Unroot 1.0.3

Features Release — August 13, 2026

A number of improvements have been made to unroot to improve usability and make the tool more seamless in real-world use, as well as more robust. Many thanks to Zeno R. R. Davatz (@zdavatz) for extended testing and comprehensive issue reporting. Unroot would not be the same without community support and feedback!

• $TERM is now preserved inside the rootfs by default which generally allows ncurses programs and colors to work seamlessly in the terminal after unroot enter.
• /sys is recursively bind-mounted read-only into the rootfs by default, allowing tools such as lscpu and htop to function properly without exposing writable sysfs control files.
• Unroot now uses libarchive through the sibling unroot-util host helper. This removes the dependency on a locally-installed GNU tar and eliminates a class of bugs.
• Parallel xz compression is now supported and enabled by default via libarchive.
• unroot unpack --id-count COUNT can create rich roots with ownership headroom beyond the default IDs 0–65535. (#19)
• New inspect host and inspect archive <archive> actions expose runtime capabilities and archive facts (#21).
• unroot unpack will always do a single pre-flight scan of the to-be-unpacked archive; when it finds POSIX ACLs or extended attributes, unpack now performs a quick check for the corresponding support on the exact destination filesystem before extraction begins. If necessary features are not present, unpack aborts prior to extraction. This can be overridden with --force.
• /etc/resolv.conf and /etc/hosts are now injected (safely copied) into managed (non-native) rootfs environments using the novel inject action. They were previously read-only bind-mounted so could not be altered. This new method allows local name resolution to work while allowing local edits to the files, and prevents local network configuration from leaking into packed rootfs tarballs. This also provides a framework for customizing other injectable files that you want inserted into the unpacked rootfs by default, but don't want to become part of the packed/distributed rootfs.
• pack and unpack now take one self-cleaning exclusive lock per rootfs, so archive reads and writes cannot overlap on the same tree. A competing command fails immediately, while process exit automatically releases ownership and leaves no stale lock file. (#27)
• For now, OCI images are rejected until we add support. (#25)
• enter rejects paths that resolve to the host /.