blog

nginx has two critical HTTP/2 flaws. Ubuntu is fixed, Debian isn't yet. Here's the stopgap

nginx released fixes for ten security flaws yesterday, and two of them are serious. Both live in its HTTP/2 code, and the worst one works on a fairly ordinary setup.

If your website runs on nginx, this one needs doing this week, not this month. Here's how to check, where the fixes are, and what to do if your Linux packages haven't caught up.

What happened

On 18 November the nginx project released versions 1.30.7 and 1.31.9. The security advisories page lists ten flaws fixed in them. Wiz, one of several research teams that reported the bugs, has published the details and scores the two HTTP/2 ones 9.2 out of 10.

CVE-2026-123007 is a counting mistake in how nginx tracks how much data it's allowed to send over an HTTP/2 connection. Wiz says it's exploitable "in default configurations with HTTP/2 enabled", and that it dates back to nginx 1.9.5 in 2015. Send the right requests and the web server crashes, or in some setups the attacker gets to run their own code.

CVE-2026-123633 is a memory bug that needs more lining up. You have to be using nginx with HTTP/2 as a reverse proxy (passing requests on to an app behind it), and the server has to be running low on free connections. The nginx team rates this one medium.

The other eight (mail proxy, HTTP/3 and so on) get fixed by the same update.

The Register quotes F5, which owns nginx, saying "we have found no evidence of active exploitation". The less good news, according to The Register, is that Rapid7 has already published proof-of-concept attack code for CVE-2026-123007. So I'd assume people are trying it.

Does it affect you?

You're on shared hosting or a managed platform. Your host runs the web server, so it's theirs to patch. It's fair to ask them when.

You run your own server or a virtual one. First, check you've got nginx:

nginx -v

Then see whether HTTP/2 is switched on:

sudo nginx -T | grep -n http2

If you see http2 on; or a listen line ending in http2, you're in scope for the serious one. If nothing comes back, your risk is much lower, but I'd still update.

You use nginx in Docker. Each container has its own nginx, so check those too. Your server's updates won't touch them.

Where the fixes are

If you installed nginx from nginx.org's own repository, 1.30.7 and 1.31.9 are there now.

If it came from your Linux distribution, the version number will look much older. That's normal. Distributions copy fixes into their own builds rather than moving to the latest release, so the number on its own tells you nothing.

Ubuntu has fixed packages out, in USN-9159-1. On 24.04 you want 1.24.0-2ubuntu7.20 or later, and on 22.04 1.18.0-6ubuntu14.23 or later. Ubuntu rates it High.

Debian isn't there yet. When I checked the security tracker this morning, Debian 12 and 13 were both still marked vulnerable, with only the development version fixed.

Red Hat lists the flaw, but I couldn't see a fixed package for RHEL 8, 9 or 10 on that page. AlmaLinux and Rocky usually follow once Red Hat ships.

What to do this week

1. If there's a fix, install it. On Ubuntu that's sudo apt update && sudo apt install --only-upgrade nginx. Then sudo nginx -t to check the config and sudo systemctl reload nginx. Check with apt-cache policy nginx that you've got the fixed version.

2. If there isn't, switch HTTP/2 off for now. This is the workaround Wiz and Red Hat both give. On nginx 1.25.1 or newer, add http2 off; to each affected server block. On older versions, take the word http2 off the end of your listen lines. Then test and reload as above.

Visitors' browsers just fall back to HTTP/1.1. Pages with lots of images might feel a touch slower, and that's it.

3. Check it took. Wiz advises confirming old worker processes have gone after the reload. ps -o pid,lstart,args -C nginx should show workers that started after your reload. A reload leaves old workers running until their connections finish, and those are still running the old code.

4. Turn HTTP/2 back on after patching. Put a note in your diary now. Otherwise it'll stay off for years and nobody will remember why.

5. Watch the error log. If /var/log/nginx/error.log starts showing workers exiting with signal 11 and you haven't changed anything, treat it as worth a closer look.

The bigger point

Wiz says it used AI to find these bugs, and attack code was public on the day the fix came out. Dev Partners wrote this week about AI turning security patches into working attacks, and this is that happening in practice. The gap between "fixed" and "being tried on your server" is now days, not months.

Last month's nginx fixes didn't need heroics. This one still doesn't, but it does need someone to look today rather than whenever.

If nginx sits in front of a bespoke app and you're not sure what switching HTTP/2 off might do to it, the team at Dev Partners look after existing software and can check.

And if you'd rather someone else watched the Debian tracker and did the reloads, that's what my support retainers are for.