EIP-6888: Arithmetic verification at EVM level

Check for math overflows and division by zero at EVM level


Metadata
Status: DraftStandards Track: CoreCreated: 2023-04-16
Authors
Renan Rodrigues de Souza (@RenanSouza2)

Abstract


This EIP adds arithmetics checks to EVM arithmetic and a new opcode jump conditionally if there were events. The list of check includes overflows, division by zero.

Motivation


The importance of math checks in smart contract projects is very clear. It was an OpenZeppelin library and then incorporated in Solidity's default behavior. Bringing this to EVM level can combine both gas efficiency and safety.

Specification


The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 and RFC 8174.

Starting from BLOCK_TIMESTAMP >= HARDFORK_TIMESTAMP

Constants

ConstantTypeValue
HARDFORK_TIMESTAMPuint64TBD
UINT_MAXuint2562 ** 256 - 1
INT_MINint256-(2**255)

Flags

VariableTypeInitial Value
carryboolfalse
overflowboolfalse

Two new flags are added to the EVM state: unsigned warning (carry) and signed warning (overflow). The scope of those flags are the same as the program counter.

Definitions

From this point forward a, b and c references the arguments in a math operation and res the output. c is only used if the operation takes 3 inputs.

The function sign(x) is defined in the set of uint256 -> {NEGATIVE, ZERO, POSITIVE}

Contidions

The carry flag MUST be set in the following circumstances:

  • When opcode is ADD (0x01) and res < a
  • When opcode is MUL (0x02) and a != 0 ∧ res / a != b
  • When opcode is SUB (0x03) and b > a
  • When opcode is DIV (0x04) or MOD (0x06); and b == 0
  • When opcode is ADDMOD (0x08) and c == 0
  • When opcode is MULMOD (0x08) and c == 0
  • When opcode is EXP (0x0A) and a ** b > UINT_MAX
  • When opcode is SHL (0x1b) and res >> a != b

The overflow flag MUST be set in the following circumstances:

  • When opcode is ADD (0x01) and a != 0 ∧ b != 0 ∧ sign(a) == sign(b) ∧ sign(a) != sign(res)
  • When opcode is SUB (0x03) and (a != 0 ∧ b != 0 ∧ sign(a) != sign(b) ∧ sign(a) != sign(res)) ∨ (a == 0 ^ b == INT_MIN)
  • When opcode is MUL (0x02) and (a == -1 ∧ b == INT_MIN) ∨ (a == INT_MIN ∧ b == -1) ∨ (a != 0 ∧ (res / a != b)) (this / represents SDIV)
  • When opcode is SDIV (0x05) or SMOD (0x06); and b == 0 ∨ (a == INT_MIN ∧ b == -1)
  • When opcode is SHL (0x1b) and res >> a != b (this >> represents SAR)

Opcodes

JUMPC

Consumes one argument from the stack, the possible pc dest, Conditionally alter the program counter depending on the carry flag. J_JUMPC = carry ? µ_s[0] : µ_pc + 1 Clears both flags. carry = overflow = false

JUMPO

Consumes one argument from the stack, the possible pc dest, Conditionally alter the program counter depending on the ovewflow flag. J_JUMPO = carry ? µ_s[0] : µ_pc + 1 Clears both flags. carry = overflow = false

gas

The gas cost for both instructions is G_high, the same as JUMPI.

Rationale


EVM uses two's complement for negative numbers. The opcodes listed above triggers one or two flags depending if they are used for signed and unsigned numbers.

The conditions described for each opcode is made with implementation friendliness in mind. The only exception is EXP as it is hard to give a concise test as most of the others relied on the inverse operation and there is no native LOG. Most EXP implementations will internally use MUL so the flag carry can be drawn from that instruction, not the overflow.

Both flags are cleaned at the same time because the instructions are expected to be used when transitioning between codes where numbers are treated as signed or unsigned.

Backwards Compatibility


This EIP introduces a new opcode and changes int EVM behavior.

Test Cases


TBD

Reference Implementation


TBD

Security Considerations


This is a new EVM behavior but each code will decide how to interact with it.

Copyright


Copyright and related rights waived via CC0.