CVE-2026-74753

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 only through local syscalls—perf_event_open(2) to create a remove_on_exec group leader and attach a sibling, execve(2) to move the leader to EXIT state, close(2) to free it, then ioctl/read on the sibling. There is no network, adjacent-wireless, or physical device entry path.

AC:LAttack complexity, low

The attacker fully controls the deterministic sequence: open leader with remove_on_exec, exec, open sibling against the leader fd, close leader, then trigger group operations on the sibling. No race or condition outside attacker control is required; the stale group_leader pointer is created reliably.

PR:LPrivileges required, low

A basic unprivileged local user can open per-task perf event groups on their own process with exclude_kernel=1 under the default sysctl_perf_event_paranoid=2, without CAP_PERFMON or init-namespace root. security_perf_event_open(PERF_SECURITY_OPEN) and perf_check_permission() allow this self-monitoring path.

UI:NUser interaction, none

Exploitation requires no action from another user or administrator beyond the attacker running their own syscalls (open group, exec, attach sibling, close leader, ioctl/read sibling). No victim must mount filesystems, open files, or interact with the system.

S:UScope, unchanged

Impact is confined to kernel perf/core heap corruption and privilege escalation within the same host kernel security authority. This is not a VM escape, IOMMU bypass, or cross-namespace boundary change; it is standard local kernel memory corruption.

C:HConfidentiality impact, high

This is a use-after-free: after the EXIT-state leader is freed, the sibling retains group_leader pointing at freed memory, and perf_event_for_each(), __perf_effective_state(), and group reads/ioctls dereference the freed leader and its context, enabling arbitrary kernel memory disclosure via controlled reallocations.

I:HIntegrity impact, high

Freed perf_event/group_leader structures can be reallocated with attacker-controlled data, providing heap grooming primitives for arbitrary kernel writes and control-flow hijack. Memory corruption from following the stale group_leader pointer is exploitable beyond a simple crash.

A:HAvailability impact, high

UAF dereferences of the freed group_leader reliably cause kernel paging faults, oops, or panic when the sibling is read, enabled, or ioctl'd (as in related perf group_leader UAF reproducers). UAF on attacker-influenceable perf_event objects causes full denial of service and potential system-wide unavailability.