惯性聚合 高效追踪和阅读你感兴趣的博客、新闻、科技资讯
阅读原文 在惯性聚合中打开

推荐订阅源

Google DeepMind News
Google DeepMind News
G
Google Developers Blog
博客园 - 三生石上(FineUI控件)
B
Blog
让小产品的独立变现更简单 - ezindie.com
让小产品的独立变现更简单 - ezindie.com
Hugging Face - Blog
Hugging Face - Blog
C
Check Point Blog
V
V2EX
Vercel News
Vercel News
U
Unit 42
Recent Announcements
Recent Announcements
Last Week in AI
Last Week in AI
J
Java Code Geeks
WordPress大学
WordPress大学
罗磊的独立博客
I
InfoQ
阮一峰的网络日志
阮一峰的网络日志
奇客Solidot–传递最新科技情报
奇客Solidot–传递最新科技情报
小众软件
小众软件
M
MIT News - Artificial intelligence
freeCodeCamp Programming Tutorials: Python, JavaScript, Git & More
Apple Machine Learning Research
Apple Machine Learning Research
Martin Fowler
Martin Fowler
云风的 BLOG
云风的 BLOG

DEV Community

Authentication Security Deep Dive: From Brute Force to Salted Hashing (With Java Examples) Why AI Systems Don’t Fail — They Drift Spilling beans for how i learn for exam😁"Reinforcement Learning Cheat Sheet" I Replaced Chrome with Safari for AI Browser Automation. Here's What Broke (and What Finally Worked) How Python Borrows Other People's Work The $40 Architecture: Processing 1 Billion API Requests with 99.99% Uptime Vibe Coding: A Workflow Guide (From Zero to SaaS) Most webhook security guides protect the wrong side. The scary part is delivery. Headless CMS for TanStack Start: Build a Blog with Cosmic EU Age Verification App "Hacked in 2 Minutes" — What Actually Happened Comfy Cloud’s delete function does not actually remove files Running AI Models on GPU Cloud Servers: A Beginner Guide Event-driven media intelligence with AWS Step Functions and Bedrock I scored 500 AI prompts across 8 quality dimensions — here's what broke How to Call Google Gemini API from Next.js (Free Tier, No Backend Needed) The Portal Protocol: Reclaiming Human Connection in the Age of AI How to Fix Your Team's Scattered Knowledge Problem With a Self-Hosted Forum Intro to tc Cloud Functors: A Graph-First Mental Model for the Modern Cloud Designing Multi-Tenant Backends With Both Ownership and Team Access I Built a Neumorphic CSS Library with 77+ Components — Here's What I Learned PostgreSQL Performance Optimization: Why Connection Pooling Is Critical at Scale Cómo construí un SaaS multi-rubro para gestionar expensas en Argentina con FastAPI + Vue 3 🚀 I Built an Ethical Hacking Scanner Tool – Open Source Project I Replaced /usage and /context in Claude Code With a Single Statusline A Pythonic Way to Handle Emails (IMAP/SMTP) with Auto-Discovery and AI-Ready Design I Collected 8.9 Million Polymarket Price Points — Here's What I Found About How Markets Really Move EcoTrack AI — Carbon Footprint Tracker & Dashboard Everyone's Using AI. No One Agrees How. 5 self-hosted ebook managers worth trying in 2026 Building Your First AI Agent with LangChain: From Chatbot to Autonomous Assistant
How Myanmar Blocks Tailscale — and How to Beat It
mariatanbobo · 2026-06-14 · via DEV Community

A government blocks a VPN with a one-line SNI rule. The fix is a custom relay on port 443. Tailscale could make this trivial for millions — but they haven't.

There's a lot of confusion about how Myanmar actually blocks Tailscale. Some say it's DNS poisoning. Others claim the coordination server is blackholed. A few insist the WireGuard protocol itself is detected and dropped.

None of that is correct. The block is simpler and stupider than most people think — and because of that, the counter is simpler too. This matters because Tailscale is genuinely important networking middleware. It's used by journalists, remote workers, distributed teams, and anyone who needs secure machine-to-machine connectivity. Blocking it isn't just censorship theater — it disrupts legitimate infrastructure.

This time, I worked on the problem with the support of a capable agentic AI. I trained its substantial capacity for research and systematic debugging on the task, and together we burned through the misconceptions, tested the actual failure points, and built a working counter. What follows is what we found.

What Myanmar Actually Blocks

Myanmar operates deep packet inspection (DPI) at the ISP level. But they're not doing anything sophisticated. They're running what amounts to a single SNI filter:

Block TLS ClientHello where SNI matches *.tailscale.com

That's it. One wildcard rule.

This hits Tailscale in three places:

Component Blocked? Why
Coordination server (controlplane.tailscale.com) No Different SNI, survived past block waves
Default DERP relays (derpN.tailscale.com) Yes All match the wildcard
Direct WireGuard (UDP 41641) Sometimes Symmetric NAT without relay = dead

When all DERP relays are unreachable, nodes behind carrier-grade NAT in Myanmar have no path to each other. The mesh collapses. Every node is an island.

The cruel part: the coordination server still works. The client can see its peers. It knows they exist. It just can't reach them. It's like being locked in a glass box — you can see everyone, but you can't touch them.

The agent and I verified this step by step: DNS resolution from inside Myanmar, successful — the IPs resolve fine. TCP handshake to the coordination server, successful — it's not IP-blocked. TLS ClientHello to derpN.tailscale.com, dropped at the SNI. TLS ClientHello to a custom domain on the same VPS, passed cleanly. The filter is exactly one rule deep.

What Doesn't Work

Peer Relays (NAT-PMP/PCP). Tailscale's own documentation suggests custom DERP isn't needed if you set up a peer relay. But peer relays use raw UDP on arbitrary ports. DPI boxes flag non-standard UDP instantly. Port 40000 looks nothing like web traffic.

Waiting for it to get better. Myanmar's filtering isn't going away. It's getting more aggressive, not less.

Commercial VPNs. Most are blocked at the same DPI layer. The ones that work today won't work tomorrow.

What Works: Your Own DERP on Port 443

The insight is simple: TLS on port 443 looks like HTTPS to a DPI box. Every website uses it. Blocking it would break the internet.

A custom DERP relay listening on TCP 443, with a valid Let's Encrypt certificate on a domain you control, is indistinguishable from a web server. The SNI matches your domain, not *.tailscale.com. The traffic is standard TLS. The DPI box shrugs and passes it through.

You can deploy this in 30 minutes:

  1. Run cmd/derper on a VPS outside the censored country
  2. Give it a Let's Encrypt certificate for a subdomain you control
  3. Tell Tailscale to use it

But here's where Tailscale's product decision bites you.

The Problem Tailscale Won't Let You Solve

You can add custom DERPs to your tailnet. But you cannot remove the default ones.

This isn't a technical limitation. Tailscale's admin console simply doesn't expose DERP controls. The ACL syntax has some undocumented, CLI-only support for DERP filtering — but it's fragile, barely documented, and not something a normal user would discover. The product team made a choice: DERP is infrastructure, not configuration. You don't get to touch it.

The consequence: your client will try the blocked default DERPs first. Each attempt times out after 5-10 seconds. Only after cycling through every blocked relay does it fall back to your custom one.

The result: Tailscale does connect eventually. But every connection attempt has a 20-40 second penalty. Every reconnect. Every network change. Every time your phone switches from WiFi to cellular.

It's functional but miserable — like a car that stalls three times before starting.

The Real Fix: Headscale

Headscale is the open-source implementation of the Tailscale coordination server. Self-hosting it gives you one thing Tailscale's SaaS doesn't: control over the DERP map.

With Headscale, you can:

  • Add your custom DERP relay
  • Remove every default Tailscale DERP
  • Serve a DERP map with exactly one entry: your relay on port 443

No timeouts. No cycling through blocked relays. Your custom DERP is the only option, so clients go there immediately.

The deployment takes an afternoon:

VPS in Singapore
├── nginx :443 (SNI router)
│   ├── derp.yourdomain.com → derper container
│   └── hs.yourdomain.com → Caddy → Headscale
├── Headscale (coordination server)
├── Custom DERP relay (port 443, LE cert)
└── Headplane (web UI for management)

On the client side, joining is one command:

tailscale up --login-server=https://hs.yourdomain.com --authkey=YOUR_KEY

Add a Tailscale node as an exit node on the Singapore VPS, and every device on the tailnet can route its internet traffic through Singapore — free of Myanmar's filtering.

Total cost: one $5/month VPS.

A note on availability

If you're reading this and planning to deploy Headscale, consider forking or mirroring the repository before you need it. The *.tailscale.com wildcard block works because it's easy. There's nothing stopping the same filter from being extended to github.com/juanfont/headscale — and after this article, that's a real possibility. Install from an alternate source. Host the binaries on your own domain. The pattern you use to beat the DERP block is the same pattern that keeps the tools themselves available.

What I'm Asking Tailscale to Do

Tailscale's engineering is excellent. The product decisions around DERP management are the problem.

Three changes would make Tailscale censorship-resistant for millions of people:

1. Let users remove default DERPs from the admin console

This is the single highest-impact change. Right now the admin console has no DERP controls at all. Adding a "DERP relays" section where users can disable defaults and add customs would solve the timeout problem without self-hosting anything.

The ACL syntax already partially supports this — but it's undocumented, CLI-only, and fragile. Make it a first-class feature.

2. Ship a one-click "censorship mode"

One toggle that:

  • Disables all default DERPs
  • Requires at least one custom DERP on port 443
  • Sets aggressive timeouts so blocked relays don't stall connections

This isn't hypothetical. Iran, China, Russia, Turkey, and Myanmar all block Tailscale infrastructure. That's hundreds of millions of people who can't use the product because of a single wildcard SNI rule.

3. Document the DPI countermeasures

Tailscale's documentation on censorship circumvention is scattered across forum posts and GitHub issues. A single page — "Using Tailscale in Censored Networks" — would tell users what they need before they spend hours debugging timeouts.

Lessons

  1. DPI is lazy. Myanmar's entire Tailscale block is one SNI wildcard. Don't assume sophisticated adversaries — they're doing the minimum that works.
  2. Port 443 is the universal blind spot. Every censorship system has to let HTTPS through. Put your tunnel traffic on 443 with a valid TLS cert and you're invisible.
  3. Headscale isn't just for homelabs. The ability to control the DERP map is the difference between "barely functional" and "instant connection." For censored networks, it's not a luxury — it's the whole point.
  4. Tailscale's defaults are a single point of failure. *.tailscale.com is a convenient wildcard for DPI boxes. Custom domains break that pattern.
  5. Exit nodes complete the picture. A relay gets you connectivity. An exit node gets you out.
  6. Test before you trust. The coordination server at controlplane.tailscale.com was reachable from Myanmar when we tested. This can change. Self-hosting Headscale removes the last dependency on tailscale.com.
  7. The gap between "works" and "works well" is 30 seconds. Without DERP map control, every connection has a built-in delay. That delay is the difference between a tool people use and a tool people abandon.
  8. Mirror before you need it. The publication of this article may accelerate blocking of the Headscale repository. Fork it. Host the binaries yourself. Your infrastructure should not depend on a GitHub URL surviving a government filter.

Written with Hermes Agent. Follow me on X: @MariaTanBoBo