Follow @Openwall on Twitter for new release announcements and other news
[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Wed, 29 Jul 2020 14:58:39 +0200
From: Solar Designer <>
Cc: Yunhai Zhang <>, Jiri Slaby <>
Subject: Re: [CVE-2020-14331] Linux Kernel: buffer over write in vgacon_scrollback_update

On Tue, Jul 28, 2020 at 11:59:14AM -0700, Eric Biggers wrote:
> On Tue, Jul 28, 2020 at 11:16:55AM +0800, ????????? wrote:
> > There is a buffer over write in drivers/video/console/vgacon.c in
> > vgacon_scrollback_update.
> > 
> > The issue is reported by Yunhai Zhang / NSFOCUS Security Team
> > <>, CVE-2020-14331 assigned via Red Hat.
> > 
> > # Affected Versions
> > The issue is found and tested on 5.7.0-rc6.
> > The issue is introduced in commit:
> > 15bdab959c9bb909c0317480dd9b35748a8f7887 ([PATCH] vgacon: Add support
> > for soft scrollback)

That was in 2006.

> > According to code review, all versions older than
> > 92ed301919932f777713b9172e525674157e983d (v5.8-rc7) are affected.
> Thanks for the writeup.  Note that there are many open syzbot reports in the
> fbdev, vt, and vgacon kernel subsystems.  These subsystems aren't actively
> maintained (receiving drive-by fixes only), and the kernel developers recommend
> to not enable these subsystems if you care about security
> (
> This particular bug, for example, appears to have been already found by someone
> running syzkaller and publicly reported over 2 years ago, with a C reproducer:
> (
> No one did anything.
> I suggest that people relying on the security of these kernel subsystems
> contribute resources to fixing the many known fuzzing bugs in them.

Wow.  I suppose the biggest risk here is services that just happen to
run on the console (or able to access it if they re-open /dev/tty) as a
result of normal system startup.  Since an ioctl() is required at least
to trigger CVE-2020-14331, at least this one is limited to attacks by
someone who already got code execution within one of such services, but
I suppose it could in some cases be used to gain ring 0 access from a
non-root pseudo-user that the service (or even merely its privsep child)
might run as.  If any other related issues are triggerable purely by
terminal escapes codes, it's much worse - could even allow for remote
attacks without a prior compromise of any service.

Do others see this same threat model or something different?  I think
non-root users with intentional console access mounting attacks is less
of a concern.

Meanwhile, Jiri Slaby brought the discussion around fixing vgacon
properly to LKML:

The patch posted in this very first LKML message (with the added check
before the loop) is already known to be insufficient - see the
follow-ups.  The patch posted by Yunhai Zhang here on oss-security
(with the added check in the loop) is still considered sufficient.


Powered by blists - more mailing lists

Please check out the Open Source Software Security Wiki, which is counterpart to this mailing list.

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.