Practical Linux guidance for safer servers and self-hosted systems.Linux • Security • Self-hosting • Practical tools
Linux Server Hardening

NoisyTCP (CVE-2026-99288): What Linux Operators Should Verify

NoisyTCP (CVE-2026-99288) is a Linux kernel TCP race that can cause a remote kernel crash under the conditions described in the public disclosure. The key operator boundary is narrower than many headlines: a demonstrated denial-of-service condition is not the same as confirmed remote code execution, active exploitation, or proof that a particular host is compromised.

This source-backed explainer does not prescribe a generic patch, reboot, sysctl, firewall, or diagnostic procedure. Vendor package status is release-specific and time-sensitive; use the advisory for the exact distribution and release under review. For broader context, see the Linux server security pillar.

What is NoisyTCP (CVE-2026-99288)?

The public disclosure describes a race in Linux kernel TCP established-socket lookup involving lockless access and reusable socket memory. Under the reported conditions, a remote sender can trigger a kernel panic, making availability the established practical impact in the sources reviewed for this draft. Researcher disclosure

The same disclosure describes a harder use-after-free path. That is important technical context, but it should not be rewritten as reliable remote code execution: AlmaLinux’s summary says the public assessment is a host crash and that no code execution has been established. AlmaLinux

Why is a kernel version number not enough?

A raw upstream or running-kernel version does not, by itself, answer whether a distribution package contains a backported fix. Distributions maintain their own package branches and publish release-specific vulnerability states, so the relevant vendor tracker is more authoritative than comparing a number copied from an upstream release note.

The Debian tracker illustrates the distinction by listing vulnerable and fixed states for named Debian releases and a fixed version for unstable. That is evidence about Debian’s package mapping—not a universal fixed-version rule for Ubuntu, RHEL-compatible systems, Alpine, or another distribution. Debian Security Tracker

Keep these four facts separate:

  1. Upstream history: when the kernel correction landed and which branches contain it.
  2. Vendor package status: whether a named distribution release marks its package fixed, affected, pending, or not affected.
  3. Installed package state: what package is present on a particular host.
  4. Running-kernel state: what the host has actually booted.

This page intentionally does not turn those distinctions into commands or maintenance instructions. The operational procedure belongs in a separately evidenced Track B article, scoped to a stated environment.

What do the current vendor records establish?

The current records establish vendor-specific boundaries, not one cross-distribution answer. Debian’s tracker provides release-specific package status. Red Hat’s CVE page describes its affected product scope and publishes a preliminary Important assessment. AlmaLinux describes its own backport and testing status, which is not automatically the same as a production release in every downstream environment.

Red Hat’s page reports a preliminary CVSS 3.1 score of 7.5 and a vector that includes high confidentiality and integrity impact. Its narrative describes an unauthenticated remote operating-system crash and says the practical effect is availability disruption. Those are both published parts of the Red Hat record; they should be attributed rather than silently reconciled into a new score or claim. Red Hat

A defensible status table therefore looks like this:

Question What the cited record can establish What it cannot establish
Did upstream fix the code? The disclosure’s stated fixed commit and release history That every vendor package contains the correction
Is Debian fixed? The Debian tracker’s state for a named Debian release Status for a different distribution or an individual running host
Is a Red Hat-family product affected? Red Hat’s stated product scope and preliminary assessment That every rebuild or downstream clone has identical status
Is an AlmaLinux package ready? AlmaLinux’s stated backport/testing boundary at the checked time That testing status equals production availability everywhere

All rows are snapshots checked on 2026-11-22. Vendor records can change.

Does this prove compromise or active exploitation?

No. A vulnerability record, a severity rating, and a demonstrated crash establish risk and urgency for source-led review; they do not prove that a particular server was breached or that attackers are exploiting the issue in the wild.

The public disclosure supports a remote availability concern. AlmaLinux says it was not aware of attacks in the wild at the time of its notice. That statement is time-bound, not a permanent guarantee. Likewise, the absence of a demonstrated code-execution path in the reviewed sources is not proof that no future research can change the assessment.

The most useful interpretation is a four-part boundary:

For related context, compare this with the Linux kernel TCP trust-boundary explainer, which covers a different CVE and should not be treated as the same incident.

What should a Linux operator verify next?

The defensible next step is source-led: identify the exact distribution and release, then read that vendor’s current record for CVE-2026-99288. Do not substitute an upstream version, another distribution’s fixed package, or a headline for release-specific status.

Keep operational choices outside this explainer. Package changes, reboots, mitigations, firewall decisions, exposure checks, and recovery planning require environment-specific prerequisites and evidence; generic instructions would move this page into Track B and are deliberately excluded.

Bottom line

NoisyTCP deserves prompt attention because the public record describes a remotely triggerable Linux kernel availability failure. The accurate conclusion is bounded: separate the demonstrated crash from unproven code execution, consult the exact vendor and release record, and do not confuse upstream history, a severity score, or a secondary headline with proof that every Linux host is affected or compromised.

Sources