Linux slab allocator: ABA race in the optimistic freelist return corrupts the partial list

Impact

The optimistic __slab_try_return_freelist() path assumed a NULL slab->freelist plus a successful cmpxchg meant nobody had freed into the slab, but another CPU can free, insert the slab onto the partial list, allocate again and be mid-removal under n->list_lock. __refill_objects_node() then re-inserts the same slab while it is being removed, corrupting the node partial list - the reporter's trace is a list_add corruption BUG from a plain msgsnd() syscall, hit as an unprivileged user (UID 65534 in the report). This is the core allocator, so it is reachable from any workload on the node, not a niche subsystem, and the known outcome is an immediate kernel panic taking every job on the GPU host with it. Note the trace is against a 7.2-rc tree - the flawed optimization arrived in commit ba7425312607, so only kernels carrying it are affected.

Who can reach it

Local unprivileged user on the node, through ordinary allocation-heavy syscalls; no GPU or device access needed. Any tenant with code execution qualifies. Requires a kernel that includes the __slab_try_return_freelist optimization.

What to do

Check whether your kernel carries commit ba7425312607; if it does, install a build with the list_lock fix and reboot the node. If your kernels predate that optimization - most current LTS lines - you are not affected and no window is needed. There is no runtime mitigation for a core allocator race.

References

Related entries

• Linux AMD IOMMU: sign-discarding error check lets nested domains use an unallocated domain IDCVE-2026-98002 · Linux kernel iommu/amd (amd_iommu_alloc_domain_nested error check)High
• Linux kernel BPF: per-CPU map updates index by logical CPU id, reading past the update bufferCVE-2026-98149 · Linux kernel BPF per-CPU map update paths (array, hash, cgroup storage)High
• Linux kernel RDMA/rxe: ODP write paths accept read-only pages, letting RDMA traffic overwrite the page cacheCVE-2026-98361 · Linux kernel RDMA/rxe (soft-RoCE on-demand paging write path)High
• Linux kernel RDMA/rxe: MR re-registration swaps the PD before validating access flags, freeing a live PDCVE-2026-98366 · Linux kernel RDMA/rxe rxe_rereg_user_mr() (access-flag validation ordering)High
• Linux kernel RDMA/siw: failed QP modify in siw_accept() races QP-to-ERROR transition into a use-after-freeCVE-2026-98367 · Linux kernel RDMA/siw siw_accept() (CEP association cleared outside state_lock)High
• Linux kernel mm/rmap: missing barrier lets a reused anon_vma hang tasks and use-after-free its rootCVE-2026-98507 · Linux kernel mm/rmap (anon_vma publish without memory barrier)High