Skip to content

Latest commit

 

History

History
1832 lines (1322 loc) · 63.2 KB

integration.rst

File metadata and controls

1832 lines (1322 loc) · 63.2 KB
Raw

Integration

If you intend to prepare your platform for using RAUC as an update framework, this chapter will guide you through the required steps and show the different ways you can choose.

To integrate RAUC, you first need to be able to build RAUC as both a host and a target application. The host application is needed for generating update bundles while the target application or service performs the core task of RAUC: updating your device.

In an update system, a lot of components have to play together and have to be configured appropriately to interact correctly. In principle, these are:

  • Hardware setup, devices, partitions, etc.
  • The bootloader
  • The Linux kernel
  • The init system
  • System utilities (mount, mkfs, ...)
  • The update tool, RAUC itself

Note

When integrating RAUC into your embedded Linux system, and in general, we highly recommend using a Linux system build system like Yocto / OpenEmbedded or PTXdist that allows you to have well defined software states while easing integration of the different components involved.

For information about how to integrate RAUC using these tools, refer to the sections :ref:`sec_int_yocto` or :ref:`sec_int_ptxdist`.

Partitioning Your Device

A basic requirement for a redundant update system is to have your storage set up properly. In a simple case, this means having two redundant partitions of equal size for an A/B setup, or a tiny and a larger partition for a recovery/A setup.

Partitioning the storage is part of the bootstrap process and not in the scope of an update tool like RAUC.

Additionally, you may also need to reserve space for your bootloader, boot state information (such as the state backend for barebox or environment partition for U-Boot), :ref:`data partition(s) <sec-data-storage>` or similar.

Since changing the partition layout is hard or even impossible to change in the field, make sure it meets both current and possible future requirements.

Note

The /etc/fstab in your system's root FS (and RAUC's system.conf) should normally use stable paths to refer to partitions or devices. Especially filesystem UUIDs (UUID=<uuid> or /dev/disk/by-uuid/<uuid>) or labels (/dev/disk/by-label/<label>) should not be used, as they depend on the partition contents. UUIDs are likely to be different after an update and labels are usually not unique in an A/B setup.

Depending on your system design and firmware, good stable paths can be:

  • plain device names: /dev/sda, /dev/mmcblk0p1, /dev/nvme0n1p1, /dev/mapper/… (may change depending on enumeration ordering on some systems)
  • topology-based symlinks: /dev/disk/by-path/… (preferable if available)
  • partition-table-UUID-based symlinks: /dev/disk/by-partuuid/… (breaks if using the same disk image on e.g. both eMMC & SD card)

For more details on how to configure udev to generate stable paths, check the FAQ entry :ref:`faq-udev-symlinks`.

SD Card

Partitioning your SD Card is quite easy as it can simply be done from your host system by either using a command-line or graphical tool (fdisk/cfdisk/gparted) or by writing a full SD Card image as generated by your embedded Linux build system.

Most modern systems should use GPT for partitioning.

eMMC

In contrast to SD cards, an eMMC is fixed to your board and can not be easily pre-programmed before soldering (except for very large production batches). Accordingly, it usually needs to be set up from a Linux factory image booted from a secondary boot source such as network (e.g. TFTP), USB (e.g. Android fastboot), or other mass storage.

A useful tool for automating partitioning at runtime is systemd-repart.

Note that an eMMC also provides dedicated boot partitions that can be selected by setting Extended CSD registers and thus, if the SoC supports it, allows :ref:`atomic bootloader updates <sec-emmc-boot>`.

The eMMC specification also supports changing the operational mode of either the entire eMMC or only parts of it to better match requirements such as write endurance or data retention, e.g. by switching to pSLC mode.

SSD

SSDs can be handled similarly to eMMCs, except that most do not provide boot partition or operational mode support.

Note that you can still make use of atomic bootloader updates here when booting from :ref:`GPT <sec-gpt-partition>` (or :ref:`MBR <sec-mbr-partition>`).

NAND

Raw NAND can either be partitioned by devicetree partitions (as a subnode of the NAND controller) or (indirectly) by using UBI, which supports creating multiple UBI volumes.

Note that when using raw NAND, responsibility for bad block and NAND quirks handling is on your side (or on side of the NAND handling layer you use). Some bugs or misconfigurations will appear to work fine and only manifest as sporadic failures much later. If in doubt, using eMMC is recommended, especially for devices with normal quantity, since debugging NAND issues can be quite time-consuming.

RAUC System Configuration

The system configuration file is the central configuration in RAUC that abstracts the loosely coupled storage setup, partitioning and boot strategy of your board to a coherent redundancy setup world view for RAUC.

RAUC configuration files are loaded from one of the listed directories in order of priority, only the first file found is used: /etc/rauc/, /run/rauc/, /usr/lib/rauc/.

The system.conf is expected to describe the system RAUC runs on in a way that all relevant information for performing updates and making decisions are given.

Note

For a full reference of the system.conf file refer to section :ref:`sec_ref_slot_config`.

Similar to other configuration files used by RAUC, the system configuration uses a key-value syntax (similar to those known from .ini files).

Slot Configuration

The most important step is to describe the slots that RAUC should use when performing updates. Which slots are required and what you have to take care of when designing your system will be covered in the chapter :ref:`sec-scenarios`. This section assumes that you have already decided on a setup and want to describe it for RAUC.

A slot is defined by a slot section. The naming of the section must follow a simple format: [slot.<slot-class>.<slot-index>] where <slot-class> describes a class of possibly multiple redundant slots (such as rootfs, recovery or appfs) and slot-index is the index of the individual slot instance, starting with index 0.

If you have two redundant slots used for the root file system, for example, you should name your sections according to this example:

[slot.rootfs.0]
device = [...]

[slot.rootfs.1]
device = [...]

RAUC does not have predefined class names. The only requirement is that the class names used in the system config match those you later use in the update manifests.

The mandatory settings for each slot are:

  • the device that holds the (device) path describing where the slot is located,
  • the type that defines how to update the target device.

If the slot is bootable, then you also need

  • the bootname which is the name the bootloader uses to refer to this slot device.

Slot Type

A list of slot storage types currently supported by RAUC:

Type Description Tar support
raw A partition holding no (known) file system. Only raw image copies may be performed.  
ext4 A block device holding an ext4 filesystem. x
nand A raw NAND flash partition.  
nor A raw NOR flash partition.  
ubivol An UBI partition in NAND.  
ubifs An UBI volume containing an UBIFS in NAND. x
vfat A block device holding a vfat filesystem. x
jffs2 A flash memory holding a JFFS2 filesystem. x

Additionally, there are specific slot types for :ref:`atomic bootloader updates <sec-advanced-updating-bootloader>`: boot-emmc, boot-mbr-switch, boot-gpt-switch, boot-raw-fallback.

Depending on this slot storage type and the slot's :ref:`image filename <image-filename>` extension, RAUC determines how to extract the image content to the target slot.

While the generic filename extension .img is supported for all filesystems, it is strongly recommended to use explicit extensions (e.g. .vfat or .ext4) when possible, as this allows checking during installation that the slot type is correct.

Grouping Slots

If multiple slots belong together in a way that they always have to be updated together with the respective other slots, you can ensure this by grouping slots.

A group must always have a single bootable slot, then all other slots define a parent relationship to this bootable slot as follows:

[slot.rootfs.0]
...

[slot.appfs.0]
parent=rootfs.0
...

[slot.rootfs.1]
...

[slot.appfs.1]
parent=rootfs.1
...

Artifact Repository Configuration

To configure :ref:`Artifact Repositories <sec-basic-artifact-repositories>`, you first need to have a shared partition which is mounted before starting the RAUC service. Each artifact repository needs its own directory on this shared partition. The directory must be created outside of RAUC.

For each repository, you need to add an :ref:`[artifacts.\<repo-name\>] section <sec_ref_artifacts>` to your system.conf. An artifact repository is referenced from the manifest using its name, so that the name needs to be unique across all slot and repository names.

[artifacts.add-ons]
path=/srv/add-ons
type=trees

This example specifies one repository stored in the add-ons directory on the shared data partition mounted on /srv. The directory must exist before starting RAUC as it will not be created automatically. The name of the repository is add-ons as well, so it could be targeted for installation with an :ref:`[image.add-ons/app-1] section <image-section>` in the manifest. In that case, RAUC would install the contents of the archive specified in the image to the repository and make it available via a link at /srv/add-ons/app-1.

A single bundle can contain images for multiple artifacts across multiple repositories. The bundle defines the intended target state of each repository mentioned in its manifest. This means that previously installed artifacts can be removed from a repository by installing a bundle which contains a different artifact for that repository. Artifacts which are currently in use (i.e. which have open files that can be detected by trying to acquire a write lease) will not be deleted, but only their symlink is removed or replaced.

Note

There is currently no way to remove all artifacts from a repository. If you need that functionality, please reach out to us!

Internally, artifacts are stored under their artifact name and image hash in the repository's directory. This means that artifacts are installed only if a given version is not yet available in the repository. A symlink is created from the image name to the actual artifact, making it available to the rest of the system atomically.

Note

Currently, you need to ensure that enough space is available on the filesystem for all installed artifacts and one temporary copy of a single artifact during installation. In the future, RAUC could be extended to check that enough space is available by itself.

You can use the :ref:`post-install handler <sec-post-install-handler>` to notify the running system about newly installed artifacts. By the time this handler is executed, the symlinks to the artifacts have been created. For example, you could restart all services or containers running from the artifact repositories.

Repository Type

Each repository is configured with a type which specifies how it stores and manages artifacts.

files

Each artifact is a single file.

Possible use-cases for this type are:

  • filesystem images for use with systemd-sysext
  • large data files such as maps, videos or read-only databases
  • disk images for virtual machines

Note

In the future, this could be combined with adaptive updates using the block-hash-index method.

trees

Each artifact is a directory tree containing files. An image should be a tar archive or a tar converted to a directory tree using convert=tar-extract in the input manifest.

Possible use-cases for this type are:

  • add-on modules consisting of binaries and some meta-data
  • container-like OS trees for use with systemd-nspawn or other runtimes.

Note

In the future, this could be combined with adaptive updates using new methods which could detect unmodified files.

Library Dependencies

The minimal requirement for RAUC regardless of whether intended for the host or target side is GLib (minimum version 2.45.8) as utility library and OpenSSL (>=1.0) for signature handling.

Note

In order to let RAUC detect mounts correctly, GLib must be compiled with libmount support (--enable-libmount) and at least be 2.49.5.

For network support (enabled with --Dnetwork=true), additionally libcurl is required. This is only useful for the target service.

For JSON-style support (enabled with -Djson=enabled), additionally libjson-glib is required.

Kernel Configuration

The kernel used on the target device must support both loop block devices and the SquashFS file system to allow installing RAUC bundles. For the recommended verity :ref:`bundle format<sec_ref_formats>`, dm-verity must be supported as well.

In kernel Kconfig you have to enable the following options as either built-in (y) or module (m):

CONFIG_MD
CONFIG_BLK_DEV_DM
CONFIG_BLK_DEV_LOOP
CONFIG_DM_VERITY
CONFIG_SQUASHFS
CONFIG_CRYPTO_SHA256

For streaming support, you have to add CONFIG_BLK_DEV_NBD.

Note

Streaming uses the NBD netlink API, which was introduced with kernel version v4.12 (released 2017-07-12). As of 2023, all LTS releases on kernel.org support this API.

For encryption support, you have to add CONFIG_DM_CRYPT, CONFIG_CRYPTO_AES.

Note

These drivers may also be loaded as modules. Kernel versions v5.0 to v5.7 will require the patch 7e81f99afd91c937f0e66dc135e26c1c4f78b003 backporting to fix a bug where the bundles cannot be mounted in a small number of cases.

Note

On ARM SoCs, there are optimized alternative SHA256 implementations available (for example CONFIG_CRYPTO_SHA2_ARM_CE, CRYPTO_SHA256_ARM or hardware accellerators such as CONFIG_CRYPTO_DEV_FSL_CAAM_AHASH_API).

Required Host Tools

To be able to generate bundles, RAUC requires at least the following host tools:

  • mksquashfs
  • unsquashfs

When using the RAUC casync integration, the casync tool and fakeroot (for converting archives to directory tree indexes) must also be available.

Required Target Tools

RAUC requires and uses a set of target tools depending on the type of supported storage and used image type.

Mandatory tools for each setup are mount and umount, either from Busybox or util-linux

Note that build systems may handle parts of these dependencies automatically, but also in this case you will have to select some of them manually as RAUC cannot fully know how you intend to use your system.

NAND Flash:

flash_erase & nandwrite (from mtd-utils)
NOR Flash:

flash_erase & flashcp (from mtd-utils)
UBIFS:

mkfs.ubifs (from mtd-utils)
TAR archives:

You may either use GNU tar or Busybox tar.

If you intend to use Busybox tar, make sure format autodetection and also the compression formats you use are enabled:

  • CONFIG_FEATURE_TAR_AUTODETECT=y
  • CONFIG_FEATURE_TAR_LONG_OPTIONS=y
  • select needed CONFIG_FEATURE_SEAMLESS_*=y options
ext4:

mkfs.ext4 (from e2fsprogs)
vfat:

mkfs.vfat (from dosfstools)

Depending on the bootloader you use on your target, RAUC also needs the right tool to interact with it:

Barebox:barebox-state (from dt-utils)
U-Boot:fw_setenv/fw_getenv (from u-boot)
GRUB:grub-editenv
EFI:efibootmgr

Note that for running rauc info on the target (as well as on the host), you also need to have the unsquashfs tool installed.

When using the RAUC casync integration, the casync tool must also be available.

Interfacing with the Bootloader

RAUC provides support for interfacing with different types of bootloaders. To select the bootloader you have or intend to use on your system, set the bootloader key in the [system] section of your device's system.conf.

Note

If in doubt about choosing the right bootloader, we recommend to use barebox as it provides a dedicated boot handling framework, called bootchooser.

To let RAUC handle a bootable slot, you have to mark it as bootable in your system.conf and configure the name under which the bootloader identifies this specific slot. This is both done by setting the bootname property.

[slot.rootfs.0]
...
bootname=system0

Amongst others, the bootname property also serves as one way to let RAUC know which slot is currently booted (running). In the following, the different options for letting RAUC detect the currently booted slot are described.

Booted Slot Detection

For RAUC it is quite essential to know from which slot the system is currently running. We will refer this as the booted slot. Only reliable detection of the booted slot enables RAUC to determine the set of currently inactive slots (that it can safely write to).

If possible, one should always prefer to signal the active slot explicitly from the bootloader to the userspace and RAUC. Only for cases where this explicit way is not possible or unwanted, some alternative approaches of automatically detecting the currently booted slot are implemented in RAUC.

A detailed list of detection mechanism follows.

Identification via Kernel Commandline

RAUC evaluates different kernel commandline parameters in the order they are listed below.

rauc.slot= and rauc.external

This is the generic way to explicitly set information about which slot was booted by the bootloader. For slots that are handled by a bootloader slot selection mechanism (such as A+B slots) you should specify the slot's configured bootname:

rauc.slot=system0

For special cases where some slots are not handled by the slot selection mechanism (such as a 'last-resort' recovery fallback that never gets explicitly selected) you can also give the name of the slot:

rauc.slot=recovery.0

When booting from a source not configured in your system.conf (for example from a USB memory stick), you can tell rauc explicitly with the flag rauc.external. This means that all slots are known to be inactive and will be valid installation targets. A possible use case for this is to use RAUC during a bootstrapping procedure to perform an initial installation.

bootchooser.active=

This is the command-line parameter used by barebox's bootchooser mechanism. It will be set automatically by the bootchooser framework and does not need any manual configuration. RAUC compares this against each slot's bootname (not the slot's name as above):

bootchooser.active=system0

root=

If none of the above parameters is given, the root= parameter is evaluated by RAUC to gain information on the currently booted system. The root= entry contains the device from which device the kernel (or initramfs) should load the rootfs. RAUC supports parsing different variants for giving these device as listed below.

root=/dev/sda1
root=/dev/ubi0_1

Giving the plain device name is supported, of course.

Note

The alternative ubi rootfs format with root=ubi0:volname is currently unsupported. If you want to refer to UBI volumes via name in your system.conf, check the FAQ entry :ref:`faq-udev-symlinks`.

root=PARTLABEL=abcde
root=PARTUUID=01234
root=UUID=01234

Parsing the PARTLABEL, PARTUUID and UUID is supported, which allows referring to a special partition / file system without having to know the enumeration-dependent sdX name.

RAUC converts the value to the corresponding /dev/disk/by-* symlink name and then to the actual device name.

root=/dev/nfs

RAUC automatically detects NFS boots (by checking if this parameter is set in the kernel command line). There is no extra slot configuration needed for this as RAUC assumes it is safe to update all available slots in case the currently running system comes from NFS.

systemd.verity_root_data=

RAUC handles the systemd.verity_root_data= parameter the same as root= above. See the systemd-veritysetup-generator documentation for details.

Identification via custom backend

When using the custom bootloader backend and the information about the currently booted slot cannot be derived from the kernel command line, RAUC will try to query the custom bootloader backend to get this information.

See the :ref:`sec-custom-bootloader-backend` bootloader section on how to implement a custom bootloader handler.

Barebox

The Barebox bootloader, which is available for many common embedded platforms, provides a dedicated boot source selection framework, called bootchooser, backed by an atomic and redundant storage backend, named state.

Barebox state allows you to save the variables required by bootchooser with memory specific storage strategies in all common storage mediums, such as block devices, mtd (NAND/NOR), EEPROM, and UEFI variables.

The Bootchooser framework maintains information about priority and remaining boot attempts while being configurable on how to deal with them for different strategies.

To enable the Barebox bootchooser support in RAUC, select it in your system.conf:

[system]
...
bootloader=barebox

Configure Barebox

As mentioned above, Barebox support requires you to have the bootchooser framework with barebox state backend enabled. In Barebox' Kconfig you can enable this by setting:

CONFIG_BOOTCHOOSER=y
CONFIG_STATE=y
CONFIG_STATE_DRV=y

To debug and interact with bootchooser and state in Barebox, you should also enable these tools:

CONFIG_CMD_STATE=y
CONFIG_CMD_BOOTCHOOSER=y

Setup Barebox Bootchooser

The barebox bootchooser framework allows you to specify a number of redundant boot targets that should be automatically selected by an algorithm, based on status information saved for each boot target.

The bootchooser itself can be used as a Barebox boot target. This is where we start by setting the barebox default boot target to bootchooser:

nv boot.default="bootchooser"

Now, when Barebox is initialized it starts the bootchooser logic to select its real boot target.

As a next step, we need to tell bootchooser which boot targets it should handle. These boot targets can have descriptive names which must not equal any of your existing boot targets, we will have a mapping for this later on.

In this example we call the virtual bootchooser boot targets system0 and system1:

nv bootchooser.targets="system0 system1"

Now connect each of these virtual boot targets to a real Barebox boot target (one of its automagical ones or custom boot scripts):

nv bootchooser.system0.boot="mmc1.1"
nv bootchooser.system1.boot="mmc1.2"

Note

For most cases, no extra boot entry needs to be configured since barebox will match the the given boot target to the corresponding device, automatically mount it and attempt to read a matching bootloader specification (bootspec) entry from /loader/entries/.

To configure bootchooser to store the variables in Barebox state, you need to configure the state_prefix:

nv bootchooser.state_prefix="state.bootstate"

Beside this very basic configuration variables, you need to set up a set of other general and slot-specific variables.

Warning

It is highly recommended to read the full Barebox bootchooser documentation in order to know about the requirements and possibilities in fine-tuning the behavior according to your needs.

Also make sure to have these nv settings in your compiled-in environment, not in your device-local environment.

Setting up Barebox State for Bootchooser

For storing its status information, the bootchooser framework requires a barebox,state instance to be set up with a set of variables matching the set of virtual boot targets defined.

To allow loading the state information in a well-defined format both from Barebox and from the kernel, we store the state data format definition in the Barebox devicetree.

Barebox fixups the information into the Linux devicetree when loading the kernel. This assures having a consistent view on the variables in Barebox and Linux.

An example devicetree node for our simple redundant setup will have the following basic structure

state {
  bootstate {
    system0 {
    ...
    };
    system1 {
    ...
    };
  };
};

In the state node, we set the appropriate compatible to tell the barebox,state driver to care for it and define where and how we want to store our data. This will look similar to this:

state: state {
        magic = <0x4d433230>;
        compatible = "barebox,state";
        backend-type = "raw";
        backend = <&state_storage>;
        backend-stridesize = <0x40>;
        backend-storage-type = "circular";
        #address-cells = <1>;
        #size-cells = <1>;

        [...]
}

where <&state_storage> is a phandle to, e.g. an EEPROM or NAND partition.

Important

The devicetree only defines where and in which format the data will be stored. By default, no data will be stored in the deviectree itself!

The rest of the variable set definition will be made in the bootstate subnode.

For each virtual boot target handled by state, two uint32 variables remaining_attempts and priority need to be defined.:

bootstate {

        system0 {
                #address-cells = <1>;
                #size-cells = <1>;

                remaining_attempts@0 {
                        reg = <0x0 0x4>;
                        type = "uint32";
                        default = <3>;
                };
                priority@4 {
                        reg = <0x4 0x4>;
                        type = "uint32";
                        default = <20>;
                };
        };

        [...]
};

Note

As the example shows, you must also specify some useful default variables the state driver will load in case of uninitialized backend storage.

Additionally one single variable for storing information about the last chosen boot target is required:

bootstate {

        [...]

        last_chosen@10 {
                reg = <0x10 0x4>;
                type = "uint32";
        };
};

Warning

This example shows only a highly condensed excerpt of setting up Barebox state for bootchooser. For a full documentation on how Barebox state works and how to properly integrate it into your platform see the official Barebox State Framework user documentation as well as the corresponding devicetree binding reference!

You can verify your setup by calling devinfo state from Barebox, which would print this for example:

barebox@board:/ devinfo state
Parameters:
bootstate.last_chosen: 2 (type: uint32)
bootstate.system0.priority: 10 (type: uint32)
bootstate.system0.remaining_attempts: 3 (type: uint32)
bootstate.system1.priority: 20 (type: uint32)
bootstate.system1.remaining_attempts: 3 (type: uint32)
dirty: 0 (type: bool)
save_on_shutdown: 1 (type: bool)

Once you have set up bootchooser properly, you finally need to enable RAUC to interact with it.

Enable Accessing Barebox State for RAUC

For this, you need to specify which (virtual) boot target belongs to which of the RAUC slots you defined. You do this by assigning the virtual boot target name to the slots bootname property:

[slot.rootfs.0]
...
bootname=system0

[slot.rootfs.1]
...
bootname=system1

For writing the bootchooser's state variables from userspace, RAUC uses the tool barebox-state from the dt-utils repository.

Note

RAUC requires dt-utils version v2017.03 or later!

Make sure to have this tool integrated on your target platform. You can verify your setup by calling it manually:

# barebox-state -d
bootstate.system0.remaining_attempts=3
bootstate.system0.priority=10
bootstate.system1.remaining_attempts=3
bootstate.system1.priority=20
bootstate.last_chosen=2

Verify Boot Slot Detection

As detecting the currently booted rootfs slot from userspace and matching it to one of the slots defined in RAUC's system.conf is not always trivial and error-prone, Barebox provides an explicit information about which slot it selected for booting adding a bootchooser.active key to the commandline of the kernel it boots. This key has the virtual bootchooser boot target assigned. In our case, if the bootchooser logic decided to boot system0 the kernel commandline will contain:

bootchooser.active=system0

RAUC uses this information for detecting the active booted slot (based on the slot's bootname property).

If the kernel commandline of your booted system contains this line, you have successfully set up bootchooser to boot your slot:

$ cat /proc/cmdline

Enable Watchdog on Boot

When enabled, Barebox will automatically set up the configured watchdog when running the boot command.

To enable this, set the boot.watchdog_timeout variable, preferably in the environment:

nv boot.watchdog_timeout=10

U-Boot

To enable handling of redundant booting in U-Boot, manual scripting is required. U-Boot allows storing and modifying variables in its Environment. Properly configured, the environment can be accessed both from U-Boot itself as well as from Linux userspace. U-Boot also supports setting up the environment redundantly for atomic modifications.

The default RAUC U-Boot boot selection implementation requires a U-Boot boot script using specific set of variables that are persisted to the environment as stateful slot selection information.

To enable U-Boot support in RAUC, select it in your system.conf:

[system]
...
bootloader=uboot

Set up U-Boot Boot Script for RAUC

U-Boot as the bootloader needs to decide which slot (partition) to boot. For this decision it needs to read and process some state information set by RAUC or previous boot attempts.

The U-Boot bootloader interface of RAUC will rely on setting the following U-Boot environment variables:

BOOT_ORDER:Contains a space-separated list of boot names in the order they should be tried, e.g. A B.
BOOT_<bootname>_LEFT:Contains the number of remaining boot attempts to perform for the respective slot.

An example U-Boot script for handling redundant A/B boot setups is located in the contrib/ folder of the RAUC source repository (contrib/uboot.sh).

Note

You must adapt the script's boot commands to match the requirements of your platform.

You should integrate your boot selection script as boot.scr default boot script into U-Boot.

For this you have to convert it to a U-boot readable default script (boot.scr) first:

mkimage -A arm -T script -C none -n "Boot script" -d <path-to-input-script> boot.scr

If you place this on a partition next to U-Boot, it will use it as its boot script.

For more details, refer the U-Boot Scripting Capabilities chapter in the U-Boot user documentation.

The example script uses the names A and B as the bootname for the two different boot targets. These names need to be set in your system.conf as the bootname of the respective slots. The resulting boot attempts variables will be BOOT_A_LEFT and BOOT_B_LEFT. The BOOT_ORDER variable will contain A B if A is the primary slot or B A if B is the primary slot to boot.

Note

For minor changes in boot logic or variable names simply change the boot script and/or the RAUC system.conf bootname settings. If you want to implement a fully different behavior, you might need to modify the uboot_set_state() and uboot_set_primary() functions in src/bootchooser.c of RAUC.

Setting up the (Fail-Safe) U-Boot Environment

The U-Boot environment is used to store stateful boot selection information and serves as the interface between userspace and bootloader. The information stored in the environment needs to be preserved, even if the bootloader should be updated. Thus the environment should be placed outside the bootloader partition!

The storage location for the environment can be controlled with CONFIG_ENV_IS_IN_* U-Boot Kconfig options like CONFIG_ENV_IS_IN_FAT or CONFIG_ENV_IS_IN_MMC. You may either select a different storage than your bootloader, or a different location/partition/volume on the same storage.

For fail-safe (atomic) updates of the environment, U-Boot can use redundant environments that allow to write to one copy while keeping the other as fallback if writing fails, e.g. due to sudden power cut.

In order to enable redundant environment storage, you have to additionally set in your U-Boot config:

CONFIG_SYS_REDUNDAND_ENVIRONMENT=y
CONFIG_ENV_SIZE=<size-of-env>
CONFIG_ENV_OFFSET=<offset-in-device>
CONFIG_ENV_OFFSET_REDUND=<copy-offset-in-device>

Note

Above switches refer to U-Boot >= v2020.01.

Refer to U-Boot source code and README for more details on this.

Enable Accessing U-Boot Environment from Userspace

To enable reading and writing of the U-Boot environment from Linux userspace, you need to have:

  • U-Boot target tools fw_printenv and fw_setenv available on your devices rootfs.
  • Environment configuration file /etc/fw_env.config in your target root filesystem.

See the corresponding HowTo section from the U-Boot documentation for more details on how to set up the environment config file for your device.

Example: Setting up U-Boot Environment on eMMC/SD Card

For this example we assume a simple redundancy boot partition layout with a bootloader partition and two rootfs partitions.

Another additional partition we use exclusively for storing the environment.

Note

It is not strictly required to have the env on an actual MBR/GPT partition, but we use this here as it better protects against accidentally overwriting relevant data of other partitions.

Partition table (excerpt with partition offsets):

/dev/mmcblk0p1 StartLBA:   8192 -> u-boot etc.
/dev/mmcblk0p2 StartLBA: 114688 -> u-boot environment
/dev/mmcblk0p3 StartLBA: 139264 -> rootfs A
/dev/mmcblk0p4 StartLBA: 475136 -> rootfs B

We enable redundant environment and storage in MMC (not in vfat/ext4 partition) in the u-boot config:

CONFIG_SYS_REDUNDAND_ENVIRONMENT=y
CONFIG_ENV_IS_IN_MMC=y

The default should be to use mmc device 0 and HW partition 0. Since U-Boot 2020.10.0 we can set this also explicitly if required:

CONFIG_SYS_MMC_ENV_DEV=0
CONFIG_SYS_MMC_ENV_PART=0

Important

With CONFIG_SYS_MMC_ENV_PART we can specify a eMMC HW partition only, not an MBR/GPT partition! HW partitions are e.g. 0=user data area, 1=boot partition.

Then we must specify the env storage size and its offset relative to the currently used device. Here the device is the eMMC user data area (or SD Card). For placing the content in partition 2 now, we must calculate the offset as offset=hex(n sector * 512 bytes/sector). With n=114688 (start of /dev/mmcblk0p2 according to above partition table) we get an offset of 0x3800000. As size we pick 0x4000 (16kB) here. The offset of the redundant copy must be the offset of the first copy + size of first copy. This results in:

CONFIG_ENV_SIZE=0x4000
CONFIG_ENV_OFFSET=0x3800000
CONFIG_ENV_OFFSET_REDUND=0x3804000

Finally, we need to configure userspace to access the same location. This can be referenced directly by its partition device name (/dev/mmcblk0p2) in the /etc/fw_env.config:

/dev/mmcblk0p2 0x0000 0x4000
/dev/mmcblk0p2 0x4000 0x4000

GRUB

[system]
...
bootloader=grub

To enable handling of redundant booting in GRUB, manual scripting is required.

The GRUB bootloader interface of RAUC uses the GRUB environment variables <bootname>_OK, <bootname>_TRY and ORDER.

An exemplary GRUB configuration for handling redundant boot setups is located in the contrib/ folder of the RAUC source repository (grub.conf). As the GRUB shell only has limited support for scripting, this example uses only one try per enabled slot.

To enable reading and writing of the GRUB environment, you need to have the tool grub-editenv available on your target.

By default RAUC expects the grubenv file to be located at /boot/grub/grubenv, you can specify a custom directory by passing grubenv=/path/to/grubenv in your system.conf [system] section.

Make sure that the grubenv file is located outside your redundant rootfs partitions as the rootfs needs to be exchangeable without affecting the environment content. For UEFI systems, a proper location would be to place it on the EFI partition, e.g. at /EFI/BOOT/grubenv. The same partition can also be used for your grub.cfg (which could be placed at /EFI/BOOT/grub.cfg).

Note that you then also need to manually tell GRUB where to load the grubenv from. You can do this in your grub.cfg by a adding the --file argument to your script's load_env and save_env calls, like:

load_env --file=(hd0,2)/grubenv

save_env --file=(hd0,2)/grubenv A_TRY A_OK B_TRY B_OK ORDER

EFI

For x86 systems that directly boot via EFI/UEFI, RAUC supports interaction with EFI boot entries by using the efibootmgr tool. To enable EFI bootloader support in RAUC, write in your system.conf:

[system]
...
bootloader=efi

To set up a system ready for pure EFI-based redundancy boot without any further bootloader or initramfs involved, you have to create an appropriate partition layout and matching boot EFI entries.

Assuming a simple A/B redundancy, you would need:

  • 2 redundant EFI partitions holding an EFI stub kernel (e.g. at EFI/LINUX/BZIMAGE.EFI)
  • 2 redundant rootfs partitions

To create boot entries for these, use the efibootmgr tool:

efibootmgr --create --disk /dev/sdaX --part 1 --label "system0" --loader \\EFI\\LINUX\\BZIMAGE.EFI --unicode "root=PARTUUID=<partuuid-of-part-1>"
efibootmgr --create --disk /dev/sdaX --part 2 --label "system1" --loader \\EFI\\LINUX\\BZIMAGE.EFI --unicode "root=PARTUUID=<partuuid-of-part-2>"

where you replace /dev/sdaX with the name of the disk you use for redundancy boot, <partuuid-of-part-1> with the PARTUUID of the first rootfs partition and <partuuid-of-part-2> with the PARTUUID of the second rootfs partition.

You can inspect and verify your settings by running:

efibootmgr -v

In your system.conf, you have to list both the EFI partitions (each containing one kernel) as well as the rootfs partitions. Make the first EFI partition a child of the first rootfs partition and the second EFI partition a child of the second rootfs partition to have valid slot groups. Set the rootfs slot bootnames to those we have defined with the --label argument in the efibootmgr call above:

[slot.efi.0]
device=/dev/sdX1
type=vfat
parent=rootfs.0

[slot.efi.1]
device=/dev/sdX2
type=vfat
parent=rootfs.1

[slot.rootfs.0]
device=/dev/sdX3
type=ext4
bootname=system0

[slot.rootfs.1]
device=/dev/sdX4
type=ext4
bootname=system1

Custom

If none of the previously mentioned approaches can be applied on the system, RAUC also offers the possibility to use customization scripts or applications as bootloader backend.

To enable the custom bootloader backend support in RAUC, select it in your system.conf:

[system]
...
bootloader=custom

Configure custom bootloader backend

The custom bootloader backed based on a handler that is called to get the desired information or set the appropriate configuration of the custom bootloader environment.

To register the custom bootloader backend handler, assign your handler to the bootloader-custom-backend key in section handlers in your system.conf:

[handlers]
...
bootloader-custom-backend=custom-bootloader-script

Custom bootloader backend interface

According to :ref:`sec-boot-slot` the custom bootloader handler is called by RAUC to trigger the following actions:

  • get the primary slot
  • set the primary slot
  • get the boot state
  • set the boot state
  • get the current booted slot (optional)

To get the primary slot, the handler is called with the argument get-primary. The handler must output the current primary slot's bootname on the stdout, and return 0 on exit, if no error occurred. In case of failure, the handler must return with non-zero value. Accordingly, in order to set the primary slot, the custom bootloader handler is called with argument set-primary <slot.bootname> where <slot.bootname> matches the bootname= key defined for the respective slot in your system.conf. If the set was successful, the handler must also return with a 0, otherwise the return value must be non-zero.

In addition to the primary slot, RAUC must also be able to determine the boot state of a specific slot. RAUC determines the necessary boot state by calling the custom bootloader handler with the argument get-state <slot.bootname>. Whereupon the handler has to output the state good or bad to stdout and exit with the return value 0. If the state cannot be determined or another error occurs, the custom bootloader handler must exit with non-zero return value. To set the boot state to the desire slot, the handler is called with argument set-state <slot.bootname> <state>. As already mentioned in the paragraph above, the <slot.bootname> matches the bootname= key defined for the respective slot in your system.conf. The <state> argument corresponds to one of the following values:

  • good if the last start of the slot was successful or
  • bad if the last start of the slot failed.

The return value must be 0 if the boot state was set successfully, or non-zero if an error occurred.

To get the current running slot, the handler must be called with the argument get-current. The handler must output the current running slot's bootname on the stdout, and return 0 on exit, if no error occurred. Implementing this is only needed when the /proc/cmdline is not providing information about current booted slot.

Init System and Service Startup

There are several ways to run the RAUC service on your target. The recommended way is to use a systemd-based system and allow to start RAUC via D-Bus activation.

You can start the RAUC service manually by executing:

$ rauc service

Keep in mind that rauc service reads the system.conf during startup and needs to be restarted for changes in the system.conf to take affect.

Systemd Integration

When building RAUC, a default systemd rauc.service file will be generated in the data/ folder.

Depending on your configuration make install will place this file in one of your system's service file folders.

It is a good idea to wait for the system to be fully started before marking it as successfully booted. In order to achieve this, a smart solution is to create a systemd service that calls rauc status mark-good and use systemd's dependency handling to assure this service will not be executed before all relevant other services came up successfully. It could look similar to this:

[Unit]
Description=RAUC Good-marking Service
ConditionKernelCommandLine=|bootchooser.active
ConditionKernelCommandLine=|rauc.slot

[Service]
ExecStart=/usr/bin/rauc status mark-good

[Install]
WantedBy=multi-user.target

D-Bus Integration

The :ref:`D-Bus <sec_ref_dbus-api>` interface RAUC provides makes it easy to integrate it into your customapplication. In order to allow sending data, make sure the D-Bus config file de.pengutronix.rauc.conf from the data/ dir gets installed properly.

To only start RAUC when required, using D-Bus activation is a smart solution. In order to enable D-Bus activation, properly install the D-Bus service file de.pengutronix.rauc.service from the data/ dir.

Watchdog Configuration

Detecting system hangs during runtime requires to have a watchdog and to have the watchdog configured and handled properly. Systemd provides a sophisticated watchdog multiplexing and handling allowing you to configure separate timeouts and handlings for each of your services.

To enable it, you need at least to have these lines in your systemd configuration:

RuntimeWatchdogSec=20
ShutdownWatchdogSec=10min

Bundle Generation

Once RAUC is set up on the target, one might want to actually create update bundles for it.

Note

Some build systems provide a high-level integration that should be used, for example in :ref:`Yocto <sec-integration-yocto-bundle>` or :ref:`PTXdist <sec-integration-ptxdist-bundle>`.

For generating a bundle, at least the following items are required:

  • signing key and certificate
  • content directory with manifest file

The signing key and cert could be created for this specific project or be supplied from somewhere else in your project or company. They can be provided as PEM files or as PKCS#11 URIs (e.g. if you use a HSM). For evaluation purposes, you can also generate a self-signed key pair. Read the :ref:`sec-security` chapter for more details.

For the bundle content, simply create a new directory:

$ mkdir install-content

Copy each image that should be installed via the bundle into the content directory, for example:

$ cp /path/to/system-image.ext3 install-content/system-image.ext4
$ cp /path/to/barebox install-content/barebox.img

Note

Since RAUC uses the image's file name extension for determining the correct update handler, make sure that the file name extension used in the content directory is :ref:`supported <sec-ref-supported-image-types>`.

Create a manifest file called manifest.raucm in the content directory:

$ vi install-content/manifest.raucm

A minimal example for a manifest could look as follows:

[update]
compatible=Test Platform
version=2023.11.0

[bundle]
format=verity

[image.rootfs]
filename=system-image.ext4

[image.bootloader]
filename=barebox.img

Ensure that compatible matches the RAUC compatible in your target's system.conf. The system-image.ext4 image will now serve as the update image for the rootfs slot class while the barebox.img will be the update image for the bootloader slot class.

Finally, invoke RAUC to create the bundle from the created content directory:

$ rauc bundle --cert=cert.pem --key=key-pem install-content/ my-update.raucb

The resulting bundle my-update.raucb is the ready for being deployed to the target.

Yocto

Yocto support for using RAUC is provided by the meta-rauc layer.

The layer supports building RAUC both for the target as well as as a host tool. With the bundle.bbclass it provides a mechanism to specify and build bundles directly with the help of Yocto.

For more information on how to use the layer, also see the layer's README file.

Note

When using the block-hash-index adaptive mode, you may need to set IMAGE_ROOTFS_ALIGNMENT = "4" in your machine.conf to ensure that the image is padded to full 4 kiB blocks.

Target System Setup

Add the meta-rauc layer to your setup:

$ git submodule add git@github.com:rauc/meta-rauc.git

Add the RAUC tool to your image recipe (or package group):

IMAGE_INSTALL_append = "rauc"

Append the RAUC recipe from your BSP layer (referred to as meta-your-bsp in the following) by creating a meta-your-bsp/recipes-core/rauc/rauc_%.bbappend with the following content:

FILESEXTRAPATHS_prepend := "${THISDIR}/files:"

Write a system.conf for your board and place it in the folder you mentioned in the recipe (meta-your-bsp/recipes-core/rauc/files). This file must provide a system compatible string to identify your system type, as well as a definition of all slots in your system. By default, the system configuration will be placed in /etc/rauc/system.conf on your target rootfs.

Also place the appropriate keyring file for your target into the directory added to FILESEXTRAPATHS above. Name it either ca.cert.pem or additionally specify the name of your custom file by setting RAUC_KEYRING_FILE. If multiple keyring certificates are required on a single system, create a keyring directory containing each certificate.

Note

For information on how to create a testing / development key/cert/keyring, please refer to scripts/README in meta-rauc.

For a reference of allowed configuration options in system.conf, see :ref:`sec_ref_slot_config`. For a more detailed instruction on how to write a system.conf, see :ref:`sec-int-system-config`.

Using RAUC on the Host System

The RAUC recipe allows to compile and use RAUC on your host system. Having RAUC available as a host tool is useful for debugging, testing or for creating bundles manually. For the preferred way of creating bundles automatically, see the chapter :ref:`sec-integration-yocto-bundle`. In order to compile RAUC for your host system, simply run:

$ bitbake rauc-native

This will place a copy of the RAUC binary in tmp/deploy/tools in your current build folder. To test it, try:

$ tmp/deploy/tools/rauc --version

Bundle Generation

Bundles can be created either manually by building and using RAUC as a native tool, or by using the bundle.bbclass that handles most of the basic steps, automatically.

First, create a bundle recipe in your BSP layer. A possible location for this could be meta-your-bsp/recipes-core/bundles/update-bundle.bb.

To create your bundle you first have to inherit the bundle class:

inherit bundle

To create the manifest file, you may either use the built-in class mechanism, or provide a custom manifest.

For using the built-in bundle generation, you need to specify some variables:

RAUC_BUNDLE_COMPATIBLE
Sets the compatible string for the bundle. This should match the compatible you specified in your system.conf or, more generally, the compatible of the target platform you intend to install this bundle on.
RAUC_BUNDLE_SLOTS
Use this to list all slot classes for which the bundle should contain images. A value of "rootfs appfs" for example will create a manifest with images for two slot classes; rootfs and appfs.
RAUC_BUNDLE_FORMAT
Use this to choose the :ref:`sec_ref_formats` for the generated bundle. It currently defaults to plain, but you should use verity if possible.
RAUC_SLOT_<slotclass>
For each slot class, set this to the recipe name which builds the image you intend to place in the slot class.
RAUC_SLOT_<slotclass>[type]
For each slot class, set this to the type of image you intend to place in this slot. Possible types are: image (default), kernel, boot, or file.

Note

For a full list of supported variables, refer to classes-recipe/bundle.bbclass in meta-rauc.

A minimal bundle recipe, such as core-bundle-minimal.bb that is contained in meta-rauc will look as follows:

inherit bundle

RAUC_BUNDLE_COMPATIBLE ?= "Demo Board"

RAUC_BUNDLE_SLOTS ?= "rootfs"

RAUC_BUNDLE_FORMAT ?= "verity"

RAUC_SLOT_rootfs ?= "core-image-minimal"

To be able to build a signed image of this, you also need to configure RAUC_KEY_FILE and RAUC_CERT_FILE to point to your key and certificate files you intend to use for signing. You may set them either from your bundle recipe or any global configuration (layer, site.conf, etc.), e.g.:

RAUC_KEY_FILE = "${COREBASE}/meta-<layername>/files/development-1.key.pem"
RAUC_CERT_FILE = "${COREBASE}/meta-<layername>/files/development-1.cert.pem"

Note

For information on how to create a testing / development key/cert/keyring, please refer to scripts/README in meta-rauc.

Based on this information, a call of:

$ bitbake core-bundle-minimal

will build all required images and generate a signed RAUC bundle from this. The created bundle can be found in ${DEPLOY_DIR_IMAGE} (defaults to tmp/deploy/images/<machine> in your build directory).

PTXdist

Note

RAUC support in PTXdist is available since version 2017.04.0.

Integration into Your RootFS Build

To enable building RAUC for your target, set:

CONFIG_RAUC=y

in your ptxconfig (by selecting RAUC via ptxdist menuconfig).

You should also customize the compatible RAUC uses for your system. To do this, set PTXCONF_RAUC_COMPATIBLE to a string that uniquely identifies your device type. The default value will be "${PTXCONF_PROJECT_VENDOR}\ ${PTXCONF_PROJECT}".

Place your system configuration file in $(PTXDIST_PLATFORMCONFIGDIR)/projectroot/etc/rauc/system.conf to let the RAUC package install it into the rootfs you build.

Note

PTXdist versions since 2020.06.0 use their code signing infrastructure for keyring creation. See PTXdist's Managing Certificate Authority Keyrings for different scenarios (refer to RAUC's :ref:`sec-ca-configuration`). Previous PTXdist versions expected the keyring in $(PTXDIST_PLATFORMCONFIGDIR)/projectroot/etc/rauc/ca.cert.pem. The keyring is installed into the rootfs to /etc/rauc/ca.cert.pem.

If using systemd, the recipes install both the default systemd.service file for RAUC as well as a rauc-mark-good.service file. This additional good-marking-service runs after user space is brought up and notifies the underlying bootloader implementation about a successful boot of the system. This is typically used in conjunction with a boot attempts counter in the bootloader that is decremented before starting the system and reset by rauc status mark-good to indicate a successful system startup.

Create Update Bundles from your RootFS

To enable building RAUC bundles, set:

CONFIG_IMAGE_RAUC=y

in your platformconfig (by using ptxdist platformconfig).

This adds a default image recipe for building a RAUC update bundle out of the system's rootfs. As for most image recipes, the genimage tool is used to configure and generate the update bundle.

PTXdist's default bundle configuration is placed in config/images/rauc.config. You may also copy this to your platform directory to use this as a base for custom bundle configuration.

RAUC enforces signing of update bundles. PTXdist versions since 2020.06.0 use its code signing infrastructure for signing and keyring verification. Previous versions expected the signing key in $(PTXDIST_PLATFORMCONFIGDIR)/config/rauc/rauc.key.pem.

Once you are done with your setup, PTXdist will automatically create a RAUC update bundle for you during the run of ptxdist images. It will be placed under $(PTXDIST_PLATFORMDIR)/images/update.raucb.

Buildroot

Note

RAUC support in Buildroot is available since version 2017.08.0.

To build RAUC using Buildroot, enable BR2_PACKAGE_RAUC in your configuration.

Package-Based Distributions

Some non-embedded-focused distributions provide RAUC packages. An overview can be found on Repology.

Note that some distributions split the service configuration in a separate rauc-service package, as the common use of RAUC on these distributions is to create and inspect bundles, for which the D-Bus service is not required.

Bundle Format Migration

Migrating from the plain to the verity :ref:`bundle format <sec_ref_formats>` should be simple in most cases and can be done in a single update. The high-level functionality of RAUC (certificate checking, update installation, hooks/handlers, …) is independent of the low-level bundle format.

The required steps are:

  • Configure your build system to build RAUC v1.5 (or newer).

  • Enable CONFIG_CRYPTO_SHA256, CONFIG_MD, CONFIG_BLK_DEV_DM and CONFIG_DM_VERITY in your kernel configuration. These may already be enabled if you are using dm-verity for verified boot.

  • Add a new bundle output configured for the verity format by adding the following to the manifest:

    [bundle]
format=verity

Note

For OE/Yocto with an up-to-date meta-rauc, you can choose the bundle format by adding the RAUC_BUNDLE_FORMAT = "verity" option in your bundle recipe. The bundle.bbclass will insert the necessary option into the manifest.

For PTXdist or Buildroot with genimage, you can add the manifest option above to the template in your genimage config file.

With these changes, the build system should produce two bundles (one in either format). A verity bundle will only be installable on systems that have already received the migration update. A plain bundle will be installable on both migrated and unmigrated systems.

You should then test that both bundle formats can be installed on a migrated system, as RAUC will now perform additional checks when installing a plain bundle to protect against potential modification during installation. This testing should include all bundle sources (USB, network, …) that you will need in the field to ensure that these new checks don't trigger in your case (which would prohibit further updates).

Note

When installing bundles from a FAT filesystem (for example on a USB memory stick), check that the mount option fmask is set to 0022 or 0133.

When you no longer need to be able to install previously built bundles in the plain format, you should also disable it in the system.conf:

[system]
…
bundle-formats=-plain
…

If you later need to support downgrades, you can use rauc extract and rauc bundle to convert a plain bundle to a verity bundle, allowing installation to systems that have already been migrated.