DIY install debian on Oracle Cloud Infrastructure ( Free Tier ) - ARM64

reinstall_VPS_from_inside.txt

# should works on any cloud-init enabled hypervisor (openstack.. )

# start from a normal ubuntu 20.04 install as minimal was not available for ARM64
# Since ARM64 machines has higher RAM, Shrinking is desired but not necessary. Instead we will increase tmpfs to 1700MB

# Getting root (if sudo -i doesn't work then set a root password beforehand using 'sudo passwd root'

sudo -i

# make sure we are on the highest kernel, so we can delete all the others ...
sudo apt update && sudo apt dist-upgrade -y && sudo apt install lsof && sudo reboot

# snap removal didn't work so proceed without it.
sudo snap remove --purge  oracle-cloud-agent && sudo snap remove --purge core18
sudo apt purge -y $(dpkg-query -Wf '${Package}\n' | grep header)  $(apt list --installed | grep -oP "^linux.*\d\d\d\d-oracle" | grep -v "$(uname -r)") linux-modules-extra-$(uname -r) lxc* lxd* vim* && sudo apt -y autoremove && sudo apt -y autoclean && sudo apt -y clean  
sudo rm -rf /var/log/* /var/lib/apt/lists/*


# use df to check the size. It should be now ~1100MB. If it is higher than 1700 MB then increase the size below
# make sure you have root here


cd /
mount -t tmpfs -o size=1700m tmpfs mnt
tar --one-file-system -c . | tar -C /mnt -x
mount --make-private -o remount,rw /
mount --move dev mnt/dev
mount --move proc mnt/proc
mount --move run mnt/run
mount --move sys mnt/sys
sed -i '/^[^#]/d;' mnt/etc/fstab
echo 'tmpfs / tmpfs defaults 0 0' >> mnt/etc/fstab
cd mnt
mkdir old_root
mount --make-private /
unshare -m
pivot_root . old_root

# commands below open 1022 port a ssh port. Please ensure the port you select is open under Oracle's VNIC 

/usr/sbin/sshd -D -p 1022 &

iptables -I INPUT -p tcp -m state --state NEW -m tcp --dport 1022 -j ACCEPT

# reconnect on port 1022: I kept the old putty session (on port 22) open and it didn't disconnect

# now the root storage is the RAM 

pkill agetty
pkill dbus-daemon
pkill atd
pkill iscsid
pkill rpcbind
pkill unattended-upgrades
kill 1 

# check with "lsof /old_root" that there is no remaining process 

umount -l /dev/sda1

# check :

df -h 
lsblk 

# the disk should be unmount ; for me lsblk showed a /dev/sda1 partiion but it went well.
# now, just copy the debian cloud image on the disk.

# I wanted an ARM64 image and Debian 10 , therefore I used this one https://cloud.debian.org/cdimage/cloud/buster/latest/debian-10-generic-arm64.tar.xz

curl -L https://cloud.debian.org/cdimage/cloud/buster/latest/debian-10-generic-arm64.tar.xz | tar -OJxvf - disk.raw | dd of=/dev/sda bs=1M

sync

reboot

# reboot command didn't work and returned an error "running in chroot, ignoring request" ;  I tried exiting as they told here (https://bbs.archlinux.org/viewtopic.php?id=184076) but it was of no use
# I forced rebooted it from the Oracle instances' GUI and it seemed stuck on "STOPPING." But after an hour I noticed that the VM was rebooted and running.
# I was able to login with the same public key as original ubuntu, but the username that worked was debian
# SUCCESS!! I checked with lsb_release -a and it showed Debian 10 (earlier it was showing ubuntu)

# Additional things: You may want to reinstall the Oracle cloud agent that was removed in the beginning.

not work for ubuntu24.04 at NOV, 2025

not work for ubuntu24.04 at NOV, 2025

works with Ubuntu 24.04. minimal, did it 3 days ago. If you used a x86, it must be minimal

i did change some things for my setup
i did drop the ssh secondary as it's basically useless
also i did fix the reboot not working at the end by using kernel magic

# elevate to root
sudo -i
# prevent any snap installation
cat <<EOF > /etc/apt/preferences.d/nosnap.pref
Package: snapd
Pin: release a=*
Pin-Priority: -10
EOF
# update 
apt update
apt dist-upgrade -y
apt install lsof -y
# make snap disappear
snap remove --purge oracle-cloud-agent
snap remove --purge core18
snap remove --purge snapd
systemctl stop snapd
systemctl disable snapd.service snapd.socket
apt purge snapd -y
rm -rf /var/cache/snapd/ /var/lib/snapd/ /var/snap/ /snap/ /home/*/snap /root/snap
# reboot so latest kernel is used
reboot
# re-elevate to root
sudo -i
# do the cleanup
apt purge $(dpkg-query -Wf '${Package}\n' | grep header) $(apt list --installed | grep -oP "^linux.*\d\d\d\d-oracle" | grep -v "$(uname -r)") linux-modules-extra-$(uname -r) lxc* lxd* vim* -y
apt autoremove -y
apt autoclean -y
apt clean -y
rm -rf /var/log/* /var/lib/apt/lists/*
# check /dev/sda1, it should be small enough
df -h
# create the ramdisk (since i'm on the arm64 ampere A1 i have 24GB ram so put 4G there)
cd /
mount -t tmpfs -o size=4000m tmpfs mnt
tar --one-file-system -c . | tar -C /mnt -x
# switch the VFS
mount --make-private -o remount,rw /
mount --move dev mnt/dev
mount --move proc mnt/proc
mount --move run mnt/run
mount --move sys mnt/sys
# rewrite the fstab to pount to our ramdisk
sed -i '/^[^#]/d;' mnt/etc/fstab
echo 'tmpfs / tmpfs defaults 0 0' >> mnt/etc/fstab
# go into our ramdisk
cd mnt
mkdir old_root
mount --make-private /
unshare -m
pivot_root . old_root
# actually fix the env so shit work
chroot /mnt
# kill processes
pkill agetty
pkill dbus-daemon
pkill atd
pkill iscsid
pkill rpcbind
pkill -f unattended-upgrades
kill 1
# check if there is no stupid process still running
lsof /old_root
# prep our disk to be overwritten
umount -l /dev/sda1
# last sanity check
df -h
lsblk
# here we gooooo (flashing latest debian)
curl -L https://cloud.debian.org/images/cloud/trixie/latest/debian-13-generic-arm64.tar.xz | tar -OJxvf - disk.raw | dd of=/dev/sda bs=1M
# sync once it's done
sync
# now to reboot we need to use magic trick as reboot won't work under chroot
echo 1 > /proc/sys/kernel/sysrq
echo b > /proc/sysrq-trigger

thanks for the tutorial tho