CVE-2026-44477: CloudNativePG: Metrics exporter allows privilege escalation to PostgreSQL superuser and OS RCE

Synopsis

This is a privilege escalation and remote code execution vulnerability in CloudNativePG, the Kubernetes-native PostgreSQL operator. The metrics exporter connects to PostgreSQL as the postgres superuser and attempts to demote itself with SET ROLE, but session_user remains postgres throughout, so any SQL expression evaluated during a scrape can call RESET ROLE to reclaim full superuser rights and then use COPY ... TO PROGRAM to run arbitrary OS commands inside the primary pod. A low-privilege authenticated attacker reachable over the network can escalate to PostgreSQL superuser and execute arbitrary processes as the postgres OS user. No fix versions have been published yet; HarborGuard tracks this advisory and will make a patched-image rebuild available the moment upstream ships a fix.

HarborGuard Coverage

Exploit Conditions

• Network reachabilityRequired

  The attacker must reach the metrics exporter endpoint over the network; the CVSS vector specifies AV:N, meaning no local or physical access is required.

• AuthenticationRequired

  PR:L indicates that any low-privilege account is sufficient; the attacker does not need admin credentials, but some level of authenticated access to the scrape or SQL evaluation surface is needed.

• Victim interactionNot required

  UI:N confirms no user action or social engineering is needed; the attacker exploits the vulnerability entirely without involving another person.

• Attack complexityDetail

  AC:L means the exploit is reliable and condition-free; no race conditions, specific memory layout, or environmental pre-conditions need to be satisfied.

Blast Radius

• The attacker reclaims PostgreSQL superuser privileges within the scrape session, gaining unrestricted read and write access to all databases on the primary pod.
• Using COPY ... TO PROGRAM, the attacker spawns arbitrary OS-level subprocesses as the postgres user inside the primary Kubernetes pod.
• With OS-level execution, the attacker reads or exfiltrates stored database contents, secrets mounted into the pod, and service account tokens accessible on the filesystem.
• The attacker modifies or destroys database state and can disrupt PostgreSQL availability, affecting all workloads dependent on the primary instance.