Canonical LXD CVE-2025-54290

Requires the LXD HTTPS REST API to be network-reachable (the images export endpoint at /1.0/images/{fingerprint}/export, e.g. on TCP 8443); the attacker then supplies a URL-encoded SQL LIKE wildcard (%25 = '%') as the fingerprint plus the 'secret' query parameter to force multiple image matches within a target project. No authentication or user interaction is needed. The key LIMITING factor is that LXD by default listens only on a local Unix socket and does not expose this API over the network - exploitation requires an operator to have enabled the HTTPS network listener (core.https_address), so default-installed hosts are not exposed. Affected versions are LXD 5.21.x before 5.21.4 and 6.0/6.x before 6.5 on Linux; impact is limited to confirming project existence (enumeration), not disclosing image or credential contents.

This is a low-to-moderate priority information disclosure issue, not a critical one, and the signals are mostly consistent. The CVSS 4.0 vector (AV:N/AC:L/PR:N/UI:N) confirms unauthenticated remote reachability with low attack complexity, but the only impact metric set is VC:L (low confidentiality) with VI/VA/SC/SI/SA all None - meaning the flaw leaks nothing beyond whether a named LXD project exists, matching CWE-200. The headline 6.9 score reflects the easy, unauthenticated vector rather than meaningful impact. EPSS is very low (0.09%, 26th percentile), indicating little expectation of mass exploitation, and the vulnerability is NOT in CISA KEV, so there is no evidence of active exploitation. Publicly available exploit code (the GHSA advisory includes working curl reproduction steps) raises the practical likelihood of opportunistic probing, but POC availability is distinct from active exploitation. The realistic risk is that this serves as a reconnaissance primitive (project-name enumeration via dictionary attack) that could precede a more serious attack, rather than a standalone high-impact bug. One important caveat: the CVSS vector implies exploitation against any network-reachable endpoint, but LXD does not expose its HTTPS API over the network by default - this materially limits real-world exposure and should be weighed against the AV:N rating.

An attacker who can reach a network-exposed LXD HTTPS API (e.g. port 8443) sends unauthenticated requests to the image export endpoint using a URL-encoded wildcard fingerprint (%25) plus the secret parameter, then observes whether the server returns HTTP 500 ('More than one image matches') versus HTTP 404 ('not found'). By cycling candidate project names through this behavioral difference, the attacker enumerates which LXD projects exist. Publicly available exploit code with working curl commands exists, and the low attack complexity makes this trivial to script against any reachable instance.