==============================================================
[CVE-2014-8178] Attacker controlled layer IDs lead to local graph content poisoning
==============================================================
Docker image layers are stored with a non-globally unique identifier vulnerable to a collision attack. These identifiers are shared during docker pull and push, allowing poisoning of a host’s image cache. This allows maliciously crafted images to poison subsequently pulled images.
Independently discovered by Florian Weimer of Red Hat and Tõnis Tiigi of the Docker Engine Team
==============================================================
[CVE-2014-8179] - Manifest validation and parsing logic errors allow pull-by-digest validation bypass
==============================================================
During Docker pulls validation and extraction of the manifest object from its JSON representation are done in separate steps. The digest that represents the manifest corresponds to a hash of the payload portion of the JSON blob returned by the remote registry. Even though the validity of the payload portion is being verified, an injection of new attributes in the surrounding JSON object is possible. This allows an override of the verified content at JSON deserialization time leading to pulling unverified layers.
Users are advised to upgrade to Docker 1.8.3.