1 LibFuzzer -- a library for coverage-guided fuzz testing.
2 ========================================================
4 This library is intended primarily for in-process coverage-guided fuzz testing
5 (fuzzing) of other libraries. The typical workflow looks like this:
7 * Build the Fuzzer library as a static archive (or just a set of .o files).
8 Note that the Fuzzer contains the main() function.
9 Preferably do *not* use sanitizers while building the Fuzzer.
10 * Build the library you are going to test with -fsanitize-coverage=[234]
11 and one of the sanitizers. We recommend to build the library in several
12 different modes (e.g. asan, msan, lsan, ubsan, etc) and even using different
13 optimizations options (e.g. -O0, -O1, -O2) to diversify testing.
14 * Build a test driver using the same options as the library.
15 The test driver is a C/C++ file containing interesting calls to the library
16 inside a single function ``extern "C" void TestOneInput(const uint8_t *Data, size_t Size);``
17 * Link the Fuzzer, the library and the driver together into an executable
18 using the same sanitizer options as for the library.
19 * Collect the initial corpus of inputs for the
20 fuzzer (a directory with test inputs, one file per input).
21 The better your inputs are the faster you will find something interesting.
22 Also try to keep your inputs small, otherwise the Fuzzer will run too slow.
23 * Run the fuzzer with the test corpus. As new interesting test cases are
24 discovered they will be added to the corpus. If a bug is discovered by
25 the sanitizer (asan, etc) it will be reported as usual and the reproducer
26 will be written to disk.
27 Each Fuzzer process is single-threaded (unless the library starts its own
28 threads). You can run the Fuzzer on the same corpus in multiple processes.
29 in parallel. For run-time options run the Fuzzer binary with '-help=1'.
32 The Fuzzer is similar in concept to AFL (http://lcamtuf.coredump.cx/afl/),
33 but uses in-process Fuzzing, which is more fragile, more restrictive, but
34 potentially much faster as it has no overhead for process start-up.
35 It uses LLVM's "Sanitizer Coverage" instrumentation to get in-process
36 coverage-feedback https://code.google.com/p/address-sanitizer/wiki/AsanCoverage
38 The code resides in the LLVM repository and is (or will be) used by various
39 parts of LLVM, but the Fuzzer itself does not (and should not) depend on any
40 part of LLVM and can be used for other projects. Ideally, the Fuzzer's code
41 should not have any external dependencies. Right now it uses STL, which may need
42 to be fixed later. See also FAQ below.
44 Examples of usage in LLVM
45 =========================
49 The inputs are random pieces of C++-like text.
51 Build (make sure to use fresh clang as the host compiler)::
53 cmake -GNinja -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DLLVM_USE_SANITIZER=Address -DLLVM_USE_SANITIZE_COVERAGE=YES -DCMAKE_BUILD_TYPE=Release /path/to/llvm
54 ninja clang-format-fuzzer
56 ./bin/clang-format-fuzzer CORPUS_DIR
58 Optionally build other kinds of binaries (asan+Debug, msan, ubsan, etc).
60 TODO: commit the pre-fuzzed corpus to svn (?).
65 See lib/Fuzzer/test/SimpleTest.cpp.
66 A simple function that does something interesting if it receives bytes "Hi!"::
68 # Build the Fuzzer with asan:
69 clang++ -std=c++11 -fsanitize=address -fsanitize-coverage=3 -O1 -g Fuzzer*.cpp test/SimpleTest.cpp
70 # Run the fuzzer with no corpus (assuming on empty input)
74 =========================
76 Q. Why Fuzzer does not use any of the LLVM support?
77 ---------------------------------------------------
79 There are two reasons.
81 First, we want this library to be used outside of the LLVM w/o users having to
82 build the rest of LLVM. This may sound unconvincing for many LLVM folks,
83 but in practice the need for building the whole LLVM frightens many potential
84 users -- and we want more users to use this code.
86 Second, there is a subtle technical reason not to rely on the rest of LLVM, or
87 any other large body of code (maybe not even STL). When coverage instrumentation
88 is enabled, it will also instrument the LLVM support code which will blow up the
89 coverage set of the process (since the fuzzer is in-process). In other words, by
90 using more external dependencies we will slow down the fuzzer while the main
91 reason for it to exist is extreme speed.
93 Q. What about Windows then? The Fuzzer contains code that does not build on Windows.
94 ------------------------------------------------------------------------------------
96 The sanitizer coverage support does not work on Windows either as of 01/2015.
97 Once it's there, we'll need to re-implement OS-specific parts (I/O, signals).
99 Q. When this Fuzzer is not a good solution for a problem?
100 ---------------------------------------------------------
102 * If the test inputs are validated by the target library and the validator
103 asserts/crashes on invalid inputs, the in-process fuzzer is not applicable
104 (we could use fork() w/o exec, but it comes with extra overhead).
105 * Bugs in the target library may accumulate w/o being detected. E.g. a memory
106 corruption that goes undetected at first and then leads to a crash while
107 testing another input. This is why it is highly recommended to run this
108 in-process fuzzer with all sanitizers to detect most bugs on the spot.
109 * It is harder to protect the in-process fuzzer from excessive memory
110 consumption and infinite loops in the target library (still possible).
111 * The target library should not have significant global state that is not
112 reset between the runs.
113 * Many interesting target libs are not designed in a way that supports
114 the in-process fuzzer interface (e.g. require a file path instead of a
116 * If a single test run takes a considerable fraction of a second (or
117 more) the speed benefit from the in-process fuzzer is negligible.
118 * If the target library runs persistent threads (that outlive
119 execution of one test) the fuzzing results will be unreliable.
121 Q. So, what exactly this Fuzzer is good for?
122 --------------------------------------------
124 This Fuzzer might be a good choice for testing libraries that have relatively
125 small inputs, each input takes < 1ms to run, and the library code is not expected
126 to crash on invalid inputs.
127 Examples: regular expression matchers, text or binary format parsers.