Security advisory · Linux kernel

CVE-2026-99288: a use-after-free in the Linux kernel's TCP socket lookup

Updated November 18, 2026By HostCrust4 min read

CVE-2026-99288 is a use-after-free in the Linux kernel's lockless TCP connection lookup. A remote packet race can make the kernel use a freed socket, crashing it — a denial of service. It affects kernels from 4.4 onward and is fixed in 6.12.117 and 6.18.63; install your distribution's kernel update once it ships, then reboot.

CVE ID
CVE-2026-99288
Published
November 10, 2026
CVSS
5.9 (Medium), availability impact
Affected
Linux kernel 4.4 up to the fixed releases
Fixed in
6.12.117 (6.12.x), 6.18.63 (6.18.x), 7.3.4 (7.3.x), mainline 7.4-rc2; install your distro's kernel update, then reboot
Assigned by
Linux kernel (kernel.org)

What is CVE-2026-99288?

CVE-2026-99288 is a use-after-free flaw in how the Linux kernel looks up established TCP connections. Those lookups are deliberately lockless for speed, and the socket structures they read use SLAB_TYPESAFE_BY_RCU, which lets the kernel reuse a freed socket's memory while another CPU is still reading it. After a tuple match, __inet_lookup_established() takes a reference and re-checks the connection's addresses, but it never checks whether the socket it landed on is still hashed — still a live entry. Under rapid address reuse, a lookup can return a reference to a socket that has already been unlinked. A proof-of-concept that drove reuse with a SYN-cookie request socket reached a NULL pointer and crashed the kernel. The maintainers describe the result as a fatal exception in interrupt context — in practice, a remotely triggerable denial of service against a machine handling TCP traffic.

Am I affected?

You're affected if a server runs a Linux kernel built from 4.4 up to — but not including — the fixed releases, and it handles TCP traffic, which nearly every server does. That covers current Debian, Ubuntu, RHEL, AlmaLinux and Rocky Linux until they ship the backported fix in a kernel update. The flaw is in mainline kernel code, so distribution kernels inherit it; your distribution backports the fix without raising the upstream version number, which means comparing the version string alone won't tell you whether you're covered.

The NVD record lists a CVSS 3.1 base score of 5.9 (Medium), with the impact scored as availability only (CVSS:3.1/AV:N/AC:H/PR:N/UI:N/S:U/C:N/I:N/A:H): a network attacker can crash the machine, but the official score does not credit data disclosure or code execution. Because it is a use-after-free in a path reachable from the network, treat it as a patch-soon item on any internet-facing host.

How do I fix it?

Install a kernel that contains the fix and reboot. The fix does not take effect until the patched kernel is the one actually running.

# Debian / Ubuntu
sudo apt update && sudo apt full-upgrade
sudo reboot

# RHEL / AlmaLinux / Rocky Linux
sudo dnf upgrade --refresh
sudo reboot

The upstream fix is the commit titled “inet: reject unhashed sockets during established lookups” (d6910034…), backported across the stable branches. Debian has shipped it (bookworm 6.1.192-1 via DLA-4906-1, trixie 6.12.117-1 via DSA-6628-1); Red Hat marks RHEL 8, 9 and 10 as affected with the fix still pending (Red Hat rates it Important); and Ubuntu tracks it across its kernels. The vulnerable path is core TCP input, so until you reboot into the fixed kernel, limit who can reach the host. A server can look patched on disk and still be running the old kernel; our guide to updating the kernel on Ubuntu and the reboot guide cover both halves.

How HostCrust helps you see it

The trap with kernel CVEs is the gap between “installed” and “running”. Patch Proof checks each server against its distribution's own security data, per CVE, and it checks the kernel that is actually booted, not just the newest one on disk. So for a flaw like CVE-2026-99288 it keeps reporting the CVE as open until the patched kernel is the one running — which is exactly the case that bites people: the package manager reports success, but the vulnerable kernel is still live until a reboot. It's read-only, needs no root, and is free to run on one server.

How we checked this: Every fact on this page — the CVE id, the affected and fixed kernel versions, the CVSS score and the remediation — was taken from the NVD record, the Linux kernel's own CVE entry and fix commits, and the Debian, Ubuntu and Red Hat security trackers, all linked under Sources, and checked on November 18, 2026. HostCrust did not independently reproduce the vulnerability. Anything we didn't run ourselves comes from the official documentation listed under Sources.

Check which CVEs are really open on your server

HostCrust Patch Proof compares every installed package with your distribution's own security data. It counts fixes your distro backported and checks the kernel that is actually running, not just the one on disk. Read-only, no root, free.

Run the free check About Patch Proof

Reboots with an approval step

By default, HostCrust holds a reboot until an approver signs off (your org's policy decides), then checks that the server reports a new boot ID. Every step is kept in a receipt you can export. Free for up to 3 servers.

See how HostCrust works

FAQ

Is CVE-2026-99288 remotely exploitable?

The vulnerable code is core TCP input, which processes packets from the network, so the race can be driven remotely. A proof-of-concept drove reuse with a SYN-cookie request socket to reproduce the crash. The NVD record scores the impact as availability only — a remote attacker can crash the kernel (denial of service); the official score does not credit data disclosure or code execution.

Which kernel versions are fixed?

The upstream fix is in mainline (7.4-rc2) and backported to specific stable branches: the kernel.org record lists fixes in the 6.12.x, 6.18.x and 7.3.x series (6.12.117, 6.18.63, 7.3.4) and none for other series. Distributions backport it into their own kernels too (Debian shipped it in its bookworm 6.1 and trixie 6.12 kernels). Because the fix lands per branch, a plain “6.12.117 and later” doesn't hold across branches; install your distribution's current kernel security update rather than matching version numbers by hand.

Do I have to reboot after patching?

Yes. A kernel update only takes effect once the new kernel is running. Until you reboot, the machine keeps running the old, vulnerable kernel even though the package manager reports the update as installed, so plan a reboot. Our Linux reboot guide covers doing it safely.

How do I check the kernel I'm actually running?

Run uname -r to see the running kernel, and compare it with what's installed on disk. They differ whenever a kernel update is waiting for a reboot. Our guide to checking your kernel version walks through it, and Patch Proof checks the running kernel for you across every server.

Sources

  1. NVD: CVE-2026-99288
  2. CVE.org record: CVE-2026-99288 (Linux kernel CNA, affected and fixed versions)
  3. kernel.org stable commit: inet: reject unhashed sockets during established lookups (d6910034…)
  4. Debian security tracker: CVE-2026-99288 (bookworm 6.1.192-1 / DLA-4906-1, trixie 6.12.117-1 / DSA-6628-1)
  5. Ubuntu security: CVE-2026-99288 (tracked across the supported kernels)
  6. Red Hat: CVE-2026-99288 (RHEL 8/9/10 affected, fix pending; severity Important)
  7. oss-security mailing list: CVE-2026-99288 announcement