Skip to content
TrustList
Security advisories

NoisyTCP: 11-year-old Linux flaw lets attackers crash servers

Editorial

By TrustList Editorial

CVE-2026-99288 needs no login and no special configuration, only a TCP port an attacker can reach. The fix went into the kernel on 3 November; two stopgaps exist for hosts that cannot take it yet.

About NoisyTCP: 11-year-old Linux flaw lets attackers crash servers

NoisyTCP: 11-year-old Linux flaw lets attackers crash servers

16 November 2026: A race condition in the Linux kernel's TCP code, tracked as CVE-2026-99288 and named NoisyTCP by the researcher who found and patched it, lets an attacker who is not logged in crash a server by sending packets to any listening TCP port. Red Hat rates it Important. The fix was merged into the kernel on 3 November, and the details became public on 16 November when the embargo agreed with Linux distributors ended.

Not yet independently verified. The technical account is the researcher's own, published on 16 November, and Red Hat's CVE record. We have not read each distribution's fixed package versions, and neither source reports attacks in the wild. We will update this when it can be confirmed, and remove this note.

A listening port is all it takes

Hyunwoo Kim, who reported the bug, says it can be triggered against what he calls plain TCP listeners: Nginx, Apache and SSH servers in their default configuration, with no unusual kernel option, firewall rule or socket type involved. The only condition is on the network card side. The machine must have two or more receive queues, or use Receive Packet Steering to spread packets across processors, which describes most servers and cloud instances.

The crash is a kernel panic, so it takes down the whole machine rather than one service. Kim points out the consequence for container hosts: an attack aimed at a web server inside a container stops the host and every other container on it. Red Hat's record describes the effect as an immediate operating system crash caused by an unauthenticated remote attacker, and classes the fault as a NULL pointer dereference.

Why the race is easier to win on virtual machines

The kernel looks up established TCP connections without taking a lock. NoisyTCP abuses the gap between two checks in that lookup: while one receive context is still holding the address of an old connection request, another frees it and builds a new SYN cookie request at the same memory address. The first context then picks up an object with no listener attached, and the receive path dereferences a NULL pointer.

How quickly that happens depends on timing noise. On QEMU/KVM virtual machines, where virtual processors can be preempted, Kim reports a crash within about five seconds to two minutes. Against Nginx on a quiet bare-metal server with eight receive queues it took about 14 hours, even with a proof of concept tuned for that machine. He also describes a second route that could give a remote use-after-free, and says it is hard to trigger because several races must succeed one inside another.

Eleven years of affected kernels

The bug arrived in a commit of 3 October 2015 and was fixed by commit 4e2b3ef19e0e on 3 November 2026, so it has been present for about 11 years and, in Kim's words, affects almost all Linux distributions. Red Hat's record carries the upstream patch title, "inet: reject unhashed sockets during established lookups", and a public date of 10 November. Each distribution ships the fix in its own kernel package, so the version to look for is in the vendor's advisory for CVE-2026-99288 rather than in a single number.

Two stopgaps, each with a cost

Applying the patched kernel is the priority. For hosts that cannot reboot yet, both sources give the same two temporary measures:

  • Turn off SYN cookies with sysctl -w net.ipv4.tcp_syncookies=0. The setting is per network namespace, so it has to be applied inside each container that has its own namespace, and it lowers the machine's protection against SYN floods.
  • Reduce the internet-facing interface to a single receive queue and disable RPS and RFS, which removes the concurrency the race needs at the price of network throughput.

Servers that sit behind a load balancer or proxy which terminates TCP are not reached directly by an attacker's packets; the balancer itself, if it runs Linux, is. Our patch management category lists the tools that track kernel updates across a fleet.

Sources

Categories & features

TrustList Weekly

The week in software and IT, in one email

The news that matters to buyers, new rankings and our own research. Every Thursday, free, and easy to leave.

We will email you to confirm. Unsubscribe with one click in any issue. Privacy policy

You may be interested in