Nocom – 2b2t Minecraft server exploit using Monte-Carlo localization

Not quite a shortcut, it was actually to patch a bug in vanilla minecraft that causes desync, see https://news.ycombinator.com/item?id=29621262 And it may seem silly that it tells you that it's block type ABC, but the reason is to tell your client to put it back as block ABC, instead of air. (in other words, it has to "undo" your client claiming to have mined that block [which replaces whatever it is with air]).

Sure!

For monitoring on a meta-level, we used Prometheus and Grafana. Here's what that looked like near the very end: https://www.dropbox.com/s/ktlvvtmppq64u8o/trimmed%20grafana%... Since the ratelimit was heavily in place by then, the numbers are pretty low. For example, the headliner "blocks per second" was well over three thousand a few weeks before then.

One of the most common queries for actual usage though, was "what bases in the last 24 hours (or whatever) have the most chests currently". We'd then travel to these locations to borrow items. Here's that query https://pastebin.com/fjhHGSYz Roughly, it grabs chunks with chests in that timeframe, then uses a recursive Postgres CTE to traverse the https://en.wikipedia.org/wiki/Disjoint-set_data_structure that we used to implement DBSCAN, from the leaf to the root. Then it groups by the root, which in practice means grouping chunks together by which ones semantically have been determined to be part of the same base, then makes a report by quantity per base.

There was also a web UI that showed all active tracks. You could click any player and see their travel history, the route they had taken to get to that location. As well as clicking any cluster and seeing the usernames associated with it. We used this a lot to grab what members were present at a given item stash or base, to make sure we weren't stepping on anyone's toes.

In terms of what it took to make this happen, the stars of the show were probably the main Postgres, which was hand-tuned on top of a hand-tuned ZFS, on top of a NVME drive that was IOMMU passthrough'd to this VM for maximum performance. As well as https://github.com/OpenHFT/Chronicle-Map, which really helped out. We needed a mapping from block coordinate to a bit of data about that location, such as the last timestamp we checked it, what Minecraft block was there last, whether the block is interesting, uninteresting, different from seed, etc. High frequency trading software was perfect for this use case, because it greatly reduced GC pressure on the JVM, since the data was stored off-heap. It ended up at around (roughly) fifty million entries, served hundreds of thousands of requests for block data per second, and requested historical data (from the Postgres) at about thirty thousands SELECTs per second (from a table with over ten billion rows). The reason why this needed to be so fast, is that the use case is downloading bases. Sometimes someone would log in to 2b2t at their base, after having not played for weeks at a time (so the data is in Postgres, not the map in RAM), we wanted to be able to pick up right where we left off on downloading what they had built. This means it needs to get back up to speed with their entire base, as fast as possible, in the seconds before they move away to other chunks, or log off the server. We are downloading bases one click at a time, and bases are huge. The area that is loaded in by a player is 144 by 144 by 256. That’s 5 million blocks. And there are about 250 players online on 2b2t at any given moment in time. That’s well over a billion blocks that we could click on, and learn which Minecraft block they are. But, which locations should we use up our clicks per second budget on? It focused on areas that are being changed rapidly, and areas that are different from the base vanilla terrain. Those are certainly bases, large Minecraft builds. We do an expanding paintbucket pattern that expands in 3 dimensions to whatever has been built there. However, we have to do this quickly, and we have to pick up where we left off. Someone might be flying around their base with an elytra, at very fast speed. We would love to scoop up the terrain that they’re flying over. If the base is particularly interesting, someone flying over a certain area could have had about twenty Minecraft accounts spread all over 2b2t’s map suddenly refocus all of their left clicks on that one area, to grab all the interesting neighbors that it wasn’t able to last time. We peaked at about three thousand block clicks per second, averaged over weeks. So this was a massive data loading and management problem.

Correct, loading a pearl and respawning would throw us off the trail. This is why our aim was to follow the initial travel to the base location, rather than a future pearl TP. The easiest advice in order to throw off nocom is to travel in a straight line in the overworld for perhaps 10k blocks, set a bed quickly, then continue another 10k blocks, then /kill to the bed, then go perpendicularly for a render distance or two before the chunks unload at your death location. Even with some hypothetical tracking improvements (that never actually were written), such as walking back the last 10 minutes whenever a track goes cold, that would still throw it off.

Generally 2 or 3. 1 point of association was given whenever a track went cold, split up among however many players logged off within -6 seconds to +1 seconds of that timestamp (because of chunks unloading on a 5 second schedule). 1 point could be a false positive, but 2 or 3 would be reliably trustworthy.

I am not done with university yet, so I don't speak confidently. But I'm afraid I have no special perspective, I just echo what I hear a lot: Learning by doing is probably the most important, but structured courses provide a theoretical and practical foundation that fills in a lot of important gaps that I would have had if I were only self taught.

Yes absolutely, for fun! Of course, it's also fun to have a secret project that no one knows about, to get an edge. 2b2t is practically the only Minecraft server that allows any kind of hacking your client (without getting banned), while also having an active community of meaningful size, so that was the clear target.

griefing

Btw Mirai botnet also came out of minecraft.

Lol! Sorry for getting it locked, it is reversible if you still have the email. Mojang, it appears, didn't like spam login from a different cloudflare warp IP every few minutes...

Fun to see you around here as well!

Hard to estimate, probably a few hundred hours from me, and a comparable amount from others. As a random data point, I sent 53000 messages in the group chat developing it. ¯\_(ツ)_/¯

Yes, Forgegradle was used. Frankly, not much interesting here, just a whole lot of work going through all the Minecraft code, and using Java reflection where possible and JVM bytecode injection otherwise. There's a lot of access transforming, using theUnsafe to allocate classes without calling their constructors, etc. As a random example that I remember, one of the last things to patch was that we had it working, but when you got kicked from the server, it would crash. This is because it would try to create a currentScreen GUI for the "disconnected from server" message, which for some reason was long enough that it was considering whether to wrap the text to the next line, which needed something like a font size, or maybe the window width, which crashed since there is no window or font.

Nothing so advanced, I'm afraid. The Monte Carlo system just creates a thousand points (2d position and velocity), and simulates them at 1 generation per second, culling points that are unlikely and reproducing points that are likely based on observations (clicking chunks and hearing back "loaded" or "unloaded").

Mostly learned as I went! I had never before implemeneted Monte Carlo localization, or DBSCAN, or a disjoint union find; it was a lot of research and trial and error. I'd honestly recommend Wikipedia! I also read a lot of blog posts such as on data clustering in order to decide how to write an algorithm to tell what is and isn't a single contiguous base.

The main expense was priority queue, which is a $20/account/month payment to 2b2t to be able to join the server faster. Probably nearing a thousand dollars on that. We lightened this by creating a proxy system where other builders could join through our headless, and it would add nocom exploit packets from their account to 2b2t, and peel off the responses on the way back. They got a more reliable experience with less disconnects, and we got some extra capacity for free (completely unbeknownst to them of course haha). By the end, all the overworld accounts were being paid for and used by other people, I only had to cover the 2 accounts in the nether.

About $300 on a new 2TB nvme drive for postgres. An unknown amount on electricity. Other than that, all the big processing ran on an existing homelab server (thanks to fr1kin). About $150 on digitalocean for a droplet to run the headless minecraft with minimal ping to 2b2t ($10/month). Backups were just WAL rsyncs to postgres replicas ran (again on homelabs) by myself and a few others.

This is one of life's great mysteries. Given the hosting provider for 2b2t, we know pretty much for a fact that the packet spam from 2018 and 2019 (before I knew about the project ;)) would have cost tens of thousands of dollars in bandwidth overages per month. Yet, he never added a packet limiter until an unrelated exploit could cause other clients to crash due to spamming an offhand swap sound effect. It took the exploit going fully public, plus another 36 hours or so, for it to actually get patched. ¯\_(ツ)_/¯

For tracking, the spatial resolution was 1 chunk, which is 16 by 16 blocks, and the temporal resolution was 1 second. So, we have a few billion rows of tracking data for every player, localizing them to roughly that distance, roughly once a second.

For downloading bases, it is an essentially perfect recreation. Some things are missing, such as the color of banners, but for 99% of blocks, just setting the correct blockstates will recreate the build. Some disconnected sections of the build might not be found by the paintbucket floodfill algorithm though, so floating parts could be missing. We counteracted this by having a few random blocks in the chunk be checked on a schedule, so eventually it would get everything.

There's a section called "The entire thing summed up in three sentences" in the linked GitHub README.

"You could punch any block coordinate in the entire map, and 2b2t would either:

-Tell you what block is placed there, if the chunk is currently loaded

-Or ignore you, if the chunk is not currently loaded.

2b2t would immediately reveal whether or not any chunk coordinate that you ask about is currently loaded. If it is currently loaded, it tells you which specific block is there at the coordinate you asked about."

This allowed the hackers to find every player's coordinates and where their bases were.

Its a typical case of Software developer not validating player input 'because thats too taxing on the server'. Minecraft server was programmed to reply to all, even stupid or impossible queries.

Hey thanks for your reply. I get how the Paper patch allowed this new exploit, but Im still curious to why Paper never even had the range check. Even for lag compensation it shouldnt be more than a chunk or 2 range.

Im actually reminded to the latest log4j exploit (main reason i aint actually gonna run Minecraft for a while) and even the Heartbleed exploit a few years back.

Did yall assume the patch to Paper would just return if the chunk was player loaded, and they wouldn't add any other sane bounds checks?

If the story from the FitMC guy is true, yall guessed/hoped that patch would be what it was. Giving yall a new leak of information without actually fixing the problem because if it was loaded you could still get all the block info in that chunk.

Pretty awesome from where i'm sitting. \(ᵔᵕᵔ)/

Thanks for the clarification!

Probably the only server experience that really felt like you had an entire world and society to explore and consequences to deal with.

I spent way too much time trying to figure out a way to scrounge up Ghasts and TNT to level Orion.