CVE-2025-38352

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 vulnerability is reached purely through local syscalls — `timer_create`/`timer_settime`/`timer_delete` (or `clock_nanosleep`) plus `ptrace`/`waitpid` to control task reaping — with the trigger arriving via the local timer tick. There is no network-facing path into posix-cpu-timers.

AC:LAttack complexity, low

The attacker controls both sides of the race: it arms the CPU timer expiry, chooses the precise moment of `release_task()` by calling `waitpid()` from its own parent/tracer, and issues the concurrent `timer_delete()` from a sibling thread — retryable in a tight loop and widenable by queueing up to 20 timers on the firing list. Real-world exploitation of this exact bug confirms it is reliably winnable.

PR:LPrivileges required, low

No capability or permission check exists anywhere on the path — `validate_clock_permissions()`/`pid_for_clock()` gate only on thread-group/pid existence, and `grep` for `capable`/`CAP_` in both posix-timers files returns nothing. Any unprivileged local user, including a seccomp-confined Android app UID, can reach it.

UI:NUser interaction, none

The attacker's own processes and threads perform every step — creating the timer, exiting the traced thread, reaping it, and deleting the timer. No victim action of any kind is needed.

S:UScope, unchanged

The use-after-free corrupts kernel heap objects and escalates within the kernel's own security authority, which is standard kernel privilege escalation. No hypervisor, IOMMU, or other security-authority boundary is crossed.

C:HConfidentiality impact, high

The freed `struct sigqueue` is linked into the signal pending list by `send_sigqueue()` and later copied out to userspace through `dequeue_signal()`/`rt_sigtimedwait()`, disclosing whatever now occupies the reclaimed slab object — an effectively arbitrary kernel-memory read. The `do_cpu_nanosleep()` variant additionally exposes reclaimed kernel stack.

I:HIntegrity impact, high

`send_sigqueue()` writes into the freed object (`q->info.si_overrun++`) and performs a `list_add_tail()` linked-list insertion on attacker-resprayable memory, followed by a double free on dequeue — a classic controlled-write primitive leading to full privilege escalation, as demonstrated by in-the-wild exploitation of this flaw.

A:HAvailability impact, high

Even without successful exploitation the race causes `BUG_ON(!(q->flags & SIGQUEUE_PREALLOC))` in `send_sigqueue()`, `WARN_ON_ONCE()` in the delete path, list corruption, and a double free — any of which oops or panic the kernel. Spinning the race is trivially repeatable as a denial of service.