EIPs/EIPS/eip-3779.md

eip	title	description	status	type	category	author	discussions-to	created	requires
3779	Safer Control Flow for the EVM	Ensure a minimal level of safety for EVM code.	Review	Standards Track	Core	Greg Colvin (@gcolvin), Greg Colvin <greg@colvin.org>	https://ethereum-magicians.org/t/eip-3779-safe-control-flow-for-the-evm/6975	2021-08-30	3540

Abstract

This EIP specifies validation rules for some important safety properties, including

• valid instructions,
• valid jump destinations,
• no stack underflows, and
• no stack overflows without recursion.

Valid contracts will not halt with an exception unless they either run out of gas, attempt to dynamically jump to an unvalidated destination, or overflow stack during a recursive subroutine call.

Code must validated at contract creation time – not runtime – by the provided algorithm or its equivalent. Therefore, valid jump destinations for dynamic jumps are specified in an EOF container section. This allows for a one-pass algorithm that is (and must be) linear in the size of the code, so as not to be a DoS vulnerability.

Motivation

Validating safe control flow at creation time has a few important advantages.

• Jump destination analysis does not need to be performed at runtime, thus improving performance and preventing denial of service attacks.
• Jump destination validity does not always need to be checked for at runtime, improving performance.
• Stack underflow does not ever need to be checked for at runtime, improving performance.

Dynamic jumps, where the destination of a JUMP or JUMPI is not known until runtime, can be an obstacle to statically proving this sort of safety, but have also been seen as necessary to implement the return jump from a subroutine. But consider this example of calling a minimal subroutine

ADD:
    RTN_ADD 
    0x02
    0x03
    ADDITION
    jump
RTN_ADD:
    jumpdest
    swap1
    jump

ADDITION:
    jumpdest
    add
    swap1
    jump 

Note that the return address and the destination address are pushed on the stack as constants, so the JUMP instructions are in fact static, not dynamic – they jump to the same PC on every run. We do not need (nor typically use) dynamic jumps to implement subroutines.

Since many of the jumps we need in practice are static we can validate their safety with a static analysis of the code. And since can, we should.

Still, providing for the safe use of dynamic jumps makes for concise and efficient implementations of language constructs like switches and virtual functions.

Dynamic jumps can be an obstacle to linear-time validation of EVM bytecode. But even where jumps are dynamic it is possible to tabulate valid destinations in advance, and the Ethereum Object Format gives us a place to store such tables. So again, we can validate their safety with a static analysis of the code, and we should.

Specification

Dependencies

We need EIP-3540: EVM Object Format (EOF) to support container sections.

EOF container changes

1. A new EOF section called jumptable (section_kind = 3) is introduced. It contains a sequence of n tuples (jumpsrc, jumpdest_i, sorted in ascending lexicographic order. Each tuple represents a valid jump from one location in the bytecode to another.
2. The jumpsrc and jumpdest values are encoded as 3-byte, MSB-first, unsigned integers.

Validity

We define safety here as avoiding exceptional halting states:

• Valid contracts will not halt with an exception unless they
  • run out of gas or
  • overflow stack while making a recursive subroutine call.

Attempts to create contracts that cannot be proven to be valid will fail.

Exceptional Halting States

Execution is as defined in the Yellow Paper a sequence of changes to the EVM state. The conditions on valid code are preserved by state changes. At runtime, if execution of an instruction would violate a condition the execution is in an exceptional halting state. The Yellow Paper defines five such states.

1. Insufficient gas
2. More than 1024 stack items
3. Insufficient stack items
4. Invalid jump destination
5. Invalid instruction

We would like to consider EVM code valid iff no execution of the program can lead to an exceptional halting state, but we must be able to validate code in linear time to avoid denial of service attacks. So in practice, we can only partially meet these requirements. Our validation algorithm does not consider the codes data and computations, only its control flow and stack use. This means we will reject programs with any invalid code paths, even if those paths are not reachable at runtime. Further, conditions 1 and 2 Insufficient gas and stack overflow must in general be checked at runtime. Conditions 3, 4, and 5 cannot occur if the code conforms to the following rules.

The Rules

This section extends the contact creation validation rules (as defined in EIP-3540.)

0. All instructions are valid.
1. Every JUMP and JUMPI address a valid JUMPDEST,
   • either static or matching a tuple in the jumptable.
2. The stack depth is
   • always positive and
   • the same on every path through a bytecode.
3. The stack pointer is always positive and at most 1024.

We need to define matching tuple. A tuple matches a jmpsrc to a jumpdest if the first element of the tuple is the offset of a JUMP or JUMPI opcode in the bytecode, and the second element is the address of a JUMPDEST opcode in the bytecode.

We need to define stack depth. The Yellow Paper has the stack pointer (SP) pointing just past the top item on the data stack. We define the stack base as the element that the SP addressed at the entry to the current basic block, or 0 on program entry. So we can define the stack depth as the number of stack elements between the current SP and the current stack base.

An Algorithm

This section specifies an algorithm for checking the above the rules. Equivalent code must be run at creation time (as defined in EIP-3540.)

The following is a pseudo-Go implementation of an algorithm for enforcing program validity. This algorithm is a symbolic execution of the program that recursively traverses the bytecode, following its control flow and stack use and checking for violations of the rules above. It uses a stack to track the slots that hold PUSHed constants, from which it pops the destinations to validate during the analysis.

This algorithm runs in time equal to O(vertices + edges) in the program's control-flow graph, where vertices represent control-flow instructions and the edges represent basic blocks – thus the algorithm takes time proportional to the size of the bytecode.

For simplicity's sake we assume a few helper functions.

• advance_pc() advances the PC, skipping any immediate data.
• imm_data() returns immediate data for an instruction.
• valid_jumpdest() tests whether
  • dynamic jumps match a valid (jumpsrc, jumpdest) in the jumptable and
  • all jump destinations are JUMPDEST bytes and not in immediate data.
• add_items() and remove_items() push or pop the const_stack items for an instruction, and return its effect on the stack pointer. add_items() will PUSH the value of constants – all other stack items are zeroed.

var bytecode [code_len]byte
var stack_depth [code_len]unsigned
var const_stack [1024]unsigned
var SP := 0
var BP := 0
struct { int src; int dst } jumps
var jumptable [n_jumps]jumps

func validate(PC :=0, stack_depth:=0) boolean {

   for ; PC < code_len; PC = advance_pc(PC, instruction) {

      // valid instructions only
      instruction := bytecode[PC]
      if !valid_instruction(instruction) {
         return false;
      }

      // if stack depth for `PC` is non-zero we have been here before 
      // so return to break cycle in control flow graph
      if stack_depth[PC] != 0 {
         if stack_depth[PC] == stack_depth
            return true
         else
            return false
      }
      stack_depth[PC] = stack_depth

      if instruction == JUMP {

         // check for valid destination
         dest = const_stack[SP]
         if !valid_jumpdest(dest) {
            return false
         }

         // reset PC to destination of jump 
         PC += jumpdest
         BP = SP
         --SP
         continue
      }
      if instruction == JUMPI {

         // check for valid destination
         jumpdest = stack[SP]
          if !valid_jumpdest(dest) {
            return false
         }
         
         // false side of if
         if stack[SP - 1] == 0 {
            BP = SP
            continue
         }

         // reset PC to destination of jump 
         BP = SP
         --SP
         PC = jumpdest

         // recurse to jump to code to validate
         if !validate(PC), stack_depth {
            return false
         }
         continue 
      }

      // check effect of instruction on stack
      SP -= remove_items(instruction)
      SP += add_items(instruction)
      stack_depth := SP - BP
      if SP < 0 || 1024 < SP || stack_depth < 0 {
         return false
      }

      // successful validation of path
      if PC > code_len
            || instruction == STOP
            || instruction == RETURN
            || instruction == SUICIDE {
         return true
      }
   }
}

Rationale

The alternative to checking validity at creation time is checking it at runtime. This hurts performance and is a denial of service vulnerability. Thus the above rules and accompanying one-pass validation algorithm.

Rule 1 – requiring static or previously tabulated destinations for JUMP and JUMPI – simplifies static jumps and constrains dynamic jumps.

• Jump destinations are currently checked at runtime, but static jumps can be validated at creation time.
• Requiring the possible destinations of dynamic jumps to be tabulated in advance allows for tractable bytecode traversal for creation-time validation: a jump table takes up space proportional to the number of jump destinations, so attempting to attack the validation algorithm with large numbers of jump destinations will also reduce the available space for code to be validated.

Rule 2 – requiring positive, consistent stack depth – ensures sufficient stack. Exceptions can be caused by irreducible paths like jumping into loops and subroutines, and by calling subroutines with insufficient numbers of arguments.

Rule 3 – bounding the stack pointer – catches all stack overflows that occur without recursion.

Taken together, these rules allow for code to be validated by traversing the control-flow graph, following each edge only once.

Backwards Compatibility

These changes affect the semantics of existing EVM code – the use of JUMP, JUMPI, and the stack are restricted, such that some existing code that run correctly will nonetheless become invalid EOF code sections.

Security Considerations

This EIP is intended to ensure a minimal level of safety for EVM code deployed on the blockchain.

Copyright

Copyright and related rights waived via CC0.