KASAN: slab-out-of-bounds Write in vgacon_scroll

[Dmitry Vyukov]

On Wed, Jan 29, 2020 at 3:40 PM Sergey Senozhatsky
<sergey.senozhatsky at gmail.com> wrote:
>
> Cc-ing Dmitry and Tetsuo
>
> Original Message-id: CAA=061EoW8AmjUrBLsJy5nTDz-1jeArLeB+z6HJuyZud0zZXug at mail.gmail.com
>
> On (20/01/29 23:17), Sergey Senozhatsky wrote:
> > > Hmm. There is something strange about it. I use vga console quite
> > > often, and scrolling happens all the time, yet I can't get the same
> > > out-of-bounds report (nor have I ever seen it in the past), even with
> > > the reproducer. Is it supposed to be executed as it is, or are there
> > > any preconditions? Any chance that something that runs prior to that
> > > reproducer somehow impacts the system? Just asking.
> >
> > These questions were addressed to anon anon (742991625abc at gmail.com),
> > not to Bartlomiej.
>
> Could this be GCC_PLUGIN related?

syzkaller repros are meant to be self-contained, but they don't
capture the image and VM setup (or actual hardware). I suspect it may
have something to do with these bugs.
syzbot has reported a bunch of similar bugs in one of our internal kernels:

KASAN: slab-out-of-bounds Read in vgacon_scroll
KASAN: slab-out-of-bounds Read in vgacon_invert_region
KASAN: use-after-free Write in vgacon_scroll
KASAN: use-after-free Read in vgacon_scroll
KASAN: use-after-free Read in vgacon_invert_region
BUG: unable to handle kernel paging request in vgacon_scroll

But none on upstream kernels. That may be some difference in config?
I actually don't know what affects these things. When I tried to get
at least some coverage of that code in syzkaller I just understood
that relations between all these
tty/pty/ptmx/vt/pt/ldisc/vcs/vcsu/fb/con/dri/drm/etc are complex to
say the least...