NoisyTCP (CVE-2026-99288) is a bug in the Linux kernel's TCP code that lets a remote attacker, with no login, crash a server that accepts TCP connections, for example nginx, Apache or SSH on their default settings. Your server is safe once it runs a fixed kernel: upstream 6.12.117, 6.18.63, 7.3.4 or newer, Debian 13's 6.12.117-1 or Debian 12's 6.1.192-1, and you have rebooted into it; on 20 November 2026 Ubuntu 24.04 and 26.04 still had no fixed kernel.
This guide shows how to check which kernel you are really running, how to patch and reboot on Debian and Ubuntu, and how to use the SYN cookie stopgap without forgetting to undo it. The commands ran in a Debian 13 test container on 20 November 2026; no attack was run. If you are choosing a server, our KVM VPS and VDS plans in Amsterdam, Dublin and Prishtina each run their own kernel, so you decide when it is patched and rebooted.
Key facts
- CVE-2026-99288 was published by the kernel CNA on 10 November 2026 under the fix title "inet: reject unhashed sockets during established lookups" (CVE record).
- The CVE record marks Linux 4.4 and later as affected and lists 6.12.117, 6.18.63, 7.3.4 and 7.4-rc2 as the first fixed versions in their series (CVE record).
- The flaw came in with commit 079096f103fa on 3 October 2015 and is fixed by commit 4e2b3ef19e0e of 3 November 2026, per the researcher, Hyunwoo Kim (oss-security advisory, NoisyTCP write-up).
- The proven impact is a kernel panic, a denial of service. The researcher also describes a use-after-free path that is "hard to trigger" and says remote code execution is "not yet" shown (write-up).
- Debian fixed trixie in DSA-6628-1 (6.12.117-1, 13 November 2026) and bookworm in DLA-4906-1 (6.1.192-1, 15 November 2026) (Debian tracker).
- Ubuntu rates it High priority and on 20 November 2026 listed 22.04, 24.04 and 26.04 as vulnerable with no fixed version (Ubuntu CVE page).
Quick check and fix
- Running kernel:
uname -r. Debian 13 needs6.12.117+deb13or newer, Debian 12 needs6.1.192or newer.- Installed fix:
zcat /usr/share/doc/linux-image-$(uname -r)/changelog.gz | grep CVE-2026-99288. A line with the fix title means the kernel you are running has it.- Patch:
sudo apt update && sudo apt full-upgrade, thensudo reboot. Installing a kernel does nothing until you boot it.- Cannot reboot today:
sudo sysctl -w net.ipv4.tcp_syncookies=0blocks the main crash path, but lowers your protection against SYN floods. Set it back to1after the reboot.- On a container VPS (LXC, OpenVZ, Docker host), the provider's host kernel decides. Ask them, or check with
systemd-detect-virt.
What NoisyTCP does, in plain words
When a packet arrives, the kernel looks up which connection it belongs to. To be fast, that lookup takes no lock, and the memory of a closed socket can be reused for a new one while the lookup is still holding it. The CVE record says the lookup could take a reference to such a reused, unhashed socket and return it, which ends in a NULL pointer dereference in tcp_v4_rcv(). A NULL pointer dereference inside the kernel is a kernel panic: the whole machine stops, not just one program.
So the realistic outcome is downtime. BleepingComputer, citing the researcher, reports that the bug does not let an attacker change data or run code (BleepingComputer, 16 November 2026). Nobody has reported data theft through NoisyTCP. The researcher's second, harder path is a use-after-free, which is the kind of bug that sometimes grows into more, so treat the fix as urgent anyway.
The attack needs nothing special from you. The write-up says it triggers "even when the server does nothing more than call listen()", so any open TCP port counts, including SSH on port 22. The fix also changes how the kernel releases references in the SYN cookie, TCP Fast Open and request migration paths, which is why SYN cookies show up in the stopgap below.
Who is affected
Almost every Linux server with an open TCP port is in scope. The researcher lists three conditions:
- a kernel from 4.4 onward without the fix;
- a plain TCP listener the attacker can reach directly;
- two or more receive (RX) queues on the network card, or RPS spreading packets over several CPUs.
The timing matters for VPS owners. The researcher reports the crash took about 5 seconds to 2 minutes against QEMU/KVM virtual machines, and about 14 hours against a quiet bare-metal server with 8 RX queues (oss-security). Virtual servers are the easy case, not the hard one.
Containers do not protect you. In the researcher's words, a crash against a web server in a container "takes down not only the web server application but the entire host". One customer's container being hit stops every container on that machine.
Check your kernel
Start with the kernel that is actually running, not the one on disk:
uname -r
6.12.111+deb13-amd64
That is a vulnerable Debian 13 kernel (below 6.12.117). Now list the kernel packages that are installed:
dpkg -l 'linux-image*' | grep ^ii
ii linux-image-6.12.117+deb13-amd64 6.12.117-1 amd64 Linux 6.12 for 64-bit PCs (signed)
ii linux-image-amd64 6.12.117-1 amd64 Linux for 64-bit PCs (meta-package)
This is the most common trap. The fixed 6.12.117 package is installed, but uname -r still says 6.12.111, so the machine is still running the old kernel. Until a reboot, you are not patched.
To see whether a kernel package contains this fix, search its changelog for the CVE ID. Debian keeps the changelog on disk:
zcat /usr/share/doc/linux-image-6.12.117+deb13-amd64/changelog.gz | grep -n CVE-2026-99288
920: - inet: reject unhashed sockets during established lookups (CVE-2026-99288)
You can do the same before installing anything, from Debian's servers:
apt changelog linux-image-6.12.117+deb13-amd64 | grep -n 99288
921: - inet: reject unhashed sockets during established lookups (CVE-2026-99288)
No output means that package does not carry the fix. To see which version apt would install and from where:
apt-cache policy linux-image-amd64
linux-image-amd64:
Installed: 6.12.117-1
Candidate: 6.12.117-1
Version table:
*** 6.12.117-1 500
500 http://deb.debian.org/debian-security trixie-security/main amd64 Packages
6.12.107-1 500
500 http://deb.debian.org/debian trixie/main amd64 Packages
If trixie-security is missing from that list, your server does not receive security updates at all. Fix the apt sources first; our Debian 13 VPS hardening guide shows a correct setup.
Is a reboot pending? On our Debian 13 test, installing the new kernel did not create /var/run/reboot-required, so do not rely on that file on Debian. Comparing uname -r with ls /boot always works. needrestart -k also reports a pending kernel on a normal VM; inside a container it prints nothing, because the kernel is not the container's to change.
Patch status on 20 November 2026
| System | Status on 20 November 2026 | Fixed kernel | What to do |
|---|---|---|---|
| Debian 13 (trixie) | Fixed, DSA-6628-1 (13 November) | 6.12.117-1 | Upgrade, reboot, check uname -r |
| Debian 12 (bookworm) | Fixed, DLA-4906-1 (15 November) | 6.1.192-1 | Upgrade, reboot; bookworm-security must be in your sources |
| Ubuntu 26.04 LTS | Vulnerable, no fix listed | None yet | Stopgap, watch the CVE page |
| Ubuntu 24.04 LTS | Vulnerable, no fix listed (generic, aws, azure, gcp, oracle) | None yet | Same as 26.04 |
| Upstream stable | Fixed | 6.12.117, 6.18.63, 7.3.4 | Custom kernels: move to these or newer |
| Upstream 5.15 | 5.15.225 not fixed; fix queued for 5.15.226 | Check | See oss-security, 20 November |
Sources: Debian security tracker, DSA-6628-1, DLA-4906-1, Ubuntu, CVE record. Distribution kernels keep their own version numbers, so on other distributions check the vendor's page for CVE-2026-99288 rather than comparing with upstream numbers.
Update and reboot safely
A kernel update is one of the few updates that always needs a reboot. On Debian 13 or 12:
sudo apt update
sudo apt full-upgrade
ls /boot | grep vmlinuz
sudo reboot
A few habits make the reboot boring:
- Take or confirm a backup first. On our servers the free weekly backup is restored with one click from the client area; elsewhere, use what your provider offers.
- Make sure the services you need are enabled to start on boot, so nginx, SSH and your database come back without you.
- Keep the VNC or serial console from your control panel open in case the network does not come back.
- After the reboot, run
uname -ragain and check it shows the new version.
The previous kernel stays installed unless you remove it, so if the new one fails to boot you can pick the old one in the boot menu from the console. To stop forgetting kernel updates in the first place, set up automatic security updates on Debian 13; they install the kernel, but you still choose when to reboot.
Live kernel patching: we found no primary source on 20 November 2026 confirming a live patch for CVE-2026-99288. If you use a live patching service, check its own notice before skipping the reboot.
Stopgap: turn off SYN cookies, and turn them back on
If you cannot reboot yet, for example on Ubuntu where no fixed kernel exists, the researcher's mitigation is to disable SYN cookies, which blocks the path that rebuilds request sockets. Check the current value:
sysctl net.ipv4.tcp_syncookies
net.ipv4.tcp_syncookies = 1
Turn it off now:
sudo sysctl -w net.ipv4.tcp_syncookies=0
net.ipv4.tcp_syncookies = 0
That lasts until the next boot. To keep it until a fixed kernel arrives:
echo 'net.ipv4.tcp_syncookies=0' | sudo tee /etc/sysctl.d/99-noisytcp.conf
sudo sysctl --system
The trade-off is real. SYN cookies are what keep a server answering during a SYN flood, and the researcher says plainly that disabling them "lowers the protection against SYN floods". You swap a crash bug for weaker flood protection, so only do it on servers you cannot patch, and watch for floods with the steps in our guide on detecting a DDoS attack on a Linux server.
The setting is per network namespace. The researcher notes it has to be set in each container that has its own network namespace, not just on the host. The researcher's other option is to cut the network card to one RX queue and disable RPS and RFS, which costs throughput on busy servers; you can see your queues with ls /sys/class/net/eth0/queues/.
To revert after you have rebooted into a fixed kernel:
sudo rm /etc/sysctl.d/99-noisytcp.conf
sudo sysctl -w net.ipv4.tcp_syncookies=1
net.ipv4.tcp_syncookies = 1
Does your VPS run its own kernel?
This is the part most NoisyTCP articles skip, and it matters when you buy a VPS. A KVM virtual server boots its own kernel: you install the update, you reboot, you are fixed. A container VPS (LXC, OpenVZ, Virtuozzo) has no kernel of its own. It shares the host's kernel with every other customer on that machine, and only the provider can patch it.
Our Debian 13 test container shows this clearly. We installed the fixed kernel package inside it:
systemd-detect-virt
uname -r
ls /boot
lxc
6.12.111+deb13-amd64
System.map-6.12.117+deb13-amd64
config-6.12.117+deb13-amd64
initrd.img-6.12.117+deb13-amd64
vmlinuz-6.12.117+deb13-amd64
The 6.12.117 files are sitting in /boot, but the container still reports the host's 6.12.111 kernel, and rebooting the container would not change that. systemd-detect-virt prints lxc, openvz or another container type in a container, and kvm or another hypervisor name in a full virtual machine (systemd-detect-virt manual). If yours says a container type, open a ticket with your provider and ask which host kernel you are on.
Every RS Computers server, VPS and VDS, is a KVM virtual machine with its own kernel and full root, in Amsterdam, Dublin and Prishtina. That also means the reboot is yours to schedule.
Frequently asked questions
What is NoisyTCP?
NoisyTCP is the name researcher Hyunwoo Kim gave to CVE-2026-99288, a race in the Linux kernel's lockless TCP socket lookup that can make the kernel dereference a NULL pointer and panic. It can be triggered remotely, without a login, against any reachable TCP listener.
Is my server affected by NoisyTCP?
It is if it runs Linux 4.4 or newer without the fix, has a TCP port open to the attacker and has two or more RX queues or RPS turned on. Check uname -r against the fixed versions: upstream 6.12.117, 6.18.63, 7.3.4, Debian 13 6.12.117-1, Debian 12 6.1.192-1.
Which kernel version fixes CVE-2026-99288?
Upstream 6.12.117, 6.18.63, 7.3.4 and 7.4-rc2 are the first fixed versions in their series, per the CVE record. Distribution kernels carry the fix under their own numbers, so search the package changelog for CVE-2026-99288.
Is NoisyTCP patched on Debian and Ubuntu?
Debian 13 is fixed in 6.12.117-1 (DSA-6628-1) and Debian 12 in 6.1.192-1 (DLA-4906-1). On 20 November 2026 Ubuntu listed 22.04, 24.04 and 26.04 as vulnerable with no fixed kernel yet.
Can NoisyTCP steal data?
No data theft has been shown. The confirmed impact is a kernel panic that takes the server offline. The researcher describes a second use-after-free path that is hard to trigger and says remote code execution is not yet shown.
Do I need to reboot after the update?
Yes. The new kernel only runs after a reboot. Until uname -r shows the fixed version, the server is still vulnerable, even with the fixed package installed.
Is it safe to disable SYN cookies?
It is a short-term trade. Setting net.ipv4.tcp_syncookies=0 blocks the main NoisyTCP path but makes the server weaker against SYN floods. Use it only until you can reboot into a fixed kernel, then set it back to 1.
Need a hand?
If you are on a container VPS and cannot get a straight answer about the host kernel, move to a KVM VPS or VDS with NVMe in Amsterdam, Dublin or Prishtina, where the kernel and the reboot are yours. Busy production servers that need steady CPU fit our VDS plans with dedicated vCPU. For custom work, such as moving services off an old server or setting up update and reboot routines across several machines, we do custom offers, custom setups and systems built for you, quoted per job: write to us on Telegram or at info@rscomputers-ks.com.