A remote crash bug sounds less scary than remote code execution until you realize the fix is a kernel update, and the hard part is not the patch, it is rebooting every box that quietly depends on staying up. That is the operational problem behind the NoisyTCP Linux kernel vulnerability: how you get a patched kernel onto a fleet you cannot take down all at once.
What the NoisyTCP Linux Kernel Vulnerability Is
In a November 15 post, researcher Hyunwoo Kim disclosed NoisyTCP as CVE-2026-99288, a pre-authentication flaw in the Linux kernel’s TCP code. An unauthenticated attacker who can send packets to a reachable TCP listener can trigger a race in a lockless socket lookup. The kernel can then follow a null pointer and panic. Kim says that can happen against “plain” TCP listeners, including Nginx, Apache, and SSH servers running with their default configuration, and that a crash in a container takes down the entire host, because the bug is in the shared kernel.
The Linux kernel CVE announcement, posted by Greg Kroah-Hartman on November 10, confirms the crash. It says a proof-of-concept “reproduced the crash,” and the resulting trace ends in “Kernel panic – not syncing: Fatal exception in interrupt.” The announcement describes a missing check during established-socket lookups, so a recycled socket can be returned in a bad state. I am not going to walk through the packet sequence. The operator-relevant fact is a remote denial of service, not a login bypass.
Red Hat, on its CVE page, rates the issue Important and says “an unauthenticated remote attacker can trigger an immediate operating system crash across the network without requiring privileges,” and that the flaw “does not allow data alteration or arbitrary code execution.” Kim also describes a second, harder race that can lead to a use-after-free, but says it is generally hard to trigger. Treat the confirmed impact as a host crash.
Eleven Years, Kernel 4.4 and Later
Kim dates the introducing commit, 079096f103fa, to October 3, 2015, and the fix, 4e2b3ef19e0e, to November 3, 2026. He writes that the vulnerability had been latent in the Linux kernel for about “11 years” and “affects almost all Linux distributions.” The kernel CVE announcement is more precise about versions: the issue was “introduced in 4.4” and is fixed in upstream stable releases 6.12.117, 6.18.63, and 7.3.4. Documented exceptions exist. Ubuntu lists 14.04 LTS as not affected. Red Hat lists RHEL 6 and 7 as not affected, which fits kernels older than 4.4.
The triggering conditions matter for triage. Kim says the race needs concurrent receive contexts, so it is a concern on hosts with two or more receive queues, or where Receive Packet Steering spreads packets across CPUs. If an attacker can reach a listening port on a typical multi-queue server, you should assume it is in scope.
Who Has a Fixed Kernel, and Who Does Not
This is the part that will change this week, so confirm it before you rely on it. As of November 20, 2026:
- Upstream: Stable kernels 6.12.117, 6.18.63, and 7.3.4 include the fix, per the kernel CVE announcement. A November 20 oss-security follow-up noted a backport in 5.15.226-rc1, expected around November 20 or 21. That is a release candidate, not a finished distro package.
- Debian: The Debian security tracker lists bookworm (Debian 12) as fixed in linux 6.1.192-1 (DLA-4906-1) and trixie (Debian 13) as fixed in 6.12.117-1 (DSA-6628-1). Forky and sid are listed as fixed starting at 7.3.4-1.
- Ubuntu: Canonical’s CVE page, last updated November 20, still lists the linux package as Vulnerable on supported releases from 16.04 LTS through 26.10, and rates the Ubuntu priority High. No fixed version is shown. Ubuntu 14.04 LTS is listed as not affected.
- Red Hat: RHEL 8, 9, and 10, including real-time kernels on 8 and 9, and OpenShift Container Platform 4 and 5, are listed as Affected, with no advisories attached. RHEL 6 and 7 are not affected.
- SUSE: The SUSE CVE page, as of this check, shows an overall state of Pending, an Important rating, and no SUSE-SU announcement. Current products including SLES 15 SP7, SLES 16.0 and 16.1, and SUSE Linux Micro 6 are listed as Affected, with some kernel-source entries In progress.
I found no entry for CVE-2026-99288 in CISA’s Known Exploited Vulnerabilities catalog (catalog version 2026.11.20), and no Canonical Livepatch notice, Red Hat kpatch advisory, or SUSE live-patch announcement that names this CVE. Those products are the right place to watch, but a livepatch can only follow a vendor fix.
Temporary Mitigations, and Their Costs
Kim’s advisory, which Red Hat repeats, offers two temporary measures if you cannot patch immediately. Neither replaces a fixed kernel.
- Disable TCP SYN cookies. This closes the request-socket path described in the advisory. The setting is per network namespace, so it has to be set inside each container that uses its own namespace. Both Kim and Red Hat warn that disabling SYN cookies lowers protection against SYN flood attacks.
- Reduce the inbound interface to a single receive queue, and disable RPS and RFS. The race needs concurrent receive contexts. This may cost network performance on busy hosts.
Patching a Fleet You Cannot Reboot All at Once
These steps are my recommendations, not guidance from the kernel team or any vendor.
- Inventory exposure first. List internet-facing Linux hosts, and any host that terminates TCP for Nginx, Apache, SSH, load balancers, or similar listeners. Note the running kernel, the distribution, and whether the box is a container host. A container host is a higher priority, because one panic is a whole-node outage.
- Install a fixed kernel where one exists. Debian operators should be moving to the versions above. Everyone else should watch their vendor CVE page daily until a package appears. Upstream stable versions help only if you actually run those kernels.
- Stage the reboot, do not wait to plan it. Build a rolling window: internet-facing hosts first, then container hosts, then internal listeners. Appliances, validators, and hosts that are not supposed to reboot need a named owner and a fallback.
- Watch livepatch channels, but do not assume they will land first. If your vendor later ships a livepatch, use it to shrink the window. Until then, the only complete fix is a patched kernel and a reboot.
- Apply temporary mitigations only with a ticket and a reversal plan. Disabling SYN cookies or collapsing receive queues buys time. It also changes how the host behaves under flood or load, so write down what you changed and when you will undo it.
The Takeaway
NoisyTCP is a confirmed, remote, unauthenticated crash in a very old path in the Linux TCP stack. The hard part for operators is the reboot. Check your kernel, install a fixed package if your distribution has one, and start scheduling the rest of the fleet.

Leave a Reply