Skip to content

Commit

Permalink
Add EIP: EOF - Prepare for Address Space Extension
Browse files Browse the repository at this point in the history
Merged by EIP-Bot.
  • Loading branch information
shemnon committed Apr 26, 2024
1 parent acac81a commit 6a0a792
Showing 1 changed file with 120 additions and 0 deletions.
120 changes: 120 additions & 0 deletions EIPS/eip-7676.md
Original file line number Diff line number Diff line change
@@ -0,0 +1,120 @@
---
eip: 7676
title: EOF - Prepare for Address Space Extension
description: Update EOF opcodes so addresses are not trimmed during execution
author: Danno Ferrin (@shemnon)
discussions-to: https://ethereum-magicians.org/t/eof-prepare-for-address-space-extension/19537
status: Draft
type: Standards Track
category: Core
created: 2024-04-03
requires: 3540, 3670, 7069
---

## Abstract

Operations in the Legacy EVM trim off the top 12 bytes of an address operand before evaluation. This
EIP changes the handling of those opcodes within EOF so that no trimming occurs and the top twelve
bytes need to be zero or an exceptional halt is raised.

## Motivation

There have been proposals to extend Ethereum Addresses from 160 bits to 256, such as one that would
use the extra bits for state expiry (such as the ethereum magicians forum topic "Increasing the
address size from 20 to 32 bytes"). One issue ground the work to a halt: EVM opcodes that accept
addresses trim all but the lowest 20 bytes out from the operand before processing. EVM Reference
tests verify this behavior in the 'stBadOpcode/invalidAddr.json' tests.

The EVM Object Format presents an opportunity to remove this address masking in a backwards
compatible way, by baking it into the format definition from the start.

Most of the work is already in place. The following 5 operations have already been banned from
EOF: `EXTCODESIZE`, `EXTCODECOPY`, `EXTCODEHASH`, `CALLCODE`, and `SELFDESTRUCT`. Three call
operations, `CALL`, `DELEGATECALL`, and `STATICCALL` are being revamped
in [EIP-7069](./eip-7069.md). That leaves only one operation, `BALANCE`, with issues.

When future uses of address space extension are specified it is expected that the exceptional halt
behavior will be modified.

## Specification

We introduce one new instruction:

- `EXTBALANCE` (`tbd`) with arguments `(target_address)`, returning `balance`

`EXTBALANCE` will pop one stack item off of the stack, the address of another account or contract.
The balance of that account or contract will be pushed onto the stack.

If `EXTBALANCE` is invoked with any of the high 12 bytes in `target_address` set to a non-zero value
the operation will cause an exceptional halt. All gas in the current frame will be consumed on
failure, no gas schedule change is needed.

The gas cost of `EXTBALANCE` will be costed according to the gas schedule for the `BALANCE`
oepration specified in [EIP-2929](./eip-2929.md). This means if the account is not
in `accessed_addresses` then it is charged 2600 gas and added to `accessed_addresses`. If the
account is in `accessed_addresses` then 100 gas is charged. The `accessed_addresses` is shared with
all other operands described in [EIP-2929](./eip-2929).

Also, for the `EXTCALL`, `EXTDELEGATECALL`, and `EXTSTATICCALL` operations a new step is added just
before the memory expansion check: halt with exceptional failure if any of the high 96 bits of
the `target_address` are set.

It is anticipated that future EIPs will alter the behavior in a way that will not result in an
exceptional halt. Contract implementors should not depend on the exceptional halt.

## Rationale

### New Opcode

There is no need to ban the `BALANCE` opcode as it does not cause any problems that would require
banning it within an EOF container. Adding a new opcode also allows the existing opcode to behave
the same in EOF and legacy code, reducing potential friction points for end user confusion and bugs.

### Revert on invalid address

There are two alternative ways to handle accounts with high bits set. The specification calls for
an exceptional halt, but the alternative was to treat the account as empty. The reason the "empty
account" approach was rejected is twofold:

- The `accessed_addresses` list could be required to track 256 bit accounts when an invalid address
is accessed.
- The `EXTCALL` series instructions could still send balance to those addresses and such accounts
would then hold an (inaccessible) balance that would need to be reflected in the merkle trie.

### No change in gas costing

Because the BALANCE operation already needs to check `accessed_addresses` there is already a good
amount of processing that must go into the operation. Hence, no changes to the gas schedule are
needed to prevent abuse of the failures. Such incremental costs will be dominated by costs related
to reverts and address checking for valid accounts.

## Backwards Compatibility

Preparing for Address Space Expansion is explicitly done inside the scope of EOF with the intent
that it will not require changes in old contracts, but with the caveat that old contracts may not be
able to use addresses in the expanded space.

Future EIPs should be mindful when adding new operation with an address operand or parameter to how
it interacts with EOF and legacy contracts. Authors should ensure that such additions remain in
harmony with this EIP.

## Test Cases

Test cases similar to `invalidAddr.json` tests in the standard reference tests will need to be
written for the EOF tests, except they would check for halts on invalid addresses.

## Reference Implementation

TBD <!-- TODO -->

## Security Considerations

This EIP only defines a revert behavior for previously stripped addresses. Compilers will need to be
aware of the need to mask addresses coming in from call data. Some of this is already present in
existing Solidity ABI standards, but more care should be taken in examining the flow
around `EXTBALANCE` and code for `EXTCALL` operations to ensure that compiled code strips the high
bytes.

## Copyright

Copyright and related rights waived via [CC0](../LICENSE.md).

0 comments on commit 6a0a792

Please sign in to comment.