Get ready for some cross-distro Linux synesthesia
Installing Ubuntu to Thin-lvm in a Very Arch-Like Way
This is a way you can install a Debian-like OS on thin-lvm. It’s 2026, and installers I’ve tried still don’t have options for thin-lvm, as far as I’m aware. Therefore, I usually install Debian-like OS using debootstrap, so I have both benefits of copy-on-write (CoW) at the volume level, while still enjoying the lower latency of using a journaling filesystem (ext4, xfs, etc.) as opposed to CoW filesystems (btrfs, zfs).
Most people who have used Proxmox will notice ext4 on thin-lvm is the default installation configuration. If CoW filesystems are universally higher latency, ext4 on thin-lvm is a great way to overcome each one’s deficiencies. Proxmox’s other recommended filesystem is zfs, which mitigates slow latency to some degree by implementing an ARC, but for a VM host (hypervisor), using valuable RAM for VMs is less ideal than not needing a cache to improve IOPS in the first place, especially for lower-end setups such as edge deployments, IoT servers, homelabs, etc. Therefore, the most straightforward way to get benefits of both CoW and journaled FS is using thin-lvm with ext4 or xfs.
Throughput + latency references:
https://chronicle.software/linux-file-systems-and-application-latency/
https://www.dimoulis.net/posts/benchmark-of-postgresql-with-ext4-xfs-btrfs-zfs/
On the flip side, some of us just prefer installing our OS the way Arch compels its users to install Arch – it allows for a lot more customization and granular control. In my case, I got used to installing distros the way I installed Arch after installing it over and over again so many times, and moreover, automated installer programs tended to lack the ability to install features I wanted. The only downside compared to using an installer to the Arch Install Scripts or debootstrap install methods, besides simplicity, is having to understand what it is you’re doing.
Disclaimer: Remember your device names, such as /dev/mapper/vg-lv should reflect the names for volume group and logical volumes you decide to use. They’re unlikely to be exactly like mine, unless you choose to use them exactly.
Another blooger named Ansemjo published his method for bootstrapping Ubuntu with debootstrap to use LUKS – it was my main source for adaptation, so I wanted to give him a shout out (thanks, Ansemjo!).
My adaptation I provide here is more simple (aka less difficult to implement), and specific to newer versions of Debian and Ubuntu, but does not include options for encryption. If you’re interested in adding a LUKS volume and encrypting your drive, I definitely recommend checking out Ansemjo’s original instructions here: https://semjonov.de/posts/2021-09/minimal-ubuntu-installation-with-debootstrap/
Let’s get started.
The easiest thing to do is use an installer ISO from Ubuntu or Debian live ISO, where one can use the operating system straight from a USB flash drive. Boot from the flash drive and open a terminal, which in Ubuntu is ctrl+alt+T
Enter sudo su so you can operate with root permissions, and install arch-install-scripts and debootstrap (updating or upgrading the live ISO’s packages not necessary, and can eventually lead to buffer overflow errors, making the live environment inoperable – we want to make sure the OS we’re installing is up-to-date, not the live ISO)
# open terminal
ctrl-alt-T
# become root
sudo su
# install the two packages we need
apt install -y arch-install-scripts debootstrapPresume we have a Samsung PM981a SSD with 256GB advertised available storage (around 230GB in reality). First, wipe it and use fdisk to make it a GPT volume, and create three partitions (EFI, BOOT, and LVM) (be sure to replace /dev/nvme0n1 with whichever device you’re using)
# CAREFUL, THIS WIPES THE ENTIRE DISK
wipefs -a /dev/nvme0n1
[... wipefs output ...]Then, follow this process while you’re inside fdisk:
fdisk /dev/nvme0n1
# in fdisk, create GPT partition structure:
g, w [enter]
# open fdisk again, and create partitions:
fdisk /dev/nvme0n1
# a 1GB EFI partition which will be /dev/nvme0n1p1
n, [enter], +1G [enter]
# set to EFI type:
t, [enter], 1
# a 2GB XBOOTLDR partition /dev/nvme0n1p2
n, [enter], +2G [enter]
t, 2 (or [enter]), 142 # extended boot partition
# then we use the rest for LVM, since we can create a swap partition that won't automatically take up its entire allocated space
n, [enter], [enter] # for the rest of the space
t, 3 (or [enter]), 44 # lvm partition
# then write changes and exit
w [enter]You’ll need to mark that 3rd partition to be an LVM volume. Start that process by configuring it as a physical volume:
pv /dev/nvme0n1p3Then you need a volume group – you can name it whatever you like. Numbers can help with collisions if you plan to have more than one (if I’m only planning to have one per system, I don’t worry about it):
# [vg command] [physical volume] [volume group name]
vg /dev/nvme0n1p3 vgThen, I like to separate the thinpool creation step from creating any logical volumes. You can double them up, but I plan on creating a few, so I prefer the 2-step granularity:
# [lv] [type] [% allocated] [pool options] [vg name]
lvcreate --type thin-pool -l 90%FREE -Zn -c64 -n thinpool vgapt update && apt install -y arch-install-scripts
mount /dev/mrvg/mrrootlv /mnt
arch-chroot /mnt mount -aIf you want to check out the current layout, try running lsblk -f (especially nice with the alias I recommended if you’re on Ubuntu)
RE: the 90%FREE allocation, I like to leave some space as an insurance policy for the chance I commit more data to a volume than space it has been allocated (a bad situation). You’re welcome to make it 100%FREE if you’re more confident (less careful).
The thin-pool is just a container, we won’t actually be writing to it directly. Inside the pool are the actual volumes we’re going to use – let’s create those:
lvcreate -V 80G -n root vg/thinpool
lvcreate -V 40G -n home vg/thinpool
lvcreate -V 40G -n containers vg/thinpool
lvcreate -V 32G -n swap vg/thinpool You can probably see right away this is getting awfully close to the physical space we’ve allocated for our pool from our ~230GB drive, but don’t worry – as long as they’re not full, they won’t be automatically taking up all the space we’re assigning to them. The beauty of using thin LVM is it automatically implements sparse volumes. Just be careful and don’t go too crazy with overcommitting. I personally like to overcommit swap since it’s rarely used, but if you need it for a crashdump log, it’ll be there if when necessary (rule of thumb is to create a swap the same size as total system RAM).
Example of lsblk with alias (removes iso and loop devices):
# use alias to reduce extraneous lsblk output:
alias lsblk="lsblk -f | grep -v -E 'loop|sr0'"
lsblk
NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINTS
nvme0n1 253:0 0 232G 0 disk
├─nvme0n1p1 253:1 0 1G 0 part /efi
├─nvme0n1p2 253:2 0 2G 0 part /boot
└─nvme0n1p3 253:3 0 228.6G 0 part
├─vg-thinpool_tmeta 252:0 0 24M 0 lvm
│ └─vg-thinpool-tpool 252:2 0 20.8G 0 lvm
│ ├─vg-thinpool 252:3 0 198.8G 1 lvm
│ ├─vg-swap 252:4 0 32G 0 lvm
│ └─vg-root 252:5 0 40G 0 lvm
│ └─vg-home 252:5 0 80G 0 lvm
│ └─vg-containers 252:5 0 40G 0 lvm
└─vg-thinpool_tdata 252:1 0 198.8G 0 lvm
└─vg-thinpool-tpool 252:2 0 20.8G 0 lvm
├─vg-thinpool 252:3 0 198.8G 1 lvm
├─vg-swap 252:4 0 32G 0 lvm
└─vg-root 252:5 0 40G 0 lvm
└─vg-home 252:5 0 80G 0 lvm
└─vg-containers 252:5 0 40G 0 lvm
# Note: These values are calculated as example
# not actual values taken from running machineCreate filesystems:
# first, create FS on raw partitions
mkfs.msdos -F 32 -n EFI /dev/nvme0n1p1
mkfs.ext4 -L BOOT /dev/nvme0n1p2
# then, create FS on logical volumes
mkswap -v -L SWAP /dev/vg/swap
# ext4 is nice since can be shrunk if necessary
mkfs.ext4 -L ROOT /dev/vg/root
mkfs.ext4 -L HOME /dev/vg/home
# I like XFS for docker's overlay2 FS, but
# be aware that XFS never be shrunk!
mkfs.xfs -L CONTAINERS /dev/vg/containersHere’s an example of some Ubuntu meta-packages on 26.04, for instance ubuntu-server (TUI, no desktop), ubuntu-desktop, vanilla-gnome-desktop (my favorite) etc. etc. You can install more meta-packages using tasksel, so I actually like to start with the smallest one possible, making ubuntu-server the best pick in that regard (sometimes debootstrap craps out while installing packages):
apt search ubuntu-desktop
Sorting... Done
Full Text Search... Done
edubuntu-desktop/resolute 26.04 amd64
educational desktop for Ubuntu
edubuntu-desktop-minimal/resolute 26.04 amd64
educational desktop for Ubuntu
kubuntu-desktop/resolute 1.451 amd64
Kubuntu Plasma Desktop/Netbook system
lubuntu-desktop/resolute 24.04.10 amd64
Lubuntu Desktop environment
ubuntu-desktop/resolute 1.539 amd64
Ubuntu desktop system
ubuntu-desktop-minimal/resolute 1.539 amd64
Ubuntu desktop minimal system
xubuntu-desktop/resolute 2.262 amd64
Xubuntu desktop system
xubuntu-desktop-minimal/resolute 2.262 amd64
Xubuntu minimal systemMount FS structure and bootstrap:
# mount the root volume to /mnt
mount /dev/vg/root /mnt
# create boot folder, mount boot partition nvme0n1p2
mkdir -pv /mnt/boot
mount -v /dev/nvme0n1p2 /mnt/boot
# create efi folder, mount efi partition nvme0n1p1
mkdir -pv /mnt/boot/efi
mount -v /dev/nvme0n1p1 /mnt/efi
# create home folder and mount home lv
mkdir -pv /mnt/home
mount -v /dev/vg/home /mnt/home
# this step optional
mkdir -pv /mnt/var/lib/docker
mount -v /dev/vg/containers /mnt/var/lib/docker
debootstrap --arch=amd64 \
--components=main,restricted,universe,multiverse \
--include=systemd-boot,systemd-boot-efi,systemd-boot-tools,arch-install-scripts,debootstrap,ubuntu-server,vim-scripts,vim-airline-themes,git,efibootmgr,bash-completion,thin-provisioning-tools,apt-file \
--keyring=/usr/share/keyrings/ubuntu-archive-keyring.gpg \
resolute /mnt https://mirror.leaseweb.com/ubuntu/If you’re using the Ubuntu 26.04 live ISO, you can copy the host’s apt sources file to your new filesystem:
cp /etc/apt/sources.list.d/ubuntu.sources /mnt/etc/apt/sources.list.dCreate /mnt/etc/kernel/cmdline with a /dev/mapper appropriate for you (note: it’s more reliable to use /dev/mapper/vg-lv for initramfs than /dev/vg/lv):
echo 'root=/dev/mapper/vg-root ro loglevel=6 intel_iommu=1 iommu=pt kvm.ignore_msrs=1 crashkernel=1024M' > /mnt/etc/kernel/cmdlineMount the swap before creating fstab – here’s an example using the partition LABEL instead of the device path:
arch-chroot /mnt swapon -v -L SWAPCreate your fstab. genfstab -t PARTUUID is nice because it should contain all the currently mounted filesystems, but additionally will replace PARTUUID with UUID or device path where appropriate (e.g. EFI partition will automatically use UUID, and logical volumes will use device mapper paths):
genfstab -t PARTUUID /mnt > /mnt/etc/fstabDelete the /mnt from the paths in the new fstab:
sed -i 's|/mnt||g' /mnt/etc/fstabHere’s a finished example – view
using cat /mnt/etc/fstab:
# <file system> <mount point> <type> <options> <dump> <pass>
# /dev/mapper/mrvg-mrrootlv UUID=37374f8c-a4bf-4577-96c0-1cd8a9fbe114 LABEL=MR_ROOTLV
/dev/mapper/vg-root / ext4 rw,relatime 0 1
# /dev/vda1 PARTUUID=a5a6570a-ea64-4480-b6d7-fb3f6a504cbf LABEL=MR_EFI
UUID=C615-2241 /efi vfat rw,relatime,fmask=0022,dmask=0022,codepage=437,iocharset=iso8859-1,shortname=mixed,errors=remount-ro 0 2
# /dev/vda2 UUID=2a466a62-2a00-4a2e-b8aa-734d71039d67 LABEL=MR_XBOOTLDR
PARTUUID=03b8e7e8-a010-4d06-b09c-5503b609d022 /boot ext4 rw,relatime 0 2
/dev/mapper/vg-home /home ext4 rw,relatime 0 1
/dev/mapper/vg-containers /var/lib/docker xfs rw,relatime 0 1
# /dev/dm-4 UUID=fb30403b-2101-4cf9-a083-a5debd8ad60a LABEL=MR_SWAPLV
/dev/mapper/vg-swap none swap defaults 0 0Important step (don’t miss!):
Try and run arch-chroot /mnt to log into your new filesystem.
If you get this error when trying to use arch-chroot:
arch-chroot /mnt
mount: /mnt/dev: udev already mounted on /dev.
dmesg(1) may have more information after failed mount system call.
==> ERROR: failed to setup chroot /mntSolution was to:
swapoff -a
umount /mnt/devTry arch-chroot /mnt again!bash-completion wasn’t initialized, so I set it up for root login (tab-completion can help with diagnoses):
sed -i 's|#if|if|g' $HOME/.bashrc
sed -i 's|#fi|fi|g' $HOME/.bashrc
sed -i 's|# . /etc/bash_completion| . /etc/bash_completion|g' $HOME/.bashrc
# if you also want to uncomment the dircolors aliases:
sed -i 's|#alias|alias|g' $HOME/.bashrcIt’s good to install the kernels while you’re chrooted into the new filesystem, but first, update dpkg‘s locale settings before installing any more packages (otherwise, lots of errors):
arch-chroot /mnt
locale-gen "en_US.UTF-8"
dpkg-reconfigure locales
(skip to OK, enter)
(select "en_US.UTF-8", OK, enter)
# update just in case - will show error
# if locale settings not resolved
apt update
apt upgrade -y Install kernel (and optional headers):
cp /usr/lib/kernel/install.conf /etc/kernel
apt install -y linux-image-generic linux-headers-genericMake sure you have dm_thin_pool modules loaded
lsmod | grep thin
dm_thin_pool 73728 0
dm_persistent_data 90112 1 dm_thin_pool
dm_bio_prison 20480 1 dm_thin_poolIf you don’t see an output like that, do this:
echo 'dm-thin-pool' > /etc/modules-load.d/dm-thin-pool.conf
# re-create your initramfs
update-initramfs -vuk allWhilst still in chroot, check and make sure /mnt/boot has your initrd with tree:
apt list --installed | grep linux-image
tree /boot
/boot
├── EFI
│ └── Linux
├── System.map-6.18.0-12-generic
├── $(cat /etc/machine-id)
│ └── 6.18.0-12-generic
│ ├── initrd.img-6.18.0-12-generic
│ └── linux
├── config-6.18.0-12-generic
├── grub
│ ├── gfxblacklist.txt
│ └── unicode.pf2
├── initrd.img -> initrd.img-6.18.0-12-generic
├── initrd.img-6.18.0-12-generic
├── loader
│ ├── entries
│ │ └── $(cat /etc/machine-id)-6.18.0-12-generic.conf
│ └── entries.srel
├── lost+found
├── vmlinuz -> vmlinuz-6.18.0-12-generic
├── vmlinuz-6.18.0-12-generic(Still in chroot) install the systemd bootloader – this is best done after linux-image-generic, since it’ll install GRUB’s BOOTx64.EFI over systemd-boot’s copy – bootctl update won’t work for this
bootctl --esp-path=/efi --boot-path=/boot installI’m not sure if you need the efifs packages (*.efi files) so systemd-boot can load your /boot partition in Ubuntu 26.04, but it’s probably a good idea to do this step just in case
Check the URLhttps://github.com/pbatard/efifs/releases first to make sure downloading latest ver, and change v1.10 to whatever that might be:
cd /efi/EFI/systemd/ && mkdir drivers && cd drivers
for EFIFS in affs afs bfs btrfs cbfs cpio_be cpio erofs exfat ext2 f2fs fat hfs hfsplus iso9660 jfs minix2_be minix2 minix3_be minix3 minix_be minix newc nilfs2 ntfs odc procfs reiserfs romfs sfs squash4 tar udf ufs1_be ufs1 ufs2 xfs zfs; do \
wget "https://github.com/pbatard/efifs/releases/download/v1.10/${EFIFS}_x64.efi"; done
cd /Create a user and a root password, give the user sudo group access:
# root
passwd
# user
useradd -m username
passwd username
usermod -aG sudo username
# default shell probably /bin/sh
chsh username -s /bin/bashYou can check to make sure sudo is working for your user
su username
sudo ls -la /
[sudo] password for avery:
(should output `ls -la /`)I’m thinking this system should boot now, so I’m going to power off, change the boot device to /dev/nvme0n1p1, and check it out
exit # chroot
poweroffTL;DR if you experienced boot failure, perhaps you’d not installed thin-provisioning-tools before building your initramfs images. If you did install it, and had no problems, feel free to skip this part unless you’re curious…
I didn’t include thin-provisioning-tools in one of the earlier lists of modules included with debootstrap, so my system didn’t boot. There is a file in thin-provisioning-tools that adds functionality to initramfs-tools package, namely the hook file /usr/share/initramfs-tools/hooks/thin-provisioning-tools that adds the dm-thin-pool module at the bottom of this example of the hook to be compiled along with the rest of the initrd. Here’s the hook:
# File: /usr/share/initramfs-tools/hooks/thin-provisioning-tools
#!/bin/sh
PREREQ=""
prereqs()
{
echo "$PREREQ"
}
case $1 in
prereqs)
prereqs
exit 0
;;
esac
. /usr/share/initramfs-tools/hook-functions
copy_exec /usr/sbin/pdata_tools
ln -s pdata_tools ${DESTDIR}/usr/sbin/cache_check
ln -s pdata_tools ${DESTDIR}/usr/sbin/thin_check
manual_add_modules dm-cache dm-cache-smq dm-thin-poolIf rebuilding your initrd after installing thin-provisioning-tools, run update-initramfs -vuk all and look for lines corresponding to the modules added at the bottom of /usr/share/initramfs-tools/hooks/thin-provisioning-tools in the output – that should be an indicator that required modules are being added now:
. . .
dracut-install: mkdir '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md'
dracut-install: cp '/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-cache.ko.zst' '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-cache.ko.zst'
dracut-install: cp '/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-bufio.ko.zst' '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-bufio.ko.zst'
dracut-install: cp '/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-bio-prison.ko.zst' '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-bio-prison.ko.zst'
dracut-install: mkdir '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md/persistent-data'
dracut-install: cp '/lib/modules/6.18.0-12-generic/kernel/drivers/md/persistent-data/dm-persistent-data.ko.zst' '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md/persistent-data/dm-persistent-data.ko.zst'
dracut-install: cp '/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-cache-smq.ko.zst' '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-cache-smq.ko.zst'
dracut-install: cp '/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-thin-pool.ko.zst' '/var/tmp/mkinitramfs_Y5NSco/lib/modules/6.18.0-12-generic/kernel/drivers/md/dm-thin-pool.ko.zst'
. . .A quick synopsis is that without these tools, initrd will lack required modules, but with it, initrd should boot fine. For a more lengthy explanation, see old workaround:
https://bugs.launchpad.net/ubuntu/+source/lvm2/+bug/1539934/comments/2 [comment 2]
https://askubuntu.com/questions/673815/how-do-i-start-my-laptop-with-root-partition-on-lvm2-thin-pool
Of course, none of this is really necessary if you install the thin-provisioning-tools package initially.
Last, but definitely not least, if you want to use snapper, you can set up a very minimal configuration for snapshots taken on each boot like this when you reboot into your new machine:
systemctl disable snapper-timeline.timer
systemctl enable snapper-boot.timer
systemctl enable snapper-cleanup.timer
snapper --no-dbus -c root create-config \
--fstype="lvm(ext4)" /
snapper --no-dbus set-config \
NUMBER_LIMIT=6 \
NUMBER_LIMIT_IMPORTANT=3 \
TIMELINE_CLEANUP=no \
TIMELINE_CREATE=no \
TIMELINE_LIMIT_DAILY=2 \
TIMELINE_LIMIT_HOURLY=1 \
TIMELINE_LIMIT_WEEKLY=3 \
TIMELINE_LIMIT_MONTHLY=5 \
TIMELINE_LIMIT_QUARTERLY=5 \
TIMELINE_LIMIT_YEARLY=5
snapper --no-dbus list-configsThat ought to do it! Hope this went well for you.
If you see anything missing, or have any information you’d like to share, especially detailing your experience trying an Archlinux-like install process with a Debian-like OS, please write the community in the comments section below!