Linux kernel io_uring: linked work items keep running after ring exit starts, stalling teardown

Impact

io_worker_handle_work() snapshots the IO_WQ_BIT_EXIT flag once before walking a chain of dependent linked work items, so if the ring starts exiting mid-chain the remaining items reuse the stale decision, skip IO_WQ_WORK_CANCEL and keep executing. A local user can queue long chains of slow linked work and immediately close the ring, making exit take an extended time while worker threads stay busy. On a shared GPU node any tenant with a normal pod can reach io_uring, so this is a noisy-neighbour availability problem rather than a containment break: no host memory is disclosed and no privilege is gained. The record reports availability impact only (CVSS 5.5, AV:L), and the fix was found by syzbot rather than in production.

Who can reach it

Local unprivileged user on the node, including any tenant inside a container where io_uring is not blocked by seccomp. No authentication beyond having code execution on the host or in a pod.

What to do

Pick up the fixed stable kernel and reboot each node; there is no runtime toggle in the commit. Where kernel maintenance is expensive, blocking io_uring_setup via the seccomp profile for untrusted workloads removes the exposure without a reboot - many container runtimes already do this. The record lists only stable git commits and does not name a released kernel version, so confirm the fixed version with your distribution before scheduling the window.

References

Related entries

• Linux io_uring: NOP with IOSQE_FIXED_FILE leaks a struct file reference on every submissionCVE-2026-64426 · Linux kernel io_uring NOP opcode (IOSQE_FIXED_FILE handling)Medium
• Linux kernel (drivers/vfio/pci): The disable_idle_d3 power-management flag was a module-wide global that could changeCVE-2026-64476 · Linux kernel (drivers/vfio/pci)Medium
• Linux kernel (drivers/iommu/intel): SVA bind and unbind are asymmetric on VT-d hardware without PCI/PRI - bind skipsCVE-2026-64591 · Linux kernel (drivers/iommu/intel)Medium
• Linux kernel (arch/x86/kvm/vmx): When a nested VM-Enter fails on invalid guest state, KVM took an open-coded exit pathCVE-2026-68081 · Linux kernel (arch/x86/kvm/vmx)Medium
• AMD SEV-ES (CacheWarp): CacheWarp: INVD lets a malicious hypervisor revert SEV-ES guest memory writes, breaking guestCVE-2023-20592 · AMD SEV-ES (CacheWarp)Medium
• Linux kernel (drivers/pci): When the kernel coalesces two adjacent host-bridge apertures it invalidates the absorbedCVE-2023-53814 · Linux kernel (drivers/pci)Medium