Daniel P. Berrangé <[email protected]> writes: > Alex, as the virtio-gpu* maintainer can you give your opinion > on the discussion below about virtio-gpu security tagging > vs our current security policy for "virtualization use case" > to ack/nack the proposal.
I'm inclined to agree with Marc-Andre here - the 3D GPU attack surface is large and for the in-QEMU handling we have a bunch of issues with GPU context handling between threads that make it tricky. Compounding that ruttabaga is not very well tested at all because its hard to enable without non-distro libs (for now). Finally virtio-gpu is Odd Fixes so it's not getting the response times of other VirtIO stuff in terms of handling security issues. We should encourage users to use vhost-user-gpu for 3D stuff because at least then the rendering pipeline and processing of guest controlled data is easier to sandbox. So I'd say: - virtio-gpu for 2d - OK - virtio-gpu for 3d via vhost - OK - virtio-gpu for 3d in QEMU - Not OK Alex. > > On Fri, Nov 06, 2026 at 05:33:31PM +0000, Daniel P. Berrangé wrote: >> On Fri, Nov 06, 2026 at 06:52:27PM +0400, Marc-André Lureau wrote: >> > Hi >> > >> > On Fri, Nov 6, 2026 at 6:40 PM Daniel P. Berrangé <[email protected]> >> > wrote: >> > > >> > > On Fri, Nov 06, 2026 at 06:35:29PM +0400, Marc-André Lureau wrote: >> > > > Hi >> > > > >> > > > On Fri, Nov 6, 2026 at 3:53 PM Thomas Huth <[email protected]> wrote: >> > > > > >> > > > > On 29/10/2026 14.12, Daniel P. Berrangé wrote: >> > > > > > These are all intended for use in a virtualization scenario and >> > > > > > must >> > > > > > provide a security boundary. >> > > > > > >> > > > > > Acked-by: Marc-André Lureau <[email protected]> >> > > > > > Signed-off-by: Daniel P. Berrangé <[email protected]> >> > > > > > --- >> > > > > Reviewed-by: Thomas Huth <[email protected]> >> > > > > >> > > > >> > > > I have a second thought about this, I'd like to exclude all >> > > > gpu-accelerated paths for now. virgl/mesa/rutabaga expose a very large >> > > > attack surface, so I'd like to drop them from the secure boundary >> > > >> > > Rather than repost, please see my response for patch 2: >> > > >> > > https://lists.gnu.org/archive/html/qemu-devel/2026-11/msg01530.html >> > > >> > > TL;DR: we have already promised anything virtio related is in scope >> > > for security handling for a long time, as they are all unambiguously >> > > "virtualization use case". >> > > >> > >> > the value of a secure policy enforcement is low if we know those >> > devices are unsecure and don't have the resources to make them secure. >> >> It isn't a great situation to be in, but we're already there for any >> virtio devices based on our existing policies for bug handling and >> security triage. We can't promise we'll get fixes for all CVEs quickly, >> but we will at least assign CVEs so that users are aware when there >> are known problems. >> >> This contrasts to other devices where problems will remain largely >> hidden for view, as we treat them as normal bugs and never call our >> security concerns when committing changes. >> >> 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 :| >> >> > > With regards, > Daniel -- Alex Bennée Virtualisation Tech Lead @ Linaro
