Skip to content
ShiftDelete.Net
Search news/ TR Türkçe
News Feed AI Smartphones Games Cars Computers Social Media Space Reviews Latest news
  1. Home
  2. News

11-Year-Old NoisyTCP Linux Bug Can Crash Servers Before Login

A race condition in the kernel's TCP code lets attackers knock SSH, Nginx and Apache hosts offline remotely. Fixed stable kernels are out, but Ubuntu and Red Hat had not listed patched packages when we checked.

Hakkı ALKANHakkı ALKAN

Published 4 min read

Server rack in a data center with one Linux server gone dark, illustrating a NoisyTCP crash

NoisyTCP is an 11-year-old Linux kernel bug that lets a remote attacker crash an exposed server before any login, simply by sending crafted TCP packets. Tracked as CVE-2026-99288, it hits ordinary TCP services such as SSH, Nginx and Apache in their default configuration. Patched stable kernels are out, but when we checked the main distribution trackers on 16 November, only Debian had shipped fixed kernels.

Security researcher Hyunwoo Kim published the NoisyTCP advisory on the oss-security mailing list at 16:00 UTC on 15 November, after the embargo agreed with Linux distribution maintainers expired. The flaw was introduced in October 2015, in the code that went into Linux 4.4, and Kim says it affects almost all Linux distributions.

What NoisyTCP does

Kim describes NoisyTCP as a "pre-auth remote corruption" flaw in the kernel's TCP code. In plain terms, the kernel looks up TCP connections without locking them. If one CPU is still holding a stale reference to a half-open connection while another CPU frees that memory and reuses it for a new connection with the same details, the kernel can follow a pointer to nothing. The result is a kernel panic: the whole machine crashes.

No special setup is needed on the victim's side. A service only has to be listening for TCP connections. The one condition is that the server spreads incoming packets across several CPUs, either because its network card has two or more receive (RX) queues or because Receive Packet Steering (RPS) is turned on.

Because the bug sits in the kernel, the damage is not limited to the targeted program. Kim notes that if the attacked web server runs in a container, the entire host goes down with it.

How easy is it to trigger?

That depends heavily on the hardware. In Kim's tests, default Nginx, Apache and SSH servers running on QEMU/KVM virtual machines crashed within about 5 seconds to 2 minutes. On a quiet bare-metal server with 8 RX queues, a proof of concept tuned for that machine needed about 14 hours to crash default Nginx.

Kim also describes a second, more complex path that could lead to a remote use-after-free, a type of memory bug that can sometimes be turned into code execution. In the advisory's FAQ, on whether NoisyTCP is a remote code execution flaw, he writes: "Not yet. and basically, no." He says he has only a limited exploit for lab VMs, while a remote denial of service "can be achieved without much difficulty". None of the sources we checked reports attacks in the wild.

Red Hat rates the issue "Important" with a draft CVSS 3.1 score of 7.5. Its CVE page says an unauthenticated attacker can crash the operating system across the network, but cannot alter data or run code.

Is there a NoisyTCP patch for my distribution?

The upstream fix is dated 3 November. According to the Linux kernel CVE team's announcement, it is included in stable kernels 6.12.117, 6.18.63 and 7.3.4. The patch makes the kernel check that a connection it has found is still valid, and restart the lookup if not. The kernel team recommends moving to the latest stable kernel rather than applying the single commit.

Distribution kernels are a different story. This is what the official trackers showed when ShiftDelete.Net checked them at about 18:25 UTC on 16 November 2026. Status can change quickly, so check your vendor's page before acting.

DistributionStatus on 2026-11-16
Debian 12 "bookworm"Fixed in security updates (DLA-4906-1, linux 6.1.192)
Debian 13 "trixie"Fixed in security updates (DSA-6628-1, linux 6.12.117-1)
Debian unstable (sid)Fixed (7.3.4-1)
Debian testing (forky)Still listed as vulnerable (7.2.15-1)
Ubuntu 16.04 LTS to 26.10"Needs evaluation", no fixed kernel listed, Ubuntu priority Medium
Red Hat Enterprise Linux 8, 9, 10Affected, no fixed package listed (RHEL 6 and 7 not affected)

Sources: Debian security tracker, Ubuntu CVE page and Red Hat's CVE page.

What to do if you cannot patch yet

Kim's advisory, repeated by Red Hat, lists two temporary mitigations. Both have trade-offs.

  • Turn off SYN cookies with sudo sysctl -w net.ipv4.tcp_syncookies=0. SYN cookies are on by default in most distributions and protect against SYN flood attacks, so disabling them weakens that defence. The setting is per network namespace, which means it has to be applied separately inside every container that has its own network namespace.
  • Serialise incoming traffic by cutting the internet-facing interface down to a single RX queue and disabling RPS and RFS. This removes the race the attack depends on, at the cost of spreading network load across fewer CPUs.

For most administrators, the practical order is: install the fixed kernel as soon as your vendor ships it and reboot, and use a mitigation only on exposed machines that cannot be updated right away. Full technical details are in Kim's NoisyTCP write-up on GitHub.

Patch gaps like this are drawing more attention as researchers and attackers automate bug hunting. OpenAI recently said AI helped it find 948 unique local root exploits built on Linux kernel vulnerabilities, and Google says attackers have started using AI to turn security patches into working exploits.

Advertisement