What is CVE-2026-53362?
CVE-2026-53362 is an out-of-bounds write in the Linux kernel's IPv6 output code. It lets an unprivileged local user corrupt kernel memory and, with the public exploit, become root on the host, even from inside a container. Red Hat calls it ipv6_frag_escape and rates it Important.
It is local only: it cannot be triggered from the network, but any unprivileged user who opens a UDPv6 socket can trigger it, without CAP_NET_RAW. Any shell account, compromised web application or container is a possible starting point for local privilege escalation (LPE).
The kernel CNA (CVE Numbering Authority) scores it CVSS 3.1 7.8 (High). The CVE was published on July 4, 2026, but the public proof of concept (PoC) had been on GitHub since June 29, 2026. CISA added it to its Known Exploited Vulnerabilities (KEV) catalog on August 27, 2026, and the CVE record marks exploitation as active.
A sibling bug, CVE-2026-53366, has the same root cause in the IPv4 UDP path.
Are you affected?
As of November 12, 2026:
- Not affected by CVE-2026-53362: RHEL 8 and 9, AlmaLinux 8 and 9, CloudLinux 7, 7 Hybrid, 8 and 9, and Ubuntu 22.04 with its default 5.15 kernel. RHEL 9 and derivatives do need the CVE-2026-53366 update (see the table); for CloudLinux 9, its status is not confirmed.
- Affected unless you have updated: RHEL 10 and derivatives (AlmaLinux, Rocky Linux, CloudLinux 10), Debian 12 and 13, Ubuntu 24.04 and 26.04, Ubuntu 22.04 with 6.8 kernels (some still unfixed: check the Ubuntu tracker) and Proxmox VE.
More distributions are affected, such as SUSE or Oracle Linux: if yours is not listed here, check its security tracker.
Containers (Docker, Podman, LXC) run on the host kernel: patch the host. Each KVM virtual machine runs its own kernel. cPanel/WHM and DirectAdmin do not ship a kernel: your operating system's kernel is what counts.
| Distribution | Fixed kernel (or newer) | Note |
|---|---|---|
| RHEL 10 and derivatives (AlmaLinux, Rocky Linux, CloudLinux 10) | kernel-6.12.0-211.30.1.el10_2 |
RHSA-2026:34911. On RHEL 10.0 EUS: kernel-6.12.0-55.86.1.el10_0
|
| RHEL 9 and derivatives (AlmaLinux, Rocky Linux) | kernel-5.14.0-687.46.1.el9_8 |
Not affected by CVE-2026-53362; this build fixes CVE-2026-53366 |
| Debian 12 and 13 |
6.1.177-1 (12), 6.12.95-1 (13) |
From the security repository |
| Ubuntu 24.04 and 26.04 |
6.8.0-147.147 (24.04, generic), 7.0.0-31.31 (26.04) |
HWE or OEM kernels: check the Ubuntu tracker |
| Proxmox VE |
6.8.12-35 or 6.17.13-17
|
Depending on the series; the opt-in 6.14 kernel is not fixed |
How to check your server
Compare the running kernel with the fixed build in the table. Because of backports, numbers are only comparable within the same distribution and stream, and an installed kernel only takes effect after a reboot.
Debian and Ubuntu:
uname -r; dpkg -l 'linux-image-*' | awk '/^ii/{print $2, $3}'
On Ubuntu, 6.8.0-147-generic corresponds to package 6.8.0-147.147; cloud kernels (linux-aws, linux-azure…) have their own numbering: look them up in the Ubuntu tracker. On Debian, this command shows the package version of the running kernel:
uname -v
RHEL, AlmaLinux, Rocky Linux and CloudLinux:
uname -r; rpm -q kernel
Proxmox VE: look for a "fix CVE-2026-53362" entry in the changelog of the proxmox-kernel package you have installed.
How to mitigate
| Option | Best for | Reboot? | Durable? |
|---|---|---|---|
| A: install the fixed kernel | Everyone; the real fix | Yes | Yes |
| B: livepatch | CloudLinux 10 with KernelCare | No | Yes |
| C: disable unprivileged user namespaces | RHEL 10 and derivatives without rootless containers, until reboot | No | Temporary and partial |
Option A: install the fixed kernel
Debian 12 and 13: enable the security repository; Debian 12's main archive still carries 6.1.176-1, which is vulnerable. This command upgrades only the kernel metapackage (replace linux-image-amd64 with yours, for example linux-image-cloud-amd64):
sudo apt update && sudo apt install --only-upgrade linux-image-amd64
Ubuntu: full-upgrade upgrades every pending package, not just the kernel, and can restart services such as databases or Docker, so review the list first:
sudo apt update
apt list --upgradable
sudo apt full-upgrade
Proxmox VE: update until you have a proxmox-kernel at or above the one in the table; if you are on the opt-in 6.14 kernel, move to a fixed series.
RHEL, AlmaLinux, Rocky Linux and CloudLinux:
sudo dnf clean metadata && sudo dnf upgrade 'kernel*'
Before rebooting, confirm with rpm -q kernel that a build equal to or newer than the one in the table for your stream is listed.
The new kernel only takes effect after a reboot. Rebooting interrupts every service, so schedule a maintenance window:
sudo reboot
If you use third-party kernel modules (DKMS), first make sure they build for the new kernel.
Option B: livepatch
On CloudLinux 10, the KernelCare livepatch K20260630_04 has been available since June 30, 2026 and covers CVE-2026-53362 and CVE-2026-53366. Verify it with CloudLinux's steps, not with kcarectl --info | grep CVE-…, which returns empty output even on patched systems.
Red Hat's kpatch for RHEL 10 targets only kernel-6.12.0-211.16.1.el10_2; on any other kernel, use option A.
Option C: disable unprivileged user namespaces (temporary)
This workaround is documented only for RHEL 10 and derivatives, not for Debian or Ubuntu. It does not fix the bug: it reduces exposure and removes the container-escape path, but a local process outside a container can still reach the bug.
Side effects: it breaks rootless Podman and Docker, and may break unprivileged LXC containers and browser or CI sandboxes.
First write down the current value, so you can revert the change:
sysctl user.max_user_namespaces
Apply it at runtime; from that moment, anything that relies on user namespaces will stop working:
sudo sysctl -w user.max_user_namespaces=0
Make it persist across reboots:
echo 'user.max_user_namespaces = 0' | sudo tee /etc/sysctl.d/99-disable-userns.conf
What does not protect you
- Blacklisting the
esp4/esp6/rxrpcmodules, as was done for Dirty Frag: it does not cover this vulnerability. - Disabling IPv6: no distribution documents it as a mitigation, and it does not cover the IPv4 sibling bug.
-
init_on_alloc=1: it turns the public technique into a crash instead of a root shell, but it does not fix the bug, and a failed attempt can crash the server.
After mitigating
- Confirm with
uname -rthat the running kernel is the fixed build (or, with KernelCare, that you verified the livepatch). - If you applied option C and need user namespaces, revert it once you have booted the fixed kernel, restoring the value you wrote down (65534 by default on RHEL, according to Red Hat):
sudo rm /etc/sysctl.d/99-disable-userns.conf sudo sysctl -w user.max_user_namespaces=YOUR_PREVIOUS_VALUE - CISA flags this flaw for forensic triage, but its entry includes no indicators of compromise (IoC). As a quick check, the public exploit finishes by hijacking
core_pattern:
Values such ascat /proc/sys/kernel/core_patterncoreor a systemd-coredump or apport handler are normal; a handler pointing into/proc/deserves investigation. A normal value does not prove the server is clean. - If you find signs of compromise on a shared server, treat it as fully compromised: with root, the attacker can reach every tenant's files, databases, credentials and backups. Rotate those credentials; in the worst case the server must be rebuilt.
Need help?
Open a ticket at soporte.telecu.cloud and reference: CVE-2026-53362.
Sources
- CVE Program: CVE-2026-53362 record (Linux kernel CNA)
- CISA: Known Exploited Vulnerabilities catalog
- Red Hat: security bulletin RHSB-2026-009
- Red Hat: CVE-2026-53362 data (RHEL 10)
- Red Hat: CVE-2026-53366 data (RHEL 9)
- Debian Security Tracker: CVE-2026-53362
- Ubuntu Security: CVE-2026-53362
- CloudLinux: mitigation and fixed kernel for CloudLinux 10