Re: [PATCH 1/7] hw/display/virtio-gpu-rutabaga: zero-init capset info response

Hi

On Mon, Jul 27, 2026 at 3:07 PM Daniel P. Berrangé <[email protected]> wrote:
>
> On Sun, Jul 26, 2026 at 12:24:52PM +0400, Marc-André Lureau wrote:
> > Hi
> >
> > On Sun, Jul 26, 2026 at 11:33 AM Akihiko Odaki
> > <[email protected]> wrote:
> > >
> > > On 2026/07/25 23:00, Marc-André Lureau wrote:
> > > > rutabaga_cmd_get_capset_info() only fills in capset_id,
> > > > capset_max_version and capset_max_size before sending the response to
> > > > the guest. The remaining fields of struct virtio_gpu_resp_capset_info,
> > > > including hdr.fence_id, hdr.ctx_id and hdr.ring_idx, are left with
> > > > stack garbage and leaked to the guest, including host pointers useful
> > > > for an ASLR bypass.
> > > >
> > > > Zero the response first, matching virgl_cmd_get_capset_info().
> > > >
> > > > Not a real risk thanks to -ftrivial-auto-var-init=zero.
> > >
> > > GCC only added -ftrivial-auto-var-init=zero in GCC 12, while QEMU still
> > > supports GCC 10.4/11.
>
> Opps, I thought we had new enough versions already.
>
> >
> > Good point, may be we should update:
> > https://patchew.org/QEMU/[email protected]/
>
> I'm inclined to bump our min GCC/Clang versions so we can assume
> zero-init.
>
> We picked 10.5 due to NetBSD, and CentOS 9 still has 11.x
>
> In both those cases, however, Clang is available in new enough
> version to support -ftrivial-auto-var-init=zero. In NetBSD
> there are also many new versions of GCC  available, only the
> base system GCC is 10.5.

Apparently it needs clang 16 from Mar 2023
(https://releases.llvm.org/18.1.6/tools/clang/docs/ReleaseNotes.html#removed-compiler-flags)

>
>
> With regards,
> Daniel
> --
> |: https://berrange.com       ~~        https://hachyderm.io/@berrange :|
> |: https://libvirt.org          ~~          https://entangle-photo.org :|
> |: https://pixelfed.art/berrange   ~~    https://fstop138.berrange.com :|
>
>