Bug 2574264 (CVE-2026-130507)

Summary: CVE-2026-130507 virglrenderer: virglrenderer: guest-triggerable OOB write/read in GBM direct-transfer path
Product: [Other] Security Response Reporter: OSIDB Bzimport <bzimport>
Component: vulnerabilityAssignee: Product Security DevOps Team <prodsec-dev>
Status: NEW --- QA Contact:
Severity: high Docs Contact:
Priority: high    
Version: unspecifiedCC: rhel-process-autobot, watson-tool-maintainers
Target Milestone: ---Keywords: Security
Target Release: ---   
Hardware: All   
OS: Linux   
Whiteboard:
Fixed In Version: Doc Type: ---
Doc Text:
A flaw was found in virglrenderer, the virtual OpenGL renderer used to provide 3D graphics acceleration to virtual machines. On the host-side buffer transfer path for GBM-backed resources, guest-supplied transfer parameters are not properly validated against the dimensions of the allocated buffer. A malicious guest with 3D acceleration enabled could use this flaw to perform an out-of-bounds write in the host renderer process, causing memory corruption, or an out-of-bounds read disclosing host memory to the guest. This could result in a denial of service of the renderer process or, potentially, a virtual machine escape. Only renderer configurations with GBM buffer allocation enabled are affected.
Story Points: ---
Clone Of: Environment:
Last Closed: Type: ---
Regression: --- Mount Type: ---
Documentation: --- CRM:
Verified Versions: Category: ---
oVirt Team: --- RHEL 7.3 requirements from Atomic Host:
Cloudforms Team: --- Target Upstream Version:
Embargoed:
Bug Depends On: 2574283    
Bug Blocks:    

Description OSIDB Bzimport 2026-11-17 09:57:32 UTC
Stuart Thomas reported via JSM (PSIRTSUPT-30502, 2026-11-13, responsible disclosure, not published elsewhere) a guest-to-host out-of-bounds write and read in virglrenderer's vrend GBM direct-transfer path.

In vrend_renderer_transfer_internal() (src/vrend/vrend_renderer.c), the GBM fast path calls virgl_gbm_transfer() and returns before the resource_contains_box() and check_iov_bounds() validations run. virgl_gbm_transfer() (src/vrend/vrend_winsys_gbm.c) derives the copy offset (host_resource_offset) and row count from the guest-controlled pipe_box without validating them against the gbm buffer object's dimensions, so a malicious guest can memcpy past the mapped bo via VIRGL_CCMD_TRANSFER_TO_HOST (host heap out-of-bounds WRITE of attacker-controlled bytes) and VIRGL_CCMD_TRANSFER_FROM_HOST (out-of-bounds READ copied back to the guest) on a GBM-backed (SCANOUT/SHARED-bound) resource. box->x/box->y are signed and unvalidated on this path, so the write start offset is attacker-arbitrary.

Confirmed by source audit and reproduced end-to-end under ASan (vtest transport, virglrenderer 1.3.0): heap-buffer-overflow WRITE and READ of 256-byte rows past a 16 KiB gbm bo mapping in virgl_gbm_transfer_internal(), reached through vrend_renderer_transfer_internal's early-return fast path. Single-variable negative control (in-bounds box) completes cleanly with data round-trip verified.

Affected configurations: builds with ENABLE_GBM_ALLOCATION + EGL and the GBM allocation path active -- minigbm builds (ChromeOS crosvm/ARCVM) always; generic libgbm builds with the gbm layout feature / Venus interop active (host/embedder configuration). Upstream main has no fix as of commit 01637fc (2026-11-11); the vulnerable code is identical in the 1.3.0 release.

Red Hat exposure: none. RHEL 8 ships virglrenderer 0.8.2 with the GBM allocation feature compiled out (meson option gbm_allocation default false, not enabled in the spec); the component is not shipped in RHEL 9 or 10.

CVSS 3.1 (vulnerable configurations): 8.8 -- AV:L/AC:L/PR:L/UI:N/S:C/C:H/I:H/A:H (PR:N / attacker controls the guest OS: 9.3).