Security in Singularity Containers

Containers are all the rage today for many good reasons. They are light weight, easy to spin-up and require reduced IT management resources as compared to hardware VM environments. More importantly, container technology facilitates advanced research computing by granting the ability to package software in highly portable and reproducible environments encapsulating all dependencies, including the operating system. But there are still some challenges to container security.

Singularity, which is a container paradigm created by necessity for scientific and application driven workloads, addresses some core missions of containers : Mobility of Compute, Reproducibility, HPC support, and Security. This document intends to inform admins of different security features supported by Singularity.

Singularity Runtime

The Singularity runtime enforces a unique security model that makes it appropriate for untrusted users to run untrusted containers safely on multi-tenant resources. Because the Singularity runtime dynamically writes UID and GID information to the appropriate files within the container, the user remains the same inside and outside the container, i.e., if you’re an unprivileged user while entering the container, you’ll remain an unprivileged user inside the container. A privilege separation model is in place to prevent users from escalating privileges once they are inside of a container. The container file system is mounted using the nosuid option, and processes are spawned with the PR_NO_NEW_PRIVS flag. Taken together, this approach provides a secure way for users to run containers and greatly simplifies things like reading and writing data to the host system with appropriate ownership.

It is also important to note that the philosophy of Singularity is Integration over Isolation. Most container run times strive to isolate your container from the host system and other containers as much as possible. Singularity, on the other hand, assumes that the user’s primary goals are portability, reproducibility, and ease of use and that isolation is often a tertiary concern. Therefore, Singularity only isolates the mount namespace by default, and will also bind mount several host directories such as $HOME and /tmp into the container at runtime. If needed, additional levels of isolation can be achieved by passing options causing Singularity to enter any or all of the other kernel namespaces and to prevent automatic bind mounting. These measures allow users to interact with the host system from within the container in sensible ways.

Singularity Image Format (SIF)

Sylabs addresses container security as a continuous process. It attempts to provide container integrity throughout the distribution pipeline.. i.e., at rest, in transit and while running. Hence, the SIF has been designed to achieve these goals.

A SIF file is an immutable container runtime image. It is a physical representation of the container environment itself. An important component of SIF that elicits security feature is the ability to cryptographically sign a container, creating a signature block within the SIF file which can guarantee immutability and provide accountability as to who signed it. Singularity follows the OpenPGP standard to create and manage these keys. After building an image within Singularity, users can singularity sign the container and push it to the Library along with its public PGP key(Stored in Keystore) which later can be verified (singularity verify) while pulling or downloading the image. This feature in particular protects collaboration within and between systems and teams.

SIF Encryption

In Singularity 3.4 and above the container root filesystem that resides in the squashFS partition of a SIF can be encrypted, rendering it’s contents inaccessible without a secret. Unlike other platforms, where encrypted layers must be assembled into an unencrypted runtime directory on disk, Singularity mounts the encrypted root file system directly from the SIF using Kernel dm-crypt/LUKS functionality, so that the content is not exposed on disk. Singularity containers provide a comparable level of security to LUKS2 full disk encryption commonly deployed on Linux server and desktop systems.

As with all matters of security, a layered approach must be taken and the system as a whole considered. For example, it is possible that decrypted memory pages could be paged out the system swap file or device, which could result in decrypted information being stored at rest on physical media. Operating system level mitigations such as encrypted swap space may be required depending on the needs of your application.

Encryption and decryption of containers requires cryptsetup version 2. The SIF root filesystem will be encrypted using the default LUKS cipher on the host. The current default cipher used by cryptsetup for LUKS2 in mainstream Linux distributions is aes-xts-plain64 with a 256 bit key size. The default key derivation function for LUKS2 is argon2i.

Singularity currently supports 2 types of secrets for encrypted containers:

• Passphrase: a text passphrase is passed directly to cryptsetup for LUKS encryption of the root fs.

• Asymmetric RSA keypair: a randomly generated 256-bit secret is used to perform LUKS encryption of the rootfs. This secret is encrypted with a user-provided RSA public key, and the ciphertext stored in the SIF file. At runtime the RSA private key must be provided to decrypt the secret and allow decryption of the root filesystem to use the container.

Note

You can verify the default LUKS2 cipher and PBKDF on your system by running cryptsetup --help.

cryptsetup sets a memory cost for the argon2i PBKDF based on the RAM available in the system used for encryption, up to a maximum of 1GiB. Encrypted containers created on systems with >2GiB RAM may be unusable on systems with <1GiB of free RAM.

Admin Configurable Files

Singularity Administrators have the ability to access various configuration files, that will let them set security restrictions, grant or revoke a user’s capabilities, manage resources and authorize containers etc. One such file interesting in this context is ecl.toml which allows blacklisting and whitelisting of containers. You can find all the configuration files and their parameters documented here.

cgroups support

Starting v3.0, Singularity added native support for cgroups, allowing users to limit the resources their containers consume without the help of a separate program like a batch scheduling system. This feature helps in preventing DoS attacks where one container seizes control of all available system resources in order to stop other containers from operating properly. To utilize this feature, a user first creates a configuration file. An example configuration file is installed by default with Singularity to provide a guide. At runtime, the --apply-cgroups option is used to specify the location of the configuration file and cgroups are configured accordingly. More about cgroups support here.

--security options

Singularity supports a number of methods for specifying the security scope and context when running Singularity containers. Additionally, it supports new flags that can be passed to the action commands; shell, exec, and run allowing fine grained control of security. Details about them are documented here.

Security in SCS

Singularity Container Services (SCS) consist of a Remote Builder, a Container Library, and a Keystore. Taken together, the Singularity Container Services provide an end-to-end solution for packaging and distributing applications in secure and trusted containers.

Remote Builder

As mentioned earlier, the Singularity runtime prevents executing code with root-level permissions on the host system. But building a container requires elevated privileges that most production environments do not grant to users. The Remote Builder solves this challenge by allowing unprivileged users a service that can be used to build containers targeting one or more CPU architectures. System administrators can use the system to monitor which users are building containers, and the contents of those containers. Starting with Singularity 3.0, the CLI has native integration with the Build Service from version 3.0 onwards. In addition, a web GUI interface to the Build Service also exists, which allows users to build containers using only a web browser.

Note

Please see the Fakeroot feature which is a secure option for admins in multi-tenant HPC environments and similar use cases where they might want to grant a user special privileges inside a container.

Container Library

The Container Library enables users to store and share Singularity container images based on the Singularity Image Format (SIF). A web front-end allows users to create new projects within the Container Library, edit documentation associated with container images, and discover container images published by their peers.

Key Store

The Key Store is a key management system offered by Sylabs that utilizes OpenPGP implementation to facilitate sharing and maintaining of PGP public keys used to sign and verify Singularity container images. This service is based on the OpenPGP HTTP Keyserver Protocol (HKP), with several enhancements:

• The Service requires connections to be secured with Transport Layer Security (TLS).

• The Service implements token-based authentication, allowing only authenticated users to add or modify PGP keys.

• A web front-end allows users to view and search for PGP keys using a web browser.

Security Considerations of Cloud Services:

1. Communications between users, the auth service and the above-mentioned services are secured via TLS.

2. The services support authentication of users via authentication tokens.

3. There is no implicit trust relationship between Auth and each of these services. Rather, each request between the services is authenticated using the authentication token supplied by the user in the associated request.

4. The services support MongoDB authentication as well as TLS/SSL.

Note

SingularityPRO is a professionally curated and licensed version of Singularity that provides added security, stability, and support beyond that offered by the open source project. Subscribers receive advanced access to security patches through regular updates so, when a CVE is announced publicly PRO subscribers are already using patched software.

Security is not a check box that one can tick and forget. It’s an ongoing process that begins with software architecture, and continues all the way through to ongoing security practices. In addition to ensuring that containers are run without elevated privileges where appropriate, and that containers are produced by trusted sources, users must monitor their containers for newly discovered vulnerabilities and update when necessary just as they would with any other software. Sylabs is constantly probing to find and patch vulnerabilities within Singularity, and will continue to do so.