An 11-Year-Old Linux Kernel Bug Lets Outsiders Crash Web and SSH Servers. Your Questions on NoisyTCP, Answered
NoisyTCP (CVE-2026-99288) is a race condition in the Linux kernel’s TCP code that lets an unauthenticated attacker panic a server running ordinary services such as Nginx, Apache or SSH. A fix is in the mainline kernel. Here is who is exposed, how hard it is to trigger, and what to do before you can patch.
A newly published Linux kernel flaw can knock a server offline from across the internet without a password, a login or any unusual setup on the target. The bug, named NoisyTCP and tracked as CVE-2026-99288, sat in the kernel’s TCP networking code for roughly 11 years. It was found and patched by security researcher Hyunwoo Kim, and it was reported by BleepingComputer; we checked the details against the researcher’s public advisory and Red Hat’s CVE page.
What exactly goes wrong?
It is a race condition. When TCP packets arrive on different receive paths at the same moment, the kernel can look up a connection-request record that is being freed and reused for a new SYN cookie request. The recycled record has no listening socket attached, the kernel follows a null pointer, and the whole operating system panics. Red Hat describes it as recycled socket memory being accessed before it has been properly set up.
Does an attacker need an account or special access?
No. The packets only have to reach a listening TCP service directly. The researcher names Nginx, Apache, SSH servers and proxies running their default settings as examples. The target does not need a rare configuration or a special firewall rule; a service that simply listens for connections is enough, as long as the machine has two or more network receive queues or uses software packet steering (RPS). The researcher says no special configuration is needed beyond that.
Can it be used to break in, or only to crash things?
As disclosed, the practical impact is denial of service. Red Hat rates the flaw Important and says it does not allow data to be changed or code to be run. The researcher notes a second, harder path that could produce memory corruption, but says he has only built a limited exploit for a lab VM setup and is still researching whether a general one is possible. Remote crashes, he says, are not difficult.
If a website runs in a container, is only the container affected?
No. Because the bug is in the host kernel, a crash triggered through a containerized web server takes down the entire host and everything else running on it.
How quickly can it be triggered?
- Virtual machines: against default Nginx, Apache and SSH setups on QEMU/KVM guests, the researcher reports a crash in about 5 seconds to 2 minutes.
- Quiet physical servers: against default Nginx on an idle bare-metal machine with eight receive queues, it took about 14 hours, even with a proof of concept tuned for that machine.
In other words, virtualized servers, which host a large share of business websites and apps, are the easier targets.
Which systems are affected?
The researcher traces the flaw to a kernel change from October 2015 and says it affects almost all Linux distributions. The fix was merged into the mainline kernel on November 3, 2026, and Red Hat’s page references corresponding fixes in the stable kernel branches. Check your distribution’s security tracker to see when a patched kernel package is available for your systems.
What if we cannot patch right away?
The advisory lists two stopgaps, each with a cost:
- Turn off SYN cookies (setting
net.ipv4.tcp_syncookiesto 0). This closes the code path the bug needs, but it weakens protection against SYN flood attacks. The setting is per network namespace, so it must be applied inside every container that has its own network stack. - Serialize incoming traffic by reducing the internet-facing interface to a single receive queue and disabling RPS and RFS. This removes the concurrency the race depends on, at some cost to throughput.
What this means for your company
If you run Linux anywhere that faces the internet, such as a website, a customer portal, a VPN or SSH jump box, or a cloud VM behind a load balancer, assume you are in scope until your provider or distribution says otherwise. A crash is not a data breach, but an outage on a revenue-producing site is still a business problem, and a host that can be rebooted at will by a stranger is an easy distraction during a larger attack. Inventory your Linux hosts, including appliances and managed hosting where someone else controls the kernel, schedule the kernel update with a reboot, and confirm the new kernel is actually running afterward. Ask hosting and SaaS providers whether they have patched.
Not sure how many Linux servers you have facing the internet, or who is responsible for rebooting them into a patched kernel? Reach out to Layer 7 Data Solutions to discuss how we’ve helped companies similar to yours.
