Skip to main content

On-Chain Randomness

Generating pseudo-random values in Move is similar to solutions in other languages. A Move function can create a new instance of RandomGenerator and use it for generating random values of different types, for example, generate_u128(&mut generator), generate_u8_in_range(&mut generator, 1, 6), or,

entry fun roll_dice(r: &Random, ctx: &mut TxContext): Dice {
let generator = new_generator(r, ctx); // generator is a PRG
Dice { value: random::generate_u8_in_range(&mut generator, 1, 6) }
}

Random has a reserved address 0x8. See random.move for the Move APIs for accessing randomness on Sui.

note

Although Random is a shared object, it is inaccessible for mutable operations, and any transaction attempting to modify it fails.

Having access to random numbers is only one part of designing secure applications, you should also pay careful attention to how you use that randomness. To securely access randomness:

  • Define your function as (private) entry.
  • Prefer generating randomness using function-local RandomGenerator.
  • Make sure that the "unhappy path" of your function does not charge more gas than the "happy path".

Use (non-public) entry functions

While composition is very powerful for smart contracts, it opens the door to attacks on functions that use randomness. Consider for example the next module:

module games::dice { 
...
struct GuessedCorrectly has drop { ... };

/// If you guess correctly the output you get a GuessedCorrectly object.
public fun play_dice(guess: u8, fee: Coin<SUI>, r: &Random, ctx: &mut TxContext): Option<GuessedCorrectly> {
// Pay for the turn
assert!(coin::value(&fee) == 1000000, EInvalidAmount);
transfer::public_transfer(fee, CREATOR_ADDRESS);
// Roll the dice
let generator = new_generator(r, ctx);
if (guess == random::generate_u8_in_range(&mut generator, 1, 6)) {
option::some(GuessedCorrectly {})
} else {
option::none()
}
}
...
}

An attacker can deploy the next function:

public fun attack(guess: u8, r: &Random, ctx: &mut TxContext): GuessedCorrectly {
let output = dice::play_dice(guess, r, ctx);
option::extract(output) // reverts the transaction if roll_dice returns option::none()
}

The attacker can now call attack with a guess, and always revert the fee transfer if the guess is incorrect.

To protect against composition attacks in this example, define play_dice as a private entry function so functions from other modules cannot call it, such as,

entry fun play_dice(guess: u8, fee: Coin<SUI>, r: &Random, ctx: &mut TxContext): Option<GuessedCorrectly> {
...
}
note

The Move compiler enforces this behavior by rejecting public functions with Random as an argument.

Programmable transaction block (PTB) restrictions

A similar attack to the one previously described involves PTBs even when play_dice is defined as a private entry function. For example, consider the entry play_dice(guess: u8, fee: Coin<SUI>, r: &Random, ctx: &mut TxContext): Option<GuessedCorrectly> { … } function defined earlier, the attacker can publish the function

public fun attack(output: Option<GuessedCorrectly>): GuessedCorrectly {
option::extract(output)
}

and send a PTB with commands play_dice(...), attack(Result(0)) where Result(0) is the output of the first command. As before, the attack takes advantage of the atomic nature of PTBs and always reverts the entire transaction if the guess was incorrect, without paying the fee. Sending multiple transactions can repeat the attck, each one executed with different randomness and reverted if the guess is incorrect.

note

To protect against PTB-based composition attacks, Sui rejects PTBs that have commands that are not TransferObjects or MergeCoins following a MoveCall command that uses Random as an input.

Instantiating RandomGenerator

RandomGenerator is secure as long as it's created by the consuming module. If passed as an argument, the caller might be able to predict the outputs of that RandomGenerator instance (for example, by calling bcs::to_bytes(&generator) and parsing its internal state).

note

The Move compiler enforces this behavior by rejecting public functions with RandomGenerator as an argument.

Limited resources and Random dependent flows

Developers should be aware that some resources that are available to transactions are limited. If a function that reads Random consumes more resources in the unhappy path flow than in the happy path flow, an attacker can use that difference to revert the transaction in the unhappy flow as previously demonstrated. Concretely, gas is such a resource. Consider the following code:

entry fun play(r: &Random, payment: Coin<SUI>, ...) {
...
let generator = new_generator(r, ctx);
let win = random::generate_bool(generator);
if (win) { // happy flow
... cheap computation ...
} else {
... very expensive computation ...
}
}

Observe that the gas costs of a transaction that calls play depends on the value of win - An attacker could call this function with a gas object that has enough balance to cover the happy flow but not the unhappy one, resulting in it either winning or reverting the transaction (but never losing the payment).

In many cases this is not an issue, like when selecting a raffle winner, lottery numbers, or a random NFT. However, in the cases where it can be problematic, you can do one of the following:

  • Write the function in a way that the happy flow consumes more gas than the unhappy one.
    • Keep in mind that external functions or native ones can change in the future, potentially resulting in different costs compared to the time you conducted your tests.
    • Use profile-transaction on Testnet transactions to verify the costs of different flows.
  • Split the logic in two: one function that fetches a random value and stores it in an object, and another function that reads that stored value and completes the operation. The latter function might indeed fail, but now the random value is fixed and cannot be modified using repeated calls.

See random_nft.move for examples.

Other limited resources per transaction are:

  • The number of new objects.
  • The number of objects that can be used (including dynamic fields).

Accessing Random from TypeScript

If you want to call roll_dice(r: &Random, ctx: &mut TxContext) in module example, use the following code snippet:

const txb = new TransactionBlock();
txb.moveCall({
target: "${PACKAGE_ID}::example::roll_dice",
arguments: [txb.object('0x8')]
});
...