Partition configuration


Flash memory types

Embedded devices almost universally use flash memory for storage. An important property of flash memory cells is that they can only handle a certain amount of writes until they fail (wear out). Wear leveling (distributing writes across the cells) and error correction (avoiding use of failed cells) are strategies used to prolong the life of flash storage devices. These strategies are either handled by the flash device itself or by software (OS and filesystem). From the software's point of view, there are two types of flash memory:

  • Block device. These flash devices will expose a linear array of blocks to the OS, just like hard drives do. This is the most common type of flash device used with Linux, except in very low-cost or older embedded devices. They are generally easy to work with and you can put block-device filesystems, like ext4 and fat, directly on top of them. Internally these devices contain a memory controller that runs a Flash Translation Layer firmware that transparently implements wear leveling and error correction. For this reason, they are also sometimes referred to as Flash Translation Layer (FTL) devices. For example, these types of flash devices expose themselves as block devices: SD, mini-SD, micro-SD, MMC, eMMC, RS-MMC, SSD, USB, CompactFlash, MemoryStick, MemoryStick Micro.

  • Raw flash. Raw flash devices do not have a memory controller that takes care of wear leveling or error correction, so this must be handled in software. Linux exposes raw flash devices as a Memory Technology Device (MTD) file. The user must take care when selecting a filesystem to ensure that it is MTD-aware and properly handles wear leveling and error correction. Popular filesystems for MTD devices include UBIFS, JFFS2, and YAFFS. Consult the raw flash section for details on setting up and configuration.

filesystem types

When building a Mender Yocto Project image the build output in tmp/deploy/images/<MACHINE> includes a binary rootfs filesystem image (e.g. with .mender extension), as well as a complete disk image (with .sdimg extension). The binary rootfs filesystem images are used when deploying updates to the device, while the .sdimg image is typically used just once during initial device provisioning to flash the entire storage, and includes the partition layout and all partitions.

In general Mender does not have dependencies on a specific filesystem type, but the version of U-Boot you are using must support the filesystem type used for rootfs because it needs to read the Linux kernel from the filesystem and start the Linux boot process.

The standard Yocto Project IMAGE_FSTYPES variable determines the image types to create in Yocto deploy directory. The meta-mender layer appends the mender type to that variable, and usually sdimg, uefiimg or a different type ending with img, depending on enabled image features for the Yocto-build. Selecting a filesystem for the individual partition files by setting the ARTIFACTIMG_FSTYPE variable. We advise that you clean up the IMAGE_FSTYPES variable to avoid creating unnecessary image files.

Configuring storage

Configuring storage for Mender requires setting two variables. The first variable, MENDER_STORAGE_DEVICE, configures the expected location on the device of the storage device. The second, MENDER_STORAGE_TOTAL_SIZE_MB is the total size of this physical storage medium. The values should be for the raw device containing the entire storage, not any single partition. For example:

# Example: Memory card storage
MENDER_STORAGE_DEVICE = "/dev/mmcblk0"
# Example: Memory card with 2GiB of storage.

These should be set either in local.conf, or preferably, in machine.conf.

A quick way to get exact storage capacity on a device is to run one of the following commands on it. This assumes that you have an existing image running on the device and that the mentioned tools are available there:

# For block based storage:
sudo blockdev --getsize64 /dev/mmcblk0 | xargs -i% expr % / 1048576

# For Flash storage:
mtdinfo -a | sed -rne '/^Amount of eraseblocks:/{s/.*[^0-9]([0-9]+) *bytes.*/\1/; p}' | awk '{s+=$0} END {print s}' | xargs -i% expr % / 1048576

The output will be in MiB, a number appropriate for MENDER_STORAGE_TOTAL_SIZE_MB.

More detailed storage configuration

If you need more fine grained control over which partitions Mender will use, you can set one or more the following variables to specific partition strings, using MENDER_STORAGE_DEVICE_BASE:


For example:


Note that the Mender image builder will not produce such images, so only set these variables if you're building partitioned images yourself, with a different layout than the default Mender layout (the example above reflects the default).

Configuring the partition sizes

When building a Mender Yocto Project image Mender defines and uses certain OpenEmbedded variables used to define the sizes of the partitions.

Mount point Purpose Default size Variable to configure size
/ Store the root filesystem and kernel. auto MENDER_STORAGE_TOTAL_SIZE_MB
<BOOT> Store the bootloader. 16 MB MENDER_BOOT_PART_SIZE_MB
/data Store persistent data, preserved during Mender updates. 128 MB MENDER_DATA_PART_SIZE_MB

Even though the default size of MENDER_DATA_PART_SIZE_MB is 128 MB, it will try to resize the partition and filesystem image on first boot to the full size of the underlying block device which is also resized to occupy remainder of available blocks on the storage medium. This functionality relies on systemd-growfs which is not available for all filesystems. See mender-growfs-data feature for more information.

The value of <BOOT> depends on what features are enabled:

  • With mender-uboot enabled: /uboot
  • With mender-grub and mender-bios enabled: /boot/grub
  • With only mender-grub enabled: /boot/efi

You can override these default values in your local.conf. For details consult Mender image variables.

Preserving data and configuration across updates

Deploying a full rootfs image update will wipe all data previously stored on that partition. To make data persist across updates, applications must use the partition mounted on /data. In fact, the Mender client itself uses /data/mender to preserve data and state across updates.

If you have data or configuration that you need to preserve across updates, the recommended approach is to create a symlink from where it gets written to somewhere within /data/. For example, if you have an application that writes to /etc/application1, then you can create a symlink /etc/application1 -> /data/application1 to ensure the data it writes is not lost during a Mender rootfs update.

Deploying files to the persistent data partition

When building a Mender Yocto Project image, if you need to include files in the persistent data partition, all you have to do is add those files to the /data directory in the root filesystem. For example:

do_install() {
    install -d ${D}/data
    install -m 0644 persistent.txt ${D}/data/

The meta-mender-demo layer includes a sample recipe, hello-mender, which deploys a text file to the persistent data partition.

Keep in mind that any files you add to the /data directory are not included in .mender artifacts, since they don't contain a data partition. Only complete partitioned images (.biosimg, .sdimg, .uefiimg, etc) will contain the files.

Producing a standalone data partition image

Although it is not needed for most work with Mender, for some flashing setups, it can be useful to have the sole data partition available as an image file. In this case, adding dataimg to the Yocto Project IMAGE_FSTYPES variable will make the resulting image file given the .dataimg suffix. Its filesystem type will be the value of ARTIFACTIMG_FSTYPE. For example:

IMAGE_FSTYPES:append = " dataimg"

We welcome contributions to improve this documentation. To submit a change, use the Edit link at the top of the page or email us at .