GPU VulnDB

Database/Kernel, userspace & hypervisor

OpenSSL: TLSv1.3 server session cache grows without bound, exhausting memory

CVSS 5.9CVE-2024-2511Kernel, userspace & hypervisorcurated

Impact

A TLS 1.3 server built on OpenSSL and configured with the non-default SSL_OP_NO_TICKET option can get its session cache into a state where it never flushes, so the cache grows until the process is out of memory. A malicious client can drive this deliberately, and OpenSSL notes it can also happen by accident under normal load. On a GPU fleet the exposure is the TLS-terminating control-plane and serving daemons that link OpenSSL - inference endpoints, registries, metrics and management APIs - where the loss is the daemon, not the GPUs. Clients are unaffected, as is OpenSSL 1.0.2 and the FIPS modules in 3.0 through 3.2.

Who can reach it

Any client that can open TLS 1.3 connections to an affected server, no authentication needed; only servers using the non-default SSL_OP_NO_TICKET option (without early_data plus default anti-replay) are affected.

What to do

Update OpenSSL to a release carrying the referenced commits and restart every service linked against it - on most distributions a shared-library update still needs each daemon bounced to pick it up, so expect a rolling restart of TLS front ends rather than a node reboot. As an interim step, review whether SSL_OP_NO_TICKET is actually needed on your TLS 1.3 servers; the issue does not arise without it.

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.