/daedaluzz

Benchmark Generator for Smart-Contract Fuzzers

Primary LanguageSolidityApache License 2.0Apache-2.0

Daedaluzz: A Benchmark Generator for Smart-Contract Fuzzers

Daedaluzz is a tool for automatically generating benchmarks for smart-contract fuzzers. The benchmark-generation approach takes inspiration from the Fuzzle benchmark generator for C-based fuzzers.

A key goal is to make it possible to compare as many different fuzzers as possible. For this reason, the benchmarks intentionally use a limited subset of Solidity to avoid language features that some tools could handle differently (or not at all).

Benchmark-Generation Approach

Each generated benchmark contract contains many assertions (some can fail, but others cannot due to infeasible path conditions). A fuzzer can create sequences of transactions that make those assertions fail. We can measure a fuzzer's performance by the number of distinct assertion violations that are found.

On a high level, each contract keeps track of the current position in a 2-dimensional maze of a fixed dimension (for instance, 7x7). Each transaction can move the current position to explore the maze. Some locations in the maze are unreachable (so-called "walls"), while others may contain a "bug" that can be found by the fuzzer when providing specific transaction inputs. Some of these bugs cannot be reached since the path conditions are infeasible.

The generated benchmarks try to capture two key challenges when fuzzing smart contracts:

  1. code that can only be reached by satisfying complex transaction-input constraints
  2. code that can only be reached by first reaching a specific state of the contract through multiple prior transactions

In the future, we may extend the benchmark generator to incorporate other challenges we often encounter when fuzzing real-world contracts. We are more than happy to discuss ideas for such extensions!

Benchmark Examples

We have used Daedaluzz to generate a set of 5 benchmark contracts (see generated-mazes folder). Yet, one can easily generate new ones by modifying the random seed and other hyperparameters (for instance, numFuncParams, dimX, and maxDepth).

One can also tweak the benchmark generator to emit custom benchmark variants for fuzzers that do not support the default benchmarks. For instance, we have used it to produce custom benchmark instances for the Foundry/Forge fuzzer; see the foundry-code-generation branch and the files ending in .foundry.sol in the generated-mazes folder.

Benchmarking Infrastructure

The repository also includes infrastructure (for instance, scripts and Dockerfiles) to run several popular fuzzers on the generated benchmarks:

The run-all-fuzzers.sh script can be used to benchmark the above fuzzers, and the show-stats.py script can be used to aggregate and visualize the results.

Frequently Asked Questions

Why do the generated benchmarks use inputs of type uint64?

There are three main reasons:

  1. We observed that some fuzzers—specifically, ones that use constraint solvers—struggle with arithmetic operations of larger bit width (e.g., uint256). We wanted the benchmarks to be compatible with as many fuzzers as possible.
  2. By defaulting to 64-bit arithmetic, we could easily tweak Daedaluzz to generate benchmarks for other languages, such as C, and compare with even more fuzzers.
  3. One can easily increase the number of input bytes (potentially increasing the difficulty of benchmarks) by adjusting the numFuncParams parameter.

We are open to revisiting this design choice in the future.

How did you come up with the current hyperparameters?

We tried several settings but did not exhaustively evaluate each hyperparameter. The main goal was to find a configuration that produces challenging benchmarks that fit into the EVM code size limit.

We are open to making changes to these hyperparameters in the future.

We want to avoid that fuzzer developers overfit their tools to this specific version of the benchmarks. For this reason, we are considering adjusting the benchmarks regularly (for instance, once or twice a year). Benchmark generation tools, such as Daedaluzz, simplify this task significantly.

Will a fuzzer that performs well on these benchmarks also perform well for my code?

Not necessarily. The only way to know is to try the fuzzer on your code. If you have the time, you should always try as many fuzzers as possible. However, it is probably not worth your time to try a fuzzer on your code if it performs poorly on these benchmarks.