How to write a Native Rust Program

To write Solana programs without leveraging the Anchor framework, we use the solana_program crate. This is the base library for writing onchain programs in Rust.

For beginners, it is recommended to start with the Anchor framework.

Program #

Below is a simple Solana program with a single instruction that creates a new account. We'll walk through it to explain the basic structure of a Solana program. Here is the program on Solana Playground.

lib.rs

use borsh::{BorshDeserialize, BorshSerialize};
use solana_program::{
    account_info::{next_account_info, AccountInfo},
    entrypoint,
    entrypoint::ProgramResult,
    msg,
    program::invoke,
    pubkey::Pubkey,
    rent::Rent,
    system_instruction::create_account,
    sysvar::Sysvar,
};
 
entrypoint!(process_instruction);
 
pub fn process_instruction(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    instruction_data: &[u8],
) -> ProgramResult {
    let instruction = Instructions::try_from_slice(instruction_data)?;
    match instruction {
        Instructions::Initialize { data } => process_initialize(program_id, accounts, data),
    }
}
 
pub fn process_initialize(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    data: u64,
) -> ProgramResult {
    let accounts_iter = &mut accounts.iter();
 
    let new_account = next_account_info(accounts_iter)?;
    let signer = next_account_info(accounts_iter)?;
    let system_program = next_account_info(accounts_iter)?;
 
    let account_data = NewAccount { data };
    let size = account_data.try_to_vec()?.len();
    let lamports = (Rent::get()?).minimum_balance(size);
 
    invoke(
        &create_account(
            signer.key,
            new_account.key,
            lamports,
            size as u64,
            program_id,
        ),
        &[signer.clone(), new_account.clone(), system_program.clone()],
    )?;
 
    account_data.serialize(&mut *new_account.data.borrow_mut())?;
    msg!("Changed data to: {:?}!", data);
    Ok(())
}
 
#[derive(BorshSerialize, BorshDeserialize)]
pub enum Instructions {
    Initialize { data: u64 },
}
 
#[derive(BorshSerialize, BorshDeserialize, Debug)]
pub struct NewAccount {
    pub data: u64,
}

Entrypoint #

Every Solana program includes a single entrypoint used to invoke the program. The process_instruction function is then used to process the data passed into the entrypoint. This function requires the following parameters:

program_id - Address of the currently executing program
accounts - Array of accounts needed to execute an instruction.
instruction_data - Serialized data specific to an instruction.

entrypoint!(process_instruction);
 
pub fn process_instruction(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    instruction_data: &[u8],
) -> ProgramResult {
    ...
}

These parameters correspond to the details required for every instruction on a transaction.

Instructions #

While there is only one entrypoint, program execution can follow different paths depending on the instruction_data. It is common to define instructions as variants within an enum, where each variant represents a distinct instruction on the program.

#[derive(BorshSerialize, BorshDeserialize)]
pub enum Instructions {
    Initialize { data: u64 },
}

The instruction_data passed into the entrypoint is deserialized to determine its corresponding enum variant.

pub fn process_instruction(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    instruction_data: &[u8],
) -> ProgramResult {
    let instruction = Instructions::try_from_slice(instruction_data)?;
    match instruction {
        Instructions::Initialize { data } => process_initialize(program_id, accounts, data),
    }
}

A match statement is then used to invoke the function including the logic to process the identified instruction. These functions are often called instruction handlers.

pub fn process_instruction(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    instruction_data: &[u8],
) -> ProgramResult {
    let instruction = Instructions::try_from_slice(instruction_data)?;
    match instruction {
        Instructions::Initialize { data } => process_initialize(program_id, accounts, data),
    }
}
 
pub fn process_initialize(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    data: u64,
) -> ProgramResult {
    ...
    Ok(())
}

Process Instruction #

For every instruction on a program, there exists a specific instruction handler function that implements the logic required to execute that instruction.

pub fn process_initialize(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    data: u64,
) -> ProgramResult {
    let accounts_iter = &mut accounts.iter();
 
    let new_account = next_account_info(accounts_iter)?;
    let signer = next_account_info(accounts_iter)?;
    let system_program = next_account_info(accounts_iter)?;
 
    let account_data = NewAccount { data };
    let size = account_data.try_to_vec()?.len();
    let lamports = (Rent::get()?).minimum_balance(size);
 
    invoke(
        &create_account(
            signer.key,
            new_account.key,
            lamports,
            size as u64,
            program_id,
        ),
        &[signer.clone(), new_account.clone(), system_program.clone()],
    )?;
 
    account_data.serialize(&mut *new_account.data.borrow_mut())?;
    msg!("Changed data to: {:?}!", data);
    Ok(())
}

To access the accounts provided to the program, use an iterator to iterate over the list of accounts passed into the entrypoint through the accounts argument. The next_account_info function is used to access the next item in the iterator.

let accounts_iter = &mut accounts.iter();
 
let new_account = next_account_info(accounts_iter)?;
let signer = next_account_info(accounts_iter)?;
let system_program = next_account_info(accounts_iter)?;

Creating a new account requires invoking the create_account instruction on the System Program. When the System Program creates a new account, it can reassign the program owner of the new account.

In this example, we use a Cross Program Invocation to invoke the System Program, creating a new account with the executing program as the owner. As part of the Solana Account Model, only the program designated as the owner of an account is allowed to modify the data on the account.

let account_data = NewAccount { data };
let size = account_data.try_to_vec()?.len();
let lamports = (Rent::get()?).minimum_balance(size);
 
invoke(
    &create_account(
        signer.key,      // payer
        new_account.key, // new account address
        lamports,        // rent
        size as u64,     // space
        program_id,      // program owner address
    ),
    &[signer.clone(), new_account.clone(), system_program.clone()],
)?;

After the account has been successfully created, the final step is to serialize data into the new account's data field. This effectively initializes the account data, storing the data passed into the program entrypoint.

account_data.serialize(&mut *new_account.data.borrow_mut())?;

State #

Structs are used to define the format of a custom data account type for a program. Serialization and deserialization of account data is commonly done using Borsh.

In this example, the NewAccount struct defines the structure of the data to store in a new account.

#[derive(BorshSerialize, BorshDeserialize, Debug)]
pub struct NewAccount {
    pub data: u64,
}

All Solana accounts include a data field that can be used to store any arbitrary data as a byte array. This flexibility enables programs to create and store customized data structures within new accounts.

In the process_initialize function, the data passed into the entrypoint is used to create an instance of the NewAccount struct. This instance is serialized and stored in the data field of the newly created account.

pub fn process_initialize(
    program_id: &Pubkey,
    accounts: &[AccountInfo],
    data: u64,
) -> ProgramResult {
 
    let account_data = NewAccount { data };
 
    invoke(
        ...
    )?;
 
    account_data.serialize(&mut *new_account.data.borrow_mut())?;
    msg!("Changed data to: {:?}!", data);
    Ok(())
}
...
 
#[derive(BorshSerialize, BorshDeserialize, Debug)]
pub struct NewAccount {
    pub data: u64,
}

Client #

Interacting with Solana programs written in native Rust involves directly building the TransactionInstruction.

Similarly, fetching and deserializing account data requires creating a schema compatible with the on-chain program's data structures.

Info

There are multiple client languages supported. You can find details for Rust and Javascript/Typescript under the Solana Clients of the documentation.

Below, we'll walk through an example demonstrating how to invoke the initialize instruction from the program above.

native.test.ts

describe("Test", () => {
  it("Initialize", async () => {
    // Generate keypair for the new account
    const newAccountKp = new web3.Keypair();
 
    const instructionIndex = 0;
    const data = 42;
 
    // Create instruction data buffer
    const instructionData = Buffer.alloc(1 + 8);
    instructionData.writeUInt8(instructionIndex, 0);
    instructionData.writeBigUInt64LE(BigInt(data), 1);
 
    const instruction = new web3.TransactionInstruction({
      keys: [
        {
          pubkey: newAccountKp.publicKey,
          isSigner: true,
          isWritable: true,
        },
        {
          pubkey: pg.wallet.publicKey,
          isSigner: true,
          isWritable: true,
        },
        {
          pubkey: web3.SystemProgram.programId,
          isSigner: false,
          isWritable: false,
        },
      ],
      programId: pg.PROGRAM_ID,
      data: instructionData,
    });
 
    const transaction = new web3.Transaction().add(instruction);
 
    const txHash = await web3.sendAndConfirmTransaction(
      pg.connection,
      transaction,
      [pg.wallet.keypair, newAccountKp],
    );
    console.log(`Use 'solana confirm -v ${txHash}' to see the logs`);
 
    // Fetch Account
    const newAccount = await pg.connection.getAccountInfo(
      newAccountKp.publicKey,
    );
 
    // Deserialize Account Data
    const deserializedAccountData = borsh.deserialize(
      AccountDataSchema,
      AccountData,
      newAccount.data,
    );
 
    console.log(Number(deserializedAccountData.data));
  });
});
 
class AccountData {
  data = 0;
  constructor(fields: { data: number }) {
    if (fields) {
      this.data = fields.data;
    }
  }
}
 
const AccountDataSchema = new Map([
  [AccountData, { kind: "struct", fields: [["data", "u64"]] }],
]);

Invoke Instructions #

To invoke an instruction, you must manually construct a TransactionInstruction that corresponds with the on-chain program. This involves specifying:

The program ID for the program being invoked
The AccountMeta for each account required by the instruction
The instruction data buffer required by the instruction

// Generate keypair for the new account
const newAccountKp = new web3.Keypair();
 
const instructionIndex = 0;
const data = 42;
 
// Create instruction data buffer
const instructionData = Buffer.alloc(1 + 8);
instructionData.writeUInt8(instructionIndex, 0);
instructionData.writeBigUInt64LE(BigInt(data), 1);
 
const instruction = new web3.TransactionInstruction({
  keys: [
    {
      pubkey: newAccountKp.publicKey,
      isSigner: true,
      isWritable: true,
    },
    {
      pubkey: pg.wallet.publicKey,
      isSigner: true,
      isWritable: true,
    },
    {
      pubkey: web3.SystemProgram.programId,
      isSigner: false,
      isWritable: false,
    },
  ],
  programId: pg.PROGRAM_ID,
  data: instructionData,
});

First, create a new keypair. The publickey from this keypair will be used as the address for the new account created by the initialize instruction.

// Generate keypair for the new account
const newAccountKp = new web3.Keypair();

Before building the instruction, prepare the instruction data buffer that the instruction expects. In this example, the buffer's first byte identifies the instruction to invoke on the program. The additional 8 bytes are allocated for the u64 type data, which is required by the initialize instruction.

const instructionIndex = 0;
const data = 42;
 
// Create instruction data buffer
const instructionData = Buffer.alloc(1 + 8);
instructionData.writeUInt8(instructionIndex, 0);
instructionData.writeBigUInt64LE(BigInt(data), 1);

After creating the instruction data buffer, use it to construct the TransactionInstruction. This involves specifying the program ID and defining the AccountMeta for each account involved in the instruction. This means specifying whether each account is writable and if it is required as a signer on the transaction.

const instruction = new web3.TransactionInstruction({
  keys: [
    {
      pubkey: newAccountKp.publicKey,
      isSigner: true,
      isWritable: true,
    },
    {
      pubkey: pg.wallet.publicKey,
      isSigner: true,
      isWritable: true,
    },
    {
      pubkey: web3.SystemProgram.programId,
      isSigner: false,
      isWritable: false,
    },
  ],
  programId: pg.PROGRAM_ID,
  data: instructionData,
});

Finally, add the instruction to a new transaction and send it to be processed by the network.

const transaction = new web3.Transaction().add(instruction);
 
const txHash = await web3.sendAndConfirmTransaction(
  pg.connection,
  transaction,
  [pg.wallet.keypair, newAccountKp],
);
console.log(`Use 'solana confirm -v ${txHash}' to see the logs`);

Fetch Accounts #

To fetch and deserialize the account data, you need to first create a scheme to match the expected on-chain account data.

class AccountData {
  data = 0;
  constructor(fields: { data: number }) {
    if (fields) {
      this.data = fields.data;
    }
  }
}
 
const AccountDataSchema = new Map([
  [AccountData, { kind: "struct", fields: [["data", "u64"]] }],
]);

Then fetch the AccountInfo for the account using its address.

const newAccount = await pg.connection.getAccountInfo(newAccountKp.publicKey);

Lastly, deserialize the AccountInfo's data field using the predefined schema.

const deserializedAccountData = borsh.deserialize(
  AccountDataSchema,
  AccountData,
  newAccount.data,
);
 
console.log(Number(deserializedAccountData.data));