Skip to main content
CVE Vulnerability Database
Vulnerability Database/CVE-2026-63845

CVE-2026-63845: Linux Kernel JPEG Ring Fence Vulnerability

CVE-2026-63845 is a user fence handling flaw in the Linux kernel's AMDGPU JPEG v4.0 ring that affects command submission processing. This article covers the technical details, affected versions, impact, and mitigation.

Updated:

CVE-2026-63845 Overview

CVE-2026-63845 is a Linux kernel vulnerability in the AMD GPU (amdgpu) driver's JPEG v4.0 ring implementation. JPEG rings do not support 64-bit user fence writes, but the driver previously accepted command submissions (CS) containing user fences. The fix sets no_user_fence for the JPEG v4.0 ring, causing the kernel to reject CS submissions containing user fences on this hardware. Exploitation requires local access with low privileges and no user interaction. The upstream fix has been backported via cherry-pick from commit 8d0cac9478a3f046279c657d6a2545de49ae675a.

Critical Impact

A local, low-privileged user submitting crafted graphics command streams to the amdgpu JPEG v4.0 ring can trigger high-impact effects on kernel confidentiality, integrity, and availability.

Affected Products

  • Linux kernel builds including the amdgpu driver with JPEG v4.0 ring support
  • Systems using AMD GPUs that expose the JPEG v4.0 engine through the Direct Rendering Manager (DRM) subsystem
  • Distributions shipping affected stable kernel branches prior to the referenced backport commits

Discovery Timeline

  • 2026-07-19 - CVE-2026-63845 published to the National Vulnerability Database (NVD)
  • 2026-07-20 - Last updated in NVD database

Technical Details for CVE-2026-63845

Vulnerability Analysis

The vulnerability resides in the drm/amdgpu/jpeg code path that handles command submissions to the JPEG v4.0 ring. The JPEG engine on this hardware generation lacks support for 64-bit user fence writes. User fences are a mechanism through which userspace can request the GPU to write a value to a specified memory location when a submission completes, enabling userspace synchronization primitives.

Because the ring did not advertise no_user_fence, the amdgpu CS ioctl accepted submissions containing user fence descriptors targeting the JPEG v4.0 ring. When such a submission executed, the engine attempted an operation it does not support, producing undefined behavior in kernel-controlled memory and ring state.

The attack vector is local with low privileges, no user interaction required, and the scope is unchanged. Confidentiality, integrity, and availability impacts are all rated high.

Root Cause

The amdgpu driver's ring descriptor for JPEG v4.0 omitted the no_user_fence flag. This missing flag allowed the command submission validation path to accept user fence write requests that the underlying JPEG hardware cannot fulfill. The fix explicitly sets no_user_fence so the CS ioctl rejects such submissions before they reach the hardware.

Attack Vector

An attacker with local shell access and permissions to open the DRM render node (typically any logged-in user or member of the video/render group) can issue DRM_IOCTL_AMDGPU_CS calls targeting the JPEG v4.0 ring with a user fence descriptor. The malformed submission drives the JPEG engine into an unsupported code path, corrupting kernel-managed fence state and potentially yielding privilege escalation or system compromise.

No public proof-of-concept exploit is available, and the vulnerability is not listed in the CISA Known Exploited Vulnerabilities catalog. See the upstream fix commits for technical details: Linux Kernel Commit 8d0cac94 and the additional backport commits.

Detection Methods for CVE-2026-63845

Indicators of Compromise

  • Unexpected kernel log entries from the amdgpu subsystem referencing JPEG ring faults, GPU resets, or CS submission errors
  • User-space processes without a legitimate media workload issuing DRM_IOCTL_AMDGPU_CS calls against the JPEG v4.0 ring
  • Repeated GPU hangs or ring timeouts correlated with a single low-privileged user session

Detection Strategies

  • Monitor dmesg and journalctl -k output for amdgpu ring errors, fence signaling failures, or GFX/JPEG resets on affected kernel versions
  • Audit process lineage for unexpected binaries invoking DRM ioctls on the render node /dev/dri/renderD*
  • Compare running kernel versions against the fixed stable branches listed in the upstream commits and flag hosts running unpatched builds

Monitoring Recommendations

  • Enable Linux audit rules on /dev/dri/renderD* device access to establish a baseline of legitimate GPU consumers
  • Forward kernel ring buffer messages to a centralized log platform and alert on amdgpu fault signatures
  • Track membership changes in the video and render groups that expand the population of users able to reach the vulnerable ioctl surface

How to Mitigate CVE-2026-63845

Immediate Actions Required

  • Apply vendor kernel updates that include the upstream fix from commit 8d0cac9478a3f046279c657d6a2545de49ae675a or any of the backport commits referenced by NVD
  • Inventory hosts with AMD GPUs exposing JPEG v4.0 rings and prioritize patching multi-user systems, workstations shared across teams, and virtual desktop infrastructure
  • Restrict interactive local access on affected systems until patched builds are deployed

Patch Information

The fix sets the no_user_fence flag on the JPEG v4.0 ring so the amdgpu CS ioctl rejects submissions containing user fences. It was cherry-picked from mainline commit 8d0cac9478a3f046279c657d6a2545de49ae675a into multiple stable branches. Reference commits: 6876d05b, a676f16e, af4b458d, d4e0172a, and e7e90b58.

Workarounds

  • Limit membership of the video and render groups to trusted users to reduce the local attacker population
  • Where feasible, unload or blacklist the amdgpu module on servers that do not require GPU acceleration
  • Apply mandatory access control policies (SELinux, AppArmor) that restrict which processes may open /dev/dri/renderD* nodes

Disclaimer: This content was generated using AI. While we strive for accuracy, please verify critical information with official sources.

Default Legacy - Prefooter | Experience the World’s Most Advanced Cybersecurity Platform

Experience the Most Advanced Cybersecurity Platform

See how the world’s most intelligent, autonomous cybersecurity platform can protect your organization today and into the future.