GPU VulnDB

Database/Kernel, userspace & hypervisor

OpenSSL: malicious TLS server can hang a client with an oversized DH prime

CVSS 7.5CVE-2018-0732Kernel, userspace & hypervisorcurated

Impact

A TLS server that a node connects to can return a very large prime for DH(E) key agreement, and the OpenSSL client will burn CPU generating a key for it until the handshake effectively hangs. On a GPU fleet the exposure is outbound: agents, package and model-registry pullers, telemetry shippers and control-plane clients that dial out over TLS can be stalled one connection at a time by whatever endpoint they talk to, including a hijacked or redirected one. There is no confidentiality or integrity loss - the CVSS vector is availability-only - but a wedged node agent looks like a dead node to a scheduler. Affected: 1.1.0-1.1.0h and 1.0.2-1.0.2o; this is a long-fixed 2018 issue that distributions patched years ago.

Who can reach it

A remote TLS server, or anyone able to impersonate or redirect one, against an outbound OpenSSL client connection. No authentication needed; the client has to initiate the handshake with a DH(E) ciphersuite enabled.

What to do

Update OpenSSL to 1.1.0i or 1.0.2p or later, or apply the distribution package (Red Hat RHSA-2018:2552/2553, Ubuntu, Oracle). Because OpenSSL is a shared library, every long-running process that loaded it has to be restarted to pick up the fix - in practice a rolling restart of node agents and daemons, or a node reboot if you would rather not chase individual services. No node drain is required for the library update itself.

References

Related entries

All Kernel, userspace & hypervisor entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.