Skip to content

Philogy/evm-glue

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

49 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🫠 EVM Glue

EVM Glue ("Glue") is an EVM assembler written in Rust. Glue gives you a simple API that allows you to insert opcodes, raw bytes, or so-called "references" (e.g. jump destinations) and assembles it into EVM bytecode.

Features

  • ✅ Base Assembly Building Blocks
    • ✅ Opcodes
    • ✅ Raw Bytecode
    • ✅ Marks & References (e.g. for jump destinations, code sections)
  • ✅ Assembly Optimizations
    • ✅ Reference push/literal size reduction (e.g. reduce PUSH2 0x0017 => PUSH1 0x17 for references)

Other TODOs:

  • Add tests
  • Add more tests
  • Simple parser / frontend

Quickstart, A full example:

use evm_glue::assembler::assemble_minimized;
use evm_glue::assembly::Asm;
use evm_glue::data;
use evm_glue::opcodes::Opcode::*;
use evm_glue::utils::MarkTracker;
use hex_literal::hex as hx;
use Asm::*;

fn main() {
    let mut runtime_marks = MarkTracker::default();
    let empty_revert = runtime_marks.next_mark();

    let mut runtime = vec![
        // Load x, y
        Op(PUSH0),
        Op(CALLDATALOAD), // x
        Op(PUSH1(hx!("20"))),
        Op(CALLDATALOAD), // x, y
        // Add and check for overflow
        Op(DUP2),                // x, y, x
        Op(ADD),                 // x, r
        Op(DUP1),                // x, r, r
        Op(SWAP2),               // r, r, x
        Op(GT),                  // r, x > r
        Asm::mref(empty_revert), // r, x > r, l
        Op(JUMPI),               // r
        // Return result.
        Op(MSIZE),
        Op(MSTORE),
        Op(MSIZE),
        Op(PUSH0),
        Op(RETURN),
        // Revert
        Mark(empty_revert),
        Op(JUMPDEST),
        Op(PUSH0),
        Op(PUSH0),
        Op(REVERT),
    ];

    let (_, runtime_bytecode) = assemble_minimized(&runtime, true).unwrap();

    let mut deploy_marks = MarkTracker::default();
    let runtime_start = deploy_marks.next_mark();
    let runtime_end = deploy_marks.next_mark();

    let mut deploy = vec![
        // Constructor
        Asm::delta_ref(runtime_start, runtime_end), // rt_size
        Op(DUP1),                                   // rt_size, rt_size
        Asm::mref(runtime_start),                   // rt_size, rt_size, rt_start
        Op(RETURNDATASIZE),                         // rt_size, rt_size, rt_start, 0
        Op(CODECOPY),                               // rt_size
        Op(RETURNDATASIZE),                         // rt_size, 0
        Op(RETURN),                                 // -- end
        // Runtime body
        Mark(runtime_start),
        Data(runtime_bytecode.clone()),
        Mark(runtime_end),
    ];

    let (_, deploy_bytecode) = assemble_minimized(&deploy, true).unwrap();

    println!("runtime bytecode: {}", hex::encode(runtime_bytecode));
    println!("deploy bytecode: {}", hex::encode(deploy_bytecode));
}

API

The overarching API is that you create a list of assembly blocks (Vec<Asm>) and feed it into assembler::assemble_minimized that then turns it into one long string of bytecode. Blocks and their types are found under evm_glue::assembly.

Assembling (assembler::assemble_minimized)

The assemble_minimized function is the main entry point for assembling a string of assembly blocks. It takes two arguments:

  • asm: &mut Vec<Asm>: This is the array containing all the assembly
  • allow_push0: bool: A boolean flag indicating whether you want to allow generated PUSHx 0 opcodes to become PUSH0. Note this will not affect any explicitly defined pushes. Values like Asm::Op(Opcode:PUSH3([0, 0, 0])) will remain untouched.

[!INFO] If you want a slightly simpler & marginally faster compilation process you can use assemble_maximized which will use the same maximal size for all references.

Op

Op "blocks" indicate individual opcodes. For PUSH1-PUSH32 operations this already includes their immediate push value e.g.:

use evm_glue::assembly::Asm::*;
use evm_glue::opcodes::Opcode::*;
// `hex_literal` is a recommended helper library
use hex_literal::hex;

let asm = vec![
    Op(PUSH1(hex!("20"))),
    Op(CALLDATALOAD),
    Op(DUP2),
    Op(PUSH5(hex!("0000138302"))),
];

Marks & References

Probably one of the most useful features of the assembler are its marks and references, these allow you to easily construct jump labels, tables and code returns.

Marks

Marks generate 0 bytecode and are simply an indicator for the assembler. They consist of a simple usize "mark id" (referenced as mid in the code):

use evm_glue::assembly::Asm::*;

let asm = vec![
    Mark(3)
];

Important

If you're manually allocating mark IDs ensure that they start at 0 and stay in a concise range. This is for performance reasons as a simple Vec is used under the hood as the mark => offset map.

MarkTracker helper

The MarkTracker is a simple helper struct that allows you to easily allocate mark IDs without worrying about duplicates:

use evm_glue::utils::MarkTracker;
use evm_glue::assembly::Asm::*;

let mut mt = MarkTracker::default();

let label1 = mt.next_mark();       // 0
let empty_revert = mt.next_mark(); // 1

let asm = vec![
    Mark(label1),
    // ...
    Mark(empty_revert),
    // ...
];

Warning

Marks must be unique, marks with duplicate Mark IDs will result in an error.

Mark References

References are objects that allow for the insertion of mark offsets into the final bytecode at compile time. This can be used to push jump labels to the stack or reference tables. There are 2 types of references:

  • Direct references: a single mark directly and insert their bytecode offset at compile time
  • Delta references: the offset between a start and end mark

Beyond that marks are also differentiated by their representation type. References can either be Pushed meaning they'll generate an actual PUSH1-PUSH32 opcode together with the reference value at compile time or Literal meaning they'll be inserted into the bytecode standalone. The former is useful for jump labels, code and data offsets while the later is useful for constructing e.g. jump tables.

These also have helpers:

use evm_glue::{assembly::Asm, utils::MarkTracker};

let mut deploy_marks = MarkTracker::default();
let runtime_start = deploy_marks.next_mark();
let runtime_end = deploy_marks.next_mark();
let asm = vec![
    Asm::delta_ref(runtime_start, runtime_end), // Pushed delta reference
    // ...,
    Asm::mref(runtime_start),                   // Pushed direct reference
    // ...,
    Asm::Mark(runtime_start),
];

Literal helpers are Asm::mref_literal and Asm::delta_ref_literal respectively.

Data

Data blocks are just continuous sequences of bytecode to be directly inserted into the final output. This can be other compiled contracts, metadata or even lookup tables.

To insert a byte stream (Vec<u8>) as data you can use the Asm::Data enum member:

use evm_glue::assembly::Asm::*;

let runtime_bytecode: Vec<u8> = vec![0x00];

let asm = vec![
    Data(runtime_bytecode.clone())
];

data! macro

You can use the evm_glue::data macro to directly specify a hex literal as data:

use evm_glue::{data, assembly::Asm};

let asm = vec![
    data!("0283")
];

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages