ITnews

Kernel-crashing NoisyTCP bug sat in Linux for 11 years


Key points

  • A high-severity Linux kernel flaw dubbed NoisyTCP, tracked as CVE-2026-99288, lets unauthenticated attackers remotely crash internet-facing servers and has existed since October 2015.
  • The bug, rated 7.5 out of 10 by Red Hat, exploits a race condition in TCP connection lookups involving SYN cookies and is fixed in kernels 6.12.117, 6.18.63 and 7.3.4.
  • Administrators unable to patch immediately can disable SYN cookies or reduce RX queues and turn off RPS and RFS, though these workarounds have trade-offs.



A security researcher has disclosed a high-severity flaw in the Linux kernel that lets unauthenticated attackers remotely crash internet-facing servers running default network configurations.

What’s more, the now-patched “NoisyTCP” bug, tracked as CVE-2026-99288 has been present in the Linux kernel since October 2015, affecting virtually every release since then.

Red Hat rates NoisyTCP as 7.5 out of 10 and it can be triggered by specially crafted packets targeting two features in the Linux network stack, namely SYN cookies.

Researcher Hyunwoo “V4bel” Kim found a race condition in how the kernel looks up TCP connections, which an attacker can exploit with specially crafted packets that abuse SYN cookies, a defence against connection flooding attacks.

Unpatched Linux-based systems that run network services are at risk if they have multiple receive (RX) queues or Receive Packet Steering (RPS) enabled, both of which spread incoming packets across processor cores.

Linux stable kernels 6.12.117, 6.18.63 and 7.3.4 contain the fix for NoisyTCP.

Kim does not claim NoisyTCP can be used for remote code execution (RCE), although he said a second, harder-to-trigger path can cause a remote use-after-free memory error.

“But remote DoS can be achieved without much difficulty,” Kim said.

How quickly the bug can be triggered depends heavily on where the target runs.

Kim said default installations of Nginx, Apache and SSH on QEMU/KVM virtual machines crashed within five seconds to two minutes.

A quiet bare-metal server running Nginx held out for about 14 hours, even against a proof-of-concept tuned for that environment.

The difference comes down to timing: hypervisors can pause a virtual processor mid-operation, widening the narrow window the attack depends on.

Applying the NoisyTCP patch is the recommended fix, but administrators who can’t do so can disable SYN cookies to block the crash path, which must be done in each container’s network namespace and weakens protection against SYN flood attacks.

That workaround does not cover the second, harder-to-trigger memory corruption path.

Alternatively, administrators can reduce the receiving network interface to a single RX queue and turn off RPS and Receive Flow Steering (RFS), which Kim said blocks the race.

Earlier this year, Kim reported the “Dirty Frag” Linux local privilege escalation flaw.



Source link