CVE-2026-63293

CVE-2026-63293 is a link-following flaw (CWE-59) in Canonical's LXD container manager: when unpacking an imported image archive, LXD does not verify that the archive's metadata.yaml is not a symbolic link, so a symlink inside the archive is resolved against the host filesystem. An attacker who can get a crafted image imported — either a low-privilege user with image-import rights or an administrator importing attacker-supplied images — can point metadata.yaml at any host path and gain arbitrary file read and write on the host (CVSS 9.9, scope changed, indicating the container/host boundary is crossed). Any LXD deployment that imports images from untrusted sources is affected, with multi-tenant or shared servers that allow untrusted users to import images at the greatest risk. A public advisory with PoC-level detail exists (GHSA-j825-cg34-5fr5), but the flaw is not on CISA KEV and EPSS is low at 0.5% (40th percentile), so no in-the-wild exploitation is currently known.

What to do: Upgrade all LXD hosts to the patched release named in Canonical's advisory GHSA-j825-cg34-5fr5 via your normal package/snap update channel. Until patched, restrict image-import permissions to trusted administrators and import images only from trusted remotes, and review which users hold import rights on multi-tenant servers. Treat any confirmed exploitation as potential host-level compromise and check hosts for unexplained host-filesystem changes.