Database/Kernel, userspace & hypervisor
OpenSSL 3.0.4: memory corruption in 2048-bit RSA private-key operations on AVX512IFMA CPUs
Impact
On x86_64 CPUs that support AVX512IFMA - which includes the Xeon Scalable parts under most GPU server platforms - OpenSSL 3.0.4 computes 2048-bit RSA private-key operations incorrectly and corrupts memory while doing so. Any process holding such a key and performing a private-key operation on attacker-reachable input is affected, and the advisory states the corruption may be escalated to remote code execution in the process doing the computation. In a datacenter that means the TLS terminators and API endpoints in front of the fleet: cluster API servers, registries, model-serving gateways and management web UIs that were built against 3.0.4. The regression was introduced in 3.0.4 itself, so only that one release is exposed and the fix is a straight forward upgrade rather than a configuration change.
Who can reach it
Unauthenticated network client that can open a TLS session (or otherwise cause a 2048-bit RSA private-key operation) against a service linked to OpenSSL 3.0.4 running on an AVX512IFMA-capable host.
What to do
Upgrade to OpenSSL 3.0.5 or later; if an upgrade cannot be scheduled immediately, roll back to 3.0.3, which does not contain the regression. Every daemon linked against the shared library must be restarted to pick it up, so this is a service-by-service restart sweep across the fleet, not a node reboot - but statically linked appliances and vendor images (Siemens has its own advisory for embedded products) need their own vendor update. Audit which hosts actually run 3.0.4 before scheduling work; most distributions never shipped it.
References
Related entries
- glibc: stack buffer overflows in the deprecated sunrpc svcunix_create and clnt_create helpersCVE-2022-23218 · glibc sunrpc compatibility functions (svcunix_create, clnt_create)Critical
- Linux kernel (net/smc): When an SMC-R link is torn down, the kernel moves the QP to Error state and then destroys theCVE-2022-48673 · Linux kernel (net/smc)Critical
- Linux kernel NVMe-oF TCP host (nvme-tcp, digest error handling in io_work): The initiator kept reading from the socketCVE-2022-48686 · Linux kernel NVMe-oF TCP host (nvme-tcp, digest error handling in io_work)Critical
- Linux kernel NVMe target core (nvmet, request completion during IO connect): KASAN-confirmed use-after-free reachedCVE-2022-48697 · Linux kernel NVMe target core (nvmet, request completion during IO connect)Critical
- Linux kernel (drivers/nvme/host): On the NVMe/RDMA initiator, an async-event command can be submitted against an adminCVE-2022-48788 · Linux kernel (drivers/nvme/host)Critical
- Linux kernel (drivers/nvme/host): Same race as the RDMA variant but on NVMe/TCP, which is the far more common fabric inCVE-2022-48789 · Linux kernel (drivers/nvme/host)Critical
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.