Docker: `PUT /containers/{id}/archive` executes container binary on the host

Summary

When a user uploads a compressed archive into a container, a malicious image can execute arbitrary code with daemon (host root) privileges.

Details

When handling PUT /containers/{id}/archive requests with compressed archives, the daemon decompresses them using external system binaries. Due to incorrect ordering of operations, these binaries are resolved from the container's filesystem rather than the host's. A container image that includes a trojanized decompression binary can achieve code execution as the daemon process whenever a compressed archive is uploaded to that container.

The executed binary runs with the daemon's full privileges, including host root UID and unrestricted capabilities.

Impact

Arbitrary code execution as host root, crossing the container-to-host trust boundary.

Conditions for exploitation

• A user must run a container from a malicious image that contains a trojanized decompression binary.
• The user must then upload a compressed archive (xz or gzip) into that container, either by piping a compressed archive via docker cp - or by calling the PUT /containers/{id}/archive API directly with compressed content.

Not affected

Standard docker cp usage is not affected, because the CLI sends uncompressed tar by default:

docker cp ./file.txt mycontainer:/file.txt

This can only be exploited when explicitly passing a xz or gzip-compressed archive to docker cp or the PUT /containers/{id}/archive API, for example:

cat archive.tar.xz | docker cp - mycontainer:/dir

Decompression formats using pure Go implementations (bzip2, zstd, and gzip when the container image does not contain an unpigz binary) are also not affected.

Workarounds

• Only run containers from trusted images.
• Use authorization plugins to limit access to the PUT /containers/{id}/archive endpoint.
• Avoid piping compressed archives into containers created from untrusted images.