CVE-2023-54088

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 in the block cgroup layer (block/blk-cgroup.c) and is reached only through local operations — cgroup v2 mkdir/rmdir, block I/O to a device, and writes to io.* cgroup control files or disk teardown. No network-facing code path touches q->blkg_list.

AC:LAttack complexity, low

The attacker drives both sides of the race: cgroup rmdir plus prior I/O queues many blkg_free_workfn() instances that do the unlocked list_del_init(&blkg->q_node), while a concurrent io.latency write (blk_iolatency_init -> blkcg_activate_policy) or disk teardown (blkg_destroy_all) walks q->blkg_list under queue_lock only. The window can be widened arbitrarily by creating thousands of blkgs and freeing them in bulk, and pd_free_fn() work inside the mutex lengthens it further.

PR:LPrivileges required, low

An unprivileged local user suffices in the standard deployment where a cgroup v2 subtree is delegated (systemd user@.service, rootless containers): the user owns the io.* files in cgroups they create, so they can create/destroy blkgs and trigger policy activation. No real root or CAP_SYS_ADMIN in the init namespace is needed.

UI:NUser interaction, none

The race is triggered entirely by the attacker's own processes issuing I/O and manipulating their cgroups; no victim action is required.

S:UScope, unchanged

Corruption stays within the kernel's own security authority — this is a standard kernel memory-corruption/privilege-escalation primitive with no crossing of a VM, IOMMU, or hypervisor boundary.

C:HConfidentiality impact, high

The unsynchronized list_del_init() lets list_for_each_entry_safe() in blkg_destroy_all() and the walkers in blkcg_activate_policy() follow stale next pointers into a blkcg_gq that blkg_free_workfn() then kfree()s, so the kernel reads and dereferences attacker-sprayable reclaimed slab contents — a use-after-free read that yields an arbitrary kernel-memory disclosure primitive.

I:HIntegrity impact, high

After the list is corrupted the kernel writes through the dangling entry: blkg->pd[pol->plid] = pd, pd->online = false, radix_tree_delete(), hlist_del_init_rcu(), rcu_assign_pointer(blkcg->blkg_hint, NULL) and percpu_ref_kill() all operate on freed/reallocated memory, giving controlled pointer writes and indirect calls suitable for privilege escalation.

A:HAvailability impact, high

The commit message states that list corruption and hard lockup are directly triggerable from blkg_destroy_all(); a corrupted doubly-linked list walked under spin_lock_irq loops forever or oopses, panicking the machine.