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,8bit-counters]`
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" int LLVMFuzzerTestOneInput(const uint8_t *Data, size_t Size);``.
25 Currently, the only expected return value is 0, others are reserved for future.
26 * Link the Fuzzer, the library and the driver together into an executable
27 using the same sanitizer options as for the library.
28 * Collect the initial corpus of inputs for the
29 fuzzer (a directory with test inputs, one file per input).
30 The better your inputs are the faster you will find something interesting.
31 Also try to keep your inputs small, otherwise the Fuzzer will run too slow.
32 By default, the Fuzzer limits the size of every input to 64 bytes
33 (use ``-max_len=N`` to override).
34 * Run the fuzzer with the test corpus. As new interesting test cases are
35 discovered they will be added to the corpus. If a bug is discovered by
36 the sanitizer (asan, etc) it will be reported as usual and the reproducer
37 will be written to disk.
38 Each Fuzzer process is single-threaded (unless the library starts its own
39 threads). You can run the Fuzzer on the same corpus in multiple processes
43 The Fuzzer is similar in concept to AFL_,
44 but uses in-process Fuzzing, which is more fragile, more restrictive, but
45 potentially much faster as it has no overhead for process start-up.
46 It uses LLVM's SanitizerCoverage_ instrumentation to get in-process
49 The code resides in the LLVM repository, requires the fresh Clang compiler to build
50 and is used to fuzz various parts of LLVM,
51 but the Fuzzer itself does not (and should not) depend on any
52 part of LLVM and can be used for other projects w/o requiring the rest of LLVM.
56 The most important flags are::
58 seed 0 Random seed. If 0, seed is generated.
59 runs -1 Number of individual test runs (-1 for infinite runs).
60 max_len 64 Maximum length of the test input.
61 cross_over 1 If 1, cross over inputs.
62 mutate_depth 5 Apply this number of consecutive mutations to each input.
63 timeout 1200 Timeout in seconds (if positive). If one unit runs more than this number of seconds the process will abort.
64 max_total_time 0 If positive, indicates the maximal total time in seconds to run the fuzzer.
66 save_minimized_corpus 0 If 1, the minimized corpus is saved into the first input directory. Example: ./fuzzer -save_minimized_corpus=1 NEW_EMPTY_DIR OLD_CORPUS
67 jobs 0 Number of jobs to run. If jobs >= 1 we spawn this number of jobs in separate worker processes with stdout/stderr redirected to fuzz-JOB.log.
68 workers 0 Number of simultaneous worker processes to run the jobs. If zero, "min(jobs,NumberOfCpuCores()/2)" is used.
69 sync_command 0 Execute an external command "<sync_command> <test_corpus>" to synchronize the test corpus.
70 sync_timeout 600 Minimum timeout between syncs.
71 use_traces 0 Experimental: use instruction traces
72 only_ascii 0 If 1, generate only ASCII (isprint+isspace) inputs.
73 test_single_input "" Use specified file content as test input. Test will be run only once. Useful for debugging a particular case.
76 For the full list of flags run the fuzzer binary with ``-help=1``.
84 A simple function that does something interesting if it receives the input "HI!"::
86 cat << EOF >> test_fuzzer.cc
87 extern "C" int LLVMFuzzerTestOneInput(const unsigned char *data, unsigned long size) {
88 if (size > 0 && data[0] == 'H')
89 if (size > 1 && data[1] == 'I')
90 if (size > 2 && data[2] == '!')
95 # Get lib/Fuzzer. Assuming that you already have fresh clang in PATH.
96 svn co http://llvm.org/svn/llvm-project/llvm/trunk/lib/Fuzzer
97 # Build lib/Fuzzer files.
98 clang -c -g -O2 -std=c++11 Fuzzer/*.cpp -IFuzzer
99 # Build test_fuzzer.cc with asan and link against lib/Fuzzer.
100 clang++ -fsanitize=address -fsanitize-coverage=edge test_fuzzer.cc Fuzzer*.o
101 # Run the fuzzer with no corpus.
104 You should get ``Illegal instruction (core dumped)`` pretty quickly.
109 Here we show how to use lib/Fuzzer on something real, yet simple: pcre2_::
111 COV_FLAGS=" -fsanitize-coverage=edge,indirect-calls,8bit-counters"
113 svn co svn://vcs.exim.org/pcre2/code/trunk pcre
114 # Get lib/Fuzzer. Assuming that you already have fresh clang in PATH.
115 svn co http://llvm.org/svn/llvm-project/llvm/trunk/lib/Fuzzer
116 # Build PCRE2 with AddressSanitizer and coverage.
117 (cd pcre; ./autogen.sh; CC="clang -fsanitize=address $COV_FLAGS" ./configure --prefix=`pwd`/../inst && make -j && make install)
118 # Build lib/Fuzzer files.
119 clang -c -g -O2 -std=c++11 Fuzzer/*.cpp -IFuzzer
120 # Build the actual function that does something interesting with PCRE2.
121 cat << EOF > pcre_fuzzer.cc
123 #include "pcre2posix.h"
124 extern "C" int LLVMFuzzerTestOneInput(const unsigned char *data, size_t size) {
125 if (size < 1) return 0;
126 char *str = new char[size+1];
127 memcpy(str, data, size);
130 if (0 == regcomp(&preg, str, 0)) {
131 regexec(&preg, str, 0, 0, 0);
138 clang++ -g -fsanitize=address $COV_FLAGS -c -std=c++11 -I inst/include/ pcre_fuzzer.cc
140 clang++ -g -fsanitize=address -Wl,--whole-archive inst/lib/*.a -Wl,-no-whole-archive Fuzzer*.o pcre_fuzzer.o -o pcre_fuzzer
142 This will give you a binary of the fuzzer, called ``pcre_fuzzer``.
143 Now, create a directory that will hold the test corpus::
147 For simple input languages like regular expressions this is all you need.
148 For more complicated inputs populate the directory with some input samples.
149 Now run the fuzzer with the corpus dir as the only parameter::
151 ./pcre_fuzzer ./CORPUS
153 You will see output like this::
156 #0 READ cov 0 bits 0 units 1 exec/s 0
157 #1 pulse cov 3 bits 0 units 1 exec/s 0
158 #1 INITED cov 3 bits 0 units 1 exec/s 0
159 #2 pulse cov 208 bits 0 units 1 exec/s 0
160 #2 NEW cov 208 bits 0 units 2 exec/s 0 L: 64
161 #3 NEW cov 217 bits 0 units 3 exec/s 0 L: 63
162 #4 pulse cov 217 bits 0 units 3 exec/s 0
164 * The ``Seed:`` line shows you the current random seed (you can change it with ``-seed=N`` flag).
165 * 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).
166 * The ``INITED`` line shows you that how many inputs will be fuzzed.
167 * 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.
168 * The ``pulse`` lines appear periodically to show the current status.
170 Now, interrupt the fuzzer and run it again the same way. You will see::
173 #0 READ cov 0 bits 0 units 564 exec/s 0
174 #1 pulse cov 502 bits 0 units 564 exec/s 0
176 #512 pulse cov 2933 bits 0 units 564 exec/s 512
177 #564 INITED cov 2991 bits 0 units 344 exec/s 564
178 #1024 pulse cov 2991 bits 0 units 344 exec/s 1024
179 #1455 NEW cov 2995 bits 0 units 345 exec/s 1455 L: 49
181 This time you were running the fuzzer with a non-empty input corpus (564 items).
182 As the first step, the fuzzer minimized the set to produce 344 interesting items (the ``INITED`` line)
184 It is quite convenient to store test corpuses in git.
185 As an example, here is a git repository with test inputs for the above PCRE2 fuzzer::
187 git clone https://github.com/kcc/fuzzing-with-sanitizers.git
188 ./pcre_fuzzer ./fuzzing-with-sanitizers/pcre2/C1/
190 You may run ``N`` independent fuzzer jobs in parallel on ``M`` CPUs::
192 N=100; M=4; ./pcre_fuzzer ./CORPUS -jobs=$N -workers=$M
194 By default (``-reload=1``) the fuzzer processes will periodically scan the CORPUS directory
195 and reload any new tests. This way the test inputs found by one process will be picked up
198 If ``-workers=$M`` is not supplied, ``min($N,NumberOfCpuCore/2)`` will be used.
202 Remember Heartbleed_?
203 As it was recently `shown <https://blog.hboeck.de/archives/868-How-Heartbleed-couldve-been-found.html>`_,
204 fuzzing with AddressSanitizer can find Heartbleed. Indeed, here are the step-by-step instructions
205 to find Heartbleed with LibFuzzer::
207 wget https://www.openssl.org/source/openssl-1.0.1f.tar.gz
208 tar xf openssl-1.0.1f.tar.gz
209 COV_FLAGS="-fsanitize-coverage=edge,indirect-calls" # -fsanitize-coverage=8bit-counters
210 (cd openssl-1.0.1f/ && ./config &&
211 make -j 32 CC="clang -g -fsanitize=address $COV_FLAGS")
212 # Get and build LibFuzzer
213 svn co http://llvm.org/svn/llvm-project/llvm/trunk/lib/Fuzzer
214 clang -c -g -O2 -std=c++11 Fuzzer/*.cpp -IFuzzer
215 # Get examples of key/pem files.
216 git clone https://github.com/hannob/selftls
217 cp selftls/server* . -v
218 cat << EOF > handshake-fuzz.cc
219 #include <openssl/ssl.h>
220 #include <openssl/err.h>
225 SSL_load_error_strings();
226 ERR_load_BIO_strings();
227 OpenSSL_add_all_algorithms();
228 assert (sctx = SSL_CTX_new(TLSv1_method()));
229 assert (SSL_CTX_use_certificate_file(sctx, "server.pem", SSL_FILETYPE_PEM));
230 assert (SSL_CTX_use_PrivateKey_file(sctx, "server.key", SSL_FILETYPE_PEM));
233 extern "C" int LLVMFuzzerTestOneInput(unsigned char *Data, size_t Size) {
234 static int unused = Init();
235 SSL *server = SSL_new(sctx);
236 BIO *sinbio = BIO_new(BIO_s_mem());
237 BIO *soutbio = BIO_new(BIO_s_mem());
238 SSL_set_bio(server, sinbio, soutbio);
239 SSL_set_accept_state(server);
240 BIO_write(sinbio, Data, Size);
241 SSL_do_handshake(server);
247 clang++ -g handshake-fuzz.cc -fsanitize=address \
248 openssl-1.0.1f/libssl.a openssl-1.0.1f/libcrypto.a Fuzzer*.o
249 # Run 20 independent fuzzer jobs.
250 ./a.out -jobs=20 -workers=20
254 #1048576 pulse cov 3424 bits 0 units 9 exec/s 24385
255 =================================================================
256 ==17488==ERROR: AddressSanitizer: heap-buffer-overflow on address 0x629000004748 at pc 0x00000048c979 bp 0x7fffe3e864f0 sp 0x7fffe3e85ca8
257 READ of size 60731 at 0x629000004748 thread T0
258 #0 0x48c978 in __asan_memcpy
259 #1 0x4db504 in tls1_process_heartbeat openssl-1.0.1f/ssl/t1_lib.c:2586:3
260 #2 0x580be3 in ssl3_read_bytes openssl-1.0.1f/ssl/s3_pkt.c:1092:4
268 LibFuzzer supports user-supplied dictionaries with input language keywords
269 or other interesting byte sequences (e.g. multi-byte magic values).
270 Use ``-dict=DICTIONARY_FILE``. For some input languages using a dictionary
271 may significantly improve the search speed.
272 The dictionary syntax is similar to that used by AFL_ for its ``-x`` option::
274 # Lines starting with '#' and empty lines are ignored.
276 # Adds "blah" (w/o quotes) to the dictionary.
278 # Use \\ for backslash and \" for quotes.
280 # Use \xAB for hex values
282 # the name of the keyword followed by '=' may be omitted:
285 Data-flow-guided fuzzing
286 ------------------------
289 With an additional compiler flag ``-fsanitize-coverage=trace-cmp`` (see SanitizerCoverageTraceDataFlow_)
290 and extra run-time flag ``-use_traces=1`` the fuzzer will try to apply *data-flow-guided fuzzing*.
291 That is, the fuzzer will record the inputs to comparison instructions, switch statements,
292 and several libc functions (``memcmp``, ``strcmp``, ``strncmp``, etc).
293 It will later use those recorded inputs during mutations.
295 This mode can be combined with DataFlowSanitizer_ to achieve better sensitivity.
299 LibFuzzer can be used in parallel with AFL_ on the same test corpus.
300 Both fuzzers expect the test corpus to reside in a directory, one file per input.
301 You can run both fuzzers on the same corpus in parallel::
303 ./afl-fuzz -i testcase_dir -o findings_dir /path/to/program -r @@
304 ./llvm-fuzz testcase_dir findings_dir # Will write new tests to testcase_dir
306 Periodically restart both fuzzers so that they can use each other's findings.
308 How good is my fuzzer?
309 ----------------------
311 Once you implement your target function ``LLVMFuzzerTestOneInput`` and fuzz it to death,
312 you will want to know whether the function or the corpus can be improved further.
313 One easy to use metric is, of course, code coverage.
314 You can get the coverage for your corpus like this::
316 ASAN_OPTIONS=coverage_pcs=1 ./fuzzer CORPUS_DIR -runs=0
318 This will run all the tests in the CORPUS_DIR but will not generate any new tests
319 and dump covered PCs to disk before exiting.
320 Then you can subtract the set of covered PCs from the set of all instrumented PCs in the binary,
321 see SanitizerCoverage_ for details.
323 User-supplied mutators
324 ----------------------
326 LibFuzzer allows to use custom (user-supplied) mutators,
327 see FuzzerInterface.h_
329 Fuzzing components of LLVM
330 ==========================
334 The inputs are random pieces of C++-like text.
336 Build (make sure to use fresh clang as the host compiler)::
338 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
339 ninja clang-format-fuzzer
341 ./bin/clang-format-fuzzer CORPUS_DIR
343 Optionally build other kinds of binaries (asan+Debug, msan, ubsan, etc).
345 Tracking bug: https://llvm.org/bugs/show_bug.cgi?id=23052
350 The behavior is very similar to ``clang-format-fuzzer``.
352 Tracking bug: https://llvm.org/bugs/show_bug.cgi?id=23057
357 Tracking bug: https://llvm.org/bugs/show_bug.cgi?id=24639
362 This tool fuzzes the MC layer. Currently it is only able to fuzz the
363 disassembler but it is hoped that assembly, and round-trip verification will be
366 When run in dissassembly mode, the inputs are opcodes to be disassembled. The
367 fuzzer will consume as many instructions as possible and will stop when it
368 finds an invalid instruction or runs out of data.
370 Please note that the command line interface differs slightly from that of other
371 fuzzers. The fuzzer arguments should follow ``--fuzzer-args`` and should have
372 a single dash, while other arguments control the operation mode and target in a
373 similar manner to ``llvm-mc`` and should have two dashes. For example::
375 llvm-mc-fuzzer --triple=aarch64-linux-gnu --disassemble --fuzzer-args -max_len=4 -jobs=10
380 We have a buildbot that runs the above fuzzers for LLVM components
381 24/7/365 at http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fuzzer .
383 Pre-fuzzed test inputs in git
384 -----------------------------
386 The buildbot occumulates large test corpuses over time.
387 The corpuses are stored in git on github and can be used like this::
389 git clone https://github.com/kcc/fuzzing-with-sanitizers.git
390 bin/clang-format-fuzzer fuzzing-with-sanitizers/llvm/clang-format/C1
391 bin/clang-fuzzer fuzzing-with-sanitizers/llvm/clang/C1/
392 bin/llvm-as-fuzzer fuzzing-with-sanitizers/llvm/llvm-as/C1 -only_ascii=1
396 =========================
398 Q. Why Fuzzer does not use any of the LLVM support?
399 ---------------------------------------------------
401 There are two reasons.
403 First, we want this library to be used outside of the LLVM w/o users having to
404 build the rest of LLVM. This may sound unconvincing for many LLVM folks,
405 but in practice the need for building the whole LLVM frightens many potential
406 users -- and we want more users to use this code.
408 Second, there is a subtle technical reason not to rely on the rest of LLVM, or
409 any other large body of code (maybe not even STL). When coverage instrumentation
410 is enabled, it will also instrument the LLVM support code which will blow up the
411 coverage set of the process (since the fuzzer is in-process). In other words, by
412 using more external dependencies we will slow down the fuzzer while the main
413 reason for it to exist is extreme speed.
415 Q. What about Windows then? The Fuzzer contains code that does not build on Windows.
416 ------------------------------------------------------------------------------------
418 The sanitizer coverage support does not work on Windows either as of 01/2015.
419 Once it's there, we'll need to re-implement OS-specific parts (I/O, signals).
421 Q. When this Fuzzer is not a good solution for a problem?
422 ---------------------------------------------------------
424 * If the test inputs are validated by the target library and the validator
425 asserts/crashes on invalid inputs, the in-process fuzzer is not applicable
426 (we could use fork() w/o exec, but it comes with extra overhead).
427 * Bugs in the target library may accumulate w/o being detected. E.g. a memory
428 corruption that goes undetected at first and then leads to a crash while
429 testing another input. This is why it is highly recommended to run this
430 in-process fuzzer with all sanitizers to detect most bugs on the spot.
431 * It is harder to protect the in-process fuzzer from excessive memory
432 consumption and infinite loops in the target library (still possible).
433 * The target library should not have significant global state that is not
434 reset between the runs.
435 * Many interesting target libs are not designed in a way that supports
436 the in-process fuzzer interface (e.g. require a file path instead of a
438 * If a single test run takes a considerable fraction of a second (or
439 more) the speed benefit from the in-process fuzzer is negligible.
440 * If the target library runs persistent threads (that outlive
441 execution of one test) the fuzzing results will be unreliable.
443 Q. So, what exactly this Fuzzer is good for?
444 --------------------------------------------
446 This Fuzzer might be a good choice for testing libraries that have relatively
447 small inputs, each input takes < 1ms to run, and the library code is not expected
448 to crash on invalid inputs.
449 Examples: regular expression matchers, text or binary format parsers.
453 * GLIBC: https://sourceware.org/glibc/wiki/FuzzingLibc
457 * http://git.musl-libc.org/cgit/musl/commit/?id=39dfd58417ef642307d90306e1c7e50aaec5a35c
458 * http://www.openwall.com/lists/oss-security/2015/03/30/3
460 * pugixml: https://github.com/zeux/pugixml/issues/39
462 * PCRE: Search for "LLVM fuzzer" in http://vcs.pcre.org/pcre2/code/trunk/ChangeLog?view=markup
464 * ICU: http://bugs.icu-project.org/trac/ticket/11838
466 * Freetype: https://savannah.nongnu.org/search/?words=LibFuzzer&type_of_search=bugs&Search=Search&exact=1#options
468 * Linux Kernel's BPF verifier: https://github.com/iovisor/bpf-fuzzer
472 * Clang: https://llvm.org/bugs/show_bug.cgi?id=23057
474 * Clang-format: https://llvm.org/bugs/show_bug.cgi?id=23052
476 * libc++: https://llvm.org/bugs/show_bug.cgi?id=24411
478 * llvm-as: https://llvm.org/bugs/show_bug.cgi?id=24639
482 * Mips: Discovered a number of untested instructions for the Mips target
483 (see valid-mips*.s in http://reviews.llvm.org/rL247405,
484 http://reviews.llvm.org/rL247414, http://reviews.llvm.org/rL247416,
485 http://reviews.llvm.org/rL247417, http://reviews.llvm.org/rL247420,
486 and http://reviews.llvm.org/rL247422) as well some instructions that
487 successfully disassembled on ISA's where they were not valid (see
488 invalid-xfail.s files in the same commits).
490 .. _pcre2: http://www.pcre.org/
492 .. _AFL: http://lcamtuf.coredump.cx/afl/
494 .. _SanitizerCoverage: http://clang.llvm.org/docs/SanitizerCoverage.html
495 .. _SanitizerCoverageTraceDataFlow: http://clang.llvm.org/docs/SanitizerCoverage.html#tracing-data-flow
496 .. _DataFlowSanitizer: http://clang.llvm.org/docs/DataFlowSanitizer.html
498 .. _Heartbleed: http://en.wikipedia.org/wiki/Heartbleed
500 .. _FuzzerInterface.h: https://github.com/llvm-mirror/llvm/blob/master/lib/Fuzzer/FuzzerInterface.h