Merge pull request #7325 from ethereum/arbitraryMemoryForInterpreter

Change interpreter memory to be non-contiguous.
This commit is contained in:
chriseth 2019-09-02 14:27:01 +02:00 committed by GitHub
commit 09b03fa07d
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
8 changed files with 49 additions and 36 deletions

View File

@ -121,7 +121,6 @@ string YulInterpreterTest::interpret()
InterpreterState state;
state.maxTraceSize = 10000;
state.maxSteps = 10000;
state.maxMemSize = 0x20000000;
Interpreter interpreter(state, EVMDialect::strictAssemblyForEVMObjects(langutil::EVMVersion{}));
try
{

View File

@ -25,14 +25,12 @@ void yulFuzzerUtil::interpret(
shared_ptr<yul::Block> _ast,
Dialect const& _dialect,
size_t _maxSteps,
size_t _maxTraceSize,
size_t _maxMemory
size_t _maxTraceSize
)
{
InterpreterState state;
state.maxTraceSize = _maxTraceSize;
state.maxSteps = _maxSteps;
state.maxMemSize = _maxMemory;
Interpreter interpreter(state, _dialect);
interpreter(*_ast);
state.dumpTraceAndState(_os);

View File

@ -30,12 +30,10 @@ struct yulFuzzerUtil
std::shared_ptr<yul::Block> _ast,
Dialect const& _dialect,
size_t _maxSteps = maxSteps,
size_t _maxTraceSize = maxTraceSize,
size_t _maxMemory = maxMemory
size_t _maxTraceSize = maxTraceSize
);
static size_t constexpr maxSteps = 100;
static size_t constexpr maxTraceSize = 75;
static size_t constexpr maxMemory = 0x200;
};
}
}

View File

@ -63,13 +63,11 @@ u256 readZeroExtended(bytes const& _data, u256 const& _offset)
/// Copy @a _size bytes of @a _source at offset @a _sourceOffset to
/// @a _target at offset @a _targetOffset. Behaves as if @a _source would
/// continue with an infinite sequence of zero bytes beyond its end.
/// Asserts the target is large enough to hold the copied segment.
void copyZeroExtended(
bytes& _target, bytes const& _source,
map<u256, uint8_t>& _target, bytes const& _source,
size_t _targetOffset, size_t _sourceOffset, size_t _size
)
{
yulAssert(_targetOffset + _size <= _target.size(), "");
for (size_t i = 0; i < _size; ++i)
_target[_targetOffset + i] = _sourceOffset + i < _source.size() ? _source[_sourceOffset + i] : 0;
}
@ -176,7 +174,7 @@ u256 EVMInstructionInterpreter::eval(
return u256("0x1234cafe1234cafe1234cafe") + arg[0];
uint64_t offset = uint64_t(arg[0] & uint64_t(-1));
uint64_t size = uint64_t(arg[1] & uint64_t(-1));
return u256(keccak256(bytesConstRef(m_state.memory.data() + offset, size)));
return u256(keccak256(readMemory(offset, size)));
}
case Instruction::ADDRESS:
return m_state.address;
@ -251,16 +249,16 @@ u256 EVMInstructionInterpreter::eval(
// --------------- memory / storage / logs ---------------
case Instruction::MLOAD:
if (accessMemory(arg[0], 0x20))
return u256(*reinterpret_cast<h256 const*>(m_state.memory.data() + size_t(arg[0])));
return readMemoryWord(arg[0]);
else
return 0x1234 + arg[0];
case Instruction::MSTORE:
if (accessMemory(arg[0], 0x20))
*reinterpret_cast<h256*>(m_state.memory.data() + size_t(arg[0])) = h256(arg[1]);
writeMemoryWord(arg[0], arg[1]);
return 0;
case Instruction::MSTORE8:
if (accessMemory(arg[0], 1))
m_state.memory[size_t(arg[0])] = uint8_t(arg[1] & 0xff);
m_state.memory[arg[0]] = uint8_t(arg[1] & 0xff);
return 0;
case Instruction::SLOAD:
return m_state.storage[h256(arg[0])];
@ -319,7 +317,7 @@ u256 EVMInstructionInterpreter::eval(
{
bytes data;
if (accessMemory(arg[0], arg[1]))
data = bytesConstRef(m_state.memory.data() + size_t(arg[0]), size_t(arg[1])).toBytes();
data = readMemory(arg[0], arg[1]);
logTrace(_instruction, arg, data);
throw ExplicitlyTerminated();
}
@ -455,12 +453,7 @@ bool EVMInstructionInterpreter::accessMemory(u256 const& _offset, u256 const& _s
{
u256 newSize = (_offset + _size + 0x1f) & ~u256(0x1f);
m_state.msize = max(m_state.msize, newSize);
if (newSize < m_state.maxMemSize)
{
if (m_state.memory.size() < newSize)
m_state.memory.resize(size_t(newSize));
return true;
}
return _size <= 0xffff;
}
else
m_state.msize = u256(-1);
@ -468,6 +461,27 @@ bool EVMInstructionInterpreter::accessMemory(u256 const& _offset, u256 const& _s
return false;
}
bytes EVMInstructionInterpreter::readMemory(u256 const& _offset, u256 const& _size)
{
yulAssert(_size <= 0xffff, "Too large read.");
bytes data(size_t(_size), uint8_t(0));
for (size_t i = 0; i < data.size(); ++i)
data[i] = m_state.memory[_offset + i];
return data;
}
u256 EVMInstructionInterpreter::readMemoryWord(u256 const& _offset)
{
return u256(h256(readMemory(_offset, 32)));
}
void EVMInstructionInterpreter::writeMemoryWord(u256 const& _offset, u256 const& _value)
{
for (size_t i = 0; i < 32; i++)
m_state.memory[_offset + i] = uint8_t((_value >> (8 * (31 - i))) & 0xff);
}
void EVMInstructionInterpreter::logTrace(dev::eth::Instruction _instruction, std::vector<u256> const& _arguments, bytes const& _data)
{
logTrace(dev::eth::instructionInfo(_instruction).name, _arguments, _data);

View File

@ -75,9 +75,19 @@ public:
dev::u256 evalBuiltin(BuiltinFunctionForEVM const& _fun, std::vector<dev::u256> const& _arguments);
private:
/// Resizes the memory to accommodate the memory access.
/// @returns false if memory would have to be expanded beyond m_state.maxMemSize.
/// Checks if the memory access is not too large for the interpreter and adjusts
/// msize accordingly.
/// @returns false if the amount of bytes read is lager than 0xffff
bool accessMemory(dev::u256 const& _offset, dev::u256 const& _size = 32);
/// @returns the memory contents at the provided address.
/// Does not adjust msize, use @a accessMemory for that
dev::bytes readMemory(dev::u256 const& _offset, dev::u256 const& _size = 32);
/// @returns the memory contents at the provided address.
/// Does not adjust msize, use @a accessMemory for that
dev::u256 readMemoryWord(dev::u256 const& _offset);
/// @returns writes a word to memory
/// Does not adjust msize, use @a accessMemory for that
void writeMemoryWord(dev::u256 const& _offset, dev::u256 const& _value);
void logTrace(dev::eth::Instruction _instruction, std::vector<dev::u256> const& _arguments = {}, dev::bytes const& _data = {});
/// Appends a log to the trace representing an instruction or similar operation by string,

View File

@ -47,13 +47,12 @@ void InterpreterState::dumpTraceAndState(ostream& _out) const
for (auto const& line: trace)
_out << " " << line << endl;
_out << "Memory dump:\n";
for (size_t i = 0; i < memory.size(); i += 0x20)
{
bytesConstRef data(memory.data() + i, 0x20);
if (boost::algorithm::all_of_equal(data, 0))
continue;
_out << " " << std::hex << std::setw(4) << i << ": " << toHex(data.toBytes()) << endl;
}
map<u256, u256> words;
for (auto const& [offset, value]: memory)
words[(offset / 0x20) * 0x20] |= u256(uint32_t(value)) << (256 - 8 - 8 * size_t(offset % 0x20));
for (auto const& [offset, value]: words)
if (value != 0)
_out << " " << std::hex << std::setw(4) << offset << ": " << h256(value).hex() << endl;
_out << "Storage dump:" << endl;
for (auto const& slot: storage)
if (slot.second != h256(0))

View File

@ -64,8 +64,7 @@ struct InterpreterState
{
dev::bytes calldata;
dev::bytes returndata;
/// TODO turn this into "vector with holes" for the randomized testing
dev::bytes memory;
std::map<dev::u256, uint8_t> memory;
/// This is different than memory.size() because we ignore gas.
dev::u256 msize;
std::map<dev::h256, dev::h256> storage;
@ -86,9 +85,6 @@ struct InterpreterState
std::vector<std::string> trace;
/// This is actually an input parameter that more or less limits the runtime.
size_t maxTraceSize = 0;
/// Memory size limit. Anything beyond this will still work, but it has
/// deterministic yet not necessarily consistent behaviour.
size_t maxMemSize = 0x200;
size_t maxSteps = 0;
size_t numSteps = 0;
LoopState loopState = LoopState::Default;

View File

@ -87,7 +87,6 @@ void interpret(string const& _source)
InterpreterState state;
state.maxTraceSize = 10000;
state.maxMemSize = 0x20000000;
Dialect const& dialect(EVMDialect::strictAssemblyForEVMObjects(langutil::EVMVersion{}));
Interpreter interpreter(state, dialect);
try