CVE-2026-99288 NoisyTCP Flaw Can Crash Linux Servers

An 11-year-old flaw in the Linux kernel can let an attacker crash a server without logging in, according to research published on 16 November 2026. The vulnerability, dubbed NoisyTCP and tracked as CVE-2026-99288, affects almost all Linux distributions and puts systems running Nginx, Apache and SSH servers with default settings at risk.
The flaw was found by security researcher Hyunwoo Kim, who also wrote the patch. He released his advisory on GitHub after an embargo agreed with Linux distributions came to an end. The bug sits in the kernel's TCP networking code, which handles almost all network traffic on Linux.
How NoisyTCP crashes a server
CVE-2026-99288 is a NULL pointer dereference caused by a race condition during lockless socket lookups. When a server handles SYN cookies and socket hashing at the same time, recycled socket memory can be reached before it has been properly initialised and hashed. A remote attacker can drive that race by sending crafted TCP packets, and the kernel then reads a NULL listener and panics, taking the whole system down.
The attack needs a server with two or more receive queues, or one with Receive Packet Steering turned on, so that packet processing is spread across more than one CPU. The attacker sends IPv4 fragments and ordinary TCP packets that land on different CPUs, creating the timing window the flaw needs. A server does not need unusual settings, firewall rules or special sockets: it only has to call listen() and receive packets from the attacker.
Timing matters. In laboratory tests on virtual machines running QEMU/KVM, a crash took between about five seconds and two minutes. On a quiet bare-metal server with eight receive queues, the tuned proof of concept needed about 14 hours. Cloud and virtualised servers are therefore the most exposed, and a crash in a containerised web server takes down the whole host, not just the application.
Patch available, but no code execution so far
The flaw does not let an attacker alter data or run arbitrary code, at least for now. Kim has built only a limited, lab-specific exploit for virtual-machine servers and is still researching a general one. A reliable crash is still serious for any internet-facing service, because it is a remote denial of service that needs no login.
The bug was introduced in October 2015 and the fix landed in the mainline Linux kernel on 3 November 2026, so updated kernels are already available from vendors. Administrators should install their vendor's updated kernel as soon as they can and reboot. Red Hat's security team described the same race in its own advisory, noting that a stale hash lookup and the freeing and reallocation of a request socket must proceed at the same time in different receive contexts.
If patching has to wait, Kim suggests two stopgaps. The first is to turn off SYN cookies, although that weakens protection against SYN flood attacks. The second is to cut the network interface down to a single receive queue and disable Receive Packet Steering and Receive Flow Steering, which stops the race. Kim disclosed two other Linux kernel flaws earlier this year: Dirty Frag, a local privilege escalation flaw reported in May, and Januscape, a virtual-machine escape reported in July. Readers can follow the researcher's advisory and the full technical reports through the source links below.
Source: BleepingComputer, with further detail from Cyber Security News.