C11-clean Keccak, SHA-3, and SHAKE implementation (targeting formal verification)
C++ C Python Objective-C Shell Go
Latest commit fa17e40 Sep 9, 2014 @coruus Update README.markdown
Permalink
Failed to load latest commit information.
build
docs
gnconfig/toolchains
keccak
libutil
ninjas
scripts
verification
.clang-format
.gitignore
.gn
API.markdown
CMakeLists.txt
LICENSE.markdown
README.markdown
build.ninja
defs.ninja
keccak.h
keccak.h.in
tool_versions_nehalem-darwin.txt

README.markdown

C11-clean SHA-3 and SHAKE

An implementation of the Keccak-f[1600] permutation, draft FIPS-202 (SHA-3 and SHAKE).

The goal of this implementation is formal verification of the C code for safety and correctness of the sponge implementation.

This is an alpha release; it has only been (very partially) verified.

Howto

#include "keccak.h"
...
shake256(out, outlen, in, inlen);

That's it. Or, if you really need to,

#include "keccak.h"
...
keccak_sponge sponge;
shake256_init(&sponge);
shake256_update(&sponge, in, inlen);
shake256_digest(&sponge, out, outlen);

Defects

It is unnecessarily complex; see https://github.com/keccak-tiny for a small implementation.

Design

Design criteria:

  • Simple API
  • Simple design
  • Lowest sloc possible
  • Pluggable choice of permutation implementation
  • Safe to incorporate in source form into client code (i.e., strict conformance with the C11 standard)
  • Built-in-self-tests

Trade-offs

It trades support for big-endian platforms and arbitrary bit-length messages for simplicity; I think this is a worthwhile tradeoff to make. (If you have a big-endian platform to test on, however, I am happy to accept pull requests in keccak/keccak to implement support.)

It trades some simplicity to get verification tools to work. This is not a worthwhile tradeoff; it is a bug (perhaps mine, perhaps the tools').

Credits

Mad props to Guido Bertoni, Joan Daemen, Michaël Peeters, and Gilles Van Assche for the very beautiful Keccak permutation, and the incredibly useful sponge construction.

And also mad props to Ronny Van Keer, whose charming assembler implementations of Keccak introduced me to the permutation.

The opt64 and varopt64 implementations of the permutation are entirely their work (and some contribution from CPP and clang-format).

Notes

This code is not endorsed by the Keccak Team or, in fact, anyone else. The official Keccak code package can be found at https://github.com/gvanas/KeccakCodePackage

Their code has been substantially cleaned up recently; I believe that it is now free of undefined behavior. It is much more flexible than this code, at the cost of a more complex interface.

Warning

Please don't use this code for anything other than testing at the moment; I will tag a release when I have sufficient confidence that the code is safe and correct.

Anything in an experimental subdirectory is just that; an experiment.

Building

To build this project, become a daimyo. You will then have enough ninjas.

The library should build in any C11-compliant environment that provides the memcpy and memset functions and defines the (optional) fixed-width uint8_t and uint64_t types.

The tests require a hosted environment at present; they use stdio.h. I consider this a bug.

Verification overview

Tools used:

Tools under consideration:

Done so far:

  1. Asan and Ubsan-clean over concrete executions.

  2. CompCert interpreter traces over concrete executions as proof of freedom from (CompCert-)undefined behavior

  3. Frama-C's value analysis plugin over set of symbolic executions (see the verification/frama-c directory for the latest analysis logs)

In progress:

Thinking about:

  • How to prove correctness of permutation implementation

Tests and metrics

Tests

TODO: Automating the test harnesses.

Build the project. Tests for each permutation implementation will be present in the build directory. Do, e.g.,

 % ./build/keccakf-opt64-test0-sha3
 Len = 0
 Msg = 00
 MD = 6B4E03423667DBB73B6E15454F0EB1ABD4597F9A1B078E3F5B5A6BC7
 SHA3-224(''): OKAY
 Len = 0
 Msg = 00
 MD = A7FFC6F8BF1ED76651C14756A061D662F580FF4DE43B49FA82D80A4B80F8434A
 SHA3-256(''): OKAY
 Len = 0
 Msg = 00
 MD = 0C63A75B845E4F7D01107D852E4C2485C51A50AAAA94FC61995E71BBEE983A2AC3713831264ADB47FB6BD1E058D5F004
 SHA3-384(''): OKAY
 Len = 0
 Msg = 00
 MD = A69F73CCA23A9AC5C8B567DC185A756E97C982164FE25859E0D1DCC1475C80A615B2123AF1F5F94C11E3E9402C3AC558F500199D95B6D3E301758586281DCD26
 SHA3-512(''): OKAY

The following files will contain CompCert's output in interpreter mode for the zero-length test-cases:

verification/compcert/test0-sha3-opt64.trace
verification/compcert/test0-shake-opt64.trace

If you do tail -1, you should see:

% tail -1 verification/compcert/test0-sha3-opt64.trace
Time 284063: program terminated (exit code = 0)
% tail -1 verification/compcert/test0-shake-opt64.trace
Time 550189: program terminated (exit code = 0)

(The files contain the output as well.)

Benchmarks

Build the code. Then run scripts/bench.sh. (TODO) Performance targets.

Lines of code

scripts/count.sh computes the sloc metric for this project. Current value: 579.

Go's implementation. Value: 194

% c_count ~/repos/code.google.com/p/go.crypto/sha3/{sha3,shake,hashes}.go

Target sloc: sloc(go) * 2 == 388

Undefined behavior

This code attempts to avoid undefined behavior. Xi Wang has several papers about why undefined behavior is dangerous in security-critical code, e.g., "Undefined behavior: What happened to my code?".

Any dynamic or static behavior that is undefined under either the C99 or C11 standard is a bug; pull requests eagerly accepted.