The Fix Only Existed in kernel6.12 — Switching Kernel Streams on Amazon Linux 2023
What happened This was during the week of the CVSS 9.8 kernel TCP vulnerability (CVE-2026-98432). When I checked the Amazon Linux Security Center for the production Amazon Linux 2023 server I manage, I found an unusual
What happened
This was during the week of the CVSS 9.8 kernel TCP vulnerability (CVE-2026-98432). When I checked the Amazon Linux Security Center for the production Amazon Linux 2023 server I manage, I found an unusual situation:
- The default
kernelpackage (6.1 series) — Pending Fix - A separate package,
kernel6.12(6.12 series) — a fixed version with its ALAS (security advisory) already published
Within the same OS, whether a fix existed or not depended on the kernel stream. AL2023 ships a newer kernel stream as a separately named package, kernel6.12, alongside the default 6.1 series. Instead of waiting for the 6.1 fix, switching streams meant I could patch today. So that's what I did.
What I did — four moves
# 1. Install the 6.12 kernel (coexists with the 6.1 series)
sudo dnf install kernel6.12 kernel6.12-modules-extra
# 2. Make 6.12 the kernel for the next boot
sudo grubby --set-default /boot/vmlinuz-6.12.*.amzn2023.x86_64
# 3. Declare that future kernel updates should keep 6.12 as default
# /etc/sysconfig/kernel
# UPDATEDEFAULT=yes
# DEFAULTKERNEL=kernel6.12
# 4. Reboot to apply
sudo reboot
dnf install doesn't replace the existing 6.1 kernel — it puts 6.12 next to it in /boot. Which one boots is decided by the grub default, which is why step 2 explicitly switches it with grubby.
Pitfalls
1. dnf update keeps you on 6.1 forever
This is the biggest trap. kernel and kernel6.12 are different package names, so no amount of dnf update will move you from 6.1 to 6.12 — it only takes you to the latest 6.1. That's why "I keep updating but the Security Center warning won't go away." A stream switch only happens through an explicit dnf install kernel6.12.
2. Without /etc/sysconfig/kernel, you'll eventually boot 6.1 again
Stopping at grubby --set-default leaves a trap armed. When the next kernel update lands, whether the new kernel becomes the default is decided by DEFAULTKERNEL in /etc/sysconfig/kernel. If that still says kernel (the 6.1 series), the default can flip back to 6.1 on an update.
# /etc/sysconfig/kernel
UPDATEDEFAULT=yes
DEFAULTKERNEL=kernel6.12
"I switched today" and "I'll keep tracking the 6.12 series" are two different settings.
3. I kept 6.1 installed as a fallback — and paid for it
I didn't uninstall the 6.1 series: it's the escape hatch if 6.12 fails to boot. That was the right call as insurance, but it has a cost. Security updates keep arriving for the unused 6.1 series, so dnf needs-restarting -r now warns "kernel updated, reboot required" — about a stream that isn't even running. Before trusting that warning, I check uname -r against grubby --default-kernel to see whether the running stream was actually updated.
4. Production reboots follow a fixed template
Production reboots always run in the same order: maintenance page ON → reboot → verify the real pages via curl with a bypass cookie → OFF. Since verification happens before reopening, users never see the 502s right after boot.
Verification
Not straight to production: I ran the whole thing on a dev server with the same setup first (about 1 minute). On production, measured by the maintenance page, downtime was 21 seconds.
Three checks afterward:
# Is the running kernel the 6.12 series?
uname -r
# → 6.12.x-x.amzn2023.x86_64
# Is the relevant ALAS now shown as installed?
sudo dnf updateinfo list --installed | grep -i <ALAS ID>
# Does the public page return 200?
curl -s -o /dev/null -w '%{http_code}\n' https://example.com/
Takeaways
- "No patch yet" is not a statement about the whole distro — read it per kernel stream. On AL2023,
kernel(6.1) was Pending Fix whilekernel6.12already had the fix - A stream switch never happens via
dnf update. It's four moves:dnf install kernel6.12 kernel6.12-modules-extra→grubby→/etc/sysconfig/kernel→ reboot - If you keep the old stream as insurance, accept the false "reboot required" warnings that come with it
The same week, an Ubuntu server I was also watching had no fix published for its kernel stream — nothing to do but wait. The AL2023 server, by switching streams, went from alert to applied fix within the day. Half of vulnerability response is locating which row in the security tracker's table is yours. A different row means different options and a different deadline.
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.