How to protect your Linux server from DirtyAH6, TUNderflow, PPPoEject and DiagSpill (kernel local privilege escalation)

What are DirtyAH6, TUNderflow, PPPoEject and DiagSpill?

They are four local privilege escalation (LPE) vulnerabilities in the Linux kernel. Each has a public proof of concept (PoC) that gives an unprivileged local user root on the systems it was written for (CVSS per Red Hat):

  • DirtyAH6 (CVE-2026-80844): out-of-bounds access in IPv6 AH (ah6). CVSS 8.3.
  • TUNderflow (CVE-2026-81000): integer underflow in the TUN/TAP driver (tun). CVSS 7.8.
  • PPPoEject (CVE-2026-68121): use-after-free in PPPoE (pppoe). CVSS 7.3.
  • DiagSpill (CVE-2026-74469): out-of-bounds write in SCTP diagnostics (sctp_diag). CVSS 8.3 (8.8 per the kernel CNA, the authority that assigns its CVEs).

Red Hat rates all four "Important". The first three need unprivileged user namespaces or else CAP_NET_ADMIN (plus CAP_NET_RAW for DirtyAH6). DiagSpill needs neither: SCTP and sctp_diag being available is enough.

The details and PoCs were published on September 18, 2026, although the fixes and the first CVE records were already public weeks earlier.

They are local. The remote scenarios need uncommon IPsec or SCTP setups and cause a crash (denial of service, DoS); remote root was only achieved with DirtyAH6, in a lab.

Exploitation: as of November 12, 2026, none of the four is in the CISA Known Exploited Vulnerabilities (KEV) catalog and we found no report of active exploitation.

Are you affected?

Short answer: if your server runs a Debian, Ubuntu, RHEL, AlmaLinux or Rocky Linux kernel older than the versions in Option A, it has at least one of these flaws. With those versions or later, it does not.

An attacker first needs to run code on the server (a hosting or shell account, or a vulnerable web application). Priority is higher if you have users you do not control or run containers: the host kernel can be corrupted from a container (with the first three, if the container holds the required capabilities; with DiagSpill, without them).

According to Red Hat, on RHEL SCTP is not loaded by default and PPPoE is rarely configured, which reduces exposure to DiagSpill and PPPoEject.

No confirmed status as of November 12, 2026: CloudLinux (it ships its own kernel) and Proxmox VE. For RHEL 7, Red Hat lists the fix only under its Extended Lifecycle Support. cPanel and DirectAdmin use the operating system's kernel.

How to check your server

What counts is the running kernel, not the newest installed one. On Debian / Ubuntu, uname -r does not show the package version and cannot be compared; these commands show the version of the running kernel and of the installed ones:

dpkg-query -W linux-image-$(uname -r)
dpkg -l 'linux-image-*' | awk '/^ii/{print $2, $3}'

On RHEL, AlmaLinux, Rocky Linux and CloudLinux, uname -r already includes version and release (for example 5.14.0-687.52.1.el9_8.x86_64):

uname -r
rpm -q kernel-core

If the fixed kernel is installed but not running, you still need to reboot.

Exposure while you update: this command shows whether each feature is a module (=m, can be blocked) or built in (=y, cannot); empty output is inconclusive. A module that is not loaded does not protect you: it can be loaded automatically.

grep -E '^CONFIG_(INET6_AH|TUN|PPPOE|IP_SCTP|INET_SCTP_DIAG)=' /boot/config-$(uname -r)

How to mitigate

Option Best for Reboot? Covers all four?
A: fixed kernel (permanent) Everyone Yes Yes
B: block modules (temporary) If you do not use those features Only if in use Only what is a module
C: no user namespaces (temporary) No rootless containers No No (DiagSpill stays open)

Option A: install the fixed kernel and reboot

It is the only complete solution. Minimum versions with all four fixes (later versions in the same branch also work):

  • Debian 12: 6.1.190-1; with the linux-6.12 package, 6.12.111-1~deb12u1.
  • Debian 13: 6.12.111-1. On Debian, the fixes are in the security repository; make sure it is enabled.
  • Ubuntu: 22.04, 5.15.0-202.212; 24.04, 6.8.0-149.149; 26.04, 7.0.0-42.42. These apply only to the generic kernel; other branches (cloud, HWE, OEM) and releases older than 22.04 have their own versions and some are still unfixed: check the Ubuntu Security Tracker.
  • RHEL, AlmaLinux and Rocky Linux 8: kernel-4.18.0-553.168.1.el8_10.
  • RHEL, AlmaLinux and Rocky Linux 9: kernel-5.14.0-687.52.1.el9_8.
  • RHEL 10.2, AlmaLinux and Rocky Linux 10: kernel-6.12.0-211.60.1.el10_2. For RHEL 10.3, Red Hat only lists a fix for TUNderflow (kernel-6.12.0-264.14.1.el10_3): install the newest 10.3 kernel.
  • CloudLinux, Proxmox VE and others: look for the four CVEs in the kernel package changelog or your provider's tracker.

On Debian / Ubuntu, full-upgrade can install and remove packages: review the list before confirming.

apt update && apt full-upgrade

On RHEL, AlmaLinux and Rocky Linux (RHEL needs an active subscription). If your dnf configuration excludes kernel*, nothing will be installed: confirm with rpm -q kernel-core.

dnf upgrade --refresh kernel

Then reboot (as of November 12, 2026, Red Hat lists no kpatch live patch for DiagSpill). This interrupts service: use a maintenance window, make sure the new kernel is the default boot entry and, on Proxmox/KVM hosts, migrate the virtual machines first.

systemctl reboot

Option B: block the modules you do not use (temporary)

Red Hat's workaround: prevent loading of ah6, tun, pppoe, sctp and sctp_diag. It does not help with code built into the kernel (=y), but it is the only workaround that covers DiagSpill.

Side effects per Red Hat: blocking tun breaks VPNs (OpenVPN, userspace WireGuard) and container and virtual machine networking: do not do it on Docker, Proxmox or KVM hosts. ah6 is used for IPv6 IPsec transport mode; pppoe, for DSL connections; sctp and sctp_diag, in telecommunications and some high-availability setups.

Example with PPPoE (repeat with ah6 or tun if you do not use them); the last line must print nothing:

echo "install pppoe /bin/true" > /etc/modprobe.d/disable-pppoe.conf
modprobe -r pppoe
lsmod | grep -E '^pppoe\b'

For SCTP, unload sctp_diag first, because it depends on sctp:

echo "install sctp_diag /bin/true" > /etc/modprobe.d/disable-sctp_diag.conf
echo "install sctp /bin/true" > /etc/modprobe.d/disable-sctp.conf
modprobe -r sctp_diag
modprobe -r sctp
lsmod | grep -E '^sctp_diag\b'
lsmod | grep -E '^sctp\b'

The block lasts until you delete the .conf file. The researcher warns that blocking only the PoC modules is not a proper mitigation: there could be other paths to root.

Option C: disable unprivileged user namespaces (temporary)

Also from Red Hat. It removes the ordinary-user path for DirtyAH6, TUNderflow and PPPoEject, but does not mitigate DiagSpill and does not stop processes or containers that already hold CAP_NET_ADMIN or CAP_NET_RAW. It can break container runtimes, sandboxed browsers and other workloads that use user namespaces. Note the current value first:

sysctl user.max_user_namespaces

To apply it now and after reboots (the second line overwrites /etc/sysctl.d/userns.conf if it exists):

sysctl -w user.max_user_namespaces=0
echo "user.max_user_namespaces=0" > /etc/sysctl.d/userns.conf

To revert, delete the file and reboot, or restore the noted value with sysctl -w.

After mitigating

  1. After the reboot, use the check commands to confirm the running kernel is at or above the minimum version.
  2. Once the fixed kernel is running, remove the workarounds if you need those features: delete the .conf files you created and restore user.max_user_namespaces.
  3. If untrusted users had access to a vulnerable kernel, look for the traces the PoCs leave. The DirtyAH6 and TUNderflow PoCs replace pam_rootok.so with pam_permit.so; what the PPPoEject PoC (root credentials in kernel memory) and the DiagSpill PoC (a sudoers rule) do is not detected by these commands. dpkg --verify or rpm -V check the packages that own PAM and /etc/pam.d/su (no output, no changes) and grep shows active pam_rootok or pam_permit lines (on Debian 12, the stock file has one pam_rootok.so line and no pam_permit.so line):
    # Debian / Ubuntu
    dpkg --verify libpam-modules util-linux
    # RHEL, AlmaLinux, Rocky Linux
    rpm -V pam util-linux
    # All
    grep -nE '^[^#]*pam_(rootok|permit)\.so' /etc/pam.d/su
  4. A clean result does not prove the server is clean: memory-only changes disappear on reboot.
  5. Red Hat's bulletin is still open and the KEV catalog changes daily: check both again.

Need help?

Open a ticket at soporte.telecu.cloud and reference: CVE-2026-80844, CVE-2026-81000, CVE-2026-68121 or CVE-2026-74469.

Sources

Tags
How to protect your Linux server from CVE-2025-39964, CVE-2026-53266 and CVE-2025-39682 (three kernel flaws in the KEV catalog)
How to protect your Linux server from CVE-2026-53362 (kernel IPv6 fragmentation flaw)