22

I am playing with QEMU. Here I found some prebuilt OS images:

http://docs.openstack.org/trunk/openstack-compute/admin/content/starting-images.html

But they are all meant for a 64-bit system while my system is 32 bit. Does anyone know if there is any 32-bit prebuilt image online?

So I can directly use them and not need to bother with the installation.

Thanks.

3
  • Are you shore you can only have a 32-bit operating system? It's only depending on the CPU.
    – Alvar
    Commented Apr 15, 2013 at 15:46
  • @Alvar I am not quite sure. My host is Fedora 12 with kernel 2.6.29. I believe my OS is 32 bit. Processor is Intel Core 2 Duo CPU E8400. I just use the command "qemu-kvm -m 1024 img". Img is downloaded from the website I gave. It just stopped at "loading initial ramdisk"....
    – Hao Shen
    Commented Apr 15, 2013 at 16:07
  • 2
    Yes your CPU has 64-bit compatibility. Source
    – Alvar
    Commented Apr 15, 2013 at 23:37

4 Answers 4

30

This answer contains detailed steps for the following setups:

  • cloud image amd64 and arm64
  • debootstrap amd64 and arm64
  • desktop image amd64

All was tested on an Ubuntu 18.04 host targeting 18.04 guests.

Cloud image amd64

The Ubuntu cloud images are pre-installed images that allow your to boot directly without doing the usual desktop system installation. See also: https://serverfault.com/questions/438611/what-are-ubuntu-cloud-images

#!/usr/bin/env bash

sudo apt-get install cloud-image-utils qemu

# This is already in qcow2 format.
img=ubuntu-18.04-server-cloudimg-amd64.img
if [ ! -f "$img" ]; then
  wget "https://cloud-images.ubuntu.com/releases/18.04/release/${img}"

  # sparse resize: does not use any extra space, just allows the resize to happen later on.
  # https://superuser.com/questions/1022019/how-to-increase-size-of-an-ubuntu-cloud-image
  qemu-img resize "$img" +128G
fi

user_data=user-data.img
if [ ! -f "$user_data" ]; then
  # For the password.
  # https://stackoverflow.com/questions/29137679/login-credentials-of-ubuntu-cloud-server-image/53373376#53373376
  # https://serverfault.com/questions/920117/how-do-i-set-a-password-on-an-ubuntu-cloud-image/940686#940686
  # https://askubuntu.com/questions/507345/how-to-set-a-password-for-ubuntu-cloud-images-ie-not-use-ssh/1094189#1094189
  cat >user-data <<EOF
#cloud-config
password: asdfqwer
chpasswd: { expire: False }
ssh_pwauth: True
EOF
  cloud-localds "$user_data" user-data
fi

qemu-system-x86_64 \
  -drive "file=${img},format=qcow2" \
  -drive "file=${user_data},format=raw" \
  -device rtl8139,netdev=net0 \
  -enable-kvm \
  -m 2G \
  -netdev user,id=net0 \
  -serial mon:stdio \
  -smp 2 \
  -vga virtio \
;

GitHub upstream.

After QEMU starts, you might have to hit enter for the boot menu to show. Select Ubuntu from there.

Then, the beginning of boot says:

error: no such device: root.

Press any key to continue...

but even if you don't press any key, boot continues after a short timeout. Go upvote this bug report: https://bugs.launchpad.net/cloud-images/+bug/1726476

Once boot finishes, login with:

  • username: ubuntu
  • password: asdfqwer

Internet works normally.

Cloud image arm64

TODO: I noticed that there is a bug that happens sometimes when using this: https://bugs.launchpad.net/cloud-images/+bug/1818197

Very similar to amd64, but we need some UEFI black magic for it to boot.

sudo apt-get install cloud-image-utils qemu-system-arm qemu-efi

# Get the image.
img=ubuntu-18.04-server-cloudimg-arm64.img
if [ ! -f "$img" ]; then
  wget "https://cloud-images.ubuntu.com/releases/18.04/release/${img}"
  qemu-img resize "$img" +128G
fi

# For the password.
user_data=user-data.img
if [ ! -f "$user_data" ]; then
  cat >user-data <<EOF
#cloud-config
password: asdfqwer
chpasswd: { expire: False }
ssh_pwauth: True
EOF
  cloud-localds "$user_data" user-data

  # Use the EFI magic. Picked up from:
  # https://wiki.ubuntu.com/ARM64/QEMU
  dd if=/dev/zero of=flash0.img bs=1M count=64
  dd if=/usr/share/qemu-efi/QEMU_EFI.fd of=flash0.img conv=notrunc
  dd if=/dev/zero of=flash1.img bs=1M count=64
fi

qemu-system-aarch64 \
  -M virt \
  -cpu cortex-a57 \
  -device rtl8139,netdev=net0 \
  -m 4096 \
  -netdev user,id=net0 \
  -nographic \
  -smp 4 \
  -drive "if=none,file=${img},id=hd0" \
  -device virtio-blk-device,drive=hd0 \
  -drive "file=${user_data},format=raw" \
  -pflash flash0.img \
  -pflash flash1.img \
;

GitHub upstream.

debootstrap amd64

Not a pre-made image, but it downloads all pre-built packages, so it is also fast, but also much more configurable and useful.

#!/usr/bin/env bash

set -eux

debootstrap_dir=debootstrap
root_filesystem=debootstrap.ext2.qcow2

sudo apt-get install \
  debootstrap \
  libguestfs-tools \
  qemu-system-x86 \
;

if [ ! -d "$debootstrap_dir" ]; then
  # Create debootstrap directory.
  # - linux-image-generic: downloads the kernel image we will use under /boot
  # - network-manager: automatically starts the network at boot for us
  sudo debootstrap \
    --include linux-image-generic \
    bionic \
    "$debootstrap_dir" \
    http://archive.ubuntu.com/ubuntu \
  ;
  sudo rm -f "$root_filesystem"
fi

linux_image="$(printf "${debootstrap_dir}/boot/vmlinuz-"*)"

if [ ! -f "$root_filesystem" ]; then
  # Set root password.
  echo 'root:root' | sudo chroot "$debootstrap_dir" chpasswd

  # Remount root filesystem as rw.
  cat << EOF | sudo tee "${debootstrap_dir}/etc/fstab"
/dev/sda / ext4 errors=remount-ro,acl 0 1
EOF

  # Automaticaly start networking.
  # Otherwise network commands fail with:
  #     Temporary failure in name resolution
  # https://askubuntu.com/questions/1045278/ubuntu-server-18-04-temporary-failure-in-name-resolution/1080902#1080902
  cat << EOF | sudo tee "$debootstrap_dir/etc/systemd/system/dhclient.service"
[Unit]
Description=DHCP Client
Documentation=man:dhclient(8)
Wants=network.target
Before=network.target
[Service]
Type=forking
PIDFile=/var/run/dhclient.pid
ExecStart=/sbin/dhclient -4 -q
[Install]
WantedBy=multi-user.target
EOF
  sudo ln -sf "$debootstrap_dir/etc/systemd/system/dhclient.service" \
    "${debootstrap_dir}/etc/systemd/system/multi-user.target.wants/dhclient.service"

  # Why Ubuntu, why.
  # https://bugs.launchpad.net/ubuntu/+source/linux/+bug/759725
  sudo chmod +r "${linux_image}"

  # Generate image file from debootstrap directory.
  # Leave 1Gb extra empty space in the image.
  sudo virt-make-fs \
    --format qcow2 \
    --size +1G \
    --type ext2 \
    "$debootstrap_dir" \
    "$root_filesystem" \
  ;
  sudo chmod 666 "$root_filesystem"
fi

qemu-system-x86_64 \
  -append 'console=ttyS0 root=/dev/sda' \
  -drive "file=${root_filesystem},format=qcow2" \
  -enable-kvm \
  -serial mon:stdio \
  -m 2G \
  -kernel "${linux_image}" \
  -device rtl8139,netdev=net0 \
  -netdev user,id=net0 \
;

GitHub upstream.

This boots without any systemd errors or warnings.

Now from the terminal, login with root / root, and then check that the Internet works with the following commands:

printf 'GET / HTTP/1.1\r\nHost: example.com\r\n\r\n' | nc example.com 80
apt-get update
apt-get install hello
hello

We used nc as explained at https://stackoverflow.com/questions/32341518/how-to-make-an-http-get-request-manually-with-netcat/52662497#52662497 because:

Analogous Debian version: https://unix.stackexchange.com/questions/275429/creating-bootable-debian-image-with-debootstrap/473256#473256

Build your own kernel

Since we are here:

git clone git://kernel.ubuntu.com/ubuntu/ubuntu-bionic.git
cd ubuntu-bionic
# Tag matches the working kernel that debootstrap downloaded for us.
git checkout Ubuntu-4.15.0-20.21
fakeroot debian/rules clean
debian/rules updateconfigs
fakeroot debian/rules build-generic
linux_image="$(pwd)/debian/build/build-generic/arch/x86_64/boot/bzImage"

This produced the exact same config and I believe used the exact same source code as the packaged Ubuntu that debootstrap downloaded as explained at: Where can I get the 11.04 kernel .config file?

Then I patched it with:

diff --git a/init/main.c b/init/main.c
index b8b121c17ff1..542229349efc 100644
--- a/init/main.c
+++ b/init/main.c
@@ -516,6 +516,8 @@ asmlinkage __visible void __init start_kernel(void)
        char *command_line;
        char *after_dashes;

+ pr_info("I'VE HACKED THE LINUX KERNEL!!!");
+
        set_task_stack_end_magic(&init_task);
        smp_setup_processor_id();
        debug_objects_early_init();

and rebuild:

fakeroot debian/rules build-generic

and it did print my message during boot:

I'VE HACKED THE LINUX KERNEL!!!

The rebuild was not very fast though, so maybe there is a better command? I just waited for it to say:

Kernel: arch/x86/boot/bzImage is ready  (#3)

and went ahead with the run.

debootstrap arm64

The procedure was similar to the amd64 one, but with the following differences:

1)

We must do a two stage debootstrap:

  • first with --foreign to just download the packages
  • then we install QEMU static into the chroot
  • then we do the package installation with --second-stage using QEMU user mode emulation + binfmt_misc

See also: What is debootstrap --second-stage for

  1. the default kernel boot fails at the end with:

    [ 0.773665] Please append a correct "root=" boot option; here are the available partitions: [ 0.774033] Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block(0,0)

The empty partition list indicates that there is a serious error with the disk driver, after a bit of trying the missing option is:

CONFIG_VIRTIO_BLK=y

I think it works when I use the ISO because the modules must get loaded from the initrd.

I tried to use other disk types, but virtio is the only valid value for -drive if= when -M virt, which is the saner machine type nowadays.

Therefore we must recompile our own kernel with that option enabled, as explained here: When cross-compiling the kernel, how can I stop it making from clean every time when I just want to modify one file?

Ubuntu devs should turn this CONFIG y by default! It is very useful!

TODO: the network is not working, the error message is:

root@ciro-p51:~# systemctl status dhclient.service
root@ciro-p51:~# cat f
● dhclient.service - DHCP Client
   Loaded: loaded (/etc/systemd/system/dhclient.service; enabled; vendor preset: enabled)
   Active: failed (Result: protocol) since Sun 2018-01-28 15:58:42 UTC; 2min 2s ago
     Docs: man:dhclient(8)
  Process: 171 ExecStart=/sbin/dhclient -4 -q (code=exited, status=0/SUCCESS)

Jan 28 15:58:40 ciro-p51 systemd[1]: Starting DHCP Client...
Jan 28 15:58:42 ciro-p51 dhclient[171]: No broadcast interfaces found - exiting.
Jan 28 15:58:42 ciro-p51 systemd[1]: dhclient.service: Can't open PID file /var/run/dhclient.pid (yet?) after start: No such file or directory
Jan 28 15:58:42 ciro-p51 systemd[1]: dhclient.service: Failed with result 'protocol'.
Jan 28 15:58:42 ciro-p51 systemd[1]: Failed to start DHCP Client.

Here is the fully automated script:

#!/usr/bin/env bash

# https://askubuntu.com/questions/281763/is-there-any-prebuilt-qemu-ubuntu-image32bit-online/1081171#1081171

set -eux

debootstrap_dir=debootstrap
root_filesystem=debootstrap.ext2.qcow2

sudo apt-get install \
  gcc-aarch64-linux-gnu \
  debootstrap \
  libguestfs-tools \
  qemu-system-aarch64 \
  qemu-user-static \
;

if [ ! -d "$debootstrap_dir" ]; then
  sudo debootstrap \
    --arch arm64 \
    --foreign \
    bionic \
    "$debootstrap_dir" \
    http://ports.ubuntu.com/ubuntu-ports \
  ;
  sudo mkdir -p "${debootstrap_dir}/usr/bin"
  sudo cp "$(which qemu-aarch64-static)" "${debootstrap_dir}/usr/bin"
  sudo chroot "$debootstrap_dir" /debootstrap/debootstrap --second-stage
  sudo rm -f "$root_filesystem"
fi

linux_image="$(printf "${debootstrap_dir}/boot/vmlinuz-"*)"

if [ ! -f "$root_filesystem" ]; then
  # Set root password.
  echo 'root:root' | sudo chroot "$debootstrap_dir" chpasswd

  # Remount root filesystem as rw.
  cat << EOF | sudo tee "${debootstrap_dir}/etc/fstab"
/dev/sda / ext4 errors=remount-ro,acl 0 1
EOF

  # Automaticaly start networking.
  # Otherwise network commands fail with:
  #     Temporary failure in name resolution
  # https://askubuntu.com/questions/1045278/ubuntu-server-18-04-temporary-failure-in-name-resolution/1080902#1080902
  cat << EOF | sudo tee "${debootstrap_dir}/etc/systemd/system/dhclient.service"
[Unit]
Description=DHCP Client
Documentation=man:dhclient(8)
Wants=network.target
Before=network.target

[Service]
Type=forking
PIDFile=/var/run/dhclient.pid
ExecStart=/sbin/dhclient -4 -q

[Install]
WantedBy=multi-user.target
EOF
  sudo ln -sf "${debootstrap_dir}/etc/systemd/system/dhclient.service" \
    "${debootstrap_dir}/etc/systemd/system/multi-user.target.wants/dhclient.service"

  # Why Ubuntu, why.
  # https://bugs.launchpad.net/ubuntu/+source/linux/+bug/759725
  sudo chmod +r "${linux_image}"

  # Generate image file from debootstrap directory.
  # Leave 1Gb extra empty space in the image.
  sudo virt-make-fs \
    --format qcow2 \
    --size +1G \
    --type ext2 \
    "$debootstrap_dir" \
    "$root_filesystem" \
  ;
  sudo chmod 666 "$root_filesystem"
fi

# Build the Linux kernel.
linux_image="$(pwd)/linux/debian/build/build-generic/arch/arm64/boot/Image"
if [ ! -f "$linux_image" ]; then
  git clone --branch Ubuntu-4.15.0-20.21 --depth 1 git://kernel.ubuntu.com/ubuntu/ubuntu-bionic.git linux
  cd linux
patch -p1 << EOF
diff --git a/debian.master/config/config.common.ubuntu b/debian.master/config/config.common.ubuntu
index 5ff32cb997e9..8a190d3a0299 100644
--- a/debian.master/config/config.common.ubuntu
+++ b/debian.master/config/config.common.ubuntu
@@ -10153,7 +10153,7 @@ CONFIG_VIDEO_ZORAN_ZR36060=m
 CONFIG_VIPERBOARD_ADC=m
 CONFIG_VIRTIO=y
 CONFIG_VIRTIO_BALLOON=y
-CONFIG_VIRTIO_BLK=m
+CONFIG_VIRTIO_BLK=y
 CONFIG_VIRTIO_BLK_SCSI=y
 CONFIG_VIRTIO_CONSOLE=y
 CONFIG_VIRTIO_INPUT=m
EOF
  export ARCH=arm64
  export $(dpkg-architecture -aarm64)
  export CROSS_COMPILE=aarch64-linux-gnu-
  fakeroot debian/rules clean
  debian/rules updateconfigs
  fakeroot debian/rules DEB_BUILD_OPTIONS=parallel=`nproc` build-generic
  cd -
fi

qemu-system-aarch64 \
  -append 'console=ttyAMA0 root=/dev/vda rootfstype=ext2' \
  -device rtl8139,netdev=net0 \
  -drive "file=${root_filesystem},format=qcow2" \
  -kernel "${linux_image}" \
  -m 2G \
  -netdev user,id=net0 \
  -serial mon:stdio \
  -M virt,highmem=off \
  -cpu cortex-a57 \
  -nographic \
;

GitHub Upstream.

Desktop image

See: How to run Ubuntu desktop on QEMU?

It does require going through the installer manually, but it is the most stable thing you can possibly do, and totally fine if you just want to get a VM for interactive use running from time to time.

For aarch64, I haven't gotten the desktop working yet maybe keep an eye out for: How to run Ubuntu 16.04 ARM in QEMU?

Debian aarch64

https://gist.github.com/philipz/04a9a165f8ce561f7ddd and change sudo mount /dev/nbd0p2 to nbd0p1.

5
  • 1
    I modified your script a little bit: gist.github.com/lnyng/8342947a1d5455303fd8730c9ca35da0 The main change is to create a dhclient systemd unit to avoid installing network-manager, which requires many UI related libraries (about 300+MB for my installation).
    – lnyng
    Commented Oct 6, 2018 at 7:23
  • @lyang thanks! You've allowed me to not learn systemd properly once more :-) Commented Oct 6, 2018 at 9:27
  • @lyang when I tried it for arm64 as explained in the updated answer, I get the error: dhclient.service: Can't open PID file /var/run/dhclient.pid (yet?) after start: No such file or directory, any clues? After boot I am able to touch /var/run/a. Commented Oct 8, 2018 at 23:28
  • Seems to be a permission error. Maybe changing the pid path to other places like /tmp/dhclient.pid? Or just remove it altogether if we don't really care to kill that process...
    – lnyng
    Commented Oct 10, 2018 at 19:52
  • 2
    Thanks! This answer really helped a lot.
    – jakob.j
    Commented Apr 15, 2022 at 13:08
16

A quick Google search revealed the following (I have not tried any of them):

Also, you can use vmbuilder(referred here as ubuntu-vmbuilder) to quickly create Ubuntu images to KVM, VirtualBox, etc.

As a last resort, you can use qemu-img command to convert disk images from VirtualBox/VMware to a format more suited to QEMU/KVM (this may not be needed: I think QEMU/KVM can work with other image types like vdi or vmdk).

$ qemu-img convert -f [vdi|vmdk|...] -O qcow2 OriginalImage NewImage

NOTE: If you are using a 32bit OS you cannot run 64bit virtual machines with KVM. But QEMU is an emulator, so it should let you run 64bit vm's on a 32bit operating system. But the performance overhead will probably be huge!

0

Refer to http://cloud-images.ubuntu.com/ which contains cloud images which can be used with qemu/kvm.

3
0

https://www.turnkeylinux.org/ has been around for ages. They have a huge catalog downloadable, premade "appliance" like images in numerous formats (ova, iso, vdmk, openstack, xen). They even can launch an image right in AWS for you.

When I want to start exploring a particular stack or need to knock out a problem, I frequently will download on of their images, convert it to a cow2 and use it.

You can also grab images from https://app.vagrantup.com/boxes/search or https://virtualboxes.org/images/ and convert them as well.

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .