GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel SUNRPC: kernel memory leak on error path when importing a krb5 GSS security context

CVSS 5.5CVE-2023-52653Kernel, userspace & hypervisorcurated

Impact

The krb5 v2 security-context import path allocated ctx->mech_used.data with kmemdup and never freed it when the import failed, leaking kernel memory on every failed context import. The fix reworks the call into gss_krb5_import_ctx_v2 so the allocation is released. This is relevant to GPU fleets that mount training data over Kerberos-authenticated NFS (sec=krb5/krb5i/krb5p): repeated failing context imports slowly consume unreclaimable kernel memory on the client node, and a node that runs out of kernel memory has to be rebooted rather than drained cheaply. There is no confidentiality or integrity effect, only exhaustion.

Who can reach it

Local to the NFS client node, via repeated GSS security-context imports from userspace (rpc.gssd path). Only reachable where krb5-secured NFS is actually configured; nodes using sec=sys are unaffected.

What to do

Take the stable kernel update for your series (commits in the record). That is a node drain and reboot. Interim mitigation if a reboot cannot be scheduled: monitor slab growth on krb5 NFS clients, or move affected mounts off krb5 security flavors.

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.