In the Linux kernel, the following vulnerability has been resolved:
drm/virtio: use the DMA API for resource backing on Xen
On a Xen PV domain page addresses bear no relation to the real machine
addresses the host would have to use to reach it.
virtio_ring.c handles this correctly, vring_use_map_api() returns true
for any xen_domain() regardless of VIRTIO_F_ACCESS_PLATFORM.
virtio-gpu makes the same decision independently, but its copy
looks only at the feature bit:
<pre>
bool use_dma_api = !virtio_has_dma_quirk(vgdev->vdev);
</pre>
QEMU does not set iommu_platform on virtio-vga by default, so
VIRTIO_F_ACCESS_PLATFORM is not negotiated, use_dma_api is false, and
virtio_gpu_object_shmem_init() describes the framebuffer's backing pages
to the host with sg_phys(). Those are guest-physical addresses. In a PV
domain they resolve, on the host side, to pages belonging to some other
domain, so the host scans out unrelated memory.
Move the decision into virtio_gpu_use_dma_api() and give it the
xen_domain() check, like vring_use_map_api() has. This
additionally enables the dma_sync_sgtable_for_device() calls in
virtgpu_vq.c, which are required for correctness whenever swiotlb
is in play.
Reproduced with a Xen 4.21 PV dom0 nested inside QEMU 8.2 with
virtio-vga, on both a distro 6.8 kernel and 6.18 LTS. A PVH dom0
works fine and doesn't need this fix because it is identity-mapped,
only PV dom0s are affected.
CVSS Vector: CVSS:3.1/AV:L/AC:H/PR:L/UI:N/S:C/C:H/I:H/A:H
CVSS Score: 7.8
AV:L - The attacker is a local process in the Xen PV dom0. It opens the virtio-gpu render node and calls VIRTGPU_RESOURCE_CREATE, which runs virtio_gpu_object_shmem_init(), and then VIRTGPU_TRANSFER_TO_HOST or TRANSFER_FROM_HOST. No remote peer supplies the bad sg_phys() addresses; the guest driver builds them itself.
AC:H - The bug only exists on a Xen PV dom0 nested inside QEMU with virtio-vga and without iommu_platform, so VIRTIO_F_ACCESS_PLATFORM is not negotiated. This is a rare deployment the attacker cannot create. The attacker also cannot choose which foreign machine frames the pseudo-physical pfns resolve to.
PR:L - VIRTGPU_RESOURCE_CREATE, TRANSFER_TO_HOST and TRANSFER_FROM_HOST are all marked DRM_RENDER_ALLOW. Any user with access to /dev/dri/renderD* (render group or a logind seat ACL) can reach them without root.
UI:N - The attacker creates the resource and issues the transfer ioctls entirely on their own. No other user has to act.
S:C - virtio_gpu_object_shmem_init() hands the host guest-pseudo-physical addresses, which the host resolves to machine frames owned by other Xen domains or by Xen itself. So a flaw in the dom0 driver affects memory outside that domain's security authority.
C:H - virtio_gpu_cmd_transfer_to_host_2d/3d make the host copy unrelated machine pages (other domains' memory) into a host resource, which is then scanned out. With virgl, the attacker could plausibly copy that resource into memory they can read back.
I:H - With virgl, VIRTGPU_TRANSFER_FROM_HOST makes the host write resource data to the wrong backing addresses. Those addresses are machine frames belonging to other domains or the hypervisor, so this is an uncontrolled cross-domain memory write.
A:H - Host writes into foreign machine frames through the bad backing entries can corrupt Xen or other domains' memory, crashing them. The unsynced swiotlb path (missing dma_sync_sgtable_for_device) also corrupts display data.
| Attack Vector |
Local |
Scope |
Changed |
| Attack Complexity |
High |
Confidentiality Impact |
High |
| Privileges Required |
Low |
Integrity Impact |
High |
| User Interaction |
None |
Availability Impact |
High |
AV:L - The attacker is a local process in the Xen PV dom0. It opens the virtio-gpu render node and calls VIRTGPU_RESOURCE_CREATE, which runs virtio_gpu_object_shmem_init(), and then VIRTGPU_TRANSFER_TO_HOST or TRANSFER_FROM_HOST. No remote peer supplies the bad sg_phys() addresses; the guest driver builds them itself.
AC:H - The bug only exists on a Xen PV dom0 nested inside QEMU with virtio-vga and without iommu_platform, so VIRTIO_F_ACCESS_PLATFORM is not negotiated. This is a rare deployment the attacker cannot create. The attacker also cannot choose which foreign machine frames the pseudo-physical pfns resolve to.
PR:L - VIRTGPU_RESOURCE_CREATE, TRANSFER_TO_HOST and TRANSFER_FROM_HOST are all marked DRM_RENDER_ALLOW. Any user with access to /dev/dri/renderD* (render group or a logind seat ACL) can reach them without root.
UI:N - The attacker creates the resource and issues the transfer ioctls entirely on their own. No other user has to act.
S:C - virtio_gpu_object_shmem_init() hands the host guest-pseudo-physical addresses, which the host resolves to machine frames owned by other Xen domains or by Xen itself. So a flaw in the dom0 driver affects memory outside that domain's security authority.
C:H - virtio_gpu_cmd_transfer_to_host_2d/3d make the host copy unrelated machine pages (other domains' memory) into a host resource, which is then scanned out. With virgl, the attacker could plausibly copy that resource into memory they can read back.
I:H - With virgl, VIRTGPU_TRANSFER_FROM_HOST makes the host write resource data to the wrong backing addresses. Those addresses are machine frames belonging to other domains or the hypervisor, so this is an uncontrolled cross-domain memory write.
A:H - Host writes into foreign machine frames through the bad backing entries can corrupt Xen or other domains' memory, crashing them. The unsynced swiotlb path (missing dma_sync_sgtable_for_device) also corrupts display data.
CVSS 3.1