Security alerts · November 22, 2026

Linux kernel: two TCP networking flaws rated CVSS 9.8, patches are available

The Linux kernel has received fixes for two flaws in how it handles TCP network connections (CVE-2026-99676 and CVE-2026-99634), both rated CVSS 9.8. No exploitation is known so far. Linux server admins should install the updated kernel from their distribution and reboot.

What happened

Linux kernel developers have released fixes for two bugs in the kernel's networking code, the part that handles TCP connections. They were published in NVD on November 20, 2026 as CVE-2026-99676 and CVE-2026-99634. Both carry a severity rating of CVSS 9.8 out of 10.

The first bug (CVE-2026-99676) concerns the order in which the processor carries out operations when the kernel counts references to network connections (sockets). An older kernel change accidentally removed the guarantee of correct ordering on certain processor types. The second bug (CVE-2026-99634) is in the MPTCP (Multipath TCP) support when combined with the TCP Fast Open feature. Here the kernel can hand an application a connection that has already been torn down and whose memory has been freed.

Neither bug is listed in the CISA KEV catalog, and the EU vulnerability database EUVD, run by ENISA, does not report any exploitation. Nothing is known about real-world attacks so far.

Why it is dangerous

Both bugs involve incorrect memory handling. CVE-2026-99634 is a use-after-free, meaning memory is used after it has been released. Think of a cloakroom: you hand back your ticket, the attendant gives the hook to someone else, but you still believe your coat is hanging there. When a program works with memory that now belongs to something else, it can crash. In the worse case, an attacker may be able to influence what ends up in that memory.

CVE-2026-99676 is a memory ordering bug. On processors with weak memory ordering, such as ARM, the kernel may touch a connection's data before it has properly “reserved” it. For a split second it can end up working with data that is not yet, or no longer, valid.

A CVSS 3.1 score of 9.8 is close to the maximum. The vector says an attack would work over the network, without logging in and without any user interaction, and could seriously affect the confidentiality, integrity and availability of the system. The score in the NVD record was not assigned by NVD itself but by another organization named in the record. The bug descriptions do not explain how they could be exploited or mention any working attack. They also say nothing about code execution or a server takeover.

EPSS estimates the probability that someone will start exploiting a bug within the next 30 days. As of November 22, 2026, it is 0.17% for both bugs, which is low. The CISA KEV catalog lists vulnerabilities whose exploitation the US agency CISA has confirmed. These two are not on it.

Who is affected

This affects systems running a Linux kernel that does not yet include these fixes. The NVD record does not list exact version ranges, it only links to fixes in the stable kernel branches on kernel.org. To find out which versions are affected, check the security advisories of your distribution, such as Debian, Ubuntu, Red Hat or SUSE.

  • CVE-2026-99676: according to its description, it shows up on architectures with weak memory ordering. That mainly means servers and devices with ARM processors, such as ARM cloud instances, network appliances or Raspberry Pi. Classic x86 servers have stricter memory ordering, but the developers do not explicitly assess this in the description.
  • CVE-2026-99634: affects systems that use MPTCP together with TCP Fast Open when accepting connections.

Typical environments are web and application servers, VPS and cloud servers, container and virtualization hosts, as well as Linux-based routers, firewalls and NAS devices. Windows and macOS are not affected. Android and many embedded devices also run the Linux kernel, but their fixes have to come from the device maker.

How to check if you are affected

All of the following commands only read information and change nothing. First, find out the kernel version and processor architecture:

uname -r
uname -m

Output of aarch64 or arm… means ARM, x86_64 means a classic Intel or AMD server. To list installed kernel packages, use the first command on Debian and Ubuntu, the second on Red Hat, Rocky and Alma:

dpkg -l 'linux-image*' | grep ^ii
rpm -q kernel

These commands show whether the system uses MPTCP and TCP Fast Open. A value of net.mptcp.enabled = 1 means MPTCP is enabled. The ss -Ma command lists MPTCP sockets, including those waiting for incoming connections. For net.ipv4.tcp_fastopen, a value of 2 or 3 means Fast Open is also enabled for server applications.

sysctl net.mptcp.enabled net.ipv4.tcp_fastopen
ss -Ma

Compare the kernel version you found with your distribution's security advisory for CVE-2026-99676 and CVE-2026-99634.

How to fix it

Kernel developers have already released fixes to the stable branches on kernel.org. The NVD record does not list exact fixed version numbers. Most companies do not build their own kernels, so wait for an updated package from your distribution or device vendor. The developers do not mention any temporary workaround. The bugs are not in the CISA KEV catalog, so CISA has not issued any required action for them.

  • Check your distribution's security advisories to see whether a fixed kernel for CVE-2026-99676 and CVE-2026-99634 is out.
  • Install the kernel update the usual way, for example apt update && apt upgrade or dnf update kernel.
  • Reboot the server. The new kernel only takes effect after a reboot. If you use live patching (such as Livepatch or kpatch), check whether it covers these two bugs.
  • After the reboot, run uname -r to confirm the new kernel is running.

Work in this order: internet-facing servers first, especially those with ARM processors and those running MPTCP with TCP Fast Open. Then internal servers, and finally other devices. For routers, firewalls and NAS devices, watch for firmware updates from the vendor.

If you do not need MPTCP, it makes sense to keep it turned off. This is not an official recommendation from the developers for these bugs, though, and it does not replace the update.

What if you may already be compromised

No indicators of compromise have been published for these bugs, and no exploitation is known. With kernel memory bugs, an attack attempt may show up as a crash or a kernel error message. It is worth checking the following:

  • Kernel messages containing Oops, BUG or general protection fault. The dmesg and journalctl -k commands only show the current boot, so use the commands below.
  • Unexplained reboots or server crashes in recent weeks.
  • New or unknown user accounts, unfamiliar SSH keys in authorized_keys files, and unknown scheduled jobs in crontab or systemd timers.
  • Unknown processes and network connections, listed for example by ss -tunap.

The first command searches kernel messages from all boots stored in the systemd journal. Older boots are only there if the journal is persistent. The second command also searches older and compressed logs: kern.log files on Debian and Ubuntu, messages files on Red Hat systems.

journalctl _TRANSPORT=kernel --no-pager | grep -iE 'oops|bug:|general protection'
zgrep -iE 'oops|bug:|general protection' /var/log/kern.log* /var/log/messages* 2>/dev/null

If you find anything suspicious, disconnect the server from the network but do not power it off, so evidence is preserved. Bring in an incident response specialist. If you suspect the kernel itself was compromised, the safest option is to reinstall the system from a trusted source.

Sources

Articles are put together by an AI agent from public sources (vendors, CISA, ENISA, NVD, SK-CERT, NÚKIB), and version numbers and CVEs are checked against them automatically. Always confirm the steps in the vendor advisory before touching production.

← All alerts · RSS

Need help patching or checking your systems?

I can help you find out which systems are affected, update them safely and check whether an attacker already got in. Let us start with an initial consultation, which is deducted from the project price if we work together.