CVE-2026-80555

Quoted from the kernel CNA's own scoring file, word for word. A highlighted metric carries the most severe value CVSS v3.1 defines for it, which is what pushes the score up. source

AV:LAttack vector, local

The bug is reached when userspace writes the vfio-ccw io_region (VFIO device mmap/write), triggering fsm_io_request() -> cp_init(); on s390 KVM this is driven by QEMU forwarding guest channel-I/O channel programs, requiring local VFIO/device interaction rather than any network protocol.

AC:LAttack complexity, low

An attacker can deterministically fail cp_init() after allocating multiple TIC-linked ccwchain segments by crafting a channel program whose early TIC targets succeed and a later one fails (invalid CPA, chain-length/count limit, or dma_rw error), leaking prior segments on every attempt without races.

PR:NPrivileges required, none

On s390 KVM with CCW passthrough, a malicious guest can submit crafted multi-TIC channel programs through normal guest I/O; QEMU forwards them to vfio-ccw without the attacker holding host CAP_SYS_ADMIN, root, or other real init-namespace privileges beyond the assigned device.

UI:NUser interaction, none

No victim interaction is required beyond the attacker (or their guest) issuing channel-I/O start requests with a malicious channel program; exploitation does not depend on another user mounting media, accepting prompts, or performing administrative actions at trigger time.

S:CScope, changed

The leaked ccwchain, CCW, and page_array allocations are host kernel slab memory; a guest VM triggering this through passthrough crosses the guest-host virtualization boundary by consuming host resources the guest security domain should not affect.

C:NConfidentiality impact, none

This is a kernel memory leak on the cp_init() error path, not a use-after-free, out-of-bounds read, or other memory corruption; leaked objects are orphaned in kernel heap and are not returned to userspace or the guest, so no information disclosure occurs.

I:NIntegrity impact, none

The flaw only fails to free kernel allocations when cp_init() errors; it does not corrupt memory, modify guest or host data, or provide write or control-flow primitives, and cp->initialized stays false so cp_free() never runs on the error or close paths.

A:HAvailability impact, high

Each failed cp_init() permanently orphans up to CCWCHAIN_COUNT_MAX ccwchain segments (large GFP_DMA allocations), and subsequent cp_init() calls INIT_LIST_HEAD without freeing prior leaks; repeated guest or VFIO I/O requests can exhaust host memory and cause system-wide OOM or severe denial of service.