Figuring out the container runtime you are in

Containers are meant to keep processes contained. But there are ways to gather information about the host – like the actual execution environment you are running in.

Containers are pretty good at keeping everyone and everything inside their boundaries – thanks to SELinux, namespaces and so on. But they are not perfect. Thanks to a recent Azure security flaw I was made aware of a nice trick via the /proc/ file system to figure out what container runtime the container is running in.

The idea is that the started container inherits some /proc/ entries – among the entry for /proc/self/. If we are able to load a malicious container, we can use this knowledge to execute the container host binary and by that get information about the runtime.

As an example, let’s take a Fedora container image with podman (and all it’s library dependencies) installed. We can run it and check the version of crun inside:

❯ podman run --rm podman:v3.2.0 crun --version
crun version 0.20.1
commit: 0d42f1109fd73548f44b01b3e84d04a279e99d2e
spec: 1.0.0

Note that I take an older version here to better highlight the difference between host and container binary!

If we now change the execution to /proc/self/exe, which points to the host binary, the result shows a different version:

❯ podman run --rm podman:v3.2.0 /proc/self/exe --version
crun version 1.0
commit: 139dc6971e2f1d931af520188763e984d6cdfbf8
spec: 1.0.0

This is actually the version string of the binary on the host system:

❯ /usr/bin/crun --version
crun version 1.0
commit: 139dc6971e2f1d931af520188763e984d6cdfbf8
spec: 1.0.0

With this nice little feature we now have insight into the version used – and as it was shown in the detailed write-up of the Azure security problem mentioned above, this insight can be crucial to identify and use the right CVEs to start lateral movement.

Of course this requires us to be able to load containers of our choice, and to have the right libraries inside the container. This example is simplified because I knew about the target system and there was a container available with everything I needed. For a more realistic attack container image, check out whoc.

[Howto] Using the new Podman API

Podman is a daemonless container engine to develop, run and manage OCI containers. In a recent version the API was rewritten and now offers a REST interface as well as a docker compatible endpoint.

Podman is a daemonless container engine to develop, run and manage OCI containers. In a recent version the API was rewritten and now offers a REST interface as well as a docker compatible endpoint.

In case you never heard of Podman before, it is certainly worth a look. Besides offering a more secure drop-in-replacement for many docker functions, it can also manage pods and thus provides a container experience more aligned with what Kubernetes uses. It even can understand Kubernetes yaml (see podman-play-kube), easing the transition from single host container development over to fully fledged container management environments. Last but not least it is among the tools supporting newest features in the container space like cgroups v2.

Background: Podman API

Of course Podman is not perfect – due to the focus on Kubernetes yaml there is no support for docker-compose files (though alternatives exist), networking and routing based on names is not as simple as on Docker (read more about Podman container networking) and last but not least, the API was different – making it hard to migrate solutions dependent on the docker API.

This changed: recently, a new API was merged:

The new API is a simpler implementation based on HTTP/REST. We provide two basic groups of endpoints. The first one is for libpod; the second is for Docker compatibility, to ease adoption. 

New API coming for Podman

So how can I access the new API and fool around with it?

If you are familiar with Podman, or read carefully, the first question is: where is this API running if Podman is daemonless? And in fact, an API service needs to be started explicitly:

$ podman system service --timeout 5000

This starts the API on a UNIX socket. Other options, like a TCP socket or to run this without a timeout are also possible, the documentation provides examples.

How to use the Docker API endpoint

Let’s use the Docker API endpoint. To talk to a UNIX socket based REST API a recent curl (version >= 7.40) is quite helpful:

$ curl --unix-socket /$XDG_RUNTIME_DIR/podman/podman.sock http://localhost/images/json
[{"Containers":1,"Created":1583300892,"Id":"8c2e0da7c436e45be5ebf2adf26b41d13939190bd186214a4d45c30485071f9f","Labels":{"license":"MIT","name":"fedora","vendor":"Fedora Project","version":"31"},"ParentId":...

Note that here we are speaking to the rootless container, thus the unix domain socket is in the user runtime directory. Also, localhost has to be provided in the URL for very recent curl versions, otherwise it does not output anything!

The answer is a JSON listing, which is not easily readable. Simplify it with the help of Python (and silence curl info with the silent flag):

$ curl -s --unix-socket /$XDG_RUNTIME_DIR/podman/podman.sock http://localhost/containers/json|python -m json.tool
        "Id": "4829e030ab1beb83db07dbc5e51481cb66562f57b79dd9eb3069dfcde91019ed",
        "Names": [

So what can you do with the API? Podman tries to recreate most of the docker API, so you can basically use the docker API documentation to see what should be possible. Note though that not all API endpoints are supported since Podman does not provide all functions Docker offers.

How to use the Podman API endpoint

As mentioned the API does provide two endpoints: the Docker endpoint, and a Podman specific endpoint. This second API is necessary for multiple reasons: first, Podman has functions which are alien to Docker and thus not part of the Docker API. The pod function is the most notable here. Another reason is that an independent API enables the Podman developers to further innovate in their own way and velocity, and to change the API when needed or wanted.

The API for Podman can be reached via curl as mentioned above. However, there are two notable differences: first, the Podman endpoint is marked via an additional “podman” string in the API URI, and second the Podman API is always versioned. To list the images as shown above, but via podman’s own API, the following call is necessary:

$ curl -s --unix-socket /$XDG_RUNTIME_DIR/podman/podman.sock http://localhost/v1.24/libpod/images/json
[{"Id":"8c2e0da7c436e45be5ebf2adf26b41d13939190bd186214a4d45c30485071f9f","RepoTags":[""],"Created":1583300892,"Size":199632198,"Labels":{"license":"MIT","name":"fedora","vendor":"Fedora ...

For pods, the endpoint is for example /pods instead of /images:

$ curl -s --unix-socket /$XDG_RUNTIME_DIR/podman/podman.sock http://localhost/v1.24/libpod/pods/json|python -m json.tool
        "Cgroup": "user.slice",
        "Containers": [
                "Id": "1510dca23d2d15ae8be1eeadcdbfb660cbf818a69d5780705cd6535d97a4a578",
                "Names": "wonderful_ardinghelli",
                "Status": "running"
                "Id": "6c05c20a42e6987ac9f78b277a9d9152ab37dd05e3bfd5ec9e675979eb93bf0e",
                "Names": "eff81a37b4b8-infra",
                "Status": "running"
        "Created": "2020-04-19T21:45:17.838549003+02:00",
        "Id": "eff81a37b4b85e92916613239001cddc2ba42f3595236586f7462492be0ac5fc",
        "InfraId": "6c05c20a42e6987ac9f78b277a9d9152ab37dd05e3bfd5ec9e675979eb93bf0e",
        "Name": "testme",
        "Namespace": "",
        "Status": "Running"

Currently there is no documentation of the API available – or at least none of the level of the current Docker API documentation. But hopefully that will change soon.


Podman providing a Docker API is a great step for people who are dependent on the Docker API but nevertheless want switch to Podman. But providing a unique, but simple to consume REST API for Podman itself is equally great because it makes it easy to integrate Podman processes into existing tools and frameworks.

Just don’t forget that the API is still in development!

Featured image by Magnascan from Pixabay