Skip to content

NoisyTCP flaw lets attackers remotely crash Linux servers

CVE-2026-99288, an 11-year-old NULL-pointer race in the Linux kernel's TCP stack, lets a remote attacker panic multi-queue servers with crafted packets. Fix is in mainline.

Published 3 min read

A newly patched bug in the Linux kernel's TCP stack, tracked as CVE-2026-99288 and named NoisyTCP, lets a remote, unauthenticated attacker crash a server with specially crafted TCP packets. The flaw is roughly 11 years old and, per BleepingComputer, reaches almost every Linux distribution. There is no code execution and no data tampering — the impact is a kernel panic and a downed host. The fix is already in mainline; the job now is pulling it through your distro.

What the bug does

NoisyTCP is a NULL-pointer dereference triggered by a race condition during the kernel's lockless socket lookups. Recycled socket memory can be reached before it is fully initialized and hashed. An attacker exploits the timing window between SYN-cookie handling and socket hashing with crafted TCP packets, dereferencing a pointer that isn't there yet and taking the kernel down.

It is not universal. Exploitation requires a server with two or more NIC receive (RX) queues, or one using Receive Packet Steering (RPS) to spread packets across CPUs — the common configuration on busy multi-core hosts, which is exactly where it hurts. The researcher who reported it, Hyunwoo Kim (handle "v4bel"), published a write-up and reproduction details at github.com/V4bel/NoisyTCP.

Severity and status

Red Hat rates the flaw Important with a preliminary CVSS of 7.5; NVD and cve.org list the score as N/A at publication. No in-the-wild exploitation has been reported. Treat it as a reliable remote reboot primitive against exposed Linux servers — an availability problem, not a breach, but a real one for anything load-bearing and internet-facing.

The fix landed in the mainline kernel about two weeks ago as commit 4e2b3ef. Distribution backports are what you actually install — track the Debian tracker and your vendor's advisory for the fixed package version.

Action checklist

  1. Patch the kernel. Update to a distribution kernel that carries the backport of commit 4e2b3ef. Confirm the fixed version in your vendor's CVE page before declaring a host clean — the mainline merge does not mean your distro has shipped it yet.
  2. If you can't patch immediately, mitigate. Red Hat's stopgaps: disable SYN cookies (sysctl -w net.ipv4.tcp_syncookies=0) and/or collapse the NIC to a single RX queue with RPS/RFS disabled. Both carry costs — turning off SYN cookies weakens SYN-flood protection, and a single RX queue caps throughput — so treat them as temporary.
  3. Prioritize internet-facing, multi-queue hosts. Load balancers, reverse proxies, edge servers and anything terminating TCP from untrusted networks are the reachable targets.
  4. Watch for unexplained panics. A spike in kernel panics on exposed hosts, with no obvious load cause, is the symptom.

Context

NoisyTCP is the latest in a steady run of remotely triggerable DoS bugs in the kernel's network path, where lockless fast-path optimizations and socket-lifetime races keep producing reachable NULL dereferences. The saving grace here is the narrow impact: no memory disclosure, no RCE, just a crash. The catch is the breadth — an 11-year-old bug in default-ish multi-queue configurations means the exposed population is large, and a single crafted-packet script is enough to knock a server over until it reboots.

Related stories