Securing Filesystem Images for Unprivileged Containers
summary
Claims that unprivileged containers can run bare‑metal in production using only user namespaces and shows prototypes for fixing UID/GID mismatch when sharing images: an in‑kernel shiftfs (a subtree bind with range maps) and a portable rootfs approach that marks superblocks for translation. Demo uses bindfs/shiftfs style mounts (module not loaded so demo partially fails), Docker currently resorts to putting real root inside containers, and neither solution is upstream yet so permissions and attack surface remain unresolved.
AI-written summary · about
chapters (12)
- 0:00Start
- 1:10Kernel Api for Containers
- 3:19Namespaces
- 4:16The Pid Namespace
- 7:41Setting Up a Container without Root Privileges
- 9:30State of Unprivileged Containers
- 13:28Set Groups
- 14:49Set Up Unprivileged Containers
- 16:36Container Images
- 20:24Bind Fs
- 25:10Portable Root File System
- 39:46Conclusion
related talks
45mMaking Containers Safer
01h 31mTutorial: Using Linux Primitives to Build Your Own Containers
31mTowards unprivileged container builds
48mState of Linux Kernel Container Features
44mRootless containers
32mThe Route To Rootless Containers
01h 32mContainers – A Look Under the Hood
45mSecuring Container Runtimes -- How Hard Can It Be?
