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.

just a tip by my trial with Oracle AMD free instance VM.Standard.E2.1.Micro, which only has 1GB RAM, you can actually use swap for the temporary chrooted os. tmpfs needs 1,7G in the guide above, which we don't have.
After the removal of the logs, I did the following (I was in Ubuntu 22 minimal)

mkdir /mnt2
fallocate -l 2G /mnt2/swapfile (mine was nearly 1,8GB in size)
chmod 600 /mnt2/swapfile
mkswap /mnt2/swapfile
swapon /mnt2/swapfile
tar --one-file-system --exclude=./mnt2/swapfile -c . | tar -C /mnt -x

full process at the end :)

I didn't change anything else in the process, also the fstab, which should be invalid at this point.
The only thing what I noticed, that after the remounts in /mnt most of the commands didn't work anymore, so I did a chroot /mnt after the pivot_root to continue. worked. I used the bookworm version for amd64, because I used the VPS for openmptcp, which requires max Debian 12. Don't forget to use the "debian" user, when login with ssh-key.
https://cloud.debian.org/cdimage/cloud/bookworm/20250814-2204/debian-12-generic-amd64-20250814-2204.tar.xz

Yes the process is still finishing, even the swapfile is overwritten by the dd :-) ignore the error, that the umount of sda didn't work. It is a surgery on the heart, but it works.

I did an additional server now with Debian trixie.

su -
snap remove --purge  oracle-cloud-agent && snap remove --purge core18
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* && apt -y autoremove && apt -y autoclean && apt -y clean
...
rm -rf /var/log/* /var/lib/apt/lists/*
mkdir /mnt2
fallocate -l 2G /mnt2/swapfile
chmod 600 /mnt2/swapfile
mkswap /mnt2/swapfile
swapon /mnt2/swapfile

Setting up swapspace version 1, size = 2 GiB (2147479552 bytes)
no label, UUID=a5d43c04-ade7-47be-b211-575eea636ad5

cd /
tar --one-file-system --exclude=./mnt2/swapfile -c . | tar -C /mnt -x

tar: ./mnt: file changed as we read it

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

chroot /mnt
iptables -I INPUT -p tcp -m state --state NEW -m tcp --dport 1022 -j ACCEPT
/usr/sbin/sshd -D -p 1022 &
pkill agetty
pkill dbus-daemon
pkill atd
pkill iscsid
pkill rpcbind
pkill unattended-upgrades
kill 1

[1] 1976091

The following commands can be skipped, it's just for checking the disk/partitions

umount -l /dev/sda1
> umount: /dev/sda1: not mounted.

df -h
> Filesystem      Size  Used Avail Use% Mounted on
> tmpfs           480M     0  480M   0% /dev/shm
> tmpfs            96M  2.3M   94M   3% /run
> tmpfs           5.0M     0  5.0M   0% /run/lock
> tmpfs            96M  4.0K   96M   1% /run/user/1001
> efivarfs        256K   21K  231K   9% /sys/firmware/efi/efivars

lsblk
> NAME    MAJ:MIN RM  SIZE RO TYPE MOUNTPOINTS
> loop0     7:0    0 49.3M  1 loop
> loop1     7:1    0 50.9M  1 loop
> sda       8:0    0 46.6G  0 disk
> ├─sda1    8:1    0 46.5G  0 part
> ├─sda14   8:14   0    4M  0 part
> └─sda15   8:15   0  106M  0 part

write Debian to the disk.

curl -L https://cloud.debian.org/cdimage/cloud/trixie/20250814-2204/debian-13-generic-amd64-20250814-2204.tar.xz | tar -OJxvf - disk.raw | dd of=/dev/sda bs=1M

% Total % Received % Xferd Average Speed Time Time Time Current
Dload Upload Total Spent Left Speed
100 374 100 374 0 0 1357 0 --:--:-- --:--:-- --:--:-- 1360
0 0 0 0 0 0 0 0 --:--:-- --:--:-- --:--:-- 0disk.raw
100 303M 100 303M 0 0 2019k 0 0:02:33 0:02:33 --:--:-- 1824k
0+68353 records in
0+68353 records out
3221225472 bytes (3.2 GB, 3.0 GiB) copied, 186.029 s, 17.3 MB/s

sync

=> force reboot via Oracle website.
Connect to your VPS again, now in debian, with the same SSH key:

ssh-keygen -f '/home/mylocaluser/.ssh/known_hosts' -R '<my oracle vps ip'
ssh -i mykeyfile debian@myvpsip

Checkin the VPS, also don't forget to change the password of debian/root:

root@ubuntu-x86:~# uname -ar
Linux ubuntu-x86 6.12.41+deb13-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.12.41-1 (2025-08-12) x86_64 GNU/Linux
root@ubuntu-x86:~# lsb_release -r 
Release:        13

(yes I know, uname showed already deb13, just additional)