A Java library for zk-SNARK circuits
This is a Java library for building circuits for preprocessing zk-SNARKs. The library uses libsnark as a backend (https://github.com/scipr-lab/libsnark), and can integrate circuits produced by the Pinocchio compiler (https://vc.codeplex.com/SourceControl/latest) when needed by the programmer. The code consists of two main parts: -
JsnarkCircuitBuilder: A Java project that has a Gadget library for building/augmenting circuits. (Check the
libsnark/jsnark_interface: A C++ interface to libsnark which accepts circuits produced by either the circuit builder or by Pinocchio's compiler directly.
Updates: - The jsnark library now has several cryptographic gadgets used in earlier work (Hawk and C0C0). Some of the gadgets like RSA and AES were improved by techniques from xJsnark. The gadgets can be found in src/examples/gadgets. - xJsnark, a high-level programming framework for zk-SNARKs is available here. xJsnark uses an enhanced version of jsnark in its back end, and aims at reducing the background/effort required by low-level libraries, while generating efficient circuits from the high-level code. Sample examples can be found in this page.
For Ubuntu 14.04, the following can be done to install the above:
To install libsnark prerequisites:
$ sudo apt-get install build-essential cmake git libgmp3-dev libprocps3-dev python-markdown libboost-all-dev libssl-dev
Note: Don't clone libsnark from
https://github.com/scipr-lab/libsnark. Make sure to use the modified libsnark submodule within the jsnark cloned repo in the next section.
To install JDK 8:
$ sudo add-apt-repository ppa:webupd8team/java
$ sudo apt-get update
$ sudo apt-get install oracle-java8-installer
Verify the installed version by
java -version. In case it is not 1.8 or later, try
$ sudo update-java-alternatives -s java-8-oracle
To install Junit4:
$ sudo apt-get install junit4
To download BouncyCastle:
$ wget https://www.bouncycastle.org/download/bcprov-jdk15on-159.jar
$ git clone --recursive https://github.com/akosba/jsnark.git
$ cd jsnark/libsnark
$ git submodule init && git submodule update
$ mkdir build && cd build && cmake ..
The CMakeLists files were modified to produce the needed executable for the interface. The executable will appear under build/libsnark/jsnark_interface
To compile the JsnarkCircuitBuilder project via command line, from the jsnark directory:
$ cd JsnarkCircuitBuilder $ mkdir -p bin $ javac -d bin -cp /usr/share/java/junit4.jar:bcprov-jdk15on-159.jar $(find ./src/* | grep ".java$")
The classpaths of junit4 and bcprov-jdk15on-159.jar may need to be adapted in case the jars are located elsewhere. The above command assumes that bcprov-jdk15on-159.jar was moved to the JsnarkCircuitBuilder directory.
Before running the following, make sure the
config.propertiespoints to the path of the
To run a simple example, the following command can be used
$ java -cp bin examples.generators.SimpleCircuitGenerator
To run one of the JUnit tests available:
$ java -cp bin:/usr/share/java/junit4.jar org.junit.runner.JUnitCore examples.tests.hash.SHA256_Test
Some of the examples and tests will require bcprov-jdk15on-159.jar as well to be added to the classpath.
Note: An IDE, e.g. Eclipse, or possibly the ant tool can be used instead to build and run the Java project more conveniently.
SimpleCircuitGenerator.java. This shows a circuit that computes very simple expressions.
DotProductGadget.java. This is a gadget for computing the dot product of two vectors. The gadget is used in
ModGadget.java. This way of writing circuits is useful when verification is more efficient than computation, and when the prover witness value can be inferred automatically in the circuit. Note the usage of
specifyProverWitnessComputation(..). This must be positioned before the steps where the witness is used.
AugmentedAuctionCircuitGenerator.java. The Pinocchio gadget can be used to use compiled circuits by the Pinocchio compiler as gagdets.
AugmentedAuctionCircuitGenerator.javashows how to use a compiled Auction circuit by Pinocchio, and augment it with other manually-developed gadgets. This can help in the cases where the programmer needs to take care only of the bottleneck parts of the circuits.
Several cryptographic gadgets spanning hashes, block ciphers, key exchange, public key encryption and signatures can be found in src/examples/gadgets. We list some notes regarding few gadgets below:
HybridEncryptionCircuitGenerator.java. The circuit uses the gadgets defined in
examples/gadgets/blockciphers, which are for key exchange (using field extension) and symmetric encryption using the speck cipher. Other variants will be added in the future.
To summarize the steps needed: - Extend the
CircuitGeneratorclass. - Override the
buildCircuit()method: Identify the inputs, outputs and prover witness wires of your circuit, and instantiate/connect gadgets inside. - Override the
generateSampleInput()method: Specify how the values of the input and possibly some of the free prover witness wires are set. This helps in quick testing. - To run a generator, the following methods should be invoked: -
generateCircuit(): generates the arithmetic circuit and the constraints. -
evalCircuit(): evaluates the circuit. -
prepFiles(): This produces two files:
.in. The first file specifies the arithemtic circuit in a way that is similar to how Pinocchio outputs arithmetic circuits, but with other kinds of instructions, like: xor, or, pack and assert. The second file outputs a file containing the values for the input and prover free witness wires. This step must be done after calling
evalCircuit()as some witness values are computed during that step. -
runLibsnark(): This runs the libsnark interface on the two files produced in the last step. By default, this method runs the r1cs_ppzksnark proof system implemented in libsnark. For other options see below. - Note: In the executing thread, use one CircuitGenerator per thread at a time. If multiple generators are used in parallel, each needs to be in a separate thread, and the corresponding property value in config.properties need to be adapted.
Given the .arith and the .in files, it's possible to use command line directly to run the jsnark-libsnark interface. You can use the executable interface
run_ppzksnarkthat appears in
jsnark/libsnark/build/libsnark/jsnark_interfaceto run the libsnark algorithms on the circuit. The executable currently allows to run the proof systems
r1cs_gg_ppzksnarkimplemented in libsnark. To run the first, the executable just takes two arguments: the arithmetic circuit file path, and a sample input file path. To run the
r1cs_gg_ppzksnarkproof system [Gro16], the first argument should be
gg, followed by the arithmetic circuit file path, and the sample input file path.
The gadget library of jsnark shares some similarities with the C++ Gadget library of libsnark, but it has some options that could possibly help with writing optimized circuits quickly without specifying all the details. If the reader is familiar with the gadget libraries of libsnark, here are some key points to minimize confusion:
- No need to maintain a distinction between Variables, LinearCombinations, ... etc. The type Wire can be used to represent Variables, LinearCombinations, Constants, .. etc. The library handles the mapping in a later stage.
- Instead of having the notion of primary input and auxiliary input for representing variables, wires in jsnark can be labeled anywhere as either input, output, prover witness wires. Both the input and output wires are public and seen by the verifier (this corresponds to the primary input in libsnark). The prover witness wires refer to the free input variables provided by the prover. This is in some sense similar to the way Pinocchio's compiler classifies wires.
- Each Gadget in libsnark requires writing and calling two methods: generateConstraints() to specify the r1cs constraints, and generateWitness() to invoke the witness computation. In jsnark's builder, applying primitive operations on wires generates constraints automatically. Additionally, the witness computation is done automatically for primitive operations, and does not need to be explicitly invoked, except in the case of prover witness computation that has to be done outside the circuit, e.g. the FieldDivisionGadget example. - Jsnark applies caching and other techniques during circuit construction to cancel unneeded constraints. This helps in code reusability when changing input variables wires to carry constant values instead. This also helps in reducing the complexity when writing optimized circuits. One example is the
majcalculation in the SHA256 gadget, in which jsnark detects similar operations across the loop iterations with little effort from the programmer, resulting in more than 1000 gates savings.
CachingTest.javaalso illustrates what cases caching can help with.
.inshould specify the hexadecimal value for each input and nizkinput wire ids, in the following format:
id value, each on a separate line.
The code is undergoing more testing and integration of other features. The future versions of this library will include more documentation, examples and optimizations.
This code is developed and maintained by Ahmed Kosba [email protected]. Please email for any questions.