A flaw in the Linux kernel’s TCP code, nicknamed NoisyTCP, lets someone with no account crash a server by sending it crafted traffic. Nginx, Apache and SSH servers on default settings are all exposed. Researcher Hyunwoo Kim says the bug has been in the kernel since October 2015.
Get the CyberExperts Daily Brief: the day's top security news in 5 minutes. Free.
A crash is the outcome. Red Hat says the flaw does not let an attacker change data or run code. A kernel panic still takes the whole host down, along with every container and site on it, and public proof-of-concept code is now on GitHub.
NoisyTCP (CVE-2026-99288) lets an outsider crash a Linux server with crafted TCP traffic. On VMs, Kim’s tests crashed default Nginx, Apache and SSH setups in seconds to minutes. Install the fixed kernel and reboot.
The 5-Minute Cyber BriefDon’t Miss the Policy Changes That Affect Security Decisions
Get the key CISA actions, new regulations, guidance, and risk shifts in a quick daily brief.
By subscribing you agree to our Privacy Policy.
Free. Weekday mornings. 5 minutes or less.
A Linux TCP bug lets outsiders crash servers with crafted packets
CVE-2026-99288 is a race condition in the kernel’s lockless lookup of established TCP connections. Socket memory can be reused before the new socket is fully set up, and with traffic built around SYN cookies the kernel ends up reading a NULL pointer and panics.
- Fixed in: Linux stable kernels 6.12.117, 6.18.63 and 7.3.4. The upstream fix, “inet: reject unhashed sockets during established lookups”, was merged about two weeks before the disclosure.
- Debian: fixed in 6.1.192-1 for bookworm and 6.12.117-1 for trixie.
- Ubuntu: kernels for every supported release were still marked “Needs evaluation” on Ubuntu’s CVE page.
- Red Hat: rates it Important, 7.5, and publishes two mitigations on its CVE page.
Internet-facing Linux servers and VMs running web or SSH services
The attack needs a server with two or more network receive queues, or Receive Packet Steering spreading packets across CPUs. Many servers and cloud VMs spread incoming packets across CPU cores this way. Kim told iTnews that default Nginx, Apache and SSH installs on QEMU/KVM virtual machines crashed within five seconds to two minutes, while a quiet bare-metal Nginx server held out for about 14 hours.
No attacks have been reported, and the flaw is not in CISA’s Known Exploited Vulnerabilities catalog. With working code public, a crash campaign against unpatched hosts costs an attacker very little.
Install your distribution’s fixed kernel and reboot
- Run
uname -ron each server and compare it with your distribution’s fixed kernel for CVE-2026-99288. A package update alone isn’t enough: the host has to reboot into the new kernel. - Patch internet-facing hosts first: web servers, SSH bastions, load balancers and Kubernetes nodes. Containers share the host’s kernel, so the node is what needs the update.
- Where no fixed kernel is out yet (Ubuntu at the time of writing), Red Hat’s first mitigation is
sysctl -w net.ipv4.tcp_syncookies=0. It must be set in every network namespace, including each container that has its own, and it weakens protection against SYN floods. - The second mitigation is to cut the outside-facing interface to a single receive queue and turn off RPS and RFS. That blocks the race but costs throughput, so test it first.
Forward this
For whoever runs our Linux servers: a newly disclosed kernel bug lets anyone on the internet crash a Linux server running web or SSH services, with code to do it already public. Fixed kernels are out for Debian and the upstream stable branches; other distributions are rolling them out. Can we schedule kernel updates and reboots for our internet-facing servers this week?
Details
- CVE-2026-99288 (NoisyTCP): NULL pointer dereference (CWE-476) after a race in TCP established-socket lookup. Red Hat CVSS 3.1: 7.5. NVD: received November 10, 2026, no NVD score yet. CVE record published November 10 by the Linux kernel CNA.
- Fixed: stable 6.12.117, 6.18.63, 7.3.4; mainline 7.4-rc2. Debian DSA-6628-1 (trixie 6.12.117-1) and DLA-4906-1 (bookworm 6.1.192-1).
- Second path: Kim says a harder-to-trigger path can cause a remote use-after-free. The SYN cookie mitigation does not cover it.
- Exploitation: none reported; not in CISA’s KEV catalog as of its November 13 update. Public proof of concept on GitHub. Status: THEORETICAL.
Hunt / verify
- Check for unexpected reboots since early November with
last -x reboot, and read the previous boot’s kernel log withjournalctl -k -b -1. - In crash logs or the cloud serial console, look for “BUG: kernel NULL pointer dereference” with
tcp_v4_rcvin the call trace. That matches the crash in the CVE record. - Use
ethtool -l <interface>to see how many receive queues each exposed interface has.
Source: Red Hat CVE-2026-99288; BleepingComputer
Sources
- Red Hat: CVE-2026-99288
- NVD: CVE-2026-99288
- Linux kernel stable commit: inet: reject unhashed sockets during established lookups
- Debian security tracker: CVE-2026-99288
- Ubuntu: CVE-2026-99288
- BleepingComputer: New NoisyTCP vulnerability lets attackers crash Linux servers
- iTnews: Kernel-crashing NoisyTCP bug sat in Linux for 11 years
- GitHub: V4bel/NoisyTCP
Start your morning with the signal that matters.
Get the biggest cybersecurity developments, why they matter, and where to go deeper on CyberExperts.
By subscribing you agree to our Privacy Policy.
Free. Weekdays. Built for operators.