On Thu, Nov 30, 2017 at 4:57 AM, Thomas Gleixner <t...@linutronix.de> wrote:
> On Thu, 30 Nov 2017, Alexey Dobriyan wrote:
>
>> [cc security@]
>> 100% oops with interrupts disabled by nobody
>> or kernel memory read
>> [nods]
>> you named the bug already
>>
>> "notify" directly comes from userspace struct sigevent::sigev_notify
>> without adult supervision.
>>
>> Reproducer is timer_create + read(/proc/self/timers)
>
> Bah. That's a really old one.
>
> Tentative fix below. That needs more though but looking at the existing
> check there is only one valid combo with SIGEV_THREAD_ID.
>
> Thanks,
>
> tglx
>
> 8<----------------
> --- a/kernel/time/posix-timers.c
> +++ b/kernel/time/posix-timers.c
> @@ -434,6 +434,16 @@ static struct pid *good_sigevent(sigeven
> {
> struct task_struct *rtn = current->group_leader;
>
> + switch (event->sigev_notify) {
> + case SIGEV_NONE:
> + case SIGEV_SIGNAL:
> + case SIGEV_SIGNAL | SIGEV_THREAD_ID:
> + case SIGEV_THREAD:
> + break;
> + default:
> + return NULL;
> + }
> +
> if ((event->sigev_notify & SIGEV_THREAD_ID ) &&
> (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) ||
> !same_thread_group(rtn, current) ||
FWIW, this looks correct to me.
-Kees
--
Kees Cook
Pixel Security
