Issue
Is there a KernelCare patch or a kernel fix for CVE-2026-99288 (NoisyTCP), a remote denial-of-service vulnerability in the Linux kernel TCP stack, and when will it be available?
- The TuxCare CVE tracker shows KernelCare patches in progress only for some Oracle Linux versions. Are patches planned for Ubuntu 22.04, Ubuntu 24.04 and CloudLinux 8?
- Is the CloudLinux kernel affected, and is there a mitigation?
Environment
- KernelCare
- CloudLinux OS 8, 9 and 10 (kernel)
- Ubuntu 22.04 (jammy)
- Ubuntu 24.04 (noble)
- Oracle Linux UEK
- CVE-2026-99288 (NoisyTCP)
Solution
As of 2026-11-16, there is no vendor kernel fix and no KernelCare patch for CVE-2026-99288 on any distribution, including Ubuntu 22.04, Ubuntu 24.04 and CloudLinux 8. The KernelCare team is tracking the CVE. There is no ETA until a fix exists, and a patch can only be built once the vendor fix is available.
Follow the KernelCare patch status. The TuxCare CVE tracker page for CVE-2026-99288 shows the KernelCare state for each system.
If your system is not in the list, no patch work has started for it yet, because the vendor has not released a fix. Check the page again later: it is updated when the state changes.
Vendor status on 2026-11-16:
-
Ubuntu listed
linuxfor jammy (22.04) and noble (24.04) as "Needs evaluation", with no fixed version. Ubuntu rates the priority as Medium. - Red Hat rates it Important (CVSS 7.5). It lists the RHEL 8, 9 and 10 kernels as affected and RHEL 6 and 7 as not affected, with no fixed kernel yet.
- The CloudLinux changelog had no entry for the CVE. CloudLinux OS 8, 9 and 10 kernels are likely affected, because they follow the RHEL kernels.
-
No public exploit or reported exploitation was known.
Temporary mitigations. Red Hat publishes two mitigations on its CVE page. TuxCare has not tested either of them, and both have drawbacks, so weigh these before applying one.
-
Disable SYN cookies. This stops the attack but also removes SYN flood protection. The setting applies to each network namespace, so apply it in the namespaces used by containers too.
sysctl -w net.ipv4.tcp_syncookies=0To revert, set the value back to
1. -
Configure the internet-facing network interface to use a single receive queue, with RPS and RFS turned off. This can reduce network throughput, especially on busy servers.
If you want TuxCare Support to confirm the status of your servers, open a ticket and attach a doctor key from one server per CloudLinux OS version (
cldiag --doctor) and from one KernelCare server (kcarectl --doctor).
Cause
The upstream bug is a race in lockless TCP socket lookups. In the established-socket lookup, the kernel takes a reference to a socket after a tuple match without checking that the socket is still hashed. A recycled socket can then be used before it is initialized, which causes a NULL pointer dereference and a kernel crash. A remote, unauthenticated attacker can trigger it through an open TCP port, and it is particularly effective against virtual machines.
The upstream fix is "inet: reject unhashed sockets during established lookups". Vendors had not yet shipped it for these distributions, so KernelCare had no vendor fix to build a patch from. Open KernelCare work at the time covered only the Oracle Linux errata (ELSA-2026-500630 and ELSA-2026-500631).
Comments
0 comments
Article is closed for comments.