GPU VulnDB

Database/Kernel, userspace & hypervisor

OpenSSL: NULL dereference crashes a TLSv1.2 server on a crafted renegotiation ClientHello

CVSS 5.9CVE-2021-3449Kernel, userspace & hypervisorcurated

Impact

A renegotiation ClientHello that omits signature_algorithms while including signature_algorithms_cert dereferences NULL and takes the server process down. Any OpenSSL 1.1.1 server with TLSv1.2 and renegotiation enabled is affected, which was the default configuration - that covers a lot of what listens on a GPU fleet: cluster API endpoints, metrics and logging collectors, registries, management web services. No authentication and no prior handshake success is needed, so a single tenant-reachable listener can be killed repeatedly until it is patched. Clients are not affected, and the issue is availability only - no key or data disclosure.

Who can reach it

Anyone who can open a TLS connection to an affected server and complete a handshake far enough to renegotiate. No authentication required. Exposure is whatever network the listener sits on - tenant network, management VLAN, or the internet.

What to do

Upgrade OpenSSL to 1.1.1k (or the distribution backport) and restart every service linked against it - statically linked and container-bundled copies will not pick up a host library update. OpenSSL 1.0.2 is not affected. Where an immediate upgrade is not possible, disabling TLSv1.2 renegotiation on the affected listeners removes the trigger.

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.