An inconsistency in the Linux kernel’s memory‑group Least-Recently-Used (mglru) page table walker caused an under‑count of resident pages during concurrent memory‑cgroup (memcg) reparenting operations. The walker updates per‑generation deltas without locking the LRU vector; later, reset_batch_size() merges these deltas while holding the lock. When the memcg reparenting path proceeds concurrently, the merge can add the delta to a child’s lru_gen after the child has been marked dying, producing a stale batch update. The result is that the accounting for active pages under a memcg underestimates the true number of resident pages, which triggers premature memory‑allocation failures (OOM) because the kernel believes the group has no reclaimable pages yet it still holds many.

All Linux kernel releases that contain the vulnerable mglru implementation are affected. No specific version range is provided, so any system running an unpatched kernel version may be susceptible. This includes all Linux distributions that ship the upstream Linux kernel.

The EPSS score of < 1% indicates a very low likelihood of exploitation. The CVSS score of 5.5 reflects moderate severity. Exploitation would require a specific race condition between a page table walk and a memcg reparenting operation, which is non‑trivial and does not require additional privileges beyond those normally granted to kernel code. The vulnerability is not listed in the CISA KEV catalog. Although the bug can trigger premature termination of processes, the likelihood of successful exploitation in the wild remains low.