ERROR driverbyte=DRIVER_OK
mai 19 11:39:25 curie kernel: INFO: task zed:1564 blocked for more than 241 seconds.
mai 19 11:39:25 curie kernel:       Tainted: P          IOE     5.10.0-14-amd64 #1 Debian 5.10.113-1
mai 19 11:39:25 curie kernel: "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
mai 19 11:39:25 curie kernel: task:zed             state:D stack:    0 pid: 1564 ppid:     1 flags:0x00000000
mai 19 11:39:25 curie kernel: Call Trace:
mai 19 11:39:25 curie kernel:  __schedule+0x282/0x870
mai 19 11:39:25 curie kernel:  ? __kmalloc_node+0x141/0x2b0
mai 19 11:39:25 curie kernel:  schedule+0x46/0xb0
mai 19 11:39:25 curie kernel:  schedule_preempt_disabled+0xa/0x10
mai 19 11:39:25 curie kernel:  __mutex_lock.constprop.0+0x133/0x460
mai 19 11:39:25 curie kernel:  ? nvlist_xalloc.part.0+0x68/0xc0 [znvpair]
mai 19 11:39:25 curie kernel:  spa_all_configs+0x41/0x120 [zfs]
mai 19 11:39:25 curie kernel:  zfs_ioc_pool_configs+0x17/0x70 [zfs]
mai 19 11:39:25 curie kernel:  zfsdev_ioctl_common+0x697/0x870 [zfs]
mai 19 11:39:25 curie kernel:  ? _copy_from_user+0x28/0x60
mai 19 11:39:25 curie kernel:  zfsdev_ioctl+0x53/0xe0 [zfs]
mai 19 11:39:25 curie kernel:  __x64_sys_ioctl+0x83/0xb0
mai 19 11:39:25 curie kernel:  do_syscall_64+0x33/0x80
mai 19 11:39:25 curie kernel:  entry_SYSCALL_64_after_hwframe+0x44/0xa9
mai 19 11:39:25 curie kernel: RIP: 0033:0x7fcf0ef32cc7
mai 19 11:39:25 curie kernel: RSP: 002b:00007fcf0e181618 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
mai 19 11:39:25 curie kernel: RAX: ffffffffffffffda RBX: 000055b212f972a0 RCX: 00007fcf0ef32cc7
mai 19 11:39:25 curie kernel: RDX: 00007fcf0e181640 RSI: 0000000000005a04 RDI: 000000000000000b
mai 19 11:39:25 curie kernel: RBP: 00007fcf0e184c30 R08: 00007fcf08016810 R09: 00007fcf08000080
mai 19 11:39:25 curie kernel: R10: 0000000000080000 R11: 0000000000000246 R12: 000055b212f972a0
mai 19 11:39:25 curie kernel: R13: 0000000000000000 R14: 00007fcf0e181640 R15: 0000000000000000
mai 19 11:39:25 curie kernel: INFO: task zpool:11815 blocked for more than 241 seconds.
mai 19 11:39:25 curie kernel:       Tainted: P          IOE     5.10.0-14-amd64 #1 Debian 5.10.113-1
mai 19 11:39:25 curie kernel: "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
mai 19 11:39:25 curie kernel: task:zpool           state:D stack:    0 pid:11815 ppid:  2621 flags:0x00004004
mai 19 11:39:25 curie kernel: Call Trace:
mai 19 11:39:25 curie kernel:  __schedule+0x282/0x870
mai 19 11:39:25 curie kernel:  schedule+0x46/0xb0
mai 19 11:39:25 curie kernel:  io_schedule+0x42/0x70
mai 19 11:39:25 curie kernel:  cv_wait_common+0xac/0x130 [spl]
mai 19 11:39:25 curie kernel:  ? add_wait_queue_exclusive+0x70/0x70
mai 19 11:39:25 curie kernel:  txg_wait_synced_impl+0xc9/0x110 [zfs]
mai 19 11:39:25 curie kernel:  txg_wait_synced+0xc/0x40 [zfs]
mai 19 11:39:25 curie kernel:  spa_export_common+0x4cd/0x590 [zfs]
mai 19 11:39:25 curie kernel:  ? zfs_log_history+0x9c/0xf0 [zfs]
mai 19 11:39:25 curie kernel:  zfsdev_ioctl_common+0x697/0x870 [zfs]
mai 19 11:39:25 curie kernel:  ? _copy_from_user+0x28/0x60
mai 19 11:39:25 curie kernel:  zfsdev_ioctl+0x53/0xe0 [zfs]
mai 19 11:39:25 curie kernel:  __x64_sys_ioctl+0x83/0xb0
mai 19 11:39:25 curie kernel:  do_syscall_64+0x33/0x80
mai 19 11:39:25 curie kernel:  entry_SYSCALL_64_after_hwframe+0x44/0xa9
mai 19 11:39:25 curie kernel: RIP: 0033:0x7fdc23be2cc7
mai 19 11:39:25 curie kernel: RSP: 002b:00007ffc8c792478 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
mai 19 11:39:25 curie kernel: RAX: ffffffffffffffda RBX: 000055942ca49e20 RCX: 00007fdc23be2cc7
mai 19 11:39:25 curie kernel: RDX: 00007ffc8c792490 RSI: 0000000000005a03 RDI: 0000000000000003
mai 19 11:39:25 curie kernel: RBP: 00007ffc8c795e80 R08: 00000000ffffffff R09: 00007ffc8c792310
mai 19 11:39:25 curie kernel: R10: 000055942ca49e30 R11: 0000000000000246 R12: 00007ffc8c792490
mai 19 11:39:25 curie kernel: R13: 000055942ca49e30 R14: 000055942aed2c20 R15: 00007ffc8c795a40
I understand those are rather extreme conditions: I would fully expect the pool to stop working if the underlying drives disappear. What doesn't seem acceptable is that a command would completely hang like this.

References See the zfs documentation for more information about ZFS, and tubman for another installation and migration procedure.

29 October 2022

Fran ois Marier: Making the mounting of an encrypted /home optional on a home server

I have a computer that serves as a home server as well as a desktop machine. It has an encrypted home directory to protect user files and, in the default configuration, that unfortunately interferes with unattended reboots since someone needs to be present to enter the encryption password. Here's how I added a timeout and made /home optional on that machine. I started by adding a one-minute timeout on the password prompt by adding timeout=60 in my /etc/crypttab:
crypt  UUID=7e12c123-abcd-5555-8c40-900d1f8cc281  none  luks,timeout=60
then I made /home optional by adding nofail to the appropriate mount point in /etc/fstab:
/dev/mapper/crypt  /home  ext4  nodev,noatime,nosuid,nofail  0  2
Before that, the password prompt would timeout but the system would be unable to boot since one of the required partitions had failed to mount. Now, to ensure that I don't accidentally re-create home directories for users when the system is mounted without a /home, I made the /home directory on the non-encrypted drive read-only:
umount /home
cd /home
chmod a-w .
Finally, with all of this in place, I was now happy to configure the machine to automatically reboot after a kernel panic by putting the following in /etc/sysctl.d/local.conf:
# Automatic reboot 10 seconds after a kernel panic
kernel.panic = 10
since I know that the machine will come back up just fine and that all services will be running. I simply won't be able to log into that machine as any other user than root until I manually unlock and mount /home.

10 September 2022

Holger Levsen: 20220910-youngest-LUKS-user

youngest LUKS user I know... So I'm in Berlin currently to attend the fourth Qubes OS Summit, also to discuss the future of the reproducible-builds.org mirror of snapshot.debian.org and in the evening I've met an old Debian friend who told a lovely story about his 5 year old daughter, who since recently is a Debian user using an old laptop with LUKS encryption, knowing her data will be lost when she forgets her passphrase... The Qubes OS Summit is also very cool, great people and exciting developments!

26 August 2022

Antoine Beaupr : How to nationalize the internet in Canada

Rogers had a catastrophic failure in July 2022. It affected emergency services (as in: people couldn't call 911, but also some 911 services themselves failed), hospitals (which couldn't access prescriptions), banks and payment systems (as payment terminals stopped working), and regular users as well. The outage lasted almost a full day, and Rogers took days to give any technical explanation on the outage, and even when they did, details were sparse. So far the only detailed account is from outside actors like Cloudflare which seem to point at an internal BGP failure. Its impact on the economy has yet to be measured, but it probably cost millions of dollars in wasted time and possibly lead to life-threatening situations. Apart from holding Rogers (criminally?) responsible for this, what should be done in the future to avoid such problems? It's not the first time something like this has happened: it happened to Bell Canada as well. The Rogers outage is also strangely similar to the Facebook outage last year, but, to its credit, Facebook did post a fairly detailed explanation only a day later. The internet is designed to be decentralised, and having large companies like Rogers hold so much power is a crucial mistake that should be reverted. The question is how. Some critics were quick to point out that we need more ISP diversity and competition, but I think that's missing the point. Others have suggested that the internet should be a public good or even straight out nationalized. I believe the solution to the problem of large, private, centralised telcos and ISPs is to replace them with smaller, public, decentralised service providers. The only way to ensure that works is to make sure that public money ends up creating infrastructure controlled by the public, which means treating ISPs as a public utility. This has been implemented elsewhere: it works, it's cheaper, and provides better service.

A modest proposal Global wireless services (like phone services) and home internet inevitably grow into monopolies. They are public utilities, just like water, power, railways, and roads. The question of how they should be managed is therefore inherently political, yet people don't seem to question the idea that only the market (i.e. "competition") can solve this problem. I disagree. 10 years ago (in french), I suggested we, in Qu bec, should nationalize large telcos and internet service providers. I no longer believe is a realistic approach: most of those companies have crap copper-based networks (at least for the last mile), yet are worth billions of dollars. It would be prohibitive, and a waste, to buy them out. Back then, I called this idea "R seau-Qu bec", a reference to the already nationalized power company, Hydro-Qu bec. (This idea, incidentally, made it into the plan of a political party.) Now, I think we should instead build our own, public internet. Start setting up municipal internet services, fiber to the home in all cities, progressively. Then interconnect cities with fiber, and build peering agreements with other providers. This also includes a bid on wireless spectrum to start competing with phone providers as well. And while that sounds really ambitious, I think it's possible to take this one step at a time.

Municipal broadband In many parts of the world, municipal broadband is an elegant solution to the problem, with solutions ranging from Stockholm's city-owned fiber network (dark fiber, layer 1) to Utah's UTOPIA network (fiber to the premises, layer 2) and municipal wireless networks like Guifi.net which connects about 40,000 nodes in Catalonia. A good first step would be for cities to start providing broadband services to its residents, directly. Cities normally own sewage and water systems that interconnect most residences and therefore have direct physical access everywhere. In Montr al, in particular, there is an ongoing project to replace a lot of old lead-based plumbing which would give an opportunity to lay down a wired fiber network across the city. This is a wild guess, but I suspect this would be much less expensive than one would think. Some people agree with me and quote this as low as 1000$ per household. There is about 800,000 households in the city of Montr al, so we're talking about a 800 million dollars investment here, to connect every household in Montr al with fiber and incidentally a quarter of the province's population. And this is not an up-front cost: this can be built progressively, with expenses amortized over many years. (We should not, however, connect Montr al first: it's used as an example here because it's a large number of households to connect.) Such a network should be built with a redundant topology. I leave it as an open question whether we should adopt Stockholm's more minimalist approach or provide direct IP connectivity. I would tend to favor the latter, because then you can immediately start to offer the service to households and generate revenues to compensate for the capital expenditures. Given the ridiculous profit margins telcos currently have 8 billion $CAD net income for BCE (2019), 2 billion $CAD for Rogers (2020) I also believe this would actually turn into a profitable revenue stream for the city, the same way Hydro-Qu bec is more and more considered as a revenue stream for the state. (I personally believe that's actually wrong and we should treat those resources as human rights and not money cows, but I digress. The point is: this is not a cost point, it's a revenue.) The other major challenge here is that the city will need competent engineers to drive this project forward. But this is not different from the way other public utilities run: we have electrical engineers at Hydro, sewer and water engineers at the city, this is just another profession. If anything, the computing science sector might be more at fault than the city here in its failure to provide competent and accountable engineers to society... Right now, most of the network in Canada is copper: we are hitting the limits of that technology with DSL, and while cable has some life left to it (DOCSIS 4.0 does 4Gbps), that is nowhere near the capacity of fiber. Take the town of Chattanooga, Tennessee: in 2010, the city-owned ISP EPB finished deploying a fiber network to the entire town and provided gigabit internet to everyone. Now, 12 years later, they are using this same network to provide the mind-boggling speed of 25 gigabit to the home. To give you an idea, Chattanooga is roughly the size and density of Sherbrooke.

Provincial public internet As part of building a municipal network, the question of getting access to "the internet" will immediately come up. Naturally, this will first be solved by using already existing commercial providers to hook up residents to the rest of the global network. But eventually, networks should inter-connect: Montr al should connect with Laval, and then Trois-Rivi res, then Qu bec City. This will require long haul fiber runs, but those links are not actually that expensive, and many of those already exist as a public resource at RISQ and CANARIE, which cross-connects universities and colleges across the province and the country. Those networks might not have the capacity to cover the needs of the entire province right now, but that is a router upgrade away, thanks to the amazing capacity of fiber. There are two crucial mistakes to avoid at this point. First, the network needs to remain decentralised. Long haul links should be IP links with BGP sessions, and each city (or MRC) should have its own independent network, to avoid Rogers-class catastrophic failures. Second, skill needs to remain in-house: RISQ has already made that mistake, to a certain extent, by selling its neutral datacenter. Tellingly, MetroOptic, probably the largest commercial dark fiber provider in the province, now operates the QIX, the second largest "public" internet exchange in Canada. Still, we have a lot of infrastructure we can leverage here. If RISQ or CANARIE cannot be up to the task, Hydro-Qu bec has power lines running into every house in the province, with high voltage power lines running hundreds of kilometers far north. The logistics of long distance maintenance are already solved by that institution. In fact, Hydro already has fiber all over the province, but it is a private network, separate from the internet for security reasons (and that should probably remain so). But this only shows they already have the expertise to lay down fiber: they would just need to lay down a parallel network to the existing one. In that architecture, Hydro would be a "dark fiber" provider.

International public internet None of the above solves the problem for the entire population of Qu bec, which is notoriously dispersed, with an area three times the size of France, but with only an eight of its population (8 million vs 67). More specifically, Canada was originally a french colony, a land violently stolen from native people who have lived here for thousands of years. Some of those people now live in reservations, sometimes far from urban centers (but definitely not always). So the idea of leveraging the Hydro-Qu bec infrastructure doesn't always work to solve this, because while Hydro will happily flood a traditional hunting territory for an electric dam, they don't bother running power lines to the village they forcibly moved, powering it instead with noisy and polluting diesel generators. So before giving me fiber to the home, we should give power (and potable water, for that matter), to those communities first. So we need to discuss international connectivity. (How else could we consider those communities than peer nations anyways?c) Qu bec has virtually zero international links. Even in Montr al, which likes to style itself a major player in gaming, AI, and technology, most peering goes through either Toronto or New York. That's a problem that we must fix, regardless of the other problems stated here. Looking at the submarine cable map, we see very few international links actually landing in Canada. There is the Greenland connect which connects Newfoundland to Iceland through Greenland. There's the EXA which lands in Ireland, the UK and the US, and Google has the Topaz link on the west coast. That's about it, and none of those land anywhere near any major urban center in Qu bec. We should have a cable running from France up to Saint-F licien. There should be a cable from Vancouver to China. Heck, there should be a fiber cable running all the way from the end of the great lakes through Qu bec, then up around the northern passage and back down to British Columbia. Those cables are expensive, and the idea might sound ludicrous, but Russia is actually planning such a project for 2026. The US has cables running all the way up (and around!) Alaska, neatly bypassing all of Canada in the process. We just look ridiculous on that map. (Addendum: I somehow forgot to talk about Teleglobe here was founded as publicly owned company in 1950, growing international phone and (later) data links all over the world. It was privatized by the conservatives in 1984, along with rails and other "crown corporations". So that's one major risk to any effort to make public utilities work properly: some government might be elected and promptly sell it out to its friends for peanuts.)

Wireless networks I know most people will have rolled their eyes so far back their heads have exploded. But I'm not done yet. I want wireless too. And by wireless, I don't mean a bunch of geeks setting up OpenWRT routers on rooftops. I tried that, and while it was fun and educational, it didn't scale. A public networking utility wouldn't be complete without providing cellular phone service. This involves bidding for frequencies at the federal level, and deploying a rather large amount of infrastructure, but it could be a later phase, when the engineers and politicians have proven their worth. At least part of the Rogers fiasco would have been averted if such a decentralized network backend existed. One might even want to argue that a separate institution should be setup to provide phone services, independently from the regular wired networking, if only for reliability. Because remember here: the problem we're trying to solve is not just technical, it's about political boundaries, centralisation, and automation. If everything is ran by this one organisation again, we will have failed. However, I must admit that phone services is where my ideas fall a little short. I can't help but think it's also an accessible goal maybe starting with a virtual operator but it seems slightly less so than the others, especially considering how closed the phone ecosystem is.

Counter points In debating these ideas while writing this article, the following objections came up.

I don't want the state to control my internet One legitimate concern I have about the idea of the state running the internet is the potential it would have to censor or control the content running over the wires. But I don't think there is necessarily a direct relationship between resource ownership and control of content. Sure, China has strong censorship in place, partly implemented through state-controlled businesses. But Russia also has strong censorship in place, based on regulatory tools: they force private service providers to install back-doors in their networks to control content and surveil their users. Besides, the USA have been doing warrantless wiretapping since at least 2003 (and yes, that's 10 years before the Snowden revelations) so a commercial internet is no assurance that we have a free internet. Quite the contrary in fact: if anything, the commercial internet goes hand in hand with the neo-colonial internet, just like businesses did in the "good old colonial days". Large media companies are the primary censors of content here. In Canada, the media cartel requested the first site-blocking order in 2018. The plaintiffs (including Qu becor, Rogers, and Bell Canada) are both content providers and internet service providers, an obvious conflict of interest. Nevertheless, there are some strong arguments against having a centralised, state-owned monopoly on internet service providers. FDN makes a good point on this. But this is not what I am suggesting: at the provincial level, the network would be purely physical, and regional entities (which could include private companies) would peer over that physical network, ensuring decentralization. Delegating the management of that infrastructure to an independent non-profit or cooperative (but owned by the state) would also ensure some level of independence.

Isn't the government incompetent and corrupt? Also known as "private enterprise is better skilled at handling this, the state can't do anything right" I don't think this is a "fait accomplit". If anything, I have found publicly ran utilities to be spectacularly reliable here. I rarely have trouble with sewage, water, or power, and keep in mind I live in a city where we receive about 2 meters of snow a year, which tend to create lots of trouble with power lines. Unless there's a major weather event, power just runs here. I think the same can happen with an internet service provider. But it would certainly need to have higher standards to what we're used to, because frankly Internet is kind of janky.

A single monopoly will be less reliable I actually agree with that, but that is not what I am proposing anyways. Current commercial or non-profit entities will be free to offer their services on top of the public network. And besides, the current "ha! diversity is great" approach is exactly what we have now, and it's not working. The pretense that we can have competition over a single network is what led the US into the ridiculous situation where they also pretend to have competition over the power utility market. This led to massive forest fires in California and major power outages in Texas. It doesn't work.

Wouldn't this create an isolated network? One theory is that this new network would be so hostile to incumbent telcos and ISPs that they would simply refuse to network with the public utility. And while it is true that the telcos currently do also act as a kind of "tier one" provider in some places, I strongly feel this is also a problem that needs to be solved, regardless of ownership of networking infrastructure. Right now, telcos often hold both ends of the stick: they are the gateway to users, the "last mile", but they also provide peering to the larger internet in some locations. In at least one datacenter in downtown Montr al, I've seen traffic go through Bell Canada that was not directly targeted at Bell customers. So in effect, they are in a position of charging twice for the same traffic, and that's not only ridiculous, it should just be plain illegal. And besides, this is not a big problem: there are other providers out there. As bad as the market is in Qu bec, there is still some diversity in Tier one providers that could allow for some exits to the wider network (e.g. yes, Cogent is here too).

What about Google and Facebook? Nationalization of other service providers like Google and Facebook is out of scope of this discussion. That said, I am not sure the state should get into the business of organising the web or providing content services however, but I will point out it already does do some of that through its own websites. It should probably keep itself to this, and also consider providing normal services for people who don't or can't access the internet. (And I would also be ready to argue that Google and Facebook already act as extensions of the state: certainly if Facebook didn't exist, the CIA or the NSA would like to create it at this point. And Google has lucrative business with the US department of defense.)

What does not work So we've seen one thing that could work. Maybe it's too expensive. Maybe the political will isn't there. Maybe it will fail. We don't know yet. But we know what does not work, and it's what we've been doing ever since the internet has gone commercial.

Subsidies The absurd price we pay for data does not actually mean everyone gets high speed internet at home. Large swathes of the Qu bec countryside don't get broadband at all, and it can be difficult or expensive, even in large urban centers like Montr al, to get high speed internet. That is despite having a series of subsidies that all avoided investing in our own infrastructure. We had the "fonds de l'autoroute de l'information", "information highway fund" (site dead since 2003, archive.org link) and "branchez les familles", "connecting families" (site dead since 2003, archive.org link) which subsidized the development of a copper network. In 2014, more of the same: the federal government poured hundreds of millions of dollars into a program called connecting Canadians to connect 280 000 households to "high speed internet". And now, the federal and provincial governments are proudly announcing that "everyone is now connected to high speed internet", after pouring more than 1.1 billion dollars to connect, guess what, another 380 000 homes, right in time for the provincial election. Of course, technically, the deadline won't actually be met until 2023. Qu bec is a big area to cover, and you can guess what happens next: the telcos threw up their hand and said some areas just can't be connected. (Or they connect their CEO but not the poor folks across the lake.) The story then takes the predictable twist of giving more money out to billionaires, subsidizing now Musk's Starlink system to connect those remote areas. To give a concrete example: a friend who lives about 1000km away from Montr al, 4km from a small, 2500 habitant village, has recently got symmetric 100 mbps fiber at home from Telus, thanks to those subsidies. But I can't get that service in Montr al at all, presumably because Telus and Bell colluded to split that market. Bell doesn't provide me with such a service either: they tell me they have "fiber to my neighborhood", and only offer me a 25/10 mbps ADSL service. (There is Vid otron offering 400mbps, but that's copper cable, again a dead technology, and asymmetric.)

Conclusion Remember Chattanooga? Back in 2010, they funded the development of a fiber network, and now they have deployed a network roughly a thousand times faster than what we have just funded with a billion dollars. In 2010, I was paying Bell Canada 60$/mth for 20mbps and a 125GB cap, and now, I'm still (indirectly) paying Bell for roughly the same speed (25mbps). Back then, Bell was throttling their competitors networks until 2009, when they were forced by the CRTC to stop throttling. Both Bell and Vid otron still explicitly forbid you from running your own servers at home, Vid otron charges prohibitive prices which make it near impossible for resellers to sell uncapped services. Those companies are not spurring innovation: they are blocking it. We have spent all this money for the private sector to build us a private internet, over decades, without any assurance of quality, equity or reliability. And while in some locations, ISPs did deploy fiber to the home, they certainly didn't upgrade their entire network to follow suit, and even less allowed resellers to compete on that network. In 10 years, when 100mbps will be laughable, I bet those service providers will again punt the ball in the public courtyard and tell us they don't have the money to upgrade everyone's equipment. We got screwed. It's time to try something new.

Updates There was a discussion about this article on Hacker News which was surprisingly productive. Trigger warning: Hacker News is kind of right-wing, in case you didn't know. Since this article was written, at least two more major acquisitions happened, just in Qu bec: In the latter case, vMedia was explicitly saying it couldn't grow because of "lack of access to capital". So basically, we have given those companies a billion dollars, and they are not using that very money to buy out their competition. At least we could have given that money to small players to even out the playing field. But this is not how that works at all. Also, in a bizarre twist, an "analyst" believes the acquisition is likely to help Rogers acquire Shaw. Also, since this article was written, the Washington Post published a review of a book bringing similar ideas: Internet for the People The Fight for Our Digital Future, by Ben Tarnoff, at Verso books. It's short, but even more ambitious than what I am suggesting in this article, arguing that all big tech companies should be broken up and better regulated:
He pulls from Ethan Zuckerman s idea of a web that is plural in purpose that just as pool halls, libraries and churches each have different norms, purposes and designs, so too should different places on the internet. To achieve this, Tarnoff wants governments to pass laws that would make the big platforms unprofitable and, in their place, fund small-scale, local experiments in social media design. Instead of having platforms ruled by engagement-maximizing algorithms, Tarnoff imagines public platforms run by local librarians that include content from public media.
(Links mine: the Washington Post obviously prefers to not link to the real web, and instead doesn't link to Zuckerman's site all and suggests Amazon for the book, in a cynical example.) And in another example of how the private sector has failed us, there was recently a fluke in the AMBER alert system where the entire province was warned about a loose shooter in Saint-Elz ar except the people in the town, because they have spotty cell phone coverage. In other words, millions of people received a strongly toned, "life-threatening", alert for a city sometimes hours away, except the people most vulnerable to the alert. Not missing a beat, the CAQ party is promising more of the same medicine again and giving more money to telcos to fix the problem, suggesting to spend three billion dollars in private infrastructure.

18 August 2022

Lukas M rdian: Netplan v0.105 is now available

I m happy to announce that Netplan version 0.105 is now available on GitHub and is soon to be deployed into an Ubuntu/Debian installation near you! Six month and exactly 100 commits after the previous version, this release is brought to you by 7 free software contributors from around the globe. Changelog Bug fixes

29 June 2022

Aigars Mahinovs: Long travel in an electric car

Since the first week of April 2022 I have (finally!) changed my company car from a plug-in hybrid to a fully electic car. My new ride, for the next two years, is a BMW i4 M50 in Aventurine Red metallic. An ellegant car with very deep and memorable color, insanely powerful (544 hp/795 Nm), sub-4 second 0-100 km/h, large 84 kWh battery (80 kWh usable), charging up to 210 kW, top speed of 225 km/h and also very efficient (which came out best in this trip) with WLTP range of 510 km and EVDB real range of 435 km. The car also has performance tyres (Hankook Ventus S1 evo3 245/45R18 100Y XL in front and 255/45R18 103Y XL in rear all at recommended 2.5 bar) that have reduced efficiency. So I wanted to document and describe how was it for me to travel ~2000 km (one way) with this, electric, car from south of Germany to north of Latvia. I have done this trip many times before since I live in Germany now and travel back to my relatives in Latvia 1-2 times per year. This was the first time I made this trip in an electric car. And as this trip includes both travelling in Germany (where BEV infrastructure is best in the world) and across Eastern/Northen Europe, I believe that this can be interesting to a few people out there. Normally when I travelled this trip with a gasoline/diesel car I would normally drive for two days with an intermediate stop somewhere around Warsaw with about 12 hours of travel time in each day. This would normally include a couple bathroom stops in each day, at least one longer lunch stop and 3-4 refueling stops on top of that. Normally this would use at least 6 liters of fuel per 100 km on average with total usage of about 270 liters for the whole trip (or about 540 just in fuel costs, nowadays). My (personal) quirk is that both fuel and recharging of my (business) car inside Germany is actually paid by my employer, so it is useful for me to charge up (or fill up) at the last station in Gemany before driving on. The plan for this trip was made in a similar way as when travelling with a gasoline car: travelling as fast as possible on German Autobahn network to last chargin stop on the A4 near G rlitz, there charging up as much as reasonable and then travelling to a hotel in Warsaw, charging there overnight and travelling north towards Ionity chargers in Lithuania from where reaching the final target in north of Latvia should be possible. How did this plan meet the reality? Travelling inside Germany with an electric car was basically perfect. The most efficient way would involve driving fast and hard with top speed of even 180 km/h (where possible due to speed limits and traffic). BMW i4 is very efficient at high speeds with consumption maxing out at 28 kWh/100km when you actually drive at this speed all the time. In real situation in this trip we saw consumption of 20.8-22.2 kWh/100km in the first legs of the trip. The more traffic there is, the more speed limits and roadworks, the lower is the average speed and also the lower the consumption. With this kind of consumption we could comfortably drive 2 hours as fast as we could and then pick any fast charger along the route and in 26 minutes at a charger (50 kWh charged total) we'd be ready to drive for another 2 hours. This lines up very well with recommended rest stops for biological reasons (bathroom, water or coffee, a bit of movement to get blood circulating) and very close to what I had to do anyway with a gasoline car. With a gasoline car I had to refuel first, then park, then go to bathroom and so on. With an electric car I can do all of that while the car is charging and in the end the total time for a stop is very similar. Also not that there was a crazy heat wave going on and temperature outside was at about 34C minimum the whole day and hitting 40C at one point of the trip, so a lot of power was used for cooling. The car has a heat pump standard, but it still was working hard to keep us cool in the sun. The car was able to plan a charging route with all the charging stops required and had all the good options (like multiple intermediate stops) that many other cars (hi Tesla) and mobile apps (hi Google and Apple) do not have yet. There are a couple bugs with charging route and display of current route guidance, those are already fixed and will be delivered with over the air update with July 2022 update. Another good alterantive is the ABRP (A Better Route Planner) that was specifically designed for electric car routing along the best route for charging. Most phone apps (like Google Maps) have no idea about your specific electric car - it has no idea about the battery capacity, charging curve and is missing key live data as well - what is the current consumption and remaining energy in the battery. ABRP is different - it has data and profiles for almost all electric cars and can also be linked to live vehicle data, either via a OBD dongle or via a new Tronity cloud service. Tronity reads data from vehicle-specific cloud service, such as MyBMW service, saves it, tracks history and also re-transmits it to ABRP for live navigation planning. ABRP allows for options and settings that no car or app offers, for example, saying that you want to stop at a particular place for an hour or until battery is charged to 90%, or saying that you have specific charging cards and would only want to stop at chargers that support those. Both the car and the ABRP also support alternate routes even with multiple intermediate stops. In comparison, route planning by Google Maps or Apple Maps or Waze or even Tesla does not really come close. After charging up in the last German fast charger, a more interesting part of the trip started. In Poland the density of high performance chargers (HPC) is much lower than in Germany. There are many chargers (west of Warsaw), but vast majority of them are (relatively) slow 50kW chargers. And that is a difference between putting 50kWh into the car in 23-26 minutes or in 60 minutes. It does not seem too much, but the key bit here is that for 20 minutes there is easy to find stuff that should be done anyway, but after that you are done and you are just waiting for the car and if that takes 4 more minutes or 40 more minutes is a big, perceptual, difference. So using HPC is much, much preferable. So we put in the Ionity charger near Lodz as our intermediate target and the car suggested an intermediate stop at a Greenway charger by Katy Wroclawskie. The location is a bit weird - it has 4 charging stations with 150 kW each. The weird bits are that each station has two CCS connectors, but only one parking place (and the connectors share power, so if two cars were to connect, each would get half power). Also from the front of the location one can only see two stations, the otehr two are semi-hidden around a corner. We actually missed them on the way to Latvia and one person actually waited for the charger behind us for about 10 minutes. We only discovered the other two stations on the way back. With slower speeds in Poland the consumption goes down to 18 kWh/100km which translates to now up to 3 hours driving between stops. At the end of the first day we drove istarting from Ulm from 9:30 in the morning until about 23:00 in the evening with total distance of about 1100 km, 5 charging stops, starting with 92% battery, charging for 26 min (50 kWh), 33 min (57 kWh + lunch), 17 min (23 kWh), 12 min (17 kWh) and 13 min (37 kW). In the last two chargers you can see the difference between a good and fast 150 kW charger at high battery charge level and a really fast Ionity charger at low battery charge level, which makes charging faster still. Arriving to hotel with 23% of battery. Overnight the car charged from a Porsche Destination Charger to 87% (57 kWh). That was a bit less than I would expect from a full power 11kW charger, but good enough. Hotels should really install 11kW Type2 chargers for their guests, it is a really significant bonus that drives more clients to you. The road between Warsaw and Kaunas is the most difficult part of the trip for both driving itself and also for charging. For driving the problem is that there will be a new highway going from Warsaw to Lithuanian border, but it is actually not fully ready yet. So parts of the way one drives on the new, great and wide highway and parts of the way one drives on temporary roads or on old single lane undivided roads. And the most annoying part is navigating between parts as signs are not always clear and the maps are either too old or too new. Some maps do not have the new roads and others have on the roads that have not been actually build or opened to traffic yet. It's really easy to loose ones way and take a significant detour. As far as charging goes, basically there is only the slow 50 kW chargers between Warsaw and Kaunas (for now). We chose to charge on the last charger in Poland, by Suwalki Kaufland. That was not a good idea - there is only one 50 kW CCS and many people decide the same, so there can be a wait. We had to wait 17 minutes before we could charge for 30 more minutes just to get 18 kWh into the battery. Not the best use of time. On the way back we chose a different charger in Lomza where would have a relaxed dinner while the car was charging. That was far more relaxing and a better use of time. We also tried charging at an Orlen charger that was not recommended by our car and we found out why. Unlike all other chargers during our entire trip, this charger did not accept our universal BMW Charging RFID card. Instead it demanded that we download their own Orlen app and register there. The app is only available in some countries (and not in others) and on iPhone it is only available in Polish. That is a bad exception to the rule and a bad example. This is also how most charging works in USA. Here in Europe that is not normal. The normal is to use a charging card - either provided from the car maker or from another supplier (like PlugSufring or Maingau Energy). The providers then make roaming arrangements with all the charging networks, so the cards just work everywhere. In the end the user gets the prices and the bills from their card provider as a single monthly bill. This also saves all any credit card charges for the user. Having a clear, separate RFID card also means that one can easily choose how to pay for each charging session. For example, I have a corporate RFID card that my company pays for (for charging in Germany) and a private BMW Charging card that I am paying myself for (for charging abroad). Having the car itself authenticate direct with the charger (like Tesla does) removes the option to choose how to pay. Having each charge network have to use their own app or token bring too much chaos and takes too much setup. The optimum is having one card that works everywhere and having the option to have additional card or cards for specific purposes. Reaching Ionity chargers in Lithuania is again a breath of fresh air - 20-24 minutes to charge 50 kWh is as expected. One can charge on the first Ionity just enough to reach the next one and then on the second charger one can charge up enough to either reach the Ionity charger in Adazi or the final target in Latvia. There is a huge number of CSDD (Road Traffic and Safety Directorate) managed chargers all over Latvia, but they are 50 kW chargers. Good enough for local travel, but not great for long distance trips. BMW i4 charges at over 50 kW on a HPC even at over 90% battery state of charge (SoC). This means that it is always faster to charge up in a HPC than in a 50 kW charger, if that is at all possible. We also tested the CSDD chargers - they worked without any issues. One could pay with the BMW Charging RFID card, one could use the CSDD e-mobi app or token and one could also use Mobilly - an app that you can use in Latvia for everything from parking to public transport tickets or museums or car washes. We managed to reach our final destination near Aluksne with 17% range remaining after just 3 charging stops: 17+30 min (18 kWh), 24 min (48 kWh), 28 min (36 kWh). Last stop we charged to 90% which took a few extra minutes that would have been optimal. For travel around in Latvia we were charging at our target farmhouse from a normal 3 kW Schuko EU socket. That is very slow. We charged for 33 hours and went from 17% to 94%, so not really full. That was perfectly fine for our purposes. We easily reached Riga, drove to the sea and then back to Aluksne with 8% still in reserve and started charging again for the next trip. If it were required to drive around more and charge faster, we could have used the normal 3-phase 440V connection in the farmhouse to have a red CEE 16A plug installed (same as people use for welders). BMW i4 comes standard with a new BMW Flexible Fast Charger that has changable socket adapters. It comes by default with a Schucko connector in Europe, but for 90 one can buy an adapter for blue CEE plug (3.7 kW) or red CEE 16A or 32A plugs (11 kW). Some public charging stations in France actually use the blue CEE plugs instead of more common Type2 electric car charging stations. The CEE plugs are also common in camping parking places. On the way back the long distance BEV travel was already well understood and did not cause us any problem. From our destination we could easily reach the first Ionity in Lithuania, on the Panevezhis bypass road where in just 8 minutes we got 19 kWh and were ready to drive on to Kaunas, there a longer 32 minute stop before the charging desert of Suwalki Gap that gave us 52 kWh to 90%. That brought us to a shopping mall in Lomzha where we had some food and charged up 39 kWh in lazy 50 minutes. That was enough to bring us to our return hotel for the night - Hotel 500W in Strykow by Lodz that has a 50kW charger on site, while we were having late dinner and preparing for sleep, the car easily recharged to full (71 kWh in 95 minutes), so I just moved it from charger to a parking spot just before going to sleep. Really easy and well flowing day. Second day back went even better as we just needed an 18 minute stop at the same Katy Wroclawskie charger as before to get 22 kWh and that was enough to get back to Germany. After that we were again flying on the Autobahn and charging as needed, 15 min (31 kWh), 23 min (48 kWh) and 31 min (54 kWh + food). We started the day on about 9:40 and were home at 21:40 after driving just over 1000 km on that day. So less than 12 hours for 1000 km travelled, including all charging, bio stops, food and some traffic jams as well. Not bad. Now let's take a look at all the apps and data connections that a technically minded customer can have for their car. Architecturally the car is a network of computers by itself, but it is very secured and normally people do not have any direct access. However, once you log in into the car with your BMW account the car gets your profile info and preferences (seat settings, navigation favorites, ...) and the car then also can start sending information to the BMW backend about its status. This information is then available to the user over multiple different channels. There is no separate channel for each of those data flow. The data only goes once to the backend and then all other communication of apps happens with the backend. First of all the MyBMW app. This is the go-to for everything about the car - seeing its current status and location (when not driving), sending commands to the car (lock, unlock, flash lights, pre-condition, ...) and also monitor and control charging processes. You can also plan a route or destination in the app in advance and then just send it over to the car so it already knows where to drive to when you get to the car. This can also integrate with calendar entries, if you have locations for appointments, for example. This also shows full charging history and allows a very easy export of that data, here I exported all charging sessions from June and then trimmed it back to only sessions relevant to the trip and cut off some design elements to have the data more visible. So one can very easily see when and where we were charging, how much power we got at each spot and (if you set prices for locations) can even show costs. I've already mentioned the Tronity service and its ABRP integration, but it also saves the information that it gets from the car and gathers that data over time. It has nice aspects, like showing the driven routes on a map, having ways to do business trip accounting and having good calendar view. Sadly it does not correctly capture the data for charging sessions (the amounts are incorrect). Update: after talking to Tronity support, it looks like the bug was in the incorrect value for the usable battery capacity for my car. They will look into getting th eright values there by default, but as a workaround one can edit their car in their system (after at least one charging session) and directly set the expected battery capacity (usable) in the car properties on the Tronity web portal settings. One other fun way to see data from your BMW is using the BMW integration in Home Assistant. This brings the car as a device in your own smart home. You can read all the variables from the car current status (and Home Asisstant makes cute historical charts) and you can even see interesting trends, for example for remaining range shows much higher value in Latvia as its prediction is adapted to Latvian road speeds and during the trip it adapts to Polish and then to German road speeds and thus to higher consumption and thus lower maximum predicted remaining range. Having the car attached to the Home Assistant also allows you to attach the car to automations, both as data and event source (like detecting when car enters the "Home" zone) and also as target, so you could flash car lights or even unlock or lock it when certain conditions are met. So, what in the end was the most important thing - cost of the trip? In total we charged up 863 kWh, so that would normally cost one about 290 , which is close to half what this trip would have costed with a gasoline car. Out of that 279 kWh in Germany (paid by my employer) and 154 kWh in the farmhouse (paid by our wonderful relatives :D) so in the end the charging that I actually need to pay adds up to 430 kWh or about 150 . Typically, it took about 400 in fuel that I had to pay to get to Latvia and back. The difference is really nice! In the end I believe that there are three different ways of charging:
  • incidental charging - this is wast majority of charging in the normal day-to-day life. The car gets charged when and where it is convinient to do so along the way. If we go to a movie or a shop and there is a chance to leave the car at a charger, then it can charge up. Works really well, does not take extra time for charging from us.
  • fast charging - charging up at a HPC during optimal charging conditions - from relatively low level to no more than 70-80% while you are still doing all the normal things one would do in a quick stop in a long travel process: bio things, cleaning the windscreen, getting a coffee or a snack.
  • necessary charging - charging from a whatever charger is available just enough to be able to reach the next destination or the next fast charger.
The last category is the only one that is really annoying and should be avoided at all costs. Even by shifting your plans so that you find something else useful to do while necessary charging is happening and thus, at least partially, shifting it over to incidental charging category. Then you are no longer just waiting for the car, you are doing something else and the car magically is charged up again. And when one does that, then travelling with an electric car becomes no more annoying than travelling with a gasoline car. Having more breaks in a trip is a good thing and makes the trips actually easier and less stressfull - I was more relaxed during and after this trip than during previous trips. Having the car air conditioning always be on, even when stopped, was a godsend in the insane heat wave of 30C-38C that we were driving trough. Final stats: 4425 km driven in the trip. Average consumption: 18.7 kWh/100km. Time driving: 2 days and 3 hours. Car regened 152 kWh. Charging stations recharged 863 kWh. Questions? You can use this i4talk forum thread or this Twitter thread to ask them to me.

22 June 2022

John Goerzen: I Finally Found a Solid Debian Tablet: The Surface Go 2

I have been looking for a good tablet for Debian for well, years. I want thin, light, portable, excellent battery life, and a servicable keyboard. For a while, I tried a Lenovo Chromebook Duet. It meets the hardware requirements, well sort of. The problem is with performance and the OS. I can run Debian inside the ChromeOS Linux environment. That works, actually pretty well. But it is slow. Terribly, terribly, terribly slow. Emacs takes minutes to launch. apt-gets also do. It has barely enough RAM to keep its Chrome foundation happy, let alone a Linux environment also. But basically it is too slow to be servicable. Not just that, but I ran into assorted issues with having it tied to a Google account particularly being unable to login unless I had Internet access after an update. That and my growing concern over Google s privacy practices led me sort of write it off. I have a wonderful System76 Lemur Pro that I m very happy with. Plenty of RAM, a good compromise size between portability and screen size at 14.1 , and so forth. But a 10 goes-anywhere it s not. I spent quite a lot of time looking at thin-and-light convertible laptops of various configurations. Many of them were quite expensive, not as small as I wanted, or had dubious Linux support. To my surprise, I wound up buying a Surface Go 2 from the Microsoft store, along with the Type Cover. They had a pretty good deal on it since the Surface Go 3 is out; the highest-processor model of the Go 2 is roughly similar to the Go 3 in terms of performance. There is an excellent linux-surface project out there that provides very good support for most Surface devices, including the Go 2 and 3. I put Debian on it. I had a fair bit of hassle with EFI, and wound up putting rEFInd on it, which mostly solved those problems. (I did keep a Windows partition, and if it comes up for some reason, the easiest way to get it back to Debian is to use the Windows settings tool to reboot into advanced mode, and then select the appropriate EFI entry to boot from there.) Researching on-screen keyboards, it seemed like Gnome had the most mature. So I wound up with Gnome (my other systems are using KDE with tiling, but I figured I d try Gnome on it.) Almost everything worked without additional tweaking, the one exception being the cameras. The cameras on the Surfaces are a known point of trouble and I didn t bother to go to all the effort to get them working. With 8GB of RAM, I didn t put ZFS on it like I do on other systems. Performance is quite satisfactory, including for Rust development. Battery life runs about 10 hours with light use; less when running a lot of cargo builds, of course. The 1920 1280 screen is nice at 10.5 . Gnome with Wayland does a decent job of adjusting to this hi-res configuration. I took this as my only computer for a trip from the USA to Germany. It was a little small at times; though that was to be expected. It let me take a nicely small bag as a carryon, and being light, it was pleasant to carry around in airports. It served its purpose quite well. One downside is that it can t be powered by a phone charger like my Chromebook Duet can. However, I found a nice slim 65W Anker charger that could charge it and phones simultaneously that did the job well enough (I left the Microsoft charger with the proprietary connector at home). The Surface Go 2 maxes out at a 128GB SSD. That feels a bit constraining, especially since I kept Windows around. However, it also has a micro SD slot, so you can put LUKS and ext4 on that and use it as another filesystem. I popped a micro SD I had lying around into there and that felt a lot better storage-wise. I could also completely zap Windows, but that would leave no way to get firmware updates and I didn t really want to do that. Still, I don t use Windows and that could be an option also. All in all, I m pretty pleased with it. Around $600 for a fully-functional Debian tablet, with a keyboard is pretty nice. I had been hoping for months that the Pinetab would come back into stock, because I d much rather support a Linux hardware vendor, but for now I think the Surface Go series is the most solid option for a Linux tablet.

8 June 2022

Laura Arjona Reina: Moving to a faster but smaller disk, encrypted setup

My work computer runs Debian 11 bullseye (the current stable release) in a mechanical 500GB disk, and I was provided with a new SDD disk but its size was 480 GB. So I had to shrink my partitions before copying the data to the new disk. It turned out to be a bit difficult because my main partition was encrypted. I write here how I did, maybe there are other simpler ways but I couldn t find them. References: I had three partitions in my old 500GB disk: /dev/sda1 is the EFI partition, /dev/sda2 the boot partition and /dev/sda3 the root partition (encrypted, with LVM, the standard way the Debian installer proposes when you choose a simple encrypted setup). First of all, I made a disk image with Clonezilla to an external USB disk, just in case I mess up things, to be able to return to a safe point and start again. Then I started my computer with a Debian 11 live USB with KDE Plasma desktop and Spanish localisation environment. I opened the KDE Partition manager and copied the non encrypted partitions (sda1, EFI and sda2, /boot) to the new disk. I shrinked the encrypted partition from the terminal with the following commands (I had enough free space so reduced my partition to a total of 300GB): Removed the swap partition and re-created it:
sudo lvremove /dev/larjona-pc-vg/swap_1
sudo pvresize --setphysicalvolumesize 380G /dev/mapper/cryptdisk
sudo pvchange -x y /dev/mapper/cryptdisk
sudo lvcreate -L 4G -n swap_1 larjona-pc-vg
sudo mkswap -L swap_1 /dev/larjona-pc-vg/swap_1
Display information about the physical volume in order to shrink it:
sudo pvs -v --segments --units s /dev/mapper/cryptdisk
sudo cryptsetup -b 838860800 resize cryptdisk
sudo cryptsetup status cryptdisk
sudo vgchange -a n vgroup
sudo vgchange -an
sudo cryptsetup luksClose cryptdisk
Then reduced the sda3 partition with the KDE partition manager (it took a while), and copy it to the new disk. Turned off the computer and unplugged the old disk. Started the computer with the Debian 11 Live USB again, UEFI boot. Now, to make my system boot:
sudo cryptsetup luksOpen /dev/sda3 crypdisk
sudo vgscan --mknodes
sudo vgchange -ay
sudo mount /dev/mapper/larjona--pc--vg-root /mnt
sudo mount /dev/sda2 /mnt/boot
sudo mount /dev/sda1 /mnt/boot/efi
mount --rbind /sys /media/linux/sys
mount -t efivarfs none /sys/firmware/efi/efivars
for i in /dev /dev/pts /proc /run; do sudo mount -B $i /mnt$i; done
sudo chroot /mnt
Then edited /mnt/etc/crypttab to reflect the name of the new encrypted partition, edited /mnt/etc/fstab to paste the UUIDs of the new partitions.
Then ran grub-install and reinstalled the kernels as noted in the reference, rebooted and logged in my Plasma desktop  (Well, the actual process was not so smooth but after several tries and errors and searching for help I managed to get the needed commands to make my system boot from the new disk).

Next.

Previous.