NoisyTCP Flaw in Linux Lets Attackers Remotely Crash Servers
The new CVE-2026-99288 flaw in Linux’s TCP subsystem can cause a kernel crash and take down an entire server under certain network conditions. A fix is already available.

The NoisyTCP flaw, tracked as CVE-2026-99288, affects the Linux kernel’s TCP subsystem and, under certain conditions, allows an unauthenticated attacker to remotely cause a system crash. Researcher Hyunwoo Kim disclosed the flaw after the embargo with Linux distributions ended. The fix was incorporated into the kernel on November 3, 2026, and administrators should verify that an updated package is available for their distribution.
The practically confirmed impact is denial of service: an attack can trigger a kernel panic, meaning a kernel crash. There is no confirmed server takeover, remote code execution, or data manipulation.
How the NoisyTCP Flaw Occurs
The vulnerability is a race condition in so-called lockless TCP socket lookup. Due to improper concurrency, the kernel may dereference a NULL pointer, causing a system crash. According to the author, the vulnerable code entered the kernel on October 3, 2015.
The attack does not require authentication or a local account on the server. However, a TCP listener—a service accepting TCP connections—must be available, and the system must process incoming packets concurrently in multiple RX contexts. According to the announcement, this means at least two network-interface receive queues or the RPS mechanism, which distributes packets among CPU cores.
Kim said he successfully tested the attack against default-configured Nginx, Apache, and SSH services. If the crash affects the host system’s kernel, containers running on that host will also stop working. The impact therefore may extend beyond the service exposing the TCP listener to the entire server.
Fix and Recommended Steps
A fix for the NoisyTCP flaw is already available. Debian, for example, lists fixed packages for the bookworm-security and trixie-security branches. However, the status should be checked in the specific distribution’s security advisories: kernel versions and backported fixes differ between distributions, so the kernel’s main version number alone may not be enough to determine vulnerability.
- Verify whether the distribution has released a fixed kernel package for the system’s branch.
- Install the kernel update.
- Reboot the server into the fixed kernel; installing the package alone does not fix the running kernel.
- Pay particular attention to Internet-facing servers with web services or SSH, as well as the configuration of network RX queues or RPS/RFS.
If an update is not immediately possible, the author describes two temporary mitigations. The first is disabling SYN cookies. However, this reduces the server’s protection against SYN flood attacks. The second option is limiting the interface to one RX queue and disabling RPS/RFS, which may reduce network performance. These measures should therefore be used only until a fixed kernel can be deployed.
What Has Not Yet Been Confirmed
There are no confirmed cases of CVE-2026-99288 being actively exploited in real-world attacks. The author also described a second, harder-to-trigger use-after-free path, but did not confirm that it enables remote code execution or access to data beyond the scope of a denial-of-service attack.
In the near term, it will be important to monitor fixed packages for additional distributions, including Ubuntu, RHEL, SUSE, and Oracle Linux, as well as any vendor or CISA announcements about exploitation. For administrators, however, the key point is already clear: an unpatched, Internet-facing Linux server with a suitable network configuration may be remotely taken offline through specially timed TCP communication.
Sources
- oss-security / Hyunwoo Kim – The author’s direct announcement describes the mechanism, triggering conditions, tested services, the flaw’s time range, and mitigations.
- Debian Security Tracker – Confirms the CVE, the technical description of the crash, and specific vulnerable and fixed Debian package versions.
- Red Hat CVE Database – Confirms remote DoS through a kernel crash, lists available mitigations, and assesses the impact on Red Hat products.
- BleepingComputer – Independently summarizes the disclosure, affected common server services, and the availability of the upstream fix.
Verified and updated: November 16, 2026 15:25



