Gave my AI this thread to deal with things, and it me asked to post this (seems I work for my AI now):
Notes from mitigating 3 Cloudron 10.1.6 / Ubuntu 26.04 servers, in case they help:
The sed for http2 on isn't enough on its own. Some app configs (here: Collabora, Woodpecker, a Matrix admin UI) use the older listen 443 ssl http2; form. I believe that keeps HTTP/2 on for every site on that port. Check with: grep -rlE 'listen .http2' /etc/nginx/applications. To strip it: sed -i -E '/^\slisten\s/ s/\s+http2(\s*;|\s)/\1/'.
Cloudron rewrites these configs. About 2 minutes after a reboot, some regenerated configs had http2 on again (I saw HTTP/2 200 from outside). App reconfigures and certificate renewals probably do the same. I run a small script from /etc/cron.d every 2 minutes. It reapplies both edits only when needed, backs up /etc/nginx, runs nginx -t (and rolls back if it fails), then reloads.
Check from outside rather than trusting the files: curl -sI --http2 --resolve your.domain:443:SERVER_IP https://your.domain/ | head -1 should show HTTP/1.1.
SCTP is in use on Cloudron: the built-in TURN service (turn container, coturn) had many kernel SCTP sockets open, so modprobe -r can't unload it until a reboot. After blocking it and rebooting, TURN came back with its normal UDP/TCP listeners and no SCTP errors, so the block looks safe. Check your call/WebRTC apps afterwards.
Other advisories: the stock configs have no mail {} block, no quic/http3 and no resolver. So only the two HTTP/2 bugs (CVE-2026-123007 and -123633) seem relevant to Cloudron's reverse proxy.
Reboots stop hand-made containers: containers you started yourself with docker run and no restart policy won't come back after the reboot.