EUVD-2026-103758
GHSA-84w9-cx8r-cxrf
Attack ChainAIDerived
Hypothetical attack flow derived from CVE metadata – not a confirmed exploit chain
-
AccessGain local low-privileged access
-
DeliveryTrigger nouveau device-memory migration spanning multiple chunks
-
ExploitResolve page address outside backing buffer object
-
ExecutionRead or corrupt out-of-bounds VRAM
-
ImpactInformation disclosure or GPU memory corruption
Vulnerability Assessment
| Exploitation Conditions | Requires a system using the Linux kernel's nouveau DRM driver with device-private (DMEM) GPU memory migration, and specifically the THP-enabled migration support added by commit c32287471077, which enlarged the registered device-private region to DMEM_CHUNK_SIZE * NR_CHUNKS while the backing VRAM buffer object stayed at DMEM_CHUNK_SIZE. The out-of-bounds condition is only reached when a migrated range extends past the first chunk, so the trigger is the migration of a region larger than DMEM_CHUNK_SIZE to GPU VRAM. Limiting factors: local access only (AV:L), a low-privileged but authenticated account (PR:L), high attack complexity (AC:H), the presence of a compatible NVIDIA GPU running nouveau rather than the proprietary driver, and a kernel built with the affected THP GPU-migration code - none of which hold on default server deployments without such hardware. |
| Impact | This is a local, higher-complexity flaw in the Linux kernel's nouveau DRM device-private memory (dmem) path, and the signals converge on a genuine-but-non-urgent priority rather than a widespread emergency. The provided CVSS 3.1 vector (AV:L/AC:H/PR:L/UI:N) indicates an authenticated local user (PR:L) must drive a non-trivial sequence (AC:H) to reach the condition, which caps the base severity at 7.0 despite full C:H/I:H/A:H impact. EPSS is very low (0.19%, 8th percentile), there is no CISA KEV listing, and no public exploit or PoC reference appears in the input (the two git.kernel.org links are the fix commits, not exploit code). The vuln.today decision label is 'this_week', which is consistent with a patched, locally-exploitable memory-safety bug that should be scheduled but is not an active-exploitation fire drill. The CWE is listed as N/A and the affected-version metadata from EUVD is internally messy (mixing '7.3-rc5', '7.2.9', '6.19' and raw commit hashes), so precise affected ranges should be confirmed against the kernel stable trees rather than taken verbatim. The 'Information Disclosure' tag understates the issue: because out-of-bounds VRAM is resolved for both reads and writes, the C:H/I:H/A:H rating (read and write corruption of GPU memory outside the buffer object) is the more accurate characterization. |
| Exploit ScenarioDerived | A local, low-privileged user on a machine with an NVIDIA GPU driven by the nouveau driver triggers migration of a memory region large enough to span more than one DMEM chunk; pages beyond the first chunk are mapped to VRAM outside the backing buffer object, allowing out-of-bounds GPU memory access. No public exploit or PoC was identified in the input, and the high attack complexity (AC:H) means reliable exploitation depends on controlling the migration layout. |
Get an e-mail when this CVE is exploited, patched or re-scored. Free account.
Alert meAnalysisAIv1 – Oct 24, 2026
Out-of-bounds VRAM access in the Linux kernel's nouveau DRM device-private memory (dmem) migration path lets a local, low-privileged user read and write GPU memory beyond the buffer object that backs the registered region. Affected systems are those running nouveau with the transparent-huge-page GPU memory migration support introduced by upstream commit c32287471077, which enlarged the registered device-private region to DMEM_CHUNK_SIZE * NR_CHUNKS while the backing VRAM buffer object remained DMEM_CHUNK_SIZE, so any page past the first chunk resolves to VRAM outside that object. Exploitation is local, requires authentication and a compatible NVIDIA GPU running nouveau rather than the proprietary driver, with high attack complexity; there is no KEV entry and no confirmed public exploit code, and EPSS is low at 0.19% (8th percentile), so the practical signal is a scheduled patch rather than an emergency.
RemediationAI
Vendor-released patch: Linux 7.2.9 and Linux 7.3-rc5, or any later stable release containing upstream commits 06c69c64686a31783c8bec6495dc22822d38c4ce or 64ca4cdd1031206424e6455f0ae4fb0560d8f46a, which size the VRAM buffer object to the full registered device-private region. Identify exposure before scheduling by confirming the running kernel postdates commit c32287471077 but predates either fix, and by checking whether an NVIDIA GPU is bound to nouveau and whether GPU memory migration is actually exercised; hosts without nouveau or without such hardware do not need emergency handling and can take the update in the normal maintenance window, consistent with the "this_week" priority rather than an act-now response. Where an immediate kernel update is not possible, the exposure surface can be removed by not binding the GPU to nouveau (for example using the proprietary NVIDIA driver or blacklisting the nouveau module), accepting the loss of nouveau functionality and any rendering or compute capability that depends on it, or by avoiding workloads that trigger device-private GPU memory migration on affected systems with the caveat that this narrows rather than eliminates the attack path. Verification of the applied fix should be against the vendor advisory and upstream commit rather than version string alone, since distribution kernels backport these changes.
Affected VersionsENISA
Affected Packages
Source: OSV (Open Source Vulnerabilities)
Affects your stack
Affects 1 package in euvd
Checking this CVE against your own stacks and servers is part of the Basic plan and up.
Technical ContextAI · show details
The flaw lives in the nouveau DRM driver's device-private memory subsystem, which registers a device-private region with the DRM layer and backs it with a VRAM buffer object. Commit c32287471077 ("gpu/drm/nouveau: enable THP support for GPU memory migration") changed nouveau_dmem_chunk_alloc() to register DMEM_CHUNK_SIZE * NR_CHUNKS of address space but did not resize the backing buffer object, which stayed at DMEM_CHUNK_SIZE. nouveau_dmem_page_addr() computes a page's VRAM address as chunk->bo->offset plus the page's offset within the registered region, so every page beyond the first chunk translates to an address outside the allocated buffer object. The resulting memory-safety class is an out-of-bounds VRAM access (no CWE identifier is assigned in the supplied data), which the assessed CVSS vector scores as high confidentiality and integrity impact with high availability impact (C:H/I:H/A:H) under a local, high-complexity, authenticated attack profile (AV:L/AC:H/PR:L/UI:N). Because the bug is a sizing mismatch between the registered region and its backing object, it is silent in normal operation and is reached only when a migration range extends past the first chunk. The affected CPE data identifies Linux kernel builds generally; the specific exposure is limited to kernels carrying the THP GPU-migration change and to machines whose NVIDIA GPU is bound to nouveau.
Know when the next one hits your stack
- Watch products and get an email when something new is published
- Check this CVE against your own stack and servers
- Exploit and PoC references, threat intelligence from 22 sources
No credit card · 7-day full trial
Severity by source
AV:L/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H
Local GPU-migration path requires an authenticated low-privileged user (PR:L) and a specific multi-chunk layout (AC:H); out-of-bounds VRAM read/write yields full C/I/A impact.
Primary rating from Vendor (Linux).
CVSS VectorVendor: Linux
Lifecycle Timeline
4DescriptionCVE.org
In the Linux kernel, the following vulnerability has been resolved:
drm/nouveau/dmem: pin VRAM for the whole registered range
Commit c32287471077 ("gpu/drm/nouveau: enable THP support for GPU memory migration") grew the device-private region that nouveau_dmem_chunk_alloc() registers from DMEM_CHUNK_SIZE to DMEM_CHUNK_SIZE * NR_CHUNKS, but left the VRAM buffer object backing that region at DMEM_CHUNK_SIZE.
nouveau_dmem_page_addr() returns chunk->bo->offset plus the page's offset within the registered region, so every page past the first chunk resolves to VRAM outside the buffer object.
Size the buffer object to the region it backs.
Executive ViewAI
A high-severity flaw in the Linux kernel's open-source NVIDIA 'nouveau' driver allows local, low-privileged users to access GPU memory beyond their authorized buffer, but only on systems using nouveau with specific memory migration support. The risk is contained to authenticated local users on affected hardware, and with low exploitation likelihood, this warrants a scheduled patch rather than an emergency response.
Apply vendor patch.
Threat Intelligence Overview
NVD
CVE.org
EPSS
How this decision was reached
The engine reads each signal below and applies a fixed list of rules from top to bottom. The first rule that matches sets the recommended action – like a statement, every line is shown.
What every action level means
SSVC AssessmentENISA
Priority Score
Attack Technique
Information disclosure occurs when an application unintentionally exposes sensitive data that aids attackers in reconnaissance or directly compromises security.
Exploit-DB