CVE-2026-68272: Linux Kernel Privilege Escalation Flaw

CVE-2026-68272 Overview

CVE-2026-68272 is a NULL pointer dereference vulnerability in the Linux kernel's AMD GPU (amdgpu) driver. The flaw resides in the command submission (CS) path, specifically in how amdgpu_cs_pass1() handles the AMDGPU_CHUNK_ID_CP_GFX_SHADOW chunk type. The function fails to validate a minimum length for this chunk, allowing an unprivileged process in the render group to submit a zero-length chunk. This triggers vmemdup_array_user() to return ZERO_SIZE_PTR, which later gets dereferenced by amdgpu_cs_p2_shadow() as a drm_amdgpu_cs_chunk_cp_gfx_shadow struct, causing a kernel NULL pointer dereference.

Critical Impact
An unprivileged local user with render group membership can trigger a kernel NULL pointer dereference, resulting in a denial-of-service condition on systems using AMD GPUs with the affected driver.

Affected Products

Linux kernel amdgpu (AMD GPU DRM driver)
Systems running affected upstream Linux kernel versions prior to the fix
Stable kernel branches receiving the cherry-picked backport of commit 7f61b2eef741

Discovery Timeline

2026-08-10 - CVE-2026-68272 published to NVD
2026-08-10 - Last updated in NVD database

Technical Details for CVE-2026-68272

Vulnerability Analysis

The vulnerability originates in the amdgpu_cs_pass1() function within the AMD GPU DRM driver. This function performs first-pass validation on chunks submitted by userspace as part of a command submission (CS) ioctl. Existing validation gates enforce a minimum length for the IB, FENCE, and BO_HANDLES chunk types before their data gets copied and later dereferenced as fixed-size structs.

The AMDGPU_CHUNK_ID_CP_GFX_SHADOW chunk was omitted from this validation. Instead, it shared a bare break statement with the dependency and syncobj chunk types, neither of which dereference a fixed-size struct. When userspace submits the shadow chunk with length_dw == 0, vmemdup_array_user() is called with size equal to zero and returns ZERO_SIZE_PTR. Because ZERO_SIZE_PTR passes the IS_ERR() check, execution continues into pass2.

In pass2, amdgpu_cs_p2_shadow() casts chunk->kdata to a pointer of type struct drm_amdgpu_cs_chunk_cp_gfx_shadow and reads shadow->flags. Dereferencing ZERO_SIZE_PTR triggers a kernel oops [CWE-476].

Root Cause

The root cause is missing input validation on a user-controlled chunk length. The kernel trusted that any chunk type reaching pass2 had already been range-checked, but the CP_GFX_SHADOW handler was placed in a code path that skipped the minimum-size gate. This produced a classic mismatch between validation state and consumer expectations.

Attack Vector

The attack requires local access. Any process belonging to the render group can open the DRM device node and issue the DRM_IOCTL_AMDGPU_CS ioctl. By crafting a submission that includes an AMDGPU_CHUNK_ID_CP_GFX_SHADOW chunk with length_dw set to zero, the attacker reaches the vulnerable code path without privilege escalation. The primary impact is denial of service through kernel crash.

No verified public exploit code is available. See the upstream fix at Kernel Patch 7f61b2ee for technical details on the corrective validation.

Detection Methods for CVE-2026-68272

Indicators of Compromise

Kernel oops or panic messages referencing amdgpu_cs_p2_shadow or amdgpu_cs_pass1 in dmesg or /var/log/kern.log
Unexpected process terminations or GPU driver resets on systems with AMD graphics
Repeated DRM_IOCTL_AMDGPU_CS calls from unprivileged users containing malformed chunk submissions

Detection Strategies

Monitor kernel ring buffer output for NULL pointer dereference traces originating in the amdgpu module
Audit membership of the render and video groups to identify accounts capable of reaching the vulnerable ioctl
Correlate GPU driver crashes with the running kernel version to identify unpatched systems

Monitoring Recommendations

Enable kdump or persistent kernel log collection to capture oops traces for post-incident analysis
Forward kernel logs to a centralized SIEM to surface repeated amdgpu faults across the fleet
Track kernel package versions against distribution security advisories to confirm patch application

How to Mitigate CVE-2026-68272

Immediate Actions Required

Apply the vendor kernel update containing the upstream fix that adds the minimum-length check for the CP_GFX_SHADOW chunk
Restrict membership of the render group to trusted users on multi-tenant systems
Reboot affected hosts after installing the patched kernel package to ensure the new module is loaded

Patch Information

The fix was cherry-picked from upstream commit 7f61b2eef7415eccdb40850aca0de94211948657 and adds a minimum-length gate for AMDGPU_CHUNK_ID_CP_GFX_SHADOW in amdgpu_cs_pass1(), rejecting undersized chunks with -EINVAL before pass2 dereferences the data. Backports are available in the following stable kernel commits:

Workarounds

Remove untrusted local users from the render group to prevent access to the DRM_IOCTL_AMDGPU_CS ioctl on affected systems
Where GPU compute is not required, unload the amdgpu module or restrict access to /dev/dri/renderD* device nodes via udev rules
Isolate workloads that require GPU access into dedicated hosts until the patched kernel is deployed

bash

# Verify kernel version and restrict render device access
uname -r
ls -l /dev/dri/renderD*
# Example udev rule to restrict render node access to a specific group
echo 'SUBSYSTEM=="drm", KERNEL=="renderD*", GROUP="trusted-render", MODE="0660"' \
  | sudo tee /etc/udev/rules.d/60-amdgpu-render.rules
sudo udevadm control --reload-rules && sudo udevadm trigger