1 ========================================================
2 LibFuzzer -- a library for coverage-guided fuzz testing.
3 ========================================================
11 This library is intended primarily for in-process coverage-guided fuzz testing
12 (fuzzing) of other libraries. The typical workflow looks like this:
14 * Build the Fuzzer library as a static archive (or just a set of .o files).
15 Note that the Fuzzer contains the main() function.
16 Preferably do *not* use sanitizers while building the Fuzzer.
17 * Build the library you are going to test with
18 `-fsanitize-coverage={bb,edge}[,indirect-calls]`
19 and one of the sanitizers. We recommend to build the library in several
20 different modes (e.g. asan, msan, lsan, ubsan, etc) and even using different
21 optimizations options (e.g. -O0, -O1, -O2) to diversify testing.
22 * Build a test driver using the same options as the library.
23 The test driver is a C/C++ file containing interesting calls to the library
24 inside a single function ``extern "C" void LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size);``
25 * Link the Fuzzer, the library and the driver together into an executable
26 using the same sanitizer options as for the library.
27 * Collect the initial corpus of inputs for the
28 fuzzer (a directory with test inputs, one file per input).
29 The better your inputs are the faster you will find something interesting.
30 Also try to keep your inputs small, otherwise the Fuzzer will run too slow.
31 * Run the fuzzer with the test corpus. As new interesting test cases are
32 discovered they will be added to the corpus. If a bug is discovered by
33 the sanitizer (asan, etc) it will be reported as usual and the reproducer
34 will be written to disk.
35 Each Fuzzer process is single-threaded (unless the library starts its own
36 threads). You can run the Fuzzer on the same corpus in multiple processes
37 in parallel. For run-time options run the Fuzzer binary with '-help=1'.
40 The Fuzzer is similar in concept to AFL_,
41 but uses in-process Fuzzing, which is more fragile, more restrictive, but
42 potentially much faster as it has no overhead for process start-up.
43 It uses LLVM's SanitizerCoverage_ instrumentation to get in-process
46 The code resides in the LLVM repository, requires the fresh Clang compiler to build
47 and is used to fuzz various parts of LLVM,
48 but the Fuzzer itself does not (and should not) depend on any
49 part of LLVM and can be used for other projects w/o requiring the rest of LLVM.
57 A simple function that does something interesting if it receives the input "HI!"::
59 cat << EOF >> test_fuzzer.cc
60 extern "C" void LLVMFuzzerTestOneInput(const unsigned char *data, unsigned long size) {
61 if (size > 0 && data[0] == 'H')
62 if (size > 1 && data[1] == 'I')
63 if (size > 2 && data[2] == '!')
67 # Get lib/Fuzzer. Assuming that you already have fresh clang in PATH.
68 svn co http://llvm.org/svn/llvm-project/llvm/trunk/lib/Fuzzer
69 # Build lib/Fuzzer files.
70 clang -c -g -O2 -std=c++11 Fuzzer/*.cpp -IFuzzer
71 # Build test_fuzzer.cc with asan and link against lib/Fuzzer.
72 clang++ -fsanitize=address -fsanitize-coverage=edge test_fuzzer.cc Fuzzer*.o
73 # Run the fuzzer with no corpus.
76 You should get ``Illegal instruction (core dumped)`` pretty quickly.
81 Here we show how to use lib/Fuzzer on something real, yet simple: pcre2_::
83 COV_FLAGS=" -fsanitize-coverage=edge,indirect-calls,8bit-counters"
85 svn co svn://vcs.exim.org/pcre2/code/trunk pcre
86 # Get lib/Fuzzer. Assuming that you already have fresh clang in PATH.
87 svn co http://llvm.org/svn/llvm-project/llvm/trunk/lib/Fuzzer
88 # Build PCRE2 with AddressSanitizer and coverage.
89 (cd pcre; ./autogen.sh; CC="clang -fsanitize=address $COV_FLAGS" ./configure --prefix=`pwd`/../inst && make -j && make install)
90 # Build lib/Fuzzer files.
91 clang -c -g -O2 -std=c++11 Fuzzer/*.cpp -IFuzzer
92 # Build the the actual function that does something interesting with PCRE2.
93 cat << EOF > pcre_fuzzer.cc
95 #include "pcre2posix.h"
96 extern "C" void LLVMFuzzerTestOneInput(const unsigned char *data, size_t size) {
98 char *str = new char[size+1];
99 memcpy(str, data, size);
102 if (0 == regcomp(&preg, str, 0)) {
103 regexec(&preg, str, 0, 0, 0);
109 clang++ -g -fsanitize=address $COV_FLAGS -c -std=c++11 -I inst/include/ pcre_fuzzer.cc
111 clang++ -g -fsanitize=address -Wl,--whole-archive inst/lib/*.a -Wl,-no-whole-archive Fuzzer*.o pcre_fuzzer.o -o pcre_fuzzer
113 This will give you a binary of the fuzzer, called ``pcre_fuzzer``.
114 Now, create a directory that will hold the test corpus::
118 For simple input languages like regular expressions this is all you need.
119 For more complicated inputs populate the directory with some input samples.
120 Now run the fuzzer with the corpus dir as the only parameter::
122 ./pcre_fuzzer ./CORPUS
124 You will see output like this::
127 #0 READ cov 0 bits 0 units 1 exec/s 0
128 #1 pulse cov 3 bits 0 units 1 exec/s 0
129 #1 INITED cov 3 bits 0 units 1 exec/s 0
130 #2 pulse cov 208 bits 0 units 1 exec/s 0
131 #2 NEW cov 208 bits 0 units 2 exec/s 0 L: 64
132 #3 NEW cov 217 bits 0 units 3 exec/s 0 L: 63
133 #4 pulse cov 217 bits 0 units 3 exec/s 0
135 * The ``Seed:`` line shows you the current random seed (you can change it with ``-seed=N`` flag).
136 * The ``READ`` line shows you how many input files were read (since you passed an empty dir there were inputs, but one dummy input was synthesised).
137 * The ``INITED`` line shows you that how many inputs will be fuzzed.
138 * The ``NEW`` lines appear with the fuzzer finds a new interesting input, which is saved to the CORPUS dir. If multiple corpus dirs are given, the first one is used.
139 * The ``pulse`` lines appear periodically to show the current status.
141 Now, interrupt the fuzzer and run it again the same way. You will see::
144 #0 READ cov 0 bits 0 units 564 exec/s 0
145 #1 pulse cov 502 bits 0 units 564 exec/s 0
147 #512 pulse cov 2933 bits 0 units 564 exec/s 512
148 #564 INITED cov 2991 bits 0 units 344 exec/s 564
149 #1024 pulse cov 2991 bits 0 units 344 exec/s 1024
150 #1455 NEW cov 2995 bits 0 units 345 exec/s 1455 L: 49
152 This time you were running the fuzzer with a non-empty input corpus (564 items).
153 As the first step, the fuzzer minimized the set to produce 344 interesting items (the ``INITED`` line)
155 It is quite convenient to store test corpuses in git.
156 As an example, here is a git repository with test inputs for the above PCRE2 fuzzer::
158 git clone https://github.com/kcc/fuzzing-with-sanitizers.git
159 ./pcre_fuzzer ./fuzzing-with-sanitizers/pcre2/C1/
161 You may run ``N`` independent fuzzer jobs in parallel on ``M`` CPUs::
163 N=100; M=4; ./pcre_fuzzer ./CORPUS -jobs=$N -workers=$M
165 By default (``-reload=1``) the fuzzer processes will periodically scan the CORPUS directory
166 and reload any new tests. This way the test inputs found by one process will be picked up
169 If ``-workers=$M`` is not supplied, ``min($N,NumberOfCpuCore/2)`` will be used.
173 Remember Heartbleed_?
174 As it was recently `shown <https://blog.hboeck.de/archives/868-How-Heartbleed-couldve-been-found.html>`_,
175 fuzzing with AddressSanitizer can find Heartbleed. Indeed, here are the step-by-step instructions
176 to find Heartbleed with LibFuzzer::
178 wget https://www.openssl.org/source/openssl-1.0.1f.tar.gz
179 tar xf openssl-1.0.1f.tar.gz
180 COV_FLAGS="-fsanitize-coverage=edge,indirect-calls" # -fsanitize-coverage=8bit-counters
181 (cd openssl-1.0.1f/ && ./config &&
182 make -j 32 CC="clang -g -fsanitize=address $COV_FLAGS")
183 # Get and build LibFuzzer
184 svn co http://llvm.org/svn/llvm-project/llvm/trunk/lib/Fuzzer
185 clang -c -g -O2 -std=c++11 Fuzzer/*.cpp -IFuzzer
186 # Get examples of key/pem files.
187 git clone https://github.com/hannob/selftls
188 cp selftls/server* . -v
189 cat << EOF > handshake-fuzz.cc
190 #include <openssl/ssl.h>
191 #include <openssl/err.h>
196 SSL_load_error_strings();
197 ERR_load_BIO_strings();
198 OpenSSL_add_all_algorithms();
199 assert (sctx = SSL_CTX_new(TLSv1_method()));
200 assert (SSL_CTX_use_certificate_file(sctx, "server.pem", SSL_FILETYPE_PEM));
201 assert (SSL_CTX_use_PrivateKey_file(sctx, "server.key", SSL_FILETYPE_PEM));
204 extern "C" void LLVMFuzzerTestOneInput(unsigned char *Data, size_t Size) {
205 static int unused = Init();
206 SSL *server = SSL_new(sctx);
207 BIO *sinbio = BIO_new(BIO_s_mem());
208 BIO *soutbio = BIO_new(BIO_s_mem());
209 SSL_set_bio(server, sinbio, soutbio);
210 SSL_set_accept_state(server);
211 BIO_write(sinbio, Data, Size);
212 SSL_do_handshake(server);
217 clang++ -g handshake-fuzz.cc -fsanitize=address \
218 openssl-1.0.1f/libssl.a openssl-1.0.1f/libcrypto.a Fuzzer*.o
219 # Run 20 independent fuzzer jobs.
220 ./a.out -jobs=20 -workers=20
224 #1048576 pulse cov 3424 bits 0 units 9 exec/s 24385
225 =================================================================
226 ==17488==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x629000004748 at pc 0x00000048c979 bp 0x7fffe3e864f0 sp 0x7fffe3e85ca8
227 READ of size 60731 at 0x629000004748 thread T0
228 #0 0x48c978 in __asan_memcpy
229 #1 0x4db504 in tls1_process_heartbeat openssl-1.0.1f/ssl/t1_lib.c:2586:3
230 #2 0x580be3 in ssl3_read_bytes openssl-1.0.1f/ssl/s3_pkt.c:1092:4
238 By default, the fuzzer is not aware of complexities of the input language
239 and when fuzzing e.g. a C++ parser it will mostly stress the lexer.
240 It is very hard for the fuzzer to come up with something like ``reinterpret_cast<int>``
241 from a test corpus that doesn't have it.
242 See a detailed discussion of this topic at
243 http://lcamtuf.blogspot.com/2015/01/afl-fuzz-making-up-grammar-with.html.
245 lib/Fuzzer implements a simple technique that allows to fuzz input languages with
246 long tokens. All you need is to prepare a text file containing up to 253 tokens, one token per line,
247 and pass it to the fuzzer as ``-tokens=TOKENS_FILE.txt``.
248 Three implicit tokens are added: ``" "``, ``"\t"``, and ``"\n"``.
249 The fuzzer itself will still be mutating a string of bytes
250 but before passing this input to the target library it will replace every byte ``b`` with the ``b``-th token.
251 If there are less than ``b`` tokens, a space will be added instead.
255 LibFuzzer can be used in parallel with AFL_ on the same test corpus.
256 Both fuzzers expect the test corpus to reside in a directory, one file per input.
257 You can run both fuzzers on the same corpus in parallel::
259 ./afl-fuzz -i testcase_dir -o findings_dir /path/to/program -r @@
260 ./llvm-fuzz testcase_dir findings_dir # Will write new tests to testcase_dir
262 Periodically restart both fuzzers so that they can use each other's findings.
264 How good is my fuzzer?
265 ----------------------
267 Once you implement your target function ``LLVMFuzzerTestOneInput`` and fuzz it to death,
268 you will want to know whether the function or the corpus can be improved further.
269 One easy to use metric is, of course, code coverage.
270 You can get the coverage for your corpus like this::
272 ASAN_OPTIONS=coverage_pcs=1 ./fuzzer CORPUS_DIR -runs=0
274 This will run all the tests in the CORPUS_DIR but will not generate any new tests
275 and dump covered PCs to disk before exiting.
276 Then you can subtract the set of covered PCs from the set of all instrumented PCs in the binary,
277 see SanitizerCoverage_ for details.
279 Fuzzing components of LLVM
280 ==========================
284 The inputs are random pieces of C++-like text.
286 Build (make sure to use fresh clang as the host compiler)::
288 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
289 ninja clang-format-fuzzer
291 ./bin/clang-format-fuzzer CORPUS_DIR
293 Optionally build other kinds of binaries (asan+Debug, msan, ubsan, etc).
295 TODO: commit the pre-fuzzed corpus to svn (?).
297 Tracking bug: https://llvm.org/bugs/show_bug.cgi?id=23052
302 The default behavior is very similar to ``clang-format-fuzzer``.
303 Clang can also be fuzzed with Tokens_ using ``-tokens=$LLVM/lib/Fuzzer/cxx_fuzzer_tokens.txt`` option.
305 Tracking bug: https://llvm.org/bugs/show_bug.cgi?id=23057
310 We have a buildbot that runs the above fuzzers for LLVM components
311 24/7/365 at http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fuzzer .
313 Pre-fuzzed test inputs in git
314 -----------------------------
316 The buildbot occumulates large test corpuses over time.
317 The corpuses are stored in git on github and can be used like this::
319 git clone https://github.com/kcc/fuzzing-with-sanitizers.git
320 bin/clang-format-fuzzer fuzzing-with-sanitizers/llvm/clang-format/C1
321 bin/clang-fuzzer fuzzing-with-sanitizers/llvm/clang/C1/
322 bin/clang-fuzzer fuzzing-with-sanitizers/llvm/clang/TOK1 -tokens=$LLVM/llvm/lib/Fuzzer/cxx_fuzzer_tokens.txt
326 =========================
328 Q. Why Fuzzer does not use any of the LLVM support?
329 ---------------------------------------------------
331 There are two reasons.
333 First, we want this library to be used outside of the LLVM w/o users having to
334 build the rest of LLVM. This may sound unconvincing for many LLVM folks,
335 but in practice the need for building the whole LLVM frightens many potential
336 users -- and we want more users to use this code.
338 Second, there is a subtle technical reason not to rely on the rest of LLVM, or
339 any other large body of code (maybe not even STL). When coverage instrumentation
340 is enabled, it will also instrument the LLVM support code which will blow up the
341 coverage set of the process (since the fuzzer is in-process). In other words, by
342 using more external dependencies we will slow down the fuzzer while the main
343 reason for it to exist is extreme speed.
345 Q. What about Windows then? The Fuzzer contains code that does not build on Windows.
346 ------------------------------------------------------------------------------------
348 The sanitizer coverage support does not work on Windows either as of 01/2015.
349 Once it's there, we'll need to re-implement OS-specific parts (I/O, signals).
351 Q. When this Fuzzer is not a good solution for a problem?
352 ---------------------------------------------------------
354 * If the test inputs are validated by the target library and the validator
355 asserts/crashes on invalid inputs, the in-process fuzzer is not applicable
356 (we could use fork() w/o exec, but it comes with extra overhead).
357 * Bugs in the target library may accumulate w/o being detected. E.g. a memory
358 corruption that goes undetected at first and then leads to a crash while
359 testing another input. This is why it is highly recommended to run this
360 in-process fuzzer with all sanitizers to detect most bugs on the spot.
361 * It is harder to protect the in-process fuzzer from excessive memory
362 consumption and infinite loops in the target library (still possible).
363 * The target library should not have significant global state that is not
364 reset between the runs.
365 * Many interesting target libs are not designed in a way that supports
366 the in-process fuzzer interface (e.g. require a file path instead of a
368 * If a single test run takes a considerable fraction of a second (or
369 more) the speed benefit from the in-process fuzzer is negligible.
370 * If the target library runs persistent threads (that outlive
371 execution of one test) the fuzzing results will be unreliable.
373 Q. So, what exactly this Fuzzer is good for?
374 --------------------------------------------
376 This Fuzzer might be a good choice for testing libraries that have relatively
377 small inputs, each input takes < 1ms to run, and the library code is not expected
378 to crash on invalid inputs.
379 Examples: regular expression matchers, text or binary format parsers.
381 .. _pcre2: http://www.pcre.org/
383 .. _AFL: http://lcamtuf.coredump.cx/afl/
385 .. _SanitizerCoverage: http://clang.llvm.org/docs/SanitizerCoverage.html
387 .. _Heartbleed: http://en.wikipedia.org/wiki/Heartbleed