Dockerfiles for seL4, CAmkES, and L4v dependencies

Requirements

• docker (See here or here for instructions)
• make

It is recommended you add yourself to the docker group, so you can run docker commands without using sudo.

Quick start

To get a running build environment for sel4 and camkes, run:

git clone https://github.com/SEL4PROJ/seL4-CAmkES-L4v-dockerfiles.git
cd seL4-CAmkES-L4v-dockerfiles
make user

Or to map a particular directory to the /host dir in the container:

make user HOST_DIR=/scratch/sel4_stuff  # as an example

What is this?

This repository contains dockerfiles which map out the dependencies for seL4, CAmkES, and L4v. It also contains some infrastructure to allow people to use the containers in a useful way.

These dockerfiles are used as the basis for regression testing in the Trustworthy Systems group, and hence should represent a well tested and up to date environment

To run

Get the repository of Dockerfiles by cloning them from GitHub:

git clone https://github.com/SEL4PROJ/seL4-CAmkES-L4v-dockerfiles.git
cd seL4-CAmkES-L4v-dockerfiles

To get an environment within the container, run:

make user

which will give you a terminal with CAmkES dependencies built. You can be more specific with:

make user_sel4
make user_camkes  # alias for 'make user'
make user_l4v

The container will map the current working directory from the host to /host within the container. You should be able to read and write files, as the container copies your username and UID.

If you want to map a different folder, you can specify it on the command line:

make user_sel4 HOST_DIR=/scratch/sel4_stuff

You can also specify commands to be executed inside the container by using EXEC:

make user EXEC="bash -c 'echo hello world'"

The images will be pulled from DockerHub if your machine does not have them.

Alternately, you can define a bash function in your bashrc, such as this:

function container() {
    if [[ $# > 0 ]]; then
        make -C /<path>/<to>/seL4-CAmkES-L4v-dockerfiles user HOST_DIR=$(pwd) EXEC="bash -c '""$@""'"
    else
        make -C /<path>/<to>/seL4-CAmkES-L4v-dockerfiles user HOST_DIR=$(pwd)
    fi
}

Where you replace the path to match where you cloned the git repo of the docker files. This then allows you to run:

container

to start the container interactively in your current directory, or:

container "echo hello && echo world"

to execute commands in the container in your current directory.

Example of compiling seL4 test

Start by creating a new workspace on your machine:

mkdir ~/sel4test

Start up the container:

make user HOST_DIR=~/sel4test
# in-container terminal
jblogs@in-container:/host$

Get seL4 test:

jblogs@in-container:/host$ repo init -u https://github.com/seL4/sel4test-manifest.git
jblogs@in-container:/host$ repo sync
jblogs@in-container:/host$ ls
apps configs Kbuild Kconfig kernel libs Makefile projects tools

Compile and simulate seL4 test for x86-64:

jblogs@in-container:/host$ mkdir build-x86
jblogs@in-container:/host$ cd build-x86
jblogs@in-container:/host$ ../init-build.sh -DPLATFORM=x86_64 -DSIMULATION=TRUE
jblogs@in-container:/host$ ninja
# ... time passes...
jblogs@in-container:/host$ ./simulate
...
Test VSPACE0002 passed

    </testcase>

    <testcase classname="sel4test" name="Test all tests ran">

    </testcase>

</testsuite>

All is well in the universe

Adding dependencies

The images and dockerfiles for seL4/CAmkES/L4v only specify enough dependencies to pass the tests in the *tests.dockerfile. The extras.dockerfile acts as a shim between the DockerHub images and the user.dockerfile.

Adding dependencies into the extras.dockerfile will build them the next time you run make user, and then be cached from then on.

To build the local dockerfiles

To build the Dockerfiles locally, you will need to use the included build.sh script. It has a help menu:

./build.sh -h
    build.sh [-r] -b [sel4|camkes|l4v] -s [binary_decomp|cakeml|camkes_vis|cogent|riscv|rust|sysinit|] -s ... -e MAKE_CACHES=no -e ...

     -r     Rebuild docker images (don't use the docker cache)
     -v     Verbose mode
     -s     Strict mode
     -e     Build arguments (NAME=VALUE) to docker build. Use a -e for each build arg.
     -p     Pull base image first. Rather than build the base image,
            get it from the web first

    Sneaky hints:
     - To build 'prebuilt' images, you can run:
           build.sh -b [riscv|cakeml]
       but it will take a while!
     - You can actually run this with '-b sel4-rust', or any other existing image,
       but it will ruin the sorting of the name.

Example builds

To build the seL4 image, run:

./build.sh -b sel4

Note that the -b flag stands for the base image. There are 3 base images: sel4, camkes, and l4v. Each base image includes the previous one, i.e.: the camkes image has everything the sel4 image has, plus the camkes dependencies.

To add additional software to the image, you can use the -s flag, to add software. For example:

./build.sh -b sel4 -s riscv # This adds the RISCV compilers

./build.sh -b sel4 -s riscv -s rust # This adds the RISCV compilers and a rust compiler

You can also pass configuration variables through to docker (in docker terms, these are BUILD_ARGS) by using the -e flag. For example, you can tell the seL4 image to use GCC version 6 (instead of 8 ) by running:

./build.sh -b sel4 -e GCC_V8_AS_DEFAULT=no

Or to turn off priming the build caches:

./build.sh -b sel4 -e MAKE_CACHES=no

Or combined:

./build.sh -b sel4 -e GCC_V8_AS_DEFAULT=no -e MAKE_CACHES=no

To speed things up, you can ask to pull the base image from DockerHub first with the -p flag:

./build.sh -p -b sel4 -s riscv # This adds the RISCV compilers

Security

Running Docker on your machine has its own security risks which you should be aware of. Be sure to read the Docker documentation.

Of particular note in this case, your UID and GID are being baked into an image. Any other user on the host who is part of the docker group could spawn a separate container of this image, and hence have read and write access to your files. Of course, if they are part of the docker group, they could do this anyway, but it just makes it a bit easier.

Use at your own risk.

Released images on DockerHub

The Trustworthy Systems group pushes "known working" images to DockerHub under the trustworthysystems/ DockerHub organisation. Images with the :latest tag are the ones currently in use in the Trustworthy Systems regression system, and so are considered to be "known working". Furthermore, each time an image is pushed out, it is tagged with a YYYY_MM_DD formatted date.

To ensure (fairly) reproducible builds of docker images, the images are built using Debian Snapshot (an apt repository that can be pinned to a date in time). When changes are made to the scripts or Dockerfiles in this repo, they are built against a "known working" date of Debian Snapshot - in other words, a date in which we were able to build all the Docker images, and they passed all of our tests. This avoids issues where something in Debian Testing or Unstable has changed and causes apt conflicts, or a newer version breaks the seL4 build process.

Internally, the Trustworthy Systems regression system will, once a week, attempt to build the docker images using regular apt (not using Snapshot), and if successful, will update the "known working" date. This means on the next build of the docker images that gets pushed out will be using this bumped Snapshot date. Typically, the further in time we get from a Debian release, the more packages we need to fetch from Testing or Unstable, and as such, the less likely this automatic bumping is to work, due to above mentioned issues. With some human intervention, it can usually be fixed up fairly easily. However, even without intervention, the "known working" images will continue to function and build.