Building and manipulating images

Building and manipulating images

This section looks into how to manipulate, inspect and create boot images. There are enough steps involved in creating a kernel which will boot the device, the steps to turn that kernel into an image to use in LAVA can be as varied.

The details of preparing a suitable kernel or configuring the selected bootloader is beyond the scope of this page. We’ll concentrate on how to look inside available images, what needs to be done to use a different operating system as the rootfs and how to mount, modify or create boot images.

This documentation relies on support present in the Linux kernel. Other kernels can be put inside boot images but using such kernels at runtime to create boot images is beyond the scope of this page.

Basics of building an image

  1. kernel - LAVA typically works with Linux but there’s nothing to say that other kernels can’t be used - just don’t expect LAVA (or LAVA developers) to be able to have direct knowledge of any issues with kernels other than Linux
  2. bootloader - lots of test jobs have been run with u-boot and increasing amounts of work are going into UEFI. Grub is also being considered for future LAVA tests.
  3. rootfs - generally a simple, minimal tarball of a filesystem created by any of the many tools available to bootstrap the operating system. Debian based distributions commonly use debootstrap or similar tools.
    • Changes to the rootfs to make it bootable - raw bootstraps are rarely bootable directly, various init changes are needed and some of these are board specific (e.g. which device to use for the serial console).

Obtaining a kernel

This often requires specialist knowledge of the particular board and you may be dependent on a landing team or other third party for a kernel configuration and patches. Some sources only provide a binary image, sometimes already combined with a bootloader.

Obtaining a bootloader

Similar to a kernel, you may have little choice over which bootloader to use, although it is entirely reasonable to chain a more limited bootloader provided by someone else into a more capable bootloader which has more functionality. Note - the Linux kernel can be used as a secondary bootloader using kexec. The details of how to do this will vary according to the board, available bootloader and boot requirements.

From here on, this page works on how to get a kernel and bootloader into an image to boot on the device.

Inspecting existing images

Tools to install and get to know

  1. parted - there are lots of sites with information on parted, the simplest way to get used to it is to use it on empty block devices - an example is at the end of this section.
  2. dd - a utility to copy a file which can take input from devices like /dev/zero, commonly used to create empty files of a known size and to copy images from one block device to another whilst preserving internal partitions. (dd always copies to the device, not the partition, so /dev/mmcblk0, not /dev/mmcblk0p3.)
  3. qemu - a wide variety of support for booting images, including images for architectures other than the host architecture.
  4. mount - already installed but there are options which will become second-nature after working with boot images.
  5. gzip - images are typically compressed for download. There are other compression algorithms but most images contain a lot of empty space (for later tests to take) so gzip is usually enough to get suitable compression. Compressed images will need gunzip before being mountable.
  6. losetup - this is part of the loop device support of the Linux kernel.
  7. chroot - change root into a directory containing a new rootfs, if using qemu, this rootfs could be of a different architecture. chroot puts you into a new shell inside the rootfs where you can modify files and execute programs without affecting the external system. (There are limitations to how much a chroot can protect the external system, but these are unlikely to affect building a boot image.)

Concepts behind boot images

  1. offsets - once decompressed, many boot images contain multiple partitions, so a simple mount operation, even using the loop option, will fail. An offset tells mount where to find the start of the partition to be mounted from inside the image. Offsets are determined by the original setup of the image and can be determined using tools like parted.
  2. loop devices - the Linux loop kernel module can allow an image to be mounted as a block device. Such mount operations need to be performed as root or with sudo. Loop devices can be limited but see Increasing the number of loop devices.
  3. boot partitions - some bootloaders require that files required to boot the device are on a partition with a particular filesystem, often FAT. To allow the rootfs to use a different filesystem like ext2, ext3 or ext4, the boot files are on a separate partition.
  4. serial console - a device to which the device can write messages during boot and provide a login prompt (which can be automated for a LAVA test job).
  5. root password - the one thing most people forget about when creating a rootfs from their favourite distribution is that the root password is typically created by an installer not a bootstrap tool. Depending on the security of the OS, you may need to chroot into the new rootfs before finishing the image and set a usable root password with the passwd command.

Find the offset

  1. First, decompress your image. These examples will assume that the resulting file is called test.img

  2. Print the partition offsets:

    $ /sbin/parted test.img -s unit b print
    Model:  (file)
    Disk /home/linaro/documents/arndale-vmgroup/test.img: 1073741824B
    Sector size (logical/physical): 512B/512B
    Partition Table: msdos
    Number  Start      End          Size         Type     File system  Flags
    1      512B       4194303B     4193792B     primary
    2      4194304B   58720255B    54525952B    primary  fat32        boot, lba
    3      58720256B  1073741823B  1015021568B  primary  ext4

    In this example, there is an unused partition starting at an offset of 512 bytes, followed by a VFAT boot partition starting at an offset of 4194304 bytes and the main rootfs in an ext4 partition starting at an offset of 58720256 bytes.

    Other tasks using parted will need root access or sudo.

Mounting partitions using loop and offset

  1. To mount the boot partition, pass the loop and offset options to mount:

    $ sudo mkdir -p /mnt/boot
    $ sudo mount -oloop,offset=4194304 test.img /mnt/boot


    Failures from mount complaining about a bad superblock can arise from a wrong offset.

  2. When you are finished with the mount, un-mount it:

    $ sudo umount /mnt/boot


    Remember to check the output of mount and avoid mounting the same partition more than once or moving the image without using umount.

Creating new images

  1. Use dd to create an empty file which can be used to host partitions and form the basis of a new boot image.

    • Using /dev/zero is recommended as it will result in much better compression if the empty file space remaining in the image is zeroed.

    dd can create a file of any size, subject to the free space on your machine. Specify the size of each block to write and the number of blocks. To create an image of 1Gb (1024Mb) use:

    $ sudo dd if=/dev/zero of=test.img bs=1M count=1024
  2. Create a partition table - whilst it is possible to use images without partition tables if all files are in a single filesystem, some devices or bootloaders may refuse to boot from such images:

    losetup /dev/loop0 test.img
    parted /dev/sda -s unit mb mktable msdos

    If you are copying the layout of a known, working, image you can use parted to replicate the partitions. If you just need a boot partition, then allow space for modification. It is very likely that you or someone using your image will want to change the kernel image or test a second kernel. There should always be enough space in your boot partition to have a second kernel image. Note that kernel images may increase in size as more functionality is supported.

    Refer to the parted documentation for how to create the partition layout you want and experiment with your empty test image file. parted has an interactive mode which can be used to get used to the tool and the options:

    $ sudo parted test.img

    One example setup could be:

    parted /dev/loop0 -s unit mb mkpart primary 1 10
    parted /dev/loop0 -s unit mb mkpart primary 11 110
    parted /dev/loop0 -s unit mb mkpart primary 111 1024
    parted /dev/loop0 unit B -s print
    Model:  (file)
    Disk /dev/loop0: 1073741824B
    Sector size (logical/physical): 512B/512B
    Partition Table: msdos
    Number  Start       End          Size        Type     File system  Flags
     1      1048576B    10485759B    9437184B    primary
     2      10485760B   110100479B   99614720B   primary
     3      110100480B  1024458751B  914358272B  primary
  3. Create a filesystem for each partition. After parted has created the partitions, the loop devices need to be set using the offsets declared by parted:

    losetup -o 10485760 /dev/loop1 /dev/loop0
    losetup -o 110100480 /dev/loop2 /dev/loop0
    mkfs.vfat /dev/loop1
    mkfs.ext3 /dev/loop2
  4. Copy your files onto the new filesystems:

    mount -oloop,offset=10485760 test.img /mnt/boot/
    pushd /mnt/boot/
    tar -xzf /tmp/boot.tar.gz
    umount /mnt/boot/
  5. Clean up your losetup operations:

    losetup -d /dev/loop2
    losetup -d /dev/loop1
    losetup -d /dev/loop0

    Ensure that there are no loopback mounts remaining:

    losetup -a

Making a bootstrap rootfs bootable

  1. set the serial console - Each device tends to have a different device used for the serial console, requiring a line to be added to the init process. For Debian, this would need to be /etc/inittab. This example is from an iMX.53 image:

    echo T0:23:respawn:/sbin/getty -L ttymxc0 115200 vt102 >> ./etc/inittab

    The bootloader settings for the board usually indicate which device is to be used as the serial console.

  2. set default networking - depending on your bootstrap tool, there may well be no network interfaces defined. For Debian, this can be implemented using a file in /etc/network/interfaces.d/, e.g.:

    echo auto lo eth0 > ./etc/network/interfaces.d/base
    echo iface lo inet loopback >> ./etc/network/interfaces.d/base
    echo iface eth0 inet dhcp >> ./etc/network/interfaces.d/base
  3. set a root password - surprisingly easy to forget until after the image has booted. Depending on the distribution, this step can involve using qemu to chroot into the rootfs to be able to execute the passwd utility. Manual changes to /etc/passwd can be ignored, depending on the shadow / authentication precautions implemented by the distribution:

    $ sudo cp /usr/bin/qemu-armhf-static ./usr/bin/
    $ sudo chroot .

Other steps which may be required

  1. enable the serial console in securetty - e.g. the arndale board has a serial console in a device which does not generally appear in /etc/securetty, so this needs to be added:

    echo ttySAC2 >> ./etc/securetty
  2. set a useful hostname - choose your board hostname and your local domain (so that a fully qualified hostname can be supported):

    echo board board.domain >> ./etc/hosts

LAVA overlays

To simplify login and use auto-serial-console, there are overlay packages available for Debian and Ubuntu images which can be installed inside the image:

chroot /mnt/sata/chroots/unstable-armhf
mount proc -t proc /proc
mount devpts -t devpts /dev/pts
wget --no-check-certificate
wget --no-check-certificate
dpkg -i linaro-overlay-minimal_1112.2_all.deb linaro-overlay_1112.2_all.deb
rm linaro-overlay-minimal_1112.2_all.deb linaro-overlay_1112.2_all.deb
umount /dev/pts
umount /proc

Increasing the number of loop devices

It can be useful to increase the number of available loopback devices from the default of 8. This can be done by adding a file in /etc/modprobe.d/:

options loop max_loop=64