Researcher Hyunwoo Kim disclosed the NoisyTCP vulnerability (CVE-2026-99288) in the Linux kernel networking subsystem, which allows a remote unauthenticated attacker to trigger a kernel panic and completely bring down the host. According to the researcher, the vulnerable code has been present in the kernel for about 11 years — since October 2015. The fix was merged into the mainline kernel on 3 November 2026. For servers running container workloads, the threat is particularly serious: a kernel crash affects not only the targeted service but the entire host with all other containers. Administrators of Linux servers with externally accessible TCP services (Nginx, Apache, SSH) are advised to prioritize installing a kernel update.
Technical nature of the vulnerability
According to the researcher’s report, the issue falls under CWE-362 — a race condition when accessing a shared resource concurrently. The vulnerability arises during the lookup of TCP sockets in the hash table of established connections (ehash). While one packet handler is working with a found request_sock object, another handler can free this object and create a new one at the same memory address. The kernel continues working with a now different object that is not associated with the listening socket, which leads to a null pointer dereference and a kernel panic.
The GitHub Advisory Database assigns the vulnerability a CVSS 5.9 (Moderate) rating with the vector AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H — that is, the attack is network-based, requires no privileges or user interaction, but has high exploitation complexity and affects only availability.
Exploitation conditions
Triggering the race condition requires parallel packet processing, which, according to the researcher, is achieved due to differences in how IPv4 fragments and regular TCP packets are distributed. The vulnerability is exploitable on servers where at least one of the following conditions holds:
- Two or more receive (RX) queues are used on the network interface
- The RPS (Receive Packet Steering) mechanism is enabled, distributing packet processing across multiple processors
No special application configuration is required — any externally accessible TCP listener is sufficient. According to the researcher, servers running Nginx, Apache and SSH with default settings are at risk. However, it should be noted that the claim that “almost all Linux distributions” are affected comes only from the researcher and has not been confirmed by independent observers. Actual vulnerability depends on the configuration of the specific host.
Demonstration and exploitation limits
The researcher reports two exploitation scenarios. The main one is remote denial of service via kernel panic. The second scenario leads to a use-after-free situation; however, only a limited exploit has been created for it, working exclusively in laboratory conditions on virtual machines. Universal remote code execution has not been demonstrated at this time.
According to the researcher’s (not independently reproduced) lab tests, the time to trigger varies significantly: from 5 seconds to 2 minutes on QEMU/KVM virtual machines and about 14 hours on a physical server with eight receive queues — even after fine-tuning the exploit timings for the specific system. These figures reflect lab conditions and may differ substantially in real environments.
The vulnerability is not listed in the CISA KEV catalog, and the analyzed sources contain no evidence of active exploitation in real-world attacks.
Impact assessment
The highest risk is to servers that meet both of the following criteria: direct exposure of TCP services to the internet and parallel processing of incoming packets (multiple RX queues or enabled RPS). The risk is amplified in containerized environments: because the vulnerability is in the kernel, a kernel panic caused by an attack on a service inside a container results in the shutdown of the entire host and all workloads running on it. We have previously analyzed Linux kernel vulnerabilities capable of affecting a broad range of infrastructure — NoisyTCP continues this trend, albeit with lower criticality on the CVSS scale.
Patch and temporary measures
The fix (commit 4e2b3ef19e0e) was merged into the Linux kernel mainline on 3 November 2026. According to the Debian tracker, the fix status by release is as follows:
- bookworm: version 6.1.176-1 is vulnerable, fixed in 6.1.192-2
- trixie: version 6.12.107-1 is vulnerable, fixed in 6.12.117-1
- forky: fixed in 7.3.4-1
- sid: fixed in 7.3.5-1
If promptly updating the kernel is not possible, the researcher suggests two temporary workarounds, each with side effects:
- Disable SYN cookies — removes the condition for the race, but weakens protection against SYN flood attacks
- Limit processing of incoming packets to a single RX queue and disable RPS/RFS — blocks parallel processing, but can significantly reduce network performance
Patch priority is highest for servers exposed to the internet and running container workloads: in this scenario, a kernel panic causes maximum damage, while temporary mitigations involve noticeable trade-offs in security or performance. Administrators of such systems are recommended to update the kernel to a fixed version without waiting for the regular update cycle and to check the RX queue and RPS configuration on interfaces that receive external traffic.












