NoisyTCP (CVE-2026-99288) is a Linux kernel TCP race condition that can be triggered remotely before authentication against ordinary TCP listeners. It requires packet processing in concurrent receive contexts, available on systems with multiple RX queues or with Receive Packet Steering enabled. By exploiting differences in how IPv4 fragments and regular TCP packets are distributed, an attacker can create a stale lookup while a request socket is freed and reused. The report describes two race paths: reuse of a request socket can make the kernel dereference a missing listener and panic, causing denial of service; a more complex race involving a full child socket can produce a use-after-free. The issue does not require special server configuration and can affect services such as web, SSH, and proxy servers.
The vulnerability was introduced in 2015 and fixed by Linux kernel commit 4e2b3ef19e0e on November 3, 2026, after roughly 11 years in the code. Its narrow timing window makes exploitation environment-dependent: tests triggered it within seconds to minutes on virtual machines, but took about 14 hours on a quiet bare-metal server despite tuning. The demonstrated practical impact is remote denial of service, which can bring down a host even when the targeted application runs in a container. The report does not establish a general-purpose remote code execution exploit. Applying the fix is the priority; temporary mitigations are disabling TCP SYN cookies, which reduces SYN-flood protection, or limiting incoming traffic to one RX queue and disabling RPS/RFS. Separate network namespaces require the SYN-cookie setting to be changed individually.
Summary generated by AI from the linked article. hn.today is not affiliated with Hacker News or Y Combinator.