I was just setting up remote runners for TankieTube when I had an epiphany:
I could ask comrades to volunteer their own computers! 
That way those who can't or don't care to donate monetarily could still contribute.
How would it work?
Conceptually, you can think of it like a crypto mining botnet. Except it transcodes videos for the community instead of producing heat for individual profit. And it's voluntary ofc.
It can run on any operating system with an internet connection. I'm going to use my gaming desktop and at least one VPS.
Tech level required: comfortable copy-and-pasting things into a CLI.
OpSec considerations: negligable as far as I can tell. There is no P2P involved. Your computer talks directly to the TankieTube server using sicko-to-HTTPS communication. The server would see your IP address, but that's always the case on every website.
Thoughts?
Video decoders are all written in C and almost certainly full of exploitable bugs, thus people participating in this are making their personal computers vulnerable to attack via manipulated video files. You'd at least want this sandboxed as much as possible, and have it run as it's own user.
Yeah, I use a designated user
prunnerlike the docs suggested.Is ffmpeg really that much of a security concern?
making their personal computers vulnerable to attack via manipulated video files
Is this any more dangerous than BitTorrenting anime? All the video files would be coming from https://tankie.tube only.
I think it is worth further research, at least. Setting aside potential exploits in ffmpeg, containerization (if not virtualization) seems necessary. A process running as root in a Docker container effectively has root access to the host, but a properly designed container should run all the work as non-privileged users. This work can be isolated using the cgroups APIs (docker should manage this, I think) and potentially reinforced with SELinux policies. Done correctly, this would effectively limit the impact of remote code execution in ffmpeg to denial of service. The attack surface for privilege escalation would then be limited to the Linux syscall API, utilities with the setuid flag, etc (highly, highly audited stuff that would allow you to root any machine if it were broken).
Alternately, it might be worth looking at bubblewrap, which is the basis of FlatPak containerization.
What is the threat model? The TankieTube server sends a malicious MP4 to the remote runner machine? Or a malicious remote runner sends a malicious MP4 to the server?
The former is easy to avoid by me not being evil. The latter is only a security concern for the TankieTube server, not the contributors.
Hostile user uploads a malicious video file, peertube instance sends it to a volunteer for transcoding, RCE occurs on volunteer's machine.
So the concern is that a volunteer could have a more vulnerable installation of ffmpeg compared to the local server? How does that happen?
I think most prospective volunteers just want ffmpeg to run in a secure context without root, and stronger vouching of security by multiple knowledgeable users. The specifics of ffmpeg having vulnerabilities is not that important.
Do you vet all videos uploaded? I thought anyone could upload files to tankietube.
Yes. Anyone can upload, and videos are published automatically. I review the videos every day that have been published recently. I also rely on user reports.
Ffmpeg is used by everybody so you'd hope people are looking at it, but I'm sure there's security bugs in there, and probably plenty of them, since it's C parser/decoder code, probably the most dangerous kind of code. I think web browsers do some kind of sandboxing around ffmpeg, plus web browser restrict the kinds of formats they support, but ffmpeg (and peertube?) supports a lot more, many of which will not be audited/fuzzed to the same degree.
Ideally this would be sandboxed so much it can't call anything but read(2) and write(2). I have no idea if any of this software does any sandboxing at all.
Is this any more dangerous than BitTorrenting anime?
Maybe, depends on the what exactly you're worried about. There's potentially political actors that might be interested in fucking with tankie.tube, whereas you can't really target anyone specifically with bittorrent. Also the attacker knows exactly what software will be used to decode the videos, which makes this easier to exploit. I assume that videos can get uploaded to tankie.tube by basically anybody, and those videos would be sent out to be transcoded on random people's machines?
If you assume tankie.tube (maybe peertube in general) is just too small to be on anyone's radar, then that's probably fine.
I assume that videos can get uploaded to tankie.tube by basically anybody,
Yes.
and those videos would be sent out to be transcoded on random people's machines?
Randomly to anyone entrusted with a token bestowed by me.
BTW I love your avatar. :3
I do have spare credit on a cloud provider but cloud network egress costs are crazy. Around 10 cents/GB. However, if the bandwidth is not too too bad I could do it.
ideally you should spin up a separated VM without access to your own network
I've got a couple machines lying around, and due to set up a server some point soon, so I'd be all down to contribute with that. Even if there are possible sec concerns (which I don't know if there are without looking harder), I'd personally sandbox it all anyway.
None. It uses the same port as the web browser you're on right now.
So basically the client periodically initiates connections to the server resolved over DNS and the server replies to these requests?
I have a sever. I'm lazy though so I would prefer a docker guide.
https://hub.docker.com/r/owntube/peertube-runner/ this seems to already exist btw
I think this would best work by writing up a simple Dockerfile for Docker or Podman (one may already exist out there) and at minimum containerizing the work. That level of constraint is likely all that's needed and the end user can also control how much CPU is used, the networking interface to possibly proxy the connection if desired, and what directories are accessible.
Alternatively, a VM could accomplish the same in terms of safety. Of course a VM requires a pre-allocated block of RAM and a full guest OS installation.
Are you sure you want to give out the direct IP address to your server? Or do you have a proxy set up for ddos protection?
Are you sure you want to give out the direct IP address to your server?
The server IP is public. I'm not using a cloudflare proxy because they're whack. What would you recommend for DDOS protection?
I have some credit on a random cloud provider, I may be interested as long as usage doesn’t raise any red flags to them.
Any usage really. If resources start getting pegged on the VPS, they may CPU limit it to preserve performance of the neighboring VPS instances.
I see. You can always throttle the CPU usage in the configuration. Or run the service only x hours per day.
I'd be happy to done some idle CPU's if you get it up and running
Would Cygwin be the way to go for Windows users?
Try either WSL or a Linux VM if you want to do this on Windows.
