Tag: CVE-2026-99288

  • NoisyTCP (CVE-2026-99288) Can Crash Linux Servers Without a Login (2026)

    NoisyTCP (CVE-2026-99288) Can Crash Linux Servers Without a Login (2026)

    Takeaway: On November 16, 2026, researcher Hyunwoo Kim publicly disclosed NoisyTCP (CVE-2026-99288), a Linux kernel TCP flaw that can crash a host over the network without a login. Red Hat’s advisory rates the issue Important and says an unauthenticated remote attacker can trigger a kernel panic and denial of service. It also states that the flaw does not allow data alteration or arbitrary code execution. Debian has already shipped updates. Ubuntu’s CVE page, last updated November 11 and unchanged at a November 17 noon recheck, still listed current releases as “Needs evaluation.” AlmaLinux said on November 16 that versions 8, 9, and 10 are affected and that patched kernels are in its testing repository, not yet as a production update. The durable fix is a patched kernel. Until you can install one, Red Hat and the public advisory document temporary workarounds, with tradeoffs.

    This is a systems-administration story, not a curiosity about kernel internals. Any Linux machine that accepts TCP from a network it does not fully control — a public Nginx or Apache site, an SSH bastion, a cloud VM with a default network setup — is in the class of systems the advisories describe. If the kernel panics, the whole host goes down, including other containers on that host.

    What official advisories say

    Red Hat describes a race in the kernel’s lockless TCP socket lookup path. Recycled socket memory can be used before it is properly initialized and hashed, leading to a NULL pointer dereference and a kernel crash. The company’s impact statement is direct: in typical Red Hat Enterprise Linux deployments, TCP networking is enabled, and a system with active network listeners can be knocked over. The same statement is careful about what the bug is not. It “does not allow data alteration or arbitrary code execution.”

    That written impact is the part operators should plan around today. Red Hat also publishes a preliminary CVSS 3.1 base score of 7.5, marked “subject to review,” with the vector CVSS:3.1/AV:N/AC:H/PR:L/UI:N/S:U/C:H/I:H/A:H. Those metrics are not a perfect match for the prose on the same page: “Privileges Required: Low” and high confidentiality and integrity impact sit next to a statement that the attacker can be unauthenticated and that the result is a crash, not data theft or code execution. Treat the 7.5 figure as a preliminary vendor score, and treat Red Hat’s written impact statement as the clearer description until the score is reviewed.

    The official CVE record, assigned by the Linux kernel CNA, was published November 10, 2026, and later updated on November 16. Kim’s oss-security post, dated November 16, 2026 at 01:00 +0900, is the named NoisyTCP public write-up. It says the bug was introduced on October 3, 2015, and that a fix landed on November 3, 2026, meaning it had been latent for about 11 years and was present in almost all Linux distributions. The public report followed a coordinated disclosure with Linux distribution maintainers. We are not reproducing the report’s trigger discussion here. If you need to assess exposure, use your distribution’s security tracker and install the kernel update it names.

    Which updates are available

    Patch status is uneven, and it is changing. Check the tracker for the image you actually run, rather than assuming a vendor has or has not shipped a fix because a neighboring distribution has.

    Debian is the clearest public case we could verify from official advisories on the morning of November 17, 2026:

    • DSA-6628-1, dated November 13, 2026, fixes CVE-2026-99288 for Debian 13 trixie in linux version 6.12.117-1, among a larger set of kernel issues.
    • DLA-4906-1, dated November 15, 2026, fixes it for Debian 12 bookworm in linux version 6.1.192-1.
    • Debian’s security tracker, reviewed the same morning, listed bookworm’s security archive at 6.1.192-2 as fixed, trixie security at 6.12.117-1 as fixed, sid at 7.3.4-1 as fixed, and forky at 7.2.15-1 as still vulnerable.

    If you run trixie or bookworm, the official advice is the usual one: upgrade the linux packages from the security archive and reboot into the new kernel. A package that is sitting in the archive but not yet running on the host does not close the issue. Forky users should watch the tracker; sid already had a fixed 7.3.4-1 package in the table we reviewed.

    Ubuntu’s CVE page is not as far along. The page’s “last updated” stamp was November 11, 2026, and it listed Ubuntu 26.10, 26.04 LTS, 24.04 LTS, 22.04 LTS, 20.04 LTS, and 18.04 LTS as “Needs evaluation” for the linux package. That date is before the public oss-security post. A noon recheck on November 17, 2026, found the same stamp and the same “Needs evaluation” status for those current releases. It does not prove Ubuntu will lack a fix, and it does not prove a later update has already shipped. It only means that, as of the page we reviewed, Ubuntu had not yet published a status beyond evaluation. Check that page again before you decide a Ubuntu host is patched or out of scope.

    AlmaLinux’s November 16, 2026 note says AlmaLinux 8, 9, and 10 are affected, and that patched kernels are available in the project’s testing repository, not as a production update. The versions it named were kernel-4.18.0-553.186.2.el8_10 (AlmaLinux 8), kernel-5.14.0-741.22.2.el9_9 (AlmaLinux 9), and kernel-6.12.0-264.16.2.el10_3 (AlmaLinux 10). The post said no fixed RHEL kernel was available at the time of writing, and that the project had backported the upstream fix ahead of a CentOS Stream / RHEL update. If you run AlmaLinux, follow that official note for the testing-repository instructions, confirm the running kernel after reboot, and treat the packages as testing until the project promotes them to production.

    Red Hat’s CVE page includes product-specific status, workarounds, and the preliminary score. We are not restating a full product matrix from a secondary source. If you run RHEL, a RHEL-derived distribution, or a layered Red Hat product, use the Red Hat CVE page and your errata feed as the authority for whether your exact product is affected and whether an erratum is out. The advisory also points to stable kernel commits for the fix; those links are useful for vendors and for people who build their own kernels, not a substitute for a distribution update.

    Temporary workarounds, and their costs

    A kernel update is the fix. If you cannot reboot yet, Red Hat’s advisory and the public disclosure list two temporary mitigations. Both have operational costs, and neither is a reason to skip the patch.

    The first is to turn off TCP SYN cookies in each affected network namespace:

    sudo sysctl -w net.ipv4.tcp_syncookies=0

    The advisories emphasize the namespace point. A setting on the host does not automatically cover a container that has its own network namespace. Apply it in each namespace that faces untrusted TCP, and make the change persist with your usual sysctl configuration if you need it to survive a reboot. The documented downside is real: disabling SYN cookies weakens protection against SYN floods. If you use this workaround, watch flood-related metrics until a patched kernel is running.

    The second workaround is to reduce receive-side concurrency on the interface that takes external traffic: a single receive queue, with Receive Packet Steering (RPS) and Receive Flow Steering (RFS) disabled. That is a performance tradeoff on busy hosts, especially machines that were tuned for multiple queues. Use it only as a stopgap, and return the interface to its normal tuning after the kernel update is in place.

    If you manage a fleet, the practical order is straightforward. Identify public TCP listeners, confirm the running kernel version against your vendor’s fixed package, patch and reboot where an update exists, and apply a documented workaround only on hosts that must stay up until a reboot window. Do not wait for a scanner that keys only off upstream version numbers. As Red Hat notes in its FAQ, backported fixes can leave a package looking “old” to a naive scanner even after the issue is addressed.

    How this fits with other recent security changes

    NoisyTCP is a kernel availability bug, not an application-layer issue in Nginx, Apache, or OpenSSH themselves. Patching the web server or SSH package will not close it. The same week’s security calendar already includes other changes that affect how teams expose services, including GitHub’s coming default block on pull_request_target. Those are separate problems. The common operational lesson is narrower: default-on network services are part of your attack surface even when you have not misconfigured the application.

    ItemWhat official sources say
    Name / CVENoisyTCP, CVE-2026-99288
    CVE recordPublished November 10, 2026, by the Linux kernel CNA; updated November 16, 2026
    Named public write-upNovember 16, 2026, oss-security post by Hyunwoo Kim (01:00 +0900)
    Impact (Red Hat statement)Unauthenticated remote kernel crash / denial of service; no data alteration or arbitrary code execution
    Red Hat ratingImportant; preliminary CVSS 3.1 score 7.5, marked subject to review
    Age (researcher)Introduced October 3, 2015; fix dated November 3, 2026; about 11 years
    Debian 13 trixieFixed in linux 6.12.117-1 (DSA-6628-1, November 13, 2026)
    Debian 12 bookwormFixed in linux 6.1.192-1 (DLA-4906-1, November 15, 2026)
    Debian forky / sid (tracker, Nov 17 review)forky 7.2.15-1 vulnerable; sid 7.3.4-1 fixed
    Ubuntu (page last updated Nov 11, 2026; rechecked Nov 17 noon)Current releases listed as “Needs evaluation”
    AlmaLinux 8/9/10 (official note, Nov 16, 2026)Affected; patched kernels in the testing repository: 4.18.0-553.186.2.el8_10, 5.14.0-741.22.2.el9_9, 6.12.0-264.16.2.el10_3; not yet a production update
    Durable fixInstall a patched kernel and reboot
    Temporary mitigationsDisable SYN cookies per network namespace, or use a single RX queue and disable RPS/RFS, per Red Hat / the public advisory

    FAQ

    What is NoisyTCP (CVE-2026-99288)?

    It is a Linux kernel TCP vulnerability. The official CVE record was published November 10, 2026; researcher Hyunwoo Kim’s named NoisyTCP write-up followed on November 16. Red Hat says a race in lockless socket lookup can lead to a NULL pointer dereference and a kernel panic, causing a denial of service.

    Can an attacker steal data or run code with this bug?

    Not according to Red Hat’s impact statement, which says the flaw does not allow data alteration or arbitrary code execution. Plan around a host crash. The preliminary CVSS vector on the same page lists high confidentiality and integrity impact, but Red Hat marks that score as subject to review and the written statement is the clearer guidance today.

    Does this affect Nginx, Apache, or SSH?

    The bug is in the Linux kernel’s TCP handling, not in those applications. Official write-ups mention common default TCP listeners such as Nginx, Apache, and SSH as examples of services that can be reached before any login. Updating only the application packages will not close the issue.

    Which Debian versions are patched?

    Debian 13 trixie is fixed in linux 6.12.117-1 (DSA-6628-1). Debian 12 bookworm is fixed in linux 6.1.192-1 (DLA-4906-1). Debian’s tracker also listed sid 7.3.4-1 as fixed and forky 7.2.15-1 as vulnerable when we reviewed it on November 17, 2026.

    Has Ubuntu released a fix?

    Ubuntu’s CVE page, last updated November 11, 2026, still listed supported releases as “Needs evaluation” when we rechecked it at noon on November 17. Check the official Ubuntu page for a later status before you conclude that a Ubuntu host is patched or unaffected.

    Has AlmaLinux released a fix?

    AlmaLinux’s November 16 note says versions 8, 9, and 10 are affected and publishes patched kernel versions in the testing repository. Those are not production updates. The same note said no fixed RHEL kernel was available at the time of writing. Follow the official AlmaLinux blog for current instructions.

    What should I do if I cannot reboot yet?

    Red Hat and the public advisory document two temporary workarounds: disable SYN cookies in every relevant network namespace, or reduce the external-facing interface to a single receive queue and disable RPS and RFS. Disabling SYN cookies weakens SYN-flood protection. These are stopgaps. Install the patched kernel and reboot when you can.

    If a web server runs in a container, is only that container at risk?

    No. This is a kernel issue. A panic takes down the host, not just one container. Workarounds that use sysctl also have to be applied in each container network namespace, because the SYN-cookie setting is per namespace.

    How long had this been in the kernel?

    The researcher’s public report says the issue was introduced on October 3, 2015, and that a fix was dated November 3, 2026, a span of about 11 years. That is why almost every mainstream distribution had to assess it.