GPU VulnDB

Database/Kernel, userspace & hypervisor

OpenSSL: out-of-bounds read and write when decrypting CMS messages that use password-based encryption

CVSS 9.8CVE-2025-9230Kernel, userspace & hypervisorcurated

Impact

Any process that decrypts a CMS/PKCS#7 message using password-based recipient info (PWRI) can be pushed past the end of a buffer, both reading and writing. The write is a memory corruption, so the realistic outcomes range from crashing the service to execution of attacker-supplied code in whatever daemon did the decryption. On a GPU fleet the exposure is indirect: OpenSSL is linked into effectively everything on the node - SSH, the kubelet and container runtime, registry clients, monitoring agents, BMC-facing tooling - so the library itself is everywhere, but almost nothing in that stack decrypts CMS messages, and PWRI in particular is rarely used. OpenSSL itself rated the issue Moderate for that reason, despite the 9.8 NVD score. The FIPS modules in 3.0 through 3.5 are not affected because the CMS implementation sits outside the FIPS module boundary.

Who can reach it

Whoever can supply a CMS message to an application that decrypts it with password-based encryption. No authentication is implied by the flaw itself - the gate is whether any component on the node performs CMS PWRI decryption on untrusted input, which for most datacenter software it does not.

What to do

Take the OpenSSL update from your distribution and restart every service that links libcrypto/libssl - in practice a rolling restart of node daemons, or a reboot if you would rather not chase every linked process. No node drain or firmware work is needed. If you run software that decrypts CMS messages with a password recipient, prioritise those hosts; otherwise this patches cleanly in the normal update cycle. The fix is in the upstream commits listed below; the record does not state a single fixed version number across all branches.

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.