Hey! I'm David, the author of the Real-World Cryptography book. I'm a crypto engineer at O(1) Labs on the Mina cryptocurrency, previously I was the security lead for Diem (formerly Libra) at Novi (Facebook), and a security consultant for the Cryptography Services of NCC Group. This is my blog about cryptography and security and other related topics that I find interesting.

# How to write a tic-tac-toe snapp posted 6 days ago

(photo by @micheile)

Snapps are zero-knowledge smart contracts that will launch on Mina this year. You can learn more about them here. This tutorial teaches you how to write a tic-tac-toe game using snarkyjs, the official library to write snapps on Mina.

## Set up

You can quickly create a project by using the Snapp CLI:

$npm install -g snapp-cli$ snapp project my-proj

you can also follow along this tutorial with the following command:

$snapp example tictactoe ## Hello world To write a smart contract, import what you need from snarkyjs and simply extend the SmartContract class. import { Field, State, PublicKey, SmartContract, state, method, PrivateKey, UInt64, Int64, Bool, Circuit, Mina, Party, shutdown, Optional, Signature, } from 'snarkyjs'; class TicTacToe extends SmartContract { // your smart contract } A zero-knowledge smart contract, or snapp, is very close in concept to the ones you see on Ethereum: • it has a state • it has a constructor that initializes the initial state when you deploy your snapp • and it has methods that can mutate the state Let's start with the state. ## A state for a tic-tac-toe Snapps, at the lowest level, only understand field elements, of type Field. Field is a type similar to Ethereum's u256 except that it is a bit smaller. Essentially it's a number between 0 and 28948022309329048855892746252171976963363056481941560715954676764349967630336. Every other type has to be derived from Field, and we provide a number of such helpful types in snarky. (You can also create your own types but we won't use that here.) First, a tic-tac-toe game needs to track the state of the game. The board. The state of a Snapp can only hold 8 field elements. This is not set in stone, but the more storage a snapp can contain, and the harder it becomes to participate in consensus (and thus the less decentralized the network becomes). Snapps that need to handle large state can do so via Merkle trees, but I won't be talking about that here. A tic-tac-toe board is made of 9 tiles, that's one too many :( so what can we do about it? Well as I said, a field element is a pretty big number, so let's just encode our board into one: class TicTacToe extends SmartContract { // The board is serialized as a single field element @state(Field) board: State<Field>; } We'll also need to keep track of who's turn is it, and if the game is finished (someone has won): class TicTacToe extends SmartContract { // The board is serialized as a single field element @state(Field) board: State<Field>; // false -> player 1 | true -> player 2 @state(Bool) nextPlayer: State<Bool>; // defaults to false, set to true when a player wins @state(Bool) gameDone: State<Bool>; // player 1's public key player1: PublicKey; // player 2's public key player2: PublicKey; } Notice that the public keys of the players are not decorated with @state. This is because we don't need to store that information on chain, we can simply hardcode it in the snapp. If you wanted to be able to start a new game with new players, you would store these on chain so that you could mutate them via a method of the snapp. But we're keeping it simple. Also, PublicKey and Bool are two types provided by the standard library and built on top of Field. Bool is built on top of a single field element, so in total our on-chain state has 3 field elements. That'll work! ## A constructor Next, we must initialize that state in a constructor. let's look at the code: class TicTacToe extends SmartContract { // The board is serialized as a single field element @state(Field) board: State<Field>; // false -> player 1 | true -> player 2 @state(Bool) nextPlayer: State<Bool>; // defaults to false, set to true when a player wins @state(Bool) gameDone: State<Bool>; // player 1's public key player1: PublicKey; // player 2's public key player2: PublicKey; // initialization constructor( initialBalance: UInt64, address: PublicKey, player1: PublicKey, player2: PublicKey ) { super(address); this.balance.addInPlace(initialBalance); this.board = State.init(Field.zero); this.nextPlayer = State.init(new Bool(false)); // player 1 starts this.gameDone = State.init(new Bool(false)); // set the public key of the players this.player1 = player1; this.player2 = player2; } } The constructor does two things that might look weird. First, it takes an address as argument, this is the address where the snapp will be deployed (this might disappear in future versions of snarky). Second, the constructor takes an initialBalance, which might be needed to pay the account creation fee (if you haven't already). The rest should be pretty straight forward. ## Adding methods to the snapp We really only need one method: a method to play. We can create one with the @method decorator: class TicTacToe extends SmartContract { // ... @method async play( pubkey: PublicKey, signature: Signature, x: Field, y: Field ) { // ... } } The method takes the player's public key, two coordinates, and a signature (to make sure that they own the public key) over the coordinates. The board can be visualized as such: ## The logic, at a high-level Let's look at the logic that we will have to implement: class TicTacToe extends SmartContract { // ... @method async play( pubkey: PublicKey, signature: Signature, x: Field, y: Field ) { // 1. if the game is already finished, abort. // 2. ensure that we know the private key associated to the public key // and that our public key is known to the snapp // 3. Make sure that it's our turn, // and set the state for the next player // 4. get and deserialize the board // 5. update the board (and the state) with our move // 6. did I just win? If so, update the state as well } Does that make sense? In the rest of this tutorial I go over every step ## Step 1: Game over is game over if the game is already finished, abort. To do this, we have to read the state of the snapp with get. When you execute the program locally, this will go and fetch the state of the snapp from the blockchain (hence the asynchronous call with await). Your execution will then only be deemed valid by the network if they see the same state (in this case the value of gameDone) on their side. const finished = await this.gameDone.get(); finished.assertEquals(false); Using assert functions like [assertEquals()] is the natural way to abort the program if it takes the wrong path. Under the hood, what happens is that the prover won't be able to satisfy the circuit and create a proof if the assert is triggered. ## Step 2: Access control ensure that we know the private key associated to the public key and that our public key is known to the snapp First, let's verify that the public key is one of the hardcoded key: Bool .or(pubkey.equals(this.player1), pubkey.equals(this.player2)) .assertEquals(true); Next, let's verify that the player truly owns the private key associated with that public key by verifying a signature over the coordinates. signature.verify(pubkey, [x, y]).assertEquals(true); ## Step 3: taking turns Make sure that it's our turn, and set the state for the next player Earlier, we encoded a player as a boolean, with false describing the first player and true the second one. So let's derive this first. In zero-knowledge proofs, we can't have branches with too much logic. The program pretty much always has to do the same thing. That being said, what we can do is to assign a different value to a variable depending on a condition. For example, this is how we can get the boolean that describes us as a player: // get player token const player = Circuit.if( pubkey.equals(this.player1), new Bool(false), new Bool(true) ); Circuit.if() takes three arguments, the first one is a condition that must be a Bool type, the other two are the returned values depending on if the condition is true or false. Now that we have the player value, we can use it to check if it our turn: // ensure its their turn const nextPlayer = await this.nextPlayer.get(); nextPlayer.assertEquals(player); And we can update the state for the next player via the set() function: // set the next player this.nextPlayer.set(player.not()); ## Step 4: fitting a board in a field element get and deserialize the board There's many strategies here, but a simple one is to deserialize the field element into bits (via toBits()) and interpret these bits as the board. const serialized_board = await this.board.get(); const bits = serialized_board.toBits(9); We can simply flatten the board like this: Yet, this is not enough. A bit can only represent two states, and we have three: empty, false (player 1), true (player 2). So we'll use two lists of size 9 instead: • one to track who played on that tile • and one to track if a tile is empty or not As you can see, it is important to track if a tile is empty or not, because in the first list 0 represents both the empty tile and a move from the player 1. const serialized_board = await this.board.get(); const bits = serialized_board.toBits(9 * 2); const bits_not_empty = bits.slice(0, 9); const bits_player = bits.slice(9, 18); let board = []; for (let i = 0; i < 3; i++) { let row = []; for (let j = 0; j < 3; j++) { const not_empty = bits_not_empty[i * 3 + j]; const player = bits_player[i * 3 + j]; row.push([not_empty, player]); } board.push(row); } To make the rest of the logic easier to read, I'm going to use the Optional type to encode the [not_empty, player] array. It looks like this: class Optional<T> { isSome: Bool; value: T; // ... } It is simply a Bool that describes if it's empty (no one has played that tile yet), and if it's not empty the value associated with it (a cross or a circle, encoded as the player Bool). Here's what the full deserialization code looks like: const serialized_board = await this.board.get(); const bits = serialized_board.toBits(9 * 2); const bits_not_empty = bits.slice(0, 9); const bits_player = bits.slice(9, 18); let board = []; for (let i = 0; i < 3; i++) { let row = []; for (let j = 0; j < 3; j++) { const not_empty = bits_not_empty[i * 3 + j]; const player = bits_player[i * 3 + j]; row.push(new Optional(not_empty, player)); } board.push(row); } ## Step 5: Updating the board update the board (and the state) with our move Before we do anything, we need to check that the coordinates are valid: x.equals(Field.zero) .or(x.equals(Field.one)) .or(x.equals(new Field(2))) .assertEquals(true); y.equals(Field.zero) .or(y.equals(Field.one)) .or(y.equals(new Field(2))) .assertEquals(true); Now here comes the trickiest part. Zero-knowledge smart contracts have some constraints: they don't allow you to dynamically index into an array, so you can't do this: board[x][y].isSome = new Bool(true); board[x][y].value = player; instead, you have to go through every tile, and check if it's the one to update. (Note that you can't have dynamic for loops as well, they need to be of constant size). for (let i = 0; i < 3; i++) { for (let j = 0; j < 3; j++) { // is this the cell the player wants to play? const to_update = Circuit.if( x.equals(new Field(i)).and(y.equals(new Field(j))), new Bool(true), new Bool(false) ); } } And you can use the same tricks to make sure you can play on the tile, and to update the board: for (let i = 0; i < 3; i++) { for (let j = 0; j < 3; j++) { // is this the cell the player wants to play? const to_update = Circuit.if( x.equals(new Field(i)).and(y.equals(new Field(j))), new Bool(true), new Bool(false) ); // make sure we can play there Circuit.if( to_update, board[i][j].isSome, new Bool(false) ).assertEquals(false); // copy the board (or update) board[i][j] = Circuit.if( to_update, new Optional(new Bool(true), player), board[i][j] ); } } Finally, we can serialize and store this update into the snapp's state: // serialize let not_empty = []; let player = []; for (let i = 0; i < 3; i++) { for (let j = 0; j < 3; j++) { not_empty.push(this.board[i][j].isSome); player.push(this.board[i][j].value); } } const new_board = Field.ofBits(not_empty.concat(player)); // update state this.board.set(new_board); ## Step 6: Did I just win? did I just win? If so, update the state as well Finally, we can check if someone won by checking all rows, columns, and diagonals. This part is a bit tedious and boring, so here's the code: let won = new Bool(false); // check rows for (let i = 0; i < 3; i++) { let row = this.board[i][0].isSome; row = row.and(this.board[i][1].isSome); row = row.and(this.board[i][2].isSome); row = row.and(this.board[i][0].value.equals(this.board[i][1].value)); row = row.and(this.board[i][1].value.equals(this.board[i][2].value)); won = won.or(row); } // check cols for (let i = 0; i < 3; i++) { let col = this.board[0][i].isSome; col = col.and(this.board[1][i].isSome); col = col.and(this.board[2][i].isSome); col = col.and(this.board[0][i].value.equals(this.board[1][i].value)); col = col.and(this.board[1][i].value.equals(this.board[2][i].value)); won = won.or(col); } // check diagonals let diag1 = this.board[0][0].isSome; diag1 = diag1.and(this.board[1][1].isSome); diag1 = diag1.and(this.board[2][2].isSome); diag1 = diag1.and(this.board[0][0].value.equals(this.board[1][1].value)); diag1 = diag1.and(this.board[1][1].value.equals(this.board[2][2].value)); won = won.or(diag1); let diag2 = this.board[0][2].isSome; diag2 = diag2.and(this.board[1][1].isSome); diag2 = diag2.and(this.board[0][2].isSome); diag2 = diag2.and(this.board[0][2].value.equals(this.board[1][1].value)); diag2 = diag2.and(this.board[1][1].value.equals(this.board[2][0].value)); won = won.or(diag2); // update the state this.gameDone.set(won); The full code is available here if you want to play with it. comment on this story # In response to Moxie's doubts on web3, and about ultra light clients posted 2 weeks ago Moxie just wrote his first impressions on web3. If you've been wondering about web3, and want a non-bullshit and technical glance at what it is, I think this is a really good post. While I do agree with his critics, I don't share his skepticism. Let's look at his description of how most users interact with decentralized apps (dapps): For example, whether it’s running on mobile or the web, a dApp like Autonomous Art or First Derivative needs to interact with the blockchain somehow – in order to modify or render state (the collectively produced work of art, the edit history for it, the NFT derivatives, etc). That’s not really possible to do from the client, though, since the blockchain can’t live on your mobile device (or in your desktop browser realistically). So the only alternative is to interact with the blockchain via a node that’s running remotely on a server somewhere. A server! But, as we know, people don’t want to run their own servers. As it happens, companies have emerged that sell API access to an ethereum node they run as a service, along with providing analytics, enhanced APIs they’ve built on top of the default ethereum APIs, and access to historical transactions. Which sounds… familiar. At this point, there are basically two companies. Almost all dApps use either Infura or Alchemy in order to interact with the blockchain. In fact, even when you connect a wallet like MetaMask to a dApp, and the dApp interacts with the blockchain via your wallet, MetaMask is just making calls to Infura! This is a real problem. To interact with the blockchain, you need to download its whole history with a program. This takes a lot of time (sometimes days) and space. This is not realistic for most users, and even less so for mobile users. The solution has been to just trust a public node. Infura is one of them and has become quite "trusted" through being the backend behind Metamask, the most popular Ethereum wallet. This service can lie to you, and you wouldn't notice. As such, the whole security of the blockchain is moot once you start interacting through a public node. “It’s early days still” is the most common refrain I see from people in the web3 space when discussing matters like these. In some ways, cryptocurrency’s failure to scale beyond relatively nascent engineering is what makes it possible to consider the days “early,” since objectively it has already been a decade or more. Did you know that most of the web was not encrypted until recently? For example, Facebook defaulted to https in 2013. Not even a decade ago. Cryptocurrencies have come a long way since the advent of Bitcoin, and research has exploded in all directions. It is early days. Now, what are the actual solutions that exist in the space? First, newer byzantine-fault tolerant (BFT) consensus protocols have straight-forward solutions to this problem. Since each block is cryptographically certified by consensus participants, and no re-organization (or forks) can happen, the certification can be reused to provide a proof to the clients (the real ones). As such, Infura could very well give you this cryptographic proof to accompany a response to your request, and the problem would be fixed. This is what Celo is doing with Plumo, but more generally any blockchain that uses a BFT consensus (Diem/Libra, Cosmos, Algorand, Dfinity, etc.) should be able to implement something like this. The security guarantee is not as high as verifying every block since the beginning of time (the genesis). (For this, cryptocurrencies like Mina have proposed recursive zero-knowledge proofs solutions that attest to all the state transition since genesis.) Plumo calls this ultra-light clients: We observe that in order to verify the latest block header in BFT networks a client only needs the public keys of the current committee. As long as no committee has had a dishonest supermajority, a client who verifies a chain of committee hand-off messages certifying the PoS election results, known as epoch messages, does not need to check each block or even the headers of each block. Instead, to make (or verify a recent) transaction, the client simply asks for the latest (or otherwise relevant) block header, and verifies that it has been signed by a supermajority of the current committee. This constitutes the simplifying assumption (SA) and light client protocol proved by Plumo. Another issue is key rotations, which increase the size of the proof (as you need to give proofs to all the key rotations before you can give a proof to the latest state of the chain), but I believe that zero-knowledge proofs can fix that as well. Bottom line: it's actually not that grim, solutions are there, but users have to care for people to implement them, apply them, and for the solutions to receive adoption. OK, this is pretty much what Moxie says: However, even if this is just the beginning (and it very well might be!), I’m not sure we should consider that any consolation. I think the opposite might be true; it seems like we should take notice that from the very beginning, these technologies immediately tended towards centralization through platforms in order for them to be realized, that this has ~zero negatively felt effect on the velocity of the ecosystem, and that most participants don’t even know or care it’s happening. This might suggest that decentralization itself is not actually of immediate practical or pressing importance to the majority of people downstream, that the only amount of decentralization people want is the minimum amount required for something to exist, and that if not very consciously accounted for, these forces will push us further from rather than closer to the ideal outcome as the days become less early. But this is not counting on Infura getting hacked. And it will get hacked. 1 comment # The Cairo SNARK CPU architecture in one example posted 4 weeks ago I spent some days over Christmas to read the paper Cairo – a Turing-complete STARK-friendly CPU architecture. First, it's a work of art. I find a lot of the papers in the field of general-purpose zero-knowledge proofs (GP-ZKPs) quite hard to read, but this one has a lot of emphasis on the flow and on examples. More than that, the ideas contained in the paper are quite mind-blowing (although everything is mind-blowing in the land of GP-ZKPs). What is this paper? It's the layout, in quite the detail, of a protocol that can be used to produce zero-knowledge proofs that a program executed correctly (what they call proof of computational integrity*). While most approaches "compile" each program into its fixed protocol (with a prover part and a verifier part), Cairo is a single protocol that works for any program (of some bounded size of course). They call this the CPU approach, as opposed to the ASIC approach. Effectively, the program you want to prove is encoded in some Cairo instruction set and passed as input to another program that will simply run it, like a VM. Translated in ZKP-lingo, a circuit will constrain the execution trace to match the execution of the program passed as public input. (This last sentence might be hard to understand if you are not familiar with at least one GP-ZKP system.) Cairo is also a higher-level language, that you can play with on the Cairo playground. If you click on "debug" you can see the program it compiles to. Essentially a series of instructions and data, which you can think of as code and data segments of a binary. The rest of the memory essentially acts as a "stack" that is used during the execution of the program. The higher-level language is out of scope of the paper. Instead, the paper details the set of instructions and its encoding, which is then unpacked in the constraint of the circuit executing the program. In the paper, they go through a simple example, the Fibonacci sequence, to illustrate how a program works and how the state transition logic works. As I had to reverse engineer it I thought I would share this here. Here's our program, which is two instructions followed by three values. Let's start simple and let's look at what the first instruction tells us. There's a lot to unpack here, there's three values of 16 bits that represent offsets in our memory. Two for the two operands (op0 and op1) of a binary operation, and one (dst) to indicate where to store the result of the operation. Let's ignore these for now and focus on the last part: the flags. To understand what each flags do, you have to take a look at the state transition function. It's a bit confusing to follow, but there's apparently a Lean proof that it is correct (this field really keeps on giving). A lot of this is not going to make sense without reading the paper. But there's three registers: • pc is the program counter, it points to an address in memory from where to fetch the next instruction. • ap is the allocation pointer, it points to a free address in memory, useful to store data globally. • fp is the frame pointer, used to store data as well when doing function calls. The logic can be summarized as: fetch data from memory; operate on it (addition or multiplication); update the registers. So what happens with our example's first instruction? Still a bit abstract but that should give you an idea. Before I reveal the result of the whole instruction let's look at the rest of the instruction: the offsets The offset leaks some of the logic: -1 and -2 are the indices in the Fibonacci formula$u_i = u_{i-1} + u_{u-2}$Let's now reveal what the first instruction is really doing. There's no doubt that if you've followed that far you must have read the paper. What happens can be summarized as: • we increment the program counter pc (to go to the next instruction) • we add the two values stored in the memory at offset -1 and -2, and store the result in memory. • we increment ap, since we stored in memory. We don't touch fp since we didn't do any function call. Let's look at the second instruction. Let's reverse it using our table. And let's look at the flags What's interesting here, is that the instruction leads to a jump! The next register values will lead to the circuit re-starting from instruction at memory address 0, because pc points to it. Thus, we're in an infinite loop, which is parameterized by the number of instructions that the prover wants to run. comment on this story # I got interviewed by Quartz posted November 2021 I got interviewed by Quartz for their Listen to the crypto converts article. Wong is a cryptography engineer at software development company O(1) Labs, working on the Mina cryptocurrency. He writes about crypto and cryptography at his blog, cryptologie.net, and was previously the security lead for Facebook’s Diem. read the full interview here comment on this story # No, cryptocurrencies are not just about Bitcoin: Part 1 posted November 2021 I've been reading a lot of negative discussions around cryptocurrencies on some of the communities that I frequent (e.g. hackernews) and I felt the urge to write something about it. I personally find cryptocurrencies to be an extremely interesting field of innovation because it tightly involves three interesting domains: • The field of cryptography • The field of economics • The current payment system and the financial world Lacking knowledge in any of these fields makes it extremely hard to have a good understanding of the added value of cryptocurrencies. Money is quite an abstract subject (what's a derivative? what's a CDO? what's a margin call? etc.) and it can take decades to understand how a new technology can impact money and society in general. In this post, I give my own opinion of what problems cryptocurrencies can address, while warning readers that I'm mostly involved in the field of cryptography, know a bit about the financial world, and know very little about economics. So take this post with the usual grain of salt. ## The point where Blockchain became mainstream Years ago, I was waiting in line on a beach in Chicago. I had been standing there for hours to try the HTC Vive, a virtual reality headset. What was waiting for me at the end of the line was the future, I believed it before I tried it, and I understood it after experiencing it. Yet, I managed to miss another hint of the coming times, a change that was unraveling in plain sight, invisible to my oblivious eyes. In line, I started chatting with some older chap who got really enthusiastic when I mentioned that I was working in cryptography. "So what do you think about the blockchain?" I had known Bitcoin for a while at that time, but this was the first someone that appeared non-technical talked about a cryptographic term directly. "You mean Bitcoin?" "Yeah" "Ah, well, I'm not sure, what else is there than the cryptocurrency itself?" The dude then started asking me questions about where I thought the "technology" would go and how revolutionary I thought it was. Odd, I remember thinking. I couldn't see why that person was so hyped and I dismissed it as misdirected enthusiasm, thinking that there was really nothing more to Bitcoin than Bitcoin. It's comical, in retrospect, that I failed hard to foresee the boom of the cryptocurrencies. I had invested some, and lost some during the first mtgox crash), I had even made a presentation to my classmates a few years ago as I was doing a masters on cryptography. I had found the idea really cool, perhaps it was one of the reason I ended up studying cryptography? But still, why would anyone see something more than the initial Bitcoin paper? For many years I continued to ignore cryptocurrencies. I didn't get why more and more people would get excited. I even created some passive aggressive page, annoyed by the new use of the term "crypto" (http://cryptoisnotcryptocurrency.com). In 2017, everything changed. At the time, I was working as a security consultant at the Cryptography Services department of NCC Group, and suddenly my team started getting more and more cryptocurrency-related work. And in turn, I got more and more interested and involved in the world of cryptocurrencies, creating the Decentralized Application Security Project Top 10 and participating in the audits of cryptocurrencies like NuCypher, Ethereum, and ZCash. The next thing I know I was leading security for the Diem cryptocurrency (formerly known as Libra) working at Novi (Facebook), two years later I was joining the cryptography team of O(1) Labs to work on the Mina cryptocurrency, and a few month ago publishing Real-World Cryptography, the first cryptography book with a chapter on cryptocurrencies. ## The state of banking today Cryptocurrencies are about the movement of money, and as such it is important to study the current ways and limitations of moving money. Without this context, cryptocurrencies do not make any sense. Today, payment works something like this: for your bank to reach someone else's bank (your friend, your landlord, or a souvenir shop in Thailand) it needs to travel through several banks and systems. In practice, not all banks have partnerships with one another, and so your bank will have to find a route to take to reach the other bank. Different destinations mean different routes. Banks allow movement of money by creating partnerships with other banks. It usually works like this: bank A has an account at bank B, and bank B has an account at bank A. We call that correspondent banking. If bank A wants to send some money to bank B, they'll either credit bank B's account, or bank B will debit bank A's account. (The direction might be decided on how much bank A trusts B, and vice versa.) Of course, not all banks have partnerships, but following the six degrees of separation concept, there'll usually be a chain of banks that connect your bank from your recipient's bank. If you're sending money within the same country, and depending on the country you're living in, you might have access to a government-operated bank that will attempt to partner with most of the territory's banks. A so-called central bank. In the US it's the Federal Reserve System (or the Fed), in France it's Banque de France, etc. Central banks have central banks too, for example, the Bank for International Settlements (BIS). Now, let's talk speed: some of these central banks sometimes have real-time settlement systems, where money can be transferred from one bank to the other almost instantaneously. These are usually called real-time gross settlements (RTGSs), where "gross" means that you're not batching several transactions together. Most banks, as I understand, use deferred net settlements to batch transactions: once (or sometimes several times) a day money moves from one bank to another bank. Thus, to transfer money from your bank to a recipient's bank, your money has to go through a chain of bank accounts. Some of these movements will be settled instantly, some of these movements will be settled after hours or days. This is why payments can be slow depending on the route your money takes. ## Money is (badly) decentralized If you dig deeper, you realize that different banks use different protocols, these protocols are old, badly designed, and errors happen constantly. The only way to make sure that money is not created out of thin air (double spending) is to constantly perform manual audits: having actual humans read balance sheets. One of my friend working at a European bank described to me one of the many incidents they had. To settle with some other bank, they use some XML protocol to communicate how much Euro should move between the banks, and that twice a day. Usually, there's always money moving. Yet, one day, the correspondent bank did not send them any second transfer before the end of the day. My friend's bank took this as a "let's replay the previous settlement" and started crediting thousands of clients with sums of money they had already received. Employees had to work nights to revert everything and make sure that money was not lost. Imagine that. And that's just one bank, every bank is doing their own thing to an extent. There are some standardization effort, as there usually is when things get bad, but they move slowly. It also doesn't fix the elephant in the room: money is too heavily and badly decentralized, and every bank is on a quest to partner with every other bank, creating an$O(n^2)$nightmare. All the solutions that attempt at fixing the system are pretty much trying to centralize more of the system. The fastest points of payments are central banks and other single points of failure like Wise. One interesting development is that some of these central banks are realizing that they need to support instant payments, better security, good interoperability, and perhaps all of that even for people like you and I (the retail). This is what you might have heard as central bank digital currencies (CBDCs). This reduces the$O(n^2)$issue, but it also introduces another one: the issue of trust. Why would central banks trust one another? The more you centralize, the more trust you need to have in that single point of failure. This is the big innovation behind cryptocurrencies: they seek to centralize the financial backbone to facilitate the movement of money. That is, cryptocurrencies are distributed systems, which ultimate goal is to simulate centralized systems while removing single points of failure. In such systems, users all connect to what appears to be, and work like, a single server. No need to connect users between them, in the peer-to-peer way I described previously, now the simulated single machine is connecting everyone together. Money takes the shortest route possible. In some sense, a cryptocurrency is like a central bank that anyone can connect to and that processes transfer in real time, but internally it has no single points of failure, provides insurance based on cryptography, and offers transparency to external users. ## A tiny introduction to the fundamentals Don't run away. I promise this is a tiny, brief, understandable, introduction to the fundamentals of cryptocurrencies. Imagine that a hundred participants decide to create this centralized bank. To do that, they all decide to start from the same ledger, a book of all account balances (and transactions). The first page (or block) might read: Bob has 3 tokens, Alice has 2 tokens This is the genesis, in crypto term. Once each participant has retrieved their copy of the ledger, they need to figure out a way to agree on who gets to add new pages in the book. For example, adding a page could look like: Alice sends 2 tokens to Charles; Bob sends 1 token to Alice And as they add pages, they also need to make absolutely sure that they are all adding the same pages. If a participant adds a different page than everyone else, we have what we call a fork. That participant would start living in a different world. A very simple way to decide on who gets to add pages to the ledger, is to do a round robin: I go first, then you, then him, then her, then me again, then you, etc. That works well if the set of participants is known and fixed (which is what Libra/Diem does, and it's called proof of authority). Other cryptocurrencies might want to dynamically choose who gets to be the next leader based on how much tokens they have (this is called proof of stake, like Tendermint). Now, Bitcoin did things differently, it let everybody have a chance at choosing the next page! To participate, you play a lottery with your computer and if you get the winning ticket you get to decide on the next set of transactions (the next block). This was called proof of work, and as you probably know it, the lottery is computationally expensive to play. Due to that, today, most modern cryptocurrencies use a proof of stake approach. comment on this story # ZK HACK: 1st puzzle write up posted November 2021 Last Tuesday was the start of ZK HACK, a "7-week virtual event featuring weekly workshops and advanced puzzle solving competitions". All related to zero-knowledge proofs, as the name suggests. The talks of the first day were really good, and you can rewatch them here. At the end of the first day, a puzzle called Let's hash it out was released. This post about solving this puzzle. ## The puzzle The puzzle is a Github repo containing a Rust program. If you run it, it displays the following message: Alice designed an authentication system in which users gain access by presenting it a signature on a username, which Alice provided. One day, Alice discovered 256 of these signatures were leaked publicly, but the secret key wasn't. Phew. The next day, she found out someone accessed her system with a username she doesn't know! This shouldn't be possible due to existential unforgeability, as she never signed such a message. Can you find out how it happend and produce a signature on your username? Looking at the code, it looks like there's indeed 256 signatures over 256 messages (that are just hexadecimal strings though, not usernames). ## The signature verification The signatures are BLS signatures (a signature scheme that makes use of pairing and that I've talked about it here). Looking at the code, there's nothing fancy in there. There's a verification function: pub fn verify(pk: G2Affine, msg: &[u8], sig: G1Affine) { let (_, h) = hash_to_curve(msg); assert!(Bls12_381::product_of_pairings(&[ ( sig.into(), G2Affine::prime_subgroup_generator().neg().into() ), (h.into(), pk.into()), ]) .is_one()); } which pretty much implements the BLS signature verification algorithm to check that $$e(\text{sig}, -G_2) \cdot e(\text{h}, \text{pk}) = 1$$ Note: if you read one of the linked resource, "BLS for the rest of us", this should make sense. If anything is confusing in this section, spend a bit of time reading that article. We know that the signature is simply the secret key$sk$multiplied with the message: $$\text{sig} = [\text{sk}]h$$ The public key is simply the secret key$\text{sk}hidden in the second group: $$\text{pk} = [\text{sk}]G_2$$ So the check gives us: \begin{align} & \;e([sk]h, -G_2) \cdot e(h, [sk]G2) \ =& \;e(h, G_2)^{-\text{sk}} \cdot e(h, G_2)^\text{sk} \ =& \;1 \end{align} ## The hash to curve Actually, the username is not signed directly. Since a signature is the first argument in the pairing it needs to be an element of the first group (so[k]G_1$for some value$k$). To transform some bytes into a field element$k$, we use what's called a hash-to-curve algorithm. Here's what the code implements: pub fn hash_to_curve(msg: &[u8]) -> (Vec<u8>, G1Affine) { let rng_pedersen = &mut ChaCha20Rng::from_seed([ 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ]); let parameters = CRH::<G1Projective, ZkHackPedersenWindow>::setup(rng_pedersen).unwrap(); let b2hash = blake2s_simd::blake2s(msg); ( b2hash.as_bytes().to_vec(), CRH::<G1Projective, ZkHackPedersenWindow>::evaluate(¶meters, b2hash.as_bytes()) .unwrap(), ) } What I'm reading is that it: 1. initializes some collision-resistant hash function (CRH) from a hardcoded seed 2. uses the BLAKE2 hash function to hash the username (msg) into 256 bits. 3. Uses the CRH to hash the BLAKE2 digest into a group element Looking at CRH, it's simply a Pedersen hashing (I've talked about that hash function here) that converts a series of bits$b_1, b_2, b_3, \cdots$into $$[b_1]G_{1,1} + [b_2]G_{1, 2} + [b_3]G_{1, 3} + \cdots$$ where the$G_{1,i}$are curve points that belongs to the first group (generated by$G_1$) and derived randomly via the hardcoded seed (in a way that prevents anyone from guessing their discrete logarithm). ## What are we doing? What are we looking for? We're trying to create a valid signature (maybe a signature forgery attack then?) on our own nickname (so more than just an existantial forgery, a chosen-message attack). We can't change the public key (so no rogue key attack), and the message is fixed. This leaves us with the signature as the only thing that can be changed. So indeed, a signature forgery attack. To recap, we have 256 valid signatures: •$e(\text{sig}_1, -G_2) \cdot e(h(m_1), \text{pk}) = 1$•$\vdots$•$e(\text{sig}_{256}, -G_2) \cdot e(h(m_{256}), \text{pk}) = 1and we want to forge a new one such that: $$e(\text{bad_sig}, -G_2) \cdot e(h(\text{"my nickname"}), \text{pk}) = 1$$ ## Forging a signature Reading on the aggregation capabilities of BLS, it seems like the whole point of that signature scheme is that we can just add things with one another. So let's try to think about adding signatures shall we? What happens if I add two signatures? \begin{align} &\; \text{sig}_1 + \text{sig}_2 \ =&\; [\text{sk}]h_1 + [\text{sk}]h_2 \end{align} if only we could factorsk$out... but wait, we know that$h_1$and$h_2are additions of the same curve points (by definition of the Pedersen hashing): \begin{align} &\; \text{sig}_1 + \text{sig}_2 \ =&\; [\text{sk}]h_1 + [\text{sk}]h_2 \ =&\; \text{sk}\ &\; + \text{sk} \end{align} where theb_{i}$(resp.$b'_{i}$) are the bits of$h_1$(resp.$h_2$). So the added signature are equal to the signature of the added bitstrings: $$[b_{1} + b'{1},\; b{2} + b'{2},\; b{3} + b'_{3},\; \cdots]$$ We just forged a signature! Now, that shouldn't mean much, because remember, these bits represent the output of a hash function and a hash function is resistant against pre-image attacks. Wait a minute... Our hash function is a collision-resistant hash function, but that's it. ## Linear combinations OK, so we forged a signature by adding two signatures. But we probably didn't get what we wanted, what we wanted is to obtain the bits$\tilde{b}_1, \tilde{b}_2, \tilde{b}_3, \cdots$that represent the hashing of my own username. Maybe if we add more signatures together we can get? Actually, we can use all the signatures and combine them. And not just by adding them, we can take any linear combination (we're in a field, not constrained by 0 and 1). So here's the system of equations that we have to solve: •$\tilde{b}_1 = x_1 b_1 + x_2 b'_1 + x_3 b''_1$•$\vdots$•$\tilde{b}_{256} = x_1 b_{256} + x_2 b'_{256} + x_3 b''_{256}$Note that we can see that as solving$xA = b$where each row of the matrix$A$represents the bits of a digest, and$b$is the bitvector of my digest (the hash of my username). Once we find that linear combinations, we just have to apply it to the signatures to obtain a signature that should work on my username :) $$\text{bad_sig} = x_1 \text{sig}_1 + x_2 \text{sig}_2 + x_3 \text{sig}_3 + \cdots$$ ## Coding the answer Because I couldn't find a way to solve a system of equations in Rust, I simply extracted what I needed and used Sage to do the complicated parts. Here's the Rust code that creates the matrix$A$and the vector$b$: // get puzzle data let (_pk, ms, _sigs) = puzzle_data(); // here's my name pedersen hashed let (digest_hash, _digest_point) = hash_to_curve("mimoo".as_bytes()); // what's that in terms of bits? use ark_crypto_primitives::crh::pedersen::bytes_to_bits; let bits = bytes_to_bits(&digest_hash); let bits: Vec<u8> = bits.into_iter().map(|b| b as u8).collect(); // order of the subgroup of G1 println!("R = GF(0x73eda753299d7d483339d80809a1d80553bda402fffe5bfeffffffff00000001)"); // xA = b print!("A = matrix(R, ["); for m in ms { let (digest, _point) = hash_to_curve(&m); let bits = bytes_to_bits(&digest); let bits: Vec<u8> = bits.into_iter().map(|x| x as u8).collect(); print!("{:?}, ", bits); } println!("b = vector(R, {:?})", bits); Note: the system of equation is over some field$R$. Why? Because eventually, the linear combination happens between curve points that are elements of the first group, generated by$G_1$, and as such are scalars that live in the field created by the order of the group generated by$G_1\$.

In sage, I simply have to write the following:

x = A.solve_left(b)

and back in Rust we can use these coefficients to add the different signatures with one another:

let mut bad_sig = G1Projective::zero();
for (coeff, sig) in coeffs.into_iter().zip(sigs) {
let coeff = string_int_to_field_el(coeff);
}

// the solution
verify(pk, "mimoo".as_bytes(), bad_sig.into());

It works!

comment on this story

# David Wong's 7 rules of programming posted September 2021

Don't worry about the pretentious title, I was just inspired by Rob Pike's 5 Rules of Programming and by the Risk Manifesto of the The Security Engineer Handbook (which is explained in more details in the book):

Threat modeling over mindless improvements. Secure early rather than later. Forcing software over forcing people. Role play over complacency.

1. Put it in writing. If you leave for a month-long spiritual trip and someone needs to fix a bug in your code, will they have to wait for you? If an expert wants to assess the algorithm you're implementing, but have no knowledge in the programming language you used, will they manage to do it? These are questions you should ask yourself when writing a complex piece of code. It happens often enough that a protocol is implemented differently from the paper. Write specifications!
2. The concorde is fast but dangerous. When writing optimized, faster code, or perhaps over-engineered or overly-clever code, you're trading off maintainability, clarity, and thus security of your code. I often hear programming language features claiming zero-overhead, but they always forget to talk about the cost paid in ambiguity and complexity.
3. Plant types and data structures, then water. As Rob Pike says, use the right data structures, from there everything should flow naturally. Well documented structures are the best abstraction you'll find. Furthermore, functions that make too many assumptions on their inputs will have trouble down the line when someone changes the code underneath them. Encoding invariants in your types is a great way to address this issue. The bottom line is that you can get most of your clarity and security through creating the right types and objects!
4. Don't let your dependencies grow on you. Dependencies easily grow like cancer cells. Beware of them. Analyse your dependencies, monitor and review the ones you use, rewrite or copy code to avoid importing large dependencies, and add friction to your process of adding new dependencies. Supply-chain attacks are being more and more of a thing, be aware.
5. Make the robots enforce it. Enforce good code patterns and prevent bad code patterns through continuous integration where you can. This is pretty much to the "forcing software over forcing people" of the Risk Manifesto.
6. Design lego cities, not real ones. The less connected your code is, the clearer it is to follow as the different parts will be self-contained, and the easier it will be to update some part of the code. Think about parts of your code as external libraries that are facing more than just your codebase.
1 comment

# Project Memento: An NFT art project posted September 2021

Believe it or not, even though I was one of the participant in the standardization of ERC-721, the Ethereum standard for NFTs, I had never read the final standard, yet alone written an ERC-721 smart contract. At the time I was against the standard, at least in its form, due to its collision with the ERC-20 standard (you can't create a smart contract that is both a compliant ERC-20 token and a compliant ERC-721 NFT). My arguments ended up losing, ERC-721 was standardized with the same function names as ERC-20, and I heard that today another standard is surfacing (ERC 1155) to fix this flaw (among others).

Anyway, I recently decided to bury the hatchet and read the damn standard. And after a few weekends of hacking with my friends Eric Khun and Simon Patole, we wrote an ERC-721 compliant dapp as an art project. It's called project memento and it boasts a grid of tiles each containing a letter, a-la million dollar page. People can acquire letters, alter them, and form lines (or diagonals) of words together. I'm curious as to what will become of this, but this was fun!

You can check out project memento here

comment on this story