EIP-7921: Skip `JUMPDEST` immediate argument check

All `JUMPDEST` bytes become valid `JUMPDEST`

Metadata

Status: DraftStandards Track: CoreCreated: 2025-03-26

Authors

William Morriss (@wjmelements)

Links

GitHub page Discussion on Ethereum Magicians

Abstract

Allow JUMP and JUMPI to arrive at any byte matching JUMPDEST (0x5b), even if that byte is an immediate argument.

Motivation

Immediate arguments are opcode parameters supplied within the code rather than the stack. Currently determining the validity of a JUMPDEST requires determining which bytes are immediate arguments to other opcodes, such as PUSH1. This presents several problems:

Codesize is a linear DoS vector because code must be preprocessed to determine JUMPDEST validity.
New opcodes with immediate arguments cannot be safely adopted.
CODECOPY data spans can invalidate subsequent JUMPDEST.

The rationale for this JUMPDEST validity check is to prevent unintended code execution. However, almost all JUMP and JUMPI target constant destinations. Removing this check allows larger programs and better opcodes. Therefore, the cost of this safety check outweighs the benefit.

Specification

When activated, all 0x5b bytes are valid JUMPDEST for JUMPI and JUMP opcodes.

Rationale

Removing the check solves several problems while reducing EVM complexity.

Backwards Compatibility

Code previously only had one interpretation for disassembly. With this change, a JUMPDEST located inside an immediate argument can cause multiple disassembly interpretations. Usually the interpretations will converge after a few bytes but the length of such a dispute can be unbounded. CODECOPY data has always been difficult to identify. It is recommended that disassemblers provide all possible interpretations in their output in order to reveal possible underhanded functionality.

Security Considerations

Current contracts performing dynamic jumps may gain new unintended functionality if it is possible to jump to an immediate argument containing JUMPDEST. It is expected that very few contracts will become vulnerable in this way. Most smart contract programming languages do not even allow dynamic jumps, and of those that do, few will have JUMPDEST in an accessible immediate argument. Therefore it is expected that few contracts will become vulnerable, and for many of them the newly possible codepaths will contain invalid opcodes. A static analysis tool should be developed and made publicly available to test if a contract might become vulnerable, and the program should be run for all current contracts in order to notify projects about potential security issues. Affected programs will have ample time to migrate.