GPU VulnDB

Database/Kernel, userspace & hypervisor

Linux kernel mm/migrate: NULL dereference panic when memory offlining dumps a page under migration

CVSS 5.5CVE-2023-52490Kernel, userspace & hypervisorcurated

Impact

During page migration the kernel stashed an anon_vma pointer in the target page's ->mapping field without setting PAGE_MAPPING_ANON. If another thread offlined that memory block at the same time, it dumped the page, page_mapping() returned what looked like a file mapping, and dump_mapping() dereferenced garbage - a hard kernel panic, reproduced with stress-ng. The fix moves the anon_vma pointer into page->private so PFN walkers (memory hotplug, compaction) can no longer be misled. On a GPU node this is a whole-host crash: every tenant pod and every in-flight training job on that node dies, and the node comes back only after a reboot. Triggering it needs concurrent memory offlining, which is an operator action, so this is a stability risk for fleets that use memory hotplug or balloon-style offlining rather than a tenant-reachable attack.

Who can reach it

Local: requires a privileged action that offlines a memory block (memory hotplug / device_offline) racing against ongoing page migration. Unprivileged tenants cannot initiate memory offlining.

What to do

Update to a stable kernel containing the fix for your series (commits in the record), which means draining the node and rebooting. Until then, avoid memory-block offlining on nodes carrying live workloads.

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.