NoisyTCP lets a stranger crash a Linux server. Debian has a fix, Ubuntu and Red Hat don't yet
A newly published Linux kernel bug, nicknamed NoisyTCP, lets someone on the internet crash your server without logging in. They can't steal data with it or take the server over. But they can switch it off, and keep switching it off.
Debian has a fix. Ubuntu and Red Hat don't yet. Here's what it is and what I'd do this week.
What happened
The bug is CVE-2026-99288. The researcher who found it posted the details on 15 November, with a fuller write-up on GitHub. The Register covered it on Monday.
It's a timing bug in how the kernel handles incoming connections. Get the timing right and the kernel panics, and the server stops dead. The code it lives in dates from October 2015, and the write-up says it "affects almost all Linux distributions".
The attacker doesn't need a password or any unusual setup on your side. Your server only has to be listening for connections, which is the whole point of a web server, and its network card needs more than one receive queue, or a feature called RPS switched on. Plenty of servers have one or the other, so don't assume yours is safe on that count.
How long it takes varies a lot. Against virtual machines in the researcher's lab, it took between about 5 seconds and 2 minutes. Against a quiet physical server, about 14 hours.
Red Hat rates it Important, 7.5 out of 10, and says "it does not allow data alteration or arbitrary code execution". So this is a crash bug, not a break-in bug. I can't find any report of attacks using it, and the exploit code hasn't been published.
Does it affect you?
You have your own server or rent a virtual one. The kernel is yours to update, so yes.
You run Docker or LXD containers. Containers share the host's kernel, and the write-up warns "it takes down not only the web server application but the entire host". One crash and every site on that box goes with it.
You're on shared hosting or a managed platform. It's your host's kernel, not yours. It's fair to ask them when they'll patch.
Where the fixes are
Debian 13 is fixed in kernel 6.12.117-1, out since 13 November.
Debian 12 is where it gets awkward. The NoisyTCP fix arrived in 6.1.192-1, the same kernel I wrote about yesterday because it was freezing some servers. The good news is that Debian has since marked the freeze fixed in 6.1.192-2, and the security tracker lists 6.1.192-2 as the current Debian 12 security kernel.
Ubuntu lists every supported release, including 24.04 and 22.04, as "Vulnerable" with no fixed version yet.
Red Hat has no fixed package listed yet either. I'd expect AlmaLinux and Rocky to follow once it does.
What to do this week
1. On Debian, update and reboot. Run sudo apt update && sudo apt full-upgrade, then check the version with dpkg -l 'linux-image-6.1.0-5*' on Debian 12. You want 6.1.192-2, not 6.1.192-1. If you set GRUB to boot 6.1.0-54 or put a hold on the kernel after yesterday's post, undo that once -2 is installed. Take a snapshot first, as always.
2. On Ubuntu or Red Hat, watch for the update. When it lands, install it and reboot. Unattended upgrades will usually install a kernel but won't reboot, and the fix doesn't count until you do.
3. Make sure a crash doesn't become an outage. This is my advice rather than the researcher's. Run sysctl kernel.panic. If it says 0, a panicked server just sits there until someone notices. Setting kernel.panic = 10 in a file under /etc/sysctl.d/ makes it reboot itself after ten seconds. It doesn't fix anything, but it turns hours of downtime into a minute or so.
4. Know about the workarounds, but think before using them. The write-up offers two. One is turning off SYN cookies with net.ipv4.tcp_syncookies=0, which in containers has to be set in each one. The catch is that SYN cookies protect you from a different kind of flood attack, so you're swapping one risk for another. The other is cutting the network card to a single receive queue and turning off RPS, which can slow a busy server down. For most small business servers I'd rather patch quickly than do either, but it depends on how much a crash would cost you.
The bigger point
The fix went into the kernel on 3 November and the details went public twelve days later. That gap is shrinking. Dev Partners wrote today about AI turning security patches into working attacks, and it's the same lesson: the time between "fix released" and "someone's using it" is getting shorter.
This week also shows why "just update" isn't the whole answer. On Debian 12 the kernel with this fix was also the one freezing servers. You need updates and a way back, every time.
If a bespoke app runs on that server and you're not sure what an update or reboot might break, the team at Dev Partners look after existing software, including the servers underneath it.
And if you'd rather someone else read the kernel bug reports and did the reboots, that's what my support retainers are for.
Liked that? Get the next one by email
When a security flaw lands that could affect your server, I'll email you what it means and what to do. Plus the odd opinion piece. Not often, no spam.