{ stdenv, fetchurl, bash, callPackage, makeWrapper }: let afl-qemu = callPackage ./qemu.nix {}; qemu-exe-name = if stdenv.system == "x86_64-linux" then "qemu-x86_64" else if stdenv.system == "i686-linux" then "qemu-i386" else throw "afl: no support for ${stdenv.system}!"; in stdenv.mkDerivation rec { name = "afl-${version}"; version = "1.58b"; src = fetchurl { url = "http://lcamtuf.coredump.cx/afl/releases/${name}.tgz"; sha256 = "1szggm4x9i9bsrcb99s5vbgncagp7jvhz8cg9amkx7p6mp2x4pld"; }; buildInputs = [ makeWrapper ]; buildPhase = "make PREFIX=$out"; installPhase = '' # Do the normal installation make install PREFIX=$out # Install the custom QEMU emulator for binary blob fuzzing. cp ${afl-qemu}/bin/${qemu-exe-name} $out/bin/afl-qemu-trace # Wrap every program with a custom $AFL_PATH; I believe there is a # bug in afl which causes it to fail to find `afl-qemu-trace` # relative to `afl-fuzz` or `afl-showmap`, so we instead set # $AFL_PATH as a workaround, which allows it to be found. for x in `ls $out/bin/afl-*`; do wrapProgram $x --prefix AFL_PATH : "$out/bin" done ''; meta = { description = "Powerful fuzzer via genetic algorithms and instrumentation"; longDescription = '' American fuzzy lop is a fuzzer that employs a novel type of compile-time instrumentation and genetic algorithms to automatically discover clean, interesting test cases that trigger new internal states in the targeted binary. This substantially improves the functional coverage for the fuzzed code. The compact synthesized corpora produced by the tool are also useful for seeding other, more labor or resource-intensive testing regimes down the road. ''; homepage = "http://lcamtuf.coredump.cx/afl/"; license = stdenv.lib.licenses.asl20; platforms = stdenv.lib.platforms.linux; maintainers = [ stdenv.lib.maintainers.thoughtpolice ]; }; }