How to protect your Linux server from CVE-2026-98432 (Linux kernel IPv4 TCP use-after-free)

What is CVE-2026-98432?

CVE-2026-98432 is a use-after-free vulnerability in the Linux kernel's IPv4 TCP code, present since Linux 4.4. According to Red Hat's bulletin RHSB-2026-016, which rates it critical, a remote, unauthenticated attacker could use it to achieve remote code execution (RCE). The CVE was published on October 15, 2026, and the bulletin on October 30.

It is a race condition in the TCP handshake: the attacker sends a SYN carrying the IPv4 Record Route or Timestamp options, which can be freed while another CPU is still using them. The Linux kernel CVE team and Red Hat assign it a CVSS 3.1 score of 9.8. According to Red Hat, exploitation requires precise timing but no privileges or user interaction, and the most certain impact is memory corruption or a system crash (denial of service, DoS).

Exploitation status as of November 12, 2026: Anthropic, which reported the flaw, says it has an exploit that runs commands as root remotely on servers with a Multipath TCP (MPTCP) listening socket, and that a failed attempt does not crash the system, so it can be retried. According to Ubuntu and CloudLinux, that exploit has not been published; CloudLinux could not reproduce the claim and says no exploitation in the wild has been reported. The CVE is not in the CISA KEV catalog. Unvetted repositories claiming to contain a PoC have appeared: do not download or run them.

Are you affected?

Very likely, yes. According to Red Hat, any system running a vulnerable kernel with IPv4 TCP networking enabled is affected: almost any Linux server with kernel 4.4 or later without the fix, including the distributions in Option A. MPTCP does not need to be enabled: RHEL and AlmaLinux ship with it disabled and are still vulnerable. Debian and Ubuntu ship with it enabled by default. The main flaw does not affect IPv6, but that does not protect you if the server also accepts TCP over IPv4.

Not affected: RHEL 7 and RHEL 6, according to Red Hat, and CloudLinux 7 (kernel 3.10), because the vulnerable code arrived in Linux 4.4.

With cPanel/WHM or DirectAdmin, the fix comes with your distribution's kernel; we found no advisory from cPanel or DirectAdmin themselves.

How to check your server

What counts is the running kernel: a fixed kernel that has not been booted yet does not protect you. Compare the output with the list in Option A.

RHEL and derivatives (AlmaLinux, Rocky Linux), CloudLinux:

uname -r
rpm -q kernel

Only compare within the same minor release (suffix el10_2, el10_3…): a RHEL 10.3 kernel older than 6.12.0-264.14.1.el10_3 is not fixed, even though its number is higher than the 10.2 one.

Debian and Ubuntu:

uname -r
uname -v
dpkg -l 'linux-image*' | grep ^ii

uname -r only shows the ABI, not the package version. On Debian, uname -v does show it. On Ubuntu, find the package for the running ABI in dpkg -l (for example, 5.15.0-203.213 for 5.15.0-203): dpkg -l lists what is installed, not what is running.

With KernelCare:

kcarectl --patch-info | grep CVE-2026-98432

A patched server prints lines naming the CVE; empty output means not patched or not covered (kcarectl --info does not list CVEs).

Finding no MPTCP listening sockets does not mean you are safe, Red Hat warns.

How to mitigate

Option Best for Durable?
A: fixed kernel Everyone; the only complete fix (requires a reboot) Yes
B: KernelCare livepatch KernelCare servers that cannot reboot yet Until you install the fixed kernel
C: nftables filter Reducing exposure until the reboot No: temporary and partial
D: disable MPTCP Hosts with MPTCP enabled (Debian, Ubuntu, Go services) No: partial

Option A: install the fixed kernel

This is the only complete fix. Fixed versions:

  • RHEL 8, AlmaLinux 8 and Rocky Linux 8: kernel-4.18.0-553.182.1.el8_10 (RHSA-2026:93414).
  • RHEL 9, AlmaLinux 9, Rocky Linux 9 and CloudLinux 9: kernel-5.14.0-687.64.1.el9_8 (RHSA-2026:93412).
  • RHEL 10.2, AlmaLinux 10 and Rocky Linux 10: kernel-6.12.0-211.72.1.el10_2 (RHSA-2026:93413).
  • RHEL 10.3: kernel-6.12.0-264.14.1.el10_3 (RHSA-2026:98608).
  • CloudLinux 10: kernel-6.12.0-211.73.1.el10_2.
  • CloudLinux 7 Hybrid and 8: 4.18.0-553.180.1.lve.1, in staged rollout as of November 3, 2026.
  • Debian 13: linux 6.12.113-1 (DSA-6568-1).
  • Debian 12: linux 6.1.190-1 (DLA-4841-1); with linux-6.12, 6.12.113-1~deb12u1.
  • Ubuntu 22.04: linux 5.15.0-203.213 (USN-9027-1).
  • Ubuntu 24.04: linux 6.8.0-149.149 (USN-9022-1).
  • Ubuntu 26.04: linux 7.0.0-43.43 (USN-9018-1).
  • Proxmox VE with the 7.0 kernel: proxmox-kernel-7.0.14-25-pve (PSA-2026-00080-1, which does not mention other series).
  • Upstream kernels: 5.10.272, 5.15.223, 6.1.190, 6.6.159, 6.12.113, 6.18.56, 7.2.10 and 7.3.

On Debian these are minimum versions; the security repository may carry a newer one. For other variants (such as Ubuntu HWE kernels, Ubuntu 20.04 or CloudLinux LTS), check your distribution's security tracker (links under Sources).

Debian and Ubuntu. This command upgrades every package, which may restart services:

sudo apt update && sudo apt upgrade

On Debian, use apt: apt-get upgrade can keep linux-image-amd64 back without installing the new kernel. On Ubuntu, if you removed the standard kernel metapackages (such as linux-generic), you must recompile third-party modules yourself.

RHEL, AlmaLinux and Rocky Linux. This command also upgrades every package:

sudo dnf clean metadata && sudo dnf upgrade

CloudLinux 9 and 10: as root, dnf update 'kernel*'. CloudLinux 7 Hybrid and 8: yum update kernel once the fixed kernel reaches your repositories; until then, use Option B. Proxmox VE: install the listed kernel or later as usual.

The new kernel is only used after a reboot. This command reboots the server; schedule it in a maintenance window:

sudo reboot

Option B: KernelCare livepatch

With KernelCare, apply the livepatch without a reboot and confirm it covers this CVE:

sudo kcarectl --update
kcarectl --patch-info | grep CVE-2026-98432

According to CloudLinux and TuxCare, as of November 12, 2026 a livepatch is available for CloudLinux 7 Hybrid, 8, 9 and 10, Debian 12, RHEL 9 and Ubuntu 24.04.

Option C: Red Hat nftables filter (temporary and partial)

This Red Hat rule drops incoming IPv4 TCP packets carrying IP options. It reduces exposure but does not eliminate it. Save it as cve-2026-98432.nft:

table inet cve_2026_98432 {
  chain input {
    type filter hook input priority -310; policy accept;
    ip protocol tcp ip hdrlength > 5 counter drop
  }
}

These commands check the syntax, load the rule (it affects incoming traffic immediately) and show its drop counter:

sudo nft --check --file cve-2026-98432.nft
sudo nft --file cve-2026-98432.nft
sudo nft list table inet cve_2026_98432
  • It breaks TCP connections that depend on IPv4 options; AlmaLinux advises against it if your network relies on CIPSO labels (NetLabel).
  • It only covers traffic addressed to the host itself, not forwarded traffic or traffic between containers on the same host.
  • It is lost on reboot. On Debian, reloading nftables.service also removes it (/etc/nftables.conf starts with flush ruleset).
  • Do not copy the persistence instructions in Red Hat's bulletin: as of November 12, 2026 they contained errors.

Option D: disable MPTCP (partial)

Anthropic's exploit needs an MPTCP listening socket. Disabling MPTCP only blocks that path: according to Red Hat, the system remains affected. On RHEL and derivatives it is already disabled. Check the current state and which services use MPTCP:

sysctl net.mptcp.enabled net.ipv4.tcp_migrate_req
sudo ss -lnMp

With net.mptcp.enabled = 1, applications can open MPTCP sockets. Leave net.ipv4.tcp_migrate_req at 0 (the default): 1 opens a second path to the flaw. Neither value tells you whether you are vulnerable, and an error because the key does not exist does not mean you are safe.

These commands disable MPTCP persistently and, immediately, for new sockets. They can break software that uses MPTCP; services already listening keep their sockets until you restart them:

echo 'net.mptcp.enabled = 0' | sudo tee /etc/sysctl.d/90-disable-mptcp.conf
sudo sysctl -w net.mptcp.enabled=0

Programs built with Go 1.24 or later enable MPTCP by default. GODEBUG=multipathtcp=0 disables it for a specific program (restart the service with it set), but the Go team calls this a fragile mitigation.

After mitigating

  1. After rebooting, confirm the running kernel with uname -r (plus uname -v on Debian; on Ubuntu, compare the ABI with dpkg -l).
  2. If you applied the livepatch, still schedule the installation of the fixed kernel.
  3. Once the fixed kernel is running, remove the temporary filter:
    sudo nft delete table inet cve_2026_98432
  4. If you disabled MPTCP and need it, delete /etc/sysctl.d/90-disable-mptcp.conf.

Need help?

Open a ticket at soporte.telecu.cloud and reference: CVE-2026-98432.

Sources

Tags
How to protect your PHP servers from PHP2Shell (CVE-2026-109618 and CVE-2026-109619)
How to protect your nginx server from CVE-2026-93479 and CVE-2026-15176 (heap overflows)