NoisyTCP Linux kernel vulnerability thumbnail
26
Views

NoisyTCP is a Linux kernel TCP flaw, tracked as CVE-2026-99288, that lets crafted packets crash servers, and this guide explains it simply. A new Linux kernel bug called NoisyTCP (CVE-2026-99288) can let a remote attacker crash a Linux server just by sending specially crafted network packets. No login, no password, no malware on the server. The machine simply panics and goes down.

⚡ Quick summary: NoisyTCP

  • NoisyTCP (CVE-2026-99288) is a flaw in the Linux kernel’s TCP handling.
  • Remote attackers can crash a server with crafted packets, a denial-of-service risk.
  • Check your distribution’s patch status and update your kernel.
NoisyTCP Linux kernel vulnerability thumbnail
NoisyTCP: the Linux kernel TCP flaw CVE-2026-99288

The good news? It does not let attackers steal data or take control of the server (as far as we know today), and a fix already exists in the upstream kernel. If you run websites, APIs or SSH servers on Linux, this one deserves ten minutes of your time. So, let’s break it down in simple words.

What is the NoisyTCP vulnerability?

TCP is the protocol that almost every internet connection uses. When you open a website or connect with SSH, your computer and the server first do a small “handshake” before any real data moves. The Linux kernel handles this handshake for every app on the server.

NoisyTCP is a race condition in this handshake code. Researcher Hyunwoo Kim disclosed it publicly on 15 November 2026. Under the right timing, the kernel ends up using a connection record that has already been freed and replaced. The result is a NULL pointer error inside the kernel, which causes a kernel panic – the Linux version of the “blue screen”.

Interestingly, the faulty code reportedly dates back to a 2015 change that made the kernel faster by handling new connections without taking a lock. So this bug has been hiding in Linux for around 11 years.

Key facts at a glance – NoisyTCP

These are the key facts on the NoisyTCP in one place.

  • CVE ID: CVE-2026-99288 (nickname: “NoisyTCP”)
  • Type: Race condition leading to a NULL pointer dereference in the Linux TCP stack
  • Impact: Remote, unauthenticated denial of service (server crash)
  • Severity: Red Hat rates it “Important” with a preliminary CVSS score of 7.5
  • Discovered by: Hyunwoo Kim
  • Affected: Almost all Linux distributions; Nginx, Apache and SSH servers in default setups are reportedly reachable
  • Fixed in: Upstream kernel commit 4e2b3ef19e0e (included in kernels 6.12.117, 6.18.63 and 7.3.4)
  • Exploited in the wild? No reports so far, and no public exploit code

How does NoisyTCP work? (Simple explanation)

Modern servers do not process all network traffic on one CPU. Instead, network cards have several receive (RX) queues, and Linux spreads incoming packets across many CPU cores. This makes servers fast. However, it also means two cores can work on related packets at the very same moment.

Here is the idea, step by step:

  • The attacker sends crafted TCP packets that play with SYN cookies, a feature that protects servers from SYN flood attacks.
  • Using a trick with IPv4 fragments, the attacker makes related packets land on different CPU cores.
  • One core frees a half-open connection record while another core is still looking it up without a lock.
  • The second core then reads an incomplete record, hits a NULL pointer, and the kernel panics.

Think of two bank clerks sharing one register. Clerk A tears up a form and puts a blank one in its place. At the same instant, Clerk B picks up the “form” and tries to process it. The blank form makes the whole system stop. That is a race condition.

The timing window is narrow. The researcher says hitting it on a quiet physical server can take a long time. However, reports suggest it is much easier on cloud VMs, which usually have multiple RX queues by default.

Who is affected? – NoisyTCP

You are most likely exposed if all of these are true:

  • Your Linux server runs an older, unpatched kernel.
  • It receives TCP traffic directly from the internet (for example, web or SSH ports are open).
  • Its network card has two or more RX queues, or RPS (Receive Packet Steering) spreads packets across CPUs.

Patch status as of 16 November 2026 (it is changing fast, so check your vendor):

  • Debian 13: fixed in linux 6.12.117-1
  • Debian 12: fixed in linux 6.1.192-1 (backport)
  • Arch Linux: fixed in linux 7.3.4 and linux-lts 6.18.63
  • Ubuntu, RHEL 8/9/10, Rocky, AlmaLinux, Fedora: reportedly no fixed package yet

What should you do? – NoisyTCP

Here is what to do about the NoisyTCP, step by step.

Step 1: Check your kernel version

uname -r
# Ubuntu (with Ubuntu Pro client)
sudo pro fix CVE-2026-99288

On Debian, search your installed kernel package changelog for “CVE-2026-99288”. On RHEL, Rocky or AlmaLinux, check dnf updateinfo for the CVE once errata are released.

Step 2: Update the kernel and reboot

A kernel fix only works after a reboot. Installing the package alone is not enough.

# Debian / Ubuntu
sudo apt update && sudo apt full-upgrade
sudo reboot

# RHEL / Rocky / AlmaLinux (once an update is released)
sudo dnf update kernel
sudo reboot

Step 3: Check your RX queues (are you in the risky setup?)

# Replace eth0 with your network interface name
ethtool -l eth0

Step 4: Temporary mitigations (only if you cannot patch yet)

Red Hat lists two stop-gap options. Both have trade-offs, so test them first:

  • Disable SYN cookies (net.ipv4.tcp_syncookies=0). This blocks the known crash path, but makes your server weaker against SYN flood attacks.
  • Use a single RX queue (for example ethtool -L eth0 combined 1, and turn off RPS). This removes the race, but can reduce network performance and does not survive a reboot.

Also, putting servers behind a load balancer, reverse proxy or DDoS protection service that terminates TCP can reduce direct exposure. In short, the real fix is a patched kernel plus a reboot. Related: read our explainer on Outlook zero-day.

In short, NoisyTCP can take Linux servers offline, so check your patch status and update the kernel.

Key takeaways on the NoisyTCP

  • NoisyTCP (CVE-2026-99288) lets a remote attacker crash Linux servers with crafted TCP packets.
  • It is a race condition in TCP handshake code that reportedly dates back to 2015.
  • Impact is denial of service, not data theft or code execution (based on current research).
  • Multi-queue cloud VMs are the most exposed setup.
  • Update the kernel and reboot as soon as your distribution ships a fix.

FAQ on the NoisyTCP vulnerability

These quick answers cover the most common questions about the NoisyTCP.

Can NoisyTCP be used to hack into my server?

Based on current reports, no. The demonstrated impact is a crash (denial of service). Researchers say deeper exploitation may be possible in theory but is hard and has not been shown.

Is my Linux laptop at risk?

The risk is much lower. Laptops usually sit behind a home router and do not accept incoming TCP connections from the internet. Servers with open ports are the main target. Still, install kernel updates when they arrive.

Do I need to reboot after updating?

Yes. The old kernel keeps running until you reboot (or use a live-patching service that covers this CVE).

Should I just turn off SYN cookies?

Only as a short-term fix if you cannot patch. SYN cookies protect you from SYN flood attacks, so disabling them trades one risk for another.

What is a race condition?

It is a bug where the result depends on the exact timing of two tasks running at the same time. If they “race” in the wrong order, the program sees inconsistent data and can crash.

GK Practice Questions on the NoisyTCP Vulnerability

Test yourself on the NoisyTCP with these practice questions.

Test your understanding with these practice questions.

Section A: Multiple Choice Questions (MCQs)

1. What is the CVE ID of the NoisyTCP flaw?
a) CVE-2026-98432 b) CVE-2026-99288 c) CVE-2026-123158 d) CVE-2026-43284

Show answer

Answer: b) CVE-2026-99288

2. NoisyTCP is mainly which type of attack impact?
a) Data theft b) Privilege escalation c) Denial of service d) Phishing

Show answer

Answer: c) Denial of service

3. Which kernel feature, meant to stop SYN floods, is involved in the crash path?
a) SELinux b) SYN cookies c) iptables d) cgroups

Show answer

Answer: b) SYN cookies

4. Who disclosed the NoisyTCP vulnerability?
a) Hyunwoo Kim b) Özgün Kültekin c) Linus Torvalds d) Greg Kroah-Hartman

Show answer

Answer: a) Hyunwoo Kim

5. What preliminary CVSS score did Red Hat give NoisyTCP?
a) 5.3 b) 7.5 c) 9.8 d) 10.0

Show answer

Answer: b) 7.5

6. The exploit needs packets to be processed on multiple CPUs. Which setup enables this?
a) Single RX queue b) Two or more RX queues or RPS c) IPv6 only d) Disabled firewall

Show answer

Answer: b) Two or more RX queues or RPS

Section B: Short Answer Questions

7. What does TCP stand for?

Show answer

Answer: Transmission Control Protocol.

8. What is a kernel panic?

Show answer

Answer: A fatal error where the Linux kernel stops working and the system crashes or reboots.

9. In which year was the code that caused NoisyTCP reportedly introduced?

Show answer

Answer: 2015.

10. Name one temporary mitigation for NoisyTCP.

Show answer

Answer: Disable SYN cookies, or serialize traffic to a single RX queue.

11. What must you do after installing a kernel update for the fix to take effect?

Show answer

Answer: Reboot the system (or apply a live patch).

Section C: Descriptive / Mains Questions

12. Explain what a race condition is and how NoisyTCP uses one to crash Linux servers.

Show answer

Define a race condition (outcome depends on timing of parallel tasks). Explain multi-queue packet processing across CPUs, lockless socket lookups added for speed, a connection record being freed and replaced on one core while another reads it, and the resulting NULL pointer dereference and kernel panic.

13. “Performance optimisations can create security risks.” Discuss with reference to the NoisyTCP flaw.

Show answer

Cover the 2015 lockless change that improved speed, how parallel processing on multi-queue NICs opened a timing window, the bug staying hidden for years, the trade-off in mitigations (SYN cookies vs SYN floods, single queue vs performance), and the need for code review, fuzzing and fast patching in critical infrastructure.

Related reading: VMware vCenter Zero-Day CVE-2026-123890: How Hackers Became Admin Without a Password – Explained Simply · Outlook Zero-Day CVE-2026-124194: How One Message File Can Leak Your Windows Password Hash – Explained Simply

Sources for the NoisyTCP

Article Categories:
Information Security

Leave a Reply

Your email address will not be published. Required fields are marked *