1 ================================
2 Fuzzing LLVM libraries and tools
3 ================================
12 The LLVM tree includes a number of fuzzers for various components. These are
13 built on top of :doc:`LibFuzzer <LibFuzzer>`. In order to build and run these
14 fuzzers, see :ref:`building-fuzzers`.
23 A |generic fuzzer| that tries to compile textual input as C++ code. Some of the
24 bugs this fuzzer has reported are `on bugzilla`__ and `on OSS Fuzz's
27 __ https://llvm.org/pr23057
28 __ https://bugs.chromium.org/p/oss-fuzz/issues/list?q=proj-llvm+clang-fuzzer
33 A |protobuf fuzzer| that compiles valid C++ programs generated from a protobuf
34 class that describes a subset of the C++ language.
36 This fuzzer accepts clang command line options after `ignore_remaining_args=1`.
37 For example, the following command will fuzz clang with a higher optimization
42 % bin/clang-proto-fuzzer <corpus-dir> -ignore_remaining_args=1 -O3
47 A |generic fuzzer| that runs clang-format_ on C++ text fragments. Some of the
48 bugs this fuzzer has reported are `on bugzilla`__
49 and `on OSS Fuzz's tracker`__.
51 .. _clang-format: https://clang.llvm.org/docs/ClangFormat.html
52 __ https://llvm.org/pr23052
53 __ https://bugs.chromium.org/p/oss-fuzz/issues/list?q=proj-llvm+clang-format-fuzzer
58 A |generic fuzzer| that tries to parse text as :doc:`LLVM assembly <LangRef>`.
59 Some of the bugs this fuzzer has reported are `on bugzilla`__.
61 __ https://llvm.org/pr24639
66 A |generic fuzzer| that interprets inputs as object files and runs
67 :doc:`llvm-dwarfdump <CommandGuide/llvm-dwarfdump>` on them. Some of the bugs
68 this fuzzer has reported are `on OSS Fuzz's tracker`__
70 __ https://bugs.chromium.org/p/oss-fuzz/issues/list?q=proj-llvm+llvm-dwarfdump-fuzzer
75 A |generic fuzzer| for the Itanium demangler used in various LLVM tools. We've
76 fuzzed __cxa_demangle to death, why not fuzz LLVM's implementation of the same
82 A |LLVM IR fuzzer| aimed at finding bugs in instruction selection.
84 This fuzzer accepts flags after `ignore_remaining_args=1`. The flags match
85 those of :doc:`llc <CommandGuide/llc>` and the triple is required. For example,
86 the following command would fuzz AArch64 with :doc:`GlobalISel`:
90 % bin/llvm-isel-fuzzer <corpus-dir> -ignore_remaining_args=1 -mtriple aarch64 -global-isel -O0
92 Some flags can also be specified in the binary name itself in order to support
93 OSS Fuzz, which has trouble with required arguments. To do this, you can copy
94 or move ``llvm-isel-fuzzer`` to ``llvm-isel-fuzzer--x-y-z``, separating options
95 from the binary name using "--". The valid options are architecture names
96 (``aarch64``, ``x86_64``), optimization levels (``O0``, ``O2``), or specific
97 keywords, like ``gisel`` for enabling global instruction selection. In this
98 mode, the same example could be run like so:
100 .. code-block:: shell
102 % bin/llvm-isel-fuzzer--aarch64-O0-gisel <corpus-dir>
107 A |LLVM IR fuzzer| aimed at finding bugs in optimization passes.
109 It receives optimzation pipeline and runs it for each fuzzer input.
111 Interface of this fuzzer almost directly mirrors ``llvm-isel-fuzzer``. Both
112 ``mtriple`` and ``passes`` arguments are required. Passes are specified in a
113 format suitable for the new pass manager. You can find some documentation about
114 this format in the doxygen for ``PassBuilder::parsePassPipeline``.
116 .. code-block:: shell
118 % bin/llvm-opt-fuzzer <corpus-dir> -ignore_remaining_args=1 -mtriple x86_64 -passes instcombine
120 Similarly to the ``llvm-isel-fuzzer`` arguments in some predefined configurations
121 might be embedded directly into the binary file name:
123 .. code-block:: shell
125 % bin/llvm-opt-fuzzer--x86_64-instcombine <corpus-dir>
127 llvm-mc-assemble-fuzzer
128 -----------------------
130 A |generic fuzzer| that fuzzes the MC layer's assemblers by treating inputs as
131 target specific assembly.
133 Note that this fuzzer has an unusual command line interface which is not fully
134 compatible with all of libFuzzer's features. Fuzzer arguments must be passed
135 after ``--fuzzer-args``, and any ``llc`` flags must use two dashes. For
136 example, to fuzz the AArch64 assembler you might use the following command:
138 .. code-block:: console
140 llvm-mc-fuzzer --triple=aarch64-linux-gnu --fuzzer-args -max_len=4
142 This scheme will likely change in the future.
144 llvm-mc-disassemble-fuzzer
145 --------------------------
147 A |generic fuzzer| that fuzzes the MC layer's disassemblers by treating inputs
148 as assembled binary data.
150 Note that this fuzzer has an unusual command line interface which is not fully
151 compatible with all of libFuzzer's features. See the notes above about
152 ``llvm-mc-assemble-fuzzer`` for details.
155 .. |generic fuzzer| replace:: :ref:`generic fuzzer <fuzzing-llvm-generic>`
157 replace:: :ref:`libprotobuf-mutator based fuzzer <fuzzing-llvm-protobuf>`
159 replace:: :ref:`structured LLVM IR fuzzer <fuzzing-llvm-ir>`
162 Mutators and Input Generators
163 =============================
165 The inputs for a fuzz target are generated via random mutations of a
166 :ref:`corpus <libfuzzer-corpus>`. There are a few options for the kinds of
167 mutations that a fuzzer in LLVM might want.
169 .. _fuzzing-llvm-generic:
171 Generic Random Fuzzing
172 ----------------------
174 The most basic form of input mutation is to use the built in mutators of
175 LibFuzzer. These simply treat the input corpus as a bag of bits and make random
176 mutations. This type of fuzzer is good for stressing the surface layers of a
177 program, and is good at testing things like lexers, parsers, or binary
180 Some of the in-tree fuzzers that use this type of mutator are `clang-fuzzer`_,
181 `clang-format-fuzzer`_, `llvm-as-fuzzer`_, `llvm-dwarfdump-fuzzer`_,
182 `llvm-mc-assemble-fuzzer`_, and `llvm-mc-disassemble-fuzzer`_.
184 .. _fuzzing-llvm-protobuf:
186 Structured Fuzzing using ``libprotobuf-mutator``
187 ------------------------------------------------
189 We can use libprotobuf-mutator_ in order to perform structured fuzzing and
190 stress deeper layers of programs. This works by defining a protobuf class that
191 translates arbitrary data into structurally interesting input. Specifically, we
192 use this to work with a subset of the C++ language and perform mutations that
193 produce valid C++ programs in order to exercise parts of clang that are more
194 interesting than parser error handling.
196 To build this kind of fuzzer you need `protobuf`_ and its dependencies
197 installed, and you need to specify some extra flags when configuring the build
198 with :doc:`CMake <CMake>`. For example, `clang-proto-fuzzer`_ can be enabled by
199 adding ``-DCLANG_ENABLE_PROTO_FUZZER=ON`` to the flags described in
200 :ref:`building-fuzzers`.
202 The only in-tree fuzzer that uses ``libprotobuf-mutator`` today is
203 `clang-proto-fuzzer`_.
205 .. _libprotobuf-mutator: https://github.com/google/libprotobuf-mutator
206 .. _protobuf: https://github.com/google/protobuf
210 Structured Fuzzing of LLVM IR
211 -----------------------------
213 We also use a more direct form of structured fuzzing for fuzzers that take
214 :doc:`LLVM IR <LangRef>` as input. This is achieved through the ``FuzzMutate``
215 library, which was `discussed at EuroLLVM 2017`_.
217 The ``FuzzMutate`` library is used to structurally fuzz backends in
220 .. _discussed at EuroLLVM 2017: https://www.youtube.com/watch?v=UBbQ_s6hNgg
226 .. _building-fuzzers:
228 Configuring LLVM to Build Fuzzers
229 ---------------------------------
231 Fuzzers will be built and linked to libFuzzer by default as long as you build
232 LLVM with sanitizer coverage enabled. You would typically also enable at least
233 one sanitizer to find bugs faster. The most common way to build the fuzzers is
234 by adding the following two flags to your CMake invocation:
235 ``-DLLVM_USE_SANITIZER=Address -DLLVM_USE_SANITIZE_COVERAGE=On``.
237 .. note:: If you have ``compiler-rt`` checked out in an LLVM tree when building
238 with sanitizers, you'll want to specify ``-DLLVM_BUILD_RUNTIME=Off``
239 to avoid building the sanitizers themselves with sanitizers enabled.
241 .. note:: You may run into issues if you build with BFD ld, which is the
242 default linker on many unix systems. These issues are being tracked
243 in https://llvm.org/PR34636.
245 Continuously Running and Finding Bugs
246 -------------------------------------
248 There used to be a public buildbot running LLVM fuzzers continuously, and while
249 this did find issues, it didn't have a very good way to report problems in an
250 actionable way. Because of this, we're moving towards using `OSS Fuzz`_ more
253 You can browse the `LLVM project issue list`_ for the bugs found by
254 `LLVM on OSS Fuzz`_. These are also mailed to the `llvm-bugs mailing
257 .. _OSS Fuzz: https://github.com/google/oss-fuzz
258 .. _LLVM project issue list:
259 https://bugs.chromium.org/p/oss-fuzz/issues/list?q=Proj-llvm
260 .. _LLVM on OSS Fuzz:
261 https://github.com/google/oss-fuzz/blob/master/projects/llvm
262 .. _llvm-bugs mailing list:
263 http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-bugs
266 Utilities for Writing Fuzzers
267 =============================
269 There are some utilities available for writing fuzzers in LLVM.
271 Some helpers for handling the command line interface are available in
272 ``include/llvm/FuzzMutate/FuzzerCLI.h``, including functions to parse command
273 line options in a consistent way and to implement standalone main functions so
274 your fuzzer can be built and tested when not built against libFuzzer.
276 There is also some handling of the CMake config for fuzzers, where you should
277 use the ``add_llvm_fuzzer`` to set up fuzzer targets. This function works
278 similarly to functions such as ``add_llvm_tool``, but they take care of linking
279 to LibFuzzer when appropriate and can be passed the ``DUMMY_MAIN`` argument to
280 enable standalone testing.