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

What are CVE-2025-39964, CVE-2026-53266 and CVE-2025-39682?

On September 18, 2026, CISA added three Linux kernel vulnerabilities to its Known Exploited Vulnerabilities (KEV) catalog, the list of flaws exploited in the wild. CISA published no attack details and no indicators of compromise (IoCs).

  • CVE-2025-39964: race condition in AF_ALG crypto sockets. CVSS 7.8 (Linux CNA); Red Hat: Moderate. Public exploit since August 11, 2026.
  • CVE-2026-53266: out-of-bounds write in the ebtables SNAT target (bridges). CVSS 8.8 (Linux CNA); Red Hat: Important. No known public exploit.
  • CVE-2025-39682: mishandled zero-length TLS records in kernel TLS (kTLS). CVSS 9.8 (Linux CNA); Red Hat: Moderate, remote denial of service (DoS). Public exploit since February 25, 2026.

CVE-2025-39964 and CVE-2026-53266 are local: the attacker must already be able to run code on the server. According to its finders, CVE-2025-39964 lets an ordinary user become root. CVE-2025-39682 can only be triggered remotely if an application uses kTLS (uncommon on web hosting, according to CloudLinux); the public exploit triggers it locally.

Are you affected?

Short answer: if your kernel is older than those in Option A, yes, at least through CVE-2025-39964, which has been present since kernel 2.6.38 and is reachable by any local user, even inside a container. The risk depends on who can run code: shared hosting accounts (cPanel, DirectAdmin), containers or a compromised web application. If only you have access, an attacker first needs another way in. On a container host, the host kernel is the one that counts; each virtual machine uses its own.

  • CVE-2025-39964: affects every CloudLinux version and unpatched Debian, Ubuntu and RHEL-and-derivatives kernels. Ubuntu 26.04 is not affected.
  • CVE-2026-53266: needs a bridge with an ebtables SNAT rule that rewrites ARP, plus CAP_NET_ADMIN, a privilege an ordinary user gets inside their own user namespace. Where users can create them (Debian kernels allow it by default), a tenant can set it all up in their own account. The kernel version does not rule it out: there are fixes even for RHEL 8. On CloudLinux 7 it is out of reach with the default configuration.
  • CVE-2025-39682: does not affect RHEL 8, CloudLinux 7, 7 Hybrid and 8, or Ubuntu 26.04. It does affect RHEL 9 (kernel 5.14) and CloudLinux 8 LTS, 9, 9 LTS and 10, despite their version numbers.

How to check your server

These commands do not change your configuration. The running kernel is the one that counts; on Debian and Ubuntu, the third line gives its package version:

uname -r
uname -v
dpkg-query -W "linux-image-$(uname -r)"   # Debian/Ubuntu only

Installed kernels (Debian and Ubuntu):

dpkg-query -W -f='${db:Status-Abbrev}${Package}\t${Version}\n' 'linux-image-*' | grep '^ii' | sort -V

On RHEL and derivatives and on CloudLinux (kernel-lts on CloudLinux 8 LTS and 9 LTS):

uname -r
rpm -q kernel kernel-lts

For CVE-2025-39682: with CONFIG_TLS=y (never shown by lsmod) or the tls module loaded, any local user can use kTLS. Because the module loads on demand, empty lsmod output guarantees nothing. TlsCurrRxSw above 0 in /proc/net/tls_stat means kTLS sessions exist, but a 0 does not prove you are safe.

lsmod | grep -w '^tls'
cat /proc/net/tls_stat
grep -E '^CONFIG_TLS=' /boot/config-$(uname -r)

For CVE-2026-53266, look for bridge SNAT rules with --snat-arp. If the ebtables command does not exist, the first command fails, which does not mean there are no rules. A clean result does not include rules a user creates inside their own namespace or container.

sudo ebtables -t nat -L        # only if the ebtables command exists
sudo nft list ruleset bridge   # nftables view of bridge-family rules

On multi-tenant servers, check whether accounts can create user namespaces: 0 closes that path and any other value leaves it open. Write the value down before you change anything:

cat /proc/sys/user/max_user_namespaces

How to mitigate

Option Best for Reboot?
A: update the kernel Everyone; the real fix Yes
B: livepatch KernelCare or kpatch users No
C (temporary): block the tls module CVE-2025-39682 with no kTLS in use No
D (temporary): remove ebtables SNAT rules with ARP CVE-2026-53266 with bridges No
E (temporary): user.max_user_namespaces=0 CVE-2026-53266 on shared hosting No

There is no workaround for CVE-2025-39964 (on CloudLinux, blocking modules does nothing): only options A and B fix it.

Option A: update the kernel

Minimum fixed versions as of November 12, 2026 (for RHEL, on 8.10, 9.8 and 10.2; other streams or rebuilds may use different builds):

  • Debian 12: 6.1.176-1; Debian 13: 6.12.94-1.
  • Ubuntu 26.04: 7.0.0-31.31; 24.04: 6.8.0-146.146; 22.04: 5.15.0-198.208. On 24.04 and 22.04 these are for the standard (GA) kernel; with an HWE kernel, check Ubuntu's page.
  • RHEL 8: kernel-4.18.0-553.166.1.el8_10; RHEL 9: kernel-5.14.0-687.50.1.el9_8; RHEL 10: kernel-6.12.0-211.59.1.el10_2.
  • CloudLinux 7 Hybrid and 8: 4.18.0-553.168.1.lve; CloudLinux 9: 5.14.0-687.51.1.el9_8; CloudLinux 10: 6.12.0-211.61.1.el10_2. CloudLinux 7 has no kernel patch planned, only livepatch.
  • Proxmox VE 9: proxmox-kernel-7.0 7.0.14-9, based on Ubuntu kernel 7.0.0-31.31, includes the CVE-2026-53266 fix.

No Rocky Linux advisory naming CVE-2026-53266 was found: check the kernel package changelog. For other versions (Ubuntu 20.04, CloudLinux 8 LTS and 9 LTS), check the Debian, Ubuntu or Red Hat tracker (change the CVE in the URL) or the CloudLinux notice.

Debian and Ubuntu: upgrade only the kernel metapackage; replace linux-image-amd64 with yours:

sudo apt update && sudo apt install --only-upgrade linux-image-amd64   # Ubuntu: e.g. linux-image-generic

RHEL and derivatives, CloudLinux 9 and 10 (on CloudLinux 7 Hybrid and 8, yum update kernel). If no new kernel shows up, check whether dnf or yum excludes kernel packages:

sudo dnf update kernel        # RHEL and derivatives, CloudLinux 9/10 (CloudLinux 7 Hybrid/8: yum update kernel)

The new kernel is only used after a reboot. This command reboots the server and stops all of its services: run it in a maintenance window. If you suspect the server is already compromised, do not reboot yet: a reboot destroys volatile evidence, and CISA recommends collecting it before you alter the system.

sudo systemctl reboot

Option B: livepatch

With KernelCare, apply the livepatches without a reboot and check which CVEs they cover with --patch-info (not --info). No line for a CVE means it is not covered or not applicable to your kernel: it does not prove you are safe. Red Hat shipped kpatch livepatches only for CVE-2026-53266, and only for specific builds.

sudo kcarectl --update
kcarectl --patch-info | grep -E 'CVE-2025-39964|CVE-2026-53266|CVE-2025-39682'

Option C: block the tls module (temporary, CVE-2025-39682)

This is Red Hat's mitigation: it unloads the module and prevents it from loading again. If the unload fails, something is using it: stop there. Anything that relies on kTLS will stop working. It has no effect if CONFIG_TLS=y.

sudo modprobe -r tls && echo 'install tls /bin/false' | sudo tee /etc/modprobe.d/disable-ktls.conf

Option D: remove ebtables SNAT rules with ARP (temporary, CVE-2026-53266)

Red Hat recommends disabling ARP address rewriting in ebtables SNAT rules, or removing the ones that operate on ARP traffic on bridge interfaces. This does not stop a user from creating their own rule inside their namespace; that is what option E is for.

Option E: disable user namespaces (temporary, shared hosting only)

This is CloudLinux's mitigation; the second command makes it persistent. Major side effect: no process, not even root's, can create user namespaces, which breaks unprivileged containers (such as LXC on Proxmox) and services that use them. Do not use it on container, virtualization or development hosts. It does not protect against processes that already hold CAP_NET_ADMIN.

sudo sysctl -w user.max_user_namespaces=0
echo 'user.max_user_namespaces = 0' | sudo tee /etc/sysctl.d/99-disable-userns.conf   # shared-hosting servers only

After mitigating

  1. After the reboot, confirm that the running kernel is at or above the version in the list.
  2. If you used KernelCare, confirm coverage with kcarectl --patch-info.
  3. Once on a fixed kernel, delete /etc/modprobe.d/disable-ktls.conf (option C) and /etc/sysctl.d/99-disable-userns.conf (option E), and restore the user.max_user_namespaces value you wrote down.
  4. If you see signs of unauthorized access, treat the server as an incident and collect evidence before you change it.

Need help?

Open a ticket at soporte.telecu.cloud and reference: CVE-2025-39964, CVE-2026-53266, CVE-2025-39682.

Sources

Tags
How to protect your WordPress site from CVE-2026-87902 (unauthenticated file inclusion)
How to protect your Linux server from DirtyAH6, TUNderflow, PPPoEject and DiagSpill (kernel local privilege escalation)