We are tracking CVE-2026-99288, known as NoisyTCP, a flaw in the Linux kernel TCP connection lookup. A remote attacker with no account on the server can trigger a kernel panic, which takes down the whole server.
- No exploit code is public and there are no reports of exploitation in the wild. The issue was disclosed in a coordinated way. - The demonstrated impact is a crash, a denial of service. No code execution has been shown. - A server is exposed when it has a reachable TCP port, SYN cookies enabled, which is the kernel default, and a network interface with two or more receive queues or RPS enabled.
Affected versions
- CloudLinux OS 8, 9 and 10: affected. - CloudLinux OS 7 Hybrid: affected. - CloudLinux OS 7: not affected. The 3.10 kernel does not contain the vulnerable code.
Fix status
- CloudLinux kernels: a fixed kernel has not been released yet. - KernelCare: a livepatch has not been released yet.
We will post release information here as soon as it is available.
Temporary mitigation
Two workarounds are published by the researcher and by Red Hat. Each has a cost, so please test before applying it across production.
Option 1: Disable SYN cookies. This closes the demonstrated crash path, but the server loses its SYN flood protection while the setting is off.
Option 2: Use a single receive queue and disable RPS and RFS on the internet-facing interface. Replace IFACE with the interface name. This lowers network throughput on busy servers.
ethtool -L IFACE combined 1 for q in /sys/class/net/IFACE/queues/rx-*/rps_cpus; do echo 0 > "$q"; done sysctl -w net.core.rps_sock_flow_entries=0
How to check your server
sysctl net.ipv4.tcp_syncookies ethtool -l IFACE
SYN cookies are enabled if the first command prints 1 or 2. The server has several receive queues if the current Combined or RX value is above 1.
Once a KernelCare livepatch is released, you can confirm it is applied with: