Linux

NoisyTCP Linux kernel vulnerability can crash servers remotely

Security researcher Hyunwoo Kim has published details of NoisyTCP, a Linux kernel vulnerability tracked as CVE-2026-99288 that lets an attacker crash a server by sending it TCP traffic, with no login required. The bug dates back about 11 years, and fixed kernels are already available from the Linux stable branches and Debian.

By Techieslite ·

Techieslite graphic with the Linux logo and the headline "NoisyTCP Linux kernel flaw can crash servers"

What NoisyTCP does

The flaw sits in the code the kernel uses to look up established TCP connections for both IPv4 and IPv6. That lookup runs without locks, so in a narrow race the kernel can grab a socket whose memory was just reused. In Kim's proof of concept, the reused socket is a SYN cookie request with no listener attached, and the kernel dereferences a NULL pointer and panics.

Kim says the bug can be triggered against "plain" TCP listeners, which means Nginx, Apache and SSH servers running their default configurations are reachable targets. The race needs a server with two or more receive (RX) queues, or with Receive Packet Steering (RPS) spreading packets across CPUs. Because the crash happens in the kernel, a web server running inside a container takes down the entire host with it.

A second path can lead to a use-after-free, but Kim says it requires several nested races and is generally hard to trigger. In the advisory's FAQ, Kim answers the question of whether NoisyTCP is a remote code execution bug with "Not yet. and basically, no." The practical risk is a remote denial of service.

How hard it is to exploit

How long an attack takes depends on the machine. Against Nginx, Apache and SSH servers on QEMU/KVM virtual machines, Kim says the crash took between 5 seconds and 2 minutes to trigger. Against Nginx on a quiet bare-metal server with 8 RX queues, it took about 14 hours, even with a proof of concept tuned for that machine.

Kim reported the bug to the linux-distros mailing list and published the write-up after the embargo ended. The flaw was introduced in a kernel change from October 3, 2015, and Kim says it affects almost all Linux distributions.

Which kernels are fixed

Kim wrote the fix, which the kernel commit titles "inet: reject unhashed sockets during established lookups." The official CVE record lists these versions as fixed:

Kernel.org released 7.3.4, 6.18.63 and 6.12.117 on Tuesday, November 10. Debian has shipped fixed kernels for Debian 12 bookworm (6.1.192-1) and Debian 13 trixie (6.12.117-1) through its security updates. Ubuntu rates the bug as medium priority and, as of its last update on November 11, lists its main supported releases as "Needs evaluation," so fixed Ubuntu kernels aren't out yet.

What it means for you

If you run a Linux server that accepts connections from the internet, install your distribution's kernel update as soon as it's available and reboot, since a new kernel only takes effect after a restart. Servers running in virtual machines deserve attention first, because that's where Kim found the crash easiest to trigger.

If you can't patch yet, Kim lists two temporary workarounds. One is to turn off SYN cookies with sudo sysctl -w net.ipv4.tcp_syncookies=0, which must be applied separately in each container that uses its own network namespace and weakens protection against SYN flood attacks. The other is to limit the internet-facing network interface to a single RX queue and turn off RPS and RFS.

The attack depends on the attacker's TCP packets reaching a listening service on the machine. A Linux laptop or an old PC running a lightweight Linux distro should still get the fix through normal system updates, but the servers Kim describes are the main targets.

Sources