HomeTech NewsCybersecurityNoisyTCP Flaw Can Remotely Crash Linux Servers Before Login

NoisyTCP Flaw Can Remotely Crash Linux Servers Before Login

Hyunwoo Kim, writing as @v4bel, has disclosed a Linux kernel TCP flaw that can crash a server from the network before any login takes place. In an oss-security post dated 16 November at 01:00 Korea time (15 November, 16:00 GMT), he called the issue NoisyTCP and said it is tracked as CVE-2026-99288. The main effect he describes is a remote denial of service: a kernel panic caused by a NULL listener being dereferenced. Because the bug sits in the kernel’s TCP stack, he wrote, a crash against a web server running in a container “takes down not only the web server application but the entire host.”

The Linux kernel CVE team had already assigned the CVE on 10 November, in a notice from Greg Kroah-Hartman titled “inet: reject unhashed sockets during established lookups.” That notice is the official kernel CNA record. Kim said an embargo agreed with linux-distros had expired before he posted the public advisory.

What is fixed, and in which kernels

Kim said the flaw was introduced in commit 079096f103fa on 3 October 2015 and fixed in mainline in commit 4e2b3ef19e0e on 3 November 2026, “about 11 years” later. The CNA notice gives the same introduction point, in kernel 4.4, and lists stable fixes in 6.12.117, 6.18.63 and 7.3.4.

Those three version numbers check out on kernel.org. As of 16 November, the site lists 7.3.4 as the current stable release and 6.18.63 and 6.12.117 as current longterm releases, all dated 10 November 2026. The changelogs for those three releases each contain the corresponding fix commit named by the CNA, under the heading “inet: reject unhashed sockets during established lookups,” authored by Hyunwoo Kim. The CNA also says unaffected versions may change as fixes are backported, and it points readers to the CVE record for updates. The older longterm branches still listed on kernel.org on 16 November — 6.6.162, 6.1.192, 5.15.225 and 5.10.274 — are dated 6 November, before those fix releases, and they are not named as fixed in the 10 November CNA notice.

The kernel CVE team’s advice is to update to the latest stable kernel. It does not recommend cherry-picking individual commits.

Kim said the issue can be triggered, pre-authentication, against servers that an attacker’s TCP packets can reach directly, including Nginx, Apache and SSH running with their default configuration, on systems with two or more receive queues or where receive packet steering spreads packets across CPUs. He reported that, in his tests, it took “about 5 seconds to 2 minutes” to trigger on QEMU/KVM virtual machines, and “about 14 hours” on a quiet bare-metal Nginx server with eight receive queues. A second path, he said, can cause a remote use-after-free but is “generally hard to trigger.” This article does not reproduce a method for reaching either path.

If a patched kernel cannot be installed at once, Kim listed two temporary mitigations: turning off SYN cookies, which he said is set per network namespace and “lowers the protection against SYN floods,” or reducing the external interface to a single receive queue and disabling receive packet steering. Those are workarounds, not the fix.

A patched stable branch is not a patched machine

The official record is now clear on two points and still open on a third. The first is the bug itself: a pre-auth TCP race in the kernel that can panic a host, including a host that is only running the vulnerable service in a container, as Kim describes it. The second is where the kernel community has put the fix: mainline on 3 November, and, as confirmed on kernel.org, in the 10 November releases of 6.12.117, 6.18.63 and 7.3.4. The third is everything that sits between a stable tarball and a running machine. Linux distributions, cloud images and appliance vendors ship their own builds, on their own timetable. A version number on kernel.org is not, by itself, a statement that a given server has been updated.

The temporary mitigations should be read in the same light. Disabling SYN cookies, Kim warns, weakens protection against SYN floods and must be applied in each network namespace, including each container that has its own. Collapsing traffic onto a single receive queue is a configuration change with its own performance cost. Neither step is the patch, and neither is a reason to treat older longterm branches as already handled. The CNA leaves the door open for later backports; until those appear in a listed release, the conservative reading is that only the three named kernels, plus mainline after 3 November, are the documented fixes.

This disclosure also sits in a familiar kernel-fix queue, alongside Microsoft’s MDASH work on Linux kernel CVEs and CISA’s three-day order to patch the earlier Fragnesia kernel flaw. How unpatched kernel reports reach a public list has its own recent TSN account, in Trend Micro’s ZDI filings. None of those is this CVE. No source reviewed here reports that NoisyTCP has been used in attacks, and Kim does not present the harder use-after-free path as a documented remote-code-execution recipe. The practical next step the kernel CVE team already named is the ordinary one: move to a current patched stable kernel, and treat the workarounds as temporary if that update has to wait.

Sources

Related stories

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

  1. Five Firms Form AI Gigafactory Northern for a Planned Bid
    Data Centres & Energy

More in this category

Latest stories

  1. Five Firms Form AI Gigafactory Northern for a Planned Bid
    Data Centres & Energy

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.