Chameth.com - posts like break-everything-fuzz-testing, docker-automatic-nginx-proxy but not debugging-beyond-the-debugger, finding-an-awkward-bug-with-claude-code, intro-to-containers, why-you-should-be-using-https

An interesting Tailscale + Docker gotcha

2026-01-29T00:00:00Z

As I’ve written about before, I use Tailscale for a lot of things. I thought I had it set up in a reasonably secure manner, but I recently noticed a problem.

I use Tailscale’s ACLs to limit what each node can access, based on the tags I apply to it. So an app node can’t access anything via Tailscale, while an integration or server node can access things tagged with either app or integration. This is expressed pretty simply in the Tailscale ACL JSON:

{
	"tagOwners": {
		// Servers that can be SSH'd into
		"tag:server": [],
		// Applications that are exposed on tailscale but never connect out
		"tag:app": [],
		// Things which talk to other services over tailscale (connecting to apps etc)
		"tag:integration": [],
	},
	"grants": [
		{
			// Users can access everything
			"src": ["autogroup:member"],
			"dst": ["*"],
			"ip":  ["*"],
		},
		{
			// Servers and integrations can access integrations and apps
			"src": ["tag:server", "tag:integration"],
			"dst": ["tag:integration", "tag:app"],
			"ip":  ["*"],
		},
	],
}

Over the past week I’ve been setting up a private Forgejo instance behind Tailscale, complete with an actions runner that runs things using a docker-in-docker container. I didn’t want the runner knowing anything about Tailscale, so I had it configured to speak to Forgejo direct over HTTP (forgejo:3000) instead of using full Tailscale HTTPS URL that I use when accessing it (http://git.example-net.ts.net/)¹.

Everything was going fine, until I forgot to do that translation… and it worked. My Forgejo action runners could access anything on my tailnet. I run several things on Tailscale that just have authentication turned off, on the basis that only authorised devices can access them. Things like the admin interface for this website. I definitely didn’t intend for any workflow I run on my git server to have access to edit my website!

This wasn’t just limited to Forgejo, either. Any docker container I was running could access the tailnet. In hindsight it’s fairly obvious why: the host is running Tailscale, connected as a node tagged with my server tag. That creates a tailscale0 interface, and automagically sets up iptables rules to route Tailscale traffic over the interface. Docker also automagically sets up iptables rules to bridge traffic, and apparently these two sets of rules interact in such a way that traffic from Docker containers is allowed to route via the tailscale0 interface.

I say it’s fairly obvious in hindsight — there’s no reason why Docker would special case any particular host interface after all — but it still feels pretty surprising. Because both bits of software inject their own iptables rules, I never really had a good mental model for how they interact. The host Tailscale node was a completely separate building block to Docker. It would be a pain to use either of them if they didn’t do these rules, but it’s also one of the reasons I don’t really like “magical” things².

There are lots of ways to fix this, but none of them feel particularly great. You can configure both Tailscale and Docker to not automatically fiddle with iptables and handle the rules yourself, but I really hate dealing with iptables³. As a stopgap I did hold my nose and add some iptables rules to drop traffic to the tailscale0 interface if it originated from the IP ranges that Docker was configured to use:

iptables -I DOCKER-USER -s 192.168.0.0/16 -o tailscale0 -j DROP
iptables -I DOCKER-USER -s 172.17.0.0/12 -o tailscale0 -j DROP

The DOCKER-USER chain is a nice little escape hatch; it comes before the main auto-generated DOCKER chain, and Docker leaves the rules in it alone.

So I added these rules, and felt pretty good about myself, and then… everything started breaking in weird ways. After some debugging I realised the problem was DNS⁴. The server’s DNS resolver is 100.100.100.100, a special Tailscale address. This is so that it can resolve tailnet hostnames via MagicDNS (and so I don’t have to configure my custom DNS servers manually on each device; Tailscale does it for me). My new iptables rules inadvertently dropped all the DNS packets coming from docker containers. D’oh.

Obviously the solution here is to double down and add MORE iptables rules:

iptables -I DOCKER-USER -s 192.168.0.0/16 -d 100.100.100.100 -o tailscale0 -j ACCEPT
iptables -I DOCKER-USER -s 172.17.0.0/12 -d 100.100.100.100 -o tailscale0 -j ACCEPT

This got DNS working again, but it all feels a bit gross. Not least because to actually make these persist you need to use iptables-save, and then you also get all the rules that both Docker and Tailscale have inserted. I came across a script to filter out the Docker ones, but… yuck⁵.

I realised a cleaner idea might just be to lock down what server tagged nodes could access within Tailscale itself. That way I can avoid touching iptables at all. It’s handy to be able to curl services from the host when debugging, but it’s not really necessary. So I removed that access, and… stuff broke again. I’m using my Forgejo instance as a registry for some of the docker images I run, so the docker daemon needs to be able to reach it. I ended up making a new tag for infrastructure, which can be accessed from server devices. This does still allow all the docker containers to reach Forgejo, but I already have it set up with appropriate access controls and public/private repository splits. Forgejo is a service designed to run publicly, so this seems a reasonable trade-off for convenience. I used the tests feature of Tailscale’s ACL config to make sure I’d got the rules right:

{
	"tests": [
		{
			// Servers can only access infrastructure
			"src":   "tag:server",
			"proto": "tcp",
			"allow": [
				"tag:infrastructure:8080",
			],
			"deny": [
				"100.84.16.43:8080",
				"me@example.com:8080",
				"tag:server:8080",
				"tag:app:8080",
				"tag:integration:8080",
			],
		},
	],
}

The whole situation still feels a bit messy. If I ever get around to switching to nftables I might loop back and manually craft some rules for routing traffic, instead of leaving Tailscale and Docker to do their own thing.

I discovered later on that logging into the container registry at forgejo:3000 actually issued a redirect to https://git.example-net.ts.net/ so this was all basically for naught… ↩︎
I didn’t use Tailscale’s MagicDNS for a long time just because the word “magic” put me off. Only when I eventually got around to learning how it worked, and seeing that it wasn’t really that magical under the hood, did I change my mind. ↩︎
nftables seems far better in lots of ways, but I can’t really be bothered migrating. Maybe next time I reimage the server for whatever reason… ↩︎
It’s always DNS… ↩︎
Again, nftables would almost certainly help here. It actually has (gasp) configuration files. But again, I really didn’t want to spend the time migrating. ↩︎

Docker reverse proxying, redux

2022-12-06T00:00:00Z

Six years ago, I described my system for configuring a reverse proxy for docker containers. It involved six containers including a key-value store and a webserver. Nothing in that system has persisted to this day. Don’t get me wrong – it worked – but there were a lot of rough edges and areas for improvement.

Microservices and their limitations

My goal was to follow the UNIX philosophy of “do one thing and do it well”. Unfortunately, that doesn’t really work when applied to network services that have to interact with one and other. UNIX tools are built upon a common file system and simple data passed over STDIN. Microservices don’t have that shared foundation. You could make one: companies that use microservices in anger often have a team that deals with the “developer experience” of creating and using microservices. But as a solo developer that’s not something I wanted to spend my time doing.

This became very apparent when trying to debug issues. In the UNIX world, if your series of commands piped together does something strange you can simply echo the data at various stages. Not so much when that data is flying around on a network, going into and out of things like etcd. Trying to figure out why a certificate hadn’t been acquired was a case of searching through logs from four containers, none of which had particularly good logging. There are many ways to get insight into what’s happening with microservices but, again, that’s not something I wanted to do myself.

Over time, and with experience in other projects, I came to realise that microservices only really make sense if you’re unable to deploy a monolith. For tech companies this naturally happens when different teams contribute to the same service: splitting it out into smaller services that are wholly owned by individual teams makes sense. For solo developers, that never happens. You can still gain the other benefits of microservices – such as code separation and having clearly defined APIs – by sticking to certain coding standards.

Proxy inconveniences

As well as being unhappy with the microservice nature of the solution, I wasn’t pleased with nginx. If you requested an unknown domain, nginx would use the first server block in its config to serve a response, instead of sending an “unrecognised name” alert as I wanted. It was a minor issue, but it irked me.

So from nginx I switched to haproxy. It has a strict-sni option when configuring TLS connections which makes it behave properly. It also performs a lot better for this type of workload than nginx. All was well for a while, but then I started getting alerts that requests were occasionally failing. I couldn’t reproduce the issue, but my nightly jobs to build and push containers managed to hit it nearly every night, causing them to fail.

After some investigation, I found that the haproxy developers had refactored the header parsing code, and neglected to properly reset flags when multiple requests were sent over the same connection. There was a patch, but it wasn’t released. No problem, I thought, I’ll just cherry-pick it onto the last release… Except that haproxy use Git in the most convoluted manner I’ve ever seen – they have one repository per release. This makes it harder to patch, but it also made me question whether I trusted them to ship stable software: there were no tests for the header parsing code (which is both fundamental and finicky, the perfect target for tests), the source code management was weird, and they didn’t seem in any rush to patch this bug.

Not long after that issue, Greg managed to encounter another bug where haproxy returned a 500 error whenever the upstream server replied with a particular, perfectly valid, header. The die was cast – it was time to move to something else.

Not Invented Here syndrome

Looking for a new solution, there were many more options than back in 2016. I’m still convinced, however, that anything exposed to the Internet should not have access to run docker containers. It’s the modern equivalent of running a CGI script as root. That single requirement eliminates most off-the-shelf solutions. What do you do when nothing quite meets your specific requirements? You make something yourself! My new solution has two components: Dotege and Centauri.

Dotege is a replacement for the microservices that monitored containers and obtained certificates. It’s fundamentally a templating engine - whenever the containers change, it evaluates a template and saves the result to disk. The template has access to details about the containers, their labels, ports, and so on. Dotege can also obtain certificates from Let’s Encrypt, and raise a signal against another process whenever the template or certificates change. I used this to generate the configuration and certificates used by haproxy for a while, and more recently changed the template so that it works for Centauri.

Centauri is my own reverse proxy. It’s configured using a simple text file and can also obtain certificates from an ACME provider. It doesn’t serve static content, has no knowledge about docker, and avoids the other bells and whistles that adorn most reverse proxies. It also has good test coverage to ensure that I don’t, say, accidentally break header parsing.

As a software engineer I enjoy writing software, but I also enjoy running simple, easy to understand software. That’s what I’ve achieved here: it’s very easy to identify where the problem is if anything goes wrong, both are small Go programs rather than vast sprawling C monstrosities, and their interaction is primarily through a file written to disk that can be inspected or edited as needed.

How to break everything by fuzz testing

2020-04-26T00:00:00Z

Fuzz testing is a bit like the infinite monkey theorem, but instead of Shakespeare you get crashes.

Fuzz testing, if you’re not aware, is a form of testing that uses procedurally generated random inputs to see how a program behaves. For instance, if you were fuzz testing a web page renderer you might generate a bunch of HTML - some valid, and some not - and make sure the rendering process didn’t unexpectedly crash.

Fuzz testing doesn’t readily lend itself to all types of software, but it particularly shines in cases where some kind of complex user input is accepted and processed in some way - like the aforementioned web page renderer. I was recently adding a library to parse EXIF data to images to an Internet-facing service and realised it was a perfect opportunity to do some fuzz testing. Even if I didn’t find any issues, I’d improve my confidence that the library was safe enough to expose to the Internet.

Breaking my EXIF library

I wrote a quick harness to run go-fuzz on the library, and gave it some pre-existing demo files as sample input. The way go-fuzz works is that it instruments your code and then mutates the inputs to try to improve the coverage. For example, if I had some sample data that had an EXIF tag with a value of 1 then go-fuzz might change it to a 2 and see if the code follows a different path. In most cases it won’t but when it does, they tend to be very interesting cases.

One of the first issues that go-fuzz found was that some values in a maker note field would cause the library to panic (i.e., crash). This happened because there was a check to see if the first six characters were “Nikon” and a null byte, without first checking to see if there were actually six characters available. This is a kind of bug that doesn’t happen much with “real” data - as the field is either not present or completed correctly - but could easily be exploited once this code is exposed to the Internet.

Another interesting bug that go-fuzz found was that if a tag had a particularly large count, the library would try to allocate an obscene amount of memory and die. There was already a check in the code that was meant to avoid this exact scenario, but go-fuzz managed to find a way around it. Each tag has a size (for example an integer tag takes a fixed number of bytes) and a count; the existing check multiplied the two together and made sure that the result wasn’t too large. For most cases this was fine but go-fuzz found a case where the count was so large that when multiplied by the size of the tag it overflowed the integer and became negative, thus passing the sanity check but then subsequently failing when it came around to actually allocating the memory.

The final bug of note that go-fuzz found was the most interesting. EXIF data is stored in IFDs (“Image File Directories”), and each IFD provides what is effectively a pointer (a byte offset) to the next one. The EXIF library already had a check to make sure that these didn’t loop, but it only checked the immediately preceding IFD - so if IFD 1 linked to IFD 2, it would catch IFD 2 linking back to IFD 1 and break the loop with an error. Go-fuzz found that having three interlinked IFDs had the same issue, though, and the guard code wasn’t triggered. This created an infinite loop, maxing out a CPU core until the process was eventually killed - one of the worst kind of bugs you could have in an Internet-facing service which doesn’t deal with private data! The fix for this was fairly straightforward - I just made the library keep a record of the previously visited IFDs and bail out if it found a loop.

Breaking my IDE

When go-fuzz detects an issue it outputs not only the details of the problem (the stack trace, error message, and so forth) but also the input that generated the problem. This is useful for reproducing and making sure the issue is fixed, but it also makes it really easy to write a test to ensure that the behaviour never regresses in the future.

As I was working through fixing the bugs that go-fuzz found, I dutifully added new tests where needed. After adding the sample input with looping IFDs to the project, I switched to IDEA to write a test to use it. I clicked on the input file to copy the file name, and then the entire IDE hung and had to be restarted. Uh oh! When I restarted IDEA, it immediately began indexing the project and again hung. It turns out IDEA parses EXIF data (presumably, even if it does nothing else with the data, to get the rotation property for images), and the library they use - an independent one written on Java - had the same bug as the Go library I was using.

In order to stop IDEA from indexing the file and becoming unusable I renamed it from a ‘.tif’ extension to ‘.dat’, and everything went back to normal. I thought I’d best report the bug to JetBrains, though, so they could put a proper fix in.

Breaking YouTrack

JetBrains use their own issue tracker called YouTrack for reporting bugs in IDEA. I dutifully went over and described the problem, attaching the log files from the IDE, a description of how the file was malformed, and carefully selected the .dat version of the file to upload so that it wouldn’t cause anyone else the same immediate problem.

After trying to upload the file I got a strange error back. Uh oh! I submitted the IDEA issue as it stood, unable to see if the attachments had even uploaded, and went and wrote up an issue for YouTrack itself about the error message. While I was doing that, YouTrack seemed to slow down and become really annoying to use. I had a sinking feeling the exact same thing was probably happening as with IDEA and my library - but this time YouTrack had content-sniffed the file instead of relying on the file extension. In hindsight, I should’ve put the file in a passworded archive to ensure no automated tools got hold of it. I marked the issue as a security problem as in a service like YouTrack it presents a denial-of-service opportunity¹ (remember when I said it was one of the worst kinds of bugs you could have in an Internet-facing service?…)

Shortly after I raised my YouTrack ticket, a notice appeared at the top of the page saying they were investigating the current performance issues. Uh oh! I was holding out hope that this was unrelated to me uploading the buggy dat file, but the timing all seemed a bit suspect. I shot support an e-mail saying I think I might be the root cause for their performance issues and linked to the ticket. In the time it took me to e-mail them, the entire site had been put into maintenance mode. I got an e-mail back a few hours later confirming the outage was in fact all my fault, as I’d feared. Within the space of days the JetBrains security team had fixed the issue in YouTrack, which was a pretty nice turnaround.

So if you were trying to access YouTrack at the start of March and couldn’t - I’m sorry, I didn’t mean to! Also, if you’re building an Internet-facing service that takes user input you should really consider running a fuzz tester against it!

“We have a problem”. “Remember, there are no such things as problems, only opportunities”. “Well then we have a DDoS opportunity.” – @J4vv4d ↩︎

Automatic reverse proxying with Docker and nginx

2016-05-21T00:00:00Z

The Docker project logo

Over the past few weeks I’ve gradually been migrating services from running in LXC containers to Docker containers. It takes a while to get into the right mindset for Docker - thinking of containers as basically immutable - especially when you’re coming from a background of running things without containers, or in “full” VM-like containers. Once you’ve got your head around that, though, it opens up a lot of opportunities: Docker doesn’t just provide a container platform, it turns software into discrete units with a defined interface.

With all of your software suddenly having a common interface, it becomes trivial to automate a lot of things that would be tedious or complicated otherwise. You don’t need to manage port forwards because the containers just declare their ports, for example. You can also apply labels to the application containers, and then query the labels through Docker’s API.

Reverse proxying and SSL termination with Nginx and Let’s Encrypt

A fairly significant chunk of the software I run has a web interface. I don’t really want to expose and remember dozens of non-standard ports, so I configure an nginx instance as a reverse proxy. I’m of the opinion that all web traffic should be encrypted, so I also have to provide nginx with trusted certificates to use for each site it reverse proxies. Let’s Encrypt makes the process of obtaining free, trusted certificates approximately a thousand times easier than it was previously, but my workflow still ends up looking like this:

Create a new config file from a template and save it in /etc/nginx/sites-available
Temporarily disable SSL for the site as there’s no valid certificate yet
Enable the site by symlinking to it from /etc/nginx/sites-enabled
Reload nginx
Run the Let’s Encrypt client to obtain certificates
Enable SSL and for the site
Reload nginx

… And that’s not including the extra steps when I miss a semi-colon, accidentally skip a step and have to spend time figuring out why it’s not working, or any of the other human-induced problems that creep in.

I’d been toying with making a script to run through these steps manually for me, if I gave it a domain name and a reverse proxy target, but I never got around to it. Now I’m moving things to Docker, though, there’s an opportunity to automate the entire thing with no human interaction at all.

Existing solutions

It seemed like this probably wasn’t a unique idea, so I had a look around for existing solutions. The most popular by far seems to be nginx-proxy by Jason Wilder. This is based on his docker-gen project that takes a template and populates values from docker containers.

It’s a good solution, but there were a few bits I didn’t like. Firstly, templates don’t really lend themselves well to every step of the process: to request Let’s Encrypt certificates, the container uses a template to create a shell script which it then sources. Each container that generates a template also needs access to the Docker socket. Both of those cause an itch in the back of my head and make me want to say phrases like “attack surface”. I don’t think there’s actually a problem, but it doesn’t really sit well with me.

Secondly, the whole system seems slightly too tightly coupled for my liking. The Let’s Encrypt component needs to modify the nginx config in order to obtain the certificate, while the main nginx component is also making different changes to add and remove sites. It feels like if it doesn’t just work, it’s going to be difficult to debug and pry apart the different components.

Another potential solution is Rancher. This is a complete platform for managing containers, and I’m fairly sure if configured right it can grab certificates from Let’s Encrypt and do SSL termination using haproxy. I tried it for a bit but the whole platform seemed a bit overkill for my purposes, and I didn’t want to invest the time I’d need to fully understand it all.

Rolling my own

In the end I decided to roll my own solution. Here’s a high-level overview of how it all works:

Diagram showing components of a reverse proxy implementation

As you probably noticed, there are quite a few containers involved. Each one performs a small, well-defined task, and its output can easily be inspected in either a volume or a database. I think there’s some similarity to piping commands together on a command line — it’s a lot easier to reason about simpler commands like head, cut and tr than it would be one giant command that combined them. And, if it does go wrong, you can inspect the pipe at each stage to see where the problem is happening.

service-reporter and etcd

The first part of the chain is my service-reporter container. This uses the Docker API to get a list of containers, and store information about them in etcd. Etcd is a distributed key-value store (similar in some ways to redis or memcached). The container also watches for containers that are added and removed, and keeps etcd updated appropriately.

As the service metadata is stored in a database, no other part of the system needs to interact with Docker. If the Docker API changes, or the host configuration changes, then only this container has to be updated.

service-letsencrypt and letsencrypt-lexicon

The left fork of the diagram deals with obtaining SSL certificates. To keep it separate from the nginx configuration, it uses DNS-based challenge to prove that we control the domains. It does this by plumbing together two great open source projects: letsencrypt.sh, a Let’s Encrypt client implemented in bash with support for the dns-01 challenge type, and Lexicon, a python library for updating DNS records using a variety of providers.

My service-letsencrypt container connects to etcd and pulls a list of containers that have a label with the key com.chameth.vhost. It uses this to build a plain text list of certificates we require (in a format understood by letsencrypt.sh), and then monitors etcd for changes and repeats as necessary.

The letsencrypt-lexicon container runs letsencrypt.sh, using Lexicon to perform the required DNS updates, and produces certificates. The nice thing about this is that it can be used in a completely standalone fashion (you can just write a domains.txt yourself). It uses iowait to watch the domains text file for updates, and automatically reruns when there are changes. It also runs once a day to renew any certs that are coming up for expiry.

service-nginx and nginx

The right fork of the diagram is concerned with nginx. My service-nginx container again connects to etcd and pulls a list of containers. It uses a couple of labels to determine the vhost, proxy port, and proxy protocol. It then feeds these values into a template to create a server block for each site, configured with SSL certificates and a reverse proxy setup. The template covers only the very minimal settings, with the expectation that everything else will be done in the global config (things such as SSL ciphers, redirection from HTTP, etc).

This container works completely independently of the Let’s Encrypt side. You can use the Let’s Encrypt containers and mount the certificate volume, or you could just provide your own certificates. It doesn’t really make any difference.

Putting it all together

The only downside to having many small containers is that it’s a bit of a nuisance to get them all set up. Fortunately, Docker has a solution for this in the form of Docker compose. This allows you to write a YAML file defining all of the services you want to run, and bring them up or down in one go. It can handle volumes, dependencies, networking, etc. I’ll be publishing a docker-compose.yml file to get this entire stack up and running soon.