Researchers Chain Cloud Infrastructure Flaws to Bypass Tenant Isolation

· Original article ↗

Summary

A bug bounty assessment of an unnamed cloud provider chained Kubernetes, control-plane and managed-service flaws to bypass tenant isolation. The researchers say the provider fixed all reported issues.

Key points

  • A privileged Kubernetes pod gave researchers root access to a worker node and its Konnectivity service token.
  • Researchers used Konnectivity packet handling to reach control-plane sockets and execute code, then exploited an OverlayFS flaw to gain root privileges.
  • A flat overlay network exposed managed services; shared logs contained root SSH keys that could be matched to other tenants’ cluster nodes.
  • An exposed Kafka JMX service, Redis exporter credential leakage and unrestricted PostgreSQL configuration storage provided additional cross-tenant attack paths.
  • SQL injection in a custom Salt extension exposed Salt master authentication tokens, potentially allowing commands across cloud minions; researchers did not develop that attack further by agreement with the vendor.
  • Researchers also examined CVE-2026-25243 in Valkey and a MySQL configuration-injection chain as alternative routes into the infrastructure.
  • The provider was advised to isolate managed services, enforce AppArmor policies and monitor potentially vulnerable services. Researchers say all reported issues were fixed.

Article Details

Attack Vectors
  • A privileged Kubernetes pod with host namespace access let researchers escape to a worker node and obtain a Konnectivity service token.
  • A crafted Konnectivity DIAL_REQ enabled server-side requests from the control plane, including access to D-Bus sockets and execution of a transient user service.
  • An OverlayFS capability-handling flaw enabled root escalation on the control-plane host.
  • A flat overlay network exposed shared metrics, logs and managed services. Logs contained root SSH private keys for other tenants’ nodes.
  • An exposed Kafka JMX registry allowed a custom MBean to be loaded and used for remote command execution.
  • The redis-exporter scrape endpoint could be directed to a researcher-controlled host, exposing credentials used to access other customers’ Redis instances.
  • Unauthenticated etcd access allowed modification of Patroni configuration to execute a command on a PostgreSQL host.
  • SQL injection through Salt minion grains exposed database records containing Salt master authentication tokens.
  • CVE-2026-25243 in managed Valkey was used to obtain host access; the researchers then used a local privilege escalation vulnerability.
  • MySQL privilege-table manipulation and configuration injection enabled execution of a custom UDF, followed by root escalation through a privileged watcher service.
Defensive Notes
  • The provider fixed all security issues reported by the researchers.
  • The researchers recommended isolating managed services to prevent network visibility between them, enforcing strict AppArmor policies and monitoring the use of potentially vulnerable services.

MITRE ATT&CK

CVE

Vendors

Products

Tools

Related Articles