Re: [oss-security] CVE-2020-25656: Linux kernel concurrency UAF in vt_do_kdgkb_ioctl

[Greg KH]

On Fri, Oct 16, 2020 at 08:58:34AM +0200, Jiri Slaby wrote:
> Cc Greg.
> 
> On 16. 10. 20, 5:39, Minh Yuan wrote:
> > Hi,
> > 
> > We recently discovered a uaf read in vt_do_kdgkb_ioctl from linux kernel
> > version 3.4 to the latest version (v5.9 for now).
> > 
> > The root cause of this vulnerability is that there exits a race in
> > KDGKBSENT and KDSKBSENT.
> > 
> > Here are details:
> > 1. use  KDSKBSENT to allocate a lager heap buffer to funcbufptr;
> > 2. use KDGKBSENT to obtain the allocated heap pointer in step1 by
> > func_table, at the same time, due to KDGKBSENT has no lock, we can use
> > KDSKBSENT again to allocate a larger buffer than step1, and the old
> > funcbufptr will be freed. However, we've obtained the heap pointer in
> > KDGKBSENT, so a uaf read will happen while executing put_user.
> 
> Hi,
> 
> this is likely the issue I am fixing at:
> https://git.kernel.org/pub/scm/linux/kernel/git/jirislaby/linux.git/commit/?h=devel&id=57c85191e788e172a446e34ef77d34473cfb1e8d
> 
> I think, it won't apply cleanly as it's a part of a larger set. I will
> reorder the patch and send something during the day.

Great, thanks for looking into this!

greg k-h