CVE-2026-68452

This CVE exposes a trust boundary failure at the kernel-coprocessor interface that should concern every engineer working with hardware security modules. The vulnerability lives in the s390/zcrypt CCA key handling path where the driver derives a copy length from a field inside a key token returned by the cryptographic coprocessor — and then uses that derived length without validating it fits the destination buffer. The fix is a single bounds check, which tells you everything about the bug's nature: not complex algorithmic failure, but a discipline failure at an abstraction boundary where developers assumed trusted hardware wouldn't send malformed responses. The analytical angle that matters: this is a 30-year-old vulnerability class that keeps surfacing in hardware abstraction layers. USB descriptors, SCSI sense handlers, PCIe config space responses, and now crypto coprocessor key tokens all share the same failure mode — deriving operational parameters from response structures without validating those parameters against kernel buffer bounds. Each instance gets patched and archived as a one-off, but each disclosure also signals to attackers: here is a proven attack surface, here is exactly what to look for in other HALs. What makes this specific to s390/zcrypt isn't that mainframes are immune to bugs — they're not. It's that the code path has accrued reviewer attention debt: a tiny population of engineers can meaningfully audit this code, the token format has evolved as CCA support expanded, and the inherited trust assumption was never challenged until now. The coprocessor is trusted hardware, but the driver must still validate that its responses conform to declared structures before using derived lengths as copy operations. Defensive priorities: first, verify your kernel includes the CCA bounds-check patch; second, treat this as a forcing function to audit other zcrypt request types for identical derived-length patterns; third, extend that audit to every kernel subsystem that derives operational parameters from hardware response structures. The exploitation surface requires local access or a token injection path, which narrows mass exploitation but creates acute risk for the high-security workloads z-series systems typically run — payment switching, government systems, enterprise critical infrastructure. The blast radius isn't 'this server is pwned'; it's 'the cryptographic infrastructure trusted by every workload on this LPAR is compromised.' That concentration of value is why this deserves prioritization attention beyond what a 7.8 CVSS score suggests in isolation.

Reviewed through automated stages and approved by a human before publication.