The Trust Layer Is Buckling: Quad9, eMudhra, and a PHP Embargo That Broke Early
This article examines a broken PHP RCE embargo alongside Quad9's exit from France and eMudhra's TLS shutdown, and explains practical DNS and certificate mitigations for operators.
The Trust Layer Is Buckling: Quad9, eMudhra, and a PHP Embargo That Broke Early
Someone broke the embargo on a PHP remote code execution bug and then confessed to it on Hacker News under the title "Tell HN: I unknowingly broke the PHP RCE embargo. I am sorry." I believe them. The apology was sincere. What stuck with me was the reaction underneath it, which was mostly a shrug from people who have watched coordinated disclosure quietly turn into a coin flip.
Three things landed in my feeds the same week, and they only look unrelated if you squint. A recursive DNS resolver walked away from an entire country. A certificate authority with roots in every major trust store announced it's leaving the global TLS business. And a security embargo collapsed before the patch was even out. Different layers, same story: the shared middle of internet trust is no longer a constant you can build on. If you run servers, that changes your weekend.
Quad9 left France, and your resolver is a policy decision now
Quad9 is the Swiss non-profit behind 9.9.9.9. Its pitch has always been narrow: block malware, resolve everything else, log almost nothing. What it doesn't do is enforce copyright filtering. France wanted it to, and Quad9 chose to leave the country rather than become a filtering chokepoint for a jurisdiction it doesn't operate in.
You can argue about whether that's principled or theatrical. What you can't argue about is the technical consequence: a resolver's location now determines which legal regimes can shape the answers it gives you. That's a meaningful change from the old model, where DNS was just a dumb lookup service you set once and forgot.
Most people I know still have their router handing out the ISP's resolvers over plaintext port 53. That's the worst of all worlds. Your ISP can transparently redirect port 53 to their own resolver, so you're not even getting the DNS you thought you configured. And whoever answers your queries sees the full map of everything your network talks to.
Two things worth doing:
- Run your own recursive resolver. Unbound on a small box is enough for a household, and with DNSSEC validation on, you stop trusting the path between you and the answer.
# Debian/Ubuntu
sudo apt install unbound
sudo tee /etc/unbound/unbound.conf.d/local.conf >/dev/null <<'EOF'
server:
interface: 0.0.0.0
access-control: 192.168.0.0/16 allow
access-control: 10.0.0.0/8 allow
hide-identity: yes
hide-version: yes
qname-minimisation: yes
auto-trust-anchor-file: "/var/lib/unbound/root.key"
EOF
sudo unbound-checkconf && sudo systemctl restart unbound
- If you'd rather use a public resolver, use DoT or DoH. Not because encryption makes the operator trustworthy, but because it makes your ISP's transparent hijack stop working. Pick the operator deliberately, based on where they're incorporated, not on benchmark latency.
If you want your resolver choice to be a deliberate part of your stack rather than a default you inherited, it belongs in the same conversation as the rest of your network build. I wrote about that in Build a Practical Homelab That Actually Works and the reasoning hasn't changed.
eMudhra quitting TLS is a different kind of problem
A DNS resolver leaving a country is a policy story. A certificate authority leaving the business is an operational one, and it has your name on it if any of your endpoints chain up to an emSign root.
eMudhra announced it's exiting the global SSL/TLS certificate business entirely. Existing certificates don't become invalid the moment the press release goes out. That's the part people get wrong. Certificates keep working until one of two things happens: the root gets removed from browser and OS trust stores, or the CA's OCSP and CRL endpoints stop answering.
That second one bites harder than people expect. If you're using OCSP stapling with hard-fail, or you've got ssl_stapling_verify on in nginx without a fallback, a CA that stops serving revocation data can take your site down. Not because the cert is distrusted, but because verification can't complete.
So the first thing to do is figure out where you actually stand. Here's a script I keep around for exactly this. Drop your hostnames in a file, one per line:
#!/usr/bin/env bash
# tls-audit.sh — issuer, expiry, days-left for every TLS endpoint in hosts.txt
set -euo pipefail
while read -r host; do
[ -z "$host" ] && continue
info=$(echo | timeout 5 openssl s_client -connect "$host:443" \
-servername "$host" 2>/dev/null \
| openssl x509 -noout -issuer -enddate 2>/dev/null) || true
if [ -z "$info" ]; then
printf '%-32s %s\n' "$host" 'unreachable / no cert'
continue
fi
issuer=$(echo "$info" | sed -n 's/^issuer=//p' | sed 's/^.*CN *= *//')
expiry=$(echo "$info" | sed -n 's/^notAfter=//p')
days=$(( ( $(date -d "$expiry" +%s) - $(date +%s) ) / 86400 ))
printf '%-32s %-42s %4s days (%s)\n' "$host" "$issuer" "$days" "$expiry"
done < "${1:-hosts.txt}"
Run it against everything, including the stuff you forgot about. The output gives you two lists that matter: hosts with an issuer you don't recognize, and hosts under 30 days to expiry. Both are the same fix, which is reissuing from somewhere else, but the second list is the one that wakes you up at 3am.
Caveat: the CN parsing is crude. Issuer strings vary between OpenSSL versions and some CAs stuff the whole RDN sequence in there. It's a triage tool, not an inventory system.
Where to move the certs
| Option | Good for | Automation | Browser-trusted | When the CA walks away |
|---|---|---|---|---|
| Let's Encrypt (ACME) | Public HTTPS | Full, 90-day renewals, certbot/acme.sh | Yes | Swap the ACME directory URL; ~5 minutes |
| Commercial CA (DigiCert, Sectigo, etc.) | EV certs, contract-driven procurement | API exists, usually clunkier | Yes | Reissue plus root-removal risk; weeks of notice |
| Internal CA (step-ca, smallstep) | Internal services, mTLS, homelab | Full, short-lived certs | No, by design | Nothing. You are the CA |
| Self-signed + manual trust | Appliances, lab boxes, one-offs | None | No | Nothing, but you'll resent it every time |
The line I'd draw: anything browser-facing should be on ACME automation, and everything internal should be on an internal CA. The commercial middle tier is where the emMudhra situation hurts, and it's also where lifetimes are shrinking anyway. Maximum certificate lifetimes have been ratcheting down for years and that trend isn't reversing. A 90-day cert that renews itself is now strictly less work than a two-year cert you have to remember.
For internal stuff, step-ca gets you a working ACME endpoint in about ten minutes:
step ca init --name "homelab" --dns ca.lan --address :9000 --deploy
step ca provisioner add acme --type ACME
step ca certificate svc.lan svc.crt svc.key --provisioner acme
Wire that into your automation and CA exits stop being a category of problem for you. I went through how I sequence this kind of thing in Practical Self-Hosting: A Lean DevOps Homelab Playbook.
The embargo problem is worse than the bug
Back to PHP. I don't have the full timeline and I'd read the thread yourself rather than take my summary. The pattern is what matters: someone preparing a fix or a report exposed the issue before the coordinated release, and the disclosure window collapsed. Whether it was a public PR, a commit that got pushed early, or an AI agent cheerfully drafting a patch description it shouldn't have is almost beside the point. The mechanism is now fragile in a way it wasn't five years ago.
Part of that is tooling. When you can hand a patch diff to a model and ask it to work backwards to the vulnerable code path, the practical gap between "patch available" and "exploit available" shrinks to hours. That's a real shift in what an embargo protects, and I don't think the disclosure process has caught up with it. I've written about the developer-side version of this in The Rise of Local Helpers: Why Stacked PRs and Local AI Are Reshaping Developer Productivity — it's genuinely useful tech, and it cuts both ways.
The practical takeaway is not "coordinate better." It's "assume the patch will be public before you're ready." That means:
- Patch on availability, not on advisory. If your distro ships a security update, take it. Waiting for the write-up is waiting for the exploit.
- Shorten your exposure window. A PHP-FPM process that can't reach the network, can't write outside
/var/www, and runs as a user with no shell is a much less fun target than root-on-shared-hosting. - Watch the right feeds.
oss-security, your distro's security tracker, and the upstream project's announcements page. Not Twitter. - Sandbox anything that executes code on your behalf. This is the same reason NVIDIA's OpenShell release is interesting. The moment an autonomous agent has shell access, it's a privileged process with a non-deterministic instruction set. Treat it like one: separate user, no credentials it doesn't need, no path to your internal network.
On that last point, I'll cop to a bias. I'm skeptical of running agents with real credentials in a homelab, and I think most people doing it are one prompt injection away from a very bad afternoon.
What I'd actually do this week
- Run the TLS audit script. Fix anything under 30 days, and note any issuer you don't recognize.
- Check whether your web servers do hard-fail OCSP or
ssl_stapling_verifyagainst a CA that might be winding down. Add a fallback or turn on soft-fail. - Turn on unattended security upgrades for the packages that matter, and make sure you're actually rebooting.
sudo apt install unattended-upgrades
sudo dpkg-reconfigure -plow unattended-upgrades
# /etc/apt/apt.conf.d/50unattended-upgrades
# Unattended-Upgrade::Automatic-Reboot "true";
# Unattended-Upgrade::Automatic-Reboot-Time "03:30";
A kernel CVE you've downloaded but not booted is not mitigated. I've lost this argument with myself more than once.
- If you don't have an internal CA, stand one up. It's the single change that makes every future "CA X is exiting the business" headline irrelevant to your internal services.
- Move your DNS off the ISP default. Self-hosted Unbound or DoT to an operator you picked on purpose.
None of this is exotic. It's just the boring hygiene that stops mattering until it suddenly, expensively matters. The difference now is that the middle layer of internet trust is visibly moving under everyone's feet, and the operators who've already stopped depending on it are the ones who won't be writing an apology post of their own.
Recommended tools & services
Disclosure: some links below are affiliate links. If you buy through them I may earn a commission at no extra cost to you. I only list tools I actually run.
| Tool | Why it is here | |
|---|---|---|
| Amazon - Managed Switches | VLAN-capable switches for a segmented homelab network | Visit |
| Amazon - Ethernet Cables | Cat6/Cat6a patch cables - buy more than you think you need | Visit |
| Amazon - Wi-Fi Access Points | Proper APs beat ISP combo routers for coverage | Visit |