CVE-2026-64569 is a Linux kernel denial-of-service flaw in the MPLS route-dump path, not a Windows vulnerability, despite appearing in Microsoft’s Security Update Guide. The immediate practical message for Windows administrators is that there is no identified Windows build, KB package, or Windows Update action attached to this record; the systems that need attention are custom Linux kernels built with MPLS enabled and IPv4 networking deliberately disabled.

Microsoft’s entry supplies the CVE identifier and the terse title, “mpls: fix NULL deref in mpls_valid_fib_dump_req() on CONFIG_INET=n,” but it does not identify an affected Microsoft product, a severity rating, a CVSS vector, an update, a mitigation, or even a kernel-version range. That omission matters here because the technical record in the Linux networking tree shows a tightly constrained bug whose relevance cannot be inferred from a generic Microsoft advisory page.

There is also a timing problem in the submitted metadata. It labels the item as published at 1:43 a.m. Pacific Daylight Time on August 9, 2026 — 08:43 UTC — which is 86 minutes after the 07:17 UTC review time. The advisory may be in the process of publication, but its stated timestamp should not be read as proof that an accompanying Windows security release exists.

Diagram showing a Linux MPLS route-dump null-pointer kernel crash, while Windows systems remain unaffected.The crash sits in a configuration-specific MPLS fallback​

The Linux networking tree records the correction as commit

56d96fededd61192cd7cc8d2b0f36adfd59036c3

, merged by network maintainer Paolo Abeni on July 17, 2026. It fixes

mpls_valid_fib_dump_req()

, the routine that validates strict netlink requests to dump MPLS forwarding information.

The defect is simple but consequential. When Linux is compiled with

CONFIG_INET=n

, the MPLS implementation cannot delegate this validation to the normal IPv4 helper,

ip_valid_fib_dump_req()

. Instead, it uses its own fallback parser and iterates through a table of netlink attributes.

For the output-interface attribute,

RTA_OIF

, that fallback code called

nla_get_u32(tb*)

even when the parsed attribute pointer was absent. A route-dump request for the

AF_MPLS

address family, made with strict validation enabled but without

RTA_OIF

, could therefore make the kernel dereference a null pointer.

The upstream commit includes a Kernel AddressSanitizer trace showing the fault in

mpls_valid_fib_dump_req()

, reached through

mpls_dump_routes()

, netlink dump handling, and

rtnetlink_rcv_msg()

. This is a crash condition: it can take down the affected kernel rather than expose memory or execute attacker-controlled code.

The repair adds a three-line guard that skips attributes not supplied in the request. That makes the MPLS-only fallback match the safer behavior already used by the IPv4-backed implementation.

An unprivileged local user can trigger the vulnerable request​

The upstream commit makes an important point absent from the Microsoft entry: the relevant

RTM_GETROUTE

operation is permitted without

CAP_NET_ADMIN

. In the vulnerable configuration, a local unprivileged process can submit the malformed-but-parseable route-dump request and trigger the null dereference.

That makes CVE-2026-64569 more than an administrator-only configuration bug. On a qualifying device, it is a local availability issue that could let an ordinary account crash the kernel. In multi-user or container-heavy deployments, a kernel panic is an outage regardless of whether the attacker can read protected data or elevate privileges.

The preconditions are unusually narrow, however:

  • The kernel must include MPLS support through CONFIG_MPLS.
  • The kernel must be compiled with CONFIG_INET=n, excluding normal IPv4 networking support.
  • The requester must reach the strict-checking MPLS route-dump validation path without including the output-interface attribute.

Linux’s own Kconfig description says MPLS is disabled by default and advises users unfamiliar with it to leave it disabled. Separately,

CONFIG_INET

is the kernel’s TCP/IP networking option and is ordinarily enabled on general-purpose Linux distributions. The affected combination is therefore plausible for specialized networking appliances, embedded images, lab builds, and stripped-down custom kernels, but it is a poor match for a conventional server, desktop, or cloud guest.

The vulnerable code dates to Linux 4.20-era work​

The fix identifies the introducing change as commit

196cfebf8972

, “net/mpls: Handle kernel side filtering of route dumps.” Linux development history places that commit on October 16, 2018, before Linux 4.20 shipped on December 23, 2018.

That means this is not a short-lived regression in a newly released kernel. The unsafe fallback has existed in the relevant source line for nearly eight years, while being effectively masked on the much more common

CONFIG_INET=y

builds. The configuration split is why the flaw escaped the earlier review of a related March 2026 patch.

That earlier patch, posted to the Linux kernel mailing lists by Yiqi Sun, identified the missing null check and initially proposed rejecting the request with

-EINVAL

. Paolo Abeni suggested a cleaner reordering: ignore absent attributes first, then process the output-interface selector only when it exists, while continuing to reject attributes that are present but unsupported. The final July fix follows that design.

This sequence changes how administrators should interpret the CVE. The flaw was visible to code review in March, but the final commit is not merely a defensive null check stapled onto an input parser. It preserves the intended route-dump semantics: no interface filter means no interface filter, while an invalid supplied attribute remains an error.

Why the Microsoft listing should not trigger Windows patching​

Microsoft’s Security Update Guide is authoritative for Microsoft’s own advisories, but the supplied entry presently does not connect this CVE to Windows, Windows Server, Azure, Hyper-V, WSL, or a Microsoft-maintained Linux kernel package. It also provides no affected build numbers and no security-update table. Treating its presence there as evidence of a Windows kernel bug would be a category error.

Windows does not execute this Linux MPLS code in its own kernel. WSL 2 uses a Linux kernel inside its utility VM, but the trigger requires a deliberately nonstandard kernel build with

CONFIG_INET=n

; ordinary WSL 2 networking depends on IP networking being available. The record therefore does not establish that stock WSL installations are affected.

Custom WSL kernel users should still verify their own

.config

rather than rely on that general conclusion. If a bespoke WSL 2 kernel has both MPLS enabled and

CONFIG_INET

disabled, it deserves the same review as any other specialized Linux image. The Microsoft entry offers no version mapping that would let administrators determine that from the Windows host build alone.

At the time of review, the National Vulnerability Database did not return an indexed public entry for CVE-2026-64569, and CVE.org did not surface a usable public record. No independent outlet appears to have published a separate assessment of affected distribution packages or stable-kernel backports. That leaves the upstream Linux commit as the primary technical record and makes vendor package tracking more important than a generic CVE dashboard status.

What Linux and Windows administrators should do now​

Linux appliance and embedded-device maintainers should inspect the running or build-time kernel configuration for

CONFIG_MPLS

and

CONFIG_INET

. Systems with

CONFIG_INET=y

do not compile the vulnerable fallback routine; systems without MPLS support do not expose this MPLS route-dump code at all.

For the narrow group that has both MPLS enabled and IPv4 support disabled, the appropriate remediation is to consume a kernel release or vendor backport containing upstream commit

56d96fededd61192cd7cc8d2b0f36adfd59036c3

. If an immediate kernel replacement is not feasible, restricting untrusted local code is the practical exposure reduction, because the upstream record says the request does not require

CAP_NET_ADMIN

.

Windows administrators should not hunt for a KB or force a Windows Update deployment for CVE-2026-64569. The concrete outstanding task is for Microsoft, the CVE record, or Linux distributors to publish an affected-version and fixed-version map; until that exists, the only defensible triage is configuration-based, and most Windows and standard WSL systems fall outside the demonstrated trigger path.


References​

  1. Primary source: MSRC
    Published: August 9, 2026 at 8:43 AM UTC
  2. Related coverage: gbmc.googlesource.com
  3. Related coverage: msrc.microsoft.com
  4. Related coverage: cve.org
  5. Related coverage: nvd.nist.gov
  6. Related coverage: nvd.nist.gov
  7. Related coverage: cve.org
  8. Related coverage: linux.googlesource.com
  9. Related coverage: coral.googlesource.com
  10. Related coverage: git.zx2c4.com
  11. Related coverage: android.googlesource.com
  12. Related coverage: git.evolution-x.org
  13. Related coverage: git.zx2c4.com
    *